Some fixes for msdosfs_rename and ufs_rename

[kib]

Summary

msdosfs_rename: Use the same locking algorithm for msdosfs_rename() as used by ufs_rename(). Convert doscheckpath() to non-sleeping version.

PR: 257522

msdosfs_rename: add a lock to ensure stability of the path check (ufs_checkpath and doscheckpath).

vfs_hash_insert: ensure that predicate is true

After vnode lock, recheck v_hash. When vfs_hash_insert() is used with a predicate, recheck it after the selected vnode is locked. Since vfs_hash_lock is dropped, vnode could be rehashed during the sleep for the vnode lock, which could go unnoticed there.

[The addition of the excl lock assert into vfs_hash_rehash()] is not going to be committed right now]

Tested by: pho

Per-commit view is available at https://kib.kiev.ua/git/gitweb.cgi?p=deviant3.git;a=shortlog;h=refs/heads/msdosfs_rename

[kib]

ufs fix was committed
Upload rebased patch, with the accumulated bug fixes.

[kib]

Upload patch which passed pho run.

[mckusick]

This looks like a correct adaptation of the UFS code to the MSDOS filesystem implementation. Given that I have been struggling with getting UFS rename correct since the mid-1980s I am sure there are still bugs lurking in this implementation, but at least they are closer to being the same bugs in both filesystem implementations.

[kib]

Fix problems with corrupted lookup cached data in directory vnodes. Basically, add non-sleeping calls to msdosfs_lookup_ino() into several places in msdosfs_rename() to refresh de offsets.