X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdclass%2Fobd_mount.c;h=e2cd921128ece8c4eaf9697ad2ec6a15e29fcb7d;hp=39fe240ca58cc3983494f690465073ac1171375f;hb=f5fb215365791507ec98796ea624abeafa31901c;hpb=11d258632de7255ae8282b0615a19bd5c9cf707a diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c index 39fe240..e2cd921 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. @@ -29,8 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011 Whamcloud, Inc. - * + * Copyright (c) 2011, 2012, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -58,6 +55,9 @@ #include #include #include +#ifdef HAVE_KERNEL_LOCKED +#include +#endif static int (*client_fill_super)(struct super_block *sb, struct vfsmount *mnt) = NULL; @@ -65,7 +65,7 @@ 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) @@ -93,7 +93,6 @@ static int server_register_mount(const char *name, struct super_block *sb, char *name_cp; ENTRY; - LASSERT(mnt); LASSERT(sb); OBD_ALLOC(lmi, sizeof(*lmi)); @@ -106,10 +105,10 @@ 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); @@ -120,10 +119,11 @@ 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, + lmi->lmi_mnt ? mnt_get_count(lmi->lmi_mnt) : -1); RETURN(0); } @@ -134,21 +134,22 @@ 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, + lmi->lmi_mnt ? mnt_get_count(lmi->lmi_mnt) : -1); 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); } @@ -162,20 +163,22 @@ 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); } lsi = s2lsi(lmi->lmi_sb); - mntget(lmi->lmi_mnt); + + if (lmi->lmi_mnt) + mntget(lmi->lmi_mnt); cfs_atomic_inc(&lsi->lsi_mounts); 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)); + lmi->lmi_mnt ? mnt_get_count(lmi->lmi_mnt) - 1 : -1); RETURN(lmi); } @@ -190,9 +193,9 @@ 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); @@ -201,13 +204,18 @@ 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); @@ -217,15 +225,18 @@ 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 = 0; ENTRY; /* This might be the last one, can't deref after this */ - unlock_mntput(mnt); + if (mnt) { + count = mnt_get_count(mnt) - 1; + 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); @@ -349,6 +360,9 @@ static int ldd_write(struct lvfs_run_ctxt *mount_ctxt, int rc = 0; ENTRY; + if (ldd->ldd_magic == 0) + RETURN(0); + LASSERT(ldd->ldd_magic == LDD_MAGIC); ldd->ldd_config_ver++; @@ -581,7 +595,7 @@ 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 * @@ -609,8 +623,12 @@ static int lustre_start_mgc(struct super_block *sb) /* Find the first non-lo MGS nid for our MGC name */ if (lsi->lsi_flags & LSI_SERVER) { ptr = lsi->lsi_ldd->ldd_params; + /* mount -o mgsnode=nid */ + if (lsi->lsi_lmd->lmd_mgs && + (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) { + i++; /* Use mgsnode= nids */ - if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) && + } else if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) && (class_parse_nid(ptr, &nid, &ptr) == 0)) { i++; } else if (IS_MGS(lsi->lsi_ldd)) { @@ -634,7 +652,7 @@ static int lustre_start_mgc(struct super_block *sb) RETURN(-EINVAL); } - cfs_mutex_down(&mgc_start_lock); + cfs_mutex_lock(&mgc_start_lock); len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1; OBD_ALLOC(mgcname, len); @@ -651,7 +669,7 @@ static int lustre_start_mgc(struct super_block *sb) 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) @@ -666,7 +684,7 @@ static int lustre_start_mgc(struct super_block *sb) int vallen = sizeof(*data); __u32 *flags = &lsi->lsi_lmd->lmd_flags; - rc = obd_get_info(obd->obd_self_export, + rc = obd_get_info(NULL, obd->obd_self_export, strlen(KEY_CONN_DATA), KEY_CONN_DATA, &vallen, data, NULL); LASSERT(rc == 0); @@ -700,7 +718,7 @@ 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); @@ -723,7 +741,11 @@ static int lustre_start_mgc(struct super_block *sb) } } else { /* Use mgsnode= nids */ - if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) { + /* mount -o mgsnode=nid */ + if (lsi->lsi_lmd->lmd_mgs) { + ptr = lsi->lsi_lmd->lmd_mgs; + } else if (class_find_param(ptr, PARAM_MGSNODE, + &ptr) != 0) { CERROR("No MGS nids given.