Do not assign boolean value to integer mask_C1

• Set signal handler to SIG_DFL for signals in sigaction[] upon fork()
• Add comment explaining how syslogd handles child process reaping

Update after rebase

Remove cleanup in die()

Update after rebase. A comment for child process handling is added in https://reviews.freebsd.org/D41433

Move deadq_enter() into close_filed()

Remove some rebasing errors (consfile got declared twice)

Update after rebase and address Mark's comments

Update after rebase

Update after rebase

Fix cleanup and exit code for sigaction()

Update to avoid rebase conflict

Address Mark's suggestions

Apply this patch after we have procdescs so we don't accidentally kill() a recycled PID.

Move procdesc closure into deadq_remove() so we avoid dangling, invalid procdescs

Adjust to accommodate the reapchild() patch being moved down the series.

Remove comment in readconfigfile()

In D41377#943266, @markj wrote:

I'm ok with this, a general comment is just that we should seek to minimize syslogd.h, i.e., try to ensure it only contains stuff that really needs to be shared. For example, I'm pretty sure casper doesn't need RCVBUF_MINSIZE?

In D41379#943056, @jfree wrote:

In D41379#942527, @markj wrote:

Can you use our funlinkat() instead of keeping track of the directory descriptor?

Yes. not sure how I missed that.

In D41379#942527, @markj wrote:

Can you use our funlinkat() instead of keeping track of the directory descriptor?

Update after rebase

Update after rebase

Update after rebase

Update after rebase

Update after rebase

Address Mark's comments

Update after rebase

Update after rebase

Update after rebase

Update after rebase

Update after rebase

Update after rebase

Update after rebase

Address Mark's comments

Address Mark's suggestions

Update after rebase

Update after rebase

Update after rebase

Address Mark's suggestions

In D41361#942459, @markj wrote:

Did you test compiling with WITHOUT_INET_SUPPORT=or WITHOUT_INET6_SUPPORT= defined? I suspect it will fail, because the #include <netinet/in.h> at the beginning of syslogd is still conditional on INET or INET6 being defined. While the ifdefs are ugly, I'm not sure we can do much better without adding some dummy definitions to replace those in netinet/in.h.

• Remove style(9) changes
• Check for blocking fd

Fix spelling mistakes and update to-do list

In D41385#942540, @markj wrote:

I'm not a big fan of reordering functions like this, since that makes git blame quite useless for little gain. In general this kind of commit should come at the end of a patch stack if at all. If it's too painful to reorder it then ok, just something to keep in mind for next time.

Process descriptors for piped processes are registered with kevent() in this patch:

In D41362#942043, @emaste wrote:

The needdofsync global variable was initialized as true, when it should be false. Fix this.

We may want to just leave the 0/false globals without explicit initialization.

Update diff with context

Update diff with context.

Update diff with context

Update diff with context

Update diff with context.

In D41362#941850, @emaste wrote:

In future when uploading patches via the web interface please generate them with -U99999 so that Phabricator can show full context (see https://wiki.freebsd.org/Phabricator for details)