Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Fri, 1 Jan 2021 05:00:18 -0500 (EST)
From:   Mikulas Patocka <mpatocka@redhat.com>
To:     Ignat Korchagin <ignat@cloudflare.com>
cc:     agk@redhat.com, snitzer@redhat.com, dm-devel@redhat.com,
        dm-crypt@saout.de, linux-kernel@vger.kernel.org,
        ebiggers@kernel.org, Damien.LeMoal@wdc.com,
        herbert@gondor.apana.org.au, kernel-team@cloudflare.com,
        nobuto.murata@canonical.com, clm@fb.com, josef@toxicpanda.com,
        dsterba@suse.com, linux-btrfs@vger.kernel.org,
        mail@maciej.szmigiero.name, stable@vger.kernel.org
Subject: Re: [PATCH v2 1/2] dm crypt: use GFP_ATOMIC when allocating crypto
 requests from softirq
In-Reply-To: <20201230214520.154793-2-ignat@cloudflare.com>
Message-ID: <alpine.LRH.2.02.2101010450460.31009@file01.intranet.prod.int.rdu2.redhat.com>
References: <20201230214520.154793-1-ignat@cloudflare.com> <20201230214520.154793-2-ignat@cloudflare.com>
User-Agent: Alpine 2.02 (LRH 1266 2009-07-14)
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Precedence: bulk


On Wed, 30 Dec 2020, Ignat Korchagin wrote:

> diff --git a/drivers/md/dm-crypt.c b/drivers/md/dm-crypt.c
> index 53791138d78b..e4fd690c70e1 100644
> --- a/drivers/md/dm-crypt.c
> +++ b/drivers/md/dm-crypt.c
> @@ -1539,7 +1549,10 @@ static blk_status_t crypt_convert(struct crypt_config *cc,
>  
>  	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
>  
> -		crypt_alloc_req(cc, ctx);
> +		r = crypt_alloc_req(cc, ctx);
> +		if (r)
> +			return BLK_STS_RESOURCE;
> +
>  		atomic_inc(&ctx->cc_pending);
>  
>  		if (crypt_integrity_aead(cc))
> -- 
> 2.20.1

I'm not quite convinced that returning BLK_STS_RESOURCE will help. The 
block layer will convert this value back to -ENOMEM and return it to the 
caller, resulting in an I/O error.

Note that GFP_ATOMIC allocations may fail anytime and you must handle 
allocation failure gracefully - i.e. process the request without any 
error.

An acceptable solution would be to punt the request to a workqueue and do 
GFP_NOIO allocation from the workqueue. Or add the request to some list 
and process the list when some other request completes.

You should write a test that simulates allocation failure and verify that 
the kernel handles it gracefully without any I/O error.

Mikulas