When starting a dummy QEMU process with virsh version, monitor_init_qmp()
enables IOThread monitoring of the QMP fd by default. However, a race
condition exists during the initialization phase: the IOThread only removes
the main thread's fd watch when it reaches qio_net_listener_set_client_func_full(),
which may be delayed under high system load.
This creates a window between monitor_qmp_setup_handlers_bh() and
qio_net_listener_set_client_func_full() where both the main thread and
IOThread are simultaneously monitoring the same fd and processing events.
This race can cause either the main thread or the IOThread to hang and
become unresponsive.
Fix this by proactively cleaning up the listener's IO sources in
monitor_init_qmp() before the IOThread initializes QMP monitoring,
ensuring exclusive fd ownership and eliminating the race condition.
Signed-off-by: Jie Song <songjie_yewu@cmss.chinamobile.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-ID: <20251125140706.114197-1-mail@jiesong.me>
An upcoming patch needs to pass more than just sioc as the opaque
pointer to an AioContext; but since our AioContext code in general
(and its QIO Channel wrapper code) lacks a notify callback present
with GSource, we do not have the trivial option of just g_malloc'ing a
small struct to hold all that data coupled with a notify of g_free.
Instead, the data pointer must outlive the registered handler; in
fact, having the data pointer have the same lifetime as QIONetListener
is adequate.
But the cleanest way to stick such a helper struct in QIONetListener
will be to rearrange internal struct members. And that in turn means
that all existing code that currently directly accesses
listener->nsioc and listener->sioc[] should instead go through
accessor functions, to be immune to the upcoming struct layout
changes. So this patch adds accessor methods qio_net_listener_nsioc()
and qio_net_listener_sioc(), and puts them to use.
While at it, notice that the pattern of grabbing an sioc from the
listener only to turn around can call
qio_channel_socket_get_local_address is common enough to also warrant
the helper of qio_net_listener_get_local_address, and fix a copy-paste
error in the corresponding documentation.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-ID: <20251113011625.878876-24-eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Directly accessing the fd member of a QIOChannelSocket is an
undesirable leaky abstraction. What's more, grabbing that fd merely
to force an eventual call to getsockname() can be wasteful, since the
channel is often able to return its cached local name.
Reported-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-ID: <20251113011625.878876-23-eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
It was deprecated in 9.2, time to remove.
Note, that (which become obvious with this commit) we forget to do some
checks for reconnect-ms options, for example, it was silently ignored
for listening server, instead of error-out. The commit fixes this, as
now we use reconnect_ms everywhere.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Daniil Tatianin <d-tatianin@yandex-team.ru>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Acked-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Michael Tokarev <mjt@tls.msk.ru>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Instead of open-coded g_unix_set_fd_nonblocking() calls, use
QEMU wrapper qemu_set_blocking().
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
[DB: fix missing closing ) in tap-bsd.c, remove now unused GError var]
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Currently, we just always pass NULL as errp argument. That doesn't
look good.
Some realizations of interface may actually report errors.
Channel-socket realization actually either ignore or crash on
errors, but we are going to straighten it out to always reporting
an errp in further commits.
So, convert all callers to either handle the error (where environment
allows) or explicitly use &error_abort.
Take also a chance to change the return value to more convenient
bool (keeping also in mind, that underlying realizations may
return -1 on failure, not -errno).
Suggested-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
[DB: fix return type mismatch in TLS/websocket channel
impls for qio_channel_set_blocking]
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Add comment, to stress that the order of operation (first drop old fds,
second check read status) is intended.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
qio_channel_readv_full() guarantees BLOCKING and CLOEXEC states for
incoming descriptors, no reason to call extra ioctls.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Compiler warning:
../chardev/baum.c:657:25: warning: comparison between pointer and integer
Use brlapi_fileDescriptor instead of int for brlapi_fd and
BRLAPI_INVALID_FILE_DESCRIPTOR instead of -1.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Reviewed-by: Michael Tokarev <mjt@tls.msk.ru>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
The 'tcp_chr_read' method has a 4k byte array used for copying
data between the socket and device. Skip the automatic zero-init
of this array to eliminate the performance overhead in the I/O
hot path.
The 'buf' array will be fully initialized when reading data off
the network socket.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20250610123709.835102-7-berrange@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The 'pty_chr_read' method has a 4k byte array used for copying
data between the PTY and device. Skip the automatic zero-init
of this array to eliminate the performance overhead in the I/O
hot path.
