* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2015, Intel Corporation.
+ * Copyright (c) 2012, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* over \a conn connection to portal \a portal.
* Returns 0 on success or error code.
*/
-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,
- unsigned int offset)
+static int ptl_send_buf(struct lnet_handle_md *mdh, void *base, int len,
+ enum lnet_ack_req ack, struct ptlrpc_cb_id *cbid,
+ lnet_nid_t self, struct lnet_process_id peer_id,
+ int portal, __u64 xid, unsigned int offset,
+ struct lnet_handle_md *bulk_cookie)
{
- int rc;
- lnet_md_t md;
- ENTRY;
+ int rc;
+ struct lnet_md md;
+ ENTRY;
- LASSERT (portal != 0);
- LASSERT (conn != NULL);
- CDEBUG (D_INFO, "conn=%p id %s\n", conn, libcfs_id2str(conn->c_peer));
- md.start = base;
- md.length = len;
- md.threshold = (ack == LNET_ACK_REQ) ? 2 : 1;
- md.options = PTLRPC_MD_OPTIONS;
- md.user_ptr = cbid;
- md.eq_handle = ptlrpc_eq_h;
+ LASSERT (portal != 0);
+ CDEBUG (D_INFO, "peer_id %s\n", libcfs_id2str(peer_id));
+ md.start = base;
+ md.length = len;
+ md.threshold = (ack == LNET_ACK_REQ) ? 2 : 1;
+ md.options = PTLRPC_MD_OPTIONS;
+ md.user_ptr = cbid;
+ md.eq_handle = ptlrpc_eq_h;
+ LNetInvalidateMDHandle(&md.bulk_handle);
- if (unlikely(ack == LNET_ACK_REQ &&
- OBD_FAIL_CHECK_ORSET(OBD_FAIL_PTLRPC_ACK, OBD_FAIL_ONCE))){
- /* don't ask for the ack to simulate failing client */
- ack = LNET_NOACK_REQ;
- }
+ if (bulk_cookie) {
+ md.bulk_handle = *bulk_cookie;
+ md.options |= LNET_MD_BULK_HANDLE;
+ }
- rc = LNetMDBind (md, LNET_UNLINK, mdh);
- if (unlikely(rc != 0)) {
- CERROR ("LNetMDBind failed: %d\n", rc);
- LASSERT (rc == -ENOMEM);
- RETURN (-ENOMEM);
- }
+ if (unlikely(ack == LNET_ACK_REQ &&
+ OBD_FAIL_CHECK_ORSET(OBD_FAIL_PTLRPC_ACK, OBD_FAIL_ONCE))){
+ /* don't ask for the ack to simulate failing client */
+ ack = LNET_NOACK_REQ;
+ }
+
+ rc = LNetMDBind (md, LNET_UNLINK, mdh);
+ if (unlikely(rc != 0)) {
+ CERROR ("LNetMDBind failed: %d\n", rc);
+ LASSERT (rc == -ENOMEM);
+ RETURN (-ENOMEM);
+ }
CDEBUG(D_NET, "Sending %d bytes to portal %d, xid %lld, offset %u\n",
- len, portal, xid, offset);
-
- rc = LNetPut (conn->c_self, *mdh, ack,
- conn->c_peer, portal, xid, offset, 0);
- if (unlikely(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! */
+ len, portal, xid, offset);
+
+ rc = LNetPut(self, *mdh, ack,
+ peer_id, portal, xid, offset, 0);
+ if (unlikely(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("LNetPut(%s, %d, %lld) failed: %d\n",
- libcfs_id2str(conn->c_peer), portal, xid, rc);
- rc2 = LNetMDUnlink(*mdh);
- LASSERTF(rc2 == 0, "rc2 = %d\n", rc2);
- }
+ libcfs_id2str(peer_id), portal, xid, rc);
+ rc2 = LNetMDUnlink(*mdh);
+ LASSERTF(rc2 == 0, "rc2 = %d\n", rc2);
+ }
- RETURN (0);
+ RETURN (0);
}
-static void mdunlink_iterate_helper(lnet_handle_md_t *bd_mds, int count)
+static void mdunlink_iterate_helper(struct lnet_handle_md *bd_mds, int count)
{
int i;
int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc)
{
struct obd_export *exp = desc->bd_export;
- struct ptlrpc_connection *conn = exp->exp_connection;
+ lnet_nid_t self_nid;
+ struct lnet_process_id peer_id;
int rc = 0;
__u64 mbits;
int posted_md;
int total_md;
- lnet_md_t md;
+ struct lnet_md md;
ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_BULK_PUT_NET))
LASSERT(desc->bd_cbid.cbid_fn == server_bulk_callback);
LASSERT(desc->bd_cbid.cbid_arg == desc);
+ /*
+ * Multi-Rail: get the preferred self and peer NIDs from the
+ * request, so they are based on the route taken by the
+ * message.
+ */
+ self_nid = desc->bd_req->rq_self;
+ peer_id = desc->bd_req->rq_source;
+
/* NB total length may be 0 for a read past EOF, so we send 0
* length bulks, since the client expects bulk events.
*
/* Network is about to get at the memory */
if (ptlrpc_is_bulk_put_source(desc->bd_type))
- rc = LNetPut(conn->c_self, desc->bd_mds[posted_md],
- LNET_ACK_REQ, conn->c_peer,
+ rc = LNetPut(self_nid, desc->bd_mds[posted_md],
+ LNET_ACK_REQ, peer_id,
desc->bd_portal, mbits, 0, 0);
else
- rc = LNetGet(conn->c_self, desc->bd_mds[posted_md],
- conn->c_peer, desc->bd_portal, mbits, 0);
+ rc = LNetGet(self_nid, desc->bd_mds[posted_md],
+ peer_id, desc->bd_portal, mbits, 0);
posted_md++;
if (rc != 0) {
CERROR("%s: failed bulk transfer with %s:%u x%llu: "
"rc = %d\n", exp->exp_obd->obd_name,
- libcfs_id2str(conn->c_peer), desc->bd_portal,
+ libcfs_id2str(peer_id), desc->bd_portal,
mbits, rc);
break;
}
CDEBUG(D_NET, "Transferring %u pages %u bytes via portal %d "
"id %s mbits %#llx-%#llx\n", desc->bd_iov_count,
- desc->bd_nob, desc->bd_portal, libcfs_id2str(conn->c_peer),
+ desc->bd_nob, desc->bd_portal, libcfs_id2str(peer_id),
mbits - posted_md, mbits - 1);
RETURN(0);
int ptlrpc_register_bulk(struct ptlrpc_request *req)
{
struct ptlrpc_bulk_desc *desc = req->rq_bulk;
- lnet_process_id_t peer;
+ struct lnet_process_id peer;
int rc = 0;
int rc2;
int posted_md;
int total_md;
__u64 mbits;
- lnet_handle_me_t me_h;
- lnet_md_t md;
+ struct lnet_handle_me me_h;
+ struct lnet_md md;
ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_BULK_GET_NET))
/* cleanup the state of the bulk for it will be reused */
if (req->rq_resend || req->rq_send_state == LUSTRE_IMP_REPLAY)
desc->bd_nob_transferred = 0;
- else
- LASSERT(desc->bd_nob_transferred == 0);
+ else if (desc->bd_nob_transferred != 0)
+ /* If the network failed after an RPC was sent, this condition
+ * could happen. Rather than assert (was here before), return
+ * an EIO error. */
+ RETURN(-EIO);
desc->bd_failure = 0;
/* Let's setup deadline for reply unlink. */
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
async && req->rq_bulk_deadline == 0 && cfs_fail_val == 0)
- req->rq_bulk_deadline = cfs_time_current_sec() + LONG_UNLINK;
+ req->rq_bulk_deadline = ktime_get_real_seconds() + LONG_UNLINK;
if (ptlrpc_client_bulk_active(req) == 0) /* completed or */
RETURN(1); /* never registered */
{
struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
struct ptlrpc_service *svc = svcpt->scp_service;
- int service_time = max_t(int, cfs_time_current_sec() -
+ int service_time = max_t(int, ktime_get_real_seconds() -
req->rq_arrival_time.tv_sec, 1);
if (!(flags & PTLRPC_REPLY_EARLY) &&
}
/* Report actual service time for client latency calc */
lustre_msg_set_service_time(req->rq_repmsg, service_time);
- /* Report service time estimate for future client reqs, but report 0
+ /* Report service time estimate for future client reqs, but report 0
* (to be ignored by client) if it's an error reply during recovery.
- * (bz15815) */
- if (req->rq_type == PTL_RPC_MSG_ERR &&
+ * b=15815
+ */
+ if (req->rq_type == PTL_RPC_MSG_ERR &&
(req->rq_export == NULL ||
req->rq_export->exp_obd->obd_recovering)) {
- lustre_msg_set_timeout(req->rq_repmsg, 0);
+ lustre_msg_set_timeout(req->rq_repmsg, 0);
} else {
- __u32 timeout;
+ time64_t timeout;
if (req->rq_export && req->rq_reqmsg != NULL &&
+ (flags & PTLRPC_REPLY_EARLY) &&
lustre_msg_get_flags(req->rq_reqmsg) &
- (MSG_REPLAY | MSG_REQ_REPLAY_DONE | MSG_LOCK_REPLAY_DONE))
- timeout = cfs_time_current_sec() -
- req->rq_arrival_time.tv_sec +
- min(at_extra,
- req->rq_export->exp_obd->
- obd_recovery_timeout / 4);
- else
+ (MSG_REPLAY | MSG_REQ_REPLAY_DONE | MSG_LOCK_REPLAY_DONE)) {
+ struct obd_device *exp_obd = req->rq_export->exp_obd;
+
+ timeout = ktime_get_real_seconds() -
+ req->rq_arrival_time.tv_sec +
+ min_t(time64_t, at_extra,
+ exp_obd->obd_recovery_timeout / 4);
+ } else {
timeout = at_get(&svcpt->scp_at_estimate);
+ }
lustre_msg_set_timeout(req->rq_repmsg, timeout);
}
*/
int ptlrpc_send_reply(struct ptlrpc_request *req, int flags)
{
- struct ptlrpc_reply_state *rs = req->rq_reply_state;
- struct ptlrpc_connection *conn;
- int rc;
+ 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). The reply generated by sptlrpc layer (e.g.
if (unlikely(rc))
goto out;
- req->rq_sent = cfs_time_current_sec();
+ req->rq_sent = ktime_get_real_seconds();
- rc = ptl_send_buf (&rs->rs_md_h, rs->rs_repbuf, rs->rs_repdata_len,
- (rs->rs_difficult && !rs->rs_no_ack) ?
- LNET_ACK_REQ : LNET_NOACK_REQ,
- &rs->rs_cb_id, conn,
- ptlrpc_req2svc(req)->srv_rep_portal,
- req->rq_xid, req->rq_reply_off);
+ rc = ptl_send_buf(&rs->rs_md_h, rs->rs_repbuf, rs->rs_repdata_len,
+ (rs->rs_difficult && !rs->rs_no_ack) ?
