1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002 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 <linux/obd_support.h>
26 #include <linux/obd_class.h>
27 #include <linux/lustre_lib.h>
28 #include <linux/lustre_ha.h>
29 #include <linux/lustre_import.h>
31 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
32 struct ptlrpc_client *cl)
34 cl->cli_request_portal = req_portal;
35 cl->cli_reply_portal = rep_portal;
39 __u8 *ptlrpc_req_to_uuid(struct ptlrpc_request *req)
41 return req->rq_connection->c_remote_uuid;
44 struct ptlrpc_connection *ptlrpc_uuid_to_connection(obd_uuid_t uuid)
46 struct ptlrpc_connection *c;
47 struct lustre_peer peer;
50 err = kportal_uuid_to_peer(uuid, &peer);
52 CERROR("cannot find peer %s!\n", uuid);
56 c = ptlrpc_get_connection(&peer, uuid);
58 memcpy(c->c_remote_uuid, uuid, sizeof(c->c_remote_uuid));
62 CDEBUG(D_INFO, "%s -> %p\n", uuid, c);
67 void ptlrpc_readdress_connection(struct ptlrpc_connection *conn,obd_uuid_t uuid)
69 struct lustre_peer peer;
72 err = kportal_uuid_to_peer(uuid, &peer);
74 CERROR("cannot find peer %s!\n", uuid);
78 memcpy(&conn->c_peer, &peer, sizeof(peer));
82 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk(struct ptlrpc_connection *conn)
84 struct ptlrpc_bulk_desc *desc;
86 OBD_ALLOC(desc, sizeof(*desc));
88 desc->bd_connection = ptlrpc_connection_addref(conn);
89 atomic_set(&desc->bd_refcount, 1);
90 init_waitqueue_head(&desc->bd_waitq);
91 INIT_LIST_HEAD(&desc->bd_page_list);
92 ptl_set_inv_handle(&desc->bd_md_h);
93 ptl_set_inv_handle(&desc->bd_me_h);
99 struct ptlrpc_bulk_page *ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc)
101 struct ptlrpc_bulk_page *bulk;
103 OBD_ALLOC(bulk, sizeof(*bulk));
105 bulk->bp_desc = desc;
106 list_add_tail(&bulk->bp_link, &desc->bd_page_list);
107 desc->bd_page_count++;
112 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *desc)
114 struct list_head *tmp, *next;
121 list_for_each_safe(tmp, next, &desc->bd_page_list) {
122 struct ptlrpc_bulk_page *bulk;
123 bulk = list_entry(tmp, struct ptlrpc_bulk_page, bp_link);
124 ptlrpc_free_bulk_page(bulk);
127 ptlrpc_put_connection(desc->bd_connection);
129 OBD_FREE(desc, sizeof(*desc));
133 void ptlrpc_free_bulk_page(struct ptlrpc_bulk_page *bulk)
141 list_del(&bulk->bp_link);
142 bulk->bp_desc->bd_page_count--;
143 OBD_FREE(bulk, sizeof(*bulk));
147 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, int opcode,
148 int count, int *lengths, char **bufs)
150 struct ptlrpc_connection *conn = imp->imp_connection;
151 struct ptlrpc_request *request;
155 OBD_ALLOC(request, sizeof(*request));
157 CERROR("request allocation out of memory\n");
161 rc = lustre_pack_msg(count, lengths, bufs,
162 &request->rq_reqlen, &request->rq_reqmsg);
164 CERROR("cannot pack request %d\n", rc);
165 OBD_FREE(request, sizeof(*request));
169 request->rq_level = LUSTRE_CONN_FULL;
170 request->rq_type = PTL_RPC_MSG_REQUEST;
171 request->rq_import = imp;
172 request->rq_connection = ptlrpc_connection_addref(conn);
174 INIT_LIST_HEAD(&request->rq_list);
175 INIT_LIST_HEAD(&request->rq_multi);
176 /* this will be dec()d once in req_finished, once in free_committed */
177 atomic_set(&request->rq_refcount, 2);
179 spin_lock(&conn->c_lock);
180 request->rq_xid = HTON__u32(++conn->c_xid_out);
181 spin_unlock(&conn->c_lock);
183 request->rq_reqmsg->magic = PTLRPC_MSG_MAGIC;
184 request->rq_reqmsg->version = PTLRPC_MSG_VERSION;
185 request->rq_reqmsg->opc = HTON__u32(opcode);
