4.2BSD netimp/if_imphost.c

[walsh%bbn-unix at sri-unix.UUCP]

From:  Bob Walsh <walsh at bbn-unix>


After thinking about it a little bit, I realize my first letter contained
code with hastily written comments.  The code corrections are correct, but
the code comments contained an incorrect analysis.  I believe this diff
contains code comments with a correct analysis of the source of the panic.

121,123d120
< #ifdef BUGFIXES
< 	register struct mbuf *nextm;
< #endif
126,132d122
< #ifdef BUGFIXES
< 	/*
< 	 * see BUGFIXES below
< 	 */
< 	for (m = hosts; m; m = nextm) {
< 		nextm = m->m_next;
< #else
134d123
< #endif
209,211d197
< #ifdef BUGFIXES
< 	register struct mbuf *nextm;
< #endif
215,236d200
< #ifdef BUGFIXES
< 	/*
< 	 * hostrelease may put the current m on the free list, so avoid
< 	 * traversing the free list instead of the host list.  Not only is
< 	 * this wrong, but it can cause a panic:  
< 	 *
< 	 *	On the mbuf free list, the m_off field is zero (due to MFREE)
< 	 *	and hm points to the m_next field of the mbuf.
< 	 *	hm_count is m_next.
< 	 *
< 	 *	m_off is h_q. m_len and m_type (0) are h_addr.
< 	 *	The qcnt, timer, rfnm, and flags fields of hp are in m_dat.
< 	 *
< 	 *	If hostrelease() is called on an item in the free list, then
< 	 *	hm_count (m_next) is decremented, trashing the free list which
< 	 *	we're following.
< 	 *
< 	 * bw 2/26/84
< 	 */
< 	for (m = hosts; m; m = nextm) {
< 		nextm = m->m_next;
< #else
238d201
< #endif

I'll try to be less sloppy,
bob walsh