[PATCH v3 0/3] md/bitmap: Optimal last page size

[PATCH v3 0/3] md/bitmap: Optimal last page size

[Jonathan Derrick]

From: Jon Derrick <jonathan.derrick@linux.dev>

Currently the last bitmap page write will size itself down to the logical block
size. This could cause less performance for devices which have atomic write
units larger than the block size, such as many NVMe devices with 4kB write
units and 512B block sizes. There is usually a large amount of space after the
bitmap and using the optimal I/O size could favor speed over size.

This was tested on an Intel/Solidigm P5520 drive with lba format 512B,
optimal I/O size of 4kB, resulting in a > 10x IOPS increase.

See patch 3 log for results.

v2->v3: Prep patch added and types fixed
Added helpers for optimal I/O sizes

Jon Derrick (3):
  md: Move sb writer loop to its own function
  md: Fix types in sb writer
  md: Use optimal I/O size for last bitmap page

 drivers/md/md-bitmap.c | 142 ++++++++++++++++++++++++-----------------
 1 file changed, 82 insertions(+), 60 deletions(-)

-- 
2.27.0

[PATCH v3 1/3] md: Move sb writer loop to its own function

[Jonathan Derrick]

From: Jon Derrick <jonathan.derrick@linux.dev>

Preparatory patch for optimal I/O size calculation. Move the sb writer
loop routine into its own function for clarity.

Signed-off-by: Jon Derrick <jonathan.derrick@linux.dev>
---
 drivers/md/md-bitmap.c | 123 +++++++++++++++++++++--------------------
 1 file changed, 63 insertions(+), 60 deletions(-)

diff --git a/drivers/md/md-bitmap.c b/drivers/md/md-bitmap.c
index e7cc6ba1b657..5c65268a2d97 100644
--- a/drivers/md/md-bitmap.c
+++ b/drivers/md/md-bitmap.c
@@ -209,76 +209,79 @@ static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mdde
 	return NULL;
 }
 
-static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
+static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
+			   struct page *page)
 {
-	struct md_rdev *rdev;
 	struct block_device *bdev;
 	struct mddev *mddev = bitmap->mddev;
 	struct bitmap_storage *store = &bitmap->storage;
+	loff_t offset = mddev->bitmap_info.offset;
+	int size = PAGE_SIZE;
+
+	bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
+	if (page->index == store->file_pages - 1) {
+		int last_page_size = store->bytes & (PAGE_SIZE - 1);
+		if (last_page_size == 0)
+			last_page_size = PAGE_SIZE;
+		size = roundup(last_page_size,
+			       bdev_logical_block_size(bdev));
+	}
+
+	/* Just make sure we aren't corrupting data or metadata */
+	if (mddev->external) {
+		/* Bitmap could be anywhere. */
+		if (rdev->sb_start + offset
+		    + (page->index * (PAGE_SIZE / SECTOR_SIZE))
+		    > rdev->data_offset &&
+		    rdev->sb_start + offset
+		    < (rdev->data_offset + mddev->dev_sectors
+		     + (PAGE_SIZE / SECTOR_SIZE)))
+			return -EINVAL;
+	} else if (offset < 0) {
+		/* DATA  BITMAP METADATA  */
+		if (offset
+		    + (long)(page->index * (PAGE_SIZE / SECTOR_SIZE))
+		    + size / SECTOR_SIZE > 0)
+			/* bitmap runs in to metadata */
+			return -EINVAL;
+
+		if (rdev->data_offset + mddev->dev_sectors
+		    > rdev->sb_start + offset)
+			/* data runs in to bitmap */
+			return -EINVAL;
+	} else if (rdev->sb_start < rdev->data_offset) {
+		/* METADATA BITMAP DATA */
+		if (rdev->sb_start + offset
+		    + page->index * (PAGE_SIZE / SECTOR_SIZE)
+		    + size / SECTOR_SIZE > rdev->data_offset)
+			/* bitmap runs in to data */
+			return -EINVAL;
+	} else {
+		/* DATA METADATA BITMAP - no problems */
+	}
 
-restart:
-	rdev = NULL;
-	while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
-		int size = PAGE_SIZE;
-		loff_t offset = mddev->bitmap_info.offset;
+	md_super_write(mddev, rdev,
+		       rdev->sb_start + offset
+		       + page->index * (PAGE_SIZE / SECTOR_SIZE),
+		       size, page);
+	return 0;
+}
 
