X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdclass%2Fobd_mount.c;h=2ca91e9a6b5d5b7f1ecf6a1168be6f28b850db0a;hp=5158bcb642692c1a4a6abbd06624161e82e1290a;hb=e3c85d27cb4b3fc59181ed52b5090e812a1ab0ae;hpb=2c74bfcb7a06addb42e79c1f561afb0acfaaa7d0 diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c index 5158bcb..2ca91e9 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. @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -55,23 +55,28 @@ #include #include #include +#ifdef HAVE_KERNEL_LOCKED +#include +#endif -static int (*client_fill_super)(struct super_block *sb) = NULL; +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 *********/ -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) { - struct list_head *tmp; + cfs_list_t *tmp; struct lustre_mount_info *lmi; ENTRY; - list_for_each(tmp, &server_mount_info_list) { - lmi = list_entry(tmp, struct lustre_mount_info, lmi_list_chain); + 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); } @@ -88,7 +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)); @@ -101,10 +105,10 @@ static int server_register_mount(const char *name, struct super_block *sb, } strcpy(name_cp, name); - down(&lustre_mount_info_lock); + cfs_mutex_lock(&lustre_mount_info_lock); if (server_find_mount(name)) { - 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); @@ -113,12 +117,11 @@ static int server_register_mount(const char *name, struct super_block *sb, lmi->lmi_name = name_cp; lmi->lmi_sb = sb; lmi->lmi_mnt = mnt; - list_add(&lmi->lmi_list_chain, &server_mount_info_list); + cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list); - 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, atomic_read(&lmi->lmi_mnt->mnt_count)); + CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name); RETURN(0); } @@ -129,21 +132,20 @@ static int server_deregister_mount(const char *name) struct lustre_mount_info *lmi; ENTRY; - down(&lustre_mount_info_lock); + cfs_mutex_lock(&lustre_mount_info_lock); lmi = server_find_mount(name); if (!lmi) { - 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, atomic_read(&lmi->lmi_mnt->mnt_count)); + CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name); OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1); - list_del(&lmi->lmi_list_chain); + cfs_list_del(&lmi->lmi_list_chain); OBD_FREE(lmi, sizeof(*lmi)); - up(&lustre_mount_info_lock); + cfs_mutex_unlock(&lustre_mount_info_lock); RETURN(0); } @@ -157,23 +159,23 @@ struct lustre_mount_info *server_get_mount(const char *name) struct lustre_sb_info *lsi; ENTRY; - down(&lustre_mount_info_lock); + cfs_mutex_lock(&lustre_mount_info_lock); lmi = server_find_mount(name); - 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); - atomic_inc(&lsi->lsi_mounts); - CDEBUG(D_MOUNT, "get_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)); + cfs_atomic_inc(&lsi->lsi_mounts); + + CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt, + name, cfs_atomic_read(&lsi->lsi_mounts)); RETURN(lmi); } +EXPORT_SYMBOL(server_get_mount); /* * Used by mdt to get mount_info from obdname. @@ -185,25 +187,15 @@ struct lustre_mount_info *server_get_mount_2(const char *name) struct lustre_mount_info *lmi; ENTRY; - down(&lustre_mount_info_lock); + cfs_mutex_lock(&lustre_mount_info_lock); lmi = server_find_mount(name); - up(&lustre_mount_info_lock); + 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()) { - unlock_kernel(); - mntput(mnt); - lock_kernel(); - } else { - mntput(mnt); - } -} +EXPORT_SYMBOL(server_get_mount_2); static int lustre_put_lsi(struct super_block *sb); @@ -212,38 +204,30 @@ 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); + cfs_mutex_lock(&lustre_mount_info_lock); lmi = server_find_mount(name); - 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); } 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, atomic_read(&lsi->lsi_mounts), count); + CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n", + lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts)); - 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); - } + if (lustre_put_lsi(lmi->lmi_sb)) + CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n", + lmi->lmi_mnt, name); /* this obd should never need the mount again */ server_deregister_mount(name); RETURN(0); } +EXPORT_SYMBOL(server_put_mount); /* Corresponding to server_get_mount_2 */ int server_put_mount_2(const char *name, struct vfsmount *mnt) @@ -251,141 +235,27 @@ 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 "LPSZ"\n", - len, 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, 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); -} - +EXPORT_SYMBOL(server_put_mount_2); /**************** config