X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lib.c;h=290005ec24338c841a6331aa7aa28c2752890517;hp=6a8f448e75c288086c0ea6ca395bafe7fc9528e8;hb=27be04d33db216e576486b2db6e1cb54aed60006;hpb=0ca7eb23309070f85f06f6e192395ed305c59108 diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c index 6a8f448..290005e 100644 --- a/lustre/ldlm/ldlm_lib.c +++ b/lustre/ldlm/ldlm_lib.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. @@ -28,15 +26,14 @@ /* * Copyright (c) 2003, 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/ * Lustre is a trademark of Sun Microsystems, Inc. */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_LDLM #ifdef __KERNEL__ @@ -190,6 +187,31 @@ out: RETURN(rc); } +/** + * Find conn uuid by peer nid. @peer is a server nid. This function is used + * to find a conn uuid of @imp which can reach @peer. + */ +int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer, + struct obd_uuid *uuid) +{ + struct obd_import_conn *conn; + int rc = -ENOENT; + ENTRY; + + cfs_spin_lock(&imp->imp_lock); + cfs_list_for_each_entry(conn, &imp->imp_conn_list, oic_item) { + /* check if conn uuid does have this peer nid */ + if (class_check_uuid(&conn->oic_uuid, peer)) { + *uuid = conn->oic_uuid; + rc = 0; + break; + } + } + cfs_spin_unlock(&imp->imp_lock); + RETURN(rc); +} +EXPORT_SYMBOL(client_import_find_conn); + void client_destroy_import(struct obd_import *imp) { /* drop security policy instance after all rpc finished/aborted @@ -199,6 +221,30 @@ void client_destroy_import(struct obd_import *imp) sptlrpc_import_sec_put(imp); class_import_put(imp); } +EXPORT_SYMBOL(client_destroy_import); + +/** + * check whether the osc is on MDT or not + * In the config log, + * osc on MDT + * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID + * osc on client + * setup 0:{fsname}-OSTxxxx-osc 1:lustre-OST0000_UUID 2:NID + * + **/ +static int osc_on_mdt(char *obdname) +{ + char *ptr; + + ptr = strrchr(obdname, '-'); + if (ptr == NULL) + return 0; + + if (strncmp(ptr + 1, "MDT", 3) == 0) + return 1; + + return 0; +} /* configure an RPC client OBD device * @@ -289,8 +335,9 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) CFS_INIT_LIST_HEAD(&cli->cl_loi_hp_ready_list); CFS_INIT_LIST_HEAD(&cli->cl_loi_write_list); CFS_INIT_LIST_HEAD(&cli->cl_loi_read_list); - CFS_INIT_LIST_HEAD(&cli->cl_loi_sync_fs_list); client_obd_list_lock_init(&cli->cl_loi_list_lock); + cfs_atomic_set(&cli->cl_pending_w_pages, 0); + cfs_atomic_set(&cli->cl_pending_r_pages, 0); cli->cl_r_in_flight = 0; cli->cl_w_in_flight = 0; @@ -328,9 +375,11 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) } else if (cfs_num_physpages >> (20 - CFS_PAGE_SHIFT) <= 512 /* MB */) { cli->cl_max_rpcs_in_flight = 4; } else { - cli->cl_max_rpcs_in_flight = OSC_MAX_RIF_DEFAULT; + if (osc_on_mdt(obddev->obd_name)) + cli->cl_max_rpcs_in_flight = MDS_OSC_MAX_RIF_DEFAULT; + else + cli->cl_max_rpcs_in_flight = OSC_MAX_RIF_DEFAULT; } - rc = ldlm_get_ref(); if (rc) { CERROR("ldlm_get_ref failed: %d\n", rc); @@ -402,6 +451,8 @@ int client_obd_cleanup(struct obd_device *obddev) ldlm_namespace_free_post(obddev->obd_namespace); obddev->obd_namespace = NULL; + LASSERT(obddev->u.cli.cl_import == NULL); + ldlm_put_ref(); RETURN(0); } @@ -444,7 +495,7 @@ int client_connect_import(const struct lu_env *env, imp->imp_connect_flags_orig = data->ocd_connect_flags; } - rc = ptlrpc_connect_import(imp, NULL); + rc = ptlrpc_connect_import(imp); if (rc != 0) { LASSERT (imp->imp_state == LUSTRE_IMP_DISCON); GOTO(out_ldlm, rc); @@ -534,13 +585,6 @@ int client_disconnect_export(struct obd_export *exp) ptlrpc_invalidate_import(imp); - if (imp->imp_rq_pool) { - ptlrpc_free_rq_pool(imp->imp_rq_pool); - imp->imp_rq_pool = NULL; - } - client_destroy_import(imp); - cli->cl_import = NULL; - EXIT; out_disconnect: @@ -555,6 +599,7 @@ int client_disconnect_export(struct obd_export *exp) RETURN(rc); } +#ifdef HAVE_SERVER_SUPPORT int server_disconnect_export(struct obd_export *exp) { int rc; @@ -575,18 +620,20 @@ int server_disconnect_export(struct obd_export *exp) struct ptlrpc_reply_state *rs = cfs_list_entry(exp->exp_outstanding_replies.next, struct ptlrpc_reply_state, rs_exp_list); - struct ptlrpc_service *svc = rs->rs_service; + struct ptlrpc_service_part *svcpt = rs->rs_svcpt; - cfs_spin_lock(&svc->srv_rs_lock); - cfs_list_del_init(&rs->rs_exp_list); - cfs_spin_lock(&rs->rs_lock); - ptlrpc_schedule_difficult_reply(rs); - cfs_spin_unlock(&rs->rs_lock); - cfs_spin_unlock(&svc->srv_rs_lock); - } - cfs_spin_unlock(&exp->exp_lock); + cfs_spin_lock(&svcpt->scp_rep_lock); - RETURN(rc); + cfs_list_del_init(&rs->rs_exp_list); + cfs_spin_lock(&rs->rs_lock); + ptlrpc_schedule_difficult_reply(rs); + cfs_spin_unlock(&rs->rs_lock); + + cfs_spin_unlock(&svcpt->scp_rep_lock); + } + cfs_spin_unlock(&exp->exp_lock); + + RETURN(rc); } /* -------------------------------------------------------------------------- @@ -598,24 +645,49 @@ static int target_handle_reconnect(struct lustre_handle *conn, struct obd_uuid *cluuid) { ENTRY; + if (exp->exp_connection && exp->exp_imp_reverse) { struct lustre_handle *hdl; + struct obd_device *target; + hdl = &exp->exp_imp_reverse->imp_remote_handle; + target = exp->exp_obd; + /* Might be a re-connect after a partition. */ if (!memcmp(&conn->cookie, &hdl->cookie, sizeof conn->cookie)) { - CWARN("%s: %s reconnecting\n", exp->exp_obd->obd_name, - cluuid->uuid); + if (target->obd_recovering) { + int timeout = cfs_duration_sec(cfs_time_sub( + cfs_timer_deadline( + &target->obd_recovery_timer), + cfs_time_current())); + + LCONSOLE_WARN("%s: Client %s (at %s) reconnect" + "ing, waiting for %d clients in recov" + "ery for %d:%.02d\n", target->obd_name, + obd_uuid2str(&exp->exp_client_uuid), + obd_export_nid2str(exp), + target->obd_max_recoverable_clients, + timeout / 60, timeout % 60); + } else { + LCONSOLE_WARN("%s: Client %s (at %s) " + "reconnecting\n", target->obd_name, + obd_uuid2str(&exp->exp_client_uuid), + obd_export_nid2str(exp)); + } + conn->cookie = exp->exp_handle.h_cookie; /* target_handle_connect() treats EALREADY and * -EALREADY differently. EALREADY means we are * doing a valid reconnect from the same client. */ RETURN(EALREADY); } else { - CERROR("%s reconnecting from %s, " - "handle mismatch (ours "LPX64", theirs " - LPX64")\n", cluuid->uuid, - exp->exp_connection->c_remote_uuid.uuid, - hdl->cookie, conn->cookie); + LCONSOLE_WARN("%s: already connected client %s (at %s) " + "with handle "LPX64". Rejecting client " + "with the same UUID trying to reconnect " + "with handle "LPX64"\n", target->obd_name, + obd_uuid2str(&exp->exp_client_uuid), + obd_export_nid2str(exp), + hdl->cookie, conn->cookie); memset(conn, 0, sizeof *conn); /* target_handle_connect() treats EALREADY and * -EALREADY differently. -EALREADY is an error @@ -645,14 +717,21 @@ void target_client_add_cb(struct obd_device *obd, __u64 transno, void *cb_data, } EXPORT_SYMBOL(target_client_add_cb); +#ifdef __KERNEL__ static void -target_start_and_reset_recovery_timer(struct obd_device *obd, - struct ptlrpc_request *req, - int new_client); +check_and_start_recovery_timer(struct obd_device *obd, + struct ptlrpc_request *req, int new_client); +#else +static inline void +check_and_start_recovery_timer(struct obd_device *obd, + struct ptlrpc_request *req, int new_client) +{ +} +#endif int target_handle_connect(struct ptlrpc_request *req) { - struct obd_device *target, *targref = NULL; + struct obd_device *target = NULL, *targref = NULL; struct obd_export *export = NULL; struct obd_import *revimp; struct lustre_handle conn; @@ -662,8 +741,11 @@ int target_handle_connect(struct ptlrpc_request *req) struct obd_uuid remote_uuid; char *str; int rc = 0; + char *target_start; + int target_len; int mds_conn = 0; struct obd_connect_data *data, *tmpdata; + int size, tmpsize; lnet_nid_t *client_nid = NULL; ENTRY; @@ -680,17 +762,30 @@ int target_handle_connect(struct ptlrpc_request *req) if (!target) target = class_name2obd(str); - if (!target || target->obd_stopping || !target->obd_set_up) { - LCONSOLE_ERROR_MSG(0x137, "UUID '%s' is not available " - " for connect (%s)\n", str, - !target ? "no target" : - (target->obd_stopping ? "stopping" : - "not set up")); - GOTO(out, rc = -ENODEV); - } + if (!target) { + deuuidify(str, NULL, &target_start, &target_len); + LCONSOLE_ERROR_MSG(0x137, "UUID '%s' is not available for " + "connect (no target)\n", str); + GOTO(out, rc = -ENODEV); + } + + cfs_spin_lock(&target->obd_dev_lock); + if (target->obd_stopping || !target->obd_set_up) { + cfs_spin_unlock(&target->obd_dev_lock); + + deuuidify(str, NULL, &target_start, &target_len); + LCONSOLE_ERROR_MSG(0x137, "%.*s: Not available for connect " + "from %s (%s)\n", target_len, target_start, + libcfs_nid2str(req->rq_peer.nid), + (target->obd_stopping ? + "stopping" : "not set up")); + GOTO(out, rc = -ENODEV); + } if (target->obd_no_conn) { - LCONSOLE_WARN("%s: temporarily refusing client connection " + cfs_spin_unlock(&target->obd_dev_lock); + + LCONSOLE_WARN("%s: Temporarily refusing client connection " "from %s\n", target->obd_name, libcfs_nid2str(req->rq_peer.nid)); GOTO(out, rc = -EAGAIN); @@ -701,6 +796,8 @@ int target_handle_connect(struct ptlrpc_request *req) Really, class_uuid2obd should take the ref. */ targref = class_incref(target, __FUNCTION__, cfs_current()); + target->obd_conn_inprogress++; + cfs_spin_unlock(&target->obd_dev_lock); str = req_capsule_client_get(&req->rq_pill, &RMF_CLUUID); if (str == NULL) { @@ -731,6 +828,8 @@ int target_handle_connect(struct ptlrpc_request *req) conn = *tmp; + size = req_capsule_get_size(&req->rq_pill, &RMF_CONNECT_DATA, + RCL_CLIENT); data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA); if (!data) GOTO(out, rc = -EPROTO); @@ -782,22 +881,23 @@ int target_handle_connect(struct ptlrpc_request *req) /* we've found an export in the hash */ if (export->exp_connecting) { /* bug 9635, et. al. */ - CWARN("%s: exp %p already connecting\n", - export->exp_obd->obd_name, export); + LCONSOLE_WARN("%s: Export %p already connecting from %s\n", + export->exp_obd->obd_name, export, + libcfs_nid2str(req->rq_peer.nid)); class_export_put(export); export = NULL; rc = -EALREADY; } else if (mds_conn && export->exp_connection) { if (req->rq_peer.nid != export->exp_connection->c_peer.nid) /* mds reconnected after failover */ - CWARN("%s: received MDS connection from NID %s," - " removing former export from NID %s\n", + LCONSOLE_WARN("%s: Received MDS connection from " + "%s, removing former export from %s\n", target->obd_name, libcfs_nid2str(req->rq_peer.nid), libcfs_nid2str(export->exp_connection->c_peer.nid)); else /* new mds connection from the same nid */ - CWARN("%s: received new MDS connection from NID %s," - " removing former export from same NID\n", + LCONSOLE_WARN("%s: Received new MDS connection from " + "%s, removing former export from same NID\n", target->obd_name, libcfs_nid2str(req->rq_peer.nid)); class_fail_export(export); class_export_put(export); @@ -809,11 +909,12 @@ int target_handle_connect(struct ptlrpc_request *req) MSG_CONNECT_INITIAL)) { /* in mds failover we have static uuid but nid can be * changed*/ - CWARN("%s: cookie %s seen on new NID %s when " - "existing NID %s is already connected\n", - target->obd_name, cluuid.uuid, - libcfs_nid2str(req->rq_peer.nid), - libcfs_nid2str(export->exp_connection->c_peer.nid)); + LCONSOLE_WARN("%s: Client %s seen on new nid %s when " + "existing nid %s is already connected\n", + target->obd_name, cluuid.uuid, + libcfs_nid2str(req->rq_peer.nid), + libcfs_nid2str( + export->exp_connection->c_peer.nid)); rc = -EALREADY; class_export_put(export); export = NULL; @@ -821,7 +922,6 @@ int target_handle_connect(struct ptlrpc_request *req) cfs_spin_lock(&export->exp_lock); export->exp_connecting = 1; cfs_spin_unlock(&export->exp_lock); - class_export_put(export); LASSERT(export->exp_obd == target); rc = target_handle_reconnect(&conn, export, &cluuid); @@ -833,18 +933,20 @@ no_export: OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_CONNECT, 2 * obd_timeout); } else if (req->rq_export == NULL && cfs_atomic_read(&export->exp_rpc_count) > 0) { - CWARN("%s: refuse connection from %s/%s to 0x%p/%d\n", - target->obd_name, cluuid.uuid, - libcfs_nid2str(req->rq_peer.nid), - export, cfs_atomic_read(&export->exp_refcount)); + LCONSOLE_WARN("%s: Client %s (at %s) refused connection, " + "still busy with %d references\n", + target->obd_name, cluuid.uuid, + libcfs_nid2str(req->rq_peer.nid), + cfs_atomic_read(&export->exp_refcount)); GOTO(out, rc = -EBUSY); } else if (req->rq_export != NULL && (cfs_atomic_read(&export->exp_rpc_count) > 1)) { /* the current connect rpc has increased exp_rpc_count */ - CWARN("%s: refuse reconnection from %s@%s to 0x%p/%d\n", - target->obd_name, cluuid.uuid, - libcfs_nid2str(req->rq_peer.nid), - export, cfs_atomic_read(&export->exp_rpc_count) - 1); + LCONSOLE_WARN("%s: Client %s (at %s) refused reconnection, " + "still busy with %d active RPCs\n", + target->obd_name, cluuid.uuid, + libcfs_nid2str(req->rq_peer.nid), + cfs_atomic_read(&export->exp_rpc_count) - 1); cfs_spin_lock(&export->exp_lock); if (req->rq_export->exp_conn_cnt < lustre_msg_get_conn_cnt(req->rq_reqmsg)) @@ -853,9 +955,12 @@ no_export: cfs_spin_unlock(&export->exp_lock); GOTO(out, rc = -EBUSY); } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1) { - CERROR("%s: NID %s (%s) reconnected with 1 conn_cnt; " - "cookies not random?\n", target->obd_name, - libcfs_nid2str(req->rq_peer.nid), cluuid.uuid); + if (!strstr(cluuid.uuid, "mdt")) + LCONSOLE_WARN("%s: Rejecting reconnect from the " + "known client %s (at %s) because it " + "is indicating it is a new client", + target->obd_name, cluuid.uuid, + libcfs_nid2str(req->rq_peer.nid)); GOTO(out, rc = -EALREADY); } else { OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_RECONNECT, 2 * obd_timeout); @@ -865,21 +970,16 @@ no_export: GOTO(out, rc); } - CWARN("%s: connection from %s@%s %st"LPU64" exp %p cur %ld last %ld\n", + CDEBUG(D_HA, "%s: connection from %s@%s %st"LPU64" exp %p cur %ld last %ld\n", target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid), target->obd_recovering ? "recovering/" : "", data->ocd_transno, export, (long)cfs_time_current_sec(), export ? (long)export->exp_last_request_time : 0); - /* Tell the client if we're in recovery. */ - if (target->obd_recovering) { - lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING); - /* If this is the first time a client connects, - reset the recovery timer */ - if (rc == 0) - target_start_and_reset_recovery_timer(target, req, - !export); - } + /* If this is the first time a client connects, reset the recovery + * timer */ + if (rc == 0 && target->obd_recovering) + check_and_start_recovery_timer(target, req, export == NULL); /* We want to handle EALREADY but *not* -EALREADY from * target_handle_reconnect(), return reconnection state in a flag */ @@ -901,13 +1001,16 @@ no_export: t = cfs_timer_deadline(&target->obd_recovery_timer); t = cfs_time_sub(t, cfs_time_current()); - CERROR("%s: denying connection for new client %s (%s): " - "%d clients in recovery for "CFS_TIME_T"s\n", - target->obd_name, - libcfs_nid2str(req->rq_peer.nid), cluuid.uuid, - cfs_atomic_read(&target-> \ - obd_lock_replay_clients), - cfs_duration_sec(t)); + t = cfs_duration_sec(t); + LCONSOLE_WARN("%s: Denying connection for new client " + "%s (at %s), waiting for %d clients in " + "recovery for %d:%.02d\n", + target->obd_name, + libcfs_nid2str(req->rq_peer.nid), + cluuid.uuid, + cfs_atomic_read(&target-> \ + obd_lock_replay_clients), + (int)t / 60, (int)t % 60); rc = -EBUSY; } else { dont_check_exports: @@ -920,19 +1023,24 @@ dont_check_exports: } else { rc = obd_reconnect(req->rq_svc_thread->t_env, export, target, &cluuid, data, client_nid); - if (rc == 0) - /* prevous done via class_conn2export */ - class_export_get(export); } if (rc) GOTO(out, rc); + + LASSERT(target->u.obt.obt_magic == OBT_MAGIC); + data->ocd_instance = target->u.obt.obt_instance; + /* Return only the parts of obd_connect_data that we understand, so the * client knows that we don't understand the rest. */ if (data) { - tmpdata = req_capsule_server_get(&req->rq_pill, - &RMF_CONNECT_DATA); - //data->ocd_connect_flags &= OBD_CONNECT_SUPPORTED; - *tmpdata = *data; + tmpsize = req_capsule_get_size(&req->rq_pill, &RMF_CONNECT_DATA, + RCL_SERVER); + tmpdata = req_capsule_server_get(&req->rq_pill, + &RMF_CONNECT_DATA); + /* Don't use struct assignment here, because the client reply + * buffer may be smaller/larger than the local struct + * obd_connect_data. */ + memcpy(tmpdata, data, min(tmpsize, size)); } /* If all else goes well, this is our RPC return code. */ @@ -950,12 +1058,14 @@ dont_check_exports: if (req->rq_export != NULL) class_export_put(req->rq_export); - req->rq_export = export; + /* request takes one export refcount */ + req->rq_export = class_export_get(export); cfs_spin_lock(&export->exp_lock); if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) { cfs_spin_unlock(&export->exp_lock); - CERROR("%s: %s already connected at higher conn_cnt: %d > %d\n", + CDEBUG(D_RPCTRACE, "%s: %s already connected at higher " + "conn_cnt: %d > %d\n", cluuid.uuid, libcfs_nid2str(req->rq_peer.nid), export->exp_conn_cnt, lustre_msg_get_conn_cnt(req->rq_reqmsg)); @@ -998,35 +1108,57 @@ dont_check_exports: &export->exp_nid_hash); } - cfs_spin_lock(&target->obd_recovery_task_lock); if (target->obd_recovering && !export->exp_in_recovery) { + int has_transno; + __u64 transno = data->ocd_transno; + cfs_spin_lock(&export->exp_lock); + /* possible race with class_disconnect_stale_exports, + * export may be already in the eviction process */ + if (export->exp_failed) { + cfs_spin_unlock(&export->exp_lock); + GOTO(out, rc = -ENODEV); + } export->exp_in_recovery = 1; export->exp_req_replay_needed = 1; export->exp_lock_replay_needed = 1; cfs_spin_unlock(&export->exp_lock); - if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_TRANSNO) - && (data->ocd_transno == 0)) + + has_transno = !!(lustre_msg_get_op_flags(req->rq_reqmsg) & + MSG_CONNECT_TRANSNO); + if (has_transno && transno == 0) CWARN("Connect with zero transno!\n"); - if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_TRANSNO) - && data->ocd_transno < target->obd_next_recovery_transno) - target->obd_next_recovery_transno = data->ocd_transno; - target->obd_connected_clients++; + if (has_transno && transno > 0 && + transno < target->obd_next_recovery_transno && + transno > target->obd_last_committed) { + /* another way is to use cmpxchg() so it will be + * lock free */ + cfs_spin_lock(&target->obd_recovery_task_lock); + if (transno < target->obd_next_recovery_transno) + target->obd_next_recovery_transno = transno; + cfs_spin_unlock(&target->obd_recovery_task_lock); + } + cfs_atomic_inc(&target->obd_req_replay_clients); cfs_atomic_inc(&target->obd_lock_replay_clients); - if (target->obd_connected_clients == + if (cfs_atomic_inc_return(&target->obd_connected_clients) == target->obd_max_recoverable_clients) cfs_waitq_signal(&target->obd_next_transno_waitq); } - cfs_spin_unlock(&target->obd_recovery_task_lock); + + /* Tell the client we're in recovery, when client is involved in it. */ + if (target->obd_recovering) + lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING); + tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN); conn = *tmp; if (export->exp_imp_reverse != NULL) { /* destroyed import can be still referenced in ctxt */ - obd_set_info_async(export, sizeof(KEY_REVIMP_UPD), - KEY_REVIMP_UPD, 0, NULL, NULL); + obd_set_info_async(req->rq_svc_thread->t_env, export, + sizeof(KEY_REVIMP_UPD), KEY_REVIMP_UPD, + 0, NULL, NULL); client_destroy_import(export->exp_imp_reverse); } @@ -1075,9 +1207,16 @@ out: cfs_spin_lock(&export->exp_lock); export->exp_connecting = 0; cfs_spin_unlock(&export->exp_lock); + + class_export_put(export); } - if (targref) + if (targref) { + cfs_spin_lock(&target->obd_dev_lock); + target->obd_conn_inprogress--; + cfs_spin_unlock(&target->obd_dev_lock); + class_decref(targref, __FUNCTION__, cfs_current()); + } if (rc) req->rq_status = rc; RETURN(rc); @@ -1182,9 +1321,17 @@ static void target_exp_dequeue_req_replay(struct ptlrpc_request *req) #ifdef __KERNEL__ static void target_finish_recovery(struct obd_device *obd) { + time_t elapsed_time = max_t(time_t, 1, cfs_time_current_sec() - + obd->obd_recovery_start); ENTRY; - LCONSOLE_INFO("%s: sending delayed replies to recovered clients\n", - obd->obd_name); + + LCONSOLE_INFO("%s: Recovery over after %d:%.02d, of %d clients " + "%d recovered and %d %s evicted.