index.tmpl,v 1.71 2001/03/01 23:53:07 wessels Exp

2.3.STABLE4 Bugs

2.3.STABLE3 Bugs

2.3.STABLE2 Bugs

2.3.STABLE1 Bugs

2.3.DEVEL3 Bugs

2.3.DEVEL2 Bugs



2.3.STABLE4 Bugs

The snmp_community ACL check segfaults if Squid fails to parse a SNMP query.

platforms All
versions 2.3.STABLE4 and earlier
synopsis Sending an unparsable SNMP query to Squid can cause a segment violation.
reported by Henrik, pogar at pik-net dot pl
configuration --enable-snmp
patch squid-2.3.stable4-snmp-community-null-pointer.patch
status Fixed

CARP assertion on sum of load factors

platforms All?
versions Squid-2.3.STABLE4
synopsis Comparing floating point numbers for equality is tricky. The old way can cause an assertion even though two numbers actually do add up to 1.
reported by Francis A. Vidal
configuration --enable-carp
patch squid-2.3.stable4-carp-assertion.patch
status Fixed

Some HTML output may not be quoted properly

platforms All
versions 2.3.STABLE4
synopsis Everywhere where Squid inserts text received from the network into a HTML page (error pages, FTP listings, Gopher listings, ...) care must be taken to ensure that the text is properly encoded as HTML, or a malicious user might be able to insert script code or other HTML tags, and exploit the web browser of any user visiting their page or clicking on that funny link received in a email..
reported by Lincoln Yeoh (lyeoh at pop dot jaring dot my)
Others...
patch squid-2.3.stable4-html_quoting.patch
status Fixed

clientAccessCheck allows proxy requests in accel_only mode.

platforms All
versions 2.3.STABLE4
synopsis clientAccessCheck incorrectly returns ACCESS_ALLOWED for proxy requests when configured as an HTTP accelerator only.
reported by Takashi Taniguchi (tani at iij dot ad dot jp)
configuration httpd_accel_host ...
httpd_accel_with_proxy off
patch squid-2.3.stable4-accel_only_access.patch
status Fixed

Invalid IP acl entry: unknown first address '144.'

platforms All
versions 2.3.STABLE4
synopsis The code that scans ACL tokens for IP addresses and hostnames couldn't tell that "123.foo.com" is a hostname rather than an IP address.
reported by Zeev Meloch (zeev at iec dot co dot il)
patch squid-2.3.stable4-invalid_ip_acl_entry.patch
status Fixed Thu Aug 10 21:38:12 GMT 2000

configure --enable-ipf-transparent fails

platforms All
versions 2.3.STABLE4
synopsis The configure script uses "==" when it should use "=" for /bin/test.
reported by Dan Larsson (dl at tyfon dot net)
configuration --enable-ipf-transparent
patch squid-2.3.stable4-ipfw_configure.patch
status Fixed Thu Aug 10 06:18:17 GMT 2000

cachemgr 'idns' RCODE table is mis-formatted

platforms ALL
versions 2.3.STABLE4
synopsis Missing a newline on cachemgr output
reported by Steve Snyder (swsnyder at home dot com)
patch squid-2.3.stable4-internal_dns_rcode_table_formatting.patch
status Fixed Sat Jul 22 18:03:01 GMT 2000

Cache Digests and FTP Icon files are not found

platforms All
versions 2.3.STABLE4
synopsis cache.log shows messages like:
2000/07/20 09:49:19| internalStart: unknown request:
GET /squid-internal-static/icons/anthony-tar.gif HTTP/1.0

There was a change in 2.3.STABLE4 that was supposed to cause FTP icons and other "internal" objects to always be cached, regardless of the 'minimum_object_size' setting. Instead it always caused them to NOT be cached.

reported by Reuben Farrelly (reuben-squid at reub dot net)
Ulrich Seidl (uis at Regent dot E-Technik dot TU-Muenchen dot DE)
Krzysztof Czuma (czuma at Elektron dot pl)
patch squid-2.3.stable4-ftp_icon_not_found.patch
status Fixed

2.3.STABLE3 Bugs

GET requests with ``Content-length: 0'' are rejected

platforms All
versions 2.3.STABLE3
synopsis Some user agents, such as FrontPage 2000, send GET requests with a ``Content-Length: 0'' header. RFC 2616 only states that GET requests should not have a body, but says nothing about the header. This patch allows such requests.
reported by Richard Birkby (rbirkby at thundermain dot com)
patch squid-2.3.stable3-zero-content-length.patch
status Fixed

Disk space over the limit

platforms All
versions 2.3.STABLE3
synopsis If you use LRU replacement (the default), Squid doesn't actually delete any old objects. Your disks fill up, and Squid complains, but disk usage never goes down.

Apply this patch and restart your Squid process. You may still get some warnings, but the disk usage should start to go down. The warnings stop when disk usage reaches a safe level.

reported by Petr Sebor (petr at scssoft dot com)
patch squid-2.3.stable3-storeExpiredReferenceAge.patch
status Fixed

CARP module does not compile

platforms All
versions 2.3.STABLE3
synopsis There is a typo for sizeof that gives this error:
carp.c:98: warning: implicit declaration of function `size'
reported by Denyo Atanasov (denis at mgu dot bg)
configuration --enable-carp
patch squid-2.3.stable3-carp_compile.patch
status Fixed

2.3.STABLE2 Bugs

Client-side persistent connections have hard-coded 15 second timeout

platforms All
versions 2.3.STABLE2
synopsis The timeout for idle persistent connections to clients was hard-coded to 15 seconds. This patch changes the timeout to the pconn_timeout value, which is also used for the server-side.
patch squid-2.3.stable2-client_pconn_timeout.patch
status Fixed for 2.3.STABLE3

Bogus Age header on IMS_HIT replies

platforms All
versions Squid-2.3.STABLE2
synopsis When Squid generates a 304 reply itself (an IMS_HIT), the Age header is bogus. Its set to the current number of seconds since 1970.
reported by Henrik Nordstrom
patch squid-2.3.stable2-ims_hit_age.patch
status Fixed for 2.3.STABLE3

Solaris x86 fails to link with mallopt, mallinfo errors

platforms Solaris/x86
versions 2.3.STABLE2, earlier
synopsis The configure script enables 'dlmalloc' by default for Solaris/x86. However, things recently got shuffled around which prevented LIBDLMALLOC from being set. Thus, libdlmalloc.a was never built by lib/Makefile.
reported by Greg Robinson (Greg dot Robinson at dsto dot defence dot gov dot au)
patch squid-2.3.stable2-Solaris_malloc_link.patch
status Fixed for 2.3.STABLE3

FTP and Gopher don't check return value of mimeGetIconURL

platforms Solaris
versions 2.3.STABLE2
synopsis Due to Solaris' 256 FD limit for stdio, mimeInit may fail to load icons during a reconfigure. In this case, mimeGetIconURL would return NULL. FTP listings call mimeGetIconURL without checking its return value. This patch changes mimeGetIconURL to return an empty string, instead of NULL.
reported by John Line
patch squid-2.3.stable2-mimeGetIconURL_returns_NULL.patch
status Fixed for 2.3.STABLE3

Asyncufs causes slow filedescriptor leak

platforms All?
versions 2.3.STABLE2
synopsis Asyncufs slowly leaks filedescriptors, leaving cache files open with no bytes read or written. Looks like they're probably files that have been opened for reading.
reported by Philip Guenther (guenther at gac dot edu)
configuration --enable-async-io
cache_dir asyncufs ...
patch squid-2.3.stable2-asyncufs_fd_leak.patch
status Fixed for 2.3.STABLE3

Failure to complain about getpwnam() return value.

platforms Solaris for sure, others?
versions anything before 2.3.STABLE2
synopsis The leave_suid function was calling getpwnam(3). On Solaris, getpwnam requires an unused filedescriptor less than 256 to open a FILE *. If getpwnam fails, Squid silently returns from leave_suid, leaving the process with root priveleges.

This bug explains why some cache swap files become owned by root.

reported by Andres Kroonmaa (andre at online dot ee).
patch squid-2.3.stable2-getpwnam_return_value.patch
status Fixed for 2.3.STABLE3

SNMP related coredumps and problems

platforms All
versions 2.3.STABLE2
synopsis snmpwalk ... public .1.3.6.1.4.1.3495 make squid coredump.
reported by Fritz Buetikofer (Fritz dot Buetikofer at id dot unibe dot ch)
configuration --enable-snmp
patch squid-2.3.stable2-snmp-problems.patch
status Fixed for 2.3.STABLE3

GCC opimization may cause ``random'' coredumps

platforms GCC 2.95.x on OpenBSD, FreeBSD, other Intel?
versions Any
synopsis Pointers magically become NULL. This can result in Unexpected ``FATAL: Received Segment Violation...dying'' messages, and various ``assertion'' messages.

We have received a number of reports like this. They seem to be caused by a GCC compiler optimization bug. Disabling the optimization fixes the problem.

patch squid-2.3.stable2-disable-gcc-2.95.2-optimization.patch, or:

Edit all Makefiles and remove '-O2' or any other optimization flags from the AC_CFLAGS macro. Then, from the top src directory:

% make clean
% make

Proxy auth user not logged for redirected URLs

versions 2.3.STABLE2, earlier?
synopsis For URLs that get redirected, the proxy auth username does not appear in access.log
reported by Ivan E. Moore II (rkrusty at tdyc dot com)
patch squid-2.3.stable2-redirected_username_logging.patch
status Fixed for 2.3.STABLE3

assertion failed: internal.c:97: "host && port && name"

platforms All
versions 2.3.STABLE2, earlier
synopsis Squid may assert in internalRemoteUri() if it can't figure out its fully-qualified domain name. This could happen if http_port is set to an IP address that does not resolve to a hostname. With this patch, getMyHostname() tries a little harder to get a falid FQDN. If not, it exits with a FATAL message and tells you to set visible_hostname
reported by Reuben Farrelly (reuben at reub dot net)
patch squid-2.3.stable2-getMyHostname.patch
status Fixed for 2.3.STABLE3

POST requests hang with WCCP.

platforms FreeBSD
versions 2.3.STABLE2, earlier
synopsis POST requests, and others with large message bodies, fail or hang. The FreeBSD GRE code incorrectly drops large GRE packets that are split across a chain of mbufs.
reported by Riccardo Pizzi (pizzi at adhoc dot net)
patch New gre.c
status Fixed

netdbExchangeHandleReply may loop on corrupt data

platforms All
versions 2.3.STABLE2 and earlier
synopsis A corrupt netdb exhange message can cause Squid to loop and run at 100% CPU utilization.
configuration --enable-icmp
patch squid-2.3.stable2-netdb_exchange_loop.patch
status Fixed for 2.3.STABLE3

Whitespace not stripped from hostnames

platforms All
versions 2.3.stable2 and earlier
synopsis Whitespace characters are not stripped from hostnames, regardless of the 'uri_whitespace' setting.
reported by Ahsan Khan (ahsank at one dot net dot pk)
patch squid-2.3.stable2-hostname_whitespace.patch
status Fixed for 2.3.stable3

Extra EOF in cf.data.pre

platforms All
versions 2.3.STABLE2
synopsis When merging CVS branches, an extraneous "EOF" got committed to cf.data.pre. This prevents the server_persistent_connections and client_persistent_connections from being compiled into Squid.
reported by Juergen Moellenhoff (jurgen at oic dot de) @stacktrace
patch squid-2.3.stable2-EOF_in_cf.data.pre.patch
status Fixed for 2.3.STABLE3

Compile fails with --disable-internal-dns

platforms All
versions 2.3.STABLE2
synopsis Squid fails to compile with --disable-internal-dns because some source files have "USE_DNSSERVER" and others have "USE_DNSSERVERS".
reported by Sverre H. Huseby (shh at online dot no)
configuration --disable-internal-dns
patch squid-2.3.stable2-USE_DNSSERVER.patch
squid-2.3.stable2-USE_DNSSERVER_part2.patch
status Fixed for 2.3.STABLE3

Problems with internal DNS code

platforms Solaris 7, gcc 2.95.2 19991024 and Sun Workshop compilers.
Debian GNU/Linux 2.1, gcc 2.7.2.3, kernel 2.0.38
Linux 2.2.14 Redhat6.0
Linux 2.2.14 Redhat6.2
versions 2.3.STABLE2
synopsis For some reason, squid will periodically decide that it can't resolve hostnames. It doesn't do this consistently and when it does, it does seem to be at time when load is lower than normal - for example, the cache will behave itself during weekdays, but during evenings and weekends, when the load on the cache drops, DNS resolution problems start.

We seem to be getting errors of either (from yesterday)

Server Failure: The name server was unable to process this query.

or

No Address records.

http://news.bbc.co.uk/ seems to be a common error case.

Also www.whitehouse.gov.

reported by Chris Tilbury (Chris dot Tilbury at warwick dot ac dot uk)
Richard Kettlewell (richard at elmail dot co dot uk)
BaRT (bart11 at dingoblue dot net dot au)
patch squid-2.3.stable3-idns_retry_srvfail.patch
status Possibly fixed

2.3.STABLE1 Bugs

Simplify Chineese error pages omitted

platforms N/A
versions 2.3.STABLE1
synopsis Simplify Chineese error pages were omitted from the distribution.
reported by wang_daqing (wang_daqing at 163 dot net)
patch Simplify_Chinese.tar.gz
status Comitted for 2.3.STABLE3

PURGE method doesn't get disabled after reconfigure

platforms All
versions 2.3.STABLE1 and earlier
synopsis The PURGE method is dangerous because it allows clients to remove cached objects. Since its just another request method, PURGE requests would be allowed unless specifically denied with an access control entry. Thus, we have a flag that disables PURGE unless there is an ACL that references the PURGE method.

This patch includes a fix that resets the PURGE enable flag upon reconfigure.

This patch also includes a fix that enables PURGE only if its ACL is used in the http_access list, whereas before it just needed to be defined in an acl.

reported by John Line
patch squid-2.3.stable1-purge_reconfigure.patch
status Fixed in 2.3.STABLE2

Cookie header is not known

platforms All
versions 2.3.STABLE1
synopsis The "Cookie" HTTP header is not in the list of known headers. This makes it impossible to use the Cookie header with the http_anonymize option.
patch squid-2.3.stable1-cookie_header.patch
status Fixed in 2.3.STABLE2

assertion "no <= ns" failed: file "rfc1035.c", line 296

platforms FreeBSD 2.2.8-RELEASE
versions 2.3.STABLE1
reported by Peter Olsson (pol at leissner dot se)
configuration N/A
stack trace N/A
patch N/A
status Reported, waiting for more details

Async/UFS uses incorrect 'errflag' values in callbacks

platforms all
versions 2.3.STABLE1
synopsis When 'errno' values are used in storeAufsIOCallback, it can trigger an assertion in storeSwapInFileClosed.
configuration --enable-async-io
cache_dir asyncufs ...
patch squid-2.3.stable1-asyncufs_callback_args.patch
status Fixed in 2.3.STABLE2

rfc1035Unpack() fails to properly handle truncated messages

platforms All, when using internal DNS lookups
versions 2.3.STABLE1
synopsis A truncated DNS reply causes an assertion in rfc1035Unpack(). This fix stops unpacking resource records when if a record appears to be truncated. The 'dnscache' name server always truncates replies at 512 octets, as per RFC 1035. BIND, however, does not.
reported by Stuart Henderson (sh at eclipse dot net dot uk)
patch squid-2.3.stable1-rfc1035RRUnpack_truncate.patch
status Fixed in 2.3.STABLE2

``anonymize_headers allow'' breaks after reconfigure

platforms All
versions 2.3.STABLE1
synopsis The paranoid anonymization with "anonymize_headers allow" won't work after sending squid a reconfigure signal. All headers will remain in the outgoing queries. The "http_header_first" variable should be reset to its default value (0) when reconfiguring squid, before parsing "anonymize_headers" directives. If not, ``httpHeaderMask'' isn't reset to a correct default value.
reported by Jean-Gabriel Dick (Jean-Gabriel dot Dick at curie dot fr)
patch squid-2.3.stable1-anonymize_headers_reconfigure.patch
status Fixed in 2.3.STABLE2

All requests are denied for accelerator-only squid installations.

platforms All
versions 2.3.STABLE1
synopsis If you run Squid exclusively as an accelerator (i.e., httpd_accel_with_proxy is off), then all requests are denied.
reported by Gregg Kellogg (gregg at cafex dot com)
patch squid-2.3.stable1-accel_only_access_denied.patch
status Fixed in 2.3.STABLE2

Internal DNS ignores 'dns_nameservers' in squid.conf

platforms ALL
versions 2.3.STABLE1
synopsis If you define some DNS nameservers in squid.conf, Squid does not use them by default. It would use them if you disabled the internal DNS and used the external dnsserver processes.
reported by Henrik Nordstrom
Lawrence Decker (ldecker at pbchcd dot state dot fl dot us)
Alexey Nogin (ayn2 at cornell dot edu)
patch squid-2.3.stable1-internal_dns_ignores_dns_nameservers.patch
status Fixed in 2.3.STABLE2

Viewing delay pools in cachemgr causes 100% CPU utilization

platforms Linux 2.2.14
versions 2.3.STABLE1
synopsis When the system is running if i go to the cachemanager and as for "Delay Pool Levels" a few times ("Refresh" in the browser) the squid process stops handling requests and jump to 99% CPU usage.
reported by Pedro Ribeiro (pribeiro at isel dot pt)
configuration --enable-delay-pools
stack trace N/A
patch N/A
status Reported

The 'maxconn' ACL works the opposite way

versions 2.3.STABLE1
synopsis Instead of returning a match for a client that is over the limit, it returns a match for a client when it is under the limit.
reported by Pedro Ribeiro (pribeiro at isel dot pt)
patch squid-2.3.stable1-acl_maxconn.patch
status Fixed in 2.3.STABLE2

When squid is started as daemon, syslog messages go to /dev/console.

platforms Linux 2.2.14 (Slackware 7.0), Others?
versions 2.3.STABLE1
synopsis In src/main.c:watch_child(), all filedescriptors are closed after calling openlog(). After that, the filedescriptor for /dev/log is not automatically reopened by syslog(), and no log entries are produced by the parent process, only the console screen is cluttered.
patch squid-2.3.stable1-daemon_syslog.patch
status Fixed in 2.3.STABLE2

Squid complains about 'http_port' values that should be accepted.

platforms Linux, others?
versions 2.3.STABLE1
synopsis Some valid 'http_port' values are rejected, especially if the value includes a hostname:port pair.
reported by Martin Brooks (martin at gointernet dot co dot uk)
CarloX (cbote at el-mundo dot es)
Kendall Lister (kendall at charon dot net dot au)
patch squid-2.3.stable1-http_port_parsing.patch
status Fixed in 2.3.STABLE2

Byte-ordering bug for internal DNS PTR queries

platforms This bug affects ``bigendian'' systems like MIPS and SPARC CPU's
versions 2.3.STABLE1
synopsis IP address octets are in the wrong order for bigendian systems.
reported by Per Hedeland (per at erix dot ericsson dot se)
patch squid-2.3.stable1-internal_dns_ptr_query.patch
status Fixed in 2.3.STABLE2

Internal DNS pending queue may get stuck

platforms All
versions 2.3.STABLE1
synopsis Pending lookup queue gets "stuck" if there is no steady DNS traffic. idnsCheckQueue() failed to reschedule itself if timeouts for pending queries hadn't been reached.
reported by Henrik Nordstrom
patch squid-2.3.stable1-internal_dns_queue.patch
status Fixed in 2.3.STABLE2

Example domain-based ACL's are incorrect

platforms All
versions 2.3.STABLE1
synopsis The examples for 'dstdomain' and 'srcdomain' ACL's say to use "foo.com" but most users probably want it to work as a wildcard, so its better to use ".foo.com."
reported by Bertold Kolics
patch squid-2.3.stable1-domain-acl-examples.patch
status Fixed in 2.3.STABLE2

Integer conversion bugs for EPLF listings on systems with 64-bit longs

platforms Alpha and MIPS (?) CPU's
versions 2.3.STABLE1
synopsis When squid runs on an alpha (where sizeof(long) > sizeof(int) and time_t is long), EPLF listings have wrong date. This happens because a pointer to int is casted to time_t*.
reported by Alexander V. Lukyanov (lav at yars dot free dot net)
patch squid-2.3.stable1-ftp_eplf_int_conversion.patch
status Fixed in 2.3.STABLE2

maximum_icp_query_timeout squid.conf example usage is wrong

platforms All
versions 2.3.STABLE1
synopsis In the coments for maximum_icp_query_timeout in squid.conf, the option is written as just icp_query_timeout.
reported by Thomas Wahyudi (thomas at home dot unpar dot ac dot id)
patch squid-2.3.stable1-maximum_icp_query_timeout.patch
status Fixed in 2.3.STABLE2

Setting request_body_max_size to zero denies all POST, PUT requests.

