Age | Commit message (Collapse) | Author |
|
|
|
(Reported by Loïc Minier.)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
to early
The current code turns of PTS tagging as soon as a match is found. But depending
on picture reordering, there may be later frames which still have the tag. The result
is that most likely the highest bit is set which makes the PTS values large negative
numbers which cause a clock error and make streams unplayable.
To fix this issue, a stable counter is introduced. The two passes of PTS tagging are
now switched after the tag has been seen stable for 100 frames. This should protect
us from picture reordering issues.
--HG--
extra : transplant_source : I%2A%BBi%A5nb/%5E%12%9Ay%7B%BAj%7D%0B%16%0By
|
|
|
|
Fixes bug #194.
|
|
|
|
This covers the internal snapshot and the version in Debian lenny.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
implementations.
Some implementations are buggy and lock resources (for example the display or
internal data structures) in different order, which results in deadlocks.
As XVMC_LOCKDISPLAY_SAFE is not defined by default, most API functions will
now be guarded by a LockDisplay()/UnlockDisplay() pair, which imposes a lock
order at least for the resource display and hence avoids those deadlocks.
(transplanted from 580a2a9148618131cedfbc9058ac7979ca16f69b)
--HG--
extra : transplant_source : X%0A%2A%91Ha%811%CE%DF%BC%90X%ACyy%CA%16%F6%9B
|
|
Am 04.01.2009 um 04:55 schrieb Adrian Bunk:
> ....
> In file included from ../../../src/xine-engine/xine_internal.h:33,
> from noise.c:24:
> .../../../include/xine.h:2230: warning: 'xine_tvsystem' is deprecated
> noise.c: Assembler messages:
> noise.c:155: Error: bad register name `%rax'
> noise.c:161: Error: bad register name `%rax)'
> <-- snip -->
One problem is that the configure script thinks we're running a 64-bit
system:
,----
| checking build system type... x86_64-unknown-linux-gnu
| checking host system type... x86_64-unknown-linux-gnu
| checking build system type... (cached) x86_64-unknown-linux-gnu
`----
This is bad, build and host type should be passed explicitly in
debian/rules.
|
|
now invalid.
But as PTS values are stored in FFmpeg's decoder, there is no way to reset them to 0.
Therefore PTS tagging has been introduced. At discontinuity a tag is generated and
applied to all new PTS values. Any returned PTS value is checked for this tag and
outdated PTS values are reset to 0. When the tag appears on returned PTS values then
tagging is reset.
|
|
the buffer
|
|
order.
Attaching buffer PTS, which are in decoding order, to decoded images is
simply wrong. FFmpeg meanwhile provides a way to pass PTS values through
its decoder too. As a result they get reordered to display order and can
be attached to the decoded frames.
|
|
Currently, once the tvtime plugin has locked onto a telecine pattern, it
will wait PULLDOWN_ERROR_WAIT (a hardcoded #defined value) number of frames
before switching to filmmode.
This sensitivity is excessively high (i.e. the value is too low) for certain
content -- the kind of content that was shot on film but edited in video
mode, so telecine patterns are constantly breaking (examples like Buffy,
Simpsons and Family Guy are especially egregious offenders).
The attached patch turns this constant into a modifiable post plugin
parameter called pulldown_error_wait.
Xine helpfully emits a XINE_EVENT_POST_TVTIME_FILMMODE_CHANGE event when
film mode changes (a patch I submitted some years back). With the attached
patch, a front-end can monitor the frequency of these events, and
dynamically adjust pulldown_error_wait in a sensible way.
|
|
input->read may return negative error codes or read less than we want
so we should check for the right return value instead of just not 0
|
|
do not forward data if there is not enough
|
|
check buffer lengths to avoid out of bound access when
decoding the header.
Based on a patch by Matthias Hopf <mhopf@suse.de>.
|
|
if the atom size is shorter than the header size, do not try
to decompress anything, as this would lead to zlib reading
out of bound data.
|
|
check the size of allocated buffers to prevent out of bound access
|
|
Based on a patch by Matthias Hopf <mhopf@suse.de>.
|
|
|
|
return error when the allocation function returns NULL
Otherwise xine might be induced to segfault by bad user data.
|
|
Some input plugins (e.g. file) return negative error codes from read,
this should be treated as no (more) data available.
|
|
Currently, this is satisfied in all locations where it is called,
but it is more prudent to add the check.
|
|
get_size might return -1 (e.g. for streams whose size is unknown),
but demux_mod is not able to handle this.
