X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fobdclass%2Fobd_mount.c;h=e89a4d4f8ca7f1f75890b1cbaf374066e4386ce2;hb=0a714ba01bbcdb43fa2d07e88652be2b8fb1c52f;hp=cfd04edcdb0b5dd88633cd1986d290413cdcc443;hpb=d2d56f38da01001c92a09afc6b52b5acbd9bc13c;p=fs%2Flustre-release.git diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c index cfd04ed..e89a4d4 100644 --- a/lustre/obdclass/obd_mount.c +++ b/lustre/obdclass/obd_mount.c @@ -1,26 +1,43 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * lustre/obdclass/obd_mount.c - * Client/server mount routines + * GPL HEADER START * - * Copyright (c) 2006 Cluster File Systems, Inc. - * Author: Nathan Rutman + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * - * This file is part of Lustre, http://www.lustre.org/ + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. * - * Lustre is free software; you can redistribute it and/or - * modify it under the terms of version 2 of the GNU General Public - * License as published by the Free Software Foundation. + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * Lustre is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. + * 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 * - * You should have received a copy of the GNU General Public License - * along with Lustre; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + * 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. + * + * GPL HEADER END + */ +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Use is subject to license terms. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. + * + * lustre/obdclass/obd_mount.c + * + * Client/server mount routines + * + * Author: Nathan Rutman */ @@ -40,11 +57,12 @@ #include static int (*client_fill_super)(struct super_block *sb) = NULL; +static void (*kill_super_cb)(struct super_block *sb) = NULL; /*********** mount lookup *********/ DECLARE_MUTEX(lustre_mount_info_lock); -struct list_head server_mount_info_list = LIST_HEAD_INIT(server_mount_info_list); +static CFS_LIST_HEAD(server_mount_info_list); static struct lustre_mount_info *server_find_mount(const char *name) { @@ -272,11 +290,11 @@ static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt, GOTO(out, rc); } - len = file->f_dentry->d_inode->i_size; + len = i_size_read(file->f_dentry->d_inode); CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len); if (len != sizeof(*ldd)) { - CERROR("disk data size does not match: see %lu expect "LPSZ"\n", - len, sizeof(*ldd)); + CERROR("disk data size does not match: see %lu expect %u\n", + len, (int)sizeof(*ldd)); GOTO(out_close, rc = -EINVAL); } @@ -398,7 +416,7 @@ int lustre_process_log(struct super_block *sb, char *logname, "communication errors between this node and " "the MGS, a bad configuration, or other " "errors. See the syslog for more " - "information.\n", mgc->obd_name, logname, + "information.\n", mgc->obd_name, logname, rc); /* class_obd_list(); */ @@ -503,10 +521,13 @@ static int server_start_mgs(struct super_block *sb) 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)))) - server_deregister_mount(LUSTRE_MGS_OBDNAME); + if (!rc) { + rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME, + LUSTRE_MGS_OBDNAME, 0, 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). " @@ -538,18 +559,22 @@ static int server_stop_mgs(struct super_block *sb) DECLARE_MUTEX(mgc_start_lock); -/* Set up a mgcobd to process startup logs */ +/** Set up a mgc obd to process startup logs + * + * \param sb [in] super block of the mgc obd + * + * \retval 0 success, otherwise error code + */ static int lustre_start_mgc(struct super_block *sb) { - struct lustre_handle mgc_conn = {0, }; - struct obd_connect_data ocd = { 0 }; + struct obd_connect_data *data = NULL; struct lustre_sb_info *lsi = s2lsi(sb); struct obd_device *obd; struct obd_export *exp; struct obd_uuid *uuid; class_uuid_t uuidc; lnet_nid_t nid; - char *mgcname, *niduuid; + char *mgcname, *niduuid, *mgssec; char *ptr; int recov_bk; int rc = 0, i = 0, j, len; @@ -592,10 +617,18 @@ static int lustre_start_mgc(struct super_block *sb) GOTO(out_free, rc = -ENOMEM); sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid)); + mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : ""; + mutex_down(&mgc_start_lock); obd = class_name2obd(mgcname); - if (obd) { + if (obd && !obd->obd_stopping) { + rc = obd_set_info_async(obd->obd_self_export, + strlen(KEY_MGSSEC), KEY_MGSSEC, + strlen(mgssec), mgssec, NULL); + if (rc) + GOTO(out_free, rc); + /* Re-using an existing MGC */ atomic_inc(&obd->u.cli.cl_mgc_refcount); @@ -614,7 +647,7 @@ static int lustre_start_mgc(struct super_block *sb) recov_bk++; CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk); rc = obd_set_info_async(obd->obd_self_export, - strlen(KEY_INIT_RECOV_BACKUP), + sizeof(KEY_INIT_RECOV_BACKUP), KEY_INIT_RECOV_BACKUP, sizeof(recov_bk), &recov_bk, NULL); GOTO(out, rc = 0); @@ -708,6 +741,12 @@ static int lustre_start_mgc(struct super_block *sb) GOTO(out_free, rc = -ENOTCONN); } + rc = obd_set_info_async(obd->obd_self_export, + strlen(KEY_MGSSEC), KEY_MGSSEC, + strlen(mgssec), mgssec, NULL); + if (rc) + GOTO(out_free, rc); + /* Keep a refcount of servers/clients who started with "mount", so we know when we can get rid of the mgc. */ atomic_set(&obd->u.cli.cl_mgc_refcount, 1); @@ -715,20 +754,26 @@ 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, - strlen(KEY_INIT_RECOV_BACKUP), + sizeof(KEY_INIT_RECOV_BACKUP), KEY_INIT_RECOV_BACKUP, sizeof(recov_bk), &recov_bk, NULL); if (rc) /* nonfatal */ - CERROR("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc); + 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 */ - rc = obd_connect(NULL, &mgc_conn, obd, &(obd->obd_uuid), &ocd); + OBD_ALLOC_PTR(data); + if (data == NULL) + GOTO(out, rc = -ENOMEM); + data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID | + OBD_CONNECT_AT; + 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); } - exp = class_conn2export(&mgc_conn); obd->u.cli.cl_mgc_mgsexp = exp; out: @@ -749,8 +794,8 @@ static int lustre_stop_mgc(struct super_block *sb) { struct lustre_sb_info *lsi = s2lsi(sb); struct obd_device *obd; - char *niduuid, *ptr = 0; - int i, rc = 0, len; + char *niduuid = 0, *ptr = 0; + int i, rc = 0, len = 0; ENTRY; if (!lsi) @@ -758,9 +803,10 @@ static int lustre_stop_mgc(struct super_block *sb) obd = lsi->lsi_mgc; if (!obd) RETURN(-ENOENT); - lsi->lsi_mgc = NULL; + mutex_down(&mgc_start_lock); + 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. */ @@ -769,15 +815,17 @@ static int lustre_stop_mgc(struct super_block *sb) GOTO(out, rc = -EBUSY); } - /* MGC must always stop */ - obd->obd_force = 1; - /* client_disconnect_export uses the no_recov flag to decide whether it - should disconnect or just invalidate. (The MGC has no - recoverable data in any case.) */ + /* The MGC has no recoverable data in any case. + * force shotdown set in umount_begin */ obd->obd_no_recov = 1; - if (obd->u.cli.cl_mgc_mgsexp) - obd_disconnect(obd->u.cli.cl_mgc_mgsexp); + if (obd->u.cli.cl_mgc_mgsexp) { + /* An error is not fatal, if we are unable to send the + disconnect mgs ping evictor cleans up the export */ + rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp); + if (rc) + CDEBUG(D_MOUNT, "disconnect failed %d\n", rc); + } /* Save the obdname for cleaning the nid uuids, which are obdname_XX */ @@ -794,7 +842,8 @@ static int lustre_stop_mgc(struct super_block *sb) /* Clean the nid uuids */ if (!niduuid) - RETURN(-ENOMEM); + GOTO(out, rc = -ENOMEM); + 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, @@ -803,10 +852,11 @@ static int lustre_stop_mgc(struct super_block *sb) CERROR("del MDC UUID %s failed: rc = %d\n", niduuid, rc); } - OBD_FREE(niduuid, len); - /* class_import_put will get rid of the additional connections */ - out: + if (niduuid) + OBD_FREE(niduuid, len); + + /* class_import_put will get rid of the additional connections */ mutex_up(&mgc_start_lock); RETURN(rc); } @@ -823,7 +873,7 @@ static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb) /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */ rc = obd_set_info_async(mgc->obd_self_export, - strlen("set_fs"), "set_fs", + sizeof(KEY_SET_FS), KEY_SET_FS, sizeof(*sb), sb, NULL); if (rc) { CERROR("can't set_fs %d\n", rc); @@ -840,7 +890,7 @@ 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, - strlen("clear_fs"), "clear_fs", + sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS, 0, NULL, NULL); RETURN(rc); } @@ -895,81 +945,10 @@ int server_mti_print(char *title, struct mgs_target_info *mti) return(0); } -static -int mti_set_sec_opts(struct mgs_target_info *mti, struct lustre_mount_data *lmd) -{ - char *s1, *s2; - - if (lmd->lmd_sec_mdt == NULL && lmd->lmd_sec_cli == NULL) { - /* just let on-disk params do its work. but we have an - * assumption that any changes of on-disk data by tune2fs - * should lead to server rewrite log. - */ - return 0; - } - - /* filter out existing sec options */ - s1 = mti->mti_params; - while (*s1) { - int clear; - - while (*s1 == ' ') - s1++; - - if (strncmp(s1, PARAM_SEC_RPC_MDT, - sizeof(PARAM_SEC_RPC_MDT) - 1) == 0 || - strncmp(s1, PARAM_SEC_RPC_CLI, - sizeof(PARAM_SEC_RPC_CLI) - 1) == 0) - clear = 1; - else - clear = 0; - - s2 = strchr(s1, ' '); - if (s2 == NULL) { - if (clear) - *s1 = '\0'; - break; - } - s2++; - if (clear) - memmove(s1, s2, strlen(s2) + 1); - else - s1 = s2; - } - - /* append sec options from lmd */ - /* FIXME add flag LDD_F_UPDATE after mountconf start supporting - * log updating. - */ - if (lmd->lmd_sec_mdt) { - if (strlen(mti->mti_params) + strlen(lmd->lmd_sec_mdt) + - sizeof(PARAM_SEC_RPC_MDT) + 1 >= sizeof(mti->mti_params)) { - CERROR("security params too big for mti\n"); - return -ENOMEM; - } - strcat(mti->mti_params, " "PARAM_SEC_RPC_MDT); - strcat(mti->mti_params, lmd->lmd_sec_mdt); - //mti->mti_flags |= LDD_F_UPDATE; - } - if (lmd->lmd_sec_cli) { - if (strlen(mti->mti_params) + strlen(lmd->lmd_sec_cli) + - sizeof(PARAM_SEC_RPC_CLI) + 2 > sizeof(mti->mti_params)) { - CERROR("security params too big for mti\n"); - return -ENOMEM; - } - strcat(mti->mti_params, " "PARAM_SEC_RPC_CLI); - strcat(mti->mti_params, lmd->lmd_sec_cli); - //mti->mti_flags |= LDD_F_UPDATE; - } - - return 0; -} - static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti) { struct lustre_sb_info *lsi = s2lsi(sb); struct lustre_disk_data *ldd = lsi->lsi_ldd; - struct lustre_mount_data *lmd = lsi->lsi_lmd; lnet_process_id_t id; int i = 0; ENTRY; @@ -1005,8 +984,7 @@ static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti) RETURN(-ENOMEM); } memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params)); - - RETURN(mti_set_sec_opts(mti, lmd)); + RETURN(0); } /* Register an old or new target with the MGS. If needed MGS will construct @@ -1040,12 +1018,10 @@ int server_register_target(struct super_block *sb) /* Register the target */ /* FIXME use mgc_process_config instead */ rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp, - strlen("register_target"), "register_target", + sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET, sizeof(*mti), mti, NULL); - if (rc) { - CERROR("registration with the MGS failed (%d)\n", rc); + if (rc) GOTO(out, rc); - } /* Always update our flags */ ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD; @@ -1158,6 +1134,9 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) 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); /* Let the target look up the mount using the target's name (we can't pass the sb or mnt through class_process_config.) */ @@ -1171,6 +1150,9 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) if (rc) { CERROR("failed to start server %s: %d\n", lsi->lsi_ldd->ldd_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); } @@ -1193,7 +1175,7 @@ out_mgc: } /* log has been fully processed */ - obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, 0); + obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG); } RETURN(rc); @@ -1203,15 +1185,15 @@ out_mgc: struct lustre_sb_info *lustre_init_lsi(struct super_block *sb) { - struct lustre_sb_info *lsi = NULL; + struct lustre_sb_info *lsi; ENTRY; - OBD_ALLOC(lsi, sizeof(*lsi)); + OBD_ALLOC_PTR(lsi); if (!lsi) RETURN(NULL); - OBD_ALLOC(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd)); + OBD_ALLOC_PTR(lsi->lsi_lmd); if (!lsi->lsi_lmd) { - OBD_FREE(lsi, sizeof(*lsi)); + OBD_FREE_PTR(lsi); RETURN(NULL); } @@ -1223,8 +1205,6 @@ struct lustre_sb_info *lustre_init_lsi(struct super_block *sb) /* Default umount style */ lsi->lsi_flags = LSI_UMOUNT_FAILOVER; - lsi->lsi_lmd->lmd_nllu = NOBODY_UID; - lsi->lsi_lmd->lmd_nllg = NOBODY_GID; RETURN(lsi); } @@ -1233,10 +1213,8 @@ static int lustre_free_lsi(struct super_block *sb) struct lustre_sb_info *lsi = s2lsi(sb); ENTRY; - if (!lsi) - RETURN(0); - - CDEBUG(D_MOUNT, "Freeing lsi\n"); + 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); @@ -1251,12 +1229,9 @@ static int lustre_free_lsi(struct super_block *sb) 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_sec_mdt != NULL) - OBD_FREE(lsi->lsi_lmd->lmd_sec_mdt, - strlen(lsi->lsi_lmd->lmd_sec_mdt) + 1); - if (lsi->lsi_lmd->lmd_sec_cli != NULL) - OBD_FREE(lsi->lsi_lmd->lmd_sec_cli, - strlen(lsi->lsi_lmd->lmd_sec_cli) + 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); @@ -1281,10 +1256,9 @@ static int lustre_put_lsi(struct super_block *sb) struct lustre_sb_info *lsi = s2lsi(sb); ENTRY; - LASSERT(lsi); + LASSERT(lsi != NULL); CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts)); - if (atomic_dec_and_test(&lsi->lsi_mounts)) { lustre_free_lsi(sb); RETURN(1); @@ -1304,6 +1278,7 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb) struct vfsmount *mnt; char *options = NULL; unsigned long page, s_flags; + struct page *__page; int rc; ENTRY; @@ -1315,30 +1290,29 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb) Note ext3/ldiskfs can't be mounted ro. */ s_flags = sb->s_flags; + /* allocate memory for options */ + OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD); + if (!