DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 2888920854
MIME-Version: 1.0
References: <20181115083817.20986-1-gbenhaim@augury.com> <CABBYNZJ_TEM18yfO6O4HFdyo7sm-cE8iL4Y8_DEa1e6xZ7a=Bg@mail.gmail.com>
In-Reply-To: <CABBYNZJ_TEM18yfO6O4HFdyo7sm-cE8iL4Y8_DEa1e6xZ7a=Bg@mail.gmail.com>
From:   Gal Ben Haim <gbenhaim@augury.com>
Date:   Thu, 15 Nov 2018 12:16:09 +0200
Message-ID: <CAHotPr-jS457EKX=7dvcwA6eB12zr7d2dZG5Prjh6+fMdAT=+Q@mail.gmail.com>
Subject: Re: [PATCH BlueZ] gatt: Ignore SIGPIPE
To:     luiz.dentz@gmail.com
Cc:     linux-bluetooth@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-bluetooth-owner@vger.kernel.org
Precedence: bulk

io_send can use poll and check for POLLOUT and POLLRDHUP before
writing to the pipe

On Thu, Nov 15, 2018 at 12:12 PM Luiz Augusto von Dentz
<luiz.dentz@gmail.com> wrote:
>
> Hi,
> On Thu, Nov 15, 2018 at 10:41 AM Gal <gbenhaim@augury.com> wrote:
> >
> > bluetoothd receives a SIGPIPE and terminates if writing to a pipe that
> > was acquired by AcquireNotify and there are no readers. it can be
> > reproduced by terminating the reader process without closing the reader
> > end of the pipe.
> >
> > Ignoring the SIGPIPE will cause the write operation to return a
> > "io_send: Broken pipe" error which will be logged.
> > ---
> >  src/gatt-client.c | 3 +++
> >  1 file changed, 3 insertions(+)
> >
> > diff --git a/src/gatt-client.c b/src/gatt-client.c
> > index 234f46ed7..236f38ad5 100644
> > --- a/src/gatt-client.c
> > +++ b/src/gatt-client.c
> > @@ -1137,6 +1137,9 @@ static DBusMessage *characteristic_create_pipe(struct characteristic *chrc,
> >         if (pipe2(pipefd, O_DIRECT | O_NONBLOCK | O_CLOEXEC) < 0)
> >                 return btd_error_failed(msg, strerror(errno));
> >
> > +       /* Ignore SIGPIPE, a broken pipe error will be returned if the pipe has no readers */
> > +       signal(SIGPIPE, SIG_IGN);
> > +
> >         dir = dbus_message_has_member(msg, "AcquireWrite");
> >
> >         io = io_new(pipefd[!dir]);
> > --
> > 2.19.1
>
> I wonder if there is a way to detect that on bluetoothd actually, that
> way even if the client don't set it properly it would not make
> bluetoothd to exit.
>
>
> --
> Luiz Augusto von Dentz