Skip to contentSkip to conversation

2.4.21: Optionally Configure UAC Policy via Sysctl (Alpha)

PATCH In review Tavis Ormandy 2 participants 3 messages Jul 10, 2003

lore
Share & tools
lore

Tavis Ormandy

author
#0 lore raw

PATCH 1/1 2.4.21: Optionally Configure UAC Policy via Sysctl (Alpha)

Hi there, this patch adds a configuration option in General Setup for Alpha users, "CONFIG_ALPHA_UAC_SYSCTL", which will allow the kernel UAC policy to be modified at runtime via sysctl's.

The sysctl's created are:

/proc/sys/kernel/uac/noprint
/proc/sys/kernel/uac/sigbus
/proc/sys/kernel/uac/nofix

which are initialised to the existing policy. I've added some text for the Help screen.

Benefits:

* No longer need to recompile to change UAC policy.
* Disabling the 'unaligned trap' messages usually requires changing the loglevel, which sacrifices other useful messages.
* UAC_SIGBUS can be turned on and off at will, which is quite handy for debugging.

Suggestions/comments/flames welcome, im considering submitting to Marcelo (so any tips welcome). Patch (against 2.4.21) below sig.

Thanks, Tavis.

Signature · 209 lines
-------------------------------------
[email protected] | finger me for my gpg key.
-------------------------------------------------------

diff -ruN linux-2.4.21.orig/arch/alpha/config.in linux-2.4.21/arch/alpha/config.in
--- linux-2.4.21.orig/arch/alpha/config.in	2003-06-13 15:51:29.000000000 +0100
+++ linux-2.4.21/arch/alpha/config.in	2003-07-10 07:45:21.000000000 +0100
@@ -294,6 +294,9 @@
 fi
 if [ "$CONFIG_PROC_FS" != "n" ]; then
    tristate 'SRM environment through procfs' CONFIG_SRM_ENV
+   if [ "$CONFIG_SYSCTL" != "n" ]; then
+	   bool 'Configure uac policy via sysctl' CONFIG_ALPHA_UAC_SYSCTL
+   fi 
 fi
  
 tristate 'Kernel support for a.out (ECOFF) binaries' CONFIG_BINFMT_AOUT
diff -ruN linux-2.4.21.orig/arch/alpha/kernel/traps.c linux-2.4.21/arch/alpha/kernel/traps.c
--- linux-2.4.21.orig/arch/alpha/kernel/traps.c	2003-06-13 15:51:29.000000000 +0100
+++ linux-2.4.21/arch/alpha/kernel/traps.c	2003-07-10 07:43:38.000000000 +0100
@@ -14,6 +14,8 @@
 #include <linux/tty.h>
 #include <linux/delay.h>
 #include <linux/smp_lock.h>
+#include <linux/sysctl.h>
+#include <linux/init.h>
 
 #include <asm/gentrap.h>
 #include <asm/uaccess.h>
@@ -32,6 +34,43 @@
 static int opDEC_checked = 0;
 static unsigned long opDEC_test_pc = 0;
 
+#ifdef CONFIG_ALPHA_UAC_SYSCTL
+static struct ctl_table_header *uac_sysctl_header;
+
+static int enabled_noprint = 0;
+static int enabled_sigbus = 0;
+static int enabled_nofix = 0;
+
+ctl_table uac_table[] = {
+	{KERN_UAC_NOPRINT, "noprint", &enabled_noprint, sizeof (int), 0644, NULL, &proc_dointvec},
+	{KERN_UAC_SIGBUS, "sigbus", &enabled_sigbus, sizeof (int), 0644, NULL, &proc_dointvec},
+	{KERN_UAC_NOFIX, "nofix", &enabled_nofix, sizeof (int), 0644, NULL, &proc_dointvec},
+        {0}
+};
+
+static int __init init_uac_sysctl(void)
+{
+	unsigned long uac_bits;
+
+	/* get the defined UAC_POLICY */
+        uac_bits = (current->thread.flags >> UAC_SHIFT) & UAC_BITMASK;
+
+	/* now initialize sysctl's with that policy */
+	enabled_noprint = (uac_bits & UAC_NOPRINT) ? 1 : 0;
+	enabled_sigbus = (uac_bits & UAC_SIGBUS) ? 1 : 0;
+	enabled_nofix = (uac_bits & UAC_NOFIX) ? 1 : 0;
+
+	/* save this for later so we can clean up */	
+	uac_sysctl_header = register_sysctl_table(uac_table, 0);
+	return 0;
+}
+
+static void __exit exit_uac_sysctl(void)
+{
+	unregister_sysctl_table(uac_sysctl_header);
+}
+#endif
+
 static void
 opDEC_check(void)
 {
@@ -710,19 +749,24 @@
 do_entUnaUser(void * va, unsigned long opcode,
 	      unsigned long reg, struct pt_regs *regs)
 {
-	static int cnt = 0;
-	static long last_time = 0;
-
 	unsigned long tmp1, tmp2, tmp3, tmp4;
 	unsigned long fake_reg, *reg_addr = &fake_reg;
-	unsigned long uac_bits;
 	long error;
+	static int cnt = 0;
+	static long last_time = 0;
 
+#ifndef CONFIG_ALPHA_UAC_SYSCTL
+	unsigned long uac_bits;
+	
 	/* Check the UAC bits to decide what the user wants us to do
 	   with the unaliged access.  */
 
 	uac_bits = (current->thread.flags >> UAC_SHIFT) & UAC_BITMASK;
 	if (!(uac_bits & UAC_NOPRINT)) {
+#else
+	/* check systcls for uac policy */
+	if (!(enabled_noprint)) {
+#endif
 		if (cnt >= 5 && jiffies - last_time > 5*HZ) {
 			cnt = 0;
 		}
@@ -733,10 +777,18 @@
 		}
 		last_time = jiffies;
 	}
+#ifndef CONFIG_ALPHA_UAC_SYSCTL
 	if (uac_bits & UAC_SIGBUS) {
+#else
+	if (enabled_sigbus) {
+#endif
 		goto give_sigbus;
 	}
+#ifndef CONFIG_ALPHA_UAC_SYSCTL
 	if (uac_bits & UAC_NOFIX) {
+#else
+	if (enabled_nofix) {
+#endif
 		/* Not sure why you'd want to use this, but... */
 		return;
 	}
diff -ruN linux-2.4.21.orig/Documentation/Configure.help linux-2.4.21/Documentation/Configure.help
--- linux-2.4.21.orig/Documentation/Configure.help	2003-06-13 15:51:29.000000000 +0100
+++ linux-2.4.21/Documentation/Configure.help	2003-07-10 07:40:03.000000000 +0100
@@ -3372,6 +3372,29 @@
 
   Say N unless you know you need gobs and gobs of vmalloc space.
 
+Configure UAC policy via sysctl
+CONFIG_ALPHA_UAC_SYSCTL
+  Configuring the UAC policy on a Linux system usually involves
+  setting a compile time define, if you say Y here and also to 
+  sysctl support above, you will be able to modify the UAC policy
+  at runtime using the /proc interface. 
+
+  The UAC Policy defines the action Linux should take when an 
+  unaligned memory access occurs, the action can include printing
+  a warning message (NOPRINT), sending a signal to help developers 
+  debug their applications (SIGBUS), and you can disable the 
+  transparent fixing (NOFIX).
+
+  The sysctl's will be initialized to the defined UAC policy, you
+  can change these manually, or with the sysctl(8) userspace 
+  utility.
+
+  To disable the warning messages at runtime, you might use
+
+  echo 1 > /proc/sys/kernel/uac/noprint
+
+  Say N if your not sure.
+
 Non-standard serial port support
 CONFIG_SERIAL_NONSTANDARD
   Say Y here if you have any non-standard serial boards -- boards
diff -ruN linux-2.4.21.orig/include/linux/sysctl.h linux-2.4.21/include/linux/sysctl.h
--- linux-2.4.21.orig/include/linux/sysctl.h	2003-06-13 15:51:39.000000000 +0100
+++ linux-2.4.21/include/linux/sysctl.h	2003-07-10 06:58:04.000000000 +0100
@@ -125,6 +125,9 @@
 	KERN_TAINTED=53,	/* int: various kernel tainted flags */
 	KERN_CADPID=54,		/* int: PID of the process to notify on CAD */
  	KERN_CORE_PATTERN=56,	/* string: pattern for core-files */
+#ifdef CONFIG_ALPHA_UAC_SYSCTL
+	KERN_UAC_POLICY=57,	/* uac policy */
+#endif
 };
 
