linux-ext4.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Ross Zwisler <ross.zwisler@linux.intel.com>
To: Jan Kara <jack@suse.cz>
Cc: linux-fsdevel@vger.kernel.org, linux-ext4@vger.kernel.org,
	linux-mm@kvack.org, Ross Zwisler <ross.zwisler@linux.intel.com>,
	Dan Williams <dan.j.williams@intel.com>,
	linux-nvdimm@lists.01.org, Matthew Wilcox <willy@linux.intel.com>
Subject: Re: [PATCH 08/18] ext4: Pre-zero allocated blocks for DAX IO
Date: Fri, 29 Apr 2016 12:01:58 -0600	[thread overview]
Message-ID: <20160429180158.GC5888@linux.intel.com> (raw)
In-Reply-To: <1461015341-20153-9-git-send-email-jack@suse.cz>

On Mon, Apr 18, 2016 at 11:35:31PM +0200, Jan Kara wrote:
> Currently ext4 treats DAX IO the same way as direct IO. I.e., it
> allocates unwritten extents before IO is done and converts unwritten
> extents afterwards. However this way DAX IO can race with page fault to
> the same area:
> 
> ext4_ext_direct_IO()				dax_fault()
>   dax_io()
>     get_block() - allocates unwritten extent
>     copy_from_iter_pmem()
> 						  get_block() - converts
> 						    unwritten block to
> 						    written and zeroes it
> 						    out
>   ext4_convert_unwritten_extents()
> 
> So data written with DAX IO gets lost. Similarly dax_new_buf() called
> from dax_io() can overwrite data that has been already written to the
> block via mmap.
> 
> Fix the problem by using pre-zeroed blocks for DAX IO the same way as we
> use them for DAX mmap. The downside of this solution is that every
> allocating write writes each block twice (once zeros, once data). Fixing
> the race with locking is possible as well however we would need to
> lock-out faults for the whole range written to by DAX IO. And that is
> not easy to do without locking-out faults for the whole file which seems
> too aggressive.
> 
> Signed-off-by: Jan Kara <jack@suse.cz>

Just a couple of simplifications - feel free to ignore them if you don't think
they are worth the effort.

> ---
>  fs/ext4/ext4.h  | 11 +++++++++--
>  fs/ext4/file.c  |  4 ++--
>  fs/ext4/inode.c | 42 +++++++++++++++++++++++++++++++++---------
>  3 files changed, 44 insertions(+), 13 deletions(-)
> 
> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> index 35792b430fb6..173da8faff81 100644
> --- a/fs/ext4/ext4.h
> +++ b/fs/ext4/ext4.h
> @@ -2521,8 +2521,8 @@ struct buffer_head *ext4_getblk(handle_t *, struct inode *, ext4_lblk_t, int);
>  struct buffer_head *ext4_bread(handle_t *, struct inode *, ext4_lblk_t, int);
>  int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
>  			     struct buffer_head *bh_result, int create);
> -int ext4_dax_mmap_get_block(struct inode *inode, sector_t iblock,
> -			    struct buffer_head *bh_result, int create);
> +int ext4_dax_get_block(struct inode *inode, sector_t iblock,
> +		       struct buffer_head *bh_result, int create);
>  int ext4_get_block(struct inode *inode, sector_t iblock,
>  		   struct buffer_head *bh_result, int create);
>  int ext4_dio_get_block(struct inode *inode, sector_t iblock,
> @@ -3328,6 +3328,13 @@ static inline void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
>  	}
>  }
>  
> +static inline bool ext4_aligned_io(struct inode *inode, loff_t off, loff_t len)
> +{
> +	int blksize = 1 << inode->i_blkbits;
> +
> +	return IS_ALIGNED(off, blksize) && IS_ALIGNED(off + len, blksize);

This could be just a tiny bit simpler by doing

	return IS_ALIGNED(off, blksize) && IS_ALIGNED(len, blksize);
						      ^^^

You've already made sure 'off' is aligned, so if 'len' is aligned 'off+len'
will be aligned.

> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index 23fd0e0a9223..6d5d5c1db293 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -3215,12 +3215,17 @@ static int ext4_releasepage(struct page *page, gfp_t wait)
>  }
>  
>  #ifdef CONFIG_FS_DAX
> -int ext4_dax_mmap_get_block(struct inode *inode, sector_t iblock,
> -			    struct buffer_head *bh_result, int create)
> +/*
> + * Get block function for DAX IO and mmap faults. It takes care of converting
> + * unwritten extents to written ones and initializes new / converted blocks
> + * to zeros.
> + */
> +int ext4_dax_get_block(struct inode *inode, sector_t iblock,
> +		       struct buffer_head *bh_result, int create)
>  {
>  	int ret;
>  
> -	ext4_debug("ext4_dax_mmap_get_block: inode %lu, create flag %d\n",
> +	ext4_debug("ext4_dax_get_block: inode %lu, create flag %d\n",
>  		   inode->i_ino, create);

This pattern could be improved by using "%s" and __func__ for the function
name.  That way you don't have to hunt through all your debug code and update
strings when you rename a function. More importantly it prevents the strings
from getting out of sync with the function name, resulting in confusing debug
messages.

>  	if (!create)
>  		return _ext4_get_block(inode, iblock, bh_result, 0);
> @@ -3233,9 +3238,9 @@ int ext4_dax_mmap_get_block(struct inode *inode, sector_t iblock,
>  
>  	if (buffer_unwritten(bh_result)) {
>  		/*
> -		 * We are protected by i_mmap_sem so we know block cannot go
> -		 * away from under us even though we dropped i_data_sem.
> -		 * Convert extent to written and write zeros there.
> +		 * We are protected by i_mmap_sem or i_mutex so we know block
> +		 * cannot go away from under us even though we dropped
> +		 * i_data_sem. Convert extent to written and write zeros there.
>  		 */
>  		ret = ext4_get_block_trans(inode, iblock, bh_result,
>  					   EXT4_GET_BLOCKS_CONVERT |
> @@ -3250,6 +3255,14 @@ int ext4_dax_mmap_get_block(struct inode *inode, sector_t iblock,
>  	clear_buffer_new(bh_result);
>  	return 0;
>  }
> +#else
> +/* Just define empty function, it will never get called. */
> +int ext4_dax_get_block(struct inode *inode, sector_t iblock,
> +		       struct buffer_head *bh_result, int create)
> +{
> +	BUG();
> +	return 0;
> +}

You don't need this stub.  All the uses of ext4_dax_get_block() are either
within their own '#ifdef CONFIG_FS_DAX' sections, or they are in an 
"if (IS_DAX)" conditional.  The latter will also be compiled out if
CONFIG_FS_DAX isn't defined.  This is because of the way that S_DAX is
defined:

  #define S_DAX		8192	/* Direct Access, avoiding the page cache */
  #else
  #define S_DAX		0	/* Make all the DAX code disappear */
  #endif

  reply	other threads:[~2016-04-29 18:01 UTC|newest]

Thread overview: 49+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-04-18 21:35 [RFC v3] [PATCH 0/18] DAX page fault locking Jan Kara
2016-04-18 21:35 ` [PATCH 01/18] ext4: Handle transient ENOSPC properly for DAX Jan Kara
2016-04-18 21:35 ` [PATCH 02/18] ext4: Fix race in transient ENOSPC detection Jan Kara
2016-04-18 21:35 ` [PATCH 03/18] DAX: move RADIX_DAX_ definitions to dax.c Jan Kara
2016-04-18 21:35 ` [PATCH 04/18] dax: Remove complete_unwritten argument Jan Kara
2016-04-18 21:35 ` [PATCH 05/18] ext2: Avoid DAX zeroing to corrupt data Jan Kara
2016-04-29 16:30   ` Ross Zwisler
2016-04-18 21:35 ` [PATCH 06/18] dax: Remove dead zeroing code from fault handlers Jan Kara
2016-04-29 16:48   ` Ross Zwisler
2016-04-18 21:35 ` [PATCH 07/18] ext4: Refactor direct IO code Jan Kara
2016-04-18 21:35 ` [PATCH 08/18] ext4: Pre-zero allocated blocks for DAX IO Jan Kara
2016-04-29 18:01   ` Ross Zwisler [this message]
2016-05-02 13:09     ` Jan Kara
2016-04-18 21:35 ` [PATCH 09/18] dax: Remove zeroing from dax_io() Jan Kara
2016-04-29 18:56   ` Ross Zwisler
2016-04-18 21:35 ` [PATCH 10/18] dax: Remove pointless writeback from dax_do_io() Jan Kara
2016-04-29 19:00   ` Ross Zwisler
2016-04-18 21:35 ` [PATCH 11/18] dax: Fix condition for filling of PMD holes Jan Kara
2016-04-29 19:08   ` Ross Zwisler
2016-05-02 13:16     ` Jan Kara
2016-04-18 21:35 ` [PATCH 12/18] dax: Remove redundant inode size checks Jan Kara
2016-04-18 21:35 ` [PATCH 13/18] dax: Make huge page handling depend of CONFIG_BROKEN Jan Kara
2016-04-29 19:53   ` Ross Zwisler
2016-05-02 13:19     ` Jan Kara
2016-04-18 21:35 ` [PATCH 14/18] dax: Define DAX lock bit for radix tree exceptional entry Jan Kara
2016-04-29 20:03   ` Ross Zwisler
2016-04-18 21:35 ` [PATCH 15/18] dax: Allow DAX code to replace exceptional entries Jan Kara
2016-04-29 20:29   ` Ross Zwisler
2016-04-18 21:35 ` [PATCH 16/18] dax: New fault locking Jan Kara
2016-04-27  4:27   ` NeilBrown
2016-05-06  4:13   ` Ross Zwisler
2016-05-10 12:27     ` Jan Kara
2016-05-11 19:26       ` Ross Zwisler
2016-05-12  7:58         ` Jan Kara
2016-04-18 21:35 ` [PATCH 17/18] dax: Use radix tree entry lock to protect cow faults Jan Kara
2016-04-19 11:46   ` Jerome Glisse
2016-04-19 14:33     ` Jan Kara
2016-04-19 15:19       ` Jerome Glisse
2016-04-18 21:35 ` [PATCH 18/18] dax: Remove i_mmap_lock protection Jan Kara
2016-05-06  3:35 ` [RFC v3] [PATCH 0/18] DAX page fault locking Ross Zwisler
2016-05-06 20:33 ` Ross Zwisler
2016-05-09  9:38   ` Jan Kara
2016-05-10 15:28     ` Jan Kara
2016-05-10 20:30       ` Ross Zwisler
2016-05-10 22:39         ` Ross Zwisler
2016-05-11  9:19           ` Jan Kara
2016-05-11 15:52             ` Ross Zwisler
2016-05-09 21:28 ` Verma, Vishal L
2016-05-10 11:52   ` Jan Kara

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20160429180158.GC5888@linux.intel.com \
    --to=ross.zwisler@linux.intel.com \
    --cc=dan.j.williams@intel.com \
    --cc=jack@suse.cz \
    --cc=linux-ext4@vger.kernel.org \
    --cc=linux-fsdevel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=linux-nvdimm@lists.01.org \
    --cc=willy@linux.intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).