[XFS] Implement support for unwritten extents in XFS.

SGI Modid: 2.5.x-xfs:slinx:141508a

fs/xfs/linux/xfs_iops.h

@@ -66,5 +66,6 @@ extern struct file_operations linvfs_dir_operations;
 extern struct address_space_operations linvfs_aops;
 
 extern int linvfs_get_block(struct inode *, sector_t, struct buffer_head *, int);
+extern void linvfs_unwritten_done(struct buffer_head *, int);
 
 #endif /* __XFS_IOPS_H__ */

fs/xfs/linux/xfs_linux.h

@@ -69,6 +69,20 @@
 #define STATIC static
 #endif
 
+/*
+ * State flag for unwritten extent buffers.
+ *
+ * We need to be able to distinguish between these and delayed
+ * allocate buffers within XFS.  The generic IO path code does
+ * not need to distinguish - we use the BH_Delay flag for both
+ * delalloc and these ondisk-uninitialised buffers.
+ */
+BUFFER_FNS(PrivateStart, unwritten);
+static inline void set_buffer_unwritten_io(struct buffer_head *bh)
+{
+	bh->b_end_io = linvfs_unwritten_done;
+}
+
 #define restricted_chown	xfs_params.restrict_chown
 #define irix_sgid_inherit	xfs_params.sgid_inherit
 #define irix_symlink_mode	xfs_params.symlink_mode

fs/xfs/pagebuf/page_buf.c

@@ -121,7 +121,8 @@ pb_trace_func(
 STATIC kmem_cache_t *pagebuf_cache;
 STATIC void pagebuf_daemon_wakeup(int);
 STATIC void pagebuf_delwri_queue(page_buf_t *, int);
-STATIC struct workqueue_struct *pagebuf_workqueue;
+STATIC struct workqueue_struct *pagebuf_logio_workqueue;
+STATIC struct workqueue_struct *pagebuf_dataio_workqueue;
 
 /*
  * Pagebuf module configuration parameters, exported via
@@ -785,6 +786,25 @@ pagebuf_get(				/* allocate a buffer		*/
 	return (pb);
 }
 
+/*
+ * Create a skeletal pagebuf (no pages associated with it).
+ */
+page_buf_t *
+pagebuf_lookup(
+	struct pb_target	*target,
+	loff_t			ioff,
+	size_t			isize,
+	page_buf_flags_t	flags)
+{
+	page_buf_t		*pb;
+
+	pb = pagebuf_allocate(flags);
+	if (pb) {
+		_pagebuf_initialize(pb, target, ioff, isize, flags);
+	}
+	return pb;
+}
+
 /*
  * If we are not low on memory then do the readahead in a deadlock
  * safe manner.
@@ -1131,6 +1151,7 @@ pagebuf_iodone_work(
 void
 pagebuf_iodone(
 	page_buf_t		*pb,
+	int			dataio,
 	int			schedule)
 {
 	pb->pb_flags &= ~(PBF_READ | PBF_WRITE);
@@ -1143,7 +1164,8 @@ pagebuf_iodone(
 	if ((pb->pb_iodone) || (pb->pb_flags & PBF_ASYNC)) {
 		if (schedule) {
 			INIT_WORK(&pb->pb_iodone_work, pagebuf_iodone_work, pb);
-			queue_work(pagebuf_workqueue, &pb->pb_iodone_work);
+			queue_work(dataio ? pagebuf_dataio_workqueue :
+				pagebuf_logio_workqueue, &pb->pb_iodone_work);
 		} else {
 			pagebuf_iodone_work(pb);
 		}
@@ -1268,7 +1290,7 @@ bio_end_io_pagebuf(
 
 	if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
 		pb->pb_locked = 0;
-		pagebuf_iodone(pb, 1);
+		pagebuf_iodone(pb, 0, 1);
 	}
 
 	bio_put(bio);
@@ -1412,7 +1434,7 @@ pagebuf_iorequest(			/* start real I/O		*/
 
 	if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
 		pb->pb_locked = 0;
-		pagebuf_iodone(pb, 0);
+		pagebuf_iodone(pb, 0, 0);
 	}
 
 	return 0;
@@ -1734,13 +1756,21 @@ pagebuf_daemon_start(void)
 {
 	int		rval;
 
-	pagebuf_workqueue = create_workqueue("pagebuf");
-	if (!pagebuf_workqueue)
+	pagebuf_logio_workqueue = create_workqueue("xfslogd");
+	if (!pagebuf_logio_workqueue)
 		return -ENOMEM;
 
+	pagebuf_dataio_workqueue = create_workqueue("xfsdatad");
+	if (!pagebuf_dataio_workqueue) {
+		destroy_workqueue(pagebuf_logio_workqueue);
+		return -ENOMEM;
+	}
+
 	rval = kernel_thread(pagebuf_daemon, NULL, CLONE_FS|CLONE_FILES);
-	if (rval < 0)
-		destroy_workqueue(pagebuf_workqueue);
+	if (rval < 0) {
+		destroy_workqueue(pagebuf_logio_workqueue);
+		destroy_workqueue(pagebuf_dataio_workqueue);
+	}
 
 	return rval;
 }
@@ -1756,7 +1786,8 @@ pagebuf_daemon_stop(void)
 	pbd_active = 0;
 	wake_up_interruptible(&pbd_waitq);
 	wait_event_interruptible(pbd_waitq, pbd_active);
-	destroy_workqueue(pagebuf_workqueue);
+	destroy_workqueue(pagebuf_logio_workqueue);
+	destroy_workqueue(pagebuf_dataio_workqueue);
 }
 
 

fs/xfs/pagebuf/page_buf.h

@@ -169,8 +169,8 @@ typedef page_buf_bmap_t pb_bmap_t;
  * This buffer structure is used by the page cache buffer management routines
  * to refer to an assembly of pages forming a logical buffer.  The actual
  * I/O is performed with buffer_head or bio structures, as required by drivers,
- * for drivers which do not understand this structure. The buffer structure is
- * used on temporary basis only, and discarded when released.  
+ * for drivers which do not understand this structure.	The buffer structure is
+ * used on temporary basis only, and discarded when released.
  *
  * The real data storage is recorded in the page cache.	 Metadata is
  * hashed to the inode for the block device on which the file system resides.
@@ -245,6 +245,13 @@ extern page_buf_t *pagebuf_get(		/* allocate a buffer		*/
 		page_buf_flags_t);	/* PBF_LOCK, PBF_READ,		*/
 					/* PBF_ASYNC			*/	
 
+extern page_buf_t *pagebuf_lookup(
+		struct pb_target *,
+		loff_t,			/* starting offset of range	*/
+		size_t,			/* length of range		*/
+		page_buf_flags_t);	/* PBF_READ, PBF_WRITE,		*/
+					/* PBF_FORCEIO, _PBF_LOCKABLE	*/
+
 extern page_buf_t *pagebuf_get_empty(	/* allocate pagebuf struct with */
 					/*  no memory or disk address	*/
 		struct pb_target *);	/* mount point "fake" inode	*/
@@ -300,6 +307,7 @@ static inline int pagebuf_geterror(page_buf_t *pb)
 
 extern void pagebuf_iodone(		/* mark buffer I/O complete	*/
 		page_buf_t *,		/* buffer to mark		*/
+		int,			/* use data/log helper thread.	*/
 		int);			/* run completion locally, or in
 					 * a helper thread. 		*/
 

fs/xfs/xfs_buf.h

@@ -163,15 +163,17 @@ extern inline xfs_caddr_t xfs_buf_offset(page_buf_t *bp, size_t offset)
 #define XFS_BUF_SET_PTR(bp, val, count)		\
 				pagebuf_associate_memory(bp, val, count)
 #define XFS_BUF_ADDR(bp)	((bp)->pb_bn)
-#define XFS_BUF_OFFSET(bp)	((bp)->pb_file_offset >> 9)
 #define XFS_BUF_SET_ADDR(bp, blk)		\
 			((bp)->pb_bn = (page_buf_daddr_t)(blk))
+#define XFS_BUF_OFFSET(bp)	((bp)->pb_file_offset)
+#define XFS_BUF_SET_OFFSET(bp, off)		\
+			((bp)->pb_file_offset = (off))
 #define XFS_BUF_COUNT(bp)	((bp)->pb_count_desired)
 #define XFS_BUF_SET_COUNT(bp, cnt)		\
-			((bp)->pb_count_desired = cnt)
+			((bp)->pb_count_desired = (cnt))
 #define XFS_BUF_SIZE(bp)	((bp)->pb_buffer_length)
 #define XFS_BUF_SET_SIZE(bp, cnt)		\
-			((bp)->pb_buffer_length = cnt)
+			((bp)->pb_buffer_length = (cnt))
 #define XFS_BUF_SET_VTYPE_REF(bp, type, ref)
 #define XFS_BUF_SET_VTYPE(bp, type)
 #define XFS_BUF_SET_REF(bp, ref)
@@ -242,7 +244,7 @@ static inline void	xfs_buf_relse(page_buf_t *bp)
 
 
 #define xfs_biodone(pb)		    \
-	    pagebuf_iodone(pb, 0)
+	    pagebuf_iodone(pb, 0, 0)
 
 #define xfs_incore(buftarg,blkno,len,lockit) \
 	    pagebuf_find(buftarg, blkno ,len, lockit)