If the length of a message exceeds the maximum length of the buffer, the
buffer size will reach its maximum value and stay there forever, with no
message ever being successfully processed. Since libwayland uses
level-triggered epoll, this will cause the compositor to loop forever
and consume CPU time. In libwayland 1.22 and below, there was an
explicit check that caused messages exceeding 4096 bytes to result in an
EOVERFLOW error, preventing the loop. However, this check was removed
between d074d52902 ("connection: Dynamically resize connection buffers").
To prevent this problem, always limit the size of messages to 4096 bytes.
Since the default and minimum buffer size is 4096 bytes, this ensures
that a single message will always fit in the buffer. It would be
possible to allow larger messages if the buffer size was larger, but the
maximum size of a message should not depend on the buffer size chosen by
the compositor.
Rejecting messages that exceed 4092 bytes seems to have the advantage of
reserving 4 bits, not 3, in the size field for future use. However,
message sizes in the range [0x0, 0x7] are invalid, so one can obtain a
fourth bit by negating the meaning of bit 12 if bits 0 through 11
(inclusive) are 0. Allowing 4096-byte messages provides the far more
important advantage that regressions compared to 1.22 are impossible
and regressions compared to 1.23 are extremely unlikely. The only case
where a regression is possible is:
- The receiving side is using libwayland 1.23.
- The sending side is either using libwayland 1.23 or is not using
libwayland.
- The sender sends a message exceeding 4096 bytes.
- If the sender of the large message is the client, the server has
increased the buffer size from the default value.
This combination is considered extremely unlikely, as libwayland 1.22
and below would disconnect upon receiving such a large message.
4096-byte messages, however, have always worked, so there was no reason
to avoid sending them.
Fixes: d074d52902 ("connection: Dynamically resize connection buffers").
Fixes: #494
Signed-off-by: Demi Marie Obenour <demi@invisiblethingslab.com>
When using fixed size connection buffers, if either the client or the
server is sending requests faster than the other end can cope with, the
connection buffers will fill up, eventually killing the connection.
This can be a problem for example with Xwayland mapping a lot of
windows, faster than the Wayland compositor can cope with, or a
high-rate mouse flooding the Wayland client with pointer events.
To avoid the issue, resize the connection buffers dynamically when they
get full.
Both data and fd buffers are resized on demand.
The default max buffer size is controlled via the wl_display interface
while each client's connection buffer size is adjustable for finer
control.
The purpose is to explicitly have larger connection buffers for specific
clients such as Xwayland, or set a larger buffer size for the client
with pointer focus to deal with a higher input events rate.
v0: Manuel:
Dynamically resize connection buffers - Both data and fd buffers are
resized on demand.
v1: Olivier
1. Add support for unbounded buffers on the client side and growable
(yet limited) connection buffers on the server side.
2. Add the API to set the default maximum size and a limit for a given
client.
3. Add tests for growable connection buffers and adjustable limits.
v2: Additional fixes by John:
1. Fix the size calculation in ring_buffer_check_space()
2. Fix wl_connection_read() to return gracefully once it has read up to
the max buffer size, rather than returning an error.
3. If wl_connection_flush() fails with EAGAIN but the transmit
ring-buffer has space remaining (or can be expanded),
wl_connection_queue() should store the message rather than
returning an error.
4. When the receive ring-buffer is at capacity but more data is
available to be read, wl_connection_read() should attempt to
expand the ring-buffer in order to read the remaining data.
v3: Thomas Lukaszewicz <tluk@chromium.org>
Add a test for unbounded buffers
v4: Add a client API as well to force bounded buffers (unbounded
by default (Olivier)
v5: Simplify ring_buffer_ensure_space() (Sebastian)
Co-authored-by: Olivier Fourdan <ofourdan@redhat.com>
Co-authored-by: John Lindgren <john@jlindgren.net>
Co-authored-by: Sebastian Wick <sebastian@sebastianwick.net>
Signed-off-by: Manuel Stoeckl <code@mstoeckl.com>
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Signed-off-by: John Lindgren <john@jlindgren.net>
Signed-off-by: Sebastian Wick <sebastian@sebastianwick.net>
Closes: https://gitlab.freedesktop.org/wayland/wayland/-/issues/237
This is less cryptic to read than letters, and allows the compiler
to check switch statements exhaustiveness.
Signed-off-by: Simon Ser <contact@emersion.fr>
Allow setting a name for an event queue. The queue is used only for
printing additional debug information.
Debug output can now show the name of the event queue an event is
dispatched from, or the event queue of a proxy when a request is made.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Since server IDs are basically indistinguishable from really big client
IDs at many points in the source, it's theoretically possible to overflow
a map and either overflow server IDs into the client ID space, or grow
client IDs into the server ID space. This would currently take a massive
amount of RAM, but the definition of massive changes yearly.
Prevent this by placing a ridiculous but arbitrary upper bound on the
number of items we can put in a map: 0xF00000, somewhere over 15 million.
This should satisfy pathological clients without restriction, but stays
well clear of the 0xFF000000 transition point between server and client
IDs. It will still take an improbable amount of RAM to hit this, and a
client could still exhaust all RAM in this way, but our goal is to prevent
overflow and undefined behaviour.
Fixes#224
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
The client side closure traces have incorrect object ids for new server
generated objects. This is because create_proxies() overwrites the id in
'n' type arguments by storing a pointer to the actual object in the 'o'
field of the union.
Getting back to an id from this pointer requires accessing a structure
that isn't visible outside of wayland-client.c.
Add a function pointer to fish the correct value out of the argument and
pass it to wl_closure_print.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Before this patch, setting WAYLAND_DEBUG=1 or WAYLAND_DEBUG=client made
a program log all requests sent and events that it processes. However,
some events received are not processed. This can happen when a Wayland
server sends an event to an object that does not exist, or was recently
destroyed by the client program (either before the event was decoded,
or after being decoded but before being dispatched.)
This commit prints all discarded messages in the debug log, producing
lines like:
[1234567.890] discarded [unknown]@42.[event 0](0 fd, 12 byte)
[1234567.890] discarded wl_callback@3.done(34567)
[1234567.890] discarded [zombie]@13.[event 1](3 fd, 8 byte)
The first indicates an event to an object that does not exist; the
second, an event to an object that was deleted after decoding, but
before dispatch; the third, an event to an object that left a
'zombie' marker behind to indicate which events have associated
file descriptors.
Signed-off-by: Manuel Stoeckl <code@mstoeckl.com>
Including wayland-server-core.h in wayland-private.h is problematic
because wayland-private.h is included by wayland-scanner which should be
able to build against non-POSIX platforms (e.g. MinGW). The only reason
that wayland-server-core.h was included in wayland-private.h was for the
wl_private_signal definitions, so move those to a
wayland-server-private.h file that can be included by both
wayland-server.c and the tests.
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Rather than have two versions of the macro with slightly different
interfaces, just use wl_container_of internally.
This also removes use of statement expressions, a GNU C extension.
Signed-off-by: Michael Forney <mforney@mforney.org>
In the past much code (weston, efl/enlightenment, mutter) has
freed structures containing wl_listeners from destroy handlers
without first removing the listener from the signal. As the
destroy notifier only fires once, this has largely gone
unnoticed until recently.
Other code does not (Qt, wlroots) - and removes itself from
the signal before free.
If somehow a destroy signal is listened to by code from both
kinds of callers, those that free will corrupt the lists for
those that don't, and Bad Things will happen.
To avoid these bad things, remove every item from the signal list
during destroy emit, and put it in a list all its own. This way
whether the listener is removed or not has no impact on the
following emits.
Signed-off-by: Derek Foreman <derekf@osg.samsung.com>
Reviewed-by: Simon Ser <contact@emersion.fr>
Reviewed-by: Markus Ongyerth <wl@ongy.net>
We need to close file descriptors sent to zombie proxies to avoid leaking
them, and perhaps more importantly, to prevent them from being dispatched
in events on other objects (since they would previously be left in the
buffer and potentially fed to following events destined for live proxies)
Signed-off-by: Derek Foreman <derekf@osg.samsung.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Since we now have the WL_MAP_ENTRY_ZOMBIE flag to determine whether or
not a client-side object is a zombie, we can remove the faux object.
[daniels: Extracted from Derek's bespoke-zombie patch as an intermediate
step.]
Signed-off-by: Derek Foreman <derekf@osg.samsung.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Add a new map entry flag to indicate that the object received is valid,
but a zombie. Previously this relied on a fixed object pointer, but
future patches in this series will have map entries returning either
NULL, or a different structure type entirely, for zombie objects.
wl_object_is_zombie() now solely uses the new flag to determine whether
or not the object is a zombie.
[daniels: Extracted from Derek's bespoke-zombie patch as an intermediate
step.]
Signed-off-by: Derek Foreman <derekf@osg.samsung.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Add a helper function which determines whether or not an object is a
zombie.
[daniels: Extracted from Derek's bespoke-zombie patch as an intermediate
step.]
Signed-off-by: Derek Foreman <derekf@osg.samsung.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
On the client side we're going to need to know if an object from the
map is a zombie before we attempt to dereference it, so we need to
pass this to the iterator.
Reviewed-by: Daniel Stone <daniels@collabora.com>
Signed-off-by: Derek Foreman <derekf@osg.samsung.com>
wl_list_for_each_safe, which is used by wl_signal_emit is not really
safe. If a signal has two listeners, and the first one removes and
re-inits the second one, it would enter an infinite loop, which was hit
in weston on resource destruction, which emits a signal.
This commit adds a new version of wl_signal, called wl_priv_signal,
which is private in wayland-server.c and which does not have this problem.
The old wl_signal cannot be improved without breaking backwards compatibility.
Signed-off-by: Giulio Camuffo <giulio.camuffo@kdab.com>
Reviewed-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
message-test.c did not cover wl_message_count_arrays, so add one test that
specifically tests this method. Note that this exposes wl_message_count_arrays
in a private header (wayland-private.h), and removes the `static` modifier of
the implementation.
Signed-off-by: Yong Bakos <ybakos@humanoriented.com>
Reviewed-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
Explicitly set the data member to an invalid memory address during
wl_array_release, such that re-using a freed wl_array without re-initializing
causes a crash. In addition, this pointer assignment makes wl_array_release
testable.
Define a constant for the invalid memory address, and add documentation about
this behavior, starting at libwayland version 1.13.
See https://lists.freedesktop.org/archives/wayland-devel/2016-September/031116.html
Signed-off-by: Yong Bakos <ybakos@humanoriented.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
[Pekka: remove the doc about crashing]
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
To complement on the new resource created signal, this allows to
iterate over the existing resources of a client.
Signed-off-by: Giulio Camuffo <giulio.camuffo@kdab.com>
Reviewed-by: Jonas Ådahl <jadahl@gmail.com>
[Pekka: added empty lines, init ret in for_each_helper()]
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
Some headers and source files have been using types such as uint32_t
without explicitly including stdint.h.
Explicitly include stdint.h where appropriate.
Signed-off-by: Yong Bakos <ybakos@humanoriented.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Declarations for wl_connection and wl_closure are not needed here.
wl_closure already has a complete definition.
Removing these forward declarations results in a clean, warning-free compile.
Signed-off-by: Yong Bakos <ybakos@humanoriented.com>
Reviewed-by: Derek Foreman <derekf@osg.samsung.com>
[Updated to apply to trunk]
Reviewed-by: Bryce Harrington <bryce@osg.samsung.com>
Move the wl_interface_equal prototype to the top of wayland-private, where
it is not buried in the middle of map, connection and closure functions.
Move the implementation out of connection and into util. This is a utility
function, not specific to connections, and has call sites within connection,
wayland-client and wayland-server.
Signed-off-by: Yong Bakos <ybakos@humanoriented.com>
Reviewed-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
Reviewed-by: Bryce Harrington <bryce@osg.samsung.com>
This adds an API to get the file descriptor for a client.
The client file descriptor can be used for a wayland compositor to validate
a request from a client if there are any additional information provided from
the client's file descriptor.
For instance, this will be helpful in some linux distributions, in which SELinux
or SMACK is enabled. In those environments, each file (including socket) will have
each security contexts in its inode as xattr member variable. A wayland compositor
can validate a client request by getting the file descriptor of the client and
by checking the security contexts associated with the file descriptor.
Signed-off-by: Sung-Jin Park <input.hacker@gmail.com>
When processing a closure, data in the connection can be consumed again
if the closure itself invokes extra event dispatch. In that case the
remaining data size is also altered, so the variable len should be
updated after the closure is processed.
Signed-off-by: Jaeyoon Jung <jaeyoon.jung@lge.com>
Reviewed-by: Jonas Ådahl <jadahl@gmail.com>
Reviewed-by: Derek Foreman <derekf@osg.samsung.com>
Signed-off-by: Jonas Ådahl <jadahl@gmail.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Marek Chalupa <mchqwerty@gmail.com>
Reviewed-by: Derek Foreman <derekf@osg.samsung.com>
On many places in the code we use wl_log + abort or wl_log + assert(0).
Replace these with one call to wl_abort, so that we don't mix abort(),
assert(0) and we'll save few lines
Signed-off-by: Marek Chalupa <mchqwerty@gmail.com>
Reviewed-by: Derek Foreman <derekf@osg.samsung.com>
A few of the header files had function prototypes that were not
following project conventions, sometimes even in the same file.
Corrected these to follow as per wayland-os.h.
Signed-off-by: Jon A. Cruz <jonc@osg.samsung.com>
Calling close() on the same file descriptor that a previous call to
close() already closed is wrong, and racy if another thread received
that same file descriptor as a eg. new socket or actual file.
There are two situations where wl_connection_destroy() would close its
file descriptor and then another function up in the call chain would
close the same file descriptor:
* When wl_client_create() fails after calling wl_connection_create(),
it will call wl_connection_destroy() before returning. However, its
caller will always close the file descriptor if wl_client_create()
fails.
* wl_display_disconnect() unconditionally closes the display file
descriptor and also calls wl_connection_destroy().
So these two seem to expect wl_connection_destroy() to leave the file
descriptor open. The other caller of wl_connection_destroy(),
wl_client_destroy(), does however expect wl_connection_destroy() to
close its file descriptor, alas.
This patch changes wl_connection_destroy() to indulge this majority of
two callers by simply not closing the file descriptor. For the benefit
of wl_client_destroy(), wl_connection_destroy() then returns the
unclosed file descriptor so that wl_client_destroy() can close it
itself.
Since wl_connection_destroy() is a private function called from few
places, changing its semantics seemed like the more expedient way to
address the double-close() problem than shuffling around the logic in
wl_client_create() to somehow enable it to always avoid calling
wl_connection_destroy().
Signed-off-by: Benjamin Herr <ben@0x539.de>
Reviewed-by: Marek Chalupa <mchqwerty@gmail.com>
Reviewed-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
This commit adds support for server-side languages bindings. This is done
in two ways:
1. Adding a wl_resource_set_dispatcher function that corresponds to
wl_resource_set_interface. The only difference between the two functions
is that the new version takes a dispatcher along with the implementation,
data, and destructor. This allows for runtime calling of native language
functions for callbacks instead of having to generate function pointers.
2. Adding versions of wl_resource_post_event and wl_resource_queue_event
that take an array of wl_argument instead of a variable argument list.
This allows for easier run-time argument conversion and removes the need
for libffi-based calling of variadic functions.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
This commit adds version information to wl_message signatures and a
wl_message_get_since function to retrieve. The since version comes in the
form of a (possible) integer at the begining of the message. If the
message starts with an integer, then it specifies the "since" version of
that message. Messages present in version one do not get this "since"
information. In this way we can run-time detect the version information
for a structure on a per-message basis.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
With the work to add wl_resource accessors and port weston to use them,
we're ready to make wl_resource and wl_object opaque structs. We keep
wl_buffer in the header for EGL stacks to use, but don't expose it by
default. In time we'll remove it completely, but for now it provides a
transition paths for code that still uses wl_buffer.
Reviewed-by: Jason Ekstrand<jason@jlekstrand.net>
In order to use the second-lowest bit of each pointer in wl_map for the
WL_MAP_ENTRY_LEGACY flag, every pointer has to be a multiple of 4. This
was a good assumption, except with WL_ZOMBIE_OBJECT. This commit creates
an actual static variable to which WL_ZOMBIE_OBJECT now points. Since
things are only every compared to WL_ZOMBIE_OBJECT with "==" or "!=", the
only thing that matters is that it is unique.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
The implementation in this commit allows for one bit worth of flags. If
more flags are desired at a future date, then the wl_map implementation
will have to change but the wl_map API will not.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
The original wl_map implementation did no checking to ensures that ids fell
on the correct side of the WL_SERVER_ID_START line. This meant that a
client could send the server a server ID and it would happily try to use
it. Also, there was no distinction between server-side and client-side in
wl_map_remove. Because wl_map_remove added the entry to the free list
regardless of which side it came from, the following set of actions would
break the map:
1. Client creates a bunch of objects
2. Client deletes one or more of those objects
3. Client does something that causes the server to create an object
Because of the problem in wl_map_remove, the server would take an old
client-side id, apply the WL_SERVER_ID_START offset, and try to use it as a
server-side id regardless of whether or not it was valid.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
This commit adds a flags parameter to wl_closure_invoke(). The so far
added flags are ment to specify if the invokation is client side or
server side. When on the server side, closure arguments of type 'new_id'
should be invoked as a integer id while on the client side they should
be invoked as a pointer to a proxy object.
This fixes a bug happening when the address of a client side 'new_id'
proxy object did not fit in a 32 bit integer.
krh: Squashed test suite compile fix from Jason Ekstrand.
Signed-off-by: Jonas Ådahl <jadahl@gmail.com>
The primary purpose of this patch is to clean up wl_closure and separate
closure storage, libffi, and the wire format. To that end, a number of changes
have been made:
- The maximum number of closure arguments has been changed from a magic number
to a #define WL_CLOSURE_MAX_ARGS
- A wl_argument union has been added for storing a generalized closure
argument and wl_closure has been converted to use wl_argument instead of the
combination of libffi, the wire format, and a dummy extra buffer. As of
now, the "extra" field in wl_closure should be treated as bulk storage and
never direclty referenced outside of wl_connection_demarshal.
- Everything having to do with libffi has been moved into wl_closure_invoke
and the convert_arguments_to_ffi helper function.
- Everything having to do with the wire format has been restricted to
wl_connection_demarshal and the new static serialize_closure function. The
wl_closure_send and wl_closure_queue functions are now light wrappers around
serialize_closure.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
When events are queued, the associated proxy objects (target proxy and
potentially closure argument proxies) are verified being valid. However,
as any event may destroy some proxy object, validity needs to be
verified again before dispatching. Before this change this was done by
again looking up the object via the display object map, but that did not
work because a delete_id event could be dispatched out-of-order if it
was queued in another queue, causing the object map to either have a new
proxy object with the same id or none at all, had it been destroyed in
an earlier event in the queue.
Instead, make wl_proxy reference counted and increase the reference
counter of every object associated with an event when it is queued. In
wl_proxy_destroy() set a flag saying the proxy has been destroyed by the
application and only free the proxy if the reference counter reaches
zero after decreasing it.
Before dispatching, verify that a proxy object still is valid by
checking that the flag set in wl_proxy_destroy() has not been set. When
dequeuing the event, all associated proxy objects are dereferenced and
free:ed if the reference counter reaches zero. As proxy reference counter
is initiated to 1, when dispatching an event it can never reach zero
without having the destroyed flag set.
Signed-off-by: Jonas Ådahl <jadahl@gmail.com>
Exporting unprefixed symbols is a pretty bad idea so don't do that.
Instea of redefining it WL_ARRAY_LENGTH, we just move the define to
our private header. The scanner generates code that uses ARRAY_LENGTH,
but we can just make it count the number elements and emit an integer
constant instead.
On the client side where we queue up multiple events before dispatching, we
need to look up the receiving proxy and argument proxies immediately before
calling the handler. Between queueing up multiple events and eventually
invoking the handler, previous handlers may have destroyed some of the
proxies.
The update callback for the file descriptors was always a bit awkward and
un-intuitive. The idea was that whenever the protocol code needed to
write data to the fd it would call the 'update' function. This function
would adjust the mainloop so that it polls for POLLOUT on the fd so we
can eventually flush the data to the socket.
The problem is that in multi-threaded applications, any thread can issue
a request, which writes data to the output buffer and thus triggers the
update callback. Thus, we'll be calling out with the display mutex
held and may call from any thread.
The solution is to eliminate the udpate callback and just require that
the application or server flushes all connection buffers before blocking.
This turns out to be a simpler API, although we now require clients to
deal with EAGAIN and non-blocking writes. It also saves a few syscalls,
since the socket will be writable most of the time and most writes will
complete, so we avoid changing epoll to poll for POLLOUT, then write and
then change it back for each write.
Expose these to other files using wayland-private.h, so wayland-client.c
can walk NULLables properly.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
