X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdclass%2Fobd_mount.c;h=0f70f44280ed4c757ce9827302f5eea8d905ece7;hp=9f02b7db436c4de707258550592b7a797142ba68;hb=11317196203c26f8c9e4d2a8261859b1c7a76c9d;hpb=e3a7c58aebafce40323db54bf6056029e5af4a70 diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c index 9f02b7d..0f70f44 100644 --- a/lustre/obdclass/obd_mount.c +++ b/lustre/obdclass/obd_mount.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Whamcloud, Inc. + * Copyright (c) 2011, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -35,20 +35,17 @@ * * lustre/obdclass/obd_mount.c * - * Client/server mount routines + * Client mount routines * * Author: Nathan Rutman */ #define DEBUG_SUBSYSTEM S_CLASS -#define D_MOUNT D_SUPER|D_CONFIG /*|D_WARNING */ +#define D_MOUNT (D_SUPER|D_CONFIG/*|D_WARNING */) #define PRINT_CMD CDEBUG -#define PRINT_MASK D_SUPER|D_CONFIG #include -#include -#include #include #include #include @@ -57,324 +54,9 @@ #include static int (*client_fill_super)(struct super_block *sb, - struct vfsmount *mnt) = NULL; -static void (*kill_super_cb)(struct super_block *sb) = NULL; - -/*********** mount lookup *********/ - -CFS_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) -{ - cfs_list_t *tmp; - struct lustre_mount_info *lmi; - ENTRY; - - cfs_list_for_each(tmp, &server_mount_info_list) { - lmi = cfs_list_entry(tmp, struct lustre_mount_info, - lmi_list_chain); - if (strcmp(name, lmi->lmi_name) == 0) - RETURN(lmi); - } - RETURN(NULL); -} - -/* 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(const char *name, struct super_block *sb, - struct vfsmount *mnt) -{ - struct lustre_mount_info *lmi; - char *name_cp; - ENTRY; - - LASSERT(mnt); - LASSERT(sb); - - OBD_ALLOC(lmi, sizeof(*lmi)); - if (!lmi) - RETURN(-ENOMEM); - OBD_ALLOC(name_cp, strlen(name) + 1); - if (!name_cp) { - OBD_FREE(lmi, sizeof(*lmi)); - RETURN(-ENOMEM); - } - strcpy(name_cp, name); - - cfs_mutex_lock(&lustre_mount_info_lock); - - if (server_find_mount(name)) { - 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); - RETURN(-EEXIST); - } - lmi->lmi_name = name_cp; - lmi->lmi_sb = sb; - lmi->lmi_mnt = mnt; - cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list); - - cfs_mutex_unlock(&lustre_mount_info_lock); - - CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n", - lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt)); - - RETURN(0); -} - -/* when an obd no longer needs a mount */ -static int server_deregister_mount(const char *name) -{ - struct lustre_mount_info *lmi; - ENTRY; - - cfs_mutex_lock(&lustre_mount_info_lock); - lmi = server_find_mount(name); - if (!lmi) { - 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, mnt_get_count(lmi->lmi_mnt)); - - OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1); - cfs_list_del(&lmi->lmi_list_chain); - OBD_FREE(lmi, sizeof(*lmi)); - cfs_mutex_unlock(&lustre_mount_info_lock); - - RETURN(0); -} - -/* 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(const char *name) -{ - struct lustre_mount_info *lmi; - struct lustre_sb_info *lsi; - ENTRY; - - cfs_mutex_lock(&lustre_mount_info_lock); - lmi = server_find_mount(name); - 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); - 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), - mnt_get_count(lmi->lmi_mnt)); - - 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; - - cfs_mutex_lock(&lustre_mount_info_lock); - lmi = server_find_mount(name); - cfs_mutex_unlock(&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()) { - cfs_unlock_kernel(); - mntput(mnt); - cfs_lock_kernel(); - } else { - mntput(mnt); - } -} - -static int lustre_put_lsi(struct super_block *sb); - -/* to be called from obd_cleanup methods */ -int server_put_mount(const char *name, struct vfsmount *mnt) -{ - struct lustre_mount_info *lmi; - struct lustre_sb_info *lsi; - int count = mnt_get_count(mnt) - 1; - ENTRY; - - /* This might be the last one, can't deref after this */ - unlock_mntput(mnt); - - cfs_mutex_lock(&lustre_mount_info_lock); - lmi = server_find_mount(name); - cfs_mutex_unlock(&lustre_mount_info_lock); - if (!lmi) { - CERROR("Can't find mount for %s\n", name); - RETURN(-ENOENT); - } - lsi = s2lsi(lmi->lmi_sb); - LASSERT(lmi->lmi_mnt == mnt); - - CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n", - lmi->lmi_mnt, name, cfs_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, count); - /* last mount is the One True Mount */ - if (count > 1) - CERROR("%s: mount busy, vfscount=%d!\n", name, count); - } - - /* this obd should never need the mount again */ - server_deregister_mount(name); - - 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"); - PRINT_CMD(PRINT_MASK, "server: %s\n", ldd->ldd_svname); - PRINT_CMD(PRINT_MASK, "uuid: %s\n", (char *)ldd->ldd_uuid); - PRINT_CMD(PRINT_MASK, "fs: %s\n", ldd->ldd_fsname); - PRINT_CMD(PRINT_MASK, "index: %04x\n", ldd->ldd_svindex); - PRINT_CMD(PRINT_MASK, "config: %d\n", ldd->ldd_config_ver); - PRINT_CMD(PRINT_MASK, "flags: %#x\n", ldd->ldd_flags); - PRINT_CMD(PRINT_MASK, "diskfs: %s\n", MT_STR(ldd)); - PRINT_CMD(PRINT_MASK, "options: %s\n", ldd->ldd_mount_opts); - 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) -{ - struct lvfs_run_ctxt saved; - struct file *file; - loff_t off = 0; - unsigned long len; - int rc; - ENTRY; - - push_ctxt(&saved, mount_ctxt, NULL); - - file = filp_open(MOUNT_DATA_FILE, O_RDONLY, 0644); - if (IS_ERR(file)) { - rc = PTR_ERR(file); - CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc); - GOTO(out, rc); - } - - 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 %u\n", - len, (int)sizeof(*ldd)); - GOTO(out_close, rc = -EINVAL); - } - - rc = lustre_fread(file, ldd, len, &off); - if (rc != len) { - CERROR("error reading %s: read %d of %lu\n", - MOUNT_DATA_FILE, rc, len); - GOTO(out_close, rc = -EINVAL); - } - rc = 0; - - if (ldd->ldd_magic != LDD_MAGIC) { - /* FIXME add swabbing support */ - CERROR("Bad magic in %s: %x!=%x\n", MOUNT_DATA_FILE, - ldd->ldd_magic, LDD_MAGIC); - GOTO(out_close, rc = -EINVAL); - } - - if (ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP) { - CERROR("%s: unsupported incompat filesystem feature(s) %x\n", - ldd->ldd_svname, - ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP); - GOTO(out_close, rc = -EINVAL); - } - if (ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP) { - CERROR("%s: unsupported read-only filesystem feature(s) %x\n", - ldd->ldd_svname, - ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP); - /* Do something like remount filesystem read-only */ - GOTO(out_close, rc = -EINVAL); - } - -out_close: - filp_close(file, 0); -out: - pop_ctxt(&saved, mount_ctxt, NULL); - RETURN(rc); -} - -static int ldd_write(struct lvfs_run_ctxt *mount_ctxt, - struct lustre_disk_data *ldd) -{ - struct lvfs_run_ctxt saved; - struct file *file; - loff_t off = 0; - unsigned long len = sizeof(struct lustre_disk_data); - int rc = 0; - ENTRY; - - LASSERT(ldd->ldd_magic == LDD_MAGIC); - - ldd->ldd_config_ver++; - - push_ctxt(&saved, mount_ctxt, NULL); - - file = filp_open(MOUNT_DATA_FILE, O_RDWR|O_SYNC, 0644); - if (IS_ERR(file)) { - rc = PTR_ERR(file); - CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc); - GOTO(out, rc); - } - - rc = lustre_fwrite(file, ldd, len, &off); - if (rc != len) { - CERROR("error writing %s: read %d of %lu\n", - MOUNT_DATA_FILE, rc, len); - GOTO(out_close, rc = -EINVAL); - } - - rc = 0; - -out_close: - filp_close(file, 0); -out: - pop_ctxt(&saved, mount_ctxt, NULL); - RETURN(rc); -} + struct vfsmount *mnt); +static void (*kill_super_cb)(struct super_block *sb); /**************** config llog ********************/ @@ -424,8 +106,7 @@ int lustre_process_log(struct super_block *sb, char *logname, "client and the MGS are running compatible " "versions of Lustre.