The user calling himself "John Doe" reported a deadlock when
attempting to use qemu-storage-daemon to serve both a base file over
NBD, and a qcow2 file with that NBD export as its backing file, from
the same process, even though it worked just fine when there were two
q-s-d processes. The bulk of the NBD server code properly uses
coroutines to make progress in an event-driven manner, but the code
for spawning a new coroutine at the point when listen(2) detects a new
client was hard-coded to use the global GMainContext; in other words,
the callback that triggers nbd_client_new to let the server start the
negotiation sequence with the client requires the main loop to be
making progress. However, the code for bdrv_open of a qcow2 image
with an NBD backing file uses an AIO_WAIT_WHILE nested event loop to
ensure that the entire qcow2 backing chain is either fully loaded or
rejected, without any side effects from the main loop causing unwanted
changes to the disk being loaded (in short, an AioContext represents
the set of actions that are known to be safe while handling block
layer I/O, while excluding any other pending actions in the global
main loop with potentially larger risk of unwanted side effects).
This creates a classic case of deadlock: the server can't progress to
the point of accept(2)ing the client to write to the NBD socket
because the main loop is being starved until the AIO_WAIT_WHILE
completes the bdrv_open, but the AIO_WAIT_WHILE can't progress because
it is blocked on the client coroutine stuck in a read() of the
expected magic number from the server side of the socket.
This patch adds a new API to allow clients to opt in to listening via
an AioContext rather than a GMainContext. This will allow NBD to fix
the deadlock by performing all actions during bdrv_open in the main
loop AioContext.
Technical debt warning: I would have loved to utilize a notify
function with AioContext to guarantee that we don't finalize listener
due to an object_unref if there is any callback still running (the way
GSource does), but wiring up notify functions into AioContext is a
bigger task that will be deferred to a later QEMU release. But for
solving the NBD deadlock, it is sufficient to note that the QMP
commands for enabling and disabling the NBD server are really the only
points where we want to change the listener's callback. Furthermore,
those commands are serviced in the main loop, which is the same
AioContext that is also listening for connections. Since a thread
cannot interrupt itself, we are ensured that at the point where we are
changing the watch, there are no callbacks active. This is NOT as
powerful as the GSource cross-thread safety, but sufficient for the
needs of today.
An upcoming patch will then add a unit test (kept separate to make it
easier to rearrange the series to demonstrate the deadlock without
this patch).
Fixes: https://gitlab.com/qemu-project/qemu/-/issues/3169
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-ID: <20251113011625.878876-26-eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
For ease of review, this patch adds an AioContext pointer to the
QIONetListener struct, the code to trace it, and refactors
listener->io_source to instead be an array of utility structs; but the
aio_context pointer is always NULL until the next patch adds an API to
set it. There should be no semantic change in this patch.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-ID: <20251113011625.878876-25-eblake@redhat.com>
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>
The code had three similar repetitions of an iteration over one or all
of nsiocs to set up a GSource, and likewise for teardown. Since an
upcoming patch wants to tweak whether GSource or AioContext is used,
it's better to consolidate that into one helper function for fewer
places to edit later.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-ID: <20251113011625.878876-22-eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
In qemu-nbd and other NBD server setups where parallel clients are
supported, it is common that the caller will re-register the same
callback function as long as it has not reached its limit on
simultaneous clients. In that case, there is no need to tear down and
reinstall GSource watches in the GMainContext.
In practice, all existing callers currently pass NULL for notify, and
no caller ever changes context across calls (for async uses, either
the caller consistently uses qio_net_listener_set_client_func_full
with the same context, or the caller consistently uses only
qio_net_listener_set_client_func which always uses the global
context); but the time spent checking these two fields in addition to
the more important func and data is still less than the savings of not
churning through extra GSource manipulations when the result will be
unchanged.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-ID: <20251113011625.878876-21-eblake@redhat.com>
Without a mutex, NetListener can run into this data race between a
thread changing the async callback callback function to use when a
client connects, and the thread servicing polling of the listening
sockets:
Thread 1:
qio_net_listener_set_client_func(lstnr, f1, ...);
=> foreach sock: socket
=> object_ref(lstnr)
=> sock_src = qio_channel_socket_add_watch_source(sock, ...., lstnr, object_unref);
Thread 2:
poll()
=> event POLLIN on socket
=> ref(GSourceCallback)
=> if (lstnr->io_func) // while lstnr->io_func is f1
...interrupt..
Thread 1:
qio_net_listener_set_client_func(lstnr, f2, ...);
=> foreach sock: socket
=> g_source_unref(sock_src)
=> foreach sock: socket
=> object_ref(lstnr)
=> sock_src = qio_channel_socket_add_watch_source(sock, ...., lstnr, object_unref);
Thread 2:
=> call lstnr->io_func(lstnr->io_data) // now sees f2
=> return dispatch(sock)
=> unref(GSourceCallback)
=> destroy-notify
=> object_unref
Found by inspection; I did not spend the time trying to add sleeps or
execute under gdb to try and actually trigger the race in practice.
This is a SEGFAULT waiting to happen if f2 can become NULL because
thread 1 deregisters the user's callback while thread 2 is trying to
service the callback. Other messes are also theoretically possible,
such as running callback f1 with an opaque pointer that should only be
passed to f2 (if the client code were to use more than just a binary
choice between a single async function or NULL).
Mitigating factor: if the code that modifies the QIONetListener can
only be reached by the same thread that is executing the polling and
async callbacks, then we are not in a two-thread race documented above
(even though poll can see two clients trying to connect in the same
window of time, any changes made to the listener by the first async
callback will be completed before the thread moves on to the second
client). However, QEMU is complex enough that this is hard to
generically analyze. If QMP commands (like nbd-server-stop) are run
in the main loop and the listener uses the main loop, things should be
okay. But when a client uses an alternative GMainContext, or if
servicing a QMP command hands off to a coroutine to avoid blocking, I
am unable to state with certainty whether a given net listener can be
modified by a thread different from the polling thread running
callbacks.
At any rate, it is worth having the API be robust. To ensure that
modifying a NetListener can be safely done from any thread, add a
mutex that guarantees atomicity to all members of a listener object
related to callbacks. This problem has been present since
QIONetListener was introduced.
Note that this does NOT prevent the case of a second round of the
user's old async callback being invoked with the old opaque data, even
when the user has already tried to change the async callback during
the first async callback; it is only about ensuring that there is no
sharding (the eventual io_func(io_data) call that does get made will
correspond to a particular combination that the user had requested at
some point in time, and not be sharded to a combination that never
existed in practice). In other words, this patch maintains the status
quo that a user's async callback function already needs to be robust
to parallel clients landing in the same window of poll servicing, even
when only one client is desired, if that particular listener can be
amended in a thread other than the one doing the polling.
CC: qemu-stable@nongnu.org
Fixes: 53047392 ("io: introduce a network socket listener API", v2.12.0)
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-ID: <20251113011625.878876-20-eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
[eblake: minor commit message wording improvements]
Signed-off-by: Eric Blake <eblake@redhat.com>
io/net-listener.c has two modes of use: asynchronous (the user calls
qio_net_listener_set_client_func to wake up the callback via the
global GMainContext, or qio_net_listener_set_client_func_full to wake
up the callback via the caller's own alternative GMainContext), and
synchronous (the user calls qio_net_listener_wait_client which creates
its own GMainContext and waits for the first client connection before
returning, with no need for a user's callback). But commit 938c8b79
has a latent logic flaw: when qio_net_listener_wait_client finishes on
its temporary context, it reverts all of the siocs back to the global
GMainContext rather than the potentially non-NULL context they might
have been originally registered with. Similarly, if the user creates
a net-listener, adds initial addresses, registers an async callback
with a non-default context (which ties to all siocs for the initial
addresses), then adds more addresses with qio_net_listener_add, the
siocs for later addresses are blindly placed in the global context,
rather than sharing the context of the earlier ones.
In practice, I don't think this has caused issues. As pointed out by
the original commit, all async callers prior to that commit were
already okay with the NULL default context; and the typical usage
pattern is to first add ALL the addresses the listener will pay
attention to before ever setting the async callback. Likewise, if a
file uses only qio_net_listener_set_client_func instead of
qio_net_listener_set_client_func_full, then it is never using a custom
context, so later assignments of async callbacks will still be to the
same global context as earlier ones. Meanwhile, any callers that want
to do the sync operation to grab the first client are unlikely to
register an async callback; altogether bypassing the question of
whether later assignments of a GSource are being tied to a different
context over time.
I do note that chardev/char-socket.c is the only file that calls both
qio_net_listener_wait_client (sync for a single client in
tcp_chr_accept_server_sync), and qio_net_listener_set_client_func_full
(several places, all with chr->gcontext, but sometimes with a NULL
callback function during teardown). But as far as I can tell, the two
uses are mutually exclusive, based on the is_waitconnect parameter to
qmp_chardev_open_socket_server.
That said, it is more robust to remember when an async callback
function is tied to a non-default context, and have both the sync wait
and any late address additions honor that same context. That way, the
code will be robust even if a later user performs a sync wait for a
specific client in the middle of servicing a longer-lived
QIONetListener that has an async callback for all other clients.
CC: qemu-stable@nongnu.org
Fixes: 938c8b79 ("qio: store gsources for net listeners", v2.12.0)
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-ID: <20251113011625.878876-19-eblake@redhat.com>
When changing the callback registered with QIONetListener, the code
was calling notify on the old opaque data prior to actually removing
the old GSource objects still pointing to that data. Similarly,
during finalize, it called notify before tearing down the various
GSource objects tied to the data.
