xfs: do not read the AGI buffer in xfs_dialloc until nessecary

Refactor the AG selection loop in xfs_dialloc to operate on the in-memory
perag data as much as possible.  We only read the AGI buffer once we have
selected an AG to allocate inodes now instead of for every AG considered.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>

fs/xfs/xfs_ialloc.c

@@ -900,11 +900,10 @@ xfs_dialloc(
 	struct xfs_mount	*mp = tp->t_mountp;
 	struct xfs_buf		*agbp;
 	xfs_agnumber_t		agno;
-	struct xfs_agi		*agi;
 	int			error;
 	int			ialloced;
 	int			noroom = 0;
-	xfs_agnumber_t		tagno;
+	xfs_agnumber_t		start_agno;
 	struct xfs_perag	*pag;
 
 	if (*IO_agbp) {
@@ -921,25 +920,17 @@ xfs_dialloc(
 	 * We do not have an agbp, so select an initial allocation
 	 * group for inode allocation.
 	 */
-	agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
+	start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
-	if (agno == NULLAGNUMBER) {
+	if (start_agno == NULLAGNUMBER) {
 		*inop = NULLFSINO;
 		return 0;
 	}
 
-	error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
-	if (error)
-		return XFS_ERROR(error);
-	agi = XFS_BUF_TO_AGI(agbp);
-
-	tagno = agno;
-
 	/*
 	 * If we have already hit the ceiling of inode blocks then clear
 	 * okalloc so we scan all available agi structures for a free
 	 * inode.
 	 */
-
 	if (mp->m_maxicount &&
 	    mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
 		noroom = 1;
@@ -951,24 +942,56 @@ xfs_dialloc(
 	 * or in which we can allocate some inodes.  Iterate through the
 	 * allocation groups upward, wrapping at the end.
 	 */
-	while (!agi->agi_freecount) {
+	agno = start_agno;
+	for (;;) {
+		pag = xfs_perag_get(mp, agno);
+		if (!pag->pagi_inodeok) {
+			xfs_ialloc_next_ag(mp);
+			goto nextag;
+		}
+
+		if (!pag->pagi_init) {
+			error = xfs_ialloc_pagi_init(mp, tp, agno);
+			if (error)
+				goto out_error;
+		}
+
 		/*
-		 * Don't do anything if we're not supposed to allocate
+		 * Do a first racy fast path check if this AG is usable.
-		 * any blocks, just go on to the next ag.
 		 */
-		if (okalloc) {
+		if (!pag->pagi_freecount && !okalloc)
+			goto nextag;
+
+		error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
+		if (error)
+			goto out_error;
+
 		/*
-			 * Try to allocate some new inodes in the allocation
+		 * Once the AGI has been read in we have to recheck
-			 * group.
+		 * pagi_freecount with the AGI buffer lock held.
 		 */
-			if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
+		if (pag->pagi_freecount) {
+			xfs_perag_put(pag);
+			goto out_alloc;
+		}
+
+		if (!okalloc) {
 			xfs_trans_brelse(tp, agbp);
-				if (error == ENOSPC) {
+			goto nextag;
+		}
+
+		error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
+		if (error) {
+			xfs_trans_brelse(tp, agbp);
+
+			if (error != ENOSPC)
+				goto out_error;
+
+			xfs_perag_put(pag);
 			*inop = NULLFSINO;
 			return 0;
-				} else
-					return error;
 		}
+
 		if (ialloced) {
 			/*
 			 * We successfully allocated some inodes, return
@@ -976,42 +999,30 @@ xfs_dialloc(
 			 * can commit the current transaction and call
 			 * us again where we left off.
 			 */
-				ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
+			ASSERT(pag->pagi_freecount > 0);
+			xfs_perag_put(pag);
+
 			*IO_agbp = agbp;
 			*inop = NULLFSINO;
 			return 0;
 		}
-		}
+
-		/*
-		 * If it failed, give up on this ag.
-		 */
-		xfs_trans_brelse(tp, agbp);
-		/*
-		 * Go on to the next ag: get its ag header.
-		 */
 nextag:
-		if (++tagno == mp->m_sb.sb_agcount)
+		xfs_perag_put(pag);
-			tagno = 0;
+		if (++agno == mp->m_sb.sb_agcount)
-		if (tagno == agno) {
+			agno = 0;
+		if (agno == start_agno) {
 			*inop = NULLFSINO;
 			return noroom ? ENOSPC : 0;
 		}
-		pag = xfs_perag_get(mp, tagno);
-		if (pag->pagi_inodeok == 0) {
-			xfs_perag_put(pag);
-			goto nextag;
-		}
-		error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
-		xfs_perag_put(pag);
-		if (error)
-			goto nextag;
-		agi = XFS_BUF_TO_AGI(agbp);
-		ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
 	}
 
 out_alloc:
 	*IO_agbp = NULL;
 	return xfs_dialloc_ag(tp, agbp, parent, inop);
+out_error:
+	xfs_perag_put(pag);
+	return XFS_ERROR(error);
 }
 
 /*