On Mon, Feb 03, 2003 at 12:06:28PM +0100, Pawel Jakub Dawidek wrote: +> We got lock order reversal here: +> +> 1st 0xc0384800 arp mutex (arp mutex) @ /usr/src/sys/netinet/if_ether.c:151 +> 2nd 0xc1886b7c radix node head (radix node head) @ /usr/src/sys/net/route.c:549 +> +> Simple backtrace: +> rtreqest1() [route.c] +> rtreqest() [route.c] +> arptfree() [if_ether.c] +> arptimer() [if_ether.c]
I think that MTX_DUPOK is needed here, so: --- radix.h.orig Sun Feb 2 20:07:42 2003 +++ radix.h Mon Feb 3 21:48:30 2003 @@ -159,7 +159,7 @@ #define RADIX_NODE_HEAD_LOCK_INIT(rnh) \ - mtx_init(&(rnh)->rnh_mtx, "radix node head", NULL, MTX_DEF | MTX_RECURSE) + mtx_init(&(rnh)->rnh_mtx, "radix node head", NULL, MTX_DEF | MTX_RECURSE | +MTX_DUPOK) #define RADIX_NODE_HEAD_LOCK(rnh) mtx_lock(&(rnh)->rnh_mtx) #define RADIX_NODE_HEAD_UNLOCK(rnh) mtx_unlock(&(rnh)->rnh_mtx) #define RADIX_NODE_HEAD_DESTROY(rnh) mtx_destroy(&(rnh)->rnh_mtx) Am I right? "radix node head" is locked first time in arptimer() and 2nd in rtrequest1(). And (if I understand code well) those locks should be in both functions, because rtrequest1() is not only called through arptimer(), but also through other functions that don't lock it eariler. -- Pawel Jakub Dawidek UNIX Systems Administrator http://garage.freebsd.pl Am I Evil? Yes, I Am.
msg51703/pgp00000.pgp
Description: PGP signature
