* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2010, 2012, Intel Corporation.
+ * Copyright (c) 2010, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
static void ptlrpc_server_hpreq_fini(struct ptlrpc_request *req);
static void ptlrpc_at_remove_timed(struct ptlrpc_request *req);
-static CFS_LIST_HEAD(ptlrpc_all_services);
-spinlock_t ptlrpc_all_services_lock;
+/** Holds a list of all PTLRPC services */
+CFS_LIST_HEAD(ptlrpc_all_services);
+/** Used to protect the \e ptlrpc_all_services list */
+struct mutex ptlrpc_all_services_mutex;
struct ptlrpc_request_buffer_desc *
ptlrpc_alloc_rqbd(struct ptlrpc_service_part *svcpt)
struct ptlrpc_hr_thread {
int hrt_id; /* thread ID */
spinlock_t hrt_lock;
- cfs_waitq_t hrt_waitq;
+ wait_queue_head_t hrt_waitq;
cfs_list_t hrt_queue; /* RS queue */
struct ptlrpc_hr_partition *hrt_partition;
};
/* CPU partition table, it's just cfs_cpt_table for now */
struct cfs_cpt_table *hr_cpt_table;
/** controller sleep waitq */
- cfs_waitq_t hr_waitq;
+ wait_queue_head_t hr_waitq;
unsigned int hr_stopping;
/** roundrobin rotor for non-affinity service */
unsigned int hr_rotor;
cfs_list_splice_init(&b->rsb_replies, &hrt->hrt_queue);
spin_unlock(&hrt->hrt_lock);
- cfs_waitq_signal(&hrt->hrt_waitq);
+ wake_up(&hrt->hrt_waitq);
b->rsb_n_replies = 0;
}
}
cfs_list_add_tail(&rs->rs_list, &hrt->hrt_queue);
spin_unlock(&hrt->hrt_lock);
- cfs_waitq_signal(&hrt->hrt_waitq);
+ wake_up(&hrt->hrt_waitq);
EXIT;
#else
cfs_list_add_tail(&rs->rs_list, &rs->rs_svcpt->scp_rep_queue);
{
ENTRY;
- LASSERT_SPIN_LOCKED(&rs->rs_svcpt->scp_rep_lock);
- LASSERT_SPIN_LOCKED(&rs->rs_lock);
- LASSERT (rs->rs_difficult);
- rs->rs_scheduled_ever = 1; /* flag any notification attempt */
+ LASSERT(spin_is_locked(&rs->rs_svcpt->scp_rep_lock));
+ LASSERT(spin_is_locked(&rs->rs_lock));
+ LASSERT (rs->rs_difficult);
+ rs->rs_scheduled_ever = 1; /* flag any notification attempt */
- if (rs->rs_scheduled) { /* being set up or already notified */
- EXIT;
- return;
- }
+ if (rs->rs_scheduled) { /* being set up or already notified */
+ EXIT;
+ return;
+ }
- rs->rs_scheduled = 1;
- cfs_list_del_init(&rs->rs_list);
- ptlrpc_dispatch_difficult_reply(rs);
- EXIT;
+ rs->rs_scheduled = 1;
+ cfs_list_del_init(&rs->rs_list);
+ ptlrpc_dispatch_difficult_reply(rs);
+ EXIT;
}
EXPORT_SYMBOL(ptlrpc_schedule_difficult_reply);
svcpt->scp_at_check = 1;
svcpt->scp_at_checktime = cfs_time_current();
- cfs_waitq_signal(&svcpt->scp_waitq);
+ wake_up(&svcpt->scp_waitq);
}
static void
svc->srv_nthrs_cpt_init = init;
if (nthrs * svc->srv_ncpts > tc->tc_nthrs_max) {
- LCONSOLE_WARN("%s: This service may have more threads (%d) "
- "than the given soft limit (%d)\n",
- svc->srv_name, nthrs * svc->srv_ncpts,
- tc->tc_nthrs_max);
+ CDEBUG(D_OTHER, "%s: This service may have more threads (%d) "
+ "than the given soft limit (%d)\n",
+ svc->srv_name, nthrs * svc->srv_ncpts,
+ tc->tc_nthrs_max);
}
#endif
}
CFS_INIT_LIST_HEAD(&svcpt->scp_rqbd_idle);
CFS_INIT_LIST_HEAD(&svcpt->scp_rqbd_posted);
CFS_INIT_LIST_HEAD(&svcpt->scp_req_incoming);
- cfs_waitq_init(&svcpt->scp_waitq);
+ init_waitqueue_head(&svcpt->scp_waitq);
/* history request & rqbd list */
CFS_INIT_LIST_HEAD(&svcpt->scp_hist_reqs);
CFS_INIT_LIST_HEAD(&svcpt->scp_hist_rqbds);
/* acitve requests and hp requests */
spin_lock_init(&svcpt->scp_req_lock);
- CFS_INIT_LIST_HEAD(&svcpt->scp_req_pending);
- CFS_INIT_LIST_HEAD(&svcpt->scp_hreq_pending);
/* reply states */
spin_lock_init(&svcpt->scp_rep_lock);
CFS_INIT_LIST_HEAD(&svcpt->scp_rep_queue);
#endif
CFS_INIT_LIST_HEAD(&svcpt->scp_rep_idle);
- cfs_waitq_init(&svcpt->scp_rep_waitq);
+ init_waitqueue_head(&svcpt->scp_rep_waitq);
cfs_atomic_set(&svcpt->scp_nreps_difficult, 0);
/* adaptive timeout */
*/
struct ptlrpc_service *
ptlrpc_register_service(struct ptlrpc_service_conf *conf,
- cfs_proc_dir_entry_t *proc_entry)
+ struct proc_dir_entry *proc_entry)
{
struct ptlrpc_service_cpt_conf *cconf = &conf->psc_cpt;
struct ptlrpc_service *service;
CFS_INIT_LIST_HEAD(&service->srv_list); /* for safty of cleanup */
/* buffer configuration */
- service->srv_nbuf_per_group = test_req_buffer_pressure ? 1 :
- max(conf->psc_buf.bc_nbufs /
- service->srv_ncpts, 1U);
+ service->srv_nbuf_per_group = test_req_buffer_pressure ?
