linux-kernel - Re: [PATCH v2 09/12] md/raid5: Keep a reference to last stripe

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: <56208e2a-5035-eb8e-c468-70b4dae66d5c@linux.dev>
Date:   Wed, 27 Apr 2022 09:36:59 +0800
From:   Guoqing Jiang <guoqing.jiang@...ux.dev>
To:     Logan Gunthorpe <logang@...tatee.com>,
        linux-kernel@...r.kernel.org, linux-raid@...r.kernel.org,
        Song Liu <song@...nel.org>
Cc:     Christoph Hellwig <hch@...radead.org>,
        Stephen Bates <sbates@...thlin.com>,
        Martin Oliveira <Martin.Oliveira@...eticom.com>,
        David Sloan <David.Sloan@...eticom.com>
Subject: Re: [PATCH v2 09/12] md/raid5: Keep a reference to last stripe_head
 for batch



On 4/21/22 3:54 AM, Logan Gunthorpe wrote:
> When batching, every stripe head has to find the previous stripe head to
> add to the batch list. This involves taking the hash lock which is
> highly contended during IO.
>
> Instead of finding the previous stripe_head each time, store a
> reference to the previous stripe_head in a pointer so that it doesn't
> require taking the contended lock another time.
>
> The reference to the previous stripe must be released before scheduling
> and waiting for work to get done. Otherwise, it can hold up
> raid5_activate_delayed() and deadlock.
>
> Signed-off-by: Logan Gunthorpe<logang@...tatee.com>
> ---
>   drivers/md/raid5.c | 51 +++++++++++++++++++++++++++++++++++-----------
>   1 file changed, 39 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
> index 0c250cc3bfff..28ea7b9b6ab6 100644
> --- a/drivers/md/raid5.c
> +++ b/drivers/md/raid5.c
> @@ -843,7 +843,8 @@ static bool stripe_can_batch(struct stripe_head *sh)
>   }
>   
>   /* we only do back search */
> -static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh)
> +static void stripe_add_to_batch_list(struct r5conf *conf,
> +		struct stripe_head *sh, struct stripe_head *last_sh)

Nit, from stripe_add_to_batch_list's view, I think "head_sh" makes more 
sense than
"last_sh".

>   {
>   	struct stripe_head *head;
>   	sector_t head_sector, tmp_sec;
> @@ -856,15 +857,20 @@ static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh
>   		return;
>   	head_sector = sh->sector - RAID5_STRIPE_SECTORS(conf);
>   
> -	hash = stripe_hash_locks_hash(conf, head_sector);
> -	spin_lock_irq(conf->hash_locks + hash);
> -	head = find_get_stripe(conf, head_sector, conf->generation, hash);
> -	spin_unlock_irq(conf->hash_locks + hash);
> -
> -	if (!head)
> -		return;
> -	if (!stripe_can_batch(head))
> -		goto out;
> +	if (last_sh && head_sector == last_sh->sector) {
> +		head = last_sh;
> +		atomic_inc(&head->count);
> +	} else {
> +		hash = stripe_hash_locks_hash(conf, head_sector);
> +		spin_lock_irq(conf->hash_locks + hash);
> +		head = find_get_stripe(conf, head_sector, conf->generation,
> +				       hash);
> +		spin_unlock_irq(conf->hash_locks + hash);
> +		if (!head)
> +			return;
> +		if (!stripe_can_batch(head))
> +			goto out;
> +	}
>   
>   	lock_two_stripes(head, sh);
>   	/* clear_batch_ready clear the flag */
> @@ -5800,6 +5806,7 @@ enum stripe_result {
>   
>   struct stripe_request_ctx {
>   	bool do_flush;
> +	struct stripe_head *batch_last;
>   };
>   
>   static enum stripe_result make_stripe_request(struct mddev *mddev,
> @@ -5889,8 +5896,13 @@ static enum stripe_result make_stripe_request(struct mddev *mddev,
>   		return STRIPE_SCHEDULE_AND_RETRY;
>   	}
>   
> -	if (stripe_can_batch(sh))
> -		stripe_add_to_batch_list(conf, sh);
> +	if (stripe_can_batch(sh)) {
> +		stripe_add_to_batch_list(conf, sh, ctx->batch_last);
> +		if (ctx->batch_last)
> +			raid5_release_stripe(ctx->batch_last);
> +		atomic_inc(&sh->count);
> +		ctx->batch_last = sh;
> +	}
>   
>   	if (ctx->do_flush) {
>   		set_bit(STRIPE_R5C_PREFLUSH, &sh->state);
> @@ -5979,6 +5991,18 @@ static bool raid5_make_request(struct mddev *mddev, struct bio * bi)
>   		} else if (res == STRIPE_RETRY) {
>   			continue;
>   		} else if (res == STRIPE_SCHEDULE_AND_RETRY) {
> +			/*
> +			 * Must release the reference to batch_last before
> +			 * scheduling and waiting for work to be done,
> +			 * otherwise the batch_last stripe head could prevent
> +			 * raid5_activate_delayed() from making progress
> +			 * and thus deadlocking.
> +			 */
> +			if (ctx.batch_last) {
> +				raid5_release_stripe(ctx.batch_last);
> +				ctx.batch_last = NULL;
> +			}
> +
>   			schedule();
>   			prepare_to_wait(&conf->wait_for_overlap, &w,
>   					TASK_UNINTERRUPTIBLE);
> @@ -5990,6 +6014,9 @@ static bool raid5_make_request(struct mddev *mddev, struct bio * bi)
>   
>   	finish_wait(&conf->wait_for_overlap, &w);
>   
> +	if (ctx.batch_last)
> +		raid5_release_stripe(ctx.batch_last);
> +
>   	if (rw == WRITE)
>   		md_write_end(mddev);
>   	bio_endio(bi);

Otherwise looks good, Acked-by: Guoqing Jiang <guoqing.jiang@...ux.dev>

Thanks,
Guoqing