-		bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
+static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
+{
+	struct md_rdev *rdev;
+	struct mddev *mddev = bitmap->mddev;
+	int ret;
 
-		if (page->index == store->file_pages-1) {
-			int last_page_size = store->bytes & (PAGE_SIZE-1);
-			if (last_page_size == 0)
-				last_page_size = PAGE_SIZE;
-			size = roundup(last_page_size,
-				       bdev_logical_block_size(bdev));
-		}
-		/* Just make sure we aren't corrupting data or
-		 * metadata
-		 */
-		if (mddev->external) {
-			/* Bitmap could be anywhere. */
-			if (rdev->sb_start + offset + (page->index
-						       * (PAGE_SIZE/512))
-			    > rdev->data_offset
-			    &&
-			    rdev->sb_start + offset
-			    < (rdev->data_offset + mddev->dev_sectors
-			     + (PAGE_SIZE/512)))
-				goto bad_alignment;
-		} else if (offset < 0) {
-			/* DATA  BITMAP METADATA  */
-			if (offset
-			    + (long)(page->index * (PAGE_SIZE/512))
-			    + size/512 > 0)
-				/* bitmap runs in to metadata */
-				goto bad_alignment;
-			if (rdev->data_offset + mddev->dev_sectors
-			    > rdev->sb_start + offset)
-				/* data runs in to bitmap */
-				goto bad_alignment;
-		} else if (rdev->sb_start < rdev->data_offset) {
-			/* METADATA BITMAP DATA */
-			if (rdev->sb_start
-			    + offset
-			    + page->index*(PAGE_SIZE/512) + size/512
-			    > rdev->data_offset)
-				/* bitmap runs in to data */
-				goto bad_alignment;
-		} else {
-			/* DATA METADATA BITMAP - no problems */
+	do {
+		while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
+			ret = __write_sb_page(rdev, bitmap, page);
+			if (ret)
+				return ret;
 		}
-		md_super_write(mddev, rdev,
-			       rdev->sb_start + offset
-			       + page->index * (PAGE_SIZE/512),
-			       size,
-			       page);
-	}
+	} while (wait && md_super_wait(mddev) < 0);
 
-	if (wait && md_super_wait(mddev) < 0)
-		goto restart;
 	return 0;
-
- bad_alignment:
-	return -EINVAL;
 }
 
 static void md_bitmap_file_kick(struct bitmap *bitmap);
-- 
2.27.0

Re: [PATCH v3 1/3] md: Move sb writer loop to its own function

[Christoph Hellwig]

I was going to complain about the formatting, but as that gets
fixed up in the next patch:

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH v3 1/3] md: Move sb writer loop to its own function

[kernel test robot]

Hi Jonathan,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on song-md/md-next]
[also build test WARNING on linus/master v6.2 next-20230222]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Jonathan-Derrick/md-Move-sb-writer-loop-to-its-own-function/20230223-060300
base:   git://git.kernel.org/pub/scm/linux/kernel/git/song/md.git md-next
patch link:    https://lore.kernel.org/r/20230222215828.225-2-jonathan.derrick%40linux.dev
patch subject: [PATCH v3 1/3] md: Move sb writer loop to its own function
config: alpha-allyesconfig (https://download.01.org/0day-ci/archive/20230223/202302231124.oHGZxi0D-lkp@intel.com/config)
compiler: alpha-linux-gcc (GCC) 12.1.0
reproduce (this is a W=1 build):
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # https://github.com/intel-lab-lkp/linux/commit/17d8d09a65e91fada0801ca9bf4e3560780bb543
        git remote add linux-review https://github.com/intel-lab-lkp/linux
        git fetch --no-tags linux-review Jonathan-Derrick/md-Move-sb-writer-loop-to-its-own-function/20230223-060300
        git checkout 17d8d09a65e91fada0801ca9bf4e3560780bb543
        # save the config file
        mkdir build_dir && cp config build_dir/.config
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-12.1.0 make.cross W=1 O=build_dir ARCH=alpha olddefconfig
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-12.1.0 make.cross W=1 O=build_dir ARCH=alpha SHELL=/bin/bash drivers/

If you fix the issue, kindly add following tag where applicable
| Reported-by: kernel test robot <lkp@intel.com>
| Link: https://lore.kernel.org/oe-kbuild-all/202302231124.oHGZxi0D-lkp@intel.com/

All warnings (new ones prefixed by >>):

   In function 'next_active_rdev',
       inlined from 'write_sb_page' at drivers/md/md-bitmap.c:277:18:
>> drivers/md/md-bitmap.c:192:12: warning: 'rdev' is used uninitialized [-Wuninitialized]
     192 |         if (rdev == NULL)
         |            ^
   drivers/md/md-bitmap.c: In function 'write_sb_page':
   drivers/md/md-bitmap.c:272:25: note: 'rdev' was declared here
     272 |         struct md_rdev *rdev;
         |                         ^~~~


vim +/rdev +192 drivers/md/md-bitmap.c

a654b9d8f851f4 drivers/md/bitmap.c    NeilBrown        2005-06-21  175  
fd01b88c75a718 drivers/md/bitmap.c    NeilBrown        2011-10-11  176  static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  177  {
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  178  	/* Iterate the disks of an mddev, using rcu to protect access to the
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  179  	 * linked list, and raising the refcount of devices we return to ensure
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  180  	 * they don't disappear while in use.
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  181  	 * As devices are only added or removed when raid_disk is < 0 and
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  182  	 * nr_pending is 0 and In_sync is clear, the entries we return will
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  183  	 * still be in the same position on the list when we re-enter
fd177481b440c3 drivers/md/bitmap.c    Michael Wang     2012-10-11  184  	 * list_for_each_entry_continue_rcu.
8532e3439087de drivers/md/bitmap.c    NeilBrown        2015-05-20  185  	 *
8532e3439087de drivers/md/bitmap.c    NeilBrown        2015-05-20  186  	 * Note that if entered with 'rdev == NULL' to start at the
8532e3439087de drivers/md/bitmap.c    NeilBrown        2015-05-20  187  	 * beginning, we temporarily assign 'rdev' to an address which
8532e3439087de drivers/md/bitmap.c    NeilBrown        2015-05-20  188  	 * isn't really an rdev, but which can be used by
8532e3439087de drivers/md/bitmap.c    NeilBrown        2015-05-20  189  	 * list_for_each_entry_continue_rcu() to find the first entry.
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  190  	 */
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  191  	rcu_read_lock();
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01 @192  	if (rdev == NULL)
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  193  		/* start at the beginning */
8532e3439087de drivers/md/bitmap.c    NeilBrown        2015-05-20  194  		rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  195  	else {
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  196  		/* release the previous rdev and start from there. */
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  197  		rdev_dec_pending(rdev, mddev);
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  198  	}
fd177481b440c3 drivers/md/bitmap.c    Michael Wang     2012-10-11  199  	list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  200  		if (rdev->raid_disk >= 0 &&
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  201  		    !test_bit(Faulty, &rdev->flags)) {
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  202  			/* this is a usable devices */
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  203  			atomic_inc(&rdev->nr_pending);
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  204  			rcu_read_unlock();
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  205  			return rdev;
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  206  		}
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  207  	}
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  208  	rcu_read_unlock();
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  209  	return NULL;
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  210  }
b2d2c4ceaddc30 drivers/md/bitmap.c    NeilBrown        2008-09-01  211  
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  212  static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  213  			   struct page *page)
a654b9d8f851f4 drivers/md/bitmap.c    NeilBrown        2005-06-21  214  {
a6ff7e089c7fca drivers/md/bitmap.c    Jonathan Brassow 2011-01-14  215  	struct block_device *bdev;
fd01b88c75a718 drivers/md/bitmap.c    NeilBrown        2011-10-11  216  	struct mddev *mddev = bitmap->mddev;
1ec885cdd01a9a drivers/md/bitmap.c    NeilBrown        2012-05-22  217  	struct bitmap_storage *store = &bitmap->storage;
f6af949c567211 drivers/md/bitmap.c    NeilBrown        2009-12-14  218  	loff_t offset = mddev->bitmap_info.offset;
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  219  	int size = PAGE_SIZE;
a6ff7e089c7fca drivers/md/bitmap.c    Jonathan Brassow 2011-01-14  220  
a6ff7e089c7fca drivers/md/bitmap.c    Jonathan Brassow 2011-01-14  221  	bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
9b1215c102d4b1 drivers/md/bitmap.c    NeilBrown        2012-05-22  222  	if (page->index == store->file_pages - 1) {
9b1215c102d4b1 drivers/md/bitmap.c    NeilBrown        2012-05-22  223  		int last_page_size = store->bytes & (PAGE_SIZE - 1);
9b1215c102d4b1 drivers/md/bitmap.c    NeilBrown        2012-05-22  224  		if (last_page_size == 0)
9b1215c102d4b1 drivers/md/bitmap.c    NeilBrown        2012-05-22  225  			last_page_size = PAGE_SIZE;
9b1215c102d4b1 drivers/md/bitmap.c    NeilBrown        2012-05-22  226  		size = roundup(last_page_size,
a6ff7e089c7fca drivers/md/bitmap.c    Jonathan Brassow 2011-01-14  227  			       bdev_logical_block_size(bdev));
9b1215c102d4b1 drivers/md/bitmap.c    NeilBrown        2012-05-22  228  	}
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  229  
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  230  	/* Just make sure we aren't corrupting data or metadata */
f6af949c567211 drivers/md/bitmap.c    NeilBrown        2009-12-14  231  	if (mddev->external) {
f6af949c567211 drivers/md/bitmap.c    NeilBrown        2009-12-14  232  		/* Bitmap could be anywhere. */
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  233  		if (rdev->sb_start + offset
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  234  		    + (page->index * (PAGE_SIZE / SECTOR_SIZE))
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  235  		    > rdev->data_offset &&
ac2f40be46ce6a drivers/md/bitmap.c    NeilBrown        2010-06-01  236  		    rdev->sb_start + offset
ac2f40be46ce6a drivers/md/bitmap.c    NeilBrown        2010-06-01  237  		    < (rdev->data_offset + mddev->dev_sectors
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  238  		     + (PAGE_SIZE / SECTOR_SIZE)))
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  239  			return -EINVAL;
f6af949c567211 drivers/md/bitmap.c    NeilBrown        2009-12-14  240  	} else if (offset < 0) {
f0d76d70bc77b9 drivers/md/bitmap.c    NeilBrown        2007-07-17  241  		/* DATA  BITMAP METADATA  */
42a04b5078ce73 drivers/md/bitmap.c    NeilBrown        2009-12-14  242  		if (offset
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  243  		    + (long)(page->index * (PAGE_SIZE / SECTOR_SIZE))
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  244  		    + size / SECTOR_SIZE > 0)
f0d76d70bc77b9 drivers/md/bitmap.c    NeilBrown        2007-07-17  245  			/* bitmap runs in to metadata */
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  246  			return -EINVAL;
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  247  
58c0fed400603a drivers/md/bitmap.c    Andre Noll       2009-03-31  248  		if (rdev->data_offset + mddev->dev_sectors
42a04b5078ce73 drivers/md/bitmap.c    NeilBrown        2009-12-14  249  		    > rdev->sb_start + offset)
f0d76d70bc77b9 drivers/md/bitmap.c    NeilBrown        2007-07-17  250  			/* data runs in to bitmap */
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  251  			return -EINVAL;
0f420358e3a2ab drivers/md/bitmap.c    Andre Noll       2008-07-11  252  	} else if (rdev->sb_start < rdev->data_offset) {
f0d76d70bc77b9 drivers/md/bitmap.c    NeilBrown        2007-07-17  253  		/* METADATA BITMAP DATA */
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  254  		if (rdev->sb_start + offset
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  255  		    + page->index * (PAGE_SIZE / SECTOR_SIZE)
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  256  		    + size / SECTOR_SIZE > rdev->data_offset)
f0d76d70bc77b9 drivers/md/bitmap.c    NeilBrown        2007-07-17  257  			/* bitmap runs in to data */
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  258  			return -EINVAL;
f0d76d70bc77b9 drivers/md/bitmap.c    NeilBrown        2007-07-17  259  	} else {
f0d76d70bc77b9 drivers/md/bitmap.c    NeilBrown        2007-07-17  260  		/* DATA METADATA BITMAP - no problems */
f0d76d70bc77b9 drivers/md/bitmap.c    NeilBrown        2007-07-17  261  	}
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  262  
a654b9d8f851f4 drivers/md/bitmap.c    NeilBrown        2005-06-21  263  	md_super_write(mddev, rdev,
42a04b5078ce73 drivers/md/bitmap.c    NeilBrown        2009-12-14  264  		       rdev->sb_start + offset
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  265  		       + page->index * (PAGE_SIZE / SECTOR_SIZE),
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  266  		       size, page);
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  267  	return 0;
ab6085c795a71b drivers/md/bitmap.c    NeilBrown        2007-05-23  268  }
a654b9d8f851f4 drivers/md/bitmap.c    NeilBrown        2005-06-21  269  
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  270  static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  271  {
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  272  	struct md_rdev *rdev;
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  273  	struct mddev *mddev = bitmap->mddev;
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  274  	int ret;
4b80991c6cb9ef drivers/md/bitmap.c    NeilBrown        2008-07-21  275  
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  276  	do {
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22 @277  		while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  278  			ret = __write_sb_page(rdev, bitmap, page);
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  279  			if (ret)
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  280  				return ret;
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  281  		}
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  282  	} while (wait && md_super_wait(mddev) < 0);
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  283  
17d8d09a65e91f drivers/md/md-bitmap.c Jon Derrick      2023-02-22  284  	return 0;
a654b9d8f851f4 drivers/md/bitmap.c    NeilBrown        2005-06-21  285  }
a654b9d8f851f4 drivers/md/bitmap.c    NeilBrown        2005-06-21  286  

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests

[PATCH v3 2/3] md: Fix types in sb writer

[Jonathan Derrick]

From: Jon Derrick <jonathan.derrick@linux.dev>

Page->index is a pgoff_t and multiplying could cause overflows on a
32-bit architecture. In the sb writer, this is used to calculate and
verify the sector being used, and is multiplied by a sector value. Using
sector_t will cast it to a u64 type and is the more appropriate type for
the unit. Additionally, the integer size unit is converted to a sector
unit in later calculations, and is now corrected to be an unsigned type.

Finally, clean up the calculations using variable aliases to improve
readabiliy.

Signed-off-by: Jon Derrick <jonathan.derrick@linux.dev>
---
 drivers/md/md-bitmap.c | 36 +++++++++++++++---------------------
 1 file changed, 15 insertions(+), 21 deletions(-)

diff --git a/drivers/md/md-bitmap.c b/drivers/md/md-bitmap.c
index 5c65268a2d97..11f4453775ee 100644
--- a/drivers/md/md-bitmap.c
+++ b/drivers/md/md-bitmap.c
@@ -215,55 +215,49 @@ static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
 	struct block_device *bdev;
 	struct mddev *mddev = bitmap->mddev;
 	struct bitmap_storage *store = &bitmap->storage;
-	loff_t offset = mddev->bitmap_info.offset;
-	int size = PAGE_SIZE;
+	sector_t offset = mddev->bitmap_info.offset;
+	sector_t ps, sboff, doff;
+	unsigned int size = PAGE_SIZE;
 
 	bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
 	if (page->index == store->file_pages - 1) {
-		int last_page_size = store->bytes & (PAGE_SIZE - 1);
+		unsigned int last_page_size = store->bytes & (PAGE_SIZE - 1);
 		if (last_page_size == 0)
 			last_page_size = PAGE_SIZE;
 		size = roundup(last_page_size,
 			       bdev_logical_block_size(bdev));
 	}
 
+	ps = page->index * (PAGE_SIZE / SECTOR_SIZE);
+	sboff = rdev->sb_start + offset;
+	doff = rdev->data_offset;
+
 	/* Just make sure we aren't corrupting data or metadata */
 	if (mddev->external) {
 		/* Bitmap could be anywhere. */
-		if (rdev->sb_start + offset
-		    + (page->index * (PAGE_SIZE / SECTOR_SIZE))
-		    > rdev->data_offset &&
-		    rdev->sb_start + offset
-		    < (rdev->data_offset + mddev->dev_sectors
-		     + (PAGE_SIZE / SECTOR_SIZE)))
+		if (sboff + ps > doff &&
+		    sboff < (doff + mddev->dev_sectors + (PAGE_SIZE / SECTOR_SIZE)))
 			return -EINVAL;
 	} else if (offset < 0) {
 		/* DATA  BITMAP METADATA  */
-		if (offset
-		    + (long)(page->index * (PAGE_SIZE / SECTOR_SIZE))
-		    + size / SECTOR_SIZE > 0)
+		if (offset + ps + size / SECTOR_SIZE > 0)
 			/* bitmap runs in to metadata */
 			return -EINVAL;
 
-		if (rdev->data_offset + mddev->dev_sectors
-		    > rdev->sb_start + offset)
+		if (doff + mddev->dev_sectors > sboff)
 			/* data runs in to bitmap */
 			return -EINVAL;
 	} else if (rdev->sb_start < rdev->data_offset) {
 		/* METADATA BITMAP DATA */
-		if (rdev->sb_start + offset
-		    + page->index * (PAGE_SIZE / SECTOR_SIZE)
-		    + size / SECTOR_SIZE > rdev->data_offset)
+		if (sboff + ps + size / SECTOR_SIZE > doff)
 			/* bitmap runs in to data */
 			return -EINVAL;
 	} else {
 		/* DATA METADATA BITMAP - no problems */
 	}
 
-	md_super_write(mddev, rdev,
-		       rdev->sb_start + offset
-		       + page->index * (PAGE_SIZE / SECTOR_SIZE),
-		       size, page);
+	md_super_write(mddev, rdev, sboff + ps,
+		       (int) size, page);
 	return 0;
 }
 
-- 
2.27.0

Re: [PATCH v3 2/3] md: Fix types in sb writer

[Christoph Hellwig]

On Wed, Feb 22, 2023 at 02:58:27PM -0700, Jonathan Derrick wrote:
> From: Jon Derrick <jonathan.derrick@linux.dev>
> 
> Page->index is a pgoff_t and multiplying could cause overflows on a
> 32-bit architecture. In the sb writer, this is used to calculate and
> verify the sector being used, and is multiplied by a sector value. Using
> sector_t will cast it to a u64 type and is the more appropriate type for
> the unit. Additionally, the integer size unit is converted to a sector
> unit in later calculations, and is now corrected to be an unsigned type.
> 
> Finally, clean up the calculations using variable aliases to improve
> readabiliy.
> 
> Signed-off-by: Jon Derrick <jonathan.derrick@linux.dev>
> ---
>  drivers/md/md-bitmap.c | 36 +++++++++++++++---------------------
>  1 file changed, 15 insertions(+), 21 deletions(-)
> 
> diff --git a/drivers/md/md-bitmap.c b/drivers/md/md-bitmap.c
> index 5c65268a2d97..11f4453775ee 100644
> --- a/drivers/md/md-bitmap.c
> +++ b/drivers/md/md-bitmap.c
> @@ -215,55 +215,49 @@ static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
>  	struct block_device *bdev;
>  	struct mddev *mddev = bitmap->mddev;
>  	struct bitmap_storage *store = &bitmap->storage;
> -	loff_t offset = mddev->bitmap_info.offset;
> -	int size = PAGE_SIZE;
> +	sector_t offset = mddev->bitmap_info.offset;
> +	sector_t ps, sboff, doff;
> +	unsigned int size = PAGE_SIZE;
>  
>  	bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
>  	if (page->index == store->file_pages - 1) {
> -		int last_page_size = store->bytes & (PAGE_SIZE - 1);
> +		unsigned int last_page_size = store->bytes & (PAGE_SIZE - 1);
>  		if (last_page_size == 0)

This should probably grow an empty line after the variable
declaration (here or in the previous patch).

> +		    sboff < (doff + mddev->dev_sectors + (PAGE_SIZE / SECTOR_SIZE)))

No need for either pair of braces here.

> -		       rdev->sb_start + offset
> -		       + page->index * (PAGE_SIZE / SECTOR_SIZE),
> -		       size, page);
> +	md_super_write(mddev, rdev, sboff + ps,
> +		       (int) size, page);

This easily fits into a single line now.

Otherwise looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>

[PATCH v3 3/3] md: Use optimal I/O size for last bitmap page

[Jonathan Derrick]

From: Jon Derrick <jonathan.derrick@linux.dev>

If the bitmap space has enough room, size the I/O for the last bitmap
page write to the optimal I/O size for the storage device. The expanded
write is checked that it won't overrun the data or metadata.

The drive this was tested against has higher latencies when there are
sub-4k writes due to device-side read-mod-writes of its atomic 4k write
unit. This change helps increase performance by sizing the last bitmap
page I/O for the device's preferred write unit, if it is given.

Example Intel/Solidigm P5520
Raid10, Chunk-size 64M, bitmap-size 57228 bits

$ mdadm --create /dev/md0 --level=10 --raid-devices=4 /dev/nvme{0,1,2,3}n1
        --assume-clean --bitmap=internal --bitmap-chunk=64M
$ fio --name=test --direct=1 --filename=/dev/md0 --rw=randwrite --bs=4k --runtime=60

Without patch:
  write: IOPS=1676, BW=6708KiB/s (6869kB/s)(393MiB/60001msec); 0 zone resets

With patch:
  write: IOPS=15.7k, BW=61.4MiB/s (64.4MB/s)(3683MiB/60001msec); 0 zone resets

Biosnoop:
Without patch:
Time        Process        PID     Device      LBA        Size      Lat
1.410377    md0_raid10     6900    nvme0n1   W 16         4096      0.02
1.410387    md0_raid10     6900    nvme2n1   W 16         4096      0.02
1.410374    md0_raid10     6900    nvme3n1   W 16         4096      0.01
1.410381    md0_raid10     6900    nvme1n1   W 16         4096      0.02
1.410411    md0_raid10     6900    nvme1n1   W 115346512  4096      0.01
1.410418    md0_raid10     6900    nvme0n1   W 115346512  4096      0.02
1.410915    md0_raid10     6900    nvme2n1   W 24         3584      0.43 <--
1.410935    md0_raid10     6900    nvme3n1   W 24         3584      0.45 <--
1.411124    md0_raid10     6900    nvme1n1   W 24         3584      0.64 <--
1.411147    md0_raid10     6900    nvme0n1   W 24         3584      0.66 <--
1.411176    md0_raid10     6900    nvme3n1   W 2019022184 4096      0.01
1.411189    md0_raid10     6900    nvme2n1   W 2019022184 4096      0.02

With patch:
Time        Process        PID     Device      LBA        Size      Lat
5.747193    md0_raid10     727     nvme0n1   W 16         4096      0.01
5.747192    md0_raid10     727     nvme1n1   W 16         4096      0.02
5.747195    md0_raid10     727     nvme3n1   W 16         4096      0.01
5.747202    md0_raid10     727     nvme2n1   W 16         4096      0.02
5.747229    md0_raid10     727     nvme3n1   W 1196223704 4096      0.02
5.747224    md0_raid10     727     nvme0n1   W 1196223704 4096      0.01
5.747279    md0_raid10     727     nvme0n1   W 24         4096      0.01 <--
5.747279    md0_raid10     727     nvme1n1   W 24         4096      0.02 <--
5.747284    md0_raid10     727     nvme3n1   W 24         4096      0.02 <--
5.747291    md0_raid10     727     nvme2n1   W 24         4096      0.02 <--
5.747314    md0_raid10     727     nvme2n1   W 2234636712 4096      0.01
5.747317    md0_raid10     727     nvme1n1   W 2234636712 4096      0.02

Signed-off-by: Jon Derrick <jonathan.derrick@linux.dev>
---
 drivers/md/md-bitmap.c | 33 +++++++++++++++++++++++++++++----
 1 file changed, 29 insertions(+), 4 deletions(-)

diff --git a/drivers/md/md-bitmap.c b/drivers/md/md-bitmap.c
index 11f4453775ee..196984fc3776 100644
--- a/drivers/md/md-bitmap.c
+++ b/drivers/md/md-bitmap.c
@@ -209,6 +209,28 @@ static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mdde
 	return NULL;
 }
 
+static unsigned int optimal_io_size(struct block_device *bdev,
+				    unsigned int last_page_size,
+				    unsigned int io_size)
+{
+	if (bdev_io_opt(bdev) > bdev_logical_block_size(bdev))
+		return roundup(last_page_size, bdev_io_opt(bdev));
+	return io_size;
+}
+
+static unsigned int bitmap_io_size(unsigned int io_size, unsigned int opt_size,
+				   sector_t start, sector_t boundary)
+{
+	if (io_size != opt_size &&
+	    start + opt_size / SECTOR_SIZE <= boundary)
+		return opt_size;
+	else if (start + io_size / SECTOR_SIZE <= boundary)
+		return io_size;
+
+	/* Overflows boundary */
+	return 0;
+}
+
 static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
 			   struct page *page)
 {
@@ -218,14 +240,15 @@ static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
 	sector_t offset = mddev->bitmap_info.offset;
 	sector_t ps, sboff, doff;
 	unsigned int size = PAGE_SIZE;
+	unsigned int opt_size = PAGE_SIZE;
 
 	bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
 	if (page->index == store->file_pages - 1) {
 		unsigned int last_page_size = store->bytes & (PAGE_SIZE - 1);
 		if (last_page_size == 0)
 			last_page_size = PAGE_SIZE;
-		size = roundup(last_page_size,
-			       bdev_logical_block_size(bdev));
+		size = roundup(last_page_size, bdev_logical_block_size(bdev));
+		opt_size = optimal_io_size(bdev, last_page_size, size);
 	}
 
 	ps = page->index * (PAGE_SIZE / SECTOR_SIZE);
@@ -240,7 +263,8 @@ static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
 			return -EINVAL;
 	} else if (offset < 0) {
 		/* DATA  BITMAP METADATA  */
-		if (offset + ps + size / SECTOR_SIZE > 0)
+		size = bitmap_io_size(size, opt_size, offset + ps, 0);
+		if (size == 0)
 			/* bitmap runs in to metadata */
 			return -EINVAL;
 
@@ -249,7 +273,8 @@ static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
 			return -EINVAL;
 	} else if (rdev->sb_start < rdev->data_offset) {
 		/* METADATA BITMAP DATA */
-		if (sboff + ps + size / SECTOR_SIZE > doff)
+		size = bitmap_io_size(size, opt_size, sboff + ps, doff);
+		if (size == 0)
 			/* bitmap runs in to data */
 			return -EINVAL;
 	} else {
-- 
2.27.0

Re: [PATCH v3 3/3] md: Use optimal I/O size for last bitmap page

[Christoph Hellwig]

On Wed, Feb 22, 2023 at 02:58:28PM -0700, Jonathan Derrick wrote:
> +	if (io_size != opt_size &&
> +	    start + opt_size / SECTOR_SIZE <= boundary)
> +		return opt_size;
> +	else if (start + io_size / SECTOR_SIZE <= boundary)

No need for an else after a return.

Otherwise looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH v3 3/3] md: Use optimal I/O size for last bitmap page

[Reindl Harald]

Am 23.02.23 um 00:42 schrieb Christoph Hellwig:
> On Wed, Feb 22, 2023 at 02:58:28PM -0700, Jonathan Derrick wrote:
>> +	if (io_size != opt_size &&
>> +	    start + opt_size / SECTOR_SIZE <= boundary)
>> +		return opt_size;
>> +	else if (start + io_size / SECTOR_SIZE <= boundary)
> 
> No need for an else after a return.
> 
> Otherwise looks good:
> 
> Reviewed-by: Christoph Hellwig <hch@lst.de>

the "return" is within the if-condition and has nothing to do with the 
else - with {} it would be clearly visible

Re: [PATCH v3 3/3] md: Use optimal I/O size for last bitmap page

[Christoph Hellwig]

On Thu, Feb 23, 2023 at 12:48:58AM +0100, Reindl Harald wrote:
> > > +	if (io_size != opt_size &&
> > > +	    start + opt_size / SECTOR_SIZE <= boundary)
> > > +		return opt_size;
> > > +	else if (start + io_size / SECTOR_SIZE <= boundary)
> > 
> > No need for an else after a return.
> > 
> > Otherwise looks good:
> > 
> > Reviewed-by: Christoph Hellwig <hch@lst.de>
> 
> the "return" is within the if-condition and has nothing to do with the else
> - with {} it would be clearly visible

That doesn't change the fact that it's not actually needed.