Skip to contentSkip to conversation

dhcp: fix overflow causing retries to stop

PATCH Superseded James Prestwood 2 participants 4 messages Jan 9, 2024

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 dhcp: fix overflow causing retries to stop

If DHCP is in a SELECTING/REQUESTING state and the number of attempts reached a value where 2 << attempts overflowed an unsigned int the next timeout would become zero, causing DHCP to never retry.

Since 5 attempts results in a value of 64 we can instead just limit the attempts to 5, and set next_timeout to 64 after that as the spec requires.

Fixes f130c448 ("dhcp: Introduce timeout fuzzing")
---
 ell/dhcp.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
ell/dhcp.c +7 -2
index ece3e23..4f8b369 100644
@@ -557,9 +557,14 @@ static void dhcp_client_timeout_resend(struct l_timeout *timeout,
 		 * RFC 2131 Section 4.1:
 		 * "The retransmission delay SHOULD be doubled with subsequent
 		 * retransmissions up to a maximum of 64 seconds.
+		 *
+		 * More than 5 attempts will trigger the maximum, don't increase
+		 * any more to avoid overflowing next_timeout.
 		 */
-		client->attempt += 1;
-		next_timeout = minsize(2 << client->attempt, 64);
+		if (client->attempt > 5)
+			next_timeout = 64;
+		else
+			next_timeout = 2 << client->attempt++;
 		break;
 	case DHCP_STATE_INIT:
 	case DHCP_STATE_INIT_REBOOT:
Signature · 1 line
2.34.1
#1 lore raw

Hi James,

Quoted 7 lines · James Prestwood

If DHCP is in a SELECTING/REQUESTING state and the number of attempts reached a value where 2 << attempts overflowed an unsigned int the next timeout would become zero, causing DHCP to never retry.

Since 5 attempts results in a value of 64 we can instead just limit the attempts to 5, and set next_timeout to 64 after that as the spec requires.

Overflowing next_timeout implies that discovery is taking ~25 minutes? I think I'd rather we gave up and sent a NO_LEASE event after retries hits a certain max number?

Regards,
-Denis

#2 lore raw

Hi Denis,

Quoted 7 lines · Denis Kenzior

Hi James,

Quoted 4 lines · James Prestwood

If DHCP is in a SELECTING/REQUESTING state and the number of attempts reached a value where 2 << attempts overflowed an unsigned int the next timeout would become zero, causing DHCP to never retry.

Quoted 5 lines
Quoted 3 lines

Since 5 attempts results in a value of 64 we can instead just limit the attempts to 5, and set next_timeout to 64 after that as the spec requires.

Overflowing next_timeout implies that discovery is taking ~25 minutes?  I think I'd rather we gave up and sent a NO_LEASE event after retries hits a certain max number?

Yep, this is whats happening, the DHCP server is down. Obviously we could change the logic, but NO_LEASE appears to just restart the DHCP client, i.e. no real difference between trying discovery forever. Should we add another event, L_DHCP_CLIENT_EVENT_TIMEOUT?

Thanks,

james

Quoted 3 lines

Regards,
-Denis

#3 lore raw

Hi James,

Quoted 10 lines · James Prestwood

Hi Denis,

Quoted 7 lines · Denis Kenzior

Hi James,

Quoted 4 lines · James Prestwood

If DHCP is in a SELECTING/REQUESTING state and the number of attempts reached a value where 2 << attempts overflowed an unsigned int the next timeout would become zero, causing DHCP to never retry.

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

Since 5 attempts results in a value of 64 we can instead just limit the attempts to 5, and set next_timeout to 64 after that as the spec requires.

Overflowing next_timeout implies that discovery is taking ~25 minutes?  I think I'd rather we gave up and sent a NO_LEASE event after retries hits a certain max number?

Yep, this is whats happening, the DHCP server is down. Obviously we could change the logic, but NO_LEASE appears to just restart the DHCP client, i.e. no real

I'm not sure why netconfig does that actually. From the comment it seems like it expects NO_LEASE event when the old address could not be obtained. But, if I recall correctly, ell's dhcp client implementation only includes the old address when using RENEW or REBIND, and never on a fresh dhcp_client_start()-up due to privacy policy.

In the end, it is up to netconfig to implement a policy it wants. Re-starting discovery is a valid policy, but I think in iwd's case, we should have netconfig treat 'NO_LEASE' as a DHCP server going down / server not present. Having a max number of retries can facilitate that. For example, if we're connecting and reach max_attempts of 6 or so, then give up and have iwd fail the connection.

Quoted 1 line

difference between trying discovery forever. Should we add another event, L_DHCP_CLIENT_EVENT_TIMEOUT?

l_dhcp_client sends NO_LEASE when it gets a NAK and there's no point in continuing without further intervention (or configuration change). Server not responding seems to fall into the same category, but we can also add an event explicitly for this. Maybe MAX_DISCOVER_ATTEMPTS_REACHED?

Regards,
-Denis