+ LNET_ACK_REQ : LNET_NOACK_REQ,
+ &rs->rs_cb_id, req->rq_self, req->rq_source,
+ ptlrpc_req2svc(req)->srv_rep_portal,
+ req->rq_xid, req->rq_reply_off, NULL);
out:
if (unlikely(rc != 0))
ptlrpc_req_drop_rs(req);
*/
int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
{
- int rc;
- int rc2;
- int mpflag = 0;
- struct ptlrpc_connection *connection;
- lnet_handle_me_t reply_me_h;
- lnet_md_t reply_md;
+ int rc;
+ int rc2;
+ int mpflag = 0;
+ struct lnet_handle_md bulk_cookie;
+ struct ptlrpc_connection *connection;
+ struct lnet_handle_me reply_me_h;
+ struct lnet_md reply_md;
struct obd_import *imp = request->rq_import;
struct obd_device *obd = imp->imp_obd;
- ENTRY;
+ ENTRY;
+
+ LNetInvalidateMDHandle(&bulk_cookie);
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_DROP_RPC))
RETURN(0);
mpflag = cfs_memory_pressure_get_and_set();
rc = sptlrpc_cli_wrap_request(request);
- if (rc == -ENOMEM)
- /* set rq_sent so that this request is treated
- * as a delayed send in the upper layers */
- request->rq_sent = cfs_time_current_sec();
if (rc)
GOTO(out, rc);
- /* bulk register should be done after wrap_request() */
- if (request->rq_bulk != NULL) {
- rc = ptlrpc_register_bulk (request);
- if (rc != 0)
- GOTO(out, rc);
- }
+ /* bulk register should be done after wrap_request() */
+ if (request->rq_bulk != NULL) {
+ rc = ptlrpc_register_bulk (request);
+ if (rc != 0)
+ GOTO(out, rc);
+ /*
+ * All the mds in the request will have the same cpt
+ * encoded in the cookie. So we can just get the first
+ * one.
+ */
+ bulk_cookie = request->rq_bulk->bd_mds[0];
+ }
if (!noreply) {
LASSERT (request->rq_replen != 0);
OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_SEND, request->rq_timeout + 5);
- do_gettimeofday(&request->rq_sent_tv);
- request->rq_sent = cfs_time_current_sec();
+ request->rq_sent_ns = ktime_get_real();
+ request->rq_sent = ktime_get_real_seconds();
/* We give the server rq_timeout secs to process the req, and
add the network latency for our local timeout. */
request->rq_deadline = request->rq_sent + request->rq_timeout +
ptlrpc_pinger_sending_on_import(imp);
- DEBUG_REQ(D_INFO, request, "send flg=%x",
- lustre_msg_get_flags(request->rq_reqmsg));
- rc = ptl_send_buf(&request->rq_req_md_h,
- request->rq_reqbuf, request->rq_reqdata_len,
- LNET_NOACK_REQ, &request->rq_req_cbid,
- connection,
- request->rq_request_portal,
- request->rq_xid, 0);
+ DEBUG_REQ(D_INFO, request, "send flg=%x",
+ lustre_msg_get_flags(request->rq_reqmsg));
+ rc = ptl_send_buf(&request->rq_req_md_h,
+ request->rq_reqbuf, request->rq_reqdata_len,
+ LNET_NOACK_REQ, &request->rq_req_cbid,
+ LNET_NID_ANY, connection->c_peer,
+ request->rq_request_portal,
+ request->rq_xid, 0, &bulk_cookie);
if (likely(rc == 0))
GOTO(out, rc);
* the chance to have long unlink to sluggish net is smaller here. */
ptlrpc_unregister_bulk(request, 0);
out:
- if (request->rq_memalloc)
- cfs_memory_pressure_restore(mpflag);
- return rc;
+ if (rc == -ENOMEM) {
+ /* set rq_sent so that this request is treated
+ * as a delayed send in the upper layers */
+ request->rq_sent = ktime_get_real_seconds();
+ }
+
+ if (request->rq_memalloc)
+ cfs_memory_pressure_restore(mpflag);
+
+ return rc;
}
EXPORT_SYMBOL(ptl_send_rpc);
*/
int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd)
{
- struct ptlrpc_service *service = rqbd->rqbd_svcpt->scp_service;
- static lnet_process_id_t match_id = {LNET_NID_ANY, LNET_PID_ANY};
- int rc;
- lnet_md_t md;
- lnet_handle_me_t me_h;
+ struct ptlrpc_service *service = rqbd->rqbd_svcpt->scp_service;
+ static struct lnet_process_id match_id = {
+ .nid = LNET_NID_ANY,
+ .pid = LNET_PID_ANY
+ };
+ int rc;
+ struct lnet_md md;
+ struct lnet_handle_me me_h;
CDEBUG(D_NET, "LNetMEAttach: portal %d\n",
service->srv_req_portal);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff