X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fobdclass%2Fobd_mount_server.c;h=da645341bc3d5f236914dbf734f4d56e0bb51b1e;hb=2d1be90507fc1f42f565d9e672c634b3855cb978;hp=a236d162a7d845bd28de4db1b7b607cebce4442d;hpb=42b7ddf41614c4cdf9b7d54758636c319d1560da;p=fs%2Flustre-release.git diff --git a/lustre/obdclass/obd_mount_server.c b/lustre/obdclass/obd_mount_server.c index a236d16..da64534 100644 --- a/lustre/obdclass/obd_mount_server.c +++ b/lustre/obdclass/obd_mount_server.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) 2013, 2014, Intel Corporation. + * Copyright (c) 2013, 2016, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -46,21 +42,24 @@ #define PRINT_CMD CDEBUG #define PRINT_MASK (D_SUPER | D_CONFIG) +#include +#include #include -#include -#include -#include #include -#include -#include -#include -#include #ifdef HAVE_KERNEL_LOCKED #include #endif -#ifdef HAVE_SELINUX_IS_ENABLED -#include -#endif + +#include +#include + +#include +#include +#include +#include +#include +#include +#include /*********** mount lookup *********/ @@ -289,7 +288,8 @@ static int server_stop_mgs(struct super_block *sb) /* 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) +static int server_mgc_set_fs(const struct lu_env *env, + struct obd_device *mgc, struct super_block *sb) { struct lustre_sb_info *lsi = s2lsi(sb); int rc; @@ -298,7 +298,7 @@ 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(NULL, mgc->obd_self_export, + rc = obd_set_info_async(env, mgc->obd_self_export, sizeof(KEY_SET_FS), KEY_SET_FS, sizeof(*sb), sb, NULL); if (rc != 0) @@ -307,14 +307,15 @@ static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb) RETURN(rc); } -static int server_mgc_clear_fs(struct obd_device *mgc) +static int server_mgc_clear_fs(const struct lu_env *env, + struct obd_device *mgc) { int rc; ENTRY; CDEBUG(D_MOUNT, "Unassign mgc disk\n"); - rc = obd_set_info_async(NULL, mgc->obd_self_export, + rc = obd_set_info_async(env, mgc->obd_self_export, sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS, 0, NULL, NULL); RETURN(rc); @@ -384,13 +385,25 @@ EXPORT_SYMBOL(tgt_name2lwp_name); static struct list_head lwp_register_list = LIST_HEAD_INIT(lwp_register_list); -static DEFINE_MUTEX(lwp_register_list_lock); +static DEFINE_SPINLOCK(lwp_register_list_lock); + +static void lustre_put_lwp_item(struct lwp_register_item *lri) +{ + if (atomic_dec_and_test(&lri->lri_ref)) { + LASSERT(list_empty(&lri->lri_list)); + + if (*lri->lri_exp != NULL) + class_export_put(*lri->lri_exp); + OBD_FREE_PTR(lri); + } +} int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp, register_lwp_cb cb_func, void *cb_data) { struct obd_device *lwp; struct lwp_register_item *lri; + bool cb = false; ENTRY; LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n", @@ -401,15 +414,12 @@ int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp, if (lri == NULL) RETURN(-ENOMEM); - mutex_lock(&lwp_register_list_lock); - lwp = class_name2obd(lwpname); if (lwp != NULL && lwp->obd_set_up == 1) { struct obd_uuid *uuid; OBD_ALLOC_PTR(uuid); if (uuid == NULL) { - mutex_unlock(&lwp_register_list_lock); OBD_FREE_PTR(lri); RETURN(-ENOMEM); } @@ -423,31 +433,42 @@ int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp, lri->lri_cb_func = cb_func; lri->lri_cb_data = cb_data; INIT_LIST_HEAD(&lri->lri_list); + /* + * Initialize the lri_ref at 2, one will be released before + * current function returned via lustre_put_lwp_item(), the + * other will be released in lustre_deregister_lwp_item(). + */ + atomic_set(&lri->lri_ref, 2); + + spin_lock(&lwp_register_list_lock); list_add(&lri->lri_list, &lwp_register_list); + if (*exp != NULL) + cb = true; + spin_unlock(&lwp_register_list_lock); - if (*exp != NULL && cb_func != NULL) + if (cb && cb_func != NULL) cb_func(cb_data); + lustre_put_lwp_item(lri); - mutex_unlock(&lwp_register_list_lock); RETURN(0); } EXPORT_SYMBOL(lustre_register_lwp_item); void lustre_deregister_lwp_item(struct obd_export **exp) { - struct lwp_register_item *lri, *tmp; + struct lwp_register_item *lri; - mutex_lock(&lwp_register_list_lock); - list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) { + spin_lock(&lwp_register_list_lock); + list_for_each_entry(lri, &lwp_register_list, lri_list) { if (exp == lri->lri_exp) { - if (*exp) - class_export_put(*exp); - list_del(&lri->lri_list); - OBD_FREE_PTR(lri); - break; + list_del_init(&lri->lri_list); + spin_unlock(&lwp_register_list_lock); + + lustre_put_lwp_item(lri); + return; } } - mutex_unlock(&lwp_register_list_lock); + spin_unlock(&lwp_register_list_lock); } EXPORT_SYMBOL(lustre_deregister_lwp_item); @@ -478,7 +499,7 @@ struct obd_export *lustre_find_lwp_by_index(const char *dev, __u32 idx) list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) { char *ptr = strstr(lwp->obd_name, lwp_name); - if (ptr != NULL) { + if (ptr != NULL && lwp->obd_lwp_export != NULL) { exp = class_export_get(lwp->obd_lwp_export); break; } @@ -492,23 +513,37 @@ err_lmi: } EXPORT_SYMBOL(lustre_find_lwp_by_index); -static void lustre_notify_lwp_list(struct obd_export *exp) +void lustre_notify_lwp_list(struct obd_export *exp) { - struct lwp_register_item *lri, *tmp; + struct lwp_register_item *lri; LASSERT(exp != NULL); - mutex_lock(&lwp_register_list_lock); - list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) { +again: + spin_lock(&lwp_register_list_lock); + list_for_each_entry(lri, &lwp_register_list, lri_list) { if (strcmp(exp->exp_obd->obd_name, lri->lri_name)) continue; if (*lri->lri_exp != NULL) continue; *lri->lri_exp = class_export_get(exp); - if (lri->lri_cb_func != NULL) - lri->lri_cb_func(lri->lri_cb_data); + if (lri->lri_cb_func == NULL) + continue; + atomic_inc(&lri->lri_ref); + spin_unlock(&lwp_register_list_lock); + + lri->lri_cb_func(lri->lri_cb_data); + lustre_put_lwp_item(lri); + + /* Others may have changed the list after we unlock, we have + * to rescan the list from the beginning. Usually, the list + * 'lwp_register_list' is very short, and there is 'guard' + * lri::lri_exp that will prevent the callback to be done + * repeatedly. So rescanning the list has no problem. */ + goto again; } - mutex_unlock(&lwp_register_list_lock); + spin_unlock(&lwp_register_list_lock); } +EXPORT_SYMBOL(lustre_notify_lwp_list); static int lustre_lwp_connect(struct obd_device *lwp) { @@ -560,7 +595,6 @@ static int lustre_lwp_connect(struct obd_device *lwp) if (unlikely(lwp->obd_lwp_export != NULL)) class_export_put(lwp->obd_lwp_export); lwp->obd_lwp_export = class_export_get(exp); - lustre_notify_lwp_list(exp); } GOTO(out, rc); @@ -747,7 +781,7 @@ static int client_lwp_config_process(const struct lu_env *env, struct llog_handle *handle, struct llog_rec_hdr *rec, void *data) { - struct config_llog_instance *clli = data; + struct config_llog_instance *cfg = data; int cfg_len = rec->lrh_len; char *cfg_buf = (char *) (rec + 1); struct lustre_cfg *lcfg = NULL; @@ -761,8 +795,9 @@ static int client_lwp_config_process(const struct lu_env *env, RETURN(-EINVAL); } - LASSERT(clli->cfg_sb != NULL); - lsi = s2lsi(clli->cfg_sb); + if (cfg->cfg_sb == NULL) + GOTO(out, rc = -EINVAL); + lsi = s2lsi(cfg->cfg_sb); lcfg = (struct lustre_cfg *)cfg_buf; if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) { @@ -784,35 +819,35 @@ static int client_lwp_config_process(const struct lu_env *env, marker->cm_flags & CM_EXCLUDE) GOTO(out, rc = 0); - if (!tgt_is_mdt(marker->cm_tgtname, &clli->cfg_lwp_idx)) + if (!tgt_is_mdt(marker->cm_tgtname, &cfg->cfg_lwp_idx)) GOTO(out, rc = 0); - if (IS_MDT(lsi) && clli->cfg_lwp_idx != 0) + if (IS_MDT(lsi) && cfg->cfg_lwp_idx != 0) 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; + cfg->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; + cfg->cfg_flags |= CFG_F_SKIP; } else if (marker->cm_flags & CM_END) { - clli->cfg_flags = 0; + cfg->cfg_flags = 0; } } break; } case LCFG_ADD_UUID: { - if (clli->cfg_flags == CFG_F_MARKER) { - rc = lustre_lwp_setup(lcfg, lsi, clli->cfg_lwp_idx); + if (cfg->cfg_flags == CFG_F_MARKER) { + rc = lustre_lwp_setup(lcfg, lsi, cfg->cfg_lwp_idx); /* XXX: process only the first nid as * we don't need another instance of lwp */ - clli->cfg_flags |= CFG_F_SKIP; - } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) { + cfg->cfg_flags |= CFG_F_SKIP; + } else if (cfg->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) { rc = class_add_uuid(lustre_cfg_string(lcfg, 1), lcfg->lcfg_nid); if (rc) @@ -1097,6 +1132,13 @@ static int server_lsi2mti(struct lustre_sb_info *lsi, } } + if (mti->mti_nid_count == 0) { + CERROR("Failed to get NID for server %s, please check whether " + "the target is specifed with improper --servicenode or " + "--network options.\n", mti->mti_svname); + RETURN(-EINVAL); + } + mti->mti_lustre_ver = LUSTRE_VERSION_CODE; mti->mti_config_ver = 0; @@ -1236,7 +1278,7 @@ static int server_start_targets(struct super_block *sb) struct obd_device *obd; struct lustre_sb_info *lsi = s2lsi(sb); struct config_llog_instance cfg; - struct lu_env env; + struct lu_env mgc_env; struct lu_device *dev; int rc; ENTRY; @@ -1280,11 +1322,15 @@ static int server_start_targets(struct super_block *sb) mutex_unlock(&server_start_lock); } + rc = lu_env_init(&mgc_env, LCT_MG_THREAD); + if (rc != 0) + GOTO(out_stop_service, rc); + /* 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. */ - rc = server_mgc_set_fs(lsi->lsi_mgc, sb); + rc = server_mgc_set_fs(&mgc_env, lsi->lsi_mgc, sb); if (rc) - GOTO(out_stop_service, rc); + GOTO(out_env, rc); /* Register with MGS */ rc = server_register_target(lsi); @@ -1336,6 +1382,8 @@ static int server_start_targets(struct super_block *sb) /* log has been fully processed, let clients connect */ dev = obd->obd_lu_dev; if (dev && dev->ld_ops->ldo_prepare) { + struct lu_env env; + rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags); if (rc == 0) { struct lu_context session_ctx; @@ -1363,8 +1411,9 @@ static int server_start_targets(struct super_block *sb) out_mgc: /* Release the mgc fs for others to use */ - server_mgc_clear_fs(lsi->lsi_mgc); - + server_mgc_clear_fs(&mgc_env, lsi->lsi_mgc); +out_env: + lu_env_fini(&mgc_env); out_stop_service: if (rc != 0) server_stop_servers(lsi->lsi_flags); @@ -1487,9 +1536,13 @@ static void server_put_super(struct super_block *sb) 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); + if (lprof != NULL) { + if (lprof->lp_dt != NULL) { + OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1); + strncpy(extraname, lprof->lp_dt, + strlen(lprof->lp_dt) + 1); + } + class_put_profile(lprof); } obd = class_name2obd(lsi->lsi_svname); @@ -1711,8 +1764,6 @@ static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags) rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL); - OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_CONNECT, 10); - if (rc) { obd->obd_force = 1; class_manual_cleanup(obd); @@ -1745,6 +1796,9 @@ int server_fill_super(struct super_block *sb) int rc; ENTRY; + /* to simulate target mount race */ + OBD_RACE(OBD_FAIL_TGT_MOUNT_RACE); + rc = lsi_prepare(lsi); if (rc) RETURN(rc); @@ -1850,23 +1904,20 @@ void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd) 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); + obd->obd_name, soft, hard, new_soft, hard); soft = new_soft; - hard = new_hard; + } else { + LCONSOLE_INFO("%s: Imperative Recovery not enabled, recovery " + "window %d-%d\n", obd->obd_name, soft, hard); } /* we're done */