*
* Copyright (c) 2002, 2003 Cluster File Systems, Inc.
*
- * This file is part of Lustre, http://www.lustre.org.
+ * This file is part of the Lustre file system, http://www.lustre.org
+ * Lustre is a trademark of Cluster File Systems, Inc.
*
- * Lustre is free software; you can redistribute it and/or
- * modify it under the terms of version 2 of the GNU General Public
- * License as published by the Free Software Foundation.
+ * You may have signed or agreed to another license before downloading
+ * this software. If so, you are bound by the terms and conditions
+ * of that agreement, and the following does not apply to you. See the
+ * LICENSE file included with this distribution for more information.
*
- * Lustre 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 for more details.
+ * If you did not agree to a different license, then this copy of Lustre
+ * is open source software; you can redistribute it and/or modify it
+ * under the terms of version 2 of the GNU General Public License as
+ * published by the Free Software Foundation.
*
- * You should have received a copy of the GNU General Public License
- * along with Lustre; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * In either case, Lustre 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
+ * license text for more details.
*
*/
#ifndef __KERNEL__
#include <liblustre.h>
#endif
-#include <linux/obd_class.h>
-#include <linux/obd_support.h>
-#include <linux/lustre_net.h>
-#include <linux/lustre_lib.h>
-#include <linux/obd.h>
-#include <linux/lustre_sec.h>
+#include <obd_support.h>
+#include <lustre_net.h>
+#include <lustre_lib.h>
+#include <obd.h>
#include "ptlrpc_internal.h"
-static int ptl_send_buf (ptl_handle_md_t *mdh, void *base, int len,
- ptl_ack_req_t ack, struct ptlrpc_cb_id *cbid,
+static int ptl_send_buf (lnet_handle_md_t *mdh, void *base, int len,
+ lnet_ack_req_t ack, struct ptlrpc_cb_id *cbid,
struct ptlrpc_connection *conn, int portal, __u64 xid)
{
int rc;
- ptl_md_t md;
- char str[PTL_NALFMT_SIZE];
+ lnet_md_t md;
ENTRY;
LASSERT (portal != 0);
LASSERT (conn != NULL);
- CDEBUG (D_INFO, "conn=%p ni %s id %s on %s\n",
- conn, conn->c_peer.peer_ni->pni_name,
- ptlrpc_id2str(&conn->c_peer, str),
- conn->c_peer.peer_ni->pni_name);
+ CDEBUG (D_INFO, "conn=%p id %s\n", conn, libcfs_id2str(conn->c_peer));
md.start = base;
md.length = len;
- md.threshold = (ack == PTL_ACK_REQ) ? 2 : 1;
+ md.threshold = (ack == LNET_ACK_REQ) ? 2 : 1;
md.options = PTLRPC_MD_OPTIONS;
md.user_ptr = cbid;
- md.eq_handle = conn->c_peer.peer_ni->pni_eq_h;
+ md.eq_handle = ptlrpc_eq_h;
- if (ack == PTL_ACK_REQ &&
+ if (ack == LNET_ACK_REQ &&
OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_ACK | OBD_FAIL_ONCE)) {
/* don't ask for the ack to simulate failing client */
- ack = PTL_NOACK_REQ;
+ ack = LNET_NOACK_REQ;
obd_fail_loc |= OBD_FAIL_ONCE | OBD_FAILED;
}
- rc = PtlMDBind (conn->c_peer.peer_ni->pni_ni_h, md,
- PTL_UNLINK, mdh);
- if (rc != PTL_OK) {
- CERROR ("PtlMDBind failed: %d\n", rc);
- LASSERT (rc == PTL_NO_SPACE);
+ rc = LNetMDBind (md, LNET_UNLINK, mdh);
+ if (rc != 0) {
+ CERROR ("LNetMDBind failed: %d\n", rc);
+ LASSERT (rc == -ENOMEM);
RETURN (-ENOMEM);
}
CDEBUG(D_NET, "Sending %d bytes to portal %d, xid "LPD64"\n",
len, portal, xid);
- rc = PtlPut (*mdh, ack, conn->c_peer.peer_id, portal, 0, xid, 0, 0);
- if (rc != PTL_OK) {
- int rc2;
+ rc = LNetPut (conn->c_self, *mdh, ack,
+ conn->c_peer, portal, xid, 0, 0);
+ if (rc != 0) {
+ int rc2;
/* We're going to get an UNLINK event when I unlink below,
* which will complete just like any other failed send, so
* I fall through and return success here! */
- CERROR("PtlPut(%s, %d, "LPD64") failed: %d\n",
- ptlrpc_id2str(&conn->c_peer, str),
- portal, xid, rc);
- rc2 = PtlMDUnlink(*mdh);
- LASSERTF(rc2 == PTL_OK, "rc2 = %d\n", rc2);
+ CERROR("LNetPut(%s, %d, "LPD64") failed: %d\n",
+ libcfs_id2str(conn->c_peer), portal, xid, rc);
+ rc2 = LNetMDUnlink(*mdh);
+ LASSERTF(rc2 == 0, "rc2 = %d\n", rc2);
}
RETURN (0);
int ptlrpc_start_bulk_transfer (struct ptlrpc_bulk_desc *desc)
{
- int rc;
