From owner-svn-src-stable-10@freebsd.org Sat Oct 3 07:59:28 2015 Return-Path: Delivered-To: svn-src-stable-10@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id E843BA0D1A6; Sat, 3 Oct 2015 07:59:27 +0000 (UTC) (envelope-from mav@FreeBSD.org) Received: from repo.freebsd.org (repo.freebsd.org [IPv6:2001:1900:2254:2068::e6a:0]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id D4F311012; Sat, 3 Oct 2015 07:59:27 +0000 (UTC) (envelope-from mav@FreeBSD.org) Received: from repo.freebsd.org ([127.0.1.70]) by repo.freebsd.org (8.15.2/8.15.2) with ESMTP id t937xR8s079153; Sat, 3 Oct 2015 07:59:27 GMT (envelope-from mav@FreeBSD.org) Received: (from mav@localhost) by repo.freebsd.org (8.15.2/8.15.2/Submit) id t937xQoa079143; Sat, 3 Oct 2015 07:59:26 GMT (envelope-from mav@FreeBSD.org) Message-Id: <201510030759.t937xQoa079143@repo.freebsd.org> X-Authentication-Warning: repo.freebsd.org: mav set sender to mav@FreeBSD.org using -f From: Alexander Motin Date: Sat, 3 Oct 2015 07:59:26 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-10@freebsd.org Subject: svn commit: r288570 - in stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs: . sys X-SVN-Group: stable-10 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-stable-10@freebsd.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: SVN commit messages for only the 10-stable src tree List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 03 Oct 2015 07:59:28 -0000 Author: mav Date: Sat Oct 3 07:59:25 2015 New Revision: 288570 URL: https://svnweb.freebsd.org/changeset/base/288570 Log: MFC r286689: 5981 Deadlock in dmu_objset_find_dp illumos/illumos-gate@1d3f896f5469c69c1339890ec3d68e9feddb0343 https://www.illumos.org/issues/5981 When dmu_objset_find_dp gets called with a read lock held, it fans out the work to the task queue. Each task in turn acquires its own read lock before calling the callback. If during this process anyone tries to a acquire a write lock, it will stall all read lock requests.Thus the tasks will never finish, the read lock of the caller will never get freed and the write lock never acquired. deadlock. Reviewed by: Matthew Ahrens Reviewed by: Dan McDonald Approved by: Robert Mustacchi Author: Arne Jansen Modified: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_objset.c stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_pool.c stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/rrwlock.c stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/dsl_pool.h stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/rrwlock.h Directory Properties: stable/10/ (props changed) Modified: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_objset.c ============================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_objset.c Sat Oct 3 07:58:28 2015 (r288569) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_objset.c Sat Oct 3 07:59:25 2015 (r288570) @@ -1746,7 +1746,15 @@ dmu_objset_find_dp_cb(void *arg) dmu_objset_find_ctx_t *dcp = arg; dsl_pool_t *dp = dcp->dc_dp; - dsl_pool_config_enter(dp, FTAG); + /* + * We need to get a pool_config_lock here, as there are several + * asssert(pool_config_held) down the stack. Getting a lock via + * dsl_pool_config_enter is risky, as it might be stalled by a + * pending writer. This would deadlock, as the write lock can + * only be granted when our parent thread gives up the lock. + * The _prio interface gives us priority over a pending writer. + */ + dsl_pool_config_enter_prio(dp, FTAG); dmu_objset_find_dp_impl(dcp); Modified: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_pool.c ============================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_pool.c Sat Oct 3 07:58:28 2015 (r288569) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_pool.c Sat Oct 3 07:59:25 2015 (r288570) @@ -1143,6 +1143,13 @@ dsl_pool_config_enter(dsl_pool_t *dp, vo } void +dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag) +{ + ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); + rrw_enter_read_prio(&dp->dp_config_rwlock, tag); +} + +void dsl_pool_config_exit(dsl_pool_t *dp, void *tag) { rrw_exit(&dp->dp_config_rwlock, tag); Modified: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/rrwlock.c ============================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/rrwlock.c Sat Oct 3 07:58:28 2015 (r288569) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/rrwlock.c Sat Oct 3 07:59:25 2015 (r288570) @@ -159,8 +159,8 @@ rrw_destroy(rrwlock_t *rrl) refcount_destroy(&rrl->rr_linked_rcount); } -void -rrw_enter_read(rrwlock_t *rrl, void *tag) +static void +rrw_enter_read_impl(rrwlock_t *rrl, boolean_t prio, void *tag) { mutex_enter(&rrl->rr_lock); #if !defined(DEBUG) && defined(_KERNEL) @@ -176,7 +176,7 @@ rrw_enter_read(rrwlock_t *rrl, void *tag ASSERT(refcount_count(&rrl->rr_anon_rcount) >= 0); while (rrl->rr_writer != NULL || (rrl->rr_writer_wanted && - refcount_is_zero(&rrl->rr_anon_rcount) && + refcount_is_zero(&rrl->rr_anon_rcount) && !prio && rrn_find(rrl) == NULL)) cv_wait(&rrl->rr_cv, &rrl->rr_lock); @@ -192,6 +192,25 @@ rrw_enter_read(rrwlock_t *rrl, void *tag } void +rrw_enter_read(rrwlock_t *rrl, void *tag) +{ + rrw_enter_read_impl(rrl, B_FALSE, tag); +} + +/* + * take a read lock even if there are pending write lock requests. if we want + * to take a lock reentrantly, but from different threads (that have a + * relationship to each other), the normal detection mechanism to overrule + * the pending writer does not work, so we have to give an explicit hint here. + */ +void +rrw_enter_read_prio(rrwlock_t *rrl, void *tag) +{ + rrw_enter_read_impl(rrl, B_TRUE, tag); +} + + +void rrw_enter_write(rrwlock_t *rrl) { mutex_enter(&rrl->rr_lock); Modified: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/dsl_pool.h ============================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/dsl_pool.h Sat Oct 3 07:58:28 2015 (r288569) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/dsl_pool.h Sat Oct 3 07:59:25 2015 (r288570) @@ -152,6 +152,7 @@ void dsl_pool_upgrade_dir_clones(dsl_poo void dsl_pool_mos_diduse_space(dsl_pool_t *dp, int64_t used, int64_t comp, int64_t uncomp); void dsl_pool_config_enter(dsl_pool_t *dp, void *tag); +void dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag); void dsl_pool_config_exit(dsl_pool_t *dp, void *tag); boolean_t dsl_pool_config_held(dsl_pool_t *dp); boolean_t dsl_pool_config_held_writer(dsl_pool_t *dp); Modified: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/rrwlock.h ============================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/rrwlock.h Sat Oct 3 07:58:28 2015 (r288569) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/rrwlock.h Sat Oct 3 07:59:25 2015 (r288570) @@ -69,6 +69,7 @@ void rrw_init(rrwlock_t *rrl, boolean_t void rrw_destroy(rrwlock_t *rrl); void rrw_enter(rrwlock_t *rrl, krw_t rw, void *tag); void rrw_enter_read(rrwlock_t *rrl, void *tag); +void rrw_enter_read_prio(rrwlock_t *rrl, void *tag); void rrw_enter_write(rrwlock_t *rrl); void rrw_exit(rrwlock_t *rrl, void *tag); boolean_t rrw_held(rrwlock_t *rrl, krw_t rw);