The SSL_get_peer_certificate openSSL function increases the lock for X509
object it returns so X509 object retrieved using this function must be
released with X509_free after use.
This patch uses the Ssl::X509_Pointer TidyPointer to release X509 object
retrieved with the SSL_get_peer_certificate function inside the
Ssl::PeerConnector::handleNegotiateError method
Unexpected SQUID_X509_V_ERR_DOMAIN_MISMATCH errors while accessing sites with valid certificates
A "const char *" pointer retrieved using the SBuf::c_str() method may attached
to an SSL object using the SSL_set_ex_data method as server name used to
validate server certificates. This pointer may become invalid, causing
the SQUID_X509_V_ERR_DOMAIN_MISMATCH errors.
This patch changes the type of the ssl_ex_index_server index used with the
SSL_set_ex_data function to be an SBuf object.
Portability: Add hacks to define C++11 explicit N-bit type limits
Add cstdint and stdint.h to libcompat headers and ensure that type limits
used by Squid are always available. Mostly this involves shuffling
existing hacks into the compat headers but the UINT32_* limits are new.
Despite the "must match" comment, MAX_AUTHTOKEN_LEN in
auth/UserRequest.h got out of sync with similar constants in Negotiate helpers.
A 32KB buffer cannot fit some helper requests (e.g., those carrying Privilege
Account Certificate information in the client's Kerberos ticket). Each truncated
request blocks the negotiate helper channel, eventually causing helper queue
overflow and possibly killing Squid.
This patch increases MAX_AUTHTOKEN_LEN in UserRequest.h to 65535 which
is also the maximum used by the negotiate helpers. The patch also adds checks
to avoid sending truncated requests, treating them as helper errors instead.
The fix for Bug 3664 "ssl_crtd fails to build on OpenSolaris/OpenIndiana/Solaris 11"
introduced a regression on BSD and Linux where lockf() implementations appear not to
lock the entire file correctly or as reliably as flock().
Reverting the flock/lockf change for non-Solaris OS.
This patch adds code in squid to control SslBump behavior when dealing with
"resuming SSL/TLS sessions". Without these changes, SslBump usually terminates
all resuming sessions with an error because such sessions do not include
server certificates, preventing Squid from successfully validating the server
identity.
After these changes, Squid splices resuming sessions. Splicing is the right
because Squid most likely has spliced the original connections that the client
and server are trying to resume now.
Without SslBump, session resumption would just work, and SslBump behaviour
should approach that ideal.
Future projects may add ACL checks for allowing resuming sessions and may
add more complex algorithms, including maintaining an SMP-shared
cache of sessions that may be resumed in the future and evaluating
client/server attempts to resume a session using that cache.
This patch also makes SSL client Hello message parsing more robust and
adds an SSL server Hello message parser.
Also add support for NPN (next protocol negotiation) and ALPN (Application-Layer Protocol Negotiation) tls extensions, required to correctly bump web clients
support these extensions
Technical details
-----------------
In Peek mode, the old Squid code would forward the client Hello message to the
server. If the server tries to resume the previous (spliced) SSL session with
the client, then Squid SSL code gets an ssl/PeerConnector.cc "ccs received
early" error (or similar) because the Squid SSL object expects a server
certificate and does not know anything about the session being resumed.
With this patch, Squid detects session resumption attempts and splices
There are two mechanism in SSL/TLS for resuming sessions. The traditional
shared session IDs and the TLS ticket extensions:
* If Squid detects a shared ID in both client and server Hello messages, then
Squid decides whether the session is being resumed by comparing those client
and server shared IDs. If (and only if) the IDs are the same, then Squid
assumes that it is dealing with a resuming session (using session IDs).
* If Squid detects a TLS ticket in the client Hello message and TLS ticket
support in the server Hello message as well as a Change Cipher Spec or a New
TLS Ticket message (following the server Hello message), then (and only then)
Squid assumes that it is dealing with a resuming session (using TLS tickets).
The TLS tickets check is not performed if Squid detects a shared session ID
in both client and server Hello messages.
NPN and ALPN tls extensions
---------------------------
Even if squid has some SSL hello messages parsing code, we are relying to
openSSL for full parsing. The openSSL used in peek and splice mode to parse
server hello message, check for errors and verify server certificates.
If the openSSL, while parses the server hello message, find an extension enabled
in the server hello message, which is not enabled in its side, fails with an
error ("...parse tlsext...").
OpenSSL supports NPN tls extension and from 1.0.2 release supports also the
ALPN tls extensions. In peek mode we are forwading the client SSL hello message
as is, and if this message include support for NPN or ALPN tls extension is
possible that the SSL server support them and include related extensions
in its response. The openSSL will fail if support for these extensions is
not enabled in its side.
This patch handles the NPN (TLSEXT_TYPE_next_proto_neg) as follows:
Try to select the http/1.1 protocol from the server protocols list. If the
http/1.1 is not supported then the SSL bumping will fail. This is valid
because only http protocol we are supporting in squid.
Splicing is not affected.
