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>
33 static int ptl_send_buf(struct ptlrpc_request *request,
34 struct ptlrpc_connection *conn, int portal)
37 ptl_process_id_t remote_id;
39 ptl_ack_req_t ack_req;
42 CDEBUG (D_INFO, "conn=%p ni %s nid "LPX64" on %s\n",
43 conn, conn->c_peer.peer_ni->pni_name,
44 conn->c_peer.peer_nid, conn->c_peer.peer_ni->pni_name);
46 request->rq_req_md.user_ptr = request;
48 switch (request->rq_type) {
49 case PTL_RPC_MSG_REQUEST:
50 request->rq_reqmsg->type = HTON__u32(request->rq_type);
51 request->rq_req_md.start = request->rq_reqmsg;
52 request->rq_req_md.length = request->rq_reqlen;
53 request->rq_req_md.eventq = conn->c_peer.peer_ni->pni_request_out_eq_h;
56 case PTL_RPC_MSG_REPLY:
57 request->rq_repmsg->type = HTON__u32(request->rq_type);
58 request->rq_req_md.start = request->rq_repmsg;
59 request->rq_req_md.length = request->rq_replen;
60 request->rq_req_md.eventq = conn->c_peer.peer_ni->pni_reply_out_eq_h;
64 return -1; /* notreached */
66 if (request->rq_flags & PTL_RPC_FL_WANT_ACK) {
67 request->rq_req_md.threshold = 2; /* SENT and ACK */
68 ack_req = PTL_ACK_REQ;
70 request->rq_req_md.threshold = 1;
71 ack_req = PTL_NOACK_REQ;
73 request->rq_req_md.options = PTL_MD_OP_PUT;
74 request->rq_req_md.user_ptr = request;
76 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_ACK | OBD_FAIL_ONCE)) {
77 request->rq_req_md.options |= PTL_MD_ACK_DISABLE;
78 obd_fail_loc |= OBD_FAIL_ONCE | OBD_FAILED;
81 rc = PtlMDBind(conn->c_peer.peer_ni->pni_ni_h, request->rq_req_md, &md_h);
83 CERROR("PtlMDBind failed: %d\n", rc);
88 remote_id.nid = conn->c_peer.peer_nid;
91 CDEBUG(D_NET, "Sending %d bytes to portal %d, xid "LPD64"\n",
92 request->rq_req_md.length, portal, request->rq_xid);
96 rc = PtlPut(md_h, ack_req, remote_id, portal, 0, request->rq_xid, 0, 0);
98 CERROR("PtlPut("LPU64", %d, "LPD64") failed: %d\n",
99 remote_id.nid, portal, request->rq_xid, rc);
106 static inline struct iovec *
107 ptlrpc_get_bulk_iov (struct ptlrpc_bulk_desc *desc)
111 if (desc->bd_page_count <= sizeof (desc->bd_iov)/sizeof (struct iovec))
112 return (desc->bd_iov);
114 OBD_ALLOC (iov, desc->bd_page_count * sizeof (struct iovec));
122 ptlrpc_put_bulk_iov (struct ptlrpc_bulk_desc *desc, struct iovec *iov)
124 if (desc->bd_page_count <= sizeof (desc->bd_iov)/sizeof (struct iovec))
127 OBD_FREE (iov, desc->bd_page_count * sizeof (struct iovec));
130 int ptlrpc_bulk_put(struct ptlrpc_bulk_desc *desc)
133 struct ptlrpc_peer *peer;
134 struct list_head *tmp, *next;
135 ptl_process_id_t remote_id;
140 iov = ptlrpc_get_bulk_iov (desc);
144 peer = &desc->bd_connection->c_peer;
146 desc->bd_md.start = iov;
147 desc->bd_md.niov = 0;
148 desc->bd_md.length = 0;
149 desc->bd_md.eventq = peer->peer_ni->pni_bulk_put_source_eq_h;
150 desc->bd_md.threshold = 2; /* SENT and ACK */
151 desc->bd_md.options = PTL_MD_OP_PUT | PTL_MD_IOV;
152 desc->bd_md.user_ptr = desc;
154 atomic_set(&desc->bd_source_callback_count, 2);
