We hit this assertion with bug 32868, but I'm stymied figuring out
how we wound up with a routerstatus like this. This patch is a
diagnostic to attempt to figure out what is going on, and to avoid a
crash in the meantime.
When looking up an entry in the rend_cache, stop asserting that it exists but
rather confirm it exists and if not, return that no entry was found.
The reason for that is because the hs_circ_cleanup_on_free() function (which
can end up looking at the rend_cache) can be called from the
circuit_free_all() function that is called _after_ the rend cache is cleaned
up in tor_free_all().
We could fix the free all ordering but then it will just hide a future bug.
Instead, handle a missing rend_cache as a valid use case as in while we are in
the cleanup process.
As Tor becomes more modular, it is getting more and more difficult to ensure
subsystem callstack ordering thus this fix aims at making the HSv2 subsystem
more robust at being called while tor is pretty much in any kind of state.
Fixes#32847.
Signed-off-by: David Goulet <dgoulet@torproject.org>
This is an automated commit, generated by this command:
./scripts/maint/rename_c_identifier.py \
EXPOSE_CLEAN_BACKTRACE BACKTRACE_PRIVATE \
TOR_CHANNEL_INTERNAL_ CHANNEL_OBJECT_PRIVATE \
CHANNEL_PRIVATE_ CHANNEL_FILE_PRIVATE \
EXPOSE_ROUTERDESC_TOKEN_TABLE ROUTERDESC_TOKEN_TABLE_PRIVATE \
SCHEDULER_PRIVATE_ SCHEDULER_PRIVATE
Some ".c" files define *_PRIVATE macros, but those macros are
not used in any header file. Delete them.
These changes were created using the "make autostyle" from
32522, and then split into commits.
I've chosen the "AuthDirMaxServersPerAddr" option here for
simplicity, since it is used literally nowhere else besides the dirauth
module. Once we have all the infrastructure in place for this, we
can move more options into this structure.
These modules are only built when the selected modules are disabled.
The provide stub implementations of the subsystem blocks. Later,
other stub implementations could move here.
Having real subsystem blocks here will let us handle disabled
configuration options better.
With v3, the "pending_final_cpath" of a circuit is always NULL which means
that for v3, established client rendezvous circuit waiting for the intro point
to ACK, will always end up timing out quickly.
This can increase the delays to which you connect to a service since in order
to succeed, the rendezvous circuit needs to fully established
(CIRCUIT_PURPOSE_C_REND_JOINED) within the cutoff of the introduction circuit
as well which is these days around 2-3 seconds.
Fixes#32021
Signed-off-by: David Goulet <dgoulet@torproject.org>
Simplify handle_control_getinfo() by using the new reply lines
abstraction. Previously, this function explicitly checked for whether
it should generate a MidReplyLine, a DataReplyLine, or an
EndReplyLine. control_write_reply_lines() now abstracts this check.
Part of #30984.
In handle_control_getconf(), use the new control reply line
abstraction to simplify output generation. Previously, this function
explicitly checked for whether it should generate a MidReplyLine or an
EndReplyLine. control_write_reply_lines() now abstracts this check.
Part of #30984.
All of these files contain "*.h", except for:
* src/app/config/.may_include
* src/test/.may_include
which also contain "*.inc".
This change prevents includes of "*.c" files, and other
unusually named files.
Part of 32609.
- Remove key_dir which is useless.
- Kill an indentation layer.
We want to make it cleaner and slimmer so that we can reuse parts of it in the
REMOVE command for removing the right client auth file.
Because the function that parses client auth credentials saved on
disk (parse_auth_file_content()) is not future compatible, there is no way to
add support for storing the nickname on the disk. Hence, nicknames cannot
persist after Tor restart making them pretty much useless.
In the future we can introduce nicknames by adding a new file format for client
auth credentials, but this was not deemed worth doing at this stage.
- See hs_client_register_auth_credentials() for the entry point.
- Also set the permanent flag for credentials we read from the filesystem.
- Also add some missing documentation.
Since the removal of ip->circuit_established, this function does litterally
nothing so clean it up.
Part of #32020
Signed-off-by: David Goulet <dgoulet@torproject.org>
By centralizing the circuit cleanup type that is: on close, free and
repurpose, some actions on the circuit can not happen for a certain cleanup
type or for all types.
This passes a cleanup type so the HS subsystem (v2 and v3) can take actions
based on the type of cleanup.
For instance, there is slow code that we do not run on a circuit close but
rather only on free.
Part of #32020
Signed-off-by: David Goulet <dgoulet@torproject.org>
Report back to the v3 subsystem any introduction point client circuit failure
so they can be noted down in the failure cache.
Fixes#32020
Signed-off-by: David Goulet <dgoulet@torproject.org>
Old and messy code path. Structure it in a more pleasant and readable way. No
behavior change with this refactor.
Part of #32020
Signed-off-by: David Goulet <dgoulet@torproject.org>
The FP_ identifiers referred to fingerprints, but they also applied
to address ranges. The router_status_t name invited confusion with
routerstasus_t. Fixes ticket 29826.
This is an automated commit, generated by this command:
./scripts/maint/rename_c_identifier.py \
router_status_t rtr_flags_t \
FP_INVALID RTR_INVALID \
FP_BADEXIT RTR_BADEXIT \
FP_REJECT RTR_REJECT
Refactor to decomplexify circuit_about_to_free() and finally have one single
entry point into the HS subsystems (v2 and v3) for when a circuit is freed.
With this, hs_circ_cleanup() becomes the one and only entry point when a
circuit is freed which then routes to the right subsystem version for any
actions to be taken.
This moves a big chunk of code from circuituse.c to rendclient.c. No behavior
change. Next commit will refactor it to reduce our technical debt.
Part of #32020
Signed-off-by: David Goulet <dgoulet@torproject.org>
We now keep descriptor that we can't decode due to missing client
authorization in the cache.
This new function is used when new client authorization are added and to tell
the client cache to retry decoding.
Part of #30382
Signed-off-by: David Goulet <dgoulet@torproject.org>
This commit extract most of the code that dirclient.c had to handle the end of
a descriptor directory requests (fetch). It is moved into hs_client.c in order
to have one single point of entry and the rest is fully handled by the HS
subsystem.
As part of #30382, depending on how the descriptor ended up stored (decoded or
not), different SOCKS error code can be returned.
Signed-off-by: David Goulet <dgoulet@torproject.org>
This will allow us to callback into the HS subsytem depending on the decoding
status and return an extended SOCKS5 error code depending on the decoding
issue.
This is how we'll be able to tell the SocksPort connection if we are missing
or have bad client authorization for a service.
Part of #30382
Signed-off-by: David Goulet <dgoulet@torproject.org>
We now keep the descriptor in the cache, obviously not decoded, if it can't be
decrypted for which we believe client authorization is missing or unusable
(bad).
This way, it can be used later once the client authorization are added or
updated.
Part of #30382
Signed-off-by: David Goulet <dgoulet@torproject.org>
Using a standard ending here will let other tools that expect
markdown understand our output here.
This commit was automatically generated with:
for fn in $(find src -name '*.dox'); do \
git mv "$fn" "${fn%.dox}.md"; \
done
