Use M_WRITABLE() rather than direct M_EXT checks

[rwatson]

Summary

Use M_WRITABLE() rather than checking explicitly (and only) for M_EXT in
cases where this was likely the intention (e.g., when deciding whether to
m_pullup() even if there is notionally sufficient data in the mbuf for a
read-write data structure). This reduces knowledge of extern-storage
mechanics outside of the mbuf allocator.

[rwatson]

Closed by commit rS272559 (authored by @rwatson).

[rwatson]

Sigh. Wrong URL in my commit message laid to premature closing of this patch.

[bz]

This is the wrong diff, right?

[bz]

Argh, I see you committed with the wring D???? number and Phrabicator is too dumb. Can we make diff1 the default again?

[rwatson]

Unfortunately, my ability to wield phabricator is weak. Maybe I should just create a new review request. I'll try resubmitting the original patch.

[rwatson]

Re-upload original diff following error in commit message with tricky consequences.

[bz]

So let's see if I do understand the difference between m_pullup() and m_copyup() correctly.

sys/netinet/igmp.c

1469 ↗

(On Diff #1895)

This logic doesn't entirely make sense to me. If the mbuf is not writeable how would you do a m_pullup() on it? You would need an m_copyup(), right? It wouldn't matter much in this code path to my understanding as we assume the mbuf to be gone (and no other copy being around) by the time we leave the function to my understanding.

The same applies to all other netinet/ files, ip6_mroute.c, and 2nd case in ip6_output.c.

Side question: is minlen actually smaller than MHLEN? I know this is unrelated ... so just assume that for the moment ;-)

sys/netinet6/icmp6.c

587 ↗

(On Diff #1895)

Here it seems good.

sys/netinet6/ip6_output.c

1187 ↗

(On Diff #1895)

M_TRAINLINGSPACE already does the M_WRITEABLE check for you. You can nuke it.

2910 ↗

(On Diff #1895)

And that's another one of the m_pulls on possibly not writeable mbufs

[rwatson]

Closed by commit rS272984 (authored by @rwatson).