| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <0b9efdbb-e897-45d2-8005-bd3d6f6fa2db@kernel.org>
Date: Mon, 25 Nov 2024 19:13:48 +0800
From: Chao Yu <chao@...nel.org>
To: 王秀红 (Xiuhong Wang) <Xiuhong.Wang@...soc.com>,
"jaegeuk@...nel.org" <jaegeuk@...nel.org>
Cc: Chao Yu <chao@...nel.org>,
"linux-f2fs-devel@...ts.sourceforge.net"
<linux-f2fs-devel@...ts.sourceforge.net>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
牛志国 (Zhiguo Niu) <Zhiguo.Niu@...soc.com>
Subject: Re: 答复: [PATCH v2 1/2] f2fs: fix to shrink read extent node in batches
On 2024/11/25 11:11, 王秀红 (Xiuhong Wang) wrote:
> Hi Chao,
>
> after tested in this weekend with these patch base on the orginal case, no issue reproduced, so
> Tested-by: Xiuhong Wang <xiuhong.wang@...soc.com>
Hi Xiuhong,
Thanks for helping to test.
Thanks,
>
> thanks!
>
> -----邮件原件-----
> 发件人: Chao Yu <chao@...nel.org>
> 发送时间: 2024年11月22日 14:50
> 收件人: jaegeuk@...nel.org
> 抄送: linux-f2fs-devel@...ts.sourceforge.net; linux-kernel@...r.kernel.org; Chao Yu <chao@...nel.org>; 王秀红 (Xiuhong Wang) <Xiuhong.Wang@...soc.com>; 牛志国 (Zhiguo Niu) <Zhiguo.Niu@...soc.com>
> 主题: [PATCH v2 1/2] f2fs: fix to shrink read extent node in batches
>
>
> 注意: 这封邮件来自于外部。除非你确定邮件内容安全，否则不要点击任何链接和附件。
> CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.
>
>
>
> We use rwlock to protect core structure data of extent tree during its shrink, however, if there is a huge number of extent nodes in extent tree, during shrink of extent tree, it may hold rwlock for a very long time, which may trigger kernel hang issue.
>
> This patch fixes to shrink read extent node in batches, so that, critical region of the rwlock can be shrunk to avoid its extreme long time hold.
>
> Reported-by: Xiuhong Wang <xiuhong.wang@...soc.com>
> Closes: https://lore.kernel.org/linux-f2fs-devel/20241112110627.1314632-1-xiuhong.wang@unisoc.com/
> Signed-off-by: Xiuhong Wang <xiuhong.wang@...soc.com>
> Signed-off-by: Zhiguo Niu <zhiguo.niu@...soc.com>
> Signed-off-by: Chao Yu <chao@...nel.org>
> ---
> v2:
> - no updates.
> fs/f2fs/extent_cache.c | 69 +++++++++++++++++++++++++-----------------
> 1 file changed, 41 insertions(+), 28 deletions(-)
>
> diff --git a/fs/f2fs/extent_cache.c b/fs/f2fs/extent_cache.c index 019c1f7b7fa5..b7a6817b44b0 100644
> --- a/fs/f2fs/extent_cache.c
> +++ b/fs/f2fs/extent_cache.c
> @@ -379,21 +379,22 @@ static struct extent_tree *__grab_extent_tree(struct inode *inode, }
>
> static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
> - struct extent_tree *et)
> + struct extent_tree *et, unsigned int
> + nr_shrink)
> {
> struct rb_node *node, *next;
> struct extent_node *en;
> - unsigned int count = atomic_read(&et->node_cnt);
> + unsigned int count;
>
> node = rb_first_cached(&et->root);
> - while (node) {
> +
> + for (count = 0; node && count < nr_shrink; count++) {
> next = rb_next(node);
> en = rb_entry(node, struct extent_node, rb_node);
> __release_extent_node(sbi, et, en);
> node = next;
> }
>
> - return count - atomic_read(&et->node_cnt);
> + return count;
> }
>
> static void __drop_largest_extent(struct extent_tree *et, @@ -622,6 +623,30 @@ static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
> return en;
> }
>
> +static unsigned int __destroy_extent_node(struct inode *inode,
> + enum extent_type type) {
> + struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
> + struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
> + unsigned int nr_shrink = type == EX_READ ?
> + READ_EXTENT_CACHE_SHRINK_NUMBER :
> + AGE_EXTENT_CACHE_SHRINK_NUMBER;
> + unsigned int node_cnt = 0;
> +
> + if (!et || !atomic_read(&et->node_cnt))
> + return 0;
> +
> + while (atomic_read(&et->node_cnt)) {
> + write_lock(&et->lock);
> + node_cnt += __free_extent_tree(sbi, et, nr_shrink);
> + write_unlock(&et->lock);
> + }
> +
> + f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
> +
> + return node_cnt;
> +}
> +
> static void __update_extent_tree_range(struct inode *inode,
> struct extent_info *tei, enum extent_type type) { @@ -760,9 +785,6 @@ static void __update_extent_tree_range(struct inode *inode,
> }
> }
>
> - if (is_inode_flag_set(inode, FI_NO_EXTENT))
> - __free_extent_tree(sbi, et);
> -
> if (et->largest_updated) {
> et->largest_updated = false;
> updated = true;
> @@ -780,6 +802,9 @@ static void __update_extent_tree_range(struct inode *inode,
> out_read_extent_cache:
> write_unlock(&et->lock);
>
> + if (is_inode_flag_set(inode, FI_NO_EXTENT))
> + __destroy_extent_node(inode, EX_READ);
> +
> if (updated)
> f2fs_mark_inode_dirty_sync(inode, true); } @@ -942,10 +967,14 @@ static unsigned int __shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink
> list_for_each_entry_safe(et, next, &eti->zombie_list, list) {
> if (atomic_read(&et->node_cnt)) {
> write_lock(&et->lock);
> - node_cnt += __free_extent_tree(sbi, et);
> + node_cnt += __free_extent_tree(sbi, et,
> + nr_shrink - node_cnt -
> + tree_cnt);
> write_unlock(&et->lock);
> }
> - f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
> +
> + if (atomic_read(&et->node_cnt))
> + goto unlock_out;
> +
> list_del_init(&et->list);
> radix_tree_delete(&eti->extent_tree_root, et->ino);
> kmem_cache_free(extent_tree_slab, et); @@ -1084,23 +1113,6 @@ unsigned int f2fs_shrink_age_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink
> return __shrink_extent_tree(sbi, nr_shrink, EX_BLOCK_AGE); }
>
> -static unsigned int __destroy_extent_node(struct inode *inode,
> - enum extent_type type)
> -{
> - struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
> - struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
> - unsigned int node_cnt = 0;
> -
> - if (!et || !atomic_read(&et->node_cnt))
> - return 0;
> -
> - write_lock(&et->lock);
> - node_cnt = __free_extent_tree(sbi, et);
> - write_unlock(&et->lock);
> -
> - return node_cnt;
> -}
> -
> void f2fs_destroy_extent_node(struct inode *inode) {
> __destroy_extent_node(inode, EX_READ); @@ -1109,7 +1121,6 @@ void f2fs_destroy_extent_node(struct inode *inode)
>
> static void __drop_extent_tree(struct inode *inode, enum extent_type type) {
> - struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
> struct extent_tree *et = F2FS_I(inode)->extent_tree[type];
> bool updated = false;
>
> @@ -1117,7 +1128,6 @@ static void __drop_extent_tree(struct inode *inode, enum extent_type type)
> return;
>
> write_lock(&et->lock);
> - __free_extent_tree(sbi, et);
> if (type == EX_READ) {
> set_inode_flag(inode, FI_NO_EXTENT);
> if (et->largest.len) {
> @@ -1126,6 +1136,9 @@ static void __drop_extent_tree(struct inode *inode, enum extent_type type)
> }
> }
> write_unlock(&et->lock);
> +
> + __destroy_extent_node(inode, type);
> +
> if (updated)
> f2fs_mark_inode_dirty_sync(inode, true); }
> --
> 2.40.1
>
Powered by blists - more mailing lists