187 ptlrpc_hdl2req(request, &imp->imp_handle);
191 void ptlrpc_req_finished(struct ptlrpc_request *request)
196 if (request->rq_repmsg != NULL) {
197 OBD_FREE(request->rq_repmsg, request->rq_replen);
198 request->rq_repmsg = NULL;
199 request->rq_reply_md.start = NULL;
202 if (atomic_dec_and_test(&request->rq_refcount))
203 ptlrpc_free_req(request);
206 void ptlrpc_free_req(struct ptlrpc_request *request)
209 if (request == NULL) {
214 if (request->rq_repmsg != NULL)
215 OBD_FREE(request->rq_repmsg, request->rq_replen);
216 if (request->rq_reqmsg != NULL)
217 OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
219 if (request->rq_connection) {
220 spin_lock(&request->rq_connection->c_lock);
221 list_del_init(&request->rq_list);
222 spin_unlock(&request->rq_connection->c_lock);
225 ptlrpc_put_connection(request->rq_connection);
226 list_del(&request->rq_multi);
227 OBD_FREE(request, sizeof(*request));
231 static int ptlrpc_check_reply(struct ptlrpc_request *req)
235 if (req->rq_repmsg != NULL) {
236 req->rq_transno = NTOH__u64(req->rq_repmsg->transno);
237 req->rq_flags |= PTL_RPC_FL_REPLIED;
241 if (req->rq_flags & PTL_RPC_FL_RESEND) {
242 CERROR("-- RESTART --\n");
247 CDEBUG(D_NET, "req = %p, rc = %d\n", req, rc);
251 int ptlrpc_check_status(struct ptlrpc_request *req, int err)
256 CERROR("err is %d\n", err);
261 CERROR("req == NULL\n");
265 if (req->rq_repmsg == NULL) {
266 CERROR("req->rq_repmsg == NULL\n");
270 err = req->rq_repmsg->status;
271 if (req->rq_repmsg->type == NTOH__u32(PTL_RPC_MSG_ERR)) {
272 CERROR("req->rq_repmsg->type == PTL_RPC_MSG_ERR\n");
273 RETURN(err ? err : -EINVAL);
278 CERROR("req->rq_repmsg->status is %d\n", err);
280 CDEBUG(D_INFO, "req->rq_repmsg->status is %d\n", err);
281 /* XXX: translate this error from net to host */
288 static void ptlrpc_cleanup_request_buf(struct ptlrpc_request *request)
290 OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
291 request->rq_reqmsg = NULL;
292 request->rq_reqlen = 0;
295 /* Abort this request and cleanup any resources associated with it. */
296 static int ptlrpc_abort(struct ptlrpc_request *request)
298 /* First remove the ME for the reply; in theory, this means
299 * that we can tear down the buffer safely. */
300 PtlMEUnlink(request->rq_reply_me_h);
301 OBD_FREE(request->rq_reply_md.start, request->rq_replen);
302 request->rq_repmsg = NULL;
303 request->rq_replen = 0;
307 /* caller must hold conn->c_lock */
308 void ptlrpc_free_committed(struct ptlrpc_connection *conn)
310 struct list_head *tmp, *saved;
311 struct ptlrpc_request *req;
314 list_for_each_safe(tmp, saved, &conn->c_sending_head) {
315 req = list_entry(tmp, struct ptlrpc_request, rq_list);
317 if (req->rq_flags & PTL_RPC_FL_REPLAY) {
318 CDEBUG(D_INFO, "Keeping req %p xid "LPD64" for replay\n",
323 if (!(req->rq_flags & PTL_RPC_FL_REPLIED)) {
324 CDEBUG(D_INFO, "Keeping in-flight req %p xid "LPD64
325 " for replay\n", req, req->rq_xid);
329 /* not yet committed */
330 if (req->rq_transno > conn->c_last_committed)
333 CDEBUG(D_INFO, "Marking request %p xid %Ld as committed "
334 "transno=%Lu, last_committed=%Lu\n", req,
335 (long long)req->rq_xid, (long long)req->rq_transno,
336 (long long)conn->c_last_committed);
337 if (atomic_dec_and_test(&req->rq_refcount)) {
338 req->rq_import = NULL;
340 /* We do this to prevent free_req deadlock. Restarting
341 * after each removal is not so bad, as we are almost
342 * always deleting the first item in the list.
