Skip to contentSkip to conversation

Modutils can't handle long kernel names

DISCUSSION Mike Fedyk 5 participants 8 messages Nov 9, 2001 – Nov 14, 2001

lore
Share & tools
lore

Mike Fedyk

author
#0 lore raw

Hello,

I've gotten into the habbit of adding the names of the patches I add to my kernel to the extraversion string in the top level Makefile in my kernels.

Here's my latest example:
VERSION = 2
PATCHLEVEL = 4
SUBLEVEL = 15 EXTRAVERSION=-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+ext3-mem_acct+elevator

Unfortunately, with this long kernel version number, modutils (I've noticed depmod and modutils so far...) choke on it.

depmod:
depmod: Can't open /lib/modules/2.4.15-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+e#1 SMP Thu Nov 8 20:18:04 PST 2001/modules.dep for writing

uname -r:
2.4.15-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+e#1 SMP Thu Nov 8 20:18:04 PST 2001

uname -a:
Linux mikef-linux.matchmail.com 2.4.15-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+e#1 SMP Thu Nov 8 20:18:04 PST 2001 #1 SMP Thu Nov 8 20:18:04 PST 2001 i686 unknown

Yep, I know, "don't do that!". Still, can this be fixed easily, or is it one of those things that kbuild 2.5 will fix, and make everything rosy and do my laundry too?

TIA,

Mike

author
#1 lore raw
Quoted 14 lines · Mike Fedyk

Hello,

I've gotten into the habbit of adding the names of the patches I add to my kernel to the extraversion string in the top level Makefile in my kernels.

Here's my latest example:
VERSION = 2
PATCHLEVEL = 4
SUBLEVEL = 15 EXTRAVERSION=-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+ext3-mem_acct+elevator

Unfortunately, with this long kernel version number, modutils (I've noticed depmod and modutils so far...) choke on it.

Quoted 9 lines

depmod:
depmod: Can't open /lib/modules/2.4.15-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+e#1 SMP Thu Nov 8 20:18:04 PST 2001/modules.dep for writing

uname -r:
2.4.15-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+e#1 SMP Thu Nov 8 20:18:04 PST 2001

uname -a:
Linux mikef-linux.matchmail.com 2.4.15-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+e#1 SMP Thu Nov 8 20:18:04 PST 2001 #1 SMP Thu Nov 8 20:18:04 PST 2001 i686 unknown

Oh, /proc/version looks good:
Linux version 2.4.15-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+ext3-mem_acct+elevator ([email protected]) (gcc version 2.95.4 20011006 (Debian prerelease)) #1 SMP Thu Nov 8 20:18:04 PST 2001

Mike

#2 lore raw
Quoted 2 lines · Mike Fedyk

I've gotten into the habbit of adding the names of the patches I add to my kernel to the extraversion string in the top level Makefile in my kernels.

Everything in-kernel is going to be OK because we set the version string like so:

	char version[] = UTS_RELEASE;

where UTS_RELEASE is set from the variables in your Makefile. Thus, the string is set to the exact size of your version on compile. If you find anywhere _inside the kernel_ that breaks with long strings, it is probably a bug.

Userspace is a different story. In general, most programs probably do not have dynamic off-the-heap storage for the version string size. Added such a feature may be advantageous in some cases, but in many it is overkill and not worth the dynamic memory management.

I would consider the case where a large string crashes a user space program a bug. While the program may choose to set some limit, it should certainly check its buffers. I would also consider an abnormally small legal string size a bug, but I honestly don't see your version size fitting that description. :)

Anyhow, if you want to change it in a given app, its a simple size that needs to be changed and then recompile.

Robert Love

#3 lore raw
In reply to Mike Fedyk
Quoted 10 lines · Mike Fedyk

EXTRAVERSION=-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+ext3-mem_acct+elevator

Unfortunately, with this long kernel version number, modutils (I've noticed depmod and modutils so far...) choke on it.

depmod:
depmod: Can't open /lib/modules/2.4.15-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+e#1 SMP Thu Nov 8 20:18:04 PST 2001/modules.dep for writing

uname -r:
2.4.15-pre1+freeswan-1.91+xsched+netdev_random+ext3-0.9.15-2414+e#1 SMP Thu Nov 8 20:18:04 PST 2001

It is not a modutils problem, it is a fixed restriction on the size of the uname() fields, modutils just uses what uname -r gives it.

         struct utsname {
                      char sysname[SYS_NMLN];
                      char nodename[SYS_NMLN];
                      char release[SYS_NMLN];
                      char version[SYS_NMLN];
                      char machine[SYS_NMLN];
                      char domainname[SYS_NMLN];
                      };

SYS_NMLN maps to _UTSNAME_LENGTH.
/* Length of the entries in `struct utsname' is 65. */ #define _UTSNAME_LENGTH 65

Like you said, don't do that :).

#4 lore raw
In reply to Mike Fedyk
Quoted 2 lines · Keith Owens

It is not a modutils problem, it is a fixed restriction on the size of the uname() fields, modutils just uses what uname -r gives it.

this patch would catch it at compiletime, wouldn't it?

Signature · 13 lines
//anders/g

--- Makefile.orig	Fri Nov  9 23:17:36 2001
+++ Makefile	Fri Nov  9 23:18:25 2001
@@ -338,7 +338,7 @@
 	@mv -f .ver $@
 
 init/version.o: init/version.c include/linux/compile.h include/config/MARKER
-	$(CC) $(CFLAGS) $(CFLAGS_KERNEL) -DUTS_MACHINE='"$(ARCH)"' -c -o init/version.o init/version.c
+	$(CC) $(CFLAGS) -Werror $(CFLAGS_KERNEL) -DUTS_MACHINE='"$(ARCH)"' -c -o init/version.o init/version.c
 
 init/main.o: init/main.c include/config/MARKER
 	$(CC) $(CFLAGS) $(CFLAGS_KERNEL) $(PROFILING) -c -o $*.o $<
#5 lore raw
Quoted 14 lines · On Fri, 9 Nov 2001 23:23:43 +0100, [email protected] wrote:
Quoted 2 lines · Keith Owens

It is not a modutils problem, it is a fixed restriction on the size of the uname() fields, modutils just uses what uname -r gives it.

this patch would catch it at compiletime, wouldn't it?

Makefile renamed from Makefile.orig +0 -0
@@ -338,7 +338,7 @@
 	@mv -f .ver $@
 

init/version.o: init/version.c include/linux/compile.h include/config/MARKER

Makefile renamed from Makefile.orig +1 -1
-	$(CC) $(CFLAGS) $(CFLAGS_KERNEL) -DUTS_MACHINE='"$(ARCH)"' -c -o init/version.o init/version.c
+	$(CC) $(CFLAGS) -Werror $(CFLAGS_KERNEL) -DUTS_MACHINE='"$(ARCH)"' -c -o init/version.o init/version.c
Quoted 2 lines

init/main.o: init/main.c include/config/MARKER

	$(CC) $(CFLAGS) $(CFLAGS_KERNEL) $(PROFILING) -c -o $*.o $<

Not quite. If any of the fields are exactly 65 characters long excluding the trailing NUL, gcc does not flag a warning. Alas 65 characters trips the uname() bug. It is arguable if
char foo[65] = "65_printable_characters..........................................";
should cause a gcc warning or not, but it does not.

Also only some of the fields should cause a warning if they are too long. A domain name can legally be 255 characters and must not cause an error when compiling the kernel. So some fields must be truncated, others must cause an error if they are too long. The length checking has to be done in the Makefile.

This patch fixes the problem of overlong utsname values. Against 2.4.14, Linus, please apply.

Index: 14.1/Makefile

Makefile Sat, 10 Nov 2001 15:23:44 +1100 kaos (linux-2.4/T/c/50_Makefile 1.1.2.15.1.2.2.25.2.2.1.17.1.4.1.29.1.19 644) renamed from Makefile Tue, 06 Nov 2001 12:07:56 +1100 kaos (linux-2.4/T/c/50_Makefile 1.1.2.15.1.2.2.25.2.2.1.17.1.4.1.29.1.19 644) +16 -12
@@ -304,25 +304,29 @@ newversion:
 	. scripts/mkversion > .tmpversion
 	@mv -f .tmpversion .version
 
+uts_len		:= 64
+uts_truncate	:= sed -e 's/\(.\{1,$(uts_len)\}\).*/\1/'
+
 include/linux/compile.h: $(CONFIGURATION) include/linux/version.h newversion
-	@echo -n \#define UTS_VERSION \"\#`cat .version` > .ver
-	@if [ -n "$(CONFIG_SMP)" ] ; then echo -n " SMP" >> .ver; fi
-	@if [ -f .name ]; then  echo -n \-`cat .name` >> .ver; fi
-	@echo ' '`date`'"' >> .ver
+	@echo -n \#`cat .version` > .ver1
+	@if [ -n "$(CONFIG_SMP)" ] ; then echo -n " SMP" >> .ver1; fi
+	@if [ -f .name ]; then  echo -n \-`cat .name` >> .ver1; fi
+	@echo ' '`date` >> .ver1
+	@echo \#define UTS_VERSION \"`cat .ver1 | $(uts_truncate)`\" > .ver
 	@echo \#define LINUX_COMPILE_TIME \"`date +%T`\" >> .ver
 	@echo \#define LINUX_COMPILE_BY \"`whoami`\" >> .ver
-	@echo \#define LINUX_COMPILE_HOST \"`hostname`\" >> .ver
-	@if [ -x /bin/dnsdomainname ]; then \
-	   echo \#define LINUX_COMPILE_DOMAIN \"`dnsdomainname`\"; \
-	 elif [ -x /bin/domainname ]; then \
-	   echo \#define LINUX_COMPILE_DOMAIN \"`domainname`\"; \
-	 else \
-	   echo \#define LINUX_COMPILE_DOMAIN ; \
-	 fi >> .ver
+	@echo \#define LINUX_COMPILE_HOST \"`hostname | $(uts_truncate)`\" >> .ver
+	@([ -x /bin/dnsdomainname ] && /bin/dnsdomainname > .ver1) || \
+	 ([ -x /bin/domainname ] && /bin/domainname > .ver1) || \
+	 echo > .ver1
+	@echo \#define LINUX_COMPILE_DOMAIN \"`cat .ver1 | $(uts_truncate)`\" >> .ver
 	@echo \#define LINUX_COMPILER \"`$(CC) $(CFLAGS) -v 2>&1 | tail -1`\" >> .ver
 	@mv -f .ver $@
+	@rm -f .ver1
 