- int rc2;
- struct ptlrpc_peer *peer;
- ptl_md_t md;
- __u64 xid;
- char str[PTL_NALFMT_SIZE];
+ struct ptlrpc_connection *conn = desc->bd_export->exp_connection;
+ int rc;
+ int rc2;
+ lnet_md_t md;
+ __u64 xid;
ENTRY;
if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_PTLRPC_BULK_PUT_NET))
LASSERT (desc->bd_type == BULK_PUT_SOURCE ||
desc->bd_type == BULK_GET_SINK);
desc->bd_success = 0;
- peer = &desc->bd_export->exp_connection->c_peer;
md.user_ptr = &desc->bd_cbid;
- md.eq_handle = peer->peer_ni->pni_eq_h;
+ md.eq_handle = ptlrpc_eq_h;
md.threshold = 2; /* SENT and ACK/REPLY */
md.options = PTLRPC_MD_OPTIONS;
ptlrpc_fill_bulk_md(&md, desc);
/* NB total length may be 0 for a read past EOF, so we send a 0
* length bulk, since the client expects a bulk event. */
- rc = PtlMDBind(peer->peer_ni->pni_ni_h, md,
- PTL_UNLINK, &desc->bd_md_h);
- if (rc != PTL_OK) {
- CERROR("PtlMDBind failed: %d\n", rc);
- LASSERT (rc == PTL_NO_SPACE);
+ rc = LNetMDBind(md, LNET_UNLINK, &desc->bd_md_h);
+ if (rc != 0) {
+ CERROR("LNetMDBind failed: %d\n", rc);
+ LASSERT (rc == -ENOMEM);
RETURN(-ENOMEM);
}
/* Client's bulk and reply matchbits are the same */
xid = desc->bd_req->rq_xid;
- CDEBUG(D_NET, "Transferring %u pages %u bytes via portal %d on %s "
- "nid %s pid %d xid "LPX64"\n", desc->bd_iov_count,
- desc->bd_nob, desc->bd_portal, peer->peer_ni->pni_name,
- ptlrpc_id2str(peer, str), peer->peer_id.pid, xid);
+ CDEBUG(D_NET, "Transferring %u pages %u bytes via portal %d "
+ "id %s xid "LPX64"\n", desc->bd_iov_count,
+ desc->bd_nob, desc->bd_portal,
+ libcfs_id2str(conn->c_peer), xid);
/* Network is about to get at the memory */
desc->bd_network_rw = 1;
if (desc->bd_type == BULK_PUT_SOURCE)
- rc = PtlPut (desc->bd_md_h, PTL_ACK_REQ, peer->peer_id,
- desc->bd_portal, 0, xid, 0, 0);
+ rc = LNetPut (conn->c_self, desc->bd_md_h, LNET_ACK_REQ,
+ conn->c_peer, desc->bd_portal, xid, 0, 0);
else
- rc = PtlGet (desc->bd_md_h, peer->peer_id,
- desc->bd_portal, 0, xid, 0);
+ rc = LNetGet (conn->c_self, desc->bd_md_h,
+ conn->c_peer, desc->bd_portal, xid, 0);
- if (rc != PTL_OK) {
+ if (rc != 0) {
/* Can't send, so we unlink the MD bound above. The UNLINK
* event this creates will signal completion with failure,
* so we return SUCCESS here! */
CERROR("Transfer(%s, %d, "LPX64") failed: %d\n",
- ptlrpc_id2str(peer, str),
- desc->bd_portal, xid, rc);
- rc2 = PtlMDUnlink(desc->bd_md_h);
- LASSERT (rc2 == PTL_OK);
+ libcfs_id2str(conn->c_peer), desc->bd_portal, xid, rc);
+ rc2 = LNetMDUnlink(desc->bd_md_h);
+ LASSERT (rc2 == 0);
}
RETURN(0);
if (!ptlrpc_bulk_active(desc)) /* completed or */
return; /* never started */
+ /* Do not send any meaningful data over the wire for evicted clients */
+ if (desc->bd_export && desc->bd_export->exp_failed)
+ ptl_rpc_wipe_bulk_pages(desc);
+
/* The unlink ensures the callback happens ASAP and is the last
* one. If it fails, it must be because completion just happened,
* but we must still l_wait_event() in this case, to give liblustre
* a chance to run server_bulk_callback()*/
- PtlMDUnlink (desc->bd_md_h);
+ LNetMDUnlink (desc->bd_md_h);
for (;;) {
/* Network access will complete in finite time but the HUGE
* timeout lets us CWARN for visibility of sluggish NALs */
- lwi = LWI_TIMEOUT (300 * HZ, NULL, NULL);
+ lwi = LWI_TIMEOUT (cfs_time_seconds(300), NULL, NULL);
rc = l_wait_event(desc->bd_waitq,
!ptlrpc_bulk_active(desc), &lwi);
if (rc == 0)
int ptlrpc_register_bulk (struct ptlrpc_request *req)
{
struct ptlrpc_bulk_desc *desc = req->rq_bulk;
- struct ptlrpc_peer *peer;
+ lnet_process_id_t peer;
int rc;
int rc2;
- ptl_handle_me_t me_h;
- ptl_md_t md;
+ lnet_handle_me_t me_h;
+ lnet_md_t md;
ENTRY;
if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_PTLRPC_BULK_GET_NET))
desc->bd_success = 0;
- peer = &desc->bd_import->imp_connection->c_peer;
+ peer = desc->bd_import->imp_connection->c_peer;
md.user_ptr = &desc->bd_cbid;
- md.eq_handle = peer->peer_ni->pni_eq_h;
+ md.eq_handle = ptlrpc_eq_h;
md.threshold = 1; /* PUT or GET */
md.options = PTLRPC_MD_OPTIONS |
((desc->bd_type == BULK_GET_SOURCE) ?
