Message-ID: <68ef14b2-4954-4bfc-1e1f-f44ded02bba3@uls.co.za>
Date:   Wed, 22 Aug 2018 04:57:50 +0200
From:   Jaco Kroon <jaco@....co.za>
To:     Andreas Dilger <adilger@...ger.ca>,
        "Theodore Y. Ts'o" <tytso@....edu>
Cc:     linux-ext4 <linux-ext4@...r.kernel.org>
Subject: Re: [PATCH] Add block_high_watermark sysfs tunable.

Hi,

The below is based on suggestion from Andreas.  I opted for the sysfs
tunable rather than an ioctl (which could also be used) because this
enables to set the value without a special tool.

Andreas mentioned a number of functions to adjust for allocating blocks,
but only the two below seems to relate.  I've also worked through all
uses of ext4_blocks_count() and could not find further use cases that
needs adjustment.  Some of the uses is during mount only (cluster
reservations), and since this change is per-mount and nt superblock
persistent adjustments won't have any effect there.  If this is
desirable I'll try putting this in the superblock instead but this would
require allocating feature bits and I'm not sure this change is worth a
feature bit.

I'll attempt limiting inode allocation in ialloc based on this next,
just wanted to get feedback on the below first.

My big question is this:  how do I build a test case for this code?

As an aside, the resize2fs from 64T to 56T eventually finished some time
during the night.  So just under 17 days total.  If we can get an online
resize for the same to be double or even triple that in total it'll
still be a massive win for me.  Even if we need to for the final stages
take the filesystem offline for two days - I know a fsck on this system
takes ~18 hours (will probably be a bit less now), so a shrink without
having to move data blocks will take at least that time in all
likelihood, a debugfs ncheck took ~12 hours on 64TB (which is needed to
migrate inodes), icheck was ~12 minutes, most of the time spent to open
the filesystem.

Kind Regards,
Jaco

On 22/08/2018 04:21, Jaco Kroon wrote:
> NOT READY FOR MERGE!!!!!!
>
> Limiting block allocations to a high watermark will eventually enable us
> to perform online shrinks of an ext4 filesystem.  As an immediate
> benefit it'll prevent allocation of blocks in the high range, which if
> performed as a precursor to an offline filesystem shrink will help to
> reduce the overall time a filesystem needs to be taken offline in order
> to shrink it.
>
> (possible) shortcomings:
>
> Currently this tunable does not get stored to the superblock, and thus
> needs to be set again after each mount.
>
> The ext4_statfs function doesn't adjust the f_bavail value currently, as
> such df will report incorrect results.
>
> The inode allocator hasn't been synced yet.
> ---
>  fs/ext4/balloc.c  |  2 +-
>  fs/ext4/ext4.h    | 10 ++++++++++
>  fs/ext4/mballoc.c |  2 +-
>  fs/ext4/sysfs.c   | 19 +++++++++++++++++++
>  4 files changed, 31 insertions(+), 2 deletions(-)
>
> diff --git a/fs/ext4/balloc.c b/fs/ext4/balloc.c
> index e5d6ee61ff48..4f723c7a9c88 100644
> --- a/fs/ext4/balloc.c
> +++ b/fs/ext4/balloc.c
> @@ -883,7 +883,7 @@ ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
>  			block_group++;
>  	}
>  	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
> -	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
> +	last_block = ext4_blocks_max_allocatable(EXT4_SB(inode->i_sb)) - 1;
>  
>  	/*
>  	 * If we are doing delayed allocation, we don't need take
> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> index 0f0edd1cd0cd..dc30ea107c55 100644
> --- a/fs/ext4/ext4.h
> +++ b/fs/ext4/ext4.h
> @@ -1423,6 +1423,7 @@ struct ext4_sb_info {
>  	unsigned int s_mb_order2_reqs;
>  	unsigned int s_mb_group_prealloc;
>  	unsigned int s_max_dir_size_kb;
> +	ext4_fsblk_t s_block_high_watermark; /* allocators must not allocate blocks above this */
>  	/* where last allocation was done - for stream allocation */
>  	unsigned long s_mb_last_group;
>  	unsigned long s_mb_last_start;
> @@ -2711,6 +2712,15 @@ static inline ext4_fsblk_t ext4_blocks_count(struct ext4_super_block *es)
>  		le32_to_cpu(es->s_blocks_count_lo);
>  }
>  
> +static inline ext4_fsblk_t ext4_blocks_max_allocatable(struct ext4_sb_info *sbi)
> +{
> +	ext4_fsblk_t blocks = ext4_blocks_count(sbi->s_es);
> +	if (sbi->s_block_high_watermark && sbi->s_block_high_watermark < blocks)
> +		return sbi->s_block_high_watermark;
> +	else
> +		return blocks;
> +}
> +
>  static inline ext4_fsblk_t ext4_r_blocks_count(struct ext4_super_block *es)
>  {
>  	return ((ext4_fsblk_t)le32_to_cpu(es->s_r_blocks_count_hi) << 32) |
> diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
> index e29fce2fbf25..a158c2c9de10 100644
> --- a/fs/ext4/mballoc.c
> +++ b/fs/ext4/mballoc.c
> @@ -4232,7 +4232,7 @@ ext4_mb_initialize_context(struct ext4_allocation_context *ac,
>  	/* start searching from the goal */
>  	goal = ar->goal;
>  	if (goal < le32_to_cpu(es->s_first_data_block) ||
> -			goal >= ext4_blocks_count(es))
> +			goal >= ext4_blocks_max_allocatable(sbi))
>  		goal = le32_to_cpu(es->s_first_data_block);
>  	ext4_get_group_no_and_offset(sb, goal, &group, &block);
>  
> diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
> index 9212a026a1f1..2a1a955c2c0b 100644
> --- a/fs/ext4/sysfs.c
> +++ b/fs/ext4/sysfs.c
> @@ -30,6 +30,7 @@ typedef enum {
>  	attr_feature,
>  	attr_pointer_ui,
>  	attr_pointer_atomic,
> +	attr_block_high_watermark,
>  } attr_id_t;
>  
>  typedef enum {
> @@ -167,6 +168,7 @@ EXT4_ATTR_FUNC(delayed_allocation_blocks, 0444);
>  EXT4_ATTR_FUNC(session_write_kbytes, 0444);
>  EXT4_ATTR_FUNC(lifetime_write_kbytes, 0444);
>  EXT4_ATTR_FUNC(reserved_clusters, 0644);
> +EXT4_ATTR_FUNC(block_high_watermark, 0600);
>  
>  EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, inode_readahead,
>  		 ext4_sb_info, s_inode_readahead_blks);
> @@ -217,6 +219,7 @@ static struct attribute *ext4_attrs[] = {
>  	ATTR_LIST(errors_count),
>  	ATTR_LIST(first_error_time),
>  	ATTR_LIST(last_error_time),
> +	ATTR_LIST(block_high_watermark),
>  	NULL,
>  };
>  
> @@ -304,6 +307,9 @@ static ssize_t ext4_attr_show(struct kobject *kobj,
>  		return print_tstamp(buf, sbi->s_es, s_first_error_time);
>  	case attr_last_error_time:
>  		return print_tstamp(buf, sbi->s_es, s_last_error_time);
> +	case attr_block_high_watermark:
> +		return snprintf(buf, PAGE_SIZE, "%llu\n",
> +				(s64) sbi->s_block_high_watermark);
>  	}
>  
>  	return 0;
> @@ -318,6 +324,7 @@ static ssize_t ext4_attr_store(struct kobject *kobj,
>  	struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
>  	void *ptr = calc_ptr(a, sbi);
>  	unsigned long t;
> +	unsigned long long t2;
>  	int ret;
>  
>  	switch (a->attr_id) {
> @@ -338,6 +345,18 @@ static ssize_t ext4_attr_store(struct kobject *kobj,
>  		return inode_readahead_blks_store(sbi, buf, len);
>  	case attr_trigger_test_error:
>  		return trigger_test_error(sbi, buf, len);
> +	case attr_block_high_watermark:
> +		if (!ptr)
> +			return 0;
> +		ret = kstrtoull(skip_spaces(buf), 0, &t2);
> +		if (ret)
> +			return ret;
> +		if (t2 > ext4_blocks_count(sbi->s_es))
> +			return -EINVAL;
> +		if (t2 && t2 < le32_to_cpu(sbi->s_es->s_first_data_block))
> +			return -EINVAL;
> +		sbi->s_block_high_watermark = t2;
> +		return len;
>  	}
>  	return 0;
>  }

Powered by blists - more mailing lists