Commit 143a1df24e changed red_worker_main
from epoll to poll. But epoll has edge triggered semantics (when requested
and we requested them), where as poll is always level triggered. And
red_worker was relying on the edge triggered semantics, as it was always
polling for POLLOUT, which, when edge triggered, would only cause poll
to register an event after we had blocked on a write. But after the
switch to regular poll, with its level triggered semantics, the POLLOUT
condition would almost always be true, causing red_worker_main to not
block on the poll and burn CPU as fast as it can as soon as a client was
connected.
Luckily we already have a mechanism to switch from polling for read only
to polling for read+write and back again in the form of watches. So this
patch changes the red_worker dummy watch implementation into a proper watch
implementation, and drops the entire EventListener concept since that then is
no longer needed.
This fixes spice-server using 400% CPU on my quad core machine as soon as
a client was connected to a multi head vm, and as an added bonus is a nice
cleanup IMHO.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
We allow channels to have different core implementations, but we were
relying on reds_stream_free to remove the stream watch on disconnect,
and reds_stream_free always uses the qemu core implementation.
So far we were getting away with this since all the alternative core
implementations always return NULL from watch_add.
But:
1) The code before this patch clearly was not correct, since it was matching
a channel-core watch_add with a qemu-core watch_remove
2) I plan to move red_worker over to actually using an alternative watch
implementation at which point this becomes a real problem
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
The red_worker EventListener struct is either embedded in one of:
1) DisplayChannelClient
2) CursorChannelClient
3) RedWorker
And as such gets destroyed when these get destroyed, in case 1 & 2 through
a call to red_channel_client_destroy().
So free-ing it when the ref-count becomes 0 is wrong, for cases:
1) and 2) this will lead to a double free;
3) this will lead to passing memory to free which was not returned by malloc.
This is not causing any issues as the ref-count never gets decremented, other
then in red_worker_main where it gets incremented before it gets decremented,
so it never becomes 0.
So we might just as well completely remove it.
Notes:
1) This is mainly a preparation patch for fixing issues introduced by
the move from epoll to poll
2) Since removing the ref-counting removes the one code path where listeners
would get set to NULL, this patch moves the setting of NULL to
pre_disconnect, where it should have been done in the first place since
red_client_destroy calls red_channel_client_disconnect
(through the dispatcher) followed by red_channel_client_destroy, so
after pre_disconnect the listener may be gone.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
While git-bisecting another issue I ended up hitting and not recognizing
the bug fixed by commit 7a079b452b.
While fixing this (again) I noticed that (even after the fix) not all
users of ChannelCbs first zero it. So this patch ensures that all users of
ChannelCbs first zero it, and does the same for ClientCbs while at it.
Since before this patch there were multiple zero-ing styles, some using
memset and other using a zero initializer this patch also unifies all
the zero-ing to use a NULL initializer for the first element.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
This fixes a core dumped observed once by repeated migration. So far 100
migrations and no recurrence.
Core was generated by `/home/alon/spice/upstream/bin/qemu-system-x86_64 --enable-kvm -qmp unix:/tmp/mi'.
Program terminated with signal 11, Segmentation fault.
11197 if (evt_listener && evt_listener->refs > 1) {
Missing separate debuginfos, use: debuginfo-install bluez-libs-4.98-3.fc17.x86_64 brlapi-0.5.6-4.fc17.x86_64 bzip2-libs-1.0.6-4.fc17.x86_64 cryptopp-5.6.1-6.fc17.x86_64 keyutils-libs-1.5.5-2.fc17.x86_64 libssh2-1.4.0-1.fc17.x86_64 nss-softokn-freebl-3.13.1-20.fc17.x86_64 xen-libs-4.1.2-11.fc17.x86_64 xz-libs-5.1.1-2alpha.fc17.x86_64
(gdb) bt
(gdb) l
11192 for (i = 0; i < MAX_EVENT_SOURCES; i++) {
11193 struct pollfd *pfd = worker.poll_fds + i;
11194 if (pfd->revents) {
11195 EventListener *evt_listener = worker.listeners[i];
11196
11197 if (evt_listener && evt_listener->refs > 1) {
11198 evt_listener->action(evt_listener, pfd);
11199 if (--evt_listener->refs) {
11200 continue;
11201 }
(gdb) p evt_listener
$1 = (EventListener *) 0x7f15a9a5d1e0
(gdb) p *evt_listener
Cannot access memory at address 0x7f15a9a5d1e0
(gdb) p i
$2 = 2
(gdb) p worker.listeners
$3 = {0x7f15bc832520, 0x7f15a406e1a0, 0x7f15a9a5d1e0, 0x0 <repeats 17 times>}
If specified, use SPICE_FOREIGN_MENU_SOCKET environment variable over
snprintf(pipe_name, PIPE_NAME_MAX_LEN, PIPE_NAME,
Platform::get_process_id()); since it's impossible to guess the client
pid when spice-xpi launches the client as a grand-child (via script
etc).
This is also more aligned with the way we handle SPICE_XPI_SOCKET
location.
Add spice_server_set_name() and spice_server_set_uuid() that allows
the client to identify a Spice server (useful to associate settings
with a particular server)
The SPICE_MSG_MAIN_NAME and SPICE_MSG_MAIN_UUID messages are only sent
to capable clients, announcing SPICE_MAIN_CAP_NAME_AND_UUID.
A message with a fixed-size array, such as uint8 uuid[16] will
generate an invalid code, missing the __nelements variable. Make sure
that variable is defined.
Currently, when a ticket has already expired, or is invalid, there is
no qemu log to tell what went wrong. This commit adds such a log.
Fixes rhbz#787669
If X509_NAME isn't undefined before including x509v3.h, very
weird compilation error occurs. It seems to be caused by duplicate
definitions for this symbols coming from wincrypto.h
mingw has a _ftime_s prototype in its headers, but no corresponding
symbol available at link time. Workaround this issue for now by
#defining it to _ftime. This is untested on win64 where the workaround
may not be needed.
common.h has some #define when doing win32 build to workaround a few
missing functions on these systems. However, since mingw32 has some
of these, this causes either warnings about redefining preprocessor
symbols or wreak havoc in mingw headers trying to use these symbols.
This commit wraps these symbols in an #ifndef __MINGW32__ to avoid
using them on this platform.
- Do not refer to .c files managed by another makefile (this will fail
make distclean)
- Do not refer to files by relative path (should use $top_srcdir for ex)
- Use LDADD for object linking instead of LDFLAGS, for linker flags
MSDN says the following about SetClipboardViewer(): "If an error occurs or there
are no other windows in the clipboard viewer chain, the return value is NULL".
Seems like the buggy case was "no other windows in the clipboard viewer chain",
which explains the 3rd party clipboard manager workaround detailed in the bug
description.
It also seems like SetClipboardViewer() does not clear the error state on
succcess. Calling SetLastError(0) before SetClipboardViewer() seems to solves
this issue.
Since we could not reproduce the bug on our env, the customer has verified on
several of their systems that a private build resolved the issue.
Add support for sending volume keys scancodes to the guest
RHBZ #552539
A good reference for mapping keymaps to scancodes can be found in
spice-gtk/gtk/keymaps.csv
Signed-off-by: Yonit Halperin <yhalperi@redhat.com>
The addition of smartcard control to the controller doesn't handle
the case when smartcard support is disabled at compile time. When
this is the case, this causes compile errors.
Now, cursor is being shown in all tests as a white rectangle and is
running in the screen doing a diagonal movement. It's a very simple
way to test cursor commands and is sufficient for our tests.
This is provided by <limits.h> on all platforms as long as _XOPEN_SOURCE
is defined. On Linux, this is 1024, on Solaris, this is 16, and on any
other platform, we now respect the value supported by the OS.
Signed-off-by: Dan McGee <dpmcgee@gmail.com>
Solaris has a pitiful maximum writev vector size of only 16, so the ping
request at initial startup destroyed this call and broke things
immediately. Reimplement stream_writev_cb() to respect IOV_MAX and break
the writev() calls into chunks as necessary. Care was taken to return
the correct values as necessary so the EAGAIN handling logic can
determine where to resume the writev call the next time around.
Signed-off-by: Dan McGee <dpmcgee@gmail.com>
This removes the epoll dependency we had in red_worker, which was the
last Linux-specific call we were using in the entire Spice server. Given
we never have more than 10 file descriptors involved, there is little
performance gain had here by using epoll() over poll().
The biggest change is introduction of a new pre_disconnect callback;
this is because poll, unlike epoll, cannot automatically remove file
descriptors as they are closed from the pollfd set. This cannot be done
in the existing on_disconnect callback; that is too late as the stream
has already been closed and the file descriptor lost. The on_disconnect
callback can not be moved before the close and other operations easily
because of some behavior that relies on client_num being set to a
certain value.
Signed-off-by: Dan McGee <dpmcgee@gmail.com>
Rather than assign the callbacks one-by-one, we can just memcpy the
struct into the one we have allocated in our RedChannel object, which is
much more efficient, not to mention future-proof when more callbacks are
added.
Signed-off-by: Dan McGee <dpmcgee@gmail.com>
We had multiple stub methods that simply called other disconnect
methods, making my head hurt with the indirection. Call the right
methods at the right time and rip out the stub methods; if they are
truely needed later they can be added again.
Signed-off-by: Dan McGee <dpmcgee@gmail.com>
With future patches in mind that will allow for some other
non-Linux-specific event polling sytem to be used, rename this to a more
generic name. All of the select/poll/epoll/kqueue family of calls are
related to evented I/O, so 'event_' makes sense in this case.
Signed-off-by: Dan McGee <dpmcgee@gmail.com>
This is supported by the GNU linker, but not the Solaris linker, which
is used as the default on that platform even when compiling with GCC.
Omit passing the option to the linker on platforms that do not support
it.
Signed-off-by: Dan McGee <dpmcgee@gmail.com>
spicevmc calls red_channel_client_destroy() on the rcc when it disconnects
since we don't want to delay the destroy until the session gets closed as
spicevmc channels can be opened, closed and opened again during a single
session.
This causes red_channel_client_destroy() to get called twice, triggering
an assert, when a connected channel gets destroyed.
This was fixed with commit ffc4de01e6 for
the case where: a spicevmc channel was open on client disconnected, and
the main channel disconnect gets handled first.
But the channel can also be destroyed when the chardev gets unregistered
with the spice-server. This path still triggers the assert.
This patch fixes this by adding a destroying flag to the rcc struct, and
also moves the previous fix over to the same, more clean, method of
detecting this special case.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
fixes rhbz 790749 use after free of SpiceChannelEventInfo.
The lifetime of the SpiceChannelEventInfo was that of RedsStream, but it
is used by main_dispatcher_handle_channel_event after the RedsStream is
freed for the cursor and display channels. Making SCEI allocation be at
RedsStream allocation, and deallocation after the DESTROY event is
processed by core->channel_event, fixes use after free.
red_printf() takes care of adding a newline to all messages; remove the
extra newline from all messages and macros that were doubling them up.
Signed-off-by: Dan McGee <dpmcgee@gmail.com>
As suggested by Alon, a simple automated test to try to find
regressions in Spice code.
To use this, compile Spice with --enable-automated-tests and
run test_display_streaming passing --automated-tests as parameter.
Some non-Linux platforms return a (caddr_t *) result for the return
value of mmap(), which is very unfortunate. Add a (void *) cast to
explicitly avoid the warning when compiling with -Werror.
For the IO vector related stuff, signed vs. unsigned comes into play so
adding a (void *) cast here is technically correct for all platforms.
Signed-off-by: Dan McGee <dpmcgee@gmail.com>
