Re: [PATCH] f2fs: fix long latency due to discard during umount

[Sahitya Tummala]

Hi Chao,

On Thu, Mar 26, 2020 at 05:00:18PM +0800, Chao Yu wrote:
> Hi Sahitya,
> 
> On 2020/3/18 12:44, Sahitya Tummala wrote:
> > F2FS already has a default timeout of 5 secs for discards that
> > can be issued during umount, but it can take more than the 5 sec
> > timeout if the underlying UFS device queue is already full and there
> > are no more available free tags to be used. In that case, submit_bio()
> > will wait for the already queued discard requests to complete to get
> > a free tag, which can potentially take way more than 5 sec.
> > 
> > Fix this by submitting the discard requests with REQ_NOWAIT
> > flags during umount. This will return -EAGAIN for UFS queue/tag full
> > scenario without waiting in the context of submit_bio(). The FS can
> > then handle these requests by retrying again within the stipulated
> > discard timeout period to avoid long latencies.
> > 
> > Signed-off-by: Sahitya Tummala <stummala@xxxxxxxxxxxxxx>
> > ---
> > v2:
> > - Handle the case where a dc can have multiple bios associated with it
> > 
> >  fs/f2fs/f2fs.h    |  1 +
> >  fs/f2fs/segment.c | 83 ++++++++++++++++++++++++++++++++++++++++++++++++-------
> >  2 files changed, 74 insertions(+), 10 deletions(-)
> > 
> > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > index 12a197e..67b8dcc 100644
> > --- a/fs/f2fs/f2fs.h
> > +++ b/fs/f2fs/f2fs.h
> > @@ -340,6 +340,7 @@ struct discard_cmd_control {
> >  	struct list_head pend_list[MAX_PLIST_NUM];/* store pending entries */
> >  	struct list_head wait_list;		/* store on-flushing entries */
> >  	struct list_head fstrim_list;		/* in-flight discard from fstrim */
> > +	struct list_head retry_list;		/* list of cmds to retry */
> >  	wait_queue_head_t discard_wait_queue;	/* waiting queue for wake-up */
> >  	unsigned int discard_wake;		/* to wake up discard thread */
> >  	struct mutex cmd_lock;
> > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > index fb3e531..4162c76 100644
> > --- a/fs/f2fs/segment.c
> > +++ b/fs/f2fs/segment.c
> > @@ -1029,13 +1029,16 @@ static void f2fs_submit_discard_endio(struct bio *bio)
> >  	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
> >  	unsigned long flags;
> >  
> > -	dc->error = blk_status_to_errno(bio->bi_status);
> > -
> >  	spin_lock_irqsave(&dc->lock, flags);
> > +	if (!dc->error)
> > +		dc->error = blk_status_to_errno(bio->bi_status);
> > +
> >  	dc->bio_ref--;
> > -	if (!dc->bio_ref && dc->state == D_SUBMIT) {
> > -		dc->state = D_DONE;
> > -		complete_all(&dc->wait);
> > +	if (!dc->bio_ref) {
> > +		if (dc->error || dc->state == D_SUBMIT) {
> > +			dc->state = D_DONE;
> > +			complete_all(&dc->wait);
> > +		}
> >  	}
> >  	spin_unlock_irqrestore(&dc->lock, flags);
> >  	bio_put(bio);
> > @@ -1124,10 +1127,13 @@ static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> >  	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
> >  	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
> >  					&(dcc->fstrim_list) : &(dcc->wait_list);
> > -	int flag = dpolicy->sync ? REQ_SYNC : 0;
> > -	block_t lstart, start, len, total_len;
> > +	int flag;
> > +	block_t lstart, start, len, total_len, orig_len;
> >  	int err = 0;
> >  
> > +	flag = dpolicy->sync ? REQ_SYNC : 0;
> > +	flag |= dpolicy->type == DPOLICY_UMOUNT ? REQ_NOWAIT : 0;
> > +
> >  	if (dc->state != D_PREP)
> >  		return 0;
> >  
> > @@ -1139,7 +1145,7 @@ static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> >  	lstart = dc->lstart;
> >  	start = dc->start;
> >  	len = dc->len;
> > -	total_len = len;
> > +	orig_len = total_len = len;
> >  
> >  	dc->len = 0;
> >  
> > @@ -1203,6 +1209,14 @@ static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> >  		bio->bi_end_io = f2fs_submit_discard_endio;
> >  		bio->bi_opf |= flag;
> >  		submit_bio(bio);
> > +		if (flag & REQ_NOWAIT) {
> > +			if (dc->error == -EAGAIN) {
> > +				dc->len = orig_len;
> > +				list_move(&dc->list, &dcc->retry_list);
> > +				err = dc->error;
> 
> I encounter lots of dmesg, which should be printed by __remove_discard_cmd()
> 
> F2FS-fs (dm-0): Issue discard(23552, 23552, 2) failed, ret: -11
> 
> This should happen only if we didn't handle all discard in 5 seconds during
> umount.
> 
> So I doubt we failed to move dc to retry_list, because after submit_bio(),
> end_io() is not called synchronously as the bio was just pluged?
> 
This can happen if a discard cmd has multiple bios and at least 1 bio is 
already submitted and when submitting other bios, we encounter -EAGAIN.
In this case, this dc will be moved to retry_list and will be moved back
to dcc->pend_list only if the dc->bio_ref becomes 0 within 5 sec timeout.
If it doesn't become zero, then it will be left in retry_list itself, which
will be later removed from retry_list. Before removing from retry_list we 
will however ensure that submitted bio is done i.e., dc->bio_ref is 0, but
dc->error will be -EAGAIN as this dc could not be requeued/retried.

