Messages in this thread | | | Date | Wed, 2 Mar 2022 11:32:27 +0800 | Subject | Re: [f2fs-dev] [PATCH] f2fs: fix to avoid potential deadlock | From | Chao Yu <> |
| |
ping,
On 2022/2/25 11:02, Chao Yu wrote: > On 2022/2/3 22:57, Chao Yu wrote: >> On 2022/2/3 9:51, Jaegeuk Kim wrote: >>> On 01/29, Chao Yu wrote: >>>> On 2022/1/29 8:37, Jaegeuk Kim wrote: >>>>> On 01/28, Chao Yu wrote: >>>>>> On 2022/1/28 5:59, Jaegeuk Kim wrote: >>>>>>> On 01/27, Chao Yu wrote: >>>>>>>> Quoted from Jing Xia's report, there is a potential deadlock may happen >>>>>>>> between kworker and checkpoint as below: >>>>>>>> >>>>>>>> [T:writeback] [T:checkpoint] >>>>>>>> - wb_writeback >>>>>>>> - blk_start_plug >>>>>>>> bio contains NodeA was plugged in writeback threads >>>>>>> >>>>>>> I'm still trying to understand more precisely. So, how is it possible to >>>>>>> have bio having node write in this current context? >>>>>> >>>>>> IMO, after above blk_start_plug(), it may plug some inode's node page in kworker >>>>>> during writebacking node_inode's data page (which should be node page)? >>>>> >>>>> Wasn't that added into a different task->plug? >>>> >>>> I'm not sure I've got your concern correctly... >>>> >>>> Do you mean NodeA and other IOs from do_writepages() were plugged in >>>> different local plug variables? >>> >>> I think so. >> >> I guess block plug helper says it doesn't allow to use nested plug, so there >> is only one plug in kworker thread? >> >> void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) >> { >> struct task_struct *tsk = current; >> >> /* >> * If this is a nested plug, don't actually assign it. >> */ >> if (tsk->plug) >> return; >> ... >> } > > Any further comments? > > Thanks, > >> >> Thanks, >> >>> >>>> >>>> Thanks, >>>> >>>>> >>>>>> >>>>>> Thanks, >>>>>> >>>>>>> >>>>>>>> - do_writepages -- sync write inodeB, inc wb_sync_req[DATA] >>>>>>>> - f2fs_write_data_pages >>>>>>>> - f2fs_write_single_data_page -- write last dirty page >>>>>>>> - f2fs_do_write_data_page >>>>>>>> - set_page_writeback -- clear page dirty flag and >>>>>>>> PAGECACHE_TAG_DIRTY tag in radix tree >>>>>>>> - f2fs_outplace_write_data >>>>>>>> - f2fs_update_data_blkaddr >>>>>>>> - f2fs_wait_on_page_writeback -- wait NodeA to writeback here >>>>>>>> - inode_dec_dirty_pages >>>>>>>> - writeback_sb_inodes >>>>>>>> - writeback_single_inode >>>>>>>> - do_writepages >>>>>>>> - f2fs_write_data_pages -- skip writepages due to wb_sync_req[DATA] >>>>>>>> - wbc->pages_skipped += get_dirty_pages() -- PAGECACHE_TAG_DIRTY is not set but get_dirty_pages() returns one >>>>>>>> - requeue_inode -- requeue inode to wb->b_dirty queue due to non-zero.pages_skipped >>>>>>>> - blk_finish_plug >>>>>>>> >>>>>>>> Let's try to avoid deadlock condition by forcing unplugging previous bio via >>>>>>>> blk_finish_plug(current->plug) once we'v skipped writeback in writepages() >>>>>>>> due to valid sbi->wb_sync_req[DATA/NODE]. >>>>>>>> >>>>>>>> Fixes: 687de7f1010c ("f2fs: avoid IO split due to mixed WB_SYNC_ALL and WB_SYNC_NONE") >>>>>>>> Signed-off-by: Zhiguo Niu <zhiguo.niu@unisoc.com> >>>>>>>> Signed-off-by: Jing Xia <jing.xia@unisoc.com> >>>>>>>> Signed-off-by: Chao Yu <chao@kernel.org> >>>>>>>> --- >>>>>>>> fs/f2fs/data.c | 6 +++++- >>>>>>>> fs/f2fs/node.c | 6 +++++- >>>>>>>> 2 files changed, 10 insertions(+), 2 deletions(-) >>>>>>>> >>>>>>>> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c >>>>>>>> index 76d6fe7b0c8f..932a4c81acaf 100644 >>>>>>>> --- a/fs/f2fs/data.c >>>>>>>> +++ b/fs/f2fs/data.c >>>>>>>> @@ -3174,8 +3174,12 @@ static int __f2fs_write_data_pages(struct address_space *mapping, >>>>>>>> /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ >>>>>>>> if (wbc->sync_mode == WB_SYNC_ALL) >>>>>>>> atomic_inc(&sbi->wb_sync_req[DATA]); >>>>>>>> - else if (atomic_read(&sbi->wb_sync_req[DATA])) >>>>>>>> + else if (atomic_read(&sbi->wb_sync_req[DATA])) { >>>>>>>> + /* to avoid potential deadlock */ >>>>>>>> + if (current->plug) >>>>>>>> + blk_finish_plug(current->plug); >>>>>>>> goto skip_write; >>>>>>>> + } >>>>>>>> if (__should_serialize_io(inode, wbc)) { >>>>>>>> mutex_lock(&sbi->writepages); >>>>>>>> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c >>>>>>>> index 556fcd8457f3..69c6bcaf5aae 100644 >>>>>>>> --- a/fs/f2fs/node.c >>>>>>>> +++ b/fs/f2fs/node.c >>>>>>>> @@ -2106,8 +2106,12 @@ static int f2fs_write_node_pages(struct address_space *mapping, >>>>>>>> if (wbc->sync_mode == WB_SYNC_ALL) >>>>>>>> atomic_inc(&sbi->wb_sync_req[NODE]); >>>>>>>> - else if (atomic_read(&sbi->wb_sync_req[NODE])) >>>>>>>> + else if (atomic_read(&sbi->wb_sync_req[NODE])) { >>>>>>>> + /* to avoid potential deadlock */ >>>>>>>> + if (current->plug) >>>>>>>> + blk_finish_plug(current->plug); >>>>>>>> goto skip_write; >>>>>>>> + } >>>>>>>> trace_f2fs_writepages(mapping->host, wbc, NODE); >>>>>>>> -- >>>>>>>> 2.32.0 >> >> >> _______________________________________________ >> Linux-f2fs-devel mailing list >> Linux-f2fs-devel@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel > > > _______________________________________________ > Linux-f2fs-devel mailing list > Linux-f2fs-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
| |