Was it ? Also te UCB copyright assignment looks doubtful.

So globally this looks like a copy/paste of fdescfs(5). Why it is needed ? Please explain where the behavior of the new fs differs from native fdescfs. Then I will ask another question.

Why this protection is needed ?

Also, there is no double-include protection.

Blank line after { is needed.

return (kernel_mount());

Why this cast is needed ?

Remove init from declaration, collapse two int vars declarations into single.

What is the purpose of taking the lock there ?

Why this dependency is needed ?

Can you explain this ?

Replaced by my copyright, is this good?

Probably no, you should added your copyright to the existing statement, but I think that this is moot.

fdesc_allocvp() is much more suitable place for the flag assignment.

You need to filter out Froot fntype when setting the flag, otherwise panic() in readlink would be called.

Why don't you use vnode v_vflag there ? It would avoid two pointers indirections. I.e. I would write this fragment as

vap->va_type = (vp->v_vflag & VV_LINRDLNK) != 0 ? VLNK : VCHR;

Same there.

Do not put initialization into declaration block.

fdesc_readlink

At this moment, the vn vnode might be reclaimed after which vn->v_data points to freed memory.

Use strlen().

I suggest the VV_READLINK name.

VV_READLINK only for Fdesc node

I would prefer this assignment written with the ?:.

This cannot work as well. The process filedesc table is locked inside fget_cap(), but the thread already owns the vnode lock. So you get LOR there. Did you run this under WITNESS ?

Arguably, you need to read fd->fd_fd into a local variable and _then_ drop the vnode lock.

This vref() is useless, while fp is refed, vnode cannot be freed since fp also vrefed the vnode.

Again, vn_fullpath() cannot be called with a vnode locked, since it might need to lock some vnodes during the patch reconstruction, and this would break the vnode locks order.

I still do not understand why don't you move this check and flag set to fdesc_allocvp().

Mostly due to two ways of success out from fdesc_allocvp(), one of them I do not understand how to test

I strongly prefer to have the vnode fully initialized before pushing it into the place where it can be found by other threads, even while it is locked.

There is clearly preferrable place where fs-private data for the vnode is constructed.

No, please re-read what I wrote.

Move the flag setting to line 200 (patched), where all other initialization of the fs-specific data is performed.