Skip to contentSkip to conversation

ipvs: fix shift-out-of-bounds in ip_vs_rht_desired_size

PATCH In review Julian Anastasov 2 participants 4 messages Apr 27 – Apr 30

Subsystems (4)
lore
Share & tools
lore

Julian Anastasov

author maintainer
#0 lore raw

PATCH 1/1 ipvs: fix shift-out-of-bounds in ip_vs_rht_desired_size

Calling roundup_pow_of_two() with 0 has undefined result:

UBSAN: shift-out-of-bounds in ./include/linux/log2.h:57:13
shift exponent 64 is too large for 64-bit type 'unsigned long'
CPU: 1 UID: 0 PID: 77 Comm: kworker/u8:4 Not tainted syzkaller #0 PREEMPT(full)
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 04/18/2026
Workqueue: events_unbound conn_resize_work_handler
Call Trace:
 <TASK>
 dump_stack_lvl+0xe8/0x150 lib/dump_stack.c:120
 ubsan_epilogue+0xa/0x30 lib/ubsan.c:233
 __ubsan_handle_shift_out_of_bounds+0x385/0x410 lib/ubsan.c:494
 __roundup_pow_of_two include/linux/log2.h:57 [inline]
 ip_vs_rht_desired_size+0x2cf/0x410 net/netfilter/ipvs/ip_vs_core.c:240
 ip_vs_conn_desired_size net/netfilter/ipvs/ip_vs_conn.c:765 [inline]
 conn_resize_work_handler+0x1b6/0x14c0 net/netfilter/ipvs/ip_vs_conn.c:822
 process_one_work kernel/workqueue.c:3302 [inline]
 process_scheduled_works+0xb5d/0x1860 kernel/workqueue.c:3385
 worker_thread+0xa53/0xfc0 kernel/workqueue.c:3466
 kthread+0x388/0x470 kernel/kthread.c:436
 ret_from_fork+0x514/0xb70 arch/x86/kernel/process.c:158
 ret_from_fork_asm+0x1a/0x30 arch/x86/entry/entry_64.S:245
 </TASK>
Reported-by [email protected]
Fixes b655388111cf ("ipvs: add resizable hash tables")
---
 net/netfilter/ipvs/ip_vs_core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/netfilter/ipvs/ip_vs_core.c +1 -1
index f5b7a2047291..ea450944465f 100644
@@ -237,7 +237,7 @@ int ip_vs_rht_desired_size(struct netns_ipvs *ipvs, struct ip_vs_rht *t, int n,
 {
 	if (!t)
 		return 1 << min_bits;
-	n = roundup_pow_of_two(n);
+	n = n ? roundup_pow_of_two(n) : 1;
 	if (lfactor < 0) {
 		int factor = min(-lfactor, max_bits);
 
Signature · 1 line
2.53.0
reviewer
#1 lore raw

From: 'Simon Horman' <[email protected]>

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible. Full review at: https://sashiko.dev
---
ipvs: fix shift-out-of-bounds in ip_vs_rht_desired_size

This commit addresses a shift-out-of-bounds issue when roundup_pow_of_two() is called with 0 in ip_vs_rht_desired_size. It updates the assignment to fall back to 1 if n is 0.

net/netfilter/ipvs/ip_vs_core.c int ip_vs_rht_desired_size(struct netns_ipvs *ipvs, struct ip_vs_rht *t, int n, · line 240

Quoted 10 lines
net/netfilter/ipvs/ip_vs_core.c +1 -1
index f5b7a2047291..ea450944465f 100644
@@ -237,7 +237,7 @@ int ip_vs_rht_desired_size(struct netns_ipvs *ipvs, struct ip_vs_rht *t, int n,
 {
 	if (!t)
 		return 1 << min_bits;
-	n = roundup_pow_of_two(n);
+	n = n ? roundup_pow_of_two(n) : 1;

The parameter n is a signed int. If n happens to be negative, it will evaluate to true and be passed to roundup_pow_of_two().

Because roundup_pow_of_two() casts the input to unsigned long, a negative value like -1 becomes ULONG_MAX. This results in a shift exponent of 64, triggering a shift-out-of-bounds regression. Would it be safer to check for n <= 0 instead?

Additionally, if n exceeds 1.07 billion (1 << 30) on systems with large amounts of RAM, roundup_pow_of_two(n) evaluates to 1UL << 31. Since n is a 32-bit signed int, assigning 1UL << 31 back to n overflows and results in a negative number.

Since subsequent bounds checks use signed comparisons:

Quoted 4 lines
net/netfilter/ipvs/ip_vs_core.c +0 -0
 	if (lfactor < 0) {
 		int factor = min(-lfactor, max_bits);
 

n = min(n, 1 << (max_bits - factor));

These checks will fail to constrain the size correctly.

[ ... ]

net/netfilter/ipvs/ip_vs_core.c

Quoted 3 lines
net/netfilter/ipvs/ip_vs_core.c +0 -0
 	/* Shrink but keep it n * 2 to prevent frequent resizing */
 	return clamp(n << 1, 1 << min_bits, 1 << max_bits);
 }

Eventually, the function returns clamp(n << 1, 1 << min_bits, 1 << max_bits). Shifting a negative number causes undefined behavior, and the clamp operation will force the hash table to its absolute minimum size.

Could this cause millions of connections to be placed into a minimally sized hash table, causing severe collisions during RCU hash lookups?

author maintainer
#2 lore raw

Hello,

Quoted 12 lines · Simon Horman

From: 'Simon Horman' <[email protected]>

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible. Full review at: https://sashiko.dev
---
ipvs: fix shift-out-of-bounds in ip_vs_rht_desired_size

This commit addresses a shift-out-of-bounds issue when roundup_pow_of_two() is called with 0 in ip_vs_rht_desired_size. It updates the assignment to fall back to 1 if n is 0.

Quoted 10 lines

net/netfilter/ipvs/ip_vs_core.c int ip_vs_rht_desired_size(struct netns_ipvs *ipvs, struct ip_vs_rht *t, int n, · line 240

Quoted 10 lines
net/netfilter/ipvs/ip_vs_core.c +1 -1
index f5b7a2047291..ea450944465f 100644
@@ -237,7 +237,7 @@ int ip_vs_rht_desired_size(struct netns_ipvs *ipvs, struct ip_vs_rht *t, int n,
 {
 	if (!t)
 		return 1 << min_bits;
-	n = roundup_pow_of_two(n);
+	n = n ? roundup_pow_of_two(n) : 1;

To overflow it on 64-bit we need 600GB+ memory, the
conns are ~300 bytes.

In v2 I changed it to n > 0. To solve it completely
we may need to introduce some kind of "conn_max" sysctl, now per-netns. It can help to limit the conns which are created GFP_ATOMIC while for other structures we will switch to GFP_KERNEL_ACCOUNT.

Quoted 28 lines

The parameter n is a signed int. If n happens to be negative, it will evaluate to true and be passed to roundup_pow_of_two().

Because roundup_pow_of_two() casts the input to unsigned long, a negative value like -1 becomes ULONG_MAX. This results in a shift exponent of 64, triggering a shift-out-of-bounds regression. Would it be safer to check for n <= 0 instead?

Additionally, if n exceeds 1.07 billion (1 << 30) on systems with large amounts of RAM, roundup_pow_of_two(n) evaluates to 1UL << 31. Since n is a 32-bit signed int, assigning 1UL << 31 back to n overflows and results in a negative number.

Since subsequent bounds checks use signed comparisons:

Quoted 4 lines
net/netfilter/ipvs/ip_vs_core.c +0 -0
 	if (lfactor < 0) {
 		int factor = min(-lfactor, max_bits);
 

n = min(n, 1 << (max_bits - factor));

These checks will fail to constrain the size correctly.

[ ... ]

net/netfilter/ipvs/ip_vs_core.c

Quoted 3 lines
net/netfilter/ipvs/ip_vs_core.c +0 -0
 	/* Shrink but keep it n * 2 to prevent frequent resizing */
 	return clamp(n << 1, 1 << min_bits, 1 << max_bits);
 }
Quoted 6 lines

Eventually, the function returns clamp(n << 1, 1 << min_bits, 1 << max_bits). Shifting a negative number causes undefined behavior, and the clamp operation will force the hash table to its absolute minimum size.

Could this cause millions of connections to be placed into a minimally sized hash table, causing severe collisions during RCU hash lookups?

All that is valid without conn/mem limits.

Regards

--
Julian Anastasov <[email protected]>

reviewer
#3 lore raw
Quoted 31 lines · Julian Anastasov

Hello,

Quoted 12 lines · Simon Horman

From: 'Simon Horman' <[email protected]>

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible. Full review at: https://sashiko.dev
---
ipvs: fix shift-out-of-bounds in ip_vs_rht_desired_size

This commit addresses a shift-out-of-bounds issue when roundup_pow_of_two() is called with 0 in ip_vs_rht_desired_size. It updates the assignment to fall back to 1 if n is 0.

Quoted 10 lines

net/netfilter/ipvs/ip_vs_core.c int ip_vs_rht_desired_size(struct netns_ipvs *ipvs, struct ip_vs_rht *t, int n, · line 240

Quoted 10 lines
net/netfilter/ipvs/ip_vs_core.c +1 -1
index f5b7a2047291..ea450944465f 100644
@@ -237,7 +237,7 @@ int ip_vs_rht_desired_size(struct netns_ipvs *ipvs, struct ip_vs_rht *t, int n,
 {
 	if (!t)
 		return 1 << min_bits;
-	n = roundup_pow_of_two(n);
+	n = n ? roundup_pow_of_two(n) : 1;

To overflow it on 64-bit we need 600GB+ memory, the
conns are ~300 bytes.

Quoted 5 lines

In v2 I changed it to n > 0. To solve it completely
we may need to introduce some kind of "conn_max" sysctl, now per-netns. It can help to limit the conns which are created GFP_ATOMIC while for other structures we will switch to GFP_KERNEL_ACCOUNT.

Thanks, I agree with this approach.

Quoted 29 lines
Quoted 29 lines

The parameter n is a signed int. If n happens to be negative, it will evaluate to true and be passed to roundup_pow_of_two().

Because roundup_pow_of_two() casts the input to unsigned long, a negative value like -1 becomes ULONG_MAX. This results in a shift exponent of 64, triggering a shift-out-of-bounds regression. Would it be safer to check for n <= 0 instead?

Additionally, if n exceeds 1.07 billion (1 << 30) on systems with large amounts of RAM, roundup_pow_of_two(n) evaluates to 1UL << 31. Since n is a 32-bit signed int, assigning 1UL << 31 back to n overflows and results in a negative number.

Since subsequent bounds checks use signed comparisons:

Quoted 4 lines
net/netfilter/ipvs/ip_vs_core.c +0 -0
 	if (lfactor < 0) {
 		int factor = min(-lfactor, max_bits);
 

n = min(n, 1 << (max_bits - factor));

These checks will fail to constrain the size correctly.

[ ... ]

Quoted 3 lines
net/netfilter/ipvs/ip_vs_core.c +0 -0
 	/* Shrink but keep it n * 2 to prevent frequent resizing */
 	return clamp(n << 1, 1 << min_bits, 1 << max_bits);
 }
Quoted 8 lines
Quoted 6 lines

Eventually, the function returns clamp(n << 1, 1 << min_bits, 1 << max_bits). Shifting a negative number causes undefined behavior, and the clamp operation will force the hash table to its absolute minimum size.

Could this cause millions of connections to be placed into a minimally sized hash table, causing severe collisions during RCU hash lookups?

All that is valid without conn/mem limits.

Sorry, my previous email was supposed to come with some commentary that I think that this is a pre-existing issue that can be treated separately from this patch.