Branch b1_4_mountconf

b=8007
vfsmount sharing

diff --git a/lustre/include/linux/lustre_disk.h b/lustre/include/linux/lustre_disk.h
index 0232eb5..870305e 100644 (file)
--- a/lustre/include/linux/lustre_disk.h
+++ b/lustre/include/linux/lustre_disk.h
@@ -28,6 +28,7 @@
 #define _LUSTRE_DISK_H
 
 #include <linux/types.h>
+
 #include <lnet/types.h>
 
 
@@ -176,7 +177,8 @@ struct lustre_sb_info {
         struct lustre_disk_data  *lsi_ldd;     /* mount info on-disk */
         //struct fsfilt_operations *lsi_fsops;
         struct ll_sb_info        *lsi_llsbi;   /* add'l client sbi info */
-        atomic_t                  lsi_mounts;  /* mount references to this sb */
+        struct vfsmount          *lsi_srv_mnt; /* the one server mount */
+        atomic_t                  lsi_mounts;  /* references to the srv_mnt */
 };
 
 #define LSI_SERVER                       0x00000001
@@ -212,7 +214,7 @@ struct lustre_mount_info {
 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb));
 void lustre_common_put_super(struct super_block *sb);
 struct lustre_mount_info *lustre_get_mount(char *name);
-int lustre_put_mount(char *name);
+int lustre_put_mount(char *name, struct vfsmount *mnt);
 int lustre_get_process_log(struct super_block *, char *, 
                            struct config_llog_instance *);
 

diff --git a/lustre/mds/handler.c b/lustre/mds/handler.c
index 0d3623f..dab146a 100644 (file)
--- a/lustre/mds/handler.c
+++ b/lustre/mds/handler.c
@@ -1561,7 +1561,7 @@ err_ns:
         obd->obd_namespace = NULL;
 err_put:
         if (lmi) {
-                lustre_put_mount(obd->obd_name);
+                lustre_put_mount(obd->obd_name, mds->mds_vfsmnt);
         } else {
                 /* old method */
                 unlock_kernel();
@@ -1752,7 +1752,8 @@ static int mds_cleanup(struct obd_device *obd)
                 CERROR("%s: mount busy, mnt_count %d != 2\n", obd->obd_name,
                        atomic_read(&obd->u.mds.mds_vfsmnt->mnt_count));
 
-        must_put = lustre_put_mount(obd->obd_name);
+        must_put = lustre_put_mount(obd->obd_name, mds->mds_vfsmnt);
+        /* must_put is for old method (l_p_m returns non-0 on err) */
 
         /* We can only unlock kernel if we are in the context of sys_ioctl,
            otherwise we never called lock_kernel */

diff --git a/lustre/mgc/mgc_request.c b/lustre/mgc/mgc_request.c
index 3ffe254..21b5136 100644 (file)
--- a/lustre/mgc/mgc_request.c
+++ b/lustre/mgc/mgc_request.c
@@ -128,7 +128,7 @@ static int mgc_cleanup(struct obd_device *obd)
         if (cli->cl_mgc_vfsmnt) {
                 /* if we're a server, eg. something's mounted */
                 mgc_fs_cleanup(obd);
-                if ((rc = lustre_put_mount(obd->obd_name)))
+                if ((rc = lustre_put_mount(obd->obd_name, cli->cl_mgc_vfsmnt)))
                      CERROR("mount_put failed %d\n", rc);
         }
 
@@ -240,26 +240,16 @@ static int mgc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                 RETURN(rc);
         }
         case OBD_IOC_START: {
-                char *conf_prof;
                 char *name = data->ioc_inlbuf1;
-                int len = strlen(name) + sizeof("-conf");
-
-                OBD_ALLOC(conf_prof, len);
-                if (!conf_prof) {
-                        CERROR("no memory\n");
-                        RETURN(-ENOMEM);
-                }
-                sprintf(conf_prof, "%s-conf", name);
+                CERROR("MGS starting config log %s\n", name);
+                /* FIXME Get llog from MGS */
 
                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
-
                 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
-                rc = class_config_parse_llog(ctxt, conf_prof, NULL);
+                rc = class_config_parse_llog(ctxt, name, NULL);
                 if (rc < 0)
-                        CERROR("Unable to process log: %s\n", conf_prof);
-
+                        CERROR("Unable to process log: %s\n", name);
                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
-                OBD_FREE(conf_prof, len);
 
