Re: [PATCH -v6 2/2] Updating ctime and mtime for memory-mapped files

[Miklos Szeredi <miklos@szeredi.hu>]

> Possibly, I didn't see a quick way to break that iteration.
> >From a quick glance at prio_tree.c the iterator isn't valid anymore
> after releasing i_mmap_lock. Fixing that would be,.. 'fun'.

Maybe i_mmap_lock isn't needed at all, since msync holds mmap_sem,
which protects the prio tree as well, no?

> I also realized I forgot to copy/paste the prio_tree_iter declaration
> and ought to make all these functions static.
> 
> But for a quick draft it conveys the idea pretty well, I guess :-)

Yes :)

There could also be nasty performance corner cases, like having a huge
file mapped thousands of times, and having only a couple of pages
dirtied between MS_ASYNC invocations.  Then most of that page table
walking would be just unnecessary overhead.

There's something to be said for walking only the dirty pages, and
doing page_mkclean on them, even if in some cases that would be
slower.

But I have a strong feeling of deja vu, and last time it ended with
Andrew not liking the whole thing...

Miklos