Whamcloud - gitweb
landing b_cmobd_merge on HEAD
[fs/lustre-release.git] / lustre / llite / llite_lib.c
index af7cbb7..14dd91a 100644 (file)
@@ -24,6 +24,7 @@
 #define DEBUG_SUBSYSTEM S_LLITE
 
 #include <linux/module.h>
+#include <linux/types.h>
 #include <linux/random.h>
 #include <linux/version.h>
 
@@ -37,13 +38,12 @@ kmem_cache_t *ll_file_data_slab;
 
 extern struct address_space_operations ll_aops;
 extern struct address_space_operations ll_dir_aops;
-extern struct super_operations ll_super_operations;
 
 #ifndef log2
 #define log2(n) ffz(~(n))
 #endif
 
-struct ll_sb_info *lustre_init_sbi(struct super_block *sb) 
+struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
 {
         struct ll_sb_info *sbi = NULL;
         class_uuid_t uuid;
@@ -53,29 +53,53 @@ 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;
+        ll_set_sbi(sb, sbi);
 
         generate_random_uuid(uuid);
         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
         RETURN(sbi);
 }
 
-void lustre_free_sbi(struct super_block *sb) 
+void lustre_free_sbi(struct super_block *sb)
 {
         struct ll_sb_info *sbi = ll_s2sbi(sb);
         ENTRY;
 
         if (sbi != NULL)
                 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 == 0) {
+                obd_init_ea_size(sbi->ll_mdc_exp,
+                                 obd_size_diskmd(sbi->ll_osc_exp, NULL),
+                                 desc.ld_tgt_count*sizeof(struct llog_cookie));
+
+                /* declare ourself as real client. not connection
+                 * from another MDS 
+                 * FIXME: remove fake valsize, mdsize --bzzz */
+                rc = obd_set_info(sbi->ll_mdc_exp, strlen("client"),
+                                  "client", valsize, &desc);
+        }
+        RETURN(rc);
+}
+
 int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
 {
         struct inode *root = 0;
@@ -103,8 +127,6 @@ 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);
         if (err == -EBUSY) {
                 CERROR("An MDS (mdc %s) is performing recovery, of which this"
@@ -126,9 +148,10 @@ 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(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid, 
-                             strlen(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid));
+       
+        devno = get_uuid2int(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid, 
+                             strlen(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid));
+
         sb->s_dev = devno;
 
         obd = class_name2obd(osc);
@@ -149,7 +172,9 @@ 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);
@@ -159,16 +184,17 @@ int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
 
         sb->s_op = &lustre_super_operations;
 
-        /* make root inode 
+        /* 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);
@@ -192,6 +218,14 @@ 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. */
+#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);
         RETURN(err);
 
@@ -230,6 +264,8 @@ 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);
@@ -355,19 +391,6 @@ out:
         RETURN(err);
 } /* ll_read_super */
 
-void ll_put_super(struct super_block *sb)
-{
-        ENTRY;
-
-        CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
-
-        lustre_common_put_super(sb);
-
-        lustre_free_sbi(sb);
-
-        EXIT;
-} /* ll_put_super */
-
 int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
                        struct config_llog_instance *cfg, int allow_recov)
 {
@@ -385,6 +408,9 @@ int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
         int err;
         ENTRY;
 
+        if (lmd_bad_magic(lmd))
+                RETURN(-EINVAL);
+
         generate_random_uuid(uuid);
         class_uuid_unparse(uuid, &mdc_uuid);
 
@@ -392,7 +418,7 @@ int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
                 pcfg.pcfg_nal = lmd->lmd_nal;
                 pcfg.pcfg_nid = lmd->lmd_local_nid;
-                err = kportal_nal_cmd(&pcfg);
+                err = libcfs_nal_cmd(&pcfg);
                 if (err <0)
                         GOTO(out, err);
         }
@@ -405,7 +431,7 @@ int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
                 pcfg.pcfg_misc    = lmd->lmd_port;
                 pcfg.pcfg_size    = 8388608;
                 pcfg.pcfg_flags   = 0x4; /*share*/
-                err = kportal_nal_cmd(&pcfg);
+                err = libcfs_nal_cmd(&pcfg);
                 if (err <0)
                         GOTO(out, err);
         }
@@ -436,14 +462,14 @@ int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
         err = class_process_config(&lcfg);
         if (err < 0)
                 GOTO(out_detach, err);
-        
+
         obd = class_name2obd(name);
         if (obd == NULL)
                 GOTO(out_cleanup, err = -EINVAL);
 
         /* Disable initial recovery on this import */
-        err = obd_set_info(obd->obd_self_export, 
-                           strlen("initial_recov"), "initial_recov", 
+        err = obd_set_info(obd->obd_self_export,
+                           strlen("initial_recov"), "initial_recov",
                            sizeof(allow_recov), &allow_recov);
         if (err)
                 GOTO(out_cleanup, err);
@@ -453,11 +479,18 @@ int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
                 GOTO(out_cleanup, err);
         }
-        
+
         exp = class_conn2export(&mdc_conn);
-        
-        ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
+
+        ctxt = llog_get_context(&exp->exp_obd->obd_llogs, 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);
         }
@@ -489,14 +522,14 @@ out_del_conn:
                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
                 pcfg.pcfg_flags   = 1; /*share*/
-                err = kportal_nal_cmd(&pcfg);
+                err = libcfs_nal_cmd(&pcfg);
                 if (err <0)
                         GOTO(out, err);
         }
 out:
         if (rc == 0)
                 rc = err;
-        
+
         RETURN(rc);
 }
 
@@ -510,6 +543,9 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
         ENTRY;
 
         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
+        if (lmd_bad_magic(lmd))
+                RETURN(-EINVAL);
+
         sbi = lustre_init_sbi(sb);
         if (!sbi)
                 RETURN(-ENOMEM);
@@ -521,28 +557,28 @@ 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);
                 }
 
                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
-                if (sbi->ll_lmd == NULL) 
+                if (sbi->ll_lmd == NULL)
                         GOTO(out_free, err = -ENOMEM);
                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 
                 /* generate a string unique to this super, let's try
                  the address of the super itself.*/
-                len = (sizeof(sb) * 2) + 1; 
+                len = (sizeof(sb) * 2) + 1;
                 OBD_ALLOC(sbi->ll_instance, len);
-                if (sbi->ll_instance == NULL) 
+                if (sbi->ll_instance == NULL)
                         GOTO(out_free, err = -ENOMEM);
                 sprintf(sbi->ll_instance, "%p", sb);
 
                 cfg.cfg_instance = sbi->ll_instance;
                 cfg.cfg_uuid = sbi->ll_sb_uuid;
                 cfg.cfg_local_nid = lmd->lmd_local_nid;
-                err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 1);
+                err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
                 if (err < 0) {
                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 
@@ -556,13 +592,13 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
                 }
                 if (osc)
                         OBD_FREE(osc, strlen(osc) + 1);
-                OBD_ALLOC(osc, strlen(lprof->lp_osc) + 
+                OBD_ALLOC(osc, strlen(lprof->lp_osc) +
                           strlen(sbi->ll_instance) + 2);
                 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
 
                 if (mdc)
                         OBD_FREE(mdc, strlen(mdc) + 1);
-                OBD_ALLOC(mdc, strlen(lprof->lp_mdc) + 
+                OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
                           strlen(sbi->ll_instance) + 2);
                 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
         }
@@ -576,9 +612,9 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
                 CERROR("no mdc\n");
                 GOTO(out_free, err = -EINVAL);
         }
-        
+
         err = lustre_common_fill_super(sb, mdc, osc);
-        
+
         if (err)
                 GOTO(out_free, err);
 
@@ -605,9 +641,9 @@ out_free:
                         OBD_ALLOC(cln_prof, len);
                         sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 
-                        err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 
+                        err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg,
                                                  0);
-                        if (err < 0) 
+                        if (err < 0)
                                 CERROR("Unable to process log: %s\n", cln_prof);
                         OBD_FREE(cln_prof, len);
                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
@@ -619,11 +655,11 @@ out_free:
         goto out_dev;
 } /* lustre_fill_super */
 
-static void lustre_manual_cleanup(struct ll_sb_info *sbi) 
+static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 {
         struct lustre_cfg lcfg;
         struct obd_device *obd;
-        int next = 0; 
+        int next = 0;
 
         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
         {
@@ -644,7 +680,7 @@ static void lustre_manual_cleanup(struct ll_sb_info *sbi)
                 }
         }
 
-        if (sbi->ll_lmd != NULL) 
+        if (sbi->ll_lmd != NULL)
                 class_del_profile(sbi->ll_lmd->lmd_profile);
 }
 
