*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_RPC
-#ifndef __KERNEL__
-# include <liblustre.h>
-#else
-# include <libcfs/libcfs.h>
-# ifdef __mips64__
-# include <linux/kernel.h>
-# endif
-#endif
-
+#include <libcfs/libcfs.h>
+#include <linux/kernel.h>
+#include <linux/delay.h>
#include <obd_class.h>
#include <lustre_net.h>
#include <lustre_sec.h>
#include "ptlrpc_internal.h"
-lnet_handle_eq_t ptlrpc_eq_h;
+struct lnet_handle_eq ptlrpc_eq_h;
/*
* Client's outgoing request callback
*/
-void request_out_callback(lnet_event_t *ev)
+void request_out_callback(struct lnet_event *ev)
{
- struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
- struct ptlrpc_request *req = cbid->cbid_arg;
- ENTRY;
+ struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
+ struct ptlrpc_request *req = cbid->cbid_arg;
+ bool wakeup = false;
+ ENTRY;
- LASSERT (ev->type == LNET_EVENT_SEND ||
- ev->type == LNET_EVENT_UNLINK);
- LASSERT (ev->unlinked);
+ LASSERT(ev->type == LNET_EVENT_SEND || ev->type == LNET_EVENT_UNLINK);
+ LASSERT(ev->unlinked);
- DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status);
+ DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status);
- sptlrpc_request_out_callback(req);
- spin_lock(&req->rq_lock);
- req->rq_real_sent = cfs_time_current_sec();
- if (ev->unlinked)
- req->rq_req_unlink = 0;
+ sptlrpc_request_out_callback(req);
- if (ev->type == LNET_EVENT_UNLINK || ev->status != 0) {
+ spin_lock(&req->rq_lock);
+ req->rq_real_sent = ktime_get_real_seconds();
+ req->rq_req_unlinked = 1;
+ /* reply_in_callback happened before request_out_callback? */
+ if (req->rq_reply_unlinked)
+ wakeup = true;
+
+ if (ev->type == LNET_EVENT_UNLINK || ev->status != 0) {
+ /* Failed send: make it seem like the reply timed out, just
+ * like failing sends in client.c does currently... */
+ req->rq_net_err = 1;
+ wakeup = true;
+ }
- /* Failed send: make it seem like the reply timed out, just
- * like failing sends in client.c does currently... */
+ if (wakeup)
+ ptlrpc_client_wake_req(req);
- req->rq_net_err = 1;
- ptlrpc_client_wake_req(req);
- }
spin_unlock(&req->rq_lock);
- ptlrpc_req_finished(req);
-
- EXIT;
+ ptlrpc_req_finished(req);
+ EXIT;
}
/*
* Client's incoming reply callback
*/
-void reply_in_callback(lnet_event_t *ev)
+void reply_in_callback(struct lnet_event *ev)
{
struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
struct ptlrpc_request *req = cbid->cbid_arg;
spin_lock(&req->rq_lock);
- req->rq_receiving_reply = 0;
- req->rq_early = 0;
- if (ev->unlinked)
- req->rq_reply_unlink = 0;
+ req->rq_receiving_reply = 0;
+ req->rq_early = 0;
+ if (ev->unlinked)
+ req->rq_reply_unlinked = 1;
if (ev->status)
goto out_wake;
if (ev->mlength < ev->rlength ) {
CDEBUG(D_RPCTRACE, "truncate req %p rpc %d - %d+%d\n", req,
req->rq_replen, ev->rlength, ev->offset);
- req->rq_reply_truncate = 1;
+ req->rq_reply_truncated = 1;
req->rq_replied = 1;
req->rq_status = -EOVERFLOW;
req->rq_nob_received = ev->rlength + ev->offset;
goto out_wake;
}
- if ((ev->offset == 0) &&
- ((lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT))) {
- /* Early reply */
- DEBUG_REQ(D_ADAPTTO, req,
- "Early reply received: mlen=%u offset=%d replen=%d "
- "replied=%d unlinked=%d", ev->mlength, ev->offset,
- req->rq_replen, req->rq_replied, ev->unlinked);
+ if ((ev->offset == 0) &&
+ ((lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT))) {
+ /* Early reply */
+ DEBUG_REQ(D_ADAPTTO, req,
+ "Early reply received, mlen=%u offset=%d replen=%d replied=%d unlinked=%d",
+ ev->mlength, ev->offset,
+ req->rq_replen, req->rq_replied, ev->unlinked);
- req->rq_early_count++; /* number received, client side */
+ req->rq_early_count++; /* number received, client side */
- if (req->rq_replied) /* already got the real reply */
- goto out_wake;
+ /* already got the real reply or buffers are already unlinked */
+ if (req->rq_replied ||
+ req->rq_reply_unlinked == 1)
+ goto out_wake;
req->rq_early = 1;
req->rq_reply_off = ev->offset;
ev->mlength, ev->offset, req->rq_replen);
}
- req->rq_import->imp_last_reply_time = cfs_time_current_sec();
+ if (lustre_msg_get_opc(req->rq_reqmsg) != OBD_PING)
+ req->rq_import->imp_last_reply_time = ktime_get_real_seconds();
out_wake:
- /* NB don't unlock till after wakeup; req can disappear under us
- * since we don't have our own ref */
- ptlrpc_client_wake_req(req);
+ /* NB don't unlock till after wakeup; req can disappear under us
+ * since we don't have our own ref */
+ ptlrpc_client_wake_req(req);
spin_unlock(&req->rq_lock);
EXIT;
}
/*
* Client's bulk has been written/read
*/
-void client_bulk_callback (lnet_event_t *ev)
+void client_bulk_callback(struct lnet_event *ev)
{
struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
struct ptlrpc_bulk_desc *desc = cbid->cbid_arg;
struct ptlrpc_request *req;
ENTRY;
- LASSERT ((desc->bd_type == BULK_PUT_SINK &&
- ev->type == LNET_EVENT_PUT) ||
- (desc->bd_type == BULK_GET_SOURCE &&
- ev->type == LNET_EVENT_GET) ||
- ev->type == LNET_EVENT_UNLINK);
- LASSERT (ev->unlinked);
+ LASSERT((ptlrpc_is_bulk_put_sink(desc->bd_type) &&
+ ev->type == LNET_EVENT_PUT) ||
+ (ptlrpc_is_bulk_get_source(desc->bd_type) &&
+ ev->type == LNET_EVENT_GET) ||
+ ev->type == LNET_EVENT_UNLINK);
+ LASSERT(ev->unlinked);
if (CFS_FAIL_CHECK_ORSET(OBD_FAIL_PTLRPC_CLIENT_BULK_CB, CFS_FAIL_ONCE))
ev->status = -EIO;
static void ptlrpc_req_add_history(struct ptlrpc_service_part *svcpt,
struct ptlrpc_request *req)
{
- __u64 sec = req->rq_arrival_time.tv_sec;
- __u32 usec = req->rq_arrival_time.tv_usec >> 4; /* usec / 16 */
- __u64 new_seq;
+ u64 sec = req->rq_arrival_time.tv_sec;
+ u32 usec = req->rq_arrival_time.tv_nsec / NSEC_PER_USEC / 16; /* usec / 16 */
+ u64 new_seq;
/* set sequence ID for request and add it to history list,
* it must be called with hold svcpt::scp_lock */
req->rq_history_seq = new_seq;
- cfs_list_add_tail(&req->rq_history_list, &svcpt->scp_hist_reqs);
+ list_add_tail(&req->rq_history_list, &svcpt->scp_hist_reqs);
}
/*
* Server's incoming request callback
*/
-void request_in_callback(lnet_event_t *ev)
+void request_in_callback(struct lnet_event *ev)
{
struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
struct ptlrpc_request_buffer_desc *rqbd = cbid->cbid_arg;
/* We moaned above already... */
return;
}
- req = ptlrpc_request_cache_alloc(ALLOC_ATOMIC_TRY);
+ req = ptlrpc_request_cache_alloc(GFP_ATOMIC);
if (req == NULL) {
CERROR("Can't allocate incoming request descriptor: "
"Dropping %s RPC from %s\n",
}
}
- /* NB we ABSOLUTELY RELY on req being zeroed, so pointers are NULL,
- * flags are reset and scalars are zero. We only set the message
- * size to non-zero if this was a successful receive. */
- req->rq_xid = ev->match_bits;
- req->rq_reqbuf = ev->md.start + ev->offset;
+ ptlrpc_srv_req_init(req);
+ /* NB we ABSOLUTELY RELY on req being zeroed, so pointers are NULL,
+ * flags are reset and scalars are zero. We only set the message
+ * size to non-zero if this was a successful receive. */
+ req->rq_xid = ev->match_bits;
+ req->rq_reqbuf = ev->md.start + ev->offset;
if (ev->type == LNET_EVENT_PUT && ev->status == 0)
req->rq_reqdata_len = ev->mlength;
- do_gettimeofday(&req->rq_arrival_time);
+ ktime_get_real_ts64(&req->rq_arrival_time);
+ /* Multi-Rail: keep track of both initiator and source NID. */
req->rq_peer = ev->initiator;
+ req->rq_source = ev->source;
req->rq_self = ev->target.nid;
req->rq_rqbd = rqbd;
- req->rq_phase = RQ_PHASE_NEW;
- spin_lock_init(&req->rq_lock);
- CFS_INIT_LIST_HEAD(&req->rq_timed_list);
- CFS_INIT_LIST_HEAD(&req->rq_exp_list);
- atomic_set(&req->rq_refcount, 1);
+ req->rq_phase = RQ_PHASE_NEW;
if (ev->type == LNET_EVENT_PUT)
- CDEBUG(D_INFO, "incoming req@%p x"LPU64" msgsize %u\n",
+ CDEBUG(D_INFO, "incoming req@%p x%llu msgsize %u\n",
req, req->rq_xid, ev->mlength);
- CDEBUG(D_RPCTRACE, "peer: %s\n", libcfs_id2str(req->rq_peer));
+ CDEBUG(D_RPCTRACE, "peer: %s (source: %s)\n",
+ libcfs_id2str(req->rq_peer), libcfs_id2str(req->rq_source));
spin_lock(&svcpt->scp_lock);
rqbd->rqbd_refcount++;
}
- cfs_list_add_tail(&req->rq_list, &svcpt->scp_req_incoming);
+ list_add_tail(&req->rq_list, &svcpt->scp_req_incoming);
svcpt->scp_nreqs_incoming++;
/* NB everything can disappear under us once the request
/*
* Server's outgoing reply callback
*/
-void reply_out_callback(lnet_event_t *ev)
+void reply_out_callback(struct lnet_event *ev)
{
struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
struct ptlrpc_reply_state *rs = cbid->cbid_arg;
rs->rs_on_net = 0;
if (!rs->rs_no_ack ||
rs->rs_transno <=
- rs->rs_export->exp_obd->obd_last_committed)
+ rs->rs_export->exp_obd->obd_last_committed ||
+ list_empty(&rs->rs_obd_list))
ptlrpc_schedule_difficult_reply(rs);
spin_unlock(&rs->rs_lock);
/*
* Server's bulk completion callback
*/
-void server_bulk_callback (lnet_event_t *ev)
+void server_bulk_callback(struct lnet_event *ev)
{
struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
struct ptlrpc_bulk_desc *desc = cbid->cbid_arg;
LASSERT(ev->type == LNET_EVENT_SEND ||
ev->type == LNET_EVENT_UNLINK ||
- (desc->bd_type == BULK_PUT_SOURCE &&
+ (ptlrpc_is_bulk_put_source(desc->bd_type) &&
ev->type == LNET_EVENT_ACK) ||
- (desc->bd_type == BULK_GET_SINK &&
+ (ptlrpc_is_bulk_get_sink(desc->bd_type) &&
ev->type == LNET_EVENT_REPLY));
CDEBUG((ev->status == 0) ? D_NET : D_ERROR,
}
#endif
-static void ptlrpc_master_callback(lnet_event_t *ev)
+static void ptlrpc_master_callback(struct lnet_event *ev)
{
struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
- void (*callback)(lnet_event_t *ev) = cbid->cbid_fn;
+ void (*callback)(struct lnet_event *ev) = cbid->cbid_fn;
/* Honestly, it's best to find out early. */
LASSERT (cbid->cbid_arg != LP_POISON);
callback (ev);
}
-int ptlrpc_uuid_to_peer (struct obd_uuid *uuid,
- lnet_process_id_t *peer, lnet_nid_t *self)
+int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
+ struct lnet_process_id *peer, lnet_nid_t *self)
{
- int best_dist = 0;
- __u32 best_order = 0;
- int count = 0;
- int rc = -ENOENT;
- int portals_compatibility;
- int dist;
- __u32 order;
- lnet_nid_t dst_nid;
- lnet_nid_t src_nid;
-
- portals_compatibility = LNetCtl(IOC_LIBCFS_PORTALS_COMPATIBILITY, NULL);
-
- peer->pid = LUSTRE_SRV_LNET_PID;
-
- /* Choose the matching UUID that's closest */
- while (lustre_uuid_to_peer(uuid->uuid, &dst_nid, count++) == 0) {
- dist = LNetDist(dst_nid, &src_nid, &order);
- if (dist < 0)
- continue;
-
- if (dist == 0) { /* local! use loopback LND */
- peer->nid = *self = LNET_MKNID(LNET_MKNET(LOLND, 0), 0);
- rc = 0;
- break;
- }
+ int best_dist = 0;
+ __u32 best_order = 0;
+ int count = 0;
+ int rc = -ENOENT;
+ int dist;
+ __u32 order;
+ lnet_nid_t dst_nid;
+ lnet_nid_t src_nid;
+
+ peer->pid = LNET_PID_LUSTRE;
+
+ /* Choose the matching UUID that's closest */
+ while (lustre_uuid_to_peer(uuid->uuid, &dst_nid, count++) == 0) {
+ if (peer->nid != LNET_NID_ANY && LNET_NIDADDR(peer->nid) == 0 &&
+ LNET_NIDNET(dst_nid) != LNET_NIDNET(peer->nid))
+ continue;
+
+ dist = LNetDist(dst_nid, &src_nid, &order);
+ if (dist < 0)
+ continue;
+
+ if (dist == 0) { /* local! use loopback LND */
+ peer->nid = *self = LNET_MKNID(LNET_MKNET(LOLND, 0), 0);
+ rc = 0;
+ break;
+ }
- if (rc < 0 ||
- dist < best_dist ||
- (dist == best_dist && order < best_order)) {
- best_dist = dist;
- best_order = order;
-
- if (portals_compatibility > 1) {
- /* Strong portals compatibility: Zero the nid's
- * NET, so if I'm reading new config logs, or
- * getting configured by (new) lconf I can
- * still talk to old servers. */
- dst_nid = LNET_MKNID(0, LNET_NIDADDR(dst_nid));
- src_nid = LNET_MKNID(0, LNET_NIDADDR(src_nid));
- }
- peer->nid = dst_nid;
- *self = src_nid;
- rc = 0;
- }
- }
+ if (rc < 0 ||
+ dist < best_dist ||
+ (dist == best_dist && order < best_order)) {
+ best_dist = dist;
+ best_order = order;
- CDEBUG(D_NET,"%s->%s\n", uuid->uuid, libcfs_id2str(*peer));
- return rc;
+ peer->nid = dst_nid;
+ *self = src_nid;
+ rc = 0;
+ }
+ }
+
+ CDEBUG(D_NET, "%s->%s\n", uuid->uuid, libcfs_id2str(*peer));
+ return rc;
}
void ptlrpc_ni_fini(void)
{
- wait_queue_head_t waitq;
- struct l_wait_info lwi;
int rc;
int retries;
CWARN("Event queue still busy\n");
/* Wait for a bit */
- init_waitqueue_head(&waitq);
- lwi = LWI_TIMEOUT(cfs_time_seconds(2), NULL, NULL);
- l_wait_event(waitq, 0, &lwi);
+ ssleep(2);
break;
}
}
lnet_pid_t ptl_get_pid(void)
{
- lnet_pid_t pid;
-
-#ifndef __KERNEL__
- pid = getpid();
-#else
- pid = LUSTRE_SRV_LNET_PID;
-#endif
- return pid;
+ return LNET_PID_LUSTRE;
}
int ptlrpc_ni_init(void)
rc = LNetNIInit(pid);
if (rc < 0) {
CDEBUG (D_NET, "Can't init network interface: %d\n", rc);
- return (-ENOENT);
+ return rc;
}
/* CAVEAT EMPTOR: how we process portals events is _radically_
* different depending on... */
-#ifdef __KERNEL__
/* kernel LNet calls our master callback when there are new event,
* because we are guaranteed to get every event via callback,
* so we just set EQ size to 0 to avoid overhread of serializing
* enqueue/dequeue operations in LNet. */
rc = LNetEQAlloc(0, ptlrpc_master_callback, &ptlrpc_eq_h);
-#else
- /* liblustre calls the master callback when it removes events from the
- * event queue. The event queue has to be big enough not to drop
- * anything */
- rc = LNetEQAlloc(10240, LNET_EQ_HANDLER_NONE, &ptlrpc_eq_h);
-#endif
if (rc == 0)
return 0;
CERROR ("Failed to allocate event queue: %d\n", rc);
LNetNIFini();
- return (-ENOMEM);
+ return rc;
}
-#ifndef __KERNEL__
-CFS_LIST_HEAD(liblustre_wait_callbacks);
-CFS_LIST_HEAD(liblustre_idle_callbacks);
-void *liblustre_services_callback;
-
-void *
-liblustre_register_waitidle_callback (cfs_list_t *callback_list,
- const char *name,
- int (*fn)(void *arg), void *arg)
-{
- struct liblustre_wait_callback *llwc;
-
- OBD_ALLOC(llwc, sizeof(*llwc));
- LASSERT (llwc != NULL);
-
- llwc->llwc_name = name;
- llwc->llwc_fn = fn;
- llwc->llwc_arg = arg;
- cfs_list_add_tail(&llwc->llwc_list, callback_list);
-
- return (llwc);
-}
-
-void
-liblustre_deregister_waitidle_callback (void *opaque)
-{
- struct liblustre_wait_callback *llwc = opaque;
-
- cfs_list_del(&llwc->llwc_list);
- OBD_FREE(llwc, sizeof(*llwc));
-}
-
-void *
-liblustre_register_wait_callback (const char *name,
- int (*fn)(void *arg), void *arg)
-{
- return liblustre_register_waitidle_callback(&liblustre_wait_callbacks,
- name, fn, arg);
-}
-
-void
-liblustre_deregister_wait_callback (void *opaque)
-{
- liblustre_deregister_waitidle_callback(opaque);
-}
-
-void *
-liblustre_register_idle_callback (const char *name,
- int (*fn)(void *arg), void *arg)
-{
- return liblustre_register_waitidle_callback(&liblustre_idle_callbacks,
- name, fn, arg);
-}
-
-void
-liblustre_deregister_idle_callback (void *opaque)
-{
- liblustre_deregister_waitidle_callback(opaque);
-}
-
-int
-liblustre_check_events (int timeout)
-{
- lnet_event_t ev;
- int rc;
- int i;
- ENTRY;
-
- rc = LNetEQPoll(&ptlrpc_eq_h, 1, timeout * 1000, &ev, &i);
- if (rc == 0)
- RETURN(0);
-
- LASSERT (rc == -EOVERFLOW || rc == 1);
-
- /* liblustre: no asynch callback so we can't afford to miss any
- * events... */
- if (rc == -EOVERFLOW) {
- CERROR ("Dropped an event!!!\n");
- abort();
- }
-
- ptlrpc_master_callback (&ev);
- RETURN(1);
-}
-
-int liblustre_waiting = 0;
-
-int
-liblustre_wait_event (int timeout)
-{
- cfs_list_t *tmp;
- struct liblustre_wait_callback *llwc;
- int found_something = 0;
-
- /* single threaded recursion check... */
- liblustre_waiting = 1;
-
- for (;;) {
- /* Deal with all pending events */
- while (liblustre_check_events(0))
- found_something = 1;
-
- /* Give all registered callbacks a bite at the cherry */
- cfs_list_for_each(tmp, &liblustre_wait_callbacks) {
- llwc = cfs_list_entry(tmp,
- struct liblustre_wait_callback,
- llwc_list);
-
- if (llwc->llwc_fn(llwc->llwc_arg))
- found_something = 1;
- }
-
- if (found_something || timeout == 0)
- break;
-
- /* Nothing so far, but I'm allowed to block... */
- found_something = liblustre_check_events(timeout);
- if (!found_something) /* still nothing */
- break; /* I timed out */
- }
-
- liblustre_waiting = 0;
-
- return found_something;
-}
-
-void
-liblustre_wait_idle(void)
-{
- static int recursed = 0;
-
- cfs_list_t *tmp;
- struct liblustre_wait_callback *llwc;
- int idle = 0;
-
- LASSERT(!recursed);
- recursed = 1;
-
- do {
- liblustre_wait_event(0);
-
- idle = 1;
-
- cfs_list_for_each(tmp, &liblustre_idle_callbacks) {
- llwc = cfs_list_entry(tmp,
- struct liblustre_wait_callback,
- llwc_list);
-
- if (!llwc->llwc_fn(llwc->llwc_arg)) {
- idle = 0;
- break;
- }
- }
-
- } while (!idle);
-
- recursed = 0;
-}
-
-#endif /* __KERNEL__ */
int ptlrpc_init_portals(void)
{
if (rc != 0) {
CERROR("network initialisation failed\n");
- return -EIO;
+ return rc;
}
-#ifndef __KERNEL__
- liblustre_services_callback =
- liblustre_register_wait_callback("liblustre_check_services",
- &liblustre_check_services,
- NULL);
- init_completion_module(liblustre_wait_event);
-#endif
rc = ptlrpcd_addref();
if (rc == 0)
return 0;
CERROR("rpcd initialisation failed\n");
-#ifndef __KERNEL__
- liblustre_deregister_wait_callback(liblustre_services_callback);
-#endif
ptlrpc_ni_fini();
return rc;
}
void ptlrpc_exit_portals(void)
{
-#ifndef __KERNEL__
- liblustre_deregister_wait_callback(liblustre_services_callback);
-#endif
ptlrpcd_decref();
ptlrpc_ni_fini();
}