                 RETURN(rc);
         }

diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c
index 69d73de..fc3a991 100644 (file)
--- a/lustre/obdclass/obd_mount.c
+++ b/lustre/obdclass/obd_mount.c
@@ -58,7 +58,9 @@ static struct lustre_mount_info *lustre_find_mount(char *name)
         return(NULL);
 }
 
-/* obd's using a mount must be preregistered so they can find it. */
+/* obd's using a mount must register for it before they call
+   their setup routine.  They call lustre_get_mount to get the mnt struct
+   by name, since we can't pass the pointer to setup. */
 int lustre_register_mount(char *name, struct super_block *sb,
                           struct vfsmount *mnt)
 {
@@ -78,6 +80,7 @@ int lustre_register_mount(char *name, struct super_block *sb,
         strcpy(name_cp, name);
 
         down(&lustre_mount_info_lock);
+        
         if (lustre_find_mount(name)) {
                 up(&lustre_mount_info_lock);
                 OBD_FREE(lmi, sizeof(*lmi));
@@ -89,6 +92,7 @@ int lustre_register_mount(char *name, struct super_block *sb,
         lmi->lmi_sb = sb;
         lmi->lmi_mnt = mnt;
         list_add(&lmi->lmi_list_chain, &lustre_mount_info_list);
+         
         up(&lustre_mount_info_lock);
         return 0;
 }
@@ -115,13 +119,15 @@ static int lustre_deregister_mount(char *name)
 }
 
 /* obd's look up a registered mount using their name. This is just
-   for initial setup; should not be called every time you want to mntget */
+   for initial obd setup to find the mount struct.  It should not be
+   called every time you want to mntget. */
 struct lustre_mount_info *lustre_get_mount(char *name)
 {
         struct lustre_mount_info *lmi;
         struct lustre_sb_info *lsi;
 
         down(&lustre_mount_info_lock);
+
         lmi = lustre_find_mount(name);
         if (!lmi) {
                 up(&lustre_mount_info_lock);
@@ -131,6 +137,7 @@ struct lustre_mount_info *lustre_get_mount(char *name)
         lsi = s2lsi(lmi->lmi_sb);
         mntget(lmi->lmi_mnt);
         atomic_inc(&lsi->lsi_mounts);
+        
         up(&lustre_mount_info_lock);
         CERROR("got mount for %s\n", name);
         return lmi;
@@ -148,7 +155,7 @@ static void unlock_mntput(struct vfsmount *mnt)
 }
 
 /* to be called from obd_cleanup methods */
-int lustre_put_mount(char *name)
+int lustre_put_mount(char *name, struct vfsmount *mnt)
 {
         struct lustre_mount_info *lmi;
         struct lustre_sb_info *lsi;
@@ -160,11 +167,12 @@ int lustre_put_mount(char *name)
                 CERROR("Can't find mount for %s\n", name);
                 return -ENOENT;
         }
+
+        CERROR("put mount for %s, #%d\n", name, atomic_read(&lsi->lsi_mounts));
         lsi = s2lsi(lmi->lmi_sb);
+        LASSERT(lmi->lmi_mnt == mnt);
         unlock_mntput(lmi->lmi_mnt);
         if (atomic_dec_and_test(&lsi->lsi_mounts)) {
-                /* The mntput from lustre_kern_mount */
-                unlock_mntput(lmi->lmi_mnt);
                 CERROR("Last put of mnt %p from %s, mount count %d\n", 
                        lmi->lmi_mnt, name, 
                        atomic_read(&lmi->lmi_mnt->mnt_count));
@@ -369,6 +377,8 @@ static int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
 static int lustre_start_simple(char *obdname, char *type, char *s1, char *s2)
 {
         int err;
+        CERROR("Starting %s\n", obdname);
+
         err = do_lcfg(obdname, 0, LCFG_ATTACH, type, obdname/*uuid*/, 0, 0);
         if (err) {
                 CERROR("%s attach error %d\n", obdname, err);
@@ -395,19 +405,17 @@ static int lustre_start_mgs(struct super_block *sb, struct vfsmount *mnt)
         sprintf(mgsname, "MGS_%p", sb);
 
         err = lustre_register_mount(mgsname, sb, mnt);
-        if (err)
-               GOTO(out_free, err);
 
