xfs: implement busy extent tracking for rtgroups

For rtgroups filesystems, track newly freed (rt) space through the log
until the rt EFIs have been committed to disk.  This way we ensure that
space cannot be reused until all traces of the old owner are gone.

As a fringe benefit, we now support -o discard on the realtime device.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>

1 files changed, 122 insertions, 5 deletions

diff --git a/fs/xfs/xfs_rtalloc.c b/fs/xfs/xfs_rtalloc.c
index 8520d72afac0..7ecea7623a15 100644
--- a/fs/xfs/xfs_rtalloc.c
+++ b/fs/xfs/xfs_rtalloc.c
@@ -29,6 +29,7 @@
 #include "xfs_metafile.h"
 #include "xfs_rtgroup.h"
 #include "xfs_error.h"
+#include "xfs_trace.h"
 
 /*
  * Return whether there are any free extents in the size range given
@@ -1659,6 +1660,114 @@ xfs_rtalloc_align_minmax(
 	*raminlen = newminlen;
 }
 
+/* Given a free extent, find any part of it that isn't busy, if possible. */
+STATIC bool
+xfs_rtalloc_check_busy(
+	struct xfs_rtalloc_args	*args,
+	xfs_rtxnum_t		start,
+	xfs_rtxlen_t		minlen_rtx,
+	xfs_rtxlen_t		maxlen_rtx,
+	xfs_rtxlen_t		len_rtx,
+	xfs_rtxlen_t		prod,
+	xfs_rtxnum_t		rtx,
+	xfs_rtxlen_t		*reslen,
+	xfs_rtxnum_t		*resrtx,
+	unsigned		*busy_gen)
+{
+	struct xfs_rtgroup	*rtg = args->rtg;
+	struct xfs_mount	*mp = rtg_mount(rtg);
+	xfs_agblock_t		rgbno = xfs_rtx_to_rgbno(rtg, rtx);
+	xfs_rgblock_t		min_rgbno = xfs_rtx_to_rgbno(rtg, start);
+	xfs_extlen_t		minlen = xfs_rtxlen_to_extlen(mp, minlen_rtx);
+	xfs_extlen_t		len = xfs_rtxlen_to_extlen(mp, len_rtx);
+	xfs_extlen_t		diff;
+	bool			busy;
+
+	busy = xfs_extent_busy_trim(rtg_group(rtg), minlen,
+			xfs_rtxlen_to_extlen(mp, maxlen_rtx), &rgbno, &len,
+			busy_gen);
+
+	/*
+	 * If we have a largish extent that happens to start before min_rgbno,
+	 * see if we can shift it into range...
+	 */
+	if (rgbno < min_rgbno && rgbno + len > min_rgbno) {
+		diff = min_rgbno - rgbno;
+		if (len > diff) {
+			rgbno += diff;
+			len -= diff;
+		}
+	}
+
+	if (prod > 1 && len >= minlen) {
+		xfs_rgblock_t	aligned_rgbno = roundup(rgbno, prod);
+
+		diff = aligned_rgbno - rgbno;
+
+		*resrtx = xfs_rgbno_to_rtx(mp, aligned_rgbno);
+		*reslen = xfs_extlen_to_rtxlen(mp,
+				diff >= len ? 0 : len - diff);
+	} else {
+		*resrtx = xfs_rgbno_to_rtx(mp, rgbno);
+		*reslen = xfs_extlen_to_rtxlen(mp, len);
+	}
+
+	return busy;
+}
+
+/*
+ * Adjust the given free extent so that it isn't busy, or flush the log and
+ * wait for the space to become unbusy.  Only needed for rtgroups.
+ */
+STATIC int
+xfs_rtallocate_adjust_for_busy(
+	struct xfs_rtalloc_args	*args,
+	xfs_rtxnum_t		start,
+	xfs_rtxlen_t		minlen,
+	xfs_rtxlen_t		maxlen,
+	xfs_rtxlen_t		*len,
+	xfs_rtxlen_t		prod,
+	xfs_rtxnum_t		*rtx)
+{
+	xfs_rtxnum_t		resrtx;
+	xfs_rtxlen_t		reslen;
+	unsigned		busy_gen;
+	bool			busy;
+	int			error;
+
+again:
+	busy = xfs_rtalloc_check_busy(args, start, minlen, maxlen, *len, prod,
+			*rtx, &reslen, &resrtx, &busy_gen);
+	if (!busy)
+		return 0;
+
+	if (reslen < minlen || (start != 0 && resrtx != *rtx)) {
+		/*
+		 * Enough of the extent was busy that we cannot satisfy the
+		 * allocation, or this is a near allocation and the start of
+		 * the extent is busy.  Flush the log and wait for the busy
+		 * situation to resolve.
+		 */
+		trace_xfs_rtalloc_extent_busy(args->rtg, start, minlen, maxlen,
+				*len, prod, *rtx, busy_gen);
+
+		error = xfs_extent_busy_flush(args->tp, rtg_group(args->rtg),
+				busy_gen, 0);
+		if (error)
+			return error;
+
+		goto again;
+	}
+
+	/* Some of the free space wasn't busy, hand that back to the caller. */
+	trace_xfs_rtalloc_extent_busy_trim(args->rtg, *rtx, *len, resrtx,
+			reslen);
+	*len = reslen;
+	*rtx = resrtx;
+
+	return 0;
+}
+
 static int
 xfs_rtallocate_rtg(
 	struct xfs_trans	*tp,
@@ -1740,15 +1849,19 @@ xfs_rtallocate_rtg(
 	}
 
 	if (error) {
-		if (xfs_has_rtgroups(args.mp)) {
-			xfs_rtgroup_unlock(args.rtg, XFS_RTGLOCK_BITMAP);
-			*rtlocked = false;
-		}
+		if (xfs_has_rtgroups(args.mp))
+			goto out_unlock;
 		goto out_release;
 	}
 
-	if (xfs_has_rtgroups(args.mp))
+	if (xfs_has_rtgroups(args.mp)) {
+		error = xfs_rtallocate_adjust_for_busy(&args, start, minlen,
+				maxlen, &len, prod, &rtx);
+		if (error)
+			goto out_unlock;
+
 		xfs_rtgroup_trans_join(tp, args.rtg, XFS_RTGLOCK_BITMAP);
+	}
 
 	error = xfs_rtallocate_range(&args, rtx, len);
 	if (error)
@@ -1764,6 +1877,10 @@ out_release:
 	xfs_rtgroup_rele(args.rtg);
 	xfs_rtbuf_cache_relse(&args);
 	return error;
+out_unlock:
+	xfs_rtgroup_unlock(args.rtg, XFS_RTGLOCK_BITMAP);
+	*rtlocked = false;
+	goto out_release;
 }
 
 static int