Re: [PATCH V3 4/6] erofs: introduce fscache-based domain

[JeffleXu <jefflexu@linux.alibaba.com>]

On 9/14/22 6:50 PM, Jia Zhu wrote:
> A new fscache-based shared domain mode is going to be introduced for
> erofs. In which case, same data blobs in same domain will be shared
> and reused to reduce on-disk space usage.
> 
> As the first step, we use pseudo mnt to manage and maintain domain's
> lifecycle.

The commit message needs to be updated, since the pseudo mnt is not
introduced yet in this patch.

> 
> The implementation of sharing blobs will be introduced in subsequent
> patches.
> 
> Signed-off-by: Jia Zhu <zhujia.zj@bytedance.com>
> ---
>  fs/erofs/fscache.c  | 134 ++++++++++++++++++++++++++++++++++++++------
>  fs/erofs/internal.h |   9 +++
>  2 files changed, 127 insertions(+), 16 deletions(-)
> 
> diff --git a/fs/erofs/fscache.c b/fs/erofs/fscache.c
> index 4159cf781924..b2100dc67cde 100644
> --- a/fs/erofs/fscache.c
> +++ b/fs/erofs/fscache.c
> @@ -1,10 +1,14 @@
>  // SPDX-License-Identifier: GPL-2.0-or-later
>  /*
>   * Copyright (C) 2022, Alibaba Cloud
> + * Copyright (C) 2022, Bytedance Inc. All rights reserved.
>   */
>  #include <linux/fscache.h>
>  #include "internal.h"
>  
> +static DEFINE_MUTEX(erofs_domain_list_lock);
> +static LIST_HEAD(erofs_domain_list);
> +
>  static struct netfs_io_request *erofs_fscache_alloc_request(struct address_space *mapping,
>  					     loff_t start, size_t len)
>  {
> @@ -417,6 +421,106 @@ const struct address_space_operations erofs_fscache_access_aops = {
>  	.readahead = erofs_fscache_readahead,
>  };
>  
> +static
> +struct erofs_domain *erofs_fscache_domain_get(struct erofs_domain *domain)
> +{
> +	refcount_inc(&domain->ref);

refcount_inc_not_zero() is prefered here.

Considering the following time sequence:

CPU1				CPU2
------				------
erofs_fscache_domain_put
  refcount decreased to 0
  				erofs_fscache_register_domain
				  mutex_lock
				  erofs_fscache_domain_get
				    inc refcount to 1
				  mutex_unlock
  mutex_lock
  remove the domain from list
  mutex_unlock



> +	return domain;
> +}
> +
> +static void erofs_fscache_domain_put(struct erofs_domain *domain)
> +{
> +	if (!domain)
> +		return;
> +	if (refcount_dec_and_test(&domain->ref)) {
> +		fscache_relinquish_volume(domain->volume, NULL, false);
> +		mutex_lock(&erofs_domain_list_lock);
> +		list_del(&domain->list);
> +		mutex_unlock(&erofs_domain_list_lock);

We need to delete the domain from the list first, and then calling
fscache_relinquish_volume(), so that others won't race with this.


> +		kfree(domain->domain_id);
> +		kfree(domain);
> +	}
> +}
> +
> +static int erofs_fscache_register_volume(struct super_block *sb)
> +{
> +	struct erofs_sb_info *sbi = EROFS_SB(sb);
> +	char *domain_id = sbi->opt.domain_id;
> +	struct fscache_volume *volume;
> +	char *name;
> +	int ret = 0;
> +
> +	if (domain_id)
> +		name = kasprintf(GFP_KERNEL, "erofs,%s", domain_id);
> +	else
> +		name = kasprintf(GFP_KERNEL, "erofs,%s", sbi->opt.fsid);
> +	if (!name)
> +		return -ENOMEM;
> +
> +	volume = fscache_acquire_volume(name, NULL, NULL, 0);
> +	if (IS_ERR_OR_NULL(volume)) {
> +		erofs_err(sb, "failed to register volume for %s", name);
> +		ret = volume ? PTR_ERR(volume) : -EOPNOTSUPP;
> +		volume = NULL;
> +	}
> +
> +	sbi->volume = volume;
> +	kfree(name);
> +	return ret;
> +}
> +
> +static int erofs_fscache_init_domain(struct super_block *sb)
> +{
> +	int err;
> +	struct erofs_domain *domain;
> +	struct erofs_sb_info *sbi = EROFS_SB(sb);
> +
> +	domain = kzalloc(sizeof(struct erofs_domain), GFP_KERNEL);
> +	if (!domain)
> +		return -ENOMEM;
> +
> +	domain->domain_id = kstrdup(sbi->opt.domain_id, GFP_KERNEL);
> +	if (!domain->domain_id) {
> +		kfree(domain);
> +		return -ENOMEM;
> +	}
> +	sbi->domain = domain;

Why bothering setting sbi->domain here? Can't we set sbi->domain finnaly
when the domain has been fully initialized?


> +	err = erofs_fscache_register_volume(sb);
> +	if (err)
> +		goto out;
> +
> +	domain->volume = sbi->volume;
> +	refcount_set(&domain->ref, 1);
> +	mutex_init(&domain->mutex);
> +	list_add(&domain->list, &erofs_domain_list);
> +	return 0;
> +out:
> +	kfree(domain->domain_id);
> +	kfree(domain);
> +	sbi->domain = NULL;
> +	return err;
> +}
> +
> +static int erofs_fscache_register_domain(struct super_block *sb)
> +{
> +	int err;
> +	struct erofs_domain *domain;
> +	struct erofs_sb_info *sbi = EROFS_SB(sb);
> +
> +	mutex_lock(&erofs_domain_list_lock);
> +	list_for_each_entry(domain, &erofs_domain_list, list) {
> +		if (!strcmp(domain->domain_id, sbi->opt.domain_id)) {
> +			sbi->domain = erofs_fscache_domain_get(domain);
> +			sbi->volume = domain->volume;
> +			mutex_unlock(&erofs_domain_list_lock);
> +			return 0;
> +		}
> +	}
> +	err = erofs_fscache_init_domain(sb);
> +	mutex_unlock(&erofs_domain_list_lock);
> +	return err;
> +}
> +
>  struct erofs_fscache *erofs_fscache_register_cookie(struct super_block *sb,
>  						     char *name, bool need_inode)
>  {
> @@ -486,24 +590,16 @@ void erofs_fscache_unregister_cookie(struct erofs_fscache *ctx)
>  int erofs_fscache_register_fs(struct super_block *sb)
>  {
>  	struct erofs_sb_info *sbi = EROFS_SB(sb);
> -	struct fscache_volume *volume;
>  	struct erofs_fscache *fscache;
> -	char *name;
> -	int ret = 0;
> +	int ret;
>  
> -	name = kasprintf(GFP_KERNEL, "erofs,%s", sbi->opt.fsid);
> -	if (!name)
> -		return -ENOMEM;
> +	if (sbi->opt.domain_id)
> +		ret = erofs_fscache_register_domain(sb);
> +	else
> +		ret = erofs_fscache_register_volume(sb);
>  
> -	volume = fscache_acquire_volume(name, NULL, NULL, 0);
> -	if (IS_ERR_OR_NULL(volume)) {
> -		erofs_err(sb, "failed to register volume for %s", name);
> -		ret = volume ? PTR_ERR(volume) : -EOPNOTSUPP;
> -		volume = NULL;
> -	}
> -
> -	sbi->volume = volume;
> -	kfree(name);
> +	if (ret)
> +		return ret;
>  
>  	fscache = erofs_fscache_register_cookie(sb, sbi->opt.fsid, true);
>  	if (IS_ERR(fscache))
> @@ -518,7 +614,13 @@ void erofs_fscache_unregister_fs(struct super_block *sb)
>  	struct erofs_sb_info *sbi = EROFS_SB(sb);
>  
>  	erofs_fscache_unregister_cookie(sbi->s_fscache);
> -	fscache_relinquish_volume(sbi->volume, NULL, false);
>  	sbi->s_fscache = NULL;
> +
> +	if (sbi->domain)
> +		erofs_fscache_domain_put(sbi->domain);
> +	else
> +		fscache_relinquish_volume(sbi->volume, NULL, false);
> +
>  	sbi->volume = NULL;
> +	sbi->domain = NULL;
>  }
> diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
> index 2d129c6b3027..5ce6889d6f1d 100644
> --- a/fs/erofs/internal.h
> +++ b/fs/erofs/internal.h
> @@ -99,6 +99,14 @@ struct erofs_sb_lz4_info {
>  	u16 max_pclusterblks;
>  };
>  
> +struct erofs_domain {
> +	refcount_t ref;
> +	struct mutex mutex;
> +	struct list_head list;
> +	struct fscache_volume *volume;
> +	char *domain_id;
> +};
> +
>  struct erofs_fscache {
>  	struct fscache_cookie *cookie;
>  	struct inode *inode;
> @@ -158,6 +166,7 @@ struct erofs_sb_info {
>  	/* fscache support */
>  	struct fscache_volume *volume;
>  	struct erofs_fscache *s_fscache;
> +	struct erofs_domain *domain;
>  };
>  
>  #define EROFS_SB(sb) ((struct erofs_sb_info *)(sb)->s_fs_info)

-- 
Thanks,
Jingbo