\n", mgc->obd_name, logname, rc); - - if (rc) + else if (rc != 0) LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' " "failed (%d). This may be the result of " "communication errors between this node and " @@ -437,6 +118,7 @@ int lustre_process_log(struct super_block *sb, char *logname, /* class_obd_list(); */ RETURN(rc); } +EXPORT_SYMBOL(lustre_process_log); /* Stop watching this config log for updates */ int lustre_end_log(struct super_block *sb, char *logname, @@ -462,6 +144,7 @@ int lustre_end_log(struct super_block *sb, char *logname, lustre_cfg_free(lcfg); RETURN(rc); } +EXPORT_SYMBOL(lustre_end_log); /**************** obd start *******************/ @@ -494,91 +177,31 @@ int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd, lustre_cfg_free(lcfg); return(rc); } +EXPORT_SYMBOL(do_lcfg); /** Call class_attach and class_setup. These methods in turn call * obd type-specific methods. */ -static int lustre_start_simple(char *obdname, char *type, char *uuid, - char *s1, char *s2) +int lustre_start_simple(char *obdname, char *type, char *uuid, + char *s1, char *s2, char *s3, char *s4) { - int rc; - CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type); - - rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0); - if (rc) { - CERROR("%s attach error %d\n", obdname, rc); - return(rc); - } - rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0); - if (rc) { - CERROR("%s setup error %d\n", obdname, rc); - do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0); - } - return rc; -} - -/* Set up a MGS to serve startup logs */ -static int server_start_mgs(struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - struct vfsmount *mnt = lsi->lsi_srv_mnt; - struct lustre_mount_info *lmi; - int rc = 0; - ENTRY; - LASSERT(mnt); - - /* It is impossible to have more than 1 MGS per node, since - MGC wouldn't know which to connect to */ - lmi = server_find_mount(LUSTRE_MGS_OBDNAME); - if (lmi) { - lsi = s2lsi(lmi->lmi_sb); - LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started" - " from server %s\n", - lsi->lsi_ldd->ldd_svname); - RETURN(-EALREADY); - } - - CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME); - - 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); - /* 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). " - "Is the 'mgs' module loaded?\n", - LUSTRE_MGS_OBDNAME, rc); - RETURN(rc); -} - -static int server_stop_mgs(struct super_block *sb) -{ - struct obd_device *obd; - int rc; - ENTRY; - - CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME); - - /* There better be only one MGS */ - obd = class_name2obd(LUSTRE_MGS_OBDNAME); - if (!obd) { - CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME); - RETURN(-EALREADY); - } - - /* The MGS should always stop when we say so */ - obd->obd_force = 1; - rc = class_manual_cleanup(obd); - RETURN(rc); + int rc; + CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type); + + rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0); + if (rc) { + CERROR("%s attach error %d\n", obdname, rc); + return rc; + } + rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4); + if (rc) { + CERROR("%s setup error %d\n", obdname, rc); + do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0); + } + return rc; } -CFS_DEFINE_MUTEX(mgc_start_lock); +DEFINE_MUTEX(mgc_start_lock); /** Set up a mgc obd to process startup logs * @@ -586,7 +209,7 @@ CFS_DEFINE_MUTEX(mgc_start_lock); * * \retval 0 success, otherwise error code */ -static int lustre_start_mgc(struct super_block *sb) +int lustre_start_mgc(struct super_block *sb) { struct obd_connect_data *data = NULL; struct lustre_sb_info *lsi = s2lsi(sb); @@ -604,13 +227,13 @@ static int lustre_start_mgc(struct super_block *sb) LASSERT(lsi->lsi_lmd); /* Find the first non-lo MGS nid for our MGC name */ - if (lsi->lsi_flags & LSI_SERVER) { - ptr = lsi->lsi_ldd->ldd_params; - /* Use mgsnode= nids */ - if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) && - (class_parse_nid(ptr, &nid, &ptr) == 0)) { - i++; - } else if (IS_MGS(lsi->lsi_ldd)) { + if (IS_SERVER(lsi)) { + /* mount -o mgsnode=nid */ + ptr = lsi->lsi_lmd->lmd_mgs; + if (lsi->lsi_lmd->lmd_mgs && + (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) { + i++; + } else if (IS_MGS(lsi)) { lnet_process_id_t id; while ((rc = LNetGetId(i++, &id)) != -ENOENT) { if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND) @@ -631,7 +254,7 @@ static int lustre_start_mgc(struct super_block *sb) RETURN(-EINVAL); } - cfs_mutex_lock(&mgc_start_lock); + mutex_lock(&mgc_start_lock); len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1; OBD_ALLOC(mgcname, len); @@ -648,14 +271,14 @@ 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) GOTO(out_free, rc); /* Re-using an existing MGC */ - cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount); + atomic_inc(&obd->u.cli.cl_mgc_refcount); /* IR compatibility check, only for clients */ if (lmd_is_client(lsi->lsi_lmd)) { @@ -663,7 +286,7 @@ 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); @@ -686,7 +309,7 @@ static int lustre_start_mgc(struct super_block *sb) recov_bk = 0; /* If we are restarting the MGS, don't try to keep the MGC's old connection, or registration will fail. */ - if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) { + if (IS_MGS(lsi)) { CDEBUG(D_MOUNT, "New MGS with live MGC\n"); recov_bk = 1; } @@ -697,7 +320,7 @@ 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); @@ -709,34 +332,50 @@ static int lustre_start_mgc(struct super_block *sb) /* Add the primary nids for the MGS */ i = 0; sprintf(niduuid, "%s_%x", mgcname, i); - if (lsi->lsi_flags & LSI_SERVER) { - ptr = lsi->lsi_ldd->ldd_params; - if (IS_MGS(lsi->lsi_ldd)) { - /* Use local nids (including LO) */ - lnet_process_id_t id; - while ((rc = LNetGetId(i++, &id)) != -ENOENT) { - rc = do_lcfg(mgcname, id.nid, - LCFG_ADD_UUID, niduuid, 0,0,0); - } - } else { - /* Use mgsnode= nids */ - if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) { - CERROR("No MGS nids given.\n"); - GOTO(out_free, rc = -EINVAL); - } - while (class_parse_nid(ptr, &nid, &ptr) == 0) { - rc = do_lcfg(mgcname, nid, - LCFG_ADD_UUID, niduuid, 0,0,0); - i++; - } - } + if (IS_SERVER(lsi)) { + ptr = lsi->lsi_lmd->lmd_mgs; + CDEBUG(D_MOUNT, "mgs nids %s.\n", ptr); + if (IS_MGS(lsi)) { + /* Use local nids (including LO) */ + lnet_process_id_t id; + while ((rc = LNetGetId(i++, &id)) != -ENOENT) { + rc = do_lcfg(mgcname, id.nid, LCFG_ADD_UUID, + niduuid, 0, 0, 0); + } + } else { + /* Use mgsnode= nids */ + /* 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); + } + /* + * LU-3829. + * Here we only take the first mgsnid as its primary + * serving mgs node, the rest mgsnid will be taken as + * failover mgs node, otherwise they would be takens + * as multiple nids of a single mgs node. + */ + while (class_parse_nid(ptr, &nid, &ptr) == 0) { + rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID, + niduuid, 0, 0, 0); + if (rc == 0) { + i = 1; + break; + } + } + } } else { /* client */ /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */ ptr = lsi->lsi_lmd->lmd_dev; while (class_parse_nid(ptr, &nid, &ptr) == 0) { - rc = do_lcfg(mgcname, nid, - LCFG_ADD_UUID, niduuid, 0,0,0); - i++; + rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID, + niduuid, 0, 0, 0); + if (rc == 0) + ++i; /* Stop at the first failover nid */ if (*ptr == ':') break; @@ -755,30 +394,32 @@ static int lustre_start_mgc(struct super_block *sb) /* Start the MGC */ rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME, - (char *)uuid->uuid, LUSTRE_MGS_OBDNAME, - niduuid); + (char *)uuid->uuid, LUSTRE_MGS_OBDNAME, + niduuid, 0, 0); OBD_FREE_PTR(uuid); if (rc) GOTO(out_free, rc); /* Add any failover MGS nids */ i = 1; - while ((*ptr == ':' || - class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) { - /* New failover node */ - sprintf(niduuid, "%s_%x", mgcname, i); - j = 0; - while (class_parse_nid(ptr, &nid, &ptr) == 0) { - j++; - rc = do_lcfg(mgcname, nid, - LCFG_ADD_UUID, niduuid, 0,0,0); - if (*ptr == ':') - break; - } - if (j > 0) { - rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN, - niduuid, 0, 0, 0); - i++; + while (ptr && ((*ptr == ':' || + class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) { + /* New failover node */ + sprintf(niduuid, "%s_%x", mgcname, i); + j = 0; + while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) { + rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID, + niduuid, 0, 0, 0); + if (rc == 0) + ++j; + if (*ptr == ':') + break; + } + if (j > 0) { + rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN, + niduuid, 0, 0, 0); + if (rc == 0) + ++i; } else { /* at ":/fsname" */ break; @@ -792,7 +433,7 @@ 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) @@ -800,21 +441,29 @@ static int lustre_start_mgc(struct super_block *sb) /* Keep a refcount of servers/clients who started with "mount", so we know when we can get rid of the mgc. */ - cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1); + atomic_set(&obd->u.cli.cl_mgc_refcount, 1); /* 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); if (rc) /* nonfatal */ CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc); - /* We connect to the MGS at setup, and don't disconnect until cleanup */ - data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID | - OBD_CONNECT_AT | OBD_CONNECT_FULL20 | - OBD_CONNECT_IMP_RECOV; + + /* We connect to the MGS at setup, and don't disconnect until cleanup */ + data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT | + OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | + OBD_CONNECT_LVB_TYPE; + +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0) + data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB; +#else +#warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab" +#endif + if (lmd_is_client(lsi->lsi_lmd) && lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR) data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV; @@ -832,7 +481,7 @@ out: to the same mgc.*/ lsi->lsi_mgc = obd; out_free: - cfs_mutex_unlock(&mgc_start_lock); + mutex_unlock(&mgc_start_lock); if (data) OBD_FREE_PTR(data); @@ -858,13 +507,13 @@ static int lustre_stop_mgc(struct super_block *sb) RETURN(-ENOENT); lsi->lsi_mgc = NULL; - 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)) { + mutex_lock(&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. */ CDEBUG(D_MOUNT, "mgc still has %d references.\n", - cfs_atomic_read(&obd->u.cli.cl_mgc_refcount)); + atomic_read(&obd->u.cli.cl_mgc_refcount)); GOTO(out, rc = -EBUSY); } @@ -910,395 +559,7 @@ out: OBD_FREE(niduuid, len); /* class_import_put will get rid of the additional connections */ - cfs_mutex_unlock(&mgc_start_lock); - RETURN(rc); -} - -/* Since there's only one mgc per node, we have to change it's fs to get - access to the right disk. */ -static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - int rc; - ENTRY; - - 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, - sizeof(KEY_SET_FS), KEY_SET_FS, - sizeof(*sb), sb, NULL); - if (rc) { - CERROR("can't set_fs %d\n", rc); - } - - RETURN(rc); -} - -static int server_mgc_clear_fs(struct obd_device *mgc) -{ - int rc; - ENTRY; - - CDEBUG(D_MOUNT, "Unassign mgc disk\n"); - - rc = obd_set_info_async(mgc->obd_self_export, - sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS, - 0, NULL, NULL); - RETURN(rc); -} - -CFS_DEFINE_MUTEX(server_start_lock); - -/* Stop MDS/OSS if nobody is using them */ -static int server_stop_servers(int lddflags, int lsiflags) -{ - struct obd_device *obd = NULL; - struct obd_type *type = NULL; - int rc = 0; - ENTRY; - - 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 */ - if ((lddflags & LDD_F_SV_TYPE_MDT) && - (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) { - /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/ - type = class_search_type(LUSTRE_MDS_NAME); - } - /* if this was an OST, and there are no more OST's, clean up the OSS */ - if ((lddflags & LDD_F_SV_TYPE_OST) && - (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) { - type = class_search_type(LUSTRE_OST_NAME); - } - - if (obd && (!type || !type->typ_refcnt)) { - int err; - obd->obd_force = 1; - /* obd_fail doesn't mean much on a server obd */ - err = class_manual_cleanup(obd); - if (!rc) - rc = err; - } - - cfs_mutex_unlock(&server_start_lock); - - RETURN(rc); -} - -int server_mti_print(char *title, struct mgs_target_info *mti) -{ - PRINT_CMD(PRINT_MASK, "mti %s\n", title); - PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname); - PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname); - PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid); - PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n", - mti->mti_config_ver, mti->mti_flags); - 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; - lnet_process_id_t id; - int i = 0; - ENTRY; - - if (!