\n", obd->obd_name, + (int)elapsed_time / 60, (int)elapsed_time % 60, + obd->obd_max_recoverable_clients, + cfs_atomic_read(&obd->obd_connected_clients), + obd->obd_stale_clients, + obd->obd_stale_clients == 1 ? "was" : "were"); ldlm_reprocess_all_ns(obd->obd_namespace); cfs_spin_lock(&obd->obd_recovery_task_lock); @@ -1255,7 +1402,6 @@ static void abort_lock_replay_queue(struct obd_device *obd) target_request_copy_put(req); } } -#endif /* Called from a cleanup function if the device is being cleaned up forcefully. The exports should all have been disconnected already, @@ -1313,56 +1459,83 @@ void target_cancel_recovery_timer(struct obd_device *obd) cfs_timer_disarm(&obd->obd_recovery_timer); } -/* extend = 1 means require at least "duration" seconds left in the timer, - extend = 0 means set the total duration (start_recovery_timer) */ -static void reset_recovery_timer(struct obd_device *obd, int duration, - int extend) +static void target_start_recovery_timer(struct obd_device *obd) { - cfs_time_t now = cfs_time_current_sec(); - cfs_duration_t left; + if (obd->obd_recovery_start != 0) + return; - cfs_spin_lock(&obd->obd_recovery_task_lock); + cfs_spin_lock(&obd->obd_dev_lock); if (!obd->obd_recovering || obd->obd_abort_recovery) { - cfs_spin_unlock(&obd->obd_recovery_task_lock); + cfs_spin_unlock(&obd->obd_dev_lock); return; } - left = cfs_time_sub(obd->obd_recovery_end, now); + LASSERT(obd->obd_recovery_timeout != 0); - if (extend && (duration > left)) - obd->obd_recovery_timeout += duration - left; - else if (!extend && (duration > obd->obd_recovery_timeout)) - /* Track the client's largest expected replay time */ - obd->obd_recovery_timeout = duration; + if (obd->obd_recovery_start != 0) { + cfs_spin_unlock(&obd->obd_dev_lock); + return; + } - /* Hard limit of obd_recovery_time_hard which should not happen */ - if (obd->obd_recovery_timeout > obd->obd_recovery_time_hard) - obd->obd_recovery_timeout = obd->obd_recovery_time_hard; + cfs_timer_arm(&obd->obd_recovery_timer, + cfs_time_shift(obd->obd_recovery_timeout)); + obd->obd_recovery_start = cfs_time_current_sec(); + cfs_spin_unlock(&obd->obd_dev_lock); - obd->obd_recovery_end = obd->obd_recovery_start + - obd->obd_recovery_timeout; - if (!cfs_timer_is_armed(&obd->obd_recovery_timer) || - cfs_time_before(now, obd->obd_recovery_end)) { - left = cfs_time_sub(obd->obd_recovery_end, now); - cfs_timer_arm(&obd->obd_recovery_timer, cfs_time_shift(left)); - } - cfs_spin_unlock(&obd->obd_recovery_task_lock); - CDEBUG(D_HA, "%s: recovery timer will expire in %u seconds\n", - obd->obd_name, (unsigned)left); + LCONSOLE_WARN("%s: Will be in recovery for at least %d:%.02d, " + "or until %d client%s reconnect%s\n", + obd->obd_name, + obd->obd_recovery_timeout / 60, + obd->obd_recovery_timeout % 60, + obd->obd_max_recoverable_clients, + (obd->obd_max_recoverable_clients == 1) ? "" : "s", + (obd->obd_max_recoverable_clients == 1) ? "s": ""); } -static void check_and_start_recovery_timer(struct obd_device *obd) +/** + * extend recovery window. + * + * if @extend is true, extend recovery window to have @drt remaining at least; + * otherwise, make sure the recovery timeout value is not less than @drt. + */ +static void extend_recovery_timer(struct obd_device *obd, int drt, bool extend) { - cfs_spin_lock(&obd->obd_recovery_task_lock); - if (cfs_timer_is_armed(&obd->obd_recovery_timer)) { - cfs_spin_unlock(&obd->obd_recovery_task_lock); + cfs_time_t now; + cfs_time_t end; + cfs_duration_t left; + int to; + + cfs_spin_lock(&obd->obd_dev_lock); + if (!obd->obd_recovering || obd->obd_abort_recovery) { + cfs_spin_unlock(&obd->obd_dev_lock); return; } - CDEBUG(D_HA, "%s: starting recovery timer\n", obd->obd_name); - obd->obd_recovery_start = cfs_time_current_sec(); - cfs_spin_unlock(&obd->obd_recovery_task_lock); + LASSERT(obd->obd_recovery_start != 0); + + now = cfs_time_current_sec(); + to = obd->obd_recovery_timeout; + end = obd->obd_recovery_start + to; + left = cfs_time_sub(end, now); - reset_recovery_timer(obd, obd->obd_recovery_timeout, 0); + if (extend && (drt > left)) { + to += drt - left; + } else if (!extend && (drt > to)) { + to = drt; + /* reduce drt by already passed time */ + drt -= obd->obd_recovery_timeout - left; + } + + if (to > obd->obd_recovery_time_hard) + to = obd->obd_recovery_time_hard; + if (obd->obd_recovery_timeout < to) { + obd->obd_recovery_timeout = to; + cfs_timer_arm(&obd->obd_recovery_timer, + cfs_time_shift(drt)); + } + cfs_spin_unlock(&obd->obd_dev_lock); + + CDEBUG(D_HA, "%s: recovery timer will expire in %u seconds\n", + obd->obd_name, (unsigned)drt); } /* Reset the timer with each new client connection */ @@ -1377,19 +1550,21 @@ static void check_and_start_recovery_timer(struct obd_device *obd) */ static void -target_start_and_reset_recovery_timer(struct obd_device *obd, - struct ptlrpc_request *req, - int new_client) +check_and_start_recovery_timer(struct obd_device *obd, + struct ptlrpc_request *req, + int new_client) { int service_time = lustre_msg_get_service_time(req->rq_reqmsg); + struct obd_device_target *obt = &obd->u.obt; + struct lustre_sb_info *lsi; if (!new_client && service_time) /* Teach server about old server's estimates, as first guess * at