/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Copyright (C) 2001-2003 Cluster File Systems, Inc.
- * Author Peter Braam <braam@clusterfs.com>
+ * GPL HEADER START
*
- * This file is part of the Lustre file system, http://www.lustre.org
- * Lustre is a trademark of Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * 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.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
*
- * 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.
+ * This program 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 version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
*
- * 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.
+ * 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.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/ptlrpc/ptlrpcd.c
*/
#define DEBUG_SUBSYSTEM S_RPC
# include <ctype.h>
#endif
-#include <libcfs/kp30.h>
#include <lustre_net.h>
# include <lustre_lib.h>
#include <lustre_ha.h>
#include <obd_class.h> /* for obd_zombie */
#include <obd_support.h> /* for OBD_FAIL_CHECK */
+#include <cl_object.h> /* cl_env_{get,put}() */
#include <lprocfs_status.h>
-#define LIOD_STOP 0
-struct ptlrpcd_ctl {
- unsigned long pc_flags;
- spinlock_t pc_lock;
- struct completion pc_starting;
- struct completion pc_finishing;
- struct list_head pc_req_list;
- cfs_waitq_t pc_waitq;
- struct ptlrpc_request_set *pc_set;
- char pc_name[16];
-#ifndef __KERNEL__
- int pc_recurred;
- void *pc_callback;
-#endif
+enum pscope_thread {
+ PT_NORMAL,
+ PT_RECOVERY,
+ PT_NR
};
-static struct ptlrpcd_ctl ptlrpcd_pc;
-static struct ptlrpcd_ctl ptlrpcd_recovery_pc;
+struct ptlrpcd_scope_ctl {
+ struct ptlrpcd_thread {
+ const char *pt_name;
+ struct ptlrpcd_ctl pt_ctl;
+ } pscope_thread[PT_NR];
+};
+
+static struct ptlrpcd_scope_ctl ptlrpcd_scopes[PSCOPE_NR] = {
+ [PSCOPE_BRW] = {
+ .pscope_thread = {
+ [PT_NORMAL] = {
+ .pt_name = "ptlrpcd-brw"
+ },
+ [PT_RECOVERY] = {
+ .pt_name = "ptlrpcd-brw-rcv"
+ }
+ }
+ },
+ [PSCOPE_OTHER] = {
+ .pscope_thread = {
+ [PT_NORMAL] = {
+ .pt_name = "ptlrpcd"
+ },
+ [PT_RECOVERY] = {
+ .pt_name = "ptlrpcd-rcv"
+ }
+ }
+ }
+};
-struct semaphore ptlrpcd_sem;
+cfs_semaphore_t ptlrpcd_sem;
static int ptlrpcd_users = 0;
void ptlrpcd_wake(struct ptlrpc_request *req)
{
- struct ptlrpcd_ctl *pc = req->rq_ptlrpcd_data;
+ struct ptlrpc_request_set *rq_set = req->rq_set;
- LASSERT(pc != NULL);
+ LASSERT(rq_set != NULL);
- cfs_waitq_signal(&pc->pc_waitq);
+ cfs_waitq_signal(&rq_set->set_waitq);
}
-/* requests that are added to the ptlrpcd queue are sent via
- * ptlrpcd_check->ptlrpc_check_set() */
-void ptlrpcd_add_req(struct ptlrpc_request *req)
+/*
+ * Move all request from an existing request set to the ptlrpcd queue.
+ * All requests from the set must be in phase RQ_PHASE_NEW.