__page) + GOTO(out_free, rc = -ENOMEM); + page = (unsigned long)cfs_page_address(__page); + options = (char *)page; + memset(options, 0, CFS_PAGE_SIZE); + + /* mount-line options must be added for pre-mount because it may + * contain mount options such as journal_dev which are required + * to mount successfuly the underlying filesystem */ + if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) + strncat(options, lmd->lmd_opts, CFS_PAGE_SIZE - 1); + /* 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, 0); + mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, (void *)options); if (IS_ERR(mnt)) { rc = PTR_ERR(mnt); -#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0)) - /* 2.6 kernels: if ldiskfs fails, try ldiskfs2 */ - mnt = ll_kern_mount("ldiskfs2", s_flags, lmd->lmd_dev, 0); - if (IS_ERR(mnt)) { - int rc2 = PTR_ERR(mnt); - CERROR("premount %s:%#lx ldiskfs failed: %d, ldiskfs2 " - "failed: %d. Is the ldiskfs module available?\n", - lmd->lmd_dev, s_flags, rc, rc2); - GOTO(out_free, rc); - } -#else - /* 2.4 kernels: if ldiskfs fails, try ext3 */ - mnt = ll_kern_mount("ext3", s_flags, lmd->lmd_dev, 0); - if (IS_ERR(mnt)) { - rc = PTR_ERR(mnt); - CERROR("premount ext3 failed: rc = %d\n", rc); - GOTO(out_free, rc); - } -#endif + CERROR("premount %s:%#lx ldiskfs failed: %d " + "Is the ldiskfs module available?\n", + lmd->lmd_dev, s_flags, rc ); + GOTO(out_free, rc); } OBD_SET_CTXT_MAGIC(&mount_ctxt); @@ -1357,11 +1331,6 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb) /* Done with our pre-mount, now do the real mount. */ /* Glom up mount options */ - page = __get_free_page(GFP_KERNEL); - if (!page) - GOTO(out_free, rc = -ENOMEM); - - options = (char *)page; memset(options, 0, CFS_PAGE_SIZE); strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2); @@ -1381,45 +1350,65 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb) MT_STR(ldd), lmd->lmd_dev, options); mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev, (void *)options); - free_page(page); if (IS_ERR(mnt)) { rc = PTR_ERR(mnt); CERROR("ll_kern_mount failed: rc = %d\n", rc); GOTO(out_free, rc); } + if (lmd->lmd_flags & LMD_FLG_ABORT_RECOV) + simple_truncate(mnt->mnt_sb->s_root, mnt, LAST_RCVD, + LR_CLIENT_START); + + OBD_PAGE_FREE(__page); lsi->lsi_ldd = ldd; /* freed at lsi cleanup */ CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt); RETURN(mnt); out_free: + if (__page) + OBD_PAGE_FREE(__page); OBD_FREE(ldd, sizeof(*ldd)); lsi->lsi_ldd = NULL; RETURN(ERR_PTR(rc)); } +/* Wait here forever until the mount refcount is 0 before completing umount, + * else we risk dereferencing a null pointer. + * LNET may take e.g. 165s before killing zombies. + */ static void server_wait_finished(struct vfsmount *mnt) { - wait_queue_head_t waitq; - struct l_wait_info lwi; - int retries = 120; - - init_waitqueue_head(&waitq); - - while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) { - LCONSOLE_WARN("Mount still busy with %d refs, waiting for " - "%d secs...\n", - atomic_read(&mnt->mnt_count), retries); - - /* Wait for a bit */ - retries -= 5; - lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL); - l_wait_event(waitq, 0, &lwi); - } - if (atomic_read(&mnt->mnt_count) > 1) { - CERROR("Mount %p is still busy (%d refs), giving up.\n", - mnt, atomic_read(&mnt->mnt_count)); - } + cfs_waitq_t waitq; + int rc, waited = 0; + cfs_sigset_t blocked; + + cfs_waitq_init(&waitq); + + while (cfs_atomic_read(&mnt->mnt_count) > 1) { + if (waited && (waited % 30 == 0)) + LCONSOLE_WARN("Mount still busy with %d refs after " + "%d secs.\n", + atomic_read(&mnt->mnt_count), + waited); + /* Cannot use l_event_wait() for an interruptible sleep. */ + waited += 3; + blocked = l_w_e_set_sigs(sigmask(SIGKILL)); + cfs_waitq_wait_event_interruptible_timeout( + waitq, + (cfs_atomic_read(&mnt->mnt_count) == 1), + cfs_time_seconds(3), + rc); + cfs_block_sigs(blocked); + if (rc < 0) { + LCONSOLE_EMERG("Danger: interrupted umount %s with " + "%d refs!