X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdclass%2Fobd_mount.c;h=ab7b6ba2021b44a198b4bfce7962c016931db08f;hp=5be2a5e941a1e0014037d4303b587c9e2c5bf749;hb=6867d7abdb9905da98d8e90b8dc65d84a7656363;hpb=c3f22bfae4ca23664959996a7d154f311f694271 diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c index 5be2a5e..ab7b6ba 100644 --- a/lustre/obdclass/obd_mount.c +++ b/lustre/obdclass/obd_mount.c @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Intel Corporation. + * Copyright (c) 2011, 2016, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -35,207 +31,28 @@ * * lustre/obdclass/obd_mount.c * - * Client/server mount routines + * Client mount routines * * Author: Nathan Rutman */ #define DEBUG_SUBSYSTEM S_CLASS -#define D_MOUNT D_SUPER|D_CONFIG /*|D_WARNING */ +#define D_MOUNT (D_SUPER|D_CONFIG/*|D_WARNING */) #define PRINT_CMD CDEBUG -#define PRINT_MASK D_SUPER|D_CONFIG #include -#include -#include #include #include #include #include #include #include -#ifdef HAVE_KERNEL_LOCKED -#include -#endif static int (*client_fill_super)(struct super_block *sb, - struct vfsmount *mnt) = NULL; -static void (*kill_super_cb)(struct super_block *sb) = NULL; - -/*********** mount lookup *********/ - -DEFINE_MUTEX(lustre_mount_info_lock); -static CFS_LIST_HEAD(server_mount_info_list); - -static struct lustre_mount_info *server_find_mount(const char *name) -{ - cfs_list_t *tmp; - struct lustre_mount_info *lmi; - ENTRY; - - cfs_list_for_each(tmp, &server_mount_info_list) { - lmi = cfs_list_entry(tmp, struct lustre_mount_info, - lmi_list_chain); - if (strcmp(name, lmi->lmi_name) == 0) - RETURN(lmi); - } - RETURN(NULL); -} - -/* we must register an obd for a mount before we call the setup routine. - *_setup will call lustre_get_mount to get the mnt struct - by obd_name, since we can't pass the pointer to setup. */ -static int server_register_mount(const char *name, struct super_block *sb, - struct vfsmount *mnt) -{ - struct lustre_mount_info *lmi; - char *name_cp; - ENTRY; - - LASSERT(sb); - - OBD_ALLOC(lmi, sizeof(*lmi)); - if (!lmi) - RETURN(-ENOMEM); - OBD_ALLOC(name_cp, strlen(name) + 1); - if (!name_cp) { - OBD_FREE(lmi, sizeof(*lmi)); - RETURN(-ENOMEM); - } - strcpy(name_cp, name); - - mutex_lock(&lustre_mount_info_lock); - - if (server_find_mount(name)) { - mutex_unlock(&lustre_mount_info_lock); - OBD_FREE(lmi, sizeof(*lmi)); - OBD_FREE(name_cp, strlen(name) + 1); - CERROR("Already registered %s\n", name); - RETURN(-EEXIST); - } - lmi->lmi_name = name_cp; - lmi->lmi_sb = sb; - lmi->lmi_mnt = mnt; - cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list); - - mutex_unlock(&lustre_mount_info_lock); - - CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name); - - RETURN(0); -} - -/* when an obd no longer needs a mount */ -static int server_deregister_mount(const char *name) -{ - struct lustre_mount_info *lmi; - ENTRY; - - mutex_lock(&lustre_mount_info_lock); - lmi = server_find_mount(name); - if (!lmi) { - mutex_unlock(&lustre_mount_info_lock); - CERROR("%s not registered\n", name); - RETURN(-ENOENT); - } - - CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name); - - OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1); - cfs_list_del(&lmi->lmi_list_chain); - OBD_FREE(lmi, sizeof(*lmi)); - mutex_unlock(&lustre_mount_info_lock); - - RETURN(0); -} - -/* obd's look up a registered mount using their obdname. This is just - for initial obd setup to find the mount struct. It should not be - called every time you want to mntget. */ -struct lustre_mount_info *server_get_mount(const char *name) -{ - struct lustre_mount_info *lmi; - struct lustre_sb_info *lsi; - ENTRY; - - mutex_lock(&lustre_mount_info_lock); - lmi = server_find_mount(name); - mutex_unlock(&lustre_mount_info_lock); - if (!lmi) { - CERROR("Can't find mount for %s\n", name); - RETURN(NULL); - } - lsi = s2lsi(lmi->lmi_sb); - - cfs_atomic_inc(&lsi->lsi_mounts); - - CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt, - name, cfs_atomic_read(&lsi->lsi_mounts)); - - RETURN(lmi); -} -EXPORT_SYMBOL(server_get_mount); - -/* - * Used by mdt to get mount_info from obdname. - * There are no blocking when using the mount_info. - * Do not use server_get_mount for this purpose. - */ -struct lustre_mount_info *server_get_mount_2(const char *name) -{ - struct lustre_mount_info *lmi; - ENTRY; - - mutex_lock(&lustre_mount_info_lock); - lmi = server_find_mount(name); - mutex_unlock(&lustre_mount_info_lock); - if (!lmi) - CERROR("Can't find mount for %s\n", name); - - RETURN(lmi); -} -EXPORT_SYMBOL(server_get_mount_2); - -static int lustre_put_lsi(struct super_block *sb); - -/* to be called from obd_cleanup methods */ -int server_put_mount(const char *name, struct vfsmount *mnt) -{ - struct lustre_mount_info *lmi; - struct lustre_sb_info *lsi; - ENTRY; - - mutex_lock(&lustre_mount_info_lock); - lmi = server_find_mount(name); - mutex_unlock(&lustre_mount_info_lock); - if (!lmi) { - CERROR("Can't find mount for %s\n", name); - RETURN(-ENOENT); - } - lsi = s2lsi(lmi->lmi_sb); + struct vfsmount *mnt); - CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n", - lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts)); - - if (lustre_put_lsi(lmi->lmi_sb)) - CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n", - lmi->lmi_mnt, name); - - /* this obd should never need the mount again */ - server_deregister_mount(name); - - RETURN(0); -} -EXPORT_SYMBOL(server_put_mount); - -/* Corresponding to server_get_mount_2 */ -int server_put_mount_2(const char *name, struct vfsmount *mnt) -{ - ENTRY; - RETURN(0); -} -EXPORT_SYMBOL(server_put_mount_2); +static void (*kill_super_cb)(struct super_block *sb); /**************** config llog ********************/ @@ -274,9 +91,11 @@ int lustre_process_log(struct super_block *sb, char *logname, lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg)); lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb)); lcfg = lustre_cfg_new(LCFG_LOG_START, bufs); - rc = obd_process_config(mgc, sizeof(*lcfg), lcfg); - lustre_cfg_free(lcfg); - + if (lcfg == NULL) + GOTO(out, rc = -ENOMEM); + rc = obd_process_config(mgc, sizeof(*lcfg), lcfg); + lustre_cfg_free(lcfg); +out: OBD_FREE_PTR(bufs); if (rc == -EINVAL) @@ -285,8 +104,7 @@ int lustre_process_log(struct super_block *sb, char *logname, "client and the MGS are running compatible " "versions of Lustre.\n", mgc->obd_name, logname, rc); - - if (rc) + else if (rc != 0) LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' " "failed (%d). This may be the result of " "communication errors between this node and " @@ -320,6 +138,8 @@ int lustre_end_log(struct super_block *sb, char *logname, if (cfg) lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg)); lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs); + if (lcfg == NULL) + RETURN(-ENOMEM); rc = obd_process_config(mgc, sizeof(*lcfg), lcfg); lustre_cfg_free(lcfg); RETURN(rc); @@ -352,96 +172,37 @@ int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd, lustre_cfg_bufs_set_string(&bufs, 4, s4); lcfg = lustre_cfg_new(cmd, &bufs); + if (lcfg == NULL) + return -ENOMEM; lcfg->lcfg_nid = nid; rc = class_process_config(lcfg); lustre_cfg_free(lcfg); return(rc); } -EXPORT_SYMBOL(do_lcfg); /** Call class_attach and class_setup. These methods in turn call * obd type-specific methods. */ -static int lustre_start_simple(char *obdname, char *type, char *uuid, - char *s1, char *s2, char *s3, char *s4) +int lustre_start_simple(char *obdname, char *type, char *uuid, + char *s1, char *s2, char *s3, char *s4) { int rc; CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type); - rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0); + rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, NULL, NULL); if (rc) { CERROR("%s attach error %d\n", obdname, rc); - return(rc); + return rc; } rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4); if (rc) { CERROR("%s setup error %d\n", obdname, rc); - do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0); + do_lcfg(obdname, 0, LCFG_DETACH, NULL, NULL, NULL, NULL); } return rc; } -/* Set up a MGS to serve startup logs */ -static int server_start_mgs(struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - struct vfsmount *mnt = lsi->lsi_srv_mnt; - struct lustre_mount_info *lmi; - int rc = 0; - ENTRY; - - /* It is impossible to have more than 1 MGS per node, since - MGC wouldn't know which to connect to */ - lmi = server_find_mount(LUSTRE_MGS_OBDNAME); - if (lmi) { - lsi = s2lsi(lmi->lmi_sb); - LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started" - " from server\n"); - RETURN(-EALREADY); - } - - CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME); - - rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt); - - if (!rc) { - rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME, - LUSTRE_MGS_OBDNAME, 0, 0, - lsi->lsi_osd_obdname, 0); - /* Do NOT call server_deregister_mount() here. This leads to - * inability cleanup cleanly and free lsi and other stuff when - * mgs calls server_put_mount() in error handling case. -umka */ - } - - if (rc) - LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). " - "Is the 'mgs' module loaded?\n", - LUSTRE_MGS_OBDNAME, rc); - RETURN(rc); -} - -static int server_stop_mgs(struct super_block *sb) -{ - struct obd_device *obd; - int rc; - ENTRY; - - CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME); - - /* There better be only one MGS */ - obd = class_name2obd(LUSTRE_MGS_OBDNAME); - if (!obd) { - CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME); - RETURN(-EALREADY); - } - - /* The MGS should always stop when we say so */ - obd->obd_force = 1; - rc = class_manual_cleanup(obd); - RETURN(rc); -} - -DEFINE_MUTEX(mgc_start_lock); +static DEFINE_MUTEX(mgc_start_lock); /** Set up a mgc obd to process startup logs * @@ -449,7 +210,7 @@ DEFINE_MUTEX(mgc_start_lock); * * \retval 0 success, otherwise error code */ -static int lustre_start_mgc(struct super_block *sb) +int lustre_start_mgc(struct super_block *sb) { struct obd_connect_data *data = NULL; struct lustre_sb_info *lsi = s2lsi(sb); @@ -458,15 +219,16 @@ static int lustre_start_mgc(struct super_block *sb) struct obd_uuid *uuid; class_uuid_t uuidc; lnet_nid_t nid; + char nidstr[LNET_NIDSTR_SIZE]; char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL; char *ptr; - int recov_bk; - int rc = 0, i = 0, j, len; - ENTRY; + int rc = 0, i = 0, j; + size_t len; + ENTRY; - LASSERT(lsi->lsi_lmd); + LASSERT(lsi->lsi_lmd); - /* Find the first non-lo MGS nid for our MGC name */ + /* Find the first non-lo MGS nid for our MGC name */ if (IS_SERVER(lsi)) { /* mount -o mgsnode=nid */ ptr = lsi->lsi_lmd->lmd_mgs; @@ -496,12 +258,13 @@ static int lustre_start_mgc(struct super_block *sb) mutex_lock(&mgc_start_lock); - len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1; - OBD_ALLOC(mgcname, len); - OBD_ALLOC(niduuid, len + 2); - if (!mgcname || !niduuid) - GOTO(out_free, rc = -ENOMEM); - sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid)); + libcfs_nid2str_r(nid, nidstr, sizeof(nidstr)); + len = strlen(LUSTRE_MGC_OBDNAME) + strlen(nidstr) + 1; + OBD_ALLOC(mgcname, len); + OBD_ALLOC(niduuid, len + 2); + if (mgcname == NULL || niduuid == NULL) + GOTO(out_free, rc = -ENOMEM); + snprintf(mgcname, len, "%s%s", LUSTRE_MGC_OBDNAME, nidstr); mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : ""; @@ -511,6 +274,8 @@ static int lustre_start_mgc(struct super_block *sb) obd = class_name2obd(mgcname); if (obd && !obd->obd_stopping) { + int recov_bk; + rc = obd_set_info_async(NULL, obd->obd_self_export, strlen(KEY_MGSSEC), KEY_MGSSEC, strlen(mgssec), mgssec, NULL); @@ -518,7 +283,7 @@ static int lustre_start_mgc(struct super_block *sb) GOTO(out_free, rc); /* Re-using an existing MGC */ - cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount); + atomic_inc(&obd->u.cli.cl_mgc_refcount); /* IR compatibility check, only for clients */ if (lmd_is_client(lsi->lsi_lmd)) { @@ -526,9 +291,9 @@ static int lustre_start_mgc(struct super_block *sb) int vallen = sizeof(*data); __u32 *flags = &lsi->lsi_lmd->lmd_flags; - rc = obd_get_info(NULL, obd->obd_self_export, - strlen(KEY_CONN_DATA), KEY_CONN_DATA, - &vallen, data, NULL); + rc = obd_get_info(NULL, obd->obd_self_export, + strlen(KEY_CONN_DATA), KEY_CONN_DATA, + &vallen, data); LASSERT(rc == 0); has_ir = OCD_HAS_FLAG(data, IMP_RECOV); if (has_ir ^ !(*flags & LMD_FLG_NOIR)) { @@ -571,39 +336,51 @@ static int lustre_start_mgc(struct super_block *sb) /* Add the primary nids for the MGS */ i = 0; - sprintf(niduuid, "%s_%x", mgcname, i); + snprintf(niduuid, len + 2, "%s_%x", mgcname, i); if (IS_SERVER(lsi)) { ptr = lsi->lsi_lmd->lmd_mgs; + CDEBUG(D_MOUNT, "mgs nids %s.\n", ptr); if (IS_MGS(lsi)) { - /* Use local nids (including LO) */ - lnet_process_id_t id; - while ((rc = LNetGetId(i++, &id)) != -ENOENT) { - rc = do_lcfg(mgcname, id.nid, - LCFG_ADD_UUID, niduuid, 0,0,0); - } - } else { - /* Use mgsnode= nids */ + /* Use local nids (including LO) */ + lnet_process_id_t id; + while ((rc = LNetGetId(i++, &id)) != -ENOENT) { + rc = do_lcfg(mgcname, id.nid, LCFG_ADD_UUID, + niduuid, NULL, NULL, NULL); + } + } else { + /* Use mgsnode= nids */ /* mount -o mgsnode=nid */ if (lsi->lsi_lmd->lmd_mgs) { ptr = lsi->lsi_lmd->lmd_mgs; } else if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) { - CERROR("No MGS nids given.\n"); - GOTO(out_free, rc = -EINVAL); - } - while (class_parse_nid(ptr, &nid, &ptr) == 0) { - rc = do_lcfg(mgcname, nid, - LCFG_ADD_UUID, niduuid, 0,0,0); - i++; - } - } + CERROR("No MGS nids given.\n"); + GOTO(out_free, rc = -EINVAL); + } + /* + * LU-3829. + * Here we only take the first mgsnid as its primary + * serving mgs node, the rest mgsnid will be taken as + * failover mgs node, otherwise they would be takens + * as multiple nids of a single mgs node. + */ + while (class_parse_nid(ptr, &nid, &ptr) == 0) { + rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID, + niduuid, NULL, NULL, NULL); + if (rc == 0) { + i = 1; + break; + } + } + } } else { /* client */ /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */ ptr = lsi->lsi_lmd->lmd_dev; while (class_parse_nid(ptr, &nid, &ptr) == 0) { - rc = do_lcfg(mgcname, nid, - LCFG_ADD_UUID, niduuid, 0,0,0); - i++; + rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID, + niduuid, NULL, NULL, NULL); + if (rc == 0) + ++i; /* Stop at the first failover nid */ if (*ptr == ':') break; @@ -617,13 +394,16 @@ static int lustre_start_mgc(struct super_block *sb) /* Random uuid for MGC allows easier reconnects */ OBD_ALLOC_PTR(uuid); + if (uuid == NULL) + GOTO(out_free, rc = -ENOMEM); + ll_generate_random_uuid(uuidc); class_uuid_unparse(uuidc, uuid); /* Start the MGC */ rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME, - (char *)uuid->uuid, LUSTRE_MGS_OBDNAME, - niduuid, 0, 0); + (char *)uuid->uuid, LUSTRE_MGS_OBDNAME, + niduuid, NULL, NULL); OBD_FREE_PTR(uuid); if (rc) GOTO(out_free, rc); @@ -632,20 +412,22 @@ static int lustre_start_mgc(struct super_block *sb) i = 1; while (ptr && ((*ptr == ':' || class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) { - /* New failover node */ - sprintf(niduuid, "%s_%x", mgcname, i); - j = 0; + /* New failover node */ + sprintf(niduuid, "%s_%x", mgcname, i); + j = 0; while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) { - j++; - rc = do_lcfg(mgcname, nid, - LCFG_ADD_UUID, niduuid, 0,0,0); - if (*ptr == ':') - break; - } - if (j > 0) { - rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN, - niduuid, 0, 0, 0); - i++; + rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID, + niduuid, NULL, NULL, NULL); + if (rc == 0) + ++j; + if (*ptr == ':') + break; + } + if (j > 0) { + rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN, + niduuid, NULL, NULL, NULL); + if (rc == 0) + ++i; } else { /* at ":/fsname" */ break; @@ -667,27 +449,15 @@ static int lustre_start_mgc(struct super_block *sb) /* Keep a refcount of servers/clients who started with "mount", so we know when we can get rid of the mgc. */ - cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1); - - /* Try all connections, but only once. */ - recov_bk = 1; - rc = obd_set_info_async(NULL, obd->obd_self_export, - sizeof(KEY_INIT_RECOV_BACKUP), - KEY_INIT_RECOV_BACKUP, - sizeof(recov_bk), &recov_bk, NULL); - if (rc) - /* nonfatal */ - CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc); + atomic_set(&obd->u.cli.cl_mgc_refcount, 1); /* We connect to the MGS at setup, and don't disconnect until cleanup */ data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT | OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | - OBD_CONNECT_LVB_TYPE; + OBD_CONNECT_LVB_TYPE | OBD_CONNECT_BULK_MBITS; -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0) +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0) data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB; -#else -#warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab" #endif if (lmd_is_client(lsi->lsi_lmd) && @@ -722,7 +492,7 @@ static int lustre_stop_mgc(struct super_block *sb) { struct lustre_sb_info *lsi = s2lsi(sb); struct obd_device *obd; - char *niduuid = 0, *ptr = 0; + char *niduuid = NULL, *ptr = NULL; int i, rc = 0, len = 0; ENTRY; @@ -734,12 +504,12 @@ static int lustre_stop_mgc(struct super_block *sb) lsi->lsi_mgc = NULL; mutex_lock(&mgc_start_lock); - LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0); - if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) { + LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0); + if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) { /* This is not fatal, every client that stops will call in here. */ CDEBUG(D_MOUNT, "mgc still has %d references.\n", - cfs_atomic_read(&obd->u.cli.cl_mgc_refcount)); + atomic_read(&obd->u.cli.cl_mgc_refcount)); GOTO(out, rc = -EBUSY); } @@ -775,7 +545,7 @@ static int lustre_stop_mgc(struct super_block *sb) for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) { sprintf(ptr, "_%x", i); rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID, - niduuid, 0, 0, 0); + niduuid, NULL, NULL, NULL); if (rc) CERROR("del MDC UUID %s failed: rc = %d\n", niduuid, rc); @@ -789,40 +559,124 @@ out: RETURN(rc); } -/* Since there's only one mgc per node, we have to change it's fs to get - access to the right disk. */ -static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb) +/***************** lustre superblock **************/ + +static struct lustre_sb_info *lustre_init_lsi(struct super_block *sb) { - struct lustre_sb_info *lsi = s2lsi(sb); - int rc; + struct lustre_sb_info *lsi; ENTRY; - CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev); - - /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */ - rc = obd_set_info_async(NULL, mgc->obd_self_export, - sizeof(KEY_SET_FS), KEY_SET_FS, - sizeof(*sb), sb, NULL); - if (rc) { - CERROR("can't set_fs %d\n", rc); + OBD_ALLOC_PTR(lsi); + if (!lsi) + RETURN(NULL); + OBD_ALLOC_PTR(lsi->lsi_lmd); + if (!lsi->lsi_lmd) { + OBD_FREE_PTR(lsi); + RETURN(NULL); } - RETURN(rc); + lsi->lsi_lmd->lmd_exclude_count = 0; + lsi->lsi_lmd->lmd_recovery_time_soft = 0; + lsi->lsi_lmd->lmd_recovery_time_hard = 0; + s2lsi_nocast(sb) = lsi; + /* we take 1 extra ref for our setup */ + atomic_set(&lsi->lsi_mounts, 1); + + /* Default umount style */ + lsi->lsi_flags = LSI_UMOUNT_FAILOVER; + INIT_LIST_HEAD(&lsi->lsi_lwp_list); + spin_lock_init(&lsi->lsi_lwp_lock); + + RETURN(lsi); } -static int server_mgc_clear_fs(struct obd_device *mgc) +static int lustre_free_lsi(struct super_block *sb) { - int rc; - ENTRY; + struct lustre_sb_info *lsi = s2lsi(sb); + ENTRY; + + LASSERT(lsi != NULL); + CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi); + + /* someone didn't call server_put_mount. */ + LASSERT(atomic_read(&lsi->lsi_mounts) == 0); + + if (lsi->lsi_lmd != NULL) { + if (lsi->lsi_lmd->lmd_dev != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_dev, + strlen(lsi->lsi_lmd->lmd_dev) + 1); + if (lsi->lsi_lmd->lmd_profile != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_profile, + strlen(lsi->lsi_lmd->lmd_profile) + 1); + if (lsi->lsi_lmd->lmd_fileset != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_fileset, + strlen(lsi->lsi_lmd->lmd_fileset) + 1); + if (lsi->lsi_lmd->lmd_mgssec != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_mgssec, + strlen(lsi->lsi_lmd->lmd_mgssec) + 1); + if (lsi->lsi_lmd->lmd_opts != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_opts, + strlen(lsi->lsi_lmd->lmd_opts) + 1); + if (lsi->lsi_lmd->lmd_exclude_count) + OBD_FREE(lsi->lsi_lmd->lmd_exclude, + sizeof(lsi->lsi_lmd->lmd_exclude[0]) * + lsi->lsi_lmd->lmd_exclude_count); + if (lsi->lsi_lmd->lmd_mgs != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_mgs, + strlen(lsi->lsi_lmd->lmd_mgs) + 1); + if (lsi->lsi_lmd->lmd_osd_type != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_osd_type, + strlen(lsi->lsi_lmd->lmd_osd_type) + 1); + if (lsi->lsi_lmd->lmd_params != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_params, 4096); + if (lsi->lsi_lmd->lmd_nidnet != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_nidnet, + strlen(lsi->lsi_lmd->lmd_nidnet) + 1); - CDEBUG(D_MOUNT, "Unassign mgc disk\n"); + OBD_FREE_PTR(lsi->lsi_lmd); + } - rc = obd_set_info_async(NULL, mgc->obd_self_export, - sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS, - 0, NULL, NULL); - RETURN(rc); + LASSERT(lsi->lsi_llsbi == NULL); + OBD_FREE_PTR(lsi); + s2lsi_nocast(sb) = NULL; + + RETURN(0); +} + +/* The lsi has one reference for every server that is using the disk - + e.g. MDT, MGS, and potentially MGC */ +int lustre_put_lsi(struct super_block *sb) +{ + struct lustre_sb_info *lsi = s2lsi(sb); + ENTRY; + + LASSERT(lsi != NULL); + + CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts)); + if (atomic_dec_and_test(&lsi->lsi_mounts)) { + if (IS_SERVER(lsi) && lsi->lsi_osd_exp) { + lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev); + lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL; + lsi->lsi_dt_dev = NULL; + obd_disconnect(lsi->lsi_osd_exp); + /* wait till OSD is gone */ + obd_zombie_barrier(); + } + lustre_free_lsi(sb); + RETURN(1); + } + RETURN(0); } +/*** SERVER NAME *** + * + * FSNAME is between 1 and 8 characters (inclusive). + * Excluded characters are '/' and ':' + * SEPARATOR is either ':' or '-' + * TYPE: "OST", "MDT", etc. + * INDEX: Hex representation of the index + */ + /** Get the fsname ("lustre") from the server name ("lustre-OST003F"). * @param [in] svname server name including type and index * @param [out] fsname Buffer to copy filesystem name prefix into. @@ -830,24 +684,15 @@ static int server_mgc_clear_fs(struct obd_device *mgc) * @param [out] endptr if endptr isn't NULL it is set to end of fsname * rc < 0 on error */ -int server_name2fsname(char *svname, char *fsname, char **endptr) +int server_name2fsname(const char *svname, char *fsname, const char **endptr) { - char *dash = strrchr(svname, '-'); - if (!dash) { - dash = strrchr(svname, ':'); - if (!dash) - return -EINVAL; - } + const char *dash; - /* interpret -MDTXXXXX-mdc as mdt, the better way is to pass - * in the fsname, then determine the server index */ - if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) { - dash--; - for (; dash > svname && *dash != '-' && *dash != ':'; dash--) - ; - if (dash == svname) - return -EINVAL; - } + dash = svname + strnlen(svname, 8); /* max fsname length is 8 */ + for (; dash > svname && *dash != '-' && *dash != ':'; dash--) + ; + if (dash == svname) + return -EINVAL; if (fsname != NULL) { strncpy(fsname, svname, dash - svname); @@ -861,1651 +706,112 @@ int server_name2fsname(char *svname, char *fsname, char **endptr) } EXPORT_SYMBOL(server_name2fsname); -static int is_mdc_device(char *devname) +/** + * Get service name (svname) from string + * rc < 0 on error + * if endptr isn't NULL it is set to end of fsname * + */ +int server_name2svname(const char *label, char *svname, const char **endptr, + size_t svsize) { - char *ptr; - ptr = strrchr(devname, '-'); - if (ptr != NULL && strcmp(ptr, "-mdc") == 0) - return 1; - return 0; -} + int rc; + const char *dash; -static int inline tgt_is_mdt0(char *tgtname) -{ - __u32 idx; - int type; + /* We use server_name2fsname() just for parsing */ + rc = server_name2fsname(label, NULL, &dash); + if (rc != 0) + return rc; + + if (endptr != NULL) + *endptr = dash; - type = server_name2index(tgtname, &idx, NULL); - if (type != LDD_F_SV_TYPE_MDT) - return 0; + if (strlcpy(svname, dash + 1, svsize) >= svsize) + return -E2BIG; - return (idx == 0) ? 1 :0; + return 0; } +EXPORT_SYMBOL(server_name2svname); -static int inline is_mdc_for_mdt0(char *devname) +/** + * check server name is OST. + **/ +int server_name_is_ost(const char *svname) { - char *ptr; + const char *dash; + int rc; - if (!is_mdc_device(devname)) - return 0; + /* We use server_name2fsname() just for parsing */ + rc = server_name2fsname(svname, NULL, &dash); + if (rc != 0) + return rc; - ptr = strrchr(devname, '-'); - if (ptr == NULL) - return 0; + dash++; - *ptr = 0; - if (tgt_is_mdt0(devname)) { - *ptr = '-'; + if (strncmp(dash, "OST", 3) == 0) return 1; - } - *ptr = '-'; return 0; } +EXPORT_SYMBOL(server_name_is_ost); /** - * Convert OST/MDT name(fsname-OSTxxxx) to an osp name - * (fsname-MDT0000-osp-OSTxxxx), which will be used to - * communicate with MDT0 for this target. + * Get the index from the target name MDTXXXX/OSTXXXX + * rc = server type, or rc < 0 on error **/ -int tgt_name2ospname(char *svname, char *ospname) +int target_name2index(const char *tgtname, __u32 *idx, const char **endptr) { - char *fsname; - char *tgt; - int rc; - ENTRY; + const char *dash = tgtname; + unsigned long index; + int rc; - OBD_ALLOC(fsname, MTI_NAME_MAXLEN); - if (fsname == NULL) - RETURN(-ENOMEM); + if (strncmp(dash, "MDT", 3) == 0) + rc = LDD_F_SV_TYPE_MDT; + else if (strncmp(dash, "OST", 3) == 0) + rc = LDD_F_SV_TYPE_OST; + else + return -EINVAL; - rc = server_name2fsname(svname, fsname, &tgt); - if (rc != 0) { - CERROR("%s change fsname error: rc %d\n", svname, rc); - GOTO(cleanup, rc); - } + dash += 3; - if (*tgt != '-' && *tgt != ':') { - CERROR("%s wrong svname name!\n", svname); - GOTO(cleanup, rc = -EINVAL); + if (strncmp(dash, "all", 3) == 0) { + if (endptr != NULL) + *endptr = dash + 3; + return rc | LDD_F_SV_ALL; } - tgt++; - if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) { - CERROR("%s is not an OST or MDT target!\n", svname); - GOTO(cleanup, rc = -EINVAL); - } - sprintf(ospname, "%s-MDT0000-%s-%s", fsname, LUSTRE_OSP_NAME, tgt); -cleanup: - if (fsname != NULL) - OBD_FREE(fsname, MTI_NAME_MAXLEN); - RETURN(rc); + index = simple_strtoul(dash, (char **)endptr, 16); + if (idx != NULL) + *idx = index; + return rc; } -EXPORT_SYMBOL(tgt_name2ospname); +EXPORT_SYMBOL(target_name2index); -static CFS_LIST_HEAD(osp_register_list); -DEFINE_MUTEX(osp_register_list_lock); - -int lustre_register_osp_item(char *ospname, struct obd_export **exp, - register_osp_cb cb_func, void *cb_data) +/* Get the index from the obd name. + rc = server type, or + rc < 0 on error + if endptr isn't NULL it is set to end of name */ +int server_name2index(const char *svname, __u32 *idx, const char **endptr) { - struct obd_device *osp; - struct osp_register_item *ori; - ENTRY; - - LASSERTF(strlen(ospname) < MTI_NAME_MAXLEN, "ospname is too long %s\n", - ospname); - LASSERT(exp != NULL && *exp == NULL); + const char *dash; + int rc; - OBD_ALLOC_PTR(ori); - if (ori == NULL) - RETURN(-ENOMEM); - - mutex_lock(&osp_register_list_lock); - - osp = class_name2obd(ospname); - if (osp != NULL && osp->obd_set_up == 1) { - struct obd_uuid *uuid; - - OBD_ALLOC_PTR(uuid); - if (uuid == NULL) { - mutex_unlock(&osp_register_list_lock); - OBD_FREE_PTR(ori); - RETURN(-ENOMEM); - } - memcpy(uuid->uuid, ospname, strlen(ospname)); - *exp = cfs_hash_lookup(osp->obd_uuid_hash, uuid); - OBD_FREE_PTR(uuid); - } - - memcpy(ori->ori_name, ospname, strlen(ospname)); - ori->ori_exp = exp; - ori->ori_cb_func = cb_func; - ori->ori_cb_data = cb_data; - CFS_INIT_LIST_HEAD(&ori->ori_list); - cfs_list_add(&ori->ori_list, &osp_register_list); - - if (*exp != NULL && cb_func != NULL) - cb_func(cb_data); - - mutex_unlock(&osp_register_list_lock); - RETURN(0); -} -EXPORT_SYMBOL(lustre_register_osp_item); - -void lustre_deregister_osp_item(struct obd_export **exp) -{ - struct osp_register_item *ori, *tmp; - - mutex_lock(&osp_register_list_lock); - cfs_list_for_each_entry_safe(ori, tmp, &osp_register_list, ori_list) { - if (exp == ori->ori_exp) { - if (*exp) - class_export_put(*exp); - cfs_list_del(&ori->ori_list); - OBD_FREE_PTR(ori); - break; - } - } - mutex_unlock(&osp_register_list_lock); -} -EXPORT_SYMBOL(lustre_deregister_osp_item); - -static void lustre_notify_osp_list(struct obd_export *exp) -{ - struct osp_register_item *ori, *tmp; - LASSERT(exp != NULL); - - mutex_lock(&osp_register_list_lock); - cfs_list_for_each_entry_safe(ori, tmp, &osp_register_list, ori_list) { - if (strcmp(exp->exp_obd->obd_name, ori->ori_name)) - continue; - if (*ori->ori_exp != NULL) - continue; - *ori->ori_exp = class_export_get(exp); - if (ori->ori_cb_func != NULL) - ori->ori_cb_func(ori->ori_cb_data); - } - mutex_unlock(&osp_register_list_lock); -} - -static int lustre_osp_connect(struct obd_device *osp) -{ - struct lu_env env; - struct lu_context session_ctx; - struct obd_export *exp; - struct obd_uuid *uuid = NULL; - struct obd_connect_data *data = NULL; - int rc; - ENTRY; - - /* log has been fully processed, let clients connect */ - rc = lu_env_init(&env, osp->obd_lu_dev->ld_type->ldt_ctx_tags); - if (rc != 0) - RETURN(rc); - - lu_context_init(&session_ctx, LCT_SESSION); - session_ctx.lc_thread = NULL; - lu_context_enter(&session_ctx); - env.le_ses = &session_ctx; - - OBD_ALLOC_PTR(data); - if (data == NULL) - GOTO(out, rc = -ENOMEM); - - data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX; - data->ocd_version = LUSTRE_VERSION_CODE; - data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID | - OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE | - OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE | - OBD_CONNECT_LIGHTWEIGHT; - OBD_ALLOC_PTR(uuid); - if (uuid == NULL) - GOTO(out, rc = -ENOMEM); - - if (strlen(osp->obd_name) > sizeof(uuid->uuid)) { - CERROR("%s: Too long osp name %s, max_size is %d\n", - osp->obd_name, osp->obd_name, (int)sizeof(uuid->uuid)); - GOTO(out, rc = -EINVAL); - } - /* Use osp name as the uuid, so we find the export by - * osp name later */ - memcpy(uuid->uuid, osp->obd_name, strlen(osp->obd_name)); - rc = obd_connect(&env, &exp, osp, uuid, data, NULL); - if (rc != 0) - CERROR("%s: connect failed: rc = %d\n", osp->obd_name, rc); - else - lustre_notify_osp_list(exp); - -out: - if (data != NULL) - OBD_FREE_PTR(data); - if (uuid != NULL) - OBD_FREE_PTR(uuid); - - lu_env_fini(&env); - lu_context_exit(&session_ctx); - lu_context_fini(&session_ctx); - - RETURN(rc); -} - -/** - * osp-on-ost is used by slaves (Non-MDT0 targets) to manage the connection - * to MDT0. - * - * The OSTs will communicate with MDT0 by the connection established by the - * osp-on-ost to get quota and fid sequence. - * - **/ -static int lustre_osp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi) -{ - struct obd_connect_data *data = NULL; - struct obd_device *obd; - char *ospname = NULL; - char *ospuuid = NULL; - int rc; - ENTRY; - - rc = class_add_uuid(lustre_cfg_string(lcfg, 1), - lcfg->lcfg_nid); - if (rc) { - CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc); - GOTO(out, rc); - } - - OBD_ALLOC(ospname, MTI_NAME_MAXLEN); - if (ospname == NULL) - GOTO(out, rc = -ENOMEM); - - rc = tgt_name2ospname(lsi->lsi_svname, ospname); - if (rc != 0) { - CERROR("%s change ospname error: rc %d\n", - lsi->lsi_svname, rc); - GOTO(out, rc); - } - - OBD_ALLOC(ospuuid, MTI_NAME_MAXLEN); - if (ospuuid == NULL) - GOTO(out, rc = -ENOMEM); - - sprintf(ospuuid, "%s_UUID", ospname); - rc = lustre_start_simple(ospname, LUSTRE_OSP_NAME, - ospuuid, lustre_cfg_string(lcfg, 1), - 0, 0, 0); - if (rc) { - CERROR("%s: setup up failed: rc %d\n", ospname, rc); - GOTO(out, rc); - } - - obd = class_name2obd(ospname); - LASSERT(obd != NULL); - - rc = lustre_osp_connect(obd); - if (rc != 0) - CERROR("%s: connect failed: rc = %d\n", ospname, rc); -out: - if (data != NULL) - OBD_FREE_PTR(data); - if (ospname != NULL) - OBD_FREE(ospname, MTI_NAME_MAXLEN); - if (ospuuid != NULL) - OBD_FREE(ospuuid, MTI_NAME_MAXLEN); - - RETURN(rc); -} - -static int lustre_osp_add_conn(struct lustre_cfg *cfg, - struct lustre_sb_info *lsi) -{ - struct lustre_cfg_bufs *bufs = NULL; - struct lustre_cfg *lcfg = NULL; - char *ospname = NULL; - struct obd_device *osp; - int rc; - ENTRY; - - OBD_ALLOC(ospname, MTI_NAME_MAXLEN); - if (ospname == NULL) - GOTO(out, rc = -ENOMEM); - - rc = tgt_name2ospname(lsi->lsi_svname, ospname); - if (rc != 0) { - CERROR("%s change ospname error: rc %d\n", - lsi->lsi_svname, rc); - GOTO(out, rc); - } - - OBD_ALLOC_PTR(bufs); - if (bufs == NULL) - GOTO(out, rc = -ENOMEM); - - lustre_cfg_bufs_reset(bufs, ospname); - lustre_cfg_bufs_set_string(bufs, 1, - lustre_cfg_string(cfg, 1)); - - lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs); - - osp = class_name2obd(ospname); - if (osp == NULL) { - CERROR("Can not find %s\n", ospname); - GOTO(out, rc = -EINVAL); - } - - rc = class_add_conn(osp, lcfg); - if (rc) - CERROR("%s: can't add conn: rc = %d\n", ospname, rc); - -out: - if (bufs != NULL) - OBD_FREE_PTR(bufs); - if (lcfg != NULL) - lustre_cfg_free(lcfg); - if (ospname != NULL) - OBD_FREE(ospname, MTI_NAME_MAXLEN); - - RETURN(rc); -} - -/** - * Retrieve MDT nids from the client log, then start the osp-on-ost device. - * there are only two scenarios which would include mdt nid. - * 1. - * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx- - * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp - * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID - * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp - * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp - * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp - * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx - * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx- - * 2. - * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx- - * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp - * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp - * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx- -**/ -static int client_osp_config_process(const struct lu_env *env, - struct llog_handle *handle, - struct llog_rec_hdr *rec, void *data) -{ - struct config_llog_instance *clli = data; - int cfg_len = rec->lrh_len; - char *cfg_buf = (char *) (rec + 1); - struct lustre_cfg *lcfg = NULL; - struct lustre_sb_info *lsi; - int rc = 0, swab = 0; - ENTRY; - - if (rec->lrh_type != OBD_CFG_REC) { - CERROR("Unknown llog record type %#x encountered\n", - rec->lrh_type); - RETURN(-EINVAL); - } - - LASSERT(clli->cfg_sb != NULL); - lsi = s2lsi(clli->cfg_sb); - - lcfg = (struct lustre_cfg *)cfg_buf; - if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) { - lustre_swab_lustre_cfg(lcfg); - swab = 1; - } - - rc = lustre_cfg_sanity_check(cfg_buf, cfg_len); - if (rc) - GOTO(out, rc); - - switch (lcfg->lcfg_command) { - case LCFG_MARKER: { - struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1); - - lustre_swab_cfg_marker(marker, swab, - LUSTRE_CFG_BUFLEN(lcfg, 1)); - if (marker->cm_flags & CM_SKIP || - marker->cm_flags & CM_EXCLUDE) - GOTO(out, rc = 0); - - if (!tgt_is_mdt0(marker->cm_tgtname)) - GOTO(out, rc = 0); - - if(!strncmp(marker->cm_comment, "add mdc", 7) || - !strncmp(marker->cm_comment, "add failnid", 11)) { - if (marker->cm_flags & CM_START) { - clli->cfg_flags = CFG_F_MARKER; - /* This hack is to differentiate the - * ADD_UUID is come from "add mdc" record - * or from "add failnid" record. */ - if (!strncmp(marker->cm_comment, - "add failnid", 11)) - clli->cfg_flags |= CFG_F_SKIP; - } else if (marker->cm_flags & CM_END) { - clli->cfg_flags = 0; - } - } - break; - } - case LCFG_ADD_UUID: { - if (clli->cfg_flags == CFG_F_MARKER) { - rc = lustre_osp_setup(lcfg, lsi); - /* XXX: process only the first nid as - * we don't need another instance of osp */ - clli->cfg_flags |= CFG_F_SKIP; - } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) { - rc = class_add_uuid(lustre_cfg_string(lcfg, 1), - lcfg->lcfg_nid); - if (rc) - CERROR("%s: Fail to add uuid, rc:%d\n", - lsi->lsi_svname, rc); - } - break; - } - case LCFG_ADD_CONN: { - if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0))) - rc = lustre_osp_add_conn(lcfg, lsi); - break; - } - default: - break; - } -out: - RETURN(rc); -} - -static int lustre_disconnect_osp(struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - struct obd_device *osp; - char *ospname = NULL; - char *logname = NULL; - struct lustre_cfg *lcfg = NULL; - struct lustre_cfg_bufs *bufs = NULL; - struct config_llog_instance *cfg = NULL; - int rc; - ENTRY; - - LASSERT(IS_OST(lsi) || IS_MDT(lsi)); - if (IS_MDT(lsi)) { - int index; - - /* Only disconnect MDT0-osp-MDT0 here, other osp on MDT - * will be disconnect during MDT stack cleanup. - * FIXME: remove later when quota on DNE is finished */ - rc = server_name2index(lsi->lsi_svname, &index, NULL); - if (rc < 0) - RETURN(rc); - if (index != 0) - RETURN(0); - } - OBD_ALLOC(logname, MTI_NAME_MAXLEN); - if (logname == NULL) - RETURN(-ENOMEM); - - OBD_ALLOC(ospname, MTI_NAME_MAXLEN); - if (ospname == NULL) - GOTO(out, rc = -ENOMEM); - - rc = server_name2fsname(lsi->lsi_svname, ospname, NULL); - if (rc != 0) { - CERROR("%s: get fsname error: %d\n", - lsi->lsi_svname, rc); - GOTO(out, rc); - } - sprintf(logname, "%s-client", ospname); - - OBD_ALLOC_PTR(cfg); - if (cfg == NULL) - GOTO(out, rc = -ENOMEM); - - /* end log first */ - cfg->cfg_instance = sb; - rc = lustre_end_log(sb, logname, cfg); - if (rc != 0) { - CERROR("Can't end config log %s\n", ospname); - GOTO(out, rc); - } - - rc = tgt_name2ospname(lsi->lsi_svname, ospname); - if (rc != 0) { - CERROR("%s: get osp name error: %d\n", - lsi->lsi_svname, rc); - GOTO(out, rc); - } - - osp = class_name2obd(ospname); - if (osp == NULL) { - CERROR("Can't find osp-on-ost %s\n", ospname); - GOTO(out, rc = -ENOENT); - } - - OBD_ALLOC_PTR(bufs); - if (bufs == NULL) - GOTO(out, rc = -ENOMEM); - - lustre_cfg_bufs_reset(bufs, osp->obd_name); - lustre_cfg_bufs_set_string(bufs, 1, NULL); - lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs); - if (!lcfg) - GOTO(out, rc = -ENOMEM); - - /* Disconnect import first. NULL is passed for the '@env', since - * it will not be used for the 'osp-on-ost'. (see osp_shutdown()) */ - rc = osp->obd_lu_dev->ld_ops->ldo_process_config(NULL, osp->obd_lu_dev, - lcfg); -out: - if (lcfg) - lustre_cfg_free(lcfg); - if (bufs) - OBD_FREE_PTR(bufs); - if (cfg) - OBD_FREE_PTR(cfg); - if (ospname) - OBD_FREE(ospname, MTI_NAME_MAXLEN); - if (logname) - OBD_FREE(logname, MTI_NAME_MAXLEN); - RETURN(rc); -} - -/** - * Stop the osp(fsname-MDT0000-osp-OSTxxxx) or (fsname-MDT0000-osp-MDT0000) for an OST target. - **/ -static int lustre_stop_osp(struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - struct obd_device *osp = NULL; - char *ospname = NULL; - int rc = 0; - ENTRY; - - LASSERT(IS_OST(lsi) || IS_MDT(lsi)); - if (IS_MDT(lsi)) { - int index; - - /* Only disconnect MDT0-osp-MDT0 here, other osp on MDT - * will be disconnect during MDT stack cleanup. - * FIXME: remove later when quota on DNE is finished */ - rc = server_name2index(lsi->lsi_svname, &index, NULL); - if (rc < 0) - RETURN(rc); - if (index != 0) - RETURN(0); - } - - OBD_ALLOC(ospname, MTI_NAME_MAXLEN); - rc = tgt_name2ospname(lsi->lsi_svname, ospname); - if (rc != 0) { - CERROR("%s get fsname error: rc %d\n", - lsi->lsi_svname, rc); - GOTO(cleanup, rc); - } - - osp = class_name2obd(ospname); - if (osp == NULL) { - CERROR("Can not find osp-on-ost %s\n", ospname); - GOTO(cleanup, rc = -ENOENT); - } - - osp->obd_force = 1; - rc = class_manual_cleanup(osp); - -cleanup: - if (ospname != NULL) - OBD_FREE(ospname, MTI_NAME_MAXLEN); - RETURN(rc); -} - -DEFINE_MUTEX(server_start_lock); - -/* Stop MDS/OSS if nobody is using them */ -static int server_stop_servers(int lsiflags) -{ - struct obd_device *obd = NULL; - struct obd_type *type = NULL; - int rc = 0; - ENTRY; - - mutex_lock(&server_start_lock); - - /* Either an MDT or an OST or neither */ - /* if this was an MDT, and there are no more MDT's, clean up the MDS */ - if ((lsiflags & LDD_F_SV_TYPE_MDT) && - (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) { - type = class_search_type(LUSTRE_MDT_NAME); - } - /* if this was an OST, and there are no more OST's, clean up the OSS */ - if ((lsiflags & LDD_F_SV_TYPE_OST) && - (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) { - type = class_search_type(LUSTRE_OST_NAME); - } - - if (obd && (!type || !type->typ_refcnt)) { - int err; - obd->obd_force = 1; - /* obd_fail doesn't mean much on a server obd */ - err = class_manual_cleanup(obd); - if (!rc) - rc = err; - } - - mutex_unlock(&server_start_lock); - - RETURN(rc); -} - -int server_mti_print(char *title, struct mgs_target_info *mti) -{ - PRINT_CMD(PRINT_MASK, "mti %s\n", title); - PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname); - PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname); - PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid); - PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n", - mti->mti_config_ver, mti->mti_flags); - return(0); -} - -/** - * Get service name (svname) from string - * rc < 0 on error - * if endptr isn't NULL it is set to end of fsname * - */ -int server_name2svname(char *label, char *svname, char **endptr) -{ - int rc; - char *dash; - - /* We use server_name2fsname() just for parsing */ - rc = server_name2fsname(label, NULL, &dash); - if (rc != 0) - return rc; - - if (*dash != '-') - return -1; - - strncpy(svname, dash + 1, MTI_NAME_MAXLEN); - - return 0; -} -EXPORT_SYMBOL(server_name2svname); - - -/* Get the index from the obd name. - rc = server type, or - rc < 0 on error - if endptr isn't NULL it is set to end of name */ -int server_name2index(char *svname, __u32 *idx, char **endptr) -{ - unsigned long index; - int rc; - char *dash; - - /* We use server_name2fsname() just for parsing */ - rc = server_name2fsname(svname, NULL, &dash); - if (rc != 0) - return rc; - - if (*dash != '-') - return -EINVAL; - - dash++; - - if (strncmp(dash, "MDT", 3) == 0) - rc = LDD_F_SV_TYPE_MDT; - else if (strncmp(dash, "OST", 3) == 0) - rc = LDD_F_SV_TYPE_OST; - else - return -EINVAL; - - dash += 3; - - if (strcmp(dash, "all") == 0) - return rc | LDD_F_SV_ALL; - - index = simple_strtoul(dash, endptr, 16); - *idx = index; - - return rc; -} -EXPORT_SYMBOL(server_name2index); - -/* Generate data for registration */ -static int server_lsi2mti(struct lustre_sb_info *lsi, - struct mgs_target_info *mti) -{ - lnet_process_id_t id; - int rc, i = 0; - ENTRY; - - if (!IS_SERVER(lsi)) - RETURN(-EINVAL); - - strncpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname)); - - mti->mti_nid_count = 0; - while (LNetGetId(i++, &id) != -ENOENT) { - if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND) - continue; - - /* server use --servicenode param, only allow specified - * nids be registered */ - if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 && - class_match_nid(lsi->lsi_lmd->lmd_params, - PARAM_FAILNODE, id.nid) < 1) - continue; - - /* match specified network */ - if (!class_match_net(lsi->lsi_lmd->lmd_params, - PARAM_NETWORK, LNET_NIDNET(id.nid))) - continue; - - mti->mti_nids[mti->mti_nid_count] = id.nid; - mti->mti_nid_count++; - if (mti->mti_nid_count >= MTI_NIDS_MAX) { - CWARN("Only using first %d nids for %s\n", - mti->mti_nid_count, mti->mti_svname); - break; - } - } - - mti->mti_lustre_ver = LUSTRE_VERSION_CODE; - mti->mti_config_ver = 0; - - rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL); - if (rc != 0) - return rc; - - rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL); - if (rc < 0) - return rc; - /* Orion requires index to be set */ - LASSERT(!(rc & LDD_F_NEED_INDEX)); - /* keep only LDD flags */ - mti->mti_flags = lsi->lsi_flags & LDD_F_MASK; - if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN)) - mti->mti_flags |= LDD_F_UPDATE; - strncpy(mti->mti_params, lsi->lsi_lmd->lmd_params, - sizeof(mti->mti_params)); - return 0; -} - -/* Register an old or new target with the MGS. If needed MGS will construct - startup logs and assign index */ -static int server_register_target(struct lustre_sb_info *lsi) -{ - struct obd_device *mgc = lsi->lsi_mgc; - struct mgs_target_info *mti = NULL; - bool writeconf; - int rc; - ENTRY; - - LASSERT(mgc); - - if (!IS_SERVER(lsi)) - RETURN(-EINVAL); - - OBD_ALLOC_PTR(mti); - if (!mti) - RETURN(-ENOMEM); - - rc = server_lsi2mti(lsi, mti); - if (rc) - GOTO(out, rc); - - CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n", - mti->mti_svname, mti->mti_fsname, - libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index, - mti->mti_flags); - - /* if write_conf is true, the registration must succeed */ - writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE)); - mti->mti_flags |= LDD_F_OPC_REG; - - /* Register the target */ - /* FIXME use mgc_process_config instead */ - rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp, - sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET, - sizeof(*mti), mti, NULL); - if (rc) { - if (mti->mti_flags & LDD_F_ERROR) { - LCONSOLE_ERROR_MSG(0x160, - "The MGS is refusing to allow this " - "server (%s) to start. Please see messages" - " on the MGS node.