This doesn't avoid the race.
You just added a huge race in the in-memory scoreboards. I'll be
re-instating the second chunk of this patch.
Dean
On 10 Dec 1998 [EMAIL PROTECTED] wrote:
> jim 98/12/10 07:14:10
>
> Modified: src/main http_main.c
> Log:
> Add 'reset slot vhost' logic to the
> ap_update_child_status function and avoid the race as well.
>
> Revision Changes Path
> 1.409 +11 -9 apache-1.3/src/main/http_main.c
>
> Index: http_main.c
> ===================================================================
> RCS file: /export/home/cvs/apache-1.3/src/main/http_main.c,v
> retrieving revision 1.408
> retrieving revision 1.409
> diff -u -r1.408 -r1.409
> --- http_main.c 1998/12/04 19:12:16 1.408
> +++ http_main.c 1998/12/10 15:14:08 1.409
> @@ -2069,6 +2069,17 @@
> sizeof(ss->request));
> }
> ss->vhostrec = r->server;
> + } else if (status == SERVER_STARTING) {
> + /* clean up the slot's vhostrec pointer (maybe re-used)
> + * and mark the slot as belonging to a new generation.
> + */
> + ss->vhostrec = NULL;
> + ap_scoreboard_image->parent[child_num].generation =
> ap_my_generation;
> +#ifdef SCOREBOARD_FILE
> + lseek(scoreboard_fd, XtOffsetOf(scoreboard, parent[child_num]), 0);
> + force_write(scoreboard_fd, &ap_scoreboard_image->parent[child_num],
> + sizeof(parent_score));
> +#endif
> }
> }
> put_scoreboard_info(child_num, ss);
> @@ -3917,15 +3928,6 @@
> Explain1("Starting new child in slot %d", slot);
> (void) ap_update_child_status(slot, SERVER_STARTING, (request_rec *)
> NULL);
>
> - /* clean up the slot's vhostrec pointer now that it is being re-used,
> - * and mark the slot as beloging to a new generation.
> - */
> - /* XXX: there's still a race condition here for file-based
> scoreboards...
> - * but... like, do we really care to spend yet another write()
> operation
> - * here? -djg
> - */
> - ap_scoreboard_image->servers[slot].vhostrec = NULL;
> - ap_scoreboard_image->parent[slot].generation = ap_my_generation;
>
> #ifndef _OSD_POSIX
> if ((pid = fork()) == -1) {
>
>
>
>
