How did the afs_shl_ext_XXXX.dll module ever load on any platform?
Since the name afs_shl_ext.dll is not 8.3 we must specify the base name
explicitly. Otherwise, it won't load.
FIXES 37887
if you manage to patch the syscall table, then the setpag() in the
patched setgroups syscall and creates a new session keyring. it does
contain the right pag id but this isnt the intended behavior.
so setpag() becomes __setpag() which just does the group insertion.
(and perhaps it should be called something else?). the stat count for
the setpag operation might be more correct now (instead of also counting
the pag restores).
The previous code failed to specify the correct size of the REG_MULTI_SZ
being written to the registry when appending an additional name to the
list of Back Connection Host names. This would cause the written string
to contain
FOO\0B
instead of
FOO\0\BAR\0\0
this caused problems for Windows 2000 lsass.exe which failed to properly
allocate the buffer for RegQueryValueEx and failed to check that it did
not read beyond the buffer. If the list was not terminated with two
nuls an exception would be thrown that could result in a lock being
orphaned on the NTLMNameTable. This would in turn prevent authentications
to UNC resources from completing. As a result, the startup of
afsd_service.exe would fail if a GlobalAutoMapper drive was configured.
The work around is to set the correct value in the registry before
switching machine names or installing/removing the Microsoft Loopback
Adapter.
If afsd takes longer than the cifs timeout to complete an operation
the cifs client may break the virtual circuit and create a new one.
this forces all file handles to be closed and locks to be released.
Try to prevent the circuit from being destroyed by allowing the
CIFS lanmanworkstation SessTimeout value to enforce upper limits
on the Conn and Hard Dead Timeouts. Permit this automatic configuration
to be disabled by setting specific configuration values for timeouts
in the registry.
Move the establishment of the GlobalAutoMapper drives to a separate
thread in order to prevent OpenAFS from reporting that the service did
not start.
OpenAFS considers the GlobalAutoMapper as deprecated functionality so
it is ok to avoid the service failure report.
Kill the remaining references to afs/longc_procs.h so that head will
actually compile (whoops). For a header that no longer did anything, we
sure included it in a lot of places.
Take advantage of the new platform infrastructure to move fs_conv_sol26 out
of the vol directory and remove the annoying Autoconf support for
optionally build it in favor of the platform support. We no longer care
about upgrading inode file servers on DUX, so make fs_conv_DUX4 go away.
The param.*_usr.h files are obsolete and no longer referenced. All of them
except for param.ia64_hpux1123_usr.h contained #error directives saying not
to use them.