+ 1 : conf->psc_buf.bc_nbufs;
service->srv_max_req_size = conf->psc_buf.bc_req_max_size +
SPTLRPC_MAX_PAYLOAD;
service->srv_buf_size = conf->psc_buf.bc_buf_size;
rc = LNetSetLazyPortal(service->srv_req_portal);
LASSERT(rc == 0);
- spin_lock(&ptlrpc_all_services_lock);
- cfs_list_add (&service->srv_list, &ptlrpc_all_services);
- spin_unlock(&ptlrpc_all_services_lock);
+ mutex_lock(&ptlrpc_all_services_mutex);
+ cfs_list_add (&service->srv_list, &ptlrpc_all_services);
+ mutex_unlock(&ptlrpc_all_services_mutex);
+
+ if (proc_entry != NULL)
+ ptlrpc_lprocfs_register_service(proc_entry, service);
- if (proc_entry != NULL)
- ptlrpc_lprocfs_register_service(proc_entry, service);
+ rc = ptlrpc_service_nrs_setup(service);
+ if (rc != 0)
+ GOTO(failed, rc);
- CDEBUG(D_NET, "%s: Started, listening on portal %d\n",
- service->srv_name, service->srv_req_portal);
+ CDEBUG(D_NET, "%s: Started, listening on portal %d\n",
+ service->srv_name, service->srv_req_portal);
#ifdef __KERNEL__
rc = ptlrpc_start_threads(service);
sptlrpc_svc_ctx_decref(req);
if (req != &req->rq_rqbd->rqbd_req) {
- /* NB request buffers use an embedded
- * req if the incoming req unlinked the
- * MD; this isn't one of them! */
- OBD_FREE(req, sizeof(*req));
- }
+ /* NB request buffers use an embedded
+ * req if the incoming req unlinked the
+ * MD; this isn't one of them! */
+ ptlrpc_request_cache_free(req);
+ }
}
/**
}
}
+/** Change request export and move hp request from old export to new */
+void ptlrpc_request_change_export(struct ptlrpc_request *req,
+ struct obd_export *export)
+{
+ if (req->rq_export != NULL) {
+ LASSERT(!list_empty(&req->rq_exp_list));
+ /* remove rq_exp_list from last export */
+ spin_lock_bh(&req->rq_export->exp_rpc_lock);
+ list_del_init(&req->rq_exp_list);
+ spin_unlock_bh(&req->rq_export->exp_rpc_lock);
+ /* export has one reference already, so it`s safe to
+ * add req to export queue here and get another
+ * reference for request later */
+ spin_lock_bh(&export->exp_rpc_lock);
+ if (req->rq_ops != NULL) /* hp request */
+ list_add(&req->rq_exp_list, &export->exp_hp_rpcs);
+ else
+ list_add(&req->rq_exp_list, &export->exp_reg_rpcs);
+ spin_unlock_bh(&export->exp_rpc_lock);
+
+ class_export_rpc_dec(req->rq_export);
+ class_export_put(req->rq_export);
+ }
+
+ /* request takes one export refcount */
+ req->rq_export = class_export_get(export);
+ class_export_rpc_inc(export);
+
+ return;
+}
+
/**
* to finish a request: stop sending more early replies, and release
- * the request. should be called after we finished handling the request.
+ * the request.
*/
static void ptlrpc_server_finish_request(struct ptlrpc_service_part *svcpt,
struct ptlrpc_request *req)
{
ptlrpc_server_hpreq_fini(req);
+ if (req->rq_session.lc_thread != NULL) {
+ lu_context_exit(&req->rq_session);
+ lu_context_fini(&req->rq_session);
+ }
+
+ ptlrpc_server_drop_request(req);
+}
+
+/**
+ * to finish a active request: stop sending more early replies, and release
+ * the request. should be called after we finished handling the request.
+ */
+static void ptlrpc_server_finish_active_request(
+ struct ptlrpc_service_part *svcpt,
+ struct ptlrpc_request *req)
+{
spin_lock(&svcpt->scp_req_lock);
+ ptlrpc_nrs_req_stop_nolock(req);
svcpt->scp_nreqs_active--;
if (req->rq_hp)
svcpt->scp_nhreqs_active--;
spin_unlock(&svcpt->scp_req_lock);
- ptlrpc_server_drop_request(req);
+ ptlrpc_nrs_req_finalize(req);
+
+ if (req->rq_export != NULL)
+ class_export_rpc_dec(req->rq_export);
+
+ ptlrpc_server_finish_request(svcpt, req);
}
/**
}
newdl = cfs_time_current_sec() + at_get(&svcpt->scp_at_estimate);
- OBD_ALLOC(reqcopy, sizeof *reqcopy);
- if (reqcopy == NULL)
- RETURN(-ENOMEM);
- OBD_ALLOC_LARGE(reqmsg, req->rq_reqlen);
- if (!reqmsg) {
- OBD_FREE(reqcopy, sizeof *reqcopy);
- RETURN(-ENOMEM);
- }
+ reqcopy = ptlrpc_request_cache_alloc(__GFP_IO);
+ if (reqcopy == NULL)
+ RETURN(-ENOMEM);
+ OBD_ALLOC_LARGE(reqmsg, req->rq_reqlen);
+ if (!reqmsg)
+ GOTO(out_free, rc = -ENOMEM);
*reqcopy = *req;
reqcopy->rq_reply_state = NULL;
GOTO(out, rc = -ENODEV);
/* RPC ref */
- class_export_rpc_get(reqcopy->rq_export);
+ class_export_rpc_inc(reqcopy->rq_export);
if (reqcopy->rq_export->exp_obd &&
reqcopy->rq_export->exp_obd->obd_fail)
GOTO(out_put, rc = -ENODEV);
ptlrpc_req_drop_rs(reqcopy);
out_put:
- class_export_rpc_put(reqcopy->rq_export);
- class_export_put(reqcopy->rq_export);
+ class_export_rpc_dec(reqcopy->rq_export);
+ class_export_put(reqcopy->rq_export);
out:
- sptlrpc_svc_ctx_decref(reqcopy);
- OBD_FREE_LARGE(reqmsg, req->rq_reqlen);
- OBD_FREE(reqcopy, sizeof *reqcopy);
- RETURN(rc);
+ sptlrpc_svc_ctx_decref(reqcopy);
+ OBD_FREE_LARGE(reqmsg, req->rq_reqlen);
+out_free:
+ ptlrpc_request_cache_free(reqcopy);
+ RETURN(rc);
}
/* Send early replies to everybody expiring within at_early_margin
RETURN(1); /* return "did_something" for liblustre */
}
+/* Check if we are already handling earlier incarnation of this request.
+ * Called under &req->rq_export->exp_rpc_lock locked */
+static int ptlrpc_server_check_resend_in_progress(struct ptlrpc_request *req)
+{
+ struct ptlrpc_request *tmp = NULL;
+
+ if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) ||
+ (cfs_atomic_read(&req->rq_export->exp_rpc_count) == 0))
+ return 0;
+
+ /* bulk request are aborted upon reconnect, don't try to
+ * find a match */
+ if (req->rq_bulk_write || req->rq_bulk_read)
+ return 0;
+
+ /* This list should not be longer than max_requests in
+ * flights on the client, so it is not all that long.
+ * Also we only hit this codepath in case of a resent
+ * request which makes it even more rarely hit */
+ cfs_list_for_each_entry(tmp, &req->rq_export->exp_reg_rpcs,
+ rq_exp_list) {
+ /* Found duplicate one */
+ if (tmp->rq_xid == req->rq_xid)
+ goto found;
+ }
+ cfs_list_for_each_entry(tmp, &req->rq_export->exp_hp_rpcs,
+ rq_exp_list) {
+ /* Found duplicate one */
+ if (tmp->rq_xid == req->rq_xid)
+ goto found;
+ }
+ return 0;
+
+found:
+ DEBUG_REQ(D_HA, req, "Found duplicate req in processing\n");
+ DEBUG_REQ(D_HA, tmp, "Request being processed\n");
+ return -EBUSY;
+}
+
/**
* Put the request to the export list if the request may become
* a high priority one.
*/
-static int ptlrpc_server_hpreq_init(struct ptlrpc_service *svc,
+static int ptlrpc_server_hpreq_init(struct ptlrpc_service_part *svcpt,
struct ptlrpc_request *req)
{
- int rc = 0;
- ENTRY;
+ cfs_list_t *list;
+ int rc, hp = 0;
- if (svc->srv_ops.so_hpreq_handler) {
- rc = svc->srv_ops.so_hpreq_handler(req);
- if (rc)
- RETURN(rc);
- }
- if (req->rq_export && req->rq_ops) {
- /* Perform request specific check. We should do this check
- * before the request is added into exp_hp_rpcs list otherwise
- * it may hit swab race at LU-1044. */
- if (req->rq_ops->hpreq_check)
- rc = req->rq_ops->hpreq_check(req);
+ ENTRY;
+
+ if (svcpt->scp_service->srv_ops.so_hpreq_handler) {
+ rc = svcpt->scp_service->srv_ops.so_hpreq_handler(req);
+ if (rc < 0)
+ RETURN(rc);
+ LASSERT(rc == 0);
+ }
+ if (req->rq_export) {
+ if (req->rq_ops) {
+ /* Perform request specific check. We should do this
+ * check before the request is added into exp_hp_rpcs
+ * list otherwise it may hit swab race at LU-1044. */
+ if (req->rq_ops->hpreq_check) {
+ rc = req->rq_ops->hpreq_check(req);
+ /**
+ * XXX: Out of all current
+ * ptlrpc_hpreq_ops::hpreq_check(), only
+ * ldlm_cancel_hpreq_check() can return an
+ * error code; other functions assert in
+ * similar places, which seems odd.
+ * What also does not seem right is that
+ * handlers for those RPCs do not assert
+ * on the same checks, but rather handle the
+ * error cases. e.g. see ost_rw_hpreq_check(),
+ * and ost_brw_read(), ost_brw_write().
+ */
+ if (rc < 0)
+ RETURN(rc);
+ LASSERT(rc == 0 || rc == 1);
+ hp = rc;
+ }
+ list = &req->rq_export->exp_hp_rpcs;
+ } else {
+ list = &req->rq_export->exp_reg_rpcs;
+ }
+ /* do search for duplicated xid and the adding to the list
+ * atomically */
spin_lock_bh(&req->rq_export->exp_rpc_lock);
- cfs_list_add(&req->rq_exp_list,
- &req->rq_export->exp_hp_rpcs);
+ rc = ptlrpc_server_check_resend_in_progress(req);
+ if (rc < 0) {
+ spin_unlock_bh(&req->rq_export->exp_rpc_lock);
+ RETURN(rc);
+ }
+ cfs_list_add(&req->rq_exp_list, list);
spin_unlock_bh(&req->rq_export->exp_rpc_lock);
}
- RETURN(rc);
+ ptlrpc_nrs_req_initialize(svcpt, req, !!hp);
+
+ RETURN(hp);
}
/** Remove the request from the export list. */
static void ptlrpc_server_hpreq_fini(struct ptlrpc_request *req)
{
- ENTRY;
- if (req->rq_export && req->rq_ops) {
- /* refresh lock timeout again so that client has more
- * room to send lock cancel RPC. */
- if (req->rq_ops->hpreq_fini)
- req->rq_ops->hpreq_fini(req);
+ ENTRY;
+ if (req->rq_export) {
+ /* refresh lock timeout again so that client has more
+ * room to send lock cancel RPC. */
+ if (req->rq_ops && req->rq_ops->hpreq_fini)
+ req->rq_ops->hpreq_fini(req);
spin_lock_bh(&req->rq_export->exp_rpc_lock);
cfs_list_del_init(&req->rq_exp_list);
}
EXPORT_SYMBOL(ptlrpc_hpreq_handler);
-/**
- * Make the request a high priority one.
- *
- * All the high priority requests are queued in a separate FIFO
- * ptlrpc_service_part::scp_hpreq_pending list which is parallel to
- * ptlrpc_service_part::scp_req_pending list but has a higher priority
- * for handling.
- *
- * \see ptlrpc_server_handle_request().
- */
-static void ptlrpc_hpreq_reorder_nolock(struct ptlrpc_service_part *svcpt,
- struct ptlrpc_request *req)
-{
- ENTRY;
-
- spin_lock(&req->rq_lock);
- if (req->rq_hp == 0) {
- int opc = lustre_msg_get_opc(req->rq_reqmsg);
-
- /* Add to the high priority queue. */
- cfs_list_move_tail(&req->rq_list, &svcpt->scp_hreq_pending);
- req->rq_hp = 1;
- if (opc != OBD_PING)
- DEBUG_REQ(D_RPCTRACE, req, "high priority req");
- }
- spin_unlock(&req->rq_lock);
- EXIT;
-}
-
-/**
- * \see ptlrpc_hpreq_reorder_nolock
- */
-void ptlrpc_hpreq_reorder(struct ptlrpc_request *req)
-{
- struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
- ENTRY;
-
- spin_lock(&svcpt->scp_req_lock);
- /* It may happen that the request is already taken for the processing
- * but still in the export list, or the request is not in the request
- * queue but in the export list already, do not add it into the
- * HP list. */
- if (!cfs_list_empty(&req->rq_list))
- ptlrpc_hpreq_reorder_nolock(svcpt, req);
- spin_unlock(&svcpt->scp_req_lock);
- EXIT;
-}
-EXPORT_SYMBOL(ptlrpc_hpreq_reorder);
-
-/**
- * Add a request to the regular or HP queue; optionally perform HP request
- * initialization.
- */
static int ptlrpc_server_request_add(struct ptlrpc_service_part *svcpt,
struct ptlrpc_request *req)
{
int rc;
ENTRY;
- rc = ptlrpc_server_hpreq_init(svcpt->scp_service, req);
+ rc = ptlrpc_server_hpreq_init(svcpt, req);
if (rc < 0)
RETURN(rc);
- spin_lock(&svcpt->scp_req_lock);
-
- if (rc)
- ptlrpc_hpreq_reorder_nolock(svcpt, req);
- else
- cfs_list_add_tail(&req->rq_list, &svcpt->scp_req_pending);
-
- spin_unlock(&svcpt->scp_req_lock);
+ ptlrpc_nrs_req_add(svcpt, req, !!rc);
RETURN(0);
}
* User can call it w/o any lock but need to hold
* ptlrpc_service_part::scp_req_lock to get reliable result
*/
-static int ptlrpc_server_allow_high(struct ptlrpc_service_part *svcpt,
- int force)
+static bool ptlrpc_server_allow_high(struct ptlrpc_service_part *svcpt,
+ bool force)
{
int running = svcpt->scp_nthrs_running;
+ if (!nrs_svcpt_has_hp(svcpt))
+ return false;
+
if (force)
- return 1;
+ return true;
if (unlikely(svcpt->scp_service->srv_req_portal == MDS_REQUEST_PORTAL &&
CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CANCEL_RESEND))) {
}
if (svcpt->scp_nreqs_active >= running - 1)
- return 0;
+ return false;
if (svcpt->scp_nhreqs_active == 0)
- return 1;
+ return true;
- return cfs_list_empty(&svcpt->scp_req_pending) ||
+ return !ptlrpc_nrs_req_pending_nolock(svcpt, false) ||
svcpt->scp_hreq_count < svcpt->scp_service->srv_hpreq_ratio;
}
-static int ptlrpc_server_high_pending(struct ptlrpc_service_part *svcpt,
- int force)
+static bool ptlrpc_server_high_pending(struct ptlrpc_service_part *svcpt,
+ bool force)
{
return ptlrpc_server_allow_high(svcpt, force) &&
- !cfs_list_empty(&svcpt->scp_hreq_pending);
+ ptlrpc_nrs_req_pending_nolock(svcpt, true);
}
/**
* User can call it w/o any lock but need to hold
* ptlrpc_service_part::scp_req_lock to get reliable result
*/
-static int ptlrpc_server_allow_normal(struct ptlrpc_service_part *svcpt,
- int force)
+static bool ptlrpc_server_allow_normal(struct ptlrpc_service_part *svcpt,
+ bool force)
{
int running = svcpt->scp_nthrs_running;
#ifndef __KERNEL__
if (1) /* always allow to handle normal request for liblustre */
- return 1;
+ return true;
#endif
if (unlikely(svcpt->scp_service->srv_req_portal == MDS_REQUEST_PORTAL &&
CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CANCEL_RESEND))) {
if (force ||
svcpt->scp_nreqs_active < running - 2)
- return 1;
+ return true;
if (svcpt->scp_nreqs_active >= running - 1)
- return 0;
+ return false;
- return svcpt->scp_nhreqs_active > 0 ||
- svcpt->scp_service->srv_ops.so_hpreq_handler == NULL;
+ return svcpt->scp_nhreqs_active > 0 || !nrs_svcpt_has_hp(svcpt);
}
-static int ptlrpc_server_normal_pending(struct ptlrpc_service_part *svcpt,
- int force)
+static bool ptlrpc_server_normal_pending(struct ptlrpc_service_part *svcpt,
+ bool force)
{
return ptlrpc_server_allow_normal(svcpt, force) &&
- !cfs_list_empty(&svcpt->scp_req_pending);
+ ptlrpc_nrs_req_pending_nolock(svcpt, false);
}
/**
* \see ptlrpc_server_allow_normal
* \see ptlrpc_server_allow high
*/
-static inline int
-ptlrpc_server_request_pending(struct ptlrpc_service_part *svcpt, int force)
+static inline bool
+ptlrpc_server_request_pending(struct ptlrpc_service_part *svcpt, bool force)
{
return ptlrpc_server_high_pending(svcpt, force) ||
ptlrpc_server_normal_pending(svcpt, force);
* Returns a pointer to fetched request.
*/
static struct ptlrpc_request *
-ptlrpc_server_request_get(struct ptlrpc_service_part *svcpt, int force)
+ptlrpc_server_request_get(struct ptlrpc_service_part *svcpt, bool force)
{
- struct ptlrpc_request *req;
+ struct ptlrpc_request *req = NULL;
ENTRY;
+ spin_lock(&svcpt->scp_req_lock);
+#ifndef __KERNEL__
+ /* !@%$# liblustre only has 1 thread */
+ if (cfs_atomic_read(&svcpt->scp_nreps_difficult) != 0) {
+ spin_unlock(&svcpt->scp_req_lock);
+ RETURN(NULL);
+ }
+#endif
+
if (ptlrpc_server_high_pending(svcpt, force)) {
- req = cfs_list_entry(svcpt->scp_hreq_pending.next,
- struct ptlrpc_request, rq_list);
- svcpt->scp_hreq_count++;
- RETURN(req);
+ req = ptlrpc_nrs_req_get_nolock(svcpt, true, force);
+ if (req != NULL) {
+ svcpt->scp_hreq_count++;
+ goto got_request;
+ }
}
if (ptlrpc_server_normal_pending(svcpt, force)) {
- req = cfs_list_entry(svcpt->scp_req_pending.next,
- struct ptlrpc_request, rq_list);
- svcpt->scp_hreq_count = 0;
- RETURN(req);
+ req = ptlrpc_nrs_req_get_nolock(svcpt, false, force);
+ if (req != NULL) {
+ svcpt->scp_hreq_count = 0;
+ goto got_request;
+ }
}
+
+ spin_unlock(&svcpt->scp_req_lock);
RETURN(NULL);
+
+got_request:
+ svcpt->scp_nreqs_active++;
+ if (req->rq_hp)
+ svcpt->scp_nhreqs_active++;
+
+ spin_unlock(&svcpt->scp_req_lock);
+
+ if (likely(req->rq_export))
+ class_export_rpc_inc(req->rq_export);
+
+ RETURN(req);
}
/**
* ptlrpc_server_handle_req later on.
*/
static int
-ptlrpc_server_handle_req_in(struct ptlrpc_service_part *svcpt)
+ptlrpc_server_handle_req_in(struct ptlrpc_service_part *svcpt,
+ struct ptlrpc_thread *thread)
{
struct ptlrpc_service *svc = svcpt->scp_service;
struct ptlrpc_request *req;
req->rq_export = class_conn2export(
lustre_msg_get_handle(req->rq_reqmsg));
if (req->rq_export) {
- class_export_rpc_get(req->rq_export);
rc = ptlrpc_check_req(req);
if (rc == 0) {
rc = sptlrpc_target_export_check(req->rq_export, req);
goto err_req;
}
- ptlrpc_at_add_timed(req);
+ req->rq_svc_thread = thread;
+ if (thread != NULL) {
+ /* initialize request session, it is needed for request
+ * processing by target */
+ rc = lu_context_init(&req->rq_session, LCT_SERVER_SESSION |
+ LCT_NOREF);
+ if (rc) {
+ CERROR("%s: failure to initialize session: rc = %d\n",
+ thread->t_name, rc);
+ goto err_req;
+ }
+ req->rq_session.lc_thread = thread;
+ lu_context_enter(&req->rq_session);
+ req->rq_svc_thread->t_env->le_ses = &req->rq_session;
+ }
+
+ ptlrpc_at_add_timed(req);
- /* Move it over to the request processing queue */
+ /* Move it over to the request processing queue */
rc = ptlrpc_server_request_add(svcpt, req);
- if (rc) {
- ptlrpc_server_hpreq_fini(req);
+ if (rc)
GOTO(err_req, rc);
- }
- cfs_waitq_signal(&svcpt->scp_waitq);
+
+ wake_up(&svcpt->scp_waitq);
RETURN(1);
err_req:
- if (req->rq_export)
- class_export_rpc_put(req->rq_export);
- spin_lock(&svcpt->scp_req_lock);
- svcpt->scp_nreqs_active++;
- spin_unlock(&svcpt->scp_req_lock);
ptlrpc_server_finish_request(svcpt, req);
RETURN(1);
ptlrpc_server_handle_request(struct ptlrpc_service_part *svcpt,
struct ptlrpc_thread *thread)
{
- struct ptlrpc_service *svc = svcpt->scp_service;
- struct obd_export *export = NULL;
- struct ptlrpc_request *request;
- struct timeval work_start;
- struct timeval work_end;
- long timediff;
- int rc;
- int fail_opc = 0;
- ENTRY;
+ struct ptlrpc_service *svc = svcpt->scp_service;
+ struct ptlrpc_request *request;
+ struct timeval work_start;
+ struct timeval work_end;
+ long timediff;
+ int fail_opc = 0;
- spin_lock(&svcpt->scp_req_lock);
-#ifndef __KERNEL__
- /* !@%$# liblustre only has 1 thread */
- if (cfs_atomic_read(&svcpt->scp_nreps_difficult) != 0) {
- spin_unlock(&svcpt->scp_req_lock);
+ ENTRY;
+
+ request = ptlrpc_server_request_get(svcpt, false);
+ if (request == NULL)
RETURN(0);
- }
-#endif
- request = ptlrpc_server_request_get(svcpt, 0);
- if (request == NULL) {
- spin_unlock(&svcpt->scp_req_lock);
- RETURN(0);
- }
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_HPREQ_NOTIMEOUT))
fail_opc = OBD_FAIL_PTLRPC_HPREQ_NOTIMEOUT;
fail_opc = OBD_FAIL_PTLRPC_HPREQ_TIMEOUT;
if (unlikely(fail_opc)) {
- if (request->rq_export && request->rq_ops) {
- spin_unlock(&svcpt->scp_req_lock);
-
+ if (request->rq_export && request->rq_ops)
OBD_FAIL_TIMEOUT(fail_opc, 4);
-
- spin_lock(&svcpt->scp_req_lock);
- request = ptlrpc_server_request_get(svcpt, 0);
- if (request == NULL) {
- spin_unlock(&svcpt->scp_req_lock);
- RETURN(0);
- }
- }
}
- cfs_list_del_init(&request->rq_list);
- svcpt->scp_nreqs_active++;
- if (request->rq_hp)
- svcpt->scp_nhreqs_active++;
-
- spin_unlock(&svcpt->scp_req_lock);
-
ptlrpc_rqphase_move(request, RQ_PHASE_INTERPRET);
- if(OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_DUMP_LOG))
- libcfs_debug_dumplog();
+ if(OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_DUMP_LOG))
+ libcfs_debug_dumplog();
- cfs_gettimeofday(&work_start);
- timediff = cfs_timeval_sub(&work_start, &request->rq_arrival_time,NULL);
- if (likely(svc->srv_stats != NULL)) {
+ do_gettimeofday(&work_start);
+ timediff = cfs_timeval_sub(&work_start, &request->rq_arrival_time,NULL);
+ if (likely(svc->srv_stats != NULL)) {
lprocfs_counter_add(svc->srv_stats, PTLRPC_REQWAIT_CNTR,
timediff);
lprocfs_counter_add(svc->srv_stats, PTLRPC_REQQDEPTH_CNTR,
at_get(&svcpt->scp_at_estimate));
}
- export = request->rq_export;
- rc = lu_context_init(&request->rq_session, LCT_SESSION | LCT_NOREF);
- if (rc) {
- CERROR("Failure to initialize session: %d\n", rc);
- goto out_req;
- }
- request->rq_session.lc_thread = thread;
- request->rq_session.lc_cookie = 0x5;
- lu_context_enter(&request->rq_session);
-
- CDEBUG(D_NET, "got req "LPU64"\n", request->rq_xid);
-
- request->rq_svc_thread = thread;
- if (thread)
- request->rq_svc_thread->t_env->le_ses = &request->rq_session;
-
- if (likely(request->rq_export)) {
+ if (likely(request->rq_export)) {
if (unlikely(ptlrpc_check_req(request)))
goto put_conn;
ptlrpc_update_export_timer(request->rq_export, timediff >> 19);
goto put_conn;
}
- CDEBUG(D_RPCTRACE, "Handling RPC pname:cluuid+ref:pid:xid:nid:opc "
- "%s:%s+%d:%d:x"LPU64":%s:%d\n", cfs_curproc_comm(),
- (request->rq_export ?
- (char *)request->rq_export->exp_client_uuid.uuid : "0"),
- (request->rq_export ?
- cfs_atomic_read(&request->rq_export->exp_refcount) : -99),
- lustre_msg_get_status(request->rq_reqmsg), request->rq_xid,
- libcfs_id2str(request->rq_peer),
- lustre_msg_get_opc(request->rq_reqmsg));
+ CDEBUG(D_RPCTRACE, "Handling RPC pname:cluuid+ref:pid:xid:nid:opc "
+ "%s:%s+%d:%d:x"LPU64":%s:%d\n", current_comm(),
+ (request->rq_export ?
+ (char *)request->rq_export->exp_client_uuid.uuid : "0"),
+ (request->rq_export ?
+ cfs_atomic_read(&request->rq_export->exp_refcount) : -99),
+ lustre_msg_get_status(request->rq_reqmsg), request->rq_xid,
+ libcfs_id2str(request->rq_peer),
+ lustre_msg_get_opc(request->rq_reqmsg));
if (lustre_msg_get_opc(request->rq_reqmsg) != OBD_PING)
CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_PAUSE_REQ, cfs_fail_val);
- rc = svc->srv_ops.so_req_handler(request);
+ CDEBUG(D_NET, "got req "LPU64"\n", request->rq_xid);
- ptlrpc_rqphase_move(request, RQ_PHASE_COMPLETE);
+ /* re-assign request and sesson thread to the current one */
+ request->rq_svc_thread = thread;
+ if (thread != NULL) {
+ LASSERT(request->rq_session.lc_thread != NULL);
+ request->rq_session.lc_thread = thread;
+ request->rq_session.lc_cookie = 0x55;
+ thread->t_env->le_ses = &request->rq_session;
+ }
+ svc->srv_ops.so_req_handler(request);
-put_conn:
- lu_context_exit(&request->rq_session);
- lu_context_fini(&request->rq_session);
+ ptlrpc_rqphase_move(request, RQ_PHASE_COMPLETE);
+put_conn:
if (unlikely(cfs_time_current_sec() > request->rq_deadline)) {
DEBUG_REQ(D_WARNING, request, "Request took longer "
"than estimated ("CFS_DURATION_T":"CFS_DURATION_T"s);"
request->rq_deadline));
}
- cfs_gettimeofday(&work_end);
- timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
- CDEBUG(D_RPCTRACE, "Handled RPC pname:cluuid+ref:pid:xid:nid:opc "
- "%s:%s+%d:%d:x"LPU64":%s:%d Request procesed in "
- "%ldus (%ldus total) trans "LPU64" rc %d/%d\n",
- cfs_curproc_comm(),
- (request->rq_export ?
- (char *)request->rq_export->exp_client_uuid.uuid : "0"),
+ do_gettimeofday(&work_end);
+ timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
+ CDEBUG(D_RPCTRACE, "Handled RPC pname:cluuid+ref:pid:xid:nid:opc "
+ "%s:%s+%d:%d:x"LPU64":%s:%d Request procesed in "
+ "%ldus (%ldus total) trans "LPU64" rc %d/%d\n",
+ current_comm(),
+ (request->rq_export ?
+ (char *)request->rq_export->exp_client_uuid.uuid : "0"),
(request->rq_export ?
cfs_atomic_read(&request->rq_export->exp_refcount) : -99),
lustre_msg_get_status(request->rq_reqmsg),
request->rq_arrival_time.tv_sec));
}
-out_req:
- if (export != NULL)
- class_export_rpc_put(export);
- ptlrpc_server_finish_request(svcpt, request);
+ ptlrpc_server_finish_active_request(svcpt, request);
RETURN(1);
}
ptlrpc_rs_decref (rs);
if (cfs_atomic_dec_and_test(&svcpt->scp_nreps_difficult) &&
svc->srv_is_stopping)
- cfs_waitq_broadcast(&svcpt->scp_waitq);
+ wake_up_all(&svcpt->scp_waitq);
RETURN(1);
}
svcpt->scp_nthrs_running++;
do {
- rc = ptlrpc_server_handle_req_in(svcpt);
+ rc = ptlrpc_server_handle_req_in(svcpt, NULL);
rc |= ptlrpc_server_handle_reply(svcpt);
rc |= ptlrpc_at_check_timed(svcpt);
rc |= ptlrpc_server_handle_request(svcpt, NULL);
lc_watchdog_disable(thread->t_watchdog);
- cfs_cond_resched();
+ cond_resched();
l_wait_event_exclusive_head(svcpt->scp_waitq,
ptlrpc_thread_stopping(thread) ||
ptlrpc_server_request_incoming(svcpt) ||
- ptlrpc_server_request_pending(svcpt, 0) ||
+ ptlrpc_server_request_pending(svcpt, false) ||
ptlrpc_rqbd_pending(svcpt) ||
ptlrpc_at_check(svcpt), &lwi);
struct ptlrpc_service *svc = svcpt->scp_service;
struct ptlrpc_reply_state *rs;
#ifdef WITH_GROUP_INFO
- cfs_group_info_t *ginfo = NULL;
+ struct group_info *ginfo = NULL;
#endif
- struct lu_env *env;
- int counter = 0, rc = 0;
- ENTRY;
+ struct lu_env *env;
+ int counter = 0, rc = 0;
+ ENTRY;
- thread->t_pid = cfs_curproc_pid();
- cfs_daemonize_ctxt(thread->t_name);
+ thread->t_pid = current_pid();
+ unshare_fs_struct();
/* NB: we will call cfs_cpt_bind() for all threads, because we
* might want to run lustre server only on a subset of system CPUs,
}
#ifdef WITH_GROUP_INFO
- ginfo = cfs_groups_alloc(0);
- if (!ginfo) {
- rc = -ENOMEM;
- goto out;
- }
+ ginfo = groups_alloc(0);
+ if (!ginfo) {
+ rc = -ENOMEM;
+ goto out;
+ }
- cfs_set_current_groups(ginfo);
- cfs_put_group_info(ginfo);
+ set_current_groups(ginfo);
+ put_group_info(ginfo);
#endif
if (svc->srv_ops.so_thr_init != NULL) {
spin_unlock(&svcpt->scp_lock);
/* wake up our creator in case he's still waiting. */
- cfs_waitq_signal(&thread->t_ctl_waitq);
+ wake_up(&thread->t_ctl_waitq);
thread->t_watchdog = lc_watchdog_add(ptlrpc_server_get_timeout(svcpt),
NULL, NULL);
spin_lock(&svcpt->scp_rep_lock);
cfs_list_add(&rs->rs_list, &svcpt->scp_rep_idle);
- cfs_waitq_signal(&svcpt->scp_rep_waitq);
+ wake_up(&svcpt->scp_rep_waitq);
spin_unlock(&svcpt->scp_rep_lock);
CDEBUG(D_NET, "service thread %d (#%d) started\n", thread->t_id,
/* Process all incoming reqs before handling any */
if (ptlrpc_server_request_incoming(svcpt)) {
- ptlrpc_server_handle_req_in(svcpt);
+ lu_context_enter(&env->le_ctx);
+ env->le_ses = NULL;
+ ptlrpc_server_handle_req_in(svcpt, thread);
+ lu_context_exit(&env->le_ctx);
+
/* but limit ourselves in case of flood */
if (counter++ < 100)
continue;
if (ptlrpc_at_check(svcpt))
ptlrpc_at_check_timed(svcpt);
- if (ptlrpc_server_request_pending(svcpt, 0)) {
+ if (ptlrpc_server_request_pending(svcpt, false)) {
lu_context_enter(&env->le_ctx);
ptlrpc_server_handle_request(svcpt, thread);
lu_context_exit(&env->le_ctx);
thread->t_id = rc;
thread_add_flags(thread, SVC_STOPPED);
- cfs_waitq_signal(&thread->t_ctl_waitq);
+ wake_up(&thread->t_ctl_waitq);
spin_unlock(&svcpt->scp_lock);
return rc;
snprintf(threadname, sizeof(threadname), "ptlrpc_hr%02d_%03d",
hrp->hrp_cpt, hrt->hrt_id);
- cfs_daemonize_ctxt(threadname);
+ unshare_fs_struct();
rc = cfs_cpt_bind(ptlrpc_hr.hr_cpt_table, hrp->hrp_cpt);
if (rc != 0) {
}
cfs_atomic_inc(&hrp->hrp_nstarted);
- cfs_waitq_signal(&ptlrpc_hr.hr_waitq);
+ wake_up(&ptlrpc_hr.hr_waitq);
while (!ptlrpc_hr.hr_stopping) {
l_wait_condition(hrt->hrt_waitq, hrt_dont_sleep(hrt, &replies));
}
cfs_atomic_inc(&hrp->hrp_nstopped);
- cfs_waitq_signal(&ptlrpc_hr.hr_waitq);
+ wake_up(&ptlrpc_hr.hr_waitq);
return 0;
}
if (hrp->hrp_thrs == NULL)
continue; /* uninitialized */
for (j = 0; j < hrp->hrp_nthrs; j++)
- cfs_waitq_broadcast(&hrp->hrp_thrs[j].hrt_waitq);
+ wake_up_all(&hrp->hrp_thrs[j].hrt_waitq);
}
cfs_percpt_for_each(hrp, i, ptlrpc_hr.hr_partitions) {
if (hrp->hrp_thrs == NULL)
continue; /* uninitialized */
- cfs_wait_event(ptlrpc_hr.hr_waitq,
+ wait_event(ptlrpc_hr.hr_waitq,
cfs_atomic_read(&hrp->hrp_nstopped) ==
cfs_atomic_read(&hrp->hrp_nstarted));
}
int rc = 0;
for (j = 0; j < hrp->hrp_nthrs; j++) {
- rc = cfs_create_thread(ptlrpc_hr_main,
- &hrp->hrp_thrs[j],
- CLONE_VM | CLONE_FILES);
- if (rc < 0)
+ struct ptlrpc_hr_thread *hrt = &hrp->hrp_thrs[j];
+ rc = PTR_ERR(kthread_run(ptlrpc_hr_main,
+ &hrp->hrp_thrs[j],
+ "ptlrpc_hr%02d_%03d",
+ hrp->hrp_cpt,
+ hrt->hrt_id));
+ if (IS_ERR_VALUE(rc))
break;
}
- cfs_wait_event(ptlrpc_hr.hr_waitq,
+ wait_event(ptlrpc_hr.hr_waitq,
cfs_atomic_read(&hrp->hrp_nstarted) == j);
- if (rc >= 0)
+ if (!IS_ERR_VALUE(rc))
continue;
CERROR("Reply handling thread %d:%d Failed on starting: "
thread_add_flags(thread, SVC_STOPPING);
}
- cfs_waitq_broadcast(&svcpt->scp_waitq);
+ wake_up_all(&svcpt->scp_waitq);
while (!cfs_list_empty(&svcpt->scp_threads)) {
thread = cfs_list_entry(svcpt->scp_threads.next,
OBD_CPT_ALLOC_PTR(thread, svc->srv_cptable, svcpt->scp_cpt);
if (thread == NULL)
RETURN(-ENOMEM);
- cfs_waitq_init(&thread->t_ctl_waitq);
+ init_waitqueue_head(&thread->t_ctl_waitq);
spin_lock(&svcpt->scp_lock);
if (!ptlrpc_threads_increasable(svcpt)) {
if (wait) {
CDEBUG(D_INFO, "Waiting for creating thread %s #%d\n",
svc->srv_thread_name, svcpt->scp_thr_nextid);
- cfs_schedule();
+ schedule();
goto again;
}
}
CDEBUG(D_RPCTRACE, "starting thread '%s'\n", thread->t_name);
- /*
- * CLONE_VM and CLONE_FILES just avoid a needless copy, because we
- * just drop the VM and FILES in cfs_daemonize_ctxt() right away.
- */
- rc = cfs_create_thread(ptlrpc_main, thread, CFS_DAEMON_FLAGS);
- if (rc < 0) {
+ rc = PTR_ERR(kthread_run(ptlrpc_main, thread, thread->t_name));
+ if (IS_ERR_VALUE(rc)) {
CERROR("cannot start thread '%s': rc %d\n",
thread->t_name, rc);
spin_lock(&svcpt->scp_lock);
- cfs_list_del(&thread->t_link);
--svcpt->scp_nthrs_starting;
- spin_unlock(&svcpt->scp_lock);
-
- OBD_FREE(thread, sizeof(*thread));
+ if (thread_is_stopping(thread)) {
+ /* this ptlrpc_thread is being hanled
+ * by ptlrpc_svcpt_stop_threads now
+ */
+ thread_add_flags(thread, SVC_STOPPED);
+ wake_up(&thread->t_ctl_waitq);
+ spin_unlock(&svcpt->scp_lock);
+ } else {
+ cfs_list_del(&thread->t_link);
+ spin_unlock(&svcpt->scp_lock);
+ OBD_FREE_PTR(thread);
+ }
RETURN(rc);
}
if (ptlrpc_hr.hr_partitions == NULL)
RETURN(-ENOMEM);
- cfs_waitq_init(&ptlrpc_hr.hr_waitq);
+ init_waitqueue_head(&ptlrpc_hr.hr_waitq);
cfs_percpt_for_each(hrp, i, ptlrpc_hr.hr_partitions) {
hrp->hrp_cpt = i;
hrt->hrt_id = j;
hrt->hrt_partition = hrp;
- cfs_waitq_init(&hrt->hrt_waitq);
+ init_waitqueue_head(&hrt->hrt_waitq);
spin_lock_init(&hrt->hrt_lock);
CFS_INIT_LIST_HEAD(&hrt->hrt_queue);
}
cfs_list_del(&req->rq_list);
svcpt->scp_nreqs_incoming--;
- svcpt->scp_nreqs_active++;
ptlrpc_server_finish_request(svcpt, req);
}
- while (ptlrpc_server_request_pending(svcpt, 1)) {
- req = ptlrpc_server_request_get(svcpt, 1);
- cfs_list_del(&req->rq_list);
- svcpt->scp_nreqs_active++;
- ptlrpc_server_hpreq_fini(req);
-
- if (req->rq_export != NULL)
- class_export_rpc_put(req->rq_export);
- ptlrpc_server_finish_request(svcpt, req);
+ while (ptlrpc_server_request_pending(svcpt, true)) {
+ req = ptlrpc_server_request_get(svcpt, true);
+ ptlrpc_server_finish_active_request(svcpt, req);
}
LASSERT(cfs_list_empty(&svcpt->scp_rqbd_posted));
service->srv_is_stopping = 1;
- spin_lock(&ptlrpc_all_services_lock);
+ mutex_lock(&ptlrpc_all_services_mutex);
cfs_list_del_init(&service->srv_list);
- spin_unlock(&ptlrpc_all_services_lock);
-
- ptlrpc_lprocfs_unregister_service(service);
+ mutex_unlock(&ptlrpc_all_services_mutex);
ptlrpc_service_del_atimer(service);
ptlrpc_stop_all_threads(service);
ptlrpc_service_unlink_rqbd(service);
ptlrpc_service_purge_all(service);
+ ptlrpc_service_nrs_cleanup(service);
+
+ ptlrpc_lprocfs_unregister_service(service);
+
ptlrpc_service_free(service);
RETURN(0);
* to be shot, so it's intentionally non-aggressive. */
int ptlrpc_svcpt_health_check(struct ptlrpc_service_part *svcpt)
{
- struct ptlrpc_request *request;
+ struct ptlrpc_request *request = NULL;
struct timeval right_now;
long timediff;
- cfs_gettimeofday(&right_now);
+ do_gettimeofday(&right_now);
spin_lock(&svcpt->scp_req_lock);
- if (!ptlrpc_server_request_pending(svcpt, 1)) {
+ /* How long has the next entry been waiting? */
+ if (ptlrpc_server_high_pending(svcpt, true))
+ request = ptlrpc_nrs_req_peek_nolock(svcpt, true);
+ else if (ptlrpc_server_normal_pending(svcpt, true))
+ request = ptlrpc_nrs_req_peek_nolock(svcpt, false);
+
+ if (request == NULL) {
spin_unlock(&svcpt->scp_req_lock);
return 0;
}
- /* How long has the next entry been waiting? */
- if (cfs_list_empty(&svcpt->scp_req_pending)) {
- request = cfs_list_entry(svcpt->scp_hreq_pending.next,
- struct ptlrpc_request, rq_list);
- } else {
- request = cfs_list_entry(svcpt->scp_req_pending.next,
- struct ptlrpc_request, rq_list);
- }
-
timediff = cfs_timeval_sub(&right_now, &request->rq_arrival_time, NULL);
spin_unlock(&svcpt->scp_req_lock);
struct ptlrpc_service_part *svcpt;
int i;
- if (svc == NULL || svc->srv_parts == NULL)
+ if (svc == NULL)
return 0;
ptlrpc_service_for_each_part(svcpt, i, svc) {
git://git.whamcloud.com / fs / lustre-release.git / blobdiff