(lsi->lsi_flags & LSI_SERVER)) - RETURN(-EINVAL); - - strncpy(mti->mti_fsname, ldd->ldd_fsname, - sizeof(mti->mti_fsname)); - strncpy(mti->mti_svname, ldd->ldd_svname, - sizeof(mti->mti_svname)); - - mti->mti_nid_count = 0; - while (LNetGetId(i++, &id) != -ENOENT) { - if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND) - continue; - - /* server use --servicenode param, only allow specified - * nids be registered */ - if ((ldd->ldd_flags & LDD_F_NO_PRIMNODE) != 0 && - class_match_nid(ldd->ldd_params, - PARAM_FAILNODE, id.nid) < 1) - continue; - - /* match specified network */ - if (!class_match_net(ldd->ldd_params, - PARAM_NETWORK, LNET_NIDNET(id.nid))) - continue; - - mti->mti_nids[mti->mti_nid_count] = id.nid; - mti->mti_nid_count++; - if (mti->mti_nid_count >= MTI_NIDS_MAX) { - CWARN("Only using first %d nids for %s\n", - mti->mti_nid_count, mti->mti_svname); - break; - } - } - - mti->mti_lustre_ver = LUSTRE_VERSION_CODE; - mti->mti_config_ver = 0; - if (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) - ldd->ldd_flags |= LDD_F_WRITECONF; - mti->mti_flags = ldd->ldd_flags; - mti->mti_stripe_index = ldd->ldd_svindex; - memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid)); - if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) { - CERROR("params too big for mti\n"); - RETURN(-ENOMEM); - } - memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params)); - RETURN(0); -} - -/* Register an old or new target with the MGS. If needed MGS will construct - startup logs and assign index */ -int server_register_target(struct super_block *sb) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - struct obd_device *mgc = lsi->lsi_mgc; - struct lustre_disk_data *ldd = lsi->lsi_ldd; - struct mgs_target_info *mti = NULL; - bool writeconf; - int rc; - ENTRY; - - LASSERT(mgc); - - if (!(lsi->lsi_flags & LSI_SERVER)) - RETURN(-EINVAL); - - OBD_ALLOC_PTR(mti); - if (!mti) - RETURN(-ENOMEM); - rc = server_sb2mti(sb, mti); - if (rc) - GOTO(out, rc); - - CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n", - mti->mti_svname, mti->mti_fsname, - libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index, - mti->mti_flags); - - /* if write_conf is true, the registration must succeed */ - writeconf = !!(ldd->ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE)); - mti->mti_flags |= LDD_F_OPC_REG; - - /* Register the target */ - /* FIXME use mgc_process_config instead */ - rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp, - sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET, - sizeof(*mti), mti, NULL); - if (rc) { - if (mti->mti_flags & LDD_F_ERROR) { - LCONSOLE_ERROR_MSG(0x160, - "The MGS is refusing to allow this " - "server (%s) to start. Please see messages" - " on the MGS node.\n", ldd->ldd_svname); - } else if (writeconf) { - LCONSOLE_ERROR_MSG(0x15f, - "Communication to the MGS return error %d. " - "Is the MGS running?\n", rc); - } else { - CERROR("Cannot talk to the MGS: %d, not fatal\n", rc); - /* reset the error code for non-fatal error. */ - rc = 0; - } - GOTO(out, rc); - } - - /* Always update our flags */ - ldd->ldd_flags = mti->mti_flags & LDD_F_ONDISK_MASK; - - /* If this flag is set, it means the MGS wants us to change our - on-disk data. (So far this means just the index.) */ - if (mti->mti_flags & LDD_F_REWRITE_LDD) { - char *label; - int err; - CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x " - "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index, - mti->mti_svname); - ldd->ldd_svindex = mti->mti_stripe_index; - strncpy(ldd->ldd_svname, mti->mti_svname, - 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) - CERROR("Label set error %d\n", err); - label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb); - if (label) - CDEBUG(D_MOUNT, "Disk label changed to %s\n", label); - - /* Flush the new ldd to disk */ - fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb); - } - -out: - if (mti) - OBD_FREE_PTR(mti); - RETURN(rc); -} - -/** - * Notify the MGS that this target is ready. - * Used by IR - if the MGS receives this message, it will notify clients. - */ -static int server_notify_target(struct super_block *sb, struct obd_device *obd) -{ - struct lustre_sb_info *lsi = s2lsi(sb); - struct obd_device *mgc = lsi->lsi_mgc; - struct mgs_target_info *mti = NULL; - int rc; - ENTRY; - - LASSERT(mgc); - - if (!(lsi->lsi_flags & LSI_SERVER)) - RETURN(-EINVAL); - - OBD_ALLOC_PTR(mti); - if (!mti) - RETURN(-ENOMEM); - rc = server_sb2mti(sb, mti); - if (rc) - GOTO(out, rc); - - mti->mti_instance = obd->u.obt.obt_instance; - mti->mti_flags |= LDD_F_OPC_READY; - - /* FIXME use mgc_process_config instead */ - rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp, - sizeof(KEY_REGISTER_TARGET), - KEY_REGISTER_TARGET, - sizeof(*mti), mti, NULL); - - /* Imperative recovery: if the mgs informs us to use IR? */ - if (!rc && !(mti->mti_flags & LDD_F_ERROR) && - (mti->mti_flags & LDD_F_IR_CAPABLE)) - lsi->lsi_flags |= LSI_IR_CAPABLE; - -out: - if (mti) - OBD_FREE_PTR(mti); - RETURN(rc); - -} - -/** Start server targets: MDTs and OSTs - */ -static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) -{ - struct obd_device *obd; - struct lustre_sb_info *lsi = s2lsi(sb); - struct config_llog_instance cfg; - int rc; - ENTRY; - - 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 */ - cfs_mutex_lock(&server_start_lock); - obd = class_name2obd(LUSTRE_MDS_OBDNAME); - if (!obd) { - rc = lustre_start_simple(LUSTRE_MDS_OBDNAME, - /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */ - LUSTRE_MDT_NAME, - LUSTRE_MDS_OBDNAME"_uuid", - 0, 0); - if (rc) { - cfs_mutex_unlock(&server_start_lock); - CERROR("failed to start MDS: %d\n", rc); - RETURN(rc); - } - } - 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_lock(&server_start_lock); - obd = class_name2obd(LUSTRE_OSS_OBDNAME); - if (!obd) { - rc = lustre_start_simple(LUSTRE_OSS_OBDNAME, - LUSTRE_OSS_NAME, - LUSTRE_OSS_OBDNAME"_uuid", - 0, 0); - if (rc) { - cfs_mutex_unlock(&server_start_lock); - CERROR("failed to start OSS: %d\n", rc); - RETURN(rc); - } - } - 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); - - /* Register with MGS */ - rc = server_register_target(sb); - if (rc) - GOTO(out_mgc, rc); - - /* Let the target look up the mount using the target's name - (we can't pass the sb or mnt through class_process_config.) */ - rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt); - if (rc) - GOTO(out_mgc, rc); - - /* Start targets using the llog named for the target */ - memset(&cfg, 0, sizeof(cfg)); - rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg); - 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); - } - -out_mgc: - /* Release the mgc fs for others to use */ - server_mgc_clear_fs(lsi->lsi_mgc); - - if (!rc) { - obd = class_name2obd(lsi->lsi_ldd->ldd_svname); - if (!obd) { - CERROR("no server named %s was started\n", - lsi->lsi_ldd->ldd_svname); - RETURN(-ENXIO); - } - - if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) && - (OBP(obd, iocontrol))) { - obd_iocontrol(OBD_IOC_ABORT_RECOVERY, - obd->obd_self_export, 0, NULL, NULL); - } - - server_notify_target(sb, obd); - - /* calculate recovery timeout, do it after lustre_process_log */ - server_calc_timeout(lsi, obd); - - /* log has been fully processed */ - obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG); - } - + mutex_unlock(&mgc_start_lock); RETURN(rc); } @@ -1323,12 +584,14 @@ struct lustre_sb_info *lustre_init_lsi(struct super_block *sb) lsi->lsi_lmd->lmd_recovery_time_hard = 0; s2lsi_nocast(sb) = lsi; /* we take 1 extra ref for our setup */ - cfs_atomic_set(&lsi->lsi_mounts, 1); + atomic_set(&lsi->lsi_mounts, 1); /* Default umount style */ lsi->lsi_flags = LSI_UMOUNT_FAILOVER; + INIT_LIST_HEAD(&lsi->lsi_lwp_list); + spin_lock_init(&lsi->lsi_lwp_lock); - RETURN(lsi); + RETURN(lsi); } static int lustre_free_lsi(struct super_block *sb) @@ -1340,10 +603,7 @@ static int lustre_free_lsi(struct super_block *sb) CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi); /* someone didn't call server_put_mount. */ - LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0); - - if (lsi->lsi_ldd != NULL) - OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd)); + LASSERT(atomic_read(&lsi->lsi_mounts) == 0); if (lsi->lsi_lmd != NULL) { if (lsi->lsi_lmd->lmd_dev != NULL) @@ -1362,6 +622,15 @@ 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); + if (lsi->lsi_lmd->lmd_params != NULL) + OBD_FREE(lsi->lsi_lmd->lmd_params, 4096); + OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd)); } @@ -1374,601 +643,173 @@ static int lustre_free_lsi(struct super_block *sb) /* The lsi has one reference for every server that is using the disk - e.g. MDT, MGS, and potentially MGC */ -static int lustre_put_lsi(struct super_block *sb) +int lustre_put_lsi(struct super_block *sb) { struct lustre_sb_info *lsi = s2lsi(sb); ENTRY; LASSERT(lsi != NULL); - CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts)); - if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) { + CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts)); + if (atomic_dec_and_test(&lsi->lsi_mounts)) { + if (IS_SERVER(lsi) && lsi->lsi_osd_exp) { + lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev); + lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL; + lsi->lsi_dt_dev = NULL; + obd_disconnect(lsi->lsi_osd_exp); + /* wait till OSD is gone */ + obd_zombie_barrier(); + } lustre_free_lsi(sb); RETURN(1); } RETURN(0); } -/*************** server mount ******************/ - -/** Kernel mount using mount options in MOUNT_DATA_FILE. - * Since this file lives on the disk, we pre-mount using a common - * type, read the file, then re-mount using the type specified in the - * file. - */ -static struct vfsmount *server_kernel_mount(struct super_block *sb) -{ - struct lvfs_run_ctxt mount_ctxt; - struct lustre_sb_info *lsi = s2lsi(sb); - struct lustre_disk_data *ldd; - struct lustre_mount_data *lmd = lsi->lsi_lmd; - struct vfsmount *mnt; - struct file_system_type *type; - char *options = NULL; - unsigned long page, s_flags; - struct page *__page; - int len; - int rc; - ENTRY; - - OBD_ALLOC(ldd, sizeof(*ldd)); - if (!ldd) - RETURN(ERR_PTR(-ENOMEM)); - - /* In the past, we have always used flags = 0. - 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); - 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 " - "Is the ldiskfs module available?\n", - lmd->lmd_dev, s_flags, rc ); - GOTO(out_free, rc); - } - - OBD_SET_CTXT_MAGIC(&mount_ctxt); - mount_ctxt.pwdmnt = mnt; - mount_ctxt.pwd = mnt->mnt_root; - mount_ctxt.fs = get_ds(); - - rc = ldd_parse(&mount_ctxt, ldd); - unlock_mntput(mnt); - - if (rc) { - CERROR("premount parse options failed: rc = %d\n", rc); - GOTO(out_free, rc); - } - - /* Done with our pre-mount, now do the real mount. */ - - /* Glom up mount options */ - memset(options, 0, CFS_PAGE_SIZE); - strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2); - - len = CFS_PAGE_SIZE - strlen(options) - 2; - if (*options != 0) - strcat(options, ","); - strncat(options, "no_mbcache", len); - - /* Add in any mount-line options */ - if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) { - len = CFS_PAGE_SIZE - strlen(options) - 2; - strcat(options, ","); - strncat(options, lmd->lmd_opts, len); - } - - /* Special permanent mount flags */ - if (IS_OST(ldd)) - s_flags |= MS_NOATIME | MS_NODIRATIME; - - CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n", - MT_STR(ldd), lmd->lmd_dev, 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("vfs_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. +/*** SERVER NAME *** + * + * FSNAME is between 1 and 8 characters (inclusive). + * Excluded characters are '/' and ':' + * SEPERATOR is either ':' or '-' + * TYPE: "OST", "MDT", etc. + * INDEX: Hex representation of the index */ -static void server_wait_finished(struct vfsmount *mnt) -{ - cfs_waitq_t waitq; - int rc, waited = 0; - cfs_sigset_t blocked; - - cfs_waitq_init(&waitq); - - while (mnt_get_count(mnt) > 1) { - if (waited && (waited % 30 == 0)) - LCONSOLE_WARN("Mount still busy with %d refs after " - "%d secs.\n", - 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, - (mnt_get_count(mnt) == 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, - mnt_get_count(mnt)); - break; - } - - } -} -/** Start the shutdown of servers at umount. +/** Get the fsname ("lustre") from the server name ("lustre-OST003F"). + * @param [in] svname server name including type and index + * @param [out] fsname Buffer to copy filesystem name prefix into. + * Must have at least 'strlen(fsname) + 1' chars. + * @param [out] endptr if endptr isn't NULL it is set to end of fsname + * rc < 0 on error */ -static void server_put_super(struct super_block *sb) +int server_name2fsname(const char *svname, char *fsname, const char **endptr) { - struct lustre_sb_info *lsi = s2lsi(sb); - struct obd_device *obd; - struct vfsmount *mnt = lsi->lsi_srv_mnt; - char *tmpname, *extraname = NULL; - int tmpname_sz; - int lddflags = lsi->lsi_ldd->ldd_flags; - int lsiflags = lsi->lsi_flags; - ENTRY; + const char *dash; - LASSERT(lsiflags & LSI_SERVER); - - tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1; - 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 (!(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 */ - lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL); - - /* COMPAT_146 - profile may get deleted in mgc_cleanup. - 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_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_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! */ - 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); - } - } + dash = svname + strnlen(svname, 8); /* max fsname length is 8 */ + for (; dash > svname && *dash != '-' && *dash != ':'; dash--) + ; + if (dash == svname) + return -EINVAL; - /* 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)) { - /* if MDS start with --nomgs, don't stop MGS then */ - if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) { - char *logname; - - OBD_ALLOC(logname, MGS_PARAM_MAXLEN); - if (!logname) { - LCONSOLE_WARN("Stopping mgs failed %d, please " - "try again.", -ENOMEM); - return; - } - strcpy(logname, lsi->lsi_ldd->ldd_fsname); - strcat(logname, "-params"); - /* tell the mgc to drop parameter config log */ - lustre_end_log(sb, logname, NULL); - OBD_FREE(logname, MGS_PARAM_MAXLEN); + if (fsname != NULL) { + strncpy(fsname, svname, dash - svname); + fsname[dash - svname] = '\0'; + } - server_stop_mgs(sb); - } - } - - /* Clean the mgc and sb */ - 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 */ - server_wait_finished(mnt); - - /* drop the One True Mount */ - 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); - - /* 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; - class_manual_cleanup(obd); - } - OBD_FREE(extraname, strlen(extraname) + 1); - } + if (endptr != NULL) + *endptr = dash; - LCONSOLE_WARN("server umount %s complete\n", tmpname); - OBD_FREE(tmpname, tmpname_sz); - EXIT; + return 0; } +EXPORT_SYMBOL(server_name2fsname); -/** Called only for 'umount -f' +/** + * Get service name (svname) from string + * rc < 0 on error + * if endptr isn't NULL it is set to end of fsname * */ -#ifdef HAVE_UMOUNTBEGIN_VFSMOUNT -static void server_umount_begin(struct vfsmount *vfsmnt, int flags) -{ - struct super_block *sb = vfsmnt->mnt_sb; -#else -static void server_umount_begin(struct super_block *sb) +int server_name2svname(const char *label, char *svname, const char **endptr, + size_t svsize) { -#endif - struct lustre_sb_info *lsi = s2lsi(sb); - ENTRY; - -#ifdef HAVE_UMOUNTBEGIN_VFSMOUNT - if (!(flags & MNT_FORCE)) { - EXIT; - return; - } -#endif + int rc; + const char *dash; - CDEBUG(D_MOUNT, "umount -f\n"); - /* umount = failover - umount -f = force - no third way to do non-force, non-failover */ - lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER; - lsi->lsi_flags |= LSI_UMOUNT_FORCE; - EXIT; -} + /* We use server_name2fsname() just for parsing */ + rc = server_name2fsname(label, NULL, &dash); + if (rc != 0) + return rc; -#ifndef HAVE_STATFS_DENTRY_PARAM -static int server_statfs (struct super_block *sb, cfs_kstatfs_t *buf) -{ -#else -static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf) -{ - struct super_block *sb = dentry->d_sb; -#endif - struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt; - ENTRY; + if (endptr != NULL) + *endptr = dash; - if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) { -#ifdef HAVE_STATFS_DENTRY_PARAM - int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf); -#else - int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf); -#endif - if (!rc) { - buf->f_type = sb->s_magic; - RETURN(0); - } - } + if (strlcpy(svname, dash + 1, svsize) >= svsize) + return -E2BIG; - /* just return 0 */ - buf->f_type = sb->s_magic; - buf->f_bsize = sb->s_blocksize; - buf->f_blocks = 1; - buf->f_bfree = 0; - buf->f_bavail = 0; - buf->f_files = 1; - buf->f_ffree = 0; - buf->f_namelen = NAME_MAX; - RETURN(0); + return 0; } +EXPORT_SYMBOL(server_name2svname); -/** The operations we support directly on the superblock: - * mount, umount, and df. - */ -static struct super_operations server_ops = -{ - .put_super = server_put_super, - .umount_begin = server_umount_begin, /* umount -f */ - .statfs = server_statfs, -}; - -#define log2(n) cfs_ffz(~(n)) -#define LUSTRE_SUPER_MAGIC 0x0BD00BD1 - -static int server_fill_super_common(struct super_block *sb) +/** + * check server name is OST. + **/ +int server_name_is_ost(const char *svname) { - struct inode *root = 0; - ENTRY; - - CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev); + const char *dash; + int rc; - sb->s_blocksize = 4096; - sb->s_blocksize_bits = log2(sb->s_blocksize); - sb->s_magic = LUSTRE_SUPER_MAGIC; - sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */ - sb->s_flags |= MS_RDONLY; - sb->s_op = &server_ops; + /* We use server_name2fsname() just for parsing */ + rc = server_name2fsname(svname, NULL, &dash); + if (rc != 0) + return rc; - root = new_inode(sb); - if (!root) { - CERROR("Can't make root inode\n"); - RETURN(-EIO); - } - - /* returns -EIO for every operation */ - /* make_bad_inode(root); -- badness - can't umount */ - /* apparently we need to be a directory for the mount to finish */ - root->i_mode = S_IFDIR; - - sb->s_root = d_alloc_root(root); - if (!sb->s_root) { - CERROR("Can't make root dentry\n"); - iput(root); - RETURN(-EIO); - } + dash++; - RETURN(0); + if (strncmp(dash, "OST", 3) == 0) + return 1; + return 0; } +EXPORT_SYMBOL(server_name_is_ost); -/** Fill in the superblock info for a Lustre server. - * Mount the device with the correct options. - * Read the on-disk config file. - * Start the services. - */ -static int server_fill_super(struct super_block *sb) +/** + * Get the index from the target name MDTXXXX/OSTXXXX + * rc = server type, or rc < 0 on error + **/ +int target_name2index(const char *tgtname, __u32 *idx, const char **endptr) { - struct lustre_sb_info *lsi = s2lsi(sb); - struct vfsmount *mnt; - int rc; - ENTRY; - - /* the One True Mount */ - mnt = server_kernel_mount(sb); - if (IS_ERR(mnt)) { - rc = PTR_ERR(mnt); - CERROR("Unable to mount device %s: %d\n", - lsi->lsi_lmd->lmd_dev, rc); - lustre_put_lsi(sb); - RETURN(rc); - } - lsi->lsi_srv_mnt = mnt; - - LASSERT(lsi->lsi_ldd); - CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n", - lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname, - lsi->lsi_lmd->lmd_dev); - - if (class_name2obd(lsi->lsi_ldd->ldd_svname)) { - 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); - unlock_mntput(mnt); - RETURN(-EALREADY); - } - - /* 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); - } - - /* Start MGC before servers */ - rc = lustre_start_mgc(sb); - if (rc) - GOTO(out_mnt, rc); - - /* Set up all obd devices for service */ - if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) && - (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) { - rc = server_start_targets(sb, mnt); - if (rc < 0) { - CERROR("Unable to start targets: %d\n", rc); - GOTO(out_mnt, rc); - } - /* FIXME overmount client here, - or can we just start a client log and client_fill_super on this sb? - We need to make sure server_put_super gets called too - ll_put_super - calls lustre_common_put_super; check there for LSI_SERVER flag, - call s_p_s if so. - Probably should start client from new thread so we can return. - Client will not finish until all servers are connected. - Note - MGS-only server does NOT get a client, since there is no - lustre fs associated - the MGS is for all lustre fs's */ - } else if (IS_MGS(lsi->lsi_ldd) && - !