/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 only, * as published by the Free Software Foundation. * * This program 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 version 2 for more details (a copy is included * in the LICENSE file that accompanied this code). * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, * CA 95054 USA or visit www.sun.com if you need additional information or * have any questions. * * GPL HEADER END */ /* * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. */ /* * This file is part of Lustre, http://www.lustre.org/ * Lustre is a trademark of Sun Microsystems, Inc. */ #define DEBUG_SUBSYSTEM S_RPC #ifndef __KERNEL__ # include #else # include # ifdef __mips64__ # include # endif #endif #include #include #include #include "ptlrpc_internal.h" lnet_handle_eq_t ptlrpc_eq_h; /* * Client's outgoing request callback */ void request_out_callback(lnet_event_t *ev) { struct ptlrpc_cb_id *cbid = ev->md.user_ptr; struct ptlrpc_request *req = cbid->cbid_arg; ENTRY; LASSERT (ev->type == LNET_EVENT_SEND || ev->type == LNET_EVENT_UNLINK); LASSERT (ev->unlinked); DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status); sptlrpc_request_out_callback(req); req->rq_real_sent = cfs_time_current_sec(); if (ev->type == LNET_EVENT_UNLINK || ev->status != 0) { /* Failed send: make it seem like the reply timed out, just * like failing sends in client.c does currently... */ spin_lock(&req->rq_lock); req->rq_net_err = 1; spin_unlock(&req->rq_lock); ptlrpc_client_wake_req(req); } ptlrpc_req_finished(req); EXIT; } /* * Client's incoming reply callback */ void reply_in_callback(lnet_event_t *ev) { struct ptlrpc_cb_id *cbid = ev->md.user_ptr; struct ptlrpc_request *req = cbid->cbid_arg; ENTRY; DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status); LASSERT (ev->type == LNET_EVENT_PUT || ev->type == LNET_EVENT_UNLINK); LASSERT (ev->md.start == req->rq_repbuf); LASSERT (ev->offset + ev->mlength <= req->rq_repbuf_len); /* We've set LNET_MD_MANAGE_REMOTE for all outgoing requests for adaptive timeouts' early reply. */ LASSERT((ev->md.options & LNET_MD_MANAGE_REMOTE) != 0); spin_lock(&req->rq_lock); req->rq_receiving_reply = 0; req->rq_early = 0; if (ev->unlinked) req->rq_must_unlink = 0; if (ev->status) goto out_wake; if (ev->type == LNET_EVENT_UNLINK) { LASSERT(ev->unlinked); DEBUG_REQ(D_NET, req, "unlink"); goto out_wake; } if (ev->mlength < ev->rlength ) { CDEBUG(D_RPCTRACE, "truncate req %p rpc %d - %d+%d\n", req, req->rq_replen, ev->rlength, ev->offset); req->rq_reply_truncate = 1; req->rq_replied = 1; req->rq_status = -EOVERFLOW; req->rq_nob_received = ev->rlength + ev->offset; goto out_wake; } if ((ev->offset == 0) && ((lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT))) { /* Early reply */ DEBUG_REQ(D_ADAPTTO, req, "Early reply received: mlen=%u offset=%d replen=%d " "replied=%d unlinked=%d", ev->mlength, ev->offset, req->rq_replen, req->rq_replied, ev->unlinked); req->rq_early_count++; /* number received, client side */ if (req->rq_replied) /* already got the real reply */ goto out_wake; req->rq_early = 1; req->rq_reply_off = ev->offset; req->rq_nob_received = ev->mlength; /* And we're still receiving */ req->rq_receiving_reply = 1; } else { /* Real reply */ req->rq_rep_swab_mask = 0; req->rq_replied = 1; req->rq_reply_off = ev->offset; req->rq_nob_received = ev->mlength; /* LNetMDUnlink can't be called under the LNET_LOCK, so we must unlink in ptlrpc_unregister_reply */ DEBUG_REQ(D_INFO, req, "reply in flags=%x mlen=%u offset=%d replen=%d", lustre_msg_get_flags(req->rq_reqmsg), ev->mlength, ev->offset, req->rq_replen); } req->rq_import->imp_last_reply_time = cfs_time_current_sec(); out_wake: /* NB don't unlock till after wakeup; req can disappear under us * since we don't have our own ref */ ptlrpc_client_wake_req(req); spin_unlock(&req->rq_lock); EXIT; } /* * Client's bulk has been written/read */ void client_bulk_callback (lnet_event_t *ev) { struct ptlrpc_cb_id *cbid = ev->md.user_ptr; struct ptlrpc_bulk_desc *desc = cbid->cbid_arg; struct ptlrpc_request *req; ENTRY; LASSERT ((desc->bd_type == BULK_PUT_SINK && ev->type == LNET_EVENT_PUT) || (desc->bd_type == BULK_GET_SOURCE && ev->type == LNET_EVENT_GET) || ev->type == LNET_EVENT_UNLINK); LASSERT (ev->unlinked); if (CFS_FAIL_CHECK_ORSET(OBD_FAIL_PTLRPC_CLIENT_BULK_CB, CFS_FAIL_ONCE)) ev->status = -EIO; if (CFS_FAIL_CHECK_ORSET(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2,CFS_FAIL_ONCE)) ev->status = -EIO; CDEBUG((ev->status == 0) ? D_NET : D_ERROR, "event type %d, status %d, desc %p\n", ev->type, ev->status, desc); spin_lock(&desc->bd_lock); req = desc->bd_req; LASSERT(desc->bd_network_rw); desc->bd_network_rw = 0; if (ev->type != LNET_EVENT_UNLINK && ev->status == 0) { desc->bd_success = 1; desc->bd_nob_transferred = ev->mlength; desc->bd_sender = ev->sender; } else { /* start reconnect and resend if network error hit */ spin_lock(&req->rq_lock); req->rq_net_err = 1; spin_unlock(&req->rq_lock); } /* release the encrypted pages for write */ if (desc->bd_req->rq_bulk_write) sptlrpc_enc_pool_put_pages(desc); /* NB don't unlock till after wakeup; desc can disappear under us * otherwise */ ptlrpc_client_wake_req(req); spin_unlock(&desc->bd_lock); EXIT; } /* * We will have percpt request history list for ptlrpc service in upcoming * patches because we don't want to be serialized by current per-service * history operations. So we require history ID can (somehow) show arriving * order w/o grabbing global lock, and user can sort them in userspace. * * This is how we generate history ID for ptlrpc_request: * ---------------------------------------------------- * | 32 bits | 16 bits | (16 - X)bits | X bits | * ---------------------------------------------------- * | seconds | usec / 16 | sequence | CPT id | * ---------------------------------------------------- * * it might not be precise but should be good enough. */ #define REQS_CPT_BITS(svcpt) ((svcpt)->scp_service->srv_cpt_bits) #define REQS_SEC_SHIFT 32 #define REQS_USEC_SHIFT 16 #define REQS_SEQ_SHIFT(svcpt) REQS_CPT_BITS(svcpt) static void ptlrpc_req_add_history(struct ptlrpc_service_part *svcpt, struct ptlrpc_request *req) { __u64 sec = req->rq_arrival_time.tv_sec; __u32 usec = req->rq_arrival_time.tv_usec >> 4; /* usec / 16 */ __u64 new_seq; /* set sequence ID for request and add it to history list, * it must be called with hold svcpt::scp_lock */ new_seq = (sec << REQS_SEC_SHIFT) | (usec << REQS_USEC_SHIFT) | svcpt->scp_cpt; if (new_seq > svcpt->scp_hist_seq) { /* This handles the initial case of scp_hist_seq == 0 or * we just jumped into a new time window */ svcpt->scp_hist_seq = new_seq; } else { LASSERT(REQS_SEQ_SHIFT(svcpt) < REQS_USEC_SHIFT); /* NB: increase sequence number in current usec bucket, * however, it's possible that we used up all bits for * sequence and jumped into the next usec bucket (future time), * then we hope there will be less RPCs per bucket at some * point, and sequence will catch up again */ svcpt->scp_hist_seq += (1U << REQS_SEQ_SHIFT(svcpt)); new_seq = svcpt->scp_hist_seq; } req->rq_history_seq = new_seq; cfs_list_add_tail(&req->rq_history_list, &svcpt->scp_hist_reqs); } /* * Server's incoming request callback */ void request_in_callback(lnet_event_t *ev) { struct ptlrpc_cb_id *cbid = ev->md.user_ptr; struct ptlrpc_request_buffer_desc *rqbd = cbid->cbid_arg; struct ptlrpc_service_part *svcpt = rqbd->rqbd_svcpt; struct ptlrpc_service *service = svcpt->scp_service; struct ptlrpc_request *req; ENTRY; LASSERT (ev->type == LNET_EVENT_PUT || ev->type == LNET_EVENT_UNLINK); LASSERT ((char *)ev->md.start >= rqbd->rqbd_buffer); LASSERT ((char *)ev->md.start + ev->offset + ev->mlength <= rqbd->rqbd_buffer + service->srv_buf_size); CDEBUG((ev->status == 0) ? D_NET : D_ERROR, "event type %d, status %d, service %s\n", ev->type, ev->status, service->srv_name); if (ev->unlinked) { /* If this is the last request message to fit in the * request buffer we can use the request object embedded in * rqbd. Note that if we failed to allocate a request, * we'd have to re-post the rqbd, which we can't do in this * context. */ req = &rqbd->rqbd_req; memset(req, 0, sizeof (*req)); } else { LASSERT (ev->type == LNET_EVENT_PUT); if (ev->status != 0) { /* We moaned above already... */ return; } OBD_ALLOC_GFP(req, sizeof(*req), CFS_ALLOC_ATOMIC_TRY); if (req == NULL) { CERROR("Can't allocate incoming request descriptor: " "Dropping %s RPC from %s\n", service->srv_name, libcfs_id2str(ev->initiator)); return; } } /* NB we ABSOLUTELY RELY on req being zeroed, so pointers are NULL, * flags are reset and scalars are zero. We only set the message * size to non-zero if this was a successful receive. */ req->rq_xid = ev->match_bits; req->rq_reqbuf = ev->md.start + ev->offset; if (ev->type == LNET_EVENT_PUT && ev->status == 0) req->rq_reqdata_len = ev->mlength; cfs_gettimeofday(&req->rq_arrival_time); req->rq_peer = ev->initiator; req->rq_self = ev->target.nid; req->rq_rqbd = rqbd; req->rq_phase = RQ_PHASE_NEW; spin_lock_init(&req->rq_lock); CFS_INIT_LIST_HEAD(&req->rq_timed_list); CFS_INIT_LIST_HEAD(&req->rq_exp_list); cfs_atomic_set(&req->rq_refcount, 1); if (ev->type == LNET_EVENT_PUT) CDEBUG(D_INFO, "incoming req@%p x"LPU64" msgsize %u\n", req, req->rq_xid, ev->mlength); CDEBUG(D_RPCTRACE, "peer: %s\n", libcfs_id2str(req->rq_peer)); spin_lock(&svcpt->scp_lock); ptlrpc_req_add_history(svcpt, req); if (ev->unlinked) { svcpt->scp_nrqbds_posted--; CDEBUG(D_INFO, "Buffer complete: %d buffers still posted\n", svcpt->scp_nrqbds_posted); /* Normally, don't complain about 0 buffers posted; LNET won't * drop incoming reqs since we