[XFS] Implement the di_extsize allocator hint for non-realtime files as

well.  Also provides a mechanism for inheriting this property from the
parent directory for new files.

SGI-PV: 945264
SGI-Modid: xfs-linux-melb:xfs-kern:24367a

Signed-off-by: Nathan Scott <nathans@sgi.com>

fs/xfs/xfs_bmap.c

@@ -2146,13 +2146,176 @@ xfs_bmap_add_extent_hole_real(
 	return 0; /* keep gcc quite */
 }
 
+/*
+ * Adjust the size of the new extent based on di_extsize and rt extsize.
+ */
+STATIC int
+xfs_bmap_extsize_align(
+	xfs_mount_t	*mp,
+	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
+	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
+	xfs_extlen_t	extsz,		/* align to this extent size */
+	int		rt,		/* is this a realtime inode? */
+	int		eof,		/* is extent at end-of-file? */
+	int		delay,		/* creating delalloc extent? */
+	int		convert,	/* overwriting unwritten extent? */
+	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
+	xfs_extlen_t	*lenp)		/* in/out: aligned length */
+{
+	xfs_fileoff_t	orig_off;	/* original offset */
+	xfs_extlen_t	orig_alen;	/* original length */
+	xfs_fileoff_t	orig_end;	/* original off+len */
+	xfs_fileoff_t	nexto;		/* next file offset */
+	xfs_fileoff_t	prevo;		/* previous file offset */
+	xfs_fileoff_t	align_off;	/* temp for offset */
+	xfs_extlen_t	align_alen;	/* temp for length */
+	xfs_extlen_t	temp;		/* temp for calculations */
+
+	if (convert)
+		return 0;
+
+	orig_off = align_off = *offp;
+	orig_alen = align_alen = *lenp;
+	orig_end = orig_off + orig_alen;
+
+	/*
+	 * If this request overlaps an existing extent, then don't
+	 * attempt to perform any additional alignment.
+	 */
+	if (!delay && !eof &&
+	    (orig_off >= gotp->br_startoff) &&
+	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
+		return 0;
+	}
+
+	/*
+	 * If the file offset is unaligned vs. the extent size
+	 * we need to align it.  This will be possible unless
+	 * the file was previously written with a kernel that didn't
+	 * perform this alignment, or if a truncate shot us in the
+	 * foot.
+	 */
+	temp = do_mod(orig_off, extsz);
+	if (temp) {
+		align_alen += temp;
+		align_off -= temp;
+	}
+	/*
+	 * Same adjustment for the end of the requested area.
+	 */
+	if ((temp = (align_alen % extsz))) {
+		align_alen += extsz - temp;
+	}
+	/*
+	 * If the previous block overlaps with this proposed allocation
+	 * then move the start forward without adjusting the length.
+	 */
+	if (prevp->br_startoff != NULLFILEOFF) {
+		if (prevp->br_startblock == HOLESTARTBLOCK)
+			prevo = prevp->br_startoff;
+		else
+			prevo = prevp->br_startoff + prevp->br_blockcount;
+	} else
+		prevo = 0;
+	if (align_off != orig_off && align_off < prevo)
+		align_off = prevo;
+	/*
+	 * If the next block overlaps with this proposed allocation
+	 * then move the start back without adjusting the length,
+	 * but not before offset 0.
+	 * This may of course make the start overlap previous block,
+	 * and if we hit the offset 0 limit then the next block
+	 * can still overlap too.
+	 */
+	if (!eof && gotp->br_startoff != NULLFILEOFF) {
+		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
+		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
+			nexto = gotp->br_startoff + gotp->br_blockcount;
+		else
+			nexto = gotp->br_startoff;
+	} else
+		nexto = NULLFILEOFF;
+	if (!eof &&
+	    align_off + align_alen != orig_end &&
+	    align_off + align_alen > nexto)
+		align_off = nexto > align_alen ? nexto - align_alen : 0;
+	/*
+	 * If we're now overlapping the next or previous extent that
+	 * means we can't fit an extsz piece in this hole.  Just move
+	 * the start forward to the first valid spot and set
+	 * the length so we hit the end.
+	 */
+	if (align_off != orig_off && align_off < prevo)
+		align_off = prevo;
+	if (align_off + align_alen != orig_end &&
+	    align_off + align_alen > nexto &&
+	    nexto != NULLFILEOFF) {
+		ASSERT(nexto > prevo);
+		align_alen = nexto - align_off;
+	}
+
+	/*
+	 * If realtime, and the result isn't a multiple of the realtime
+	 * extent size we need to remove blocks until it is.
+	 */
+	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
+		/*
+		 * We're not covering the original request, or
+		 * we won't be able to once we fix the length.
+		 */
+		if (orig_off < align_off ||
+		    orig_end > align_off + align_alen ||
+		    align_alen - temp < orig_alen)
+			return XFS_ERROR(EINVAL);
+		/*
+		 * Try to fix it by moving the start up.
+		 */
+		if (align_off + temp <= orig_off) {
+			align_alen -= temp;
+			align_off += temp;
+		}
+		/*
+		 * Try to fix it by moving the end in.
+		 */
+		else if (align_off + align_alen - temp >= orig_end)
+			align_alen -= temp;
+		/*
+		 * Set the start to the minimum then trim the length.
+		 */
+		else {
+			align_alen -= orig_off - align_off;
+			align_off = orig_off;
+			align_alen -= align_alen % mp->m_sb.sb_rextsize;
+		}
+		/*
+		 * Result doesn't cover the request, fail it.
+		 */
+		if (orig_off < align_off || orig_end > align_off + align_alen)
+			return XFS_ERROR(EINVAL);
+	} else {
+		ASSERT(orig_off >= align_off);
+		ASSERT(orig_end <= align_off + align_alen);
+	}
+
+#ifdef DEBUG
+	if (!eof && gotp->br_startoff != NULLFILEOFF)
+		ASSERT(align_off + align_alen <= gotp->br_startoff);
+	if (prevp->br_startoff != NULLFILEOFF)
+		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
+#endif
+
+	*lenp = align_alen;
+	*offp = align_off;
+	return 0;
+}
+
 #define XFS_ALLOC_GAP_UNITS	4
 
 /*
  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
  * It figures out where to ask the underlying allocator to put the new extent.
  */
-STATIC int				/* error */
+STATIC int
 xfs_bmap_alloc(
 	xfs_bmalloca_t	*ap)		/* bmap alloc argument struct */
 {
@@ -2163,10 +2326,10 @@ xfs_bmap_alloc(
 	xfs_mount_t	*mp;		/* mount point structure */
 	int		nullfb;		/* true if ap->firstblock isn't set */
 	int		rt;		/* true if inode is realtime */
-#ifdef __KERNEL__
-	xfs_extlen_t	prod=0;		/* product factor for allocators */
-	xfs_extlen_t	ralen=0;	/* realtime allocation length */
-#endif
+	xfs_extlen_t	prod = 0;	/* product factor for allocators */
+	xfs_extlen_t	ralen = 0;	/* realtime allocation length */
+	xfs_extlen_t	align;		/* minimum allocation alignment */
+	xfs_rtblock_t	rtx;
 
 #define	ISVALID(x,y)	\
 	(rt ? \
@@ -2182,125 +2345,25 @@ xfs_bmap_alloc(
 	nullfb = ap->firstblock == NULLFSBLOCK;
 	rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
-#ifdef __KERNEL__
 	if (rt) {
-		xfs_extlen_t	extsz;		/* file extent size for rt */
-		xfs_fileoff_t	nexto;		/* next file offset */
-		xfs_extlen_t	orig_alen;	/* original ap->alen */
-		xfs_fileoff_t	orig_end;	/* original off+len */
-		xfs_fileoff_t	orig_off;	/* original ap->off */
-		xfs_extlen_t	mod_off;	/* modulus calculations */
-		xfs_fileoff_t	prevo;		/* previous file offset */
-		xfs_rtblock_t	rtx;		/* realtime extent number */
-		xfs_extlen_t	temp;		/* temp for rt calculations */
-
-		/*
-		 * Set prod to match the realtime extent size.
-		 */
-		if (!(extsz = ap->ip->i_d.di_extsize))
-			extsz = mp->m_sb.sb_rextsize;
-		prod = extsz / mp->m_sb.sb_rextsize;
-		orig_off = ap->off;
-		orig_alen = ap->alen;
-		orig_end = orig_off + orig_alen;
-		/*
-		 * If the file offset is unaligned vs. the extent size
-		 * we need to align it.  This will be possible unless
-		 * the file was previously written with a kernel that didn't
-		 * perform this alignment.
-		 */
-		mod_off = do_mod(orig_off, extsz);
-		if (mod_off) {
-			ap->alen += mod_off;
-			ap->off -= mod_off;
-		}
-		/*
-		 * Same adjustment for the end of the requested area.
-		 */
-		if ((temp = (ap->alen % extsz)))
-			ap->alen += extsz - temp;
-		/*
-		 * If the previous block overlaps with this proposed allocation
-		 * then move the start forward without adjusting the length.
-		 */
-		prevo =
-			ap->prevp->br_startoff == NULLFILEOFF ?
-				0 :
-				(ap->prevp->br_startoff +
-				 ap->prevp->br_blockcount);
-		if (ap->off != orig_off && ap->off < prevo)
-			ap->off = prevo;
-		/*
-		 * If the next block overlaps with this proposed allocation
-		 * then move the start back without adjusting the length,
-		 * but not before offset 0.
-		 * This may of course make the start overlap previous block,
-		 * and if we hit the offset 0 limit then the next block
-		 * can still overlap too.
-		 */
-		nexto = (ap->eof || ap->gotp->br_startoff == NULLFILEOFF) ?
-			NULLFILEOFF : ap->gotp->br_startoff;
-		if (!ap->eof &&
-		    ap->off + ap->alen != orig_end &&
-		    ap->off + ap->alen > nexto)
-			ap->off = nexto > ap->alen ? nexto - ap->alen : 0;
-		/*
-		 * If we're now overlapping the next or previous extent that
-		 * means we can't fit an extsz piece in this hole.  Just move
-		 * the start forward to the first valid spot and set
-		 * the length so we hit the end.
-		 */
-		if ((ap->off != orig_off && ap->off < prevo) ||
-		    (ap->off + ap->alen != orig_end &&
-		     ap->off + ap->alen > nexto)) {
-			ap->off = prevo;
-			ap->alen = nexto - prevo;
-		}
-		/*
-		 * If the result isn't a multiple of rtextents we need to
-		 * remove blocks until it is.
-		 */
-		if ((temp = (ap->alen % mp->m_sb.sb_rextsize))) {
-			/*
-			 * We're not covering the original request, or
-			 * we won't be able to once we fix the length.
-			 */
-			if (orig_off < ap->off ||
-			    orig_end > ap->off + ap->alen ||
-			    ap->alen - temp < orig_alen)
-				return XFS_ERROR(EINVAL);
-			/*
-			 * Try to fix it by moving the start up.
-			 */
-			if (ap->off + temp <= orig_off) {
-				ap->alen -= temp;
-				ap->off += temp;
-			}
-			/*
-			 * Try to fix it by moving the end in.
-			 */
-			else if (ap->off + ap->alen - temp >= orig_end)
-				ap->alen -= temp;
-			/*
-			 * Set the start to the minimum then trim the length.
-			 */
-			else {
-				ap->alen -= orig_off - ap->off;
-				ap->off = orig_off;
-				ap->alen -= ap->alen % mp->m_sb.sb_rextsize;
-			}
-			/*
-			 * Result doesn't cover the request, fail it.
-			 */
-			if (orig_off < ap->off || orig_end > ap->off + ap->alen)
-				return XFS_ERROR(EINVAL);
-		}
+		align = ap->ip->i_d.di_extsize ?
+			ap->ip->i_d.di_extsize : mp->m_sb.sb_rextsize;
+		/* Set prod to match the extent size */
+		prod = align / mp->m_sb.sb_rextsize;
+
+		error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
+						align, rt, ap->eof, 0,
+						ap->conv, &ap->off, &ap->alen);
+		if (error)
+			return error;
+		ASSERT(ap->alen);
 		ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
+
 		/*
 		 * If the offset & length are not perfectly aligned
 		 * then kill prod, it will just get us in trouble.
 		 */
-		if (do_mod(ap->off, extsz) || ap->alen % extsz)
+		if (do_mod(ap->off, align) || ap->alen % align)
 			prod = 1;
 		/*
 		 * Set ralen to be the actual requested length in rtextents.
@@ -2326,15 +2389,24 @@ xfs_bmap_alloc(
 			ap->rval = rtx * mp->m_sb.sb_rextsize;
 		} else
 			ap->rval = 0;
+	} else {
+		align = (ap->userdata && ap->ip->i_d.di_extsize &&
+			(ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)) ?
+			ap->ip->i_d.di_extsize : 0;
+		if (unlikely(align)) {
+			error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
+							align, rt,
+							ap->eof, 0, ap->conv,
+							&ap->off, &ap->alen);
+			ASSERT(!error);
+			ASSERT(ap->alen);
+		}
+		if (nullfb)
+			ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
+		else
+			ap->rval = ap->firstblock;
 	}
-#else
-	if (rt)
-		ap->rval = 0;
-#endif	/* __KERNEL__ */
-	else if (nullfb)
-		ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
-	else
-		ap->rval = ap->firstblock;
+
 	/*
 	 * If allocating at eof, and there's a previous real block,
 	 * try to use it's last block as our starting point.
@@ -2598,11 +2670,12 @@ xfs_bmap_alloc(
 			args.total = ap->total;
 			args.minlen = ap->minlen;
 		}
-		if (ap->ip->i_d.di_extsize) {
+		if (unlikely(ap->userdata && ap->ip->i_d.di_extsize &&
+			    (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE))) {
 			args.prod = ap->ip->i_d.di_extsize;
 			if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
 				args.mod = (xfs_extlen_t)(args.prod - args.mod);
-		} else if (mp->m_sb.sb_blocksize >= NBPP) {
+		} else if (unlikely(mp->m_sb.sb_blocksize >= NBPP)) {
 			args.prod = 1;
 			args.mod = 0;
 		} else {
@@ -4590,6 +4663,7 @@ xfs_bmapi(
 	char		contig;		/* allocation must be one extent */
 	char		delay;		/* this request is for delayed alloc */
 	char		exact;		/* don't do all of wasdelayed extent */
+	char		convert;	/* unwritten extent I/O completion */
 	xfs_bmbt_rec_t	*ep;		/* extent list entry pointer */
 	int		error;		/* error return */
 	xfs_bmbt_irec_t	got;		/* current extent list record */
@@ -4643,7 +4717,7 @@ xfs_bmapi(
 	}
 	if (XFS_FORCED_SHUTDOWN(mp))
 		return XFS_ERROR(EIO);
-	rt = XFS_IS_REALTIME_INODE(ip);
+	rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
 	ifp = XFS_IFORK_PTR(ip, whichfork);
 	ASSERT(ifp->if_ext_max ==
 	       XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
@@ -4654,6 +4728,7 @@ xfs_bmapi(
 	delay = (flags & XFS_BMAPI_DELAY) != 0;
 	trim = (flags & XFS_BMAPI_ENTIRE) == 0;
 	userdata = (flags & XFS_BMAPI_METADATA) == 0;
+	convert = (flags & XFS_BMAPI_CONVERT) != 0;
 	exact = (flags & XFS_BMAPI_EXACT) != 0;
 	rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
 	contig = (flags & XFS_BMAPI_CONTIG) != 0;
@@ -4748,16 +4823,26 @@ xfs_bmapi(
 			}
 			minlen = contig ? alen : 1;
 			if (delay) {
-				xfs_extlen_t	extsz = 0;
+				xfs_extlen_t	extsz;
 
 				/* Figure out the extent size, adjust alen */
 				if (rt) {
 					if (!(extsz = ip->i_d.di_extsize))
 						extsz = mp->m_sb.sb_rextsize;
-					alen = roundup(alen, extsz);
-					extsz = alen / mp->m_sb.sb_rextsize;
+				} else {
+					extsz = ip->i_d.di_extsize;
+				}
+				if (extsz) {
+					error = xfs_bmap_extsize_align(mp,
+							&got, &prev, extsz,
+							rt, eof, delay, convert,
+							&aoff, &alen);
+					ASSERT(!error);
 				}
 
+				if (rt)
+					extsz = alen / mp->m_sb.sb_rextsize;
+
 				/*
 				 * Make a transaction-less quota reservation for
 				 * delayed allocation blocks. This number gets
@@ -4785,14 +4870,15 @@ xfs_bmapi(
 					xfs_bmap_worst_indlen(ip, alen);
 				ASSERT(indlen > 0);
 
-				if (rt)
+				if (rt) {
 					error = xfs_mod_incore_sb(mp,
 							XFS_SBS_FREXTENTS,
 							-(extsz), rsvd);
-				else
+				} else {
 					error = xfs_mod_incore_sb(mp,
 							XFS_SBS_FDBLOCKS,
 							-(alen), rsvd);
+				}
 				if (!error) {
 					error = xfs_mod_incore_sb(mp,
 							XFS_SBS_FDBLOCKS,
@@ -4811,6 +4897,7 @@ xfs_bmapi(
 				if (error) {
 					if (XFS_IS_QUOTA_ON(ip->i_mount))
 						/* unreserve the blocks now */
+						(void)
 						XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
 							mp, NULL, ip,
 							(long)alen, 0, rt ?
@@ -4849,6 +4936,7 @@ xfs_bmapi(
 				bma.firstblock = *firstblock;
 				bma.alen = alen;
 				bma.off = aoff;
+				bma.conv = convert;
 				bma.wasdel = wasdelay;
 				bma.minlen = minlen;
 				bma.low = flist->xbf_low;
@@ -5270,8 +5358,7 @@ xfs_bunmapi(
 		return 0;
 	}
 	XFS_STATS_INC(xs_blk_unmap);
-	isrt = (whichfork == XFS_DATA_FORK) &&
-	       (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
+	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
 	start = bno;
 	bno = start + len - 1;
 	ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
@@ -5443,7 +5530,7 @@ xfs_bunmapi(
 		}
 		if (wasdel) {
 			ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
-			/* Update realtim/data freespace, unreserve quota */
+			/* Update realtime/data freespace, unreserve quota */
 			if (isrt) {
 				xfs_filblks_t rtexts;
 
@@ -5451,14 +5538,14 @@ xfs_bunmapi(
 				do_div(rtexts, mp->m_sb.sb_rextsize);
 				xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
 						(int)rtexts, rsvd);
-				XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, NULL, ip,
-					-((long)del.br_blockcount), 0,
+				(void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
+					NULL, ip, -((long)del.br_blockcount), 0,
 					XFS_QMOPT_RES_RTBLKS);
 			} else {
 				xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
 						(int)del.br_blockcount, rsvd);
-				XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, NULL, ip,
-					-((long)del.br_blockcount), 0,
+				(void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
+					NULL, ip, -((long)del.br_blockcount), 0,
 					XFS_QMOPT_RES_REGBLKS);
 			}
 			ip->i_delayed_blks -= del.br_blockcount;
@@ -5652,7 +5739,9 @@ xfs_getbmap(
 		   ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
 		return XFS_ERROR(EINVAL);
 	if (whichfork == XFS_DATA_FORK) {
-		if (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC) {
+		if ((ip->i_d.di_extsize && (ip->i_d.di_flags &
+				(XFS_DIFLAG_REALTIME|XFS_DIFLAG_EXTSIZE))) ||
+		    ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
 			prealloced = 1;
 			fixlen = XFS_MAXIOFFSET(mp);
 		} else {

fs/xfs/xfs_bmap.h

@@ -62,6 +62,10 @@ typedef	struct xfs_bmap_free
 #define	XFS_BMAPI_IGSTATE	0x200	/* Ignore state - */
 					/* combine contig. space */
 #define	XFS_BMAPI_CONTIG	0x400	/* must allocate only one extent */
+/*	XFS_BMAPI_DIRECT_IO	0x800	*/
+#define XFS_BMAPI_CONVERT	0x1000	/* unwritten extent conversion - */
+					/* need write cache flushing and no */
+					/* additional allocation alignments */
 
 #define	XFS_BMAPI_AFLAG(w)	xfs_bmapi_aflag(w)
 static inline int xfs_bmapi_aflag(int w)
@@ -101,7 +105,8 @@ typedef struct xfs_bmalloca {
 	char			wasdel;	/* replacing a delayed allocation */
 	char			userdata;/* set if is user data */
 	char			low;	/* low on space, using seq'l ags */
-	char			aeof;   /* allocated space at eof */
+	char			aeof;	/* allocated space at eof */
+	char			conv;	/* overwriting unwritten extents */
 } xfs_bmalloca_t;
 
 #ifdef __KERNEL__

fs/xfs/xfs_dinode.h

@@ -246,8 +246,10 @@ typedef enum xfs_dinode_fmt
 #define XFS_DIFLAG_NOATIME_BIT   6	/* do not update atime */
 #define XFS_DIFLAG_NODUMP_BIT    7	/* do not dump */
 #define XFS_DIFLAG_RTINHERIT_BIT 8	/* create with realtime bit set */
-#define XFS_DIFLAG_PROJINHERIT_BIT  9	/* create with parents projid */
-#define XFS_DIFLAG_NOSYMLINKS_BIT  10	/* disallow symlink creation */
+#define XFS_DIFLAG_PROJINHERIT_BIT   9	/* create with parents projid */
+#define XFS_DIFLAG_NOSYMLINKS_BIT   10	/* disallow symlink creation */
+#define XFS_DIFLAG_EXTSIZE_BIT      11	/* inode extent size allocator hint */
+#define XFS_DIFLAG_EXTSZINHERIT_BIT 12	/* inherit inode extent size */
 #define XFS_DIFLAG_REALTIME      (1 << XFS_DIFLAG_REALTIME_BIT)
 #define XFS_DIFLAG_PREALLOC      (1 << XFS_DIFLAG_PREALLOC_BIT)
 #define XFS_DIFLAG_NEWRTBM       (1 << XFS_DIFLAG_NEWRTBM_BIT)
@@ -259,11 +261,14 @@ typedef enum xfs_dinode_fmt
 #define XFS_DIFLAG_RTINHERIT     (1 << XFS_DIFLAG_RTINHERIT_BIT)
 #define XFS_DIFLAG_PROJINHERIT   (1 << XFS_DIFLAG_PROJINHERIT_BIT)
 #define XFS_DIFLAG_NOSYMLINKS    (1 << XFS_DIFLAG_NOSYMLINKS_BIT)
+#define XFS_DIFLAG_EXTSIZE       (1 << XFS_DIFLAG_EXTSIZE_BIT)
+#define XFS_DIFLAG_EXTSZINHERIT  (1 << XFS_DIFLAG_EXTSZINHERIT_BIT)
 
 #define XFS_DIFLAG_ANY \
 	(XFS_DIFLAG_REALTIME | XFS_DIFLAG_PREALLOC | XFS_DIFLAG_NEWRTBM | \
 	 XFS_DIFLAG_IMMUTABLE | XFS_DIFLAG_APPEND | XFS_DIFLAG_SYNC | \
 	 XFS_DIFLAG_NOATIME | XFS_DIFLAG_NODUMP | XFS_DIFLAG_RTINHERIT | \
-	 XFS_DIFLAG_PROJINHERIT | XFS_DIFLAG_NOSYMLINKS)
+	 XFS_DIFLAG_PROJINHERIT | XFS_DIFLAG_NOSYMLINKS | XFS_DIFLAG_EXTSIZE | \
+	 XFS_DIFLAG_EXTSZINHERIT)
 
 #endif	/* __XFS_DINODE_H__ */

fs/xfs/xfs_fs.h

@@ -65,6 +65,8 @@ struct fsxattr {
 #define XFS_XFLAG_RTINHERIT	0x00000100	/* create with rt bit set */
 #define XFS_XFLAG_PROJINHERIT	0x00000200	/* create with parents projid */
 #define XFS_XFLAG_NOSYMLINKS	0x00000400	/* disallow symlink creation */
+#define XFS_XFLAG_EXTSIZE	0x00000800	/* extent size allocator hint */
+#define XFS_XFLAG_EXTSZINHERIT	0x00001000	/* inherit inode extent size */
 #define XFS_XFLAG_HASATTR	0x80000000	/* no DIFLAG for this	*/
 
 /*

fs/xfs/xfs_inode.c

@@ -809,6 +809,10 @@ _xfs_dic2xflags(
 			flags |= XFS_XFLAG_PROJINHERIT;
 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
 			flags |= XFS_XFLAG_NOSYMLINKS;
+		if (di_flags & XFS_DIFLAG_EXTSIZE)
+			flags |= XFS_XFLAG_EXTSIZE;
+		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
+			flags |= XFS_XFLAG_EXTSZINHERIT;
 	}
 
 	return flags;
@@ -1192,11 +1196,19 @@ xfs_ialloc(
 			if ((mode & S_IFMT) == S_IFDIR) {
 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
 					di_flags |= XFS_DIFLAG_RTINHERIT;
-			} else {
+				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
+					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
+					ip->i_d.di_extsize = pip->i_d.di_extsize;
+				}
+			} else if ((mode & S_IFMT) == S_IFREG) {
 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
 					di_flags |= XFS_DIFLAG_REALTIME;
 					ip->i_iocore.io_flags |= XFS_IOCORE_RT;
 				}
+				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
+					di_flags |= XFS_DIFLAG_EXTSIZE;
+					ip->i_d.di_extsize = pip->i_d.di_extsize;
+				}
 			}
 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
 			    xfs_inherit_noatime)
@@ -1262,7 +1274,7 @@ xfs_isize_check(
 	if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
 		return;
 
-	if ( ip->i_d.di_flags & XFS_DIFLAG_REALTIME )
+	if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
 		return;
 
 	nimaps = 2;

fs/xfs/xfs_iomap.c

@@ -263,7 +263,7 @@ phase2:
 	case BMAPI_WRITE:
 		/* If we found an extent, return it */
 		if (nimaps &&
-		    (imap.br_startblock != HOLESTARTBLOCK) && 
+		    (imap.br_startblock != HOLESTARTBLOCK) &&
 		    (imap.br_startblock != DELAYSTARTBLOCK)) {
 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 					offset, count, iomapp, &imap, flags);
@@ -317,6 +317,58 @@ out:
 	return XFS_ERROR(error);
 }
 
+STATIC int
+xfs_iomap_eof_align_last_fsb(
+	xfs_mount_t	*mp,
+	xfs_iocore_t	*io,
+	xfs_fsize_t	isize,
+	xfs_extlen_t	extsize,
+	xfs_fileoff_t	*last_fsb)
+{
+	xfs_fileoff_t	new_last_fsb = 0;
+	xfs_extlen_t	align;
+	int		eof, error;
+
+	if (io->io_flags & XFS_IOCORE_RT)
+		;
+	/*
+	 * If mounted with the "-o swalloc" option, roundup the allocation
+	 * request to a stripe width boundary if the file size is >=
+	 * stripe width and we are allocating past the allocation eof.
+	 */
+	else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
+	        (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)))
+		new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
+	/*
+	 * Roundup the allocation request to a stripe unit (m_dalign) boundary
+	 * if the file size is >= stripe unit size, and we are allocating past
+	 * the allocation eof.
+	 */
+	else if (mp->m_dalign && (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)))
+		new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
+
+	/*
+	 * Always round up the allocation request to an extent boundary
+	 * (when file on a real-time subvolume or has di_extsize hint).
+	 */
+	if (extsize) {
+		if (new_last_fsb)
+			align = roundup_64(new_last_fsb, extsize);
+		else
+			align = extsize;
+		new_last_fsb = roundup_64(*last_fsb, align);
+	}
+
+	if (new_last_fsb) {
+		error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
+		if (error)
+			return error;
+		if (eof)
+			*last_fsb = new_last_fsb;
+	}
+	return 0;
+}
+
 STATIC int
 xfs_flush_space(
 	xfs_inode_t	*ip,
@@ -363,19 +415,20 @@ xfs_iomap_write_direct(
 	xfs_iocore_t	*io = &ip->i_iocore;
 	xfs_fileoff_t	offset_fsb;
 	xfs_fileoff_t	last_fsb;
-	xfs_filblks_t	count_fsb;
+	xfs_filblks_t	count_fsb, resaligned;
 	xfs_fsblock_t	firstfsb;
+	xfs_extlen_t	extsz, temp;
+	xfs_fsize_t	isize;
 	int		nimaps;
-	int		error;
 	int		bmapi_flag;
 	int		quota_flag;
 	int		rt;
 	xfs_trans_t	*tp;
 	xfs_bmbt_irec_t imap;
 	xfs_bmap_free_t free_list;
-	xfs_filblks_t	qblocks, resblks;
+	uint		qblocks, resblks, resrtextents;
 	int		committed;
-	int		resrtextents;
+	int		error;
 
 	/*
 	 * Make sure that the dquots are there. This doesn't hold
@@ -385,37 +438,52 @@ xfs_iomap_write_direct(
 	if (error)
 		return XFS_ERROR(error);
 
-	offset_fsb = XFS_B_TO_FSBT(mp, offset);
-	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
-	count_fsb = last_fsb - offset_fsb;
-	if (found && (ret_imap->br_startblock == HOLESTARTBLOCK)) {
-		xfs_fileoff_t	map_last_fsb;
-
-		map_last_fsb = ret_imap->br_blockcount + ret_imap->br_startoff;
-		if (map_last_fsb < last_fsb) {
-			last_fsb = map_last_fsb;
-			count_fsb = last_fsb - offset_fsb;
-		}
-		ASSERT(count_fsb > 0);
+	rt = XFS_IS_REALTIME_INODE(ip);
+	if (unlikely(rt)) {
+		if (!(extsz = ip->i_d.di_extsize))
+			extsz = mp->m_sb.sb_rextsize;
+	} else {
+		extsz = ip->i_d.di_extsize;
 	}
 
-	/*
-	 * Determine if reserving space on the data or realtime partition.
-	 */
-	if ((rt = XFS_IS_REALTIME_INODE(ip))) {
-		xfs_extlen_t	extsz;
+	isize = ip->i_d.di_size;
+	if (io->io_new_size > isize)
+		isize = io->io_new_size;
 
-		if (!(extsz = ip->i_d.di_extsize))
-			extsz = mp->m_sb.sb_rextsize;
-		resrtextents = qblocks = (count_fsb + extsz - 1);
-		do_div(resrtextents, mp->m_sb.sb_rextsize);
-		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
-		quota_flag = XFS_QMOPT_RES_RTBLKS;
+  	offset_fsb = XFS_B_TO_FSBT(mp, offset);
+  	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
+	if ((offset + count) > isize) {
+		error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
+							&last_fsb);
+		if (error)
+			goto error_out;
 	} else {
-		resrtextents = 0;
-		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, count_fsb);
-		quota_flag = XFS_QMOPT_RES_REGBLKS;
+		if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
+			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
+					ret_imap->br_blockcount +
+					ret_imap->br_startoff);
 	}
+	count_fsb = last_fsb - offset_fsb;
+	ASSERT(count_fsb > 0);
+
+	resaligned = count_fsb;
+	if (unlikely(extsz)) {
+		if ((temp = do_mod(offset_fsb, extsz)))
+			resaligned += temp;
+		if ((temp = do_mod(resaligned, extsz)))
+			resaligned += extsz - temp;
+	}
+
+	if (unlikely(rt)) {
+		resrtextents = qblocks = resaligned;
+		resrtextents /= mp->m_sb.sb_rextsize;
+  		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
+  		quota_flag = XFS_QMOPT_RES_RTBLKS;
+  	} else {
+  		resrtextents = 0;
+		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
+  		quota_flag = XFS_QMOPT_RES_REGBLKS;
+  	}
 
 	/*
 	 * Allocate and setup the transaction
@@ -426,7 +494,6 @@ xfs_iomap_write_direct(
 			XFS_WRITE_LOG_RES(mp), resrtextents,
 			XFS_TRANS_PERM_LOG_RES,
 			XFS_WRITE_LOG_COUNT);
-
 	/*
 	 * Check for running out of space, note: need lock to return
 	 */
@@ -436,20 +503,20 @@ xfs_iomap_write_direct(
 	if (error)
 		goto error_out;
 
-	if (XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag)) {
-		error = (EDQUOT);
+	error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
+					      qblocks, 0, quota_flag);
+	if (error)
 		goto error1;
-	}
 
-	bmapi_flag = XFS_BMAPI_WRITE;
 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 	xfs_trans_ihold(tp, ip);
 
-	if (!(flags & BMAPI_MMAP) && (offset < ip->i_d.di_size || rt))
+	bmapi_flag = XFS_BMAPI_WRITE;
+	if ((flags & BMAPI_DIRECT) && (offset < ip->i_d.di_size || extsz))
 		bmapi_flag |= XFS_BMAPI_PREALLOC;
 
 	/*
-	 * Issue the bmapi() call to allocate the blocks
+	 * Issue the xfs_bmapi() call to allocate the blocks
 	 */
 	XFS_BMAP_INIT(&free_list, &firstfsb);
 	nimaps = 1;
@@ -501,6 +568,62 @@ error_out:
 	return XFS_ERROR(error);
 }
 
+/*
+ * If the caller is doing a write at the end of the file,
+ * then extend the allocation out to the file system's write
+ * iosize.  We clean up any extra space left over when the
+ * file is closed in xfs_inactive().
+ *
+ * For sync writes, we are flushing delayed allocate space to
+ * try to make additional space available for allocation near
+ * the filesystem full boundary - preallocation hurts in that
+ * situation, of course.
+ */
+STATIC int
+xfs_iomap_eof_want_preallocate(
+	xfs_mount_t	*mp,
+	xfs_iocore_t	*io,
+	xfs_fsize_t	isize,
+	xfs_off_t	offset,
+	size_t		count,
+	int		ioflag,
+	xfs_bmbt_irec_t *imap,
+	int		nimaps,
+	int		*prealloc)
+{
+	xfs_fileoff_t   start_fsb;
+	xfs_filblks_t   count_fsb;
+	xfs_fsblock_t	firstblock;
+	int		n, error, imaps;
+
+	*prealloc = 0;
+	if ((ioflag & BMAPI_SYNC) || (offset + count) <= isize)
+		return 0;
+
+	/*
+	 * If there are any real blocks past eof, then don't
+	 * do any speculative allocation.
+	 */
+	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
+	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
+	while (count_fsb > 0) {
+		imaps = nimaps;
+		error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
+				  0, &firstblock, 0, imap, &imaps, NULL);
+		if (error)
+			return error;
+		for (n = 0; n < imaps; n++) {
+			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
+			    (imap[n].br_startblock != DELAYSTARTBLOCK))
+				return 0;
+			start_fsb += imap[n].br_blockcount;
+			count_fsb -= imap[n].br_blockcount;
+		}
+	}
+	*prealloc = 1;
+	return 0;
+}
+
 int
 xfs_iomap_write_delay(
 	xfs_inode_t	*ip,
@@ -514,13 +637,15 @@ xfs_iomap_write_delay(
 	xfs_iocore_t	*io = &ip->i_iocore;
 	xfs_fileoff_t	offset_fsb;
 	xfs_fileoff_t	last_fsb;
-	xfs_fsize_t	isize;
+	xfs_off_t	aligned_offset;
+	xfs_fileoff_t	ioalign;
 	xfs_fsblock_t	firstblock;
+	xfs_extlen_t	extsz;
+	xfs_fsize_t	isize;
 	int		nimaps;
-	int		error;
 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
-	int		aeof;
-	int		fsynced = 0;
+	int		prealloc, fsynced = 0;
+	int		error;
 
 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
 
@@ -528,152 +653,57 @@ xfs_iomap_write_delay(
 	 * Make sure that the dquots are there. This doesn't hold
 	 * the ilock across a disk read.
 	 */
-
 	error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
 	if (error)
 		return XFS_ERROR(error);
 
+	if (XFS_IS_REALTIME_INODE(ip)) {
+		if (!(extsz = ip->i_d.di_extsize))
+			extsz = mp->m_sb.sb_rextsize;
+	} else {
+		extsz = ip->i_d.di_extsize;
+	}
+
+	offset_fsb = XFS_B_TO_FSBT(mp, offset);
+
 retry:
 	isize = ip->i_d.di_size;
-	if (io->io_new_size > isize) {
+	if (io->io_new_size > isize)
 		isize = io->io_new_size;
-	}
 
-	aeof = 0;
-	offset_fsb = XFS_B_TO_FSBT(mp, offset);
-	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
-	/*
-	 * If the caller is doing a write at the end of the file,
-	 * then extend the allocation (and the buffer used for the write)
-	 * out to the file system's write iosize.  We clean up any extra
-	 * space left over when the file is closed in xfs_inactive().
-	 *
-	 * For sync writes, we are flushing delayed allocate space to
-	 * try to make additional space available for allocation near
-	 * the filesystem full boundary - preallocation hurts in that
-	 * situation, of course.
-	 */
-	if (!(ioflag & BMAPI_SYNC) && ((offset + count) > ip->i_d.di_size)) {
-		xfs_off_t	aligned_offset;
-		xfs_filblks_t   count_fsb;
-		unsigned int	iosize;
-		xfs_fileoff_t	ioalign;
-		int		n;
-		xfs_fileoff_t   start_fsb;
+	error = xfs_iomap_eof_want_preallocate(mp, io, isize, offset, count,
+				ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
+	if (error)
+		return error;
 
-		/*
-		 * If there are any real blocks past eof, then don't
-		 * do any speculative allocation.
-		 */
-		start_fsb = XFS_B_TO_FSBT(mp,
-					((xfs_ufsize_t)(offset + count - 1)));
-		count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
-		while (count_fsb > 0) {
-			nimaps = XFS_WRITE_IMAPS;
-			error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
-					0, &firstblock, 0, imap, &nimaps, NULL);
-			if (error) {
-				return error;
-			}
-			for (n = 0; n < nimaps; n++) {
-				if ( !(io->io_flags & XFS_IOCORE_RT)  && 
-					!imap[n].br_startblock) {
-					cmn_err(CE_PANIC,"Access to block "
-						"zero:  fs <%s> inode: %lld "
-						"start_block : %llx start_off "
-						": %llx blkcnt : %llx "
-						"extent-state : %x \n",
-						(ip->i_mount)->m_fsname,
-						(long long)ip->i_ino,
-						imap[n].br_startblock,
-						imap[n].br_startoff,
-						imap[n].br_blockcount,
-						imap[n].br_state);
-        			}
-				if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
-				    (imap[n].br_startblock != DELAYSTARTBLOCK)) {
-					goto write_map;
-				}
-				start_fsb += imap[n].br_blockcount;
-				count_fsb -= imap[n].br_blockcount;
-			}
-		}
-		iosize = mp->m_writeio_blocks;
+	if (prealloc) {
 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
-		last_fsb = ioalign + iosize;
-		aeof = 1;
+		last_fsb = ioalign + mp->m_writeio_blocks;
+	} else {
+		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 	}
-write_map:
-	nimaps = XFS_WRITE_IMAPS;
-	firstblock = NULLFSBLOCK;
 
-	/*
-	 * If mounted with the "-o swalloc" option, roundup the allocation
-	 * request to a stripe width boundary if the file size is >=
-	 * stripe width and we are allocating past the allocation eof.
-	 */
-	if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_swidth 
-	    && (mp->m_flags & XFS_MOUNT_SWALLOC)
-	    && (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)) && aeof) {
-		int eof;
-		xfs_fileoff_t new_last_fsb;
-
-		new_last_fsb = roundup_64(last_fsb, mp->m_swidth);
-		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
-		if (error) {
-			return error;
-		}
-		if (eof) {
-			last_fsb = new_last_fsb;
-		}
-	/*
-	 * Roundup the allocation request to a stripe unit (m_dalign) boundary
-	 * if the file size is >= stripe unit size, and we are allocating past
-	 * the allocation eof.
-	 */
-	} else if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_dalign &&
-		   (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)) && aeof) {
-		int eof;
-		xfs_fileoff_t new_last_fsb;
-		new_last_fsb = roundup_64(last_fsb, mp->m_dalign);
-		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
-		if (error) {
-			return error;
-		}
-		if (eof) {
-			last_fsb = new_last_fsb;
-		}
-	/*
-	 * Round up the allocation request to a real-time extent boundary
-	 * if the file is on the real-time subvolume.
-	 */
-	} else if (io->io_flags & XFS_IOCORE_RT && aeof) {
-		int eof;
-		xfs_fileoff_t new_last_fsb;
-
-		new_last_fsb = roundup_64(last_fsb, mp->m_sb.sb_rextsize);
-		error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
-		if (error) {
+	if (prealloc || extsz) {
+		error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
+							&last_fsb);
+		if (error)
 			return error;
-		}
-		if (eof)
-			last_fsb = new_last_fsb;
 	}
+
+	nimaps = XFS_WRITE_IMAPS;
+	firstblock = NULLFSBLOCK;
 	error = xfs_bmapi(NULL, ip, offset_fsb,
 			  (xfs_filblks_t)(last_fsb - offset_fsb),
 			  XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
 			  XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
 			  &nimaps, NULL);
-	/*
-	 * This can be EDQUOT, if nimaps == 0
-	 */
-	if (error && (error != ENOSPC)) {
+	if (error && (error != ENOSPC))
 		return XFS_ERROR(error);
-	}
+
 	/*
 	 * If bmapi returned us nothing, and if we didn't get back EDQUOT,
-	 * then we must have run out of space.
+	 * then we must have run out of space - flush delalloc, and retry..
 	 */
 	if (nimaps == 0) {
 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
@@ -685,9 +715,7 @@ write_map:
 		goto retry;
 	}
 
-	*ret_imap = imap[0];
-	*nmaps = 1;
-	if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
+	if (!(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 		cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
                         "start_block : %llx start_off : %llx blkcnt : %llx "
                         "extent-state : %x \n",
@@ -696,6 +724,10 @@ write_map:
                         ret_imap->br_startblock, ret_imap->br_startoff,
                         ret_imap->br_blockcount,ret_imap->br_state);
 	}
+
+	*ret_imap = imap[0];
+	*nmaps = 1;
+
 	return 0;
 }
 
@@ -868,17 +900,17 @@ xfs_iomap_write_unwritten(
 {
 	xfs_mount_t	*mp = ip->i_mount;
 	xfs_iocore_t    *io = &ip->i_iocore;
-	xfs_trans_t	*tp;
 	xfs_fileoff_t	offset_fsb;
 	xfs_filblks_t	count_fsb;
 	xfs_filblks_t	numblks_fsb;
-	xfs_bmbt_irec_t	imap;
+	xfs_fsblock_t	firstfsb;
+	int		nimaps;
+	xfs_trans_t	*tp;
+	xfs_bmbt_irec_t imap;
+	xfs_bmap_free_t free_list;
+	uint		resblks;
 	int		committed;
 	int		error;
-	int		nres;
-	int		nimaps;
-	xfs_fsblock_t	firstfsb;
-	xfs_bmap_free_t	free_list;
 
 	xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
 				&ip->i_iocore, offset, count);
@@ -887,9 +919,9 @@ xfs_iomap_write_unwritten(
 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 
-	do {
-		nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
+	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 
+	do {
 		/*
 		 * set up a transaction to convert the range of extents
 		 * from unwritten to real. Do allocations in a loop until
@@ -897,7 +929,7 @@ xfs_iomap_write_unwritten(
 		 */
 
 		tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
-		error = xfs_trans_reserve(tp, nres,
+		error = xfs_trans_reserve(tp, resblks,
 				XFS_WRITE_LOG_RES(mp), 0,
 				XFS_TRANS_PERM_LOG_RES,
 				XFS_WRITE_LOG_COUNT);
@@ -916,7 +948,7 @@ xfs_iomap_write_unwritten(
 		XFS_BMAP_INIT(&free_list, &firstfsb);
 		nimaps = 1;
 		error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
-				  XFS_BMAPI_WRITE, &firstfsb,
+				  XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
 				  1, &imap, &nimaps, &free_list);
 		if (error)
 			goto error_on_bmapi_transaction;
@@ -930,7 +962,7 @@ xfs_iomap_write_unwritten(
 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 		if (error)
 			goto error0;
-		
+
 		if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
 			cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
 				"inode: %lld start_block : %llx start_off : "

fs/xfs/xfs_vnodeops.c

@@ -540,24 +540,6 @@ xfs_setattr(
 			goto error_return;
 		}
 
-		/*
-		 * Can't set extent size unless the file is marked, or
-		 * about to be marked as a realtime file.
-		 *
-		 * This check will be removed when fixed size extents
-		 * with buffered data writes is implemented.
-		 *
-		 */
-		if ((mask & XFS_AT_EXTSIZE)			&&
-		    ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
-		     vap->va_extsize) &&
-		    (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
-		       ((mask & XFS_AT_XFLAGS) &&
-			(vap->va_xflags & XFS_XFLAG_REALTIME))))) {
-			code = XFS_ERROR(EINVAL);
-			goto error_return;
-		}
-
 		/*
 		 * Can't change realtime flag if any extents are allocated.
 		 */
@@ -820,13 +802,17 @@ xfs_setattr(
 					di_flags |= XFS_DIFLAG_RTINHERIT;
 				if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
 					di_flags |= XFS_DIFLAG_NOSYMLINKS;
-			} else {
+				if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
+					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
+			} else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
 				if (vap->va_xflags & XFS_XFLAG_REALTIME) {
 					di_flags |= XFS_DIFLAG_REALTIME;
 					ip->i_iocore.io_flags |= XFS_IOCORE_RT;
 				} else {
 					ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
 				}
+				if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
+					di_flags |= XFS_DIFLAG_EXTSIZE;
 			}
 			ip->i_d.di_flags = di_flags;
 		}
@@ -1568,7 +1554,8 @@ xfs_release(
 		if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
 		     ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
 		     (ip->i_df.if_flags & XFS_IFEXTENTS))  &&
-		    (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
+		    (!(ip->i_d.di_flags &
+				(XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
 			if ((error = xfs_inactive_free_eofblocks(mp, ip)))
 				return (error);
 			/* Update linux inode block count after free above */
@@ -1644,9 +1631,10 @@ xfs_inactive(
 	if (ip->i_d.di_nlink != 0) {
 		if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
 		     ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
-		     (ip->i_df.if_flags & XFS_IFEXTENTS))  &&
-		    (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
-		     (ip->i_delayed_blks != 0))) {
+		      (ip->i_df.if_flags & XFS_IFEXTENTS) &&
+		     (!(ip->i_d.di_flags &
+				(XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
+		      (ip->i_delayed_blks != 0)))) {
 			if ((error = xfs_inactive_free_eofblocks(mp, ip)))
 				return (VN_INACTIVE_CACHE);
 			/* Update linux inode block count after free above */
@@ -3998,42 +3986,36 @@ xfs_alloc_file_space(
 	int			alloc_type,
 	int			attr_flags)
 {
+	xfs_mount_t		*mp = ip->i_mount;
+	xfs_off_t		count;
 	xfs_filblks_t		allocated_fsb;
 	xfs_filblks_t		allocatesize_fsb;
-	int			committed;
-	xfs_off_t		count;
-	xfs_filblks_t		datablocks;
-	int			error;
+	xfs_extlen_t		extsz, temp;
+	xfs_fileoff_t		startoffset_fsb;
 	xfs_fsblock_t		firstfsb;
-	xfs_bmap_free_t		free_list;
-	xfs_bmbt_irec_t		*imapp;
-	xfs_bmbt_irec_t		imaps[1];
-	xfs_mount_t		*mp;
-	int			numrtextents;
-	int			reccount;
-	uint			resblks;
+	int			nimaps;
+	int			bmapi_flag;
+	int			quota_flag;
 	int			rt;
-	int			rtextsize;
-	xfs_fileoff_t		startoffset_fsb;
 	xfs_trans_t		*tp;
-	int			xfs_bmapi_flags;
+	xfs_bmbt_irec_t		imaps[1], *imapp;
+	xfs_bmap_free_t		free_list;
+	uint			qblocks, resblks, resrtextents;
+	int			committed;
+	int			error;
 
 	vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
-	mp = ip->i_mount;
 
 	if (XFS_FORCED_SHUTDOWN(mp))
 		return XFS_ERROR(EIO);
 
-	/*
-	 * determine if this is a realtime file
-	 */
-	if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
-		if (ip->i_d.di_extsize)
-			rtextsize = ip->i_d.di_extsize;
-		else
-			rtextsize = mp->m_sb.sb_rextsize;
-	} else
-		rtextsize = 0;
+	rt = XFS_IS_REALTIME_INODE(ip);
+	if (unlikely(rt)) {
+		if (!(extsz = ip->i_d.di_extsize))
+			extsz = mp->m_sb.sb_rextsize;
+	} else {
+		extsz = ip->i_d.di_extsize;
+	}
 
 	if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
 		return error;
@@ -4044,8 +4026,8 @@ xfs_alloc_file_space(
 	count = len;
 	error = 0;
 	imapp = &imaps[0];
-	reccount = 1;
-	xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
+	nimaps = 1;
+	bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
 	startoffset_fsb	= XFS_B_TO_FSBT(mp, offset);
 	allocatesize_fsb = XFS_B_TO_FSB(mp, count);
 
@@ -4066,43 +4048,51 @@ xfs_alloc_file_space(
 	}
 
 	/*
-	 * allocate file space until done or until there is an error
+	 * Allocate file space until done or until there is an error
 	 */
 retry:
 	while (allocatesize_fsb && !error) {
+		xfs_fileoff_t	s, e;
+
 		/*
-		 * determine if reserving space on
-		 * the data or realtime partition.
+		 * Determine space reservations for data/realtime,
 		 */
-		if (rt) {
-			xfs_fileoff_t s, e;
-
+		if (unlikely(extsz)) {
 			s = startoffset_fsb;
-			do_div(s, rtextsize);
-			s *= rtextsize;
-			e = roundup_64(startoffset_fsb + allocatesize_fsb,
-				rtextsize);
-			numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
-			datablocks = 0;
+			do_div(s, extsz);
+			s *= extsz;
+			e = startoffset_fsb + allocatesize_fsb;
+			if ((temp = do_mod(startoffset_fsb, extsz)))
+				e += temp;
+			if ((temp = do_mod(e, extsz)))
+				e += extsz - temp;
+		} else {
+			s = 0;
+			e = allocatesize_fsb;
+		}
+
+		if (unlikely(rt)) {
+			resrtextents = qblocks = (uint)(e - s);
+			resrtextents /= mp->m_sb.sb_rextsize;
+			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
+			quota_flag = XFS_QMOPT_RES_RTBLKS;
 		} else {
-			datablocks = allocatesize_fsb;
-			numrtextents = 0;
+			resrtextents = 0;
+			resblks = qblocks = \
+				XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
+			quota_flag = XFS_QMOPT_RES_REGBLKS;
 		}
 
 		/*
-		 * allocate and setup the transaction
+		 * Allocate and setup the transaction.
 		 */
 		tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
-		resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
-		error = xfs_trans_reserve(tp,
-					  resblks,
-					  XFS_WRITE_LOG_RES(mp),
-					  numrtextents,
+		error = xfs_trans_reserve(tp, resblks,
+					  XFS_WRITE_LOG_RES(mp), resrtextents,
 					  XFS_TRANS_PERM_LOG_RES,
 					  XFS_WRITE_LOG_COUNT);
-
 		/*
-		 * check for running out of space
+		 * Check for running out of space
 		 */
 		if (error) {
 			/*
@@ -4113,8 +4103,8 @@ retry:
 			break;
 		}
 		xfs_ilock(ip, XFS_ILOCK_EXCL);
-		error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
-				ip->i_udquot, ip->i_gdquot, resblks, 0, 0);
+		error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
+						      qblocks, 0, quota_flag);
 		if (error)
 			goto error1;
 
@@ -4122,19 +4112,19 @@ retry:
 		xfs_trans_ihold(tp, ip);
 
 		/*
-		 * issue the bmapi() call to allocate the blocks
+		 * Issue the xfs_bmapi() call to allocate the blocks
 		 */
 		XFS_BMAP_INIT(&free_list, &firstfsb);
 		error = xfs_bmapi(tp, ip, startoffset_fsb,
-				  allocatesize_fsb, xfs_bmapi_flags,
-				  &firstfsb, 0, imapp, &reccount,
+				  allocatesize_fsb, bmapi_flag,
+				  &firstfsb, 0, imapp, &nimaps,
 				  &free_list);
 		if (error) {
 			goto error0;
 		}
 
 		/*
-		 * complete the transaction
+		 * Complete the transaction
 		 */
 		error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
 		if (error) {
@@ -4149,7 +4139,7 @@ retry:
 
 		allocated_fsb = imapp->br_blockcount;
 
-		if (reccount == 0) {
+		if (nimaps == 0) {
 			error = XFS_ERROR(ENOSPC);
 			break;
 		}
@@ -4172,9 +4162,11 @@ dmapi_enospc_check:
 
 	return error;
 
- error0:
+error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
 	xfs_bmap_cancel(&free_list);
- error1:
+	XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
+
+error1:	/* Just cancel transaction */
 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 	goto dmapi_enospc_check;
@@ -4419,8 +4411,8 @@ xfs_free_file_space(
 		}
 		xfs_ilock(ip, XFS_ILOCK_EXCL);
 		error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
-				ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
-				XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
+				ip->i_udquot, ip->i_gdquot, resblks, 0,
+				XFS_QMOPT_RES_REGBLKS);
 		if (error)
 			goto error1;