X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fobdclass%2Fobd_mount.c;h=890d409737d09a72837ce47cfcb7d60d3d26fe0c;hb=921fa0f21c1722e127fd072089ee35ddb0135653;hp=b27034450caf6c08ab13122e2ac4826f00e31038;hpb=1cbf25caab01ac561bb35053ca34fb6eb7336b3d;p=fs%2Flustre-release.git diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c index b270344..890d409 100644 --- a/lustre/obdclass/obd_mount.c +++ b/lustre/obdclass/obd_mount.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -28,6 +26,8 @@ /* * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -56,12 +56,13 @@ #include #include -static int (*client_fill_super)(struct super_block *sb) = NULL; +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 *********/ -CFS_DECLARE_MUTEX(lustre_mount_info_lock); +CFS_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) @@ -102,10 +103,10 @@ static int server_register_mount(const char *name, struct super_block *sb, } strcpy(name_cp, name); - cfs_down(&lustre_mount_info_lock); + cfs_mutex_lock(&lustre_mount_info_lock); if (server_find_mount(name)) { - cfs_up(&lustre_mount_info_lock); + cfs_mutex_unlock(&lustre_mount_info_lock); OBD_FREE(lmi, sizeof(*lmi)); OBD_FREE(name_cp, strlen(name) + 1); CERROR("Already registered %s\n", name); @@ -116,10 +117,10 @@ static int server_register_mount(const char *name, struct super_block *sb, lmi->lmi_mnt = mnt; cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list); - cfs_up(&lustre_mount_info_lock); + cfs_mutex_unlock(&lustre_mount_info_lock); CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n", - lmi->lmi_mnt, name, cfs_atomic_read(&lmi->lmi_mnt->mnt_count)); + lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt)); RETURN(0); } @@ -130,21 +131,21 @@ static int server_deregister_mount(const char *name) struct lustre_mount_info *lmi; ENTRY; - cfs_down(&lustre_mount_info_lock); + cfs_mutex_lock(&lustre_mount_info_lock); lmi = server_find_mount(name); if (!lmi) { - cfs_up(&lustre_mount_info_lock); + cfs_mutex_unlock(&lustre_mount_info_lock); CERROR("%s not registered\n", name); RETURN(-ENOENT); } CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n", - lmi->lmi_mnt, name, cfs_atomic_read(&lmi->lmi_mnt->mnt_count)); + lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt)); OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1); cfs_list_del(&lmi->lmi_list_chain); OBD_FREE(lmi, sizeof(*lmi)); - cfs_up(&lustre_mount_info_lock); + cfs_mutex_unlock(&lustre_mount_info_lock); RETURN(0); } @@ -158,9 +159,9 @@ struct lustre_mount_info *server_get_mount(const char *name) struct lustre_sb_info *lsi; ENTRY; - cfs_down(&lustre_mount_info_lock); + cfs_mutex_lock(&lustre_mount_info_lock); lmi = server_find_mount(name); - cfs_up(&lustre_mount_info_lock); + cfs_mutex_unlock(&lustre_mount_info_lock); if (!lmi) { CERROR("Can't find mount for %s\n", name); RETURN(NULL); @@ -171,7 +172,7 @@ struct lustre_mount_info *server_get_mount(const char *name) CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n", lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts), - cfs_atomic_read(&lmi->lmi_mnt->mnt_count)); + mnt_get_count(lmi->lmi_mnt)); RETURN(lmi); } @@ -186,9 +187,9 @@ struct lustre_mount_info *server_get_mount_2(const char *name) struct lustre_mount_info *lmi; ENTRY; - cfs_down(&lustre_mount_info_lock); + cfs_mutex_lock(&lustre_mount_info_lock); lmi = server_find_mount(name); - cfs_up(&lustre_mount_info_lock); + cfs_mutex_unlock(&lustre_mount_info_lock); if (!lmi) CERROR("Can't find mount for %s\n", name); @@ -197,13 +198,18 @@ struct lustre_mount_info *server_get_mount_2(const char *name) static void unlock_mntput(struct vfsmount *mnt) { - if (kernel_locked()) { - cfs_unlock_kernel(); - mntput(mnt); - cfs_lock_kernel(); - } else { - mntput(mnt); - } +#ifdef HAVE_KERNEL_LOCKED + /* for kernel < 2.6.37 */ + if (kernel_locked()) { + unlock_kernel(); + mntput(mnt); + lock_kernel(); + } else { + mntput(mnt); + } +#else + mntput(mnt); +#endif } static int lustre_put_lsi(struct super_block *sb); @@ -213,15 +219,15 @@ int server_put_mount(const char *name, struct vfsmount *mnt) { struct lustre_mount_info *lmi; struct lustre_sb_info *lsi; - int count = atomic_read(&mnt->mnt_count) - 1; + int count = mnt_get_count(mnt) - 1; ENTRY; /* This might be the last one, can't deref after this */ unlock_mntput(mnt); - cfs_down(&lustre_mount_info_lock); + cfs_mutex_lock(&lustre_mount_info_lock); lmi = server_find_mount(name); - cfs_up(&lustre_mount_info_lock); + cfs_mutex_unlock(&lustre_mount_info_lock); if (!lmi) { CERROR("Can't find mount for %s\n", name); RETURN(-ENOENT); @@ -351,7 +357,7 @@ static int ldd_write(struct lvfs_run_ctxt *mount_ctxt, push_ctxt(&saved, mount_ctxt, NULL); - file = filp_open(MOUNT_DATA_FILE, O_RDWR, 0644); + file = filp_open(MOUNT_DATA_FILE, O_RDWR|O_SYNC, 0644); if (IS_ERR(file)) { rc = PTR_ERR(file); CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc); @@ -393,7 +399,7 @@ int lustre_process_log(struct super_block *sb, char *logname, struct config_llog_instance *cfg) { struct lustre_cfg *lcfg; - struct lustre_cfg_bufs bufs; + struct lustre_cfg_bufs *bufs; struct lustre_sb_info *lsi = s2lsi(sb); struct obd_device *mgc = lsi->lsi_mgc; int rc; @@ -402,15 +408,21 @@ int lustre_process_log(struct super_block *sb, char *logname, LASSERT(mgc); LASSERT(cfg); + OBD_ALLOC_PTR(bufs); + if (bufs == NULL) + RETURN(-ENOMEM); + /* mgc_process_config */ - lustre_cfg_bufs_reset(&bufs, mgc->obd_name); - lustre_cfg_bufs_set_string(&bufs, 1, 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); + lustre_cfg_bufs_reset(bufs, mgc->obd_name); + lustre_cfg_bufs_set_string(bufs, 1, 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); + OBD_FREE_PTR(bufs); + if (rc == -EINVAL) LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'" "failed from the MGS (%d). Make sure this " @@ -571,7 +583,7 @@ static int server_stop_mgs(struct super_block *sb) RETURN(rc); } -CFS_DECLARE_MUTEX(mgc_start_lock); +CFS_DEFINE_MUTEX(mgc_start_lock); /** Set up a mgc obd to process startup logs * @@ -588,7 +600,7 @@ static int lustre_start_mgc(struct super_block *sb) struct obd_uuid *uuid; class_uuid_t uuidc; lnet_nid_t nid; - char *mgcname, *niduuid, *mgssec; + char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL; char *ptr; int recov_bk; int rc = 0, i = 0, j, len; @@ -624,6 +636,8 @@ static int lustre_start_mgc(struct super_block *sb) RETURN(-EINVAL); } + cfs_mutex_lock(&mgc_start_lock); + len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1; OBD_ALLOC(mgcname, len); OBD_ALLOC(niduuid, len + 2); @@ -633,11 +647,13 @@ static int lustre_start_mgc(struct super_block *sb) mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : ""; - cfs_mutex_down(&mgc_start_lock); + OBD_ALLOC_PTR(data); + if (data == NULL) + GOTO(out_free, rc = -ENOMEM); obd = class_name2obd(mgcname); if (obd && !obd->obd_stopping) { - rc = obd_set_info_async(obd->obd_self_export, + rc = obd_set_info_async(NULL, obd->obd_self_export, strlen(KEY_MGSSEC), KEY_MGSSEC, strlen(mgssec), mgssec, NULL); if (rc) @@ -646,6 +662,32 @@ static int lustre_start_mgc(struct super_block *sb) /* Re-using an existing MGC */ cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount); + /* IR compatibility check, only for clients */ + if (lmd_is_client(lsi->lsi_lmd)) { + int has_ir; + 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); + LASSERT(rc == 0); + has_ir = OCD_HAS_FLAG(data, IMP_RECOV); + if (has_ir ^ !(*flags & LMD_FLG_NOIR)) { + /* LMD_FLG_NOIR is for test purpose only */ + LCONSOLE_WARN( + "Trying to mount a client with IR setting " + "not compatible with current mgc. " + "Force to use current mgc setting that is " + "IR %s.