From: Jan Kiszka <jan.kiszka@web.de>
To: zwu.kernel@gmail.com
Cc: aliguori@us.ibm.com, Zhi Yong Wu <wuzhy@linux.vnet.ibm.com>,
qemu-devel@nongnu.org, stefanha@linux.vnet.ibm.com,
mst@redhat.com
Subject: Re: [Qemu-devel] [PATCH 2/2] slirp: fix packet requeue issue in batchq
Date: Wed, 15 Feb 2012 09:30:46 +0100 [thread overview]
Message-ID: <4F3B6D36.1030808@web.de> (raw)
In-Reply-To: <1329293587-16246-1-git-send-email-zwu.kernel@gmail.com>
[-- Attachment #1: Type: text/plain, Size: 3762 bytes --]
On 2012-02-15 09:13, zwu.kernel@gmail.com wrote:
> From: Zhi Yong Wu <wuzhy@linux.vnet.ibm.com>
>
> This patch fixes the slirp crash in current QEMU upstream.
>
> Signed-off-by: Zhi Yong Wu <wuzhy@linux.vnet.ibm.com>
> ---
> slirp/if.c | 37 ++++++++++++++++++++++++++++++-------
> slirp/mbuf.c | 3 +--
> 2 files changed, 31 insertions(+), 9 deletions(-)
>
> diff --git a/slirp/if.c b/slirp/if.c
> index 8e0cac2..f7f8577 100644
> --- a/slirp/if.c
> +++ b/slirp/if.c
> @@ -20,8 +20,15 @@ ifs_insque(struct mbuf *ifm, struct mbuf *ifmhead)
> static void
> ifs_remque(struct mbuf *ifm)
> {
> - ifm->ifs_prev->ifs_next = ifm->ifs_next;
> - ifm->ifs_next->ifs_prev = ifm->ifs_prev;
> + if (ifm->ifs_next->ifs_next == ifm
> + && ifm->ifs_next->ifs_prev == ifm) {
> + ifs_init(ifm->ifs_next);
> + } else {
> + ifm->ifs_prev->ifs_next = ifm->ifs_next;
> + ifm->ifs_next->ifs_prev = ifm->ifs_prev;
> + }
> +
> + ifs_init(ifm);
This looks, well, interesting. Can you explain the logic? We really need
to document the queuing mechanics.
> }
>
> void
> @@ -154,14 +161,18 @@ if_start(Slirp *slirp)
> {
> uint64_t now = qemu_get_clock_ns(rt_clock);
> int requeued = 0;
> - struct mbuf *ifm, *ifqt;
> + struct mbuf *ifm, *ifqt, *ifm_next;
>
> - DEBUG_CALL("if_start");
> + DEBUG_CALL("if_start");
>
> - if (slirp->if_queued == 0)
> - return; /* Nothing to do */
> + if (slirp->if_queued == 0)
> + return; /* Nothing to do */
Even if the result looks funny, let's not touch lines just for indention
(and braces would be missing anyway).
> +
> + slirp->next_m = &slirp->if_batchq;
Have you understood the difference between the natural order of
if_batchq and next_m? I still wonder what the latter is good for.
>
> again:
> + ifm_next = NULL;
> +
> /* check if we can really output */
> if (!slirp_can_output(slirp->opaque))
> return;
> @@ -190,6 +201,7 @@ if_start(Slirp *slirp)
> /* If there are more packets for this session, re-queue them */
> if (ifm->ifs_next != /* ifm->ifs_prev != */ ifm) {
> insque(ifm->ifs_next, ifqt);
> + ifm_next = ifm->ifs_next;
> ifs_remque(ifm);
> }
>
> @@ -209,7 +221,18 @@ if_start(Slirp *slirp)
> m_free(ifm);
> } else {
> /* re-queue */
> - insque(ifm, ifqt);
> + if (ifm_next) {
> + /*restore the original state of bachq*/
> + remque(ifm_next);
> + insque(ifm, ifqt);
> + ifm_next->ifs_prev->ifs_next = ifm;
> + ifm->ifs_prev = ifm_next->ifs_prev;
> + ifm->ifs_next = ifm_next;
> + ifm_next->ifs_prev = ifm;
So is this only about the correct ordering or also about pointer
correctness?
> + } else {
> + insque(ifm, ifqt);
> + }
> +
> requeued++;
> }
> }
> diff --git a/slirp/mbuf.c b/slirp/mbuf.c
> index c699c75..f429c0a 100644
> --- a/slirp/mbuf.c
> +++ b/slirp/mbuf.c
> @@ -68,8 +68,7 @@ m_get(Slirp *slirp)
> m->m_size = SLIRP_MSIZE - offsetof(struct mbuf, m_dat);
> m->m_data = m->m_dat;
> m->m_len = 0;
> - m->m_nextpkt = NULL;
> - m->m_prevpkt = NULL;
> + ifs_init(m);
> m->arp_requested = false;
> m->expiration_date = (uint64_t)-1;
> end_error:
Thanks for digging into this, but I really think it needs more comments
and potentially even some cleanups.
Jan
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]
next prev parent reply other threads:[~2012-02-15 8:31 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-02-15 8:13 [Qemu-devel] [PATCH 2/2] slirp: fix packet requeue issue in batchq zwu.kernel
2012-02-15 8:30 ` Jan Kiszka [this message]
2012-02-16 7:34 ` Zhi Yong Wu
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=4F3B6D36.1030808@web.de \
--to=jan.kiszka@web.de \
--cc=aliguori@us.ibm.com \
--cc=mst@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=stefanha@linux.vnet.ibm.com \
--cc=wuzhy@linux.vnet.ibm.com \
--cc=zwu.kernel@gmail.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).