344 spin_unlock(&conn->c_lock);
345 ptlrpc_free_req(req);
346 spin_lock(&conn->c_lock);
349 list_del(&req->rq_list);
350 list_add(&req->rq_list, &conn->c_dying_head);
358 void ptlrpc_cleanup_client(struct obd_import *imp)
360 struct list_head *tmp, *saved;
361 struct ptlrpc_request *req;
362 struct ptlrpc_connection *conn = imp->imp_connection;
368 spin_lock(&conn->c_lock);
369 list_for_each_safe(tmp, saved, &conn->c_sending_head) {
370 req = list_entry(tmp, struct ptlrpc_request, rq_list);
371 if (req->rq_import != imp)
373 CDEBUG(D_INFO, "Cleaning req %p from sending list.\n", req);
374 list_del_init(&req->rq_list);
375 req->rq_import = NULL;
376 spin_unlock(&conn->c_lock);
377 ptlrpc_free_req(req);
381 list_for_each_safe(tmp, saved, &conn->c_dying_head) {
382 req = list_entry(tmp, struct ptlrpc_request, rq_list);
383 if (req->rq_import != imp)
385 CERROR("Request %p is on the dying list at cleanup!\n", req);
386 list_del_init(&req->rq_list);
387 req->rq_import = NULL;
388 spin_unlock(&conn->c_lock);
389 ptlrpc_free_req(req);
390 spin_lock(&conn->c_lock);
393 spin_unlock(&conn->c_lock);
399 void ptlrpc_continue_req(struct ptlrpc_request *req)
402 CDEBUG(D_INODE, "continue delayed request "LPD64" opc %d\n",
403 req->rq_xid, req->rq_reqmsg->opc);
404 req->rq_reqmsg->addr = req->rq_import->imp_handle.addr;
405 req->rq_reqmsg->cookie = req->rq_import->imp_handle.cookie;
406 wake_up(&req->rq_wait_for_rep);
410 void ptlrpc_resend_req(struct ptlrpc_request *req)
413 CDEBUG(D_INODE, "resend request "LPD64", opc %d\n",
414 req->rq_xid, req->rq_reqmsg->opc);
415 req->rq_reqmsg->addr = req->rq_import->imp_handle.addr;
416 req->rq_reqmsg->cookie = req->rq_import->imp_handle.cookie;
417 req->rq_status = -EAGAIN;
418 req->rq_level = LUSTRE_CONN_RECOVD;
419 req->rq_flags |= PTL_RPC_FL_RESEND;
420 req->rq_flags &= ~PTL_RPC_FL_TIMEOUT;
421 wake_up(&req->rq_wait_for_rep);
425 void ptlrpc_restart_req(struct ptlrpc_request *req)
428 CDEBUG(D_INODE, "restart completed request "LPD64", opc %d\n",
429 req->rq_xid, req->rq_reqmsg->opc);
430 req->rq_status = -ERESTARTSYS;
431 req->rq_flags |= PTL_RPC_FL_RECOVERY;
432 req->rq_flags &= ~PTL_RPC_FL_TIMEOUT;
433 wake_up(&req->rq_wait_for_rep);
437 static int expired_request(void *data)
439 struct ptlrpc_request *req = data;
442 CERROR("req timeout on connid %d xid %Ld\n", req->rq_connid,
443 (unsigned long long)req->rq_xid);
444 req->rq_flags |= PTL_RPC_FL_TIMEOUT;
445 if (!req->rq_import->imp_connection->c_recovd_data.rd_recovd)
449 req->rq_connection->c_level = LUSTRE_CONN_RECOVD;
450 recovd_conn_fail(req->rq_import->imp_connection);
452 /* If this request is for recovery or other primordial tasks,
453 * don't go back to sleep.
