X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fptlrpc%2Frecover.c;h=6bf1d35e4bd43afbd5e688ab8781dc7a7511a900;hb=56a32318ff0509aaddb69466b4f17069c3f207b6;hp=0e44c86d1c81da7396fafb741b3d203c31dd09fa;hpb=cc4bd2873eab77210c69df32be466dae78479ad1;p=fs%2Flustre-release.git diff --git a/lustre/ptlrpc/recover.c b/lustre/ptlrpc/recover.c index 0e44c86..6bf1d35 100644 --- a/lustre/ptlrpc/recover.c +++ b/lustre/ptlrpc/recover.c @@ -1,219 +1,440 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Lustre Light Super operations + * Portal-RPC reconnection and replay operations, for use in recovery. * - * This code is issued under the GNU General Public License. - * See the file COPYING in this distribution + * Copyright (c) 2002, 2003 Cluster File Systems, Inc. + * Author: Mike Shaver * - * Copryright (C) 1996 Peter J. Braam - * Copryright (C) 1999 Stelias Computing Inc. - * Copryright (C) 1999 Seagate Technology Inc. - * Copryright (C) 2001 Mountain View Data, Inc. - * Copryright (C) 2002 Cluster File Systems, Inc. + * This file is part of Lustre, http://www.lustre.org. * + * Lustre is free software; you can redistribute it and/or + * modify it under the terms of version 2 of the GNU General Public + * License as published by the Free Software Foundation. + * + * Lustre is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with Lustre; if not, write to the Free Software + * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ -#include -#include -#include - -#define DEBUG_SUBSYSTEM S_LLITE +#define DEBUG_SUBSYSTEM S_RPC +#ifdef __KERNEL__ +# include +# include +# include +#else +# include +#endif -#include +#include #include +#include +#include +#include +#include +#include +#include +#include /* for IOC_LOV_SET_OSC_ACTIVE */ -int ll_reconnect(struct ptlrpc_connection *conn) +#include "ptlrpc_internal.h" + +static int ptlrpc_recover_import_no_retry(struct obd_import *, char *); + +void ptlrpc_run_recovery_over_upcall(struct obd_device *obd) { - struct ptlrpc_request *request; - struct list_head *tmp; - int rc = -EINVAL; - - /* XXX c_lock semantics! */ - conn->c_level = LUSTRE_CONN_CON; - - /* XXX this code MUST be shared with class_obd_connect! */ - list_for_each(tmp, &conn->c_imports) { - struct obd_import *imp = list_entry(tmp, struct obd_import, - imp_chain); - struct obd_device *obd = imp->imp_obd; - struct client_obd *cli = &obd->u.cli; - int rq_opc = (obd->obd_type->typ_ops->o_brw) - ? OST_CONNECT : MDS_CONNECT; - int size[] = { sizeof(cli->cl_target_uuid), - sizeof(obd->obd_uuid) }; - char *tmp[] = {cli->cl_target_uuid, obd->obd_uuid }; - struct lustre_handle old_hdl; - - LASSERT(imp->imp_connection == conn); - request = ptlrpc_prep_req(imp, rq_opc, 2, size, tmp); - request->rq_level = LUSTRE_CONN_NEW; - request->rq_replen = lustre_msg_size(0, NULL); - /* XXX are (addr, cookie) right? */ - request->rq_reqmsg->addr = imp->imp_handle.addr; - request->rq_reqmsg->cookie = imp->imp_handle.cookie; - rc = ptlrpc_queue_wait(request); - rc = ptlrpc_check_status(request, rc); - if (rc) { - CERROR("cannot connect to %s@%s: rc = %d\n", - cli->cl_target_uuid, conn->c_remote_uuid, rc); - ptlrpc_free_req(request); - GOTO(out_disc, rc = -ENOTCONN); - } + char *argv[4]; + char *envp[3]; + int rc; + ENTRY; - old_hdl = imp->imp_handle; - imp->imp_handle.addr = request->rq_repmsg->addr; - imp->imp_handle.cookie = request->rq_repmsg->cookie; - CERROR("reconnected to %s@%s (%Lx/%Lx, was %Lx/%Lx)!