platforms All
versions 2.3.STABLE1
synopsis Setting request_body_max_size to zero did not work as it was advertised. It is supposed to disable size checking for all HTTP message bodies.
reported by Aleksandr Konstantinov (sanjak at tipas dot lt),
wang_daqing (wang_daqing at 163 dot net)
patch squid-2.3.stable1-request_body_max_size.patch
status Fixed in 2.3.STABLE2

myport ACL type does not work with redirector_access list

platforms All
versions 2.3.STABLE1
synopsis The local port number was not being assigned for redirector access checks.
reported by Denis Shaposhnikov (dsh at neva dot vlink dot ru)
patch squid-2.3.stable1-redirector_access_with_myport_acl.patch
status Fixed in 2.3.STABLE2

2.3.DEVEL3 Bugs

Trusting client's Host: header may result in cache poisoning.

platforms All
versions 2.3.DEVEL3 and earlier, including 2.0-2.2
synopsis Prior to this patch, Squid copied the Host: header from the client's request into the request headers being sent to an origin server. Squid does not complain if the Host: header does not match the URL hostname.

When the two hostnames do not match, and both hostnames are served from the same physical HTTP server, it may result in cache poisoning. Squid stores the response using the URL as a the cache key, but an origin server sends a response based on the Host: header.

This patch causes Squid to not use the client's Host: header unless (a) the URL went through a Squid redirector, and (b) the ``redirect_rewrites_host_header'' option is turned on. Otherwise, Squid always copies the URL hostname into the forwarded Host: header.

reported by Brian Ristuccia (brianr at osiris dot 978 dot org)
patch squid-2.3.devel3-host-header-mismatch.patch
status committed to 2.3 branch

RFC 2616 says whitespace in URL's should be stripped

platforms all
versions 2.3.DEVEL2 and earlier
synopsis By default, Squid would deny requests when the URL contained a whitespace character. RFC 2616 says that applications should strip out whitespace. This patch adds a new ``strip'' option for the ``uri_whitespace'' directive and makes it the default.
patch squid-2.3.devel3-strip-whitespace.patch
status committed to 2.3 branch

HTTP responses that don't have a Content-length header don't get cached.

platforms all
versions 2.3.DEVEL3
synopsis When a reponse doesn't have a Content-length header, then mem->reply->content_length is negative. Comparing a negative value to a signed value (Config.Store.maxObjectSize) is not such a good idea.
patch squid-2.3.devel3-not-storing-replies-without-clen.patch
status committed to 2.3 branch

Filemap runs out of swap file numbers

platforms all
versions 2.2, 2.3
synopsis It is increasingly common for filemap to run out of swap file numbers. With new replacement algorithms, it is possible to have very small average object sizes. People won't set avg_object_size, and shouldn't have to.
patch squid-2.3.devel3-filemap-fixes.patch
status Committed to 2.3 branch

ftp.netscape.com sometimes hangs on PASV command

platforms all?
versions 2.3
synopsis This is not a Squid bug fix, but rather a workaround for netscape.com's buggy ftp server. Sometimes it just hangs when you send it a PASV command. This patch lowers the read timeout after sending a PASV command (to 15 seconds) and sets a flag to try PORT next time if the timeout occurs.
patch squid-2.3.devel3-ftp-pasv-timeout-hack.patch
status Committed to 2.3 branch

2.3.DEVEL2 Bugs

FATAL : unknown cache_dir type '/foo'

platforms All
versions 2.3.DEVEL2
synopsis The cache_dir directive now requires a ``type'' field. The default squid.conf cache_dir line does not work. Furthermore, the cache_dir comments were not updated.
patch squid-2.3.devel2-cache_dir-conf.patch
status Fixed

rfc1035AnswersUnpack() may assert at line 427

platforms All
versions 2.3.DEVEL2
synopsis rfc1035AnswersUnpack() has a byte-to-integer conversion bug that could result in a negative length.
reported by DW
patch squid-2.3.devel2-rfc1035AnswersUnpack.patch
status Fixed

Shrinking cache_dir size with Async I/O can cause an abort

platforms All
versions 2.3.x
synopsis When using Async I/O, if you try to shrink the cache_dir size and then reconfigure or restart Squid, you may get a message about ``request queue growing uncontrollably!'' This is because the current Async I/O code has no feedback mechanism and Squid tries to remove a lot of objects in a short amount of time. The filesystem can't keep up and Squid quickly saturates all the thread processes.
reported by tom minchin (tom at interact dot net dot au)
configuration --enable-async-io
status Reported

When using cache digests, Squid doesn't fetch a digest immediately when it detects a neighbor is again reachable.

platforms all
versions 2.3, probably earlier as well
synopsis When a dialup link is down, Squid displays a message in the cache log to say that digests are unavailable from the parent proxy, because its not connected.