\n"); GOTO(out_free, rc = -EINVAL); } @@ -795,7 +817,7 @@ 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) @@ -807,7 +829,7 @@ 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); @@ -835,7 +857,7 @@ 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); @@ -861,7 +883,7 @@ 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 @@ -913,7 +935,7 @@ 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); } @@ -928,7 +950,7 @@ 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) { @@ -945,13 +967,13 @@ 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) @@ -961,7 +983,7 @@ 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 */ @@ -985,7 +1007,7 @@ 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); } @@ -1093,7 +1115,7 @@ 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, + 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) { @@ -1130,6 +1152,8 @@ 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); + if (lsi->lsi_lmd->lmd_osd_type) + goto out; err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb, mti->mti_svname); if (err) @@ -1176,7 +1200,7 @@ static int server_notify_target(struct super_block *sb, struct obd_device *obd) mti->mti_flags |= LDD_F_OPC_READY; /* 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); @@ -1209,7 +1233,7 @@ 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, @@ -1218,19 +1242,19 @@ 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 (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, @@ -1238,19 +1262,21 @@ 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 * read and write configs locally, in case it can't talk to the MGS. */ - rc = server_mgc_set_fs(lsi->lsi_mgc, sb); - if (rc) - RETURN(rc); + if (lsi->lsi_lmd->lmd_osd_type == NULL) { + rc = server_mgc_set_fs(lsi->lsi_mgc, sb); + if (rc) + RETURN(rc); + } /* Register with MGS */ rc = server_register_target(sb); @@ -1277,7 +1303,8 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) out_mgc: /* Release the mgc fs for others to use */ - server_mgc_clear_fs(lsi->lsi_mgc); + if (lsi->lsi_lmd->lmd_osd_type == NULL) + server_mgc_clear_fs(lsi->lsi_mgc); if (!rc) { obd = class_name2obd(lsi->lsi_ldd->ldd_svname); @@ -1365,6 +1392,13 @@ static int lustre_free_lsi(struct super_block *sb) OBD_FREE(lsi->lsi_lmd->lmd_exclude, sizeof(lsi->lsi_lmd->lmd_exclude[0]) * lsi->lsi_lmd->lmd_exclude_count); + if (lsi->lsi_lmd->lmd_mgs != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_mgs, + strlen(lsi->lsi_lmd->lmd_mgs) + 1); + if (lsi->lsi_lmd->lmd_osd_type != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_osd_type, + strlen(lsi->lsi_lmd->lmd_osd_type) + 1); + OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd)); } @@ -1392,6 +1426,78 @@ static int lustre_put_lsi(struct super_block *sb) RETURN(0); } +static int lsi_prepare(struct lustre_sb_info *lsi) +{ + struct lustre_disk_data *ldd; + __u32 index; + int rc; + ENTRY; + + LASSERT(lsi); + LASSERT(lsi->lsi_lmd); + + OBD_ALLOC(ldd, sizeof(*ldd)); + if (ldd == NULL) + RETURN(-ENOMEM); + strcpy(lsi->lsi_osd_type, LUSTRE_OSD_NAME); + + /* The server name is given as a mount line option */ + if (lsi->lsi_lmd->lmd_profile == NULL) { + LCONSOLE_ERROR("Can't determine server name\n"); + RETURN(-EINVAL); + } + + if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(ldd->ldd_svname)) + RETURN(-ENAMETOOLONG); + + strcpy(ldd->ldd_svname, lsi->lsi_lmd->lmd_profile); + strcpy(ldd->ldd_fsname, "lustre"); + + /* Determine osd type */ + if (lsi->lsi_lmd->lmd_osd_type != NULL) { + if (strlen(lsi->lsi_lmd->lmd_osd_type) >= + sizeof(lsi->lsi_osd_type)) + RETURN (-ENAMETOOLONG); + + strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type); + } + + /* Determine server type */ + rc = server_name2index(ldd->ldd_svname, &index, NULL); + if (rc < 0) { + if (0 /*lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS*/) { + /* Assume we're a bare MGS */ + rc = 0; + lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC; + } else { + LCONSOLE_ERROR("Can't determine server type of '%s'\n", + lsi->lsi_svname); + RETURN(rc); + } + } + ldd->ldd_svindex = index; + //lsi->lsi_flags |= rc; + ldd->ldd_flags = rc | LDD_F_WRITECONF; + + lsi->lsi_ldd = ldd; + + /* Add mount line flags that used to be in ldd: + * writeconf, mgs, iam, anything else? + */ +#if 0 + lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ? + LDD_F_WRITECONF : 0; + lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ? + LDD_F_SV_TYPE_MGS : 0; + lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_IAM) ? + LDD_F_IAM_DIR : 0; + lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ? + LDD_F_NO_PRIMNODE : 0; +#endif + + RETURN(0); +} + /*************** server mount ******************/ /** Kernel mount using mount options in MOUNT_DATA_FILE. @@ -1406,6 +1512,7 @@ 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; @@ -1413,9 +1520,15 @@ static struct vfsmount *server_kernel_mount(struct super_block *sb) int rc; ENTRY; + if (lsi->lsi_lmd->lmd_osd_type) { + rc = lsi_prepare(lsi); + RETURN(ERR_PTR(rc)); + } + OBD_ALLOC(ldd, sizeof(*ldd)); if (!ldd) RETURN(ERR_PTR(-ENOMEM)); + strcpy(lsi->lsi_osd_type, LUSTRE_OSD_NAME); /* In the past, we have always used flags = 0. Note ext3/ldiskfs can't be mounted ro. */ @@ -1437,7 +1550,13 @@ 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 " @@ -1483,11 +1602,16 @@ 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); } @@ -1518,27 +1642,35 @@ static void server_wait_finished(struct vfsmount *mnt) int rc, waited = 0; cfs_sigset_t blocked; + if (mnt == NULL) { + cfs_waitq_init(&waitq); + cfs_waitq_wait_event_interruptible_timeout(waitq, 0, + cfs_time_seconds(3), rc); + return; + } + + LASSERT(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 = 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)); + mnt_get_count(mnt)); break; } @@ -1615,16 +1747,17 @@ static void server_put_super(struct super_block *sb) server_wait_finished(mnt); /* drop the One True Mount */ - unlock_mntput(mnt); + if (mnt) + unlock_mntput(mnt); - /* Stop the servers (MDS, OSS) if no longer needed. We must wait - until the target is really gone so that our type refcount check - is right. */ - server_stop_servers(lddflags, lsiflags); + /* Stop the servers (MDS, OSS) if no longer needed. We must wait + until the target is really gone so that our type refcount check + is right. */ + server_stop_servers(lddflags, lsiflags); - /* In case of startup or cleanup err, stop related obds */ - if (extraname) { - obd = class_name2obd(extraname); + /* In case of startup or cleanup err, stop related obds */ + if (extraname) { + obd = class_name2obd(extraname); if (obd) { CWARN("Cleaning orphaned obd %s\n", extraname); obd->obd_force = 1; @@ -1725,7 +1858,7 @@ 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; @@ -2087,6 +2220,73 @@ static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr) return 0; } +static int lmd_parse_string(char **handle, char *ptr) +{ + char *tail; + int length; + + if ((handle == NULL) || (ptr == NULL)) + return -EINVAL; + + if (*handle != NULL) { + OBD_FREE(*handle, strlen(*handle) + 1); + *handle = NULL; + } + + tail = strchr(ptr, ','); + if (tail == NULL) + length = strlen(ptr); + else + length = tail - ptr; + + OBD_ALLOC(*handle, length + 1); + if (*handle == NULL) + return -ENOMEM; + + memcpy(*handle, ptr, length); + (*handle)[length] = '\0'; + + return 0; +} + +/* Collect multiple values for mgsnid specifiers */ +static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr) +{ + lnet_nid_t nid; + char *tail = *ptr; + char *mgsnid; + int length; + int oldlen = 0; + + /* Find end of nidlist */ + while (class_parse_nid(tail, &nid, &tail) == 0) {} + length = tail - *ptr; + if (length == 0) { + LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr); + return -EINVAL; + } + + if (lmd->lmd_mgs != NULL) + oldlen = strlen(lmd->lmd_mgs) + 1; + + OBD_ALLOC(mgsnid, oldlen + length + 1); + if (mgsnid == NULL) + return -ENOMEM; + + if (lmd->lmd_mgs != NULL) { + /* Multiple mgsnid= are taken to mean failover locations */ + memcpy(mgsnid, lmd->lmd_mgs, oldlen); + mgsnid[oldlen - 1] = ':'; + OBD_FREE(lmd->lmd_mgs, oldlen); + } + memcpy(mgsnid + oldlen, *ptr, length); + mgsnid[oldlen + length] = '\0'; + lmd->lmd_mgs = mgsnid; + *ptr = tail; + + return 0; +} + /** Parse mount line options * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre * dev is passed as device=uml1:/lustre by mount.lustre @@ -2150,6 +2350,18 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) } 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, PARAM_MGSNODE, + sizeof(PARAM_MGSNODE) - 1) == 0) { + s2 = s1 + sizeof(PARAM_MGSNODE) - 1; + /* Assume the next mount opt is the first + invalid nid we get to. */ + rc = lmd_parse_mgs(lmd, &s2); + if (rc) + goto invalid; + clear++; } else if (strncmp(s1, "writeconf", 9) == 0) { lmd->lmd_flags |= LMD_FLG_WRITECONF; clear++; @@ -2164,6 +2376,23 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) if (rc) goto invalid; clear++; + } else if (strncmp(s1, "svname=", 7) == 0) { + rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7); + if (rc) + goto invalid; + clear++; + } else if (strncmp(s1, "osd=", 4) == 0) { + rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4); + if (rc) + goto invalid; + /* with ldiskfs we're still doing ldd parsing + * in the kernel space */ + if (!strcmp(lmd->lmd_osd_type, "osd-ldiskfs")) { + OBD_FREE(lmd->lmd_osd_type, + strlen(lmd->lmd_osd_type) + 1); + lmd->lmd_osd_type = NULL; + } + clear++; } /* Linux 2.4 doesn't pass the device, so we stuck it at the end of the options. */