Russ Allbery [Wed, 6 Feb 2008 18:44:37 +0000 (18:44 +0000)]
* Fix FTBFS caused by a change in dh_strip behavior. It no longer
creates empty files for already stripped binaries, so we should no
longer remove them. Thanks, Lucas Nussbaum. (Closes: #464274)
Russ Allbery [Tue, 18 Dec 2007 02:27:16 +0000 (02:27 +0000)]
* Build bosserver with --enable-bos-new-config. On startup, if
BosConfig.new exists, it will be renamed to BosConfig before the
configuration is read. (Closes: #453398)
Russ Allbery [Tue, 18 Dec 2007 02:20:30 +0000 (02:20 +0000)]
* Apply upstream patch from the 1.5 series to allow multiple realms to
be listed in the server krb.conf and hence treated as local.
* Mention in README.Debian that the server is built with supergroups
enabled and with the multiple local realm patch. (Closes: #453399)
Russ Allbery [Tue, 18 Dec 2007 00:39:45 +0000 (00:39 +0000)]
* Apply upstream patch to allocate vnodes from the malloc pool rather
than off the stack, fixing file server problems on SPARC. Thanks to
Barney Sowood for tracking down the patch. (Closes: #448380)
Russ Allbery [Sun, 11 Nov 2007 23:52:18 +0000 (23:52 +0000)]
* Fix man pages for subcommands to use an underscore in the NAME section
instead of a space so that they'll be indexed properly by the current
groff and man-db.
Russ Allbery [Sat, 10 Nov 2007 03:48:47 +0000 (03:48 +0000)]
* Use uname -m rather than dpkg --print-architecture to determine the
sysname for kernel module builds. Otherwise, builds with a different
kernel architecture than userspace (amd64 kernels with i386 userspace,
for example) don't work. Thanks, Anders Kaseorg. (Closes: #448452)
Russ Allbery [Sat, 10 Nov 2007 03:37:28 +0000 (03:37 +0000)]
- Even with a new enough Linux kernel to support keyrings, we still
may have found the system call table and need to clean it up when
unloading the module. (Closes: #448793)
Russ Allbery [Sat, 10 Nov 2007 03:30:57 +0000 (03:30 +0000)]
* Fix signing the generated the changes file on a kernel module build.
The module build file reorganization required an additional change to
genchanges. Thanks, Anders Kaseorg. (Closes: #448449)
Russ Allbery [Tue, 16 Oct 2007 21:53:22 +0000 (21:53 +0000)]
* Scan /var/log/openafs for core files with a PID appended when a server
run by bosserver exits with a signal. bosserver was unable to
correctly handle core files for multithreaded servers since it
expected a filename of core and Linux always appends a PID to core
dumps by multithreaded processes.
* Stop creating /var/lib/openafs/cores. Nothing actually uses it.
Russ Allbery [Tue, 16 Oct 2007 06:22:00 +0000 (06:22 +0000)]
* Remove the workaround added for 2.6.20 and 2.6.21 kernels with
CONFIG_PARAVIRT enabled. The mistaken GPL-labelling of core
interfaces was fixed in 2.6.22, which has now migrated to testing.
Russ Allbery [Mon, 20 Aug 2007 19:52:13 +0000 (19:52 +0000)]
Better handling of stamp files to squash more duplicate work by make.
Every target now has a stamp file. Also remove the debugging symbols
for upserver and upclient, since they're empty. Some additional fixes
for generation of the kernel source package.
Russ Allbery [Mon, 20 Aug 2007 04:48:31 +0000 (04:48 +0000)]
* Restructure the source for the module package. The build files for
the module build are now in debian/module in the source package and
there is a completely separate rules file. This way, the module build
rules don't have to depend on quilt and the structure is clearer.
* Fix debhelper dependencies for the module build.
* Update module standards version to 3.7.2 (no changes required).
Russ Allbery [Fri, 3 Aug 2007 02:53:09 +0000 (02:53 +0000)]
* Apply upstream patch for the OpenAFS file server to treat clients that
have an all-zero UUID as if they don't have UUID support, working
around bugs in various client versions that can cause multiple hosts
to use all-zero UUIDs and conflict with each other.
Russ Allbery [Fri, 3 Aug 2007 02:49:40 +0000 (02:49 +0000)]
* Use dpkg --print-architecture to determine the AFS sysname for both
regular builds (avoiding --print-installation-architecture, which is
deprecated) and kernel builds (avoiding arch, which is temporarily not
in unstable and which wouldn't do the right thing for cross-compiles).
* Add hppa as a recognized architecture for kernel module builds,
although OpenAFS currently only supports 2.4 kernels on HPPA so this
isn't that useful.
* Add a union to des_ks_struct and reorder includes to get the correct
alignment for DES code on SPARC. Thanks, Chaskiel Grundman. Partly
addresses #415952.
* Fix a typo in debian/kern-sysname that caused openafs-modules-source
to still assume all ia64 systems used Linux 2.4. Thanks, Jakob
Haufe. (Closes: #432344)
Russ Allbery [Wed, 27 Jun 2007 14:33:29 +0000 (14:33 +0000)]
* Clean up more completely after a purge of openafs-fileserver. Remove
the files created by the server in /var/lib/openafs/local, remove
KeyFile and CellServDB.old from /etc/openafs/server, and remove
BosConfig from /etc/openafs.
