Fix assertion on bodyPipe != NULL inside ConnStateData::handleFtpRequestData()
ftp-native branch hits this assertion when ICAP server respond with error on
a ftp request.
This patch closes the data connection on these cases and also add a check in
ConnStateData::clientReadFtpData to avoid calling handleFtpRequestData when
the bodyPipe is gone for a reason.
Amos Jeffries [Mon, 2 Jun 2014 11:49:56 +0000 (04:49 -0700)]
Cleanup: rewrite Http::One::Parser::getHeaderField() using Tokenizer
Fixes performance regression in SBuf usage.
Also, fixes one nasty bug where it would return the value of a line
containign the named header despite obs-fold and quoting. This was
particualarly bad as this method is used primarily to retrieve Host:
header before full aprsing of the mime block takes place.
The bug causes "foo.invalid" to be detected as hostname in:
GET / HTTP/1.1
Foo: bar="\r Host: fake.invalid\n "
Host: example.com
Alex Rousskov [Mon, 2 Jun 2014 05:26:17 +0000 (22:26 -0700)]
Do not leak ex_data for SSL state that survived reconfigure.
SSL_get_ex_new_index() allocates a new index on every call, even if its
parameters remain unchanged. It should be called once per process
lifetime.
Besides leaking, this 12 year-old(!) bug could probably make some SSL
code misbehave during reconfigure because reconfigure would change the
supposedly constant ex_data indexes.
Alex Rousskov [Mon, 2 Jun 2014 05:16:35 +0000 (22:16 -0700)]
Do not register the same Cache Manager action more than once
... to avoid wrong mgr:menu output and the impression of a reconfigure
memory leak.
The old code was comparing action object pointers, which could not work,
and was adding the same action on every reconfigure call for modules that
register with Cache Manager during [re]configuration.
We already have a working method for finding registered actions. Use it.
Amos Jeffries [Sun, 1 Jun 2014 11:53:02 +0000 (04:53 -0700)]
Update HTTP-version parsing with RFC 7230 octet magics
RFC 7230 replaces RFC 2616 and defines HTTP-version for HTTP/1 protocol
as having exact case-sensitive octets "HTTP/1." and a variable minor
version consisting of exactly one DIGIT.
This allows us to use magic-octet matching to detect the HTTP-version
field and remove slow matching logics for unknown version and HTTP major
version number (DIGIT '1').
Amos Jeffries [Sat, 31 May 2014 17:00:05 +0000 (10:00 -0700)]
Cleanup: de-duplicate auth_param program parameter code
Moves the "program" parse and dump code into Auth::Config.
Also, changes API to Auth::Config::dump() to not dump any config settings
for schemes which are not configured with a "program". Including scheme
specific settings.
Also, fixes missing Digest "utf8" parameter in config dump.
Move realm parse and config dump logics to Auth::Config base object.
This de-duplicates Basic, Digest (and future schemes ie Bearer) config
processing code. Also makes realm available to NTLM and Negotiate
schemes, although at present it remains unused by those schemes.
Also, convert the realm parameter string to an SBuf. Removing the need
for some memory maintenance code.
Alex Rousskov [Fri, 23 May 2014 06:34:59 +0000 (00:34 -0600)]
Revised ftp-gw timeout handling to cope with very long data downloads/uploads
that triggered bogus ctrl connection timeouts due to ctrl channel inactivity.
Unset ctrl timeout when we are done waiting for the ctrl response
(ServerStateData::readControlReply).
Removed code setting data timeout from Ftp::ServerStateData::dataRead()
because the data timeout is set in ServerStateData::maybeReadVirginBody() that
dataRead calls to read data.
Removed switchTimeoutToDataChannel() from
Ftp::Gateway::ServerStateData::startDataDownload because
* ctrl timeout should be cleared when we are done waiting for the ctrl
response (ServerStateData::readControlReply) and
* data timeout should be set when we start waiting for the data
(ServerStateData::maybeReadVirginBody)
- Defines three steps of the SSL bumping processing:
step1: Get TCP-level and CONNECT info. Evaluate ssl_bump and perform
the first matching action (splice, bump, peek, stare, terminate,
or err)
step2: Get SSL Client Hello info. Evaluate ssl_bump and perform the
first matching action (splice, bump, peek, stare, terminate,
or err). Peeking usually prevents future bumping. Staring
usually prevents future splicing.
step3: Get SSL Server Hello info. Evaluate ssl_bump and perform the
first matching action (splice, bump, terminate, or err).
In most cases, the only remaining choice at this step is
whether to terminate the connection. The splicing or bumping
decision is usually dictated by either peeking or staring at the
previous step.
- The ssl_bump ACLs list may evaluated in all SSL Bumping processing steps to
take a decision for the next step:
splice or none: Become a TCP tunnel without decoding the connection.
bump: Establish a secure connection with the server and, using a
mimicked server certificate, with the client
peek: Receive client (step1) or server (step2) certificate while
preserving the possibility of splicing the connection. Peeking
at the server certificate usually precludes future bumping of
the connection.
stare: Receive client (step1) or server (step2) certificate while
preserving the possibility of bumping the connection. Staring at
the server certificate usually precludes future splicing of the
connection.
terminate or err: Close client and server connections.
All actions except peek and stare correspond to final decisions: Once an
ssl_bump directive with a final action matches, no further ssl_bump
evaluations will take place, regardless of the current processing step.
- Add the atstep acl to match against SSL bumping step: "step1", "step2" or
"step3"
Current Implementation details:
---------------------------------
1) If the "peek" mode selected in step2 then the client hello message
forwarded to server. If this mode selected in step2 the splice is always
possible and bump maybe is not possible (in most cases where the client uses
different SSL client library implementation)
2) If the "stare" mode selected in step2 then the squid builds a new
hello message, which try to mimic, if it is possible , client hello message.
If stare selected in step2 the bump is always possible, but splice maybe is
not possible any more.
3) In step3 if bump decided, and bump is not possible any more then squid
is always splicing.
4) In step3 if splice decided but splice is not possible any more then
squid is always bumping.
5) Because of (3) and (4), in practice, if firefox browser used with
peek mode, squid always splice the connection, because squid/openSSL
does not support the firefox SSL features reported in client hello message.
6) In step2 if ACL list evaluation result to terminate or err then we just
close client connection. If the check result to ssl-bump then just bump.
If check result to client-first, server-first, then bump the connection
else do peek/stare.
7) In step3 the ssl_bump ACL list evakuation result client-first, server-first,
bump or peek result to bumping (if bumping is possible).
Amos Jeffries [Thu, 22 May 2014 06:04:05 +0000 (23:04 -0700)]
Cleanup: drop Auth::User::proxy_auth_list header cache
This list/cache was originally used to short-circuit auth helper lookups
based on previousy seen [Proxy-]Authorization header strings.
However, that permitted replay attacks in most auth schemes and has been
replaced by scheme-specific mechanisms:
* Basic and Digest credentials are cached in the global user name cache
wih additional nonce/password comparisons to verify located entries.
* NTLM and Negotiate credentials are cached in the ConnStateData with
exact-match comparison done to verify tokens.
After r13324 patch the SBuf argument of the ConnStateData::handleReadData member
is used only to check if ConnStateData::In::buf is correctly filled with read
data. ConnStateData::handleReadData considers that the data already written
in ConnStateData::in.buf and checks if the passed Sbuf argument is the
ConnStateData::in.buf:
The httpsSslBumpAccessCheckDone function needs to write the CONNECT request
generated internally to force tunnel mode, in ConnStateData::In::buf and then
call ConnStateData::handleReadData method.
Amos Jeffries [Tue, 20 May 2014 11:00:04 +0000 (04:00 -0700)]
Cleanup: drop parsedCount_ tracking
Now that parse() is receiving a buffer directly we no longer have to
track how many bytes have been consumed by the parse. It can be
calculated by comparing the current and original SBuf.
Amos Jeffries [Sun, 18 May 2014 10:36:05 +0000 (03:36 -0700)]
Fix infinite parse loop on partial request reads
parseHttpRequest() returns NULL on incomplete parse. This case was not
exiting the loop to parse multiple requests. As a result traffic would
only receive a response if the request headers were received entirely
within one read(2) event. Pipelined requests received over multiple hung.
Amos Jeffries [Thu, 15 May 2014 10:44:05 +0000 (03:44 -0700)]
Fix outstanding build issues and parser audit results
* Give SBuf I/O buffer directly to Http1::RequestParser
* Redesign parser state engine to represent the current state
being parsed instead of previous completed. This allows much
more incremental resume of a parse and reliable consume() of
the input buffer as sections complete instead of complex byte
accounting outide the parser.
* Maintain an internal counter of bytes parsed and consumed by
the parser instead of a buffer offset. This allows much more
reliable positioning of the state/section boundaries.
* Remove erroneous fprintf debug left in previous commit.
* Redesign HttpRequestMethod constructor to drop end parameter.
* Redesign all parser unit tests. Marking RFC non-compliance
for future fixing.
author: Alex Rousskov <rousskov@measurement-factory.com>
Avoid on-exit crashes when adaptation is enabled.
After trunk r13269 (Vector refactor) destroyed vector objects still have
positive item counts. This exposes use-after-delete bugs. In this particular
case, global adaptation rule/group/service arrays are destructed by global
destruction sequence first and then again by Adaptation::*::TheConfig objects
destructors.
This change avoiding static destruction order dependencies by storing those
global adaptation arrays on heap.
Alex Rousskov [Thu, 8 May 2014 22:43:01 +0000 (16:43 -0600)]
Temporary fix for segmentation faults in FwdState::serverClosed.
r13388 (cache_peer standby=N) moved noteUses() call from Comm to FwdState, to
avoid exposing Comm to pconn pools. Unfortunately, the closing handler does
not get a valid FD value when the closing callback shares the Connection
object with the code that called conn->close(). It gets -1. The FD of the
FwdState connection itself is already -1 at that point, for similar reasons.
The code thinks it got a matching FD and calls noteUses() with an invalid FD.
This temporary workaround prevents noteUses() calls when FD is unknown.
Without those calls, pconn usage statistics will be wrong. A different
long-term solution is needed.