Re: [PATCH 1/1] infiniband: ulp/iser, fix error retval in iser_create_ib_conn_res

[Roland Dreier <rdreier@cisco.com>]

So looking at merging this finally, I think I see one problem with the
proposed patch.  We have:

 > @@ -183,7 +180,7 @@  static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
 >  	ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, &params);
 >  	if (IS_ERR(ib_conn->fmr_pool)) {
 >  		ret = PTR_ERR(ib_conn->fmr_pool);
 > -		goto fmr_pool_err;
 > +		goto out_err;
 >  	}

and

 > @@ -209,12 +206,7 @@  static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
 >  		 ib_conn->fmr_pool, ib_conn->cma_id->qp);
 >  	return ret;
 >  
 > -qp_err:
 > -	(void)ib_destroy_fmr_pool(ib_conn->fmr_pool);
 > -fmr_pool_err:
 > -	kfree(ib_conn->page_vec);
 > -	kfree(ib_conn->login_buf);
 > -alloc_err:
 > +out_err:
 >  	iser_err("unable to alloc mem or create resource, err %d\n", ret);
 >  	return ret;
 >  }

so if ib_create_fmr_pool() fails, we're left with ib_conn->fmr_pool
holding an error pointer, right?  But we're relying on
iser_free_ib_conn_res() to clean up after us, and that has:

	if (ib_conn->fmr_pool != NULL)
		ib_destroy_fmr_pool(ib_conn->fmr_pool);

so we're going to end up trying to free an error pointer, which will
probably crash.

I think.

Dan or Or, am I wrong here or do we need another iteration of this
patch? (the login_buf and page_vec changes do look correct to me, since
a failed kmalloc() will leave us with a NULL pointer that it is safe to
kfree() later)

 - R.
-- 
Roland Dreier <rolandd@cisco.com> || For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html