\n", - cli->cl_target_uuid, conn->c_remote_uuid, - imp->imp_handle.addr, imp->imp_handle.cookie, - old_hdl.addr, old_hdl.cookie); - ptlrpc_free_req(request); - } - conn->c_level = LUSTRE_CONN_RECOVD; + argv[0] = obd_lustre_upcall; + argv[1] = "RECOVERY_OVER"; + argv[2] = (char *)obd->obd_uuid.uuid; + argv[3] = NULL; + + envp[0] = "HOME=/"; + envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; + envp[2] = NULL; + + rc = USERMODEHELPER(argv[0], argv, envp); + if (rc < 0) { + CERROR("Error invoking recovery upcall %s %s %s: %d; check " + "/proc/sys/lustre/upcall\n", + argv[0], argv[1], argv[2], rc); - out_disc: - return rc; + } else { + CWARN("Invoked upcall %s %s %s\n", + argv[0], argv[1], argv[2]); + } } -static int ll_recover_upcall(struct ptlrpc_connection *conn) +void ptlrpc_run_failed_import_upcall(struct obd_import* imp) { - char *argv[3]; +#ifdef __KERNEL__ + unsigned long flags; + char *argv[7]; char *envp[3]; - + int rc; ENTRY; - conn->c_level = LUSTRE_CONN_RECOVD; - argv[0] = obd_recovery_upcall; - argv[1] = conn->c_remote_uuid; - argv[2] = NULL; + spin_lock_irqsave(&imp->imp_lock, flags); + if (imp->imp_state == LUSTRE_IMP_CLOSED) { + spin_unlock_irqrestore(&imp->imp_lock, flags); + EXIT; + return; + } + spin_unlock_irqrestore(&imp->imp_lock, flags); + + argv[0] = obd_lustre_upcall; + argv[1] = "FAILED_IMPORT"; + argv[2] = (char *)imp->imp_target_uuid.uuid; + argv[3] = imp->imp_obd->obd_name; + argv[4] = (char *)imp->imp_connection->c_remote_uuid.uuid; + argv[5] = (char *)imp->imp_obd->obd_uuid.uuid; + argv[6] = NULL; envp[0] = "HOME=/"; envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; envp[2] = NULL; - RETURN(call_usermodehelper(argv[0], argv, envp)); + rc = USERMODEHELPER(argv[0], argv, envp); + if (rc < 0) { + CERROR("Error invoking recovery upcall %s %s %s %s %s: %d; " + "check /proc/sys/lustre/upcall\n", + argv[0], argv[1], argv[2], argv[3], argv[4],rc); + + } else { + CWARN("Invoked upcall %s %s %s %s %s\n", + argv[0], argv[1], argv[2], argv[3], argv[4]); + } +#else + if (imp->imp_state == LUSTRE_IMP_CLOSED) { + EXIT; + return; + } + ptlrpc_recover_import(imp, NULL); +#endif +} + +/* This might block waiting for the upcall to start, so it should + * not be called from a thread that shouldn't block. (Like ptlrpcd) */ +void ptlrpc_initiate_recovery(struct obd_import *imp) +{ + ENTRY; + + LASSERT (obd_lustre_upcall != NULL); + + if (strcmp(obd_lustre_upcall, "DEFAULT") == 0) { + CDEBUG(D_HA, "%s: starting recovery without upcall\n", + imp->imp_target_uuid.uuid); + ptlrpc_connect_import(imp, NULL); + } + else if (strcmp(obd_lustre_upcall, "NONE") == 0) { + CDEBUG(D_HA, "%s: recovery disabled\n", + imp->imp_target_uuid.uuid); + } + else { + CDEBUG(D_HA, "%s: calling upcall to start recovery\n", + imp->imp_target_uuid.uuid); + ptlrpc_run_failed_import_upcall(imp); + } + + EXIT; } -static int ll_recover_reconnect(struct ptlrpc_connection *conn) +int ptlrpc_replay_next(struct obd_import *imp, int *inflight) { int rc = 0; struct list_head *tmp, *pos; - struct ptlrpc_request *req; + struct ptlrpc_request *req = NULL; + unsigned long flags; + __u64 last_transno; ENTRY; - /* 1. reconnect */ - rc = ll_reconnect(conn); - if (rc) - RETURN(rc); - - /* 2. walk the request list */ - spin_lock(&conn->c_lock); + *inflight = 0; - list_for_each_safe(tmp, pos, &conn->c_sending_head) { - req = list_entry(tmp, struct ptlrpc_request, rq_list); - - /* replay what needs to be replayed */ - if (req->rq_flags & PTL_RPC_FL_REPLAY) { - CDEBUG(D_NET, "req %Ld needs replay [last rcvd %Ld]\n", - req->rq_xid, conn->c_last_xid); - rc = ptlrpc_replay_req(req); -#if 0 -#error We should not hold a spinlock over such a lengthy operation. -#error If necessary, drop spinlock, do operation, re-get spinlock, restart loop. -#error If we need to avoid re-processint items, then delete them from the list -#error as they are replayed and re-add at the tail of this list, so the next -#error item to process will always be at the head of the list. -#endif - if (rc) { - CERROR("recovery replay error %d for req %Ld\n", - rc, req->rq_xid); - GOTO(out, rc); - } - } + /* It might have committed some after we last spoke, so make sure we + * get rid of them now. + */ + spin_lock_irqsave(&imp->imp_lock, flags); + ptlrpc_free_committed(imp); + last_transno = imp->imp_last_replay_transno; + spin_unlock_irqrestore(&imp->imp_lock, flags); - /* server has seen req, we have reply: skip */ - if ((req->rq_flags & PTL_RPC_FL_REPLIED) && - req->rq_xid <= conn->c_last_xid) { - CDEBUG(D_NET, - "req %Ld was complete: skip [last rcvd %Ld]\n", - req->rq_xid, conn->c_last_xid); - continue; - } + CDEBUG(D_HA, "import %p from %s committed "LPU64" last "LPU64"\n", + imp, imp->imp_target_uuid.uuid, imp->imp_peer_committed_transno, + last_transno); + + /* Do I need to hold a lock across this iteration? We shouldn't be + * racing with any additions to the list, because we're in recovery + * and are therefore not processing additional requests to add. Calls + * to ptlrpc_free_committed might commit requests, but nothing "newer" + * than the one we're replaying (it can't be committed until it's + * replayed, and we're doing that here). l_f_e_safe protects against + * problems with the current request being committed, in the unlikely + * event of that race. So, in conclusion, I think that it's safe to + * perform this list-walk without the imp_lock held. + * + * But, the {mdc,osc}_replay_open callbacks both iterate + * request lists, and have comments saying they assume the + * imp_lock is being held by ptlrpc_replay, but it's not. it's + * just a little race... + */ + list_for_each_safe(tmp, pos, &imp->imp_replay_list) { + req = list_entry(tmp, struct ptlrpc_request, rq_replay_list); - /* server has lost req, we have reply: resend, ign reply */ - if ((req->rq_flags & PTL_RPC_FL_REPLIED) && - req->rq_xid > conn->c_last_xid) { - CDEBUG(D_NET, "lost req %Ld have rep: replay [last " - "rcvd %Ld]\n", req->rq_xid, conn->c_last_xid); - rc = ptlrpc_replay_req(req); - if (rc) { - CERROR("request resend error %d for req %Ld\n", - rc, req->rq_xid); - GOTO(out, rc); - } + /* If need to resend, stop on the matching one first. It's + possible though it's already been committed, so in that case + we'll just continue with replay */ + if (imp->imp_resend_replay && + req->rq_transno == last_transno) { + lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT); + break; } - /* server has seen req, we have lost reply: -ERESTARTSYS */ - if ( !(req->rq_flags & PTL_RPC_FL_REPLIED) && - req->rq_xid <= conn->c_last_xid) { - CDEBUG(D_NET, "lost rep %Ld srv did req: restart " - "[last rcvd %Ld]\n", - req->rq_xid, conn->c_last_xid); - ptlrpc_restart_req(req); + if (req->rq_transno > last_transno) { + imp->imp_last_replay_transno = req->rq_transno; + break; } - /* service has not seen req, no reply: resend */ - if ( !