diff options
| -rw-r--r-- | threadproc/beos/.cvsignore | 1 | ||||
| -rw-r--r-- | threadproc/beos/proc.c | 32 |
2 files changed, 33 insertions, 0 deletions
diff --git a/threadproc/beos/.cvsignore b/threadproc/beos/.cvsignore index 11f54a843..f82426797 100644 --- a/threadproc/beos/.cvsignore +++ b/threadproc/beos/.cvsignore @@ -1,5 +1,6 @@ Makefile apr_proc_stub *.lo +*.slo .libs .deps diff --git a/threadproc/beos/proc.c b/threadproc/beos/proc.c index ecbb302ca..71dd5a88f 100644 --- a/threadproc/beos/proc.c +++ b/threadproc/beos/proc.c @@ -149,6 +149,38 @@ APR_DECLARE(apr_status_t) apr_proc_fork(apr_proc_t *proc, apr_pool_t *pool) return errno; } else if (pid == 0) { + /* This is really ugly... + * The semantics of BeOS's fork() are that areas (used for shared + * memory) get COW'd :-( The only way we can make shared memory + * work across fork() is therefore to find any areas that have + * been created and then clone them into our address space. + * Thankfully only COW'd areas have the lock variable set at + * anything but 0, so we can use that to find the areas we need to + * copy. Of course what makes it even worse is that the loop through + * the area's will go into an infinite loop, eating memory and then + * eventually segfault unless we know when we reach then end of the + * "original" areas and stop. Why? Well, we delete the area and then + * add another to the end of the list... + */ + area_info ai; + int32 cookie = 0; + area_id highest = 0; + + while (get_next_area_info(0, &cookie, &ai) == B_OK) + if (ai.area > highest) + highest = ai.area; + cookie = 0; + while (get_next_area_info(0, &cookie, &ai) == B_OK) { + if (ai.area > highest) + break; + if (ai.lock > 0) { + area_id original = find_area(ai.name); + delete_area(ai.area); + clone_area(ai.name, &ai.address, B_CLONE_ADDRESS, + ai.protection, original); + } + } + proc->pid = pid; proc->in = NULL; proc->out = NULL; |