Re: [PATCH 2/2] exfat: add support FITRIM ioctl

[Chaitanya Kulkarni <Chaitanya.Kulkarni@wdc.com>]

On 2/14/21 20:28, Hyeongseok Kim wrote:
> add FITRIM ioctl to support trimming mounted filesystem
>
> Signed-off-by: Hyeongseok Kim <hyeongseok@gmail.com>
> ---
>  fs/exfat/balloc.c   | 89 +++++++++++++++++++++++++++++++++++++++++++++
>  fs/exfat/exfat_fs.h |  1 +
>  fs/exfat/file.c     | 33 +++++++++++++++++
>  3 files changed, 123 insertions(+)
>
> diff --git a/fs/exfat/balloc.c b/fs/exfat/balloc.c
> index 761c79c3a4ba..edd0f6912e8e 100644
> --- a/fs/exfat/balloc.c
> +++ b/fs/exfat/balloc.c
> @@ -273,3 +273,92 @@ int exfat_count_used_clusters(struct super_block *sb, unsigned int *ret_count)
>  	*ret_count = count;
>  	return 0;
>  }
> +
> +int exfat_trim_fs(struct inode *inode, struct fstrim_range *range)
> +{
> +	struct super_block *sb = inode->i_sb;
Reverse tree style for function variable declaration would be nice which you
partially have it here.
> +	struct exfat_sb_info *sbi = EXFAT_SB(sb);
> +	u64 clu_start, clu_end, trim_minlen, trimmed_total = 0;
> +	unsigned int trim_begin, trim_end, count;
> +	unsigned int next_free_clu;
> +	int err = 0;
> +
> +	clu_start = max_t(u64, range->start >> sbi->cluster_size_bits,
> +				EXFAT_FIRST_CLUSTER);
> +	clu_end = clu_start + (range->len >> sbi->cluster_size_bits) - 1;
> +	trim_minlen = range->minlen >> sbi->cluster_size_bits;
> +
> +	if (clu_start >= sbi->num_clusters || range->len < sbi->cluster_size)
> +		return -EINVAL;
> +
> +	if (clu_end >= sbi->num_clusters)
> +		clu_end = sbi->num_clusters - 1;
> +
> +	mutex_lock(&EXFAT_SB(inode->i_sb)->s_lock);
> +
> +	trim_begin = trim_end = exfat_find_free_bitmap(sb, clu_start);
> +	if (trim_begin == EXFAT_EOF_CLUSTER)
> +		goto unlock;
> +
> +	next_free_clu = exfat_find_free_bitmap(sb, trim_end + 1);
> +	if (next_free_clu == EXFAT_EOF_CLUSTER)
> +		goto unlock;
> +
> +	do {
> +		if (next_free_clu == trim_end + 1)
> +			/* extend trim range for continuous free cluster */
> +			trim_end++;
> +		else {
> +			/* trim current range if it's larger than trim_minlen */
> +			count = trim_end - trim_begin + 1;
> +			if (count >= trim_minlen) {
> +				err = sb_issue_discard(sb,
> +					exfat_cluster_to_sector(sbi, trim_begin),
> +					count * sbi->sect_per_clus, GFP_NOFS, 0);
You are specifying the last argument as 0 to sb_issue_disacrd() i.e.
flags == 0 this will propagate to :-

sb_issue_discard()
    blkdev_issue_discard()
        __blkdev__issue_discard()

Now blkdev_issue_disacrd() returns -ENOTSUPP in 3 cases :-

1. If flags arg is set to BLKDEV_DISCARD_SECURE and queue doesn't support
   secure erase. In this case you have not set BLKDEV_DISCARD_SECURE that.
   So it should not return -EOPNOTSUPP.
2. If queue doesn't support discard. In this case caller of this function
   already set that. So it should not return -EOPNOTSUPP.
3. If q->limits.discard_granularity is not but LLD which I think caller of
   this function already used that to calculate the range->minlen.

If above is true then err will not have value of -EOPNOTSUPP ?

> +				if (err && err != -EOPNOTSUPP)
> +					goto unlock;
> +				if (!err)
> +					trimmed_total += count;
> +			}
> +
> +			/* set next start point of the free hole */
> +			trim_begin = trim_end = next_free_clu;
> +		}
> +
> +		if (next_free_clu >= clu_end)
> +			break;
> +
> +		if (fatal_signal_pending(current)) {
> +			err = -ERESTARTSYS;
> +			goto unlock;
> +		}
> +
> +		if (need_resched()) {
> +			mutex_unlock(&EXFAT_SB(inode->i_sb)->s_lock);
sb_issue_discard() ->blkdev_issue_discard() will call cond_resced().
1. The check for need_resched() will ever be true since
blkdev_issue_discard()
   is already calling cond_sched() ?
2. If so do you still need to drop the mutex before calling
   sb_issue_discard() ?
> +			cond_resched();
> +			mutex_lock(&EXFAT_SB(inode->i_sb)->s_lock);
> +		}
> +
> +		next_free_clu = exfat_find_free_bitmap(sb, next_free_clu + 1);
> +
> +	} while (next_free_clu != EXFAT_EOF_CLUSTER &&
> +			next_free_clu > trim_end);
> +
> +	/* try to trim remainder */
> +	count = trim_end - trim_begin + 1;
> +	if (count >= trim_minlen) {
> +		err = sb_issue_discard(sb, exfat_cluster_to_sector(sbi, trim_begin),
> +			count * sbi->sect_per_clus, GFP_NOFS, 0);
> +		if (err && err != -EOPNOTSUPP)
> +			goto unlock;
> +
> +		if (!err)
> +			trimmed_total += count;
> +	}
> +
> +unlock:
> +	mutex_unlock(&EXFAT_SB(inode->i_sb)->s_lock);
> +	range->len = trimmed_total << sbi->cluster_size_bits;
> +
> +	return err;
> +}
> diff --git a/fs/exfat/exfat_fs.h b/fs/exfat/exfat_fs.h
> index a183021ae31d..e050aea0b639 100644
> --- a/fs/exfat/exfat_fs.h
> +++ b/fs/exfat/exfat_fs.h
> @@ -411,6 +411,7 @@ int exfat_set_bitmap(struct inode *inode, unsigned int clu);
>  void exfat_clear_bitmap(struct inode *inode, unsigned int clu, bool sync);
>  unsigned int exfat_find_free_bitmap(struct super_block *sb, unsigned int clu);
>  int exfat_count_used_clusters(struct super_block *sb, unsigned int *ret_count);
> +int exfat_trim_fs(struct inode *inode, struct fstrim_range *range);
>  
>  /* file.c */
>  extern const struct file_operations exfat_file_operations;
> diff --git a/fs/exfat/file.c b/fs/exfat/file.c
> index 679828e7be07..61a64a4d4e6a 100644
> --- a/fs/exfat/file.c
> +++ b/fs/exfat/file.c
> @@ -351,7 +351,40 @@ int exfat_setattr(struct dentry *dentry, struct iattr *attr)
>  
>  long exfat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
>  {
> +	struct inode *inode = file_inode(filp);
> +	struct super_block *sb = inode->i_sb;
> +
>  	switch (cmd) {
> +	case FITRIM:
> +	{
> +		struct request_queue *q = bdev_get_queue(sb->s_bdev);
> +		struct fstrim_range range;
> +		int ret = 0;
> +
> +		if (!capable(CAP_SYS_ADMIN))
> +			return -EPERM;
> +
> +		if (!blk_queue_discard(q))
> +			return -EOPNOTSUPP;
> +
> +		if (copy_from_user(&range, (struct fstrim_range __user *)arg,
> +			sizeof(range)))
> +			return -EFAULT;
> +
> +		range.minlen = max_t(unsigned int, range.minlen,
> +					q->limits.discard_granularity);
> +
> +		ret = exfat_trim_fs(inode, &range);
> +		if (ret < 0)
> +			return ret;
> +
> +		if (copy_to_user((struct fstrim_range __user *)arg, &range,
> +			sizeof(range)))
> +			return -EFAULT;
> +
> +		return 0;
> +	}
> +
Is {} really needed for switch case ?
Also, code related to FITRIM needs to be moved to a helper otherwise it
will bloat
the ioctl function, unless that is the objective here.
>  	default:
>  		return -ENOTTY;
>  	}