X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fobdclass%2Fobd_mount.c;h=72b5128a665747e2106debb512ebf7c4fcf466f8;hb=2c3286fa318daa55b5887274a4550d6669f2e56d;hp=e1b1ed6fe4fb77cc81db53fdc95bbb4253d13f13;hpb=113303973ec9f8484eb2355a1a6ef3c4c7fd6a56;p=fs%2Flustre-release.git diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c index e1b1ed6..72b5128 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 */ @@ -40,13 +57,14 @@ #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(char *name) +static struct lustre_mount_info *server_find_mount(const char *name) { struct list_head *tmp; struct lustre_mount_info *lmi; @@ -63,7 +81,7 @@ static struct lustre_mount_info *server_find_mount(char *name) /* 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(char *name, struct super_block *sb, +static int server_register_mount(const char *name, struct super_block *sb, struct vfsmount *mnt) { struct lustre_mount_info *lmi; @@ -106,7 +124,7 @@ static int server_register_mount(char *name, struct super_block *sb, } /* when an obd no longer needs a mount */ -static int server_deregister_mount(char *name) +static int server_deregister_mount(const char *name) { struct lustre_mount_info *lmi; ENTRY; @@ -133,7 +151,7 @@ static int server_deregister_mount(char *name) /* 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(char *name) +struct lustre_mount_info *server_get_mount(const char *name) { struct lustre_mount_info *lmi; struct lustre_sb_info *lsi; @@ -157,6 +175,25 @@ struct lustre_mount_info *server_get_mount(char *name) RETURN(lmi); } +/* + * 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; + + down(&lustre_mount_info_lock); + lmi = server_find_mount(name); + up(&lustre_mount_info_lock); + if (!lmi) + CERROR("Can't find mount for %s\n", name); + + RETURN(lmi); +} + static void unlock_mntput(struct vfsmount *mnt) { if (kernel_locked()) { @@ -171,12 +208,16 @@ static void unlock_mntput(struct vfsmount *mnt) static int lustre_put_lsi(struct super_block *sb); /* to be called from obd_cleanup methods */ -int server_put_mount(char *name, struct vfsmount *mnt) +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; ENTRY; + /* This might be the last one, can't deref after this */ + unlock_mntput(mnt); + down(&lustre_mount_info_lock); lmi = server_find_mount(name); up(&lustre_mount_info_lock); @@ -186,20 +227,16 @@ int server_put_mount(char *name, struct vfsmount *mnt) } lsi = s2lsi(lmi->lmi_sb); LASSERT(lmi->lmi_mnt == mnt); - unlock_mntput(lmi->lmi_mnt); CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n", - lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts), - atomic_read(&lmi->lmi_mnt->mnt_count)); + lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts), count); if (lustre_put_lsi(lmi->lmi_sb)) { CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n", - lmi->lmi_mnt, name, - atomic_read(&lmi->lmi_mnt->mnt_count)); + lmi->lmi_mnt, name, count); /* last mount is the One True Mount */ - if (atomic_read(&lmi->lmi_mnt->mnt_count) > 1) - CERROR("%s: mount busy, vfscount=%d!\n", name, - atomic_read(&lmi->lmi_mnt->mnt_count)); + if (count > 1) + CERROR("%s: mount busy, vfscount=%d!\n", name, count); } /* this obd should never need the mount again */ @@ -208,9 +245,16 @@ int server_put_mount(char *name, struct vfsmount *mnt) RETURN(0); } +/* Corresponding to server_get_mount_2 */ +int server_put_mount_2(const char *name, struct vfsmount *mnt) +{ + ENTRY; + RETURN(0); +} /******* mount helper utilities *********/ +#if 0 static void ldd_print(struct lustre_disk_data *ldd) { PRINT_CMD(PRINT_MASK, " disk data:\n"); @@ -225,6 +269,7 @@ static void ldd_print(struct lustre_disk_data *ldd) PRINT_CMD(PRINT_MASK, "params: %s\n", ldd->ldd_params); PRINT_CMD(PRINT_MASK, "comment: %s\n", ldd->ldd_userdata); } +#endif static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt, struct lustre_disk_data *ldd) @@ -245,7 +290,7 @@ 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", @@ -282,8 +327,6 @@ static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt, GOTO(out_close, rc = -EINVAL); } - ldd_print(ldd); - out_close: filp_close(file, 0); out: @@ -322,7 +365,6 @@ static int ldd_write(struct lvfs_run_ctxt *mount_ctxt, } rc = 0; - ldd_print(ldd); out_close: filp_close(file, 0); @@ -362,19 +404,20 @@ int lustre_process_log(struct super_block *sb, char *logname, lustre_cfg_free(lcfg); if (rc == -EINVAL) - LCONSOLE_ERROR("%s: The configuration from log '%s' failed " - "(%d). Make sure this client and " - "the MGS are running compatible versions of " - "Lustre.\n", - mgc->obd_name, logname, rc); + LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'" + "failed from the MGS (%d). Make sure this " + "client and the MGS are running compatible " + "versions of Lustre.\n", + mgc->obd_name, logname, rc); if (rc) - LCONSOLE_ERROR("%s: The configuration from log '%s' failed " - "(%d). This may be the result of " - "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, rc); + LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' " + "failed (%d). This may be the result of " + "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, + rc); /* class_obd_list(); */ RETURN(rc); @@ -468,8 +511,9 @@ static int server_start_mgs(struct super_block *sb) lmi = server_find_mount(LUSTRE_MGS_OBDNAME); if (lmi) { lsi = s2lsi(lmi->lmi_sb); - LCONSOLE_ERROR("The MGS service was already started from " - "server %s\n", lsi->lsi_ldd->ldd_svname); + LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started" + " from server %s\n", + lsi->lsi_ldd->ldd_svname); RETURN(-EALREADY); } @@ -477,15 +521,18 @@ 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("Failed to start MGS '%s' (%d). Is the 'mgs' " - "module loaded?\n", LUSTRE_MGS_OBDNAME, rc); - + LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). " + "Is the 'mgs' module loaded?\n", + LUSTRE_MGS_OBDNAME, rc); RETURN(rc); } @@ -512,18 +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; @@ -566,10 +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); @@ -588,7 +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); @@ -640,7 +699,7 @@ static int lustre_start_mgc(struct super_block *sb) /* Random uuid for MGC allows easier reconnects */ OBD_ALLOC_PTR(uuid); - class_generate_random_uuid(uuidc); + ll_generate_random_uuid(uuidc); class_uuid_unparse(uuidc, uuid); /* Start the MGC */ @@ -682,6 +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); @@ -689,21 +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(&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: @@ -724,8 +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) @@ -733,9 +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. */ @@ -744,15 +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 */ @@ -769,7 +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, @@ -778,10 +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); } @@ -798,7 +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); @@ -815,7 +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); } @@ -872,9 +947,9 @@ int server_mti_print(char *title, struct mgs_target_info *mti) 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; - lnet_process_id_t id; + struct lustre_sb_info *lsi = s2lsi(sb); + struct lustre_disk_data *ldd = lsi->lsi_ldd; + lnet_process_id_t id; int i = 0; ENTRY; @@ -943,12 +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; @@ -966,7 +1039,6 @@ int server_register_target(struct super_block *sb) sizeof(ldd->ldd_svname)); /* or ldd_make_sv_name(ldd); */ ldd_write(&mgc->obd_lvfs_ctxt, ldd); - err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb, mti->mti_svname); if (err) @@ -996,6 +1068,7 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname); +#if 0 /* 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 */ @@ -1015,6 +1088,7 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) } mutex_up(&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) { @@ -1048,17 +1122,21 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) CERROR("Required registration failed for %s: %d\n", lsi->lsi_ldd->ldd_svname, rc); if (rc == -EIO) { - LCONSOLE_ERROR("Communication error with the MGS. Is " - "the MGS running?\n"); + LCONSOLE_ERROR_MSG(0x15f, "Communication error with " + "the MGS. Is the MGS running?