Skip to contentSkip to conversation

lib/vdso: guard clockid before building u32 bitmask

PATCH In review Sun Jian 2 participants 6 messages Jan 14 Generic VDSO library

lore
Share & tools
lore

Sun Jian

author
#0 lore raw

PATCH 1/1 lib/vdso: guard clockid before building u32 bitmask

__cvdso_clock_gettime_common() and __cvdso_clock_getres_common() build a u32 bitmask from clockid_t using "1U << clock". This requires the shift amount to be < 32, otherwise leading to undefined behaviour.

Add a guard to reject out-of-range clockids before building the bitmask. This avoids undefined shifts and silences sparse "shift too big" warnings.

No functional change intended.

---
 lib/vdso/gettimeofday.c | 6 ++++++
 1 file changed, 6 insertions(+)
lib/vdso/gettimeofday.c +6 -0
index 95df0153f05a..eee30f2bc6c6 100644
@@ -298,6 +298,9 @@ __cvdso_clock_gettime_common(const struct vdso_time_data *vd, clockid_t clock,
 	 * Convert the clockid to a bitmask and use it to check which
 	 * clocks are handled in the VDSO directly.
 	 */
+	if ((u32)clock >= 32)
+		return false;
+
 	msk = 1U << clock;
 	if (likely(msk & VDSO_HRES))
 		vc = &vc[CS_HRES_COARSE];
@@ -440,6 +443,9 @@ bool __cvdso_clock_getres_common(const struct vdso_time_data *vd, clockid_t cloc
 	 * Convert the clockid to a bitmask and use it to check which
 	 * clocks are handled in the VDSO directly.
 	 */
+	if ((u32)clock >= 32)
+		return false;
+
 	msk = 1U << clock;
 	if (msk & (VDSO_HRES | VDSO_RAW)) {
 		/*
Signature · 1 line
2.43.0
#1 lore raw
Quoted 19 lines · Sun Jian

__cvdso_clock_gettime_common() and __cvdso_clock_getres_common() build a u32 bitmask from clockid_t using "1U << clock". This requires the shift amount to be < 32, otherwise leading to undefined behaviour.

Add a guard to reject out-of-range clockids before building the bitmask. This avoids undefined shifts and silences sparse "shift too big" warnings.

No functional change intended.

---
 lib/vdso/gettimeofday.c | 6 ++++++
 1 file changed, 6 insertions(+)
lib/vdso/gettimeofday.c +0 -0
index 95df0153f05a..eee30f2bc6c6 100644

lib/vdso/gettimeofday.c __cvdso_clock_gettime_common(const struct vdso_time_data *vd, clockid_t clock, · line 302

Quoted 5 lines
lib/vdso/gettimeofday.c +2 -0
@@ -298,6 +298,9 @@ __cvdso_clock_gettime_common(const struct vdso_time_data *vd, clockid_t clock,
 	 * Convert the clockid to a bitmask and use it to check which
 	 * clocks are handled in the VDSO directly.
 	 */
+	if ((u32)clock >= 32)
+		return false;

Right above is this check:

if (!vdso_clockid_valid(clock))

	return false;

where:

static __always_inline bool vdso_clockid_valid(clockid_t clock)

{
	/* Check for negative values or invalid clocks */
	/* CLOCK_AUX_LAST == 23 */
	return likely((u32) clock <= CLOCK_AUX_LAST);
}

So this looks like a false-positive. Can you post the sparse warning?

Thomas

author
#2 lore raw
Quoted 1 line · Thomas Weißschuh

So this looks like a false-positive. Can you post the sparse warning?

Here is the sparse warning I saw:

arch/x86/entry/vdso/vdso32/../../../../../lib/vdso/gettimeofday.c:454:26:
warning: shift too big (40) for type unsigned long

It is emitted when checking the vdso32 include path via vclock_gettime.c.

Given that vdso_clockid_valid() bounds clock to <= CLOCK_AUX_LAST (23), the shift cannot reach that value in practice. So I agree this is a sparse false positive in this analysis context.

I'll drop this patch rather than adding a redundant runtime check.

Thanks,
Sun

#3 lore raw
Quoted 9 lines · sun jian
Quoted 1 line · Thomas Weißschuh

So this looks like a false-positive. Can you post the sparse warning?

Here is the sparse warning I saw:

arch/x86/entry/vdso/vdso32/../../../../../lib/vdso/gettimeofday.c:454:26:
warning: shift too big (40) for type unsigned long

It is emitted when checking the vdso32 include path via vclock_gettime.c.

On which specific kernel commit are you?

The message points to '} else if (msg & VDSO_AUX) {'. This means that it probably complains about the shift in '#define VDSO_AUX __GENMASK(CLOCK_AUX_LAST, CLOCK_AUX)'. We had problems in that area before, so I'd like to get all the details.

Which sparse version are you using?

(...)

Thomas

author
#4 lore raw
Quoted 1 line · Thomas Weißschuh

On which specific kernel commit are you?

I'm on:
b71e635feefc852405b14620a7fc58c4c80c0f73
("Merge tag 'cgroup-for-6.19-rc5-fixes'")
, no tags available.

Quoted 5 lines

The message points to '} else if (msg & VDSO_AUX) {'. This means that it probably complains about the shift in '#define VDSO_AUX __GENMASK(CLOCK_AUX_LAST, CLOCK_AUX)'. We had problems in that area before, so I'd like to get all the details.

Which sparse version are you using?

Sparse version:

v0.6.4-73-gfbdde312

The warning was reproduced with gcc + CHECK=sparse on x86, via the vdso32 include path (arch/x86/entry/vdso/vdso32/vclock_gettime.c).

Thanks,
Sun

#5 lore raw
Quoted 11 lines · sun jian
Quoted 1 line · Thomas Weißschuh

On which specific kernel commit are you?

I'm on:
b71e635feefc852405b14620a7fc58c4c80c0f73
("Merge tag 'cgroup-for-6.19-rc5-fixes'")
, no tags available.

Quoted 4 lines

The message points to '} else if (msg & VDSO_AUX) {'. This means that it probably complains about the shift in '#define VDSO_AUX __GENMASK(CLOCK_AUX_LAST, CLOCK_AUX)'. We had problems in that area before, so I'd like to get all the details.

Quoted 7 lines
Quoted 1 line

Which sparse version are you using?

Sparse version:

v0.6.4-73-gfbdde312

The warning was reproduced with gcc + CHECK=sparse on x86, via the vdso32 include path (arch/x86/entry/vdso/vdso32/vclock_gettime.c).

I was able to reproduce the warning. It is a false positive due to the way sparse is configured. The 32-bit compat vDSO is using the 64-bit sparse flags.

See also:
https://lore.kernel.org/lkml/20251107155158-90fb8c9c-59bf-47b3-8756-7c406166db70@linutronix.de/

Thomas