Alexander Hansen Færøy authored 7 years ago and Nick Mathewson committed 7 years ago

This patch adds support for enabling support for LZMA to our configure
script. By default, the --enable-lzma option is set to "auto" which
means if liblzma is available we'll build Tor with LZMA support.

See: https://bugs.torproject.org/21662

Alexander Hansen Færøy authored 7 years ago and Nick Mathewson committed 7 years ago

This patch changes the way `tor_compress_new()`,
`tor_compress_process()`, and `tor_compress_free()` handles different
compression methods. This should give us compiler warnings in case an
additional compression method is added, but the developer forgets to add
handlers in the three aforementioned functions.

See https://bugs.torproject.org/21663

Alexander Hansen Færøy authored 7 years ago and Nick Mathewson committed 7 years ago

See https://bugs.torproject.org/21663

Alexander Hansen Færøy authored 7 years ago and Nick Mathewson committed 7 years ago

See https://bugs.torproject.org/21663

Alexander Hansen Færøy authored 7 years ago and Nick Mathewson committed 7 years ago

This patch refactors the `torgzip` module to allow us to extend a common
compression API to support multiple compression backends.

Additionally we move the gzip/zlib code into its own module under the
name `compress_zlib`.

See https://bugs.torproject.org/21664

This patch exposes the old `is_compression_bomb()` function as a public
API as part of the compression module.

See https://bugs.torproject.org/21663

This patch makes the internal `get_memlevel()` a part of the public
compression API as `tor_compress_memory_level()`.

See https://bugs.torproject.org/21663

See https://bugs.torproject.org/21663

See https://bugs.torproject.org/21663

This patch renames the `zlib_state` field in `dir_connection_t` to
`compress_state`.

See https://bugs.torproject.org/21663

This patch refactors our streaming compression code to allow us to
extend it with non-zlib/non-gzip based compression schemas.

See https://bugs.torproject.org/21663

To allow us to use the API name `tor_compress` and `tor_uncompress` as
the main entry-point for all compression/uncompression and not just gzip
and zlib.

See https://bugs.torproject.org/21663

See https://bugs.torproject.org/21663

This patch removes the unused `is_gzip_supported()` and changes the
documentation string around the `compress_method_t` enumeration to
explicitly state that both `ZLIB_METHOD` and `GZIP_METHOD` are both
always supported.

Zlib version 1.2.0 was released on the 9'th of March, 2003 according to
their ChangeLog.

See https://bugs.torproject.org/21663

See https://bugs.torproject.org/21663

This patch changes some of the tt_assert() usage in test_util_gzip() to
use tt_int_op() to get better error messages upon failure.

Additionally we move to use explicit NULL checks.

See https://bugs.torproject.org/21663

David Goulet authored 8 years ago and Nick Mathewson committed 7 years ago

The reason for making the temporary list public is to keep it encapsulated in
the rendservice subsystem so the prop224 code does not have direct access to
it and can only affect it through the rendservice pruning function.

It also has been modified to not take list as arguments but rather use the
global lists (main and temporary ones) because prop224 code will call it to
actually prune the rendservice's lists. The function does the needed rotation
of pointers between those lists and then prune if needed.

In order to make the unit test work and not completely horrible, there is a
"impl_" version of the function that doesn't free memory, it simply moves
pointers around. It is directly used in the unit test and two setter functions
for those lists' pointer have been added only for unit test.

Signed-off-by: David Goulet <dgoulet@torproject.org>

George Kadianakis authored 8 years ago and Nick Mathewson committed 7 years ago

Now we have separate getters and setters for service-side and relay-side. I
took this approach over adding arguments to the already existing methods to
have more explicit type-checking, and also because some functions would grow
too large and dirty.

This commit also fixes every callsite to use the new function names which
modifies the legacy HS (v2) and the prop224 (v3) code.

Signed-off-by: David Goulet <dgoulet@torproject.org>

George Kadianakis authored 8 years ago and Nick Mathewson committed 7 years ago

This test is identical to the ./src/test/test_ntor.sh integration test.

George Kadianakis authored 8 years ago and Nick Mathewson committed 7 years ago

The test checks that introduce1/rendezvous1 key material is generated
correctly both for client-side and service-side.

George Kadianakis authored 8 years ago and Nick Mathewson committed 7 years ago

and also does the key expansion.

Signed-off-by: David Goulet <dgoulet@torproject.org>

One of the goals of this change is to have trunnel API/ABI being more explicit
so we namespace them with "trn_*". Furthermore, we can now create
hs_cells.[ch] without having to confuse it with trunnel which used to be
"hs_cell_*" before that change.

Here are the perl line that were used for this rename:

  perl -i -pe 's/cell_extension/trn_cell_extension/g;' src/*/*.[ch]
  perl -i -pe 's/cell_extension/trn_cell_extension/g;' src/trunnel/hs/*.trunnel
  perl -i -pe 's/hs_cell_/trn_cell_/g;' src/*/*.[ch]
  perl -i -pe 's/hs_cell_/trn_cell_/g;' src/trunnel/hs/*.trunnel

  And then "./scripts/codegen/run_trunnel.sh" with trunnel commit id
  613fb1b98e58504e2b84ef56b1602b6380629043.

Fixes #21919

Signed-off-by: David Goulet <dgoulet@torproject.org>

Pinning EntryNodes along with hidden services can be possibly harmful (for
instance #14917 and #21155) so at the very least warn the operator if this is
the case.

Fixes #21155

Signed-off-by: David Goulet <dgoulet@torproject.org>

These tests tried to use ridiculously large buffer sizes to check
the sanity-checking in the code; but since the sanity-checking
changed, these need to change too.

Now that base64_decode() checks the destination buffer length against
the actual number of bytes as they're produced, shared_random.c no
longer needs the "SR_COMMIT_LEN+2" workaround.

Remove base64_decode_nopad() because it is redundant now that
base64_decode() correctly handles both padded and unpadded base64
encodings with "right-sized" output buffers.

Test base64_decode() with odd sized decoded lengths, including
unpadded encodings and padded encodings with "right-sized" output
buffers.  Convert calls to base64_decode_nopad() to base64_decode()
because base64_decode_nopad() is redundant.

base64_decode() was applying an overly conservative check on the
output buffer length that could incorrectly produce an error if the
input encoding contained padding or newlines.  Fix this by checking
the output buffer length against the actual decoded length produced
during decoding.