+ */
+void ptlrpcd_add_rqset(struct ptlrpc_request_set *set)
+{
+ cfs_list_t *tmp, *pos;
+
+ cfs_list_for_each_safe(pos, tmp, &set->set_requests) {
+ struct ptlrpc_request *req =
+ cfs_list_entry(pos, struct ptlrpc_request,
+ rq_set_chain);
+
+ LASSERT(req->rq_phase == RQ_PHASE_NEW);
+ cfs_list_del_init(&req->rq_set_chain);
+ req->rq_set = NULL;
+ ptlrpcd_add_req(req, PSCOPE_OTHER);
+ set->set_remaining--;
+ }
+ LASSERT(set->set_remaining == 0);
+}
+EXPORT_SYMBOL(ptlrpcd_add_rqset);
+
+/*
+ * Requests that are added to the ptlrpcd queue are sent via
+ * ptlrpcd_check->ptlrpc_check_set().
+ */
+int ptlrpcd_add_req(struct ptlrpc_request *req, enum ptlrpcd_scope scope)
{
struct ptlrpcd_ctl *pc;
+ enum pscope_thread pt;
+ int rc;
- if (req->rq_send_state == LUSTRE_IMP_FULL)
- pc = &ptlrpcd_pc;
- else
- pc = &ptlrpcd_recovery_pc;
+ LASSERT(scope < PSCOPE_NR);
+ pt = req->rq_send_state == LUSTRE_IMP_FULL ? PT_NORMAL : PT_RECOVERY;
+ pc = &ptlrpcd_scopes[scope].pscope_thread[pt].pt_ctl;
+ rc = ptlrpc_set_add_new_req(pc, req);
+ /*
+ * XXX disable this for CLIO: environment is needed for interpreter.
+ * add debug temporary to check rc.
+ */
+ LASSERTF(rc == 0, "ptlrpcd_add_req failed (rc = %d)\n", rc);
+ if (rc && 0) {
+ /*
+ * Thread is probably in stop now so we need to
+ * kill this rpc as it was not added. Let's call
+ * interpret for it to let know we're killing it
+ * so that higher levels might free associated
+ * resources.
+ */
+ ptlrpc_req_interpret(NULL, req, -EBADR);
+ req->rq_set = NULL;
+ ptlrpc_req_finished(req);
+ } else if (req->rq_send_state == LUSTRE_IMP_CONNECTING) {
+ /*
+ * The request is for recovery, should be sent ASAP.
+ */
+ cfs_waitq_signal(&pc->pc_set->set_waitq);
+ }
- req->rq_ptlrpcd_data = pc;
- ptlrpc_set_add_new_req(pc->pc_set, req);
- wake_up(&pc->pc_waitq);
+ return rc;
}
-static int ptlrpcd_check(struct ptlrpcd_ctl *pc)
+static int ptlrpcd_check(const struct lu_env *env, struct ptlrpcd_ctl *pc)
{
- struct list_head *tmp, *pos;
+ cfs_list_t *tmp, *pos;
struct ptlrpc_request *req;
int rc = 0;
ENTRY;
- if (test_bit(LIOD_STOP, &pc->pc_flags))
- RETURN(1);
-
- obd_zombie_impexp_cull();
-
- spin_lock(&pc->pc_set->set_new_req_lock);
- list_for_each_safe(pos, tmp, &pc->pc_set->set_new_requests) {
- req = list_entry(pos, struct ptlrpc_request, rq_set_chain);
- list_del_init(&req->rq_set_chain);
+ cfs_spin_lock(&pc->pc_set->set_new_req_lock);
+ cfs_list_for_each_safe(pos, tmp, &pc->pc_set->set_new_requests) {
+ req = cfs_list_entry(pos, struct ptlrpc_request, rq_set_chain);
+ cfs_list_del_init(&req->rq_set_chain);
ptlrpc_set_add_req(pc->pc_set, req);
- rc = 1; /* need to calculate its timeout */
+ /*
+ * Need to calculate its timeout.