- PTL_MD_OP_GET : PTL_MD_OP_PUT);
+ LNET_MD_OP_GET : LNET_MD_OP_PUT);
ptlrpc_fill_bulk_md(&md, desc);
LASSERT (desc->bd_cbid.cbid_fn == client_bulk_callback);
/* XXX Registering the same xid on retried bulk makes my head
* explode trying to understand how the original request's bulk
* might interfere with the retried request -eeb */
- LASSERT (!desc->bd_registered || req->rq_xid != desc->bd_last_xid);
+ LASSERTF (!desc->bd_registered || req->rq_xid != desc->bd_last_xid,
+ "registered: %d rq_xid: "LPU64" bd_last_xid: "LPU64"\n",
+ desc->bd_registered, req->rq_xid, desc->bd_last_xid);
desc->bd_registered = 1;
desc->bd_last_xid = req->rq_xid;
- rc = PtlMEAttach(peer->peer_ni->pni_ni_h, desc->bd_portal,
- desc->bd_import->imp_connection->c_peer.peer_id,
- req->rq_xid, 0, PTL_UNLINK, PTL_INS_AFTER, &me_h);
- if (rc != PTL_OK) {
- CERROR("PtlMEAttach failed: %d\n", rc);
- LASSERT (rc == PTL_NO_SPACE);
+ rc = LNetMEAttach(desc->bd_portal, peer,
+ req->rq_xid, 0, LNET_UNLINK, LNET_INS_AFTER, &me_h);
+ if (rc != 0) {
+ CERROR("LNetMEAttach failed: %d\n", rc);
+ LASSERT (rc == -ENOMEM);
RETURN (-ENOMEM);
}
/* About to let the network at it... */
desc->bd_network_rw = 1;
- rc = PtlMDAttach(me_h, md, PTL_UNLINK, &desc->bd_md_h);
- if (rc != PTL_OK) {
- CERROR("PtlMDAttach failed: %d\n", rc);
- LASSERT (rc == PTL_NO_SPACE);
+ rc = LNetMDAttach(me_h, md, LNET_UNLINK, &desc->bd_md_h);
+ if (rc != 0) {
+ CERROR("LNetMDAttach failed: %d\n", rc);
+ LASSERT (rc == -ENOMEM);
desc->bd_network_rw = 0;
- rc2 = PtlMEUnlink (me_h);
- LASSERT (rc2 == PTL_OK);
+ rc2 = LNetMEUnlink (me_h);
+ LASSERT (rc2 == 0);
RETURN (-ENOMEM);
}
CDEBUG(D_NET, "Setup bulk %s buffers: %u pages %u bytes, xid "LPX64", "
- "portal %u on %s\n",
+ "portal %u\n",
desc->bd_type == BULK_GET_SOURCE ? "get-source" : "put-sink",
desc->bd_iov_count, desc->bd_nob,
- req->rq_xid, desc->bd_portal, peer->peer_ni->pni_name);
+ req->rq_xid, desc->bd_portal);
RETURN(0);
}
/* Disconnect a bulk desc from the network. Idempotent. Not
* thread-safe (i.e. only interlocks with completion callback). */
struct ptlrpc_bulk_desc *desc = req->rq_bulk;
- wait_queue_head_t *wq;
+ cfs_waitq_t *wq;
struct l_wait_info lwi;
int rc;
- LASSERT (!in_interrupt ()); /* might sleep */
+ LASSERT (!in_interrupt ()); /* might sleep */
- if (!ptlrpc_bulk_active(desc)) /* completed or */
- return; /* never registered */
-
- LASSERT (desc->bd_req == req); /* bd_req NULL until registered */
+ if (!ptlrpc_bulk_active(desc)) /* completed or */
+ return; /* never registered */
+
+ LASSERT (desc->bd_req == req); /* bd_req NULL until registered */
/* the unlink ensures the callback happens ASAP and is the last
* one. If it fails, it must be because completion just happened,
* but we must still l_wait_event() in this case to give liblustre
* a chance to run client_bulk_callback() */
- PtlMDUnlink (desc->bd_md_h);
-
- if (req->rq_set != NULL)
+ LNetMDUnlink (desc->bd_md_h);
+
+ if (req->rq_set != NULL)
wq = &req->rq_set->set_waitq;
else
wq = &req->rq_reply_waitq;
for (;;) {
/* Network access will complete in finite time but the HUGE
* timeout lets us CWARN for visibility of sluggish NALs */
- lwi = LWI_TIMEOUT (300 * HZ, NULL, NULL);
+ lwi = LWI_TIMEOUT (cfs_time_seconds(300), NULL, NULL);
rc = l_wait_event(*wq, !ptlrpc_bulk_active(desc), &lwi);
if (rc == 0)
return;
-
+
LASSERT (rc == -ETIMEDOUT);
- CWARN("Unexpectedly long timeout: desc %p\n", desc);
+ DEBUG_REQ(D_WARNING,req,"Unexpectedly long timeout: desc %p\n",
+ desc);
}
}
int ptlrpc_send_reply (struct ptlrpc_request *req, int may_be_difficult)
{