\n", lsi->lsi_svname); - } else if (writeconf) { - LCONSOLE_ERROR_MSG(0x15f, - "Communication to the MGS return error %d. " - "Is the MGS running?\n", rc); - } else { - CERROR("Cannot talk to the MGS: %d, not fatal\n", rc); - /* reset the error code for non-fatal error. */ - rc = 0; - } - GOTO(out, rc); - } - -out: - if (mti) - OBD_FREE_PTR(mti); - RETURN(rc); -} - -/** - * Notify the MGS that this target is ready. - * Used by IR - if the MGS receives this message, it will notify clients. - */ -static int server_notify_target(struct super_block *sb, struct obd_device *obd) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - struct obd_device *mgc = lsi->lsi_mgc; - struct mgs_target_info *mti = NULL; - int rc; - ENTRY; - - LASSERT(mgc); - - if (!(IS_SERVER(lsi))) - RETURN(-EINVAL); - - OBD_ALLOC_PTR(mti); - if (!mti) - RETURN(-ENOMEM); - rc = server_lsi2mti(lsi, mti); - if (rc) - GOTO(out, rc); - - mti->mti_instance = obd->u.obt.obt_instance; - mti->mti_flags |= LDD_F_OPC_READY; - - /* FIXME use mgc_process_config instead */ - rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp, - sizeof(KEY_REGISTER_TARGET), - KEY_REGISTER_TARGET, - sizeof(*mti), mti, NULL); - - /* Imperative recovery: if the mgs informs us to use IR? */ - if (!rc && !(mti->mti_flags & LDD_F_ERROR) && - (mti->mti_flags & LDD_F_IR_CAPABLE)) - lsi->lsi_flags |= LDD_F_IR_CAPABLE; - -out: - if (mti) - OBD_FREE_PTR(mti); - RETURN(rc); - -} - -/** - * Start the osp(fsname-MDT0000-osp-OSTxxxx) for an OST target, - * which would be used to communicate with MDT0 for quota and FID. - **/ -static int lustre_start_osp(struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - struct config_llog_instance *cfg = NULL; - struct obd_device *osp; - char *ospname = NULL; - char *logname = NULL; - char *tgt; - int rc; - ENTRY; - - LASSERT(IS_OST(lsi) || IS_MDT(lsi)); - OBD_ALLOC(ospname, MTI_NAME_MAXLEN); - OBD_ALLOC(logname, MTI_NAME_MAXLEN); - if (ospname == NULL || logname == NULL) - GOTO(cleanup, rc = -ENOMEM); - - rc = server_name2fsname(lsi->lsi_svname, ospname, &tgt); - if (rc != 0) { - CERROR("%s change fsname error: rc %d\n", - lsi->lsi_svname, rc); - GOTO(cleanup, rc); - } - sprintf(logname, "%s-client", ospname); - - rc = tgt_name2ospname(lsi->lsi_svname, ospname); - if (rc != 0) { - CERROR("%s change ospname error: rc %d\n", - lsi->lsi_svname, rc); - GOTO(cleanup, rc); - } - - osp = class_name2obd(ospname); - if (osp != NULL) - GOTO(cleanup, rc = 0); - - OBD_ALLOC_PTR(cfg); - if (cfg == NULL) - GOTO(cleanup, rc = -ENOMEM); - - cfg->cfg_callback = client_osp_config_process; - cfg->cfg_instance = sb; - - rc = lustre_process_log(sb, logname, cfg); - -cleanup: - if (ospname != NULL) - OBD_FREE(ospname, MTI_NAME_MAXLEN); - if (logname != NULL) - OBD_FREE(logname, MTI_NAME_MAXLEN); - if (cfg != NULL) - OBD_FREE_PTR(cfg); - - RETURN(rc); -} - -/** Start server targets: MDTs and OSTs - */ -static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) -{ - struct obd_device *obd; - struct lustre_sb_info *lsi = s2lsi(sb); - struct config_llog_instance cfg; - struct lu_env env; - struct lu_device *dev; - int rc; - ENTRY; - - CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname); - - if (IS_MDT(lsi)) { - /* make sure the MDS is started */ - mutex_lock(&server_start_lock); - obd = class_name2obd(LUSTRE_MDS_OBDNAME); - if (!obd) { - rc = lustre_start_simple(LUSTRE_MDS_OBDNAME, - LUSTRE_MDS_NAME, - LUSTRE_MDS_OBDNAME"_uuid", - 0, 0, 0, 0); - if (rc) { - mutex_unlock(&server_start_lock); - CERROR("failed to start MDS: %d\n", rc); - RETURN(rc); - } - } - mutex_unlock(&server_start_lock); - } - - /* If we're an OST, make sure the global OSS is running */ - if (IS_OST(lsi)) { - /* make sure OSS is started */ - mutex_lock(&server_start_lock); - obd = class_name2obd(LUSTRE_OSS_OBDNAME); - if (!obd) { - rc = lustre_start_simple(LUSTRE_OSS_OBDNAME, - LUSTRE_OSS_NAME, - LUSTRE_OSS_OBDNAME"_uuid", - 0, 0, 0, 0); - if (rc) { - mutex_unlock(&server_start_lock); - CERROR("failed to start OSS: %d\n", rc); - RETURN(rc); - } - } - mutex_unlock(&server_start_lock); - } - - /* Set the mgc fs to our server disk. This allows the MGC to - * read and write configs locally, in case it can't talk to the MGS. */ - if (lsi->lsi_srv_mnt) { - rc = server_mgc_set_fs(lsi->lsi_mgc, sb); - if (rc) - GOTO(out_stop_service, rc); - } - - /* Register with MGS */ - rc = server_register_target(lsi); - if (rc) - GOTO(out_mgc, rc); - - /* Let the target look up the mount using the target's name - (we can't pass the sb or mnt through class_process_config.) */ - rc = server_register_mount(lsi->lsi_svname, sb, mnt); - if (rc) - GOTO(out_mgc, rc); - - /* Start targets using the llog named for the target */ - memset(&cfg, 0, sizeof(cfg)); - cfg.cfg_callback = class_config_llog_handler; - rc = lustre_process_log(sb, lsi->lsi_svname, &cfg); - if (rc) { - CERROR("failed to start server %s: %d\n", - lsi->lsi_svname, rc); - /* Do NOT call server_deregister_mount() here. This makes it - * impossible to find mount later in cleanup time and leaves - * @lsi and othder stuff leaked. -umka */ - GOTO(out_mgc, rc); - } - - obd = class_name2obd(lsi->lsi_svname); - if (!obd) { - CERROR("no server named %s was started\n", lsi->lsi_svname); - GOTO(out_mgc, rc = -ENXIO); - } - - if (IS_OST(lsi) || IS_MDT(lsi)) { - rc = lustre_start_osp(sb); - if (rc) { - CERROR("%s: failed to start OSP: %d\n", - lsi->lsi_svname, rc); - GOTO(out_mgc, rc); - } - } - - server_notify_target(sb, obd); - - /* calculate recovery timeout, do it after lustre_process_log */ - server_calc_timeout(lsi, obd); - - /* log has been fully processed */ - obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG); - - /* log has been fully processed, let clients connect */ - dev = obd->obd_lu_dev; - if (dev && dev->ld_ops->ldo_prepare) { - rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags); - if (rc == 0) { - struct lu_context session_ctx; - - lu_context_init(&session_ctx, LCT_SESSION); - session_ctx.lc_thread = NULL; - lu_context_enter(&session_ctx); - env.le_ses = &session_ctx; - - dev->ld_ops->ldo_prepare(&env, NULL, dev); - - lu_env_fini(&env); - lu_context_exit(&session_ctx); - lu_context_fini(&session_ctx); - } - } - - /* abort recovery only on the complete stack: - * many devices can be involved */ - if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) && - (OBP(obd, iocontrol))) { - obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0, - NULL, NULL); - } - -out_mgc: - /* Release the mgc fs for others to use */ - if (lsi->lsi_srv_mnt) - server_mgc_clear_fs(lsi->lsi_mgc); - -out_stop_service: - if (rc != 0) - server_stop_servers(lsi->lsi_flags); - - RETURN(rc); -} - -/***************** lustre superblock **************/ - -struct lustre_sb_info *lustre_init_lsi(struct super_block *sb) -{ - struct lustre_sb_info *lsi; - ENTRY; - - OBD_ALLOC_PTR(lsi); - if (!lsi) - RETURN(NULL); - OBD_ALLOC_PTR(lsi->lsi_lmd); - if (!lsi->lsi_lmd) { - OBD_FREE_PTR(lsi); - RETURN(NULL); - } - - lsi->lsi_lmd->lmd_exclude_count = 0; - lsi->lsi_lmd->lmd_recovery_time_soft = 0; - lsi->lsi_lmd->lmd_recovery_time_hard = 0; - s2lsi_nocast(sb) = lsi; - /* we take 1 extra ref for our setup */ - cfs_atomic_set(&lsi->lsi_mounts, 1); - - /* Default umount style */ - lsi->lsi_flags = LSI_UMOUNT_FAILOVER; - - RETURN(lsi); -} - -static int lustre_free_lsi(struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - ENTRY; - - LASSERT(lsi != NULL); - CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi); - - /* someone didn't call server_put_mount. */ - LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0); - - if (lsi->lsi_lmd != NULL) { - if (lsi->lsi_lmd->lmd_dev != NULL) - OBD_FREE(lsi->lsi_lmd->lmd_dev, - strlen(lsi->lsi_lmd->lmd_dev) + 1); - if (lsi->lsi_lmd->lmd_profile != NULL) - OBD_FREE(lsi->lsi_lmd->lmd_profile, - strlen(lsi->lsi_lmd->lmd_profile) + 1); - if (lsi->lsi_lmd->lmd_mgssec != NULL) - OBD_FREE(lsi->lsi_lmd->lmd_mgssec, - strlen(lsi->lsi_lmd->lmd_mgssec) + 1); - if (lsi->lsi_lmd->lmd_opts != NULL) - OBD_FREE(lsi->lsi_lmd->lmd_opts, - strlen(lsi->lsi_lmd->lmd_opts) + 1); - if (lsi->lsi_lmd->lmd_exclude_count) - OBD_FREE(lsi->lsi_lmd->lmd_exclude, - sizeof(lsi->lsi_lmd->lmd_exclude[0]) * - lsi->lsi_lmd->lmd_exclude_count); - if (lsi->lsi_lmd->lmd_mgs != NULL) - OBD_FREE(lsi->lsi_lmd->lmd_mgs, - strlen(lsi->lsi_lmd->lmd_mgs) + 1); - if (lsi->lsi_lmd->lmd_osd_type != NULL) - OBD_FREE(lsi->lsi_lmd->lmd_osd_type, - strlen(lsi->lsi_lmd->lmd_osd_type) + 1); - if (lsi->lsi_lmd->lmd_params != NULL) - OBD_FREE(lsi->lsi_lmd->lmd_params, 4096); - - OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd)); - } - - LASSERT(lsi->lsi_llsbi == NULL); - OBD_FREE(lsi, sizeof(*lsi)); - s2lsi_nocast(sb) = NULL; - - RETURN(0); -} - -/* The lsi has one reference for every server that is using the disk - - e.g. MDT, MGS, and potentially MGC */ -static int lustre_put_lsi(struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - ENTRY; - - LASSERT(lsi != NULL); - - CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts)); - if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) { - if (IS_SERVER(lsi) && lsi->lsi_osd_exp) { - obd_disconnect(lsi->lsi_osd_exp); - /* wait till OSD is gone */ - obd_zombie_barrier(); - } - lustre_free_lsi(sb); - RETURN(1); - } - RETURN(0); -} - -static int lsi_prepare(struct lustre_sb_info *lsi) -{ - __u32 index; - int rc; - ENTRY; - - LASSERT(lsi); - LASSERT(lsi->lsi_lmd); - - /* The server name is given as a mount line option */ - if (lsi->lsi_lmd->lmd_profile == NULL) { - LCONSOLE_ERROR("Can't determine server name\n"); - RETURN(-EINVAL); - } - - if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname)) - RETURN(-ENAMETOOLONG); - - strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile); - - /* Determine osd type */ - if (lsi->lsi_lmd->lmd_osd_type != NULL) { - if (strlen(lsi->lsi_lmd->lmd_osd_type) >= - sizeof(lsi->lsi_osd_type)) - RETURN(-ENAMETOOLONG); - - strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type); - } else { - strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME); - } - - /* XXX: a temp. solution for components using fsfilt - * to be removed in one of the subsequent patches */ - if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs")) { - strcpy(lsi->lsi_fstype, "ldiskfs"); - } else { - strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type); - } - - /* Determine server type */ - rc = server_name2index(lsi->lsi_svname, &index, NULL); - if (rc < 0) { - if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) { - /* Assume we're a bare MGS */ - rc = 0; - lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC; - } else { - LCONSOLE_ERROR("Can't determine server type of '%s'\n", - lsi->lsi_svname); - RETURN(rc); - } - } - lsi->lsi_flags |= rc; - - /* Add mount line flags that used to be in ldd: - * writeconf, mgs, anything else? - */ - lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ? - LDD_F_WRITECONF : 0; - lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ? - LDD_F_VIRGIN : 0; - lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ? - LDD_F_SV_TYPE_MGS : 0; - lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ? - LDD_F_NO_PRIMNODE : 0; - - RETURN(0); -} - -/*************** server mount ******************/ - -/** Start the shutdown of servers at umount. - */ -static void server_put_super(struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - struct obd_device *obd; - char *tmpname, *extraname = NULL; - int tmpname_sz; - int lsiflags = lsi->lsi_flags; - ENTRY; - - LASSERT(IS_SERVER(lsi)); - - tmpname_sz = strlen(lsi->lsi_svname) + 1; - OBD_ALLOC(tmpname, tmpname_sz); - memcpy(tmpname, lsi->lsi_svname, tmpname_sz); - CDEBUG(D_MOUNT, "server put_super %s\n", tmpname); - if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC)) - snprintf(tmpname, tmpname_sz, "MGS"); - - /* disconnect the osp-on-ost first to drain off the inflight request */ - if (IS_OST(lsi) || IS_MDT(lsi)) { - int rc; - - rc = lustre_disconnect_osp(sb); - if (rc && rc != ETIMEDOUT) - CERROR("%s: failed to disconnect osp-on-ost (rc=%d)!