Russ Allbery [Sun, 14 Aug 2005 04:15:05 +0000 (04:15 +0000)]
Redo all of the patch sets and move them into a subdirectory. Add
explanations of the status of the various patches relative to upstream.
Add the new PAM patch.
Russ Allbery [Fri, 12 Aug 2005 04:00:29 +0000 (04:00 +0000)]
Remove the old PAM patches and the makefile logic to build the PIC
objects. Also remove some other now-unnecessary divergences from
upstream to reduce the size of the diff.
Russ Allbery [Thu, 11 Aug 2005 17:40:23 +0000 (17:40 +0000)]
Point readers at the comments in /etc/openafs/afs.conf as well. Update
the status of the man page work. Make slight updates to the wording and
capitalization in a few other places.
Russ Allbery [Thu, 11 Aug 2005 17:34:35 +0000 (17:34 +0000)]
- Add an AFS_PRE_SHUTDOWN hook called before shutting down the client.
- Provide a shutdown hook, disabled by default, that kills all
processes with files open in /afs. Enabling this ensures a more
reliable clean shutdown and umounting of the cache partition, but
may not always be what the admin wants.
This duplicates the functionality in the upstream init script (where it
calls /usr/vice/etc/killafs if present), but in a more general way and
with better documentation.
Russ Allbery [Thu, 11 Aug 2005 07:58:44 +0000 (07:58 +0000)]
* Rework how PAM modules are handled.
- Build against the pthread-aware libraries. (Closes: #297804)
- Use the shared library objects instead of building separate PIC
objects, eliminating the need for many Makefile modifications.
- Include pam_afs.krb.so as well. (Closes: #165972)
This is still a work in progress, but I want to get it into the repository
as a checkpoint since everything seems to work. Remaining to be done is
the cleanup of the old Makefile rules to build the pic directory, which
should no longer be needed.
Tested with unstable ssh-krb5 and login so far, with no problems.
Russ Allbery [Thu, 11 Aug 2005 01:35:25 +0000 (01:35 +0000)]
* Don't include the empty /usr/src/modules directory in
openafs-modules-source, only the tarball, to match how other kernel
module source packages behave. (Closes: #244685)
Russ Allbery [Thu, 11 Aug 2005 01:12:16 +0000 (01:12 +0000)]
- Update the default client options based on the cache tuning synopsis
by Jeffrey Hutzelman, including setting a larger chunksize for all
default option sets. (Closes: #303944)
Russ Allbery [Thu, 11 Aug 2005 00:29:20 +0000 (00:29 +0000)]
* Improve the openafs-client init script.
- Choose the client options based on the cache size by default.
- Move the code to build the options string to the init script from
afs.conf so that afs.conf is pure configuration.
- Improve the readability of the output on start.
- Significantly improve the comments in afs.conf.
- Remove the code to set -nosettime as it's now the default.
* Move the openafs-client init script and afs.conf file we use into the
debian directory rather than patching the upstream version. We've
diverged so much that there's no point in continuing to merge changes.
Russ Allbery [Wed, 10 Aug 2005 21:48:42 +0000 (21:48 +0000)]
* Diagnose unsupported architectures earlier and with a clearer error.
* Suppress error messages from a missing kernel version header since
they occur normally when doing the regular package build and are just
confusing. Instead, print an error in kern-sysname and abort if the
kernel version could not be determined.
====================
This delta was composed from multiple commits as part of the CVS->Git migration.
The checkin message with each commit was inconsistent.
The following are the additional commit messages.
====================
FIXES 20392
The attached patch restores the behaviour of the 1.2.x client: writes
appending to a file do not trigger a 0-length read which at best updates the
current status. If another cache manager wrote to the same file in the
meantime, the file status is updated only after the StoreData RPC (and for
the changes the last one wins).
The attached patch restores the behaviour of the 1.2.x client: writes
appending to a file do not trigger a 0-length read which at best updates the
current status. If another cache manager wrote to the same file in the
meantime, the file status is updated only after the StoreData RPC (and for
the changes the last one wins).
addresses:
byte-range locks work (mildly tested)
fixes a reference counting error
can shutdown/restart client
code formatting
major speed improvements
fixes delete operation problem
internal locking in more places
i have reviewed this patch myself carefully. specifically, please review
the changes to cm_buf.c and cm_callback.c. in cm_buf, i added the looping
code because i ran into the following assert once. i am not sure why, and
my attempt to diagnose the problem was not successful.
apart from the byte-range locking code, the code has not change for quite
a bit.
use the generic read/write functions, but set the superblock up correctly.
====================
This delta was composed from multiple commits as part of the CVS->Git migration.
The checkin message with each commit was inconsistent.
The following are the additional commit messages.
====================
FIXES 20396
I have built openafs 1.3.85 on an emt64 platform on RedHat Enterprise 4,
and found that the aklog supplied with 1.3.85 (not the one in the afs-krb5
conversion kit) consistently segfaults when aklog is called with the
"-path ...." parm. It works properly when called with no parms.
I traced this down to the call to the readlink() call in aklog_main.c,
when the params structure references were removed the nesting was
changed on an if statement that changed the logic and allowed the
do while loop to loop past the end and de-reference a NULL pointer.
Fix various minor problems with aklog, including the return type from main,
removal of #ifndef __STDC__ code, missing includes, removal of unused
variables, correct number of arguments to pr_Initialize, correct argument
type for pr_SNameToId, and use of getcwd instead of getwd.
i slipped a rollback on the linux 2.4 dentry stuff for osi_UFS* in here
at the same time. this patch lets us use kernel inodes instead of our own pool.
* Make bosserver less picky about permissions and remove the attempt to
set /etc/openafs/server-local not-world-readable since dh_fixperms was
overridding it anyway. (Closes: #312921)
====================
This delta was composed from multiple commits as part of the CVS->Git migration.
The checkin message with each commit was inconsistent.
The following are the additional commit messages.
====================
The previous patch to pr_Initialize() opened an opportunity for
afsconf_Check() to be called with a NULL pointer if tdir == 0.
Prevent this occurrance.
====================
This delta was composed from multiple commits as part of the CVS->Git migration.
The checkin message with each commit was inconsistent.
The following are the additional commit messages.
====================
fixed one null pointer reference but not the other.