1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #define DEBUG_SUBSYSTEM S_RPC
25 #include <liblustre.h>
26 #include <portals/lib-types.h>
28 #include <linux/obd_support.h>
29 #include <linux/lustre_net.h>
30 #include <linux/lustre_lib.h>
31 #include <linux/obd.h>
32 #include "ptlrpc_internal.h"
34 static int ptl_send_buf (ptl_handle_md_t *mdh, void *base, int len,
35 ptl_ack_req_t ack, struct ptlrpc_cb_id *cbid,
36 struct ptlrpc_connection *conn, int portal, __u64 xid)
38 ptl_process_id_t remote_id;
42 char str[PTL_NALFMT_SIZE];
45 LASSERT (portal != 0);
46 LASSERT (conn != NULL);
47 CDEBUG (D_INFO, "conn=%p ni %s nid "LPX64" (%s) on %s\n",
48 conn, conn->c_peer.peer_ni->pni_name,
49 conn->c_peer.peer_nid,
50 portals_nid2str(conn->c_peer.peer_ni->pni_number,
51 conn->c_peer.peer_nid, str),
52 conn->c_peer.peer_ni->pni_name);
54 remote_id.nid = conn->c_peer.peer_nid,
59 md.threshold = (ack == PTL_ACK_REQ) ? 2 : 1;
62 md.eventq = conn->c_peer.peer_ni->pni_eq_h;
64 if (ack == PTL_ACK_REQ &&
65 OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_ACK | OBD_FAIL_ONCE)) {
66 /* don't ask for the ack to simulate failing client */
68 obd_fail_loc |= OBD_FAIL_ONCE | OBD_FAILED;
71 rc = PtlMDBind (conn->c_peer.peer_ni->pni_ni_h, md, mdh);
73 CERROR ("PtlMDBind failed: %d\n", rc);
74 LASSERT (rc == PTL_NOSPACE);
78 CDEBUG(D_NET, "Sending %d bytes to portal %d, xid "LPD64"\n",
81 rc2 = PtlPut (*mdh, ack, remote_id, portal, 0, xid, 0, 0);
83 /* We're going to get an UNLINK event when I unlink below,
84 * which will complete just like any other failed send, so
85 * I fall through and return success here! */
86 CERROR("PtlPut("LPU64", %d, "LPD64") failed: %d\n",
87 remote_id.nid, portal, xid, rc);
88 rc2 = PtlMDUnlink(*mdh);
89 LASSERT (rc2 == PTL_OK);
95 int ptlrpc_start_bulk_transfer (struct ptlrpc_bulk_desc *desc)
99 struct ptlrpc_peer *peer;
100 ptl_process_id_t remote_id;
105 /* NB no locking required until desc is on the network */
106 LASSERT (!desc->bd_network_rw);
107 LASSERT (desc->bd_type == BULK_PUT_SOURCE ||
108 desc->bd_type == BULK_GET_SINK);
109 desc->bd_success = 0;
110 peer = &desc->bd_export->exp_connection->c_peer;
112 md.start = &desc->bd_iov[0];
113 md.niov = desc->bd_page_count;
114 md.length = desc->bd_nob;
115 md.eventq = peer->peer_ni->pni_eq_h;
116 md.threshold = 2; /* SENT and ACK/REPLY */
118 md.options = PTL_MD_KIOV;
120 md.options = PTL_MD_IOV;
122 md.user_ptr = &desc->bd_cbid;
123 LASSERT (desc->bd_cbid.cbid_fn == server_bulk_callback);
124 LASSERT (desc->bd_cbid.cbid_arg == desc);
126 /* NB total length may be 0 for a read past EOF, so we send a 0
127 * length bulk, since the client expects a bulk event. */
129 rc = PtlMDBind(peer->peer_ni->pni_ni_h, md, &desc->bd_md_h);
131 CERROR("PtlMDBind failed: %d\n", rc);
132 LASSERT (rc == PTL_NOSPACE);
136 /* Client's bulk and reply matchbits are the same */
137 xid = desc->bd_req->rq_xid;
138 remote_id.nid = peer->peer_nid;
141 CDEBUG(D_NET, "Transferring %u pages %u bytes via portal %d on %s "
142 "nid "LPX64" pid %d xid "LPX64"\n",
143 md.niov, md.length, desc->bd_portal, peer->peer_ni->pni_name,
144 remote_id.nid, remote_id.pid, xid);
146 /* Network is about to get at the memory */
147 desc->bd_network_rw = 1;
149 if (desc->bd_type == BULK_PUT_SOURCE)
150 rc = PtlPut (desc->bd_md_h, PTL_ACK_REQ, remote_id,
151 desc->bd_portal, 0, xid, 0, 0);
153 rc = PtlGet (desc->bd_md_h, remote_id,
154 desc->bd_portal, 0, xid, 0);
157 /* Can't send, so we unlink the MD bound above. The UNLINK
158 * event this creates will signal completion with failure,
159 * so we return SUCCESS here! */
160 CERROR("Transfer("LPU64", %d, "LPX64") failed: %d\n",
161 remote_id.nid, desc->bd_portal, xid, rc);
162 rc2 = PtlMDUnlink(desc->bd_md_h);
163 LASSERT (rc2 == PTL_OK);
169 void ptlrpc_abort_bulk (struct ptlrpc_bulk_desc *desc)
171 /* Server side bulk abort. Idempotent. Not thread-safe (i.e. only
172 * serialises with completion callback) */
173 struct l_wait_info lwi;
176 LASSERT (!in_interrupt ()); /* might sleep */
178 if (!ptlrpc_bulk_active(desc)) /* completed or */
179 return; /* never started */
181 /* The unlink ensures the callback happens ASAP and is the last
182 * one. If it fails, it must be because completion just
185 rc = PtlMDUnlink (desc->bd_md_h);
186 if (rc == PTL_INV_MD) {
187 LASSERT(!ptlrpc_bulk_active(desc));
191 LASSERT (rc == PTL_OK);
194 /* Network access will complete in finite time but the HUGE
195 * timeout lets us CWARN for visibility of sluggish NALs */
196 lwi = LWI_TIMEOUT (300 * HZ, NULL, NULL);
197 rc = l_wait_event(desc->bd_waitq,
198 !ptlrpc_bulk_active(desc), &lwi);
202 LASSERT(rc == -ETIMEDOUT);
203 CWARN("Unexpectedly long timeout: desc %p\n", desc);
207 int ptlrpc_register_bulk (struct ptlrpc_request *req)
209 struct ptlrpc_bulk_desc *desc = req->rq_bulk;
210 struct ptlrpc_peer *peer;
213 ptl_process_id_t source_id;
214 ptl_handle_me_t me_h;
218 /* NB no locking required until desc is on the network */
219 LASSERT (desc->bd_nob > 0);
220 LASSERT (!desc->bd_network_rw);
221 LASSERT (desc->bd_page_count <= PTL_MD_MAX_PAGES);
222 LASSERT (desc->bd_req != NULL);
223 LASSERT (desc->bd_type == BULK_PUT_SINK ||
224 desc->bd_type == BULK_GET_SOURCE);
226 desc->bd_success = 0;
228 peer = &desc->bd_import->imp_connection->c_peer;
230 md.start = &desc->bd_iov[0];
231 md.niov = desc->bd_page_count;
232 md.length = desc->bd_nob;
233 md.eventq = peer->peer_ni->pni_eq_h;
234 md.threshold = 1; /* PUT or GET */
235 md.options = (desc->bd_type == BULK_GET_SOURCE) ?
