- Added configure check for libbind.a. If found, it will be
used instead of libresolv.a.
- Changed fwdStart() to always allow internally generated
- requests, such as for peer digests.
+ requests, such as for peer digests. These requests are
+ known to fwdStart() because the address arg is set to
+ 'no_addr'.
- Completed initial HTCP implementation. It works, but is not
tested much.
- Added counters for I/O syscalls.
(netapp) Squid doesn't use private keys. This caused us
to remove almost every object from the cache.
- Added 'asndb' cachemgr stats to show AS Number tree.
+ - Fixed AS Number byte-order bug for netmasks.
- Fixed comm_incoming calling rate for high loads (Stewart
Forster).
- Give always_direct higher precedence than never_direct
(Henrik Nordstrom).
- - experimental attempt to fix "peer disappeared" coredumps
+ - Changed PORT ACL type to accept ranges. Now you can easily
+ deny, for example, all priveleged ports except 80, 70, 21,
+ etc.
+ - ARP ACL fixes for Linux (David Luyer).
+ - Replaced various "EBIT" flags bitfileds with structures of
+ "int:1" members.
+ - Changed storeKeyPrivate and storeKeyPublic to be a bit more
+ efficient by removing snprintf(). This causes an
+ incompatibility with old cache keys, however. To transition,
+ we will look up both the new and old style keys for about the
+ next 30 days. After that, if you haven't run this (or a
+ future) version, your cache contents will be lost.
+ - Made the client-side write buffer size configurable with
+ a #define in defines.h. By default it is still 4096 bytes.
+ - Removed redirectUnregister(). It should be unnecessary
+ because of cbdata locks.
+ - Fixed multiple HEAD request brokennesses (Henrik Nordstrom).
+ - Changed non-blocking connect(2) code to call getsockopt()
+ instead of connect() again. This is the approach recommended
+ by Stevens, and fixes bugs on BSD-ish systems when subsequent
+ connect() calls loop with EAGAIN status.
+ - Added MD5 cache keys to memory pool accounting.
+ - Added code to track number of open DISK descriptors and stop
+ swapping out objects if the number of disk descriptors becomes
+ too large. For now the limit must be manually configured with
+ the 'max_open_disk_fds'. By default, there is no limit.
+ - Stopped encoding a request method in the high byte of the ICP
+ reqnum field. Instead queried cache keys are copied to a
+ static array, indexed by the reqnum, modulo the array size.
+ Now we just use the request number to lookup a cache key,
+ instead of rebuilding it from the ICP reply URL and method,
+ unless we have netapp neighbors--they don't do reqnum
+ properly.
+ - Fixed reconfigure memory access bugs in redirect.c.
Changes to squid-1.2.beta24 (August 21, 1998):