[v2] binfmt_elf: Allow .bss in any interp PT_LOAD

Message ID 20221111061315.gonna.703-kees@kernel.org
State New
Headers
Series [v2] binfmt_elf: Allow .bss in any interp PT_LOAD |

Commit Message

Kees Cook Nov. 11, 2022, 6:13 a.m. UTC
  Traditionally, only the final PT_LOAD for load_elf_interp() supported
having p_memsz > p_filesz. Recently, lld's construction of musl's
libc.so on PowerPC64 started having two PT_LOAD program headers with
p_memsz > p_filesz.

As the least invasive change possible, check for p_memsz > p_filesz for
each PT_LOAD in load_elf_interp.

Reported-by: Rich Felker <dalias@libc.org>
Link: https://maskray.me/blog/2022-11-05-lld-musl-powerpc64
Cc: Pedro Falcato <pedro.falcato@gmail.com>
Cc: Fangrui Song <maskray@google.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: linux-fsdevel@vger.kernel.org
Cc: linux-mm@kvack.org
Signed-off-by: Kees Cook <keescook@chromium.org>
---
v2: I realized we need to retain the final padding call.
v1: https://lore.kernel.org/linux-hardening/20221111055747.never.202-kees@kernel.org/
---
 fs/binfmt_elf.c | 18 ++++++++++++++----
 1 file changed, 14 insertions(+), 4 deletions(-)
  

Comments

Fangrui Song Nov. 11, 2022, 7:42 a.m. UTC | #1
(+ sam@gentoo.org from Pedro Falcato's patch)

On 2022-11-10, Kees Cook wrote:
>Traditionally, only the final PT_LOAD for load_elf_interp() supported
>having p_memsz > p_filesz. Recently, lld's construction of musl's
>libc.so on PowerPC64 started having two PT_LOAD program headers with
>p_memsz > p_filesz.
>
>As the least invasive change possible, check for p_memsz > p_filesz for
>each PT_LOAD in load_elf_interp.
>
>Reported-by: Rich Felker <dalias@libc.org>
>Link: https://maskray.me/blog/2022-11-05-lld-musl-powerpc64
>Cc: Pedro Falcato <pedro.falcato@gmail.com>
>Cc: Fangrui Song <maskray@google.com>
>Cc: Alexander Viro <viro@zeniv.linux.org.uk>
>Cc: Eric Biederman <ebiederm@xmission.com>
>Cc: linux-fsdevel@vger.kernel.org
>Cc: linux-mm@kvack.org
>Signed-off-by: Kees Cook <keescook@chromium.org>
>---
>v2: I realized we need to retain the final padding call.
>v1: https://lore.kernel.org/linux-hardening/20221111055747.never.202-kees@kernel.org/
>---
> fs/binfmt_elf.c | 18 ++++++++++++++----
> 1 file changed, 14 insertions(+), 4 deletions(-)
>
>diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
>index 528e2ac8931f..0a24bbbef1d6 100644
>--- a/fs/binfmt_elf.c
>+++ b/fs/binfmt_elf.c
>@@ -673,15 +673,25 @@ static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
> 				last_bss = k;
> 				bss_prot = elf_prot;
> 			}
>+
>+			/*
>+			 * Clear any p_memsz > p_filesz area up to the end
>+			 * of the page to wipe anything left over from the
>+			 * loaded file contents.
>+			 */
>+			if (last_bss > elf_bss && padzero(elf_bss))

Missing {

But after fixing this, I get a musl ld.so error.

