Samba works on a wide range of platforms but the interface all the platforms provide is not always compatible. This chapter contains platform-specific information about compiling and using samba.
HP's implementation of supplementary groups is, er, non-standard (for hysterical reasons). There are two group files, /etc/group and /etc/logingroup; the system maps UIDs to numbers using the former, but initgroups() reads the latter. Most system admins who know the ropes symlink /etc/group to /etc/logingroup (hard link doesn't work for reasons too stupid to go into here). initgroups() will complain if one of the groups you're in in /etc/logingroup has what it considers to be an invalid ID, which means outside the range [0..UID_MAX], where UID_MAX is (I think) 60000 currently on HP-UX. This precludes -2 and 65534, the usual 'nobody' GIDs.
If you encounter this problem, make sure that the programs that are failing to initgroups() be run as users not in any groups with GIDs outside the allowed range.
This is documented in the HP manual pages under setgroups(2) and passwd(4).
On HPUX you must use gcc or the HP Ansi compiler. The free compiler that comes with HP-UX is not Ansi compliant and cannot compile Samba.
If you run an old version of SCO Unix then you may need to get important TCP/IP patches for Samba to work correctly. Without the patch, you may encounter corrupt data transfers using samba.
The patch you need is UOD385 Connection Drivers SLS. It is available from SCO (ftp.sco.com, directory SLS, files uod385a.Z and uod385a.ltr.Z).
DNIX has a problem with seteuid() and setegid(). These routines are needed for Samba to work correctly, but they were left out of the DNIX C library for some reason.
For this reason Samba by default defines the macro NO_EID in the DNIX section of includes.h. This works around the problem in a limited way, but it is far from ideal, some things still won't work right.
To fix the problem properly you need to assemble the following two functions and then either add them to your C library or link them into Samba.
put this in the file setegid.s:
.globl _setegid _setegid: moveq #47,d0 movl #100,a0 moveq #1,d1 movl 4(sp),a1 trap #9 bccs 1$ jmp cerror 1$: clrl d0 rts
put this in the file seteuid.s:
.globl _seteuid _seteuid: moveq #47,d0 movl #100,a0 moveq #0,d1 movl 4(sp),a1 trap #9 bccs 1$ jmp cerror 1$: clrl d0 rts
after creating the above files you then assemble them using
as seteuid.s
as setegid.s
that should produce the files seteuid.o and setegid.o
then you need to add these to the LIBSM line in the DNIX section of the Samba Makefile. Your LIBSM line will then look something like this:
LIBSM = setegid.o seteuid.o -ln
You should then remove the line:
#define NO_EID
from the DNIX section of includes.h
By default RedHat Rembrandt-II during installation adds an entry to /etc/hosts as follows:
127.0.0.1 loopback "hostname"."domainname"
This causes Samba to loop back onto the loopback interface. The result is that Samba fails to communicate correctly with the world and therefor may fail to correctly negotiate who is the master browse list holder and who is the master browser.
Corrective Action: Delete the entry after the word loopback in the line starting 127.0.0.1
Disabling Sequential Read Ahead using vmtune -r 0 improves samba performance significally.
Some people have been experiencing problems with F_SETLKW64/fcntl when running samba on solaris. The built in file locking mechanism was not scalable. Performance would degrade to the point where processes would get into loops of trying to lock a file. It woul try a lock, then fail, then try again. The lock attempt was failing before the grant was occurring. So the visible manifestation of this would be a handful of processes stealing all of the CPU, and when they were trussed they would be stuck if F_SETLKW64 loops.
Sun released patches for Solaris 2.6, 8, and 9. The patch for Solaris 7 has not been released yet.
The patch revision for 2.6 is 105181-34 for 8 is 108528-19 and for 9 is 112233-04
After the install of these patches it is recommended to reconfigure and rebuild samba.
Thanks to Joe Meslovich for reporting