Re: [f2fs-dev] [PATCH 2/2] f2fs: avoid f2fs_bug_on if f2fs_get_meta_page_nofail got EIO

[Jaegeuk Kim]

On 09/18, Chao Yu wrote:
> On 2018/9/18 10:18, Jaegeuk Kim wrote:
> > This patch avoids BUG_ON when f2fs_get_meta_page_nofail got EIO during
> > xfstests/generic/475.
> > 
> > Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
> > ---
> >  fs/f2fs/checkpoint.c |  2 +-
> >  fs/f2fs/gc.c         |  2 ++
> >  fs/f2fs/node.c       | 12 ++++++++++--
> >  fs/f2fs/recovery.c   |  2 ++
> >  fs/f2fs/segment.c    |  3 +++
> >  5 files changed, 18 insertions(+), 3 deletions(-)
> > 
> > diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
> > index 01e0d8f5bbbe..6ce3cb6502dd 100644
> > --- a/fs/f2fs/checkpoint.c
> > +++ b/fs/f2fs/checkpoint.c
> > @@ -121,7 +121,7 @@ struct page *f2fs_get_meta_page_nofail(struct f2fs_sb_info *sbi, pgoff_t index)
> >  			goto retry;
> >  
> >  		f2fs_stop_checkpoint(sbi, false);
> > -		f2fs_bug_on(sbi, 1);
> > +		return NULL;
> 
> How about propagate PTR_ERR(page) to caller?

Done.

> 
> >  	}
> >  
> >  	return page;
> > diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
> > index 4bcc8a59fdef..d049865887cf 100644
> > --- a/fs/f2fs/gc.c
> > +++ b/fs/f2fs/gc.c
> > @@ -1070,6 +1070,8 @@ static int do_garbage_collect(struct f2fs_sb_info *sbi,
> >  	/* reference all summary page */
> >  	while (segno < end_segno) {
> >  		sum_page = f2fs_get_sum_page(sbi, segno++);
> > +		if (!sum_page)
> > +			return -EIO;
> 
> Well, for large section, we need to release all referenced sum page by
> f2fs_put_page().

Done.

> 
> >  		unlock_page(sum_page);
> >  	}
> >  
> > diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> > index fa2381c0bc47..b3595522c35b 100644
> > --- a/fs/f2fs/node.c
> > +++ b/fs/f2fs/node.c
> > @@ -126,6 +126,8 @@ static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
> >  
> >  	/* get current nat block page with lock */
> >  	src_page = get_current_nat_page(sbi, nid);
> > +	if (!src_page)
> > +		return NULL;
> >  	dst_page = f2fs_grab_meta_page(sbi, dst_off);
> >  	f2fs_bug_on(sbi, PageDirty(src_page));
> >  
> > @@ -2265,8 +2267,12 @@ static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
> >  						nm_i->nat_block_bitmap)) {
> >  			struct page *page = get_current_nat_page(sbi, nid);
> >  
> > -			ret = scan_nat_page(sbi, page, nid);
> > -			f2fs_put_page(page, 1);
> > +			if (page) {
> > +				ret = scan_nat_page(sbi, page, nid);
> > +				f2fs_put_page(page, 1);
> > +			} else {
> > +				ret = -EIO;
> > +			}
> >  
> >  			if (ret) {
> >  				up_read(&nm_i->nat_tree_lock);
> 
> Should propagate the error to f2fs_alloc_nid()?

Done.

> 
> > @@ -2724,6 +2730,8 @@ static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
> >  		down_write(&curseg->journal_rwsem);
> >  	} else {
> >  		page = get_next_nat_page(sbi, start_nid);
> > +		if (!page)
> > +			return;
> 
> Ditto, propagate such error to write_checkpoint()?

Done.

> 
> >  		nat_blk = page_address(page);
> >  		f2fs_bug_on(sbi, !nat_blk);
> >  	}
> > diff --git a/fs/f2fs/recovery.c b/fs/f2fs/recovery.c
> > index 56d34193a74b..a3dce16bfd6c 100644
> > --- a/fs/f2fs/recovery.c
> > +++ b/fs/f2fs/recovery.c
> > @@ -355,6 +355,8 @@ static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
> >  	}
> >  
> >  	sum_page = f2fs_get_sum_page(sbi, segno);
> > +	if (!sum_page)
> > +		return -EIO;
> >  	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
> >  	sum = sum_node->entries[blkoff];
> >  	f2fs_put_page(sum_page, 1);
> > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > index aa96a371aaf8..cfc9eb492da1 100644
> > --- a/fs/f2fs/segment.c
> > +++ b/fs/f2fs/segment.c
> > @@ -2487,6 +2487,7 @@ static void change_curseg(struct f2fs_sb_info *sbi, int type)
> >  	__next_free_blkoff(sbi, curseg, 0);
> >  
> >  	sum_page = f2fs_get_sum_page(sbi, new_segno);
> > +	f2fs_bug_on(sbi, !sum_page);
> 
> Well, next time we may panic here...

This is the same as before, and it's almost impossible to hit anyway in
production.

Thanks,

> 
> In product, for EIO case, usually we just reboot cell phone directly to avoid
> potential data loss later.
> 
> So I just set DEFAULT_RETRY_IO_COUNT to 32 temporarily to pass xfstest IO error
> injection cases.
> 
> Thanks,
> 
> >  	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
> >  	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
> >  	f2fs_put_page(sum_page, 1);
> > @@ -3971,6 +3972,8 @@ static int build_sit_entries(struct f2fs_sb_info *sbi)
> >  
> >  			se = &sit_i->sentries[start];
> >  			page = get_current_sit_page(sbi, start);
> > +			if (!page)
> > +				return err;
> >  			sit_blk = (struct f2fs_sit_block *)page_address(page);
> >  			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
> >  			f2fs_put_page(page, 1);
> >