Merge tag 'realtime-groups-6.13_2024-11-05' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into staging-merge

xfs: shard the realtime section [v5.5 06/10]

Right now, the realtime section uses a single pair of metadata inodes to
store the free space information.  This presents a scalability problem
since every thread trying to allocate or free rt extents have to lock
these files.  Solve this problem by sharding the realtime section into
separate realtime allocation groups.

While we're at it, define a superblock to be stamped into the start of
the rt section.  This enables utilities such as blkid to identify block
devices containing realtime sections, and avoids the situation where
anything written into block 0 of the realtime extent can be
misinterpreted as file data.

The best advantage for rtgroups will become evident later when we get to
adding rmap and reflink to the realtime volume, since the geometry
constraints are the same for rt groups and AGs.  Hence we can reuse all
that code directly.

This is a very large patchset, but it catches us up with 20 years of
technical debt that have accumulated.

With a bit of luck, this should all go splendidly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>

1 files changed, 45 insertions, 19 deletions

diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c
index cee7f0564409..4db022c189e1 100644
--- a/fs/xfs/xfs_trans.c
+++ b/fs/xfs/xfs_trans.c
@@ -25,6 +25,8 @@
 #include "xfs_dquot.h"
 #include "xfs_icache.h"
 #include "xfs_rtbitmap.h"
+#include "xfs_rtgroup.h"
+#include "xfs_sb.h"
 
 struct kmem_cache	*xfs_trans_cache;
 
@@ -420,6 +422,8 @@ xfs_trans_mod_sb(
 			ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
 		}
 		tp->t_frextents_delta += delta;
+		if (xfs_has_rtgroups(mp))
+			flags &= ~XFS_TRANS_SB_DIRTY;
 		break;
 	case XFS_TRANS_SB_RES_FREXTENTS:
 		/*
@@ -429,6 +433,8 @@ xfs_trans_mod_sb(
 		 */
 		ASSERT(delta < 0);
 		tp->t_res_frextents_delta += delta;
+		if (xfs_has_rtgroups(mp))
+			flags &= ~XFS_TRANS_SB_DIRTY;
 		break;
 	case XFS_TRANS_SB_DBLOCKS:
 		tp->t_dblocks_delta += delta;
@@ -455,6 +461,10 @@ xfs_trans_mod_sb(
 	case XFS_TRANS_SB_REXTSLOG:
 		tp->t_rextslog_delta += delta;
 		break;
+	case XFS_TRANS_SB_RGCOUNT:
+		ASSERT(delta > 0);
+		tp->t_rgcount_delta += delta;
+		break;
 	default:
 		ASSERT(0);
 		return;
@@ -497,20 +507,22 @@ xfs_trans_apply_sb_deltas(
 	}
 
 	/*
-	 * Updating frextents requires careful handling because it does not
-	 * behave like the lazysb counters because we cannot rely on log
-	 * recovery in older kenels to recompute the value from the rtbitmap.
-	 * This means that the ondisk frextents must be consistent with the
-	 * rtbitmap.
+	 * sb_frextents was added to the lazy sb counters when the rt groups
+	 * feature was introduced.  This is possible because we know that all
+	 * kernels supporting rtgroups will also recompute frextents from the
+	 * realtime bitmap.
+	 *
+	 * For older file systems, updating frextents requires careful handling
+	 * because we cannot rely on log recovery in older kernels to recompute
+	 * the value from the rtbitmap.  This means that the ondisk frextents
+	 * must be consistent with the rtbitmap.
 	 *
 	 * Therefore, log the frextents change to the ondisk superblock and
 	 * update the incore superblock so that future calls to xfs_log_sb
 	 * write the correct value ondisk.
-	 *
-	 * Don't touch m_frextents because it includes incore reservations,
-	 * and those are handled by the unreserve function.
 	 */
-	if (tp->t_frextents_delta || tp->t_res_frextents_delta) {
+	if ((tp->t_frextents_delta || tp->t_res_frextents_delta) &&
+	    !xfs_has_rtgroups(tp->t_mountp)) {
 		struct xfs_mount	*mp = tp->t_mountp;
 		int64_t			rtxdelta;
 
@@ -536,6 +548,18 @@ xfs_trans_apply_sb_deltas(
 	}
 	if (tp->t_rextsize_delta) {
 		be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
+
+		/*
+		 * Because the ondisk sb records rtgroup size in units of rt
+		 * extents, any time we update the rt extent size we have to
+		 * recompute the ondisk rtgroup block log.  The incore values
+		 * will be recomputed in xfs_trans_unreserve_and_mod_sb.
+		 */
+		if (xfs_has_rtgroups(tp->t_mountp)) {
+			sbp->sb_rgblklog = xfs_compute_rgblklog(
+						be32_to_cpu(sbp->sb_rgextents),
+						be32_to_cpu(sbp->sb_rextsize));
+		}
 		whole = 1;
 	}
 	if (tp->t_rbmblocks_delta) {
@@ -554,6 +578,10 @@ xfs_trans_apply_sb_deltas(
 		sbp->sb_rextslog += tp->t_rextslog_delta;
 		whole = 1;
 	}
+	if (tp->t_rgcount_delta) {
+		be32_add_cpu(&sbp->sb_rgcount, tp->t_rgcount_delta);
+		whole = 1;
+	}
 
 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
 	if (whole)
@@ -618,7 +646,7 @@ xfs_trans_unreserve_and_mod_sb(
 	}
 
 	ASSERT(tp->t_rtx_res || tp->t_frextents_delta >= 0);
-	if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
+	if (xfs_has_rtgroups(mp) || (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
 		rtxdelta += tp->t_frextents_delta;
 		ASSERT(rtxdelta >= 0);
 	}
@@ -651,23 +679,21 @@ xfs_trans_unreserve_and_mod_sb(
 	mp->m_sb.sb_icount += idelta;
 	mp->m_sb.sb_ifree += ifreedelta;
 	/*
-	 * Do not touch sb_frextents here because we are dealing with incore
-	 * reservation.  sb_frextents is not part of the lazy sb counters so it
-	 * must be consistent with the ondisk rtbitmap and must never include
-	 * incore reservations.
+	 * Do not touch sb_frextents here because it is handled in
+	 * xfs_trans_apply_sb_deltas for file systems where it isn't a lazy
+	 * counter anyway.
 	 */
 	mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
 	mp->m_sb.sb_agcount += tp->t_agcount_delta;
 	mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
-	mp->m_sb.sb_rextsize += tp->t_rextsize_delta;
-	if (tp->t_rextsize_delta) {
-		mp->m_rtxblklog = log2_if_power2(mp->m_sb.sb_rextsize);
-		mp->m_rtxblkmask = mask64_if_power2(mp->m_sb.sb_rextsize);
-	}
+	if (tp->t_rextsize_delta)
+		xfs_mount_sb_set_rextsize(mp, &mp->m_sb,
+				mp->m_sb.sb_rextsize + tp->t_rextsize_delta);
 	mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
 	mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
 	mp->m_sb.sb_rextents += tp->t_rextents_delta;
 	mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
+	mp->m_sb.sb_rgcount += tp->t_rgcount_delta;
 	spin_unlock(&mp->m_sb_lock);
 
 	/*