[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <4fc164e8-f307-0217-3977-39f3fa412a21@bytedance.com>
Date: Sun, 16 Oct 2022 23:05:39 +0800
From: Jia Zhu <zhujia.zj@...edance.com>
To: Dawei Li <set_pte_at@...look.com>, xiang@...nel.org,
chao@...nel.org
Cc: huyue2@...lpad.com, linux-erofs@...ts.ozlabs.org,
linux-kernel@...r.kernel.org, zhujia.zj@...edance.com
Subject: Re: [External] [PATCH] erofs: protect s_inodes with s_inode_list_lock
在 2022/10/16 20:37, Dawei Li 写道:
> s_inodes is superblock-specific resource, which should be
> protected by sb's specific lock s_inode_list_lock.
> > base-commit: 8436c4a57bd147b0bd2943ab499bb8368981b9e1
>
> Signed-off-by: Dawei Li <set_pte_at@...look.com>
> ---
> fs/erofs/fscache.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/fs/erofs/fscache.c b/fs/erofs/fscache.c
> index 998cd26a1b3b..bbf5268440df 100644
> --- a/fs/erofs/fscache.c
> +++ b/fs/erofs/fscache.c
> @@ -589,15 +589,18 @@ struct erofs_fscache *erofs_domain_register_cookie(struct super_block *sb,
> struct erofs_domain *domain = EROFS_SB(sb)->domain;
> struct super_block *psb = erofs_pseudo_mnt->mnt_sb;
>
> - mutex_lock(&erofs_domain_cookies_lock);
Hi Dawei,
Thanks for catching this.
I would suggest holding this mutex lock during inode searches and
inserts to avoid the following case:
P1 P2
lock inode_list lock
traverse sb->s_inodes
unlock inode_list lock
lock inode_list
traverse sb->s_inodes
unlock inode_list
domain_init_cookie
domain_init_cookie
Thanks,
Jia
> + spin_lock(&psb->s_inode_list_lock);
> list_for_each_entry(inode, &psb->s_inodes, i_sb_list) {
> ctx = inode->i_private;
> if (!ctx || ctx->domain != domain || strcmp(ctx->name, name))
> continue;
> igrab(inode);
> - mutex_unlock(&erofs_domain_cookies_lock);
> + spin_unlock(&psb->s_inode_list_lock);
> return ctx;
> }
> + spin_unlock(&psb->s_inode_list_lock);
> +
> + mutex_lock(&erofs_domain_cookies_lock);
> ctx = erofs_fscache_domain_init_cookie(sb, name, need_inode);
> mutex_unlock(&erofs_domain_cookies_lock);
> return ctx;
Powered by blists - more mailing lists