From: "Longpeng (Mike, Cloud Infrastructure Service Product Dept.)" <longpeng2@huawei.com>
To: "Daniel P. Berrangé" <berrange@redhat.com>,
"Markus Armbruster" <armbru@redhat.com>
Cc: chenjiashang@huawei.com, qemu-devel@nongnu.org,
Peter Xu <peterx@redhat.com>,
"Gonglei \(Arei\)" <arei.gonglei@huawei.com>,
pbonzini@redhat.com, marcandre.lureau@redhat.com
Subject: Re: A bug of Monitor Chardev ?
Date: Wed, 9 Jun 2021 08:20:56 +0800 [thread overview]
Message-ID: <6528dfe1-7cc7-c530-a56f-06517a627cda@huawei.com> (raw)
In-Reply-To: <YL+Op6F7wFoVYum2@redhat.com>
在 2021/6/8 23:37, Daniel P. Berrangé 写道:
> On Tue, Jun 08, 2021 at 04:07:30PM +0200, Markus Armbruster wrote:
>> "Longpeng (Mike, Cloud Infrastructure Service Product Dept.)"
>> <longpeng2@huawei.com> writes:
>>
>>> We find a race during QEMU starting, which would case the QEMU process coredump.
>>>
>>> <main loop> | <MON iothread>
>>> |
>>> [1] create MON chardev |
>>> qemu_create_early_backends |
>>> chardev_init_func |
>>> |
>>> [2] create MON iothread |
>>> qemu_create_late_backends |
>>> mon_init_func |
>>> aio_bh_schedule-----------------------> monitor_qmp_setup_handlers_bh
>>> [3] enter main loog | tcp_chr_update_read_handler
>>> (* A client come in, e.g. Libvirt *) | update_ioc_handlers
>>> tcp_chr_new_client |
>>> update_ioc_handlers |
>>> |
>>> [4] create new hup_source |
>>> s->hup_source = *PTR1* |
>>> g_source_attach(s->hup_source)|
>>> | [5] remove_hup_source(*PTR1*)
>>> | (create new hup_source)
>>> | s->hup_source = *PTR2*
>>> [6] g_source_attach_unlocked |
>>> *PTR1* is freed by [5] |
>>>
>>> Do you have any suggestion to fix this bug ? Thanks!
>>
>> Do we? We talked, but I'm not sure we reached a conclusion.
>
> Seems like we ended up with two options.
>
> 1. A workaround for the current specific problem by rearranging
> the initilization code in the monitor a little.
>
> 2. A design fix of splitting the chardev creation into two
> parts, one creation, and one activation.
>
> The latter is significantly more work, but is a better long term bet IMHO.
> But what we really is someone motivated to actually implement one of the
> two options.
>
How about the following implementation of option-1 ? We've tested it for several
weeks, it works fine.
diff --git a/chardev/char-socket.c b/chardev/char-socket.c
index a484641..ecb3db9 100644
--- a/chardev/char-socket.c
+++ b/chardev/char-socket.c
@@ -722,6 +722,19 @@ static void tcp_chr_update_read_handler(Chardev *chr)
update_ioc_handlers(s);
}
+static void tcp_chr_disable_handler(Chardev *chr)
+{
+ SocketChardev *s = SOCKET_CHARDEV(chr);
+
+ if (s->listener && s->state == TCP_CHARDEV_STATE_DISCONNECTED) {
+ qio_net_listener_set_client_func_full(s->listener, NULL, NULL,
+ NULL, chr->gcontext);
+ }
+
+ remove_fd_in_watch(chr);
+ remove_hup_source(s);
+}
+
static bool tcp_chr_is_connected(Chardev *chr)
{
SocketChardev *s = SOCKET_CHARDEV(chr);
@@ -1703,6 +1716,7 @@ static void char_socket_class_init(ObjectClass *oc, void
*data)
cc->chr_add_watch = tcp_chr_add_watch;
cc->chr_set_reconnect_time = tcp_chr_set_reconnect_time;
cc->chr_update_read_handler = tcp_chr_update_read_handler;
+ cc->chr_disable_handler = tcp_chr_disable_handler;
cc->chr_is_connected = tcp_chr_is_connected;
cc->chr_get_connect_id = tcp_chr_get_connect_id;
diff --git a/chardev/char.c b/chardev/char.c
index ff0a3cf..990fe4f 100644
--- a/chardev/char.c
+++ b/chardev/char.c
@@ -238,6 +238,15 @@ void qemu_chr_be_update_read_handlers(Chardev *s,
}
}
+void qemu_chr_be_disable_handlers(Chardev *s)
+{
+ ChardevClass *cc = CHARDEV_GET_CLASS(s);
+
+ if (cc->chr_disable_handler) {
+ cc->chr_disable_handler(s);
+ }
+}
+
int qemu_chr_add_client(Chardev *s, int fd)
{
return CHARDEV_GET_CLASS(s)->chr_add_client ?
diff --git a/include/chardev/char.h b/include/chardev/char.h
index d1ec628..7a8c740 100644
--- a/include/chardev/char.h
+++ b/include/chardev/char.h
@@ -212,6 +212,8 @@ void qemu_chr_be_write_impl(Chardev *s, uint8_t *buf, int len);
void qemu_chr_be_update_read_handlers(Chardev *s,
GMainContext *context);
+void qemu_chr_be_disable_handlers(Chardev *s);
+
/**
* qemu_chr_be_event:
* @event: the event to send
@@ -282,6 +284,7 @@ typedef struct ChardevClass {
int (*chr_sync_read)(Chardev *s, const uint8_t *buf, int len);
GSource *(*chr_add_watch)(Chardev *s, GIOCondition cond);
void (*chr_update_read_handler)(Chardev *s);
+ void (*chr_disable_handler)(Chardev *s);
int (*chr_ioctl)(Chardev *s, int cmd, void *arg);
int (*get_msgfds)(Chardev *s, int* fds, int num);
int (*set_msgfds)(Chardev *s, int *fds, int num);
diff --git a/monitor/qmp.c b/monitor/qmp.c
index 9a69ae4..2c2248c 100644
--- a/monitor/qmp.c
+++ b/monitor/qmp.c
@@ -413,11 +413,13 @@ void monitor_init_qmp(Chardev *chr, bool pretty)
* e.g. the chardev is in client mode, with wait=on.
*/
remove_fd_in_watch(chr);
+
/*
- * We can't call qemu_chr_fe_set_handlers() directly here
- * since chardev might be running in the monitor I/O
- * thread. Schedule a bottom half.
+ * Before schedule a bottom half, we should clean up the handler in the
+ * default context to prevent the race between main thread and iothread
*/
+ qemu_chr_be_disable_handlers(chr);
+
aio_bh_schedule_oneshot(iothread_get_aio_context(mon_iothread),
monitor_qmp_setup_handlers_bh, mon);
/* The bottom half will add @mon to @mon_list */
--
1.8.3.1
> Regards,
> Daniel
>
next prev parent reply other threads:[~2021-06-09 0:21 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-05-17 6:56 A bug of Monitor Chardev ? Longpeng (Mike, Cloud Infrastructure Service Product Dept.)
2021-05-19 16:17 ` Marc-André Lureau
2021-05-19 16:40 ` Daniel P. Berrangé
2021-05-21 7:25 ` Markus Armbruster
2021-05-21 14:43 ` Peter Xu
2021-05-21 16:33 ` Daniel P. Berrangé
2021-05-21 16:56 ` Daniel P. Berrangé
2021-05-21 16:59 ` Marc-André Lureau
2021-05-21 17:07 ` Daniel P. Berrangé
2021-05-21 17:14 ` Peter Xu
2021-05-25 6:53 ` Longpeng (Mike, Cloud Infrastructure Service Product Dept.)
2021-05-21 17:09 ` Peter Xu
2021-05-21 17:15 ` Daniel P. Berrangé
2021-06-08 14:07 ` Markus Armbruster
2021-06-08 15:37 ` Daniel P. Berrangé
2021-06-09 0:20 ` Longpeng (Mike, Cloud Infrastructure Service Product Dept.) [this message]
2021-06-09 10:13 ` Marc-André Lureau
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=6528dfe1-7cc7-c530-a56f-06517a627cda@huawei.com \
--to=longpeng2@huawei.com \
--cc=arei.gonglei@huawei.com \
--cc=armbru@redhat.com \
--cc=berrange@redhat.com \
--cc=chenjiashang@huawei.com \
--cc=marcandre.lureau@redhat.com \
--cc=pbonzini@redhat.com \
--cc=peterx@redhat.com \
--cc=qemu-devel@nongnu.org \
/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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.