Skip to contentSkip to conversation

security/dirtyc0w_shmem: Fix remaining cases where UFFD_FEATURE_MINOR_SHMEM is absent

PATCH In review David Hildenbrand 3 participants 6 messages Nov 25, 2022

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 security/dirtyc0w_shmem: Fix remaining cases where UFFD_FEATURE_MINOR_SHMEM is absent

When UFFD_FEATURE_MINOR_SHMEM is not defined, we still have to make the checkpoint happy, otherwise our parent process will run into a timeout. Further, we have to test if UFFD_FEATURE_MINOR_SHMEM is really returned by the UFFD_API ioctl: if the kernel knows about the feature but doesn't support it, it will be masked off.

---
 .../dirtyc0w_shmem/dirtyc0w_shmem_child.c        | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +12 -4
index cb2e9df0c..c117c6f39 100644
@@ -24,12 +24,12 @@
 #include <linux/userfaultfd.h>
 #endif
 
-#ifdef UFFD_FEATURE_MINOR_SHMEM
-
 #define TST_NO_DEFAULT_MAIN
 #include "tst_test.h"
 #include "tst_safe_macros.h"
 #include "tst_safe_pthread.h"
+
+#ifdef UFFD_FEATURE_MINOR_SHMEM
 #include "lapi/syscalls.h"
 
 #define TMP_DIR "tmp_dirtyc0w_shmem"
@@ -162,6 +162,10 @@ retry:
 			"Could not create userfault file descriptor");
 	}
 
+	if (!(uffdio_api.features & UFFD_FEATURE_MINOR_SHMEM))
+		tst_brk(TCONF,
+			"System does not have userfaultfd minor fault support for shmem");
+
 	uffdio_register.range.start = (unsigned long) map;
 	uffdio_register.range.len = page_size;
 	uffdio_register.mode = UFFDIO_REGISTER_MODE_MINOR;
@@ -236,6 +240,10 @@ int main(void)
 	return 0;
 }
 #else /* UFFD_FEATURE_MINOR_SHMEM */
-#include "tst_test.h"
-TST_TEST_TCONF("System does not have userfaultfd minor fault support for shmem");
+int main(void)
+{
+	tst_reinit();
+	TST_CHECKPOINT_WAKE(0);
+	tst_brk(TCONF, "System does not have userfaultfd minor fault support for shmem");
+}
 #endif /* UFFD_FEATURE_MINOR_SHMEM */
Signature · 5 lines
2.38.1


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#1 lore raw

Hi David,

Quoted 19 lines

When UFFD_FEATURE_MINOR_SHMEM is not defined, we still have to make the checkpoint happy, otherwise our parent process will run into a timeout. Further, we have to test if UFFD_FEATURE_MINOR_SHMEM is really returned by the UFFD_API ioctl: if the kernel knows about the feature but doesn't support it, it will be masked off.

---
 .../dirtyc0w_shmem/dirtyc0w_shmem_child.c        | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +0 -0
index cb2e9df0c..c117c6f39 100644

testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c line 27

Quoted 4 lines
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +0 -1
@@ -24,12 +24,12 @@
 #include <linux/userfaultfd.h>
 #endif
 
-#ifdef UFFD_FEATURE_MINOR_SHMEM

Shouldn't be the check and TST_TEST_TCONF() actually be in dirtyc0w_shmem.c? I overlooked that, but IMHO test does not make sense at all if UFFD_FEATURE_MINOR_SHMEM not defined, right?

Also Martin noted that ("The parent process should not even fork() when UFFD_FEATURE_MINOR_SHMEM is not defined in config.h.").

Therefore one fix would be to move the check to parent and second (maybe better in separate commit) to add check for uffdio_api.features.

Kind regards,
Petr

Quoted 34 lines
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +12 -3
-
 #define TST_NO_DEFAULT_MAIN
 #include "tst_test.h"
 #include "tst_safe_macros.h"
 #include "tst_safe_pthread.h"
+
+#ifdef UFFD_FEATURE_MINOR_SHMEM
 #include "lapi/syscalls.h"
 
 #define TMP_DIR "tmp_dirtyc0w_shmem"
@@ -162,6 +162,10 @@ retry:
 			"Could not create userfault file descriptor");
 	}
 
+	if (!(uffdio_api.features & UFFD_FEATURE_MINOR_SHMEM))
+		tst_brk(TCONF,
+			"System does not have userfaultfd minor fault support for shmem");
+
 	uffdio_register.range.start = (unsigned long) map;
 	uffdio_register.range.len = page_size;
 	uffdio_register.mode = UFFDIO_REGISTER_MODE_MINOR;
@@ -236,6 +240,10 @@ int main(void)
 	return 0;
 }
 #else /* UFFD_FEATURE_MINOR_SHMEM */
-#include "tst_test.h"
-TST_TEST_TCONF("System does not have userfaultfd minor fault support for shmem");
+int main(void)
+{
+	tst_reinit();
+	TST_CHECKPOINT_WAKE(0);
+	tst_brk(TCONF, "System does not have userfaultfd minor fault support for shmem");
+}
 #endif /* UFFD_FEATURE_MINOR_SHMEM */
Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#2 lore raw
Quoted 24 lines · Petr Vorel

Hi David,

Quoted 21 lines

When UFFD_FEATURE_MINOR_SHMEM is not defined, we still have to make the checkpoint happy, otherwise our parent process will run into a timeout. Further, we have to test if UFFD_FEATURE_MINOR_SHMEM is really returned by the UFFD_API ioctl: if the kernel knows about the feature but doesn't support it, it will be masked off.

---
  .../dirtyc0w_shmem/dirtyc0w_shmem_child.c        | 16 ++++++++++++----
  1 file changed, 12 insertions(+), 4 deletions(-)
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +0 -0
index cb2e9df0c..c117c6f39 100644
@@ -24,12 +24,12 @@
  #include <linux/userfaultfd.h>
  #endif
Quoted 8 lines

testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c line 27

Quoted 1 line
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +0 -1
-#ifdef UFFD_FEATURE_MINOR_SHMEM

Shouldn't be the check and TST_TEST_TCONF() actually be in dirtyc0w_shmem.c? I overlooked that, but IMHO test does not make sense at all if UFFD_FEATURE_MINOR_SHMEM not defined, right?

Also Martin noted that ("The parent process should not even fork() when UFFD_FEATURE_MINOR_SHMEM is not defined in config.h.").

I tried that first, but then we can still run into the runtime absence of UFFD_FEATURE_MINOR_SHMEM. Checking that also in the parent resulted in some IMHO unpleasant code while I worked on that.

This is certainly the easiest approach, because we still have to make the child program compile either way.

Anyhow, I'll do whatever you decide, because I want to cross this off my list. So any guidance on how to complete this would be appreciated.

Signature · 7 lines
Thanks,

David / dhildenb


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#3 lore raw
Quoted 34 lines · Petr Vorel
Quoted 24 lines

Hi David,

Quoted 21 lines

When UFFD_FEATURE_MINOR_SHMEM is not defined, we still have to make the checkpoint happy, otherwise our parent process will run into a timeout. Further, we have to test if UFFD_FEATURE_MINOR_SHMEM is really returned by the UFFD_API ioctl: if the kernel knows about the feature but doesn't support it, it will be masked off.

---
  .../dirtyc0w_shmem/dirtyc0w_shmem_child.c        | 16 ++++++++++++----
  1 file changed, 12 insertions(+), 4 deletions(-)
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +0 -0
index cb2e9df0c..c117c6f39 100644
@@ -24,12 +24,12 @@
  #include <linux/userfaultfd.h>
  #endif
Quoted 7 lines

testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c line 27

Quoted 1 line
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +0 -1
-#ifdef UFFD_FEATURE_MINOR_SHMEM

Shouldn't be the check and TST_TEST_TCONF() actually be in dirtyc0w_shmem.c? I overlooked that, but IMHO test does not make sense at all if UFFD_FEATURE_MINOR_SHMEM not defined, right?

Also Martin noted that ("The parent process should not even fork() when UFFD_FEATURE_MINOR_SHMEM is not defined in config.h.").

Quoted 6 lines

I tried that first, but then we can still run into the runtime absence of UFFD_FEATURE_MINOR_SHMEM. Checking that also in the parent resulted in some IMHO unpleasant code while I worked on that.

This is certainly the easiest approach, because we still have to make the child program compile either way.

Right, it needs to be in child. Using TST_TEST_TCONF() also in master does not look to me as too unpleasant code, but take it just a suggestion. Obviously the only requirement is code compiles and works on both defined and undefined UFFD_FEATURE_MINOR_SHMEM.

Quoted 2 lines

Anyhow, I'll do whatever you decide, because I want to cross this off my list. So any guidance on how to complete this would be appreciated.

Understand.

Kind regards,
Petr

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#4 lore raw
In reply to David Hildenbrand

Hi,

Quoted 47 lines · David Hildenbrand

When UFFD_FEATURE_MINOR_SHMEM is not defined, we still have to make the checkpoint happy, otherwise our parent process will run into a timeout. Further, we have to test if UFFD_FEATURE_MINOR_SHMEM is really returned by the UFFD_API ioctl: if the kernel knows about the feature but doesn't support it, it will be masked off.