455 if (req->rq_level < LUSTRE_CONN_FULL)
460 static int interrupted_request(void *data)
462 struct ptlrpc_request *req = data;
464 req->rq_flags |= PTL_RPC_FL_INTR;
465 RETURN(1); /* ignored, as of this writing */
468 int ptlrpc_queue_wait(struct ptlrpc_request *req)
471 struct l_wait_info lwi;
472 struct ptlrpc_client *cli = req->rq_import->imp_client;
473 struct ptlrpc_connection *conn = req->rq_import->imp_connection;
476 init_waitqueue_head(&req->rq_wait_for_rep);
477 CDEBUG(D_NET, "subsys: %s req "LPD64" opc %d level %d, conn level %d\n",
478 cli->cli_name, req->rq_xid, req->rq_reqmsg->opc, req->rq_level,
479 req->rq_connection->c_level);
481 /* XXX probably both an import and connection level are needed */
482 if (req->rq_level > conn->c_level) {
483 CERROR("process %d waiting for recovery (%d > %d)\n",
484 current->pid, req->rq_level, conn->c_level);
486 spin_lock(&conn->c_lock);
487 list_del(&req->rq_list);
488 list_add_tail(&req->rq_list, &conn->c_delayed_head);
489 spin_unlock(&conn->c_lock);
491 lwi = LWI_INTR(NULL, NULL);
492 rc = l_wait_event(req->rq_wait_for_rep,
493 req->rq_level <= conn->c_level, &lwi);
495 spin_lock(&conn->c_lock);
496 list_del_init(&req->rq_list);
497 spin_unlock(&conn->c_lock);
502 CERROR("process %d resumed\n", current->pid);
505 req->rq_time = CURRENT_TIME;
506 req->rq_timeout = obd_timeout;
507 spin_lock(&conn->c_lock);
508 list_del(&req->rq_list);
509 list_add_tail(&req->rq_list, &conn->c_sending_head);
510 spin_unlock(&conn->c_lock);
511 rc = ptl_send_rpc(req);
513 CERROR("error %d, opcode %d, need recovery\n", rc,
514 req->rq_reqmsg->opc);
515 /* the sleep below will time out, triggering recovery */
518 CDEBUG(D_OTHER, "-- sleeping\n");
519 lwi = LWI_TIMEOUT_INTR(req->rq_timeout * HZ, expired_request,
520 interrupted_request,req);
521 l_wait_event(req->rq_wait_for_rep, ptlrpc_check_reply(req), &lwi);
522 CDEBUG(D_OTHER, "-- done\n");
524 /* Don't resend if we were interrupted. */
525 if ((req->rq_flags & (PTL_RPC_FL_RESEND | PTL_RPC_FL_INTR)) ==
527 req->rq_flags &= ~PTL_RPC_FL_RESEND;
528 CDEBUG(D_OTHER, "resending req %p xid "LPD64"\n",
533 // up(&cli->cli_rpc_sem);
534 if (req->rq_flags & PTL_RPC_FL_INTR) {
535 if (!(req->rq_flags & PTL_RPC_FL_TIMEOUT))
536 LBUG(); /* should only be interrupted if we timed out */
537 /* Clean up the dangling reply buffers */
539 GOTO(out, rc = -EINTR);
542 if (req->rq_flags & PTL_RPC_FL_TIMEOUT)
543 GOTO(out, rc = -ETIMEDOUT);
545 if (!(req->rq_flags & PTL_RPC_FL_REPLIED))
546 GOTO(out, rc = req->rq_status);
548 rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
550 CERROR("unpack_rep failed: %d\n", rc);
554 /* FIXME: Enable when BlueArc makes new release */
555 if (req->rq_repmsg->type != PTL_RPC_MSG_REPLY &&
556 req->rq_repmsg->type != PTL_RPC_MSG_ERR) {
557 CERROR("invalid packet type received (type=%u)\n",
558 req->rq_repmsg->type);
560 GOTO(out, rc = -EINVAL);
563 CDEBUG(D_NET, "got rep "LPD64"\n", req->rq_xid);
564 if (req->rq_repmsg->status == 0)
565 CDEBUG(D_NET, "--> buf %p len %d status %d\n", req->rq_repmsg,
566 req->rq_replen, req->rq_repmsg->status);
568 spin_lock(&conn->c_lock);
569 conn->c_last_xid = req->rq_repmsg->last_xid;
570 conn->c_last_committed = req->rq_repmsg->last_committed;
571 ptlrpc_free_committed(conn);
572 spin_unlock(&conn->c_lock);
579 int ptlrpc_replay_req(struct ptlrpc_request *req)
582 // struct ptlrpc_client *cli = req->rq_import->imp_client;
583 struct l_wait_info lwi;
586 init_waitqueue_head(&req->rq_wait_for_rep);
587 CDEBUG(D_NET, "req "LPD64" opc %d level %d, conn level %d\n",
588 req->rq_xid, req->rq_reqmsg->opc, req->rq_level,
589 req->rq_connection->c_level);
591 req->rq_time = CURRENT_TIME;
592 req->rq_timeout = obd_timeout;
593 req->rq_reqmsg->addr = req->rq_import->imp_handle.addr;
594 req->rq_reqmsg->cookie = req->rq_import->imp_handle.cookie;
595 rc = ptl_send_rpc(req);
597 CERROR("error %d, opcode %d\n", rc, req->rq_reqmsg->opc);
598 ptlrpc_cleanup_request_buf(req);
599 // up(&cli->cli_rpc_sem);
603 CDEBUG(D_OTHER, "-- sleeping\n");
604 lwi = LWI_INTR(NULL, NULL);
605 l_wait_event(req->rq_wait_for_rep, ptlrpc_check_reply(req), &lwi);
606 CDEBUG(D_OTHER, "-- done\n");
608 // up(&cli->cli_rpc_sem);
610 if (!(req->rq_flags & PTL_RPC_FL_REPLIED)) {
611 CERROR("Unknown reason for wakeup\n");
612 /* XXX Phil - I end up here when I kill obdctl */
614 GOTO(out, rc = -EINTR);
617 rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
619 CERROR("unpack_rep failed: %d\n", rc);
623 CDEBUG(D_NET, "got rep "LPD64"\n", req->rq_xid);
624 if (req->rq_repmsg->status == 0)
625 CDEBUG(D_NET, "--> buf %p len %d status %d\n", req->rq_repmsg,
626 req->rq_replen, req->rq_repmsg->status);
628 CERROR("recovery failed: ");
629 CERROR("req "LPD64" opc %d level %d, conn level %d\n",
630 req->rq_xid, req->rq_reqmsg->opc, req->rq_level,
631 req->rq_connection->c_level);
635 if (req->rq_replay_cb)
636 req->rq_replay_cb(req, req->rq_replay_cb_data);