From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:45124) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1YXWJS-0007SI-4A for qemu-devel@nongnu.org; Mon, 16 Mar 2015 10:48:15 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1YXWJN-0007hz-53 for qemu-devel@nongnu.org; Mon, 16 Mar 2015 10:48:14 -0400 Message-ID: <5506ED27.6020507@redhat.com> Date: Mon, 16 Mar 2015 10:48:07 -0400 From: Max Reitz MIME-Version: 1.0 References: <1424887718-10800-1-git-send-email-mreitz@redhat.com> <1424887718-10800-6-git-send-email-mreitz@redhat.com> <55002568.2070201@redhat.com> <5506DFE3.2000104@redhat.com> <5506EBE9.7070309@redhat.com> In-Reply-To: <5506EBE9.7070309@redhat.com> Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH 05/25] nbd: Avoid generic -EINVAL List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Paolo Bonzini , qemu-block@nongnu.org Cc: Kevin Wolf , qemu-devel@nongnu.org, Stefan Hajnoczi On 2015-03-16 at 10:42, Paolo Bonzini wrote: > > On 16/03/2015 14:51, Max Reitz wrote: >>> Propagating the return value from write_sync is uglier, but it is even >>> better in terms of returned value. >> We can only return -errno values, but write_sync() may do partial writes >> so it may return non-negative values which still indicate an error. So >> we'd have to check whether the return value is negative, if it is, >> return that, if it isn't but if it's still below what we wanted to >> write, return a fixed error (such as -EIO). I'd rather just return -EIO >> and be done with it, but if you really want me to, I can of course do it >> differently. > nbd_wr_sync doesn't do that, it always returns negative errno for a partial > error: qemu_send() might do a partial send, returning a value smaller than len. nbd_wr_sync() will try iterating until everything has been sent, but if send() returns 0, the loop is aborted and a value smaller than len may be returned. Maybe send() never returns 0, in which case nbd_wr_sync() will actually return either len or -errno, but this isn't clear from the structure of nbd_wr_sync(). If you really want to pass the value returned from nbd_wr_sync(), I'd rather restructure that function so that it always returns either len or -errno. Max > > if (len < 0) { > err = socket_error(); > > /* recoverable error */ > if (err == EINTR || (offset > 0 && (err == EAGAIN || err == EWOULDBLOCK))) { > continue; > } > > /* unrecoverable error */ > return -err; > } > > The precise error can be useful to distinguish a network error from something > else. I'm just in doubt about partial reads; those can return a positive error, > in which case you can return ESHUTDOWN (in read_sync). > > Paolo