Rename the 'flags' argument to getfsstat() to 'flag' and validate it.

[jhb]

Summary

Rename the 'flags' argument to getfsstat() to 'flag' and validate it.

This argument is not a bitmask of flags, but only accepts a single value.
Fail with EINVAL if an invalid value is passed to 'flag'.

This is a followup to r308088. I will include the regen'd bits in a
separate followup. Will also bump .Dd on manpage when I actually
commit.

Test Plan

• compiled and booted, but we do not have an existing test for getfsstat

[kib]

As we discussed this with trasz, the callers must be inspected.

lib/libc/sys/getfsstat.2

44 ↗

(On Diff #23094)

flag->mode or wait_mode.

sys/kern/vfs_syscalls.c

455 ↗

(On Diff #23094)

What about MNT_LAZY ? Either it should be enabled there, or check for != MNT_LAZY removed below ?

[trasz]

As discussed, the callers have been inspected :-)

MNT_LAZY is an invalid flag there, and other BSDs don't allow anything other than MNT_WAIT, MNT_NOWAIT, or 0.

[jhb]

sys/kern/vfs_syscalls.c

455 ↗

(On Diff #23094)

I'm guessing it needs removing as it has never been documented. I looked in history and MNT_LAZY was added long ago in rS34266 which has a completely useless commit message.

Hmm, there is a followup e-mail saying that this was added for soft-updates:

https://docs.freebsd.org/cgi/getmsg.cgi?fetch=8550+0+archive/1998/cvs-sys/19980308.cvs-sys

Nothing in the rest of that commit (i.e. the userland bits) uses MNT_LAZY with getfsstat(). It seems to primarily be required for fsync(), so I wonder if it's addition to getfsstat() was an error to begin with?

[kib]

sys/kern/vfs_syscalls.c

455 ↗

(On Diff #23094)

I am fine with the removal, but removal must be done in the same commit which adds the flags check for MNT_WAIT/NOWAIT, IMO.

[mckusick]

I did some sleuthing through BSDI sources and the addition of the MNT_LAZY flag check was put into getfsstat by one of the BSDI engineers apparently in response to a customer request. From there it got carried over to FreeBSD when I made the diffs for Julian who did the import.

I do not think that it is wrong to treat MNT_LAZY the same as MNT_NOWAIT, but I agree that there is nothing to be gained from doing that other than possibly keeping some programs from breaking.

[jhb]

• Don't check for MNT_LAZY in getfsstat().
• Rename 'flag' to 'mode'.
• Regen.

[jhb]

• Rename flags argument to getmntinfo to mode to match getfsstat.

[jhb]

While doing a check via grep for all callers of getfsstat I found getmntinfo. All the in-tree callers pass either MNT_WAIT or MNT_NOWAIT.