Wanted: a limit on kernel log buffer size

Wanted: a limit on kernel log buffer size

[Chuck Ebbert]

 Some people (who will mercifully go unnamed) just will _not_
read the documentation, and set the kernel log buffer shift
to 31 on a 256MB machine.  This attempt to allocate 2GB of memory
for the buffer results in an unbootable kernel.

 Suggestions?

--
 Chuck
 I am not a number!

Re: Wanted: a limit on kernel log buffer size

[Randy.Dunlap]

>  Some people (who will mercifully go unnamed) just will _not_
> read the documentation, and set the kernel log buffer shift
> to 31 on a 256MB machine.  This attempt to allocate 2GB of memory
> for the buffer results in an unbootable kernel.
>
>  Suggestions?

This is a multi-part answer.  Say, 5 parts.

a.  If someone won't read the help text, how can we help them?

b.  If we make a 2 GB log buffer size a compile-time error, will
they read that?

c.  If we make it a compile-time warning, will they read that?

d.  What limit(s) do you suggest?  I can try to add some limits.

e.  This kind of config limiting should be done in the config system IMO.
I've asked Roman for that capability....

~Randy

Re: Wanted: a limit on kernel log buffer size

[Randy.Dunlap]

>>  Some people (who will mercifully go unnamed) just will _not_
>> read the documentation, and set the kernel log buffer shift
>> to 31 on a 256MB machine.  This attempt to allocate 2GB of memory for the
>> buffer results in an unbootable kernel.
>>
>>  Suggestions?
>
> This is a multi-part answer.  Say, 5 parts.
>
> a.  If someone won't read the help text, how can we help them?
>
> b.  If we make a 2 GB log buffer size a compile-time error, will
> they read that?
>
> c.  If we make it a compile-time warning, will they read that?
>
> d.  What limit(s) do you suggest?  I can try to add some limits.
>
> e.  This kind of config limiting should be done in the config system IMO.
> I've asked Roman for that capability....


Here's a patch that limits kernel log buffer size to 1 MB max.
Comments?

I'm inserting it here via cut-and-paste, so it might not be all clean.
Patch is to 2.5.66-PV (plain vanilla).


patch_name:	logbuf-limit.patch
patch_version:	2003-04-06.20:28:43
author:		Randy.Dunlap <rddunlap@osdl.org>
description:	limit kernel log buffer size to 1 MB
product:	Linux
product_versions: 2.5.66
diffstat:	=
 kernel/printk.c |    3 +++
 1 files changed, 3 insertions(+)


diff -Naur ./kernel/printk.c%LBLIM ./kernel/printk.c
--- ./kernel/printk.c%LBLIM	2003-04-06 20:27:28.000000000 -0700
+++ ./kernel/printk.c	2003-04-06 20:27:53.000000000 -0700
@@ -34,6 +34,9 @@

 #define LOG_BUF_LEN	(1 << CONFIG_LOG_BUF_SHIFT)
 #define LOG_BUF_MASK	(LOG_BUF_LEN-1)
+#if (CONFIG_LOG_BUF_SHIFT > 20)
+#error CONFIG_LOG_BUF_SHIFT is ridiculously large (more than 1 MB).
+#endif

 /* printk's without a loglevel use this.. */
 #define DEFAULT_MESSAGE_LOGLEVEL 4 /* KERN_WARNING */

Re: Wanted: a limit on kernel log buffer size

[Robert Love]

On Sun, 2003-04-06 at 23:35, Randy.Dunlap wrote:

> > a.  If someone won't read the help text, how can we help them?
> >
> > b.  If we make a 2 GB log buffer size a compile-time error, will
> > they read that?
> >
> > c.  If we make it a compile-time warning, will they read that?
> >
> > d.  What limit(s) do you suggest?  I can try to add some limits.
> >
> > e.  This kind of config limiting should be done in the config system IMO.
> > I've asked Roman for that capability....
> 
> Here's a patch that limits kernel log buffer size to 1 MB max.
> Comments?

I liked points (a) and (e) above.

I say if users cannot bother to read the documentation and understanding
things, why are they compiling a kernel?

And if we are going to implement parameters bounds checking it should be
done in kconfig.  There are a few other places that want it, too.

	Robert Love

Re: Wanted: a limit on kernel log buffer size

[Randy.Dunlap]

> On Sun, 2003-04-06 at 23:35, Randy.Dunlap wrote:
>
>> > a.  If someone won't read the help text, how can we help them?
>> >
>> > b.  If we make a 2 GB log buffer size a compile-time error, will they
>> read that?
>> >
>> > c.  If we make it a compile-time warning, will they read that?
>> >
>> > d.  What limit(s) do you suggest?  I can try to add some limits.
>> >
>> > e.  This kind of config limiting should be done in the config system
>> IMO. I've asked Roman for that capability....
>>
>> Here's a patch that limits kernel log buffer size to 1 MB max.
>> Comments?
>
> I liked points (a) and (e) above.
>
> I say if users cannot bother to read the documentation and understanding
> things, why are they compiling a kernel?
>
> And if we are going to implement parameters bounds checking it should be
> done in kconfig.  There are a few other places that want it, too.

I forgot another point:  don't change default config settings unless
you are willing to read the help text.

~Randy  [sorry if you get this multiple times; i had to resend it]

Re: Wanted: a limit on kernel log buffer size

[Roman Zippel]

Hi,

On Sun, 6 Apr 2003, Randy.Dunlap wrote:

> This is a multi-part answer.  Say, 5 parts.
> 
> a.  If someone won't read the help text, how can we help them?
> 
> b.  If we make a 2 GB log buffer size a compile-time error, will
> they read that?
> 
> c.  If we make it a compile-time warning, will they read that?
> 
> d.  What limit(s) do you suggest?  I can try to add some limits.
> 
> e.  This kind of config limiting should be done in the config system IMO.
> I've asked Roman for that capability....

While I don't mind adding limits, checking it at compile time certainly 
won't hurt.
Even better would be a more dynamic solution, which can release unused 
print buffer after booting.

bye, Roman

[PATCH] Wanted: a limit on kernel log buffer size

[Randy.Dunlap]

On Sun, 6 Apr 2003 20:35:59 -0700 (PDT) "Randy.Dunlap" <rddunlap@osdl.org> wrote:

| >>  Some people (who will mercifully go unnamed) just will _not_
| >> read the documentation, and set the kernel log buffer shift
| >> to 31 on a 256MB machine.  This attempt to allocate 2GB of memory for the
| >> buffer results in an unbootable kernel.
| >>
| >>  Suggestions?
| >
| > This is a multi-part answer.  Say, 5 parts.
| >
| > a.  If someone won't read the help text, how can we help them?
| >
| > b.  If we make a 2 GB log buffer size a compile-time error, will
| > they read that?
| >
| > c.  If we make it a compile-time warning, will they read that?
| >
| > d.  What limit(s) do you suggest?  I can try to add some limits.
| >
| > e.  This kind of config limiting should be done in the config system IMO.
| > I've asked Roman for that capability....


Here's a [modified] patch that limits kernel log buffer size to 1 MB max
and 4 KB min.

To me, ideally the config system would allow limits to be specified,
and then advanced users could edit .config to get around those limits.
By putting limits checking in kernel source files, there is no way
around them other than by editing the source files.

I've made this patch because there are some cases where it shouldn't be
possible to shoot oneself in the foot IMO -- at least not using the
config system...it's OK to do that when editing .config.

--
~Randy


patch_name:	logbuf_limits.patch
patch_version:	2003-04-07.15:54:36
author:		Randy.Dunlap <rddunlap@osdl.org>
description:	enforce kernel log buffer limits (min and max);
product:	Linux
product_versions: 2.5.67
changelog:	add min and max kernel log buffer limits checking;
URL:		http://www.osdl.org/archive/rddunlap/patches/logbuf_limits.patch
diffstat:	=
 kernel/printk.c |    6 ++++++
 1 files changed, 6 insertions(+)


diff -Naur ./kernel/printk.c%LBLIM ./kernel/printk.c
--- ./kernel/printk.c%LBLIM	Mon Apr  7 11:12:38 2003
+++ ./kernel/printk.c	Mon Apr  7 14:44:16 2003
@@ -34,6 +34,12 @@
 
 #define LOG_BUF_LEN	(1 << CONFIG_LOG_BUF_SHIFT)
 #define LOG_BUF_MASK	(LOG_BUF_LEN-1)
+#if (LOG_BUF_LEN > (1024 * 1024))
+#error CONFIG_LOG_BUF_SHIFT is ridiculously large (more than 20 [1 MB]).
+#endif
+#if (LOG_BUF_LEN < (4 * 1024))
+#error CONFIG_LOG_BUF_SHIFT is ridiculously small (less than 12 [4 KB]).
+#endif
 
 /* printk's without a loglevel use this.. */
 #define DEFAULT_MESSAGE_LOGLEVEL 4 /* KERN_WARNING */

Re: Wanted: a limit on kernel log buffer size

[Chuck Ebbert]

Randy Dunlap wrote:

>> Just how far does boot get?  What messages are printed before death?


I don't have it exactly, but it's along the lines of:

   Decompressing.... OK, now booting the kernel

i.e., none.


--
 Chuck
 I am not an octal number!

Re: Wanted: a limit on kernel log buffer size

[Randy.Dunlap]

On Mon, 7 Apr 2003 04:55:27 -0400 Chuck Ebbert <76306.1226@compuserve.com> wrote:

| Robert Love
| 
| > I say if users cannot bother to read the documentation
| >and understanding things, why are they compiling a kernel?
| 
| 
|  True, but then they'll come bleating to l-k when all they get is
| a blank, black screen on boot.  (At least _I_ figured out what
| went wrong first...)

Good that you figured it out.

Just how far does boot get?  What messages are printed before death?

--
~Randy

Re: Wanted: a limit on kernel log buffer size

[Chuck Ebbert]

Randy Dunlap:


>+#if (CONFIG_LOG_BUF_SHIFT > 20)
>+#error CONFIG_LOG_BUF_SHIFT is ridiculously large (more than 1 MB).
>+#endif


That ought to do it.  Anyone who needs more than that can just change
the source.


--
 Chuck
 I am not an octal number!

Re: Wanted: a limit on kernel log buffer size

[Chuck Ebbert]

Robert Love

> I say if users cannot bother to read the documentation
>and understanding things, why are they compiling a kernel?


 True, but then they'll come bleating to l-k when all they get is
a blank, black screen on boot.  (At least _I_ figured out what
went wrong first...)

--
 Chuck
 I am not a number!