[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 10 Oct 2019 16:39:24 +1100
From: Matthew Bobrowski <mbobrowski@...browski.org>
To: Jan Kara <jack@...e.cz>
Cc: tytso@....edu, adilger.kernel@...ger.ca,
linux-ext4@...r.kernel.org, linux-fsdevel@...r.kernel.org,
hch@...radead.org, david@...morbit.com, darrick.wong@...cle.com
Subject: Re: [PATCH v4 1/8] ext4: move out iomap field population into
separate helper
On Wed, Oct 09, 2019 at 03:06:09PM +0200, Jan Kara wrote:
> On Wed 09-10-19 19:57:23, Matthew Bobrowski wrote:
> > On Tue, Oct 08, 2019 at 12:27:09PM +0200, Jan Kara wrote:
> > > On Thu 03-10-19 21:33:09, Matthew Bobrowski wrote:
> > > > +static int ext4_set_iomap(struct inode *inode, struct iomap *iomap, u16 type,
> > > > + unsigned long first_block, struct ext4_map_blocks *map)
> > > > +{
> > > > + u8 blkbits = inode->i_blkbits;
> > > > +
> > > > + iomap->flags = 0;
> > > > + if (ext4_inode_datasync_dirty(inode))
> > > > + iomap->flags |= IOMAP_F_DIRTY;
> > > > + iomap->bdev = inode->i_sb->s_bdev;
> > > > + iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
> > > > + iomap->offset = (u64) first_block << blkbits;
> > > > + iomap->length = (u64) map->m_len << blkbits;
> > > > +
> > > > + if (type) {
> > > > + iomap->type = type;
> > > > + iomap->addr = IOMAP_NULL_ADDR;
> > > > + } else {
> > > > + if (map->m_flags & EXT4_MAP_MAPPED) {
> > > > + iomap->type = IOMAP_MAPPED;
> > > > + } else if (map->m_flags & EXT4_MAP_UNWRITTEN) {
> > > > + iomap->type = IOMAP_UNWRITTEN;
> > > > + } else {
> > > > + WARN_ON_ONCE(1);
> > > > + return -EIO;
> > > > + }
> > > > + iomap->addr = (u64) map->m_pblk << blkbits;
> > > > + }
> > >
> > > Looking at this function now, the 'type' argument looks a bit weird. Can we
> > > perhaps just remove the 'type' argument and change the above to:
> >
> > We can, but refer to the point below.
> >
> > > if (map->m_flags & (EXT4_MAP_MAPPED | EXT4_MAP_UNWRITTEN)) {
> > > if (map->m_flags & EXT4_MAP_MAPPED)
> > > iomap->type = IOMAP_MAPPED;
> > > else if (map->m_flags & EXT4_MAP_UNWRITTEN)
> > > iomap->type = IOMAP_UNWRITTEN;
> > > iomap->addr = (u64) map->m_pblk << blkbits;
> > > } else {
> > > iomap->type = IOMAP_HOLE;
> > > iomap->addr = IOMAP_NULL_ADDR;
> > > }
> > >
> > > And then in ext4_iomap_begin() we overwrite the type to:
> > >
> > > if (delalloc && iomap->type == IOMAP_HOLE)
> > > iomap->type = IOMAP_DELALLOC;
> > >
> > > That would IMO make ext4_set_iomap() arguments harder to get wrong.
> >
> > I was thinking about this while doing a bunch of other things at work
> > today. I'm kind of aligned with what Christoph mentioned around
> > possibly duplicating some of the post 'iomap->type' setting from both
> > current and any future ext4_set_iomap() callers. In addition to this,
> > my thought was that if we're populating the iomap structure with
> > values respectively, then it would make most sense to encapsulate
> > those routines, if possible, within the ext4_set_iomap() as that's the
> > sole purpose of the function.
>
> Well, what I dislike about 'type' argument is the inconsistency in it's
> handling. It is useful only for HOLE/DELALLOC, anything else will just give
> you invalid iomap and you have to be careful to pass 0 in that case.
>
> I understand the concern about possible duplication but since only
> IOMAP_REPORT cares about IOMAP_DELALLOC, I'm not much concerned about it.
> But another sensible API would be to optionally pass 'struct extent_status
> *es' argument to ext4_set_iomap() and if this argument is non-NULL,
> ext4_set_iomap() will handle the intersection of ext4_map_blocks and
> extent_status and deduce appropriate resulting type from that.
I see and thank you for sharing your view point. Let's go with what you
originally proposed, which is dropping the 'type' argument and then handling
IOMAP_DELALLOC within ext4_set_iomap(). No real hard objections. :)
--<M>--
Powered by blists - more mailing lists