- struct ptlrpc_service *svc = req->rq_rqbd->rqbd_srv_ni->sni_service;
+ struct ptlrpc_service *svc = req->rq_rqbd->rqbd_service;
struct ptlrpc_reply_state *rs = req->rq_reply_state;
struct ptlrpc_connection *conn;
int rc;
/* We must already have a reply buffer (only ptlrpc_error() may be
- * called without one). We usually also have a request buffer which
+ * called without one). We must also have a request buffer which
* is either the actual (swabbed) incoming request, or a saved copy
- * if this is a req saved in target_queue_final_reply(). but this
- * will not be true since some security handling may skip the reqmsg
- * setting and prepare reply under normal ptlrpc layer */
- LASSERT (rs != NULL);
+ * if this is a req saved in target_queue_final_reply(). */
+ LASSERT (req->rq_reqmsg != NULL);
LASSERT (req->rq_repmsg != NULL);
- LASSERT (may_be_difficult || !rs->rs_difficult);
+ LASSERT (rs != NULL);
LASSERT (req->rq_repmsg == rs->rs_msg);
+ LASSERT (may_be_difficult || !rs->rs_difficult);
LASSERT (rs->rs_cb_id.cbid_fn == reply_out_callback);
LASSERT (rs->rs_cb_id.cbid_arg == rs);
- LASSERT (req->rq_repmsg != NULL);
+ if (req->rq_export && req->rq_export->exp_obd &&
+ req->rq_export->exp_obd->obd_fail) {
+ /* Failed obd's only send ENODEV */
+ req->rq_type = PTL_RPC_MSG_ERR;
+ req->rq_status = -ENODEV;
+ CDEBUG(D_HA, "sending ENODEV from failed obd %d\n",
+ req->rq_export->exp_obd->obd_minor);
+ }
+
if (req->rq_type != PTL_RPC_MSG_ERR)
req->rq_type = PTL_RPC_MSG_REPLY;
- req->rq_repmsg->type = req->rq_type;
- req->rq_repmsg->status = req->rq_status;
- req->rq_repmsg->opc = req->rq_reqmsg ? req->rq_reqmsg->opc : 0;
+ lustre_msg_set_type(req->rq_repmsg, req->rq_type);
+ lustre_msg_set_status(req->rq_repmsg, req->rq_status);
+ lustre_msg_set_opc(req->rq_repmsg, lustre_msg_get_opc(req->rq_reqmsg));
- if (req->rq_export == NULL)
- conn = ptlrpc_get_connection(&req->rq_peer, NULL);
+ if (req->rq_export == NULL || req->rq_export->exp_connection == NULL)
+ conn = ptlrpc_get_connection(req->rq_peer, req->rq_self, NULL);
else
conn = ptlrpc_connection_addref(req->rq_export->exp_connection);
- atomic_inc (&svc->srv_outstanding_replies);
-
- rc = svcsec_authorize(req);
- if (rc) {
- CDEBUG(D_ERROR, "Error wrap reply message "LPU64"\n", req->rq_xid);
- goto out;
+ if (conn == NULL) {
+ CERROR("not replying on NULL connection\n"); /* bug 9635 */
+ return -ENOTCONN;
}
+ atomic_inc (&svc->srv_outstanding_replies);
+ ptlrpc_rs_addref(rs); /* +1 ref for the network */
- rc = ptl_send_buf (&rs->rs_md_h, rs->rs_repbuf, rs->rs_repdata_len,
- rs->rs_difficult ? PTL_ACK_REQ : PTL_NOACK_REQ,
+ rc = ptl_send_buf (&rs->rs_md_h, req->rq_repmsg, req->rq_replen,
+ rs->rs_difficult ? LNET_ACK_REQ : LNET_NOACK_REQ,
&rs->rs_cb_id, conn,
svc->srv_rep_portal, req->rq_xid);
-out:
if (rc != 0) {
atomic_dec (&svc->srv_outstanding_replies);
-
- if (!rs->rs_difficult) {
- /* Callers other than target_send_reply() expect me
- * to clean up on a comms error */
- lustre_free_reply_state (rs);
- req->rq_reply_state = NULL;
- req->rq_repmsg = NULL;
- }
+ ptlrpc_rs_decref(rs);
}
ptlrpc_put_connection(conn);
return rc;
ENTRY;
if (!req->rq_repmsg) {
- rc = lustre_pack_reply(req, 0, NULL, NULL);
+ rc = lustre_pack_reply(req, 1, NULL, NULL);
if (rc)
RETURN(rc);
}
req->rq_type = PTL_RPC_MSG_ERR;
- rc = ptlrpc_send_reply (req, 0);
+ rc = ptlrpc_send_reply(req, 0);
RETURN(rc);
}
-int ptl_send_rpc(struct ptlrpc_request *request)
+int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
{
int rc;
int rc2;
struct ptlrpc_connection *connection;
- unsigned long flags;