-        if ((err = lustre_start_simple(mgsname, LUSTRE_MGS_NAME, 0, 0)))
-                GOTO(out_dereg, err);
-
-out_free:
+        if (!err &&
+            ((err = lustre_start_simple(mgsname, LUSTRE_MGS_NAME, 0, 0)))) 
+                lustre_deregister_mount(mgsname);
+        
         OBD_FREE(mgsname, mgsname_size);
 out:
+        if (err)                                
+                LCONSOLE_ERROR("Failed to start the MGS (%d).  Is the 'mgs' "
+                               "module loaded?\n", err);
         return err;
-out_dereg:
-        lustre_deregister_mount(mgsname);
-        goto out_free;
 }
 
 static void lustre_stop_mgs(struct super_block *sb)
@@ -452,7 +460,7 @@ static int lustre_start_mgc(struct super_block *sb, struct vfsmount *mnt)
         nid = lsi->lsi_lmd->lmd_mgsnid[0];
         err = do_lcfg(mgcname, nid, LCFG_ADD_UUID, libcfs_nid2str(nid), 0,0,0);
         if (err < 0)
-                GOTO(out_free, err);
+                GOTO(out_dereg, err);
 
         /* Start the MGC */
         if ((err = lustre_start_simple(mgcname, LUSTRE_MDC_NAME/*LUSTRE_MGC_NAME*/, "MGS", 
@@ -504,37 +512,16 @@ static void lustre_stop_mgc(struct super_block *sb)
 }
           
 /* Start targets */
-static int server_start_targets(struct super_block *sb)
+static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
 {
         struct obd_ioctl_data ioc_data = { 0 };
         struct obd_device *obd;
         struct lustre_sb_info *lsi = s2lsi(sb);
         int err;
                                                                                        
-        obd = lsi->lsi_mgc;
-        LASSERT(obd);
                                         
         CERROR("starting target %s\n", lsi->lsi_ldd->ldd_svname);
         
-        /* Get a new index if needed */
-        if (lsi->lsi_ldd->ldd_flags & LDD_F_NEED_INDEX) {
-                // FIXME implement
-                CERROR("Need new target index from MGS!\n");
-                // FIXME rewrite last_rcvd, ldd (for new svname)
-        }
-
-        /* The MGC starts targets using the svname llog */
-        ioc_data.ioc_inllen1 = strlen(lsi->lsi_ldd->ldd_svname) + 1;
-        ioc_data.ioc_inlbuf1 = lsi->lsi_ldd->ldd_svname;
-        
-        err = obd_iocontrol(OBD_IOC_START, obd->obd_self_export,
-                            sizeof ioc_data, &ioc_data, NULL);
-        if (err) {
-                CERROR("failed to start server %s: %d\n",
-                       lsi->lsi_ldd->ldd_svname, err);
-                return (err);
-        }
-                                                                                       
         /* If we're an MDT, make sure the global MDS is running */
         if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
                 /* make sure (what will be called) the MDS is started */
@@ -555,6 +542,33 @@ static int server_start_targets(struct super_block *sb)
                         err = lustre_start_simple("OSS", LUSTRE_OSS_NAME, 0, 0);
         }
 
+        /* Get a new index if needed */
+        if (lsi->lsi_ldd->ldd_flags & LDD_F_NEED_INDEX) {
+                // FIXME implement
+                CERROR("Need new target index from MGS!\n");
+                // FIXME rewrite last_rcvd, ldd (for new svname)
+        }
+
+        /* The MGC starts targets using the svname llog */
+        obd = lsi->lsi_mgc;
+        LASSERT(obd);
+
+        err = lustre_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
+        if (err) 
+                return (err);
+
+        ioc_data.ioc_inllen1 = strlen(lsi->lsi_ldd->ldd_svname) + 1;
+        ioc_data.ioc_inlbuf1 = lsi->lsi_ldd->ldd_svname;
+        
+        err = obd_iocontrol(OBD_IOC_START, obd->obd_self_export,
+                            sizeof ioc_data, &ioc_data, NULL);
+        if (err) {
+                CERROR("failed to start server %s: %d\n",
+                       lsi->lsi_ldd->ldd_svname, err);
+                lustre_deregister_mount(lsi->lsi_ldd->ldd_svname);
+                return (err);
+        }
+
         return(err);
 }
 
@@ -614,16 +628,20 @@ static struct vfsmount *lustre_kern_mount(struct super_block *sb)
         struct lustre_mount_data *lmd = lsi->lsi_lmd;
         struct vfsmount *mnt;
         char *options = NULL;
-        unsigned long page;
+        unsigned long page, s_flags;
         int err;
 
         OBD_ALLOC(ldd, sizeof(*ldd));
         if (!ldd)
                 return(ERR_PTR(-ENOMEM));
 
+        /* In the past, we have always used flags = 0.
+           Note ext3/ldiskfs can't be mounted ro. */
+        s_flags = sb->s_flags;
+
         /* Pre-mount ext3 to read the MOUNT_DATA_FILE */
         CERROR("Pre-mount ext3 %s\n", lmd->lmd_dev);
-        mnt = do_kern_mount("ext3", sb->s_flags, lmd->lmd_dev, 0);
+        mnt = do_kern_mount("ext3", s_flags, lmd->lmd_dev, 0);
         if (IS_ERR(mnt)) {
                 err = PTR_ERR(mnt);
                 CERROR("premount failed: err = %d\n", err);
@@ -661,9 +679,12 @@ static struct vfsmount *lustre_kern_mount(struct super_block *sb)
                 strcat(options, lmd->lmd_opts);
         }
 
-        CERROR("kern_mount: %s %s %s\n", MT_STR(ldd), lmd->lmd_dev, options);
+        /* Special permanent mount flags */
+        if (IS_OST(ldd)) 
+            s_flags |= MS_NOATIME | MS_NODIRATIME;
 
-        mnt = do_kern_mount(MT_STR(ldd), sb->s_flags, lmd->lmd_dev, 
+        CERROR("kern_mount: %s %s %s\n", MT_STR(ldd), lmd->lmd_dev, options);
+        mnt = do_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev, 
                             (void *)options);
         free_page(page);
         if (IS_ERR(mnt)) {
@@ -671,7 +692,7 @@ static struct vfsmount *lustre_kern_mount(struct super_block *sb)
                 CERROR("do_kern_mount failed: err = %d\n", err);
                 goto out_free;
         }
-        
+
         lsi->lsi_ldd = ldd;   /* freed at lsi cleanup */
         CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
         return(mnt);
@@ -702,7 +723,8 @@ static void server_put_super(struct super_block *sb)
         } else {
                 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
         }
-        class_del_profile(lsi->lsi_ldd->ldd_svname); /* if it exists */
+
+        //class_del_profile(lsi->lsi_ldd->ldd_svname); /* if it exists */
                                                                                        
         /* if this was an MDT, and there are no more MDT's, clean up the MDS */
         if ((lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) &&
@@ -737,9 +759,12 @@ static void server_put_super(struct super_block *sb)
            should have put it on a different device. */ 
         if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MGMT) 
                 lustre_stop_mgs(sb);
-        
+
         /* clean the mgc and sb */
         lustre_common_put_super(sb);
+        
+        /* drop the kernel mount from server_fill_super */
+        unlock_mntput(lsi->lsi_srv_mnt);
 }
 
 static void server_umount_begin(struct super_block *sb)
@@ -806,7 +831,7 @@ static int server_fill_super(struct super_block *sb)
         int mgs_service = 0, err;
         ENTRY;
 
-        /* mount to read server info */
+        /* the One True Mount */
         mnt = lustre_kern_mount(sb);
         if (IS_ERR(mnt)) {
                 err = PTR_ERR(mnt);
@@ -819,50 +844,49 @@ static int server_fill_super(struct super_block *sb)
                lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname, 
                lsi->lsi_lmd->lmd_dev);
 
-        err = lustre_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
-        if (err) 
-                GOTO(out, err);
-
         if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MGMT) {
                 CERROR("Found MGS service for fs %s on device %s\n",
                        lsi->lsi_ldd->ldd_fsname, lsi->lsi_lmd->lmd_dev);
-                mgs_service++;
                 err = lustre_start_mgs(sb, mnt);
-                if (err) 
-                       GOTO(out_dereg, err);
+                if (err) {
+                        CERROR("Ignoring failed MGS start!!\n");
+                        //GOTO(out_mnt, err);
+                } else {
+                        mgs_service++;
+                }
         }
 
         err = lustre_start_mgc(sb, mnt);
         if (err) 
-                GOTO(out_dereg, err);
+                GOTO(out_mnt, err);
 
         /* Set up all obd devices for service */
-        err = server_start_targets(sb);
+        err = server_start_targets(sb, mnt);
         if (err < 0) {
                 CERROR("Unable to process logs: %d\n", err);
-                GOTO(out_dereg, err);
+                GOTO(out_mnt, err);
         }
         
-        CERROR("Mounting server\n");
-        err = server_fill_super_common(sb);
-        if (err) 
-                GOTO(out_dereg, err);
-
         /* FIXME overmount client here,
            or can we just start a client log and client_fill_super on this sb? 
            have to fix up the s_ops after! */
-out:
-        //FIXME mntput
-        if (lsi->lsi_ldd)
-                // FIXME ??
-                class_del_profile(lsi->lsi_ldd->ldd_svname);
-        RETURN(err);
+        err = server_fill_super_common(sb);
+        if (err) 
+                GOTO(out_mnt, err);
 
-out_dereg:
+        lsi->lsi_srv_mnt = mnt;
+
+        RETURN(0);
+
+out_mnt:
         if (mgs_service)
                 lustre_stop_mgs(sb);
-        lustre_deregister_mount(lsi->lsi_ldd->ldd_svname);
-        goto out;     
+        if (lsi->lsi_mgc)
+                lustre_stop_mgc(sb);
+        unlock_mntput(mnt);
+out:
+        //if (lsi->lsi_ldd) class_del_profile(lsi->lsi_ldd->ldd_svname);
+        RETURN(err);
 }
 
 
@@ -1057,7 +1081,7 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
         } else {
                 CERROR("Mounting server\n");
                 err = server_fill_super(sb);
-                /* calls lustre_start_mgc */
+                /* calls lustre_start_mgc after the mount */
         }
                                                                                 
         if (err){

diff --git a/lustre/obdfilter/filter.c b/lustre/obdfilter/filter.c
index c7554f0..5c2e9db 100644 (file)
--- a/lustre/obdfilter/filter.c
+++ b/lustre/obdfilter/filter.c
@@ -1338,7 +1338,6 @@ int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
         lmi = lustre_get_mount(obd->obd_name);
         if (lmi) {
                 /* We already mounted in lustre_fill_super */
-                /* FIXME did we mount with the flags below?*/
                 mnt = lmi->lmi_mnt;
         } else {
                 /* old path - used by lctl */
@@ -1457,7 +1456,7 @@ err_post:
         filter_post(obd);
 err_mntput:
         if (lmi) {
-                lustre_put_mount(obd->obd_name);
+                lustre_put_mount(obd->obd_name, mnt);
         } else {
                 /* old method */
                 unlock_kernel();
@@ -1621,7 +1620,8 @@ static int filter_cleanup(struct obd_device *obd)
                        obd->obd_name, filter->fo_vfsmnt,
                        atomic_read(&filter->fo_vfsmnt->mnt_count));
 
-        must_put = lustre_put_mount(obd->obd_name);
+        must_put = lustre_put_mount(obd->obd_name, filter->fo_vfsmnt);
+        /* must_put is for old method (l_p_m returns non-0 on err) */
 
         /* We can only unlock kernel if we are in the context of sys_ioctl,
            otherwise we never called lock_kernel */

diff --git a/lustre/utils/mkfs_lustre.c b/lustre/utils/mkfs_lustre.c
index c5769b8..b143f24 100644 (file)
--- a/lustre/utils/mkfs_lustre.c
+++ b/lustre/utils/mkfs_lustre.c
@@ -136,39 +136,6 @@ int get_os_version()
         return version;
 }
 
-static int load_module(char *module_name)
-{
-        char buf[256];
-        int rc;
-        
-        vprint("loading %s\n", module_name);
-        sprintf(buf, "/sbin/modprobe %s", module_name);
-        rc = system(buf);
-        if (rc) {
-                fprintf(stderr, "%s: failed to modprobe %s (%d)\n", 
-                        progname, module_name, rc);
-                fprintf(stderr, "Check /etc/modules.conf\n");
-        }
-        return rc;
-}
-
-static int load_modules(struct mkfs_opts *mop)
-{
-        int rc = 0;
-
-        //client: rc = load_module("lustre");
-        vprint("Loading modules...");
-
-        /* portals, ksocknal, fsfilt, etc. in modules.conf */
-        rc = load_module("lustre");
-        if (rc) return rc;
-
-        /* FIXME currently use the MDT to write llogs, should be a MGS */
-        rc = load_module("mds");
-        vprint("done\n");
-        return rc;
-}
-
 //Ugly implement. FIXME 
 int run_command(char *cmd)
 {
@@ -673,8 +640,6 @@ int write_llog_files(struct mkfs_opts *mop)
         char *dev;
         int  ret = 0;
 
-        load_modules(mop);
-
         vprint("Creating Lustre logs\n"); 
         if ((ret = jt_setup()))
                 return ret;