156 list_for_each_safe(tmp, next, &desc->bd_page_list) {
157 struct ptlrpc_bulk_page *bulk;
158 bulk = list_entry(tmp, struct ptlrpc_bulk_page, bp_link);
160 LASSERT(desc->bd_md.niov < desc->bd_page_count);
162 if (desc->bd_md.niov == 0)
164 LASSERT(xid == bulk->bp_xid); /* should all be the same */
166 iov[desc->bd_md.niov].iov_base = bulk->bp_buf;
167 iov[desc->bd_md.niov].iov_len = bulk->bp_buflen;
168 if (iov[desc->bd_md.niov].iov_len <= 0) {
169 CERROR("bad bp_buflen[%d] @ %p: %d\n", desc->bd_md.niov,
170 bulk->bp_buf, bulk->bp_buflen);
171 CERROR("desc: xid %u, pages %d, ptl %d, ref %d\n",
172 xid, desc->bd_page_count, desc->bd_portal,
173 atomic_read(&desc->bd_refcount));
177 desc->bd_md.length += bulk->bp_buflen;
180 LASSERT(desc->bd_md.niov == desc->bd_page_count);
181 LASSERT(desc->bd_md.niov != 0);
183 rc = PtlMDBind(peer->peer_ni->pni_ni_h, desc->bd_md,
186 ptlrpc_put_bulk_iov (desc, iov); /*move down to reduce latency to send*/
189 CERROR("PtlMDBind failed: %d\n", rc);
194 remote_id.nid = peer->peer_nid;
197 CDEBUG(D_NET, "Sending %u pages %u bytes to portal %d on %s "
198 "nid "LPX64" pid %d xid %d\n",
199 desc->bd_md.niov, desc->bd_md.length,
200 desc->bd_portal, peer->peer_ni->pni_name,
201 remote_id.nid, remote_id.pid, xid);
203 rc = PtlPut(desc->bd_md_h, PTL_ACK_REQ, remote_id,
204 desc->bd_portal, 0, xid, 0, 0);
206 CERROR("PtlPut("LPU64", %d, %d) failed: %d\n",
207 remote_id.nid, desc->bd_portal, xid, rc);
208 PtlMDUnlink(desc->bd_md_h);
216 int ptlrpc_bulk_get(struct ptlrpc_bulk_desc *desc)
219 struct ptlrpc_peer *peer;
220 struct list_head *tmp, *next;
221 ptl_process_id_t remote_id;
226 iov = ptlrpc_get_bulk_iov (desc);
230 peer = &desc->bd_connection->c_peer;
232 desc->bd_md.start = iov;
233 desc->bd_md.niov = 0;
234 desc->bd_md.length = 0;
235 desc->bd_md.eventq = peer->peer_ni->pni_bulk_get_sink_eq_h;
236 desc->bd_md.threshold = 2; /* SENT and REPLY */
237 desc->bd_md.options = PTL_MD_OP_GET | PTL_MD_IOV;
238 desc->bd_md.user_ptr = desc;
240 atomic_set(&desc->bd_source_callback_count, 2);
242 list_for_each_safe(tmp, next, &desc->bd_page_list) {
243 struct ptlrpc_bulk_page *bulk;
244 bulk = list_entry(tmp, struct ptlrpc_bulk_page, bp_link);
246 LASSERT(desc->bd_md.niov < desc->bd_page_count);
248 if (desc->bd_md.niov == 0)
250 LASSERT(xid == bulk->bp_xid); /* should all be the same */
252 iov[desc->bd_md.niov].iov_base = bulk->bp_buf;
253 iov[desc->bd_md.niov].iov_len = bulk->bp_buflen;
254 if (iov[desc->bd_md.niov].iov_len <= 0) {
255 CERROR("bad bulk %p bp_buflen[%d] @ %p: %d\n", bulk,
256 desc->bd_md.niov, bulk->bp_buf, bulk->bp_buflen);
257 CERROR("desc %p: xid %u, pages %d, ptl %d, ref %d\n",
258 desc, xid, desc->bd_page_count, desc->bd_portal,
259 atomic_read(&desc->bd_refcount));
263 desc->bd_md.length += bulk->bp_buflen;
266 LASSERT(desc->bd_md.niov == desc->bd_page_count);
267 LASSERT(desc->bd_md.niov != 0);
269 rc = PtlMDBind(peer->peer_ni->pni_ni_h, desc->bd_md,
272 ptlrpc_put_bulk_iov (desc, iov); /*move down to reduce latency to send*/
275 CERROR("PtlMDBind failed: %d\n", rc);
280 remote_id.nid = desc->bd_connection->c_peer.peer_nid;
283 CDEBUG(D_NET, "Sending %u pages %u bytes to portal %d on %s "
284 "nid "LPX64" pid %d xid %d\n",
285 desc->bd_md.niov, desc->bd_md.length,
286 desc->bd_portal, peer->peer_ni->pni_name,
287 remote_id.nid, remote_id.pid, xid);
289 rc = PtlGet(desc->bd_md_h, remote_id, desc->bd_portal, 0, xid, 0);
291 CERROR("PtlGet("LPU64", %d, %d) failed: %d\n",
292 remote_id.nid, desc->bd_portal, xid, rc);
293 PtlMDUnlink(desc->bd_md_h);
301 static int ptlrpc_register_bulk_shared(struct ptlrpc_bulk_desc *desc)
303 struct ptlrpc_peer *peer;
304 struct list_head *tmp, *next;
308 ptl_process_id_t source_id;
311 if (desc->bd_page_count > PTL_MD_MAX_IOV) {
312 CERROR("iov longer than %d pages not supported (count=%d)\n",
313 PTL_MD_MAX_IOV, desc->bd_page_count);
317 iov = ptlrpc_get_bulk_iov (desc);
321 peer = &desc->bd_connection->c_peer;
323 desc->bd_md.start = iov;
324 desc->bd_md.niov = 0;
325 desc->bd_md.length = 0;
326 desc->bd_md.threshold = 1;
327 desc->bd_md.user_ptr = desc;
329 list_for_each_safe(tmp, next, &desc->bd_page_list) {
330 struct ptlrpc_bulk_page *bulk;
331 bulk = list_entry(tmp, struct ptlrpc_bulk_page, bp_link);
333 LASSERT(desc->bd_md.niov < desc->bd_page_count);
335 if (desc->bd_md.niov == 0)
337 LASSERT(xid == bulk->bp_xid); /* should all be the same */
339 iov[desc->bd_md.niov].iov_base = bulk->bp_buf;
340 iov[desc->bd_md.niov].iov_len = bulk->bp_buflen;
342 desc->bd_md.length += bulk->bp_buflen;
345 LASSERT(desc->bd_md.niov == desc->bd_page_count);
346 LASSERT(desc->bd_md.niov != 0);
348 source_id.nid = desc->bd_connection->c_peer.peer_nid;
349 source_id.pid = PTL_PID_ANY;
351 rc = PtlMEAttach(peer->peer_ni->pni_ni_h,
352 desc->bd_portal, source_id, xid, 0,
353 PTL_UNLINK, PTL_INS_AFTER, &desc->bd_me_h);
356 CERROR("PtlMEAttach failed: %d\n", rc);
361 rc = PtlMDAttach(desc->bd_me_h, desc->bd_md, PTL_UNLINK,
364 CERROR("PtlMDAttach failed: %d\n", rc);
369 ptlrpc_put_bulk_iov (desc, iov);
371 CDEBUG(D_NET, "Setup bulk sink buffers: %u pages %u bytes, xid %u, "
372 "portal %u on %s\n", desc->bd_md.niov, desc->bd_md.length,
373 xid, desc->bd_portal, peer->peer_ni->pni_name);
378 ptlrpc_put_bulk_iov (desc, iov);
379 ptlrpc_abort_bulk(desc);
384 int ptlrpc_register_bulk_get(struct ptlrpc_bulk_desc *desc)
386 desc->bd_md.options = PTL_MD_OP_GET | PTL_MD_IOV;
388 desc->bd_connection->c_peer.peer_ni->pni_bulk_get_source_eq_h;
390 return ptlrpc_register_bulk_shared(desc);
393 int ptlrpc_register_bulk_put(struct ptlrpc_bulk_desc *desc)
395 desc->bd_md.options = PTL_MD_OP_PUT | PTL_MD_IOV;
397 desc->bd_connection->c_peer.peer_ni->pni_bulk_put_sink_eq_h;
399 return ptlrpc_register_bulk_shared(desc);
402 int ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc)
405 /* This should be safe: these handles are initialized to be
406 * invalid in ptlrpc_prep_bulk() */
407 rc1 = PtlMDUnlink(desc->bd_md_h);
409 CERROR("PtlMDUnlink: %d\n", rc1);
410 rc2 = PtlMEUnlink(desc->bd_me_h);
412 CERROR("PtlMEUnlink: %d\n", rc2);
414 return rc1 ? rc1 : rc2;
417 void obd_brw_set_addref(struct obd_brw_set *set)
419 atomic_inc(&set->brw_refcount);
422 void obd_brw_set_add(struct obd_brw_set *set, struct ptlrpc_bulk_desc *desc)
424 LASSERT(list_empty(&desc->bd_set_chain));
426 ptlrpc_bulk_addref(desc);
427 atomic_inc(&set->brw_desc_count);
428 desc->bd_brw_set = set;
429 list_add(&desc->bd_set_chain, &set->brw_desc_head);
432 void obd_brw_set_del(struct ptlrpc_bulk_desc *desc)
434 atomic_dec(&desc->bd_brw_set->brw_desc_count);
435 list_del_init(&desc->bd_set_chain);
436 ptlrpc_bulk_decref(desc);
439 struct obd_brw_set *obd_brw_set_new(void)
441 struct obd_brw_set *set;
443 OBD_ALLOC(set, sizeof(*set));
446 init_waitqueue_head(&set->brw_waitq);
447 INIT_LIST_HEAD(&set->brw_desc_head);
448 atomic_set(&set->brw_refcount, 1);
449 atomic_set(&set->brw_desc_count, 0);
455 static void obd_brw_set_free(struct obd_brw_set *set)
457 struct list_head *tmp, *next;
460 list_for_each_safe(tmp, next, &set->brw_desc_head) {
461 struct ptlrpc_bulk_desc *desc =
462 list_entry(tmp, struct ptlrpc_bulk_desc, bd_set_chain);
464 CERROR("Unfinished bulk descriptor: %p\n", desc);
466 ptlrpc_abort_bulk(desc);
468 OBD_FREE(set, sizeof(*set));
473 void obd_brw_set_decref(struct obd_brw_set *set)
476 if (atomic_dec_and_test(&set->brw_refcount))
477 obd_brw_set_free(set);
481 int ptlrpc_reply(struct ptlrpc_service *svc, struct ptlrpc_request *req)
483 if (req->rq_repmsg == NULL) {
484 CERROR("bad: someone called ptlrpc_reply when they meant "
489 /* FIXME: we need to increment the count of handled events */
490 if (req->rq_type != PTL_RPC_MSG_ERR)
491 req->rq_type = PTL_RPC_MSG_REPLY;
492 //req->rq_repmsg->conn = req->rq_connection->c_remote_conn;
493 //req->rq_repmsg->token = req->rq_connection->c_remote_token;
494 req->rq_repmsg->status = HTON__u32(req->rq_status);
495 return ptl_send_buf(req, req->rq_connection, svc->srv_rep_portal);
498 int ptlrpc_error(struct ptlrpc_service *svc, struct ptlrpc_request *req)
503 if (!req->rq_repmsg) {
504 rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen,
511 req->rq_type = PTL_RPC_MSG_ERR;
513 rc = ptlrpc_reply(svc, req);
517 int ptl_send_rpc(struct ptlrpc_request *request)
521 ptl_process_id_t source_id;
525 if (request->rq_type != PTL_RPC_MSG_REQUEST) {
526 CERROR("wrong packet type sent %d\n",
527 NTOH__u32(request->rq_reqmsg->type));
532 source_id.nid = request->rq_connection->c_peer.peer_nid;
533 source_id.pid = PTL_PID_ANY;
535 /* add a ref, which will be balanced in request_out_callback */
536 ptlrpc_request_addref(request);
537 if (request->rq_replen != 0) {
538 if (request->rq_reply_md.start != NULL) {
539 rc = PtlMEUnlink(request->rq_reply_me_h);
540 if (rc != PTL_OK && rc != PTL_INV_ME) {
541 CERROR("rc %d\n", rc);
544 repbuf = (char *)request->rq_reply_md.start;
545 request->rq_repmsg = NULL;
547 OBD_ALLOC(repbuf, request->rq_replen);
554 rc = PtlMEAttach(request->rq_connection->c_peer.peer_ni->pni_ni_h,
555 request->rq_reply_portal,/* XXX FIXME bug 625069 */
556 source_id, request->rq_xid, 0, PTL_UNLINK,
557 PTL_INS_AFTER, &request->rq_reply_me_h);
559 CERROR("PtlMEAttach failed: %d\n", rc);
564 request->rq_reply_md.start = repbuf;
565 request->rq_reply_md.length = request->rq_replen;
566 request->rq_reply_md.threshold = 1;
567 request->rq_reply_md.options = PTL_MD_OP_PUT;
568 request->rq_reply_md.user_ptr = request;
569 request->rq_reply_md.eventq =
570 request->rq_connection->c_peer.peer_ni->pni_reply_in_eq_h;
572 rc = PtlMDAttach(request->rq_reply_me_h, request->rq_reply_md,
575 CERROR("PtlMDAttach failed: %d\n", rc);
580 CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid "LPU64
581 ", portal %u on %s\n",
582 request->rq_replen, request->rq_xid,
583 request->rq_reply_portal,
584 request->rq_connection->c_peer.peer_ni->pni_name);
587 /* Clear any flags that may be present from previous sends,
588 * except for REPLAY, NO_RESEND and WANT_ACK. */
589 request->rq_flags &= (PTL_RPC_FL_REPLAY | PTL_RPC_FL_NO_RESEND |
590 PTL_RPC_FL_WANT_ACK);
591 rc = ptl_send_buf(request, request->rq_connection,
592 request->rq_request_portal);
596 PtlMEUnlink(request->rq_reply_me_h);
598 OBD_FREE(repbuf, request->rq_replen);
599 // up(&request->rq_client->cli_rpc_sem);
604 void ptlrpc_link_svc_me(struct ptlrpc_request_buffer_desc *rqbd)
606 struct ptlrpc_srv_ni *srv_ni = rqbd->rqbd_srv_ni;
607 struct ptlrpc_service *service = srv_ni->sni_service;
608 static ptl_process_id_t match_id = {PTL_NID_ANY, PTL_PID_ANY};
611 ptl_handle_md_t md_h;
613 LASSERT(atomic_read(&rqbd->rqbd_refcount) == 0);
615 CDEBUG(D_NET, "PtlMEAttach: portal %d on %s h %lx.%lx\n",
616 service->srv_req_portal, srv_ni->sni_ni->pni_name,
617 srv_ni->sni_ni->pni_ni_h.nal_idx,
618 srv_ni->sni_ni->pni_ni_h.handle_idx);
620 /* Attach the leading ME on which we build the ring */
621 rc = PtlMEAttach(srv_ni->sni_ni->pni_ni_h, service->srv_req_portal,
623 PTL_UNLINK, PTL_INS_AFTER, &rqbd->rqbd_me_h);
625 CERROR("PtlMEAttach failed: %d\n", rc);
629 dummy.start = rqbd->rqbd_buffer;
630 dummy.length = service->srv_buf_size;
631 dummy.max_size = service->srv_max_req_size;
632 dummy.threshold = PTL_MD_THRESH_INF;
633 dummy.options = PTL_MD_OP_PUT | PTL_MD_MAX_SIZE | PTL_MD_AUTO_UNLINK;
634 dummy.user_ptr = rqbd;
635 dummy.eventq = srv_ni->sni_eq_h;
637 atomic_inc(&srv_ni->sni_nrqbds_receiving);
638 atomic_set(&rqbd->rqbd_refcount, 1); /* 1 ref for portals */
640 rc = PtlMDAttach(rqbd->rqbd_me_h, dummy, PTL_UNLINK, &md_h);
642 CERROR("PtlMDAttach failed: %d\n", rc);
644 #warning proper cleanup required
645 PtlMEUnlink (rqbd->rqbd_me_h);
646 atomic_set(&rqbd->rqbd_refcount, 0);
647 atomic_dec(&srv_ni->sni_nrqbds_receiving);