Re: [PATCH] sysv: convert pointers_lock from rw_lock to rw_sem

[Al Viro <viro@zeniv.linux.org.uk>]

On Mon, Mar 27, 2023 at 07:19:47PM +0900, Tetsuo Handa wrote:

> I feel worried about
> 
> 	/*
> 	 * Indirect block might be removed by truncate while we were
> 	 * reading it. Handling of that case (forget what we've got and
> 	 * reread) is taken out of the main path.
> 	 */
> 	if (err == -EAGAIN)
> 		goto changed;
> 
> in get_block()...

Look at the caller of find_shared(); there won't be other truncate messing
up with the indirect blocks.  sysv_truncate() is called by sysv_write_failed()
(from ->write_begin()), sysv_setattr() (->setattr(), with ATTR_SIZE) and
sysv_evict_inode() (if there's no links left to on-disk inode; it's an
->evict_inode() instance, so we are dropping the last reference to in-core one).

The first two have i_mutex held by callers, serializing them against each
other, and both have the in-core inode pinned, which gives exclusion with
the third one...

IOW, sysv_truncate() (as well as its minixfs counterpart) relies upon having
serialization wrt other callers of sysv_truncate().