Amos Jeffries [Fri, 20 Jun 2008 05:14:34 +0000 (17:14 +1200)]
Bug 1566 (also bug 975): esi:include aborts with lock assert
Proper fix requires callers inserting HTTPMSGLOCK() when the base rep
pointers are set, and HTTPMSGUNLOCK() macros when done. The lock/unlock
mechanism will take care of garbage collection in the background if used
properly.
The function this patches has no need to perform any of that itself
either way (it's a * not a ** ptr parameter so _cannot_ be safely deleted).
I'm cementing the temporary fix for 975 and 1566 as a permanent one and
documenting the correct requirements of the functions callers to prevent
memory leaks.
If leaks are found it will be separate bugs in the calling code related
to bad refcounting.
Amos Jeffries [Fri, 20 Jun 2008 04:43:01 +0000 (16:43 +1200)]
Author: Alex Rousskov <rousskov@measurement-factory.com>
Bug #425 fix: purge matching entries on PUT, POST, and DELETE requests.
This patch focuses on purging URLs in Location and Content-Location headers
of PUT, POST, and DELETE responses.
Purging Request-URIs was already supported for PUT and DELETE, but needed
polishing. I moved all code related to method-based purging into one Server
method and outside the neighbors_do_private_keys guard (and store entry key is
private guard). We may purge more related entries than before.
I also implemented Amos' TODO to purge related entries when receiving a
request with an unknown request method. Again, we may now purge more related
entries than before.
My primary concern about the polishing part of the change is that the old code
used to check that the cache entry being purged is not the current entry:
assert(e != pe);
The new code does not check for that but appears to work in my limited tests.
I am not quite sure why we needed that check if all purging methods are not
cachable anyway. Perhaps it is unsafe to call e->release() for some entries?
TODO: We cannot find Vary-controlled entries by URL and, hence, we cannot
purge them, right?
TODO: Optimize method category "search" in HttpRequestMethod by using
precomputed method_id:category maps.
Henrik Nordstrom [Wed, 18 Jun 2008 18:49:32 +0000 (20:49 +0200)]
Fix SNMP reporting of counters with a value >= 0xFF800000
The ASN.1 encoding of unsigned integers incorrectly compressed the value
as if it was a signed integer, truncating leading 0xFF octest, causing
0xFF800000 to be send ax 0x800000, 0xFFFF8000 as 0x8000 etc..
Amos Jeffries [Thu, 5 Jun 2008 03:54:46 +0000 (15:54 +1200)]
Update ACL docs
- Some regex types were missing -i mentions.
- Altered auth type to mention intercept alongside transparent
now that they are distinct in squid
- Also cosmetic re-order of a few items to provide a some type grouping
Amos Jeffries [Thu, 22 May 2008 12:05:45 +0000 (00:05 +1200)]
Silence secondary errors on fatal shutdowns.
This patch causes fatal() errors to set the shutdown flag before aborting.
The result of this is that secondary errors not related to the fatal condition
but caused during the shutdown sequence no longer confuse the error traces.
It also cleans up a little broken whitespace formatting in the snmp_core.cc
Guido Serassio [Sat, 17 May 2008 11:27:47 +0000 (13:27 +0200)]
Windows port: Added new mswin_check_ad_group external ACL helper
This helper allow the lookup of users's group membership in a Windows
Active Directory domain.
It overcomes the Lan Manager limits of mswin_check_lm_group, but it can be
used only with native Windows Active Directory domains, so mswin_check_lm_group
will not removed from Squid.
Amos Jeffries [Tue, 13 May 2008 08:56:04 +0000 (20:56 +1200)]
Add cachemgr.conf.default for easier maintenance
cachemgr.conf has for a long while been created conditionally if none
already existed, but no current new default file was added to compare
for new configuration options.
This patch reverts the behaviour to what was reportedy available in 2.5.
Adding a file named cachemgr.conf.default next to the cachemgr.conf
cachemgr.conf.default file is replaces unconditionally so that it remains
current with the most recently installed build.
cachemgr.conf itself is only created if not already present so as not to
loose local configuration changes.
Amos Jeffries [Thu, 8 May 2008 03:47:53 +0000 (15:47 +1200)]
Bug 2222 part 2: ipv4 client trying to view an ipv6 website crashes FreeBSD squid
This one:
- omits the v6-specific socket options on new addrinfo
- correctly counts the failed connection
- Resets the socket when a protocol error is found
- retries without any delay
Amos Jeffries [Wed, 7 May 2008 09:41:50 +0000 (03:41 -0600)]
Bug 2206: Build error caused by incorrect configure include file detection
Adds type-definitions for the *BSD family networking OS header files which
depend on them without including the necessary definition headers themselves.
Amos Jeffries [Wed, 7 May 2008 07:50:28 +0000 (01:50 -0600)]
Bug 2196: configure: net*/*.h present but cannot be compiled
This moves critical net*/*.h file tests from the general location
and use testign to the *BSD special testing which accounts for
system file dependencies.
TODO: some files are still missing dependencies even with the *BSD test
their brokenness will be fixed at a later point.
Make --with-large-files try to build 64-bit if possible
--with-large-files for some reason tried to make a 32-bit build with
large file offsets even if the host supports 64-bit applications, making
--with-large-files degrade Squid capabilities on 64-bit OS:es..
Amos Jeffries [Fri, 2 May 2008 10:37:01 +0000 (22:37 +1200)]
Author: Christos Tsantilas <chtsanti@users.sourceforge.net>
Bug 2308: Segmentation fault in AuthDigestUserRequest::authUser
In this patch:
- In method AuthDigestConfig::decode just do not delete the digest_request on
errors but use it as is in the authDigestLogUsername functions.
- In the method AuthDigestConfig::fixHeader change the line "int stale = 1;"
to "int stale = 0;" to make squid respond with "stale=false" in the first
unauthenticated request of web client.
Guido Serassio [Thu, 1 May 2008 16:19:06 +0000 (18:19 +0200)]
Removed the advertisement clause from BSD license
According to the new revised (3-clause) BSD license:
ftp://ftp.cs.berkeley.edu/pub/4bsd/README.Impt.License.Change
the advertisement clause (3) of old (4-clause) BSD license can
be deleted.
Also added forgotten reference to lib/strnstr.cc into CREDITS
Adds --enable-zph-qos options to turn on the following:
- Allows you to select a TOS/Diffserv value to mark local hits.
- Allows you to select a TOS/Diffserv value to mark peer hits.
- Allows you to selectively set only sibling or sibling+parent requests
- Allows any HTTP response towards clients will
have the TOS value of the response comming from the remote
server masked with the value of zph_preserve_miss_tos_mask.
For this to work correctly, you will need to patch your linux
kernel with the TOS preserving ZPH patch.
The kernel patch can be downloaded from http://zph.bratcheda.org
- Allows you to mask certain bits in the TOS received from the
remote server, before copying the value to the TOS send towards
clients.
Import strnstr from FreeBSD sources. Needed for some string-safe operations.
strnstr() is not provided on all OS (Linux with gcc 3.x for one).
And some OS are known to bundle an unsafe version (MacOS X 10.4 has a buffer overrun)
So code should use the function named squid_strnstr() and auto-tools will
test to see if the OS provided version is usable.
This patch merges part 2 of the TPROXY-related updates.
- Makes interception handling flags and options always-present
- Updates squid.conf http_port options for clarity
- Builds structure for sequential lookup of multiple interception methods
- Performs sequential lookups for IPFW and multiple Netfilter targets
if Squid configured to enable those transparency methods.
- Pulls most of the TPROXYv2 related code out of Comm into IPInterception
IPFW changes are still experimental, but Netfilter targets have been tested.
TODO: depending on the anoyance levels a better logging method for NAT
failures may need to be implemented. The existing methods of logging
one-per-N seconds, for all lookup methods may prove annoying.
Alex Rousskov [Thu, 17 Apr 2008 05:50:09 +0000 (23:50 -0600)]
Avoid segfaults when scheduling an async call for a non-existent job.
Scheduling a call for invalidated or non-existent job should be a no-op
because (a) it simplifies the code and (b) the job may disappear while
the call is queued so checking job status at the call time is
pointless from the correctness point of view.
Existing code already relies on this guarantee, but apparently not too much.
The bug was in a JobDialer. I have not yet verified whether other dialers
need a similar fix.
projects / thirdparty / squid.git / log
