linux-kernel - Re: [V9fs-developer] [PATCH 2/5] 9p: store req details and callback in struct p9_req

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 for Android: free password hash cracker in your pocket

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

Message-ID: <20161209071824.GC18158@nautica>
Date:   Fri, 9 Dec 2016 08:18:24 +0100
From:   Dominique Martinet <dominique.martinet@....fr>
To:     Stefano Stabellini <sstabellini@...nel.org>
Cc:     v9fs-developer@...ts.sourceforge.net, ericvh@...il.com,
        rminnich@...dia.gov, linux-kernel@...r.kernel.org, lucho@...kov.net
Subject: Re: [V9fs-developer] [PATCH 2/5] 9p: store req details and callback
 in struct p9_req_t


Nice. I like the idea of async I/Os :)

Stefano Stabellini wrote on Thu, Dec 08, 2016:
> Add a few fields to struct p9_req_t. Callback is the function which will
> be called upon requestion completion. offset, rsize, pagevec and kiocb
> store important information regarding the read or write request,
> essential to complete the request.
> 
> Currently not utilized, but they will be used in a later patch.
> 
> Signed-off-by: Stefano Stabellini <sstabellini@...nel.org>
> ---
>  include/net/9p/client.h | 8 ++++++++
>  net/9p/client.c         | 9 ++++++++-
>  2 files changed, 16 insertions(+), 1 deletion(-)
> 
> diff --git a/include/net/9p/client.h b/include/net/9p/client.h
> index aef19c6..69fc2f0 100644
> --- a/include/net/9p/client.h
> +++ b/include/net/9p/client.h
> @@ -110,6 +110,7 @@ enum p9_req_status_t {
>   *
>   */
>  
> +struct p9_client;
>  struct p9_req_t {
>  	int status;
>  	int t_err;
> @@ -118,6 +119,13 @@ struct p9_req_t {
>  	struct p9_fcall *rc;
>  	void *aux;
>  
> +    /* Used for async requests */
> +	void (*callback)(struct p9_client *c, struct p9_req_t *req, int status);
> +	size_t offset;
> +	u64 rsize;
> +	struct page **pagevec;
> +	struct kiocb *kiocb;
> +
>  	struct list_head req_list;
>  };
>  
> diff --git a/net/9p/client.c b/net/9p/client.c
> index b5ea9a3..bfe1715 100644
> --- a/net/9p/client.c
> +++ b/net/9p/client.c
> @@ -405,6 +405,10 @@ static void p9_free_req(struct p9_client *c, struct p9_req_t *r)
>  	int tag = r->tc->tag;
>  	p9_debug(P9_DEBUG_MUX, "clnt %p req %p tag: %d\n", c, r, tag);
>  
> +	r->offset = 0;
> +	r->rsize = 0;
> +	r->kiocb = NULL;
> +	r->callback = NULL;

Probably want to cleanup r->pagevec here too, even if that doesn't seem
to have any implication short-term (e.g. only looked at if callback is
not empty from what I've seen)

>  	r->status = REQ_STATUS_IDLE;
>  	if (tag != P9_NOTAG && p9_idpool_check(tag, c->tagpool))
>  		p9_idpool_put(tag, c->tagpool);
> @@ -427,7 +431,10 @@ void p9_client_cb(struct p9_client *c, struct p9_req_t *req, int status)
>  	smp_wmb();
>  	req->status = status;
>  
> -	wake_up(req->wq);
> +	if (req->callback != NULL)
> +		req->callback(c, req, status);
> +	else
> +		wake_up(req->wq);
>  	p9_debug(P9_DEBUG_MUX, "wakeup: %d\n", req->tc->tag);
>  }
>  EXPORT_SYMBOL(p9_client_cb);

Mostly a warning here, but p9_client_cb is called from an interrupt
context in 9P/RDMA.
This has been working up till now because we only do a wake_up and
there's no waiting, but (looking at later patches),
p9_client_read_complete for example does allocations and possibly other
unsafe operations from an interrupt context.

I don't know if the way forward is to move p9_client_cb from that
context or to have the callback be scheduled in a work queue instead;
but we'll need to fix that later.

-- 
Dominique Martinet