set the portal lazy */ if (test_req_buffer_pressure && ev->type != LNET_EVENT_UNLINK && svcpt->scp_nrqbds_posted == 0) CWARN("All %s request buffers busy\n", service->srv_name); /* req takes over the network's ref on rqbd */ } else { /* req takes a ref on rqbd */ rqbd->rqbd_refcount++; } cfs_list_add_tail(&req->rq_list, &svcpt->scp_req_incoming); svcpt->scp_nreqs_incoming++; /* NB everything can disappear under us once the request * has been queued and we unlock, so do the wake now... */ cfs_waitq_signal(&svcpt->scp_waitq); spin_unlock(&svcpt->scp_lock); EXIT; } /* * Server's outgoing reply callback */ void reply_out_callback(lnet_event_t *ev) { struct ptlrpc_cb_id *cbid = ev->md.user_ptr; struct ptlrpc_reply_state *rs = cbid->cbid_arg; struct ptlrpc_service_part *svcpt = rs->rs_svcpt; ENTRY; LASSERT (ev->type == LNET_EVENT_SEND || ev->type == LNET_EVENT_ACK || ev->type == LNET_EVENT_UNLINK); if (!rs->rs_difficult) { /* 'Easy' replies have no further processing so I drop the * net's ref on 'rs' */ LASSERT (ev->unlinked); ptlrpc_rs_decref(rs); EXIT; return; } LASSERT (rs->rs_on_net); if (ev->unlinked) { /* Last network callback. The net's ref on 'rs' stays put * until ptlrpc_handle_rs() is done with it */ spin_lock(&svcpt->scp_rep_lock); spin_lock(&rs->rs_lock); rs->rs_on_net = 0; if (!rs->rs_no_ack || rs->rs_transno <= rs->rs_export->exp_obd->obd_last_committed) ptlrpc_schedule_difficult_reply(rs); spin_unlock(&rs->rs_lock); spin_unlock(&svcpt->scp_rep_lock); } EXIT; } #ifdef HAVE_SERVER_SUPPORT /* * Server's bulk completion callback */ void server_bulk_callback (lnet_event_t *ev) { struct ptlrpc_cb_id *cbid = ev->md.user_ptr; struct ptlrpc_bulk_desc *desc = cbid->cbid_arg; ENTRY; LASSERT (ev->type == LNET_EVENT_SEND || ev->type == LNET_EVENT_UNLINK || (desc->bd_type == BULK_PUT_SOURCE && ev->type == LNET_EVENT_ACK) || (desc->bd_type == BULK_GET_SINK && ev->type == LNET_EVENT_REPLY)); CDEBUG((ev->status == 0) ? D_NET : D_ERROR, "event type %d, status %d, desc %p\n", ev->type, ev->status, desc); spin_lock(&desc->bd_lock); if ((ev->type == LNET_EVENT_ACK || ev->type == LNET_EVENT_REPLY) && ev->status == 0) { /* We heard back from the peer, so even if we get this * before the SENT event (oh yes we can), we know we * read/wrote the peer buffer and how much... */ desc->bd_success = 1; desc->bd_nob_transferred = ev->mlength; desc->bd_sender = ev->sender; } if (ev->unlinked) { /* This is the last callback no matter what... */ desc->bd_network_rw = 0; cfs_waitq_signal(&desc->bd_waitq); } spin_unlock(&desc->bd_lock); EXIT; } #endif static void ptlrpc_master_callback(lnet_event_t *ev) { struct ptlrpc_cb_id *cbid = ev->md.user_ptr; void (*callback)(lnet_event_t *ev) = cbid->cbid_fn; /* Honestly, it's best to find out early. */ LASSERT (cbid->cbid_arg != LP_POISON); LASSERT (callback == request_out_callback || callback == reply_in_callback || callback == client_bulk_callback || callback == request_in_callback || callback == reply_out_callback #ifdef HAVE_SERVER_SUPPORT || callback == server_bulk_callback #endif ); callback (ev); } int ptlrpc_uuid_to_peer (struct obd_uuid *uuid, lnet_process_id_t *peer, lnet_nid_t *self) { int best_dist = 0; __u32 best_order = 0; int count = 0; int rc = -ENOENT; int portals_compatibility; int dist; __u32 order; lnet_nid_t dst_nid; lnet_nid_t src_nid; portals_compatibility = LNetCtl(IOC_LIBCFS_PORTALS_COMPATIBILITY, NULL); peer->pid = LUSTRE_SRV_LNET_PID; /* Choose the matching UUID that's closest */ while (lustre_uuid_to_peer(uuid->uuid, &dst_nid, count++) == 0) { dist = LNetDist(dst_nid, &src_nid, &order); if (dist < 0) continue; if (dist == 0) { /* local! use loopback LND */ peer->nid = *self = LNET_MKNID(LNET_MKNET(LOLND, 0), 0); rc = 0; break; } if (rc < 0 || dist < best_dist || (dist == best_dist && order < best_order)) { best_dist = dist; best_order = order; if (portals_compatibility > 1) { /* Strong portals compatibility: Zero the nid's * NET, so if I'm reading new config logs, or * getting configured by (new) lconf I can * still talk to old servers. */ dst_nid = LNET_MKNID(0, LNET_NIDADDR(dst_nid)); src_nid = LNET_MKNID(0, LNET_NIDADDR(src_nid)); } peer->nid = dst_nid; *self = src_nid; rc = 0; } } CDEBUG(D_NET,"%s->%s\n", uuid->uuid, libcfs_id2str(*peer)); return rc; } void ptlrpc_ni_fini(void) { cfs_waitq_t waitq; struct l_wait_info lwi; int rc; int retries; /* Wait for the event queue to become idle since there may still be * messages in flight with pending events (i.e. the fire-and-forget * messages == client requests and "non-difficult" server * replies */ for (retries = 0;; retries++) { rc = LNetEQFree(ptlrpc_eq_h); switch (rc) { default: LBUG(); case 0: LNetNIFini(); return; case -EBUSY: if (retries != 0) CWARN("Event queue still busy\n"); /* Wait for a bit */ cfs_waitq_init(&waitq); lwi = LWI_TIMEOUT(cfs_time_seconds(2), NULL, NULL); l_wait_event(waitq, 0, &lwi); break; } } /* notreached */ } lnet_pid_t ptl_get_pid(void) { lnet_pid_t pid; #ifndef __KERNEL__ pid = getpid(); #else pid = LUSTRE_SRV_LNET_PID; #endif return pid; } int ptlrpc_ni_init(void) { int rc; lnet_pid_t pid; pid = ptl_get_pid(); CDEBUG(D_NET, "My pid is: %x\n", pid); /* We're not passing any limits yet... */ rc = LNetNIInit(pid); if (rc < 0) { CDEBUG (D_NET, "Can't init network interface: %d\n", rc); return (-ENOENT); } /* CAVEAT EMPTOR: how we process portals events is _radically_ * different depending on... */ #ifdef __KERNEL__ /* kernel LNet calls our master callback when there are new event, * because we are guaranteed to get every event via callback, * so we just set EQ size to 0 to avoid overhread of serializing * enqueue/dequeue operations in LNet. */ rc = LNetEQAlloc(0, ptlrpc_master_callback, &ptlrpc_eq_h); #else /* liblustre calls the master callback when it removes events from the * event queue. The event queue has to be big enough not to drop * anything */ rc = LNetEQAlloc(10240, LNET_EQ_HANDLER_NONE, &ptlrpc_eq_h); #endif if (rc == 0) return 0; CERROR ("Failed to allocate event queue: %d\n", rc); LNetNIFini(); return (-ENOMEM); } #ifndef __KERNEL__ CFS_LIST_HEAD(liblustre_wait_callbacks); CFS_LIST_HEAD(liblustre_idle_callbacks); void *liblustre_services_callback; void * liblustre_register_waitidle_callback (cfs_list_t *callback_list, const char *name, int (*fn)(void *arg), void *arg) { struct liblustre_wait_callback *llwc; OBD_ALLOC(llwc, sizeof(*llwc)); LASSERT (llwc != NULL); llwc->llwc_name = name; llwc->llwc_fn = fn; llwc->llwc_arg = arg; cfs_list_add_tail(&llwc->llwc_list, callback_list); return (llwc); } void liblustre_deregister_waitidle_callback (void *opaque) { struct liblustre_wait_callback *llwc = opaque; cfs_list_del(&llwc->llwc_list); OBD_FREE(llwc, sizeof(*llwc)); } void * liblustre_register_wait_callback (const char *name, int (*fn)(void *arg), void *arg) { return liblustre_register_waitidle_callback(&liblustre_wait_callbacks, name, fn, arg); } void liblustre_deregister_wait_callback (void *opaque) { liblustre_deregister_waitidle_callback(opaque); } void * liblustre_register_idle_callback (const char *name, int (*fn)(void *arg), void *arg) { return liblustre_register_waitidle_callback(&liblustre_idle_callbacks, name, fn, arg); } void liblustre_deregister_idle_callback (void *opaque) { liblustre_deregister_waitidle_callback(opaque); } int liblustre_check_events (int timeout) { lnet_event_t ev; int rc; int i; ENTRY; rc = LNetEQPoll(&ptlrpc_eq_h, 1, timeout * 1000, &ev, &i); if (rc == 0) RETURN(0); LASSERT (rc == -EOVERFLOW || rc == 1); /* liblustre: no asynch callback so we can't afford to miss any * events... */ if (rc == -EOVERFLOW) { CERROR ("Dropped an event!!!\n"); abort(); } ptlrpc_master_callback (&ev); RETURN(1); } int liblustre_waiting = 0; int liblustre_wait_event (int timeout) { cfs_list_t *tmp; struct liblustre_wait_callback *llwc; int found_something = 0; /* single threaded recursion check... */ liblustre_waiting = 1; for (;;) { /* Deal with all pending events */ while (liblustre_check_events(0)) found_something = 1; /* Give all registered callbacks a bite at the cherry */ cfs_list_for_each(tmp, &liblustre_wait_callbacks) { llwc = cfs_list_entry(tmp, struct liblustre_wait_callback, llwc_list); if (llwc->llwc_fn(llwc->llwc_arg)) found_something = 1; } if (found_something || timeout == 0) break; /* Nothing so far, but I'm allowed to block... */ found_something = liblustre_check_events(timeout); if (!found_something) /* still nothing */ break; /* I timed out */ } liblustre_waiting = 0; return found_something; } void liblustre_wait_idle(void) { static int recursed = 0; cfs_list_t *tmp; struct liblustre_wait_callback *llwc; int idle = 0; LASSERT(!recursed); recursed = 1; do { liblustre_wait_event(0); idle = 1; cfs_list_for_each(tmp, &liblustre_idle_callbacks) { llwc = cfs_list_entry(tmp, struct liblustre_wait_callback, llwc_list); if (!llwc->llwc_fn(llwc->llwc_arg)) { idle = 0; break; } } } while (!idle); recursed = 0; } #endif /* __KERNEL__ */ int ptlrpc_init_portals(void) { int rc = ptlrpc_ni_init(); if (rc != 0) { CERROR("network initialisation failed\n"); return -EIO; } #ifndef __KERNEL__ liblustre_services_callback = liblustre_register_wait_callback("liblustre_check_services", &liblustre_check_services, NULL); init_completion_module(liblustre_wait_event); #endif rc = ptlrpcd_addref(); if (rc == 0) return 0; CERROR("rpcd initialisation failed\n"); #ifndef __KERNEL__ liblustre_deregister_wait_callback(liblustre_services_callback); #endif ptlrpc_ni_fini(); return rc; } void ptlrpc_exit_portals(void) { #ifndef __KERNEL__ liblustre_deregister_wait_callback(liblustre_services_callback); #endif ptlrpcd_decref(); ptlrpc_ni_fini(); }