236 PTL_MD_OP_GET : PTL_MD_OP_PUT;
238 md.options |= PTL_MD_KIOV;
240 md.options |= PTL_MD_IOV;
242 md.user_ptr = &desc->bd_cbid;
243 LASSERT (desc->bd_cbid.cbid_fn == client_bulk_callback);
244 LASSERT (desc->bd_cbid.cbid_arg == desc);
246 /* XXX Registering the same xid on retried bulk makes my head
247 * explode trying to understand how the original request's bulk
248 * might interfere with the retried request -eeb */
249 LASSERT (!desc->bd_registered || req->rq_xid != desc->bd_last_xid);
250 desc->bd_registered = 1;
251 desc->bd_last_xid = req->rq_xid;
253 source_id.nid = desc->bd_import->imp_connection->c_peer.peer_nid;
254 source_id.pid = PTL_PID_ANY;
256 rc = PtlMEAttach(peer->peer_ni->pni_ni_h,
257 desc->bd_portal, source_id, req->rq_xid, 0,
258 PTL_UNLINK, PTL_INS_AFTER, &me_h);
260 CERROR("PtlMEAttach failed: %d\n", rc);
261 LASSERT (rc == PTL_NOSPACE);
265 /* About to let the network at it... */
266 desc->bd_network_rw = 1;
267 rc = PtlMDAttach(me_h, md, PTL_UNLINK, &desc->bd_md_h);
269 CERROR("PtlMDAttach failed: %d\n", rc);
270 LASSERT (rc == PTL_NOSPACE);
271 desc->bd_network_rw = 0;
272 rc2 = PtlMEUnlink (me_h);
273 LASSERT (rc2 == PTL_OK);
277 CDEBUG(D_NET, "Setup bulk %s buffers: %u pages %u bytes, xid "LPX64", "
279 desc->bd_type == BULK_GET_SOURCE ? "get-source" : "put-sink",
281 req->rq_xid, desc->bd_portal, peer->peer_ni->pni_name);
285 void ptlrpc_unregister_bulk (struct ptlrpc_request *req)
287 /* Disconnect a bulk desc from the network. Idempotent. Not
288 * thread-safe (i.e. only interlocks with completion callback). */
289 struct ptlrpc_bulk_desc *desc = req->rq_bulk;
290 wait_queue_head_t *wq;
291 struct l_wait_info lwi;
294 LASSERT (!in_interrupt ()); /* might sleep */
296 if (!ptlrpc_bulk_active(desc)) /* completed or */
297 return; /* never registered */
299 LASSERT (desc->bd_req == req); /* bd_req NULL until registered */
301 /* the unlink ensures the callback happens ASAP and is the last
302 * one. If it fails, it must be because completion just
305 rc = PtlMDUnlink (desc->bd_md_h);
306 if (rc == PTL_INV_MD) {
307 LASSERT(!ptlrpc_bulk_active(desc));
311 LASSERT (rc == PTL_OK);
313 if (desc->bd_req->rq_set != NULL)
314 wq = &req->rq_set->set_waitq;
316 wq = &req->rq_reply_waitq;
319 /* Network access will complete in finite time but the HUGE
320 * timeout lets us CWARN for visibility of sluggish NALs */
321 lwi = LWI_TIMEOUT (300 * HZ, NULL, NULL);
322 rc = l_wait_event(*wq, !ptlrpc_bulk_active(desc), &lwi);
326 LASSERT (rc == -ETIMEDOUT);
327 CWARN("Unexpectedly long timeout: desc %p\n", desc);
331 int ptlrpc_send_reply (struct ptlrpc_request *req, int may_be_difficult)
333 struct ptlrpc_service *svc = req->rq_rqbd->rqbd_srv_ni->sni_service;
334 struct ptlrpc_reply_state *rs = req->rq_reply_state;
335 struct ptlrpc_connection *conn;
338 /* We must already have a reply buffer (only ptlrpc_error() may be
339 * called without one). We must also have a request buffer which
340 * is either the actual (swabbed) incoming request, or a saved copy
