Hi Christos,

I’ve tried your patch (I had to go back to nightly r14487 for the patch to 
apply cleanly so this may have something to do with my findings)

http_port / CONNECT proxy works correctly with peek/splice decision making and 
all the correct information being available - of note is the %>ru field is now 
represented as FQDN:443 instead of just the FQDN (I don’t know if this is 
intentional or expected, but may be misleading).

https_port/transparent on the other hand immediately crashes (see log/bt below) 
- although this may be related to running on the older nightly. I’ll try again 
if you have a clean patch for latest nightly and/or it’s being committed to 
trunk.

2016/01/29 15:21:48| assertion failed: ../src/base/Lock.h:48: "count_ > 0"

#0  0x00007ffff53da5d7 in raise () from /lib64/libc.so.6
#1  0x00007ffff53dbcc8 in abort () from /lib64/libc.so.6
#2  0x00000000005858cf in xassert (msg=<optimized out>, file=<optimized out>, 
line=<optimized out>) at debug.cc:556
#3  0x000000000050d054 in unlock (this=0x2059d40) at ../src/base/Lock.h:48
#4  AccessLogEntry::~AccessLogEntry (this=0x205a2c0, __in_chrg=<optimized out>, 
__vtt_parm=<optimized out>) at AccessLogEntry.cc:82
#5  0x000000000050d0f9 in AccessLogEntry::~AccessLogEntry (this=0x205a2c0, 
__in_chrg=<optimized out>, __vtt_parm=<optimized out>) at AccessLogEntry.cc:87
#6  0x00000000006c1434 in dereference (newP=0x0, this=0x205a078) at 
../../src/base/RefCount.h:97
#7  ~RefCount (this=0x205a078, __in_chrg=<optimized out>) at 
../../src/base/RefCount.h:35
#8  ACLFilledChecklist::~ACLFilledChecklist (this=0x2059ec8, 
__in_chrg=<optimized out>) at FilledChecklist.cc:50
#9  0x00000000006c15a9 in ACLFilledChecklist::~ACLFilledChecklist 
(this=0x2059ec8, __in_chrg=<optimized out>) at FilledChecklist.cc:67
#10 0x00000000006f69ce in ACLChecklist::checkCallback (this=0x2059ec8, 
answer=...) at Checklist.cc:174
#11 0x00000000005a483a in fqdncacheCallback (f=f@entry=0x205cb20, 
wait=wait@entry=10) at fqdncache.cc:316
#12 0x00000000005a4c4a in fqdncacheHandleReply (data=<optimized out>, 
answers=<optimized out>, na=<optimized out>, error_message=<optimized out>) at 
fqdncache.cc:407
#13 0x000000000058ad2d in idnsCallback (q=q@entry=0x205cbe8, 
error=error@entry=0x0) at dns_internal.cc:1144
#14 0x000000000058fd66 in idnsGrokReply (buf=buf@entry=0xbc07e0 <idnsRead(int, 
void*)::rbuf> "\036<\201\200", sz=sz@entry=220) at dns_internal.cc:1313
#15 0x0000000000590a4f in idnsRead (fd=20) at dns_internal.cc:1400
#16 0x0000000000811c70 in Comm::DoSelect (msec=<optimized out>) at 
ModEpoll.cc:273
#17 0x000000000075836e in CommSelectEngine::checkEvents (this=<optimized out>, 
timeout=<optimized out>) at comm.cc:1829
#18 0x0000000000599b39 in EventLoop::checkEngine 
(this=this@entry=0x7fffffffe3b0, engine=engine@entry=0x7fffffffe330, 
primary=primary@entry=true) at EventLoop.cc:36
#19 0x0000000000599d75 in EventLoop::runOnce (this=this@entry=0x7fffffffe3b0) 
at EventLoop.cc:115
#20 0x0000000000599f80 in EventLoop::run (this=this@entry=0x7fffffffe3b0) at 
EventLoop.cc:83
#21 0x0000000000606d71 in SquidMain (argc=<optimized out>, argv=<optimized 
out>) at main.cc:1638
#22 0x00000000004ff12d in SquidMainSafe (argv=<optimized out>, argc=<optimized 
out>) at main.cc:1363








On 29/01/2016 12:46, "squid-dev on behalf of Amos Jeffries" 
<[email protected] on behalf of [email protected]> 
wrote:

>On 29/01/2016 8:10 a.m., Christos Tsantilas wrote:
>> Hi all,
>> 
>> After the patch r14351 created the following problems:
>>  - external_acl requires AccessLogEntry but ALE is not available
>>    in many cases such as ssl_bump ACLs.
>>  - The %<cert_subject stopped working because it was supported by
>>    external_acl code and not by logformat code.
>> 
>> This patch:
>>   - Passes AccessLogEntry in most cases.
>>     For example, PeerConnector-related classes are now covered.
>>   - Implements the %<cert_subject formating code for logformat.
>> 
>> 
>> Still there are cases which are not handled correctly:
>>   - In the case of transparent SSL bumping, the patch uses a local
>> AccessLogEntry to allow external_acl work with the ssl_bump access list.
>> 
>>  - The slow acls inside Ssl::PeerConnector can not support external_acl
>> in the case of PeerPoolMgr
>> 
>>   - Most of the fast acls does not support ALE based acls. I know that
>> currently the only ALE based acl is the external_acl, which is slow acl,
>> but my opinion is that it is not bad idea  to support cases the
>> external_acl result is stored in cache.
>> 
>>   - Also we need to check and review if the informations passed with the
>> ALE is the same passed using the FilledChecklist object. This is not
>> obvious.
>> 
>> 
>> This is a Measurement Factory project.
>> 
>
>+1. Please apply.
>
>But note that PeerConnector child classes are now in different files
>when merging to trunk.
>
>Amos
>
>_______________________________________________
>squid-dev mailing list
>[email protected]
>http://lists.squid-cache.org/listinfo/squid-dev
_______________________________________________
squid-dev mailing list
[email protected]
http://lists.squid-cache.org/listinfo/squid-dev

Reply via email to