Hi Jaegeuk,
Please drop this patch first, as it still can trigger deadlock as below, sorry
about this mistake.
- f2fs_put_super
- f2fs_wait_discard_bios
- __issue_discard_cmd
- lock(&dcc->cmd_lock); A
- __check_sit_bitmap
- down_read(&sit_i->sentry_lock) B
- do_write_page
- allocate_data_block
- down_write(&sit_i->sentry_lock) B
- locate_dirty_segment
- lock(&dirty_i->seglist_lock) C
- write_checkpoint
- clear_prefree_segments
- lock(&dirty_i->seglist_lock) C
- f2fs_issue_discard
- __issue_discard_async
- __queue_discard_cmd
- __update_discard_tree_range
- lock(&dcc->cmd_lock) A
Thanks,
On 2017/11/13 17:54, Chao Yu wrote:
> This patch protects sit bitmap accessing in __check_sit_bitmap with
> sentry_lock, meanwhile, in allocate_data_block, f2fs_wait_discard_bio
> will not access/update sit cache, let's exclude it from sentry_lock's
> region.>
> Signed-off-by: Chao Yu <yuch...@huawei.com>
> ---
> fs/f2fs/segment.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> index c117e0913f2a..464696b1612a 100644
> --- a/fs/f2fs/segment.c
> +++ b/fs/f2fs/segment.c
> @@ -894,6 +894,8 @@ void __check_sit_bitmap(struct f2fs_sb_info *sbi,
> unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
> unsigned long *map;
>
> + down_read(&SIT_I(sbi)->sentry_lock);
> +
> while (blk < end) {
> segno = GET_SEGNO(sbi, blk);
> sentry = get_seg_entry(sbi, segno);
> @@ -908,6 +910,8 @@ void __check_sit_bitmap(struct f2fs_sb_info *sbi,
> f2fs_bug_on(sbi, offset != size);
> blk = START_BLOCK(sbi, segno + 1);
> }
> +
> + up_read(&SIT_I(sbi)->sentry_lock);
> #endif
> }
>
> @@ -2527,12 +2531,13 @@ void allocate_data_block(struct f2fs_sb_info *sbi,
> struct page *page,
> down_read(&SM_I(sbi)->curseg_lock);
>
> mutex_lock(&curseg->curseg_mutex);
> - down_write(&sit_i->sentry_lock);
>
> *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
>
> f2fs_wait_discard_bio(sbi, *new_blkaddr);
>
> + down_write(&sit_i->sentry_lock);
> +
> /*
> * __add_sum_entry should be resided under the curseg_mutex
> * because, this function updates a summary entry in the
>