 include/linux/version.h: ./Makefile
+	@expr length "$(KERNELRELEASE)" \<= $(uts_len) > /dev/null || \
+	  (echo KERNELRELEASE \"$(KERNELRELEASE)\" exceeds $(uts_len) characters >&2; false)
 	@echo \#define UTS_RELEASE \"$(KERNELRELEASE)\" > .ver
 	@echo \#define LINUX_VERSION_CODE `expr $(VERSION) \\* 65536 + $(PATCHLEVEL) \\* 256 + $(SUBLEVEL)` >> .ver
 	@echo '#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))' >>.ver
#6 lore raw

Makefile Sat, 10 Nov 2001 15:23:44 +1100 kaos (linux-2.4/T/c/50_Makefile 1.1.2.15.1.2.2.25.2.2.1.17.1.4.1.29.1.19 644) newversion: · line 308

Quoted 8 lines · Keith Owens
Quoted 3 lines · On Fri, 9 Nov 2001 23:23:43 +0100, [email protected] wrote:
Quoted 2 lines · Keith Owens

It is not a modutils problem, it is a fixed restriction on the size of the uname() fields, modutils just uses what uname -r gives it.

Makefile Sat, 10 Nov 2001 15:23:44 +1100 kaos (linux-2.4/T/c/50_Makefile 1.1.2.15.1.2.2.25.2.2.1.17.1.4.1.29.1.19 644) renamed from Makefile Tue, 06 Nov 2001 12:07:56 +1100 kaos (linux-2.4/T/c/50_Makefile 1.1.2.15.1.2.2.25.2.2.1.17.1.4.1.29.1.19 644) +2 -0
+uts_len		:= 64
+uts_truncate	:= sed -e 's/\(.\{1,$(uts_len)\}\).*/\1/'

Should this be a fixed length of 64?
Or should it be grabbed form a header somewhere? So when/if SYS_NMLN/_UTSNAME_LENGTH is changed longer strings can be used? I check and Solaris 8 defines SYS_NMLN as 257.

Would this break cross-comiling badly? Are other libc headers needed in the build?

-Thomas

#7 lore raw
Quoted 10 lines · Thomas Dodd

Makefile Sat, 10 Nov 2001 15:23:44 +1100 kaos (linux-2.4/T/c/50_Makefile 1.1.2.15.1.2.2.25.2.2.1.17.1.4.1.29.1.19 644) newversion: · line 308

Quoted 2 lines · Keith Owens
Makefile Sat, 10 Nov 2001 15:23:44 +1100 kaos (linux-2.4/T/c/50_Makefile 1.1.2.15.1.2.2.25.2.2.1.17.1.4.1.29.1.19 644) renamed from Makefile Tue, 06 Nov 2001 12:07:56 +1100 kaos (linux-2.4/T/c/50_Makefile 1.1.2.15.1.2.2.25.2.2.1.17.1.4.1.29.1.19 644) +2 -0
+uts_len		:= 64
+uts_truncate	:= sed -e 's/\(.\{1,$(uts_len)\}\).*/\1/'

Should this be a fixed length of 64?
Or should it be grabbed form a header somewhere? So when/if SYS_NMLN/_UTSNAME_LENGTH is changed longer strings can be used? I check and Solaris 8 defines SYS_NMLN as 257.

Quoted 2 lines

Would this break cross-comiling badly? Are other libc headers needed in the build?

This is the kernel's idea of uts length, not glibc, it is independent of where you are compiling. It could be extracted from the kernel header,
uts_len := $(shell sed -ne 's/#define __NEW_UTS_LEN //p' include/linux/utsname.h)
but why bother? The value has been fixed at 64 since at least 2.0 and is embedded in glibc so it is unlikely to change. Even if it does change, it must be upwards so the worst case is fail safe.