---
  .../dirtyc0w_shmem/dirtyc0w_shmem_child.c        | 16 ++++++++++++----
  1 file changed, 12 insertions(+), 4 deletions(-)
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +6 -2
index cb2e9df0c..c117c6f39 100644
@@ -24,12 +24,12 @@
  #include <linux/userfaultfd.h>
  #endif
 
-#ifdef UFFD_FEATURE_MINOR_SHMEM
-
  #define TST_NO_DEFAULT_MAIN
  #include "tst_test.h"
  #include "tst_safe_macros.h"
  #include "tst_safe_pthread.h"
+
+#ifdef UFFD_FEATURE_MINOR_SHMEM
  #include "lapi/syscalls.h"
 
  #define TMP_DIR "tmp_dirtyc0w_shmem"
@@ -162,6 +162,10 @@ retry:
  			"Could not create userfault file descriptor");
  	}
 
+	if (!(uffdio_api.features & UFFD_FEATURE_MINOR_SHMEM))
+		tst_brk(TCONF,
+			"System does not have userfaultfd minor fault support for shmem");
+
  	uffdio_register.range.start = (unsigned long) map;
  	uffdio_register.range.len = page_size;
  	uffdio_register.mode = UFFDIO_REGISTER_MODE_MINOR;
@@ -236,6 +240,10 @@ int main(void)
  	return 0;
  }

testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c int main(void) · line 249

Quoted 10 lines
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +6 -2
  #else /* UFFD_FEATURE_MINOR_SHMEM */
-#include "tst_test.h"
-TST_TEST_TCONF("System does not have userfaultfd minor fault support for shmem");
+int main(void)
+{
+	tst_reinit();
+	TST_CHECKPOINT_WAKE(0);
+	tst_brk(TCONF, "System does not have userfaultfd minor fault support for shmem");
+}
  #endif /* UFFD_FEATURE_MINOR_SHMEM */

This would work as a workaround but I'd recommend adding the necessary structures and constants to include/lapi/userfaultfd.h instead. Then you won't need this conditional compilation at all.

I also recommend looking at the fuzzy sync library we use for race conditions:
https://github.com/linux-test-project/ltp/wiki/C-Test-API#133-reproducing-race-conditions

The original dirtyc0w test was written before this library but using it generally makes race condition reproducers faster, simpler and more reliable.

Signature · 11 lines
Martin Doucha   [email protected]
QA Engineer for Software Maintenance
SUSE LINUX, s.r.o.
CORSO IIa
Krizikova 148/34
186 00 Prague 8
Czech Republic


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#5 lore raw
Quoted 56 lines · Martin Doucha

Hi,

Quoted 53 lines · David Hildenbrand

When UFFD_FEATURE_MINOR_SHMEM is not defined, we still have to make the checkpoint happy, otherwise our parent process will run into a timeout. Further, we have to test if UFFD_FEATURE_MINOR_SHMEM is really returned by the UFFD_API ioctl: if the kernel knows about the feature but doesn't support it, it will be masked off.

---
   .../dirtyc0w_shmem/dirtyc0w_shmem_child.c        | 16 ++++++++++++----
   1 file changed, 12 insertions(+), 4 deletions(-)
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +9 -4
index cb2e9df0c..c117c6f39 100644
@@ -24,12 +24,12 @@
   #include <linux/userfaultfd.h>
   #endif
 
-#ifdef UFFD_FEATURE_MINOR_SHMEM
-
   #define TST_NO_DEFAULT_MAIN
   #include "tst_test.h"
   #include "tst_safe_macros.h"
   #include "tst_safe_pthread.h"
+
+#ifdef UFFD_FEATURE_MINOR_SHMEM
   #include "lapi/syscalls.h"
 
   #define TMP_DIR "tmp_dirtyc0w_shmem"
@@ -162,6 +162,10 @@ retry:
   			"Could not create userfault file descriptor");
   	}
 
+	if (!(uffdio_api.features & UFFD_FEATURE_MINOR_SHMEM))
+		tst_brk(TCONF,
+			"System does not have userfaultfd minor fault support for shmem");
+
   	uffdio_register.range.start = (unsigned long) map;
   	uffdio_register.range.len = page_size;
   	uffdio_register.mode = UFFDIO_REGISTER_MODE_MINOR;
@@ -236,6 +240,10 @@ int main(void)
   	return 0;
   }
   #else /* UFFD_FEATURE_MINOR_SHMEM */
-#include "tst_test.h"
-TST_TEST_TCONF("System does not have userfaultfd minor fault support for shmem");
+int main(void)
+{
+	tst_reinit();
Quoted 8 lines

testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c int main(void) · line 249

Quoted 4 lines
testcases/kernel/security/dirtyc0w_shmem/dirtyc0w_shmem_child.c +3 -0
+	TST_CHECKPOINT_WAKE(0);
+	tst_brk(TCONF, "System does not have userfaultfd minor fault support for shmem");
+}
   #endif /* UFFD_FEATURE_MINOR_SHMEM */

This would work as a workaround but I'd recommend adding the necessary structures and constants to include/lapi/userfaultfd.h instead. Then you won't need this conditional compilation at all.

That seems to work as well, although I'm still a bit unsure what to include in that file, what not, what to strip, ...

I'll use include/lapi/io_uring.h as an orientation but will most probably mess it up on the first attempt.

Quoted 5 lines

I also recommend looking at the fuzzy sync library we use for race conditions:
https://github.com/linux-test-project/ltp/wiki/C-Test-API#133-reproducing-race-conditions

I'll look into that as well.

Quoted 3 lines

The original dirtyc0w test was written before this library but using it generally makes race condition reproducers faster, simpler and more reliable.

Thanks

Signature · 7 lines
Thanks,

David / dhildenb


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp