| Message ID | 20231221150558.2740823-5-libaokun1@huawei.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | ext4: fix divide error in mb_update_avg_fragment_size() | expand |
On Thu 21-12-23 23:05:54, Baokun Li wrote: > After updating bb_free in mb_free_blocks, it is possible to return without > updating bb_fragments because the block being freed is found to have > already been freed, which leads to inconsistency between bb_free and > bb_fragments. > > Since the group may be unlocked in ext4_grp_locked_error(), this can lead > to problems such as dividing by zero when calculating the average fragment > length. Hence move the update of bb_free to after the block double-free > check guarantees that the corresponding statistics are updated only after > the core block bitmap is modified. > > Fixes: eabe0444df90 ("ext4: speed-up releasing blocks on commit") > CC: stable@vger.kernel.org # 3.10 > Signed-off-by: Baokun Li <libaokun1@huawei.com> Just one nit below but regardless of that feel free to add: Reviewed-by: Jan Kara <jack@suse.cz> > @@ -1941,10 +1936,16 @@ static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b, > EXT4_GROUP_INFO_BBITMAP_CORRUPT); > } else { > mb_regenerate_buddy(e4b); > + goto check; > } > - goto done; > + return; > } I think this might be more readable when we revert the condition like: /* * Fastcommit replay can free already freed blocks which * corrupts allocation info. Regenerate it. */ if (sbi->s_mount_state & EXT4_FC_REPLAY) { mb_regenerate_buddy(e4b); goto check; } ext4_grp_locked_error(sb, e4b->bd_group, inode ? inode->i_ino : 0, blocknr, "freeing already freed block (bit %u); block bitmap corrupt.", block); ext4_mark_group_bitmap_corrupted(sb, e4b->bd_group, EXT4_GROUP_INFO_BBITMAP_CORRUPT); return; } Honza
On 2024/1/4 18:42, Jan Kara wrote: > On Thu 21-12-23 23:05:54, Baokun Li wrote: >> After updating bb_free in mb_free_blocks, it is possible to return without >> updating bb_fragments because the block being freed is found to have >> already been freed, which leads to inconsistency between bb_free and >> bb_fragments. >> >> Since the group may be unlocked in ext4_grp_locked_error(), this can lead >> to problems such as dividing by zero when calculating the average fragment >> length. Hence move the update of bb_free to after the block double-free >> check guarantees that the corresponding statistics are updated only after >> the core block bitmap is modified. >> >> Fixes: eabe0444df90 ("ext4: speed-up releasing blocks on commit") >> CC: stable@vger.kernel.org # 3.10 >> Signed-off-by: Baokun Li <libaokun1@huawei.com> > Just one nit below but regardless of that feel free to add: > > Reviewed-by: Jan Kara <jack@suse.cz> > >> @@ -1941,10 +1936,16 @@ static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b, >> EXT4_GROUP_INFO_BBITMAP_CORRUPT); >> } else { >> mb_regenerate_buddy(e4b); >> + goto check; >> } >> - goto done; >> + return; >> } > I think this might be more readable when we revert the condition like: > > /* > * Fastcommit replay can free already freed blocks which > * corrupts allocation info. Regenerate it. > */ > if (sbi->s_mount_state & EXT4_FC_REPLAY) { > mb_regenerate_buddy(e4b); > goto check; > } > ext4_grp_locked_error(sb, e4b->bd_group, > inode ? inode->i_ino : 0, blocknr, > "freeing already freed block (bit %u); block bitmap corrupt.", > block); > ext4_mark_group_bitmap_corrupted(sb, e4b->bd_group, > EXT4_GROUP_INFO_BBITMAP_CORRUPT); > return; > } > > Honza Yes, it looks much clearer that way! I will switch to it in the next version. Thanks a lot!
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c index f6131ba514c8..1f15774971d7 100644 --- a/fs/ext4/mballoc.c +++ b/fs/ext4/mballoc.c @@ -1910,11 +1910,6 @@ static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b, mb_check_buddy(e4b); mb_free_blocks_double(inode, e4b, first, count); - this_cpu_inc(discard_pa_seq); - e4b->bd_info->bb_free += count; - if (first < e4b->bd_info->bb_first_free) - e4b->bd_info->bb_first_free = first; - /* access memory sequentially: check left neighbour, * clear range and then check right neighbour */ @@ -1941,10 +1936,16 @@ static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b, EXT4_GROUP_INFO_BBITMAP_CORRUPT); } else { mb_regenerate_buddy(e4b); + goto check; } - goto done; + return; } + this_cpu_inc(discard_pa_seq); + e4b->bd_info->bb_free += count; + if (first < e4b->bd_info->bb_first_free) + e4b->bd_info->bb_first_free = first; + /* let's maintain fragments counter */ if (left_is_free && right_is_free) e4b->bd_info->bb_fragments--; @@ -1969,9 +1970,9 @@ static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b, if (first <= last) mb_buddy_mark_free(e4b, first >> 1, last >> 1); -done: mb_set_largest_free_order(sb, e4b->bd_info); mb_update_avg_fragment_size(sb, e4b->bd_info); +check: mb_check_buddy(e4b); }
After updating bb_free in mb_free_blocks, it is possible to return without updating bb_fragments because the block being freed is found to have already been freed, which leads to inconsistency between bb_free and bb_fragments. Since the group may be unlocked in ext4_grp_locked_error(), this can lead to problems such as dividing by zero when calculating the average fragment length. Hence move the update of bb_free to after the block double-free check guarantees that the corresponding statistics are updated only after the core block bitmap is modified. Fixes: eabe0444df90 ("ext4: speed-up releasing blocks on commit") CC: stable@vger.kernel.org # 3.10 Signed-off-by: Baokun Li <libaokun1@huawei.com> --- fs/ext4/mballoc.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-)