On Nov 11, 2019 / 11:27, Chao Yu wrote:
> On 2019/11/8 12:27, Shinichiro Kawasaki wrote:
> > On Nov 05, 2019 / 20:22, Chao Yu wrote:
> >> On 2019/10/28 14:58, Shin'ichiro Kawasaki wrote:
> >>> To catch f2fs bugs in write pointer handling code for zoned block
> >>> devices, check write pointers of non-open zones that current segments do
> >>> not point to. Do this check at mount time, after the fsync data recovery
> >>> and current segments' write pointer consistency fix. Check two items
> >>> comparing write pointers with valid block maps in SIT.
> >>>
> >>> The first item is check for zones with no valid blocks. When there is no
> >>> valid blocks in a zone, the write pointer should be at the start of the
> >>> zone. If not, next write operation to the zone will cause unaligned write
> >>> error. If write pointer is not at the zone start, make mount fail and ask
> >>> users to run fsck.
> >>>
> >>> The second item is check between the write pointer position and the last
> >>> valid block in the zone. It is unexpected that the last valid block
> >>> position is beyond the write pointer. In such a case, report as the bug.
> >>> Fix is not required for such zone, because the zone is not selected for
> >>> next write operation until the zone get discarded.
> >>>
> >>> Also move a constant F2FS_REPORT_ZONE from super.c to f2fs.h to use it
> >>> in segment.c also.
> >>>
> >>> Signed-off-by: Shin'ichiro Kawasaki <[email protected]>
> >>> ---
> >>> fs/f2fs/f2fs.h | 3 +
> >>> fs/f2fs/segment.c | 147 ++++++++++++++++++++++++++++++++++++++++++++++
> >>> fs/f2fs/super.c | 11 ++--
> >>> 3 files changed, 157 insertions(+), 4 deletions(-)
> >>>
> >>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> >>> index 0216282c5b80..e8524be17852 100644
> >>> --- a/fs/f2fs/f2fs.h
> >>> +++ b/fs/f2fs/f2fs.h
> >>> @@ -3137,6 +3137,7 @@ int f2fs_lookup_journal_in_cursum(struct
> >>> f2fs_journal *journal, int type,
> >>> unsigned int val, int alloc);
> >>> void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control
> >>> *cpc);
> >>> int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi, bool
> >>> check_only);
> >>> +int f2fs_check_write_pointer(struct f2fs_sb_info *sbi);
> >>> int f2fs_build_segment_manager(struct f2fs_sb_info *sbi);
> >>> void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi);
> >>> int __init f2fs_create_segment_manager_caches(void);
> >>> @@ -3610,6 +3611,8 @@ static inline bool f2fs_blkz_is_seq(struct
> >>> f2fs_sb_info *sbi, int devi,
> >>>
> >>> return test_bit(zno, FDEV(devi).blkz_seq);
> >>> }
> >>> +
> >>> +#define F2FS_REPORT_NR_ZONES 4096
> >>> #endif
> >>>
> >>> static inline bool f2fs_hw_should_discard(struct f2fs_sb_info *sbi)
> >>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> >>> index 2b6e637dd6d3..20ef5b3705e1 100644
> >>> --- a/fs/f2fs/segment.c
> >>> +++ b/fs/f2fs/segment.c
> >>> @@ -4333,6 +4333,131 @@ static int sanity_check_curseg(struct
> >>> f2fs_sb_info *sbi)
> >>>
> >>> #ifdef CONFIG_BLK_DEV_ZONED
> >>>
> >>> +static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
> >>> + struct f2fs_dev_info *fdev,
> >>> + struct blk_zone *zone)
> >>> +{
> >>> + unsigned int s, wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
> >>> + block_t zone_block, wp_block, last_valid_block, b;
> >>> + unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
> >>> + int i;
> >>> + struct seg_entry *se;
> >>> +
> >>> + wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
> >>> + wp_segno = GET_SEGNO(sbi, wp_block);
> >>> + wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
> >>> + zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
> >>> + zone_segno = GET_SEGNO(sbi, zone_block);
> >>> + zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
> >>> +
> >>> + if (zone_segno >= MAIN_SEGS(sbi))
> >>> + return 0;
> >>> +
> >>> + /*
> >>> + * If a curseg points to the zone, skip check because the zone
> >>> + * may have fsync data that valid block map does not mark.
> >>
> >> None-curseg zone may also contain fsynced data as well? Maybe we can only
> >> verify
> >> on clean image or recovered image?
> >
> > Right. This function for none-curseg zones should be called after fsync data
>
> If so, any place to check recovery is done? You know, user can choose to skip
> recovery by using disable_roll_forward/norecovery mount option.
Ah, disable_roll_forward/norecovery mount option needs some care. The patch I
posted did not care it well enough.
When disable_roll_forward/norecovery mount option is set, I think the function
for none-curseg zones should be called only if there is no fsync data. If fsync
data exists, mount fails regardless of write pointer status.
I will modify the function calling step and conditions in the next patch.
>
> > recovery. I think my comment above is confusing. The point is that this
> > function is for none-curseg zones, and other function covers check for
> > curseg
> > zones. Let me revise the comment as follows:
> >
> > Skip check of zones cursegs point to, since fix_curseg_write_pointer()
> > checks them.
> >
> >>
> >>> + */
> >>> + for (i = 0; i < NO_CHECK_TYPE; i++)
> >>> + if (zone_secno == GET_SEC_FROM_SEG(sbi,
> >>> + CURSEG_I(sbi, i)->segno))
> >>> + return 0;
> >>> +
> >>> + /*
> >>> + * Get last valid block of the zone.
> >>> + */
> >>> + last_valid_block = zone_block - 1;
> >>> + for (s = 0; s < sbi->segs_per_sec; s++) {
> >>> + segno = zone_segno + s;
> >>> + se = get_seg_entry(sbi, segno);
> >>> + for (b = 0; b < sbi->blocks_per_seg; b++)
> >>> + if (f2fs_test_bit(b, se->cur_valid_map))
> >>> + last_valid_block = START_BLOCK(sbi, segno) + b;
> >>> + }
> >>
> >> We search bitmap table reversedly.
> >
> > Yes, will reverse the loops in the next post.
> >
> >>
> >>> +
> >>> + /*
> >>> + * If last valid block is beyond the write pointer, report the
> >>> + * inconsistency. This inconsistency does not cause write error
> >>> + * because the zone will not be selected for write operation until
> >>> + * it get discarded. Just report it.
> >>> + */
> >>> + if (last_valid_block >= wp_block) {
> >>> + f2fs_notice(sbi, "Valid block beyond write pointer: "
> >>> + "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
> >>> + GET_SEGNO(sbi, last_valid_block),
> >>> + GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
> >>> + wp_segno, wp_blkoff);
> >>> + return 0;
> >>> + }
> >>> +
> >>> + /*
> >>> + * If there is no valid block in the zone and if write pointer is
> >>> + * not at zone start, report the error to run fsck and mark the
> >>> + * zone as used.
> >>> + */
> >>> + if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
> >>> + f2fs_notice(sbi,
> >>> + "Zone without valid block has non-zero write "
> >>> + "pointer, run fsck to fix: wp[0x%x,0x%x]",
> >>> + wp_segno, wp_blkoff);
> >>> + __set_inuse(sbi, zone_segno);
> >>
> >> Why do we need to set it inused? if this is necessary, we need to call this
> >> under free_i->segmap_lock.
> >
> > I once thought that I need to set inconsistent zones in use, because I
> > observed
> > that write operation happened after zone consistency check failure (in
> > fill_super() after free_meta label). It caused unaligned writer error. To
> > avoid
> > it, I added __set_inuse() to keep inconsistent zones not selected for the
> > write
> > target.
> >
> > But that write operation happened because the write pointer fix curseg was
> > done
> > out of the SBI_POR_DOING protection. Now I learned SBI_POR_DOING can protect
> > write operation, and I don't think set in use for the inconsistent zones is
> > required. Will remove __set_inuse() calls from this patch and the first
> > patch.
>
> Also f2fs_stop_checkpoint() will stop any data/node/meta writeback, so it'd be
> safe here.
Ok, I note it. Now I'm looking at the code again, and I think the write pointer
fix for cursegs can be done within SBI_POR_DOING protection. At this moment, I
think f2fs_stop_checkpoint() is not necessary.
Thanks!
--
Best Regards,
Shin'ichiro Kawasaki
_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
