Skip to contentSkip to conversation

Disable execmem for sparc

PATCH In review Tom "spot" Callaway 5 participants 17 messages Mar 24, 2010 – Apr 28, 2010 SELinux security module

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Disable execmem for sparc

Attached is a patch which disables execmem for sparc. Without it, selinux does not work at all on SPARC64.

This patch should be reasonably non-controversial, because this is already being done for PPC32.

Dennis Gilmore <[email protected]>

Thanks in advance,

~spot

security/selinux/hooks.c renamed from security/selinux/hooks.c.mprotect-sparc +2 -2
diff -up linux-2.6.32.noarch/security/selinux/hooks.c.mprotect-sparc linux-2.6.32.noarch/security/selinux/hooks.c
@@ -3010,7 +3010,7 @@ static int file_map_prot_check(struct fi
 	const struct cred *cred = current_cred();
 	int rc = 0;
 
-#ifndef CONFIG_PPC32
+#if !defined(CONFIG_PPC32) && !defined(CONFIG_SPARC)
 	if ((prot & PROT_EXEC) && (!file || (!shared && (prot & PROT_WRITE)))) {
 		/*
 		 * We are making executable an anonymous mapping or a
@@ -3082,7 +3082,7 @@ static int selinux_file_mprotect(struct 
 	if (selinux_checkreqprot)
 		prot = reqprot;
 
-#ifndef CONFIG_PPC32
+#if !defined(CONFIG_PPC32) && !defined(CONFIG_SPARC)
 	if ((prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
 		int rc = 0;
 		if (vma->vm_start >= vma->vm_mm->start_brk &&
#1 lore raw

From: "Tom \"spot\" Callaway" <[email protected]>
Date: Wed, 24 Mar 2010 17:52:57 -0400

Quoted 9 lines

Attached is a patch which disables execmem for sparc. Without it, selinux does not work at all on SPARC64.

This patch should be reasonably non-controversial, because this is already being done for PPC32.

Dennis Gilmore <[email protected]>

What is the reason why it doesn't work, I'm just curious?

Is there some dependency upon executable stacks or executable data segments always working? Why can't SELINUX protect be used with that correctly?

And since we're touching selinux code we need to at a minimum
CC: them so they can have a look at your change.

--------------------

security/selinux/hooks.c renamed from security/selinux/hooks.c.mprotect-sparc +2 -2
diff -up linux-2.6.32.noarch/security/selinux/hooks.c.mprotect-sparc linux-2.6.32.noarch/security/selinux/hooks.c
@@ -3010,7 +3010,7 @@ static int file_map_prot_check(struct fi
 	const struct cred *cred = current_cred();
 	int rc = 0;
 
-#ifndef CONFIG_PPC32
+#if !defined(CONFIG_PPC32) && !defined(CONFIG_SPARC)
 	if ((prot & PROT_EXEC) && (!file || (!shared && (prot & PROT_WRITE)))) {
 		/*
 		 * We are making executable an anonymous mapping or a
@@ -3082,7 +3082,7 @@ static int selinux_file_mprotect(struct 
 	if (selinux_checkreqprot)
 		prot = reqprot;
 
-#ifndef CONFIG_PPC32
+#if !defined(CONFIG_PPC32) && !defined(CONFIG_SPARC)
 	if ((prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
 		int rc = 0;
 		if (vma->vm_start >= vma->vm_mm->start_brk &&
#2 lore raw
In reply to Tom "spot" Callaway
Quoted 13 lines · David Miller

From: "Tom \"spot\" Callaway" <[email protected]>
Date: Wed, 24 Mar 2010 17:52:57 -0400

Quoted 10 lines

Attached is a patch which disables execmem for sparc. Without it, selinux does not work at all on SPARC64.

This patch should be reasonably non-controversial, because this is already being done for PPC32.

Dennis Gilmore <[email protected]>

Quoted 7 lines
Quoted 1 line

What is the reason why it doesn't work, I'm just curious?

Is there some dependency upon executable stacks or executable data segments always working? Why can't SELINUX protect be used with that correctly?

what happens is that almost all binaries end up with execmem set and selinux prevents them from running. the system fails to even get close to coming up in a usable state

Dec 31 18:00:40 sparcbook kernel: type=1400 audit(8.160:3): avc:  denied  { 
execmem } for  pid=208 comm="consoletype" 
scontext=system_u:system_r:consoletype_t:s0 
tcontext=system_u:system_r:consoletype_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type=1400 audit(8.315:4): avc:  denied  { 
execmem } for  pid=211 comm="hostname" 
scontext=system_u:system_r:hostname_t:s0 
tcontext=system_u:system_r:hostname_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type=1400 audit(8.520:5): avc:  denied  { 
execmem } for  pid=213 comm="mount" scontext=system_u:system_r:mount_t:s0 
tcontext=system_u:system_r:mount_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type=1400 audit(8.570:6): avc:  denied  { 

execmem } for pid=203 comm="readahead-colle" scontext=system_u:system_r:readahead_t:s0 tcontext=system_u:system_r:readahead_t:s0 tclass=process

is a small sample of the logs you get not everything fails but almost everything

Quoted 32 lines

And since we're touching selinux code we need to at a minimum
CC: them so they can have a look at your change.

--------------------
diff -up linux-2.6.32.noarch/security/selinux/hooks.c.mprotect-sparc linux-2.6.32.noarch/security/selinux/hooks.c ---
linux-2.6.32.noarch/security/selinux/hooks.c.mprotect-sparc 2010-03-10
08:28:20.957571926 -0500 +++
linux-2.6.32.noarch/security/selinux/hooks.c 2010-03-10 08:29:15.732698763
-0500 @@ -3010,7 +3010,7 @@ static int file_map_prot_check(struct fi

security/selinux/hooks.c renamed from security/selinux/hooks.c.mprotect-sparc +2 -3
 	const struct cred *cred = current_cred();
 	int rc = 0;
 
-#ifndef CONFIG_PPC32
+#if !defined(CONFIG_PPC32) && !defined(CONFIG_SPARC)
 	if ((prot & PROT_EXEC) && (!file || (!shared && (prot & PROT_WRITE)))) {
 		/*
 		 * We are making executable an anonymous mapping or a
@@ -3082,7 +3082,7 @@ static int selinux_file_mprotect(struct
 	if (selinux_checkreqprot)
 		prot = reqprot;
 
-#ifndef CONFIG_PPC32
+#if !defined(CONFIG_PPC32) && !defined(CONFIG_SPARC)
 	if ((prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
 		int rc = 0;
 		if (vma->vm_start >= vma->vm_mm->start_brk &&
--

To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#3 lore raw
In reply to Tom "spot" Callaway
Quoted 38 lines · Dennis Gilmore
Quoted 13 lines · David Miller

From: "Tom \"spot\" Callaway" <[email protected]>
Date: Wed, 24 Mar 2010 17:52:57 -0400

Quoted 10 lines

Attached is a patch which disables execmem for sparc. Without it, selinux does not work at all on SPARC64.

This patch should be reasonably non-controversial, because this is already being done for PPC32.

Dennis Gilmore <[email protected]>

Quoted 7 lines
Quoted 1 line

What is the reason why it doesn't work, I'm just curious?

Is there some dependency upon executable stacks or executable data segments always working? Why can't SELINUX protect be used with that correctly?

what happens is that almost all binaries end up with execmem set and selinux prevents them from running. the system fails to even get close to coming up in a usable state

Dec 31 18:00:40 sparcbook kernel: type00 audit(8.160:3): avc:  denied  { 
execmem } for  pid 8 comm="consoletype" 
scontext=system_u:system_r:consoletype_t:s0 
tcontext=system_u:system_r:consoletype_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type00 audit(8.315:4): avc:  denied  { 
execmem } for  pid!1 comm="hostname" 
scontext=system_u:system_r:hostname_t:s0 
tcontext=system_u:system_r:hostname_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type00 audit(8.520:5): avc:  denied  { 
execmem } for  pid!3 comm="mount" scontext=system_u:system_r:mount_t:s0 
tcontext=system_u:system_r:mount_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type00 audit(8.570:6): avc:  denied  { 
Quoted 6 lines

execmem } for pid 3 comm="readahead-colle" scontext=system_u:system_r:readahead_t:s0 tcontext=system_u:system_r:readahead_t:s0 tclass=process

is a small sample of the logs you get not everything fails but almost everything

I think we need to understand why this is happening - it usually reflects a toolchain problem (that was the case in the ppc32 situation, and was later fixed in Fedora through an updated toolchain and rebuilt userland). eu-readelf -l /bin/hostname shows what?

Quoted 34 lines
Quoted 32 lines

And since we're touching selinux code we need to at a minimum
CC: them so they can have a look at your change.

--------------------
diff -up linux-2.6.32.noarch/security/selinux/hooks.c.mprotect-sparc linux-2.6.32.noarch/security/selinux/hooks.c ---
linux-2.6.32.noarch/security/selinux/hooks.c.mprotect-sparc 2010-03-10
08:28:20.957571926 -0500 +++
linux-2.6.32.noarch/security/selinux/hooks.c 2010-03-10 08:29:15.732698763
-0500 @@ -3010,7 +3010,7 @@ static int file_map_prot_check(struct fi

security/selinux/hooks.c renamed from security/selinux/hooks.c.mprotect-sparc +2 -3
 	const struct cred *cred = current_cred();
 	int rc = 0;
 
-#ifndef CONFIG_PPC32
+#if !defined(CONFIG_PPC32) && !defined(CONFIG_SPARC)
 	if ((prot & PROT_EXEC) && (!file || (!shared && (prot & PROT_WRITE)))) {
 		/*
 		 * We are making executable an anonymous mapping or a
@@ -3082,7 +3082,7 @@ static int selinux_file_mprotect(struct
 	if (selinux_checkreqprot)
 		prot = reqprot;
 
-#ifndef CONFIG_PPC32
+#if !defined(CONFIG_PPC32) && !defined(CONFIG_SPARC)
 	if ((prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
 		int rc = 0;
 		if (vma->vm_start >= vma->vm_mm->start_brk &&
--

To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Signature · 2 lines
Stephen Smalley
National Security Agency
#4 lore raw
In reply to Tom "spot" Callaway
Quoted 45 lines · Stephen Smalley
Quoted 44 lines · Dennis Gilmore
Quoted 14 lines · David Miller

From: "Tom \"spot\" Callaway" <[email protected]>
Date: Wed, 24 Mar 2010 17:52:57 -0400

Quoted 11 lines

Attached is a patch which disables execmem for sparc. Without it, selinux does not work at all on SPARC64.

This patch should be reasonably non-controversial, because this is already being done for PPC32.

T5220)

Dennis Gilmore <[email protected]>

Quoted 7 lines
Quoted 1 line

What is the reason why it doesn't work, I'm just curious?

Is there some dependency upon executable stacks or executable data segments always working? Why can't SELINUX protect be used with that correctly?

what happens is that almost all binaries end up with execmem set and selinux prevents them from running. the system fails to even get close to coming up in a usable state

Dec 31 18:00:40 sparcbook kernel: type=1400 audit(8.160:3): avc:  denied 
{ execmem } for  pid=208 comm="consoletype"
scontext=system_u:system_r:consoletype_t:s0
tcontext=system_u:system_r:consoletype_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type=1400 audit(8.315:4): avc:  denied 
{ execmem } for  pid=211 comm="hostname"
scontext=system_u:system_r:hostname_t:s0
tcontext=system_u:system_r:hostname_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type=1400 audit(8.520:5): avc:  denied 
{ execmem } for  pid=213 comm="mount"
scontext=system_u:system_r:mount_t:s0
tcontext=system_u:system_r:mount_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type=1400 audit(8.570:6): avc:  denied 

{ execmem } for pid=203 comm="readahead-colle" scontext=system_u:system_r:readahead_t:s0 tcontext=system_u:system_r:readahead_t:s0 tclass=process

Quoted 7 lines
Quoted 2 lines

is a small sample of the logs you get not everything fails but almost everything

I think we need to understand why this is happening - it usually reflects a toolchain problem (that was the case in the ppc32 situation, and was later fixed in Fedora through an updated toolchain and rebuilt userland). eu-readelf -l /bin/hostname shows what?

eu-readelf -l /bin/hostname
Program Headers:

  Type           Offset   VirtAddr   PhysAddr   FileSiz  MemSiz   Flg Align
  PHDR           0x000034 0x00010034 0x00010034 0x000100 0x000100 R E 0x4
  INTERP         0x000134 0x00010134 0x00010134 0x000013 0x000013 R   0x1
        [Requesting program interpreter: /lib/ld-linux.so.2]
  LOAD           0x000000 0x00010000 0x00010000 0x002204 0x002204 R E 0x10000
  LOAD           0x002204 0x00022204 0x00022204 0x000284 0x0002ac RWE 0x10000
  DYNAMIC        0x002218 0x00022218 0x00022218 0x0000d0 0x0000d0 RW  0x4
  NOTE           0x000148 0x00010148 0x00010148 0x000044 0x000044 R   0x4
  GNU_EH_FRAME   0x002110 0x00012110 0x00012110 0x00003c 0x00003c R   0x4
  GNU_STACK      0x000000 0x00000000 0x00000000 0x000000 0x000000 RW  0x4

Section to Segment mapping:
Segment Sections...
00
01 [RO: .interp]
02 [RO: .interp .note.ABI-tag .note.gnu.build-id .gnu.hash .dynsym
.dynstr .gnu.version .gnu.version_r .rela.dyn .rela.plt .init .text .fini .rodata .eh_frame_hdr .eh_frame]

   03      .ctors .dtors .jcr .dynamic .got .plt .data .bss
   04      .dynamic
   05      [RO: .note.ABI-tag .note.gnu.build-id]
   06      [RO: .eh_frame_hdr]

07

Quoted 44 lines
Quoted 44 lines
Quoted 44 lines

And since we're touching selinux code we need to at a minimum
CC: them so they can have a look at your change.

--------------------
diff -up linux-2.6.32.noarch/security/selinux/hooks.c.mprotect-sparc linux-2.6.32.noarch/security/selinux/hooks.c ---
linux-2.6.32.noarch/security/selinux/hooks.c.mprotect-sparc 2010-03-10
08:28:20.957571926 -0500 +++
linux-2.6.32.noarch/security/selinux/hooks.c 2010-03-10
08:29:15.732698763 -0500 @@ -3010,7 +3010,7 @@ static int file_map_prot_check(struct fi

security/selinux/hooks.c renamed from security/selinux/hooks.c.mprotect-sparc +2 -2
 	const struct cred *cred = current_cred();
 	int rc = 0;
 
-#ifndef CONFIG_PPC32
+#if !defined(CONFIG_PPC32) && !defined(CONFIG_SPARC)
 
 	if ((prot & PROT_EXEC) && (!file || (!shared && (prot &
 	PROT_WRITE)))) {
 
 		/*
 
 		 * We are making executable an anonymous mapping or a
@@ -3082,7 +3082,7 @@ static int selinux_file_mprotect(struct
 
 	if (selinux_checkreqprot)
 
 		prot = reqprot;
 
-#ifndef CONFIG_PPC32
+#if !defined(CONFIG_PPC32) && !defined(CONFIG_SPARC)
 
security/selinux/hooks.c renamed from security/selinux/hooks.c.mprotect-sparc +0 -1
 	if ((prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
 
 		int rc = 0;
 		if (vma->vm_start >= vma->vm_mm->start_brk &&
 
--

To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#6 lore raw
In reply to Tom "spot" Callaway
Quoted 43 lines · Stephen Smalley
Quoted 42 lines · Dennis Gilmore
Quoted 13 lines · David Miller

From: "Tom \"spot\" Callaway" <[email protected]>
Date: Wed, 24 Mar 2010 17:52:57 -0400

Quoted 10 lines

Attached is a patch which disables execmem for sparc. Without it, selinux does not work at all on SPARC64.

This patch should be reasonably non-controversial, because this is already being done for PPC32.

Dennis Gilmore <[email protected]>

Quoted 7 lines
Quoted 1 line

What is the reason why it doesn't work, I'm just curious?

Is there some dependency upon executable stacks or executable data segments always working? Why can't SELINUX protect be used with that correctly?

what happens is that almost all binaries end up with execmem set and selinux prevents them from running. the system fails to even get close to coming up in a usable state

Dec 31 18:00:40 sparcbook kernel: type00 audit(8.160:3): avc:  denied  { 
execmem } for  pid 8 comm="consoletype" 
scontext=system_u:system_r:consoletype_t:s0 
tcontext=system_u:system_r:consoletype_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type00 audit(8.315:4): avc:  denied  { 
execmem } for  pid!1 comm="hostname" 
scontext=system_u:system_r:hostname_t:s0 
tcontext=system_u:system_r:hostname_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type00 audit(8.520:5): avc:  denied  { 
execmem } for  pid!3 comm="mount" scontext=system_u:system_r:mount_t:s0 
tcontext=system_u:system_r:mount_t:s0 tclass=process
Dec 31 18:00:40 sparcbook kernel: type00 audit(8.570:6): avc:  denied  { 

execmem } for pid 3 comm="readahead-colle" scontext=system_u:system_r:readahead_t:s0 tcontext=system_u:system_r:readahead_t:s0 tclass=process

Quoted 7 lines
Quoted 2 lines

is a small sample of the logs you get not everything fails but almost everything

I think we need to understand why this is happening - it usually reflects a toolchain problem (that was the case in the ppc32 situation, and was later fixed in Fedora through an updated toolchain and rebuilt userland). eu-readelf -l /bin/hostname shows what?