Also add support for the ALPN TLS extension. This extension is a replacement
for the NPN extension. The client sends a list of supported protocols. In the
case of stare mode squid now sends only http as supported protocol. In the
case of server-first or client-first bumbing modes, squid does enable this
extension.
The NPN supported by chromium browser the ALPN supported by firefox.
Support for ALPN is added to openSSL 1.0.2 release.
These extensions are used to support SPDY and similar protocols.
Add server_name ACL matching server name(s) obtained from various sources such as CONNECT request URI, client SNI, and SSL server certificate CN.
During each SslBump step, Squid improves its understanding of a "true server
name", with a bias towards server-provided (and Squid-validated) information.
The server-provided server names are retrieved from the server certificate CN
and Subject Alternate Names. The new server_name ACL matches any of alternate
names and CN. If the CN or an alternate name is a wildcard, then the new ACL
matches any domain that matches the domain with the wildcard.
Other than supporting many sources of server name information (including
sources that may supply Squid with multiple server name variants and
wildcards), the new ACL is similar to dstdomain.
Fix HttpStateData::readReply to retry read from server in the case of EINPROGRESS, EAGAIN or similar errors
This bug mostly affects SSL bumped connections.
The HttpStateData::readReply will not retry read from server in the case of an
EINPROGRESS or similar comm errors and the connection will hang, until the
timeout handler called.
The Comm::ReadNow method calls ignoreErrno function to test if the comm error
should be ignored and in this case return Comm::INPROGRESS value.
In this case we need to set flags.do_next_read to true to force
HttpStateData::maybeReadVirginBody() method retry read.
Fix: An invalid request->clientConnectionManager object can be used inside Ssl::PeerConnector::handleNegotiateError method
This patch adds the Ssl::ServerBio::bumpMode() method to retrieve the configured
mode from a ServerBio object, and uses this method for checking the bumping
mode inside Ssl::PeerConnector::handleNegotiateError method
After a failed http_access acl check of an HTTP request, tunneled through a
SSL bumped connection, ssl bumping code try to re-setup the connection for a
client-first bumping mode to serve the error crashing squid.
Amos Jeffries [Sat, 28 Mar 2015 14:53:59 +0000 (07:53 -0700)]
Parser-NG: Convert the ICAP read buffer to an SBuf.
* Remove the double-buffering hack used to comm_read() ICAP responses as
c-string then convert to MemBuf for parsing.
* Revert the HttpMsg parser API from MemBuf to c-string parameters.
The internals did not make much use of the MemBuf abilities and it is
simpler to retrieve c-string values directly from an SBuf than to go
via a MemBuf conversion.
When squid generated an error page which contains the "%m" formating code
but the authentication information is not available squid dies with
segfault.
Amos Jeffries [Sat, 21 Mar 2015 08:25:19 +0000 (01:25 -0700)]
Crypto-NG: Move Ssl::PeerConnectorAnswer to Security::EncryptorAnswer
This class was not actually depending on OpenSSL API symbols and by
abstracting it out we can unify the callback handlers for encrypted and
non-encrypted logic paths for several classes that setup connections.
Amos Jeffries [Fri, 20 Mar 2015 15:10:07 +0000 (08:10 -0700)]
Move Ssl::PeerConnectorAnswer to Security::EncryptorAnswer
This class was not actually depending on OpenSSL API symbols and by
abstracting it out we can unify the callback handlers for encrypted and
non-encrypted logic paths for several classes that setup connections.
SMP workers in trunk start without root privileges. This results in startup
failures when workers need to use a privileged port (e.g., 443) or other
root-only features such as TPROXY.
The watch_child function, responsible to watch and start squid workers for
the squid monitor process, called after a enter_suid() call, but the
writePidFile() call, inside the watch_child(), will leave suid mode before exit.
This patch add enter_suid() cals after the writePidFile and removePidFile()
inside the watch_child() function.
Amos Jeffries [Sun, 15 Mar 2015 18:13:19 +0000 (11:13 -0700)]
Cleanup: extend SBuf debugging information
It can be hard determining what simple operations (ie cow(), grow()) are
being done no what SBuf object. Add the SBuf::id to debugs() output on
many more operations.
Amos Jeffries [Fri, 13 Mar 2015 11:26:27 +0000 (04:26 -0700)]
Portability: check 64-bit GNU atomic operators are useable
Sometimes (namely 32-bit OpenBSD libstdc++) do not fully implement the
GNU atomic operators for both 32-bit and 64-bit. But Squid makes use of
both types if the compiler deems them required.
We need to check them all before declaring the atomics usable, or not.
Thanks to Stuart Henderson for identifying the issue.
Amos Jeffries [Thu, 12 Mar 2015 01:30:21 +0000 (18:30 -0700)]
Tests: extend pre-compiler unit tests to check macro permutations
We are getting come complaints about precompiler issues building with
#if FOO && FOO || FOO boolean constructs using undefined macros.
The particualr form reported so far dies when building the test, so will
be found earlier than these checks are run. This is to ensure its not a
widespread subtle error in other "working" installations.