(req->rq_flags & PTL_RPC_FL_REPLIED) && - req->rq_xid > conn->c_last_xid) { - CDEBUG(D_NET, - "lost rep/req %Ld: resend [last rcvd %Ld]\n", - req->rq_xid, conn->c_last_xid); - ptlrpc_resend_req(req); + req = NULL; + } + + imp->imp_resend_replay = 0; + + if (req != NULL) { + rc = ptlrpc_replay_req(req); + if (rc) { + CERROR("recovery replay error %d for req " + LPD64"\n", rc, req->rq_xid); + RETURN(rc); } + imp->imp_reqs_replayed++; + *inflight = 1; + } + RETURN(rc); +} + +int ptlrpc_resend(struct obd_import *imp) +{ + struct list_head *tmp, *pos; + struct ptlrpc_request *req; + unsigned long flags; + ENTRY; + + /* As long as we're in recovery, nothing should be added to the sending + * list, so we don't need to hold the lock during this iteration and + * resend process. + */ + /* Well... what if lctl recover is called twice at the same time? + */ + spin_lock_irqsave(&imp->imp_lock, flags); + if (imp->imp_state != LUSTRE_IMP_RECOVER) { + spin_unlock_irqrestore(&imp->imp_lock, flags); + RETURN(-1); } + spin_unlock_irqrestore(&imp->imp_lock, flags); - conn->c_level = LUSTRE_CONN_FULL; - recovd_conn_fixed(conn); + list_for_each_safe(tmp, pos, &imp->imp_sending_list) { + req = list_entry(tmp, struct ptlrpc_request, rq_list); + ptlrpc_resend_req(req); + } - CDEBUG(D_NET, "recovery complete on conn %p(%s), waking delayed reqs\n", - conn, conn->c_remote_uuid); - /* Finally, continue what we delayed since recovery started */ - list_for_each_safe(tmp, pos, &conn->c_delayed_head) { + RETURN(0); +} + +void ptlrpc_wake_delayed(struct obd_import *imp) +{ + unsigned long flags; + struct list_head *tmp, *pos; + struct ptlrpc_request *req; + + spin_lock_irqsave(&imp->imp_lock, flags); + list_for_each_safe(tmp, pos, &imp->imp_delayed_list) { req = list_entry(tmp, struct ptlrpc_request, rq_list); - ptlrpc_continue_req(req); + + DEBUG_REQ(D_HA, req, "waking (set %p):", req->rq_set); + ptlrpc_wake_client_req(req); + } + spin_unlock_irqrestore(&imp->imp_lock, flags); +} + +void ptlrpc_request_handle_notconn(struct ptlrpc_request *failed_req) +{ + int rc; + struct obd_import *imp= failed_req->rq_import; + unsigned long flags; + ENTRY; + + CDEBUG(D_HA, "import %s of %s@%s abruptly disconnected: reconnecting\n", + imp->imp_obd->obd_name, + imp->imp_target_uuid.uuid, + imp->imp_connection->c_remote_uuid.uuid); + + if (ptlrpc_set_import_discon(imp)) { + if (!imp->imp_replayable) { + CDEBUG(D_HA, "import %s@%s for %s not replayable, " + "auto-deactivating\n", + imp->imp_target_uuid.uuid, + imp->imp_connection->c_remote_uuid.uuid, + imp->imp_obd->obd_name); + ptlrpc_deactivate_import(imp); + } + + /* to control recovery via lctl {disable|enable}_recovery */ + if (imp->imp_deactive == 0) + rc = ptlrpc_connect_import(imp, NULL); } + /* Wait for recovery to complete and resend. If evicted, then + this request will be errored out later.*/ + spin_lock_irqsave(&failed_req->rq_lock, flags); + failed_req->rq_resend = 1; + spin_unlock_irqrestore(&failed_req->rq_lock, flags); + EXIT; - out: - spin_unlock(&conn->c_lock); - return rc; } -static int ll_retry_recovery(struct ptlrpc_connection *conn) +/* + * This should only be called by the ioctl interface, currently + * with the lctl deactivate and activate commands. + */ +int ptlrpc_set_import_active(struct obd_import *imp, int active) { - /* XXX use a timer, sideshow bob */ - recovd_conn_fail(conn); - return 0; + struct obd_device *obd = imp->imp_obd; + int rc = 0; + + LASSERT(obd); + + /* When deactivating, mark import