(never received your reply but found it in the mailing list archives)

Your eu-readelf output showed why SELinux is checking execmem - the data segment has flags RWE and thus a private file mapping is being created with PROT_WRITE and PROT_EXEC. That's a problem with the compiler toolchain - report it to them please. This was a problem with ppc32 binaries before secure-plt was introduced.

In the meantime, I'd be interested in knowing whether we truly need to disable all of the exec* checking or if we can in fact just disable this specific case, ala:

Re-enable full SELinux exec* checking on ppc32 since the compiler toolchain has long since been fixed and userland has been rebuilt, and introduce a special case for sparc that only disables the execmem check for private file mappings but retains it for anonymous mappings and retains the other exec* checks.

security/selinux/hooks.c +4 -4
index 63c2d36..9a4d0e4 100644
@@ -3004,8 +3004,11 @@ static int file_map_prot_check(struct file *file, unsigned long prot, int shared
 	const struct cred *cred = current_cred();
 	int rc = 0;
 
-#ifndef CONFIG_PPC32
+#if defined(CONFIG_SPARC)
+	if ((prot & PROT_EXEC) && !file) {
+#else
 	if ((prot & PROT_EXEC) && (!file || (!shared && (prot & PROT_WRITE)))) {
+#endif
 		/*
 		 * We are making executable an anonymous mapping or a
 		 * private file mapping that will also be writable.
@@ -3015,7 +3018,6 @@ static int file_map_prot_check(struct file *file, unsigned long prot, int shared
 		if (rc)
 			goto error;
 	}
-#endif
 
 	if (file) {
 		/* read access is always possible with a mapping */
@@ -3076,7 +3078,6 @@ static int selinux_file_mprotect(struct vm_area_struct *vma,
 	if (selinux_checkreqprot)
 		prot = reqprot;
 
-#ifndef CONFIG_PPC32
 	if ((prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
 		int rc = 0;
 		if (vma->vm_start >= vma->vm_mm->start_brk &&
@@ -3099,7 +3100,6 @@ static int selinux_file_mprotect(struct vm_area_struct *vma,
 		if (rc)
 			return rc;
 	}
-#endif
 
 	return file_map_prot_check(vma->vm_file, prot, vma->vm_flags&VM_SHARED);
 }
Signature · 2 lines
Stephen Smalley
National Security Agency
#7 lore raw
In reply to Tom "spot" Callaway

From: Stephen Smalley <[email protected]>
Date: Thu, 15 Apr 2010 08:43:05 -0400

Quoted 5 lines

Your eu-readelf output showed why SELinux is checking execmem - the data segment has flags RWE and thus a private file mapping is being created with PROT_WRITE and PROT_EXEC. That's a problem with the compiler toolchain - report it to them please. This was a problem with ppc32 binaries before secure-plt was introduced.

I don't really intend to implement secure-plt any time soon on sparc because there simply is no way to do it efficiently.

And when you talk about "toolchain issues" that all goes my way anyways, so just direct such queries to me directly since I handle both the kernel and toolchain bits entirely myself these days.

So you'll always have to deal with the PLT section on sparc having write and execute permission.

#8 lore raw
In reply to Tom "spot" Callaway
Quoted 12 lines · David Miller

From: Stephen Smalley <[email protected]>
Date: Thu, 15 Apr 2010 08:43:05 -0400

Quoted 5 lines

Your eu-readelf output showed why SELinux is checking execmem - the data segment has flags RWE and thus a private file mapping is being created with PROT_WRITE and PROT_EXEC. That's a problem with the compiler toolchain - report it to them please. This was a problem with ppc32 binaries before secure-plt was introduced.

I don't really intend to implement secure-plt any time soon on sparc because there simply is no way to do it efficiently.

Quoted 6 lines

And when you talk about "toolchain issues" that all goes my way anyways, so just direct such queries to me directly since I handle both the kernel and toolchain bits entirely myself these days.

So you'll always have to deal with the PLT section on sparc having write and execute permission.

Ok. Can someone with sparc hardware try the patch I posted to see if it suffices?

Signature · 2 lines
Stephen Smalley
National Security Agency
#9 lore raw
In reply to Tom "spot" Callaway

From: Stephen Smalley <[email protected]>
Date: Fri, 16 Apr 2010 08:36:42 -0400

Quoted 17 lines · David Miller
Quoted 16 lines

From: Stephen Smalley <[email protected]>
Date: Thu, 15 Apr 2010 08:43:05 -0400

Quoted 5 lines

Your eu-readelf output showed why SELinux is checking execmem - the data segment has flags RWE and thus a private file mapping is being created with PROT_WRITE and PROT_EXEC. That's a problem with the compiler toolchain - report it to them please. This was a problem with ppc32 binaries before secure-plt was introduced.

I don't really intend to implement secure-plt any time soon on sparc because there simply is no way to do it efficiently.

And when you talk about "toolchain issues" that all goes my way anyways, so just direct such queries to me directly since I handle both the kernel and toolchain bits entirely myself these days.

Quoted 5 lines
Quoted 2 lines

So you'll always have to deal with the PLT section on sparc having write and execute permission.

Ok. Can someone with sparc hardware try the patch I posted to see if it suffices?

Tom, please give Stephen's patch a test. Thanks!

#10 lore raw
In reply to Tom "spot" Callaway
Quoted 17 lines · Stephen Smalley
Quoted 16 lines · David Miller

From: Stephen Smalley <[email protected]>
Date: Thu, 15 Apr 2010 08:43:05 -0400

Quoted 5 lines

Your eu-readelf output showed why SELinux is checking execmem - the data segment has flags RWE and thus a private file mapping is being created with PROT_WRITE and PROT_EXEC. That's a problem with the compiler toolchain - report it to them please. This was a problem with ppc32 binaries before secure-plt was introduced.

I don't really intend to implement secure-plt any time soon on sparc because there simply is no way to do it efficiently.

And when you talk about "toolchain issues" that all goes my way anyways, so just direct such queries to me directly since I handle both the kernel and toolchain bits entirely myself these days.

Quoted 5 lines
Quoted 2 lines

So you'll always have to deal with the PLT section on sparc having write and execute permission.

Ok. Can someone with sparc hardware try the patch I posted to see if it suffices?

Apologies for the delay. Your patch does not suffice.

With your patch applied, this is the result:

dracut: Mounted root filesystem /dev/mapper/vg_apollo-lv_root
dracut: Loading SELinux policy
type04 audit(1272381939.416:2): enforcing=1 old_enforcing=0 auidB94967295 sesB94967295
type03 audit(1272381940.696:3): policy loaded auidB94967295 sesB94967295
dracut: Switching root
type00 audit(1272381942.195:4): avc: denied { execmem } for pid55 comm="consoletype" scontext=system_u:system_r:consoletype_t:s0 tcontext=system_u:system_r:consoletype_t:s0 tclass=process type00 audit(1272381942.245:5): avc: denied { execmem } for pid59 comm="consoletype" scontext=system_u:system_r:consoletype_t:s0 tcontext=system_u:system_r:consoletype_t:s0 tclass=process type00 audit(1272381942.315:6): avc: denied { execmem } for pid60 comm="hostname" scontext=system_u:system_r:hostname_t:s0 tcontext=system_u:system_r:hostname_t:s0 tclass=process type00 audit(1272381942.356:7): avc: denied { execmem } for pid50 comm="readahead-colle" scontext=system_u:system_r:readahead_t:s0 tcontext=system_u:system_r:readahead_t:s0 tclass=process type00 audit(1272381942.376:8): avc: denied { execmem } for pid63 comm="mount" scontext=system_u:system_r:mount_t:s0 tcontext=system_u:system_r:mount_t:s0 tclass=process type00 audit(1272381942.385:9): avc: denied { execmem } for pid65 comm="consoletype" scontext=system_u:system_r:consoletype_t:s0 tcontext=system_u:system_r:consoletype_t:s0 tclass=process type00 audit(1272381942.396:10): avc: denied { execmem } for pid68 comm="consoletype" scontext=system_u:system_r:consoletype_t:s0 tcontext=system_u:system_r:consoletype_t:s0 tclass=process type00 audit(1272381942.466:11): avc: denied { execmem } for pid77 comm="restorecon" scontext=system_u:system_r:setfiles_t:s0 tcontext=system_u:system_r:setfiles_t:s0 tclass=process
udev: starting version 145
e1000e: Intel(R) PRO/1000 Network Driver - 1.0.2-k2
e1000e: Copyright (c) 1999-2008 Intel Corporation. 0000:08:00.0: eth0: (PCI Express:2.5GB/s:Width x4) 00:14:4f:d4:8a:5a 0000:08:00.0: eth0: Intel(R) PRO/1000 Network Connection 0000:08:00.0: eth0: MAC: 0, PHY: 4, PBA No: ffffff-0ff 0000:08:00.1: eth1: (PCI Express:2.5GB/s:Width x4) 00:14:4f:d4:8a:5b 0000:08:00.1: eth1: Intel(R) PRO/1000 Network Connection 0000:08:00.1: eth1: MAC: 0, PHY: 4, PBA No: ffffff-0ff 0000:09:00.0: eth2: (PCI Express:2.5GB/s:Width x4) 00:14:4f:d4:8a:5c 0000:09:00.0: eth2: Intel(R) PRO/1000 Network Connection 0000:09:00.0: eth2: MAC: 0, PHY: 4, PBA No: ffffff-0ff 0000:09:00.1: eth3: (PCI Express:2.5GB/s:Width x4) 00:14:4f:d4:8a:5d 0000:09:00.1: eth3: Intel(R) PRO/1000 Network Connection 0000:09:00.1: eth3: MAC: 0, PHY: 4, PBA No: ffffff-0ff __ratelimit: 24 callbacks suppressed
type00 audit(1272381946.637:20): avc: denied { execmem } for pid32 comm="mount" scontext=system_u:system_r:mount_t:s0 tcontext=system_u:system_r:mount_t:s0 tclass=process type00 audit(1272381946.637:21): avc: denied { execmem } for pid33 comm="restorecon" scontext=system_u:system_r:setfiles_t:s0 tcontext=system_u:system_r:setfiles_t:s0 tclass=process type00 audit(1272381946.654:22): avc: denied { execmem } for pid34 comm="plymouth" scontext=system_u:system_r:plymouth_t:s0 tcontext=system_u:system_r:plymouth_t:s0 tclass=process type00 audit(1272381946.687:23): avc: denied { execmem } for pid37 comm="hostname" scontext=system_u:system_r:hostname_t:s0 tcontext=system_u:system_r:hostname_t:s0 tclass=process
device-mapper: multipath: version 1.1.0 loaded