But when the dialup link is reconnected, Squid does not seem to retry and obtain a digest as soon as the link to the remote proxy comes up. It will obtain one eventually, but often does not seem to realise that since the remote proxy is up, it can obtain a digest immediately.

Would it be possible to either: (a) force, via some command, to get my local squid on the end of my dialup link to try again and retrieve a proxy (manual intervention)

reported by Reuben Farrelly (reuben dot farrelly at bigpond dot com)
status Reported. Low priority. There is probably no ``bug'' here. This is how it is programmed to work. Whether or not this is desirable or not is debatable.

Setting icp_port to 0 causes Squid to ignore cache_peer lines.

platforms all?
versions 2.2.devel2
synopsis I've just found that setting icp_port to 0 means you can't use cache_peer (you can, but Squid ignores any cache_peer lines).
reported by tom minchin (tom at interact dot net dot au)
status Reported.

FATAL messages appear in access.log

platforms all?
versions 2.2.X, 2.3.DEVEL2
synopsis Sometimes, FATAL messages will appear in access.log. It only happens when Squid is run as a daemon process (i.e. without the -N option). In this case, stderr gets closed and access.log gets opened as filedescriptor #2. But the real bug is that were writing to ``stderr'' when opt_debug_stderr was nonzero, instead of when it is greater than zero.
reported by Arjan de Vet (Arjan dot deVet at nl dot origin-it dot com)
patch squid-2.3.devel2-fatal-msgs-in-access-log.patch
status Fixed

Empty first line of response is discarded

platforms all
versions 2.3.DEVEL2, 2.2.X, probably earlier
synopsis When there are NO reply headers, and the first character of an HTTP body is CR ("\r") or NL ("\n"), Squid may treat it as a part of the headers and fail to send that line to a client. For example:
	HTTP/1.0 200 Ok

		
	foo
	
reported by Mike Dixon (mdixon at placeware dot com)
status confirmed

Async I/O leaks lots of memory

platforms Anything that supports Async I/O
versions 2.3.DEVEL2
reported by Walter Klomp (walter at swiftech dot net dot sg)
configuration --enable-async-io
patch squid-2.3.devel2-async-io-memory-leak.patch
status Probably fixed.

Unlinkd process doesn't start

platforms Redhat Linux 6.0, kernel 2.2.10
versions 2.3.DEVEL2
synopsis cache.log does not show the "Unlinkd pipe opened on FD XX" message. The cache_dir partition fills up because old objects are not removed.
reported by Marcelo M. Sosa Lugones (marcelo at sosa dot com dot ar)
configuration --enable-wccp --enable-delay-pools
status Reported

assertion failed: store.c:446: "e->store_status == STORE_PENDING"

platforms FreeBSD 2.2.8-STABLE
versions 2.3.DEVEL2, 2.2.STABLE3
reported by Brad Meier (blm at its dot uct dot ac dot za)
configuration N/A
stack trace N/A
patch N/A
status Reported

Compile time bugs with HEAP_REPLACEMENT defined

platforms ALL
versions 2.3.DEVEL2
synopsis store.c and snmp_agent.c have compile bugs when HEAP_REPLACEMENT is defined.
reported by Arjan de Vet (Arjan dot deVet at nl dot origin-it dot com)
patch squid-2.3.devel2-heap-replacement-compile.patch
status Fixed

Coredump if /etc/resolv.conf has blank lines

versions 2.3.DEVEL2
synopsis If /etc/resolv.conf has a blank line, squid coredumps trying to access a NULL pointer in idnsParseResolvConf().
reported by Markus Storm (Markus dot Storm at mediaWays dot net)
patch squid-2.3.devel2-resolv.conf-coredump.patch
status Fixed

Delay pools causes memPoolAlloc() to fail with an assertion

platforms All
versions 2.3.DEVEL2
synopsis Parsing delay pools options adds an event, but the event module isn't initialized yet.
reported by Daniel Kalchev (daniel at digsys dot bg)
configuration --enable-delay-pools
patch squid-2.3.devel2-event-add-assertion.patch
status Fixed

index.tmpl,v 1.71 2001/03/01 23:53:07 wessels Exp