From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751413AbdBRVu2 (ORCPT ); Sat, 18 Feb 2017 16:50:28 -0500 Received: from [160.91.203.10] ([160.91.203.10]:51102 "EHLO smtp1.ccs.ornl.gov" rhost-flags-FAIL-FAIL-OK-OK) by vger.kernel.org with ESMTP id S1751214AbdBRVuZ (ORCPT ); Sat, 18 Feb 2017 16:50:25 -0500 From: James Simmons To: Greg Kroah-Hartman , devel@driverdev.osuosl.org, Andreas Dilger , Oleg Drokin Cc: Linux Kernel Mailing List , Lustre Development List , Niu Yawei , James Simmons Subject: [PATCH 11/14] staging: lustre: ldlm: fix race of starting bl threads Date: Sat, 18 Feb 2017 16:47:12 -0500 Message-Id: <1487454435-4895-12-git-send-email-jsimmons@infradead.org> X-Mailer: git-send-email 1.8.3.1 In-Reply-To: <1487454435-4895-1-git-send-email-jsimmons@infradead.org> References: <1487454435-4895-1-git-send-email-jsimmons@infradead.org> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org From: Niu Yawei There is race in the code of starting bl threads which leads to thread number exceeds the maximum number when race happened, it can also lead to duplicated thread name. This patch fixes the race and cleanup the code a bit. Signed-off-by: Niu Yawei Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-7330 Reviewed-on: http://review.whamcloud.com/17026 Reviewed-by: Bobi Jam Reviewed-by: Andreas Dilger Signed-off-by: James Simmons --- drivers/staging/lustre/lustre/ldlm/ldlm_lockd.c | 49 ++++++++++++++----------- 1 file changed, 28 insertions(+), 21 deletions(-) diff --git a/drivers/staging/lustre/lustre/ldlm/ldlm_lockd.c b/drivers/staging/lustre/lustre/ldlm/ldlm_lockd.c index 4c21b9b..6f9d540 100644 --- a/drivers/staging/lustre/lustre/ldlm/ldlm_lockd.c +++ b/drivers/staging/lustre/lustre/ldlm/ldlm_lockd.c @@ -714,7 +714,6 @@ static int ldlm_bl_get_work(struct ldlm_bl_pool *blp, /* This only contains temporary data until the thread starts */ struct ldlm_bl_thread_data { - char bltd_name[CFS_CURPROC_COMM_MAX]; struct ldlm_bl_pool *bltd_blp; struct completion bltd_comp; int bltd_num; @@ -722,19 +721,32 @@ struct ldlm_bl_thread_data { static int ldlm_bl_thread_main(void *arg); -static int ldlm_bl_thread_start(struct ldlm_bl_pool *blp) +static int ldlm_bl_thread_start(struct ldlm_bl_pool *blp, bool check_busy) { struct ldlm_bl_thread_data bltd = { .bltd_blp = blp }; struct task_struct *task; init_completion(&bltd.bltd_comp); - bltd.bltd_num = atomic_read(&blp->blp_num_threads); - snprintf(bltd.bltd_name, sizeof(bltd.bltd_name), - "ldlm_bl_%02d", bltd.bltd_num); - task = kthread_run(ldlm_bl_thread_main, &bltd, "%s", bltd.bltd_name); + + bltd.bltd_num = atomic_inc_return(&blp->blp_num_threads); + if (bltd.bltd_num >= blp->blp_max_threads) { + atomic_dec(&blp->blp_num_threads); + return 0; + } + + LASSERTF(bltd.bltd_num > 0, "thread num:%d\n", bltd.bltd_num); + if (check_busy && + atomic_read(&blp->blp_busy_threads) < (bltd.bltd_num - 1)) { + atomic_dec(&blp->blp_num_threads); + return 0; + } + + task = kthread_run(ldlm_bl_thread_main, &bltd, "ldlm_bl_%02d", + bltd.bltd_num); if (IS_ERR(task)) { CERROR("cannot start LDLM thread ldlm_bl_%02d: rc %ld\n", - atomic_read(&blp->blp_num_threads), PTR_ERR(task)); + bltd.bltd_num, PTR_ERR(task)); + atomic_dec(&blp->blp_num_threads); return PTR_ERR(task); } wait_for_completion(&bltd.bltd_comp); @@ -746,12 +758,11 @@ static int ldlm_bl_thread_start(struct ldlm_bl_pool *blp) static int ldlm_bl_thread_need_create(struct ldlm_bl_pool *blp, struct ldlm_bl_work_item *blwi) { - int busy = atomic_read(&blp->blp_busy_threads); - - if (busy >= blp->blp_max_threads) + if (atomic_read(&blp->blp_num_threads) >= blp->blp_max_threads) return 0; - if (busy < atomic_read(&blp->blp_num_threads)) + if (atomic_read(&blp->blp_busy_threads) < + atomic_read(&blp->blp_num_threads)) return 0; if (blwi && (!blwi->blwi_ns || blwi->blwi_mem_pressure)) @@ -815,9 +826,6 @@ static int ldlm_bl_thread_main(void *arg) blp = bltd->bltd_blp; - atomic_inc(&blp->blp_num_threads); - atomic_inc(&blp->blp_busy_threads); - complete(&bltd->bltd_comp); /* cannot use bltd after this, it is only on caller's stack */ @@ -828,27 +836,26 @@ static int ldlm_bl_thread_main(void *arg) int rc; rc = ldlm_bl_get_work(blp, &blwi, &exp); - if (!rc) { - atomic_dec(&blp->blp_busy_threads); + if (!rc) l_wait_event_exclusive(blp->blp_waitq, ldlm_bl_get_work(blp, &blwi, &exp), &lwi); - atomic_inc(&blp->blp_busy_threads); - } + atomic_inc(&blp->blp_busy_threads); if (ldlm_bl_thread_need_create(blp, blwi)) /* discard the return value, we tried */ - ldlm_bl_thread_start(blp); + ldlm_bl_thread_start(blp, true); if (blwi) rc = ldlm_bl_thread_blwi(blp, blwi); + atomic_dec(&blp->blp_busy_threads); + if (rc == LDLM_ITER_STOP) break; } - atomic_dec(&blp->blp_busy_threads); atomic_dec(&blp->blp_num_threads); complete(&blp->blp_comp); return 0; @@ -1028,7 +1035,7 @@ static int ldlm_setup(void) } for (i = 0; i < blp->blp_min_threads; i++) { - rc = ldlm_bl_thread_start(blp); + rc = ldlm_bl_thread_start(blp, false); if (rc < 0) goto out; } -- 1.8.3.1