This is particularly bad because it is later assigned to unsigned types
(demux_mod_t.filesize is size_t).
Based on a patch by Matthias Hopf <mhopf@suse.de>.
|
|
Some input plugins (e.g. file) return negative error codes from read,
this should be treated as no (more) data available.
This is particularly bad because the error code is assigned to an
unsigned integer variable for use by the caller.
Based on a patch by Matthias Hopf <mhopf@suse.de>
|
|
matroska
while codec_private_len is unsigned, the size is later used to calculate
the signed xine_bmiheader.size
|
|
Add checks for negative return values in aac,ac3,dts,mpc,
nsf,ogg,shn,slave,ts,tta,vox demuxers.
Some input plugins (e.g. file) return negative error codes from read,
this should be treated as no (more) data available.
This is particularly the negative size is then assigned to buf->size,
potentially causing overflows elsewhere.
The patch also removes the duplication of the (previously) == 0 handler
in demux_ac3.
|
|
When a track's fifo is not set up (typically because the track type is invalid),
do not call init_codec, as all implementations dereference track->fifo,
segfaulting if it is NULL.
|
|
The real_parse_headers function in demux_real.c in xine-lib 1.1.12,
and other 1.1.15 and earlier versions, relies on an untrusted input
length value to "reindex into an allocated buffer," which allows
remote attackers to cause a denial of service (crash) via a crafted
value, probably an array index error.
|
|
xine-lib 1.1.12, and other 1.1.15 and earlier versions, relies on an
untrusted input value to determine the memory allocation and does not
check the result for (1) the MATROSKA_ID_TR_CODECPRIVATE track entry
element processed by demux_matroska.c; and (2) PROP_TAG, (3) MDPR_TAG,
and (4) CONT_TAG chunks processed by the real_parse_headers function
in demux_real.c; which allows remote attackers to cause a denial of
service (NULL pointer dereference and crash) or possibly execute
arbitrary code via a crafted value.
|
|
xine-lib 1.1.12, and other 1.1.15 and earlier versions, does not
properly handle (a) negative and (b) zero values during unspecified
read function calls in input_file.c, input_net.c, input_smb.c, and
input_http.c, which allows remote attackers to cause a denial of
service (crash) or possibly execute arbitrary code via vectors such as
(1) a file or (2) an HTTP response, which triggers consequences such
as out-of-bounds reads and heap-based buffer overflows.
|
|
Multiple integer overflows in xine-lib 1.1.12, and other 1.1.15 and
earlier versions, allow remote attackers to cause a denial of service
(crash) or possibly execute arbitrary code via (1) crafted width and
height values that are not validated by the mymng_process_header
function in demux_mng.c before use in an allocation calculation or (2)
crafted current_atom_size and string_size values processed by the
parse_reference_atom function in demux_qt.c.
|
|
Multiple heap-based buffer overflows in xine-lib 1.1.12, and other
1.1.15 and earlier versions, allow remote attackers to execute
arbitrary code via vectors related to (1) a crafted EBML element
length processed by the parse_block_group function in
demux_matroska.c; (2) a certain combination of sps, w, and h values
processed by the real_parse_audio_specific_data and
demux_real_send_chunk functions in demux_real.c; and (3) an
unspecified combination of three values processed by the open_ra_file
function in demux_realaudio.c. NOTE: vector 2 reportedly exists
because of an incomplete fix in 1.1.15.
|
|
Multiple heap-based buffer overflows in xine-lib 1.1.12, and other
versions before 1.1.15, allow remote attackers to execute arbitrary
code via vectors related to (1) a crafted metadata atom size processed
by the parse_moov_atom function in demux_qt.c and (2) frame reading in
the id3v23_interp_frame function in id3.c. NOTE: as of 20081122, it is
possible that vector 1 has not been fixed in 1.1.15.
case ( FOURCC_TAG('C', 'O', 'M', 'M') ):
_x_meta_info_set_generic(stream, XINE_META_INFO_COMMENT, buf + 1 + 3, id3_encoding[enc]);
|
|
On XINE_GUI_SEND_DRAWABLE_CHANGED, the clipping area is adjusted
to the new HWND and the frame shown again. This allows to switch
the window where the video is shown at runtime, simliar to the X11
drivers.
|