>+				error = -EFAULT;
>+				goto out;
>+			}
> 		}
> 	}
>
> 	/*
>-	 * Now fill out the bss section: first pad the last page from
>-	 * the file up to the page boundary, and zero it from elf_bss
>-	 * up to the end of the page.
>+	 * Finally, pad the last page from the file up to the page boundary,
>+	 * and zero it from elf_bss up to the end of the page, if this did
>+	 * not already happen with the last PT_LOAD.
> 	 */
>-	if (padzero(elf_bss)) {
>+	if (last_bss == elf_bss && padzero(elf_bss)) {
> 		error = -EFAULT;
> 		goto out;
> 	}
>-- 
>2.34.1
>

I added a new section to https://maskray.me/blog/2022-11-05-lld-musl-powerpc64
Copying here:

To test that the kernel ELF loader can handle more RW `PT_LOAD` program headers, we can create an executable with more RW `PT_LOAD` program headers with `p_filesz < p_memsz`.
We can place a read-only section after `.bss` followed by a `SHT_NOBITS` `SHF_ALLOC|SHF_WRITE` section. The read-only section will form a read-only `PT_LOAD` while the RW section will form a RW `PT_LOAD`.

```text
#--- a.c
#include <assert.h>
#include <stdio.h>

extern const char toc[];
char nobits0[0] __attribute__((section(".nobits0")));
char nobits1[0] __attribute__((section(".nobits1")));

int main(void) {
   assert(toc[4096-1] == 0);
   for (int i = 0; i < 1024; i++)
     assert(nobits0[i] == 0);
   nobits0[0] = nobits0[1024-1] = 1;
   for (int i = 0; i < 4096; i++)
     assert(nobits1[i] == 0);
   nobits1[0] = nobits1[4096-1] = 1;

   puts("hello");
}

#--- toc.s
.section .toc,"aw",@nobits
.globl toc
toc:
.space 4096

.section .ro0,"a"; .byte 255
.section .nobits0,"aw",@nobits; .space 1024
.section .ro1,"a"; .byte 255
.section .nobits1,"aw",@nobits; .space 4096

#--- a.lds
SECTIONS { .ro0 : {} .nobits0 : {} .ro1 : {} .nobits1 : {} } INSERT AFTER .bss;
```

```sh
split-file a.txt a
path/to/musl-gcc -Wl,--dynamic-linker=/lib/libc.so a/a.c a/a.lds -o toy
```

split-file is a utility in llvm-project.
  
Kees Cook Nov. 11, 2022, 8:13 p.m. UTC | #2
On Thu, Nov 10, 2022 at 11:42:34PM -0800, Fangrui Song wrote:
> (+ sam@gentoo.org from Pedro Falcato's patch)
> 
> On 2022-11-10, Kees Cook wrote:
> > Traditionally, only the final PT_LOAD for load_elf_interp() supported
> > having p_memsz > p_filesz. Recently, lld's construction of musl's
> > libc.so on PowerPC64 started having two PT_LOAD program headers with
> > p_memsz > p_filesz.
> > 
> > As the least invasive change possible, check for p_memsz > p_filesz for
> > each PT_LOAD in load_elf_interp.
> > 
> > Reported-by: Rich Felker <dalias@libc.org>
> > Link: https://maskray.me/blog/2022-11-05-lld-musl-powerpc64
> > Cc: Pedro Falcato <pedro.falcato@gmail.com>
> > Cc: Fangrui Song <maskray@google.com>
> > Cc: Alexander Viro <viro@zeniv.linux.org.uk>
> > Cc: Eric Biederman <ebiederm@xmission.com>
> > Cc: linux-fsdevel@vger.kernel.org
> > Cc: linux-mm@kvack.org
> > Signed-off-by: Kees Cook <keescook@chromium.org>
> > ---
> > v2: I realized we need to retain the final padding call.
> > v1: https://lore.kernel.org/linux-hardening/20221111055747.never.202-kees@kernel.org/
> > ---
> > fs/binfmt_elf.c | 18 ++++++++++++++----
> > 1 file changed, 14 insertions(+), 4 deletions(-)
> > 
> > diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> > index 528e2ac8931f..0a24bbbef1d6 100644
> > --- a/fs/binfmt_elf.c
> > +++ b/fs/binfmt_elf.c
> > @@ -673,15 +673,25 @@ static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
> > 				last_bss = k;
> > 				bss_prot = elf_prot;
> > 			}
> > +
> > +			/*
> > +			 * Clear any p_memsz > p_filesz area up to the end
> > +			 * of the page to wipe anything left over from the
> > +			 * loaded file contents.
> > +			 */
> > +			if (last_bss > elf_bss && padzero(elf_bss))
> 
> Missing {
> 
> But after fixing this, I get a musl ld.so error.
> 
> > +				error = -EFAULT;
> > +				goto out;
> > +			}
> > 		}
> > 	}
> > 
> > 	/*
> > -	 * Now fill out the bss section: first pad the last page from
> > -	 * the file up to the page boundary, and zero it from elf_bss
> > -	 * up to the end of the page.
> > +	 * Finally, pad the last page from the file up to the page boundary,
> > +	 * and zero it from elf_bss up to the end of the page, if this did
> > +	 * not already happen with the last PT_LOAD.
> > 	 */
> > -	if (padzero(elf_bss)) {
> > +	if (last_bss == elf_bss && padzero(elf_bss)) {
> > 		error = -EFAULT;
> > 		goto out;
> > 	}
> > -- 
> > 2.34.1
> > 
> 
> I added a new section to https://maskray.me/blog/2022-11-05-lld-musl-powerpc64
> Copying here:
> 
> To test that the kernel ELF loader can handle more RW `PT_LOAD` program headers, we can create an executable with more RW `PT_LOAD` program headers with `p_filesz < p_memsz`.
> We can place a read-only section after `.bss` followed by a `SHT_NOBITS` `SHF_ALLOC|SHF_WRITE` section. The read-only section will form a read-only `PT_LOAD` while the RW section will form a RW `PT_LOAD`.
> 
> ```text
> #--- a.c
> #include <assert.h>
> #include <stdio.h>
> 
> extern const char toc[];
> char nobits0[0] __attribute__((section(".nobits0")));
> char nobits1[0] __attribute__((section(".nobits1")));
> 
> int main(void) {
>   assert(toc[4096-1] == 0);
>   for (int i = 0; i < 1024; i++)
>     assert(nobits0[i] == 0);
>   nobits0[0] = nobits0[1024-1] = 1;
>   for (int i = 0; i < 4096; i++)
>     assert(nobits1[i] == 0);
>   nobits1[0] = nobits1[4096-1] = 1;
> 
>   puts("hello");
> }
> 
> #--- toc.s
> .section .toc,"aw",@nobits
> .globl toc
> toc:
> .space 4096
> 
> .section .ro0,"a"; .byte 255
> .section .nobits0,"aw",@nobits; .space 1024
> .section .ro1,"a"; .byte 255
> .section .nobits1,"aw",@nobits; .space 4096
> 
> #--- a.lds
> SECTIONS { .ro0 : {} .nobits0 : {} .ro1 : {} .nobits1 : {} } INSERT AFTER .bss;
> ```
> 
> ```sh
> split-file a.txt a
> path/to/musl-gcc -Wl,--dynamic-linker=/lib/libc.so a/a.c a/a.lds -o toy
> ```
> 
> split-file is a utility in llvm-project.

Where is a.txt? Also, it'd be nice to have this without needing the
musl-gcc.
  
Fangrui Song Nov. 11, 2022, 8:27 p.m. UTC | #3
On 2022-11-11, Kees Cook wrote:
>On Thu, Nov 10, 2022 at 11:42:34PM -0800, Fangrui Song wrote:
>> (+ sam@gentoo.org from Pedro Falcato's patch)
>>
>> On 2022-11-10, Kees Cook wrote:
>> > Traditionally, only the final PT_LOAD for load_elf_interp() supported
>> > having p_memsz > p_filesz. Recently, lld's construction of musl's
>> > libc.so on PowerPC64 started having two PT_LOAD program headers with
>> > p_memsz > p_filesz.
>> >
>> > As the least invasive change possible, check for p_memsz > p_filesz for
>> > each PT_LOAD in load_elf_interp.
>> >
>> > Reported-by: Rich Felker <dalias@libc.org>
>> > Link: https://maskray.me/blog/2022-11-05-lld-musl-powerpc64
>> > Cc: Pedro Falcato <pedro.falcato@gmail.com>
>> > Cc: Fangrui Song <maskray@google.com>
>> > Cc: Alexander Viro <viro@zeniv.linux.org.uk>
>> > Cc: Eric Biederman <ebiederm@xmission.com>
>> > Cc: linux-fsdevel@vger.kernel.org
>> > Cc: linux-mm@kvack.org
>> > Signed-off-by: Kees Cook <keescook@chromium.org>
>> > ---
>> > v2: I realized we need to retain the final padding call.
>> > v1: https://lore.kernel.org/linux-hardening/20221111055747.never.202-kees@kernel.org/
>> > ---
>> > fs/binfmt_elf.c | 18 ++++++++++++++----
>> > 1 file changed, 14 insertions(+), 4 deletions(-)
>> >
>> > diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
>> > index 528e2ac8931f..0a24bbbef1d6 100644
>> > --- a/fs/binfmt_elf.c
>> > +++ b/fs/binfmt_elf.c
>> > @@ -673,15 +673,25 @@ static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
>> > 				last_bss = k;
>> > 				bss_prot = elf_prot;
>> > 			}
>> > +
>> > +			/*
>> > +			 * Clear any p_memsz > p_filesz area up to the end
>> > +			 * of the page to wipe anything left over from the
>> > +			 * loaded file contents.
>> > +			 */
>> > +			if (last_bss > elf_bss && padzero(elf_bss))
>>
>> Missing {
>>
>> But after fixing this, I get a musl ld.so error.
>>
>> > +				error = -EFAULT;
>> > +				goto out;
>> > +			}
>> > 		}
>> > 	}
>> >
>> > 	/*
>> > -	 * Now fill out the bss section: first pad the last page from
>> > -	 * the file up to the page boundary, and zero it from elf_bss
>> > -	 * up to the end of the page.
>> > +	 * Finally, pad the last page from the file up to the page boundary,
>> > +	 * and zero it from elf_bss up to the end of the page, if this did
>> > +	 * not already happen with the last PT_LOAD.
>> > 	 */
>> > -	if (padzero(elf_bss)) {
>> > +	if (last_bss == elf_bss && padzero(elf_bss)) {
>> > 		error = -EFAULT;
>> > 		goto out;
>> > 	}
>> > --
>> > 2.34.1
>> >
>>
>> I added a new section to https://maskray.me/blog/2022-11-05-lld-musl-powerpc64
>> Copying here:
>>
>> To test that the kernel ELF loader can handle more RW `PT_LOAD` program headers, we can create an executable with more RW `PT_LOAD` program headers with `p_filesz < p_memsz`.
>> We can place a read-only section after `.bss` followed by a `SHT_NOBITS` `SHF_ALLOC|SHF_WRITE` section. The read-only section will form a read-only `PT_LOAD` while the RW section will form a RW `PT_LOAD`.
>>
>> ```text
>> #--- a.c
>> #include <assert.h>
>> #include <stdio.h>
>>
>> extern const char toc[];
>> char nobits0[0] __attribute__((section(".nobits0")));
>> char nobits1[0] __attribute__((section(".nobits1")));
>>
>> int main(void) {
>>   assert(toc[4096-1] == 0);
>>   for (int i = 0; i < 1024; i++)
>>     assert(nobits0[i] == 0);
>>   nobits0[0] = nobits0[1024-1] = 1;
>>   for (int i = 0; i < 4096; i++)
>>     assert(nobits1[i] == 0);
>>   nobits1[0] = nobits1[4096-1] = 1;
>>
>>   puts("hello");
>> }
>>
>> #--- toc.s
>> .section .toc,"aw",@nobits
>> .globl toc
>> toc:
>> .space 4096
>>
>> .section .ro0,"a"; .byte 255
>> .section .nobits0,"aw",@nobits; .space 1024
>> .section .ro1,"a"; .byte 255
>> .section .nobits1,"aw",@nobits; .space 4096
>>
>> #--- a.lds
>> SECTIONS { .ro0 : {} .nobits0 : {} .ro1 : {} .nobits1 : {} } INSERT AFTER .bss;
>> ```
>>
>> ```sh
>> split-file a.txt a
>> path/to/musl-gcc -Wl,--dynamic-linker=/lib/libc.so a/a.c a/a.lds -o toy
>> ```
>>
>> split-file is a utility in llvm-project.
>
>Where is a.txt? Also, it'd be nice to have this without needing the
>musl-gcc.

