Skip to contentSkip to conversation

tst_taint: TCONF when kernel is alreay tainted

RFC PATCH New Li Wang 3 participants 7 messages Sep 18, 2019 – Oct 15, 2019 Library code

lore
Share & tools
lore

Li Wang

author
#0 lore raw

RFC PATCH 1/1 tst_taint: TCONF when kernel is alreay tainted

As the tst_taint_init comments described, If the tainted-flags are already set by the kernel, there is no reason to continue and TCONF is generated. But in the function achieve, it uses TBROK.

cmdline="cve-2017-17053"
tst_test.c:1096: INFO: Timeout per run is 0h 10m 00s tst_taint.c:88: BROK: Kernel is already tainted: 536871424

Cc Chang Yin <[email protected]>
Cc Michael Moese <[email protected]>

---

Notes:
Hi Cyril & Michael,

I'm not sure if that's a good choice to skip whole test if the kernel already tainted, so maybe we could also perform it but not do a strict tainted-flags check after testing? The reason I think is it probably have chance to find new issue even with a tainted kernel.

Any suggestion?

 lib/tst_taint.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/tst_taint.c +1 -1
index a5dbf77d2..f7cafb96f 100644
@@ -85,7 +85,7 @@ void tst_taint_init(unsigned int mask)
 
 	taint = tst_taint_read();
 	if ((taint & mask) != 0)
-		tst_brk(TBROK, "Kernel is already tainted: %u", taint);
+		tst_brk(TCONF, "Kernel is already tainted: %u", taint);
 }
 
 
Signature · 1 line
2.20.1
author
#1 lore raw

RFC PATCH 1/1 tst_taint: TCONF when kernel is alreay tainted

Quoted 18 lines · Li Wang

As the tst_taint_init comments described, If the tainted-flags are already set
by the kernel, there is no reason to continue and TCONF is generated. But in
the function achieve, it uses TBROK.

cmdline="cve-2017-17053"
tst_test.c:1096: INFO: Timeout per run is 0h 10m 00s tst_taint.c:88: BROK: Kernel is already tainted: 536871424

Cc Chang Yin <[email protected]>
Cc Michael Moese <[email protected]>

---

Notes:
Hi Cyril & Michael,

Quoted 8 lines

I'm not sure if that's a good choice to skip whole test if the kernel
already
tainted, so maybe we could also perform it but not do a strict
tainted-flags
check after testing? The reason I think is it probably have chance to
find
new issue even with a tainted kernel.

Something maybe change like below, which way is better?

lib/tst_taint.c +6 -3
@@ -7,6 +7,7 @@
 #define TAINT_FILE "/proc/sys/kernel/tainted"
 
 static unsigned int taint_mask = -1;
+static unsigned int taint_check = 1;
 
 static unsigned int tst_taint_read(void)
 {
@@ -84,8 +85,10 @@ void tst_taint_init(unsigned int mask)
        taint_mask = mask;
 
        taint = tst_taint_read();
-       if ((taint & mask) != 0)
-               tst_brk(TCONF, "Kernel is already tainted: %u", taint);
+       if ((taint & mask) != 0) {
+               tst_res(TINFO, "Kernel is already tainted: %u", taint);
+               taint_check = 0;
+       }
 }
 
 
@@ -98,5 +101,5 @@ unsigned int tst_taint_check(void)
 
        taint = tst_taint_read();
 
-       return (taint & taint_mask);
+       return (taint & taint_mask & taint_check);
 }
Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20190918/2a2f7a28/attachment.htm>
#2 lore raw

RFC PATCH 1/1 tst_taint: TCONF when kernel is alreay tainted

In reply to Li Wang

Hi!

Quoted 7 lines

As the tst_taint_init comments described, If the tainted-flags are already set by the kernel, there is no reason to continue and TCONF is generated. But in the function achieve, it uses TBROK.

cmdline="cve-2017-17053"
tst_test.c:1096: INFO: Timeout per run is 0h 10m 00s tst_taint.c:88: BROK: Kernel is already tainted: 536871424

There is a reason for generating TBROK, we do not want the test to be skipped silently in this case. If kernel is tainted something went wrong anyways and we are looking only for a specific flags.

Quoted 37 lines
Cc Chang Yin <[email protected]>
Cc Michael Moese <[email protected]>

---

Notes:
Hi Cyril & Michael,

I'm not sure if that's a good choice to skip whole test if the kernel already tainted, so maybe we could also perform it but not do a strict tainted-flags check after testing? The reason I think is it probably have chance to find new issue even with a tainted kernel.

Any suggestion?

 lib/tst_taint.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/tst_taint.c +1 -1
index a5dbf77d2..f7cafb96f 100644
@@ -85,7 +85,7 @@ void tst_taint_init(unsigned int mask)
 
 	taint = tst_taint_read();
 	if ((taint & mask) != 0)
-		tst_brk(TBROK, "Kernel is already tainted: %u", taint);
+		tst_brk(TCONF, "Kernel is already tainted: %u", taint);
 }
 
 
Signature · 5 lines
2.20.1


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
Signature · 2 lines
Cyril Hrubis
[email protected]
author
#3 lore raw

RFC PATCH 1/1 tst_taint: TCONF when kernel is alreay tainted

Quoted 7 lines · Cyril Hrubis

Hi!

Quoted 1 line

As the tst_taint_init comments described, If the tainted-flags are

already set

Quoted 1 line

by the kernel, there is no reason to continue and TCONF is generated.