\n", + has_ir ? "enabled" : "disabled"); + if (has_ir) + *flags &= ~LMD_FLG_NOIR; + else + *flags |= LMD_FLG_NOIR; + } + } + recov_bk = 0; /* If we are restarting the MGS, don't try to keep the MGC's old connection, or registration will fail. */ @@ -660,7 +702,7 @@ static int lustre_start_mgc(struct super_block *sb) if at all possible. */ recov_bk++; CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk); - rc = obd_set_info_async(obd->obd_self_export, + 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); @@ -755,7 +797,7 @@ static int lustre_start_mgc(struct super_block *sb) GOTO(out_free, rc = -ENOTCONN); } - rc = obd_set_info_async(obd->obd_self_export, + rc = obd_set_info_async(NULL, obd->obd_self_export, strlen(KEY_MGSSEC), KEY_MGSSEC, strlen(mgssec), mgssec, NULL); if (rc) @@ -767,7 +809,7 @@ static int lustre_start_mgc(struct super_block *sb) /* Try all connections, but only once. */ recov_bk = 1; - rc = obd_set_info_async(obd->obd_self_export, + 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); @@ -775,14 +817,14 @@ static int lustre_start_mgc(struct super_block *sb) /* nonfatal */ CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc); /* We connect to the MGS at setup, and don't disconnect until cleanup */ - OBD_ALLOC_PTR(data); - if (data == NULL) - GOTO(out, rc = -ENOMEM); data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID | - OBD_CONNECT_AT | OBD_CONNECT_FULL20; + OBD_CONNECT_AT | OBD_CONNECT_FULL20 | + OBD_CONNECT_IMP_RECOV; + if (lmd_is_client(lsi->lsi_lmd) && + lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR) + data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV; data->ocd_version = LUSTRE_VERSION_CODE; rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL); - OBD_FREE_PTR(data); if (rc) { CERROR("connect failed %d\n", rc); GOTO(out, rc); @@ -795,8 +837,10 @@ out: to the same mgc.*/ lsi->lsi_mgc = obd; out_free: - cfs_mutex_up(&mgc_start_lock); + cfs_mutex_unlock(&mgc_start_lock); + if (data) + OBD_FREE_PTR(data); if (mgcname) OBD_FREE(mgcname, len); if (niduuid) @@ -819,7 +863,7 @@ static int lustre_stop_mgc(struct super_block *sb) RETURN(-ENOENT); lsi->lsi_mgc = NULL; - cfs_mutex_down(&mgc_start_lock); + cfs_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)) { /* This is not fatal, every client that stops @@ -871,7 +915,7 @@ out: OBD_FREE(niduuid, len); /* class_import_put will get rid of the additional connections */ - cfs_mutex_up(&mgc_start_lock); + cfs_mutex_unlock(&mgc_start_lock); RETURN(rc); } @@ -886,7 +930,7 @@ static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb) 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(mgc->obd_self_export, + rc = obd_set_info_async(NULL, mgc->obd_self_export, sizeof(KEY_SET_FS), KEY_SET_FS, sizeof(*sb), sb, NULL); if (rc) { @@ -903,13 +947,13 @@ static int server_mgc_clear_fs(struct obd_device *mgc) CDEBUG(D_MOUNT, "Unassign mgc disk\n"); - rc = obd_set_info_async(mgc->obd_self_export, + rc = obd_set_info_async(NULL, mgc->obd_self_export, sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS, 0, NULL, NULL); RETURN(rc); } -CFS_DECLARE_MUTEX(server_start_lock); +CFS_DEFINE_MUTEX(server_start_lock); /* Stop MDS/OSS if nobody is using them */ static int server_stop_servers(int lddflags, int lsiflags) @@ -919,7 +963,7 @@ static int server_stop_servers(int lddflags, int lsiflags) int rc = 0; ENTRY; - cfs_mutex_down(&server_start_lock); + cfs_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 */ @@ -943,7 +987,7 @@ static int server_stop_servers(int lddflags, int lsiflags) rc = err; } - cfs_mutex_up(&server_start_lock); + cfs_mutex_unlock(&server_start_lock); RETURN(rc); } @@ -980,12 +1024,17 @@ static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti) if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND) continue; - if (class_find_param(ldd->ldd_params, - PARAM_NETWORK, NULL) == 0 && - !class_match_net(ldd->ldd_params, id.nid)) { - /* can't match specified network */ + /* server use --servicenode param, only allow specified + * nids be registered */ + if ((ldd->ldd_flags & LDD_F_NO_PRIMNODE) != 0 && + class_match_nid(ldd->ldd_params, + PARAM_FAILNODE, id.nid) < 1) + continue; + + /* match specified network */ + if (!class_match_net(ldd->ldd_params, + PARAM_NETWORK, LNET_NIDNET(id.nid))) continue; - } mti->mti_nids[mti->mti_nid_count] = id.nid; mti->mti_nid_count++; @@ -1019,6 +1068,7 @@ int server_register_target(struct super_block *sb) struct obd_device *mgc = lsi->lsi_mgc; struct lustre_disk_data *ldd = lsi->lsi_ldd; struct mgs_target_info *mti = NULL; + bool writeconf; int rc; ENTRY; @@ -1039,16 +1089,35 @@ int server_register_target(struct super_block *sb) libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index, mti->mti_flags); + /* if write_conf is true, the registration must succeed */ + writeconf = !!(ldd->ldd_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(mgc->u.cli.cl_mgc_mgsexp, + 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 (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", ldd->ldd_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); + } /* Always update our flags */ - ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD; + ldd->ldd_flags = mti->mti_flags & LDD_F_ONDISK_MASK; /* If this flag is set, it means the MGS wants us to change our on-disk data. (So far this means just the index.) */ @@ -1081,6 +1150,51 @@ out: 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 (!(lsi->lsi_flags & LSI_SERVER)) + RETURN(-EINVAL); + + OBD_ALLOC_PTR(mti); + if (!mti) + RETURN(-ENOMEM); + rc = server_sb2mti(sb, 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 |= LSI_IR_CAPABLE; + +out: + if (mti) + OBD_FREE_PTR(mti); + RETURN(rc); + +} + /** Start server targets: MDTs and OSTs */ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) @@ -1097,7 +1211,7 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) /* 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 the MDS is started */ - cfs_mutex_down(&server_start_lock); + cfs_mutex_lock(&server_start_lock); obd = class_name2obd(LUSTRE_MDS_OBDNAME); if (!obd) { rc = lustre_start_simple(LUSTRE_MDS_OBDNAME, @@ -1106,19 +1220,19 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) LUSTRE_MDS_OBDNAME"_uuid", 0, 0); if (rc) { - cfs_mutex_up(&server_start_lock); + cfs_mutex_unlock(&server_start_lock); CERROR("failed to start MDS: %d\n", rc); RETURN(rc); } } - cfs_mutex_up(&server_start_lock); + cfs_mutex_unlock(&server_start_lock); } #endif /* If we're an OST, make sure the global OSS is running */ - if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) { + if (IS_OST(lsi->lsi_ldd)) { /* make sure OSS is started */ - cfs_mutex_down(&server_start_lock); + cfs_mutex_lock(&server_start_lock); obd = class_name2obd(LUSTRE_OSS_OBDNAME); if (!obd) { rc = lustre_start_simple(LUSTRE_OSS_OBDNAME, @@ -1126,12 +1240,12 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) LUSTRE_OSS_OBDNAME"_uuid", 0, 0); if (rc) { - cfs_mutex_up(&server_start_lock); + cfs_mutex_unlock(&server_start_lock); CERROR("failed to start OSS: %d\n", rc); RETURN(rc); } } - cfs_mutex_up(&server_start_lock); + cfs_mutex_unlock(&server_start_lock); } /* Set the mgc fs to our server disk. This allows the MGC to @@ -1142,26 +1256,8 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) /* Register with MGS */ rc = server_register_target(sb); - if (rc && (lsi->lsi_ldd->ldd_flags & - (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){ - CERROR("Required registration failed for %s: %d\n", - lsi->lsi_ldd->ldd_svname, rc); - if (rc == -EIO) { - LCONSOLE_ERROR_MSG(0x15f, "Communication error with " - "the MGS. Is the MGS running?\n"); - } - GOTO(out_mgc, rc); - } - if (rc == -EINVAL) { - 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_ldd->ldd_svname); - GOTO(out_mgc, rc); - } - /* non-fatal error of registeration with MGS */ if (rc) - CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", 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.) */ @@ -1199,6 +1295,11 @@ out_mgc: obd->obd_self_export, 0, NULL, NULL); } + 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); } @@ -1307,11 +1408,12 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb) struct lustre_disk_data *ldd; struct lustre_mount_data *lmd = lsi->lsi_lmd; struct vfsmount *mnt; + struct file_system_type *type; char *options = NULL; unsigned long page, s_flags; struct page *__page; - int rc; int len; + int rc; ENTRY; OBD_ALLOC(ldd, sizeof(*ldd)); @@ -1338,7 +1440,13 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb) /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */ CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev); - mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, (void *)options); + type = get_fs_type("ldiskfs"); + if (!type) { + CERROR("premount failed: cannot find ldiskfs module\n"); + GOTO(out_free, rc = -ENODEV); + } + mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options); + cfs_module_put(type->owner); if (IS_ERR(mnt)) { rc = PTR_ERR(mnt); CERROR("premount %s:%#lx ldiskfs failed: %d " @@ -1364,20 +1472,17 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb) /* Glom up mount options */ memset(options, 0, CFS_PAGE_SIZE); - if (IS_MDT(ldd)) { - /* enable 64bithash for MDS by force */ - strcpy(options, "64bithash,"); - len = CFS_PAGE_SIZE - strlen(options) - 2; - strncat(options, ldd->ldd_mount_opts, len); - } else { - strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2); - } + strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2); + + len = CFS_PAGE_SIZE - strlen(options) - 2; + if (*options != 0) + strcat(options, ","); + strncat(options, "no_mbcache", len); /* Add in any mount-line options */ if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) { len = CFS_PAGE_SIZE - strlen(options) - 2; - if (*options != 0) - strcat(options, ","); + strcat(options, ","); strncat(options, lmd->lmd_opts, len); } @@ -1387,11 +1492,16 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb) CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n", MT_STR(ldd), lmd->lmd_dev, options); - mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev, - (void *)options); + type = get_fs_type(MT_STR(ldd)); + if (!type) { + CERROR("get_fs_type failed\n"); + GOTO(out_free, rc = -ENODEV); + } + mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options); + cfs_module_put(type->owner); if (IS_ERR(mnt)) { rc = PTR_ERR(mnt); - CERROR("ll_kern_mount failed: rc = %d\n", rc); + CERROR("vfs_kern_mount failed: rc = %d\n", rc); GOTO(out_free, rc); } @@ -1424,26 +1534,25 @@ static void server_wait_finished(struct vfsmount *mnt) cfs_waitq_init(&waitq); - while (atomic_read(&mnt->mnt_count) > 1) { + while (mnt_get_count(mnt) > 1) { if (waited && (waited % 30 == 0)) LCONSOLE_WARN("Mount still busy with %d refs after " "%d secs.\n", - atomic_read(&mnt->mnt_count), + mnt_get_count(mnt), waited); /* Cannot use l_event_wait() for an interruptible sleep. */ waited += 3; - blocked = l_w_e_set_sigs(sigmask(SIGKILL)); + blocked = cfs_block_sigsinv(sigmask(SIGKILL)); cfs_waitq_wait_event_interruptible_timeout( waitq, - (atomic_read(&mnt->mnt_count) == 1), + (mnt_get_count(mnt) == 1), cfs_time_seconds(3), rc); - cfs_block_sigs(blocked); + cfs_restore_sigs(blocked); if (rc < 0) { LCONSOLE_EMERG("Danger: interrupted umount %s with " - "%d refs!\n", - mnt->mnt_devname, - atomic_read(&mnt->mnt_count)); + "%d refs!\n", mnt->mnt_devname, + mnt_get_count(mnt)); break; } @@ -1630,7 +1739,7 @@ static int server_fill_super_common(struct super_block *sb) sb->s_blocksize = 4096; sb->s_blocksize_bits = log2(sb->s_blocksize); sb->s_magic = LUSTRE_SUPER_MAGIC; - sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES; + sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */ sb->s_flags |= MS_RDONLY; sb->s_op = &server_ops; @@ -1772,6 +1881,66 @@ int server_name2index(char *svname, __u32 *idx, char **endptr) 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 & LSI_IR_CAPABLE); + int min = OBD_RECOVERY_TIME_MIN; + + LASSERT(lsi->lsi_flags & LSI_SERVER); + + 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); + + 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); + + soft = new_soft; + hard = new_hard; + } + + /* 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; +} +EXPORT_SYMBOL(server_calc_timeout); + /*************** mount common betweeen server and client ***************/ /* Common umount */ @@ -1964,8 +2133,7 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) s1 = options; while (*s1) { int clear = 0; - int time_min = 2 * (CONNECTION_SWITCH_MAX + - 2 * INITIAL_CONNECT_TIMEOUT); + int time_min = OBD_RECOVERY_TIME_MIN; /* Skip whitespace and extra commas */ while (*s1 == ' ' || *s1 == ',') @@ -1987,12 +2155,18 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) 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, "writeconf", 9) == 0) { lmd->lmd_flags |= LMD_FLG_WRITECONF; clear++; @@ -2041,7 +2215,7 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) s1 = strstr(devname, ":/"); if (s1) { ++s1; - lmd->lmd_flags = LMD_FLG_CLIENT; + lmd->lmd_flags |= LMD_FLG_CLIENT; /* Remove leading /s from fsname */ while (*++s1 == '/') ; /* Freed in lustre_free_lsi */ @@ -2079,6 +2253,10 @@ invalid: RETURN(-EINVAL); } +struct lustre_mount_data2 { + void *lmd2_data; + struct vfsmount *lmd2_mnt; +}; /** This is the entry point for the mount call into Lustre. * This is called when a server or client is mounted, @@ -2088,6 +2266,7 @@ invalid: int lustre_fill_super(struct super_block *sb, void *data, int silent) { struct lustre_mount_data *lmd; + struct lustre_mount_data2 *lmd2 = data; struct lustre_sb_info *lsi; int rc; ENTRY; @@ -2105,8 +2284,13 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent) */ cfs_lockdep_off(); + /* + * LU-639: the obd cleanup of last mount may not finish yet, wait here. + */ + obd_zombie_barrier(); + /* Figure out the lmd from the mount options */ - if (lmd_parse((char *)data, lmd)) { + if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) { lustre_put_lsi(sb); GOTO(out, rc = -EINVAL); } @@ -2127,7 +2311,7 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent) } /* Connect and start */ /* (should always be ll_fill_super) */ - rc = (*client_fill_super)(sb); + rc = (*client_fill_super)(sb, lmd2->lmd2_mnt); /* c_f_s will call lustre_common_put_super on failure */ } } else { @@ -2158,7 +2342,8 @@ out: /* We can't call ll_fill_super by name because it lives in a module that must be loaded after this one. */ -void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb)) +void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb, + struct vfsmount *mnt)) { client_fill_super = cfs; } @@ -2171,17 +2356,18 @@ void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb)) /***************** FS registration ******************/ #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18)) -struct super_block * lustre_get_sb(struct file_system_type *fs_type, - int flags, const char *devname, void * data) +struct super_block * lustre_get_sb(struct file_system_type *fs_type, int flags, + const char *devname, void * data) { return get_sb_nodev(fs_type, flags, data, lustre_fill_super); } #else -int lustre_get_sb(struct file_system_type *fs_type, - int flags, const char *devname, void * data, - struct vfsmount *mnt) +int lustre_get_sb(struct file_system_type *fs_type, int flags, + const char *devname, void * data, struct vfsmount *mnt) { - return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt); + struct lustre_mount_data2 lmd2 = {data, mnt}; + + return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt); } #endif