Skip to contentSkip to conversation

dhcp-server: Stricter start_ip/end_ip checks

PATCH In review Andrew Zaborowski 3 participants 5 messages Mar 3, 2021 – Mar 4, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 dhcp-server: Stricter start_ip/end_ip checks

Add basic checks that start_ip/end_ip are in the subnet defined by server->address and server->netmask. Exclude server->address from allowed client addresses since we don't prohibit a start_ip/end_ip combination that contains server->address (and that's Ok). Slightly modify how the default end_ip is derived from server->address to work with any prefix_length -- that seems to have been the only place that made assumptions about the prefix length.

---
 ell/dhcp-server.c | 34 ++++++++++++++++++++++++++++++----
 1 file changed, 30 insertions(+), 4 deletions(-)
ell/dhcp-server.c +30 -4
index 3bc1eac..68b6f13 100644
@@ -323,7 +323,10 @@ static bool check_requested_ip(struct l_dhcp_server *server,
 	if (ntohl(requested_nip) < server->start_ip)
 		return false;
 
-	if (htonl(requested_nip) > server->end_ip)
+	if (ntohl(requested_nip) > server->end_ip)
+		return false;
+
+	if (requested_nip == server->address)
 		return false;
 
 	lease = find_lease_by_ip(server->lease_list, requested_nip);
@@ -361,6 +364,9 @@ static uint32_t find_free_or_expired_ip(struct l_dhcp_server *server,
 		if ((ip_addr & 0xff) == 0xff)
 			continue;
 
+		if (ip_nl == server->address)
+			continue;
+
 		/*
 		 * Search both active and expired leases. If this exausts all
 		 * IP's in the range pop the expired list (oldest expired lease)
@@ -778,13 +784,33 @@ LIB_EXPORT bool l_dhcp_server_start(struct l_dhcp_server *server)
 
 	/*
 	 * Assign a default ip range if not already. This will default to
-	 * server->address + 1 ... 254
+	 * server->address + 1 ... subnet end address - 1
 	 */
 	if (!server->start_ip) {
-		server->start_ip = L_BE32_TO_CPU(server->address) + 1;
-		server->end_ip = (server->start_ip & 0xffffff00) | 0xfe;
+		server->start_ip = ntohl(server->address) + 1;
+		server->end_ip = ntohl(server->address) |
+			(~ntohl(server->netmask));
+	} else {
+		if ((server->start_ip ^ ntohl(server->address)) &
+				ntohl(server->netmask))
+			return false;
+
+		if ((server->end_ip ^ ntohl(server->address)) &
+				ntohl(server->netmask))
+			return false;
 	}
 
+	/*
+	 * We skip over IPs ending in 0 or 255 when selecting a free address
+	 * later on but make sure end_ip is not 0xffffffff so we can use
+	 * "<= server->end_ip" safely in the loop condition.
+	 */
+	if ((server->end_ip & 0xff) == 255)
+		server->end_ip--;
+
+	if (server->start_ip > server->end_ip)
+		return false;
+
 	if (!server->ifname) {
 		server->ifname = l_net_get_name(server->ifindex);
 
Signature · 1 line
2.27.0
#1 lore raw

PATCH 2/2 dhcp-server: Drop a redundant cast

---
 ell/dhcp-server.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
ell/dhcp-server.c +1 -1
index 68b6f13..67037cf 100644
@@ -892,7 +892,7 @@ LIB_EXPORT bool l_dhcp_server_set_ip_range(struct l_dhcp_server *server,
 	if (unlikely(!server || !start_ip || !end_ip))
 		return false;
 
-	if (inet_aton((const char *)start_ip, &_host_addr) == 0)
+	if (inet_aton(start_ip, &_host_addr) == 0)
 		return false;
 
 	start = ntohl(_host_addr.s_addr);
Signature · 1 line
2.27.0
#2 lore raw
In reply to Andrew Zaborowski

Hi Andrew,

Quoted 11 lines · Andrew Zaborowski

Add basic checks that start_ip/end_ip are in the subnet defined by server->address and server->netmask. Exclude server->address from allowed client addresses since we don't prohibit a start_ip/end_ip combination that contains server->address (and that's Ok). Slightly modify how the default end_ip is derived from server->address to work with any prefix_length -- that seems to have been the only place that made assumptions about the prefix length.

