Message ID | 20200512142824.13063-2-rpalethorpe@suse.com |
---|---|
State | Superseded |
Headers | show |
Series | [1/2] pty04: Remove unnecessary volatile and style fix | expand |
Hi! > +static ssize_t try_read(int fd, char *data, ssize_t size, ssize_t *n) > +{ > + ssize_t ret = read(fd, data, size); > + > + if (ret < 0) > + return -(errno != EAGAIN); > + > + return (*n += ret) >= size; > +} I had to read this piece twice, but I think it's correct. > static void read_netdev(const struct ldisc_info *ldisc) > { > - int rlen, plen = 0; > + int ret, rlen, plen = 0; > + ssize_t n; > char *data; > > switch (ldisc->n) { > @@ -305,20 +316,27 @@ static void read_netdev(const struct ldisc_info *ldisc) > > tst_res(TINFO, "Reading from socket %d", sk); > > - SAFE_READ(1, sk, data, plen); > + n = 0; > + ret = TST_RETRY_FUNC(try_read(sk, data, plen, &n), TST_RETVAL_NOTNULL); > + if (ret < 0) > + tst_brk(TBROK | TERRNO, "Read %zd of %d bytes", n, plen); I wonder if a simple loop without exponential backoff would suffice here. A least the code would probably be more readable. > check_data(ldisc, data, plen); > tst_res(TPASS, "Read netdev 1"); > > - SAFE_READ(1, sk, data, plen); > + n = 0; > + ret = TST_RETRY_FUNC(try_read(sk, data, plen, &n), TST_RETVAL_NOTNULL); > + if (ret < 0) > + tst_brk(TBROK | TERRNO, "Read %zd of %d bytes", n, plen); > check_data(ldisc, data, plen); > tst_res(TPASS, "Read netdev 2"); > > TST_CHECKPOINT_WAKE(0); > - while((rlen = read(sk, data, plen)) > 0) > + while ((rlen = read(sk, data, plen)) > 0) > check_data(ldisc, data, rlen); This should have been part of the previous cleanup patch. > tst_res(TPASS, "Reading data from netdev interrupted by hangup"); > > + close(sk); > tst_free_all(); > } > > @@ -357,6 +375,7 @@ static void cleanup(void) > { > ioctl(pts, TIOCVHANGUP); > ioctl(ptmx, TIOCVHANGUP); > + close(sk); > > tst_reap_children(); > } > -- > 2.26.1 > > > -- > Mailing list info: https://lists.linux.it/listinfo/ltp
Hi Richard, > Even though reads are blocking and packets are flipped into the netdevice > buffer whole, it seems read may return before a full packet is read into user > land. Retrying read should prevent timeouts and read failures on some > machines. > Signed-off-by: Richard Palethorpe <rpalethorpe@suse.com> > Reported-by: Jan Stancek <jstancek@redhat.com> > Cc: Petr Vorel <pvorel@suse.cz> Reviewed-by: Petr Vorel <pvorel@suse.cz> Thanks for taking care of this. It's still possible to reproduce timeout just with higher -i (-i10), but it's an improvement. Kind regards, Petr
----- Original Message ----- > Hi Richard, > > > Even though reads are blocking and packets are flipped into the netdevice > > buffer whole, it seems read may return before a full packet is read into > > user > > land. Retrying read should prevent timeouts and read failures on some > > machines. > > > Signed-off-by: Richard Palethorpe <rpalethorpe@suse.com> > > Reported-by: Jan Stancek <jstancek@redhat.com> > > Cc: Petr Vorel <pvorel@suse.cz> > > Reviewed-by: Petr Vorel <pvorel@suse.cz> > > Thanks for taking care of this. ACK, I'll update github issue when this runs for a while with recent upstream kernels. > It's still possible to reproduce timeout just with higher -i (-i10), > but it's an improvement. > > Kind regards, > Petr > >
diff --git a/testcases/kernel/pty/pty04.c b/testcases/kernel/pty/pty04.c index bfda08b2b..9b2911421 100644 --- a/testcases/kernel/pty/pty04.c +++ b/testcases/kernel/pty/pty04.c @@ -288,9 +288,20 @@ static void check_data(const struct ldisc_info *ldisc, tst_res(TINFO, "Will continue test without data checking"); } +static ssize_t try_read(int fd, char *data, ssize_t size, ssize_t *n) +{ + ssize_t ret = read(fd, data, size); + + if (ret < 0) + return -(errno != EAGAIN); + + return (*n += ret) >= size; +} + static void read_netdev(const struct ldisc_info *ldisc) { - int rlen, plen = 0; + int ret, rlen, plen = 0; + ssize_t n; char *data; switch (ldisc->n) { @@ -305,20 +316,27 @@ static void read_netdev(const struct ldisc_info *ldisc) tst_res(TINFO, "Reading from socket %d", sk); - SAFE_READ(1, sk, data, plen); + n = 0; + ret = TST_RETRY_FUNC(try_read(sk, data, plen, &n), TST_RETVAL_NOTNULL); + if (ret < 0) + tst_brk(TBROK | TERRNO, "Read %zd of %d bytes", n, plen); check_data(ldisc, data, plen); tst_res(TPASS, "Read netdev 1"); - SAFE_READ(1, sk, data, plen); + n = 0; + ret = TST_RETRY_FUNC(try_read(sk, data, plen, &n), TST_RETVAL_NOTNULL); + if (ret < 0) + tst_brk(TBROK | TERRNO, "Read %zd of %d bytes", n, plen); check_data(ldisc, data, plen); tst_res(TPASS, "Read netdev 2"); TST_CHECKPOINT_WAKE(0); - while((rlen = read(sk, data, plen)) > 0) + while ((rlen = read(sk, data, plen)) > 0) check_data(ldisc, data, rlen); tst_res(TPASS, "Reading data from netdev interrupted by hangup"); + close(sk); tst_free_all(); } @@ -357,6 +375,7 @@ static void cleanup(void) { ioctl(pts, TIOCVHANGUP); ioctl(ptmx, TIOCVHANGUP); + close(sk); tst_reap_children(); }
Even though reads are blocking and packets are flipped into the netdevice buffer whole, it seems read may return before a full packet is read into user land. Retrying read should prevent timeouts and read failures on some machines. Signed-off-by: Richard Palethorpe <rpalethorpe@suse.com> Reported-by: Jan Stancek <jstancek@redhat.com> Cc: Petr Vorel <pvorel@suse.cz> --- NOTE: When running this test repeatedly and in parallel; strange things still happen, see: https://github.com/linux-test-project/ltp/issues/674#issuecomment-625181783 However this hopefully will fix the timeout issues and read failures reported by some testers. testcases/kernel/pty/pty04.c | 27 +++++++++++++++++++++++---- 1 file changed, 23 insertions(+), 4 deletions(-)