341 * if this is a req saved in target_queue_final_reply(). */
342 LASSERT (req->rq_reqmsg != NULL);
343 LASSERT (rs != NULL);
344 LASSERT (req->rq_repmsg != NULL);
345 LASSERT (may_be_difficult || !rs->rs_difficult);
346 LASSERT (req->rq_repmsg == &rs->rs_msg);
347 LASSERT (rs->rs_cb_id.cbid_fn == reply_out_callback);
348 LASSERT (rs->rs_cb_id.cbid_arg == rs);
350 LASSERT (req->rq_repmsg != NULL);
351 if (req->rq_type != PTL_RPC_MSG_ERR)
352 req->rq_type = PTL_RPC_MSG_REPLY;
354 req->rq_repmsg->type = req->rq_type;
355 req->rq_repmsg->status = req->rq_status;
356 req->rq_repmsg->opc = req->rq_reqmsg->opc;
358 if (req->rq_export == NULL)
359 conn = ptlrpc_get_connection(&req->rq_peer, NULL);
361 conn = ptlrpc_connection_addref(req->rq_export->exp_connection);
363 atomic_inc (&svc->srv_outstanding_replies);
365 rc = ptl_send_buf (&rs->rs_md_h, req->rq_repmsg, req->rq_replen,
366 rs->rs_difficult ? PTL_ACK_REQ : PTL_NOACK_REQ,
368 svc->srv_rep_portal, req->rq_xid);
370 atomic_dec (&svc->srv_outstanding_replies);
372 if (!rs->rs_difficult) {
373 /* Callers other than target_send_reply() expect me
374 * to clean up on a comms error */
375 lustre_free_reply_state (rs);
376 req->rq_reply_state = NULL;
377 req->rq_repmsg = NULL;
380 ptlrpc_put_connection(conn);
384 int ptlrpc_reply (struct ptlrpc_request *req)
386 return (ptlrpc_send_reply (req, 0));
389 int ptlrpc_error(struct ptlrpc_request *req)
394 if (!req->rq_repmsg) {
395 rc = lustre_pack_reply(req, 0, NULL, NULL);
400 req->rq_type = PTL_RPC_MSG_ERR;
402 rc = ptlrpc_send_reply (req, 0);
406 int ptl_send_rpc(struct ptlrpc_request *request)
410 struct ptlrpc_connection *connection;
412 ptl_process_id_t source_id;
413 ptl_handle_me_t reply_me_h;
417 LASSERT (request->rq_type == PTL_RPC_MSG_REQUEST);
419 /* If this is a re-transmit, we're required to have disengaged
420 * cleanly from the previous attempt */
421 LASSERT (!request->rq_receiving_reply);
423 connection = request->rq_import->imp_connection;
425 if (request->rq_bulk != NULL) {
426 rc = ptlrpc_register_bulk (request);
431 request->rq_reqmsg->handle = request->rq_import->imp_remote_handle;
432 request->rq_reqmsg->type = PTL_RPC_MSG_REQUEST;
433 request->rq_reqmsg->conn_cnt = request->rq_import->imp_conn_cnt;
435 source_id.nid = connection->c_peer.peer_nid;
436 source_id.pid = PTL_PID_ANY;
438 LASSERT (request->rq_replen != 0);
439 if (request->rq_repmsg == NULL)
440 OBD_ALLOC(request->rq_repmsg, request->rq_replen);
441 if (request->rq_repmsg == NULL)
442 GOTO(cleanup_bulk, rc = -ENOMEM);
444 rc = PtlMEAttach(connection->c_peer.peer_ni->pni_ni_h,
445 request->rq_reply_portal, /* XXX FIXME bug 249 */
446 source_id, request->rq_xid, 0, PTL_UNLINK,
447 PTL_INS_AFTER, &reply_me_h);
449 CERROR("PtlMEAttach failed: %d\n", rc);
450 LASSERT (rc == PTL_NOSPACE);
451 GOTO(cleanup_repmsg, rc = -ENOMEM);
454 spin_lock_irqsave (&request->rq_lock, flags);
455 /* If the MD attach succeeds, there _will_ be a reply_in callback */
456 request->rq_receiving_reply = 1;
457 /* Clear any flags that may be present from previous sends. */
458 request->rq_replied = 0;
460 request->rq_timedout = 0;
461 request->rq_resend = 0;
462 request->rq_restart = 0;
463 spin_unlock_irqrestore (&request->rq_lock, flags);
465 reply_md.start = request->rq_repmsg;
466 reply_md.length = request->rq_replen;
467 reply_md.threshold = 1;
468 reply_md.options = PTL_MD_OP_PUT;
469 reply_md.user_ptr = &request->rq_reply_cbid;
470 reply_md.eventq = connection->c_peer.peer_ni->pni_eq_h;
472 rc = PtlMDAttach(reply_me_h, reply_md, PTL_UNLINK,
473 &request->rq_reply_md_h);
475 CERROR("PtlMDAttach failed: %d\n", rc);
476 LASSERT (rc == PTL_NOSPACE);
477 GOTO(cleanup_me, rc -ENOMEM);
480 CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid "LPU64
481 ", portal %u on %s\n",
482 request->rq_replen, request->rq_xid,
483 request->rq_reply_portal,
484 connection->c_peer.peer_ni->pni_name);
486 ptlrpc_request_addref(request); /* +1 ref for the SENT callback */
488 request->rq_sent = LTIME_S(CURRENT_TIME);
489 ptlrpc_pinger_sending_on_import(request->rq_import);
490 rc = ptl_send_buf(&request->rq_req_md_h,
491 request->rq_reqmsg, request->rq_reqlen,
492 PTL_NOACK_REQ, &request->rq_req_cbid,
494 request->rq_request_portal,
497 ptlrpc_lprocfs_rpc_sent(request);
501 ptlrpc_req_finished (request); /* drop callback ref */
504 /* MEUnlink is safe; the PUT didn't even get off the ground, and
505 * nobody apart from the PUT's target has the right nid+XID to
506 * access the reply buffer. */
507 rc2 = PtlMEUnlink(reply_me_h);
508 LASSERT (rc2 == PTL_OK);
509 /* UNLINKED callback called synchronously */
510 LASSERT (!request->rq_receiving_reply);
513 OBD_FREE(request->rq_repmsg, request->rq_replen);
514 request->rq_repmsg = NULL;
517 if (request->rq_bulk != NULL)
518 ptlrpc_unregister_bulk(request);
523 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd)
525 struct ptlrpc_srv_ni *srv_ni = rqbd->rqbd_srv_ni;
526 struct ptlrpc_service *service = srv_ni->sni_service;
527 static ptl_process_id_t match_id = {PTL_NID_ANY, PTL_PID_ANY};
530 ptl_handle_me_t me_h;
532 CDEBUG(D_NET, "PtlMEAttach: portal %d on %s h %lx."LPX64"\n",
533 service->srv_req_portal, srv_ni->sni_ni->pni_name,
534 srv_ni->sni_ni->pni_ni_h.nal_idx,
535 srv_ni->sni_ni->pni_ni_h.cookie);
537 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_PTLRPC_RQBD))
540 rc = PtlMEAttach(srv_ni->sni_ni->pni_ni_h, service->srv_req_portal,
541 match_id, 0, ~0, PTL_UNLINK, PTL_INS_AFTER, &me_h);
543 CERROR("PtlMEAttach failed: %d\n", rc);
547 LASSERT(rqbd->rqbd_refcount == 0);
548 rqbd->rqbd_refcount = 1;
550 md.start = rqbd->rqbd_buffer;
551 md.length = service->srv_buf_size;
552 md.max_size = service->srv_max_req_size;
553 md.threshold = PTL_MD_THRESH_INF;
554 md.options = PTL_MD_OP_PUT | PTL_MD_MAX_SIZE | PTL_MD_AUTO_UNLINK;
555 md.user_ptr = &rqbd->rqbd_cbid;
556 md.eventq = srv_ni->sni_ni->pni_eq_h;
558 rc = PtlMDAttach(me_h, md, PTL_UNLINK, &rqbd->rqbd_md_h);
562 CERROR("PtlMDAttach failed: %d; \n", rc);
563 LASSERT (rc == PTL_NOSPACE);
564 rc = PtlMEUnlink (me_h);
565 LASSERT (rc == PTL_OK);
566 rqbd->rqbd_refcount = 0;