But in

Quoted 2 lines

the function achieve, it uses TBROK.

Quoted 8 lines
Quoted 3 lines

cmdline="cve-2017-17053"
tst_test.c:1096: INFO: Timeout per run is 0h 10m 00s tst_taint.c:88: BROK: Kernel is already tainted: 536871424

There is a reason for generating TBROK, we do not want the test to be skipped silently in this case. If kernel is tainted something went wrong anyways and we are looking only for a specific flags.

That's true. The TBROK can highlight the tainted kernel as a good reminder to the tester.

But in a real testing situation, LTP sometimes not being run in the first, after we go here cve-2017-17053, the kernel very probably has already been tainted and reported by other tests. So this new TBROK will drive tester to double-check if something wrong as new.

Imagine that, if there are many test cases invoke a tainted checking(e.g. TST_TAINT_W) in LTP, then all of them will be skipped and report TBORK on such tainted-kernel, that seems not to be friendly to the tester to review the result.

Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20190920/7ebf5063/attachment.htm>
#4 lore raw

RFC PATCH 1/1 tst_taint: TCONF when kernel is alreay tainted

In reply to Cyril Hrubis

----- Original Message -----

Quoted 7 lines

Hi!

Quoted 6 lines

As the tst_taint_init comments described, If the tainted-flags are already set
by the kernel, there is no reason to continue and TCONF is generated. But in
the function achieve, it uses TBROK.

Quoted 6 lines
Quoted 3 lines

cmdline="cve-2017-17053"
tst_test.c:1096: INFO: Timeout per run is 0h 10m 00s tst_taint.c:88: BROK: Kernel is already tainted: 536871424

There is a reason for generating TBROK, we do not want the test to be skipped silently in this case.

It can still run and maybe trigger worse problem. IMO if test wants to report taint flags it should only report _new_ taint flags.

We could add a dummy test to end of runtest file, which would check selected taint flags and report WARN/FAIL, so they are guaranteed to appear on report.

Regards,
Jan

#5 lore raw

RFC PATCH 1/1 tst_taint: TCONF when kernel is alreay tainted

Hi!

Quoted 10 lines
Quoted 10 lines
Quoted 9 lines

As the tst_taint_init comments described, If the tainted-flags are already set
by the kernel, there is no reason to continue and TCONF is generated. But in
the function achieve, it uses TBROK.

cmdline="cve-2017-17053"
tst_test.c:1096: INFO: Timeout per run is 0h 10m 00s tst_taint.c:88: BROK: Kernel is already tainted: 536871424

Quoted 5 lines
Quoted 2 lines

There is a reason for generating TBROK, we do not want the test to be skipped silently in this case.

It can still run and maybe trigger worse problem. IMO if test wants to report taint flags it should only report _new_ taint flags.

IMHO this is more complex than this, we look for a specific subset of taint flags. One is the warn flag, that is not well defined and may be harmless and the second is the died flag, which is set on OOPS or BUG. I I would say that in case of the died flag any subseqent testing is pointless since the machine is in inconsistent state and has to be rebooted anyways. Hence we may weaken the check for a warning flag, i.e. ignore it if it was set at the start of the test, but certainly not for the died flag, which would solve this problem as well.

Quoted 3 lines

We could add a dummy test to end of runtest file, which would check selected taint flags and report WARN/FAIL, so they are guaranteed to appear on report.

Actually the new testrunner runltp-ng checks the taint flags after each test and reboots the machine if needed, which is, as far as I can tell, the right place for such checks.

Signature · 2 lines
Cyril Hrubis
[email protected]
#6 lore raw

RFC PATCH 1/1 tst_taint: TCONF when kernel is alreay tainted

----- Original Message -----

Quoted 19 lines

Hi!

Quoted 12 lines
Quoted 12 lines
Quoted 11 lines

As the tst_taint_init comments described, If the tainted-flags are already
set
by the kernel, there is no reason to continue and TCONF is generated. But
in
the function achieve, it uses TBROK.

cmdline="cve-2017-17053"
tst_test.c:1096: INFO: Timeout per run is 0h 10m 00s tst_taint.c:88: BROK: Kernel is already tainted: 536871424

Quoted 5 lines
Quoted 2 lines

There is a reason for generating TBROK, we do not want the test to be skipped silently in this case.

It can still run and maybe trigger worse problem. IMO if test wants to report taint flags it should only report _new_ taint flags.

Quoted 8 lines

IMHO this is more complex than this, we look for a specific subset of taint flags. One is the warn flag, that is not well defined and may be harmless and the second is the died flag, which is set on OOPS or BUG. I I would say that in case of the died flag any subseqent testing is pointless since the machine is in inconsistent state and has to be rebooted anyways. Hence we may weaken the check for a warning flag, i.e. ignore it if it was set at the start of the test, but certainly not for the died flag, which would solve this problem as well.

OK, I'd vote for getting rid of warn flag then or ignore it on test start. Upstream devel trees tend to trigger lot of random warnings during boot, e.g. a random driver warning from device that isn't even used by LTP.

Quoted 8 lines
Quoted 4 lines

We could add a dummy test to end of runtest file, which would check selected taint flags and report WARN/FAIL, so they are guaranteed to appear on report.

Actually the new testrunner runltp-ng checks the taint flags after each test and reboots the machine if needed, which is, as far as I can tell, the right place for such checks.

It is right place, but I think it (reboot) should be configurable behavior. Some harnesses might want to handle it themselves or just stop testing. For example on 1TB box that takes 45min to boot, I'd probably turn it off.