From: Richard Palethorpe Date: Wed, 25 Sep 2019 14:21:36 +0200 Subject: [LTP] [Patch v4] fzsync: revoke thread_b if parent hits accidental break In-Reply-To: <20190925103813.16078-1-liwang@redhat.com> References: <20190925103813.16078-1-liwang@redhat.com> Message-ID: <87blv8y3an.fsf@rpws.prws.suse.cz> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ltp@lists.linux.it Hello, Li Wang writes: > We shouldn't rely entirely on the pair->exit flag in tst_fzsync_pair_cleanup() > since there is possible to call tst_brk() at anyplace of thread_a, that will > lead to timeout eventually because of thread_b(tst_fzsync_wait_b) fall into > an infinite(no explicit condition to exit) loop. > > Thread_a path trace: > tst_brk() > cleanup() > tst_fzsync_pair_cleanup() > SAFE_PTHREAD_JOIN(pair->thread_b, NULL) > #Or pthread_cancel(pair->thread_b) > > We fix the problem via a way to kill thread_b with pthread_cancel. With new > wrap_run_b involves enabling thread cancel at the start of the thread B, > then do asynchronous cancellation in tst_fzsync_pair_cleanup if pair->exit > is not being set to 1. > > Workaround: [commit 2e74d996: Check recvmmsg exists before entering fuzzy loop] > Signed-off-by: Li Wang > Cc: Richard Palethorpe > Cc: Cyril Hrubis > --- > include/tst_fuzzy_sync.h | 29 +++++++++++++++++++++++++++-- > 1 file changed, 27 insertions(+), 2 deletions(-) > > diff --git a/include/tst_fuzzy_sync.h b/include/tst_fuzzy_sync.h > index f9a1947c7..00c74e951 100644 > --- a/include/tst_fuzzy_sync.h > +++ b/include/tst_fuzzy_sync.h > @@ -63,6 +63,7 @@ > #include > #include > #include > +#include > #include "tst_atomic.h" > #include "tst_timer.h" > #include "tst_safe_pthread.h" > @@ -218,12 +219,36 @@ static void tst_fzsync_pair_init(struct tst_fzsync_pair *pair) > static void tst_fzsync_pair_cleanup(struct tst_fzsync_pair *pair) > { > if (pair->thread_b) { > - tst_atomic_store(1, &pair->exit); > + /* Revoke thread B if parent hits accidental break */ > + if (!pair->exit) { > + tst_atomic_store(1, &pair->exit); > + usleep(100000); > + pthread_cancel(pair->thread_b); > + pair->thread_b = 0; We still need to join after using cancel, or? > + return; > + } > SAFE_PTHREAD_JOIN(pair->thread_b, NULL); > pair->thread_b = 0; > } > } > > +/** > + * Wrap run_b for tst_fzsync_pair_reset to enable pthread cancel > + * at the start of the thread B. > + */ > +static void *wrap_run_b(void *run_b) > +{ > + void *(*real_run_b)(void *) = run_b; The C standard doesn't actually allow this, except maybe C11. See my other e-mail to patch V3. > + > + if (real_run_b) { > + pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, NULL); > + pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL); > + (*real_run_b)(NULL); > + } > + > + return NULL; > +} > + > /** > * Zero some stat fields > * > @@ -272,7 +297,7 @@ static void tst_fzsync_pair_reset(struct tst_fzsync_pair *pair, > pair->b_cntr = 0; > pair->exit = 0; > if (run_b) > - SAFE_PTHREAD_CREATE(&pair->thread_b, 0, run_b, 0); > + SAFE_PTHREAD_CREATE(&pair->thread_b, 0, wrap_run_b, run_b); > > pair->exec_time_start = (float)tst_timeout_remaining(); > } -- Thank you, Richard.