linux-kernel - Re: [openib-general] [PATCH 04/13] Connection Manager

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]

Message-Id: <1163788000.8457.88.camel@stevo-desktop>
Date:	Fri, 17 Nov 2006 12:26:40 -0600
From:	Steve Wise <swise@...ngridcomputing.com>
To:	Bryan O'Sullivan <bos@...hscale.com>
Cc:	rdreier@...co.com, netdev@...r.kernel.org,
	linux-kernel@...r.kernel.org, openib-general@...nib.org
Subject: Re: [openib-general] [PATCH  04/13] Connection Manager

On Fri, 2006-11-17 at 10:07 -0800, Bryan O'Sullivan wrote:
> Steve Wise wrote:
> 
> > +static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
> > +{
> > +	struct cpl_tid_release *req;
> > +
> > +	skb = get_skb(skb, sizeof *req, GFP_KERNEL);
> > +	if (!skb) {
> > +		return;
> > +	}
> 
> Style micronit: no curlies for single-statement blocks.
> 

yup.

> > +void __free_ep(struct iwch_ep_common *epc)
> > +{
> > +	PDBG("%s ep %p, &refcnt %p state %s, refcnt %d\n",
> > +	     __FUNCTION__, epc, &epc->refcnt,
> > +	     states[state_read(epc)], atomic_read(&epc->refcnt));
> > +
> > +	if (atomic_read(&epc->refcnt) == 1) {
> > +		goto out;
> > +	}
> > +	if (!atomic_dec_and_test(&epc->refcnt)) {
> > +		return;
> > +	}
> > +out:
> > +	PDBG("free ep %p\n", epc);
> > +	kfree(epc);
> > +}
> 
> Whatever you're trying to do with refcounting and atomics here looks 
> extremely dodgy and race-prone to me.  Why are you using atomic ops in 
> such a scary manner, instead of just slapping a spinlock around this?
> 

This logic is the same as kfree_skb() (and kref_put() :).  It optimizes
the case where you're the last one freeing I guess and avoids the
dec-and-test in that case..

> Anyway, please drop this atomic refcounting stuff and use embedded krefs 
> instead.  You're tunnelling into a bug mine.

The kref put code looks very much like the code above.  But I can use
krefs to avoid replicating code.

> 
> By the way, it would be more consistent with normal kernel naming 
> conventions to name these refcount-diddling routines ep_get and ep_put, 
> since __ep_free doesn't actually free an object unless it feels like it.

Again, it was modeled after skb.

> 
> > +int __init iwch_cm_init(void)
> > +{
> > +	skb_queue_head_init(&rxq);
> > +
> > +	workq = create_singlethread_workqueue("iw_cxgb3");
> > +	if (!workq)
> > +		return -ENOMEM;
> > +
> > +	/*
> > +	 * All upcalls from the T3 Core go to sched() to 
> > +	 * schedule the processing on a work queue.
> > +	 */
> > +	t3c_handlers[CPL_ACT_ESTABLISH] = sched;
> > +	t3c_handlers[CPL_ACT_OPEN_RPL] = sched;
> > +	t3c_handlers[CPL_RX_DATA] = sched;
> > +	t3c_handlers[CPL_TX_DMA_ACK] = sched;
> > +	t3c_handlers[CPL_ABORT_RPL_RSS] = sched;
> > +	t3c_handlers[CPL_ABORT_RPL] = sched;
> > +	t3c_handlers[CPL_PASS_OPEN_RPL] = sched;
> > +	t3c_handlers[CPL_CLOSE_LISTSRV_RPL] = sched;
> > +	t3c_handlers[CPL_PASS_ACCEPT_REQ] = sched;
> > +	t3c_handlers[CPL_PASS_ESTABLISH] = sched;
> > +	t3c_handlers[CPL_PEER_CLOSE] = sched;
> > +	t3c_handlers[CPL_CLOSE_CON_RPL] = sched;
> > +	t3c_handlers[CPL_ABORT_REQ_RSS] = sched;
> > +	t3c_handlers[CPL_RDMA_TERMINATE] = sched;
> > +	t3c_handlers[CPL_RDMA_EC_STATUS] = sched;
> > +
> > +	/*
> > +	 * These are the real handlers that are called from a 
> > +	 * work queue.
> > +	 */
> > +	work_handlers[CPL_ACT_ESTABLISH] = act_establish;
> > +	work_handlers[CPL_ACT_OPEN_RPL] = act_open_rpl;
> > +	work_handlers[CPL_RX_DATA] = rx_data;
> > +	work_handlers[CPL_TX_DMA_ACK] = tx_ack;
> > +	work_handlers[CPL_ABORT_RPL_RSS] = abort_rpl;
> > +	work_handlers[CPL_ABORT_RPL] = abort_rpl;
> > +	work_handlers[CPL_PASS_OPEN_RPL] = pass_open_rpl;
> > +	work_handlers[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl;
> > +	work_handlers[CPL_PASS_ACCEPT_REQ] = pass_accept_req;
> > +	work_handlers[CPL_PASS_ESTABLISH] = pass_establish;
> > +	work_handlers[CPL_PEER_CLOSE] = peer_close;
> > +	work_handlers[CPL_ABORT_REQ_RSS] = peer_abort;
> > +	work_handlers[CPL_CLOSE_CON_RPL] = close_con_rpl;
> > +	work_handlers[CPL_RDMA_TERMINATE] = terminate;
> > +	work_handlers[CPL_RDMA_EC_STATUS] = ec_status;
> > +	return 0;
> > +}
> 
> This seems mighty peculiar.  Why aren't you keeping this stuff in 
> structs, instead of faking up structs via arrays?
> 

Its a function handler table.  Given an incoming message with a command
number, we can index into the table using the command number to get the
handler function for that message.



-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/