From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:41014) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1UjyzD-0008Ch-A6 for qemu-devel@nongnu.org; Tue, 04 Jun 2013 17:41:48 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1UjyzC-0005i8-8a for qemu-devel@nongnu.org; Tue, 04 Jun 2013 17:41:47 -0400 Date: Tue, 4 Jun 2013 17:41:41 -0400 (EDT) From: Alon Levy Message-ID: <378221460.14313718.1370382101151.JavaMail.root@redhat.com> In-Reply-To: <51AE5808.6000000@redhat.com> References: <1370377419-31788-1-git-send-email-alevy@redhat.com> <51AE5808.6000000@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH 1/5] oslib-posix: add qemu_pipe_non_block List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Eric Blake Cc: qemu-trivial@nongnu.org, qemu-devel@nongnu.org > On 06/04/2013 02:23 PM, Alon Levy wrote: > > Used by the followin patch. > > s/followin/following/ Thanks. > > > > > +int qemu_pipe_non_block(int pipefd[2]) > > +{ > > + int ret; > > + > > + ret = qemu_pipe(pipefd); > > qemu_pipe() already uses pipe2() when available; it seems like it would > be nicer to use pipe2's O_NONBLOCK option directly in one syscall (where > supported) instead of having to make additional syscalls after the fact. > Would it just be smarter to change the signature of qemu_pipe() to add > a bool block parameter, and then change the 5 existing callers to pass > false with your later patch in the series passing true, and do it > without creating a new wrapper? Answered below. > > > + if (ret) { > > + return ret; > > + } > > + if (fcntl(card->pipe[0], F_SETFL, O_NONBLOCK) == -1) { > > + return -errno; > > + } > > + if (fcntl(card->pipe[1], F_SETFL, O_NONBLOCK) == -1) { > > + return -errno; > > Leaks fds. If you're going to report error, then you must close the fds > already created. As Peter pointed out, I should not go here, so I'll drop these checks, instead doing naked fcntl calls, so no fd leak possible (no returns). > > > + } > > + if (fcntl(card->pipe[0], F_SETOWN, getpid()) == -1) { > > + return -errno; > > Same comment about fd leaks. > > This part seems like a useful change, IF you plan on using SIGIO and > SIGURG signals; and it is something which pipe2() cannot optimize, so I > can see why you are adding a new function instead of changing > qemu_pipe() and adjust all its callers to pass an additional parameter. > But are you really planning on using SIGIO/SIGURG? I don't plan on using those signals, so I'll add a parameter instead. > > Furthermore, this is undefined behavior. According to POSIX, use of > F_SETOWN is only portable on sockets, not pipes. It may work on Linux, > but you'll need to be aware of what it does on other platforms. > http://pubs.opengroup.org/onlinepubs/9699919799/functions/fcntl.html > -- > Eric Blake eblake redhat com +1-919-301-3266 > Libvirt virtualization library http://libvirt.org > >