Russ Allbery [Thu, 21 Jun 2007 14:15:59 +0000 (14:15 +0000)]
* Use LSB logging functions for openafs-fileserver and support passing
arguments to bosserver in /etc/default/openafs-fileserver by setting
DAEMON_ARGS.
Russ Allbery [Thu, 21 Jun 2007 09:34:55 +0000 (09:34 +0000)]
Also drop --pidfile from the openafs-fileserver init script, since this
option is for when the daemon writes out its own PID file and bosserver
doesn't do that.
Russ Allbery [Thu, 21 Jun 2007 09:31:35 +0000 (09:31 +0000)]
Also drop --pidfile from the openafs-fileserver init script, since this
option is for when the daemon writes out its own PID file and bosserver
doesn't do that.
Russ Allbery [Tue, 19 Jun 2007 11:02:27 +0000 (11:02 +0000)]
* Move upserver and upclient to /usr/lib/openafs (from /usr/sbin) to
match upstream's layout and for consistency with the other servers run
via bosserver.
Russ Allbery [Tue, 19 Jun 2007 10:59:08 +0000 (10:59 +0000)]
* Move /etc/openafs/server-local to /var/lib/openafs/local. Most of the
files here are automatically generated and hence not configuration
files, and the file server has to write to this directory to start (so
having it in /etc breaks read-only / file systems). Thanks, Alexei
Sheplyakov. (Closes: #405697)
Russ Allbery [Tue, 12 Jun 2007 18:41:41 +0000 (18:41 +0000)]
* Apply upstream patches for Linux 2.6.22 support. (Closes: #428401)
* Apply upstream patch to fix the AFS /proc structure under Linux
2.6.22.1 and later.
Russ Allbery [Sun, 20 May 2007 04:59:04 +0000 (04:59 +0000)]
* Go back to using krb5-config. The dependency problem has now been
fixed on all platforms in the krb5 package and this is the preferred
upstream method.
Russ Allbery [Thu, 17 May 2007 23:23:23 +0000 (23:23 +0000)]
Upstream fixed the problem that was causing these files to be generated
in the distribution tarball a while back and we've just been carrying
them via merges. Drop them completely.
Russ Allbery [Thu, 17 May 2007 01:14:14 +0000 (01:14 +0000)]
* When afsd is started with -rmtsys, it starts another process that afsd
-shutdown doesn't kill. Update the init script to kill this process
on stop. (Closes: #405982)
Russ Allbery [Wed, 16 May 2007 23:55:36 +0000 (23:55 +0000)]
* Always generate a UUID on client startup even if for some reason we
don't think we have any network interfaces. Otherwise, the client
will use an all-zeroes UUID, which will collide with every other
client with the same problem.
Russ Allbery [Wed, 16 May 2007 23:53:15 +0000 (23:53 +0000)]
* Don't use krb5-config to get a library list. It runs the risk of
encoding unnecessary library dependencies. Instead, hard-code the
flags and libraries we actually need. (Closes: #424448)
* Debian kernels as of 2.6.20 enable CONFIG_PARAVIRT, which redefines
several core kernel calls as redirects through a paravirt.ops table
that's marked GPL-only. This breaks all non-GPL modules that used
those (previously generally exported) calls even indirectly. Apply a
hack to switch from spin_lock_irq to spin_lock_irqsave to avoid the
paravirt redirection so that the module will build.
* Enable ucontext for all Linux builds with glibc 2.4 or later, not just
with i386 and amd64. Newer versions of glibc change internal
structures in ways that cause the old LWP code to have stack
overflows. (Closes: #419903)
* When determining the AFS sysname for kernel module builds, add
explicit cases for 2.6 kernels for all architectures that support them
(everything except non-64-bit sparc_linux). Fixes build problems with
ia64 and probably other architectures. Thanks to Ippokratis Pandis
for the debugging help. (Closes: #411818)
The missing aclocal looks like it also caused the failures on SPARC,
since it meant we missed the change that forced SPARC to use the right
thread flags.
* Build-depend on Automake since we run aclocal during the build. Not
having aclocal caused local patches to osconf.m4 to not be effective,
leading to the wrong compiler flags for shared libraries. Thanks to
Frank Lichtenheld for the report. (Closes: #416398)
Russ Allbery [Tue, 27 Mar 2007 01:54:30 +0000 (01:54 +0000)]
* Remove the rx_Init calls in the PAM module. The internal counters
that had to be initialized that way have been removed.
* Now that we're running regen.sh as part of the build process, only
patch the Autoconf source files and not the generated output to make
the diff easier to audit.
Russ Allbery [Tue, 27 Mar 2007 01:14:32 +0000 (01:14 +0000)]
* New upstream release. (Closes: #415699)
- Support newer Linux kernels. (Closes: #409797, #410120)
- Add aklog fallbacks for null realms to support the referral
capability in MIT Kerberos 1.6 and later. (Closes: #410314)
Russ Allbery [Mon, 19 Mar 2007 21:06:14 +0000 (21:06 +0000)]
* SECURITY: Apply upstream patch to disable setuid status on all cells
by default. Prior versions of AFS defaulted to honoring setuid bits
in the local cell, but since unauthenticated file access in AFS is
unencrypted, an attacker could forge packets from an AFS file server
to synthesize a setuid binary in AFS.
Russ Allbery [Sun, 18 Mar 2007 02:03:08 +0000 (02:03 +0000)]
* Apply upstream fix to show reasonable free space numbers for AFS in
df. Without this fix, some programs which use df to check free space
may think that directories in AFS are full and prevent the user from
attempting to write files. (Closes: #415294)