\n"); } GOTO(out_mgc, rc); } if (rc == -EINVAL) { - LCONSOLE_ERROR("The MGS is refusing to allow this server (%s) " - "to start. Please see messages on the MGS node." - "\n", lsi->lsi_ldd->ldd_svname); + 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); /* Let the target look up the mount using the target's name (we can't pass the sb or mnt through class_process_config.) */ @@ -1072,6 +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); } @@ -1092,6 +1173,9 @@ out_mgc: obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0, NULL, NULL); } + + /* log has been fully processed */ + obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG); } RETURN(rc); @@ -1101,15 +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); } @@ -1120,6 +1204,7 @@ struct lustre_sb_info *lustre_init_lsi(struct super_block *sb) /* Default umount style */ lsi->lsi_flags = LSI_UMOUNT_FAILOVER; + RETURN(lsi); } @@ -1128,10 +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); @@ -1146,6 +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_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); @@ -1157,7 +1243,6 @@ static int lustre_free_lsi(struct super_block *sb) } LASSERT(lsi->lsi_llsbi == NULL); - OBD_FREE(lsi, sizeof(*lsi)); s2lsi_nocast(sb) = NULL; @@ -1171,10 +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); @@ -1194,6 +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; @@ -1205,24 +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)) - CERROR("premount %s:%#lx ldiskfs failed (%d), is the ldiskfs " - "module available?\n", lmd->lmd_dev, s_flags, rc); + CERROR("premount %s:%#lx ldiskfs failed: %d " + "Is the ldiskfs module available?\n", + lmd->lmd_dev, s_flags, rc ); GOTO(out_free, rc); -#else - /* If ldisk 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 } OBD_SET_CTXT_MAGIC(&mount_ctxt); @@ -1241,11 +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); @@ -1265,18 +1350,24 @@ 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)); @@ -1286,13 +1377,13 @@ static void server_wait_finished(struct vfsmount *mnt) { wait_queue_head_t waitq; struct l_wait_info lwi; - int retries = 120; + int retries = 330; 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", + LCONSOLE_WARN("%s: Mount still busy with %d refs, waiting for " + "%d secs...\n", mnt->mnt_devname, atomic_read(&mnt->mnt_count), retries); /* Wait for a bit */ @@ -1301,8 +1392,8 @@ static void server_wait_finished(struct vfsmount *mnt) 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)); + CERROR("%s: Mount still busy (%d refs), giving up.\n", + mnt->mnt_devname, atomic_read(&mnt->mnt_count)); } } @@ -1311,12 +1402,10 @@ static void server_put_super(struct super_block *sb) struct lustre_sb_info *lsi = s2lsi(sb); struct obd_device *obd; struct vfsmount *mnt = lsi->lsi_srv_mnt; - lvfs_sbdev_type save_dev; char *tmpname, *extraname = NULL; int tmpname_sz; int lddflags = lsi->lsi_ldd->ldd_flags; int lsiflags = lsi->lsi_flags; - int rc; ENTRY; LASSERT(lsiflags & LSI_SERVER); @@ -1325,9 +1414,12 @@ 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 */ @@ -1337,43 +1429,36 @@ 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_ldd->ldd_svname); - if (lprof && lprof->lp_osc) { - OBD_ALLOC(extraname, strlen(lprof->lp_osc) + 1); - strcpy(extraname, lprof->lp_osc); + if (lprof && lprof->lp_dt) { + OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1); + strcpy(extraname, lprof->lp_dt); } 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 { CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname); server_deregister_mount(lsi->lsi_ldd->ldd_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->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); } - save_dev = lvfs_sbdev(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 */ @@ -1381,9 +1466,6 @@ static void server_put_super(struct super_block *sb) /* drop the One True Mount */ unlock_mntput(mnt); -#ifndef LUSTRE_PATCHLESS - lvfs_clear_rdonly(save_dev); -#endif /* Stop the servers (MDS, OSS) if no longer needed. We must wait until the target is really gone so that our type refcount check @@ -1405,6 +1487,7 @@ static void server_put_super(struct super_block *sb) OBD_FREE(tmpname, tmpname_sz); EXIT; } + #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT static void server_umount_begin(struct vfsmount *vfsmnt, int flags) { @@ -1524,9 +1607,9 @@ 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; @@ -1536,16 +1619,17 @@ static int server_fill_super(struct super_block *sb) lsi->lsi_lmd->lmd_dev); if (class_name2obd(lsi->lsi_ldd->ldd_svname)) { - LCONSOLE_ERROR("The target named %s is already running. " - "Double-mount may have compromised the disk " - "journal.