@@ -660,7 +696,7 @@ void lustre_put_super(struct super_block *sb)
         if (obd)
                 force_umount = obd->obd_no_recov;
         obd = NULL;
-        
+
         lustre_common_put_super(sb);
 
         if (sbi->ll_lmd != NULL) {
@@ -712,7 +748,7 @@ struct inode *ll_inode_from_lock(struct ldlm_lock *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)
                 lock->l_ast_data = NULL;
@@ -735,7 +771,7 @@ void ll_clear_inode(struct inode *inode)
 
         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,
@@ -746,6 +782,13 @@ 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);
@@ -825,17 +868,17 @@ 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);
@@ -844,7 +887,7 @@ 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)) {
@@ -860,13 +903,13 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr)
                         /* from sys_utime() */
                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
                                 if (current->fsuid != inode->i_uid &&
-                                    (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
+                                    (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
                                         RETURN(rc);
                         } else {
-                               /* from inode_change_ok() */
-                               if (current->fsuid != inode->i_uid &&
-                                   !capable(CAP_FOWNER))
-                                       RETURN(-EPERM);
+                                /* from inode_change_ok() */
+                                if (current->fsuid != inode->i_uid &&
+                                    !capable(CAP_FOWNER))
+                                        RETURN(-EPERM);
                         }
                 }
 
@@ -878,11 +921,11 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr)
          * If we don't we can race with other i_size updaters on our node, like
          * ll_file_read.  We can also race with i_size propogation to other
          * nodes through dirtying and writeback of final cached pages.  This
-         * last one is especially bad for racing o_append users on other 
+         * last one is especially bad for racing o_append users on other
          * nodes. */
         if (ia_valid & ATTR_SIZE) {
-                struct ldlm_extent extent = { .start = attr->ia_size,
-                                              .end = OBD_OBJECT_EOF };
+                ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
+                                                           OBD_OBJECT_EOF } };
                 struct lustre_handle lockh = { 0 };
                 int err, ast_flags = 0;
                 /* XXX when we fix the AST intents to pass the discard-range
@@ -894,22 +937,21 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr)
                 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
                 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
                 up(&inode->i_sem);
-                rc = ll_extent_lock_no_validate(NULL, inode, lsm, LCK_PW,
-                                                &extent, &lockh, ast_flags);
+                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);
-                if (rc == 0)
-                        set_bit(LLI_F_HAVE_OST_SIZE_LOCK,
-                                &ll_i2info(inode)->lli_flags);
-
-                //ll_try_done_writing(inode);
 
+                /* We need to drop the semaphore here, because this unlock may
+                 * result in a cancellation, which will need the i_sem */
+                up(&inode->i_sem);
                 /* unlock now as we don't mind others file lockers racing with
                  * the mds updates below? */
                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
+                down(&inode->i_sem);
                 if (err) {
                         CERROR("ll_extent_unlock failed: %d\n", err);
                         if (!rc)
@@ -921,7 +963,8 @@ 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);
@@ -951,6 +994,8 @@ int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
                 RETURN(rc);
         }
 
+        osfs->os_type = sb->s_magic;
+
         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
 
@@ -1023,13 +1068,18 @@ void dump_lsm(int level, struct lov_stripe_md *lsm)
                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;
@@ -1046,10 +1096,29 @@ void ll_update_inode(struct inode *inode, struct mds_body *body,
                                 LBUG();
                         }
                 }
+                /* bug 2844 - limit i_blksize for broken user-space apps */
+                LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
+                inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
                 if (lli->lli_smd != lsm)
                         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)
@@ -1090,8 +1159,19 @@ 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(2,5,0))
+static struct backing_dev_info ll_backing_dev_info = {
+        .ra_pages       = 0,    /* No readahead */
+        .memory_backed  = 0,    /* Does contribute to dirty memory */
+};
+#endif
+
 void ll_read_inode2(struct inode *inode, void *opaque)
 {
         struct lustre_md *md = opaque;
@@ -1112,7 +1192,9 @@ 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); */
 
@@ -1131,12 +1213,34 @@ void ll_read_inode2(struct inode *inode, void *opaque)
                 EXIT;
         } else {
                 inode->i_op = &ll_special_inode_operations;
+
 #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;
 #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))
+                        inode->i_fop = &ll_special_chr_inode_fops;
+                else if (S_ISBLK(inode->i_mode))
+                        inode->i_fop = &ll_special_blk_inode_fops;
+                else if (S_ISFIFO(inode->i_mode))
+                        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,
+                      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;
         }
 }
@@ -1156,7 +1260,7 @@ 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));
@@ -1194,26 +1298,27 @@ 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);
                 }
 
@@ -1279,25 +1384,29 @@ 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);
                 }