invalid, and abort in-flight + * requests. */ + if (!active) { + ptlrpc_invalidate_import(imp, 0); + imp->imp_deactive = 1; + } + + /* When activating, mark import valid, and attempt recovery */ + if (active) { + imp->imp_deactive = 0; + CDEBUG(D_HA, "setting import %s VALID\n", + imp->imp_target_uuid.uuid); + rc = ptlrpc_recover_import(imp, NULL); + } + + RETURN(rc); } -int ll_recover(struct recovd_data *rd, int phase) +int ptlrpc_recover_import(struct obd_import *imp, char *new_uuid) { - struct ptlrpc_connection *conn = class_rd2conn(rd); + int rc; + ENTRY; + + /* force import to be disconnected. */ + ptlrpc_set_import_discon(imp); + + rc = ptlrpc_recover_import_no_retry(imp, new_uuid); - LASSERT(conn); + RETURN(rc); +} + +int ptlrpc_import_in_recovery(struct obd_import *imp) +{ + unsigned long flags; + int in_recovery = 1; + spin_lock_irqsave(&imp->imp_lock, flags); + if (imp->imp_state == LUSTRE_IMP_FULL || + imp->imp_state == LUSTRE_IMP_CLOSED || + imp->imp_state == LUSTRE_IMP_DISCON) + in_recovery = 0; + spin_unlock_irqrestore(&imp->imp_lock, flags); + return in_recovery; +} + +int ptlrpc_import_control_recovery(struct obd_import *imp, int disable) +{ + unsigned long flags; + + /* with imp_deactivate == 1 pinger won't initiate re-connect */ + spin_lock_irqsave(&imp->imp_lock, flags); + if (disable) + imp->imp_deactive = 1; + else + imp->imp_deactive = 0; + if (imp->imp_state == LUSTRE_IMP_DISCON) { + imp->imp_force_verify = 1; + ptlrpc_pinger_wake_up(); + } + spin_unlock_irqrestore(&imp->imp_lock, flags); + RETURN(0); +} + +static int ptlrpc_recover_import_no_retry(struct obd_import *imp, + char *new_uuid) +{ + int rc; + unsigned long flags; + int in_recovery = 0; + struct l_wait_info lwi; ENTRY; - switch (phase) { - case PTLRPC_RECOVD_PHASE_PREPARE: - RETURN(ll_recover_upcall(conn)); - case PTLRPC_RECOVD_PHASE_RECOVER: - RETURN(ll_recover_reconnect(conn)); - case PTLRPC_RECOVD_PHASE_FAILURE: - RETURN(ll_retry_recovery(conn)); + spin_lock_irqsave(&imp->imp_lock, flags); + if (imp->imp_state != LUSTRE_IMP_DISCON) { + in_recovery = 1; } + spin_unlock_irqrestore(&imp->imp_lock, flags); - LBUG(); - RETURN(-ENOSYS); + if (in_recovery == 1) + RETURN(-EALREADY); + + rc = ptlrpc_connect_import(imp, new_uuid); + if (rc) + RETURN(rc); + + CDEBUG(D_HA, "%s: recovery started, waiting\n", + imp->imp_target_uuid.uuid); + + lwi = LWI_TIMEOUT(MAX(obd_timeout * HZ, 1), NULL, NULL); + rc = l_wait_event(imp->imp_recovery_waitq, + !ptlrpc_import_in_recovery(imp), &lwi); + CDEBUG(D_HA, "%s: recovery finished\n", + imp->imp_target_uuid.uuid); + + RETURN(rc); +} + +void ptlrpc_fail_export(struct obd_export *exp) +{ + int rc, already_failed; + unsigned long flags; + + spin_lock_irqsave(&exp->exp_lock, flags); + already_failed = exp->exp_failed; + exp->exp_failed = 1; + spin_unlock_irqrestore(&exp->exp_lock, flags); + + if (already_failed) { + CDEBUG(D_HA, "disconnecting dead export %p/%s; skipping\n", + exp, exp->exp_client_uuid.uuid); + return; + } + + CDEBUG(D_HA, "disconnecting export %p/%s\n", + exp, exp->exp_client_uuid.uuid); + + if (obd_dump_on_timeout) + portals_debug_dumplog(); + + /* Most callers into obd_disconnect are removing their own reference + * (request, for example) in addition to the one from the hash table. + * We don't have such a reference here, so make one. */ + class_export_get(exp); + rc = obd_disconnect(exp, 0); + if (rc) + CERROR("disconnecting export %p failed: %d\n", exp, rc); }