Sorry for the unclear description. I rewrite it.
(`char nobits0[0] __attribute__((section(".nobits0")));` is not effective. It's SHT_PROGBITS and makes the output section SHT_PROGBITS.
The new example addresses the deficiency.)



Create some files. If you have split-file (a [test utility](https://llvm.org/docs/TestingGuide.html#extra-files) from llvm-project), you may place the following content into `a.txt`.

```text
#--- a.c
#include <assert.h>
#include <stdio.h>

extern const char toc[];
extern char nobits0[], nobits1[];

int main(void) {
   assert(toc[4096-1] == 0);
   for (int i = 0; i < 1024; i++) {
     assert(nobits0[i] == 0);
     nobits0[i] = 1;
   }
   for (int i = 0; i < 8192; i++) {
     assert(nobits1[i] == 0);
     nobits1[i] = 1;
   }

   puts("hello");
}

#--- toc.s
.globl toc, nobits0, nobits1

.section .toc,"aw",@nobits; toc: .space 4096

.section .ro0,"a"; .byte 255
.section .nobits0,"aw",@nobits; nobits0: .space 1024
.section .ro1,"a"; .byte 255
.section .nobits1,"aw",@nobits; nobits1: .space 8192

#--- a.lds
SECTIONS { .ro0 : {} .nobits0 : {} .ro1 : {} .nobits1 : {} } INSERT AFTER .bss;
```

Then run:
```sh
split-file a.txt a
path/to/musl-gcc -Wl,--dynamic-linker=/lib/libc.so a/a.c a/a.lds -o toy
```

Note: when a `SHT_NOBITS` section is followed by another section, the `SHT_NOBITS` section behaves as if it occupies the file offset range. This is because ld.lld does not implement a file size optimization.


For this simple example, using glibc based gcc works as well (musl provides __assert_fail and puts referenced by the executable):

gcc -Wl,--dynamic-linker=/lib/libc.so a/a.c a/a.lds -o toy
  

Patch

diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
index 528e2ac8931f..0a24bbbef1d6 100644
--- a/fs/binfmt_elf.c
+++ b/fs/binfmt_elf.c
@@ -673,15 +673,25 @@  static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
 				last_bss = k;
 				bss_prot = elf_prot;
 			}
+
+			/*
+			 * Clear any p_memsz > p_filesz area up to the end
+			 * of the page to wipe anything left over from the
+			 * loaded file contents.
+			 */
+			if (last_bss > elf_bss && padzero(elf_bss))
+				error = -EFAULT;
+				goto out;
+			}
 		}
 	}
 
 	/*
-	 * Now fill out the bss section: first pad the last page from
-	 * the file up to the page boundary, and zero it from elf_bss
-	 * up to the end of the page.
+	 * Finally, pad the last page from the file up to the page boundary,
+	 * and zero it from elf_bss up to the end of the page, if this did
+	 * not already happen with the last PT_LOAD.
 	 */
-	if (padzero(elf_bss)) {
+	if (last_bss == elf_bss && padzero(elf_bss)) {
 		error = -EFAULT;
 		goto out;
 	}