In practice, a grep of the QEMU code base found that every existing
client of QIONetListener passes in a NULL notifier (the opaque data,
if non-NULL, outlives the NetListener and so does not need cleanup
when the NetListener is torn down), so this patch has no impact. And
even if a caller had passed in a reference-counted object with a
notifier of object_unref but kept its own reference on the data, then
the early notify would merely reduce a refcount from (say) 2 to 1, but
not free the object. However, it is a latent bug waiting to bite any
future caller that passes in data where the notifier actually frees
the object, because the GSource could then trigger a use-after-free if
it loses the race on a last-minute client connection resulting in the
data being passed to one final use of the async callback.
Better is to delay the notify call until after all GSource that have
been given a copy of the opaque data are torn down.
CC: qemu-stable@nongnu.org
Fixes: 530473924d "io: introduce a network socket listener API", v2.12.0
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-ID: <20251113011625.878876-18-eblake@redhat.com>
Upcoming patches will adjust how net_listener watches for new client
connections; adding trace points now makes it easier to debug that the
changes work as intended. For example, adding
--trace='qio_net_listener*' to the qemu-storage-daemon command line
before --nbd-server will track when the server first starts listening
for clients.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-ID: <20251113011625.878876-17-eblake@redhat.com>
- Fabiano's patch to fix snapshot crash by rejecting some caps
- Marco's mapped-ram support on snapshot save/load
- Steve's cpr maintainers entry update on retirement
- Peter's coverity fixes
- Chenyi's tdx fix on hugetlbfs regression
- Peter's doc update on migrate resume flag
- Peter's doc update on HMP set parameter for cpr-exec-command's char** parsing
- Xiaoyao's guest-memfd fix for enabling shmem
- Arun's fix on error_fatal regression for migration errors
- Bin's fix on redundant error free for add block failures
- Markus's cleanup around MigMode sets
- Peter's two patches (out of loadvm threadify) to cleanup qio read peek process
- Thomas's vmstate-static-checker update for possible deprecation of argparse use
- Stefan's fix on windows deadlock by making unassigned MMIOs lockless
-----BEGIN PGP SIGNATURE-----
iIgEABYKADAWIQS5GE3CDMRX2s990ak7X8zN86vXBgUCaQkZPBIccGV0ZXJ4QHJl
ZGhhdC5jb20ACgkQO1/MzfOr1wZhTgEA8eCBMpM7PusNSdzzeIygKnIp2A8I70ca
eIJz3ZM+FiUBAPVDrIZ59EhZA6NPcJb8Ya9OY4lT63F4BxrvN+f+uG4N
=GUBi
-----END PGP SIGNATURE-----
Merge tag 'staging-pull-request' of https://gitlab.com/peterx/qemu into staging
mem + migration pull for 10.2
- Fabiano's patch to fix snapshot crash by rejecting some caps
- Marco's mapped-ram support on snapshot save/load
- Steve's cpr maintainers entry update on retirement
- Peter's coverity fixes
- Chenyi's tdx fix on hugetlbfs regression
- Peter's doc update on migrate resume flag
- Peter's doc update on HMP set parameter for cpr-exec-command's char** parsing
- Xiaoyao's guest-memfd fix for enabling shmem
- Arun's fix on error_fatal regression for migration errors
- Bin's fix on redundant error free for add block failures
- Markus's cleanup around MigMode sets
- Peter's two patches (out of loadvm threadify) to cleanup qio read peek process
- Thomas's vmstate-static-checker update for possible deprecation of argparse use
- Stefan's fix on windows deadlock by making unassigned MMIOs lockless
# -----BEGIN PGP SIGNATURE-----
#
# iIgEABYKADAWIQS5GE3CDMRX2s990ak7X8zN86vXBgUCaQkZPBIccGV0ZXJ4QHJl
# ZGhhdC5jb20ACgkQO1/MzfOr1wZhTgEA8eCBMpM7PusNSdzzeIygKnIp2A8I70ca
# eIJz3ZM+FiUBAPVDrIZ59EhZA6NPcJb8Ya9OY4lT63F4BxrvN+f+uG4N
# =GUBi
# -----END PGP SIGNATURE-----
# gpg: Signature made Mon 03 Nov 2025 10:06:04 PM CET
# gpg: using EDDSA key B9184DC20CC457DACF7DD1A93B5FCCCDF3ABD706
# gpg: issuer "peterx@redhat.com"
# gpg: Good signature from "Peter Xu <xzpeter@gmail.com>" [unknown]
# gpg: aka "Peter Xu <peterx@redhat.com>" [unknown]
# gpg: WARNING: The key's User ID is not certified with a trusted signature!
# gpg: There is no indication that the signature belongs to the owner.
# Primary key fingerprint: B918 4DC2 0CC4 57DA CF7D D1A9 3B5F CCCD F3AB D706
* tag 'staging-pull-request' of https://gitlab.com/peterx/qemu: (36 commits)
migration: Introduce POSTCOPY_DEVICE state
migration: Make postcopy listen thread joinable
migration: Respect exit-on-error when migration fails before resuming
migration: Refactor all incoming cleanup info migration_incoming_destroy()
migration: Introduce postcopy incoming setup and cleanup functions
migration: Move postcopy_ram_listen_thread() to postcopy-ram.c
migration: Do not try to start VM if disk activation fails
migration: Flush migration channel after sending data of CMD_PACKAGED
system/physmem: mark io_mem_unassigned lockless
scripts/vmstate-static-checker: Fix deprecation warnings with latest argparse
migration: vmsd errp handlers: return bool
migration/vmstate: stop reporting error number for new _errp APIs
tmp_emulator: improve and fix use of errp
migration: vmstate_save_state_v(): fix error path
migration: Properly wait on G_IO_IN when peeking messages
io: Add qio_channel_wait_cond() helper
migration: Put Error **errp parameter last
migration: Use bitset of MigMode instead of variable arguments
migration: Use unsigned instead of int for bit set of MigMode
migration: Don't free the reason after calling migrate_add_blocker
...
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Add the helper to wait for QIO channel's IO availability in any
context (coroutine, or non-coroutine). Use it tree-wide for three
occurences.
Cc: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Link: https://lore.kernel.org/r/20251022192612.2737648-2-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
The kernel allocates extra metadata SKBs in case of a zerocopy send,
eventually used for zerocopy's notification mechanism. This metadata
memory is accounted for in the OPTMEM limit. The kernel queues
completion notifications on the socket error queue and this error queue
is freed when userspace reads it.
Usually, in the case of in-order processing, the kernel will batch the
notifications and merge the metadata into a single SKB and free the
rest. As a result, it never exceeds the OPTMEM limit. However, if there
is any out-of-order processing or intermittent zerocopy failures, this
error chain can grow significantly, exhausting the OPTMEM limit. As a
result, all new sendmsg requests fail to allocate any new SKB, leading
to an ENOBUF error. Depending on the amount of data queued before the
flush (i.e., large live migration iterations), even large OPTMEM limits
are prone to failure.
To work around this, if we encounter an ENOBUF error with a zerocopy
sendmsg, flush the error queue and retry once more.
Co-authored-by: Manish Mishra <manish.mishra@nutanix.com>
Signed-off-by: Tejus GK <tejus.gk@nutanix.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
[DB: change TRUE/FALSE to true/false for 'bool' type;
add more #ifdef QEMU_MSG_ZEROCOPY blocks]
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Add a 'blocking' boolean field to QIOChannelSocket to track whether the
underlying socket is in blocking or non-blocking mode.
Signed-off-by: Tejus GK <tejus.gk@nutanix.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
I/O channel read/write functions can operate on any area of
memory, regardless of the content their represent. Do not
restrict to array of char, use the void* type, which is also
the type of the underlying iovec::iov_base field.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
[DB: also adapt test-crypto-tlssession.c func signatures]
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
If the QIOChannelWebsock object is freed while it is waiting to
complete a handshake, a GSource is leaked. This can lead to the
callback firing later on and triggering a use-after-free in the
use of the channel. This was observed in the VNC server with the
following trace from valgrind:
==2523108== Invalid read of size 4
==2523108== at 0x4054A24: vnc_disconnect_start (vnc.c:1296)
==2523108== by 0x4054A24: vnc_client_error (vnc.c:1392)
==2523108== by 0x4068A09: vncws_handshake_done (vnc-ws.c:105)
==2523108== by 0x44863B4: qio_task_complete (task.c:197)
==2523108== by 0x448343D: qio_channel_websock_handshake_io (channel-websock.c:588)
==2523108== by 0x6EDB862: UnknownInlinedFun (gmain.c:3398)
==2523108== by 0x6EDB862: g_main_context_dispatch_unlocked.lto_priv.0 (gmain.c:4249)
