diff mbox

kbuild: escape single backslashes in macro make-cmd

Message ID 20140726163551.19857.17737.stgit@zurg (mailing list archive)
State New, archived
Headers show

Commit Message

Konstantin Khlebnikov July 26, 2014, 4:35 p.m. UTC
This already has been fixed in commit c353acba28fb3fa1fd05fd
("kbuild: make: fix if_changed when command contains backslashes")
but escaping still isn't perfect and triggers false-positive rebuilds.

For x86 problem happens every time, because rules in arch/x86/realmode/rm/
and arch/x86/boot/ contains commands like sed -n -e 's/foo\(.*\)/\1/p'.
Backslash in \1 isn't escaped and turns into ascii symbol with code 1.
Macro if_changed detects command change and rebuilds target again and again.

Backslash escaping conflicts with other passes because it's used for escaping
other symbols. To avoid that current macro handles only double backslashes.
Obviously this doesn't work for \1 like above.

This patch reorders passes. It doubles all backslashes before escaping # and '

Visible effect in rebuilding x86/defconfig without changes, before patch:

blind@zurg:~/src/linux$ make V=2
  CHK     include/config/kernel.release
  CHK     include/generated/uapi/linux/version.h
  CHK     include/generated/utsrelease.h
  CALL    scripts/checksyscalls.sh - due to target missing
  CHK     include/generated/compile.h
  PASYMS  arch/x86/realmode/rm/pasyms.h - due to command line change
  LDS     arch/x86/realmode/rm/realmode.lds - due to: arch/x86/realmode/rm/pasyms.h
  LD      arch/x86/realmode/rm/realmode.elf - due to: arch/x86/realmode/rm/realmode.lds
  RELOCS  arch/x86/realmode/rm/realmode.relocs - due to: arch/x86/realmode/rm/realmode.elf
  OBJCOPY arch/x86/realmode/rm/realmode.bin - due to: arch/x86/realmode/rm/realmode.elf arch/x86/realmode/rm/realmode.relocs
  AS      arch/x86/realmode/rmpiggy.o - due to: arch/x86/realmode/rm/realmode.bin
  LD      arch/x86/realmode/built-in.o - due to: arch/x86/realmode/rmpiggy.o
  LD      arch/x86/built-in.o - due to: arch/x86/realmode/built-in.o
  LINK    vmlinux - due to: arch/x86/built-in.o
  LD      vmlinux.o
  MODPOST vmlinux.o - due to vmlinux.o not in $(targets)
  GEN     .version
  CHK     include/generated/compile.h
  UPD     include/generated/compile.h
  CC      init/version.o - due to: include/generated/compile.h
  LD      init/built-in.o - due to: init/version.o
  KSYM    .tmp_kallsyms1.o
  KSYM    .tmp_kallsyms2.o
  LD      vmlinux
  SORTEX  vmlinux
  SYSMAP  System.map
  VOFFSET arch/x86/boot/voffset.h - due to: vmlinux
  OBJCOPY arch/x86/boot/compressed/vmlinux.bin - due to: vmlinux
  GZIP    arch/x86/boot/compressed/vmlinux.bin.gz - due to: arch/x86/boot/compressed/vmlinux.bin
  MKPIGGY arch/x86/boot/compressed/piggy.S - due to: arch/x86/boot/compressed/vmlinux.bin.gz
  AS      arch/x86/boot/compressed/piggy.o - due to: arch/x86/boot/compressed/piggy.S
  LD      arch/x86/boot/compressed/vmlinux - due to: arch/x86/boot/compressed/piggy.o
  ZOFFSET arch/x86/boot/zoffset.h - due to: arch/x86/boot/compressed/vmlinux
  AS      arch/x86/boot/header.o - due to: arch/x86/boot/voffset.h arch/x86/boot/zoffset.h
  CC      arch/x86/boot/version.o - due to: include/generated/compile.h
  LD      arch/x86/boot/setup.elf - due to: arch/x86/boot/header.o arch/x86/boot/version.o
  OBJCOPY arch/x86/boot/setup.bin - due to: arch/x86/boot/setup.elf
  OBJCOPY arch/x86/boot/vmlinux.bin - due to: arch/x86/boot/compressed/vmlinux
  BUILD   arch/x86/boot/bzImage - due to: arch/x86/boot/setup.bin arch/x86/boot/vmlinux.bin
Setup is 15676 bytes (padded to 15872 bytes).
System is 5569 kB
CRC c06bbc0c
Kernel: arch/x86/boot/bzImage is ready  (#4)
  Building modules, stage 2.
  MODPOST 12 modules - due to target is PHONY

After:

blind@zurg:~/src/linux$ make V=2
  CHK     include/config/kernel.release
  CHK     include/generated/uapi/linux/version.h
  CHK     include/generated/utsrelease.h
  CALL    scripts/checksyscalls.sh - due to target missing
  CHK     include/generated/compile.h
Kernel: arch/x86/boot/bzImage is ready  (#5)
  Building modules, stage 2.
  MODPOST 12 modules - due to target is PHONY

Signed-off-by: Konstantin Khlebnikov <koct9i@gmail.com>
---
 scripts/Kbuild.include |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)


--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Michal Marek Aug. 6, 2014, 11:45 a.m. UTC | #1
On 2014-07-26 18:35, Konstantin Khlebnikov wrote:
> This already has been fixed in commit c353acba28fb3fa1fd05fd
> ("kbuild: make: fix if_changed when command contains backslashes")
> but escaping still isn't perfect and triggers false-positive rebuilds.
> 
> For x86 problem happens every time, because rules in arch/x86/realmode/rm/
> and arch/x86/boot/ contains commands like sed -n -e 's/foo\(.*\)/\1/p'.
> Backslash in \1 isn't escaped and turns into ascii symbol with code 1.
> Macro if_changed detects command change and rebuilds target again and again.
> 
> Backslash escaping conflicts with other passes because it's used for escaping
> other symbols. To avoid that current macro handles only double backslashes.
> Obviously this doesn't work for \1 like above.
> 
> This patch reorders passes. It doubles all backslashes before escaping # and '
> 
> Visible effect in rebuilding x86/defconfig without changes, before patch:
> 
> blind@zurg:~/src/linux$ make V=2
>   CHK     include/config/kernel.release
>   CHK     include/generated/uapi/linux/version.h
>   CHK     include/generated/utsrelease.h
>   CALL    scripts/checksyscalls.sh - due to target missing
>   CHK     include/generated/compile.h
>   PASYMS  arch/x86/realmode/rm/pasyms.h - due to command line change

With which make and shell version are you seeing this? While the patch
looks correct, I can't reproduce the error here:

$ make V=2
  CHK     include/config/kernel.release
  CHK     include/generated/uapi/linux/version.h
  CHK     include/generated/utsrelease.h
  CALL    scripts/checksyscalls.sh - due to target missing
  CHK     include/generated/compile.h
Kernel: arch/x86/boot/bzImage is ready  (#1)
  Building modules, stage 2.
  MODPOST 12 modules - due to target is PHONY
$ cat arch/x86/realmode/rm/.pasyms.h.cmd
cmd_arch/x86/realmode/rm/pasyms.h := nm arch/x86/realmode/rm/header.o
arch/x86/realmode/rm/trampoline_64.o arch/x86/realmode/rm/stack.o
arch/x86/realmode/rm/reboot.o arch/x86/realmode/rm/wakeup_asm.o
arch/x86/realmode/rm/wakemain.o arch/x86/realmode/rm/video-mode.o
arch/x86/realmode/rm/copy.o arch/x86/realmode/rm/bioscall.o
arch/x86/realmode/rm/regs.o arch/x86/realmode/rm/video-vga.o
arch/x86/realmode/rm/video-vesa.o arch/x86/realmode/rm/video-bios.o |
sed -n -r -e 's/^([0-9a-fA-F]+) [ABCDGRSTVW] (.+)$$/pa_\2 = \2;/p' |
sort | uniq > arch/x86/realmode/rm/pasyms.h

This is with both 3.16 and my kbuild branch, which is based on 3.16-rc1.
Also, in the description, you say that backslash escaping conflicts with
escaping # and ', but this has nothing to do with the \2 backreference
in the sed command.

Michal
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Konstantin Khlebnikov Aug. 6, 2014, 12:19 p.m. UTC | #2
On Wed, Aug 6, 2014 at 3:45 PM, Michal Marek <mmarek@suse.cz> wrote:
> On 2014-07-26 18:35, Konstantin Khlebnikov wrote:
>> This already has been fixed in commit c353acba28fb3fa1fd05fd
>> ("kbuild: make: fix if_changed when command contains backslashes")
>> but escaping still isn't perfect and triggers false-positive rebuilds.
>>
>> For x86 problem happens every time, because rules in arch/x86/realmode/rm/
>> and arch/x86/boot/ contains commands like sed -n -e 's/foo\(.*\)/\1/p'.
>> Backslash in \1 isn't escaped and turns into ascii symbol with code 1.
>> Macro if_changed detects command change and rebuilds target again and again.
>>
>> Backslash escaping conflicts with other passes because it's used for escaping
>> other symbols. To avoid that current macro handles only double backslashes.
>> Obviously this doesn't work for \1 like above.
>>
>> This patch reorders passes. It doubles all backslashes before escaping # and '
>>
>> Visible effect in rebuilding x86/defconfig without changes, before patch:
>>
>> blind@zurg:~/src/linux$ make V=2
>>   CHK     include/config/kernel.release
>>   CHK     include/generated/uapi/linux/version.h
>>   CHK     include/generated/utsrelease.h
>>   CALL    scripts/checksyscalls.sh - due to target missing
>>   CHK     include/generated/compile.h
>>   PASYMS  arch/x86/realmode/rm/pasyms.h - due to command line change
>
> With which make and shell version are you seeing this? While the patch
> looks correct, I can't reproduce the error here:

/bin/sh points to dash (debian default setup).

I cannot reproduce this using bash. That explains why this bug is still here.

>
> $ make V=2
>   CHK     include/config/kernel.release
>   CHK     include/generated/uapi/linux/version.h
>   CHK     include/generated/utsrelease.h
>   CALL    scripts/checksyscalls.sh - due to target missing
>   CHK     include/generated/compile.h
> Kernel: arch/x86/boot/bzImage is ready  (#1)
>   Building modules, stage 2.
>   MODPOST 12 modules - due to target is PHONY
> $ cat arch/x86/realmode/rm/.pasyms.h.cmd
> cmd_arch/x86/realmode/rm/pasyms.h := nm arch/x86/realmode/rm/header.o
> arch/x86/realmode/rm/trampoline_64.o arch/x86/realmode/rm/stack.o
> arch/x86/realmode/rm/reboot.o arch/x86/realmode/rm/wakeup_asm.o
> arch/x86/realmode/rm/wakemain.o arch/x86/realmode/rm/video-mode.o
> arch/x86/realmode/rm/copy.o arch/x86/realmode/rm/bioscall.o
> arch/x86/realmode/rm/regs.o arch/x86/realmode/rm/video-vga.o
> arch/x86/realmode/rm/video-vesa.o arch/x86/realmode/rm/video-bios.o |
> sed -n -r -e 's/^([0-9a-fA-F]+) [ABCDGRSTVW] (.+)$$/pa_\2 = \2;/p' |
> sort | uniq > arch/x86/realmode/rm/pasyms.h
>
> This is with both 3.16 and my kbuild branch, which is based on 3.16-rc1.
> Also, in the description, you say that backslash escaping conflicts with
> escaping # and ', but this has nothing to do with the \2 backreference
> in the sed command.

I'm not quite understand what the difference between escaping standalone
backslash ( ' \ ' -> \' \\ \' ) that backreference (  ' \2 ' -> \' \\2 \' )

We have to double backslashes, but this must be done before escaping # and ',
otherwise doubling undoes their escaping.

>
> Michal
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Michal Marek Aug. 6, 2014, 3:01 p.m. UTC | #3
On 2014-08-06 14:19, Konstantin Khlebnikov wrote:
> On Wed, Aug 6, 2014 at 3:45 PM, Michal Marek <mmarek@suse.cz> wrote:
>> On 2014-07-26 18:35, Konstantin Khlebnikov wrote:
>>> This already has been fixed in commit c353acba28fb3fa1fd05fd
>>> ("kbuild: make: fix if_changed when command contains backslashes")
>>> but escaping still isn't perfect and triggers false-positive rebuilds.
>>>
>>> For x86 problem happens every time, because rules in arch/x86/realmode/rm/
>>> and arch/x86/boot/ contains commands like sed -n -e 's/foo\(.*\)/\1/p'.
>>> Backslash in \1 isn't escaped and turns into ascii symbol with code 1.
>>> Macro if_changed detects command change and rebuilds target again and again.
>>>
>>> Backslash escaping conflicts with other passes because it's used for escaping
>>> other symbols. To avoid that current macro handles only double backslashes.
>>> Obviously this doesn't work for \1 like above.
>>>
>>> This patch reorders passes. It doubles all backslashes before escaping # and '
>>>
>>> Visible effect in rebuilding x86/defconfig without changes, before patch:
>>>
>>> blind@zurg:~/src/linux$ make V=2
>>>   CHK     include/config/kernel.release
>>>   CHK     include/generated/uapi/linux/version.h
>>>   CHK     include/generated/utsrelease.h
>>>   CALL    scripts/checksyscalls.sh - due to target missing
>>>   CHK     include/generated/compile.h
>>>   PASYMS  arch/x86/realmode/rm/pasyms.h - due to command line change
>>
>> With which make and shell version are you seeing this? While the patch
>> looks correct, I can't reproduce the error here:
> 
> /bin/sh points to dash (debian default setup).
> 
> I cannot reproduce this using bash. That explains why this bug is still here.

So the difference between the shells is that their 'echo' builtin treats
\<number> differently:
$ ./dash -c "echo '\2'"

$ ./dash -c "echo '\2'" | xxd
0000000: 020a                                     ..
$ /bin/bash -c "echo '\2'"
\2

For some reason we fail to escape the \ and dash treats \2 as \02. POSIX
says that this is implementation-defined, so none of the shells is wrong.

I need to look into this a bit further. I'll most likely end up applying
your patch, but I'd like to understand the fix in detail first.

Thanks,
Michal
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/scripts/Kbuild.include b/scripts/Kbuild.include
index 122f95c..edd0b0d 100644
--- a/scripts/Kbuild.include
+++ b/scripts/Kbuild.include
@@ -219,7 +219,7 @@  endif
 # >$< substitution to preserve $ when reloading .cmd file
 # note: when using inline perl scripts [perl -e '...$$t=1;...']
 # in $(cmd_xxx) double $$ your perl vars
-make-cmd = $(subst \\,\\\\,$(subst \#,\\\#,$(subst $$,$$$$,$(call escsq,$(cmd_$(1))))))
+make-cmd = $(call escsq,$(subst \#,\\\#,$(subst \,\\,$(subst $$,$$$$,$(cmd_$(1))))))
 
 # Find any prerequisites that is newer than target or that does not exist.
 # PHONY targets skipped in both cases.