Subject: Re: [PATCH] tmpfs: shmem_fallocate must return ERESTARTSYS
To: Maxim Patlasov <mpatlasov@virtuozzo.com>, hughd@google.com
References: <20160304002954.19844.52266.stgit@maxim-thinkpad>
Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org, devel@openvz.org
From: Mike Kravetz <mike.kravetz@oracle.com>
Message-ID: <56D8DA86.6020803@oracle.com>
Date: Thu, 3 Mar 2016 16:44:54 -0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101
 Thunderbird/38.5.0
MIME-Version: 1.0
In-Reply-To: <20160304002954.19844.52266.stgit@maxim-thinkpad>
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1657
Lines: 49

On 03/03/2016 04:30 PM, Maxim Patlasov wrote:
> shmem_fallocate() is restartable, so it can return ERESTARTSYS if
> signal_pending(). Although fallocate(2) manpage permits EINTR,
> the more places use ERESTARTSYS the better.
> 
> Signed-off-by: Maxim Patlasov <mpatlasov@virtuozzo.com>
> ---
>  mm/shmem.c |    8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 440e2a7..60e9c8a 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -2229,11 +2229,13 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
>  		struct page *page;
>  
>  		/*
> -		 * Good, the fallocate(2) manpage permits EINTR: we may have
> -		 * been interrupted because we are using up too much memory.
> +		 * Although fallocate(2) manpage permits EINTR, the more
> +		 * places use ERESTARTSYS the better. If we have been
> +		 * interrupted because we are using up too much memory,
> +		 * oom-killer used fatal signal and we will die anyway.
>  		 */
>  		if (signal_pending(current))
> -			error = -EINTR;
> +			error = -ERESTARTSYS;
>  		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
>  			error = -ENOMEM;
>  		else

I used the shmem fallocate code as a basis for hugetlbfs fallocate.
See, hugetlbfs_fallocate().  Specifically:

		/*
		 * fallocate(2) manpage permits EINTR; we may have been
		 * interrupted because we are using up too much memory.
		 */
		if (signal_pending(current)) {
			error = -EINTR;
			break;
		}

I don't know much about the advantages of changing to ERESTARTSYS.  But,
if it is changed for shmem it should be changed for hugetlbfs as well.

-- 
Mike Kravetz