\n", - tmpname, rc); - } - - /* Stop the target */ - if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) && - (IS_MDT(lsi) || IS_OST(lsi))) { - struct lustre_profile *lprof = NULL; - - /* tell the mgc to drop the config log */ - lustre_end_log(sb, lsi->lsi_svname, NULL); - - /* COMPAT_146 - profile may get deleted in mgc_cleanup. - If there are any setup/cleanup errors, save the lov - name for safety cleanup later. */ - lprof = class_get_profile(lsi->lsi_svname); - if (lprof && lprof->lp_dt) { - OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1); - strcpy(extraname, lprof->lp_dt); - } - - obd = class_name2obd(lsi->lsi_svname); - if (obd) { - CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name); - if (lsiflags & LSI_UMOUNT_FAILOVER) - obd->obd_fail = 1; - /* We can't seem to give an error return code - * to .put_super, so we better make sure we clean up! */ - obd->obd_force = 1; - class_manual_cleanup(obd); - } else { - CERROR("no obd %s\n", lsi->lsi_svname); - server_deregister_mount(lsi->lsi_svname); - } - } - - /* If they wanted the mgs to stop separately from the mdt, they - should have put it on a different device. */ - if (IS_MGS(lsi)) { - /* if MDS start with --nomgs, don't stop MGS then */ - if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) - server_stop_mgs(sb); - } - - if (IS_OST(lsi) || IS_MDT(lsi)) { - if (lustre_stop_osp(sb) < 0) - CERROR("%s: Fail to stop osp-on-ost!\n", tmpname); - } - - /* Clean the mgc and sb */ - lustre_common_put_super(sb); - - /* wait till all in-progress cleanups are done - * specifically we're interested in ofd cleanup - * as it pins OSS */ - obd_zombie_barrier(); - - /* Stop the servers (MDS, OSS) if no longer needed. We must wait - until the target is really gone so that our type refcount check - is right. */ - server_stop_servers(lsiflags); - - /* In case of startup or cleanup err, stop related obds */ - if (extraname) { - obd = class_name2obd(extraname); - if (obd) { - CWARN("Cleaning orphaned obd %s\n", extraname); - obd->obd_force = 1; - class_manual_cleanup(obd); - } - OBD_FREE(extraname, strlen(extraname) + 1); - } - - LCONSOLE_WARN("server umount %s complete\n", tmpname); - OBD_FREE(tmpname, tmpname_sz); - EXIT; -} - -/** Called only for 'umount -f' - */ -#ifdef HAVE_UMOUNTBEGIN_VFSMOUNT -static void server_umount_begin(struct vfsmount *vfsmnt, int flags) -{ - struct super_block *sb = vfsmnt->mnt_sb; -#else -static void server_umount_begin(struct super_block *sb) -{ -#endif - struct lustre_sb_info *lsi = s2lsi(sb); - ENTRY; - -#ifdef HAVE_UMOUNTBEGIN_VFSMOUNT - if (!(flags & MNT_FORCE)) { - EXIT; - return; - } -#endif - - CDEBUG(D_MOUNT, "umount -f\n"); - /* umount = failover - umount -f = force - no third way to do non-force, non-failover */ - lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER; - EXIT; -} - -static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf) -{ - struct super_block *sb = dentry->d_sb; - struct lustre_sb_info *lsi = s2lsi(sb); - struct obd_statfs statfs; - int rc; - ENTRY; - - if (lsi->lsi_dt_dev) { - rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs); - if (rc == 0) { - statfs_unpack(buf, &statfs); - buf->f_type = sb->s_magic; - RETURN(0); - } - } - - /* just return 0 */ - buf->f_type = sb->s_magic; - buf->f_bsize = sb->s_blocksize; - buf->f_blocks = 1; - buf->f_bfree = 0; - buf->f_bavail = 0; - buf->f_files = 1; - buf->f_ffree = 0; - buf->f_namelen = NAME_MAX; - RETURN(0); -} - -/** The operations we support directly on the superblock: - * mount, umount, and df. - */ -static struct super_operations server_ops = -{ - .put_super = server_put_super, - .umount_begin = server_umount_begin, /* umount -f */ - .statfs = server_statfs, -}; - -#define log2(n) ffz(~(n)) -#define LUSTRE_SUPER_MAGIC 0x0BD00BD1 - -static int server_fill_super_common(struct super_block *sb) -{ - struct inode *root = 0; - ENTRY; - - CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev); - - sb->s_blocksize = 4096; - sb->s_blocksize_bits = log2(sb->s_blocksize); - sb->s_magic = LUSTRE_SUPER_MAGIC; - sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */ - sb->s_flags |= MS_RDONLY; - sb->s_op = &server_ops; - - root = new_inode(sb); - if (!root) { - CERROR("Can't make root inode\n"); - RETURN(-EIO); - } - - /* returns -EIO for every operation */ - /* make_bad_inode(root); -- badness - can't umount */ - /* apparently we need to be a directory for the mount to finish */ - root->i_mode = S_IFDIR; - - sb->s_root = d_make_root(root); - if (!sb->s_root) { - CERROR("%s: can't make root dentry\n", sb->s_id); - RETURN(-EIO); - } - - RETURN(0); -} - -static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags) -{ - struct lustre_mount_data *lmd = lsi->lsi_lmd; - struct obd_device *obd; - struct dt_device_param p; - char flagstr[16]; - int rc; - ENTRY; - - CDEBUG(D_MOUNT, - "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n", - lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags); - - sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname); - strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname); - strcat(lsi->lsi_osd_uuid, "_UUID"); - sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags); - - obd = class_name2obd(lsi->lsi_osd_obdname); - if (obd == NULL) { - rc = lustre_start_simple(lsi->lsi_osd_obdname, - lsi->lsi_osd_type, - lsi->lsi_osd_uuid, lmd->lmd_dev, - flagstr, lsi->lsi_lmd->lmd_opts, - lsi->lsi_svname); - if (rc) - GOTO(out, rc); - obd = class_name2obd(lsi->lsi_osd_obdname); - LASSERT(obd); - } - - rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL); - if (rc) { - obd->obd_force = 1; - class_manual_cleanup(obd); - lsi->lsi_dt_dev = NULL; - } - - /* XXX: to keep support old components relying on lsi_srv_mnt - * we get this info from OSD just started */ - LASSERT(obd->obd_lu_dev); - lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev); - LASSERT(lsi->lsi_dt_dev); - - dt_conf_get(NULL, lsi->lsi_dt_dev, &p); - - lsi->lsi_srv_mnt = p.ddp_mnt; - -out: - RETURN(rc); -} - -/** Fill in the superblock info for a Lustre server. - * Mount the device with the correct options. - * Read the on-disk config file. - * Start the services. - */ -static int server_fill_super(struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - int rc; - ENTRY; - - rc = lsi_prepare(lsi); - if (rc) - RETURN(rc); - - /* Start low level OSD */ - rc = osd_start(lsi, sb->s_flags); - if (rc) { - CERROR("Unable to start osd on %s: %d\n", - lsi->lsi_lmd->lmd_dev, rc); - lustre_put_lsi(sb); - RETURN(rc); - } - - CDEBUG(D_MOUNT, "Found service %s on device %s\n", - lsi->lsi_svname, lsi->lsi_lmd->lmd_dev); - - if (class_name2obd(lsi->lsi_svname)) { - LCONSOLE_ERROR_MSG(0x161, "The target named %s is already " - "running. Double-mount may have compromised" - " the disk journal.\n", - lsi->lsi_svname); - lustre_put_lsi(sb); - RETURN(-EALREADY); - } - - /* Start MGS before MGC */ - if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){ - rc = server_start_mgs(sb); - if (rc) - GOTO(out_mnt, rc); - } - - /* Start MGC before servers */ - rc = lustre_start_mgc(sb); - if (rc) - GOTO(out_mnt, rc); - - /* Set up all obd devices for service */ - if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) && - (IS_OST(lsi) || IS_MDT(lsi))) { - rc = server_start_targets(sb, lsi->lsi_srv_mnt); - if (rc < 0) { - CERROR("Unable to start targets: %d\n", rc); - GOTO(out_mnt, rc); - } - /* FIXME overmount client here, - or can we just start a client log and client_fill_super on this sb? - We need to make sure server_put_super gets called too - ll_put_super - calls lustre_common_put_super; check there for LSI_SERVER flag, - call s_p_s if so. - Probably should start client from new thread so we can return. - Client will not finish until all servers are connected. - Note - MGS-only server does NOT get a client, since there is no - lustre fs associated - the MGS is for all lustre fs's */ - } - - rc = server_fill_super_common(sb); - if (rc) - GOTO(out_mnt, rc); - - RETURN(0); -out_mnt: - /* We jump here in case of failure while starting targets or MGS. - * In this case we can't just put @mnt and have to do real cleanup - * with stoping targets, etc. */ - server_put_super(sb); - return rc; -} - -/* - * Calculate timeout value for a target. - */ -void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd) -{ - struct lustre_mount_data *lmd; - int soft = 0; - int hard = 0; - int factor = 0; - bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE); - int min = OBD_RECOVERY_TIME_MIN; - - LASSERT(IS_SERVER(lsi)); - - lmd = lsi->lsi_lmd; - if (lmd) { - soft = lmd->lmd_recovery_time_soft; - hard = lmd->lmd_recovery_time_hard; - has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR); - obd->obd_no_ir = !has_ir; - } - - if (soft == 0) - soft = OBD_RECOVERY_TIME_SOFT; - if (hard == 0) - hard = OBD_RECOVERY_TIME_HARD; - - /* target may have ir_factor configured. */ - factor = OBD_IR_FACTOR_DEFAULT; - if (obd->obd_recovery_ir_factor) - factor = obd->obd_recovery_ir_factor; - - if (has_ir) { - int new_soft = soft; - int new_hard = hard; - - /* adjust timeout value by imperative recovery */ - - new_soft = (soft * factor) / OBD_IR_FACTOR_MAX; - new_hard = (hard * factor) / OBD_IR_FACTOR_MAX; - - /* make sure the timeout is not too short */ - new_soft = max(min, new_soft); - new_hard = max(new_soft, new_hard); + /* We use server_name2fsname() just for parsing */ + rc = server_name2fsname(svname, NULL, &dash); + if (rc != 0) + return rc; - LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery " - "window shrunk from %d-%d down to %d-%d\n", - obd->obd_name, soft, hard, new_soft, new_hard); + dash++; + rc = target_name2index(dash, idx, endptr); + if (rc < 0) + return rc; - soft = new_soft; - hard = new_hard; - } + /* Account for -mdc after index that is possible when specifying mdt */ + if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1, + sizeof(LUSTRE_MDC_NAME)-1) == 0) + *endptr += sizeof(LUSTRE_MDC_NAME); - /* we're done */ - obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft); - obd->obd_recovery_time_hard = hard; - obd->obd_recovery_ir_factor = factor; + return rc; } -EXPORT_SYMBOL(server_calc_timeout); +EXPORT_SYMBOL(server_name2index); /*************** mount common betweeen server and client ***************/ @@ -2530,36 +836,36 @@ int lustre_common_put_super(struct super_block *sb) } /* Drop a ref to the mounted disk */ lustre_put_lsi(sb); - lu_types_stop(); - RETURN(rc); + + RETURN(rc); } EXPORT_SYMBOL(lustre_common_put_super); static void lmd_print(struct lustre_mount_data *lmd) { - int i; + int i; - PRINT_CMD(PRINT_MASK, " mount data:\n"); - if (lmd_is_client(lmd)) - PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile); - PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev); - PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags); + PRINT_CMD(D_MOUNT, " mount data:\n"); + if (lmd_is_client(lmd)) + PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile); + PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev); + PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags); - if (lmd->lmd_opts) - PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts); + if (lmd->lmd_opts) + PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts); - if (lmd->lmd_recovery_time_soft) - PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n", - lmd->lmd_recovery_time_soft); + if (lmd->lmd_recovery_time_soft) + PRINT_CMD(D_MOUNT, "recovery time soft: %d\n", + lmd->lmd_recovery_time_soft); - if (lmd->lmd_recovery_time_hard) - PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n", - lmd->lmd_recovery_time_hard); + if (lmd->lmd_recovery_time_hard) + PRINT_CMD(D_MOUNT, "recovery time hard: %d\n", + lmd->lmd_recovery_time_hard); - for (i = 0; i < lmd->lmd_exclude_count; i++) { - PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i, - lmd->lmd_exclude[i]); - } + for (i = 0; i < lmd->lmd_exclude_count; i++) { + PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i, + lmd->lmd_exclude[i]); + } } /* Is this server on the exclusion list */ @@ -2589,58 +895,61 @@ int lustre_check_exclusion(struct super_block *sb, char *svname) } /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */ -static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr) +static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr) { - char *s1 = ptr, *s2; - __u32 index, *exclude_list; - int rc = 0, devmax; - ENTRY; + const char *s1 = ptr, *s2; + __u32 *exclude_list; + __u32 index = 0; + int rc = 0, devmax; + ENTRY; - /* The shortest an ost name can be is 8 chars: -OST0000. - We don't actually know the fsname at this time, so in fact - a user could specify any fsname. */ - devmax = strlen(ptr) / 8 + 1; + /* The shortest an ost name can be is 8 chars: -OST0000. + We don't actually know the fsname at this time, so in fact + a user could specify any fsname. */ + devmax = strlen(ptr) / 8 + 1; - /* temp storage until we figure out how many we have */ - OBD_ALLOC(exclude_list, sizeof(index) * devmax); - if (!exclude_list) - RETURN(-ENOMEM); + /* temp storage until we figure out how many we have */ + OBD_ALLOC(exclude_list, sizeof(index) * devmax); + if (!exclude_list) + RETURN(-ENOMEM); - /* we enter this fn pointing at the '=' */ - while (*s1 && *s1 != ' ' && *s1 != ',') { - s1++; - rc = server_name2index(s1, &index, &s2); - if (rc < 0) { - CERROR("Can't parse server name '%s'\n", s1); - break; - } - if (rc == LDD_F_SV_TYPE_OST) - exclude_list[lmd->lmd_exclude_count++] = index; - else - CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1); - s1 = s2; - /* now we are pointing at ':' (next exclude) - or ',' (end of excludes) */ - if (lmd->lmd_exclude_count >= devmax) - break; - } - if (rc >= 0) /* non-err */ - rc = 0; - - if (lmd->lmd_exclude_count) { - /* permanent, freed in lustre_free_lsi */ - OBD_ALLOC(lmd->lmd_exclude, sizeof(index) * - lmd->lmd_exclude_count); - if (lmd->lmd_exclude) { - memcpy(lmd->lmd_exclude, exclude_list, - sizeof(index) * lmd->lmd_exclude_count); - } else { - rc = -ENOMEM; - lmd->lmd_exclude_count = 0; - } - } - OBD_FREE(exclude_list, sizeof(index) * devmax); - RETURN(rc); + /* we enter this fn pointing at the '=' */ + while (*s1 && *s1 != ' ' && *s1 != ',') { + s1++; + rc = server_name2index(s1, &index, &s2); + if (rc < 0) { + CERROR("Can't parse server name '%s': rc = %d\n", + s1, rc); + break; + } + if (rc == LDD_F_SV_TYPE_OST) + exclude_list[lmd->lmd_exclude_count++] = index; + else + CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n", + (uint)(s2-s1), s1, rc); + s1 = s2; + /* now we are pointing at ':' (next exclude) + or ',' (end of excludes) */ + if (lmd->lmd_exclude_count >= devmax) + break; + } + if (rc >= 0) /* non-err */ + rc = 0; + + if (lmd->lmd_exclude_count) { + /* permanent, freed in lustre_free_lsi */ + OBD_ALLOC(lmd->lmd_exclude, sizeof(index) * + lmd->lmd_exclude_count); + if (lmd->lmd_exclude) { + memcpy(lmd->lmd_exclude, exclude_list, + sizeof(index) * lmd->lmd_exclude_count); + } else { + rc = -ENOMEM; + lmd->lmd_exclude_count = 0; + } + } + OBD_FREE(exclude_list, sizeof(index) * devmax); + RETURN(rc); } static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr) @@ -2668,6 +977,31 @@ static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr) return 0; } +static int lmd_parse_network(struct lustre_mount_data *lmd, char *ptr) +{ + char *tail; + int length; + + if (lmd->lmd_nidnet != NULL) { + OBD_FREE(lmd->lmd_nidnet, strlen(lmd->lmd_nidnet) + 1); + lmd->lmd_nidnet = NULL; + } + + tail = strchr(ptr, ','); + if (tail == NULL) + length = strlen(ptr); + else + length = tail - ptr; + + OBD_ALLOC(lmd->lmd_nidnet, length + 1); + if (lmd->lmd_nidnet == NULL) + return -ENOMEM; + + memcpy(lmd->lmd_nidnet, ptr, length); + lmd->lmd_nidnet[length] = '\0'; + return 0; +} + static int lmd_parse_string(char **handle, char *ptr) { char *tail; @@ -2735,77 +1069,167 @@ static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr) return 0; } +/** + * Find the first delimiter (comma or colon) from the specified \a buf and + * make \a *endh point to the string starting with the delimiter. The commas + * in expression list [...] will be skipped. + * + * \param[in] buf a delimiter-separated string + * \param[in] endh a pointer to a pointer that will point to the string + * starting with the delimiter + * + * \retval 0 if delimiter is found + * \retval 1 if delimiter is not found + */ +static int lmd_find_delimiter(char *buf, char **endh) +{ + char *c = buf; + int skip = 0; + + if (buf == NULL) + return 1; + + while (*c != '\0') { + if (*c == '[') + skip++; + else if (*c == ']') + skip--; + + if ((*c == ',' || *c == ':') && skip == 0) { + if (endh != NULL) + *endh = c; + return 0; + } + + c++; + } + + return 1; +} + +/** + * Find the first valid string delimited by comma or colon from the specified + * \a buf and parse it to see whether it's a valid nid list. If yes, \a *endh + * will point to the next string starting with the delimiter. + * + * \param[in] buf a delimiter-separated string + * \param[in] endh a pointer to a pointer that will point to the string + * starting with the delimiter + * + * \retval 0 if the string is a valid nid list + * \retval 1 if the string is not a valid nid list + */ +static int lmd_parse_nidlist(char *buf, char **endh) +{ + struct list_head nidlist; + char *endp = buf; + char tmp; + int rc = 0; + + if (buf == NULL) + return 1; + while (*buf == ',' || *buf == ':') + buf++; + if (*buf == ' ' || *buf == '/' || *buf == '\0') + return 1; + + if (lmd_find_delimiter(buf, &endp) != 0) + endp = buf + strlen(buf); + + tmp = *endp; + *endp = '\0'; + + INIT_LIST_HEAD(&nidlist); + if (cfs_parse_nidlist(buf, strlen(buf), &nidlist) <= 0) + rc = 1; + cfs_free_nidlist(&nidlist); + + *endp = tmp; + if (rc != 0) + return rc; + if (endh != NULL) + *endh = endp; + return 0; +} + /** Parse mount line options * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre * dev is passed as device=uml1:/lustre by mount.lustre */ static int lmd_parse(char *options, struct lustre_mount_data *lmd) { - char *s1, *s2, *devname = NULL; - struct lustre_mount_data *raw = (struct lustre_mount_data *)options; - int rc = 0; - ENTRY; + char *s1, *s2, *devname = NULL; + struct lustre_mount_data *raw = (struct lustre_mount_data *)options; + int rc = 0; + ENTRY; - LASSERT(lmd); - if (!options) { - LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that " - "/sbin/mount.lustre is installed.\n"); - RETURN(-EINVAL); - } + LASSERT(lmd); + if (!options) { + LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that " + "/sbin/mount.lustre is installed.\n"); + RETURN(-EINVAL); + } - /* Options should be a string - try to detect old lmd data */ - if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) { - LCONSOLE_ERROR_MSG(0x163, "You're using an old version of " - "/sbin/mount.lustre. Please install " - "version %s\n", LUSTRE_VERSION_STRING); - RETURN(-EINVAL); - } - lmd->lmd_magic = LMD_MAGIC; + /* Options should be a string - try to detect old lmd data */ + if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) { + LCONSOLE_ERROR_MSG(0x163, "You're using an old version of " + "/sbin/mount.lustre. Please install " + "version %s\n", LUSTRE_VERSION_STRING); + RETURN(-EINVAL); + } + lmd->lmd_magic = LMD_MAGIC; - OBD_ALLOC(lmd->lmd_params, 4096); + OBD_ALLOC(lmd->lmd_params, LMD_PARAMS_MAXLEN); if (lmd->lmd_params == NULL) RETURN(-ENOMEM); lmd->lmd_params[0] = '\0'; - /* Set default flags here */ - - s1 = options; - while (*s1) { - int clear = 0; - int time_min = OBD_RECOVERY_TIME_MIN; - - /* Skip whitespace and extra commas */ - while (*s1 == ' ' || *s1 == ',') - s1++; - - /* Client options are parsed in ll_options: eg. flock, - user_xattr, acl */ - - /* Parse non-ldiskfs options here. Rather than modifying - ldiskfs, we just zero these out here */ - if (strncmp(s1, "abort_recov", 11) == 0) { - lmd->lmd_flags |= LMD_FLG_ABORT_RECOV; - clear++; - } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) { - lmd->lmd_recovery_time_soft = max_t(int, - simple_strtoul(s1 + 19, NULL, 10), time_min); - clear++; - } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) { - lmd->lmd_recovery_time_hard = max_t(int, - simple_strtoul(s1 + 19, NULL, 10), time_min); - clear++; - } else if (strncmp(s1, "noir", 4) == 0) { - lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */ - clear++; - } else if (strncmp(s1, "nosvc", 5) == 0) { - lmd->lmd_flags |= LMD_FLG_NOSVC; - clear++; - } else if (strncmp(s1, "nomgs", 5) == 0) { - lmd->lmd_flags |= LMD_FLG_NOMGS; - clear++; + /* Set default flags here */ + + s1 = options; + while (*s1) { + int clear = 0; + int time_min = OBD_RECOVERY_TIME_MIN; + char *s3; + + /* Skip whitespace and extra commas */ + while (*s1 == ' ' || *s1 == ',') + s1++; + s3 = s1; + + /* Client options are parsed in ll_options: eg. flock, + user_xattr, acl */ + + /* Parse non-ldiskfs options here. Rather than modifying + ldiskfs, we just zero these out here */ + if (strncmp(s1, "abort_recov", 11) == 0) { + lmd->lmd_flags |= LMD_FLG_ABORT_RECOV; + clear++; + } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) { + lmd->lmd_recovery_time_soft = + max_t(int, simple_strtoul(s1 + 19, NULL, 10), + time_min); + clear++; + } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) { + lmd->lmd_recovery_time_hard = + max_t(int, simple_strtoul(s1 + 19, NULL, 10), + time_min); + clear++; + } else if (strncmp(s1, "noir", 4) == 0) { + lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */ + clear++; + } else if (strncmp(s1, "nosvc", 5) == 0) { + lmd->lmd_flags |= LMD_FLG_NOSVC; + clear++; + } else if (strncmp(s1, "nomgs", 5) == 0) { + lmd->lmd_flags |= LMD_FLG_NOMGS; + clear++; } else if (strncmp(s1, "noscrub", 7) == 0) { lmd->lmd_flags |= LMD_FLG_NOSCRUB; clear++; + } else if (strncmp(s1, "skip_lfsck", 10) == 0) { + lmd->lmd_flags |= LMD_FLG_SKIP_LFSCK; + clear++; } else if (strncmp(s1, PARAM_MGSNODE, sizeof(PARAM_MGSNODE) - 1) == 0) { s2 = s1 + sizeof(PARAM_MGSNODE) - 1; @@ -2814,27 +1238,31 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) rc = lmd_parse_mgs(lmd, &s2); if (rc) goto invalid; + s3 = s2; + clear++; + } else if (strncmp(s1, "writeconf", 9) == 0) { + lmd->lmd_flags |= LMD_FLG_WRITECONF; + clear++; + } else if (strncmp(s1, "update", 6) == 0) { + lmd->lmd_flags |= LMD_FLG_UPDATE; clear++; - } else if (strncmp(s1, "writeconf", 9) == 0) { - lmd->lmd_flags |= LMD_FLG_WRITECONF; - clear++; } else if (strncmp(s1, "virgin", 6) == 0) { lmd->lmd_flags |= LMD_FLG_VIRGIN; clear++; } else if (strncmp(s1, "noprimnode", 10) == 0) { lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE; clear++; - } else if (strncmp(s1, "mgssec=", 7) == 0) { - rc = lmd_parse_mgssec(lmd, s1 + 7); - if (rc) - goto invalid; - clear++; - /* ost exclusion list */ - } else if (strncmp(s1, "exclude=", 8) == 0) { - rc = lmd_make_exclusion(lmd, s1 + 7); - if (rc) - goto invalid; - clear++; + } else if (strncmp(s1, "mgssec=", 7) == 0) { + rc = lmd_parse_mgssec(lmd, s1 + 7); + if (rc) + goto invalid; + clear++; + /* ost exclusion list */ + } else if (strncmp(s1, "exclude=", 8) == 0) { + rc = lmd_make_exclusion(lmd, s1 + 7); + if (rc) + goto invalid; + clear++; } else if (strncmp(s1, "mgs", 3) == 0) { /* We are an MGS */ lmd->lmd_flags |= LMD_FLG_MGS; @@ -2845,91 +1273,140 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) goto invalid; clear++; } else if (strncmp(s1, "param=", 6) == 0) { - int length; - char *tail = strchr(s1 + 6, ','); - if (tail == NULL) + size_t length, params_length; + char *tail = s1; + if (lmd_find_delimiter(s1 + 6, &tail) != 0) length = strlen(s1); - else - length = tail - s1; + else { + char *param_str = tail + 1; + int supplementary = 1; + while (lmd_parse_nidlist(param_str, + ¶m_str) == 0) { + supplementary = 0; + } + length = param_str - s1 - supplementary; + } length -= 6; + params_length = strlen(lmd->lmd_params); + if (params_length + length + 1 >= LMD_PARAMS_MAXLEN) + RETURN(-E2BIG); strncat(lmd->lmd_params, s1 + 6, length); - strcat(lmd->lmd_params, " "); + lmd->lmd_params[params_length + length] = '\0'; + strlcat(lmd->lmd_params, " ", LMD_PARAMS_MAXLEN); + s3 = s1 + 6 + length; clear++; } else if (strncmp(s1, "osd=", 4) == 0) { rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4); if (rc) goto invalid; clear++; - } - /* Linux 2.4 doesn't pass the device, so we stuck it at the - end of the options. */ - else if (strncmp(s1, "device=", 7) == 0) { - devname = s1 + 7; - /* terminate options right before device. device - must be the last one. */ - *s1 = '\0'; - break; - } + } + /* Linux 2.4 doesn't pass the device, so we stuck it at the + end of the options. */ + else if (strncmp(s1, "device=", 7) == 0) { + devname = s1 + 7; + /* terminate options right before device. device + must be the last one. */ + *s1 = '\0'; + break; + } else if (strncmp(s1, "network=", 8) == 0) { + rc = lmd_parse_network(lmd, s1 + 8); + if (rc) + goto invalid; + clear++; + } - /* Find next opt */ - s2 = strchr(s1, ','); - if (s2 == NULL) { - if (clear) - *s1 = '\0'; - break; - } - s2++; - if (clear) - memmove(s1, s2, strlen(s2) + 1); - else - s1 = s2; - } + /* Find next opt */ + s2 = strchr(s3, ','); + if (s2 == NULL) { + if (clear) + *s1 = '\0'; + break; + } + s2++; + if (clear) + memmove(s1, s2, strlen(s2) + 1); + else + s1 = s2; + } - if (!devname) { - LCONSOLE_ERROR_MSG(0x164, "Can't find the device name " - "(need mount option 'device=...')\n"); - goto invalid; - } + if (!devname) { + LCONSOLE_ERROR_MSG(0x164, "Can't find the device name " + "(need mount option 'device=...')\n"); + goto invalid; + } - s1 = strstr(devname, ":/"); - if (s1) { - ++s1; - lmd->lmd_flags |= LMD_FLG_CLIENT; - /* Remove leading /s from fsname */ - while (*++s1 == '/') ; - /* Freed in lustre_free_lsi */ - OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8); - if (!lmd->lmd_profile) - RETURN(-ENOMEM); - sprintf(lmd->lmd_profile, "%s-client", s1); - } + s1 = strstr(devname, ":/"); + if (s1) { + ++s1; + lmd->lmd_flags |= LMD_FLG_CLIENT; + /* Remove leading /s from fsname */ + while (*++s1 == '/') + ; + s2 = s1; + while (*s2 != '/' && *s2 != '\0') + s2++; + /* Freed in lustre_free_lsi */ + OBD_ALLOC(lmd->lmd_profile, s2 - s1 + 8); + if (!lmd->lmd_profile) + RETURN(-ENOMEM); - /* Freed in lustre_free_lsi */ - OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1); - if (!lmd->lmd_dev) - RETURN(-ENOMEM); - strcpy(lmd->lmd_dev, devname); - - /* Save mount options */ - s1 = options + strlen(options) - 1; - while (s1 >= options && (*s1 == ',' || *s1 == ' ')) - *s1-- = 0; - if (*options != 0) { - /* Freed in lustre_free_lsi */ - OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1); - if (!lmd->lmd_opts) - RETURN(-ENOMEM); - strcpy(lmd->lmd_opts, options); - } + strncat(lmd->lmd_profile, s1, s2 - s1); + strncat(lmd->lmd_profile, "-client", 7); + + s1 = s2; + s2 = s1 + strlen(s1) - 1; + /* Remove padding /s from fileset */ + while (*s2 == '/') + s2--; + if (s2 > s1) { + OBD_ALLOC(lmd->lmd_fileset, s2 - s1 + 2); + if (lmd->lmd_fileset == NULL) { + OBD_FREE(lmd->lmd_profile, + strlen(lmd->lmd_profile) + 1); + RETURN(-ENOMEM); + } + strncat(lmd->lmd_fileset, s1, s2 - s1 + 1); + } + } else { + /* server mount */ + if (lmd->lmd_nidnet != NULL) { + /* 'network=' mount option forbidden for server */ + OBD_FREE(lmd->lmd_nidnet, strlen(lmd->lmd_nidnet) + 1); + lmd->lmd_nidnet = NULL; + rc = -EINVAL; + CERROR("%s: option 'network=' not allowed for Lustre " + "servers: rc = %d\n", devname, rc); + RETURN(rc); + } + } - lmd_print(lmd); - lmd->lmd_magic = LMD_MAGIC; + /* Freed in lustre_free_lsi */ + OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1); + if (!lmd->lmd_dev) + RETURN(-ENOMEM); + strncpy(lmd->lmd_dev, devname, strlen(devname)+1); + + /* Save mount options */ + s1 = options + strlen(options) - 1; + while (s1 >= options && (*s1 == ',' || *s1 == ' ')) + *s1-- = 0; + if (*options != 0) { + /* Freed in lustre_free_lsi */ + OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1); + if (!lmd->lmd_opts) + RETURN(-ENOMEM); + strncpy(lmd->lmd_opts, options, strlen(options)+1); + } - RETURN(rc); + lmd_print(lmd); + lmd->lmd_magic = LMD_MAGIC; + + RETURN(rc); invalid: - CERROR("Bad mount options %s\n", options); - RETURN(-EINVAL); + CERROR("Bad mount options %s\n", options); + RETURN(-EINVAL); } struct lustre_mount_data2 { @@ -2942,7 +1419,7 @@ struct lustre_mount_data2 { * and this is where we start setting things up. * @param data Mount options (e.g. -o flock,abort_recov) */ -int lustre_fill_super(struct super_block *sb, void *data, int silent) +static int lustre_fill_super(struct super_block *sb, void *data, int silent) { struct lustre_mount_data *lmd; struct lustre_mount_data2 *lmd2 = data; @@ -2957,10 +1434,10 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent) RETURN(-ENOMEM); lmd = lsi->lsi_lmd; - /* - * Disable lockdep during mount, because mount locking patterns are - * `special'. - */ + /* + * Disable lockdep during mount, because mount locking patterns are + * `special'. + */ lockdep_off(); /* @@ -2974,33 +1451,41 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent) GOTO(out, rc = -EINVAL); } - if (lmd_is_client(lmd)) { - CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile); - if (!client_fill_super) { - LCONSOLE_ERROR_MSG(0x165, "Nothing registered for " - "client mount! Is the 'lustre' " - "module loaded?\n"); - lustre_put_lsi(sb); - rc = -ENODEV; - } else { - rc = lustre_start_mgc(sb); - if (rc) { - lustre_put_lsi(sb); - GOTO(out, rc); - } - /* Connect and start */ - /* (should always be ll_fill_super) */ - rc = (*client_fill_super)(sb, lmd2->lmd2_mnt); - /* c_f_s will call lustre_common_put_super on failure */ - } - } else { - CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev); - rc = server_fill_super(sb); - /* s_f_s calls lustre_start_mgc after the mount because we need - the MGS nids which are stored on disk. Plus, we may - need to start the MGS first. */ - /* s_f_s will call server_put_super on failure */ - } + if (lmd_is_client(lmd)) { + CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile); + if (client_fill_super == NULL) + request_module("lustre"); + if (client_fill_super == NULL) { + LCONSOLE_ERROR_MSG(0x165, "Nothing registered for " + "client mount! Is the 'lustre' " + "module loaded?\n"); + lustre_put_lsi(sb); + rc = -ENODEV; + } else { + rc = lustre_start_mgc(sb); + if (rc) { + lustre_common_put_super(sb); + GOTO(out, rc); + } + /* Connect and start */ + /* (should always be ll_fill_super) */ + rc = (*client_fill_super)(sb, lmd2->lmd2_mnt); + /* c_f_s will call lustre_common_put_super on failure */ + } + } else { +#ifdef HAVE_SERVER_SUPPORT + CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev); + rc = server_fill_super(sb); + /* s_f_s calls lustre_start_mgc after the mount because we need + the MGS nids which are stored on disk. Plus, we may + need to start the MGS first. */ + /* s_f_s will call server_put_super on failure */ +#else + CERROR("This is client-side-only module, " + "cannot handle server mount.\n"); + rc = -EINVAL; +#endif + } /* If error happens in fill_super() call, @lsi will be killed there. * This is why we do not put it here. */ @@ -3014,7 +1499,7 @@ out: lmd->lmd_dev); } lockdep_on(); - return rc; + return rc; } @@ -3035,24 +1520,29 @@ EXPORT_SYMBOL(lustre_register_kill_super_cb); /***************** FS registration ******************/ #ifdef HAVE_FSTYPE_MOUNT -struct dentry *lustre_mount(struct file_system_type *fs_type, int flags, - const char *devname, void *data) +static struct dentry *lustre_mount(struct file_system_type *fs_type, int flags, + const char *devname, void *data) { - struct lustre_mount_data2 lmd2 = { data, NULL }; + struct lustre_mount_data2 lmd2 = { + .lmd2_data = data, + }; return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super); } #else -int lustre_get_sb(struct file_system_type *fs_type, int flags, - const char *devname, void * data, struct vfsmount *mnt) +static int lustre_get_sb(struct file_system_type *fs_type, int flags, + const char *devname, void *data, struct vfsmount *mnt) { - struct lustre_mount_data2 lmd2 = { data, mnt }; + struct lustre_mount_data2 lmd2 = { + .lmd2_data = data, + .lmd2_mnt = mnt, + }; return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt); } #endif -void lustre_kill_super(struct super_block *sb) +static void lustre_kill_super(struct super_block *sb) { struct lustre_sb_info *lsi = s2lsi(sb); @@ -3064,7 +1554,7 @@ void lustre_kill_super(struct super_block *sb) /** Register the "lustre" fs type */ -struct file_system_type lustre_fs_type = { +static struct file_system_type lustre_fs_type = { .owner = THIS_MODULE, .name = "lustre", #ifdef HAVE_FSTYPE_MOUNT @@ -3073,9 +1563,9 @@ struct file_system_type lustre_fs_type = { .get_sb = lustre_get_sb, #endif .kill_sb = lustre_kill_super, - .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV | - FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE, + .fs_flags = FS_REQUIRES_DEV | FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE, }; +MODULE_ALIAS_FS("lustre"); int lustre_register_fs(void) { @@ -3086,5 +1576,3 @@ int lustre_unregister_fs(void) { return unregister_filesystem(&lustre_fs_type); } - -EXPORT_SYMBOL(server_mti_print);