==2523108== by 0x6EDBAE4: g_main_context_dispatch (gmain.c:4237)
==2523108== by 0x45EC79F: glib_pollfds_poll (main-loop.c:287)
==2523108== by 0x45EC79F: os_host_main_loop_wait (main-loop.c:310)
==2523108== by 0x45EC79F: main_loop_wait (main-loop.c:589)
==2523108== by 0x423A56D: qemu_main_loop (runstate.c:835)
==2523108== by 0x454F300: qemu_default_main (main.c:37)
==2523108== by 0x73D6574: (below main) (libc_start_call_main.h:58)
==2523108== Address 0x57a6e0dc is 28 bytes inside a block of size 103,608 free'd
==2523108== at 0x5F2FE43: free (vg_replace_malloc.c:989)
==2523108== by 0x6EDC444: g_free (gmem.c:208)
==2523108== by 0x4053F23: vnc_update_client (vnc.c:1153)
==2523108== by 0x4053F23: vnc_refresh (vnc.c:3225)
==2523108== by 0x4042881: dpy_refresh (console.c:880)
==2523108== by 0x4042881: gui_update (console.c:90)
==2523108== by 0x45EFA1B: timerlist_run_timers.part.0 (qemu-timer.c:562)
==2523108== by 0x45EFC8F: timerlist_run_timers (qemu-timer.c:495)
==2523108== by 0x45EFC8F: qemu_clock_run_timers (qemu-timer.c:576)
==2523108== by 0x45EFC8F: qemu_clock_run_all_timers (qemu-timer.c:663)
==2523108== by 0x45EC765: main_loop_wait (main-loop.c:600)
==2523108== by 0x423A56D: qemu_main_loop (runstate.c:835)
==2523108== by 0x454F300: qemu_default_main (main.c:37)
==2523108== by 0x73D6574: (below main) (libc_start_call_main.h:58)
==2523108== Block was alloc'd at
==2523108== at 0x5F343F3: calloc (vg_replace_malloc.c:1675)
==2523108== by 0x6EE2F81: g_malloc0 (gmem.c:133)
==2523108== by 0x4057DA3: vnc_connect (vnc.c:3245)
==2523108== by 0x448591B: qio_net_listener_channel_func (net-listener.c:54)
==2523108== by 0x6EDB862: UnknownInlinedFun (gmain.c:3398)
==2523108== by 0x6EDB862: g_main_context_dispatch_unlocked.lto_priv.0 (gmain.c:4249)
==2523108== by 0x6EDBAE4: g_main_context_dispatch (gmain.c:4237)
==2523108== by 0x45EC79F: glib_pollfds_poll (main-loop.c:287)
==2523108== by 0x45EC79F: os_host_main_loop_wait (main-loop.c:310)
==2523108== by 0x45EC79F: main_loop_wait (main-loop.c:589)
==2523108== by 0x423A56D: qemu_main_loop (runstate.c:835)
==2523108== by 0x454F300: qemu_default_main (main.c:37)
==2523108== by 0x73D6574: (below main) (libc_start_call_main.h:58)
==2523108==
The above can be reproduced by launching QEMU with
$ qemu-system-x86_64 -vnc localhost:0,websocket=5700
and then repeatedly running:
for i in {1..100}; do
(echo -n "GET / HTTP/1.1" && sleep 0.05) | nc -w 1 localhost 5700 &
done
CVE-2025-11234
Reported-by: Grant Millar | Cylo <rid@cylo.io>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The QIOChannelWebsock object releases all its resources in the
finalize callback. This is later than desired, as callers expect
to be able to call qio_channel_close() to fully close a channel
and release resources related to I/O.
The logic in the finalize method is at most a failsafe to handle
cases where a consumer forgets to call qio_channel_close.
This adds equivalent logic to the close method to release the
resources, using g_clear_handle_id/g_clear_pointer to be robust
against repeated invocations. The finalize method is tweaked
so that the GSource is removed before releasing the underlying
channel.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
While code is supposed to call qio_channel_close() before releasing the
last reference on an QIOChannel, this is not guaranteed. QIOChannelFile
and QIOChannelSocket both cleanup resources in their finalizer if the
close operation was missed.
This ensures the TLS channel will do the same failsafe cleanup.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The way that premature termination was handled in TLS connections was
changed to handle an ordering problem during graceful shutdown in the
migration code.
Unfortunately one of the codepaths returned -1 to indicate an error
condition, but failed to set the 'errp' parameter.
This broke error handling in the qio_channel_tls_handshake function,
as the QTask callback would no longer see that an error was raised.
As a result, the client will go on to try to use the already closed
TLS connection, resulting in misleading errors.
This was evidenced in the I/O test 233 which showed changes such as
-qemu-nbd: Certificate does not match the hostname localhost
+qemu-nbd: Failed to read initial magic: Unable to read from socket: Connection reset by peer
Fixes: 7e0c22d585
Acked-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
QCryptoTLSSession allows TLS premature termination in two cases, one of the
case is when the channel shutdown() is invoked on READ side.
It's possible the shutdown() happened after the read thread blocked at
gnutls_record_recv(). In this case, we should allow the premature
termination to happen.
The problem is by the time qcrypto_tls_session_read() was invoked,
tioc->shutdown may not have been set, so this may instead be treated as an
error if there is concurrent shutdown() calls.
To allow the flag to reflect the latest status of tioc->shutdown, move the
check upper into the QIOChannel level, so as to read the flag only after
QEMU gets an GNUTLS_E_PREMATURE_TERMINATION.
