Skip to contentSkip to conversation

KERN_SETUID_DUMPABLE in /proc/sys/fs/

PATCH In review Kurt Garloff 5 participants 13 messages Mar 10, 2006 – Mar 13, 2006 Proc sysctl

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 KERN_SETUID_DUMPABLE in /proc/sys/fs/

Hi,

likely a merge error by patch ... which makes setuid_dumpable appear in /proc/sys/fs/ rather than kernel/. I assume it belongs to kernel as the naming suggests.

Patch is against 2.6.16-rc5-git9.

From: Kurt Garloff <[email protected]>
Subject: suid-dumpable ended up in wrong sysctl dir

Diffing in sysctl.c is tricky, using more context is recommended. suid_dumpable ended up in fs/ instead of kernel/ and the reason is likely a patch with too little context.

kernel/sysctl.c +8 -8
Index: linux-2.6.15/kernel/sysctl.c
===================================================================
@@ -647,12 +647,20 @@ static ctl_table kern_table[] = {
 		.data		= &randomize_va_space,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
 		.proc_handler	= &proc_dointvec,
 	},
 #endif
+	{
+		.ctl_name	= KERN_SETUID_DUMPABLE,
+		.procname	= "suid_dumpable",
+		.data		= &suid_dumpable,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec,
+	},
 #if defined(CONFIG_S390) && defined(CONFIG_SMP)
 	{
 		.ctl_name	= KERN_SPIN_RETRY,
 		.procname	= "spin_retry",
 		.data		= &spin_retry,
 		.maxlen		= sizeof (int),
@@ -1032,20 +1040,12 @@ static ctl_table fs_table[] = {
 		.procname	= "inotify",
 		.mode		= 0555,
 		.child		= inotify_table,
 	},
 #endif	
 #endif
-	{
-		.ctl_name	= KERN_SETUID_DUMPABLE,
-		.procname	= "suid_dumpable",
-		.data		= &suid_dumpable,
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.proc_handler	= &proc_dointvec,
-	},
 	{ .ctl_name = 0 }
 };
 
 static ctl_table debug_table[] = {
 	{ .ctl_name = 0 }
 };
Signature · 1 line
Kurt Garloff, Head Architect Linux, Novell Inc.
#1 lore raw
Quoted 4 lines · Kurt Garloff

Diffing in sysctl.c is tricky, using more context is recommended. suid_dumpable ended up in fs/ instead of kernel/ and the reason is likely a patch with too little context.

It's been in kernel/ since 2.6.13. What will break if we move it?

This is security-related. If we move it we risk unsecuring people's machines...

#2 lore raw
Quoted 10 lines · Andrew Morton
Quoted 4 lines · Kurt Garloff

Diffing in sysctl.c is tricky, using more context is recommended. suid_dumpable ended up in fs/ instead of kernel/ and the reason is likely a patch with too little context.

It's been in kernel/ since 2.6.13. What will break if we move it?

This is security-related. If we move it we risk unsecuring people's machines...

only a very little bit since the default value is "secure", the option is to make it "insecure"...
but yeah by this time we should just bite the bullet and rename the variable rather than move it about

#3 lore raw
Quoted 8 lines · Arjan van de Ven
Quoted 6 lines · Andrew Morton
Quoted 4 lines · Kurt Garloff

Diffing in sysctl.c is tricky, using more context is recommended. suid_dumpable ended up in fs/ instead of kernel/ and the reason is likely a patch with too little context.

Quoted 7 lines
Quoted 4 lines

It's been in kernel/ since 2.6.13. What will break if we move it?

This is security-related. If we move it we risk unsecuring people's machines...

only a very little bit since the default value is "secure", the option is to make it "insecure"...

OK, that's a good point.

Quoted 2 lines

but yeah by this time we should just bite the bullet and rename the variable rather than move it about

That wouldn't help - we'll still break existing scripts.

crap. I tend to think we leave it where it is - it's only a cosmetic irritation, isn't it?

#4 lore raw
Quoted 6 lines
Quoted 4 lines

but yeah by this time we should just bite the bullet and rename the variable rather than move it about

That wouldn't help - we'll still break existing scripts.

why? (I mean the KERN_ to FS_ rename in the kernel, that's not exported to userland anywhere)

#5 lore raw
Quoted 10 lines · Arjan van de Ven
Quoted 7 lines
Quoted 5 lines

but yeah by this time we should just bite the bullet and rename the variable rather than move it about

That wouldn't help - we'll still break existing scripts.

why? (I mean the KERN_ to FS_ rename in the kernel, that's not exported to userland anywhere)

oic. What are you expecting here? Reading skills?

