Skip to contentSkip to conversation

ldt.h: Add workaround for x86_64

PATCH In review Ricardo B. Marlière via ltp 2 participants 7 messages May 20, 2025

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 ldt.h: Add workaround for x86_64

From: Ricardo B. Marlière <[email protected]>

The commit be0aaca2f742 ("syscalls/modify_ldt: Add lapi/ldt.h") left behind an important factor of modify_ldt(): the kernel intentionally casts the return value to unsigned int. This was handled in testcases/cve/cve-2015-3290.c but was removed. Add it back to the relevant file.

---
Changes in v3:
- EDITME: describe what is new in this series revision.
- EDITME: use bulletpoints and terse descriptions.
- Link to v2: https://lore.kernel.org/r/[email protected]

Changes in v2:
- Added TBROK for any ret != 0 in modify_ldt call in cve-2015-3290.c
- Link to v1: https://lore.kernel.org/r/[email protected]

---
 include/lapi/ldt.h | 30 +++++++++++++++++++++++++++++-
 1 file changed, 29 insertions(+), 1 deletion(-)
include/lapi/ldt.h +29 -1
index 6b5a2d59cb41bfc24eb5ac26c3d47d49fb8ff78f..2533e50b9c02a974947c6628e9fa20c9c626aa34 100644
@@ -31,7 +31,27 @@ struct user_desc {
 static inline int modify_ldt(int func, const struct user_desc *ptr,
 			     unsigned long bytecount)
 {
-	return tst_syscall(__NR_modify_ldt, func, ptr, bytecount);
+	long rval;
+
+	errno = 0;
+	rval = tst_syscall(__NR_modify_ldt, func, ptr, bytecount);
+
+#ifdef __x86_64__
+	/*
+	 * The kernel intentionally casts modify_ldt() return value
+	 * to unsigned int to prevent sign extension to 64 bits. This may
+	 * result in syscall() returning the value as is instead of setting
+	 * errno and returning -1.
+	 */
+	if (rval > 0 && (int)rval < 0) {
+		tst_res(TINFO,
+			"WARNING: Libc mishandled modify_ldt() return value");
+		errno = -(int)errno;
+		rval = -1;
+	}
+#endif /* __x86_64__ */
+
+	return rval;
 }
 
 static inline int safe_modify_ldt(const char *file, const int lineno, int func,
@@ -40,7 +60,15 @@ static inline int safe_modify_ldt(const char *file, const int lineno, int func,
 {
 	int rval;
 
+	errno = 0;
 	rval = modify_ldt(func, ptr, bytecount);
+#ifdef __x86_64__
+	if (rval == -1 && errno == EINVAL) {
+		tst_brk_(file, lineno, TCONF | TTERRNO,
+			 "modify_ldt(%d, %p, %lu): 16-bit data segments are probably disabled",
+			 func, ptr, bytecount);
+	}
+#endif
 	if (rval == -1) {
 		tst_brk_(file, lineno, TBROK | TERRNO,
 			 "modify_ldt(%d, %p, %lu)", func, ptr, bytecount);

---
base-commit: b070a5692e035ec12c3d3c7a7e9e97c270fd4d7d
change-id: 20250507-fixes-modify_ldt-ebcfdd2a7d30

Best regards,

Signature · 5 lines
Ricardo B. Marlière <[email protected]>


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#1 lore raw
Quoted 8 lines · Ricardo B. Marlière

From: Ricardo B. Marlière <[email protected]>

The commit be0aaca2f742 ("syscalls/modify_ldt: Add lapi/ldt.h") left behind an important factor of modify_ldt(): the kernel intentionally casts the return value to unsigned int. This was handled in testcases/cve/cve-2015-3290.c but was removed. Add it back to the relevant file.

Quoted 8 lines

---
Changes in v3:
- EDITME: describe what is new in this series revision.
- EDITME: use bulletpoints and terse descriptions.
- Link to v2: https://lore.kernel.org/r/[email protected]

Changes in v3:
- Moved the (rval == -1 && errno == EINVAL) check into safe_modify_ldt.

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#2 lore raw
In reply to Ricardo B. Marlière via ltp

Hi,
safe_modify_ldt() should not allow any errors. Let's merge v2.

Quoted 79 lines · Ricardo B. Marlière

From: Ricardo B. Marlière <[email protected]>

The commit be0aaca2f742 ("syscalls/modify_ldt: Add lapi/ldt.h") left behind an important factor of modify_ldt(): the kernel intentionally casts the return value to unsigned int. This was handled in testcases/cve/cve-2015-3290.c but was removed. Add it back to the relevant file.