+ */
+ rc = 1;
}
- spin_unlock(&pc->pc_set->set_new_req_lock);
+ cfs_spin_unlock(&pc->pc_set->set_new_req_lock);
if (pc->pc_set->set_remaining) {
- rc = rc | ptlrpc_check_set(pc->pc_set);
-
- /* XXX our set never completes, so we prune the completed
- * reqs after each iteration. boy could this be smarter. */
- list_for_each_safe(pos, tmp, &pc->pc_set->set_requests) {
- req = list_entry(pos, struct ptlrpc_request,
+ rc = rc | ptlrpc_check_set(env, pc->pc_set);
+
+ /*
+ * XXX: our set never completes, so we prune the completed
+ * reqs after each iteration. boy could this be smarter.
+ */
+ cfs_list_for_each_safe(pos, tmp, &pc->pc_set->set_requests) {
+ req = cfs_list_entry(pos, struct ptlrpc_request,
rq_set_chain);
if (req->rq_phase != RQ_PHASE_COMPLETE)
continue;
- list_del_init(&req->rq_set_chain);
+ cfs_list_del_init(&req->rq_set_chain);
req->rq_set = NULL;
ptlrpc_req_finished (req);
}
}
if (rc == 0) {
- /* If new requests have been added, make sure to wake up */
- spin_lock(&pc->pc_set->set_new_req_lock);
- rc = !list_empty(&pc->pc_set->set_new_requests);
- spin_unlock(&pc->pc_set->set_new_req_lock);
+ /*
+ * If new requests have been added, make sure to wake up.
+ */
+ cfs_spin_lock(&pc->pc_set->set_new_req_lock);
+ rc = !cfs_list_empty(&pc->pc_set->set_new_requests);
+ cfs_spin_unlock(&pc->pc_set->set_new_req_lock);
}
RETURN(rc);
}
#ifdef __KERNEL__
-/* ptlrpc's code paths like to execute in process context, so we have this
- * thread which spins on a set which contains the io rpcs. llite specifies
- * ptlrpcd's set when it pushes pages down into the oscs */
+/*
+ * ptlrpc's code paths like to execute in process context, so we have this
+ * thread which spins on a set which contains the io rpcs. llite specifies
+ * ptlrpcd's set when it pushes pages down into the oscs.
+ */
static int ptlrpcd(void *arg)
{
struct ptlrpcd_ctl *pc = arg;
- int rc;
+ struct lu_env env = { .le_ses = NULL };
+ int rc, exit = 0;
ENTRY;
- if ((rc = cfs_daemonize_ctxt(pc->pc_name))) {
- complete(&pc->pc_starting);
- return rc;
+ rc = cfs_daemonize_ctxt(pc->pc_name);
+ if (rc == 0) {
+ /*
+ * XXX So far only "client" ptlrpcd uses an environment. In
+ * the future, ptlrpcd thread (or a thread-set) has to given
+ * an argument, describing its "scope".
+ */
+ rc = lu_context_init(&env.le_ctx,
+ LCT_CL_THREAD|LCT_REMEMBER|LCT_NOREF);
}
- complete(&pc->pc_starting);
+ cfs_complete(&pc->pc_starting);
- /* this mainloop strongly resembles ptlrpc_set_wait except
- * that our set never completes. ptlrpcd_check calls ptlrpc_check_set
- * when there are requests in the set. new requests come in
- * on the set's new_req_list and ptlrpcd_check moves them into
- * the set. */
- while (1) {
- cfs_waitlink_t set_wait;
+ if (rc != 0)
+ RETURN(rc);
+ env.le_ctx.lc_cookie = 0x7;
+
+ /*
+ * This mainloop strongly resembles ptlrpc_set_wait() except that our
+ * set never completes. ptlrpcd_check() calls ptlrpc_check_set() when
+ * there are requests in the set. New requests come in on the set's
+ * new_req_list and ptlrpcd_check() moves them into the set.
+ */
+ do {
struct l_wait_info lwi;
- cfs_duration_t timeout;
+ int timeout;
+
+ rc = lu_env_refill(&env);
+ if (rc != 0) {
+ /*
+ * XXX This is very awkward situation, because
+ * execution can neither continue (request
+ * interpreters assume that env is set up), nor repeat
+ * the loop (as this potentially results in a tight
+ * loop of -ENOMEM's).
+ *
+ * Fortunately, refill only ever does something when
+ * new modules are loaded, i.e., early during boot up.
+ */
+ CERROR("Failure to refill session: %d\n", rc);
+ continue;
+ }
- timeout = cfs_time_seconds(ptlrpc_set_next_timeout(pc->pc_set));
- lwi = LWI_TIMEOUT(timeout, ptlrpc_expired_set, pc->pc_set);
+ timeout = ptlrpc_set_next_timeout(pc->pc_set);
+ lwi = LWI_TIMEOUT(cfs_time_seconds(timeout ? timeout : 1),
+ ptlrpc_expired_set, pc->pc_set);
+
+ lu_context_enter(&env.le_ctx);
+ l_wait_event(pc->pc_set->set_waitq,
+ ptlrpcd_check(&env, pc), &lwi);
+ lu_context_exit(&env.le_ctx);
+
+ /*
+ * Abort inflight rpcs for forced stop case.
+ */
+ if (cfs_test_bit(LIOD_STOP, &pc->pc_flags)) {
+ if (cfs_test_bit(LIOD_FORCE, &pc->pc_flags))
+ ptlrpc_abort_set(pc->pc_set);
+ exit++;
+ }
- /* ala the pinger, wait on pc's waitqueue and the set's */
- cfs_waitlink_init(&set_wait);
- cfs_waitq_add(&pc->pc_set->set_waitq, &set_wait);
- cfs_waitq_forward(&set_wait, &pc->pc_waitq);
- l_wait_event(pc->pc_waitq, ptlrpcd_check(pc), &lwi);
- cfs_waitq_del(&pc->pc_set->set_waitq, &set_wait);
+ /*
+ * Let's make one more loop to make sure that ptlrpcd_check()
+ * copied all raced new rpcs into the set so we can kill them.
+ */
+ } while (exit < 2);
- if (test_bit(LIOD_STOP, &pc->pc_flags))
- break;
- }
- /* wait for inflight requests to drain */
- if (!list_empty(&pc->pc_set->set_requests))
+ /*
+ * Wait for inflight requests to drain.
+ */
+ if (!cfs_list_empty(&pc->pc_set->set_requests))
ptlrpc_set_wait(pc->pc_set);
- complete(&pc->pc_finishing);
+ lu_context_fini(&env.le_ctx);
+ cfs_complete(&pc->pc_finishing);
+
+ cfs_clear_bit(LIOD_START, &pc->pc_flags);
+ cfs_clear_bit(LIOD_STOP, &pc->pc_flags);
+ cfs_clear_bit(LIOD_FORCE, &pc->pc_flags);
return 0;
}
-static void ptlrpcd_zombie_impexp_notify(void)
-{
- cfs_waitq_signal(&ptlrpcd_pc.pc_waitq);
-}
-#else
+#else /* !__KERNEL__ */
int ptlrpcd_check_async_rpcs(void *arg)
{
struct ptlrpcd_ctl *pc = arg;
- int rc = 0;
+ int rc = 0;
- /* single threaded!! */
+ /*
+ * Single threaded!!
+ */
pc->pc_recurred++;
if (pc->pc_recurred == 1) {
- rc = ptlrpcd_check(pc);
- if (!rc)
- ptlrpc_expired_set(pc->pc_set);
- /*XXX send replay requests */
- if (pc == &ptlrpcd_recovery_pc)
- rc = ptlrpcd_check(pc);
+ rc = lu_env_refill(&pc->pc_env);
+ if (rc == 0) {
+ lu_context_enter(&pc->pc_env.le_ctx);
+ rc = ptlrpcd_check(&pc->pc_env, pc);
+ lu_context_exit(&pc->pc_env.le_ctx);
+ if (!rc)
+ ptlrpc_expired_set(pc->pc_set);
+ /*
+ * XXX: send replay requests.
+ */
+ if (cfs_test_bit(LIOD_RECOVERY, &pc->pc_flags))
+ rc = ptlrpcd_check(&pc->pc_env, pc);
+ }
}
pc->pc_recurred--;
return rc;
}
+
+int ptlrpcd_idle(void *arg)
+{
+ struct ptlrpcd_ctl *pc = arg;
+
+ return (cfs_list_empty(&pc->pc_set->set_new_requests) &&
+ pc->pc_set->set_remaining == 0);
+}
+
#endif
-static int ptlrpcd_start(char *name, struct ptlrpcd_ctl *pc)
+int ptlrpcd_start(const char *name, struct ptlrpcd_ctl *pc)
{
int rc;
-
ENTRY;
- memset(pc, 0, sizeof(*pc));
- init_completion(&pc->pc_starting);
- init_completion(&pc->pc_finishing);
- cfs_waitq_init(&pc->pc_waitq);
- pc->pc_flags = 0;
- spin_lock_init(&pc->pc_lock);
- CFS_INIT_LIST_HEAD(&pc->pc_req_list);
- snprintf (pc->pc_name, sizeof (pc->pc_name), name);
+ /*
+ * Do not allow start second thread for one pc.
+ */
+ if (cfs_test_and_set_bit(LIOD_START, &pc->pc_flags)) {
+ CERROR("Starting second thread (%s) for same pc %p\n",
+ name, pc);
+ RETURN(-EALREADY);
+ }
+
+ cfs_init_completion(&pc->pc_starting);
+ cfs_init_completion(&pc->pc_finishing);
+ cfs_spin_lock_init(&pc->pc_lock);
+ strncpy(pc->pc_name, name, sizeof(pc->pc_name) - 1);
pc->pc_set = ptlrpc_prep_set();
if (pc->pc_set == NULL)
- RETURN(-ENOMEM);
+ GOTO(out, rc = -ENOMEM);
+ /*
+ * So far only "client" ptlrpcd uses an environment. In the future,
+ * ptlrpcd thread (or a thread-set) has to be given an argument,
+ * describing its "scope".
+ */
+ rc = lu_context_init(&pc->pc_env.le_ctx, LCT_CL_THREAD|LCT_REMEMBER);
+ if (rc != 0) {
+ ptlrpc_set_destroy(pc->pc_set);
+ GOTO(out, rc);
+ }
#ifdef __KERNEL__
- /* wake ptlrpcd when zombie imports or exports exist */
- obd_zombie_impexp_notify = ptlrpcd_zombie_impexp_notify;
-
rc = cfs_kernel_thread(ptlrpcd, pc, 0);
if (rc < 0) {
+ lu_context_fini(&pc->pc_env.le_ctx);
ptlrpc_set_destroy(pc->pc_set);
- RETURN(rc);
+ GOTO(out, rc);
}
-
- wait_for_completion(&pc->pc_starting);
+ rc = 0;
+ cfs_wait_for_completion(&pc->pc_starting);
#else
- pc->pc_callback =
- liblustre_register_wait_callback(&ptlrpcd_check_async_rpcs, pc);
- (void)rc;
+ pc->pc_wait_callback =
+ liblustre_register_wait_callback("ptlrpcd_check_async_rpcs",
+ &ptlrpcd_check_async_rpcs, pc);
+ pc->pc_idle_callback =
+ liblustre_register_idle_callback("ptlrpcd_check_idle_rpcs",
+ &ptlrpcd_idle, pc);
#endif
- RETURN(0);
+out:
+ if (rc)
+ cfs_clear_bit(LIOD_START, &pc->pc_flags);
+ RETURN(rc);
}
-static void ptlrpcd_stop(struct ptlrpcd_ctl *pc)
+void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force)
{
- set_bit(LIOD_STOP, &pc->pc_flags);
- cfs_waitq_signal(&pc->pc_waitq);
+ if (!cfs_test_bit(LIOD_START, &pc->pc_flags)) {
+ CERROR("Thread for pc %p was not started\n", pc);
+ return;
+ }
+
+ cfs_set_bit(LIOD_STOP, &pc->pc_flags);
+ if (force)
+ cfs_set_bit(LIOD_FORCE, &pc->pc_flags);
+ cfs_waitq_signal(&pc->pc_set->set_waitq);
#ifdef __KERNEL__
- obd_zombie_impexp_notify = NULL;
- wait_for_completion(&pc->pc_finishing);
+ cfs_wait_for_completion(&pc->pc_finishing);
#else
- liblustre_deregister_wait_callback(pc->pc_callback);
+ liblustre_deregister_wait_callback(pc->pc_wait_callback);
+ liblustre_deregister_idle_callback(pc->pc_idle_callback);
#endif
+ lu_context_fini(&pc->pc_env.le_ctx);
ptlrpc_set_destroy(pc->pc_set);
}
-int ptlrpcd_addref(void)
+void ptlrpcd_fini(void)
{
- int rc = 0;
+ int i;
+ int j;
+
ENTRY;
- mutex_down(&ptlrpcd_sem);
- if (++ptlrpcd_users != 1)
- GOTO(out, rc);
+ for (i = 0; i < PSCOPE_NR; ++i) {
+ for (j = 0; j < PT_NR; ++j) {
+ struct ptlrpcd_ctl *pc;
- rc = ptlrpcd_start("ptlrpcd", &ptlrpcd_pc);
- if (rc) {
- --ptlrpcd_users;
- GOTO(out, rc);
+ pc = &ptlrpcd_scopes[i].pscope_thread[j].pt_ctl;
+
+ if (cfs_test_bit(LIOD_START, &pc->pc_flags))
+ ptlrpcd_stop(pc, 0);
+ }
}
+ EXIT;
+}
- rc = ptlrpcd_start("ptlrpcd-recov", &ptlrpcd_recovery_pc);
- if (rc) {
- ptlrpcd_stop(&ptlrpcd_pc);
- --ptlrpcd_users;
- GOTO(out, rc);
+int ptlrpcd_addref(void)
+{
+ int rc = 0;
+ int i;
+ int j;
+ ENTRY;
+
+ cfs_mutex_down(&ptlrpcd_sem);
+ if (++ptlrpcd_users == 1) {
+ for (i = 0; rc == 0 && i < PSCOPE_NR; ++i) {
+ for (j = 0; rc == 0 && j < PT_NR; ++j) {
+ struct ptlrpcd_thread *pt;
+ struct ptlrpcd_ctl *pc;
+
+ pt = &ptlrpcd_scopes[i].pscope_thread[j];
+ pc = &pt->pt_ctl;
+ if (j == PT_RECOVERY)
+ cfs_set_bit(LIOD_RECOVERY, &pc->pc_flags);
+ rc = ptlrpcd_start(pt->pt_name, pc);
+ }
+ }
+ if (rc != 0) {
+ --ptlrpcd_users;
+ ptlrpcd_fini();
+ }
}
-out:
- mutex_up(&ptlrpcd_sem);
+ cfs_mutex_up(&ptlrpcd_sem);
RETURN(rc);
}
void ptlrpcd_decref(void)
{
- mutex_down(&ptlrpcd_sem);
- if (--ptlrpcd_users == 0) {
- ptlrpcd_stop(&ptlrpcd_pc);
- ptlrpcd_stop(&ptlrpcd_recovery_pc);
- }
- mutex_up(&ptlrpcd_sem);
+ cfs_mutex_down(&ptlrpcd_sem);
+ if (--ptlrpcd_users == 0)
+ ptlrpcd_fini();
+ cfs_mutex_up(&ptlrpcd_sem);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff