On Wed, Sep 02, 2020 at 09:23:36AM -0400, Brian Foster wrote:
> On Tue, Sep 01, 2020 at 07:55:58PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
> >
> > Add a btree block usage counters for both inode btrees to the AGI header
> > so that we don't have to walk the entire finobt at mount time to create
> > the per-AG reservations.
> >
> > Signed-off-by: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
> > ---
>
> A couple nits..
>
> > fs/xfs/libxfs/xfs_ag.c | 5 +++++
> > fs/xfs/libxfs/xfs_format.h | 18 +++++++++++++++++-
> > fs/xfs/libxfs/xfs_ialloc.c | 1 +
> > fs/xfs/libxfs/xfs_ialloc_btree.c | 24 ++++++++++++++++++++++++
> > fs/xfs/xfs_ondisk.h | 2 +-
> > fs/xfs/xfs_super.c | 4 ++++
> > 6 files changed, 52 insertions(+), 2 deletions(-)
> >
> >
> > diff --git a/fs/xfs/libxfs/xfs_ag.c b/fs/xfs/libxfs/xfs_ag.c
> > index 8cf73fe4338e..9331f3516afa 100644
> > --- a/fs/xfs/libxfs/xfs_ag.c
> > +++ b/fs/xfs/libxfs/xfs_ag.c
> > @@ -333,6 +333,11 @@ xfs_agiblock_init(
> > }
> > for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
> > agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
> > + if (xfs_sb_version_hasinobtcounts(&mp->m_sb)) {
> > + agi->agi_iblocks = cpu_to_be32(1);
> > + if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > + agi->agi_fblocks = cpu_to_be32(1);
> > + }
> > }
> >
> > typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
> > diff --git a/fs/xfs/libxfs/xfs_format.h b/fs/xfs/libxfs/xfs_format.h
> > index 31b7ece985bb..03cbedb7eafc 100644
> > --- a/fs/xfs/libxfs/xfs_format.h
> > +++ b/fs/xfs/libxfs/xfs_format.h
> > @@ -449,6 +449,7 @@ xfs_sb_has_compat_feature(
> > #define XFS_SB_FEAT_RO_COMPAT_FINOBT (1 << 0) /* free inode btree */
> > #define XFS_SB_FEAT_RO_COMPAT_RMAPBT (1 << 1) /* reverse map btree */
> > #define XFS_SB_FEAT_RO_COMPAT_REFLINK (1 << 2) /* reflinked files */
> > +#define XFS_SB_FEAT_RO_COMPAT_INOBTCNT (1 << 3) /* inobt block counts */
> > #define XFS_SB_FEAT_RO_COMPAT_ALL \
> > (XFS_SB_FEAT_RO_COMPAT_FINOBT | \
> > XFS_SB_FEAT_RO_COMPAT_RMAPBT | \
> > @@ -563,6 +564,17 @@ static inline bool xfs_sb_version_hasreflink(struct xfs_sb *sbp)
> > (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_REFLINK);
> > }
> >
> > +/*
> > + * Inode btree block counter. We record the number of inobt and finobt blocks
> > + * in the AGI header so that we can skip the finobt walk at mount time when
> > + * setting up per-AG reservations.
> > + */
> > +static inline bool xfs_sb_version_hasinobtcounts(struct xfs_sb *sbp)
> > +{
> > + return XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
> > + (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_INOBTCNT);
> > +}
> > +
> > /*
> > * end of superblock version macros
> > */
> > @@ -765,6 +777,9 @@ typedef struct xfs_agi {
> > __be32 agi_free_root; /* root of the free inode btree */
> > __be32 agi_free_level;/* levels in free inode btree */
> >
> > + __be32 agi_iblocks; /* inobt blocks used */
> > + __be32 agi_fblocks; /* finobt blocks used */
> > +
> > /* structure must be padded to 64 bit alignment */
> > } xfs_agi_t;
> >
> > @@ -785,7 +800,8 @@ typedef struct xfs_agi {
> > #define XFS_AGI_ALL_BITS_R1 ((1 << XFS_AGI_NUM_BITS_R1) - 1)
> > #define XFS_AGI_FREE_ROOT (1 << 11)
> > #define XFS_AGI_FREE_LEVEL (1 << 12)
> > -#define XFS_AGI_NUM_BITS_R2 13
> > +#define XFS_AGI_IBLOCKS (1 << 13) /* both inobt/finobt block counters */
> > +#define XFS_AGI_NUM_BITS_R2 14
>
> I still find it a little odd that we'd log both fields if only one might
> be supported/modified, as opposed to just tracking them both
> independently with a couple extra lines of code. That said, I don't see
> it as a functional problem that couldn't be fixed later.
Yeah. I didn't want to go burning two bits for this since in all
likelihood the finobt will be enabled anytime inobtcounts are active.
I guess we could revisit that if someone uses db to frankenstein a
filesystem into having inobtcounts without a finobt and shows that the
overhead makes a difference, but the xfs_admin and mkfs tools aren't
going to allow that combination.
> >
> > /* disk block (xfs_daddr_t) in the AG */
> > #define XFS_AGI_DADDR(mp) ((xfs_daddr_t)(2 << (mp)->m_sectbb_log))
> > diff --git a/fs/xfs/libxfs/xfs_ialloc.c b/fs/xfs/libxfs/xfs_ialloc.c
> > index f742a96a2fe1..fef1d94c60a4 100644
> > --- a/fs/xfs/libxfs/xfs_ialloc.c
> > +++ b/fs/xfs/libxfs/xfs_ialloc.c
> > @@ -2473,6 +2473,7 @@ xfs_ialloc_log_agi(
> > offsetof(xfs_agi_t, agi_unlinked),
> > offsetof(xfs_agi_t, agi_free_root),
> > offsetof(xfs_agi_t, agi_free_level),
> > + offsetof(xfs_agi_t, agi_iblocks),
> > sizeof(xfs_agi_t)
> > };
> > #ifdef DEBUG
> > diff --git a/fs/xfs/libxfs/xfs_ialloc_btree.c b/fs/xfs/libxfs/xfs_ialloc_btree.c
> > index 3c8aebc36e64..cf51b342b6ef 100644
> > --- a/fs/xfs/libxfs/xfs_ialloc_btree.c
> > +++ b/fs/xfs/libxfs/xfs_ialloc_btree.c
> > @@ -67,6 +67,28 @@ xfs_finobt_set_root(
> > XFS_AGI_FREE_ROOT | XFS_AGI_FREE_LEVEL);
> > }
> >
> > +/* Update the inode btree block counter for this btree. */
> > +static inline void
> > +xfs_inobt_mod_blockcount(
> > + struct xfs_btree_cur *cur,
> > + int howmuch)
> > +{
> > + struct xfs_buf *agbp = cur->bc_ag.agbp;
> > + struct xfs_agi *agi = agbp->b_addr;
> > +
> > + if (!xfs_sb_version_hasinobtcounts(&cur->bc_mp->m_sb))
> > + return;
> > +
> > + if (cur->bc_btnum == XFS_BTNUM_FINO &&
> > + xfs_sb_version_hasfinobt(&cur->bc_mp->m_sb)) {
>
> This check might be spurious because I suspect you wouldn't get a finobt
> cursor without the feature enabled. Those nits aside:
Heh, yeah. I'll fix that one up, at least. Thanks for the review!
--D
> Reviewed-by: Brian Foster <bfoster@xxxxxxxxxx>
>
> > + be32_add_cpu(&agi->agi_fblocks, howmuch);
> > + xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_IBLOCKS);
> > + } else if (cur->bc_btnum == XFS_BTNUM_INO) {
> > + be32_add_cpu(&agi->agi_iblocks, howmuch);
> > + xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_IBLOCKS);
> > + }
> > +}
> > +
> > STATIC int
> > __xfs_inobt_alloc_block(
> > struct xfs_btree_cur *cur,
> > @@ -102,6 +124,7 @@ __xfs_inobt_alloc_block(
> >
> > new->s = cpu_to_be32(XFS_FSB_TO_AGBNO(args.mp, args.fsbno));
> > *stat = 1;
> > + xfs_inobt_mod_blockcount(cur, 1);
> > return 0;
> > }
> >
> > @@ -134,6 +157,7 @@ __xfs_inobt_free_block(
> > struct xfs_buf *bp,
> > enum xfs_ag_resv_type resv)
> > {
> > + xfs_inobt_mod_blockcount(cur, -1);
> > return xfs_free_extent(cur->bc_tp,
> > XFS_DADDR_TO_FSB(cur->bc_mp, XFS_BUF_ADDR(bp)), 1,
> > &XFS_RMAP_OINFO_INOBT, resv);
> > diff --git a/fs/xfs/xfs_ondisk.h b/fs/xfs/xfs_ondisk.h
> > index 5f04d8a5ab2a..acb9b737fe6b 100644
> > --- a/fs/xfs/xfs_ondisk.h
> > +++ b/fs/xfs/xfs_ondisk.h
> > @@ -23,7 +23,7 @@ xfs_check_ondisk_structs(void)
> > XFS_CHECK_STRUCT_SIZE(struct xfs_acl_entry, 12);
> > XFS_CHECK_STRUCT_SIZE(struct xfs_agf, 224);
> > XFS_CHECK_STRUCT_SIZE(struct xfs_agfl, 36);
> > - XFS_CHECK_STRUCT_SIZE(struct xfs_agi, 336);
> > + XFS_CHECK_STRUCT_SIZE(struct xfs_agi, 344);
> > XFS_CHECK_STRUCT_SIZE(struct xfs_bmbt_key, 8);
> > XFS_CHECK_STRUCT_SIZE(struct xfs_bmbt_rec, 16);
> > XFS_CHECK_STRUCT_SIZE(struct xfs_bmdr_block, 4);
> > diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
> > index 71ac6c1cdc36..c7ffcb57b586 100644
> > --- a/fs/xfs/xfs_super.c
> > +++ b/fs/xfs/xfs_super.c
> > @@ -1549,6 +1549,10 @@ xfs_fc_fill_super(
> > goto out_filestream_unmount;
> > }
> >
> > + if (xfs_sb_version_hasinobtcounts(&mp->m_sb))
> > + xfs_warn(mp,
> > + "EXPERIMENTAL inode btree counters feature in use. Use at your own risk!");
> > +
> > error = xfs_mountfs(mp);
> > if (error)
> > goto out_filestream_unmount;
> >
>
