Merge pull request #77 from danielinux/iface-mtu

Add selection of MTU value per-interface

Author: gasbytes

config.h

@@ -7,6 +7,12 @@
 
 #define ETHERNET
 #define LINK_MTU 1536
+#ifndef LINK_MTU_MIN
+#define LINK_MTU_MIN 64U
+#endif
+#if LINK_MTU < LINK_MTU_MIN
+#error "LINK_MTU must be greater than or equal to LINK_MTU_MIN"
+#endif
 
 #define MAX_TCPSOCKETS 4
 #define MAX_UDPSOCKETS 2

docs/API.md

@@ -27,13 +27,15 @@ struct wolfIP_ll_dev {
     uint8_t mac[6];          // Device MAC address
     char ifname[16];         // Interface name
     uint8_t non_ethernet;    // L3-only link (no Ethernet header/ARP when set)
+    uint32_t mtu;            // Optional internal frame budget, defaults to LINK_MTU
     int (*poll)(struct wolfIP_ll_dev *ll, void *buf, uint32_t len);  // Receive function
     int (*send)(struct wolfIP_ll_dev *ll, void *buf, uint32_t len);  // Transmit function
 };
 ```
 wolfIP maintains an array of these descriptors sized by `WOLFIP_MAX_INTERFACES` (default `1`). Call `wolfIP_getdev_ex()` to access a specific slot; the legacy `wolfIP_getdev()` helper targets the first hardware slot (index `0` normally, or `1` when the optional loopback interface is enabled).
 
 When `non_ethernet` is set, the interface is treated as L3-only point-to-point: the stack skips ARP/neighbor resolution, omits Ethernet headers on transmit, and expects receive buffers to begin at the IP header.
+The `mtu` field still describes wolfIP's internal frame budget including Ethernet headroom, so on non-Ethernet links the payload passed to `ll->send()` is effectively capped at `mtu - ETH_HEADER_LEN` on Ethernet-enabled builds.
 
 ### IP Configuration
 ```c
@@ -203,8 +205,13 @@ Per-interface versions of the IP configuration helpers. The legacy functions tar
 ```c
 struct wolfIP_ll_dev *wolfIP_getdev(struct wolfIP *s);
 struct wolfIP_ll_dev *wolfIP_getdev_ex(struct wolfIP *s, unsigned int if_idx);
+int wolfIP_mtu_set(struct wolfIP *s, unsigned int if_idx, uint32_t mtu);
+int wolfIP_mtu_get(struct wolfIP *s, unsigned int if_idx, uint32_t *mtu);
 ```
 Access the link-layer descriptor(s) that should be wired to hardware drivers. `_ex` returns `NULL` if `if_idx` exceeds `WOLFIP_MAX_INTERFACES`.
+`wolfIP_mtu_set()` updates the effective per-interface MTU, treating `0` as the default `LINK_MTU` and clamping to `[LINK_MTU_MIN, LINK_MTU]`. `wolfIP_mtu_get()` returns the effective MTU currently used by the stack.
+For `non_ethernet` devices this value remains the internal frame budget; the maximum IP bytes handed to the driver are reduced by `ETH_HEADER_LEN` when Ethernet support is compiled in.
+- Returns: `wolfIP_getdev()`/`wolfIP_getdev_ex()` return a pointer to the link-layer descriptor or `NULL` on invalid interface index; `wolfIP_mtu_set()` returns `0` on success or a negative error code on failure; `wolfIP_mtu_get()` returns `0` on success or a negative error code on failure and stores the effective MTU in `*mtu`.
 
 ## DHCP Client Functions