Re: [PATCH] quota: don't let mark_dquot_dirty() fail silently

[Chao Yu]

On 2024/4/8 22:30, Jan Kara wrote:
> On Sun 07-04-24 15:31:28, Chao Yu wrote:
> > mark_dquot_dirty() will callback to specified filesystem function,
> > it may fail due to any reasons, however, no caller will check return
> > value of mark_dquot_dirty(), so, it may fail silently, let's print
> > one line message for such case.
> >
> > Signed-off-by: Chao Yu <chao@xxxxxxxxxx>
> > ---
> >   fs/quota/dquot.c | 23 +++++++++++++----------
> >   1 file changed, 13 insertions(+), 10 deletions(-)
> >
> > diff --git a/fs/quota/dquot.c b/fs/quota/dquot.c
> > index dacbee455c03..c5df7863942a 100644
> > --- a/fs/quota/dquot.c
> > +++ b/fs/quota/dquot.c
> > @@ -399,21 +399,20 @@ int dquot_mark_dquot_dirty(struct dquot *dquot)
> >   EXPORT_SYMBOL(dquot_mark_dquot_dirty);
> >   
> >   /* Dirtify all the dquots - this can block when journalling */
> > -static inline int mark_all_dquot_dirty(struct dquot __rcu * const *dquots)
> > +static inline void mark_all_dquot_dirty(struct dquot __rcu * const *dquots)
> >   {
> > -	int ret, err, cnt;
> > +	int ret, cnt;
> >   	struct dquot *dquot;
> >   
> > -	ret = err = 0;
> >   	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
> >   		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
> > -		if (dquot)
> > -			/* Even in case of error we have to continue */
> > -			ret = mark_dquot_dirty(dquot);
> > -		if (!err)
> > -			err = ret;
> > +		if (!dquot)
> > +			continue;
> > +		ret = mark_dquot_dirty(dquot);
> > +		if (ret < 0)
> > +			quota_error(dquot->dq_sb,
> > +				"mark_all_dquot_dirty fails, ret: %d", ret);
>
> Do you have any practical case you care about? Because in practice the

Actually, no.

> filesystem will usually report if there's some catastrophic error (and the
> errors from ->mark_dirty() all mean the filesystem is in unhealthy state).
> So this message just adds to the noise in the error log - and e.g. if the
> disk goes bad so we cannot write, we could spew a lot of messages like
> this.

Agreed,

I guess we can propagate the error to caller rather than printing redundant
message in log.

> >   	}
> > -	return err;
> >   }
> >   
> >   static inline void dqput_all(struct dquot **dquot)
> > @@ -2725,6 +2724,7 @@ static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
> >   {
> >   	struct mem_dqblk *dm = &dquot->dq_dqb;
> >   	int check_blim = 0, check_ilim = 0;
> > +	int ret;
> >   	struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
> >   
> >   	if (di->d_fieldmask & ~VFS_QC_MASK)
> > @@ -2807,7 +2807,10 @@ static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
> >   	else
> >   		set_bit(DQ_FAKE_B, &dquot->dq_flags);
> >   	spin_unlock(&dquot->dq_dqb_lock);
> > -	mark_dquot_dirty(dquot);
> > +	ret = mark_dquot_dirty(dquot);
> > +	if (ret < 0)
> > +		quota_error(dquot->dq_sb,
> > +			"mark_dquot_dirty fails, ret: %d", ret);
>
> Here, we can propagate the error back to userspace, which is probably
> better than spamming the logs.

Yes, let me submit a new patch for this.

Thanks,

> 								Honza