#6 lore raw
Quoted 7 lines · Andrew Morton
Quoted 6 lines · Arjan van de Ven
Quoted 6 lines
Quoted 5 lines

but yeah by this time we should just bite the bullet and rename the variable rather than move it about

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

That wouldn't help - we'll still break existing scripts.

why? (I mean the KERN_ to FS_ rename in the kernel, that's not exported to userland anywhere)

oic. What are you expecting here? Reading skills?

lesson to self: only speak in diff -u form before morning coffee

the setuid-dumpable sysctl got misplaces due to a patch glitch. But it's part of the ABI now for some time so we need to keep it. This patch at least renames the variable inside the kernel to match the new place, and puts a comment in place to indicate that it's a known glitch

include/linux/sysctl.h +3 -1
diff -purN linux-2.6.16-rc5/include/linux/sysctl.h linux-2.6.16-rc5-setuid/include/linux/sysctl.h
@@ -144,7 +144,6 @@ enum
 	KERN_UNKNOWN_NMI_PANIC=66, /* int: unknown nmi panic flag */
 	KERN_BOOTLOADER_TYPE=67, /* int: boot loader type */
 	KERN_RANDOMIZE=68, /* int: randomize virtual address space */
-	KERN_SETUID_DUMPABLE=69, /* int: behaviour of dumps for setuid core */
 	KERN_SPIN_RETRY=70,	/* int: number of spinlock retries */
 	KERN_ACPI_VIDEO_FLAGS=71, /* int: flags for setting up video after ACPI sleep */
 };
@@ -759,6 +758,9 @@ enum
 	FS_AIO_NR=18,	/* current system-wide number of aio requests */
 	FS_AIO_MAX_NR=19,	/* system-wide maximum number of aio requests */
 	FS_INOTIFY=20,	/* inotify submenu */
+			/* Note: the following got misplaced but this mistake is 
+			   now part of the ABI */
+	FS_SETUID_DUMPABLE=21, /* int: behaviour of dumps for setuid core */
 };
 
 /* /proc/sys/fs/quota/ */
kernel/sysctl.c +1 -1
diff -purN linux-2.6.16-rc5/kernel/sysctl.c linux-2.6.16-rc5-setuid/kernel/sysctl.c
@@ -1022,7 +1022,7 @@ static ctl_table fs_table[] = {
 #endif	
 #endif
 	{
-		.ctl_name	= KERN_SETUID_DUMPABLE,
+		.ctl_name	= FS_SETUID_DUMPABLE,
 		.procname	= "suid_dumpable",
 		.data		= &suid_dumpable,
 		.maxlen		= sizeof(int),
#7 lore raw
Quoted 5 lines

the setuid-dumpable sysctl got misplaces due to a patch glitch. But it's part of the ABI now for some time so we need to keep it. This patch at least renames the variable inside the kernel to match the new place, and puts a comment in place to indicate that it's a known glitch

"To be corrected to the right thing for 2.7".

#8 lore raw
In reply to Arjan van de Ven
Quoted 16 lines · Arjan van de Ven

diff -purN linux-2.6.16-rc5/include/linux/sysctl.h linux-2.6.16-rc5-setuid/include/linux/sysctl.h

include/linux/sysctl.h 2006-03-11 09:02:13.000000000 renamed from include/linux/sysctl.h 2006-02-27 09:13:31.000000000 +0100 +1 -1
+0100
@@ -144,7 +144,6 @@ enum
 	KERN_UNKNOWN_NMI_PANIC=66, /* int: unknown nmi panic flag */
 	KERN_BOOTLOADER_TYPE=67, /* int: boot loader type */
 	KERN_RANDOMIZE=68, /* int: randomize virtual address space */
-	KERN_SETUID_DUMPABLE=69, /* int: behaviour of dumps for setuid core */
 	KERN_SPIN_RETRY=70,	/* int: number of spinlock retries */
 	KERN_ACPI_VIDEO_FLAGS=71, /* int: flags for setting up video after ACPI

sleep */
};

include/linux/sysctl.h 2006-03-11 09:02:13.000000000 enum · line 763

Quoted 6 lines
include/linux/sysctl.h 2006-03-11 09:02:13.000000000 renamed from include/linux/sysctl.h 2006-02-27 09:13:31.000000000 +0100 +3 -0
@@ -759,6 +758,9 @@ enum
 	FS_AIO_NR=18,	/* current system-wide number of aio requests */
 	FS_AIO_MAX_NR=19, /* system-wide maximum number of aio requests */
 	FS_INOTIFY=20,	/* inotify submenu */
+ /* Note: the following got misplaced but this mistake is
+			   now part of the ABI */
+	FS_SETUID_DUMPABLE=21, /* int: behaviour of dumps for setuid core */