So this is expected, where it only means that this dc could not be
submitted/retried again within timeout. I think we can ignore
this -EAGAIN error in __remove_discard_cmd().

Thanks,

> Thanks,
> 
> > +				break;
> > +			}
> > +		}
> >  
> >  		atomic_inc(&dcc->issued_discard);
> >  
> > @@ -1463,6 +1477,40 @@ static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
> >  	return issued;
> >  }
> >  
> > +static bool __should_discard_retry(struct f2fs_sb_info *sbi,
s> > +		struct discard_policy *dpolicy)
> > +{
> > +	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
> > +	struct discard_cmd *dc, *tmp;
> > +	bool retry = false;
> > +	unsigned long flags;
> > +
> > +	if (dpolicy->type != DPOLICY_UMOUNT)
> > +		f2fs_bug_on(sbi, 1);
> > +
> > +	mutex_lock(&dcc->cmd_lock);
> > +	list_for_each_entry_safe(dc, tmp, &(dcc->retry_list), list) {
> > +		if (dpolicy->timeout != 0 &&
> > +			f2fs_time_over(sbi, dpolicy->timeout)) {
> > +			retry = false;
> > +			break;
> > +		}
> > +
> > +		spin_lock_irqsave(&dc->lock, flags);
> > +		if (!dc->bio_ref) {
> > +			dc->state = D_PREP;
> > +			dc->error = 0;
> > +			reinit_completion(&dc->wait);
> > +			__relocate_discard_cmd(dcc, dc);
> > +			retry = true;
> > +		}
> > +		spin_unlock_irqrestore(&dc->lock, flags);
> > +	}
> > +	mutex_unlock(&dcc->cmd_lock);
> > +
> > +	return retry;
> > +}
> > +
> >  static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
> >  					struct discard_policy *dpolicy)
> >  {
> > @@ -1470,12 +1518,13 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
> >  	struct list_head *pend_list;
> >  	struct discard_cmd *dc, *tmp;
> >  	struct blk_plug plug;
> > -	int i, issued = 0;
> > +	int i, err, issued = 0;
> >  	bool io_interrupted = false;
> >  
> >  	if (dpolicy->timeout != 0)
> >  		f2fs_update_time(sbi, dpolicy->timeout);
> >  
> > +retry:
> >  	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
> >  		if (dpolicy->timeout != 0 &&
> >  				f2fs_time_over(sbi, dpolicy->timeout))
> > @@ -1509,7 +1558,10 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
> >  				break;
> >  			}
> >  
> > -			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
> > +			err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
> > +			if (err == -EAGAIN)
> > +				congestion_wait(BLK_RW_ASYNC,
> > +						DEFAULT_IO_TIMEOUT);
> >  
> >  			if (issued >= dpolicy->max_requests)
> >  				break;
> > @@ -1522,6 +1574,10 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
> >  			break;
> >  	}
> >  
> > +	if (!list_empty(&dcc->retry_list) &&
> > +		__should_discard_retry(sbi, dpolicy))
> > +		goto retry;
> > +
> >  	if (!issued && io_interrupted)
> >  		issued = -1;
> >  
> > @@ -1613,6 +1669,12 @@ static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
> >  		goto next;
> >  	}
> >  
> > +	if (dpolicy->type == DPOLICY_UMOUNT &&
> > +		!list_empty(&dcc->retry_list)) {
> > +		wait_list = &dcc->retry_list;
> > +		goto next;
> > +	}
> > +
> >  	return trimmed;
> >  }
> >  
> > @@ -2051,6 +2113,7 @@ static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
> >  	for (i = 0; i < MAX_PLIST_NUM; i++)
> >  		INIT_LIST_HEAD(&dcc->pend_list[i]);
> >  	INIT_LIST_HEAD(&dcc->wait_list);
> > +	INIT_LIST_HEAD(&dcc->retry_list);
> >  	INIT_LIST_HEAD(&dcc->fstrim_list);
> >  	mutex_init(&dcc->cmd_lock);
> >  	atomic_set(&dcc->issued_discard, 0);
> > 

-- 
--
Sent by a consultant of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.