linux-kernel - Re: [PATCH 3/3] 9p: Add mempools for RPCs

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 4 Jul 2022 12:38:46 +0900
From:   Dominique Martinet <asmadeus@...ewreck.org>
To:     Kent Overstreet <kent.overstreet@...il.com>,
        Christian Schoenebeck <linux_oss@...debyte.com>
Cc:     linux-kernel@...r.kernel.org, v9fs-developer@...ts.sourceforge.net,
        Eric Van Hensbergen <ericvh@...il.com>,
        Latchesar Ionkov <lucho@...kov.net>
Subject: Re: [PATCH 3/3] 9p: Add mempools for RPCs

+Christian, sorry I just noticed you weren't in Ccs again --
the patches are currently there if you want a look:
https://evilpiepirate.org/git/bcachefs.git/log/?h=9p_mempool 

I think it'll conflict a bit with your 8k non-read/write RPCs but I'll
take care of that when checking it this weekend.


Kent Overstreet wrote on Sun, Jul 03, 2022 at 11:05:57PM -0400:
> > We shouldn't have any user calling with more at this point (the
> > user-provided size comes from p9_client_prepare_req arguments and it's
> > either msize or header size constants); and it probably makes sense to
> > check and error out rather than cap it.
> 
> If that's the case I think we should just switch the warning to a BUG_ON() - I
> just wasn't sure from reading the code if that was really guarded against.

yes, BUG_ON is good for me.

> > > -	if (p9_fcall_init(c, &req->tc, alloc_msize))
> > > +	if (p9_fcall_init(c, &req->tc, 0, alloc_msize))
> > >  		goto free_req;
> > > -	if (p9_fcall_init(c, &req->rc, alloc_msize))
> > > +	if (p9_fcall_init(c, &req->rc, 1, alloc_msize))
> > 
> > given the two rc/tc buffers are of same size I don't see the point of
> > using two caches either, you could just double the min number of
> > elements to the same effect?
> 
> You can't double allocate from the same mempool, that will deadlock if multiple
> threads need the last element at the same time - I should've left a comment for
> that.

hmm, looking at the code as long as min elements is big enough the
deadlock becomes increasingly difficult to hit -- but I guess there's no
guarantee we won't get 8 threads each getting their first item from the
pool and starving each other on the second... Fair enough, thank you for
the comment.

> @@ -270,10 +276,8 @@ p9_tag_alloc(struct p9_client *c, int8_t type, unsigned int max_size)
>  	if (!req)
>  		return ERR_PTR(-ENOMEM);
>  
> -	if (p9_fcall_init(c, &req->tc, alloc_msize))
> -		goto free_req;
> -	if (p9_fcall_init(c, &req->rc, alloc_msize))
> -		goto free;
> +	p9_fcall_init(c, &req->tc, 0, alloc_msize);
> +	p9_fcall_init(c, &req->rc, 1, alloc_msize);


mempool allocation never fails, correct?

(don't think this needs a comment, just making sure here)


This all looks good to me, will queue it up in my -next branch after
running some tests next weekend and hopefully submit when 5.20 opens
with the code making smaller allocs more common.

--
Dominique