Re: [RFC PATCH 1/4] vfs: Create fs_context-aware mount_bdev() replacement

[Andrew Price]

Hi David,

I've been testing gfs2 on top of this patch and it seems...

On 19/03/2019 16:23, David Howells wrote:
> Create a function, vfs_get_block_super(), that is fs_context-aware and a
> replacement for mount_bdev().  It caches the block device pointer and file
> open mode in the fs_context struct so that this information can be passed
> into sget_fc()'s test and set functions.
>
> Signed-off-by: David Howells <dhowells@xxxxxxxxxx>
> ---
>
>   fs/fs_context.c            |    2 +
>   fs/super.c                 |  106 ++++++++++++++++++++++++++++++++++++++++++++
>   include/linux/fs_context.h |    6 ++
>   3 files changed, 114 insertions(+)
>
> diff --git a/fs/fs_context.c b/fs/fs_context.c
> index 87e3546b9a52..ea027762c0b2 100644
> --- a/fs/fs_context.c
> +++ b/fs/fs_context.c
> @@ -425,6 +425,8 @@ void put_fs_context(struct fs_context *fc)
>   
>   	if (fc->need_free && fc->ops && fc->ops->free)
>   		fc->ops->free(fc);
> +	if (fc->bdev)
> +		blkdev_put(fc->bdev, fc->bdev_mode);

doing this means...

>   
>   	security_free_mnt_opts(&fc->security);
>   	put_net(fc->net_ns);
> diff --git a/fs/super.c b/fs/super.c
> index f27ee08fb26f..85851adb0f19 100644
> --- a/fs/super.c
> +++ b/fs/super.c
> @@ -1211,6 +1211,112 @@ int vfs_get_super(struct fs_context *fc,
>   EXPORT_SYMBOL(vfs_get_super);
>   
>   #ifdef CONFIG_BLOCK
> +static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
> +{
> +	s->s_bdev = fc->bdev;
> +	s->s_dev = s->s_bdev->bd_dev;
> +	s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
> +	fc->bdev = NULL;
> +	return 0;
> +}
> +
> +static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
> +{
> +	return s->s_bdev == fc->bdev;
> +}
> +
> +/**
> + * vfs_get_block_super - Get a superblock based on a single block device
> + * @fc: The filesystem context holding the parameters
> + * @keying: How to distinguish superblocks
> + * @fill_super: Helper to initialise a new superblock
> + */
> +int vfs_get_block_super(struct fs_context *fc,
> +			int (*fill_super)(struct super_block *,
> +					  struct fs_context *))
> +{
> +	struct block_device *bdev;
> +	struct super_block *s;
> +	int error = 0;
> +
> +	fc->bdev_mode = FMODE_READ | FMODE_EXCL;
> +	if (!(fc->sb_flags & SB_RDONLY))
> +		fc->bdev_mode |= FMODE_WRITE;
> +
> +	if (!fc->source)
> +		return invalf(fc, "No source specified");
> +
> +	bdev = blkdev_get_by_path(fc->source, fc->bdev_mode, fc->fs_type);
> +	if (IS_ERR(bdev)) {
> +		errorf(fc, "%s: Can't open blockdev", fc->source);
> +		return PTR_ERR(bdev);
> +	}
> +
> +	/* Once the superblock is inserted into the list by sget_fc(), s_umount
> +	 * will protect the lockfs code from trying to start a snapshot while
> +	 * we are mounting
> +	 */
> +	mutex_lock(&bdev->bd_fsfreeze_mutex);
> +	if (bdev->bd_fsfreeze_count > 0) {
> +		mutex_unlock(&bdev->bd_fsfreeze_mutex);
> +		warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
> +		error = -EBUSY;
> +		goto error_bdev;
> +	}
> +
> +	fc->bdev = bdev;
> +	fc->sb_flags |= SB_NOSEC;
> +	s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
> +	mutex_unlock(&bdev->bd_fsfreeze_mutex);
> +	if (IS_ERR(s)) {
> +		error = PTR_ERR(s);
> +		goto error_bdev;
> +	}
> +
> +	if (s->s_root) {
> +		/* Don't summarily change the RO/RW state. */
> +		if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
> +			warnf(fc, "%pg: Can't mount, would change RO state", bdev);
> +			error = -EBUSY;
> +			goto error_sb;
> +		}
> +
> +		/* s_umount nests inside bd_mutex during __invalidate_device().
> +		 * blkdev_put() acquires bd_mutex and can't be called under
> +		 * s_umount.  Drop s_umount temporarily.  This is safe as we're
> +		 * holding an active reference.
> +		 */
> +		up_write(&s->s_umount);
> +		blkdev_put(bdev, fc->bdev_mode);
> +		down_write(&s->s_umount);

fc->bdev should be NULLed here (or, on the way out of sget_fc() might be 
more appropriate) otherwise we get a double-blkdev_put() leading to NULL 
pointer derefs later. This happens when I mount a device twice and then 
unmount them, or mount it 3 times.

> +	} else {
> +		s->s_mode = fc->bdev_mode;
> +		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
> +		sb_set_blocksize(s, block_size(bdev));
> +		error = fill_super(s, fc);
> +		if (error)
> +			goto error_sb;
> +
> +		s->s_flags |= SB_ACTIVE;
> +		bdev->bd_super = s;
> +	}
> +
> +	BUG_ON(fc->root);

Maybe BUG_ON(fc->bdev); too?

Cheers,
Andy