- ptl_handle_me_t reply_me_h;
- ptl_md_t reply_md;
+ lnet_handle_me_t reply_me_h;
+ lnet_md_t reply_md;
ENTRY;
+ OBD_FAIL_RETURN(OBD_FAIL_PTLRPC_DROP_RPC, 0);
+
LASSERT (request->rq_type == PTL_RPC_MSG_REQUEST);
/* If this is a re-transmit, we're required to have disengaged
* cleanly from the previous attempt */
LASSERT (!request->rq_receiving_reply);
+ if (request->rq_import->imp_obd &&
+ request->rq_import->imp_obd->obd_fail) {
+ CDEBUG(D_HA, "muting rpc for failed imp obd %s\n",
+ request->rq_import->imp_obd->obd_name);
+ /* this prevents us from waiting in ptlrpc_queue_wait */
+ request->rq_err = 1;
+ RETURN(-ENODEV);
+ }
+
connection = request->rq_import->imp_connection;
if (request->rq_bulk != NULL) {
RETURN(rc);
}
- request->rq_reqmsg->handle = request->rq_import->imp_remote_handle;
- request->rq_reqmsg->type = PTL_RPC_MSG_REQUEST;
- request->rq_reqmsg->conn_cnt = request->rq_import->imp_conn_cnt;
-
- /* wrap_request might need to refresh gss cred, if this is called
- * in ptlrpcd then the whole daemon thread will be waiting on
- * gss negotiate rpc. FIXME
- */
- rc = ptlrpcs_cli_wrap_request(request);
- if (rc)
- GOTO(cleanup_bulk, rc);
-
- LASSERT (request->rq_replen != 0);
- if (request->rq_repbuf == NULL) {
- rc = ptlrpcs_cli_alloc_repbuf(request, request->rq_replen);
- if (rc)
- GOTO(cleanup_bulk, rc);
- }
-
- rc = PtlMEAttach(connection->c_peer.peer_ni->pni_ni_h,
- request->rq_reply_portal, /* XXX FIXME bug 249 */
- connection->c_peer.peer_id, request->rq_xid, 0,
- PTL_UNLINK, PTL_INS_AFTER, &reply_me_h);
- if (rc != PTL_OK) {
- CERROR("PtlMEAttach failed: %d\n", rc);
- LASSERT (rc == PTL_NO_SPACE);
- GOTO(cleanup_repmsg, rc = -ENOMEM);
+ lustre_msg_set_handle(request->rq_reqmsg,
+ &request->rq_import->imp_remote_handle);
+ lustre_msg_set_type(request->rq_reqmsg, PTL_RPC_MSG_REQUEST);
+ lustre_msg_set_conn_cnt(request->rq_reqmsg,
+ request->rq_import->imp_conn_cnt);
+
+ if (!noreply) {
+ LASSERT (request->rq_replen != 0);
+ if (request->rq_repmsg == NULL)
+ OBD_ALLOC(request->rq_repmsg, request->rq_replen);
+ if (request->rq_repmsg == NULL)
+ GOTO(cleanup_bulk, rc = -ENOMEM);
+
+ rc = LNetMEAttach(request->rq_reply_portal,/*XXX FIXME bug 249*/
+ connection->c_peer, request->rq_xid, 0,
+ LNET_UNLINK, LNET_INS_AFTER, &reply_me_h);
+ if (rc != 0) {
+ CERROR("LNetMEAttach failed: %d\n", rc);
+ LASSERT (rc == -ENOMEM);
+ GOTO(cleanup_repmsg, rc = -ENOMEM);
+ }
}
- spin_lock_irqsave (&request->rq_lock, flags);
+ spin_lock(&request->rq_lock);
/* If the MD attach succeeds, there _will_ be a reply_in callback */
- request->rq_receiving_reply = 1;
+ request->rq_receiving_reply = !noreply;
/* Clear any flags that may be present from previous sends. */
request->rq_replied = 0;
request->rq_err = 0;
request->rq_net_err = 0;
request->rq_resend = 0;
request->rq_restart = 0;
- request->rq_ptlrpcs_restart = 0;
- request->rq_ptlrpcs_err = 0;
- spin_unlock_irqrestore (&request->rq_lock, flags);
-
- reply_md.start = request->rq_repbuf;
- reply_md.length = request->rq_repbuf_len;
- reply_md.threshold = 1;
- reply_md.options = PTLRPC_MD_OPTIONS | PTL_MD_OP_PUT;
- reply_md.user_ptr = &request->rq_reply_cbid;
- reply_md.eq_handle = connection->c_peer.peer_ni->pni_eq_h;
-
- rc = PtlMDAttach(reply_me_h, reply_md, PTL_UNLINK,
- &request->rq_reply_md_h);
- if (rc != PTL_OK) {
- CERROR("PtlMDAttach failed: %d\n", rc);
- LASSERT (rc == PTL_NO_SPACE);
- GOTO(cleanup_me, rc -ENOMEM);
+ spin_unlock(&request->rq_lock);
+
+ if (!noreply) {
+ reply_md.start = request->rq_repmsg;
+ reply_md.length = request->rq_replen;
+ reply_md.threshold = 1;
