A quick comment: if cp.user_block_count is corrupted due to resize failure, then cp.free_segment_count will also be corrupted. Then in fsck_verify(), we will find the check "free segment_count matched with CP" fails, and still call fix_checkpoint() later.
Thanks, ________________________________________ From: Yuchao (T) Sent: Wednesday, July 03, 2019 14:29 To: sunqiuyang; [email protected]; [email protected] Cc: fangwei (I) Subject: Re: [PATCH 1/1] fsck.f2fs: allow to fix inconsistency from online resize On 2019/7/2 15:42, sunqiuyang wrote: > From: Qiuyang Sun <[email protected]> > > During F2FS online resize, if panic or poweroff happens when the new SB > has been committed but not the new CP, the FS may end up with an > inconsistent state, where user_block_count and free_segment_count in CP > can be larger than allowed by the new SB. > > This patch allows fsck.f2fs to fix such inconsistency. > > Signed-off-by: Qiuyang Sun <[email protected]> > --- > fsck/fsck.c | 2 ++ > fsck/mount.c | 2 +- > 2 files changed, 3 insertions(+), 1 deletion(-) > > diff --git a/fsck/fsck.c b/fsck/fsck.c > index 6f0f262..66eb53c 100644 > --- a/fsck/fsck.c > +++ b/fsck/fsck.c > @@ -2083,6 +2083,8 @@ static void fix_checkpoint(struct f2fs_sb_info *sbi) > set_cp(valid_block_count, fsck->chk.valid_blk_cnt); > set_cp(valid_node_count, fsck->chk.valid_node_cnt); > set_cp(valid_inode_count, fsck->chk.valid_inode_cnt); > + set_cp(user_block_count, (uint64_t)(get_sb(segment_count_main) - > + get_cp(overprov_segment_count)) * c.blks_per_seg); > > crc = f2fs_checkpoint_chksum(cp); > *((__le32 *)((unsigned char *)cp + get_cp(checksum_offset))) = > diff --git a/fsck/mount.c b/fsck/mount.c > index aecd0cd..81ab660 100644 > --- a/fsck/mount.c > +++ b/fsck/mount.c > @@ -953,7 +953,7 @@ int sanity_check_ckpt(struct f2fs_sb_info *sbi) > segment_count_main = get_sb(segment_count_main); > log_blocks_per_seg = get_sb(log_blocks_per_seg); > if (!user_block_count || user_block_count >= > - segment_count_main << log_blocks_per_seg) { > + c.total_sectors / c.sectors_per_blk) { This breaks the previous consistency check, then cp.user_block_count could still be corrupted due to we won't call fix_checkpoint later. Other concern is if user upgrade kernel only, with low version of fsck, user can't repair image suffering power-cut during online resizing, so I wonder that is there a way to find such inconsistency and give a hint to user that which version of fsck can repair such corruption. Thanks, > MSG(0, "\tWrong user_block_count(%u)\n", user_block_count); > return 1; > } > _______________________________________________ Linux-f2fs-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