llog ********************/ -/* Get a config log from the MGS and process it. - This func is called for both clients and servers. - Continue to process new statements appended to the logs - (whenever the config lock is revoked) until lustre_end_log - is called. */ +/** Get a config log from the MGS and process it. + * This func is called for both clients and servers. + * Continue to process new statements appended to the logs + * (whenever the config lock is revoked) until lustre_end_log + * is called. + * @param sb The superblock is used by the MGC to write to the local copy of + * the config log + * @param logname The name of the llog to replicate from the MGS + * @param cfg Since the same mgc may be used to follow multiple config logs + * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for + * this log, and is added to the mgc's list of logs to follow. + */ int lustre_process_log(struct super_block *sb, char *logname, struct config_llog_instance *cfg) { struct lustre_cfg *lcfg; - struct lustre_cfg_bufs bufs; + struct lustre_cfg_bufs *bufs; struct lustre_sb_info *lsi = s2lsi(sb); struct obd_device *mgc = lsi->lsi_mgc; int rc; @@ -394,15 +264,21 @@ int lustre_process_log(struct super_block *sb, char *logname, LASSERT(mgc); LASSERT(cfg); + OBD_ALLOC_PTR(bufs); + if (bufs == NULL) + RETURN(-ENOMEM); + /* mgc_process_config */ - lustre_cfg_bufs_reset(&bufs, mgc->obd_name); - lustre_cfg_bufs_set_string(&bufs, 1, logname); - lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg)); - lustre_cfg_bufs_set(&bufs, 3, &sb, sizeof(sb)); - lcfg = lustre_cfg_new(LCFG_LOG_START, &bufs); + lustre_cfg_bufs_reset(bufs, mgc->obd_name); + lustre_cfg_bufs_set_string(bufs, 1, logname); + lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg)); + lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb)); + lcfg = lustre_cfg_new(LCFG_LOG_START, bufs); rc = obd_process_config(mgc, sizeof(*lcfg), lcfg); lustre_cfg_free(lcfg); + OBD_FREE_PTR(bufs); + if (rc == -EINVAL) LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'" "failed from the MGS (%d). Make sure this " @@ -422,6 +298,7 @@ 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, @@ -447,9 +324,13 @@ int lustre_end_log(struct super_block *sb, char *logname, lustre_cfg_free(lcfg); RETURN(rc); } +EXPORT_SYMBOL(lustre_end_log); /**************** obd start *******************/ +/** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from + * lctl (and do for echo cli/srv. + */ int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd, char *s1, char *s2, char *s3, char *s4) { @@ -476,24 +357,28 @@ 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) + 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; + 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; } /* Set up a MGS to serve startup logs */ @@ -504,7 +389,6 @@ static int server_start_mgs(struct super_block *sb) 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 */ @@ -512,8 +396,7 @@ static int server_start_mgs(struct super_block *sb) 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); + " from server\n"); RETURN(-EALREADY); } @@ -523,7 +406,8 @@ static int server_start_mgs(struct super_block *sb) if (!rc) { rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME, - LUSTRE_MGS_OBDNAME, 0, 0); + LUSTRE_MGS_OBDNAME, 0, 0, + lsi->lsi_osd_obdname, 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 */ @@ -557,7 +441,7 @@ static int server_stop_mgs(struct super_block *sb) RETURN(rc); } -DECLARE_MUTEX(mgc_start_lock); +CFS_DEFINE_MUTEX(mgc_start_lock); /** Set up a mgc obd to process startup logs * @@ -574,7 +458,7 @@ static int lustre_start_mgc(struct super_block *sb) struct obd_uuid *uuid; class_uuid_t uuidc; lnet_nid_t nid; - char *mgcname, *niduuid, *mgssec; + char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL; char *ptr; int recov_bk; int rc = 0, i = 0, j, len; @@ -583,13 +467,13 @@ 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) @@ -610,6 +494,8 @@ static int lustre_start_mgc(struct super_block *sb) RETURN(-EINVAL); } + cfs_mutex_lock(&mgc_start_lock); + len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1; OBD_ALLOC(mgcname, len); OBD_ALLOC(niduuid, len + 2); @@ -619,23 +505,51 @@ static int lustre_start_mgc(struct super_block *sb) mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : ""; - mutex_down(&mgc_start_lock); + OBD_ALLOC_PTR(data); + if (data == NULL) + GOTO(out_free, rc = -ENOMEM); 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 */ - atomic_inc(&obd->u.cli.cl_mgc_refcount); + cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount); + + /* IR compatibility check, only for clients */ + if (lmd_is_client(lsi->lsi_lmd)) { + int has_ir; + int vallen = sizeof(*data); + __u32 *flags = &lsi->lsi_lmd->lmd_flags; + + rc = obd_get_info(NULL, obd->obd_self_export, + strlen(KEY_CONN_DATA), KEY_CONN_DATA, + &vallen, data, NULL); + LASSERT(rc == 0); + has_ir = OCD_HAS_FLAG(data, IMP_RECOV); + if (has_ir ^ !