Skip to contentSkip to conversation

sysctl: Undeprecate sys_sysctl

PATCH In review Eric W. Biederman 8 participants 15 messages Nov 8, 2006 – Nov 11, 2006 Documentation

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sysctl: Undeprecate sys_sysctl

The basic issue is that despite have been deprecated and warned about as a very bad thing in the man pages since its inception there are a few real users of sys_sysctl. It was my assumption that because sysctl had been deprecated for all of 2.6 there would be no user space users by this point, so I initially gave sys_sysctl a very short deprecation period.

Now that I know there are a few real users the only sane way to proceed with deprecation is to push the time limit out to a year or two work and work with distributions that have big testing pools like fedora core to find these last remaining users.

Which means that the sys_sysctl interface needs to be maintained in the meantime.

Since I have provided a technical measure that allows us to add new sysctl entries without reserving more binary numbers I believe that is enough to fix the sys_sysctl binary interface maintenance problems, because there is no longer a need to change the binary interface at all.

Since the sys_sysctl implementation needs to stay around for a while and the worst of the maintenance issues that caused us to occasionally break the ABI have been addressed I don't see any advantage in continuing with the removal of sys_sysctl.

So instead of merely increasing the deprecation period this patch removes the deprecation of sys_sysctl and modifies the kernel to compile the code in by default.

With committing to maintain sys_sysctl we get all of the advantages of a fast interface for anything that needs it. Currently sys_sysctl is about 5x faster than /proc/sys, for the same string data.

---
 Documentation/feature-removal-schedule.txt |   12 ------------
 init/Kconfig                               |    2 +-
 2 files changed, 1 insertions(+), 13 deletions(-)
Documentation/feature-removal-schedule.txt +0 -12
index 1ac3c74..d52c4aa 100644
@@ -53,18 +53,6 @@ Who:	Mauro Carvalho Chehab <mchehab@brtu
 
 ---------------------------
 
-What:	sys_sysctl
-When:	January 2007
-Why:	The same information is available through /proc/sys and that is the
-	interface user space prefers to use. And there do not appear to be
-	any existing user in user space of sys_sysctl.  The additional
-	maintenance overhead of keeping a set of binary names gets
-	in the way of doing a good job of maintaining this interface.
-
-Who:	Eric Biederman <[email protected]>
-
----------------------------
-
 What:	PCMCIA control ioctl (needed for pcmcia-cs [cardmgr, cardctl])
 When:	November 2005
 Files:	drivers/pcmcia/: pcmcia_ioctl.c
init/Kconfig +1 -1
index c8b2624..e85e554 100644
@@ -304,7 +304,7 @@ config UID16
 
 config SYSCTL_SYSCALL
 	bool "Sysctl syscall support" if EMBEDDED
-	default n
+	default y
 	select SYSCTL
 	---help---
 	  Enable the deprecated sysctl system call.  sys_sysctl uses
Signature · 1 line
1.4.2.rc3.g7e18e-dirty
#1 lore raw
Quoted 1 line · Eric W. Biederman

[...]

Quoted 10 lines

Since the sys_sysctl implementation needs to stay around for a while and the worst of the maintenance issues that caused us to occasionally break the ABI have been addressed I don't see any advantage in continuing with the removal of sys_sysctl.

So instead of merely increasing the deprecation period this patch removes the deprecation of sys_sysctl and modifies the kernel to compile the code in by default.

[...]

init/Kconfig config UID16 · line 310

Quoted 8 lines

config SYSCTL_SYSCALL
bool "Sysctl syscall support" if EMBEDDED

init/Kconfig +1 -1
-       default n
+       default y
        select SYSCTL
        ---help---
          Enable the deprecated sysctl system call.  sys_sysctl uses

Perhaps you should also change the help text here to not say "the deprecated sysctl system call". Perhaps change to something like :

"Deselect this option, to not build sysctl suport into the kernel, if you are sure you won't be running any applications that depend on the sysctl system call.
If unsure, say Y."

Signature · 3 lines
Jesper Juhl <[email protected]>
Don't top-post  http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please      http://www.expita.com/nomime.html
#2 lore raw

PATCH 1/1 sysctl: Undeprecate sys_sysctl (take 2)

The basic issue is that despite have been ``deprecated'' and warned about as a very bad thing in the man pages since it's inception there are a few real users of sys_sysctl. It was my assumption that because sysctl had been ``deprecated'' for all of 2.6 there would be no user space users by this point, so I initially gave sys_sysctl a very short deprecation period.

Now that I know there are a few real users the only sane way to proceed with deprecation is to push the time limit out to a year or two work and work with distributions that have big testing pools like fedora core to find these last remaining users.

