On 11.12.18 г. 10:23 ч., Nikolay Borisov wrote: > There is a function which clearly conveys the objective of checking > i_writecount. Additionally the usage in ext4_mb_initialize_context was > wrong, since a node would have wrongfully been reported as writable if > i_writecount had a negative value (MMAP_DENY_WRITE). > > Reviewed-by: Jan Kara <jack@xxxxxxx> > Signed-off-by: Nikolay Borisov <nborisov@xxxxxxxx> Yet another ping, if something is wrong with the patch I'm happy to fix it and get it merged. > --- > fs/ext4/inode.c | 2 +- > fs/ext4/mballoc.c | 7 +++---- > 2 files changed, 4 insertions(+), 5 deletions(-) > > diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c > index 22a9d8159720..04e5bc0b5c8d 100644 > --- a/fs/ext4/inode.c > +++ b/fs/ext4/inode.c > @@ -391,7 +391,7 @@ void ext4_da_update_reserve_space(struct inode *inode, > * inode's preallocations. > */ > if ((ei->i_reserved_data_blocks == 0) && > - (atomic_read(&inode->i_writecount) == 0)) > + !inode_is_open_for_write(inode)) > ext4_discard_preallocations(inode); > } > > diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c > index e2248083cdca..6fb76d408093 100644 > --- a/fs/ext4/mballoc.c > +++ b/fs/ext4/mballoc.c > @@ -4176,9 +4176,8 @@ static void ext4_mb_group_or_file(struct ext4_allocation_context *ac) > isize = (i_size_read(ac->ac_inode) + ac->ac_sb->s_blocksize - 1) > >> bsbits; > > - if ((size == isize) && > - !ext4_fs_is_busy(sbi) && > - (atomic_read(&ac->ac_inode->i_writecount) == 0)) { > + if ((size == isize) && !ext4_fs_is_busy(sbi) && > + !inode_is_open_for_write(ac->ac_inode)) { > ac->ac_flags |= EXT4_MB_HINT_NOPREALLOC; > return; > } > @@ -4258,7 +4257,7 @@ ext4_mb_initialize_context(struct ext4_allocation_context *ac, > (unsigned) ar->goal, ac->ac_flags, ac->ac_2order, > (unsigned) ar->lleft, (unsigned) ar->pleft, > (unsigned) ar->lright, (unsigned) ar->pright, > - atomic_read(&ar->inode->i_writecount) ? "" : "non-"); > + inode_is_open_for_write(ar->inode) ? "" : "non-"); > return 0; > > } >