\n", lsi->lsi_ldd->ldd_svname); - unlock_mntput(mnt); + LCONSOLE_ERROR_MSG(0x161, "The target named %s is already " + "running. Double-mount may have compromised" + " the disk journal.\n", + lsi->lsi_ldd->ldd_svname); 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); @@ -1579,14 +1663,17 @@ 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. @@ -1597,16 +1684,27 @@ 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; @@ -1636,9 +1734,11 @@ 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); } +#if 0 static void lmd_print(struct lustre_mount_data *lmd) { int i; @@ -1655,6 +1755,7 @@ static void lmd_print(struct lustre_mount_data *lmd) lmd->lmd_exclude[i]); } } +#endif /* Is this server on the exclusion list */ int lustre_check_exclusion(struct super_block *sb, char *svname) @@ -1715,7 +1816,6 @@ static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr) s1 = s2; /* now we are pointing at ':' (next exclude) or ',' (end of excludes) */ - if (lmd->lmd_exclude_count >= devmax) break; } @@ -1738,6 +1838,31 @@ static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr) RETURN(rc); } +static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr) +{ + char *tail; + int length; + + if (lmd->lmd_mgssec != NULL) { + OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1); + lmd->lmd_mgssec = NULL; + } + + tail = strchr(ptr, ','); + if (tail == NULL) + length = strlen(ptr); + else + length = tail - ptr; + + 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; +} + /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */ static int lmd_parse(char *options, struct lustre_mount_data *lmd) { @@ -1748,16 +1873,16 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) LASSERT(lmd); if (!options) { - LCONSOLE_ERROR("Missing mount data: check that " - "/sbin/mount.lustre is installed.\n"); + 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("You're using an old version of " - "/sbin/mount.lustre. Please install version " - "%s\n", LUSTRE_VERSION_STRING); + 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; @@ -1782,6 +1907,14 @@ 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++; + } 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++; /* ost exclusion list */ } else if (strncmp(s1, "exclude=", 8) == 0) { rc = lmd_make_exclusion(lmd, s1 + 7); @@ -1789,7 +1922,6 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) 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) { @@ -1815,13 +1947,14 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) } if (!devname) { - LCONSOLE_ERROR("Can't find the device name " - "(need mount option 'device=...')\n"); + LCONSOLE_ERROR_MSG(0x164, "Can't find the device name " + "(need mount option 'device=...')\n"); goto invalid; } - s1 = strrchr(devname, ':'); + s1 = strstr(devname, ":/"); if (s1) { + ++s1; lmd->lmd_flags = LMD_FLG_CLIENT; /* Remove leading /s from fsname */ while (*++s1 == '/') ; @@ -1852,7 +1985,6 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) lmd->lmd_magic = LMD_MAGIC; - lmd_print(lmd); RETURN(rc); invalid: @@ -1876,29 +2008,36 @@ 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)) { CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile); if (!client_fill_super) { - LCONSOLE_ERROR("Nothing registered for client mount!" - " Is the 'lustre' module loaded?\n"); + 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) */ rc = (*client_fill_super)(sb); /* c_f_s will call lustre_common_put_super on failure */ - } } else { CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev); @@ -1910,14 +2049,19 @@ 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; } @@ -1928,17 +2072,17 @@ 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 @@ -1946,42 +2090,28 @@ 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) { @@ -1994,14 +2124,15 @@ 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); EXPORT_SYMBOL(server_get_mount); +EXPORT_SYMBOL(server_get_mount_2); EXPORT_SYMBOL(server_put_mount); +EXPORT_SYMBOL(server_put_mount_2); EXPORT_SYMBOL(server_register_target); EXPORT_SYMBOL(server_name2index); EXPORT_SYMBOL(server_mti_print); EXPORT_SYMBOL(do_lcfg); - -