This must be number 69 here. Or else we break the sys_sysctl ABI.

include/linux/sysctl.h 2006-03-11 09:02:13.000000000 enum · line 765

Quoted 2 lines
include/linux/sysctl.h 2006-03-11 09:02:13.000000000 renamed from include/linux/sysctl.h 2006-02-27 09:13:31.000000000 +0100 +0 -0
 };
 

Eric

#9 lore raw

Hi Arjan, Eric,

Quoted 15 lines · Eric W. Biederman
Quoted 13 lines · Arjan van de Ven

diff -purN linux-2.6.16-rc5/include/linux/sysctl.h linux-2.6.16-rc5-setuid/include/linux/sysctl.h

include/linux/sysctl.h 2006-03-11 09:02:13.000000000 renamed from include/linux/sysctl.h 2006-02-27 09:13:31.000000000 +0100 +1 -1
+0100
@@ -144,7 +144,6 @@ enum
 	KERN_UNKNOWN_NMI_PANIC=66, /* int: unknown nmi panic flag */
 	KERN_BOOTLOADER_TYPE=67, /* int: boot loader type */
 	KERN_RANDOMIZE=68, /* int: randomize virtual address space */
-	KERN_SETUID_DUMPABLE=69, /* int: behaviour of dumps for setuid core */
 	KERN_SPIN_RETRY=70,	/* int: number of spinlock retries */
 	KERN_ACPI_VIDEO_FLAGS=71, /* int: flags for setting up video after ACPI

sleep */

Quoted 8 lines

include/linux/sysctl.h 2006-03-11 09:02:13.000000000 enum · line 763

Quoted 8 lines

};

include/linux/sysctl.h 2006-03-11 09:02:13.000000000 renamed from include/linux/sysctl.h 2006-02-27 09:13:31.000000000 +0100 +3 -0
@@ -759,6 +758,9 @@ enum
 	FS_AIO_NR=18,	/* current system-wide number of aio requests */
 	FS_AIO_MAX_NR=19, /* system-wide maximum number of aio requests */
 	FS_INOTIFY=20,	/* inotify submenu */
+ /* Note: the following got misplaced but this mistake is
+			   now part of the ABI */
+	FS_SETUID_DUMPABLE=21, /* int: behaviour of dumps for setuid core */

OK, this is the other possibility. It'll leave a hole in the KERN_ sysctl numbering, but that's not a problem AFAICT.
I would however at least annotate with FIXME and make a note that it's gonna change in 2.7.

It's hard to tell how many people depend on suid_dumpable being at the wrong location. The fact nobody noticed it being placed wrongly would make me suspect it's not widely used yet (it only got introduced 2005-06-23 by Alan Cox). So I would prefer to move it to the correct location now rather than in 2.7. By the time 2.7 comes, we may have so many users, we may not want to correct this any more at all.

Quoted 1 line

This must be number 69 here. Or else we break the sys_sysctl ABI.

I don't think we break anything we want to maintain. Kernel-internal defines?
OK to change when going from 2.6.15 to 2.6.16 IMVHO. No module compiled for 2.6.15 will work with 2.6.16 anyway.

Best,

Signature · 1 line
Kurt Garloff, Head Architect Linux, Novell Inc.
#10 lore raw
Quoted 25 lines · Kurt Garloff

Hi Arjan, Eric,

Quoted 5 lines · Eric W. Biederman
Quoted 3 lines · Arjan van de Ven

diff -purN linux-2.6.16-rc5/include/linux/sysctl.h linux-2.6.16-rc5-setuid/include/linux/sysctl.h
--- linux-2.6.16-rc5/include/linux/sysctl.h 2006-02-27 09:13:31.000000000

+0100

Quoted 1 line
Quoted 1 line

+++ linux-2.6.16-rc5-setuid/include/linux/sysctl.h 2006-03-11

09:02:13.000000000

Quoted 14 lines
Quoted 14 lines

+0100

include/linux/sysctl.h +0 -1
@@ -144,7 +144,6 @@ enum
 	KERN_UNKNOWN_NMI_PANIC=66, /* int: unknown nmi panic flag */
 	KERN_BOOTLOADER_TYPE=67, /* int: boot loader type */
 	KERN_RANDOMIZE=68, /* int: randomize virtual address space */
-	KERN_SETUID_DUMPABLE=69, /* int: behaviour of dumps for setuid core */
 	KERN_SPIN_RETRY=70,	/* int: number of spinlock retries */
 	KERN_ACPI_VIDEO_FLAGS=71, /* int: flags for setting up video after ACPI

sleep */
};

include/linux/sysctl.h +0 -0
@@ -759,6 +758,9 @@ enum
 	FS_AIO_NR=18,	/* current system-wide number of aio requests */
 	FS_AIO_MAX_NR=19, /* system-wide maximum number of aio requests */
 	FS_INOTIFY=20,	/* inotify submenu */
Quoted 8 lines
Quoted 3 lines

include/linux/sysctl.h enum · line 763

Quoted 3 lines
include/linux/sysctl.h +3 -0
+ /* Note: the following got misplaced but this mistake is
+			   now part of the ABI */
+	FS_SETUID_DUMPABLE=21, /* int: behaviour of dumps for setuid core */

OK, this is the other possibility. It'll leave a hole in the KERN_ sysctl numbering, but that's not a problem AFAICT.
I would however at least annotate with FIXME and make a note that it's gonna change in 2.7.

Making the variable show up in two places is reasonable. Removing it from it's current location is not, as there are currently no technical reasons to make the change.

Unless something dramatic shows up 2.7 isn't going to happen. If you want something like that to happen use:
Documentation/feature-removal-schedule.txt

Quoted 6 lines

It's hard to tell how many people depend on suid_dumpable being at the wrong location. The fact nobody noticed it being placed wrongly would make me suspect it's not widely used yet (it only got introduced 2005-06-23 by Alan Cox). So I would prefer to move it to the correct location now rather than in 2.7. By the time 2.7 comes, we may have so many users, we may not want to correct this any more at all.

Don't move it just add it to the correct location. But also leave it at it's old location. Then put something in Documentation/feature-removal-schedule.txt that says in six months or so it will be gone, from the bad location.

Quoted 4 lines
Quoted 1 line

This must be number 69 here. Or else we break the sys_sysctl ABI.

I don't think we break anything we want to maintain. Kernel-internal defines?

That 69 is not kernel internal it is exported to user space, which is why changing that is wrong.

Quoted 2 lines

OK to change when going from 2.6.15 to 2.6.16 IMVHO. No module compiled for 2.6.15 will work with 2.6.16 anyway.

No because it isn't kernel internal. Which is why I mentioned sys_sysctl. It uses that 69 as part of the path to that variable.

Eric

#12 lore raw
Quoted 4 lines · Arjan van de Ven
Quoted 1 line

This must be number 69 here. Or else we break the sys_sysctl ABI.

numeric sysctl abi is since 2.6.0 no longer an ABI though; anything after that.. not an ABI :)

The system call still exists, in the system call table and is still implemented. The system call still takes a binary path. I don't see any big fat deprecated warnings in Documentation. So if the status has changed it has not been well communicated.

Looking a little closer I can find this note in sysctl.h:

****************************************************************
**

 **  The values in this file are exported to user space via 
 **  the sysctl() binary interface.  However this interface
 **  is unstable and deprecated and will be removed in the future. 
 **  For a stable interface use /proc/sys.

**
****************************************************************
Which used to be:

****************************************************************
****************************************************************
**

 **  WARNING:  
 **  The values in this file are exported to user space via 
 **  the sysctl() binary interface.  Do *NOT* change the 
 **  numbering of any existing values here, and do not change
 **  any numbers within any one set of values.  If you have
 **  to redefine an existing interface, use a new number for it.
 **  The kernel will then return ENOTDIR to any application using
 **  the old binary interface.

**
** --sct
**
****************************************************************

However except for new values and old values that are now reserved because that variable is no longer supported I do not see any changes in sysctl values from 2.4 to 2.6.

Looking in the git history of 2.6.0 at sysctl.h I can only see one clear instance of a value being reused for a different purpose, and that was somewhere under SCTP.

If we are going to kill the binary ABI I'm fine with that but it should be in Documentation/feature-removal-schedule.txt The system call should generate a rate-limited warning in the logs, or be removed entirely from the syscall table.

None of that was done. Given the current state of things I am strongly tempted to just to revert the comment, at the top of sysctl.h, as that would be easier than going through the proper steps to remove the system call.

However on the practical side it looks like /sbin/sysctl only uses the /proc interface. So it looks like the number of users of the binary interface are probably quite slim. So now would not be a bad time to deprecate the binary system call entirely, and in six months or whatever submit the patch to remove it.

But until we do that we should maintain the ABI, since we have been. It's not like setting enum to number 69 is an onerous task.

Eric