+ reply_md.options = PTLRPC_MD_OPTIONS | LNET_MD_OP_PUT;
+ reply_md.user_ptr = &request->rq_reply_cbid;
+ reply_md.eq_handle = ptlrpc_eq_h;
+
+ rc = LNetMDAttach(reply_me_h, reply_md, LNET_UNLINK,
+ &request->rq_reply_md_h);
+ if (rc != 0) {
+ CERROR("LNetMDAttach failed: %d\n", rc);
+ LASSERT (rc == -ENOMEM);
+ spin_lock(&request->rq_lock);
+ /* ...but the MD attach didn't succeed... */
+ request->rq_receiving_reply = 0;
+ spin_unlock(&request->rq_lock);
+ GOTO(cleanup_me, rc -ENOMEM);
+ }
+
+ CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid "LPU64
+ ", portal %u\n",
+ request->rq_replen, request->rq_xid,
+ request->rq_reply_portal);
}
- CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid "LPU64
- ", portal %u on %s\n",
- request->rq_replen, request->rq_xid,
- request->rq_reply_portal,
- connection->c_peer.peer_ni->pni_name);
+ /* add references on request and import for request_out_callback */
+ ptlrpc_request_addref(request);
+ atomic_inc(&request->rq_import->imp_inflight);
- ptlrpc_request_addref(request); /* +1 ref for the SENT callback */
+ OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_SEND, request->rq_timeout + 5);
- request->rq_sent = LTIME_S(CURRENT_TIME);
- rc = ptl_send_buf(&request->rq_req_md_h,
- request->rq_reqbuf, request->rq_reqdata_len,
- PTL_NOACK_REQ, &request->rq_req_cbid,
+ request->rq_sent = CURRENT_SECONDS;
+ ptlrpc_pinger_sending_on_import(request->rq_import);
+ rc = ptl_send_buf(&request->rq_req_md_h,
+ request->rq_reqmsg, request->rq_reqlen,
+ LNET_NOACK_REQ, &request->rq_req_cbid,
connection,
request->rq_request_portal,
request->rq_xid);
RETURN(rc);
}
- ptlrpc_req_finished (request); /* drop callback ref */
+ /* drop request_out_callback refs, we couldn't start the send */
+ atomic_dec(&request->rq_import->imp_inflight);
+ ptlrpc_req_finished(request);
+ if (noreply)
+ RETURN(rc);
+ else
+ GOTO(cleanup_me, rc);
cleanup_me:
/* MEUnlink is safe; the PUT didn't even get off the ground, and
* nobody apart from the PUT's target has the right nid+XID to
* access the reply buffer. */
- rc2 = PtlMEUnlink(reply_me_h);
- LASSERT (rc2 == PTL_OK);
+ rc2 = LNetMEUnlink(reply_me_h);
+ LASSERT (rc2 == 0);
/* UNLINKED callback called synchronously */
LASSERT (!request->rq_receiving_reply);
cleanup_repmsg:
- ptlrpcs_cli_free_repbuf(request);
+ OBD_FREE(request->rq_repmsg, request->rq_replen);
+ request->rq_repmsg = NULL;
cleanup_bulk:
if (request->rq_bulk != NULL)
int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd)
{
- struct ptlrpc_srv_ni *srv_ni = rqbd->rqbd_srv_ni;
- struct ptlrpc_service *service = srv_ni->sni_service;
- static ptl_process_id_t match_id = {PTL_NID_ANY, PTL_PID_ANY};
+ struct ptlrpc_service *service = rqbd->rqbd_service;
+ static lnet_process_id_t match_id = {LNET_NID_ANY, LNET_PID_ANY};
int rc;
- ptl_md_t md;
- ptl_handle_me_t me_h;
+ lnet_md_t md;
+ lnet_handle_me_t me_h;
- CDEBUG(D_NET, "PtlMEAttach: portal %d on %s\n",
- service->srv_req_portal, srv_ni->sni_ni->pni_name);
+ CDEBUG(D_NET, "LNetMEAttach: portal %d\n",
+ service->srv_req_portal);
if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_PTLRPC_RQBD))
return (-ENOMEM);
- rc = PtlMEAttach(srv_ni->sni_ni->pni_ni_h, service->srv_req_portal,
- match_id, 0, ~0, PTL_UNLINK, PTL_INS_AFTER, &me_h);
- if (rc != PTL_OK) {
- CERROR("PtlMEAttach failed: %d\n", rc);
+ rc = LNetMEAttach(service->srv_req_portal,
+ match_id, 0, ~0, LNET_UNLINK, LNET_INS_AFTER, &me_h);
+ if (rc != 0) {
+ CERROR("LNetMEAttach failed: %d\n", rc);
return (-ENOMEM);
}
md.start = rqbd->rqbd_buffer;
md.length = service->srv_buf_size;
md.max_size = service->srv_max_req_size;
- md.threshold = PTL_MD_THRESH_INF;
- md.options = PTLRPC_MD_OPTIONS | PTL_MD_OP_PUT | PTL_MD_MAX_SIZE;
+ md.threshold = LNET_MD_THRESH_INF;
+ md.options = PTLRPC_MD_OPTIONS | LNET_MD_OP_PUT | LNET_MD_MAX_SIZE;
md.user_ptr = &rqbd->rqbd_cbid;
- md.eq_handle = srv_ni->sni_ni->pni_eq_h;
+ md.eq_handle = ptlrpc_eq_h;
- rc = PtlMDAttach(me_h, md, PTL_UNLINK, &rqbd->rqbd_md_h);
- if (rc == PTL_OK)
+ rc = LNetMDAttach(me_h, md, LNET_UNLINK, &rqbd->rqbd_md_h);
+ if (rc == 0)
return (0);
- CERROR("PtlMDAttach failed: %d; \n", rc);
- LASSERT (rc == PTL_NO_SPACE);
- rc = PtlMEUnlink (me_h);
- LASSERT (rc == PTL_OK);
+ CERROR("LNetMDAttach failed: %d; \n", rc);
+ LASSERT (rc == -ENOMEM);
+ rc = LNetMEUnlink (me_h);
+ LASSERT (rc == 0);
rqbd->rqbd_refcount = 0;
return (-ENOMEM);
}
-
-
-/********************************************
- * rawrpc stuff, currently only used by gss *
- ********************************************/
-
-static int rawrpc_timedout(void *data)
-{
- struct ptlrpc_request *req = (struct ptlrpc_request *) data;
- unsigned long flags;
-
- spin_lock_irqsave(&req->rq_lock, flags);
- if (!req->rq_replied)
- req->rq_timedout = 1;
- spin_unlock_irqrestore(&req->rq_lock, flags);
-
- return 1;
-}
-
-static int rawrpc_timedout_wait(void *data)
-{
- return 0;
-}
-
-/* to make things as simple as possible */
-static int rawrpc_check_reply(struct ptlrpc_request *req)
-{
- unsigned long flags;
- int rc;
-
- spin_lock_irqsave (&req->rq_lock, flags);
- rc = req->rq_replied || req->rq_net_err || req->rq_err ||
- req->rq_resend || req->rq_restart;
- spin_unlock_irqrestore(&req->rq_lock, flags);
- return rc;
-}
-
-void rawrpc_req_finished(struct ptlrpc_request *req)
-{
- struct obd_import *imp;
- unsigned long irq_flags;
-
- if (!req)
- return;
-
- if (!atomic_dec_and_test(&req->rq_refcount))
- return;
-
- LASSERT(req->rq_import);
- imp = req->rq_import;
-
- spin_lock_irqsave(&imp->imp_lock, irq_flags);
- list_del_init(&req->rq_list);
- spin_unlock_irqrestore(&imp->imp_lock, irq_flags);
-
- class_import_put(imp);
-
- if (req->rq_reqbuf) {
- LASSERT(req->rq_reqbuf_len);
- OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
- }
- if (req->rq_repbuf) {
- LASSERT(req->rq_repbuf_len);
- OBD_FREE(req->rq_repbuf, req->rq_repbuf_len);
- }
- OBD_FREE(req, sizeof(*req));
-}
-
-/*
- * if returned non-NULL, reqbuf and repbuf will be take over by the
- * request: the caller can't release them directly, instead should
- * call rawrpc_req_finished().
- */
-struct ptlrpc_request *
-ptl_do_rawrpc(struct obd_import *imp,
- char *reqbuf, int reqbuf_len, int reqlen,
- char *repbuf, int repbuf_len, int *replenp,
- int timeout, int *res)
-{
- struct ptlrpc_connection *conn;
- struct ptlrpc_request *request;
- ptl_handle_me_t reply_me_h;
- ptl_md_t reply_md, req_md;
- struct l_wait_info lwi;
- unsigned long irq_flags;
- int rc;
- ENTRY;
-
- LASSERT(imp);
- LASSERT(reqbuf && reqbuf_len);
- LASSERT(repbuf && repbuf_len);
- LASSERT(reqlen && reqlen <= reqbuf_len);
-
- OBD_ALLOC(request, sizeof(*request));
- if (!request) {
- *res = -ENOMEM;
- RETURN(NULL);
- }
-
- imp = request->rq_import = class_import_get(imp);
- conn = imp->imp_connection;
-
- if (imp->imp_state == LUSTRE_IMP_CLOSED) {
- CDEBUG(D_SEC, "raw rpc on closed imp(=>%s)\n",
- imp->imp_target_uuid.uuid);
- }
-
- /* initialize request */
- request->rq_req_cbid.cbid_fn = rawrpc_request_out_callback;
- request->rq_req_cbid.cbid_arg = request;
- request->rq_reply_cbid.cbid_fn = reply_in_callback;
- request->rq_reply_cbid.cbid_arg = request;
- request->rq_reqbuf = reqbuf;
- request->rq_reqbuf_len = reqbuf_len;
- request->rq_repbuf = repbuf;
- request->rq_repbuf_len = repbuf_len;
- request->rq_set = NULL;
- spin_lock_init(&request->rq_lock);
- init_waitqueue_head(&request->rq_reply_waitq);
- atomic_set(&request->rq_refcount, 1);
- request->rq_xid = ptlrpc_next_xid();
-
- /* add into sending list */
- spin_lock_irqsave(&imp->imp_lock, irq_flags);
- list_add_tail(&request->rq_list, &imp->imp_rawrpc_list);
- spin_unlock_irqrestore(&imp->imp_lock, irq_flags);
-
- /* prepare reply buffer */
- rc = PtlMEAttach(conn->c_peer.peer_ni->pni_ni_h,
- imp->imp_client->cli_reply_portal,
- conn->c_peer.peer_id, request->rq_xid, 0, PTL_UNLINK,
- PTL_INS_AFTER, &reply_me_h);
- if (rc != PTL_OK) {
- CERROR("PtlMEAttach failed: %d\n", rc);
- LASSERT (rc == PTL_NO_SPACE);
- GOTO(out, rc = -ENOMEM);
- }
-
- spin_lock_irqsave(&request->rq_lock, irq_flags);
- request->rq_receiving_reply = 1;
- spin_unlock_irqrestore(&request->rq_lock, irq_flags);
-
- reply_md.start = repbuf;
- reply_md.length = repbuf_len;
- reply_md.threshold = 1;
- reply_md.options = PTLRPC_MD_OPTIONS | PTL_MD_OP_PUT;
- reply_md.user_ptr = &request->rq_reply_cbid;
- reply_md.eq_handle = conn->c_peer.peer_ni->pni_eq_h;
-
- rc = PtlMDAttach(reply_me_h, reply_md, PTL_UNLINK,
- &request->rq_reply_md_h);
- if (rc != PTL_OK) {
- CERROR("PtlMDAttach failed: %d\n", rc);
- LASSERT (rc == PTL_NO_SPACE);
- GOTO(cleanup_me, rc = -ENOMEM);
- }
-
- /*
- * the extra 2 refcount will be balanced by out_callback
- */
- atomic_set(&request->rq_refcount, 3);
-
- /* prepare request buffer */
- req_md.start = reqbuf;
- req_md.length = reqlen;
- req_md.threshold = 1;
- req_md.options = PTLRPC_MD_OPTIONS;
- req_md.user_ptr = &request->rq_req_cbid;
- req_md.eq_handle = conn->c_peer.peer_ni->pni_eq_h;
-
- rc = PtlMDBind(conn->c_peer.peer_ni->pni_ni_h,
- req_md, PTL_UNLINK, &request->rq_req_md_h);
- if (rc != PTL_OK) {
- CERROR("PtlMDBind failed %d\n", rc);
- LASSERT (rc == PTL_NO_SPACE);
- atomic_set(&request->rq_refcount, 1);
- GOTO(cleanup_me, rc = -ENOMEM);
- }
-
- rc = PtlPut(request->rq_req_md_h, PTL_NOACK_REQ, conn->c_peer.peer_id,
- imp->imp_client->cli_request_portal, 0,
- request->rq_xid, 0, 0);
- if (rc != PTL_OK) {
- CERROR("PtlPut failed %d\n", rc);
- GOTO(cleanup_md, rc);
- }
-
- if (timeout)
- lwi = LWI_TIMEOUT(timeout * HZ, rawrpc_timedout, request);
- else
- lwi = LWI_TIMEOUT(100 * HZ, rawrpc_timedout_wait, request);
-
- l_wait_event(request->rq_reply_waitq,
- rawrpc_check_reply(request), &lwi);
-
- ptlrpc_unregister_reply(request);
-
- if (request->rq_replied) {
- *replenp = request->rq_nob_received;
- rc = 0;
- } else {
- CERROR("rawrpc error: err %d, neterr %d, int %d, timedout %d\n",
- request->rq_err, request->rq_net_err,
- request->rq_intr, request->rq_timedout);
-
- /* give timedout higher priority */
- rc = request->rq_timedout ? -ETIMEDOUT : -EIO;
- }
-
-out:
- *res = rc;
- RETURN(request);
-
-cleanup_md:
- PtlMDUnlink(request->rq_req_md_h);
-cleanup_me:
- PtlMEUnlink(reply_me_h);
- goto out;
-}
-
-/*
- * caller will take care the reqbuf & repbuf, and only return
- * when the rpc really finished on wire.
- */
-int ptl_do_rawrpc_simple(struct obd_import *imp,
- char *reqbuf, int reqlen,
- char *repbuf, int *replenp)
-{
- struct ptlrpc_request *req;
- int res;
-
- req = ptl_do_rawrpc(imp, reqbuf, reqlen, reqlen,
- repbuf, *replenp, replenp, 0, &res);
-
- if (req == NULL)
- return res;
-
- req->rq_reqbuf = req->rq_repbuf = NULL;
- req->rq_reqbuf_len = req->rq_repbuf_len = 0;
-
- rawrpc_req_finished(req);
- return res;
-}