Which means that the sys_sysctl interface needs to be maintained in the meantime.

Since I have provided a technical measure that allows us to add new sysctl entries without reserving more binary numbers I believe that is enough to fix the sys_sysctl binary interface maintenance problems (I.e. don't grow it any more).

Since the sys_sysctl implementation needs to stay around for a while and the worst of the maintenance issues that caused us to occassionally break the ABI have been addressed I don't see any advantage in continuing with the removal of sys_sysctl.

With committing to maintain sys_sysctl we get all of the advantages of a fast interface for anything that needs it. Currently sys_sysctl is about 5x faster than /proc/sys, for the same string data.

The Kconfig help text is now also fixed.

---
 Documentation/feature-removal-schedule.txt |   12 ------------
 init/Kconfig                               |   19 +++++++++----------
 2 files changed, 9 insertions(+), 22 deletions(-)
Documentation/feature-removal-schedule.txt +0 -12
index 1ac3c74..d52c4aa 100644
@@ -53,18 +53,6 @@ Who:	Mauro Carvalho Chehab <mchehab@brtu
 
 ---------------------------
 
-What:	sys_sysctl
-When:	January 2007
-Why:	The same information is available through /proc/sys and that is the
-	interface user space prefers to use. And there do not appear to be
-	any existing user in user space of sys_sysctl.  The additional
-	maintenance overhead of keeping a set of binary names gets
-	in the way of doing a good job of maintaining this interface.
-
-Who:	Eric Biederman <[email protected]>
-
----------------------------
-
 What:	PCMCIA control ioctl (needed for pcmcia-cs [cardmgr, cardctl])
 When:	November 2005
 Files:	drivers/pcmcia/: pcmcia_ioctl.c
init/Kconfig +9 -10
index c8b2624..ffe945a 100644
@@ -304,20 +304,19 @@ config UID16
 
 config SYSCTL_SYSCALL
 	bool "Sysctl syscall support" if EMBEDDED
-	default n
+	default y
 	select SYSCTL
 	---help---
-	  Enable the deprecated sysctl system call.  sys_sysctl uses
-	  binary paths that have been found to be a major pain to maintain
-	  and use.  The interface in /proc/sys is now the primary and what
-	  everyone uses.
+	  sys_sysctl uses binary paths that have been found challenging
+	  to properly maintain and use.  The interface in /proc/sys
+	  using paths with ascii names is now the primary path to this
+	  information. 
 
-	  Nothing has been using the binary sysctl interface for some
-	  time now so nothing should break if you disable sysctl syscall
-	  support, and your kernel will get marginally smaller.
+	  Almost nothing using the binary sysctl interface so if you are
+	  trying to save some space it is probably safe to disable this,
+	  making your kernel marginally smaller.
 
-	  Unless you have an application that uses the sys_sysctl interface
- 	  you should probably say N here.
+	  If unsure say Y here.
 
 config KALLSYMS
 	 bool "Load all symbols for debugging/kksymoops" if EMBEDDED
Signature · 1 line
1.4.2.rc3.g7e18e-dirty
#3 lore raw

Re: [PATCH] sysctl: Undeprecate sys_sysctl (take 2)

Quoted 4 lines · Eric W. Biederman

The basic issue is that despite have been ``deprecated'' and warned about as a very bad thing in the man pages since it's inception there are a few real users of sys_sysctl.

But they only seem to use a small number of actually used with sysctl(2) sysctls.
I still think just maintaining a conversion table for those is the right thing to do.

The important part really is to get rid of the crufty old infrastructure internally.

-Andi

#4 lore raw

Re: [PATCH] sysctl: Undeprecate sys_sysctl (take 2)

Quoted 10 lines · Andi Kleen
Quoted 4 lines · Eric W. Biederman

The basic issue is that despite have been ``deprecated'' and warned about as a very bad thing in the man pages since it's inception there are a few real users of sys_sysctl.

But they only seem to use a small number of actually used with sysctl(2) sysctls.
I still think just maintaining a conversion table for those is the right thing to do.

I don't know. Every distinct user of the binary sysctl interface used a different entry. So the fact that there are a small number of programs and thus a small number of sysctls used I agree with. I do not agree with the conclusion that we can predict the set of binary sysctl that are in use. We do not get good enough feedback from the user community.

I don't have a problem with the principle of a conversion table if it meant that we would never add any additional binary sysctls.

Quoted 2 lines

The important part really is to get rid of the crufty old infrastructure internally.

There is certainly some cleanup we could do there. For absolutely simple values sysctl isn't too bad. But generated values are horrible.

Certainly some of the parts are crufty and harder to use then they should be. The /proc side actually seems worse than the binary side.

That said I have some plans on attacking sysctl in a very practical sense, because it is required to implement multiple namespaces.

Eric

#5 lore raw

Re: [PATCH] sysctl: Undeprecate sys_sysctl (take 2)

Quoted 17 lines · Eric W. Biederman
Quoted 7 lines · Andi Kleen
Quoted 3 lines · Eric W. Biederman

The basic issue is that despite have been ``deprecated'' and warned about as a very bad thing in the man pages since it's inception there are a few real users of sys_sysctl.

But they only seem to use a small number of actually used with sysctl(2) sysctls.
I still think just maintaining a conversion table for those is the right thing to do.

I don't know. Every distinct user of the binary sysctl interface used a different entry. So the fact that there are a small number of programs and thus a small number of sysctls used I agree with. I do not agree with the conclusion that we can predict the set of binary sysctl that are in use. We do not get good enough feedback from the user community.

Quoted 3 lines

I don't have a problem with the principle of a conversion table if it meant that we would never add any additional binary sysctls.

Okay, my opinion now...

I think we should change the sysctl system so most sysctls simply aren't accessible through the binary interface. The rest of them should be documented in one place, preferrably machine-readable.

However, I think having the binary sysctls available as a limited last resort is better than adding ad hoc system calls all over the place, like sys_mips.

-hpa

Acked-by on v1 1/1
#6 lore raw
In reply to Eric W. Biederman
Quoted 4 lines · Eric W. Biederman

Now that I know there are a few real users the only sane way to proceed with deprecation is to push the time limit out to a year or two work and work with distributions that have big testing pools like fedora core to find these last remaining users.

Some early boot code needs to know the kernel version and it needs to do it before /proc is mounted and potentially in order to run mount. In places it has its role but only in places.

Quoted 15 lines
init/Kconfig +1 -1
index c8b2624..e85e554 100644
@@ -304,7 +304,7 @@ config UID16
 
 config SYSCTL_SYSCALL
 	bool "Sysctl syscall support" if EMBEDDED
-	default n
+	default y
 	select SYSCTL
 	---help---
 	  Enable the deprecated sysctl system call.  sys_sysctl uses
Signature · 1 line
1.4.2.rc3.g7e18e-dirty
#7 lore raw
Quoted 9 lines · Alan Cox
Quoted 4 lines · Eric W. Biederman

Now that I know there are a few real users the only sane way to proceed with deprecation is to push the time limit out to a year or two work and work with distributions that have big testing pools like fedora core to find these last remaining users.

Some early boot code needs to know the kernel version and it needs to do it before /proc is mounted and potentially in order to run mount. In places it has its role but only in places.

isn't kernel version passed as AT vector nowadays?

#8 lore raw
In reply to Alan Cox
Quoted 9 lines · Alan Cox
Quoted 4 lines · Eric W. Biederman

Now that I know there are a few real users the only sane way to proceed with deprecation is to push the time limit out to a year or two work and work with distributions that have big testing pools like fedora core to find these last remaining users.

Some early boot code needs to know the kernel version and it needs to do it before /proc is mounted and potentially in order to run mount. In places it has its role but only in places.

uname() provides the kernel version.

-hpa

#9 lore raw
In reply to Eric W. Biederman
Quoted 11 lines · you

The basic issue is that despite have been deprecated and warned about as a very bad thing in the man pages since its inception there are a few real users of sys_sysctl. It was my assumption that because sysctl had been deprecated for all of 2.6 there would be no user space users by this point, so I initially gave sys_sysctl a very short deprecation period.

Now that I know there are a few real users the only sane way to proceed with deprecation is to push the time limit out to a year or two work and work with distributions that have big testing pools like fedora core to find these last remaining users.

Eric, do you have a list of the remaining users? It'd be good to know for people using Linux in an embedded environment, where they may want to switch off the option, but only if it doesn't break their userspace.

Signature · 5 lines
Cheers,
Alistair.

Final year Computer Science undergraduate.
1F2 55 South Clerk Street, Edinburgh, UK.
#10 lore raw
Quoted 8 lines · Alistair John Strachan
Quoted 7 lines · you

The basic issue is that despite have been deprecated and warned about as a very bad thing in the man pages since its inception there are a few real users of sys_sysctl. It was my assumption that because sysctl had been deprecated for all of 2.6 there would be no user space users by this point, so I initially gave sys_sysctl a very short deprecation period.