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){ - struct config_llog_instance cfg; - char *logname; - - OBD_ALLOC(logname, MGS_PARAM_MAXLEN); - if (logname == NULL) - GOTO(out_mnt, rc = -ENOMEM); - strcpy(logname, lsi->lsi_ldd->ldd_fsname); - strcat(logname, "-params"); - - memset(&cfg, 0, sizeof(cfg)); - rc = lustre_process_log(sb, logname, &cfg); - OBD_FREE(logname, MGS_PARAM_MAXLEN); - if (rc) { - CERROR("failed to process parameters %s: %d\n", - logname, rc); - GOTO(out_mnt, rc); - } - } - - rc = server_fill_super_common(sb); - if (rc) - GOTO(out_mnt, rc); - - 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); - return rc; + const char *dash = tgtname; + unsigned long index; + int rc; + + if (strncmp(dash, "MDT", 3) == 0) + rc = LDD_F_SV_TYPE_MDT; + else if (strncmp(dash, "OST", 3) == 0) + rc = LDD_F_SV_TYPE_OST; + else + return -EINVAL; + + dash += 3; + + if (strncmp(dash, "all", 3) == 0) { + if (endptr != NULL) + *endptr = dash + 3; + return rc | LDD_F_SV_ALL; + } + + index = simple_strtoul(dash, (char **)endptr, 16); + if (idx != NULL) + *idx = index; + return rc; } +EXPORT_SYMBOL(target_name2index); /* Get the index from the obd name. rc = server type, or rc < 0 on error if endptr isn't NULL it is set to end of name */ -int server_name2index(char *svname, __u32 *idx, char **endptr) +int server_name2index(const char *svname, __u32 *idx, const char **endptr) { - unsigned long index; - int rc; - 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); - } + const char *dash; + int rc; - 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; + /* We use server_name2fsname() just for parsing */ + rc = server_name2fsname(svname, NULL, &dash); + if (rc != 0) + return rc; - index = simple_strtoul(dash + 4, endptr, 16); - *idx = index; - return rc; -} + dash++; + rc = target_name2index(dash, idx, endptr); + if (rc < 0) + return rc; -/* - * Calculate timeout value for a target. - */ -void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd) -{ - struct lustre_mount_data *lmd; - int soft = 0; - int hard = 0; - int factor = 0; - bool has_ir = !!(lsi->lsi_flags & LSI_IR_CAPABLE); - int min = OBD_RECOVERY_TIME_MIN; + /* Account for -mdc after index that is possible when specifying mdt */ + if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1, + sizeof(LUSTRE_MDC_NAME)-1) == 0) + *endptr += sizeof(LUSTRE_MDC_NAME); - LASSERT(lsi->lsi_flags & LSI_SERVER); - - lmd = lsi->lsi_lmd; - if (lmd) { - soft = lmd->lmd_recovery_time_soft; - hard = lmd->lmd_recovery_time_hard; - has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR); - obd->obd_no_ir = !has_ir; - } - - if (soft == 0) - soft = OBD_RECOVERY_TIME_SOFT; - if (hard == 0) - hard = OBD_RECOVERY_TIME_HARD; - - /* target may have ir_factor configured. */ - factor = OBD_IR_FACTOR_DEFAULT; - if (obd->obd_recovery_ir_factor) - factor = obd->obd_recovery_ir_factor; - - if (has_ir) { - int new_soft = soft; - int new_hard = hard; - - /* adjust timeout value by imperative recovery */ - - new_soft = (soft * factor) / OBD_IR_FACTOR_MAX; - new_hard = (hard * factor) / OBD_IR_FACTOR_MAX; - - /* make sure the timeout is not too short */ - new_soft = max(min, new_soft); - new_hard = max(new_soft, new_hard); - - LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery " - "window shrunk from %d-%d down to %d-%d\n", - obd->obd_name, soft, hard, new_soft, new_hard); - - soft = new_soft; - hard = new_hard; - } - - /* we're done */ - obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft); - obd->obd_recovery_time_hard = hard; - obd->obd_recovery_ir_factor = factor; + return rc; } -EXPORT_SYMBOL(server_calc_timeout); +EXPORT_SYMBOL(server_name2index); /*************** mount common betweeen server and client ***************/ @@ -1996,32 +837,33 @@ int lustre_common_put_super(struct super_block *sb) lu_types_stop(); RETURN(rc); } +EXPORT_SYMBOL(lustre_common_put_super); static void lmd_print(struct lustre_mount_data *lmd) { - int i; + int i; - PRINT_CMD(PRINT_MASK, " mount data:\n"); - if (lmd_is_client(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); + PRINT_CMD(D_MOUNT, " mount data:\n"); + if (lmd_is_client(lmd)) + PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile); + PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev); + PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags); - if (lmd->lmd_opts) - PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts); + if (lmd->lmd_opts) + PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts); - if (lmd->lmd_recovery_time_soft) - PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n", - lmd->lmd_recovery_time_soft); + if (lmd->lmd_recovery_time_soft) + PRINT_CMD(D_MOUNT, "recovery time soft: %d\n", + lmd->lmd_recovery_time_soft); - if (lmd->lmd_recovery_time_hard) - PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n", - lmd->lmd_recovery_time_hard); + if (lmd->lmd_recovery_time_hard) + PRINT_CMD(D_MOUNT, "recovery time hard: %d\n", + lmd->lmd_recovery_time_hard); - for (i = 0; i < lmd->lmd_exclude_count; i++) { - PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i, - lmd->lmd_exclude[i]); - } + for (i = 0; i < lmd->lmd_exclude_count; i++) { + PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i, + lmd->lmd_exclude[i]); + } } /* Is this server on the exclusion list */ @@ -2051,58 +893,60 @@ int lustre_check_exclusion(struct super_block *sb, char *svname) } /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */ -static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr) +static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr) { - char *s1 = ptr, *s2; - __u32 index, *exclude_list; - int rc = 0, devmax; - ENTRY; - - /* The shortest an ost name can be is 8 chars: -OST0000. - We don't actually know the fsname at this time, so in fact - a user could specify any fsname. */ - devmax = strlen(ptr) / 8 + 1; - - /* temp storage until we figure out how many we have */ - OBD_ALLOC(exclude_list, sizeof(index) * devmax); - if (!exclude_list) - RETURN(-ENOMEM); - - /* we enter this fn pointing at the '=' */ - while (*s1 && *s1 != ' ' && *s1 != ',') { - s1++; - rc = server_name2index(s1, &index, &s2); - if (rc < 0) { - CERROR("Can't parse server name '%s'\n", s1); - break; - } - if (rc == LDD_F_SV_TYPE_OST) - exclude_list[lmd->lmd_exclude_count++] = index; - else - CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1); - s1 = s2; - /* now we are pointing at ':' (next exclude) - or ',' (end of excludes) */ - if (lmd->lmd_exclude_count >= devmax) - break; - } - if (rc >= 0) /* non-err */ - rc = 0; - - if (lmd->lmd_exclude_count) { - /* permanent, freed in lustre_free_lsi */ - OBD_ALLOC(lmd->lmd_exclude, sizeof(index) * - lmd->lmd_exclude_count); - if (lmd->lmd_exclude) { - memcpy(lmd->lmd_exclude, exclude_list, - sizeof(index) * lmd->lmd_exclude_count); - } else { - rc = -ENOMEM; - lmd->lmd_exclude_count = 0; - } - } - OBD_FREE(exclude_list, sizeof(index) * devmax); - RETURN(rc); + const char *s1 = ptr, *s2; + __u32 index, *exclude_list; + int rc = 0, devmax; + ENTRY; + + /* The shortest an ost name can be is 8 chars: -OST0000. + We don't actually know the fsname at this time, so in fact + a user could specify any fsname. */ + devmax = strlen(ptr) / 8 + 1; + + /* temp storage until we figure out how many we have */ + OBD_ALLOC(exclude_list, sizeof(index) * devmax); + if (!exclude_list) + RETURN(-ENOMEM); + + /* we enter this fn pointing at the '=' */ + while (*s1 && *s1 != ' ' && *s1 != ',') { + s1++; + rc = server_name2index(s1, &index, &s2); + if (rc < 0) { + CERROR("Can't parse server name '%s': rc = %d\n", + s1, rc); + break; + } + if (rc == LDD_F_SV_TYPE_OST) + exclude_list[lmd->lmd_exclude_count++] = index; + else + CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n", + (uint)(s2-s1), s1, rc); + s1 = s2; + /* now we are pointing at ':' (next exclude) + or ',' (end of excludes) */ + if (lmd->lmd_exclude_count >= devmax) + break; + } + if (rc >= 0) /* non-err */ + rc = 0; + + if (lmd->lmd_exclude_count) { + /* permanent, freed in lustre_free_lsi */ + OBD_ALLOC(lmd->lmd_exclude, sizeof(index) * + lmd->lmd_exclude_count); + if (lmd->lmd_exclude) { + memcpy(lmd->lmd_exclude, exclude_list, + sizeof(index) * lmd->lmd_exclude_count); + } else { + rc = -ENOMEM; + lmd->lmd_exclude_count = 0; + } + } + OBD_FREE(exclude_list, sizeof(index) * devmax); + RETURN(rc); } static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr) @@ -2130,13 +974,80 @@ 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_quiet(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 */ static int lmd_parse(char *options, struct lustre_mount_data *lmd) { - char *s1, *s2, *devname = NULL; + char *s1, *s2, *s3, *devname = NULL; struct lustre_mount_data *raw = (struct lustre_mount_data *)options; int rc = 0; ENTRY; @@ -2157,6 +1068,11 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) } lmd->lmd_magic = LMD_MAGIC; + OBD_ALLOC(lmd->lmd_params, 4096); + if (lmd->lmd_params == NULL) + RETURN(-ENOMEM); + lmd->lmd_params[0] = '\0'; + /* Set default flags here */ s1 = options; @@ -2167,6 +1083,7 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) /* Skip whitespace and extra commas */ while (*s1 == ' ' || *s1 == ',') s1++; + s3 = s1; /* Client options are parsed in ll_options: eg. flock, user_xattr, acl */ @@ -2193,9 +1110,31 @@ 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; + s3 = s2; + clear++; } else if (strncmp(s1, "writeconf", 9) == 0) { lmd->lmd_flags |= LMD_FLG_WRITECONF; clear++; + } else if (strncmp(s1, "update", 6) == 0) { + lmd->lmd_flags |= LMD_FLG_UPDATE; + clear++; + } else if (strncmp(s1, "virgin", 6) == 0) { + lmd->lmd_flags |= LMD_FLG_VIRGIN; + clear++; + } else if (strncmp(s1, "noprimnode", 10) == 0) { + lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE; + clear++; } else if (strncmp(s1, "mgssec=", 7) == 0) { rc = lmd_parse_mgssec(lmd, s1 + 7); if (rc) @@ -2207,6 +1146,41 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) if (rc) goto invalid; clear++; + } else if (strncmp(s1, "mgs", 3) == 0) { + /* We are an MGS */ + lmd->lmd_flags |= LMD_FLG_MGS; + 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, "param=", 6) == 0) { + int length; + char *tail = strchr(s1 + 6, ','); + if (tail == NULL) { + length = strlen(s1); + } else { + lnet_nid_t nid; + char *param_str = tail + 1; + int supplementary = 1; + + while (class_parse_nid_quiet(param_str, &nid, + ¶m_str) == 0) { + supplementary = 0; + } + length = param_str - s1 - supplementary; + } + length -= 6; + strncat(lmd->lmd_params, s1 + 6, length); + strcat(lmd->lmd_params, " "); + s3 = s1 + 6 + length; + clear++; + } else if (strncmp(s1, "osd=", 4) == 0) { + rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4); + if (rc) + goto invalid; + clear++; } /* Linux 2.4 doesn't pass the device, so we stuck it at the end of the options. */ @@ -2218,8 +1192,8 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) break; } - /* Find next opt */ - s2 = strchr(s1, ','); + /* Find next opt */ + s2 = strchr(s3, ','); if (s2 == NULL) { if (clear) *s1 = '\0'; @@ -2304,11 +1278,11 @@ 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'. - */ - cfs_lockdep_off(); + /* + * Disable lockdep during mount, because mount locking patterns are + * `special'. + */ + lockdep_off(); /* * LU-639: the obd cleanup of last mount may not finish yet, wait here. @@ -2341,13 +1315,18 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent) /* c_f_s will call lustre_common_put_super on failure */ } } else { - CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev); - lsi->lsi_flags |= LSI_SERVER; - rc = server_fill_super(sb); - /* s_f_s calls lustre_start_mgc after the mount because we need - the MGS nids which are stored on disk. Plus, we may - need to start the MGS first. */ - /* s_f_s will call server_put_super on failure */ +#ifdef HAVE_SERVER_SUPPORT + CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev); + rc = server_fill_super(sb); + /* s_f_s calls lustre_start_mgc after the mount because we need + the MGS nids which are stored on disk. Plus, we may + need to start the MGS first. */ + /* s_f_s will call server_put_super on failure */ +#else + CERROR("This is client-side-only module, " + "cannot handle server mount.\n"); + rc = -EINVAL; +#endif } /* If error happens in fill_super() call, @lsi will be killed there. @@ -2361,8 +1340,8 @@ out: CDEBUG(D_SUPER, "Mount %s complete\n", lmd->lmd_dev); } - cfs_lockdep_on(); - return rc; + lockdep_on(); + return rc; } @@ -2373,27 +1352,30 @@ void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb, { client_fill_super = cfs; } +EXPORT_SYMBOL(lustre_register_client_fill_super); void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb)) { kill_super_cb = cfs; } +EXPORT_SYMBOL(lustre_register_kill_super_cb); /***************** FS registration ******************/ - -#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18)) -struct super_block * lustre_get_sb(struct file_system_type *fs_type, int flags, - const char *devname, void * data) +#ifdef HAVE_FSTYPE_MOUNT +struct dentry *lustre_mount(struct file_system_type *fs_type, int flags, + const char *devname, void *data) { - return get_sb_nodev(fs_type, flags, data, lustre_fill_super); + struct lustre_mount_data2 lmd2 = { data, NULL }; + + return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super); } #else int lustre_get_sb(struct file_system_type *fs_type, int flags, const char *devname, void * data, struct vfsmount *mnt) { - struct lustre_mount_data2 lmd2 = {data, mnt}; + struct lustre_mount_data2 lmd2 = { data, mnt }; - return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt); + return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt); } #endif @@ -2401,7 +1383,7 @@ void lustre_kill_super(struct super_block *sb) { struct lustre_sb_info *lsi = s2lsi(sb); - if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER)) + if (kill_super_cb && lsi && !IS_SERVER(lsi)) (*kill_super_cb)(sb); kill_anon_super(sb); @@ -2412,13 +1394,14 @@ void lustre_kill_super(struct super_block *sb) struct file_system_type lustre_fs_type = { .owner = THIS_MODULE, .name = "lustre", +#ifdef HAVE_FSTYPE_MOUNT + .mount = lustre_mount, +#else .get_sb = lustre_get_sb, - .kill_sb = lustre_kill_super, - .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV | -#ifdef FS_HAS_FIEMAP - FS_HAS_FIEMAP | #endif - LL_RENAME_DOES_D_MOVE, + .kill_sb = lustre_kill_super, + .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV | + FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE, }; int lustre_register_fs(void) @@ -2430,17 +1413,3 @@ int lustre_unregister_fs(void) { return unregister_filesystem(&lustre_fs_type); } - -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);