When at it, introduce qio_channel_tls_allow_premature_termination() helper
to make the condition checks easier to read. When doing so, change the
qatomic_load_acquire() to qatomic_read(): here we don't need any ordering
of memory accesses, but reading a flag. qatomic_read() would suffice
because it guarantees fetching from memory. Nothing else we should need to
order on memory access.
This patch will fix a qemu qtest warning when running the preempt tls test,
reporting premature termination:
QTEST_QEMU_BINARY=./qemu-system-x86_64 ./tests/qtest/migration-test --full -r /x86_64/migration/postcopy/preempt/tls/psk
...
qemu-kvm: Cannot read from TLS channel: The TLS connection was non-properly terminated.
...
In this specific case, the error was set by postcopy_preempt_thread, which
normally will be concurrently shutdown()ed by the main thread.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juraj Marcin <jmarcin@redhat.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20250918203937.200833-2-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
We added @error_warn some two years ago in commit 3ffef1a55c (error:
add global &error_warn destination). It has multiple issues:
* error.h's big comment was not updated for it.
* Function contracts were not updated for it.
* ERRP_GUARD() is unaware of @error_warn, and fails to mask it from
error_prepend() and such. These crash on @error_warn, as pointed
out by Akihiko Odaki.
All fixable. However, after more than two years, we had just of 15
uses, of which the last few patches removed seven as unclean or
otherwise undesirable, adding back five elsewhere. I didn't look
closely enough at the remaining seven to decide whether they are
desirable or not.
I don't think this feature earns its keep. Drop it.
Thanks-to: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-ID: <20250923091000.3180122-14-armbru@redhat.com>
Reviewed-by: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp>
qemu_socket_select() and its wrapper qemu_socket_unselect() treat a
null @errp as &error_warn. This is wildly inappropriate. A caller
passing null @errp specifies that errors are to be ignored. If
warnings are wanted, the caller must pass &error_warn.
Change callers to do that, and drop the inappropriate treatment of
null @errp.
This assumes that warnings are wanted. I'm not familiar with the
calling code, so I can't say whether it will work when the socket is
invalid, or WSAEventSelect() fails. If it doesn't, then this should
be an error instead of a warning. Invalid socket might even be a
programming error.
These warnings were introduced in commit f5fd677ae7 (win32/socket:
introduce qemu_socket_select() helper). I considered reverting to
silence, but Daniel Berrangé asked for the warnings to be preserved.
Cc: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20250923091000.3180122-9-armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp>
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>
We want to switch from qemu_socket_set_block() to newer
qemu_set_blocking(), which provides return status of operation,
to handle errors.
Still, we want to keep qio_channel_socket_readv() interface clean,
as currently it allocate @fds only on success.
So, in case of error, we should close all incoming fds and keep
user's @fds untouched or zero.
Let's make separate functions qio_channel_handle_fds() and
qio_channel_cleanup_fds(), to achieve what we want.
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>
Use common qemu_set_blocking() instead.
Note that pre-patch the behavior of Win32 and Linux realizations
are inconsistent: we ignore failure for Win32, and assert success
for Linux.
How do we convert the callers?
1. Most of callers call qemu_socket_set_nonblock() on a
freshly created socket fd, in conditions when we may simply
report an error. Seems correct switching to error handling
both for Windows (pre-patch error is ignored) and Linux
(pre-patch we assert success). Anyway, we normally don't
expect errors in these cases.
Still in tests let's use &error_abort for simplicity.
What are exclusions?
2. hw/virtio/vhost-user.c - we are inside #ifdef CONFIG_LINUX,
so no damage in switching to error handling from assertion.
3. io/channel-socket.c: here we convert both old calls to
qemu_socket_set_nonblock() and qemu_socket_set_block() to
one new call. Pre-patch we assert success for Linux in
qemu_socket_set_nonblock(), and ignore all other errors here.
So, for Windows switch is a bit dangerous: we may get
new errors or crashes(when error_abort is passed) in
cases where we have silently ignored the error before
(was it correct in all such cases, if they were?) Still,
there is no other way to stricter API than take
this risk.
4. util/vhost-user-server - compiled only for Linux (see
util/meson.build), so we are safe, switching from assertion to
&error_abort.
Note: In qga/channel-posix.c we use g_warning(), where g_printerr()
would actually be a better choice. Still let's for now follow
common style of qga, where g_warning() is commonly used to print
such messages, and no call to g_printerr(). Converting everything
to use g_printerr() should better be another series.
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>
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>
In migration we want to pass fd "as is", not changing its
blocking status.
The only current user of these fds is CPR state (through VMSTATE_FD),
which of-course doesn't want to modify fds on target when source is
still running and use these fds.
Suggested-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
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>
Add a QIO_CHANNEL_FEATURE_CONCURRENT_IO feature flag.
If this is set on a QIOChannelTLS session object, the TLS
session will be marked as requiring thread safety, which
will activate the workaround for GNUTLS bug 1717 if needed.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/qemu-devel/20250718150514.2635338-3-berrange@redhat.com
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Testing reading and writing from qemu-nbd using a unix domain socket
shows that the platform default send buffer size is too low, leading to
poor performance and hight cpu usage.
Add a helper for setting socket send buffer size to be used in NBD code.
It can also be used in other contexts.
We don't need a helper for receive buffer size since it is not used with
unix domain sockets. This is documented for Linux, and not documented
for macOS.
Failing to set the socket buffer size is not a fatal error, but the
caller may want to warn about the failure.
Signed-off-by: Nir Soffer <nirsof@gmail.com>
Message-ID: <20250517201154.88456-2-nirsof@gmail.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Commit aec21d3175 (qapi: Add InetSocketAddress member keep-alive)
introduces the keep-alive flag, but this flag is not copied together
with other options in qio_dns_resolver_lookup_sync_inet().
This patch fixes this issue and also prevents future ones by copying the
entire structure first and only then overriding a few attributes that
need to be different.
Fixes: aec21d3175 (qapi: Add InetSocketAddress member keep-alive)
Signed-off-by: Juraj Marcin <jmarcin@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Mechanical change using gsed, then style manually adapted
to pass checkpatch.pl script.
Suggested-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20250424194905.82506-4-philmd@linaro.org>
Add a read flag that can inform a channel that it's ok to receive an
EOF at any moment. Channels that have some form of strict EOF
tracking, such as TLS session termination, may choose to ignore EOF
errors with the use of this flag.
This is being added for compatibility with older migration streams
that do not include a TLS termination step.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
We want to pass flags into qio_channel_tls_readv() but
qio_channel_readv_full_all_eof() doesn't take a flags argument.
No functional change.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Acked-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
The correct way of calling qcrypto_tls_session_handshake() requires
calling qcrypto_tls_session_get_handshake_status() right after it so
there's no reason to have a separate method.
Refactor qcrypto_tls_session_handshake() to inform the status in its
own return value and alter the callers accordingly.
No functional change.
Suggested-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Acked-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Add a task dispatcher for gnutls_bye similar to the
qio_channel_tls_handshake_task(). The gnutls_bye() call might be
interrupted and so it needs to be rescheduled.
The migration code will make use of this to help the migration
destination identify a premature EOF. Once the session termination is
in place, any EOF that happens before the source issued gnutls_bye()
will be considered an error.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Acked-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
QAPI's 'prefix' feature can make the connection between enumeration
type and its constants less than obvious. It's best used with
restraint.
QCryptoHashAlgorithm has a 'prefix' that overrides the generated
enumeration constants' prefix to QCRYPTO_HASH_ALG.
We could simply drop 'prefix', but then the prefix becomes
QCRYPTO_HASH_ALGORITHM, which is rather long.
We could additionally rename the type to QCryptoHashAlg, but I think
the abbreviation "alg" is less than clear.
Rename the type to QCryptoHashAlgo instead. The prefix becomes to
QCRYPTO_HASH_ALGO.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Acked-by: Daniel P. Berrangé <berrange@redhat.com>
Message-ID: <20240904111836.3273842-12-armbru@redhat.com>
[Conflicts with merge commit 7bbadc60b5 resolved]
GNUTLS doesn't know how to perform I/O on anything other than plain
FDs, so the TLS session provides it with some I/O callbacks. The
GNUTLS API design requires these callbacks to return a unix errno
value, which means we're currently loosing the useful QEMU "Error"
object.
This changes the I/O callbacks in QEMU to stash the "Error" object
in the QCryptoTLSSession class, and fetch it when seeing an I/O
error returned from GNUTLS, thus preserving useful error messages.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The current TLS session I/O APIs just return a synthetic errno
value on error, which has been translated from a gnutls error
value. This looses a large amount of valuable information that
distinguishes different scenarios.
Pushing population of the "Error *errp" object into the TLS
session I/O APIs gives more detailed error information.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The function qio_channel_get_peercred() returns a pointer to the
credentials of the peer process connected to this socket.
This credentials structure is defined in <sys/socket.h> as follows:
struct ucred {
pid_t pid; /* Process ID of the sending process */
uid_t uid; /* User ID of the sending process */
gid_t gid; /* Group ID of the sending process */
};
The use of this function is possible only for connected AF_UNIX stream
sockets and for AF_UNIX stream and datagram socket pairs.
On platform other than Linux, the function return 0.
Signed-off-by: Anthony Harivel <aharivel@redhat.com>
Link: https://lore.kernel.org/r/20240522153453.1230389-2-aharivel@redhat.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This reverts commit 3eacf70bb5.
It was only needed because of duplicate objects caused by
declare_dependency(link_whole: ...), and can be dropped now
that meson.build specifies objects and dependencies separately
for the internal dependencies.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Message-ID: <20240524-objects-v1-2-07cbbe96166b@daynix.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We want to make use of the Error object to report fdset errors from
qemu_open_internal() and passing the error pointer to qemu_open_old()
would require changing all callers. Move the file channel to the new
API instead.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Since qemu 8.2, the combination of NBD + TLS + iothread crashes on an
assertion failure:
qemu-kvm: ../io/channel.c:534: void qio_channel_restart_read(void *): Assertion `qemu_get_current_aio_context() == qemu_coroutine_get_aio_context(co)' failed.
It turns out that when we removed AioContext locking, we did so by
having NBD tell its qio channels that it wanted to opt in to
qio_channel_set_follow_coroutine_ctx(); but while we opted in on the
main channel, we did not opt in on the TLS wrapper channel.
qemu-iotests has coverage of NBD+iothread and NBD+TLS, but apparently
no coverage of NBD+TLS+iothread, or we would have noticed this
regression sooner. (I'll add that in the next patch)
But while we could manually opt in to the TLS channel in nbd/server.c
(a one-line change), it is more generic if all qio channels that wrap
other channels inherit the follow status, in the same way that they
inherit feature bits.
CC: Stefan Hajnoczi <stefanha@redhat.com>
CC: Daniel P. Berrangé <berrange@redhat.com>
CC: qemu-stable@nongnu.org
Fixes: https://issues.redhat.com/browse/RHEL-34786
Fixes: 06e0f098 ("io: follow coroutine AioContext in qio_channel_yield()", v8.2.0)
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-ID: <20240518025246.791593-5-eblake@redhat.com>
Add a new helper function for creating a QIOChannelFile channel with a
duplicated file descriptor. This saves the calling code from having to
do error checking on the dup() call.
Suggested-by: "Daniel P. Berrangé" <berrange@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: "Daniel P. Berrangé" <berrange@redhat.com>
Link: https://lore.kernel.org/r/20240311233335.17299-2-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>
Commit bc38feddeb ("io: fsync before closing a file channel") added a
fsync/fdatasync at the closing point of the QIOChannelFile to ensure
integrity of the migration stream in case of QEMU crash.
The decision to do the sync at qio_channel_close() was not the best
since that function runs in the main thread and the fsync can cause
QEMU to hang for several minutes, depending on the migration size and
disk speed.
To fix the hang, remove the fsync from qio_channel_file_close().
At this moment, the migration code is the only user of the fsync and
we're taking the tradeoff of not having a sync at all, leaving the
responsibility to the upper layers.
Fixes: bc38feddeb ("io: fsync before closing a file channel")
Reviewed-by: "Daniel P. Berrangé" <berrange@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240305195629.9922-1-farosas@suse.de
Link: https://lore.kernel.org/r/20240305174332.2553-1-farosas@suse.de
[peterx: add more comment to the qio_channel_close()]
Signed-off-by: Peter Xu <peterx@redhat.com>
Make sure the data is flushed to disk before closing file
channels. This is to ensure data is on disk and not lost in the event
of a host crash.
This is currently being implemented to affect the migration code when
migrating to a file, but all QIOChannelFile users should benefit from
the change.
Reviewed-by: "Daniel P. Berrangé" <berrange@redhat.com>
Acked-by: "Daniel P. Berrangé" <berrange@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240229153017.2221-6-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>
The upcoming 'mapped-ram' feature will require qemu to write data to
(and restore from) specific offsets of the migration file.
Add a minimal implementation of pwritev/preadv and expose them via the
io_pwritev and io_preadv interfaces.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: "Daniel P. Berrangé" <berrange@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240229153017.2221-5-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>
Introduce basic pwritev/preadv support in the generic channel layer.
Specific implementation will follow for the file channel as this is
required in order to support migration streams with fixed location of
each ram page.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: "Daniel P. Berrangé" <berrange@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240229153017.2221-4-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>
Add a generic QIOChannel feature SEEKABLE which would be used by the
qemu_file* apis. For the time being this will be only implemented for
file channels.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: "Daniel P. Berrangé" <berrange@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240229153017.2221-3-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>
All callers of qio_net_listener_set_name() already add some sort of
"listen" or "listener" suffix.
For intance, we currently have "migration-socket-listener-listen" and
"vnc-listen-listen" as ioc names.
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
For SO_EE_ORIGIN_ZEROCOPY the 32-bit notification range is encoded
as [ee_info, ee_data] inclusively, so ee_info should be less or
equal to ee_data.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Maksim Davydov <davydov-max@yandex-team.ru>
Message-id: 20231017125941.810461-7-vsementsov@yandex-team.ru
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
