Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Marko Rößler [EMAIL PROTECTED] writes: Well, it seams to solve the key_sched.c problem. There is another problem with the following Backtrace: snip Starting program: /tmp/openafs-1.4.4.dfsg1/src/kauth/klog Password: stackcheck = 0: stack = 0 topstack = 0xffcfbb1c: stackptr = 0xf7c5a008: stacksize = 0x48000 Thu Jul 12 09:18:14 2007 LWP: stack overflow in process IO MANAGER! We've tracked this problem down and it goes away with either a 64-bit userland build or with some really hideous and scary pointer demangling code in OpenAFS STABLE that will probably break again with later glibc. I'm hoping to get approval from the SPARC porters to just build OpenAFS 64-bit so that it can use getcontext/savecontext like every other architecture, but failing that, this will be fixed as soon as I package a 1.4.5 release candidate (which we're waiting on a few other bug fixes for). -- Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Marko Rößler [EMAIL PROTECTED] writes: Well, it seams to solve the key_sched.c problem. There is another problem with the following Backtrace: snip Starting program: /tmp/openafs-1.4.4.dfsg1/src/kauth/klog Password: stackcheck = 0: stack = 0 topstack = 0xffcfbb1c: stackptr = 0xf7c5a008: stacksize = 0x48000 Thu Jul 12 09:18:14 2007 LWP: stack overflow in process IO MANAGER! Program received signal SIGABRT, Aborted. We've tracked this problem down, but unfortunately there isn't great news. This is a different problem, related to the glibc upgrade. OpenAFS supports two ways of implementing its LWP threading library: through knowing way too much about glibc internals, which broke with glibc 2.3 when glibc started XORing stack pointers, and through using getcontext/savecontext. We switched all platforms over to using getcontext/savecontext for glibc 2.3 and higher. However, getcontext/savecontext is not implemented for sparc32, only for sparc64. I'm currently doing a sparc64 build just to confirm that solves the problem. If it does, I'm going to check on debian-devel to see if it's okay at this point to build a sparc64 package for the Debian sparc architecture. I think that it may be at this point given that the kernel team is dropping support for the sparc32 kernels, but it's likely to cause at least some consternation. -- Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Russ Allbery schrieb: Okay, one more try at this. One of the other OpenAFS developers found a SPARC Linux box with AFS installed and spent some time tracking this down. Could you try this patch instead of all the previous patches I sent you (in other words, try applying it to a fresh copy of the source retrieved via apt-get source openafs) and see if this clears up the problem? Well, it seams to solve the key_sched.c problem. There is another problem with the following Backtrace: snip Starting program: /tmp/openafs-1.4.4.dfsg1/src/kauth/klog Password: stackcheck = 0: stack = 0 topstack = 0xffcfbb1c: stackptr = 0xf7c5a008: stacksize = 0x48000 Thu Jul 12 09:18:14 2007 LWP: stack overflow in process IO MANAGER! Program received signal SIGABRT, Aborted. 0xf7e09120 in raise () from /lib/libc.so.6 (gdb) bt #0 0xf7e09120 in raise () from /lib/libc.so.6 #1 0xf7e0a660 in abort () from /lib/libc.so.6 #2 0x000465e4 in Dispatcher () #3 0x00046e6c in savecontext () #4 0x00046b24 in LWP_CreateProcess () #5 0x00047974 in IOMGR_Initialize () #6 0x00045c3c in rxi_InitializeThreadSupport () #7 0x0003e550 in rx_InitHost () #8 0x0001378c in ka_AuthServerConn () #9 0x0001a5e4 in ka_GetAuthToken () #10 0x00013ef4 in GetTickets () #11 0x00014050 in ka_UserAuthenticateGeneral () #12 0x000120b0 in CommandProc () #13 0x0004cd1c in cmd_Dispatch () #14 0x00011bd0 in main () snipp I am not sure whether this still belongs to the bug report as it happens independently from the password length. Cheers, Marko -- --- Bitte beachten Sie die neue Telefonnummer --- _ _ __Marko Rößler @ Technische Universität Chemnitz / \ / \ | \_ \_ |_ Tel. (49) 371 53133590 *neu* \ \ | Fax(49) 371 531833590 *neu* \_/ \_/ |__Mail [EMAIL PROTECTED] WEBhttp://www.tu-chemnitz.de/etit/sse signature.asc Description: OpenPGP digital signature
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Marko Rößler [EMAIL PROTECTED] writes: Well, it seams to solve the key_sched.c problem. There is another problem with the following Backtrace: snip Starting program: /tmp/openafs-1.4.4.dfsg1/src/kauth/klog Password: stackcheck = 0: stack = 0 topstack = 0xffcfbb1c: stackptr = 0xf7c5a008: stacksize = 0x48000 Thu Jul 12 09:18:14 2007 LWP: stack overflow in process IO MANAGER! That's the standard UCONTEXT problem, and I could have sworn I already fixed that. Hm. Did you build the new version by running debian/rules build or by running configure and make directly? As near as I can tell, the sparc_linux24 parameter file should take care of this problem (by defining UCONTEXT) and so should the sparc64_linux26 parameter file, but sparc64_linux24 may still have this problem. -- Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Marko Rößler [EMAIL PROTECTED] writes:
Russ Allbery schrieb:
Could you try this patch as well and see if it helps any? It's
unfortunately less likely to help than the previous one, but it's
possible that it's the issue. There's some weird gcc optimizer issue
going on. This helps on Mac OS X, at least.
I am afraid but this patch does not help either.
The bus error happens at line 236 in src/des/key_sched.c
236 if ((unsigned)Key[(int)*kp++])
237 temp |= (1 BIT(31));
Okay, one more try at this. One of the other OpenAFS developers found a
SPARC Linux box with AFS installed and spent some time tracking this
down.
Could you try this patch instead of all the previous patches I sent you
(in other words, try applying it to a fresh copy of the source retrieved
via apt-get source openafs) and see if this clears up the problem?
diff -ru openafs-1.4.4/src/des/des.h openafs-1.4.4.patched/src/des/des.h
--- openafs-1.4.4/src/des/des.h 2003-07-15 23:14:59.0 +
+++ openafs-1.4.4.patched/src/des/des.h 2007-07-04 14:35:55.0 +
@@ -20,7 +20,10 @@
typedef unsigned char des_cblock[8]; /* crypto-block size */
/* Key schedule */
typedef struct des_ks_struct {
+union {
+afs_uint32 i[2];
des_cblock _;
+} _;
} des_key_schedule[16];
#define DES_KEY_SZ (sizeof(des_cblock))
diff -ru openafs-1.4.4/src/des/pcbc_encrypt.c
openafs-1.4.4.patched/src/des/pcbc_encrypt.c
--- openafs-1.4.4/src/des/pcbc_encrypt.c2004-10-18 07:11:56.0
+
+++ openafs-1.4.4.patched/src/des/pcbc_encrypt.c2007-07-04
14:36:33.0 +
@@ -24,9 +24,9 @@
#ifndef KERNEL
#include stdio.h
#endif
-#include des.h
#include afsconfig.h
#include afs/param.h
+#include des.h
#include des_prototypes.h
RCSID
--
Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Russ Allbery schrieb: Could you try this patch as well and see if it helps any? It's unfortunately less likely to help than the previous one, but it's possible that it's the issue. There's some weird gcc optimizer issue going on. This helps on Mac OS X, at least. I am afraid but this patch does not help either. The bus error happens at line 236 in src/des/key_sched.c 236 if ((unsigned)Key[(int)*kp++]) 237 temp |= (1 BIT(31)); Hope that helps, Marko -- Marko Rößler @ Chemnitz University of Technology _ _ __ / \ / \ | Phone (49) 371 53133590 \_ \_ |_ Fax(49) 371 531833590 \ \ | Mail [EMAIL PROTECTED] \_/ \_/ |__WEBhttp://www.tu-chemnitz.de/etit/sse GPG1024D/F81F1B7F signature.asc Description: OpenPGP digital signature
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Marko Rößler [EMAIL PROTECTED] writes: I applied the patch, but unfortunately it didn't change the behaviour. Could you try this patch as well and see if it helps any? It's unfortunately less likely to help than the previous one, but it's possible that it's the issue. There's some weird gcc optimizer issue going on. This helps on Mac OS X, at least. === RCS file: /cvs/openafs/src/des/des.c,v retrieving revision 1.11.2.5 retrieving revision 1.11.2.6 diff -u -r1.11.2.5 -r1.11.2.6 --- openafs/src/des/des.c 2005/08/18 04:56:14 1.11.2.5 +++ openafs/src/des/des.c 2007/07/03 04:24:45 1.11.2.6 @@ -37,7 +37,7 @@ #include afs/param.h RCSID -($Header: /cvs/openafs/src/des/des.c,v 1.11.2.5 2005/08/18 04:56:14 shadow Exp $); +($Header: /cvs/openafs/src/des/des.c,v 1.11.2.6 2007/07/03 04:24:45 shadow Exp $); #ifndef KERNEL #include stdio.h @@ -90,7 +90,7 @@ #error dunno how to do this machine type, you lose; #endif #endif -afs_uint32 P_temp; +volatile afs_uint32 P_temp; volatile unsigned char *P_temp_p = (unsigned char *)P_temp; #ifdef BITS16 sbox_out S_out; -- Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
I applied the patch, but unfortunately it didn't change the behaviour.
Russ Allbery schrieb:
Russ Allbery [EMAIL PROTECTED] writes:
Marko Rößler [EMAIL PROTECTED] writes:
finaly i managed to compile the openafs-client with debug symbols and
received the following BackTrace:
Sorry about the delay in responding. I've forwarded this upstream. It
looks to me like there may be some sort of alignment issue in the DES
library, although I'm not sure why this would just start happening now.
Can you try building with the following patch and let me know if it helps?
=== src/des/des.h
==
--- src/des/des.h (revision 2131)
+++ src/des/des.h (local)
@@ -19,8 +19,9 @@
typedef unsigned char des_cblock[8]; /* crypto-block size */
/* Key schedule */
+/* This hack gives us int alignment, which the DES library requires. */
typedef struct des_ks_struct {
-des_cblock _;
+int _[8/sizeof(int)];
} des_key_schedule[16];
#define DES_KEY_SZ (sizeof(des_cblock))
--
Marko Rößler @ Chemnitz University of Technology
_ _ __
/ \ / \ | Phone (49) 371 53133590
\_ \_ |_ Fax(49) 371 531833590
\ \ | Mail [EMAIL PROTECTED]
\_/ \_/ |__WEBhttp://www.tu-chemnitz.de/etit/sse
GPG1024D/F81F1B7F
signature.asc
Description: OpenPGP digital signature
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Marko Rößler [EMAIL PROTECTED] writes: finaly i managed to compile the openafs-client with debug symbols and received the following BackTrace: Sorry about the delay in responding. I've forwarded this upstream. It looks to me like there may be some sort of alignment issue in the DES library, although I'm not sure why this would just start happening now. -- Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Russ Allbery [EMAIL PROTECTED] writes:
Marko Rößler [EMAIL PROTECTED] writes:
finaly i managed to compile the openafs-client with debug symbols and
received the following BackTrace:
Sorry about the delay in responding. I've forwarded this upstream. It
looks to me like there may be some sort of alignment issue in the DES
library, although I'm not sure why this would just start happening now.
Can you try building with the following patch and let me know if it helps?
=== src/des/des.h
==
--- src/des/des.h (revision 2131)
+++ src/des/des.h (local)
@@ -19,8 +19,9 @@
typedef unsigned char des_cblock[8]; /* crypto-block size */
/* Key schedule */
+/* This hack gives us int alignment, which the DES library requires. */
typedef struct des_ks_struct {
-des_cblock _;
+int _[8/sizeof(int)];
} des_key_schedule[16];
#define DES_KEY_SZ (sizeof(des_cblock))
--
Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Hi Russ, finaly i managed to compile the openafs-client with debug symbols and received the following BackTrace: (gdb) run Starting program: /var/cache/apt-build/build/openafs-1.4.4.dfsg1/src/kauth/klog Password: Program received signal SIGBUS, Bus error. 0x0004a2a8 in des_key_sched () (gdb) bt #0 0x0004a2a8 in des_key_sched () #1 0x00012708 in ka_StringToKey () #2 0x00013f34 in ka_UserAuthenticateGeneral () #3 0x00012008 in CommandProc () #4 0x0004 in cmd_Dispatch () #5 0x00011b28 in main () Can i help any further? Cheers, Marko On Mon, 21 May 2007, Russ Allbery wrote: Marko Rößler [EMAIL PROTECTED] writes: i did send a core dump to the bugtracking system... Hm, it wasn't received, but it was possibly too large. A backtrace is more useful anyway. I am not shure how to create a backtrace. Can you please quickly point me the way out? Install gdb and then run: gdb /usr/bin/klog and then at the gdb prompt, type: run username where username is the username you're authenticating as. Enter the password that causes it to die, and gdb should catch the bus error. At that point, you should be able to type: bt to get a backtrace. -- Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/ -- Please take notice of the new telephone number Marko Rößler @ Chemnitz University of Technology _ _ __ / \ / \ | Phone (49) 371 53133590 *new* \_ \_ |_ Fax(49) 371 531833590 *new* \ \ | Mail [EMAIL PROTECTED] \_/ \_/ |__WEBhttp://www.tu-chemnitz.de/etit/sse GPG1024D/F81F1B7F -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Marko Rößler [EMAIL PROTECTED] writes: i did send a core dump to the bugtracking system... Hm, it wasn't received, but it was possibly too large. A backtrace is more useful anyway. I am not shure how to create a backtrace. Can you please quickly point me the way out? Install gdb and then run: gdb /usr/bin/klog and then at the gdb prompt, type: run username where username is the username you're authenticating as. Enter the password that causes it to die, and gdb should catch the bus error. At that point, you should be able to type: bt to get a backtrace. -- Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Hi Marko, I haven't heard anything back from you about this bug. Are you still seeing it? If so, can I get the backtrace from klog if possible? I think this problem is SPARC-specific and don't have the hardware to duplicate it, but I can at least forward it to the openafs-devel mailing list. Thanks. -- Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Package: openafs-client Version: 1.4.2-6 Severity: normal -- System Information: Debian Release: 4.0 APT prefers unstable APT policy: (500, 'unstable') Architecture: sparc (sparc64) Shell: /bin/sh linked to /bin/bash Kernel: Linux 2.6.18-4-sparc64 Locale: [EMAIL PROTECTED], [EMAIL PROTECTED] (charmap=ISO-8859-15) Versions of packages openafs-client depends on: ii debconf [debconf-2.0] 1.5.13 Debian configuration management sy ii libc6 2.3.6.ds1-13 GNU C Library: Shared libraries ii libncurses5 5.5-5Shared libraries for terminal hand Versions of packages openafs-client recommends: ii lsof4.77.dfsg.1-3List open files ii openafs-modules 1.4.2-6+2.6.18.dfsg.1-11 AFS distributed filesystem kernel ii openafs-modules 1.4.2-6 AFS distributed filesystem kernel -- debconf information: openafs-client/run-client: true openafs-client/crypt: true * openafs-client/cachesize: 5 openafs-client/cell-info: openafs-client/fakestat: true openafs-client/afsdb: true * openafs-client/dynroot: false * openafs-client/thiscell: tu-chemnitz.de -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#415952: openafs-client: klog fails with an bus access error for password longer then 8 symbols.
Marko Roessler [EMAIL PROTECTED] writes: Package: openafs-client Version: 1.4.2-6 Severity: normal My guess is that this is SPARC-specific, since I can't seem to duplicate it. Can you get a core dump from klog and get a backtrace from that? -- Russ Allbery ([EMAIL PROTECTED]) http://www.eyrie.org/~eagle/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
