From: Manos Pitsidianakis <el13635@mail.ntua.gr>
To: Kevin Wolf <kwolf@redhat.com>
Cc: qemu-devel <qemu-devel@nongnu.org>,
qemu-block <qemu-block@nongnu.org>,
Stefan Hajnoczi <stefanha@redhat.com>,
Alberto Garcia <berto@igalia.com>, Max Reitz <mreitz@redhat.com>
Subject: Re: [Qemu-devel] [PATCH v2] block: fix leaks in bdrv_open_driver()
Date: Wed, 12 Jul 2017 11:39:30 +0300 [thread overview]
Message-ID: <20170712083930.tkit2nqxsras4trw@postretch> (raw)
In-Reply-To: <20170712083337.GC4917@noname.str.redhat.com>
[-- Attachment #1: Type: text/plain, Size: 4020 bytes --]
On Wed, Jul 12, 2017 at 10:33:37AM +0200, Kevin Wolf wrote:
>Am 11.07.2017 um 20:50 hat Manos Pitsidianakis geschrieben:
>> On Tue, Jul 11, 2017 at 05:16:17PM +0200, Kevin Wolf wrote:
>> >Am 01.07.2017 um 17:39 hat Manos Pitsidianakis geschrieben:
>> >>bdrv_open_driver() is called in two places, bdrv_new_open_driver() and
>> >>bdrv_open_common(). In the latter, failure cleanup in is in its caller,
>> >>bdrv_open_inherit(), which unrefs the bs->file of the failed driver open if it
>> >>exists.
>> >>
>> >>Let's move the bs->file cleanup to bdrv_open_driver() to take care of all
>> >>callers and do not set bs->drv to NULL unless the driver's open function
>> >>failed. When bs is destroyed by removing its last reference, bdrv_close()
>> >>checks bs->drv to perform the needed cleanups and also call the driver's close
>> >>function.
>> >>
>> >>Signed-off-by: Manos Pitsidianakis <el13635@mail.ntua.gr>
>> >>---
>> >>
>> >>v2:
>> >> move bdrv_unref_child(bs, bs->file) to bdrv_open_driver
>> >> do not set bs->drv to NULL if open succeeds
>> >>
>> >> block.c | 21 +++++++++++++--------
>> >> 1 file changed, 13 insertions(+), 8 deletions(-)
>> >>
>> >>diff --git a/block.c b/block.c
>> >>index 694396281b..df2a46990c 100644
>> >>--- a/block.c
>> >>+++ b/block.c
>> >>@@ -1091,6 +1091,7 @@ static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
>> >> {
>> >> Error *local_err = NULL;
>> >> int ret;
>> >>+ bool open_failed;
>> >>
>> >> bdrv_assign_node_name(bs, node_name, &local_err);
>> >> if (local_err) {
>> >>@@ -1111,7 +1112,9 @@ static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
>> >> ret = 0;
>> >> }
>> >>
>> >>- if (ret < 0) {
>> >>+ open_failed = ret < 0;
>> >>+
>> >>+ if (open_failed) {
>> >> if (local_err) {
>> >> error_propagate(errp, local_err);
>> >> } else if (bs->filename[0]) {
>> >>@@ -1142,10 +1145,15 @@ static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
>> >> return 0;
>> >>
>> >> free_and_fail:
>> >>- /* FIXME Close bs first if already opened*/
>> >>- g_free(bs->opaque);
>> >>- bs->opaque = NULL;
>> >>- bs->drv = NULL;
>> >>+ if (open_failed) {
>> >>+ g_free(bs->opaque);
>> >>+ bs->opaque = NULL;
>> >>+ bs->drv = NULL;
>> >>+ }
>> >>+ if (bs->file != NULL) {
>> >>+ bdrv_unref_child(bs, bs->file);
>> >>+ bs->file = NULL;
>> >>+ }
>> >
>> >Is this bdrv_unref_child() safe if we leave bs->drv set? Format drivers
>> >expect that if an image is opened, it also has a valid bs->file.
>> >
>> >For example, if I add ret = -1 after refresh_total_sectors() (because I
>> >couldn't find an easier way to make it fail intentionally), I get an
>> >ugly heap corruption crash instead of a nice error message with this
>> >patch.
>> >
>> This is triggered by bdrv_open_inherit doing
>> QDECREF(bs->explicit_options) and leaving the dangling pointer. Not
>> setting bs->drv means bdrv_close was called and tried to decref it
>> again, causing the heap error. Setting bs->explicit_options = NULL;
>> right below that fixes the heap corruption for me.
>
>Wouldn't it be better to call drv->bdrv_close() instead and then set
>bs->drv/opaque = NULL like for the other error path?
That was my first approach but I thought it wouldn't look nice since
bdrv_close is called anyway on delete. I will do it in the next version.
>
>> I can send a seperate fix for this.
>
>No, this doesn't fail before this patch, so it's a regression and we
>can't merge the patch without a fix. You need to respin this one.
Yes, I meant to first fix this and then apply this patch. It's a
dangling pointer anyway.
>
>> I also saw that there's no reason to use a boolean, a label would do
>> just fine so I can change that and finalize the patch in the next
>> version if everything is okay with it.
>
>Yes, that sounds better.
>
>Kevin
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
prev parent reply other threads:[~2017-07-12 8:40 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-07-01 15:39 [Qemu-devel] [PATCH v2] block: fix leaks in bdrv_open_driver() Manos Pitsidianakis
2017-07-11 15:16 ` Kevin Wolf
2017-07-11 18:50 ` Manos Pitsidianakis
2017-07-12 8:33 ` Kevin Wolf
2017-07-12 8:39 ` Manos Pitsidianakis [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20170712083930.tkit2nqxsras4trw@postretch \
--to=el13635@mail.ntua.gr \
--cc=berto@igalia.com \
--cc=kwolf@redhat.com \
--cc=mreitz@redhat.com \
--cc=qemu-block@nongnu.org \
--cc=qemu-devel@nongnu.org \
--cc=stefanha@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).