---
  ell/dhcp-server.c | 34 ++++++++++++++++++++++++++++++----
  1 file changed, 30 insertions(+), 4 deletions(-)

Both applied, thanks.

Regards,
-Denis

#3 lore raw
In reply to Andrew Zaborowski

Em 03/03/2021 18:35, Andrew Zaborowski escreveu:

Quoted 16 lines

Add basic checks that start_ip/end_ip are in the subnet defined by server->address and server->netmask. Exclude server->address from allowed client addresses since we don't prohibit a start_ip/end_ip combination that contains server->address (and that's Ok). Slightly modify how the default end_ip is derived from server->address to work with any prefix_length -- that seems to have been the only place that made assumptions about the prefix length.

---
  ell/dhcp-server.c | 34 ++++++++++++++++++++++++++++++----
  1 file changed, 30 insertions(+), 4 deletions(-)
ell/dhcp-server.c +0 -0
index 3bc1eac..68b6f13 100644

ell/dhcp-server.c static bool check_requested_ip(struct l_dhcp_server *server, · line 326

Quoted 5 lines
ell/dhcp-server.c +1 -1
@@ -323,7 +323,10 @@ static bool check_requested_ip(struct l_dhcp_server *server,
  	if (ntohl(requested_nip) < server->start_ip)
  		return false;
 
-	if (htonl(requested_nip) > server->end_ip)
+	if (ntohl(requested_nip) > server->end_ip)

The commit message didn't mention anything about fixing wrong htonl usage, so is this change correct?

Quoted 54 lines
ell/dhcp-server.c +29 -3
+		return false;
+
+	if (requested_nip == server->address)
  		return false;
 
  	lease = find_lease_by_ip(server->lease_list, requested_nip);
@@ -361,6 +364,9 @@ static uint32_t find_free_or_expired_ip(struct l_dhcp_server *server,
  		if ((ip_addr & 0xff) == 0xff)
  			continue;
 
+		if (ip_nl == server->address)
+			continue;
+
  		/*
  		 * Search both active and expired leases. If this exausts all
  		 * IP's in the range pop the expired list (oldest expired lease)
@@ -778,13 +784,33 @@ LIB_EXPORT bool l_dhcp_server_start(struct l_dhcp_server *server)
 
  	/*
  	 * Assign a default ip range if not already. This will default to
-	 * server->address + 1 ... 254
+	 * server->address + 1 ... subnet end address - 1
  	 */
  	if (!server->start_ip) {
-		server->start_ip = L_BE32_TO_CPU(server->address) + 1;
-		server->end_ip = (server->start_ip & 0xffffff00) | 0xfe;
+		server->start_ip = ntohl(server->address) + 1;
+		server->end_ip = ntohl(server->address) |
+			(~ntohl(server->netmask));
+	} else {
+		if ((server->start_ip ^ ntohl(server->address)) &
+				ntohl(server->netmask))
+			return false;
+
+		if ((server->end_ip ^ ntohl(server->address)) &
+				ntohl(server->netmask))
+			return false;
  	}
 
+	/*
+	 * We skip over IPs ending in 0 or 255 when selecting a free address
+	 * later on but make sure end_ip is not 0xffffffff so we can use
+	 * "<= server->end_ip" safely in the loop condition.
+	 */
+	if ((server->end_ip & 0xff) == 255)
+		server->end_ip--;
+
+	if (server->start_ip > server->end_ip)
+		return false;
+
  	if (!server->ifname) {
  		server->ifname = l_net_get_name(server->ifindex);
 
#4 lore raw

Hi Érico,

Quoted 6 lines
Quoted 2 lines

-    if (htonl(requested_nip) > server->end_ip)
+    if (ntohl(requested_nip) > server->end_ip)

The commit message didn't mention anything about fixing wrong htonl usage, so is this change correct?

Eagle eyes ;)

But yes, the change is correct. In the end both ntohl and htonl resolve to a no_op or bswap32. So the end result is the same.

Not sure why it was originally written as htonl, but ntohl is more logically correct since requested_nip is is network order and end_ip is in host order.

Ideally there should have been a separate fixup commit for this...

Regards,
-Denis