Quoted 8 lines
Quoted 4 lines

Now that I know there are a few real users the only sane way to proceed with deprecation is to push the time limit out to a year or two work and work with distributions that have big testing pools like fedora core to find these last remaining users.

Eric, do you have a list of the remaining users? It'd be good to know for people using Linux in an embedded environment, where they may want to switch off the option, but only if it doesn't break their userspace.

They are very very few. The ones I recall are kudzu, radvd, and libpthreads (which doesn't care).

There is a thread a month or so ago about this where I did a request for testers, that listed all of the users we could find.

The reality is that I don't think kernel developers can seriously find them.

If someone actually wants to kill sys_sysctl more power to them. As long as we don't add more binary numbers I think it is actually easier to support it than to find those weird users and remove it.

Eric

#11 lore raw
Quoted 7 lines · Eric W. Biederman
Quoted 3 lines · Alistair John Strachan

Eric, do you have a list of the remaining users? It'd be good to know for people using Linux in an embedded environment, where they may want to switch off the option, but only if it doesn't break their userspace.

They are very very few. The ones I recall are kudzu, radvd, and libpthreads (which doesn't care).

Let's not forget that on ARM it's used to get the MMIO region which is used for PIO emulation by glibc. Without it, glibc can't detect where this region is.

Signature · 3 lines
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core
#12 lore raw
Quoted 6 lines · Russell King
Quoted 5 lines · Eric W. Biederman
Quoted 3 lines · Alistair John Strachan

Eric, do you have a list of the remaining users? It'd be good to know for people using Linux in an embedded environment, where they may want to switch off the option, but only if it doesn't break their userspace.

Quoted 6 lines
Quoted 2 lines

They are very very few. The ones I recall are kudzu, radvd, and libpthreads (which doesn't care).

Let's not forget that on ARM it's used to get the MMIO region which is used for PIO emulation by glibc. Without it, glibc can't detect where this region is.

Is that on all subarches or just some of them?

But given the scarcity of the list this is certainly one of the significant known users.

Eric

#13 lore raw
Quoted 6 lines · Eric W. Biederman
Quoted 3 lines · Russell King

Let's not forget that on ARM it's used to get the MMIO region which is used for PIO emulation by glibc. Without it, glibc can't detect where this region is.

Is that on all subarches or just some of them?

No idea - I don't have that sort of information. Those which are PCI based have the potential to use it if some userspace programmer wants to use the ioperm/inb/outb etc which glibc provides.

I have had the occasional bug report that these functions haven't worked when we haven't been exporting the right info via sysctl, so they do seem to get used.

Signature · 3 lines
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core
#14 lore raw
In reply to Eric W. Biederman
Quoted 14 lines · Eric W. Biederman
Quoted 13 lines · Alistair John Strachan
Quoted 11 lines · you

The basic issue is that despite have been deprecated and warned about as a very bad thing in the man pages since its inception there are a few real users of sys_sysctl. It was my assumption that because sysctl had been deprecated for all of 2.6 there would be no user space users by this point, so I initially gave sys_sysctl a very short deprecation period.

Now that I know there are a few real users the only sane way to proceed with deprecation is to push the time limit out to a year or two work and work with distributions that have big testing pools like fedora core to find these last remaining users.

Quoted 6 lines
Quoted 3 lines

Eric, do you have a list of the remaining users? It'd be good to know for people using Linux in an embedded environment, where they may want to switch off the option, but only if it doesn't break their userspace.

They are very very few. The ones I recall are kudzu, radvd, and libpthreads (which doesn't care).

Neither do new versions of radvd, from what I can see.

radvd.c/check_ip6_forwarding():

#ifdef __linux__
	fp = fopen(PROC_SYS_IP6_FORWARDING, "r");
	if (fp) {
		fscanf(fp, "%d", &value);
		fclose(fp);
	}
	else
		flog(LOG_DEBUG, "Correct IPv6 forwarding procfs entry not found, "
	                       "perhaps the procfs is disabled, "
	                        "or the kernel interface has changed?");

#endif /* __linux__ */

	if (!fp && sysctl(forw_sysctl, sizeof(forw_sysctl)/sizeof(forw_sysctl[0]),
	    &value, &size, NULL, 0) < 0) {
		flog(LOG_DEBUG, "Correct IPv6 forwarding sysctl branch not found, "
			"perhaps the kernel interface has changed?");
		return(0);	/* this is of advisory value only */
	}

Maybe you should just hold out until everybody fixes their programs ;-)

Signature · 5 lines
Cheers,
Alistair.

Final year Computer Science undergraduate.
1F2 55 South Clerk Street, Edinburgh, UK.