type00 audit(1272381947.536:24): avc: denied { execmem } for pid85 comm="mount" scontext=system_u:system_r:mount_t:s0 tcontext=system_u:system_r:mount_t:s0 tclass=process type00 audit(1272381947.546:25): avc: denied { execmem } for pid87 comm="restorecon" scontext=system_u:system_r:setfiles_t:s0 tcontext=system_u:system_r:setfiles_t:s0 tclass=process type00 audit(1272381947.556:26): avc: denied { execmem } for pid90 comm="mount" scontext=system_u:system_r:mount_t:s0 tcontext=system_u:system_r:mount_t:s0 tclass=process type00 audit(1272381947.566:27): avc: denied { execmem } for pid91 comm="mount" scontext=system_u:system_r:mount_t:s0 tcontext=system_u:system_r:mount_t:s0 tclass=process type00 audit(1272381947.566:28): avc: denied { execmem } for pid92 comm="mount" scontext=system_u:system_r:mount_t:s0 tcontext=system_u:system_r:mount_t:s0 tclass=process type00 audit(1272381947.576:29): avc: denied { execmem } for pid93 comm="mount" scontext=system_u:system_r:mount_t:s0 tcontext=system_u:system_r:mount_t:s0 tclass=process Adding 35241968k swap on /dev/mapper/vg_apollo-lv_swap. Priority:-1 extents:1 across:35241968k
__ratelimit: 123 callbacks suppressed
type00 audit(1272381951.656:71): avc: denied { execmem } for pid55 comm="consoletype" scontext=system_u:system_r:consoletype_t:s0 tcontext=system_u:system_r:consoletype_t:s0 tclass=process type00 audit(1272381951.726:72): avc: denied { execmem } for pid61 comm="consoletype" scontext=system_u:system_r:consoletype_t:s0 tcontext=system_u:system_r:consoletype_t:s0 tclass=process
RPC: Registered udp transport module.
RPC: Registered tcp transport module.
RPC: Registered tcp NFSv4.1 backchannel transport module. type00 audit(1272381952.934:73): avc: denied { execmem } for pid41 comm="mount" scontext=system_u:system_r:mount_t:s0 tcontext=system_u:system_r:mount_t:s0 tclass=process type00 audit(1272381952.996:74): avc: denied { execmem } for pid50 comm="consoletype" scontext=system_u:system_r:consoletype_t:s0 tcontext=system_u:system_r:consoletype_t:s0 tclass=process type00 audit(1272381953.146:75): avc: denied { execmem } for pid57 comm="consoletype" scontext=system_u:system_r:consoletype_t:s0 tcontext=system_u:system_r:consoletype_t:s0 tclass=process type00 audit(1272381953.246:76): avc: denied { execmem } for pid61 comm="sshd" scontext=system_u:system_r:sshd_t:s0-s0:c0.c1023 tcontext=system_u:system_r:sshd_t:s0-s0:c0.c1023 tclass=process type00 audit(1272381953.286:77): avc: denied { execmem } for pid68 comm="consoletype" scontext=system_u:system_r:consoletype_t:s0 tcontext=system_u:system_r:consoletype_t:s0 tclass=process type00 audit(1272381953.456:78): avc: denied { execmem } for pid77 comm="sendmail" scontext=system_u:system_r:sendmail_t:s0 tcontext=system_u:system_r:sendmail_t:s0 tclass=process type00 audit(1272381953.464:79): avc: denied { execmem } for pid78 comm="plymouth" scontext=system_u:system_r:plymouth_t:s0 tcontext=system_u:system_r:plymouth_t:s0 tclass=process type00 audit(1272381953.506:80): avc: denied { execmem } for pid82 comm="restorecon" scontext=system_u:system_r:setfiles_t:s0 tcontext=system_u:system_r:setfiles_t:s0 tclass=process __ratelimit: 21 callbacks suppressed
type00 audit(1272381957.135:88): avc: denied { execmem } for pid40 comm="mingetty" scontext=system_u:system_r:getty_t:s0 tcontext=system_u:system_r:getty_t:s0 tclass=process type00 audit(1272381957.135:90): avc: denied { execmem } for pid41 comm="mingetty" scontext=system_u:system_r:getty_t:s0 tcontext=system_u:system_r:getty_t:s0 tclass=process type00 audit(1272381957.135:91): avc: denied { execmem } for pid38 comm="mingetty" scontext=system_u:system_r:getty_t:s0 tcontext=system_u:system_r:getty_t:s0 tclass=process type00 audit(1272381957.135:92): avc: denied { execmem } for pid43 comm="mingetty" scontext=system_u:system_r:getty_t:s0 tcontext=system_u:system_r:getty_t:s0 tclass=process type00 audit(1272381957.135:89): avc: denied { execmem } for pid39 comm="mingetty" scontext=system_u:system_r:getty_t:s0 tcontext=system_u:system_r:getty_t:s0 tclass=process type00 audit(1272381957.135:93): avc: denied { execmem } for pid42 comm="mingetty" scontext=system_u:system_r:getty_t:s0 tcontext=system_u:system_r:getty_t:s0 tclass=process
init: tty4 main process (1938) terminated with status 127
init: tty4 main process ended, respawning
init: tty5 main process (1939) terminated with status 127
init: tty5 main process ended, respawning
init: tty2 main process (1940) terminated with status 127
init: tty2 main process ended, respawning
init: tty3 main process (1941) terminated with status 127
init: tty3 main process ended, respawning
init: tty1 main process (1942) terminated with status 127
init: tty1 main process ended, respawning
init: tty6 main process (1943) terminated with status 127
init: tty6 main process ended, respawning
type00 audit(1272381957.145:94): avc: denied { execmem } for pid44 comm="mingetty" scontext=system_u:system_r:getty_t:s0 tcontext=system_u:system_r:getty_t:s0 tclass=process type00 audit(1272381957.145:95): avc: denied { execmem } for pid45 comm="mingetty" scontext=system_u:system_r:getty_t:s0 tcontext=system_u:system_r:getty_t:s0 tclass=process type00 audit(1272381957.145:96): avc: denied { execmem } for pid46 comm="mingetty" scontext=system_u:system_r:getty_t:s0 tcontext=system_u:system_r:getty_t:s0 tclass=process type00 audit(1272381957.145:97): avc: denied { execmem } for pid47 comm="mingetty" scontext=system_u:system_r:getty_t:s0 tcontext=system_u:system_r:getty_t:s0 tclass=process

Init trails off and the system never goes anywhere.

~spot

#11 lore raw
In reply to Tom "spot" Callaway
Quoted 38 lines · Tom "spot" Callaway
Quoted 17 lines · Stephen Smalley
Quoted 16 lines · David Miller

From: Stephen Smalley <[email protected]>
Date: Thu, 15 Apr 2010 08:43:05 -0400

Quoted 5 lines

Your eu-readelf output showed why SELinux is checking execmem - the data segment has flags RWE and thus a private file mapping is being created with PROT_WRITE and PROT_EXEC. That's a problem with the compiler toolchain - report it to them please. This was a problem with ppc32 binaries before secure-plt was introduced.

I don't really intend to implement secure-plt any time soon on sparc because there simply is no way to do it efficiently.

And when you talk about "toolchain issues" that all goes my way anyways, so just direct such queries to me directly since I handle both the kernel and toolchain bits entirely myself these days.

Quoted 5 lines
Quoted 2 lines

So you'll always have to deal with the PLT section on sparc having write and execute permission.

Ok. Can someone with sparc hardware try the patch I posted to see if it suffices?

Apologies for the delay. Your patch does not suffice.

With your patch applied, this is the result:

dracut: Mounted root filesystem /dev/mapper/vg_apollo-lv_root
dracut: Loading SELinux policy
type04 audit(1272381939.416:2): enforcing=1 old_enforcing=0 auidB94967295 sesB94967295
type03 audit(1272381940.696:3): policy loaded auidB94967295 sesB94967295
dracut: Switching root
type00 audit(1272381942.195:4): avc: denied { execmem } for pid55 comm="consoletype" scontext=system_u:system_r:consoletype_t:s0 tcontext=system_u:system_r:consoletype_t:s0 tclass=process

<snip>

Quoted 1 line

Init trails off and the system never goes anywhere.

Ok, thanks for trying. Could you send a copy of /proc/pid/maps for one of these processes that is triggering an execmem check?

Signature · 2 lines
Stephen Smalley
National Security Agency
#12 lore raw
In reply to Tom "spot" Callaway
Quoted 2 lines · Stephen Smalley

Ok, thanks for trying. Could you send a copy of /proc/pid/maps for one of these processes that is triggering an execmem check?

Mingetty seems like one of the few that is persistent. With the system rebooted into a kernel with the original patch applied, and selinux enabled, here's the info:

[spot@apollo ~]$ ps ax |grep minge

 2174 tty4     Ss+    0:00 /sbin/mingetty tty4
 2175 tty5     Ss+    0:00 /sbin/mingetty tty5
 2176 tty2     Ss+    0:00 /sbin/mingetty tty2
 2177 tty3     Ss+    0:00 /sbin/mingetty tty3
 2178 tty1     Ss+    0:00 /sbin/mingetty tty1
 2179 tty6     Ss+    0:00 /sbin/mingetty tty6

[root@apollo ~]$ cat /proc/2174/maps
00010000-00014000 r-xp 00000000 fd:00 15466577
/sbin/mingetty
00022000-00024000 rwxp 00002000 fd:00 15466577
/sbin/mingetty
00024000-00046000 rwxp 00000000 00:00 0
[heap]
f7d58000-f7ee2000 r-xp 00000000 fd:00 13893929
/lib/libc-2.11.1.so
f7ee2000-f7ef2000 ---p 0018a000 fd:00 13893929
/lib/libc-2.11.1.so
f7ef2000-f7ef4000 r--p 0018a000 fd:00 13893929
/lib/libc-2.11.1.so
f7ef4000-f7ef8000 rwxp 0018c000 fd:00 13893929
/lib/libc-2.11.1.so
f7ef8000-f7efa000 rwxp 00000000 00:00 0
f7efa000-f7efc000 rw-p 00000000 00:00 0
f7f12000-f7f14000 rw-p 00000000 00:00 0
f7f14000-f7f36000 r-xp 00000000 fd:00 13893922
/lib/ld-2.11.1.so
f7f44000-f7f46000 r--p 00020000 fd:00 13893922
/lib/ld-2.11.1.so
f7f46000-f7f48000 rwxp 00022000 fd:00 13893922
/lib/ld-2.11.1.so
ffacc000-ffaf6000 rw-p 00000000 00:00 0
[stack]

Please let me know if this is not what you needed.

~spot

#13 lore raw
In reply to Tom "spot" Callaway

From: "Tom \"spot\" Callaway" <[email protected]>
Date: Tue, 27 Apr 2010 14:20:21 -0400

Quoted 7 lines

[root@apollo ~]$ cat /proc/2174/maps
00010000-00014000 r-xp 00000000 fd:00 15466577
/sbin/mingetty
00022000-00024000 rwxp 00002000 fd:00 15466577
/sbin/mingetty
00024000-00046000 rwxp 00000000 00:00 0
[heap]

SELINUX probably barfs on the executable heap, the PLT is in the HEAP just like powerpc32 and that's why VM_DATA_DEFAULT_FLAGS has to set both executable and writable.

You also can't remove the CONFIG_PPC32 ifdefs in selinux, since because of the VM_DATA_DEFAULT_FLAGS setting used still in that arch, the heap will always have executable permission, just like sparc does. You have to support those binaries forever, whether you like it or not.

Let's just replace the CONFIG_PPC32 ifdef in SELINUX with CONFIG_PPC32 || CONFIG_SPARC as in Tom's original patch and let's be done with this.

In fact I would go through all the arch/ header files and check the VM_DATA_DEFAULT_FLAGS settings and add the necessary new ifdefs to the SELINUX code so that other platforms don't have the pain of having to go through this process too.

Thanks.

#14 lore raw
In reply to Tom "spot" Callaway
Quoted 20 lines · David Miller

From: "Tom \"spot\" Callaway" <[email protected]>
Date: Tue, 27 Apr 2010 14:20:21 -0400

Quoted 7 lines

[root@apollo ~]$ cat /proc/2174/maps
00010000-00014000 r-xp 00000000 fd:00 15466577
/sbin/mingetty
00022000-00024000 rwxp 00002000 fd:00 15466577
/sbin/mingetty
00024000-00046000 rwxp 00000000 00:00 0
[heap]

SELINUX probably barfs on the executable heap, the PLT is in the HEAP just like powerpc32 and that's why VM_DATA_DEFAULT_FLAGS has to set both executable and writable.

You also can't remove the CONFIG_PPC32 ifdefs in selinux, since because of the VM_DATA_DEFAULT_FLAGS setting used still in that arch, the heap will always have executable permission, just like sparc does. You have to support those binaries forever, whether you like it or not.

Quoted 8 lines

Let's just replace the CONFIG_PPC32 ifdef in SELINUX with CONFIG_PPC32 || CONFIG_SPARC as in Tom's original patch and let's be done with this.

In fact I would go through all the arch/ header files and check the VM_DATA_DEFAULT_FLAGS settings and add the necessary new ifdefs to the SELINUX code so that other platforms don't have the pain of having to go through this process too.

To avoid maintaining per-arch ifdefs, it seems that we could just directly use (VM_DATA_DEFAULT_FLAGS & VM_EXEC) as the basis for deciding
whether to enable or disable these checks. VM_DATA_DEFAULT_FLAGS isn't
constant on some architectures but instead depends on current->personality, but we want this applied uniformly. So we'll just use the initial task state to determine whether or not to enable these checks.

security/selinux/hooks.c +8 -6
index ebee467..a03fd74 100644
@@ -2999,13 +2999,15 @@ static int selinux_file_ioctl(struct file *file, unsigned int cmd,
 	return file_has_perm(cred, file, av);
 }
 
+static int default_noexec;
+
 static int file_map_prot_check(struct file *file, unsigned long prot, int shared)
 {
 	const struct cred *cred = current_cred();
 	int rc = 0;
 
-#ifndef CONFIG_PPC32
-	if ((prot & PROT_EXEC) && (!file || (!shared && (prot & PROT_WRITE)))) {
+	if (default_noexec &&
+	    (prot & PROT_EXEC) && (!file || (!shared && (prot & PROT_WRITE)))) {
 		/*
 		 * We are making executable an anonymous mapping or a
 		 * private file mapping that will also be writable.
@@ -3015,7 +3017,6 @@ static int file_map_prot_check(struct file *file, unsigned long prot, int shared
 		if (rc)
 			goto error;
 	}
-#endif
 
 	if (file) {
 		/* read access is always possible with a mapping */
@@ -3076,8 +3077,8 @@ static int selinux_file_mprotect(struct vm_area_struct *vma,
 	if (selinux_checkreqprot)
 		prot = reqprot;
 
-#ifndef CONFIG_PPC32
-	if ((prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
+	if (default_noexec &&
+	    (prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
 		int rc = 0;
 		if (vma->vm_start >= vma->vm_mm->start_brk &&
 		    vma->vm_end <= vma->vm_mm->brk) {
@@ -3099,7 +3100,6 @@ static int selinux_file_mprotect(struct vm_area_struct *vma,
 		if (rc)
 			return rc;
 	}
-#endif
 
 	return file_map_prot_check(vma->vm_file, prot, vma->vm_flags&VM_SHARED);
 }
@@ -5662,6 +5662,8 @@ static __init int selinux_init(void)
 	/* Set the security state for the initial task. */
 	cred_init_security();
 
+	default_noexec = !(VM_DATA_DEFAULT_FLAGS & VM_EXEC);
+
 	sel_inode_cache = kmem_cache_create("selinux_inode_security",
 					    sizeof(struct inode_security_struct),
 					    0, SLAB_PANIC, NULL);
Signature · 2 lines
Stephen Smalley
National Security Agency
Acked-by on v1 1/1
#15 lore raw
In reply to Tom "spot" Callaway

From: Stephen Smalley <[email protected]>
Date: Wed, 28 Apr 2010 15:57:57 -0400

Quoted 9 lines

To avoid maintaining per-arch ifdefs, it seems that we could just directly use (VM_DATA_DEFAULT_FLAGS & VM_EXEC) as the basis for deciding
whether to enable or disable these checks. VM_DATA_DEFAULT_FLAGS isn't
constant on some architectures but instead depends on current->personality, but we want this applied uniformly. So we'll just use the initial task state to determine whether or not to enable these checks.

#16 lore raw
In reply to Tom "spot" Callaway
Quoted 9 lines · Stephen Smalley

To avoid maintaining per-arch ifdefs, it seems that we could just directly use (VM_DATA_DEFAULT_FLAGS & VM_EXEC) as the basis for deciding
whether to enable or disable these checks. VM_DATA_DEFAULT_FLAGS isn't
constant on some architectures but instead depends on current->personality, but we want this applied uniformly. So we'll just use the initial task state to determine whether or not to enable these checks.

Applied to
git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/security-testing-2.6#next

Signature · 2 lines
James Morris
<[email protected]>