As Anders correctly pointed out in I7d8f84b2e, constantly resetting state will
turn init_dissection into a bit of a hot path. Especially as we will already
bear the overhead of switching files, we don't want to fall any further behind
than we have to.
This change includes three unrelated optimizations that reduce the cost of
init_dissection by about 40% as measured by callgrind:
- only initialize ares/ADNS if that preference is enabled (this of course only
applies if you specify -n to tshark or otherwise disable the preference)
- use memcpy instead of a loop in sigcomp UDVM init
- use memcpy instead of a loop in bootp dissector
The only remaining obvious hot spot in this path is reassembly_table_init since
it is called by so many dissectors. Suggestions (perhaps to get rid of the
GPtrArray) welcome.
Oh, and one other change to use g_strerror instead of strerror as insisted
upon by the API pre-commit hook.
Change-Id: I18a74f2b64b25498116079bd4e7fc2b335c7703a
Reviewed-on: https://code.wireshark.org/review/2738
Reviewed-by: Anders Broman <a.broman58@gmail.com>
Two conditions were not checked, state_length == 0 && state_begin != 0
and the boundaries of the state buffer. The former is not a big deal,
but the second issue causes a buffer overrun (detected by ASAN).
The buffer size is supposed to be stored in the state buffer, that was
not the case for the initial two SIP SDP and Presence state buffers.
Fix a typo for presence_buf zero-ing while at it.
Bug: 9601
Change-Id: I41dde83185da60b670cca010ecc7b2a2aaaedeb9
Reviewed-on: https://code.wireshark.org/review/1529
Reviewed-by: Anders Broman <a.broman58@gmail.com>
(Using sed : sed -i '/^ \* \$Id\$/,+1 d')
Fix manually some typo (in export_object_dicom.c and crc16-plain.c)
Change-Id: I4c1ae68d1c4afeace8cb195b53c715cf9e1227a8
Reviewed-on: https://code.wireshark.org/review/497
Reviewed-by: Anders Broman <a.broman58@gmail.com>
obvious that the returned string is ephemeral, and opens up the original names
in the API for versions that take a wmem pool (and thus can work in any scope).
svn path=/trunk/; revision=54249
crash found while fuzzing, and hopefully fixes bug 4867.
In sigcomp_init_udvm() initialize the first 8 bytes of our buffers,
which squelches a valgrind complaint.
svn path=/trunk/; revision=33321
The svn version of wireshark does not decode a sigcomp message, if presence
static dictionary (RFC 5112) is used. I have attached a patch for adding this support.
svn path=/trunk/; revision=26550
As a workarond for the problem that accsessing a state with a state id > the stored
minimum access length fail, only store and compare the state with the minimum access
lengt specified in the standard of 6 bytes.
svn path=/trunk/; revision=20041
instead of while loops to initialize variables. Use a define instead
of magic numbers. Don't use guint8s for counters. Add a corresponding
entry to the release notes.
svn path=/trunk/; revision=15715
-use g_snprintf instead of sprintf and snprintf
-use g_strdup_printf where appropriate
-remove #include "snprintf.h" (as only g_snprintf should be used)
-replace some more alloc/realloc/calloc/free with their glib pendants
svn path=/trunk/; revision=15264
epan/sigcomp_state_hdlr.c epan/sigcomp_state_hdlr.h
---------------------------------------------------
Function udvm_state_access:
1. Removed the state_vars_valid parameter. It is no longer used.
2. The function now correctly updates each of state_length,
state_address and state_instruction with saved state value.
[Previously this was only done for state referenced in the SigComp
message header.]
3. Removed code that (incorrectly) read byte_copy_left from UDVM memory
within the loop that writes the state to UDVM memory.
4. Removed the code that set the useful values in UDVM memory. This
should only be done for state referenced in the SigComp message
header. (The decompress_sigcomp_message function, sets the useful
values.)
epan/sigcomp-udvm.c epan/sigcomp-udvm.h
---------------------------------------
Function decompress_sigcomp_message:
1. Added three new parameters; header_len, byte_code_state_len,
byte_code_id_len. The byte_code_state_len and byte_code_id_len
are used to set the useful values in UDVM memory. The
header_len is required to calculate the cycles available
for decompression.
2. Various minor corrections to spellings, etc in generated
messages.
3. Reduced the number of lines of output for print_level_1. Now
the execution trace shows the instruction name and parameter
values on one line.
[This was done because, we found that even at the lowest
level of detail ethereal became unusably slow; running on
a 500 MHz SPARC. This seems to be related to the number of
lines produced by the SigComp trace, which can easily be over
1000 with the modified code.]
4. Removed the used of some floating point functions. In all
cases they were unnecessary and could potentially cause the
code to produce the wrong results on some platforms (although
this is unlikely).
5. The useful values are now set correctly, using the new
byte_code_state_len and byte_code_id_len parameters.
6. The message header length is now included in the calculation of
maximum_UDVM_cycles. Previously, the calculation could underestimate
the value, resulting in the (small) chance that some legitimate
messages might fail to decompress.
Note: The calculation might now slightly over-estimate the cycles
but this is a reasonable thing to do.
7. Implemented the PUSH, POP, CALL and RETURN instructions.
Note: The two SORT instructions and the CRC and SHA-1 instructions
are still not implemented.
8. The COPY and COPY-LITERAL instructions now apply the
byte-copying-rules to the source address as well as the destination
address.
9. The COPY-LITERAL and COPY-OFFSET now correctly handle the
destination operand. Previously, it was possible for the destination
to be left at byte_copy_right when it should be left at
byte_copy_left.
10. All three COPY instructions will now behave correctly if the source
or destination start at byte_copy_right; i.e. they read/write to
byte_copy_right, byte_copy_right+1, etc. Previously the parameters
would wrap to byte_copy_left.
11. The COPY-OFFSET instruction now correctly calculates the source
address for all values of offset. Previously a very large offset
(greater than the circular buffer size) could cause the starting
position to be mis-calculated.
12. Modulo 65536 arithmetic has been added in some places where it was
missing.
[Only when noticed, we have not been able to check all the code.]
13. Some redundant code was removed for the INPUT-BITS instruction.
Also, INPUT-BITS (0, x, x) now correctly discards spare bits when
the P-bit, has changed.
14. Corrected the cycles used calculation for the INPUT-BITS
instruction.
15. Corrected some minor cases where the used_udvm_cycles was not
incremented.
[Not really a problem, but it allowed us to verify correctness with
some of the SigComp torture tests.]
16. Removed some redundant code in INPUT-HUFFMAN (and reorganised some
comments to keep them making sense).
Function decomp_dispatch_get_bits:
1. This has been rewritten. The original version could not correctly
handle requests for more than 8 bits. The new version is cleaner and
correctly handles all cases.
svn path=/trunk/; revision=14666
Evan Huus