how long new requests will take. */ - at_measured(&req->rq_rqbd->rqbd_service->srv_at_estimate, + at_measured(&req->rq_rqbd->rqbd_svcpt->scp_at_estimate, service_time); - check_and_start_recovery_timer(obd); + target_start_recovery_timer(obd); /* convert the service time to rpc timeout, * reuse service_time to limit stack usage */ @@ -1398,14 +1573,17 @@ target_start_and_reset_recovery_timer(struct obd_device *obd, /* We expect other clients to timeout within service_time, then try * to reconnect, then try the failover server. The max delay between * connect attempts is SWITCH_MAX + SWITCH_INC + INITIAL */ - service_time += 2 * (CONNECTION_SWITCH_MAX + CONNECTION_SWITCH_INC + - INITIAL_CONNECT_TIMEOUT); + service_time += 2 * INITIAL_CONNECT_TIMEOUT; + + LASSERT(obt->obt_magic == OBT_MAGIC); + lsi = s2lsi(obt->obt_sb); + if (!(lsi->lsi_flags | LSI_IR_CAPABLE)) + service_time += 2 * (CONNECTION_SWITCH_MAX + + CONNECTION_SWITCH_INC); if (service_time > obd->obd_recovery_timeout && !new_client) - reset_recovery_timer(obd, service_time, 0); + extend_recovery_timer(obd, service_time, false); } -#ifdef __KERNEL__ - /** Health checking routines */ static inline int exp_connect_healthy(struct obd_export *exp) { @@ -1438,12 +1616,13 @@ static inline int exp_finished(struct obd_export *exp) /** Checking routines for recovery */ static int check_for_clients(struct obd_device *obd) { + unsigned int clnts = cfs_atomic_read(&obd->obd_connected_clients); + if (obd->obd_abort_recovery || obd->obd_recovery_expired) return 1; - LASSERT(obd->obd_connected_clients <= obd->obd_max_recoverable_clients); + LASSERT(clnts <= obd->obd_max_recoverable_clients); if (obd->obd_no_conn == 0 && - obd->obd_connected_clients + obd->obd_stale_clients == - obd->obd_max_recoverable_clients) + clnts + obd->obd_stale_clients == obd->obd_max_recoverable_clients) return 1; return 0; } @@ -1464,7 +1643,7 @@ static int check_for_next_transno(struct obd_device *obd) req_transno = 0; } - connected = obd->obd_connected_clients; + connected = cfs_atomic_read(&obd->obd_connected_clients); completed = connected - cfs_atomic_read(&obd->obd_req_replay_clients); queue_len = obd->obd_requests_queued_for_recovery; next_transno = obd->obd_next_recovery_transno; @@ -1556,8 +1735,8 @@ repeat: } else if (obd->obd_recovery_expired) { obd->obd_recovery_expired = 0; /** If some clients died being recovered, evict them */ - CDEBUG(D_WARNING, - "recovery is timed out, evict stale exports\n"); + LCONSOLE_WARN("%s: recovery is timed out, " + "evict stale exports\n", obd->obd_name); /** evict cexports with no replay in queue, they are stalled */ class_disconnect_stale_exports(obd, health_check); /** continue with VBR */ @@ -1568,7 +1747,7 @@ repeat: * reset timer, recovery will proceed with versions now, * timeout is set just to handle reconnection delays */ - reset_recovery_timer(obd, RECONNECT_DELAY_MAX, 1); + extend_recovery_timer(obd, RECONNECT_DELAY_MAX, true); /** Wait for recovery events again, after evicting bad clients */ goto repeat; } @@ -1686,13 +1865,28 @@ static int handle_recovery_req(struct ptlrpc_thread *thread, lu_context_fini(&req->rq_recov_session); /* don't reset timer for final stage */ if (!exp_finished(req->rq_export)) { + int to = obd_timeout; + /** * Add request timeout to the recovery time so next request from * this client may come in recovery time */ - reset_recovery_timer(class_exp2obd(req->rq_export), - AT_OFF ? obd_timeout : - lustre_msg_get_timeout(req->rq_reqmsg), 1); + if (!AT_OFF) { + struct ptlrpc_service_part *svcpt; + + svcpt = req->rq_rqbd->rqbd_svcpt; + /* If the server sent early reply for this request, + * the client will recalculate the timeout according to + * current server estimate service time, so we will + * use the maxium timeout here for waiting the client + * sending the next req */ + to = max((int)at_est2timeout( + at_get(&svcpt->scp_at_estimate)), + (int)lustre_msg_get_timeout(req->rq_reqmsg)); + /* Add net_latency (see ptlrpc_replay_req) */ + to += lustre_msg_get_service_time(req->rq_reqmsg); + } + extend_recovery_timer(class_exp2obd(req->rq_export), to, true); } reqcopy_put: RETURN(rc); @@ -1706,8 +1900,8 @@ static int target_recovery_thread(void *arg) struct target_recovery_data *trd = &obd->obd_recovery_data; unsigned long delta; unsigned long flags; - struct lu_env env; - struct ptlrpc_thread fake_svc_thread, *thread = &fake_svc_thread; + struct lu_env *env; + struct ptlrpc_thread *thread = NULL; int rc = 0; ENTRY; @@ -1718,16 +1912,30 @@ static int target_recovery_thread(void *arg) RECALC_SIGPENDING; SIGNAL_MASK_UNLOCK(current, flags); - rc = lu_context_init(&env.le_ctx, LCT_MD_THREAD); - if (rc) + OBD_ALLOC_PTR(thread); + if (thread == NULL) + RETURN(-ENOMEM); + + OBD_ALLOC_PTR(env); + if (env == NULL) { + OBD_FREE_PTR(thread); + RETURN(-ENOMEM); + } + + rc = lu_context_init(&env->le_ctx, LCT_MD_THREAD | LCT_DT_THREAD); + if (rc) { + OBD_FREE_PTR(thread); + OBD_FREE_PTR(env); RETURN(rc); + } - thread->t_env = &env; + thread->t_env = env; thread->t_id = -1; /* force filter_iobuf_get/put to use local buffers */ - env.le_ctx.lc_thread = thread; + env->le_ctx.lc_thread = thread; thread->t_data = NULL; + thread->t_watchdog = NULL; - CERROR("%s: started recovery thread pid %d\n", obd->obd_name, + CDEBUG(D_HA, "%s: started recovery thread pid %d\n", obd->obd_name, cfs_curproc_pid()); trd->trd_processing_task = cfs_curproc_pid(); @@ -1816,9 +2024,12 @@ static int target_recovery_thread(void *arg) target_finish_recovery(obd); - lu_context_fini(&env.le_ctx); + lu_context_fini(&env->le_ctx); trd->trd_processing_task = 0; cfs_complete(&trd->trd_finishing); + + OBD_FREE_PTR(thread); + OBD_FREE_PTR(env); RETURN(rc); } @@ -1834,7 +2045,7 @@ static int target_start_recovery_thread(struct lu_target *lut, cfs_init_completion(&trd->trd_finishing); trd->trd_recovery_handler = handler; - if (cfs_kernel_thread(target_recovery_thread, lut, 0) > 0) { + if (cfs_create_thread(target_recovery_thread, lut, 0) > 0) { cfs_wait_for_completion(&trd->trd_starting); LASSERT(obd->obd_recovering != 0); } else @@ -1874,7 +2085,7 @@ static void target_recovery_expired(unsigned long castmeharder) " after %lds (%d clients connected)\n", obd->obd_name, cfs_atomic_read(&obd->obd_lock_replay_clients), cfs_time_current_sec()- obd->obd_recovery_start, - obd->obd_connected_clients); + cfs_atomic_read(&obd->obd_connected_clients)); obd->obd_recovery_expired = 1; cfs_waitq_signal(&obd->obd_next_transno_waitq); @@ -1897,17 +2108,12 @@ void target_recovery_init(struct lu_target *lut, svc_handler_t handler) obd->obd_recovery_start = 0; obd->obd_recovery_end = 0; - /* both values can be get from mount data already */ - if (obd->obd_recovery_timeout == 0) - obd->obd_recovery_timeout = OBD_RECOVERY_TIME_SOFT; - if (obd->obd_recovery_time_hard == 0) - obd->obd_recovery_time_hard = OBD_RECOVERY_TIME_HARD; cfs_timer_init(&obd->obd_recovery_timer, target_recovery_expired, obd); target_start_recovery_thread(lut, handler); } EXPORT_SYMBOL(target_recovery_init); -#endif +#endif /* __KERNEL__ */ static int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req) @@ -1998,7 +2204,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req, /* CAVEAT EMPTOR: The incoming request message has been swabbed * (i.e. buflens etc are in my own byte order), but type-dependent - * buffers (eg mds_body, ost_body etc) have NOT been swabbed. */ + * buffers (eg mdt_body, ost_body etc) have NOT been swabbed. */ if (!transno) { CFS_INIT_LIST_HEAD(&req->rq_list); @@ -2016,8 +2222,9 @@ int target_queue_recovery_request(struct ptlrpc_request *req, * Also, a resent, replayed request that has already been * handled will pass through here and be processed immediately. */ - CWARN("Next recovery transno: "LPU64", current: "LPU64", replaying\n", - obd->obd_next_recovery_transno, transno); + CDEBUG(D_HA, "Next recovery transno: "LPU64 + ", current: "LPU64", replaying\n", + obd->obd_next_recovery_transno, transno); cfs_spin_lock(&obd->obd_recovery_task_lock); if (transno < obd->obd_next_recovery_transno) { /* Processing the queue right now, don't re-add. */ @@ -2076,6 +2283,31 @@ int target_queue_recovery_request(struct ptlrpc_request *req, RETURN(0); } +int target_handle_ping(struct ptlrpc_request *req) +{ + obd_ping(req->rq_svc_thread->t_env, req->rq_export); + return req_capsule_server_pack(&req->rq_pill); +} + +void target_committed_to_req(struct ptlrpc_request *req) +{ + struct obd_export *exp = req->rq_export; + + if (!exp->exp_obd->obd_no_transno && req->rq_repmsg != NULL) + lustre_msg_set_last_committed(req->rq_repmsg, + exp->exp_last_committed); + else + DEBUG_REQ(D_IOCTL, req, "not sending last_committed update (%d/" + "%d)", exp->exp_obd->obd_no_transno, + req->rq_repmsg == NULL); + + CDEBUG(D_INFO, "last_committed "LPU64", transno "LPU64", xid "LPU64"\n", + exp->exp_last_committed, req->rq_transno, req->rq_xid); +} +EXPORT_SYMBOL(target_committed_to_req); + +#endif /* HAVE_SERVER_SUPPORT */ + /** * Packs current SLV and Limit into \a req. */ @@ -2116,7 +2348,7 @@ int target_send_reply_msg(struct ptlrpc_request *req, int rc, int fail_id) } if (unlikely(rc)) { - DEBUG_REQ(D_ERROR, req, "processing error (%d)", rc); + DEBUG_REQ(D_NET, req, "processing error (%d)", rc); req->rq_status = rc; return (ptlrpc_send_error(req, 1)); } else { @@ -2128,11 +2360,10 @@ int target_send_reply_msg(struct ptlrpc_request *req, int rc, int fail_id) void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) { + struct ptlrpc_service_part *svcpt; int netrc; struct ptlrpc_reply_state *rs; - struct obd_device *obd; struct obd_export *exp; - struct ptlrpc_service *svc; ENTRY; if (req->rq_no_reply) { @@ -2140,7 +2371,7 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) return; } - svc = req->rq_rqbd->rqbd_service; + svcpt = req->rq_rqbd->rqbd_svcpt; rs = req->rq_reply_state; if (rs == NULL || !rs->rs_difficult) { /* no notifiers */ @@ -2152,7 +2383,7 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) /* must be an export if locks saved */ LASSERT (req->rq_export != NULL); /* req/reply consistent */ - LASSERT (rs->rs_service == svc); + LASSERT(rs->rs_svcpt == svcpt); /* "fresh" reply */ LASSERT (!rs->rs_scheduled); @@ -2164,7 +2395,6 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) LASSERT (cfs_list_empty(&rs->rs_exp_list)); exp = class_export_get (req->rq_export); - obd = exp->exp_obd; /* disable reply scheduling while I'm setting up */ rs->rs_scheduled = 1; @@ -2190,9 +2420,9 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) netrc = target_send_reply_msg (req, rc, fail_id); - cfs_spin_lock(&svc->srv_rs_lock); + cfs_spin_lock(&svcpt->scp_rep_lock); - cfs_atomic_inc(&svc->srv_n_difficult_replies); + cfs_atomic_inc(&svcpt->scp_nreps_difficult); if (netrc != 0) { /* error sending: reply is off the net. Also we need +1 @@ -2212,37 +2442,14 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) CDEBUG(D_HA, "Schedule reply immediately\n"); ptlrpc_dispatch_difficult_reply(rs); } else { - cfs_list_add (&rs->rs_list, &svc->srv_active_replies); - rs->rs_scheduled = 0; /* allow notifier to schedule */ - } - cfs_spin_unlock(&rs->rs_lock); - cfs_spin_unlock(&svc->srv_rs_lock); - EXIT; + cfs_list_add(&rs->rs_list, &svcpt->scp_rep_active); + rs->rs_scheduled = 0; /* allow notifier to schedule */ + } + cfs_spin_unlock(&rs->rs_lock); + cfs_spin_unlock(&svcpt->scp_rep_lock); + EXIT; } -int target_handle_ping(struct ptlrpc_request *req) -{ - obd_ping(req->rq_export); - return req_capsule_server_pack(&req->rq_pill); -} - -void target_committed_to_req(struct ptlrpc_request *req) -{ - struct obd_export *exp = req->rq_export; - - if (!exp->exp_obd->obd_no_transno && req->rq_repmsg != NULL) - lustre_msg_set_last_committed(req->rq_repmsg, - exp->exp_last_committed); - else - DEBUG_REQ(D_IOCTL, req, "not sending last_committed update (%d/" - "%d)", exp->exp_obd->obd_no_transno, - req->rq_repmsg == NULL); - - CDEBUG(D_INFO, "last_committed "LPU64", transno "LPU64", xid "LPU64"\n", - exp->exp_last_committed, req->rq_transno, req->rq_xid); -} -EXPORT_SYMBOL(target_committed_to_req); - int target_handle_qc_callback(struct ptlrpc_request *req) { struct obd_quotactl *oqctl; @@ -2455,8 +2662,115 @@ void ldlm_dump_export_locks(struct obd_export *exp) CERROR("dumping locks for export %p," "ignore if the unmount doesn't hang\n", exp); cfs_list_for_each_entry(lock, &exp->exp_locks_list, l_exp_refs_link) - ldlm_lock_dump(D_ERROR, lock, 0); + LDLM_ERROR(lock, "lock:"); } cfs_spin_unlock(&exp->exp_locks_list_guard); } #endif + +#ifdef HAVE_SERVER_SUPPORT +static int target_bulk_timeout(void *data) +{ + ENTRY; + /* We don't fail the connection here, because having the export + * killed makes the (vital) call to commitrw very sad. + */ + RETURN(1); +} + +static inline char *bulk2type(struct ptlrpc_bulk_desc *desc) +{ + return desc->bd_type == BULK_GET_SINK ? "GET" : "PUT"; +} + +int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, + struct l_wait_info *lwi) +{ + struct ptlrpc_request *req = desc->bd_req; + int rc = 0; + ENTRY; + + /* Check if there is eviction in progress, and if so, wait for + * it to finish */ + if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) { + *lwi = LWI_INTR(NULL, NULL); + rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq, + !cfs_atomic_read(&exp->exp_obd-> + obd_evict_inprogress), + lwi); + } + + /* Check if client was evicted or tried to reconnect already */ + if (exp->exp_failed || exp->exp_abort_active_req) { + rc = -ENOTCONN; + } else { + if (desc->bd_type == BULK_PUT_SINK) + rc = sptlrpc_svc_wrap_bulk(req, desc); + if (rc == 0) + rc = ptlrpc_start_bulk_transfer(desc); + } + + if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) { + ptlrpc_abort_bulk(desc); + } else if (rc == 0) { + time_t start = cfs_time_current_sec(); + do { + long timeoutl = req->rq_deadline - cfs_time_current_sec(); + cfs_duration_t timeout = timeoutl <= 0 ? + CFS_TICK : cfs_time_seconds(timeoutl); + *lwi = LWI_TIMEOUT_INTERVAL(timeout, + cfs_time_seconds(1), + target_bulk_timeout, + desc); + rc = l_wait_event(desc->bd_waitq, + !ptlrpc_server_bulk_active(desc) || + exp->exp_failed || + exp->exp_abort_active_req, + lwi); + LASSERT(rc == 0 || rc == -ETIMEDOUT); + /* Wait again if we changed deadline */ + } while ((rc == -ETIMEDOUT) && + (req->rq_deadline > cfs_time_current_sec())); + + if (rc == -ETIMEDOUT) { + DEBUG_REQ(D_ERROR, req, + "timeout on bulk %s after %ld%+lds", + bulk2type(desc), + req->rq_deadline - start, + cfs_time_current_sec() - + req->rq_deadline); + ptlrpc_abort_bulk(desc); + } else if (exp->exp_failed) { + DEBUG_REQ(D_ERROR, req, "Eviction on bulk %s", + bulk2type(desc)); + rc = -ENOTCONN; + ptlrpc_abort_bulk(desc); + } else if (exp->exp_abort_active_req) { + DEBUG_REQ(D_ERROR, req, "Reconnect on bulk %s", + bulk2type(desc)); + /* we don't reply anyway */ + rc = -ETIMEDOUT; + ptlrpc_abort_bulk(desc); + } else if (!desc->bd_success || + desc->bd_nob_transferred != desc->bd_nob) { + DEBUG_REQ(D_ERROR, req, "%s bulk %s %d(%d)", + desc->bd_success ? + "truncated" : "network error on", + bulk2type(desc), + desc->bd_nob_transferred, + desc->bd_nob); + /* XXX should this be a different errno? */ + rc = -ETIMEDOUT; + } else if (desc->bd_type == BULK_GET_SINK) { + rc = sptlrpc_svc_unwrap_bulk(req, desc); + } + } else { + DEBUG_REQ(D_ERROR, req, "bulk %s failed: rc %d", + bulk2type(desc), rc); + } + + RETURN(rc); +} +EXPORT_SYMBOL(target_bulk_io); + +#endif /* HAVE_SERVER_SUPPORT */