 
@@ -146,6 +149,16 @@
 	VM_MAX_READAHEAD=13,    /* Max file readahead */
 };
 
+#ifdef CONFIG_ALPHA_UAC_SYSCTL
+/* /proc/sys/kernel/uac */
+enum
+{
+	/* UAC policy */
+	KERN_UAC_NOPRINT=1,    /* int: printk() on unaligned access */
+	KERN_UAC_SIGBUS=2,     /* int: send SIGBUS on unaligned access */
+	KERN_UAC_NOFIX=3,      /* int: dont fix. */
+};				
+#endif
 
 /* CTL_NET names: */
 enum
diff -ruN linux-2.4.21.orig/kernel/sysctl.c linux-2.4.21/kernel/sysctl.c
--- linux-2.4.21.orig/kernel/sysctl.c	2003-06-13 15:51:39.000000000 +0100
+++ linux-2.4.21/kernel/sysctl.c	2003-07-10 06:18:00.000000000 +0100
@@ -97,6 +97,11 @@
 extern int acct_parm[];
 #endif
 
+
+#ifdef CONFIG_ALPHA_UAC_SYSCTL
+extern ctl_table uac_table[];
+#endif
+
 extern int pgt_cache_water[];
 
 static int parse_table(int *, int, void *, size_t *, void *, size_t,
@@ -259,6 +264,10 @@
 	{KERN_S390_USER_DEBUG_LOGGING,"userprocess_debug",
 	 &sysctl_userprocess_debug,sizeof(int),0644,NULL,&proc_dointvec},
 #endif
+
+#ifdef CONFIG_ALPHA_UAC_SYSCTL
+	{KERN_UAC_POLICY, "uac", NULL, 0, 0555, uac_table},
+#endif
 	{0}
 };
#1 lore raw

Review

Quoted 3 lines · Tavis Ormandy
(quoted diff) +3 -0
+  To disable the warning messages at runtime, you might use
+
+  echo 1 > /proc/sys/kernel/uac/noprint

It's really bad idea. Unaligned traps are *bugs*, so we don't want to hide them. Especially system-wide.

You can do all the same per process, using existing OSF-compatible UAC control.

Ivan.

author
#2 lore raw
Quoted 8 lines · Ivan Kokshaysky

Review

Quoted 3 lines · Tavis Ormandy
(quoted diff) +3 -0
+  To disable the warning messages at runtime, you might use
+
+  echo 1 > /proc/sys/kernel/uac/noprint

It's really bad idea. Unaligned traps are *bugs*, so we don't want to hide them. Especially system-wide.

It's true that that they are bugs, but on such widespread errors that are transparently fixed by the kernel, do we really need to be reminded every time such an event occurs?

$ grep unaligned /proc/cpuinfo 
kernel unaligned acc    : 0 (pc=0,va=0)
user unaligned acc      : 4400 (pc=30281c8c434,va=30281db8f62)

This machine has been up just a few hours, and already a has fixed a lot of these errors. I really dont need the signal:noise ratio of my logs diluted by this, unless im searching for bugs, which is why a sysctl makes sense, imo.

Quoted 3 lines

You can do all the same per process, using existing OSF-compatible UAC control.

True, and thats _really_ useful for turning the messages _on_ for the application your working with/debugging/etc, system-wide messages just dont make sense to me.

Also, it is optional, so only users who foresee wanting to disable the messages will do so.

Signature · 3 lines
-------------------------------------
[email protected] | finger me for my gpg key.
-------------------------------------------------------