Baokun Li <libaokun1@xxxxxxxxxx> writes: > If there is a pa in the i_prealloc_list of an inode with a tmp_pa_end > of 4294967296(0x100000000), since tmp_pa_end is of type ext4_lblk_t, > tmp_pa_end will be recognized as 0 due to overflow, which causes > (ac->ac_o_ ex.fe_logical >= tmp_pa_end) always holds, so that pa will > always be skipped. This then triggers the regular allocation process, > and if the excess tail of the free extent from that allocation is put > into the i_prealloc_list again, it will again not be used. This ends up > leaving us with a lot of free physical blocks in the i_prealloc_list. > > We avoid this problem by using pa_end() to compute tmp_pa_end and > declaring tmp_pa_end to be of type loff_t. > > Signed-off-by: Baokun Li <libaokun1@xxxxxxxxxx> Looks good to me with fex_end() and pa_end() dropped. -ritesh > --- > fs/ext4/mballoc.c | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c > index 77d47af525d9..06db40fb29d6 100644 > --- a/fs/ext4/mballoc.c > +++ b/fs/ext4/mballoc.c > @@ -4765,7 +4765,8 @@ ext4_mb_use_preallocated(struct ext4_allocation_context *ac) > struct ext4_inode_info *ei = EXT4_I(ac->ac_inode); > struct ext4_locality_group *lg; > struct ext4_prealloc_space *tmp_pa, *cpa = NULL; > - ext4_lblk_t tmp_pa_start, tmp_pa_end; > + ext4_lblk_t tmp_pa_start; > + loff_t tmp_pa_end; > struct rb_node *iter; > ext4_fsblk_t goal_block; > > @@ -4784,7 +4785,7 @@ ext4_mb_use_preallocated(struct ext4_allocation_context *ac) > /* all fields in this condition don't change, > * so we can skip locking for them */ > tmp_pa_start = tmp_pa->pa_lstart; > - tmp_pa_end = tmp_pa->pa_lstart + EXT4_C2B(sbi, tmp_pa->pa_len); > + tmp_pa_end = pa_end(sbi, tmp_pa); > > /* original request start doesn't lie in this PA */ > if (ac->ac_o_ex.fe_logical < tmp_pa_start || > -- > 2.31.1