The 'buf' array will be fully initialized when reading data off
the PTY.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20250610123709.835102-6-berrange@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The 'fd_chr_read' method has a 4k byte array used for copying
data between the socket and device. Skip the automatic zero-init
of this array to eliminate the performance overhead in the I/O
hot path.
The 'buf' array will be fully initialized when reading data off
the network socket.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20250610123709.835102-5-berrange@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
If the chardev is client, the socket file path in localAddr may be NULL.
This is because the socket path comes from getsockname(), according
to man page, getsockname() returns the current address bound by the
socket sockfd. If the chardev is client, it's socket is unbound sockfd.
Therefore, when computing the client chardev socket file path, using
remoteAddr is more appropriate.
Signed-off-by: Haoqian He <haoqian.he@smartx.com>
Acked-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-ID: <20250225104526.2924175-1-haoqian.he@smartx.com>
The general expectation is that header files should follow the same
file/path naming scheme as the corresponding source file. There are
various historical exceptions to this practice in QEMU, with one of
the most notable being the include/qapi/qmp/ directory. Most of the
headers there correspond to source files in qobject/.
This patch corrects most of that inconsistency by creating
include/qobject/ and moving the headers for qobject/ there.
This also fixes MAINTAINERS for include/qapi/qmp/dispatch.h:
scripts/get_maintainer.pl now reports "QAPI" instead of "No
maintainers found".
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
Acked-by: Halil Pasic <pasic@linux.ibm.com> #s390x
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20241118151235.2665921-2-armbru@redhat.com>
[Rebased]
This patch implements a new chardev backend `hub` device, which
aggregates input from multiple backend devices and forwards it to a
single frontend device. Additionally, `hub` device takes the output
from the frontend device and sends it back to all the connected
backend devices. This allows for seamless interaction between
different backend devices and a single frontend interface.
The idea of the change is trivial: keep list of backend devices
(up to 4), init them on demand and forward data buffer back and
forth.
The following is QEMU command line example:
-chardev pty,path=/tmp/pty,id=pty0 \
-chardev vc,id=vc0 \
-chardev hub,id=hub0,chardevs.0=pty0,chardevs.1=vc0 \
-device virtconsole,chardev=hub0 \
-vnc 0.0.0.0:0
Which creates 2 backend devices: text virtual console (`vc0`) and a
pseudo TTY (`pty0`) connected to the single virtio hvc console with
the backend aggregator (`hub0`) help. `vc0` renders text to an image,
which can be shared over the VNC protocol. `pty0` is a pseudo TTY
backend which provides biderectional communication to the virtio hvc
console.
'chardevs.N' list syntax is used for the sake of compatibility with
the representation of JSON lists in 'key=val' pairs format of the
util/keyval.c, despite the fact that modern QAPI way of parsing,
namely qobject_input_visitor_new_str(), is not used. Choice of keeping
QAPI list syntax may help to smoothly switch to modern parsing in the
future.
Signed-off-by: Roman Penyaev <r.peniaev@gmail.com>
Reviewed-by: "Marc-André Lureau" <marcandre.lureau@redhat.com>
Cc: qemu-devel@nongnu.org
Message-ID: <20250123085327.965501-3-r.peniaev@gmail.com>
Headers in include/sysemu/ are not only related to system
*emulation*, they are also used by virtualization. Rename
as system/ which is clearer.
Files renamed manually then mechanical change using sed tool.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Lei Yang <leiyang@redhat.com>
Message-Id: <20241203172445.28576-1-philmd@linaro.org>
qemu_chardev_set_replay() was being called in chardev creation to
set up replay parameters even if the chardev is NULL.
A segfault can be reproduced by specifying '-serial chardev:bad' with
an rr=record mode.
Fix this with a NULL pointer check.
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Resolves: Coverity CID 1559470
Fixes: 4c193bb129 ("chardev: set record/replay on the base device of a muxed device")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-ID: <20240828043337.14587-2-npiggin@gmail.com>
Frontends can be attached and detached during run-time (although detach
is not implemented, but will follow). Counter variable of muxes is not
enough for proper attach/detach management, so this patch implements
bitset: if bit is set for the `mux_bitset` variable, then frontend
device can be found in the `backend` array (yes, huge confusion with
backend and frontends names).
Signed-off-by: Roman Penyaev <r.peniaev@gmail.com>
Cc: "Marc-André Lureau" <marcandre.lureau@redhat.com>
Cc: qemu-devel@nongnu.org
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-ID: <20241014152408.427700-7-r.peniaev@gmail.com>
Add path option to the pty char backend which will create a symbolic
link to the given path that points to the allocated PTY.
This avoids having to make QMP or HMP monitor queries to find out what
the new PTY device path is.
Based on patch from Paulo Neves:
https://patchew.org/QEMU/1548509635-15776-1-git-send-email-ptsneves@gmail.com/
Tested with the following invocations that the link is created and
removed when qemu stops:
qemu-system-x86_64 -nodefaults -mon chardev=compat_monitor \
-chardev pty,path=test,id=compat_monitor0
qemu-system-x86_64 -nodefaults -monitor pty:test
# check QMP invocation with path set
qemu-system-x86_64 -nodefaults -qmp tcp:localhost:4444,server=on,wait=off
nc localhost 4444
> {"execute": "qmp_capabilities"}
> {"execute": "chardev-add", "arguments": {"id": "bar", "backend": {
"type": "pty", "data": {"path": "test" }}}}
# check QMP invocation with path not set
qemu-system-x86_64 -nodefaults -qmp tcp:localhost:4444,server=on,wait=off
nc localhost 4444
> {"execute": "qmp_capabilities"}
> {"execute": "chardev-add", "arguments": {"id": "bar", "backend": {
"type": "pty", "data": {}}}}
Also tested that when a link path is not passed invocations still work, e.g.:
qemu-system-x86_64 -monitor pty
Co-authored-by: Paulo Neves <ptsneves@gmail.com>
Signed-off-by: Paulo Neves <ptsneves@gmail.com>
[OP: rebase and address original patch review comments]
Signed-off-by: Octavian Purdila <tavip@google.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-ID: <20240806010735.2450555-1-tavip@google.com>
The 'reconnect' option only allows to specify the time in seconds,
which is way too long for certain workflows.
We have a lightweight disk backend server, which takes about 20ms to
live update, but due to this limitation in QEMU, previously the guest
disk controller would hang for one second because it would take this
long for QEMU to reinitialize the socket connection.
Introduce a new option called 'reconnect-ms', which is the same as
'reconnect', except the value is treated as milliseconds. These are
mutually exclusive and specifying both results in an error.
'reconnect' is also deprecated by this commit to make it possible to
remove it in the future as to not keep two options that control the
same thing.
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Acked-by: Peter Krempa <pkrempa@redhat.com>
Signed-off-by: Daniil Tatianin <d-tatianin@yandex-team.ru>
Acked-by: Markus Armbruster <armbru@redhat.com>
Tested-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20240913094604.269135-1-d-tatianin@yandex-team.ru>
chardev events to a muxed device don't get recorded because e.g.,
qemu_chr_be_write() checks whether the base device has the record flag
set.
This can be seen when replaying a trace that has characters typed into
the console, an examination of the log shows they are not recorded.
Setting QEMU_CHAR_FEATURE_REPLAY on the base chardev fixes the problem.
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20240813050638.446172-8-npiggin@gmail.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240813202329.1237572-16-alex.bennee@linaro.org>
This adds trace points to every error scenario in the chardev socket
backend that can lead to termination of the connection.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Since fifo8_pop_buf() return a const buffer (which points
directly into the FIFO backing store). Rename it using the
'bufptr' suffix to better reflect that it is a pointer to
the internal buffer that is being returned. This will help
differentiate with methods *copying* the FIFO data.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Message-Id: <20240722160745.67904-6-philmd@linaro.org>
If I use `-serial stdio` on Windows, after QEMU exits, the terminal
could not handle arrow keys and tab any more. Because stdio backend
on Windows sets console mode to virtual terminal input when starts,
but does not restore the old mode when finalize.
This small patch saves the old console mode and set it back.
Signed-off-by: Ziming Song <s.ziming@hotmail.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-ID: <ME3P282MB25488BE7C39BF0C35CD0DA5D8CA82@ME3P282MB2548.AUSP282.PROD.OUTLOOK.COM>
This reverts commit 2b316774f6.
After 038b421788 ("Revert "chardev: use a child source for qio input
source"") we've been observing the "iwp->src == NULL" assertion
triggering periodically during the initial capabilities querying by
libvirtd. One of possible backtraces:
Thread 1 (Thread 0x7f16cd4f0700 (LWP 43858)):
0 __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:50
1 0x00007f16c6c21e65 in __GI_abort () at abort.c:79
2 0x00007f16c6c21d39 in __assert_fail_base at assert.c:92
3 0x00007f16c6c46e86 in __GI___assert_fail (assertion=assertion@entry=0x562e9bcdaadd "iwp->src == NULL", file=file@entry=0x562e9bcdaac8 "../chardev/char-io.c", line=line@entry=99, function=function@entry=0x562e9bcdab10 <__PRETTY_FUNCTION__.20549> "io_watch_poll_finalize") at assert.c:101
4 0x0000562e9ba20c2c in io_watch_poll_finalize (source=<optimized out>) at ../chardev/char-io.c:99
5 io_watch_poll_finalize (source=<optimized out>) at ../chardev/char-io.c:88
6 0x00007f16c904aae0 in g_source_unref_internal () from /lib64/libglib-2.0.so.0
7 0x00007f16c904baf9 in g_source_destroy_internal () from /lib64/libglib-2.0.so.0
8 0x0000562e9ba20db0 in io_remove_watch_poll (source=0x562e9d6720b0) at ../chardev/char-io.c:147
9 remove_fd_in_watch (chr=chr@entry=0x562e9d5f3800) at ../chardev/char-io.c:153
10 0x0000562e9ba23ffb in update_ioc_handlers (s=0x562e9d5f3800) at ../chardev/char-socket.c:592
11 0x0000562e9ba2072f in qemu_chr_fe_set_handlers_full at ../chardev/char-fe.c:279
12 0x0000562e9ba207a9 in qemu_chr_fe_set_handlers at ../chardev/char-fe.c:304
13 0x0000562e9ba2ca75 in monitor_qmp_setup_handlers_bh (opaque=0x562e9d4c2c60) at ../monitor/qmp.c:509
14 0x0000562e9bb6222e in aio_bh_poll (ctx=ctx@entry=0x562e9d4c2f20) at ../util/async.c:216
15 0x0000562e9bb4de0a in aio_poll (ctx=0x562e9d4c2f20, blocking=blocking@entry=true) at ../util/aio-posix.c:722
16 0x0000562e9b99dfaa in iothread_run (opaque=0x562e9d4c26f0) at ../iothread.c:63
17 0x0000562e9bb505a4 in qemu_thread_start (args=0x562e9d4c7ea0) at ../util/qemu-thread-posix.c:543
18 0x00007f16c70081ca in start_thread (arg=<optimized out>) at pthread_create.c:479
19 0x00007f16c6c398d3 in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
io_remove_watch_poll(), which makes sure that iwp->src is NULL, calls
g_source_destroy() which finds that iwp->src is not NULL in the finalize
callback. This can only happen if another thread has managed to trigger
io_watch_poll_prepare() callback in the meantime.
Move iwp->src destruction back to the finalize callback to prevent the
described race, and also remove the stale comment. The deadlock glib bug
was fixed back in 2010 by b35820285668 ("gmain: move finalization of
GSource outside of context lock").
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Sergey Dyasli <sergey.dyasli@nutanix.com>
Link: https://lore.kernel.org/r/20240712092659.216206-1-sergey.dyasli@nutanix.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This reverts commit a7077b8e35,
and add comments to explain why child sources cannot be used.
When a GSource is added as a child of another GSource, if its
'prepare' function indicates readiness, then the parent's
'prepare' function will never be run. The io_watch_poll_prepare
absolutely *must* be run on every iteration of the main loop,
to ensure that the chardev backend doesn't feed data to the
frontend that it is unable to consume.
At the time a7077b8e35 was made,
all the child GSource impls were relying on poll'ing an FD,
so their 'prepare' functions would never indicate readiness
ahead of poll() being invoked. So the buggy behaviour was
not noticed and lay dormant.
Relatively recently the QIOChannelTLS impl introduced a
level 2 child GSource, which checks with GNUTLS whether it
has cached any data that was decoded but not yet consumed:
commit ffda5db65a
Author: Antoine Damhet <antoine.damhet@shadow.tech>
Date: Tue Nov 15 15:23:29 2022 +0100
io/channel-tls: fix handling of bigger read buffers
Since the TLS backend can read more data from the underlying QIOChannel
we introduce a minimal child GSource to notify if we still have more
data available to be read.