\n", + mnt->mnt_devname, + atomic_read(&mnt->mnt_count)); + break; + } + + } } static void server_put_super(struct super_block *sb) @@ -1431,7 +1420,6 @@ static void server_put_super(struct super_block *sb) int tmpname_sz; int lddflags = lsi->lsi_ldd->ldd_flags; int lsiflags = lsi->lsi_flags; - int rc; ENTRY; LASSERT(lsiflags & LSI_SERVER); @@ -1440,9 +1428,12 @@ static void server_put_super(struct super_block *sb) OBD_ALLOC(tmpname, tmpname_sz); memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz); CDEBUG(D_MOUNT, "server put_super %s\n", tmpname); + if (IS_MDT(lsi->lsi_ldd) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC)) + snprintf(tmpname, tmpname_sz, "MGS"); /* Stop the target */ - if (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd)) { + if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) && + (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) { struct lustre_profile *lprof = NULL; /* tell the mgc to drop the config log */ @@ -1460,12 +1451,10 @@ static void server_put_super(struct super_block *sb) obd = class_name2obd(lsi->lsi_ldd->ldd_svname); if (obd) { CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name); - if (lsi->lsi_flags & LSI_UMOUNT_FORCE) - obd->obd_force = 1; if (lsi->lsi_flags & 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! */ + * to .put_super, so we better make sure we clean up! */ obd->obd_force = 1; class_manual_cleanup(obd); } else { @@ -1477,15 +1466,13 @@ static void server_put_super(struct super_block *sb) /* If they wanted the mgs to stop separately from the mdt, they should have put it on a different device. */ if (IS_MGS(lsi->lsi_ldd)) { - /* stop the mgc before the mgs so the connection gets cleaned - up */ - lustre_stop_mgc(sb); - server_stop_mgs(sb); + /* if MDS start with --nomgs, don't stop MGS then */ + if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) + server_stop_mgs(sb); } /* Clean the mgc and sb */ - rc = lustre_common_put_super(sb); - /* FIXME how can I report a failure to umount? */ + lustre_common_put_super(sb); /* Wait for the targets to really clean up - can't exit (and let the sb get destroyed) while the mount is still in use */ @@ -1634,9 +1621,9 @@ static int server_fill_super(struct super_block *sb) if (IS_ERR(mnt)) { rc = PTR_ERR(mnt); CERROR("Unable to mount device %s: %d\n", - lsi->lsi_lmd->lmd_dev, rc); + lsi->lsi_lmd->lmd_dev, rc); lustre_put_lsi(sb); - GOTO(out, rc); + RETURN(rc); } lsi->lsi_srv_mnt = mnt; @@ -1650,13 +1637,13 @@ static int server_fill_super(struct super_block *sb) "running. Double-mount may have compromised" " the disk journal.\n", lsi->lsi_ldd->ldd_svname); - unlock_mntput(mnt); lustre_put_lsi(sb); - GOTO(out, rc = -EALREADY); + unlock_mntput(mnt); + RETURN(-EALREADY); } - /* start MGS before MGC */ - if (IS_MGS(lsi->lsi_ldd)) { + /* Start MGS before MGC */ + if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){ rc = server_start_mgs(sb); if (rc) GOTO(out_mnt, rc); @@ -1690,14 +1677,17 @@ static int server_fill_super(struct super_block *sb) GOTO(out_mnt, rc); LCONSOLE_WARN("Server %s on device %s has started\n", - lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev); + ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) && + (IS_MDT(lsi->lsi_ldd))) ? "MGS" : lsi->lsi_ldd->ldd_svname, + lsi->lsi_lmd->lmd_dev); 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); -out: - RETURN(rc); + return rc; } /* Get the index from the obd name. @@ -1708,16 +1698,27 @@ int server_name2index(char *svname, __u32 *idx, char **endptr) { unsigned long index; int rc; - char *dash = strchr(svname, '-'); + char *dash = strrchr(svname, '-'); if (!dash) return(-EINVAL); + /* intepret -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--); + if (dash == svname) + return(-EINVAL); + } + if (strncmp(dash + 1, "MDT", 3) == 0) rc = LDD_F_SV_TYPE_MDT; else if (strncmp(dash + 1, "OST", 3) == 0) rc = LDD_F_SV_TYPE_OST; else return(-EINVAL); + if (strcmp(dash + 4, "all") == 0) + return rc | LDD_F_SV_ALL; index = simple_strtoul(dash + 4, endptr, 16); *idx = index; @@ -1747,6 +1748,7 @@ 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); } @@ -1760,10 +1762,6 @@ static void lmd_print(struct lustre_mount_data *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); - if (lmd->lmd_sec_mdt) - PRINT_CMD(PRINT_MASK, "sec_mdt: %s\n", lmd->lmd_sec_mdt); - if (lmd->lmd_sec_cli) - PRINT_CMD(PRINT_MASK, "sec_cli: %s\n", lmd->lmd_sec_cli); if (lmd->lmd_opts) PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts); for (i = 0; i < lmd->lmd_exclude_count; i++) { @@ -1854,42 +1852,14 @@ static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr) RETURN(rc); } -static -int lmd_set_sec_opts(char **set, char *opts, int length) +static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr) { - if (*set) - OBD_FREE(*set, strlen(*set) + 1); + char *tail; + int length; - OBD_ALLOC(*set, length + 1); - if (*set == NULL) - return -ENOMEM; - - memcpy(*set, opts, length); - (*set)[length] = '\0'; - - return 0; -} - -static -int lmd_parse_sec_opts(struct lustre_mount_data *lmd, char *ptr) -{ - char *tail; - char **set = NULL; - int length; - - /* check peer name */ - if (strncmp(ptr, "sec_mdt=", 8) == 0) { - set = &lmd->lmd_sec_mdt; - ptr += 8; - } else if (strncmp(ptr, "sec_cli=", 8) == 0) { - set = &lmd->lmd_sec_cli; - ptr += 8; - } else if (strncmp(ptr, "sec=", 4) == 0) { - /* leave 'set' be null */ - ptr += 4; - } else { - CERROR("invalid security options: %s\n", ptr); - return -EINVAL; + if (lmd->lmd_mgssec != NULL) { + OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1); + lmd->lmd_mgssec = NULL; } tail = strchr(ptr, ','); @@ -1898,19 +1868,12 @@ int lmd_parse_sec_opts(struct lustre_mount_data *lmd, char *ptr) else length = tail - ptr; - if (set) { - if (lmd_set_sec_opts(set, ptr, length)) - return -EINVAL; - } else { - if (lmd->lmd_sec_mdt == NULL && - lmd_set_sec_opts(&lmd->lmd_sec_mdt, ptr, length)) - return -EINVAL; - - if (lmd->lmd_sec_cli == NULL && - lmd_set_sec_opts(&lmd->lmd_sec_cli, ptr, length)) - return -EINVAL; - } + OBD_ALLOC(lmd->lmd_mgssec, length + 1); + if (lmd->lmd_mgssec == NULL) + return -ENOMEM; + memcpy(lmd->lmd_mgssec, ptr, length); + lmd->lmd_mgssec[length] = '\0'; return 0; } @@ -1958,20 +1921,17 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) } else if (strncmp(s1, "nosvc", 5) == 0) { lmd->lmd_flags |= LMD_FLG_NOSVC; clear++; - /* ost exclusion list */ - } else if (strncmp(s1, "exclude=", 8) == 0) { - rc = lmd_make_exclusion(lmd, s1 + 7); + } else if (strncmp(s1, "nomgs", 5) == 0) { + lmd->lmd_flags |= LMD_FLG_NOMGS; + clear++; + } else if (strncmp(s1, "mgssec=", 7) == 0) { + rc = lmd_parse_mgssec(lmd, s1 + 7); if (rc) goto invalid; clear++; - } else if (strncmp(s1, "nllu=", 5) == 0) { - lmd->lmd_nllu = simple_strtoul(s1 + 5, NULL, 10); - clear++; - } else if (strncmp(s1, "nllg=", 5) == 0) { - lmd->lmd_nllg = simple_strtoul(s1 + 5, NULL, 10); - clear++; - } else if (strncmp(s1, "sec", 3) == 0) { - rc = lmd_parse_sec_opts(lmd, s1); + /* ost exclusion list */ + } else if (strncmp(s1, "exclude=", 8) == 0) { + rc = lmd_make_exclusion(lmd, s1 + 7); if (rc) goto invalid; clear++; @@ -2006,8 +1966,9 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) goto invalid; } - s1 = strrchr(devname, ':'); + s1 = strstr(devname, ":/"); if (s1) { + ++s1; lmd->lmd_flags = LMD_FLG_CLIENT; /* Remove leading /s from fsname */ while (*++s1 == '/') ; @@ -2061,10 +2022,16 @@ 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'. + */ + lockdep_off(); + /* Figure out the lmd from the mount options */ if (lmd_parse((char *)data, lmd)) { lustre_put_lsi(sb); - RETURN(-EINVAL); + GOTO(out, rc = -EINVAL); } if (lmd_is_client(lmd)) { @@ -2073,12 +2040,13 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent) 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_stop_mgc(sb); - goto out; + lustre_put_lsi(sb); + GOTO(out, rc); } /* Connect and start */ /* (should always be ll_fill_super) */ @@ -2095,14 +2063,19 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent) /* s_f_s will call server_put_super on failure */ } + /* If error happens in fill_super() call, @lsi will be killed there. + * This is why we do not put it here. */ + GOTO(out, rc); out: - if (rc){ + if (rc) { CERROR("Unable to mount %s (%d)\n", s2lsi(sb) ? lmd->lmd_dev : "", rc); } else { - CDEBUG(D_SUPER, "mount %s complete\n", lmd->lmd_dev); + CDEBUG(D_SUPER, "Mount %s complete\n", + lmd->lmd_dev); } - RETURN(rc); + lockdep_on(); + return rc; } @@ -2113,17 +2086,17 @@ void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb)) client_fill_super = cfs; } +void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb)) +{ + kill_super_cb = cfs; +} + /***************** FS registration ******************/ -#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0)) -/* 2.5 and later */ #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) { - /* calls back in fill super */ - /* we could append devname= onto options (*data) here, - but 2.4 doesn't get devname. So we do it in mount_lustre.c */ return get_sb_nodev(fs_type, flags, data, lustre_fill_super); } #else @@ -2131,42 +2104,28 @@ int lustre_get_sb(struct file_system_type *fs_type, int flags, const char *devname, void * data, struct vfsmount *mnt) { - /* calls back in fill super */ - /* we could append devname= onto options (*data) here, - but 2.4 doesn't get devname. So we do it in mount_lustre.c */ return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt); } #endif -struct file_system_type lustre_fs_type = { - .owner = THIS_MODULE, - .name = "lustre", - .get_sb = lustre_get_sb, - .kill_sb = kill_anon_super, - .fs_flags = FS_BINARY_MOUNTDATA, -}; - -#else -/* 2.4 */ -static struct super_block *lustre_read_super(struct super_block *sb, - void *data, int silent) +void lustre_kill_super(struct super_block *sb) { - int rc; - ENTRY; + struct lustre_sb_info *lsi = s2lsi(sb); - rc = lustre_fill_super(sb, data, silent); - if (rc) - RETURN(NULL); - RETURN(sb); + if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER)) + (*kill_super_cb)(sb); + + kill_anon_super(sb); } -static struct file_system_type lustre_fs_type = { - .owner = THIS_MODULE, - .name = "lustre", - .fs_flags = FS_NFSEXP_FSID, - .read_super = lustre_read_super, +struct file_system_type lustre_fs_type = { + .owner = THIS_MODULE, + .name = "lustre", + .get_sb = lustre_get_sb, + .kill_sb = lustre_kill_super, + .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV | + LL_RENAME_DOES_D_MOVE, }; -#endif int lustre_register_fs(void) { @@ -2179,6 +2138,7 @@ int lustre_unregister_fs(void) } EXPORT_SYMBOL(lustre_register_client_fill_super); +EXPORT_SYMBOL(lustre_register_kill_super_cb); EXPORT_SYMBOL(lustre_common_put_super); EXPORT_SYMBOL(lustre_process_log); EXPORT_SYMBOL(lustre_end_log); @@ -2190,5 +2150,3 @@ EXPORT_SYMBOL(server_register_target); EXPORT_SYMBOL(server_name2index); EXPORT_SYMBOL(server_mti_print); EXPORT_SYMBOL(do_lcfg); - -