From Hanno:
When a server replies to a cookieless ClientHello with a HelloVerifyRequest,
it is supposed to reset the connection and wait for a subsequent ClientHello
which includes the cookie from the HelloVerifyRequest.
In testing environments, it might happen that the reset of the server
takes longer than for the client to replying to the HelloVerifyRequest
with the ClientHello+Cookie. In this case, the ClientHello gets lost
and the client will need retransmit. This may happen even if the underlying
datagram transport is reliable.
Previous commits introduced the field `total_bytes_buffered`
which is supposed to keep track of the cumulative size of
all heap allocated buffers used for the purpose of reassembly
and/or buffering of future messages.
However, the buffering of future epoch records were not reflected
in this field so far. This commit changes this, adding the length
of a future epoch record to `total_bytes_buffered` when it's buffered,
and subtracting it when it's freed.
This commit adds a static function ssl_buffer_make_space() which
takes a buffer size as an argument and attempts to free as many
future message bufffers as necessary to ensure that the desired
amount of buffering space is available without violating the
total buffering limit set by MBEDTLS_SSL_DTLS_MAX_BUFFERING.
This commit continues commit 47db877 by removing resend guards in the
ssl-opt.sh tests 'DTLS fragmenting: proxy MTU, XXX' which sometimes made
the tests fail in case the log showed a resend from the client.
See 47db877 for more information.
Previously, the UDP proxy could only remember one delayed message
for future transmission; if two messages were delayed in succession,
without another one being normally forwarded in between,
the message that got delayed first would be dropped.
This commit enhances the UDP proxy to allow to delay an arbitrary
(compile-time fixed) number of messages in succession.
If the next expected handshake message can't be reassembled because
buffered future messages have already used up too much of the available
space for buffering, free those future message buffers in order to
make space for the reassembly, starting with the handshake message
that's farthest in the future.
This commit adds a static function ssl_buffering_free_slot()
which allows to free a particular structure used to buffer
and/or reassembly some handshake message.
This commit introduces a compile time constant MBEDTLS_SSL_DTLS_MAX_BUFFERING
to mbedtls/config.h which allows the user to control the cumulative size of
all heap buffer allocated for the purpose of reassembling and buffering
handshake messages.
It is put to use by introducing a new field `total_bytes_buffered` to
the buffering substructure of `mbedtls_ssl_handshake_params` that keeps
track of the total size of heap allocated buffers for the purpose of
reassembly and buffering at any time. It is increased whenever a handshake
message is buffered or prepared for reassembly, and decreased when a
buffered or fully reassembled message is copied into the input buffer
and passed to the handshake logic layer.
This commit does not yet include future epoch record buffering into
account; this will be done in a subsequent commit.
Also, it is now conceivable that the reassembly of the next expected
handshake message fails because too much buffering space has already
been used up for future messages. This case currently leads to an
error, but instead, the stack should get rid of buffered messages
to be able to buffer the next one. This will need to be implemented
in one of the next commits.
A previous commit introduced the function ssl_prepare_reassembly_buffer()
which took a message length and a boolean flag indicating if a reassembly
bit map was needed, and attempted to heap-allocate a buffer of sufficient
size to hold both the message, its header, and potentially the reassembly
bitmap.
A subsequent commit is going to introduce a limit on the amount of heap
allocations allowed for the purpose of buffering, and this change will
need to know the reassembly buffer size before attempting the allocation.
To this end, this commit changes ssl_prepare_reassembly_buffer() into
ssl_get_reassembly_buffer_size() which solely computes the reassembly
buffer size, and performing the heap allocation manually in
ssl_buffer_message().
When a server replies to a cookieless ClientHello with a HelloVerifyRequest,
it is supposed to reset the connection and wait for a subsequent ClientHello
which includes the cookie from the HelloVerifyRequest.
In testing environments, it might happen that the reset of the server
takes longer than for the client to replying to the HelloVerifyRequest
with the ClientHello+Cookie. In this case, the ClientHello gets lost
and the client will need retransmit. This may happen even if the underlying
datagram transport is reliable.
This commit removes a guard in the ssl-opt.sh test
'DTLS fragmenting: proxy MTU, resumed handshake' which made
the test fail in case the log showed a resend from the client.
This commit moves the length and content check for CCS messages to
the function mbedtls_ssl_handle_message_type() which is called after
a record has been deprotected.
Previously, these checks were performed in the function
mbedtls_ssl_parse_change_cipher_spec(); however, now that
the arrival of out-of-order CCS messages is remembered
as a boolean flag, the check also has to happen when this
flag is set. Moving the length and content check to
mbedtls_ssl_handle_message_type() allows to treat both
checks uniformly.
We previously observed random-looking failures from this test. I think they
were caused by a race condition where the client tries to reconnect while the
server is still closing the connection and has not yet returned to an
accepting state. In that case, the server would fail to see and reply to the
ClientHello, and the client would have to resend it.
I believe logs of failing runs are compatible with this interpretation:
- the proxy logs show the new ClientHello and the server's closing Alert are
sent the same millisecond.
- the client logs show the server's closing Alert is received after the new
handshake has been started (discarding message from wrong epoch).
The attempted fix is for the client to wait a bit before reconnecting, which
should vastly enhance the probability of the server reaching its accepting
state before the client tries to reconnect. The value of 1 second is arbitrary
but should be more than enough even on loaded machines.
The test was run locally 100 times in a row on a slightly loaded machine (an
instance of all.sh running in parallel) without any failure after this fix.
Depends on the current transform, which might change when retransmitting a
flight containing a Finished message, so compute it only after the transform
is swapped.
Use the same values as other 3d tests: this makes the test hopefully a bit
faster than the default values, while not increasing the failure rate.
While at it:
- adjust "needs_more_time" setting for 3d interop tests (we can't set the
timeout values for other implementations, so the test might be slow)
- fix some supposedly DTLS 1.0 test that were using dtls1_2 on the command
line