Signed-off-by: Antoine Damhet <antoine.damhet@shadow.tech>
Signed-off-by: Charles Frey <charles.frey@shadow.tech>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
With this, it is now quite common for the 'prepare' function
on a QIOChannelTLS GSource to indicate immediate readiness,
bypassing the parent GSource 'prepare' function. IOW, the
critical 'io_watch_poll_prepare' is being skipped on some
iterations of the main loop. As a result chardev frontend
asserts are now being triggered as they are fed data they
are not ready to consume.
A reproducer is as follows:
* In terminal 1 run a GNUTLS *echo* server
$ gnutls-serv --echo \
--x509cafile ca-cert.pem \
--x509keyfile server-key.pem \
--x509certfile server-cert.pem \
-p 9000
* In terminal 2 run a QEMU guest
$ qemu-system-s390x \
-nodefaults \
-display none \
-object tls-creds-x509,id=tls0,dir=$PWD,endpoint=client \
-chardev socket,id=con0,host=localhost,port=9000,tls-creds=tls0 \
-device sclpconsole,chardev=con0 \
-hda Fedora-Cloud-Base-39-1.5.s390x.qcow2
After the previous patch revert, but before this patch revert,
this scenario will crash:
qemu-system-s390x: ../hw/char/sclpconsole.c:73: chr_read: Assertion
`size <= SIZE_BUFFER_VT220 - scon->iov_data_len' failed.
This assert indicates that 'tcp_chr_read' was called without
'tcp_chr_read_poll' having first been checked for ability to
receive more data
QEMU's use of a 'prepare' function to create/delete another
GSource is rather a hack and not normally the kind of thing that
is expected to be done by a GSource. There is no mechanism to
force GLib to always run the 'prepare' function of a parent
GSource. The best option is to simply not use the child source
concept, and go back to the functional approach previously
relied on.
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Tested-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
This commit results in unexpected termination of the TLS connection.
When 'fd_can_read' returns 0, the code goes on to pass a zero length
buffer to qio_channel_read. The TLS impl calls into gnutls_recv()
with this zero length buffer, at which point GNUTLS returns an error
GNUTLS_E_INVALID_REQUEST. This is treated as fatal by QEMU's TLS code
resulting in the connection being torn down by the chardev.
Simply skipping the qio_channel_read when the buffer length is zero
is also not satisfactory, as it results in a high CPU burn busy loop
massively slowing QEMU's functionality.
The proper solution is to avoid tcp_chr_read being called at all
unless the frontend is able to accept more data. This will be done
in a followup commit.
This reverts commit 462945cd22
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The socket chardev often has 2 GSource object registered against the
same FD. One is registered all the time and is just intended to handle
POLLHUP events, while the other gets registered & unregistered on the
fly as the frontend is ready to receive more data or not.
It is very common for poll() to signal a POLLHUP event at the same time
as there is pending incoming data from the disconnected client. It is
therefore essential to process incoming data prior to processing HUP.
The problem with having 2 GSource on the same FD is that there is no
guaranteed ordering of execution between them, so the chardev code may
process HUP first and thus discard data.
This failure scenario is non-deterministic but can be seen fairly
reliably by reverting a7077b8e35, and
then running 'tests/unit/test-char', which will sometimes fail with
missing data.
Ideally QEMU would only have 1 GSource, but that's a complex code
refactoring job. The next best solution is to try to ensure ordering
between the 2 GSource objects. This can be achieved by lowering the
priority of the HUP GSource, so that it is never dispatched if the
main GSource is also ready to dispatch. Counter-intuitively, lowering
the priority of a GSource is done by raising its priority number.
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Both
$ qemu-system-x86_64 -chardev null,id=chr0,mux=on -mon chardev=chr0 -mon chardev=chr0 -mon chardev=chr0 -mon chardev=chr0 -mon chardev=chr0
and
$ qemu-system-x86_64 -chardev null,id=chr0 -mon chardev=chr0 -mon chardev=chr0
fail with
qemu-system-x86_64: -mon chardev=chr0: Device 'chr0' is in use
Improve to
qemu-system-x86_64: -mon chardev=chr0: too many uses of multiplexed chardev 'chr0' (maximum is 4)
and
qemu-system-x86_64: -mon chardev=chr0: chardev 'chr0' is already in use
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Michael Tokarev <mjt@tls.msk.ru>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
