Skip to contentSkip to conversation

kill13, CVE-2018-10124: Reproduce INT_MIN negation

PATCH Superseded Richard Palethorpe 3 participants 6 messages Jul 7, 2021

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 kill13, CVE-2018-10124: Reproduce INT_MIN negation

---
 runtest/cve                               |  1 +
 runtest/syscalls                          |  1 +
 testcases/kernel/syscalls/kill/.gitignore |  1 +
 testcases/kernel/syscalls/kill/kill13.c   | 35 +++++++++++++++++++++++
 4 files changed, 38 insertions(+)
 create mode 100644 testcases/kernel/syscalls/kill/kill13.c
runtest/cve +1 -0
index 5a6ef966d..226b5ea44 100644
@@ -49,6 +49,7 @@ cve-2018-5803 sctp_big_chunk
 cve-2018-7566 snd_seq01
 cve-2018-8897 ptrace09
 cve-2018-9568 connect02
+cve-2018-10124 kill13
 cve-2018-1000001 realpath01
 cve-2018-1000199 ptrace08
 cve-2018-1000204 ioctl_sg01
runtest/syscalls +1 -0
index 98fe3c02e..0c1e16f9e 100644
@@ -644,6 +644,7 @@ kill09 kill09
 kill10 kill10
 kill11 kill11
 kill12 kill12
+kill13 kill13
 
 lchown01 lchown01
 lchown01_16 lchown01_16
testcases/kernel/syscalls/kill/.gitignore +1 -0
index 75fdaa561..810ed0200 100644
@@ -8,3 +8,4 @@
 /kill10
 /kill11
 /kill12
+/kill13
testcases/kernel/syscalls/kill/kill13.c new file +35 -0
new file mode 100644
index 000000000..b5afb653f
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2021 SUSE LLC <[email protected]>
+ */
+
+/*\
+ * [Description]
+ *
+ * Reproducer of CVE-2018-10124; INT_MIN negation.
+ *
+ * Most likely this test will always pass without UBSAN
+ * enabled. Perhaps unless negating INT_MIN results in -1 or
+ * 0. However on my computer it just results in INT_MIN.
+ *
+ */
+
+#include <limits.h>
+#include <signal.h>
+#include "tst_test.h"
+
+static void run(void)
+{
+	TST_EXP_FAIL2(kill(INT_MIN, 1000), ESRCH,
+		      "kill(INT_MIN, ...) fails with ESRCH");
+}
+
+static struct tst_test test = {
+	.test_all = run,
+	.taint_check = TST_TAINT_W | TST_TAINT_D,
+	.tags = (const struct tst_tag[]) {
+		{"linux-git", "4ea77014af0d"},
+		{"CVE", "CVE-2018-10124"},
+		{}
+	}
+};
Signature · 1 line
2.31.1
#1 lore raw

PATCH 1/1 kill13, CVE-2018-10124: Reproduce INT_MIN negation

Quoted 1 line · Richard Palethorpe
Quoted 82 lines
---
 runtest/cve                               |  1 +
 runtest/syscalls                          |  1 +
 testcases/kernel/syscalls/kill/.gitignore |  1 +
 testcases/kernel/syscalls/kill/kill13.c   | 35 +++++++++++++++++++++++
 4 files changed, 38 insertions(+)
 create mode 100644 testcases/kernel/syscalls/kill/kill13.c
runtest/cve +1 -0
index 5a6ef966d..226b5ea44 100644
@@ -49,6 +49,7 @@ cve-2018-5803 sctp_big_chunk
 cve-2018-7566 snd_seq01
 cve-2018-8897 ptrace09
 cve-2018-9568 connect02
+cve-2018-10124 kill13
 cve-2018-1000001 realpath01
 cve-2018-1000199 ptrace08
 cve-2018-1000204 ioctl_sg01
runtest/syscalls +1 -0
index 98fe3c02e..0c1e16f9e 100644
@@ -644,6 +644,7 @@ kill09 kill09
 kill10 kill10
 kill11 kill11
 kill12 kill12
+kill13 kill13
 