Amos Jeffries [Sun, 1 Mar 2015 01:44:26 +0000 (17:44 -0800)]
Cleanup: convert BodyPipe to MEMPROXY_CLASS
BodyPipe is ref-counted. Such classes should not be using CBDATA_CLASS
for smart pointer referencing and memory management. They use RefCount
for smart pointer referencing, MEMPROXY_CLASS for memory management.
Amos Jeffries [Thu, 26 Feb 2015 13:19:35 +0000 (05:19 -0800)]
Parser-NG: HTTP Response Parser upgrade
1) convert the HTTP server read buffer to an SBuf using the same design
and Comm::Read API implemented previously for the client connections.
The buffer remains default initialized at 16KB per connection but is no
longer absolutely limited to 256KB. Instead it is limited by
configuration options controlling maximum server input sizes on
read_ahead_gap and response message headers.
The Client API has been extended with a new method to estimate size
requirements of an SBuf I/O buffer. Modelled on and deprecating the
existing MemBuf estimator.
The Comm::ReadNow() API is extended to allow limited-size read(2)
operations by setting the CommIoCbParams::size parameter.
The HttpStateData buffer is partially detached from
StoreEntry::delayAwareRead() API due to requirements of the
Comm::ReadNow() API. Instead StoreEntry::bytesWanted() is used directly
to determine read(2) size, and DeferredRead are generated only when
ReadNow() is actually and immediately to be deferred. Theoretically this
means less read operations get deferred in some high load cases.
Practically it means there is no longer an AsyncCall queue plus socket
wait delay between delay_pools promising a read size, doing the
read(2), and accounting for the bytes received - accuracy should be much
improved under load.
This introduces one temporary performance regression converting the SBuf
content to MemBuf for chunked decoder to process.
2) add Http1::ResponseParser class for parsing HTTP response messages.
Modelled on the same design as used for the HTTP RequestParser, and
inheriting from a mutual parent Http1::Parser.
The Parser is Tokeniser based, incremental and 'consumes' bytes out of
the buffer as they are parsed.
The Tokenizer int64 API is updated to handle limited-length scans and
optional +/- symbols.
This Parser class recognises HTTP/1.x and ICY/1 syntax messages. Any
unknown syntax input is assumed to be HTTP "0.9" and it will
gateway/transform the response to HTTP/1.1.
NOTE: these are all semantic actions performed by the code being
replaced in (3). Only the form and OO scoping has changed.
The mime header block detection operation is generalized into the
Http1::Parser for use by both RequestParser and ResponseParser. The
request_parse_status error code has also been adapted for shared use.
3) integrate the HTTP1::ResponseParser with HttpStateData server
response processing.
This is largely code shuffling. Though I have extended the EOF \r\n hack
such that it enables Squid to parse truncated response headers now.
Amos Jeffries [Thu, 26 Feb 2015 10:37:41 +0000 (02:37 -0800)]
Bug 2741 (partial): Initial libsecurity API
The first step(s) towards a generic TLS/SSL security API for Squid.
Creates the basic security/libsecurity.la library and Security::
namespace infrastructure. Symbols provided by this API are always
available instead of conditionally compiled (unlike the ssl/* code for
OpenSSL use).
Merge the TLS/SSL context parameters into a Security::PeerOptions
object instead of maintaining multiple member variables in the
CachePeer and SquidConfig objects.
Squid now provides an error if SSL-specific squid.conf parameters are
used for a Squid without OpenSSL support, instead of silently ignoring
them.
Amos Jeffries [Thu, 26 Feb 2015 06:05:02 +0000 (22:05 -0800)]
Bug 2907: high CPU usage on CONNECT when using delay pools
When delay pools are active on a CONNECT tunnel and the pool is drained
the I/O loop cycles very often transferring 1 byte until the pool is
topped-up at the end of the second.
Instead of looping constantly trying to read 1 byte at a time, add an
asynchronous event to wait for a few I/O cycles or until more bytes can
be read.
To protect against infinite loops of waiting when many tunnels are
competing for the pool allowance we only delay for a limited number of
loops before allowing at least 1 byte through. Also, the amount of time
waited is an odd fraction of 1 second so re-tries naturally spread
across any given second fairly, with connections rotating closer or
further from the time when pool topup happens. That behaviour still
allows some variance in service times, but overall the CPU consumption
and (as a result) total proxy speed appears to be much improved.
NP: Initial production testing shows a 36% RPS speed increase,
with a 50% reduction in total CPU usage.
Squid closes the SSL client connection with "Failed to start fake CONNECT
request for ssl spliced connection". This happens especially often when
the pipeline_prefetch configuration parameter is set to "0" (i.e., default).
When a transparent SSL connection is peeked and then spliced in step2, we are
generating a fake CONNECT request. The fake CONNECT request is counted as a
new pipelined request and may exceed the configured limit. This patch solves
this problem by raising the limit for that request.
Needs more work to better identify the requests that need a different limit.
Joshua Root [Wed, 25 Feb 2015 13:32:14 +0000 (14:32 +0100)]
Bug 3805: support shared memory on MacOS X in Mem::IPC::Segment
MacOS X doesn't support the O_TRUNC flag to shm_open; it is redundant anyway
because the shared memory segment is truncated immediately after opening
as per best practices. With this support Squid can now be built and run
under MacOS X.