Re: [PATCH v2 04/18] btrfs: move space cache settings into open_ctree

[Josef Bacik <josef@toxicpanda.com>]

On Wed, Nov 15, 2023 at 08:06:02AM +0800, Anand Jain wrote:
> 
> > @@ -3287,6 +3287,12 @@ int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_device
> >   	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
> >   	fs_info->stripesize = stripesize;
> > +	/*
> > +	 * Handle the space caching options appropriately now that we have the
> > +	 * super loaded and validated.
> > +	 */
> > +	btrfs_set_free_space_cache_settings(fs_info);
> > +
> >   	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
> >   	if (ret)
> >   		goto fail_alloc;
> > @@ -3298,17 +3304,6 @@ int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_device
> >   	if (sectorsize < PAGE_SIZE) {
> >   		struct btrfs_subpage_info *subpage_info;
> > -		/*
> > -		 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
> > -		 * going to be deprecated.
> > -		 *
> > -		 * Force to use v2 cache for subpage case.
> > -		 */
> > -		btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
> > -		btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
> > -			"forcing free space tree for sector size %u with page size %lu",
> > -			sectorsize, PAGE_SIZE);
> > -
> >   		btrfs_warn(fs_info,
> >   		"read-write for sector size %u with page size %lu is experimental",
> >   			   sectorsize, PAGE_SIZE);
> > diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
> > index 639601d346d0..aef7e67538a3 100644
> > --- a/fs/btrfs/super.c
> > +++ b/fs/btrfs/super.c
> > @@ -266,6 +266,31 @@ static bool check_options(struct btrfs_fs_info *info, unsigned long flags)
> >   	return true;
> >   }
> > +void btrfs_set_free_space_cache_settings(struct btrfs_fs_info *fs_info)
> > +{
> > +	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
> > +		btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
> > +	else if (btrfs_free_space_cache_v1_active(fs_info)) {
> > +		if (btrfs_is_zoned(fs_info)) {
> > +			btrfs_info(fs_info,
> > +			"zoned: clearing existing space cache");
> > +			btrfs_set_super_cache_generation(fs_info->super_copy, 0);
> > +		} else {
> > +			btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
> > +		}
> > +	}
> > +
> > +	if (fs_info->sectorsize < PAGE_SIZE) {
> > +		btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
> > +		if (!btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
> > +			btrfs_info(fs_info,
> > +				   "forcing free space tree for sector size %u with page size %lu",
> > +				   fs_info->sectorsize, PAGE_SIZE);
> > +			btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
> > +		}
> > +	}
> > +}
> > +
> >   static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)
> >   {
> >   	char *opts;
> > @@ -345,18 +370,6 @@ int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
> >   	bool saved_compress_force;
> >   	int no_compress = 0;
> > -	if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
> > -		btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
> > -	else if (btrfs_free_space_cache_v1_active(info)) {
> > -		if (btrfs_is_zoned(info)) {
> > -			btrfs_info(info,
> > -			"zoned: clearing existing space cache");
> > -			btrfs_set_super_cache_generation(info->super_copy, 0);
> > -		} else {
> > -			btrfs_set_opt(info->mount_opt, SPACE_CACHE);
> > -		}
> > -	}
> > -
> >   	/*
> >   	 * Even the options are empty, we still need to do extra check
> >   	 * against new flags
> 
> 
> btrfs_remount() calls btrfs_parse_options(), which previously handled
> space cache/tree flags set/reset. However, with the move of this
> functionality to a separate function, btrfs_set_free_space_cache_settings(),
> the btrfs_remount() thread no longer has the space cache/tree flags
> set/reset. The changelog provides no explanation for this change.
> Could this be a bug?
> 

All this bit does is set the mount_opt's based on the current state of the file
system, so if we're mounting with no "-o space_cache=*" option, we'll derive it
from wether FREE_SPACE_TREE is set or btrfs_free_space_cache_v1_active() returns
true.

So in your case if we mounted -o nospace_cache then we'll have cleared the
v1_active bit and this won't do anything and we'll get -o nospace_cache again.
The same goes for the free space tree, we're only allowed to change the disk
option for this in certain cases.  And in those cases it'll clear the compat
flag.

But this is definitely subtle, I'll expand the changelog.  Thanks,

Josef