From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:60998) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XTPsj-0001sa-9r for qemu-devel@nongnu.org; Mon, 15 Sep 2014 02:35:31 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1XTPsd-0006TN-4k for qemu-devel@nongnu.org; Mon, 15 Sep 2014 02:35:25 -0400 Received: from mx1.redhat.com ([209.132.183.28]:18998) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XTPsc-0006TJ-UI for qemu-devel@nongnu.org; Mon, 15 Sep 2014 02:35:19 -0400 From: Markus Armbruster References: <1410549984-16110-1-git-send-email-armbru@redhat.com> <1410549984-16110-2-git-send-email-armbru@redhat.com> <54148EBA.2070305@redhat.com> Date: Mon, 15 Sep 2014 08:35:14 +0200 In-Reply-To: <54148EBA.2070305@redhat.com> (Max Reitz's message of "Sat, 13 Sep 2014 20:36:42 +0200") Message-ID: <8738bts865.fsf@blackfin.pond.sub.org> MIME-Version: 1.0 Content-Type: text/plain Subject: Re: [Qemu-devel] [PATCH 1/4] blockdev: Disentangle BlockDriverState and DriveInfo creation List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Max Reitz Cc: kwolf@redhat.com, benoit.canet@irqsave.net, qemu-devel@nongnu.org, stefanha@redhat.com Max Reitz writes: > On 12.09.2014 21:26, Markus Armbruster wrote: >> blockdev_init() mixes up BlockDriverState and DriveInfo initialization >> Finish the BlockDriverState job before starting to mess with >> DriveInfo. Easier on the eyes. >> >> Signed-off-by: Markus Armbruster >> --- >> blockdev.c | 43 +++++++++++++++++++++++-------------------- >> 1 file changed, 23 insertions(+), 20 deletions(-) >> >> diff --git a/blockdev.c b/blockdev.c >> index b361fbb..5ec4635 100644 >> --- a/blockdev.c >> +++ b/blockdev.c >> @@ -301,6 +301,7 @@ static DriveInfo *blockdev_init(const char *file, QDict *bs_opts, >> int ro = 0; >> int bdrv_flags = 0; >> int on_read_error, on_write_error; >> + BlockDriverState *bs; >> DriveInfo *dinfo; >> ThrottleConfig cfg; >> int snapshot = 0; >> @@ -456,26 +457,27 @@ static DriveInfo *blockdev_init(const char *file, QDict *bs_opts, >> } >> /* init */ >> + bs = bdrv_new(qemu_opts_id(opts), errp); >> + if (!bs) { >> + goto early_err; >> + } >> + bs->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0; >> + bs->read_only = ro; >> + bs->detect_zeroes = detect_zeroes; >> + >> + bdrv_set_on_error(bs, on_read_error, on_write_error); >> + >> + /* disk I/O throttling */ >> + if (throttle_enabled(&cfg)) { >> + bdrv_io_limits_enable(bs); >> + bdrv_set_io_limits(bs, &cfg); >> + } >> + >> dinfo = g_malloc0(sizeof(*dinfo)); > > Could've changed this to g_new0 in the process, but you're the expert > for that, so I'll leave it up to you. ;-) When I made block use g_new() & friends, I only converted patterns like p = g_malloc(sizeof(T)), not patterns like p = g_malloc(sizeof(*p)). In the former case, p = g_new(T) is a clear improvement, because now the compiler checks T matches typeof(*p). In the latter case, we trade some visible obviousness for type safety. Matter of taste. If we agree to prefer type safety in block land, I'll gladly do the conversion work. >> dinfo->id = g_strdup(qemu_opts_id(opts)); >> - dinfo->bdrv = bdrv_new(dinfo->id, &error); >> - if (error) { >> - error_propagate(errp, error); >> - goto bdrv_new_err; >> - } >> - dinfo->bdrv->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0; >> - dinfo->bdrv->read_only = ro; >> - dinfo->bdrv->detect_zeroes = detect_zeroes; >> + dinfo->bdrv = bs; >> QTAILQ_INSERT_TAIL(&drives, dinfo, next); >> - bdrv_set_on_error(dinfo->bdrv, on_read_error, >> on_write_error); >> - >> - /* disk I/O throttling */ >> - if (throttle_enabled(&cfg)) { >> - bdrv_io_limits_enable(dinfo->bdrv); >> - bdrv_set_io_limits(dinfo->bdrv, &cfg); >> - } >> - >> if (!file || !*file) { >> if (has_driver_specific_opts) { >> file = NULL; >> @@ -502,7 +504,8 @@ static DriveInfo *blockdev_init(const char *file, QDict *bs_opts, >> bdrv_flags |= ro ? 0 : BDRV_O_RDWR; >> QINCREF(bs_opts); >> - ret = bdrv_open(&dinfo->bdrv, file, NULL, bs_opts, bdrv_flags, drv, &error); >> + ret = bdrv_open(&bs, file, NULL, bs_opts, bdrv_flags, drv, &error); >> + assert(bs == dinfo->bdrv); > > Well, this is guaranteed by bdrv_open(), but of course better having > too many assertions than too few. Indeed. bdrv_open() is in another file, and its function comment doesn't quite spell out this part of its contract. Assertions do double-duty: they check assumptions are valid, and they remind the reader of assumptions. The second part can be useful even when the first part is trivial. > With or without g_new0: > > Reviewed-by: Max Reitz Thanks!