 lchown01 lchown01
 lchown01_16 lchown01_16
testcases/kernel/syscalls/kill/.gitignore +1 -0
index 75fdaa561..810ed0200 100644
@@ -8,3 +8,4 @@
 /kill10
 /kill11
 /kill12
+/kill13
testcases/kernel/syscalls/kill/kill13.c new file +35 -0
new file mode 100644
index 000000000..b5afb653f
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2021 SUSE LLC <[email protected]>
+ */
+
+/*\
+ * [Description]
+ *
+ * Reproducer of CVE-2018-10124; INT_MIN negation.
+ *
+ * Most likely this test will always pass without UBSAN
+ * enabled. Perhaps unless negating INT_MIN results in -1 or
+ * 0. However on my computer it just results in INT_MIN.
+ *
+ */
+
+#include <limits.h>
+#include <signal.h>
+#include "tst_test.h"
+
+static void run(void)
+{
+	TST_EXP_FAIL2(kill(INT_MIN, 1000), ESRCH,
+		      "kill(INT_MIN, ...) fails with ESRCH");
+}
+
+static struct tst_test test = {
+	.test_all = run,
+	.taint_check = TST_TAINT_W | TST_TAINT_D,
+	.tags = (const struct tst_tag[]) {
+		{"linux-git", "4ea77014af0d"},
+		{"CVE", "CVE-2018-10124"},
+		{}
+	}
+};
Signature · 2 lines
Thank you,
Richard.
#2 lore raw

PATCH 1/1 kill13, CVE-2018-10124: Reproduce INT_MIN negation

In reply to Richard Palethorpe

Hi Richard,

Quoted 51 lines · Richard Palethorpe via ltp
---
  runtest/cve                               |  1 +
  runtest/syscalls                          |  1 +
  testcases/kernel/syscalls/kill/.gitignore |  1 +
  testcases/kernel/syscalls/kill/kill13.c   | 35 +++++++++++++++++++++++
  4 files changed, 38 insertions(+)

create mode 100644 testcases/kernel/syscalls/kill/kill13.c

runtest/cve +1 -0
index 5a6ef966d..226b5ea44 100644
@@ -49,6 +49,7 @@ cve-2018-5803 sctp_big_chunk
  cve-2018-7566 snd_seq01
  cve-2018-8897 ptrace09
  cve-2018-9568 connect02
+cve-2018-10124 kill13
  cve-2018-1000001 realpath01
  cve-2018-1000199 ptrace08
  cve-2018-1000204 ioctl_sg01
runtest/syscalls +1 -0
index 98fe3c02e..0c1e16f9e 100644
@@ -644,6 +644,7 @@ kill09 kill09
  kill10 kill10
  kill11 kill11
  kill12 kill12
+kill13 kill13
 
  lchown01 lchown01
  lchown01_16 lchown01_16
testcases/kernel/syscalls/kill/.gitignore +1 -0
index 75fdaa561..810ed0200 100644
@@ -8,3 +8,4 @@
  /kill10
  /kill11
  /kill12
+/kill13
testcases/kernel/syscalls/kill/kill13.c new file +3 -0
new file mode 100644
index 000000000..b5afb653f
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2021 SUSE LLC <[email protected]>

testcases/kernel/syscalls/kill/kill13.c line 13

Quoted 10 lines
testcases/kernel/syscalls/kill/kill13.c new file +10 -0
+ */
+
+/*\
+ * [Description]
+ *
+ * Reproducer of CVE-2018-10124; INT_MIN negation.
+ *
+ * Most likely this test will always pass without UBSAN
+ * enabled. Perhaps unless negating INT_MIN results in -1 or
+ * 0. However on my computer it just results in INT_MIN.

Results in INT_MIN? What does this mean or did you meant to write results in ESRCH?
Default negating int min in two's complement yields 0 (INT_MIN = b1000... and negating means inverting the first bit -> thus 0).

testcases/kernel/syscalls/kill/kill13.c line 24

Quoted 11 lines
testcases/kernel/syscalls/kill/kill13.c new file +11 -0
+ *
+ */
+
+#include <limits.h>
+#include <signal.h>
+#include "tst_test.h"
+
+static void run(void)
+{
+	TST_EXP_FAIL2(kill(INT_MIN, 1000), ESRCH,
+		      "kill(INT_MIN, ...) fails with ESRCH");

If at some point the kernel decides to change the order of pid and signal validation, it will probably return EINVAL.
Maybe using signal 0 would be better here? It does not "do" anything, but validates the given pid.

Joerg

#3 lore raw

PATCH 1/1 kill13, CVE-2018-10124: Reproduce INT_MIN negation

Hi,

Quoted 8 lines · Joerg Vehlow

Hi Richard,

Quoted 1 line · Richard Palethorpe via ltp

Results in INT_MIN? What does this mean or did you meant to write results in ESRCH?
Default negating int min in two's complement yields 0 (INT_MIN = b1000... and negating means inverting the first bit -> thus 0).

Nevermind... Just checked. Looks like somehow x86 keeps -INT_MIN. Now I know what you mean with "results in INT_MIN"

Joerg

#4 lore raw

PATCH 1/1 kill13, CVE-2018-10124: Reproduce INT_MIN negation

Hello Joerg,

Quoted 7 lines · Joerg Vehlow

Hi,

Quoted 4 lines · Joerg Vehlow

Hi Richard,

Quoted 1 line · Richard Palethorpe via ltp
Quoted 8 lines
Quoted 4 lines

Results in INT_MIN? What does this mean or did you meant to write results in ESRCH?
Default negating int min in two's complement yields 0 (INT_MIN = b1000... and negating means inverting the first bit -> thus 0).

Nevermind... Just checked. Looks like somehow x86 keeps -INT_MIN. Now I know what you mean with "results in INT_MIN"

Joerg

Actually this seems to be expected unless the CPU takes action to prvent it. As you invert all the bits then add one in two's complement. So inverting results in INT_MAX then we add one and overflow back to INT_MIN.

It is undefined though so the CPU could decide to do something else. OTOH it seems quite unlikely we will get anything other than INT_MIN or INT_MAX. So I will set sig number to zero as you suggested and we will just have to rely on UBSAN.

Signature · 2 lines
Thank you,
Richard.
#5 lore raw

PATCH 1/1 kill13, CVE-2018-10124: Reproduce INT_MIN negation

Hi Richard,

Quoted 14 lines · Richard Palethorpe

Hello Joerg,

Quoted 10 lines · Joerg Vehlow

Hi,

Quoted 7 lines · Joerg Vehlow

Hi Richard,

On 7/7/2021 9:18 AM, Richard Palethorpe via ltp wrote:
Results in INT_MIN? What does this mean or did you meant to write results in ESRCH?
Default negating int min in two's complement yields 0 (INT_MIN = b1000... and negating means inverting the first bit -> thus 0).

Quoted 8 lines
Quoted 4 lines

Nevermind... Just checked. Looks like somehow x86 keeps -INT_MIN. Now I know what you mean with "results in INT_MIN"

Joerg

Actually this seems to be expected unless the CPU takes action to prvent it. As you invert all the bits then add one in two's complement. So inverting results in INT_MAX then we add one and overflow back to INT_MIN.

Ahh yes totally had the wrong model of * -1 in my head... Of course it is invert everything and add +1.
That of course means -INT_MIN == INT_MIN.

Quoted 5 lines

It is undefined though so the CPU could decide to do something else. OTOH it seems quite unlikely we will get anything other than INT_MIN or INT_MAX. So I will set sig number to zero as you suggested and we will just have to rely on UBSAN.

With this change:

J?rg