(*flags & LMD_FLG_NOIR)) { + /* LMD_FLG_NOIR is for test purpose only */ + LCONSOLE_WARN( + "Trying to mount a client with IR setting " + "not compatible with current mgc. " + "Force to use current mgc setting that is " + "IR %s.\n", + has_ir ? "enabled" : "disabled"); + if (has_ir) + *flags &= ~LMD_FLG_NOIR; + else + *flags |= LMD_FLG_NOIR; + } + } 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; } @@ -646,7 +560,7 @@ 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); @@ -658,9 +572,9 @@ 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)) { + if (IS_SERVER(lsi)) { + ptr = lsi->lsi_lmd->lmd_mgs; + if (IS_MGS(lsi)) { /* Use local nids (including LO) */ lnet_process_id_t id; while ((rc = LNetGetId(i++, &id)) != -ENOENT) { @@ -669,7 +583,11 @@ 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); } @@ -705,19 +623,19 @@ 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); + 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)) { + 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(ptr, &nid, &ptr) == 0) { + while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) { j++; rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID, niduuid, 0,0,0); @@ -741,7 +659,7 @@ 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) @@ -749,26 +667,33 @@ 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. */ - atomic_set(&obd->u.cli.cl_mgc_refcount, 1); + cfs_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 */ - OBD_ALLOC_PTR(data); - if (data == NULL) - GOTO(out, rc = -ENOMEM); - data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID | - OBD_CONNECT_AT; + + /* 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; + +#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; 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); @@ -781,8 +706,10 @@ out: to the same mgc.*/ lsi->lsi_mgc = obd; out_free: - mutex_up(&mgc_start_lock); + cfs_mutex_unlock(&mgc_start_lock); + if (data) + OBD_FREE_PTR(data); if (mgcname) OBD_FREE(mgcname, len); if (niduuid) @@ -805,13 +732,13 @@ static int lustre_stop_mgc(struct super_block *sb) 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)) { + 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 will call in here. */ CDEBUG(D_MOUNT, "mgc still has %d references.\n", - atomic_read(&obd->u.cli.cl_mgc_refcount)); + cfs_atomic_read(&obd->u.cli.cl_mgc_refcount)); GOTO(out, rc = -EBUSY); } @@ -857,7 +784,7 @@ out: OBD_FREE(niduuid, len); /* class_import_put will get rid of the additional connections */ - mutex_up(&mgc_start_lock); + cfs_mutex_unlock(&mgc_start_lock); RETURN(rc); } @@ -872,7 +799,7 @@ 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) { @@ -889,33 +816,33 @@ 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); } -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) +static int server_stop_servers(int lsiflags) { struct obd_device *obd = NULL; struct obd_type *type = NULL; int rc = 0; ENTRY; - 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 */ - if ((lddflags & LDD_F_SV_TYPE_MDT) && + if ((lsiflags & 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) && + if ((lsiflags & LDD_F_SV_TYPE_OST) && (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) { type = class_search_type(LUSTRE_OST_NAME); } @@ -929,7 +856,7 @@ static int server_stop_servers(int lddflags, int lsiflags) rc = err; } - mutex_up(&server_start_lock); + cfs_mutex_unlock(&server_start_lock); RETURN(rc); } @@ -945,26 +872,128 @@ int server_mti_print(char *title, struct mgs_target_info *mti) return(0); } -static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti) +/** 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 int server_name2fsname(char *svname, char *fsname, char **endptr) { - struct lustre_sb_info *lsi = s2lsi(sb); - struct lustre_disk_data *ldd = lsi->lsi_ldd; - lnet_process_id_t id; - int i = 0; + char *p; + + p = strstr(svname, "-OST"); + if (p == NULL) + p = strstr(svname, "-MDT"); + if (p == NULL) + return -1; + + if (fsname) { + strncpy(fsname, svname, p - svname); + fsname[p - svname] = '\0'; + } + + if (endptr != NULL) + *endptr = p; + + return 0; +} + +/** + * Get service name (svname) from string + * rc < 0 on error + * if endptr isn't NULL it is set to end of fsname * + */ +int server_name2svname(char *label, char *svname, char **endptr) +{ + int rc; + char *dash; + + /* We use server_name2fsname() just for parsing */ + rc = server_name2fsname(label, NULL, &dash); + if (rc != 0) + return rc; + + if (*dash != '-') + return -1; + + strncpy(svname, dash + 1, MTI_NAME_MAXLEN); + + return 0; +} +EXPORT_SYMBOL(server_name2svname); + + +/* 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) +{ + unsigned long index; + int rc; + char *dash; + + /* We use server_name2fsname() just for parsing */ + rc = server_name2fsname(svname, NULL, &dash); + if (rc != 0) + return rc; + + if (*dash != '-') + return -EINVAL; + + dash++; + + 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 (strcmp(dash, "all") == 0) + return rc | LDD_F_SV_ALL; + + index = simple_strtoul(dash, endptr, 16); + *idx = index; + + return rc; +} +EXPORT_SYMBOL(server_name2index); + +/* Generate data for registration */ +static int server_lsi2mti(struct lustre_sb_info *lsi, + struct mgs_target_info *mti) +{ + lnet_process_id_t id; + int rc, i = 0; ENTRY; - if (!(lsi->lsi_flags & LSI_SERVER)) + if (!IS_SERVER(lsi)) RETURN(-EINVAL); - strncpy(mti->mti_fsname, ldd->ldd_fsname, - sizeof(mti->mti_fsname)); - strncpy(mti->mti_svname, ldd->ldd_svname, - sizeof(mti->mti_svname)); + strncpy(mti->mti_svname, lsi->lsi_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 ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 && + class_match_nid(lsi->lsi_lmd->lmd_params, + PARAM_FAILNODE, id.nid) < 1) + continue; + + /* match specified network */ + if (!class_match_net(lsi->lsi_lmd->lmd_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) { @@ -976,37 +1005,44 @@ static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti) mti->mti_lustre_ver = LUSTRE_VERSION_CODE; mti->mti_config_ver = 0; - 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); + + rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL); + if (rc != 0) + return rc; + + rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL); + if (rc < 0) + return rc; + /* Orion requires index to be set */ + LASSERT(!(rc & LDD_F_NEED_INDEX)); + /* keep only LDD flags */ + mti->mti_flags = lsi->lsi_flags & LDD_F_MASK; + mti->mti_flags |= LDD_F_UPDATE; + strncpy(mti->mti_params, lsi->lsi_lmd->lmd_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) +static int server_register_target(struct lustre_sb_info *lsi) { - 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)) + if (!IS_SERVER(lsi)) RETURN(-EINVAL); OBD_ALLOC_PTR(mti); if (!mti) RETURN(-ENOMEM); - rc = server_sb2mti(sb, mti); + + rc = server_lsi2mti(lsi, mti); if (rc) GOTO(out, rc); @@ -1015,64 +1051,103 @@ int server_register_target(struct super_block *sb) libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index, mti->mti_flags); + /* if write_conf is true, the registration must succeed */ + writeconf = !!(lsi->lsi_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, + 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) { + 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", lsi->lsi_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); + } + +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 (!(IS_SERVER(lsi))) + RETURN(-EINVAL); + + OBD_ALLOC_PTR(mti); + if (!mti) + RETURN(-ENOMEM); + rc = server_lsi2mti(lsi, mti); if (rc) GOTO(out, rc); - /* Always update our flags */ - ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD; + mti->mti_instance = obd->u.obt.obt_instance; + mti->mti_flags |= LDD_F_OPC_READY; - /* 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); - } + /* FIXME use mgc_process_config instead */ + rc = obd_set_info_async(NULL, 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 |= LDD_F_IR_CAPABLE; out: if (mti) OBD_FREE_PTR(mti); RETURN(rc); + } -/* Start targets */ +/** 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; + struct lu_env env; + struct lu_device *dev; int rc; ENTRY; - CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname); + CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_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 */ - 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, @@ -1081,101 +1156,113 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt) LUSTRE_MDS_OBDNAME"_uuid", 0, 0); if (rc) { - mutex_up(&server_start_lock); + cfs_mutex_unlock(&server_start_lock); CERROR("failed to start MDS: %d\n", rc); RETURN(rc); } } - 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 (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) { + if (IS_OST(lsi)) { /* make sure OSS is started */ - 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, LUSTRE_OSS_NAME, LUSTRE_OSS_OBDNAME"_uuid", - 0, 0); + 0, 0, 0, 0); if (rc) { - mutex_up(&server_start_lock); + cfs_mutex_unlock(&server_start_lock); CERROR("failed to start OSS: %d\n", rc); RETURN(rc); } } - 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. */ - rc = server_mgc_set_fs(lsi->lsi_mgc, sb); - if (rc) - RETURN(rc); + /* Set the mgc fs to our server disk. This allows the MGC to + * read and write configs locally, in case it can't talk to the MGS. */ + if (lsi->lsi_srv_mnt) { + rc = server_mgc_set_fs(lsi->lsi_mgc, sb); + if (rc) + RETURN(rc); + } /* Register with MGS */ - rc = server_register_target(sb); - if (rc && (lsi->lsi_ldd->ldd_flags & - (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){ - CERROR("Required registration failed for %s: %d\n", - lsi->lsi_ldd->ldd_svname, rc); - if (rc == -EIO) { - LCONSOLE_ERROR_MSG(0x15f, "Communication error with " - "the MGS. Is the MGS running?\n"); - } - GOTO(out_mgc, rc); - } - if (rc == -EINVAL) { - 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 */ + rc = server_register_target(lsi); if (rc) - CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", 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); + rc = server_register_mount(lsi->lsi_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); - } + /* Start targets using the llog named for the target */ + memset(&cfg, 0, sizeof(cfg)); + rc = lustre_process_log(sb, lsi->lsi_svname, &cfg); + if (rc) { + CERROR("failed to start server %s: %d\n", + lsi->lsi_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 (lsi->lsi_srv_mnt) + server_mgc_clear_fs(lsi->lsi_mgc); if (!rc) { - obd = class_name2obd(lsi->lsi_ldd->ldd_svname); + obd = class_name2obd(lsi->lsi_svname); if (!obd) { CERROR("no server named %s was started\n", - lsi->lsi_ldd->ldd_svname); + lsi->lsi_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); + + /* log has been fully processed, let clients connect */ + dev = obd->obd_lu_dev; + if (dev && dev->ld_ops->ldo_prepare) { + rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags); + if (rc == 0) { + struct lu_context session_ctx; + + lu_context_init(&session_ctx, LCT_SESSION); + session_ctx.lc_thread = NULL; + lu_context_enter(&session_ctx); + env.le_ses = &session_ctx; + + dev->ld_ops->ldo_prepare(&env, NULL, dev); + + lu_env_fini(&env); + lu_context_exit(&session_ctx); + lu_context_fini(&session_ctx); + } + } + + /* abort recovery only on the complete stack: + * many devices can be involved */ + 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); + } } RETURN(rc); @@ -1198,9 +1285,11 @@ struct lustre_sb_info *lustre_init_lsi(struct super_block *sb) } lsi->lsi_lmd->lmd_exclude_count = 0; + lsi->lsi_lmd->lmd_recovery_time_soft = 0; + lsi->lsi_lmd->lmd_recovery_time_hard = 0; s2lsi_nocast(sb) = lsi; /* we take 1 extra ref for our setup */ - atomic_set(&lsi->lsi_mounts, 1); + cfs_atomic_set(&lsi->lsi_mounts, 1); /* Default umount style */ lsi->lsi_flags = LSI_UMOUNT_FAILOVER; @@ -1217,10 +1306,7 @@ 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(atomic_read(&lsi->lsi_mounts) == 0); - - if (lsi->lsi_ldd != NULL) - OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd)); + LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0); if (lsi->lsi_lmd != NULL) { if (lsi->lsi_lmd->lmd_dev != NULL) @@ -1239,6 +1325,15 @@ 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)); } @@ -1258,181 +1353,130 @@ static int lustre_put_lsi(struct super_block *sb) LASSERT(lsi != NULL); - CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts)); - if (atomic_dec_and_test(&lsi->lsi_mounts)) { + CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts)); + if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) { + if (IS_SERVER(lsi) && lsi->lsi_osd_exp) { + 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 */ -static struct vfsmount *server_kernel_mount(struct super_block *sb) +static int lsi_prepare(struct lustre_sb_info *lsi) { - 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; - char *options = NULL; - unsigned long page, s_flags; - struct page *__page; - 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); - mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, (void *)options); - 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); - - /* Add in any mount-line options */ - if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) { - int len = CFS_PAGE_SIZE - strlen(options) - 2; - if (*options != 0) - 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); - mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev, - (void *)options); - 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)); + __u32 index; + int rc; + ENTRY; + + LASSERT(lsi); + LASSERT(lsi->lsi_lmd); + + /* 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(lsi->lsi_svname)) + RETURN(-ENAMETOOLONG); + + strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile); + + /* 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); + } else { + strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME); + } + + /* XXX: a temp. solution for components using fsfilt + * to be removed in one of the subsequent patches */ + if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs")) { + strcpy(lsi->lsi_fstype, "ldiskfs"); + } else { + strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type); + } + + /* Determine server type */ + rc = server_name2index(lsi->lsi_svname, &index, NULL); + if (rc < 0) { + if (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); + } + } + lsi->lsi_flags |= rc; + + /* Add mount line flags that used to be in ldd: + * writeconf, mgs, iam, anything else? + */ + 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_VIRGIN) ? + LDD_F_VIRGIN : 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; + + RETURN(0); } -static void server_wait_finished(struct vfsmount *mnt) -{ - wait_queue_head_t waitq; - struct l_wait_info lwi; - int retries = 330; - - init_waitqueue_head(&waitq); - - while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) { - 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 */ - retries -= 5; - lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL); - l_wait_event(waitq, 0, &lwi); - } - if (atomic_read(&mnt->mnt_count) > 1) { - CERROR("%s: Mount still busy (%d refs), giving up.\n", - mnt->mnt_devname, atomic_read(&mnt->mnt_count)); - } -} +/*************** server mount ******************/ +/** Start the shutdown of servers at umount. + */ 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; char *tmpname, *extraname = NULL; int tmpname_sz; - int lddflags = lsi->lsi_ldd->ldd_flags; int lsiflags = lsi->lsi_flags; ENTRY; - LASSERT(lsiflags & LSI_SERVER); + LASSERT(IS_SERVER(lsi)); - tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1; + tmpname_sz = strlen(lsi->lsi_svname) + 1; OBD_ALLOC(tmpname, tmpname_sz); - memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz); + memcpy(tmpname, lsi->lsi_svname, tmpname_sz); CDEBUG(D_MOUNT, "server put_super %s\n", tmpname); + if (IS_MDT(lsi) && (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))) { + (IS_MDT(lsi) || IS_OST(lsi))) { struct lustre_profile *lprof = NULL; /* tell the mgc to drop the config log */ - lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL); + lustre_end_log(sb, lsi->lsi_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); + lprof = class_get_profile(lsi->lsi_svname); if (lprof && lprof->lp_dt) { OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1); strcpy(extraname, lprof->lp_dt); } - obd = class_name2obd(lsi->lsi_ldd->ldd_svname); + obd = class_name2obd(lsi->lsi_svname); if (obd) { CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name); if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER) @@ -1442,14 +1486,14 @@ static void server_put_super(struct super_block *sb) 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); + CERROR("no obd %s\n", lsi->lsi_svname); + server_deregister_mount(lsi->lsi_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)) { + if (IS_MGS(lsi)) { /* if MDS start with --nomgs, don't stop MGS then */ if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) server_stop_mgs(sb); @@ -1458,21 +1502,19 @@ static void server_put_super(struct super_block *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); + /* wait till all in-progress cleanups are done + * specifically we're interested in ofd cleanup + * as it pins OSS */ + obd_zombie_barrier(); - /* 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(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; @@ -1486,6 +1528,8 @@ static void server_put_super(struct super_block *sb) EXIT; } +/** Called only for 'umount -f' + */ #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT static void server_umount_begin(struct vfsmount *vfsmnt, int flags) { @@ -1509,45 +1553,41 @@ static void server_umount_begin(struct super_block *sb) 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; } -#ifndef HAVE_STATFS_DENTRY_PARAM -static int server_statfs (struct super_block *sb, struct kstatfs *buf) -{ -#else -static int server_statfs (struct dentry *dentry, struct kstatfs *buf) +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 (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); - } - } - - /* 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); + struct super_block *sb = dentry->d_sb; + struct lustre_sb_info *lsi = s2lsi(sb); + struct obd_statfs statfs; + int rc; + ENTRY; + + if (lsi->lsi_dt_dev) { + rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs); + if (rc == 0) { + statfs_unpack(buf, &statfs); + buf->f_type = sb->s_magic; + RETURN(0); + } + } + + /* 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); } +/** The operations we support directly on the superblock: + * mount, umount, and df. + */ static struct super_operations server_ops = { .put_super = server_put_super, @@ -1555,7 +1595,7 @@ static struct super_operations server_ops = .statfs = server_statfs, }; -#define log2(n) ffz(~(n)) +#define log2(n) cfs_ffz(~(n)) #define LUSTRE_SUPER_MAGIC 0x0BD00BD1 static int server_fill_super_common(struct super_block *sb) @@ -1568,7 +1608,7 @@ 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; @@ -1593,54 +1633,110 @@ static int server_fill_super_common(struct super_block *sb) RETURN(0); } +static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags) +{ + struct lustre_mount_data *lmd = lsi->lsi_lmd; + struct obd_device *obd; + struct dt_device_param p; + char flagstr[16]; + int rc; + ENTRY; + + CDEBUG(D_MOUNT, + "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n", + lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags); + + sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname); + strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname); + strcat(lsi->lsi_osd_uuid, "_UUID"); + sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags); + + obd = class_name2obd(lsi->lsi_osd_obdname); + if (obd == NULL) { + rc = lustre_start_simple(lsi->lsi_osd_obdname, + lsi->lsi_osd_type, + lsi->lsi_osd_uuid, lmd->lmd_dev, + flagstr, lsi->lsi_lmd->lmd_opts, + lsi->lsi_svname); + if (rc) + GOTO(out, rc); + obd = class_name2obd(lsi->lsi_osd_obdname); + LASSERT(obd); + } + + rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL); + if (rc) { + obd->obd_force = 1; + class_manual_cleanup(obd); + lsi->lsi_dt_dev = NULL; + } + + /* XXX: to keep support old components relying on lsi_srv_mnt + * we get this info from OSD just started */ + LASSERT(obd->obd_lu_dev); + lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev); + LASSERT(lsi->lsi_dt_dev); + + dt_conf_get(NULL, lsi->lsi_dt_dev, &p); + + lsi->lsi_srv_mnt = p.ddp_mnt; + +out: + RETURN(rc); +} + +/** 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) { 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); + rc = lsi_prepare(lsi); + if (rc) + RETURN(rc); + + /* Start low level OSD */ + rc = osd_start(lsi, sb->s_flags); + if (rc) { + CERROR("Unable to start osd on %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); + CDEBUG(D_MOUNT, "Found service %s on device %s\n", + lsi->lsi_svname, lsi->lsi_lmd->lmd_dev); - if (class_name2obd(lsi->lsi_ldd->ldd_svname)) { + if (class_name2obd(lsi->lsi_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); + lsi->lsi_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)){ + if (IS_MGS(lsi) && !(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); + (IS_OST(lsi) || IS_MDT(lsi))) { + rc = server_start_targets(sb, lsi->lsi_srv_mnt); if (rc < 0) { CERROR("Unable to start targets: %d\n", rc); GOTO(out_mnt, rc); @@ -1660,9 +1756,6 @@ static int server_fill_super(struct super_block *sb) if (rc) GOTO(out_mnt, rc); - LCONSOLE_WARN("Server %s on device %s has started\n", - 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. @@ -1672,40 +1765,65 @@ out_mnt: return rc; } -/* 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) +/* + * Calculate timeout value for a target. + */ +void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd) { - 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); + struct lustre_mount_data *lmd; + int soft = 0; + int hard = 0; + int factor = 0; + bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE); + int min = OBD_RECOVERY_TIME_MIN; + + LASSERT(IS_SERVER(lsi)); + + 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 (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; + if (soft == 0) + soft = OBD_RECOVERY_TIME_SOFT; + if (hard == 0) + hard = OBD_RECOVERY_TIME_HARD; - index = simple_strtoul(dash + 4, endptr, 16); - *idx = index; - return rc; + /* 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; } +EXPORT_SYMBOL(server_calc_timeout); /*************** mount common betweeen server and client ***************/ @@ -1733,8 +1851,8 @@ int lustre_common_put_super(struct super_block *sb) lu_types_stop(); RETURN(rc); } +EXPORT_SYMBOL(lustre_common_put_super); -#if 0 static void lmd_print(struct lustre_mount_data *lmd) { int i; @@ -1744,14 +1862,23 @@ static void lmd_print(struct lustre_mount_data *lmd) PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile); PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev); PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags); + if (lmd->lmd_opts) PRINT_CMD(PRINT_MASK, "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_hard) + PRINT_CMD(PRINT_MASK, "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]); } } -#endif /* Is this server on the exclusion list */ int lustre_check_exclusion(struct super_block *sb, char *svname) @@ -1859,7 +1986,77 @@ static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr) return 0; } -/* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */ +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; @@ -1883,11 +2080,18 @@ 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; while (*s1) { int clear = 0; + int time_min = OBD_RECOVERY_TIME_MIN; + /* Skip whitespace and extra commas */ while (*s1 == ' ' || *s1 == ',') s1++; @@ -1900,12 +2104,44 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) if (strncmp(s1, "abort_recov", 11) == 0) { lmd->lmd_flags |= LMD_FLG_ABORT_RECOV; clear++; + } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) { + lmd->lmd_recovery_time_soft = max_t(int, + simple_strtoul(s1 + 19, NULL, 10), time_min); + clear++; + } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) { + lmd->lmd_recovery_time_hard = max_t(int, + simple_strtoul(s1 + 19, NULL, 10), time_min); + clear++; + } else if (strncmp(s1, "noir", 4) == 0) { + lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */ + clear++; } 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, "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++; + } 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) @@ -1917,6 +2153,31 @@ 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 + length = tail - s1; + length -= 6; + strncat(lmd->lmd_params, s1 + 6, length); + strcat(lmd->lmd_params, " "); + 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. */ @@ -1951,7 +2212,7 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) s1 = strstr(devname, ":/"); if (s1) { ++s1; - lmd->lmd_flags = LMD_FLG_CLIENT; + lmd->lmd_flags |= LMD_FLG_CLIENT; /* Remove leading /s from fsname */ while (*++s1 == '/') ; /* Freed in lustre_free_lsi */ @@ -1979,6 +2240,7 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd) strcpy(lmd->lmd_opts, options); } + lmd_print(lmd); lmd->lmd_magic = LMD_MAGIC; RETURN(rc); @@ -1988,11 +2250,20 @@ invalid: RETURN(-EINVAL); } +struct lustre_mount_data2 { + void *lmd2_data; + struct vfsmount *lmd2_mnt; +}; -/* Common mount */ +/** This is the entry point for the mount call into Lustre. + * This is called when a server or client is mounted, + * and this is where we start setting things up. + * @param data Mount options (e.g. -o flock,abort_recov) + */ int lustre_fill_super(struct super_block *sb, void *data, int silent) { struct lustre_mount_data *lmd; + struct lustre_mount_data2 *lmd2 = data; struct lustre_sb_info *lsi; int rc; ENTRY; @@ -2008,10 +2279,15 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent) * Disable lockdep during mount, because mount locking patterns are * `special'. */ - lockdep_off(); + cfs_lockdep_off(); + + /* + * LU-639: the obd cleanup of last mount may not finish yet, wait here. + */ + obd_zombie_barrier(); /* Figure out the lmd from the mount options */ - if (lmd_parse((char *)data, lmd)) { + if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) { lustre_put_lsi(sb); GOTO(out, rc = -EINVAL); } @@ -2032,12 +2308,11 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent) } /* Connect and start */ /* (should always be ll_fill_super) */ - rc = (*client_fill_super)(sb); + rc = (*client_fill_super)(sb, lmd2->lmd2_mnt); /* 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 @@ -2056,37 +2331,42 @@ out: CDEBUG(D_SUPER, "Mount %s complete\n", lmd->lmd_dev); } - lockdep_on(); + cfs_lockdep_on(); return rc; } /* We can't call ll_fill_super by name because it lives in a module that must be loaded after this one. */ -void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb)) +void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb, + struct vfsmount *mnt)) { 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) +int lustre_get_sb(struct file_system_type *fs_type, int flags, + const char *devname, void * data, struct vfsmount *mnt) { - return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt); + struct lustre_mount_data2 lmd2 = { data, mnt }; + + return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt); } #endif @@ -2094,18 +2374,27 @@ 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); } +/** Register the "lustre" fs type + */ 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, +#endif .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, }; @@ -2119,16 +2408,4 @@ 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);