---
Changes in v3:
- EDITME: describe what is new in this series revision.
- EDITME: use bulletpoints and terse descriptions.
- Link to v2: https://lore.kernel.org/r/[email protected]

Changes in v2:
- Added TBROK for any ret != 0 in modify_ldt call in cve-2015-3290.c
- Link to v1: https://lore.kernel.org/r/[email protected]

---
  include/lapi/ldt.h | 30 +++++++++++++++++++++++++++++-
  1 file changed, 29 insertions(+), 1 deletion(-)
include/lapi/ldt.h +29 -2
index 6b5a2d59cb41bfc24eb5ac26c3d47d49fb8ff78f..2533e50b9c02a974947c6628e9fa20c9c626aa34 100644
@@ -31,7 +31,27 @@ struct user_desc {
  static inline int modify_ldt(int func, const struct user_desc *ptr,
  			     unsigned long bytecount)
  {
-	return tst_syscall(__NR_modify_ldt, func, ptr, bytecount);
+	long rval;
+
+	errno = 0;
+	rval = tst_syscall(__NR_modify_ldt, func, ptr, bytecount);
+
+#ifdef __x86_64__
+	/*
+	 * The kernel intentionally casts modify_ldt() return value
+	 * to unsigned int to prevent sign extension to 64 bits. This may
+	 * result in syscall() returning the value as is instead of setting
+	 * errno and returning -1.
+	 */
+	if (rval > 0 && (int)rval < 0) {
+		tst_res(TINFO,
+			"WARNING: Libc mishandled modify_ldt() return value");
+		errno = -(int)errno;
+		rval = -1;
+	}
+#endif /* __x86_64__ */
+
+	return rval;
  }
 
  static inline int safe_modify_ldt(const char *file, const int lineno, int func,
@@ -40,7 +60,15 @@ static inline int safe_modify_ldt(const char *file, const int lineno, int func,
  {
  	int rval;
 
+	errno = 0;
  	rval = modify_ldt(func, ptr, bytecount);
+#ifdef __x86_64__
+	if (rval == -1 && errno == EINVAL) {
+		tst_brk_(file, lineno, TCONF | TTERRNO,
+			 "modify_ldt(%d, %p, %lu): 16-bit data segments are probably disabled",
+			 func, ptr, bytecount);
+	}
+#endif
  	if (rval == -1) {
  		tst_brk_(file, lineno, TBROK | TERRNO,
  			 "modify_ldt(%d, %p, %lu)", func, ptr, bytecount);
 
---

base-commit: b070a5692e035ec12c3d3c7a7e9e97c270fd4d7d
change-id: 20250507-fixes-modify_ldt-ebcfdd2a7d30

Best regards,

Signature · 10 lines
Martin Doucha   [email protected]
SW Quality Engineer
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
#3 lore raw

Hi,

Quoted 2 lines · Martin Doucha

Hi,
safe_modify_ldt() should not allow any errors. Let's merge v2.

What about the other tests? This check should be applied to them as well eventually.

- Andrea

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#4 lore raw
Quoted 7 lines · Andrea Cervesato

Hi,

Quoted 2 lines · Martin Doucha

Hi,
safe_modify_ldt() should not allow any errors. Let's merge v2.

What about the other tests? This check should be applied to them as well eventually.

That varies by testcase. If the test expects any errors from the call, it should call modify_ldt() directly and handle the return value either directly or using one of the TST_EXP helper macros.

Signature · 10 lines
Martin Doucha   [email protected]
SW Quality Engineer
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
author Acked-by Andrea Cervesato
#5 lore raw

Ok, merged.

Quoted 11 lines · Martin Doucha
Quoted 7 lines · Andrea Cervesato

Hi,

Quoted 2 lines · Martin Doucha

Hi,
safe_modify_ldt() should not allow any errors. Let's merge v2.

What about the other tests? This check should be applied to them as well eventually.

That varies by testcase. If the test expects any errors from the call, it should call modify_ldt() directly and handle the return value either directly or using one of the TST_EXP helper macros.

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#6 lore raw
Quoted 3 lines · Andrea Cervesato

Ok, merged.

Thank you Andrea for the review and suggestion! :)

Quoted 16 lines · Martin Doucha
Quoted 14 lines · Andrea Cervesato
Quoted 8 lines

Hi,

Quoted 2 lines · Martin Doucha

Hi,
safe_modify_ldt() should not allow any errors. Let's merge v2.

What about the other tests? This check should be applied to them as well eventually.

That varies by testcase. If the test expects any errors from the call, it should call modify_ldt() directly and handle the return value either directly or using one of the TST_EXP helper macros.

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp