From: Jinshan Xiong <jinshan.xiong@whamcloud.com>
Date: Tue, 12 Jun 2012 00:03:56 +0000 (-0700)
Subject: LU-1030 osc: new IO engine implementation
X-Git-Tag: 2.2.57~7
X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=commitdiff_plain;h=9fe4b52ad2ffadf125d9b5c78bb2ff9a01725707

LU-1030 osc: new IO engine implementation

New IO engine to manage dirty pages with osc_extent.

Osc_extent is a data structure to manage a series of contiguous
blocks; however, the pages in an extent is not required to be
contiguous. An extent must be written out in one RPC.

The purpose of introducing extents are:
1. make grants work for extent-based OSD;
2. form better IO by picking up contiguous pages to compose RPC;
3. reimplement ll_writepages() with CIT_FSYNC.

Signed-off-by: Jinshan Xiong <jinshan.xiong@whamcloud.com>
Change-Id: I3ef619c1f07eefd201236ab55e5fd858791d41e0
Reviewed-on: http://review.whamcloud.com/2270
Reviewed-by: Johann Lombardi <johann@whamcloud.com>
Tested-by: Hudson
Tested-by: Maloo <whamcloud.maloo@gmail.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
---

diff --git a/libcfs/include/libcfs/linux/libcfs.h b/libcfs/include/libcfs/linux/libcfs.h
index a6c8edd..3635a7f 100644
--- a/libcfs/include/libcfs/linux/libcfs.h
+++ b/libcfs/include/libcfs/linux/libcfs.h
@@ -62,6 +62,7 @@
 #include <linux/types.h>
 #include <asm/timex.h>
 #include <linux/sched.h> /* THREAD_SIZE */
+#include <linux/rbtree.h>
 
 #define CFS_THREAD_SIZE   THREAD_SIZE
 #define LUSTRE_TRACE_SIZE (THREAD_SIZE >> 5)
diff --git a/libcfs/include/libcfs/posix/libcfs.h b/libcfs/include/libcfs/posix/libcfs.h
index 57a6239..17d37ce 100644
--- a/libcfs/include/libcfs/posix/libcfs.h
+++ b/libcfs/include/libcfs/posix/libcfs.h
@@ -415,6 +415,78 @@ static inline void radix_tree_preload_end(void)
 {
 }
 
+/***************************************************************************
+ *
+ * Linux kernel red black tree emulation.
+ *
+ ***************************************************************************/
+struct rb_node {
+	unsigned long  rb_parent_color;
+#define	RB_RED		0
+#define	RB_BLACK	1
+	struct rb_node *rb_right;
+	struct rb_node *rb_left;
+};
+
+struct rb_root {
+	struct rb_node *rb_node;
+};
+
+
+#define rb_parent(r)   ((struct rb_node *)((r)->rb_parent_color & ~3))
+#define rb_color(r)   ((r)->rb_parent_color & 1)
+#define rb_is_red(r)   (!rb_color(r))
+#define rb_is_black(r) rb_color(r)
+#define rb_set_red(r)  do { (r)->rb_parent_color &= ~1; } while (0)
+#define rb_set_black(r)  do { (r)->rb_parent_color |= 1; } while (0)
+
+static inline void rb_set_parent(struct rb_node *rb, struct rb_node *p)
+{
+	rb->rb_parent_color = (rb->rb_parent_color & 3) | (unsigned long)p;
+}
+static inline void rb_set_color(struct rb_node *rb, int color)
+{
+	rb->rb_parent_color = (rb->rb_parent_color & ~1) | color;
+}
+
+#define RB_ROOT	((struct rb_root) { NULL, })
+#define rb_entry(ptr, type, member) container_of(ptr, type, member)
+
+#define RB_EMPTY_ROOT(root)	((root)->rb_node == NULL)
+#define RB_EMPTY_NODE(node)	(rb_parent(node) == node)
+#define RB_CLEAR_NODE(node)	(rb_set_parent(node, node))
+
+static inline void rb_init_node(struct rb_node *rb)
+{
+	rb->rb_parent_color = 0;
+	rb->rb_right = NULL;
+	rb->rb_left = NULL;
+	RB_CLEAR_NODE(rb);
+}
+
+extern void rb_insert_color(struct rb_node *, struct rb_root *);
+extern void rb_erase(struct rb_node *, struct rb_root *);
+
+/* Find logical next and previous nodes in a tree */
+extern struct rb_node *rb_next(const struct rb_node *);
+extern struct rb_node *rb_prev(const struct rb_node *);
+extern struct rb_node *rb_first(const struct rb_root *);
+extern struct rb_node *rb_last(const struct rb_root *);
+static inline void rb_link_node(struct rb_node *node, struct rb_node *parent,
+				struct rb_node **rb_link)
+{
+	node->rb_parent_color = (unsigned long)parent;
+	node->rb_left = node->rb_right = NULL;
+
+	*rb_link = node;
+}
+
+/***************************************************************************
+ *
+ * End of Linux kernel red black tree emulation.
+ *
+ ***************************************************************************/
+
 typedef ssize_t (*read_actor_t)();
 
 #define CFS_IFSHIFT 12
diff --git a/libcfs/libcfs/autoMakefile.am b/libcfs/libcfs/autoMakefile.am
index 7ce8133..3b143e7 100644
--- a/libcfs/libcfs/autoMakefile.am
+++ b/libcfs/libcfs/autoMakefile.am
@@ -44,7 +44,8 @@ if LIBLUSTRE
 noinst_LIBRARIES= libcfs.a
 libcfs_a_SOURCES= posix/posix-debug.c user-prim.c user-lock.c user-tcpip.c \
 		  prng.c user-bitops.c user-mem.c hash.c kernel_user_comm.c \
-		  workitem.c fail.c libcfs_cpu.c libcfs_mem.c libcfs_lock.c
+		  workitem.c fail.c libcfs_cpu.c libcfs_mem.c libcfs_lock.c \
+		  posix/rbtree.c
 libcfs_a_CPPFLAGS = $(LLCPPFLAGS)
 libcfs_a_CFLAGS = $(LLCFLAGS)
 endif
@@ -70,7 +71,7 @@ nodist_libcfs_SOURCES := darwin/darwin-sync.c darwin/darwin-mem.c	\
 	darwin/darwin-debug.c darwin/darwin-proc.c 			\
 	darwin/darwin-tracefile.c darwin/darwin-module.c 		\
 	posix/posix-debug.c module.c tracefile.c nidstrings.c watchdog.c \
-	kernel_user_comm.c hash.c
+	kernel_user_comm.c hash.c posix/rbtree.c
 
 libcfs_CFLAGS := $(EXTRA_KCFLAGS)
 libcfs_LDFLAGS := $(EXTRA_KLDFLAGS)
diff --git a/libcfs/libcfs/posix/rbtree.c b/libcfs/libcfs/posix/rbtree.c
new file mode 100644
index 0000000..9d0e5bb
--- /dev/null
+++ b/libcfs/libcfs/posix/rbtree.c
@@ -0,0 +1,333 @@
+/*
+  Red Black Trees
+  (C) 1999  Andrea Arcangeli <andrea@suse.de>
+  (C) 2002  David Woodhouse <dwmw2@infradead.org>
+
+  This program is free software; you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  This program is distributed in the hope that it will be useful,
+  but WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+  GNU General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with this program; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+
+  linux/lib/rbtree.c
+*/
+
+#include <libcfs/libcfs.h>
+
+static void __rb_rotate_left(struct rb_node *node, struct rb_root *root)
+{
+	struct rb_node *right = node->rb_right;
+	struct rb_node *parent = rb_parent(node);
+
+	node->rb_right = right->rb_left;
+	if (node->rb_right != NULL)
+		rb_set_parent(right->rb_left, node);
+	right->rb_left = node;
+
+	rb_set_parent(right, parent);
+
+	if (parent) {
+		if (node == parent->rb_left)
+			parent->rb_left = right;
+		else
+			parent->rb_right = right;
+	} else
+		root->rb_node = right;
+	rb_set_parent(node, right);
+}
+
+static void __rb_rotate_right(struct rb_node *node, struct rb_root *root)
+{
+	struct rb_node *left = node->rb_left;
+	struct rb_node *parent = rb_parent(node);
+
+	node->rb_left = left->rb_right;
+	if (node->rb_left != NULL)
+		rb_set_parent(left->rb_right, node);
+	left->rb_right = node;
+
+	rb_set_parent(left, parent);
+
+	if (parent) {
+		if (node == parent->rb_right)
+			parent->rb_right = left;
+		else
+			parent->rb_left = left;
+	} else
+		root->rb_node = left;
+	rb_set_parent(node, left);
+}
+
+void rb_insert_color(struct rb_node *node, struct rb_root *root)
+{
+	struct rb_node *parent, *gparent;
+
+	while ((parent = rb_parent(node)) != NULL && rb_is_red(parent)) {
+		gparent = rb_parent(parent);
+
+		if (parent == gparent->rb_left) {
+			register struct rb_node *uncle = gparent->rb_right;
+			if (uncle && rb_is_red(uncle)) {
+				rb_set_black(uncle);
+				rb_set_black(parent);
+				rb_set_red(gparent);
+				node = gparent;
+				continue;
+			}
+
+			if (parent->rb_right == node) {
+				register struct rb_node *tmp;
+				__rb_rotate_left(parent, root);
+				tmp = parent;
+				parent = node;
+				node = tmp;
+			}
+
+			rb_set_black(parent);
+			rb_set_red(gparent);
+			__rb_rotate_right(gparent, root);
+		} else {
+			register struct rb_node *uncle = gparent->rb_left;
+			if (uncle && rb_is_red(uncle)) {
+				rb_set_black(uncle);
+				rb_set_black(parent);
+				rb_set_red(gparent);
+				node = gparent;
+				continue;
+			}
+
+			if (parent->rb_left == node) {
+				register struct rb_node *tmp;
+				__rb_rotate_right(parent, root);
+				tmp = parent;
+				parent = node;
+				node = tmp;
+			}
+
+			rb_set_black(parent);
+			rb_set_red(gparent);
+			__rb_rotate_left(gparent, root);
+		}
+	}
+
+	rb_set_black(root->rb_node);
+}
+
+static void __rb_erase_color(struct rb_node *node, struct rb_node *parent,
+			     struct rb_root *root)
+{
+	struct rb_node *ptr;
+
+	while ((!node || rb_is_black(node)) && node != root->rb_node) {
+		if (parent->rb_left == node) {
+			ptr = parent->rb_right;
+			if (rb_is_red(ptr)) {
+				rb_set_black(ptr);
+				rb_set_red(parent);
+				__rb_rotate_left(parent, root);
+				ptr = parent->rb_right;
+			}
+			if ((!ptr->rb_left || rb_is_black(ptr->rb_left)) &&
+			    (!ptr->rb_right || rb_is_black(ptr->rb_right))) {
+				rb_set_red(ptr);
+				node = parent;
+				parent = rb_parent(node);
+			} else {
+				if (!ptr->rb_right ||
+				    rb_is_black(ptr->rb_right)) {
+					rb_set_black(ptr->rb_left);
+					rb_set_red(ptr);
+					__rb_rotate_right(ptr, root);
+					ptr = parent->rb_right;
+				}
+				rb_set_color(ptr, rb_color(parent));
+				rb_set_black(parent);
+				rb_set_black(ptr->rb_right);
+				__rb_rotate_left(parent, root);
+				node = root->rb_node;
+				break;
+			}
+		} else {
+			ptr = parent->rb_left;
+			if (rb_is_red(ptr)) {
+				rb_set_black(ptr);
+				rb_set_red(parent);
+				__rb_rotate_right(parent, root);
+				ptr = parent->rb_left;
+			}
+			if ((!ptr->rb_left || rb_is_black(ptr->rb_left)) &&
+			    (!ptr->rb_right || rb_is_black(ptr->rb_right))) {
+				rb_set_red(ptr);
+				node = parent;
+				parent = rb_parent(node);
+			} else {
+				if (!ptr->rb_left ||
+				    rb_is_black(ptr->rb_left)) {
+					rb_set_black(ptr->rb_right);
+					rb_set_red(ptr);
+					__rb_rotate_left(ptr, root);
+					ptr = parent->rb_left;
+				}
+				rb_set_color(ptr, rb_color(parent));
+				rb_set_black(parent);
+				rb_set_black(ptr->rb_left);
+				__rb_rotate_right(parent, root);
+				node = root->rb_node;
+				break;
+			}
+		}
+	}
+	if (node)
+		rb_set_black(node);
+}
+
+void rb_erase(struct rb_node *node, struct rb_root *root)
+{
+	struct rb_node *child, *parent;
+	int color;
+
+	if (!node->rb_left)
+		child = node->rb_right;
+	else if (!node->rb_right)
+		child = node->rb_left;
+	else {
+		struct rb_node *old = node, *left;
+
+		node = node->rb_right;
+		while ((left = node->rb_left) != NULL)
+			node = left;
+
+		if (rb_parent(old)) {
+			if (rb_parent(old)->rb_left == old)
+				rb_parent(old)->rb_left = node;
+			else
+				rb_parent(old)->rb_right = node;
+		} else
+			root->rb_node = node;
+
+		child = node->rb_right;
+		parent = rb_parent(node);
+		color = rb_color(node);
+
+		if (parent == old) {
+			parent = node;
+		} else {
+			if (child)
+				rb_set_parent(child, parent);
+			parent->rb_left = child;
+
+			node->rb_right = old->rb_right;
+			rb_set_parent(old->rb_right, node);
+		}
+
+		node->rb_parent_color = old->rb_parent_color;
+		node->rb_left = old->rb_left;
+		rb_set_parent(old->rb_left, node);
+
+		goto color;
+	}
+
+	parent = rb_parent(node);
+	color = rb_color(node);
+
+	if (child)
+		rb_set_parent(child, parent);
+	if (parent) {
+		if (parent->rb_left == node)
+			parent->rb_left = child;
+		else
+			parent->rb_right = child;
+	} else
+		root->rb_node = child;
+
+ color:
+	if (color == RB_BLACK)
+		__rb_erase_color(child, parent, root);
+}
+
+/*
+ * This function returns the first node (in sort order) of the tree.
+ */
+struct rb_node *rb_first(const struct rb_root *root)
+{
+	struct rb_node	*n;
+
+	n = root->rb_node;
+	if (!n)
+		return NULL;
+	while (n->rb_left)
+		n = n->rb_left;
+	return n;
+}
+
+struct rb_node *rb_last(const struct rb_root *root)
+{
+	struct rb_node	*n;
+
+	n = root->rb_node;
+	if (!n)
+		return NULL;
+	while (n->rb_right)
+		n = n->rb_right;
+	return n;
+}
+
+struct rb_node *rb_next(const struct rb_node *node)
+{
+	struct rb_node *parent;
+
+	if (rb_parent(node) == node)
+		return NULL;
+
+	/* If we have a right-hand child, go down and then left as far
+	   as we can. */
+	if (node->rb_right) {
+		node = node->rb_right;
+		while (node->rb_left)
+			node = node->rb_left;
+		return (struct rb_node *)node;
+	}
+
+	/* No right-hand children.  Everything down and left is
+	   smaller than us, so any 'next' node must be in the general
+	   direction of our parent. Go up the tree; any time the
+	   ancestor is a right-hand child of its parent, keep going
+	   up. First time it's a left-hand child of its parent, said
+	   parent is our 'next' node. */
+	while ((parent = rb_parent(node)) && node == parent->rb_right)
+		node = parent;
+
+	return parent;
+}
+
+struct rb_node *rb_prev(const struct rb_node *node)
+{
+	struct rb_node *parent;
+
+	if (rb_parent(node) == node)
+		return NULL;
+
+	/* If we have a left-hand child, go down and then right as far
+	   as we can. */
+	if (node->rb_left) {
+		node = node->rb_left;
+		while (node->rb_right)
+			node = node->rb_right;
+		return (struct rb_node *)node;
+	}
+
+	/* No left-hand children. Go up till we find an ancestor which
+	   is a right-hand child of its parent */
+	while ((parent = rb_parent(node)) && node == parent->rb_left)
+		node = parent;
+
+	return parent;
+}
diff --git a/lustre/include/cl_object.h b/lustre/include/cl_object.h
index 1acd1ba..53bb5ca 100644
--- a/lustre/include/cl_object.h
+++ b/lustre/include/cl_object.h
@@ -1054,6 +1054,15 @@ struct cl_page_operations {
          */
         int (*cpo_cancel)(const struct lu_env *env,
                           const struct cl_page_slice *slice);
+	/**
+	 * Write out a page by kernel. This is only called by ll_writepage
+	 * right now.
+	 *
+	 * \see cl_page_flush()
+	 */
+	int (*cpo_flush)(const struct lu_env *env,
+			 const struct cl_page_slice *slice,
+			 struct cl_io *io);
         /** @} transfer */
 };
 
@@ -1960,11 +1969,6 @@ enum cl_io_state {
         CIS_FINI
 };
 
-enum cl_req_priority {
-        CRP_NORMAL,
-        CRP_CANCEL
-};
-
 /**
  * IO state private for a layer.
  *
@@ -2082,8 +2086,7 @@ struct cl_io_operations {
                 int  (*cio_submit)(const struct lu_env *env,
                                    const struct cl_io_slice *slice,
                                    enum cl_req_type crt,
-                                   struct cl_2queue *queue,
-                                   enum cl_req_priority priority);
+				   struct cl_2queue *queue);
         } req_op[CRT_NR];
         /**
          * Read missing page.
@@ -2245,6 +2248,18 @@ enum cl_io_lock_dmd {
         CILR_PEEK
 };
 
+enum cl_fsync_mode {
+	/** start writeback, do not wait for them to finish */
+	CL_FSYNC_NONE  = 0,
+	/** start writeback and wait for them to finish */
+	CL_FSYNC_LOCAL = 1,
+	/** discard all of dirty pages in a specific file range */
+	CL_FSYNC_DISCARD = 2,
+	/** start writeback and make sure they have reached storage before
+	 * return. OST_SYNC RPC must be issued and finished */
+	CL_FSYNC_ALL   = 3
+};
+
 struct cl_io_rw_common {
         loff_t      crw_pos;
         size_t      crw_count;
@@ -2291,6 +2306,7 @@ struct cl_io {
                 struct cl_wr_io {
                         struct cl_io_rw_common wr;
                         int                    wr_append;
+			int                    wr_sync;
                 } ci_wr;
                 struct cl_io_rw_common ci_rw;
                 struct cl_setattr_io {
@@ -2318,6 +2334,9 @@ struct cl_io {
 			struct obd_capa   *fi_capa;
 			/** file system level fid */
 			struct lu_fid     *fi_fid;
+			enum cl_fsync_mode fi_mode;
+			/* how many pages were written/discarded */
+			unsigned int       fi_nr_written;
 		} ci_fsync;
         } u;
         struct cl_2queue     ci_queue;
@@ -2769,6 +2788,8 @@ int  cl_page_cache_add  (const struct lu_env *env, struct cl_io *io,
 void cl_page_clip       (const struct lu_env *env, struct cl_page *pg,
                          int from, int to);
 int  cl_page_cancel     (const struct lu_env *env, struct cl_page *page);
+int  cl_page_flush      (const struct lu_env *env, struct cl_io *io,
+			 struct cl_page *pg);
 
 /** @} transfer */
 
@@ -2815,9 +2836,19 @@ struct cl_lock *cl_lock_peek(const struct lu_env *env, const struct cl_io *io,
 struct cl_lock *cl_lock_request(const struct lu_env *env, struct cl_io *io,
                                 const struct cl_lock_descr *need,
                                 const char *scope, const void *source);
-struct cl_lock *cl_lock_at_page(const struct lu_env *env, struct cl_object *obj,
-                                struct cl_page *page, struct cl_lock *except,
-                                int pending, int canceld);
+struct cl_lock *cl_lock_at_pgoff(const struct lu_env *env,
+				 struct cl_object *obj, pgoff_t index,
+				 struct cl_lock *except, int pending,
+				 int canceld);
+static inline struct cl_lock *cl_lock_at_page(const struct lu_env *env,
+					      struct cl_object *obj,
+					      struct cl_page *page,
+					      struct cl_lock *except,
+					      int pending, int canceld)
+{
+	return cl_lock_at_pgoff(env, obj, page->cp_index, except,
+				pending, canceld);
+}
 
 const struct cl_lock_slice *cl_lock_at(const struct cl_lock *lock,
                                        const struct lu_device_type *dtype);
@@ -2899,8 +2930,7 @@ int  cl_lock_mutex_try  (const struct lu_env *env, struct cl_lock *lock);
 void cl_lock_mutex_put  (const struct lu_env *env, struct cl_lock *lock);
 int  cl_lock_is_mutexed (struct cl_lock *lock);
 int  cl_lock_nr_mutexed (const struct lu_env *env);
-int  cl_lock_page_out   (const struct lu_env *env, struct cl_lock *lock,
-                         int discard);
+int  cl_lock_discard_pages(const struct lu_env *env, struct cl_lock *lock);
 int  cl_lock_ext_match  (const struct cl_lock_descr *has,
                          const struct cl_lock_descr *need);
 int  cl_lock_descr_match(const struct cl_lock_descr *has,
@@ -2958,11 +2988,10 @@ int   cl_io_prepare_write(const struct lu_env *env, struct cl_io *io,
 int   cl_io_commit_write (const struct lu_env *env, struct cl_io *io,
                           struct cl_page *page, unsigned from, unsigned to);
 int   cl_io_submit_rw    (const struct lu_env *env, struct cl_io *io,
-                          enum cl_req_type iot, struct cl_2queue *queue,
-                          enum cl_req_priority priority);
+			  enum cl_req_type iot, struct cl_2queue *queue);
 int   cl_io_submit_sync  (const struct lu_env *env, struct cl_io *io,
-                          enum cl_req_type iot, struct cl_2queue *queue,
-                          enum cl_req_priority priority, long timeout);
+			  enum cl_req_type iot, struct cl_2queue *queue,
+			  long timeout);
 void  cl_io_rw_advance   (const struct lu_env *env, struct cl_io *io,
                           size_t nob);
 int   cl_io_cancel       (const struct lu_env *env, struct cl_io *io,
@@ -2977,6 +3006,16 @@ static inline int cl_io_is_append(const struct cl_io *io)
         return io->ci_type == CIT_WRITE && io->u.ci_wr.wr_append;
 }
 
+static inline int cl_io_is_sync_write(const struct cl_io *io)
+{
+	return io->ci_type == CIT_WRITE && io->u.ci_wr.wr_sync;
+}
+
+static inline int cl_io_is_mkwrite(const struct cl_io *io)
+{
+	return io->ci_type == CIT_FAULT && io->u.ci_fault.ft_mkwrite;
+}
+
 /**
  * True, iff \a io is a truncate(2).
  */
diff --git a/lustre/include/linux/obd.h b/lustre/include/linux/obd.h
index 6b78f62..5c8024f 100644
--- a/lustre/include/linux/obd.h
+++ b/lustre/include/linux/obd.h
@@ -54,6 +54,7 @@
 # include <linux/lustre_intent.h>
 #endif
 
+#define CLIENT_OBD_LIST_LOCK_DEBUG 1
 typedef struct {
         cfs_spinlock_t          lock;
 
diff --git a/lustre/include/lustre_net.h b/lustre/include/lustre_net.h
index 0ab359f..ff325fb 100644
--- a/lustre/include/lustre_net.h
+++ b/lustre/include/lustre_net.h
@@ -244,7 +244,7 @@ union ptlrpc_async_args {
          * least big enough for that.
          */
         void      *pointer_arg[11];
-        __u64      space[6];
+	__u64      space[7];
 };
 
 struct ptlrpc_request_set;
diff --git a/lustre/include/obd.h b/lustre/include/obd.h
index 61e6649..07ef9af 100644
--- a/lustre/include/obd.h
+++ b/lustre/include/obd.h
@@ -453,10 +453,22 @@ struct client_obd {
         long                     cl_dirty_transit; /* dirty synchronous */
         long                     cl_avail_grant;   /* bytes of credit for ost */
         long                     cl_lost_grant;    /* lost credits (trunc) */
-        cfs_list_t               cl_cache_waiters; /* waiting for cache/grant */
-        cfs_time_t               cl_next_shrink_grant;   /* jiffies */
-        cfs_list_t               cl_grant_shrink_list;  /* Timeout event list */
-        int                      cl_grant_shrink_interval; /* seconds */
+
+	/* since we allocate grant by blocks, we don't know how many grant will
+	 * be used to add a page into cache. As a solution, we reserve maximum
+	 * grant before trying to dirty a page and unreserve the rest.
+	 * See osc_{reserve|unreserve}_grant for details. */
+	long                 cl_reserved_grant;
+	cfs_list_t           cl_cache_waiters; /* waiting for cache/grant */
+	cfs_time_t           cl_next_shrink_grant;   /* jiffies */
+	cfs_list_t           cl_grant_shrink_list;  /* Timeout event list */
+	int                  cl_grant_shrink_interval; /* seconds */
+
+	/* A chunk is an optimal size used by osc_extent to determine
+	 * the extent size. A chunk is max(CFS_PAGE_SIZE, OST block size) */
+	int                  cl_chunkbits;
+	int                  cl_chunk;
+	int                  cl_extent_tax; /* extent overhead, by bytes */
 
         /* keep track of objects that have lois that contain pages which
          * have been queued for async brw.  this lock also protects the
@@ -478,7 +490,7 @@ struct client_obd {
 	 *
 	 * NB by Jinshan: though field names are still _loi_, but actually
 	 * osc_object{}s are in the list.
-         */
+	 */
         client_obd_lock_t        cl_loi_list_lock;
         cfs_list_t               cl_loi_ready_list;
         cfs_list_t               cl_loi_hp_ready_list;
@@ -487,9 +499,9 @@ struct client_obd {
         int                      cl_r_in_flight;
         int                      cl_w_in_flight;
         /* just a sum of the loi/lop pending numbers to be exported by /proc */
-        int                      cl_pending_w_pages;
-        int                      cl_pending_r_pages;
-        int                      cl_max_pages_per_rpc;
+	cfs_atomic_t             cl_pending_w_pages;
+	cfs_atomic_t             cl_pending_r_pages;
+	int                      cl_max_pages_per_rpc;
         int                      cl_max_rpcs_in_flight;
         struct obd_histogram     cl_read_rpc_hist;
         struct obd_histogram     cl_write_rpc_hist;
diff --git a/lustre/include/obd_ost.h b/lustre/include/obd_ost.h
index 17d9c03..761d780 100644
--- a/lustre/include/obd_ost.h
+++ b/lustre/include/obd_ost.h
@@ -54,6 +54,7 @@ struct osc_brw_async_args {
         struct brw_page  **aa_ppga;
         struct client_obd *aa_cli;
         cfs_list_t         aa_oaps;
+	cfs_list_t         aa_exts;
         struct obd_capa   *aa_ocapa;
         struct cl_req     *aa_clerq;
 };
diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c
index 213e94d..31d55f4 100644
--- a/lustre/ldlm/ldlm_lib.c
+++ b/lustre/ldlm/ldlm_lib.c
@@ -313,6 +313,8 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
         CFS_INIT_LIST_HEAD(&cli->cl_loi_write_list);
         CFS_INIT_LIST_HEAD(&cli->cl_loi_read_list);
         client_obd_list_lock_init(&cli->cl_loi_list_lock);
+	cfs_atomic_set(&cli->cl_pending_w_pages, 0);
+	cfs_atomic_set(&cli->cl_pending_r_pages, 0);
         cli->cl_r_in_flight = 0;
         cli->cl_w_in_flight = 0;
 
diff --git a/lustre/ldlm/ldlm_resource.c b/lustre/ldlm/ldlm_resource.c
index 3fc1345..c484953 100644
--- a/lustre/ldlm/ldlm_resource.c
+++ b/lustre/ldlm/ldlm_resource.c
@@ -460,7 +460,6 @@ static void ldlm_res_hop_get_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 
         res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
         ldlm_resource_getref(res);
-        LDLM_RESOURCE_ADDREF(res);
 }
 
 static void ldlm_res_hop_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
@@ -469,7 +468,6 @@ static void ldlm_res_hop_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 
         res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
         /* cfs_hash_for_each_nolock is the only chance we call it */
-        LDLM_RESOURCE_DELREF(res);
         ldlm_resource_putref_locked(res);
 }
 
@@ -478,7 +476,6 @@ static void ldlm_res_hop_put(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
         struct ldlm_resource *res;
 
         res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
-        LDLM_RESOURCE_DELREF(res);
         ldlm_resource_putref(res);
 }
 
diff --git a/lustre/liblustre/llite_cl.c b/lustre/liblustre/llite_cl.c
index d6fbf12..d6fc898 100644
--- a/lustre/liblustre/llite_cl.c
+++ b/lustre/liblustre/llite_cl.c
@@ -558,7 +558,7 @@ static int llu_queue_pio(const struct lu_env *env, struct cl_io *io,
         if (rc == 0) {
                 enum cl_req_type iot;
                 iot = io->ci_type == CIT_READ ? CRT_READ : CRT_WRITE;
-                rc = cl_io_submit_sync(env, io, iot, queue, CRP_NORMAL, 0);
+		rc = cl_io_submit_sync(env, io, iot, queue, 0);
         }
 
         group->lig_rc = rc;
diff --git a/lustre/llite/file.c b/lustre/llite/file.c
index 109d3b9..cc6214c 100644
--- a/lustre/llite/file.c
+++ b/lustre/llite/file.c
@@ -809,8 +809,10 @@ void ll_io_init(struct cl_io *io, const struct file *file, int write)
         struct inode *inode = file->f_dentry->d_inode;
 
         io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
-        if (write)
-                io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
+	if (write) {
+		io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
+		io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || IS_SYNC(inode);
+	}
         io->ci_obj     = ll_i2info(inode)->lli_clob;
         io->ci_lockreq = CILR_MAYBE;
         if (ll_file_nolock(file)) {
@@ -1985,8 +1987,11 @@ int ll_flush(struct file *file)
 /**
  * Called to make sure a portion of file has been written out.
  * if @local_only is not true, it will send OST_SYNC RPCs to ost.
+ *
+ * Return how many pages have been written.
  */
-int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end)
+int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
+		       enum cl_fsync_mode mode)
 {
 	struct cl_env_nest nest;
 	struct lu_env *env;
@@ -1996,6 +2001,10 @@ int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end)
 	int result;
 	ENTRY;
 
+	if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
+	    mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
+		RETURN(-EINVAL);
+
 	env = cl_env_nested_get(&nest);
 	if (IS_ERR(env))
 		RETURN(PTR_ERR(env));
@@ -2011,11 +2020,15 @@ int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end)
 	fio->fi_start = start;
 	fio->fi_end = end;
 	fio->fi_fid = ll_inode2fid(inode);
+	fio->fi_mode = mode;
+	fio->fi_nr_written = 0;
 
 	if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
 		result = cl_io_loop(env, io);
 	else
 		result = io->ci_result;
+	if (result == 0)
+		result = fio->fi_nr_written;
 	cl_io_fini(env, io);
 	cl_env_nested_put(&nest, env);
 
@@ -2071,8 +2084,9 @@ int ll_fsync(struct file *file, struct dentry *dentry, int data)
                 ptlrpc_req_finished(req);
 
         if (data && lsm) {
-		err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF);
-                if (!rc)
+		err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
+					 CL_FSYNC_ALL);
+		if (rc == 0 && err < 0)
                         rc = err;
                 lli->lli_write_rc = rc < 0 ? rc : 0;
         }
diff --git a/lustre/llite/llite_internal.h b/lustre/llite/llite_internal.h
index 2ecdf02..cdf30ed 100644
--- a/lustre/llite/llite_internal.h
+++ b/lustre/llite/llite_internal.h
@@ -679,6 +679,7 @@ struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
 int ll_prepare_write(struct file *, struct page *, unsigned from, unsigned to);
 int ll_commit_write(struct file *, struct page *, unsigned from, unsigned to);
 int ll_writepage(struct page *page, struct writeback_control *wbc);
+int ll_writepages(struct address_space *, struct writeback_control *wbc);
 void ll_removepage(struct page *page);
 int ll_readpage(struct file *file, struct page *page);
 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
@@ -1407,7 +1408,8 @@ static inline int cl_merge_lvb(struct inode *inode)
 
 struct obd_capa *cl_capa_lookup(struct inode *inode, enum cl_req_type crt);
 
-int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end);
+int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
+		       enum cl_fsync_mode mode);
 
 /** direct write pages */
 struct ll_dio_pages {
diff --git a/lustre/llite/llite_lib.c b/lustre/llite/llite_lib.c
index 78ba6bb..050a9a3 100644
--- a/lustre/llite/llite_lib.c
+++ b/lustre/llite/llite_lib.c
@@ -1839,13 +1839,19 @@ void ll_read_inode2(struct inode *inode, void *opaque)
 
 void ll_delete_inode(struct inode *inode)
 {
-        struct ll_sb_info *sbi = ll_i2sbi(inode);
-        int rc;
-        ENTRY;
-
-        rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
-        if (rc)
-                CERROR("fid_delete() failed, rc %d\n", rc);
+	struct cl_inode_info *lli = cl_i2info(inode);
+	struct ll_sb_info    *sbi = ll_i2sbi(inode);
+	int rc;
+	ENTRY;
+
+	rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
+	if (rc)
+		CERROR("fid_delete() failed, rc %d\n", rc);
+
+	if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
+		/* discard all dirty pages before truncating them, required by
+		 * osc_extent implementation at LU-1030. */
+		cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
 
         truncate_inode_pages(&inode->i_data, 0);
 
diff --git a/lustre/llite/rw.c b/lustre/llite/rw.c
index e0909f5..db4d451 100644
--- a/lustre/llite/rw.c
+++ b/lustre/llite/rw.c
@@ -141,7 +141,21 @@ static struct ll_cl_context *ll_cl_init(struct file *file,
         cio = ccc_env_io(env);
         io = cio->cui_cl.cis_io;
         if (io == NULL && create) {
-                loff_t pos;
+		struct inode *inode = vmpage->mapping->host;
+		loff_t pos;
+
+		if (TRYLOCK_INODE_MUTEX(inode)) {
+			UNLOCK_INODE_MUTEX(inode);
+
+			/* this is too bad. Someone is trying to write the
+			 * page w/o holding inode mutex. This means we can
+			 * add dirty pages into cache during truncate */
+			CERROR("Proc %s is dirting page w/o inode lock, this"
+			       "will break truncate.\n", cfs_current()->comm);
+			libcfs_debug_dumpstack(NULL);
+			LBUG();
+			return ERR_PTR(-EIO);
+		}
 
                 /*
                  * Loop-back driver calls ->prepare_write() and ->sendfile()
@@ -1152,8 +1166,8 @@ int ll_writepage(struct page *vmpage, struct writeback_control *wbc)
         struct cl_io           *io;
         struct cl_page         *page;
         struct cl_object       *clob;
-        struct cl_2queue       *queue;
         struct cl_env_nest      nest;
+	int redirtied = 0;
         int result;
         ENTRY;
 
@@ -1167,7 +1181,6 @@ int ll_writepage(struct page *vmpage, struct writeback_control *wbc)
         if (IS_ERR(env))
                 RETURN(PTR_ERR(env));
 
-        queue = &vvp_env_info(env)->vti_queue;
         clob  = ll_i2info(inode)->lli_clob;
         LASSERT(clob != NULL);
 
@@ -1181,41 +1194,90 @@ int ll_writepage(struct page *vmpage, struct writeback_control *wbc)
                         lu_ref_add(&page->cp_reference, "writepage",
                                    cfs_current());
                         cl_page_assume(env, io, page);
-                        /*
-                         * Mark page dirty, because this is what
-                         * ->vio_submit()->cpo_prep_write() assumes.
-                         *
-                         * XXX better solution is to detect this from within
-                         * cl_io_submit_rw() somehow.
-                         */
-                        set_page_dirty(vmpage);
-                        cl_2queue_init_page(queue, page);
-                        result = cl_io_submit_rw(env, io, CRT_WRITE,
-                                                 queue, CRP_NORMAL);
-                        if (result != 0) {
-                                /*
-                                 * Re-dirty page on error so it retries write,
-                                 * but not in case when IO has actually
-                                 * occurred and completed with an error.
-                                 */
-                                if (!PageError(vmpage)) {
-                                        redirty_page_for_writepage(wbc, vmpage);
-                                        result = 0;
-                                }
-                        }
-                        cl_page_list_disown(env, io, &queue->c2_qin);
-                        LASSERT(!cl_page_is_owned(page, io));
+			result = cl_page_flush(env, io, page);
+			if (result != 0) {
+				/*
+				 * Re-dirty page on error so it retries write,
+				 * but not in case when IO has actually
+				 * occurred and completed with an error.
+				 */
+				if (!PageError(vmpage)) {
+					redirty_page_for_writepage(wbc, vmpage);
+					result = 0;
+					redirtied = 1;
+				}
+			}
+			cl_page_disown(env, io, page);
                         lu_ref_del(&page->cp_reference,
                                    "writepage", cfs_current());
                         cl_page_put(env, page);
-                        cl_2queue_fini(env, queue);
                 }
         }
         cl_io_fini(env, io);
+
+	if (redirtied && wbc->sync_mode == WB_SYNC_ALL) {
+		loff_t offset = cl_offset(clob, vmpage->index);
+
+		/* Flush page failed because the extent is being written out.
+		 * Wait for the write of extent to be finished to avoid
+		 * breaking kernel which assumes ->writepage should mark
+		 * PageWriteback or clean the page. */
+		result = cl_sync_file_range(inode, offset,
+					    offset + CFS_PAGE_SIZE - 1,
+					    CL_FSYNC_LOCAL);
+		if (result > 0) {
+			/* actually we may have written more than one page.
+			 * decreasing this page because the caller will count
+			 * it. */
+			wbc->nr_to_write -= result - 1;
+			result = 0;
+		}
+	}
+
         cl_env_nested_put(&nest, env);
         RETURN(result);
 }
 
+int ll_writepages(struct address_space *mapping, struct writeback_control *wbc)
+{
+	struct inode *inode = mapping->host;
+	loff_t start;
+	loff_t end;
+	enum cl_fsync_mode mode;
+	int range_whole = 0;
+	int result;
+	ENTRY;
+
+	if (wbc->range_cyclic) {
+		start = mapping->writeback_index << CFS_PAGE_SHIFT;
+		end = OBD_OBJECT_EOF;
+	} else {
+		start = wbc->range_start;
+		end = wbc->range_end;
+		if (end == LLONG_MAX) {
+			end = OBD_OBJECT_EOF;
+			range_whole = start == 0;
+		}
+	}
+
+	mode = CL_FSYNC_NONE;
+	if (wbc->sync_mode == WB_SYNC_ALL)
+		mode = CL_FSYNC_LOCAL;
+
+	result = cl_sync_file_range(inode, start, end, mode);
+	if (result > 0) {
+		wbc->nr_to_write -= result;
+		result = 0;
+	 }
+
+	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) {
+		if (end == OBD_OBJECT_EOF)
+			end = i_size_read(inode);
+		mapping->writeback_index = (end >> CFS_PAGE_SHIFT) + 1;
+	}
+	RETURN(result);
+}
+
 int ll_readpage(struct file *file, struct page *vmpage)
 {
         struct ll_cl_context *lcc;
diff --git a/lustre/llite/rw26.c b/lustre/llite/rw26.c
index 661ae5b..6335b7f 100644
--- a/lustre/llite/rw26.c
+++ b/lustre/llite/rw26.c
@@ -332,7 +332,7 @@ ssize_t ll_direct_rw_pages(const struct lu_env *env, struct cl_io *io,
         if (rc == 0 && io_pages) {
                 rc = cl_io_submit_sync(env, io,
                                        rw == READ ? CRT_READ : CRT_WRITE,
-                                       queue, CRP_NORMAL, 0);
+				       queue, 0);
         }
         if (rc == 0)
                 rc = pv->ldp_size;
@@ -553,7 +553,7 @@ struct address_space_operations ll_aops = {
 //        .readpages      = ll_readpages,
         .direct_IO      = ll_direct_IO_26,
         .writepage      = ll_writepage,
-        .writepages     = generic_writepages,
+	.writepages     = ll_writepages,
         .set_page_dirty = ll_set_page_dirty,
 #ifdef HAVE_KERNEL_WRITE_BEGIN_END
         .write_begin    = ll_write_begin,
@@ -575,7 +575,7 @@ struct address_space_operations_ext ll_aops = {
 //        .orig_aops.readpages      = ll_readpages,
         .orig_aops.direct_IO      = ll_direct_IO_26,
         .orig_aops.writepage      = ll_writepage,
-        .orig_aops.writepages     = generic_writepages,
+	.orig_aops.writepages     = ll_writepages,
         .orig_aops.set_page_dirty = ll_set_page_dirty,
         .orig_aops.prepare_write  = ll_prepare_write,
         .orig_aops.commit_write   = ll_commit_write,
diff --git a/lustre/llite/vvp_io.c b/lustre/llite/vvp_io.c
index 4a65b9e..f534184 100644
--- a/lustre/llite/vvp_io.c
+++ b/lustre/llite/vvp_io.c
@@ -36,6 +36,7 @@
  * Implementation of cl_io for VVP layer.
  *
  *   Author: Nikita Danilov <nikita.danilov@sun.com>
+ *   Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
  */
 
 #define DEBUG_SUBSYSTEM S_LLITE
@@ -347,44 +348,8 @@ static int vvp_io_setattr_trunc(const struct lu_env *env,
                                 const struct cl_io_slice *ios,
                                 struct inode *inode, loff_t size)
 {
-        struct vvp_io        *vio   = cl2vvp_io(env, ios);
-        struct cl_io         *io    = ios->cis_io;
-        struct cl_object     *obj   = ios->cis_obj;
-        pgoff_t               start = cl_index(obj, size);
-        int                   result;
-
-        DOWN_WRITE_I_ALLOC_SEM(inode);
-
-        result = vvp_do_vmtruncate(inode, size);
-
-        /*
-         * If a page is partially truncated, keep it owned across truncate to
-         * prevent... races.
-         *
-         * XXX this properly belongs to osc, because races in question are OST
-         * specific.
-         */
-        if (cl_offset(obj, start) != size) {
-                struct cl_object_header *hdr;
-
-                hdr = cl_object_header(obj);
-                cfs_spin_lock(&hdr->coh_page_guard);
-                vio->cui_partpage = cl_page_lookup(hdr, start);
-                cfs_spin_unlock(&hdr->coh_page_guard);
-
-                if (vio->cui_partpage != NULL)
-                        /*
-                         * Wait for the transfer completion for a partially
-                         * truncated page to avoid dead-locking an OST with
-                         * the concurrent page-wise overlapping WRITE and
-                         * PUNCH requests. BUG:17397.
-                         *
-                         * Partial page is disowned in vvp_io_trunc_end().
-                         */
-                        cl_page_own(env, io, vio->cui_partpage);
-        } else
-                vio->cui_partpage = NULL;
-        return result;
+	DOWN_WRITE_I_ALLOC_SEM(inode);
+	return 0;
 }
 
 static int vvp_io_setattr_time(const struct lu_env *env,
@@ -434,23 +399,15 @@ static int vvp_io_setattr_start(const struct lu_env *env,
 static void vvp_io_setattr_end(const struct lu_env *env,
                                const struct cl_io_slice *ios)
 {
-        struct vvp_io        *vio   = cl2vvp_io(env, ios);
         struct cl_io         *io    = ios->cis_io;
         struct inode         *inode = ccc_object_inode(io->ci_obj);
 
         if (!cl_io_is_trunc(io))
                 return;
-        if (vio->cui_partpage != NULL) {
-                cl_page_disown(env, ios->cis_io, vio->cui_partpage);
-                cl_page_put(env, vio->cui_partpage);
-                vio->cui_partpage = NULL;
-        }
 
-        /*
-         * Do vmtruncate again, to remove possible stale pages populated by
-         * competing read threads. bz20645.
-         */
-        vvp_do_vmtruncate(inode, io->u.ci_setattr.sa_attr.lvb_size);
+	/* Truncate in memory pages - they must be clean pages because osc
+	 * has already notified to destroy osc_extents. */
+	vvp_do_vmtruncate(inode, io->u.ci_setattr.sa_attr.lvb_size);
 }
 
 static void vvp_io_setattr_fini(const struct lu_env *env,
@@ -721,6 +678,11 @@ static int vvp_io_fault_start(const struct lu_env *env,
 
         /* must return locked page */
         if (fio->ft_mkwrite) {
+		/* we grab alloc_sem to exclude truncate case.
+		 * Otherwise, we could add dirty pages into osc cache
+		 * while truncate is on-going. */
+		DOWN_READ_I_ALLOC_SEM(inode);
+
                 LASSERT(cfio->ft_vmpage != NULL);
                 lock_page(cfio->ft_vmpage);
         } else {
@@ -766,15 +728,22 @@ static int vvp_io_fault_start(const struct lu_env *env,
                          * started before the page is really made dirty, we
                          * still have chance to detect it. */
                         result = cl_page_cache_add(env, io, page, CRT_WRITE);
-                        if (result < 0) {
-                                cl_page_unassume(env, io, page);
-                                cl_page_put(env, page);
-
-                                /* we're in big trouble, what can we do now? */
-                                if (result == -EDQUOT)
-                                        result = -ENOSPC;
-                                GOTO(out, result);
-                        }
+			LASSERT(cl_page_is_owned(page, io));
+
+			vmpage = NULL;
+			if (result < 0) {
+				cl_page_unmap(env, io, page);
+				cl_page_discard(env, io, page);
+				cl_page_disown(env, io, page);
+
+				cl_page_put(env, page);
+
+				/* we're in big trouble, what can we do now? */
+				if (result == -EDQUOT)
+					result = -ENOSPC;
+				GOTO(out, result);
+			} else
+				cl_page_disown(env, io, page);
                 }
         }
 
@@ -795,18 +764,23 @@ static int vvp_io_fault_start(const struct lu_env *env,
 
 out:
         /* return unlocked vmpage to avoid deadlocking */
-        unlock_page(vmpage);
+	if (vmpage != NULL)
+		unlock_page(vmpage);
+	if (fio->ft_mkwrite)
+		UP_READ_I_ALLOC_SEM(inode);
 #ifdef HAVE_VM_OP_FAULT
-        cfio->fault.ft_flags &= ~VM_FAULT_LOCKED;
+	cfio->fault.ft_flags &= ~VM_FAULT_LOCKED;
 #endif
-        return result;
+	return result;
 }
 
-static void vvp_io_fsync_end(const struct lu_env *env,
-			     const struct cl_io_slice *ios)
+static int vvp_io_fsync_start(const struct lu_env *env,
+			      const struct cl_io_slice *ios)
 {
-	/* never try to verify there is no dirty pages in sync range
-	 * because page_mkwrite() can generate new dirty pages any time. */
+	/* we should mark TOWRITE bit to each dirty page in radix tree to
+	 * verify pages have been written, but this is difficult because of
+	 * race. */
+	return 0;
 }
 
 static int vvp_io_read_page(const struct lu_env *env,
@@ -874,7 +848,7 @@ static int vvp_page_sync_io(const struct lu_env *env, struct cl_io *io,
         queue = &io->ci_queue;
         cl_2queue_init_page(queue, page);
 
-        result = cl_io_submit_sync(env, io, crt, queue, CRP_NORMAL, 0);
+	result = cl_io_submit_sync(env, io, crt, queue, 0);
         LASSERT(cl_page_is_owned(page, io));
 
         if (crt == CRT_READ)
@@ -1049,6 +1023,7 @@ static int vvp_io_commit_write(const struct lu_env *env,
                         }
                         if (need_clip)
                                 cl_page_clip(env, pg, 0, to);
+			clear_page_dirty_for_io(vmpage);
                         result = vvp_page_sync_io(env, io, pg, cp, CRT_WRITE);
                         if (result)
                                 CERROR("Write page %lu of inode %p failed %d\n",
@@ -1108,7 +1083,7 @@ static const struct cl_io_operations vvp_io_ops = {
                         .cio_end       = ccc_io_end
                 },
 		[CIT_FSYNC] = {
-			.cio_end    = vvp_io_fsync_end,
+			.cio_start  = vvp_io_fsync_start,
 			.cio_fini   = vvp_io_fini
 		},
                 [CIT_MISC] = {
diff --git a/lustre/llite/vvp_page.c b/lustre/llite/vvp_page.c
index bc0e2b1..990d59b 100644
--- a/lustre/llite/vvp_page.c
+++ b/lustre/llite/vvp_page.c
@@ -36,6 +36,7 @@
  * Implementation of cl_page for VVP layer.
  *
  *   Author: Nikita Danilov <nikita.danilov@sun.com>
+ *   Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
  */
 
 #define DEBUG_SUBSYSTEM S_LLITE
@@ -225,21 +226,15 @@ static int vvp_page_prep_write(const struct lu_env *env,
                                const struct cl_page_slice *slice,
                                struct cl_io *unused)
 {
-        struct cl_page *cp     = slice->cpl_page;
-        cfs_page_t     *vmpage = cl2vm_page(slice);
-        int result;
+	cfs_page_t *vmpage = cl2vm_page(slice);
 
-        if (clear_page_dirty_for_io(vmpage)) {
-                set_page_writeback(vmpage);
-                vvp_write_pending(cl2ccc(slice->cpl_obj), cl2ccc_page(slice));
-                result = 0;
+	LASSERT(PageLocked(vmpage));
+	LASSERT(!PageDirty(vmpage));
 
-                /* only turn on writeback for async write. */
-                if (cp->cp_sync_io == NULL)
-                        unlock_page(vmpage);
-        } else
-                result = -EALREADY;
-        return result;
+	set_page_writeback(vmpage);
+	vvp_write_pending(cl2ccc(slice->cpl_obj), cl2ccc_page(slice));
+
+	return 0;
 }
 
 /**
@@ -342,37 +337,32 @@ static void vvp_page_completion_write(const struct lu_env *env,
  * truncated. Skip it.
  */
 static int vvp_page_make_ready(const struct lu_env *env,
-                               const struct cl_page_slice *slice)
+			       const struct cl_page_slice *slice)
 {
-        cfs_page_t *vmpage = cl2vm_page(slice);
-        struct cl_page *pg = slice->cpl_page;
-        int result;
-
-        result = -EAGAIN;
-        /* we're trying to write, but the page is locked.. come back later */
-        if (!TestSetPageLocked(vmpage)) {
-                if (pg->cp_state == CPS_CACHED) {
-                        /*
-                         * We can cancel IO if page wasn't dirty after all.
-                         */
-                        clear_page_dirty_for_io(vmpage);
-                        /*
-                         * This actually clears the dirty bit in the radix
-                         * tree.
-                         */
-                        set_page_writeback(vmpage);
-                        vvp_write_pending(cl2ccc(slice->cpl_obj),
-                                          cl2ccc_page(slice));
-                        CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
-                        result = 0;
-                } else
-                        /*
-                         * Page was concurrently truncated.
-                         */
-                        LASSERT(pg->cp_state == CPS_FREEING);
-                unlock_page(vmpage);
-        }
-        RETURN(result);
+	cfs_page_t *vmpage = cl2vm_page(slice);
+	struct cl_page *pg = slice->cpl_page;
+	int result = 0;
+
+	lock_page(vmpage);
+	if (clear_page_dirty_for_io(vmpage)) {
+		LASSERT(pg->cp_state == CPS_CACHED);
+		/* This actually clears the dirty bit in the radix
+		 * tree. */
+		set_page_writeback(vmpage);
+		vvp_write_pending(cl2ccc(slice->cpl_obj),
+				cl2ccc_page(slice));
+		CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
+	} else if (pg->cp_state == CPS_PAGEOUT) {
+		/* is it possible for osc_flush_async_page() to already
+		 * make it ready? */
+		result = -EALREADY;
+	} else {
+		CL_PAGE_DEBUG(D_ERROR, env, pg, "Unexpecting page state %d.\n",
+			      pg->cp_state);
+		LBUG();
+	}
+	unlock_page(vmpage);
+	RETURN(result);
 }
 
 static int vvp_page_print(const struct lu_env *env,
diff --git a/lustre/lov/lov_io.c b/lustre/lov/lov_io.c
index 3cf97dd..ab0e41b 100644
--- a/lustre/lov/lov_io.c
+++ b/lustre/lov/lov_io.c
@@ -114,10 +114,12 @@ static void lov_io_sub_inherit(struct cl_io *io, struct lov_io *lio,
 		io->u.ci_fsync.fi_end = end;
 		io->u.ci_fsync.fi_capa = parent->u.ci_fsync.fi_capa;
 		io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
+		io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
 		break;
 	}
-        case CIT_READ:
-        case CIT_WRITE: {
+	case CIT_READ:
+	case CIT_WRITE: {
+		io->u.ci_wr.wr_sync = cl_io_is_sync_write(parent);
                 if (cl_io_is_append(parent)) {
                         io->u.ci_wr.wr_append = 1;
                 } else {
@@ -573,8 +575,7 @@ static struct cl_page_list *lov_io_submit_qin(struct lov_device *ld,
  */
 static int lov_io_submit(const struct lu_env *env,
                          const struct cl_io_slice *ios,
-                         enum cl_req_type crt, struct cl_2queue *queue,
-                         enum cl_req_priority priority)
+			 enum cl_req_type crt, struct cl_2queue *queue)
 {
         struct lov_io          *lio = cl2lov_io(env, ios);
         struct lov_object      *obj = lio->lis_object;
@@ -605,7 +606,7 @@ static int lov_io_submit(const struct lu_env *env,
                 LASSERT(!IS_ERR(sub));
                 LASSERT(sub->sub_io == &lio->lis_single_subio);
                 rc = cl_io_submit_rw(sub->sub_env, sub->sub_io,
-                                     crt, queue, priority);
+				     crt, queue);
                 lov_sub_put(sub);
                 RETURN(rc);
         }
@@ -646,7 +647,7 @@ static int lov_io_submit(const struct lu_env *env,
                 sub = lov_sub_get(env, lio, stripe);
                 if (!IS_ERR(sub)) {
                         rc = cl_io_submit_rw(sub->sub_env, sub->sub_io,
-                                             crt, cl2q, priority);
+					     crt, cl2q);
                         lov_sub_put(sub);
                 } else
                         rc = PTR_ERR(sub);
@@ -743,6 +744,28 @@ static int lov_io_fault_start(const struct lu_env *env,
         RETURN(lov_io_start(env, ios));
 }
 
+static void lov_io_fsync_end(const struct lu_env *env,
+			     const struct cl_io_slice *ios)
+{
+	struct lov_io *lio = cl2lov_io(env, ios);
+	struct lov_io_sub *sub;
+	unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
+	ENTRY;
+
+	*written = 0;
+	cfs_list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
+		struct cl_io *subio = sub->sub_io;
+
+		lov_sub_enter(sub);
+		lov_io_end_wrapper(sub->sub_env, subio);
+		lov_sub_exit(sub);
+
+		if (subio->ci_result == 0)
+			*written += subio->u.ci_fsync.fi_nr_written;
+	}
+	RETURN_EXIT;
+}
+
 static const struct cl_io_operations lov_io_ops = {
         .op = {
                 [CIT_READ] = {
@@ -788,7 +811,7 @@ static const struct cl_io_operations lov_io_ops = {
 			.cio_lock      = lov_io_lock,
 			.cio_unlock    = lov_io_unlock,
 			.cio_start     = lov_io_start,
-			.cio_end       = lov_io_end
+			.cio_end       = lov_io_fsync_end
 		},
                 [CIT_MISC] = {
                         .cio_fini   = lov_io_fini
diff --git a/lustre/lov/lov_page.c b/lustre/lov/lov_page.c
index ca3548f..03335ff 100644
--- a/lustre/lov/lov_page.c
+++ b/lustre/lov/lov_page.c
@@ -109,6 +109,30 @@ static void lov_page_assume(const struct lu_env *env,
         lov_page_own(env, slice, io, 0);
 }
 
+static int lov_page_cache_add(const struct lu_env *env,
+			      const struct cl_page_slice *slice,
+			      struct cl_io *io)
+{
+	struct lov_io     *lio = lov_env_io(env);
+	struct lov_io_sub *sub;
+	int rc = 0;
+
+	LINVRNT(lov_page_invariant(slice));
+	LINVRNT(!cl2lov_page(slice)->lps_invalid);
+	ENTRY;
+
+	sub = lov_page_subio(env, lio, slice);
+	if (!IS_ERR(sub)) {
+		rc = cl_page_cache_add(sub->sub_env, sub->sub_io,
+				       slice->cpl_page->cp_child, CRT_WRITE);
+		lov_sub_put(sub);
+	} else {
+		rc = PTR_ERR(sub);
+		CL_PAGE_DEBUG(D_ERROR, env, slice->cpl_page, "rc = %d\n", rc);
+	}
+	RETURN(rc);
+}
+
 static int lov_page_print(const struct lu_env *env,
                           const struct cl_page_slice *slice,
                           void *cookie, lu_printer_t printer)
@@ -122,6 +146,11 @@ static const struct cl_page_operations lov_page_ops = {
         .cpo_fini   = lov_page_fini,
         .cpo_own    = lov_page_own,
         .cpo_assume = lov_page_assume,
+	.io = {
+		[CRT_WRITE] = {
+			.cpo_cache_add = lov_page_cache_add
+		}
+	},
         .cpo_print  = lov_page_print
 };
 
diff --git a/lustre/obdclass/cl_internal.h b/lustre/obdclass/cl_internal.h
index b17ec42..48ba8da 100644
--- a/lustre/obdclass/cl_internal.h
+++ b/lustre/obdclass/cl_internal.h
@@ -110,7 +110,7 @@ struct cl_thread_info {
          */
         struct cl_sync_io    clt_anchor;
         /**
-         * Fields used by cl_lock_page_out().
+	 * Fields used by cl_lock_discard_pages().
          */
         pgoff_t              clt_next_index;
         pgoff_t              clt_fn_index; /* first non-overlapped index */
diff --git a/lustre/obdclass/cl_io.c b/lustre/obdclass/cl_io.c
index 4ac5087..2e8619a 100644
--- a/lustre/obdclass/cl_io.c
+++ b/lustre/obdclass/cl_io.c
@@ -790,7 +790,7 @@ int cl_io_read_page(const struct lu_env *env, struct cl_io *io,
                 }
         }
         if (result == 0)
-                result = cl_io_submit_rw(env, io, CRT_READ, queue, CRP_NORMAL);
+		result = cl_io_submit_rw(env, io, CRT_READ, queue);
         /*
          * Unlock unsent pages in case of error.
          */
@@ -886,8 +886,7 @@ EXPORT_SYMBOL(cl_io_commit_write);
  * \see cl_io_operations::cio_submit()
  */
 int cl_io_submit_rw(const struct lu_env *env, struct cl_io *io,
-                    enum cl_req_type crt, struct cl_2queue *queue,
-                    enum cl_req_priority priority)
+		    enum cl_req_type crt, struct cl_2queue *queue)
 {
         const struct cl_io_slice *scan;
         int result = 0;
@@ -899,7 +898,7 @@ int cl_io_submit_rw(const struct lu_env *env, struct cl_io *io,
                 if (scan->cis_iop->req_op[crt].cio_submit == NULL)
                         continue;
                 result = scan->cis_iop->req_op[crt].cio_submit(env, scan, crt,
-                                                               queue, priority);
+							       queue);
                 if (result != 0)
                         break;
         }
@@ -917,21 +916,19 @@ EXPORT_SYMBOL(cl_io_submit_rw);
  */
 int cl_io_submit_sync(const struct lu_env *env, struct cl_io *io,
                       enum cl_req_type iot, struct cl_2queue *queue,
-                      enum cl_req_priority prio, long timeout)
+		      long timeout)
 {
         struct cl_sync_io *anchor = &cl_env_info(env)->clt_anchor;
         struct cl_page *pg;
         int rc;
 
-        LASSERT(prio == CRP_NORMAL || prio == CRP_CANCEL);
-
         cl_page_list_for_each(pg, &queue->c2_qin) {
                 LASSERT(pg->cp_sync_io == NULL);
                 pg->cp_sync_io = anchor;
         }
 
         cl_sync_io_init(anchor, queue->c2_qin.pl_nr);
-        rc = cl_io_submit_rw(env, io, iot, queue, prio);
+	rc = cl_io_submit_rw(env, io, iot, queue);
         if (rc == 0) {
                 /*
                  * If some pages weren't sent for any reason (e.g.,
diff --git a/lustre/obdclass/cl_lock.c b/lustre/obdclass/cl_lock.c
index 677d652..fdfeed8 100644
--- a/lustre/obdclass/cl_lock.c
+++ b/lustre/obdclass/cl_lock.c
@@ -1830,12 +1830,13 @@ void cl_lock_cancel(const struct lu_env *env, struct cl_lock *lock)
 EXPORT_SYMBOL(cl_lock_cancel);
 
 /**
- * Finds an existing lock covering given page and optionally different from a
+ * Finds an existing lock covering given index and optionally different from a
  * given \a except lock.
  */
-struct cl_lock *cl_lock_at_page(const struct lu_env *env, struct cl_object *obj,
-                                struct cl_page *page, struct cl_lock *except,
-                                int pending, int canceld)
+struct cl_lock *cl_lock_at_pgoff(const struct lu_env *env,
+				 struct cl_object *obj, pgoff_t index,
+				 struct cl_lock *except,
+				 int pending, int canceld)
 {
         struct cl_object_header *head;
         struct cl_lock          *scan;
@@ -1850,7 +1851,7 @@ struct cl_lock *cl_lock_at_page(const struct lu_env *env, struct cl_object *obj,
 
         need->cld_mode = CLM_READ; /* CLM_READ matches both READ & WRITE, but
                                     * not PHANTOM */
-        need->cld_start = need->cld_end = page->cp_index;
+	need->cld_start = need->cld_end = index;
         need->cld_enq_flags = 0;
 
         cfs_spin_lock(&head->coh_lock_guard);
@@ -1879,7 +1880,7 @@ struct cl_lock *cl_lock_at_page(const struct lu_env *env, struct cl_object *obj,
         cfs_spin_unlock(&head->coh_lock_guard);
         RETURN(lock);
 }
-EXPORT_SYMBOL(cl_lock_at_page);
+EXPORT_SYMBOL(cl_lock_at_pgoff);
 
 /**
  * Calculate the page offset at the layer of @lock.
@@ -1935,60 +1936,45 @@ static int check_and_discard_cb(const struct lu_env *env, struct cl_io *io,
         return CLP_GANG_OKAY;
 }
 
-static int pageout_cb(const struct lu_env *env, struct cl_io *io,
+static int discard_cb(const struct lu_env *env, struct cl_io *io,
                       struct cl_page *page, void *cbdata)
 {
-        struct cl_thread_info *info  = cl_env_info(env);
-        struct cl_page_list   *queue = &info->clt_queue.c2_qin;
-        struct cl_lock        *lock  = cbdata;
-        typeof(cl_page_own)   *page_own;
-        int rc = CLP_GANG_OKAY;
-
-        page_own = queue->pl_nr ? cl_page_own_try : cl_page_own;
-        if (page_own(env, io, page) == 0) {
-                cl_page_list_add(queue, page);
-                info->clt_next_index = pgoff_at_lock(page, lock) + 1;
-        } else if (page->cp_state != CPS_FREEING) {
-                /* cl_page_own() won't fail unless
-                 * the page is being freed. */
-                LASSERT(queue->pl_nr != 0);
-                rc = CLP_GANG_AGAIN;
-        }
+	struct cl_thread_info *info = cl_env_info(env);
+	struct cl_lock *lock   = cbdata;
+
+	LASSERT(lock->cll_descr.cld_mode >= CLM_WRITE);
+	KLASSERT(ergo(page->cp_type == CPT_CACHEABLE,
+		      !PageWriteback(cl_page_vmpage(env, page))));
+	KLASSERT(ergo(page->cp_type == CPT_CACHEABLE,
+		      !PageDirty(cl_page_vmpage(env, page))));
+
+	info->clt_next_index = pgoff_at_lock(page, lock) + 1;
+	if (cl_page_own(env, io, page) == 0) {
+		/* discard the page */
+		cl_page_unmap(env, io, page);
+		cl_page_discard(env, io, page);
+		cl_page_disown(env, io, page);
+	} else {
+		LASSERT(page->cp_state == CPS_FREEING);
+	}
 
-        return rc;
+	return CLP_GANG_OKAY;
 }
 
 /**
- * Invalidate pages protected by the given lock, sending them out to the
- * server first, if necessary.
- *
- * This function does the following:
- *
- *     - collects a list of pages to be invalidated,
- *
- *     - unmaps them from the user virtual memory,
- *
- *     - sends dirty pages to the server,
- *
- *     - waits for transfer completion,
- *
- *     - discards pages, and throws them out of memory.
- *
- * If \a discard is set, pages are discarded without sending them to the
- * server.
+ * Discard pages protected by the given lock. This function traverses radix
+ * tree to find all covering pages and discard them. If a page is being covered
+ * by other locks, it should remain in cache.
  *
  * If error happens on any step, the process continues anyway (the reasoning
  * behind this being that lock cancellation cannot be delayed indefinitely).
  */
-int cl_lock_page_out(const struct lu_env *env, struct cl_lock *lock,
-                     int discard)
+int cl_lock_discard_pages(const struct lu_env *env, struct cl_lock *lock)
 {
         struct cl_thread_info *info  = cl_env_info(env);
         struct cl_io          *io    = &info->clt_io;
-        struct cl_2queue      *queue = &info->clt_queue;
         struct cl_lock_descr  *descr = &lock->cll_descr;
         cl_page_gang_cb_t      cb;
-        long page_count;
         int res;
         int result;
 
@@ -2000,38 +1986,12 @@ int cl_lock_page_out(const struct lu_env *env, struct cl_lock *lock,
         if (result != 0)
                 GOTO(out, result);
 
-        cb = descr->cld_mode == CLM_READ ? check_and_discard_cb : pageout_cb;
+	cb = descr->cld_mode == CLM_READ ? check_and_discard_cb : discard_cb;
         info->clt_fn_index = info->clt_next_index = descr->cld_start;
         do {
-                cl_2queue_init(queue);
                 res = cl_page_gang_lookup(env, descr->cld_obj, io,
                                           info->clt_next_index, descr->cld_end,
                                           cb, (void *)lock);
-                page_count = queue->c2_qin.pl_nr;
-                if (page_count > 0) {
-                        /* must be writeback case */
-                        LASSERTF(descr->cld_mode >= CLM_WRITE, "lock mode %s\n",
-                                 cl_lock_mode_name(descr->cld_mode));
-
-                        result = cl_page_list_unmap(env, io, &queue->c2_qin);
-                        if (!discard) {
-                                long timeout = 600; /* 10 minutes. */
-                                /* for debug purpose, if this request can't be
-                                 * finished in 10 minutes, we hope it can
-                                 * notify us.
-                                 */
-                                result = cl_io_submit_sync(env, io, CRT_WRITE,
-                                                           queue, CRP_CANCEL,
-                                                           timeout);
-                                if (result)
-                                        CWARN("Writing %lu pages error: %d\n",
-                                              page_count, result);
-                        }
-                        cl_2queue_discard(env, io, queue);
-                        cl_2queue_disown(env, io, queue);
-                        cl_2queue_fini(env, queue);
-                }
-
                 if (info->clt_next_index > descr->cld_end)
                         break;
 
@@ -2042,7 +2002,7 @@ out:
         cl_io_fini(env, io);
         RETURN(result);
 }
-EXPORT_SYMBOL(cl_lock_page_out);
+EXPORT_SYMBOL(cl_lock_discard_pages);
 
 /**
  * Eliminate all locks for a given object.
diff --git a/lustre/obdclass/cl_page.c b/lustre/obdclass/cl_page.c
index 217ee00..e2f6f93 100644
--- a/lustre/obdclass/cl_page.c
+++ b/lustre/obdclass/cl_page.c
@@ -1406,32 +1406,61 @@ EXPORT_SYMBOL(cl_page_make_ready);
  * its queues.
  *
  * \pre  cl_page_is_owned(pg, io)
- * \post ergo(result == 0,
- *            pg->cp_state == CPS_PAGEIN || pg->cp_state == CPS_PAGEOUT)
+ * \post cl_page_is_owned(pg, io)
  *
  * \see cl_page_operations::cpo_cache_add()
  */
 int cl_page_cache_add(const struct lu_env *env, struct cl_io *io,
                       struct cl_page *pg, enum cl_req_type crt)
 {
-        int result;
+	const struct cl_page_slice *scan;
+	int result = 0;
 
-        PINVRNT(env, pg, crt < CRT_NR);
-        PINVRNT(env, pg, cl_page_is_owned(pg, io));
-        PINVRNT(env, pg, cl_page_invariant(pg));
+	PINVRNT(env, pg, crt < CRT_NR);
+	PINVRNT(env, pg, cl_page_is_owned(pg, io));
+	PINVRNT(env, pg, cl_page_invariant(pg));
 
-        ENTRY;
-        result = cl_page_invoke(env, io, pg, CL_PAGE_OP(io[crt].cpo_cache_add));
-        if (result == 0) {
-                cl_page_owner_clear(pg);
-                cl_page_state_set(env, pg, CPS_CACHED);
-        }
-        CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result);
-        RETURN(result);
+	ENTRY;
+
+	cfs_list_for_each_entry(scan, &pg->cp_layers, cpl_linkage) {
+		if (scan->cpl_ops->io[crt].cpo_cache_add == NULL)
+			continue;
+
+		result = scan->cpl_ops->io[crt].cpo_cache_add(env, scan, io);
+		if (result != 0)
+			break;
+	}
+	CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result);
+	RETURN(result);
 }
 EXPORT_SYMBOL(cl_page_cache_add);
 
 /**
+ * Called if a pge is being written back by kernel's intention.
+ *
+ * \pre  cl_page_is_owned(pg, io)
+ * \post ergo(result == 0, pg->cp_state == CPS_PAGEOUT)
+ *
+ * \see cl_page_operations::cpo_flush()
+ */
+int cl_page_flush(const struct lu_env *env, struct cl_io *io,
+		  struct cl_page *pg)
+{
+	int result;
+
+	PINVRNT(env, pg, cl_page_is_owned(pg, io));
+	PINVRNT(env, pg, cl_page_invariant(pg));
+
+	ENTRY;
+
+	result = cl_page_invoke(env, io, pg, CL_PAGE_OP(cpo_flush));
+
+	CL_PAGE_HEADER(D_TRACE, env, pg, "%d\n", result);
+	RETURN(result);
+}
+EXPORT_SYMBOL(cl_page_flush);
+
+/**
  * Checks whether page is protected by any extent lock is at least required
  * mode.
  *
diff --git a/lustre/obdclass/lprocfs_status.c b/lustre/obdclass/lprocfs_status.c
index 4b957f7..b8b6a38 100644
--- a/lustre/obdclass/lprocfs_status.c
+++ b/lustre/obdclass/lprocfs_status.c
@@ -2457,32 +2457,6 @@ int lprocfs_obd_rd_max_pages_per_rpc(char *page, char **start, off_t off,
 }
 EXPORT_SYMBOL(lprocfs_obd_rd_max_pages_per_rpc);
 
-int lprocfs_obd_wr_max_pages_per_rpc(struct file *file, const char *buffer,
-                                     unsigned long count, void *data)
-{
-        struct obd_device *dev = data;
-        struct client_obd *cli = &dev->u.cli;
-        struct obd_connect_data *ocd = &cli->cl_import->imp_connect_data;
-        int val, rc;
-
-        rc = lprocfs_write_helper(buffer, count, &val);
-        if (rc)
-                return rc;
-
-        LPROCFS_CLIMP_CHECK(dev);
-        if (val < 1 || val > ocd->ocd_brw_size >> CFS_PAGE_SHIFT) {
-                LPROCFS_CLIMP_EXIT(dev);
-                return -ERANGE;
-        }
-        client_obd_list_lock(&cli->cl_loi_list_lock);
-        cli->cl_max_pages_per_rpc = val;
-        client_obd_list_unlock(&cli->cl_loi_list_lock);
-
-        LPROCFS_CLIMP_EXIT(dev);
-        return count;
-}
-EXPORT_SYMBOL(lprocfs_obd_wr_max_pages_per_rpc);
-
 int lprocfs_target_rd_instance(char *page, char **start, off_t off,
                                int count, int *eof, void *data)
 {
diff --git a/lustre/obdclass/lu_ref.c b/lustre/obdclass/lu_ref.c
index 84afc67..3471bd3 100644
--- a/lustre/obdclass/lu_ref.c
+++ b/lustre/obdclass/lu_ref.c
@@ -57,18 +57,17 @@
  * Asserts a condition for a given lu_ref. Must be called with
  * lu_ref::lf_guard held.
  */
-#define REFASSERT(ref, expr)                            \
-  do {                                                  \
-          struct lu_ref *__ref = (ref);                 \
-                                                        \
-          if (unlikely(!(expr))) {                      \
-                  lu_ref_print(__ref);                  \
-                  cfs_spin_unlock(&__ref->lf_guard);    \
-                  lu_ref_print_all();                   \
-                  LASSERT(0);                           \
-                  cfs_spin_lock(&__ref->lf_guard);      \
-          }                                             \
-  } while (0)
+#define REFASSERT(ref, expr)  do {                      \
+	struct lu_ref *__tmp = (ref);                   \
+							\
+	if (unlikely(!(expr))) {                        \
+		lu_ref_print(__tmp);                    \
+		cfs_spin_unlock(&__tmp->lf_guard);      \
+		lu_ref_print_all();                     \
+		LASSERT(0);                             \
+		cfs_spin_lock(&__tmp->lf_guard);        \
+	}                                               \
+} while (0)
 
 struct lu_ref_link {
         struct lu_ref    *ll_ref;
diff --git a/lustre/obdecho/echo_client.c b/lustre/obdecho/echo_client.c
index 0cc6d1d..65d7429e 100644
--- a/lustre/obdecho/echo_client.c
+++ b/lustre/obdecho/echo_client.c
@@ -1336,8 +1336,7 @@ static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
                 if (async)
                         rc = cl_echo_async_brw(env, io, typ, queue);
                 else
-                        rc = cl_io_submit_sync(env, io, typ, queue,
-                                               CRP_NORMAL, 0);
+			rc = cl_io_submit_sync(env, io, typ, queue, 0);
                 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
                        async ? "async" : "sync", rc);
         }
diff --git a/lustre/osc/lproc_osc.c b/lustre/osc/lproc_osc.c
index 4018132..742733c 100644
--- a/lustre/osc/lproc_osc.c
+++ b/lustre/osc/lproc_osc.c
@@ -565,6 +565,35 @@ static int osc_rd_destroys_in_flight(char *page, char **start, off_t off,
                         cfs_atomic_read(&obd->u.cli.cl_destroy_in_flight));
 }
 
+static int lprocfs_osc_wr_max_pages_per_rpc(struct file *file,
+	const char *buffer, unsigned long count, void *data)
+{
+	struct obd_device *dev = data;
+	struct client_obd *cli = &dev->u.cli;
+	struct obd_connect_data *ocd = &cli->cl_import->imp_connect_data;
+	int chunk_mask, val, rc;
+
+	rc = lprocfs_write_helper(buffer, count, &val);
+	if (rc)
+		return rc;
+
+	LPROCFS_CLIMP_CHECK(dev);
+
+	chunk_mask = ~((1 << (cli->cl_chunkbits - CFS_PAGE_SHIFT)) - 1);
+	/* max_pages_per_rpc must be chunk aligned */
+	val = (val + ~chunk_mask) & chunk_mask;
+	if (val == 0 || val > ocd->ocd_brw_size >> CFS_PAGE_SHIFT) {
+		LPROCFS_CLIMP_EXIT(dev);
+		return -ERANGE;
+	}
+	client_obd_list_lock(&cli->cl_loi_list_lock);
+	cli->cl_max_pages_per_rpc = val;
+	client_obd_list_unlock(&cli->cl_loi_list_lock);
+
+	LPROCFS_CLIMP_EXIT(dev);
+	return count;
+}
+
 static struct lprocfs_vars lprocfs_osc_obd_vars[] = {
         { "uuid",            lprocfs_rd_uuid,        0, 0 },
         { "ping",            0, lprocfs_wr_ping,     0, 0, 0222 },
@@ -581,7 +610,7 @@ static struct lprocfs_vars lprocfs_osc_obd_vars[] = {
         { "active",          osc_rd_active,
                              osc_wr_active, 0 },
         { "max_pages_per_rpc", lprocfs_obd_rd_max_pages_per_rpc,
-                               lprocfs_obd_wr_max_pages_per_rpc, 0 },
+			       lprocfs_osc_wr_max_pages_per_rpc, 0 },
         { "max_rpcs_in_flight", osc_rd_max_rpcs_in_flight,
                                 osc_wr_max_rpcs_in_flight, 0 },
         { "destroys_in_flight", osc_rd_destroys_in_flight, 0, 0 },
@@ -638,9 +667,9 @@ static int osc_rpc_stats_seq_show(struct seq_file *seq, void *v)
         seq_printf(seq, "write RPCs in flight: %d\n",
                    cli->cl_w_in_flight);
         seq_printf(seq, "pending write pages:  %d\n",
-                   cli->cl_pending_w_pages);
+		   cfs_atomic_read(&cli->cl_pending_w_pages));
         seq_printf(seq, "pending read pages:   %d\n",
-                   cli->cl_pending_r_pages);
+		   cfs_atomic_read(&cli->cl_pending_r_pages));
 
         seq_printf(seq, "\n\t\t\tread\t\t\twrite\n");
         seq_printf(seq, "pages per rpc         rpcs   %% cum %% |");
diff --git a/lustre/osc/osc_cache.c b/lustre/osc/osc_cache.c
index 9f12ad6..b67911b 100644
--- a/lustre/osc/osc_cache.c
+++ b/lustre/osc/osc_cache.c
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -46,37 +44,1137 @@
 #include "osc_cl_internal.h"
 #include "osc_internal.h"
 
-static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli,
-			   struct osc_async_page *oap);
-static int osc_enter_cache_try(const struct lu_env *env, struct client_obd *cli,
-			       struct osc_async_page *oap, int transient);
-static void osc_exit_cache(struct client_obd *cli, struct osc_async_page *oap,
-			   int sent);
+static int extent_debug; /* set it to be true for more debug */
+
+static void osc_update_pending(struct osc_object *obj, int cmd, int delta);
+static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext,
+			   int state);
+static void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
+			      struct osc_async_page *oap, int sent, int rc);
+static int osc_make_ready(const struct lu_env *env, struct osc_async_page *oap,
+			  int cmd);
+static int osc_refresh_count(const struct lu_env *env,
+			     struct osc_async_page *oap, int cmd);
+static int osc_io_unplug_async(const struct lu_env *env,
+			       struct client_obd *cli, struct osc_object *osc);
+static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages,
+			   unsigned int lost_grant);
+
+static void osc_extent_tree_dump0(int level, struct osc_object *obj,
+				  const char *func, int line);
+#define osc_extent_tree_dump(lvl, obj) \
+	osc_extent_tree_dump0(lvl, obj, __func__, __LINE__)
 
 /** \addtogroup osc
  *  @{
  */
 
-#define OSC_IO_DEBUG(OSC, STR, args...)				  \
-	CDEBUG(D_INODE, "loi ready %d wr %d:%d rd %d:%d " STR,	   \
-	       !cfs_list_empty(&(OSC)->oo_ready_item) ||		 \
-	       !cfs_list_empty(&(OSC)->oo_hp_ready_item),		\
-	       (OSC)->oo_write_pages.oop_num_pending,		    \
-	       !cfs_list_empty(&(OSC)->oo_write_pages.oop_urgent),       \
-	       (OSC)->oo_read_pages.oop_num_pending,		     \
-	       !cfs_list_empty(&(OSC)->oo_read_pages.oop_urgent),	\
-	       args)
+/* ------------------ osc extent ------------------ */
+static inline char *ext_flags(struct osc_extent *ext, char *flags)
+{
+	char *buf = flags;
+	*buf++ = ext->oe_rw ? 'r' : 'w';
+	if (ext->oe_intree)
+		*buf++ = 'i';
+	if (ext->oe_srvlock)
+		*buf++ = 's';
+	if (ext->oe_hp)
+		*buf++ = 'h';
+	if (ext->oe_urgent)
+		*buf++ = 'u';
+	if (ext->oe_memalloc)
+		*buf++ = 'm';
+	if (ext->oe_trunc_pending)
+		*buf++ = 't';
+	if (ext->oe_fsync_wait)
+		*buf++ = 'Y';
+	*buf = 0;
+	return flags;
+}
+
+static inline char list_empty_marker(cfs_list_t *list)
+{
+	return cfs_list_empty(list) ? '-' : '+';
+}
+
+#define EXTSTR       "[%lu -> %lu/%lu]"
+#define EXTPARA(ext) (ext)->oe_start, (ext)->oe_end, (ext)->oe_max_end
+
+#define OSC_EXTENT_DUMP(lvl, extent, fmt, ...) do {			      \
+	struct osc_extent *__ext = (extent);				      \
+	const char *__str[] = OES_STRINGS;				      \
+	char __buf[16];							      \
+									      \
+	CDEBUG(lvl,							      \
+		"extent %p@{" EXTSTR ", "				      \
+		"[%d|%d|%c|%s|%s|%p], [%d|%d|%c|%c|%p|%u|%p]} " fmt,	      \
+		/* ----- extent part 0 ----- */			              \
+		__ext, EXTPARA(__ext),					      \
+		/* ----- part 1 ----- */				      \
+		cfs_atomic_read(&__ext->oe_refc),			      \
+		cfs_atomic_read(&__ext->oe_users),			      \
+		list_empty_marker(&__ext->oe_link),			      \
+		__str[__ext->oe_state], ext_flags(__ext, __buf),	      \
+		__ext->oe_obj,						      \
+		/* ----- part 2 ----- */				      \
+		__ext->oe_grants, __ext->oe_nr_pages,			      \
+		list_empty_marker(&__ext->oe_pages),			      \
+		cfs_waitq_active(&__ext->oe_waitq) ? '+' : '-',		      \
+		__ext->oe_osclock, __ext->oe_mppr, __ext->oe_owner,	      \
+		/* ----- part 4 ----- */				      \
+		## __VA_ARGS__);					      \
+} while (0)
+
+#undef EASSERTF
+#define EASSERTF(expr, ext, fmt, args...) do {				\
+	if (!(expr)) {							\
+		OSC_EXTENT_DUMP(D_ERROR, (ext), fmt, ##args);		 \
+		osc_extent_tree_dump(D_ERROR, (ext)->oe_obj);		 \
+		LASSERT(expr);						\
+	}								     \
+} while (0)
+
+#undef EASSERT
+#define EASSERT(expr, ext) EASSERTF(expr, ext, "\n")
+
+static inline struct osc_extent *rb_extent(struct rb_node *n)
+{
+	if (n == NULL)
+		return NULL;
+
+	return container_of(n, struct osc_extent, oe_node);
+}
+
+static inline struct osc_extent *next_extent(struct osc_extent *ext)
+{
+	if (ext == NULL)
+		return NULL;
+
+	LASSERT(ext->oe_intree);
+	return rb_extent(rb_next(&ext->oe_node));
+}
+
+static inline struct osc_extent *prev_extent(struct osc_extent *ext)
+{
+	if (ext == NULL)
+		return NULL;
+
+	LASSERT(ext->oe_intree);
+	return rb_extent(rb_prev(&ext->oe_node));
+}
+
+static inline struct osc_extent *first_extent(struct osc_object *obj)
+{
+	return rb_extent(rb_first(&obj->oo_root));
+}
+
+/* object must be locked by caller. */
+static int osc_extent_sanity_check0(struct osc_extent *ext,
+				    const char *func, const int line)
+{
+	struct osc_object *obj = ext->oe_obj;
+	struct osc_async_page *oap;
+	int page_count;
+	int rc = 0;
+
+	if (!osc_object_is_locked(obj))
+		GOTO(out, rc = 9);
+
+	if (ext->oe_state >= OES_STATE_MAX)
+		GOTO(out, rc = 10);
+
+	if (cfs_atomic_read(&ext->oe_refc) <= 0)
+		GOTO(out, rc = 20);
+
+	if (cfs_atomic_read(&ext->oe_refc) < cfs_atomic_read(&ext->oe_users))
+		GOTO(out, rc = 30);
+
+	switch (ext->oe_state) {
+	case OES_INV:
+		if (ext->oe_nr_pages > 0 || !cfs_list_empty(&ext->oe_pages))
+			GOTO(out, rc = 35);
+		GOTO(out, rc = 0);
+		break;
+	case OES_ACTIVE:
+		if (cfs_atomic_read(&ext->oe_users) == 0)
+			GOTO(out, rc = 40);
+		if (ext->oe_hp)
+			GOTO(out, rc = 50);
+		if (ext->oe_fsync_wait && !ext->oe_urgent)
+			GOTO(out, rc = 55);
+		break;
+	case OES_CACHE:
+		if (ext->oe_grants == 0)
+			GOTO(out, rc = 60);
+		if (ext->oe_fsync_wait && !ext->oe_urgent && !ext->oe_hp)
+			GOTO(out, rc = 65);
+	default:
+		if (cfs_atomic_read(&ext->oe_users) > 0)
+			GOTO(out, rc = 70);
+	}
+
+	if (ext->oe_max_end < ext->oe_end || ext->oe_end < ext->oe_start)
+		GOTO(out, rc = 80);
+
+	if (ext->oe_osclock == NULL && ext->oe_grants > 0)
+		GOTO(out, rc = 90);
+
+	if (ext->oe_osclock) {
+		struct cl_lock_descr *descr;
+		descr = &ext->oe_osclock->cll_descr;
+		if (!(descr->cld_start <= ext->oe_start &&
+		      descr->cld_end >= ext->oe_max_end))
+			GOTO(out, rc = 100);
+	}
+
+	if (ext->oe_nr_pages > ext->oe_mppr)
+		GOTO(out, rc = 105);
+
+	/* Do not verify page list if extent is in RPC. This is because an
+	 * in-RPC extent is supposed to be exclusively accessible w/o lock. */
+	if (ext->oe_state > OES_CACHE)
+		GOTO(out, rc = 0);
+
+	if (!extent_debug)
+		GOTO(out, rc = 0);
+
+	page_count = 0;
+	cfs_list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
+		pgoff_t index = oap2cl_page(oap)->cp_index;
+		++page_count;
+		if (index > ext->oe_end || index < ext->oe_start)
+			GOTO(out, rc = 110);
+	}
+	if (page_count != ext->oe_nr_pages)
+		GOTO(out, rc = 120);
+
+out:
+	if (rc != 0)
+		OSC_EXTENT_DUMP(D_ERROR, ext,
+				"%s:%d sanity check %p failed with rc = %d\n",
+				func, line, ext, rc);
+	return rc;
+}
+
+#define sanity_check_nolock(ext) \
+	osc_extent_sanity_check0(ext, __func__, __LINE__)
+
+#define sanity_check(ext) ({                                                   \
+	int __res;                                                             \
+	osc_object_lock((ext)->oe_obj);                                        \
+	__res = sanity_check_nolock(ext);                                      \
+	osc_object_unlock((ext)->oe_obj);                                      \
+	__res;                                                                 \
+})
+
+
+/**
+ * sanity check - to make sure there is no overlapped extent in the tree.
+ */
+static int osc_extent_is_overlapped(struct osc_object *obj,
+				    struct osc_extent *ext)
+{
+	struct osc_extent *tmp;
+
+	LASSERT(osc_object_is_locked(obj));
+
+	if (!extent_debug)
+		return 0;
+
+	for (tmp = first_extent(obj); tmp != NULL; tmp = next_extent(tmp)) {
+		if (tmp == ext)
+			continue;
+		if (tmp->oe_end >= ext->oe_start &&
+		    tmp->oe_start <= ext->oe_end)
+			return 1;
+	}
+	return 0;
+}
+
+static void osc_extent_state_set(struct osc_extent *ext, int state)
+{
+	LASSERT(osc_object_is_locked(ext->oe_obj));
+	LASSERT(state >= OES_INV && state < OES_STATE_MAX);
+
+	/* Never try to sanity check a state changing extent :-) */
+	/* LASSERT(sanity_check_nolock(ext) == 0); */
+
+	/* TODO: validate the state machine */
+	ext->oe_state = state;
+	cfs_waitq_broadcast(&ext->oe_waitq);
+}
+
+static struct osc_extent *osc_extent_alloc(struct osc_object *obj)
+{
+	struct osc_extent *ext;
+
+	OBD_SLAB_ALLOC_PTR_GFP(ext, osc_extent_kmem, CFS_ALLOC_STD);
+	if (ext == NULL)
+		return NULL;
+
+	RB_CLEAR_NODE(&ext->oe_node);
+	ext->oe_obj = obj;
+	cfs_atomic_set(&ext->oe_refc, 1);
+	cfs_atomic_set(&ext->oe_users, 0);
+	CFS_INIT_LIST_HEAD(&ext->oe_link);
+	ext->oe_state = OES_INV;
+	CFS_INIT_LIST_HEAD(&ext->oe_pages);
+	cfs_waitq_init(&ext->oe_waitq);
+	ext->oe_osclock = NULL;
+
+	return ext;
+}
+
+static void osc_extent_free(struct osc_extent *ext)
+{
+	OBD_SLAB_FREE_PTR(ext, osc_extent_kmem);
+}
+
+static struct osc_extent *osc_extent_get(struct osc_extent *ext)
+{
+	LASSERT(cfs_atomic_read(&ext->oe_refc) >= 0);
+	cfs_atomic_inc(&ext->oe_refc);
+	return ext;
+}
+
+static void osc_extent_put(const struct lu_env *env, struct osc_extent *ext)
+{
+	LASSERT(cfs_atomic_read(&ext->oe_refc) > 0);
+	if (cfs_atomic_dec_and_test(&ext->oe_refc)) {
+		LASSERT(cfs_list_empty(&ext->oe_link));
+		LASSERT(cfs_atomic_read(&ext->oe_users) == 0);
+		LASSERT(ext->oe_state == OES_INV);
+		LASSERT(!ext->oe_intree);
+
+		if (ext->oe_osclock) {
+			cl_lock_put(env, ext->oe_osclock);
+			ext->oe_osclock = NULL;
+		}
+		osc_extent_free(ext);
+	}
+}
+
+/**
+ * osc_extent_put_trust() is a special version of osc_extent_put() when
+ * it's known that the caller is not the last user. This is to address the
+ * problem of lacking of lu_env ;-).
+ */
+static void osc_extent_put_trust(struct osc_extent *ext)
+{
+	LASSERT(cfs_atomic_read(&ext->oe_refc) > 1);
+	LASSERT(osc_object_is_locked(ext->oe_obj));
+	cfs_atomic_dec(&ext->oe_refc);
+}
+
+/**
+ * Return the extent which includes pgoff @index, or return the greatest
+ * previous extent in the tree.
+ */
+static struct osc_extent *osc_extent_search(struct osc_object *obj,
+					    pgoff_t index)
+{
+	struct rb_node    *n = obj->oo_root.rb_node;
+	struct osc_extent *tmp, *p = NULL;
+
+	LASSERT(osc_object_is_locked(obj));
+	while (n != NULL) {
+		tmp = rb_extent(n);
+		if (index < tmp->oe_start) {
+			n = n->rb_left;
+		} else if (index > tmp->oe_end) {
+			p = rb_extent(n);
+			n = n->rb_right;
+		} else {
+			return tmp;
+		}
+	}
+	return p;
+}
+
+/*
+ * Return the extent covering @index, otherwise return NULL.
+ * caller must have held object lock.
+ */
+static struct osc_extent *osc_extent_lookup(struct osc_object *obj,
+					    pgoff_t index)
+{
+	struct osc_extent *ext;
+
+	ext = osc_extent_search(obj, index);
+	if (ext != NULL && ext->oe_start <= index && index <= ext->oe_end)
+		return osc_extent_get(ext);
+	return NULL;
+}
+
+/* caller must have held object lock. */
+static void osc_extent_insert(struct osc_object *obj, struct osc_extent *ext)
+{
+	struct rb_node   **n      = &obj->oo_root.rb_node;
+	struct rb_node    *parent = NULL;
+	struct osc_extent *tmp;
+
+	LASSERT(ext->oe_intree == 0);
+	LASSERT(ext->oe_obj == obj);
+	LASSERT(osc_object_is_locked(obj));
+	while (*n != NULL) {
+		tmp = rb_extent(*n);
+		parent = *n;
+
+		if (ext->oe_end < tmp->oe_start)
+			n = &(*n)->rb_left;
+		else if (ext->oe_start > tmp->oe_end)
+			n = &(*n)->rb_right;
+		else
+			EASSERTF(0, tmp, EXTSTR, EXTPARA(ext));
+	}
+	rb_link_node(&ext->oe_node, parent, n);
+	rb_insert_color(&ext->oe_node, &obj->oo_root);
+	osc_extent_get(ext);
+	ext->oe_intree = 1;
+}
+
+/* caller must have held object lock. */
+static void osc_extent_erase(struct osc_extent *ext)
+{
+	struct osc_object *obj = ext->oe_obj;
+	LASSERT(osc_object_is_locked(obj));
+	if (ext->oe_intree) {
+		rb_erase(&ext->oe_node, &obj->oo_root);
+		ext->oe_intree = 0;
+		/* rbtree held a refcount */
+		osc_extent_put_trust(ext);
+	}
+}
+
+static struct osc_extent *osc_extent_hold(struct osc_extent *ext)
+{
+	struct osc_object *obj = ext->oe_obj;
+
+	LASSERT(osc_object_is_locked(obj));
+	LASSERT(ext->oe_state == OES_ACTIVE || ext->oe_state == OES_CACHE);
+	if (ext->oe_state == OES_CACHE) {
+		osc_extent_state_set(ext, OES_ACTIVE);
+		osc_update_pending(obj, OBD_BRW_WRITE, -ext->oe_nr_pages);
+	}
+	cfs_atomic_inc(&ext->oe_users);
+	cfs_list_del_init(&ext->oe_link);
+	return osc_extent_get(ext);
+}
+
+static void __osc_extent_remove(struct osc_extent *ext)
+{
+	LASSERT(osc_object_is_locked(ext->oe_obj));
+	LASSERT(cfs_list_empty(&ext->oe_pages));
+	osc_extent_erase(ext);
+	cfs_list_del_init(&ext->oe_link);
+	osc_extent_state_set(ext, OES_INV);
+	OSC_EXTENT_DUMP(D_CACHE, ext, "destroyed.\n");
+}
+
+static void osc_extent_remove(struct osc_extent *ext)
+{
+	struct osc_object *obj = ext->oe_obj;
+
+	osc_object_lock(obj);
+	__osc_extent_remove(ext);
+	osc_object_unlock(obj);
+}
+
+/**
+ * This function is used to merge extents to get better performance. It checks
+ * if @cur and @victim are contiguous at chunk level.
+ */
+static int osc_extent_merge(const struct lu_env *env, struct osc_extent *cur,
+			    struct osc_extent *victim)
+{
+	struct osc_object *obj = cur->oe_obj;
+	pgoff_t chunk_start;
+	pgoff_t chunk_end;
+	int ppc_bits;
+
+	LASSERT(cur->oe_state == OES_CACHE);
+	LASSERT(osc_object_is_locked(obj));
+	if (victim == NULL)
+		return -EINVAL;
+
+	if (victim->oe_state != OES_CACHE || victim->oe_fsync_wait)
+		return -EBUSY;
+
+	if (cur->oe_max_end != victim->oe_max_end)
+		return -ERANGE;
+
+	LASSERT(cur->oe_osclock == victim->oe_osclock);
+	ppc_bits = osc_cli(obj)->cl_chunkbits - CFS_PAGE_SHIFT;
+	chunk_start = cur->oe_start >> ppc_bits;
+	chunk_end   = cur->oe_end   >> ppc_bits;
+	if (chunk_start   != (victim->oe_end >> ppc_bits) + 1 &&
+	    chunk_end + 1 != victim->oe_start >> ppc_bits)
+		return -ERANGE;
+
+	OSC_EXTENT_DUMP(D_CACHE, victim, "will be merged by %p.\n", cur);
+
+	cur->oe_start     = min(cur->oe_start, victim->oe_start);
+	cur->oe_end       = max(cur->oe_end,   victim->oe_end);
+	cur->oe_grants   += victim->oe_grants;
+	cur->oe_nr_pages += victim->oe_nr_pages;
+	/* only the following bits are needed to merge */
+	cur->oe_urgent   |= victim->oe_urgent;
+	cur->oe_memalloc |= victim->oe_memalloc;
+	cfs_list_splice_init(&victim->oe_pages, &cur->oe_pages);
+	cfs_list_del_init(&victim->oe_link);
+	victim->oe_nr_pages = 0;
+
+	osc_extent_get(victim);
+	__osc_extent_remove(victim);
+	osc_extent_put(env, victim);
+
+	OSC_EXTENT_DUMP(D_CACHE, cur, "after merging %p.\n", victim);
+	return 0;
+}
+
+/**
+ * Drop user count of osc_extent, and unplug IO asynchronously.
+ */
+int osc_extent_release(const struct lu_env *env, struct osc_extent *ext)
+{
+	struct osc_object *obj = ext->oe_obj;
+	int rc = 0;
+	ENTRY;
+
+	LASSERT(cfs_atomic_read(&ext->oe_users) > 0);
+	LASSERT(sanity_check(ext) == 0);
+	LASSERT(ext->oe_grants > 0);
+
+	if (cfs_atomic_dec_and_lock(&ext->oe_users, &obj->oo_lock)) {
+		LASSERT(ext->oe_state == OES_ACTIVE);
+		if (ext->oe_trunc_pending) {
+			/* a truncate process is waiting for this extent.
+			 * This may happen due to a race, check
+			 * osc_cache_truncate_start(). */
+			osc_extent_state_set(ext, OES_TRUNC);
+			ext->oe_trunc_pending = 0;
+		} else {
+			osc_extent_state_set(ext, OES_CACHE);
+			osc_update_pending(obj, OBD_BRW_WRITE,
+					   ext->oe_nr_pages);
+
+			/* try to merge the previous and next extent. */
+			osc_extent_merge(env, ext, prev_extent(ext));
+			osc_extent_merge(env, ext, next_extent(ext));
+
+			if (ext->oe_urgent)
+				cfs_list_move_tail(&ext->oe_link,
+						   &obj->oo_urgent_exts);
+		}
+		osc_object_unlock(obj);
+
+		osc_io_unplug_async(env, osc_cli(obj), obj);
+	}
+	osc_extent_put(env, ext);
+	RETURN(rc);
+}
+
+static inline int overlapped(struct osc_extent *ex1, struct osc_extent *ex2)
+{
+	return !(ex1->oe_end < ex2->oe_start || ex2->oe_end < ex1->oe_start);
+}
+
+/**
+ * Find or create an extent which includes @index, core function to manage
+ * extent tree.
+ */
+struct osc_extent *osc_extent_find(const struct lu_env *env,
+				   struct osc_object *obj, pgoff_t index,
+				   int *grants)
+
+{
+	struct client_obd *cli = osc_cli(obj);
+	struct cl_lock    *lock;
+	struct osc_extent *cur;
+	struct osc_extent *ext;
+	struct osc_extent *conflict = NULL;
+	struct osc_extent *found = NULL;
+	pgoff_t    chunk;
+	pgoff_t    max_end;
+	int        max_pages; /* max_pages_per_rpc */
+	int        chunksize;
+	int        ppc_bits; /* pages per chunk bits */
+	int        chunk_mask;
+	int        rc;
+	ENTRY;
+
+	cur = osc_extent_alloc(obj);
+	if (cur == NULL)
+		RETURN(ERR_PTR(-ENOMEM));
+
+	lock = cl_lock_at_pgoff(env, osc2cl(obj), index, NULL, 1, 0);
+	LASSERT(lock != NULL);
+	LASSERT(lock->cll_descr.cld_mode >= CLM_WRITE);
+
+	LASSERT(cli->cl_chunkbits >= CFS_PAGE_SHIFT);
+	ppc_bits   = cli->cl_chunkbits - CFS_PAGE_SHIFT;
+	chunk_mask = ~((1 << ppc_bits) - 1);
+	chunksize  = 1 << cli->cl_chunkbits;
+	chunk      = index >> ppc_bits;
+
+	/* align end to rpc edge, rpc size may not be a power 2 integer. */
+	max_pages = cli->cl_max_pages_per_rpc;
+	LASSERT((max_pages & ~chunk_mask) == 0);
+	max_end = index - (index % max_pages) + max_pages - 1;
+	max_end = min_t(pgoff_t, max_end, lock->cll_descr.cld_end);
+
+	/* initialize new extent by parameters so far */
+	cur->oe_max_end = max_end;
+	cur->oe_start   = index & chunk_mask;
+	cur->oe_end     = ((index + ~chunk_mask + 1) & chunk_mask) - 1;
+	if (cur->oe_start < lock->cll_descr.cld_start)
+		cur->oe_start = lock->cll_descr.cld_start;
+	if (cur->oe_end > max_end)
+		cur->oe_end = max_end;
+	cur->oe_osclock = lock;
+	cur->oe_grants  = 0;
+	cur->oe_mppr    = max_pages;
+
+	/* grants has been allocated by caller */
+	LASSERTF(*grants >= chunksize + cli->cl_extent_tax,
+		 "%u/%u/%u.\n", *grants, chunksize, cli->cl_extent_tax);
+	LASSERTF((max_end - cur->oe_start) < max_pages, EXTSTR, EXTPARA(cur));
+
+restart:
+	osc_object_lock(obj);
+	ext = osc_extent_search(obj, cur->oe_start);
+	if (ext == NULL)
+		ext = first_extent(obj);
+	while (ext != NULL) {
+		loff_t ext_chk_start = ext->oe_start >> ppc_bits;
+		loff_t ext_chk_end   = ext->oe_end   >> ppc_bits;
+
+		LASSERT(sanity_check_nolock(ext) == 0);
+		if (chunk > ext_chk_end + 1)
+			break;
+
+		/* if covering by different locks, no chance to match */
+		if (lock != ext->oe_osclock) {
+			EASSERTF(!overlapped(ext, cur), ext,
+				 EXTSTR, EXTPARA(cur));
+
+			ext = next_extent(ext);
+			continue;
+		}
+
+		/* discontiguous chunks? */
+		if (chunk + 1 < ext_chk_start) {
+			ext = next_extent(ext);
+			continue;
+		}
+
+		/* ok, from now on, ext and cur have these attrs:
+		 * 1. covered by the same lock
+		 * 2. contiguous at chunk level or overlapping. */
+
+		if (overlapped(ext, cur)) {
+			/* cur is the minimum unit, so overlapping means
+			 * full contain. */
+			EASSERTF((ext->oe_start <= cur->oe_start &&
+				  ext->oe_end >= cur->oe_end),
+				 ext, EXTSTR, EXTPARA(cur));
+
+			if (ext->oe_state > OES_CACHE || ext->oe_fsync_wait) {
+				/* for simplicity, we wait for this extent to
+				 * finish before going forward. */
+				conflict = osc_extent_get(ext);
+				break;
+			}
+
+			found = osc_extent_hold(ext);
+			break;
+		}
+
+		/* non-overlapped extent */
+		if (ext->oe_state != OES_CACHE || ext->oe_fsync_wait) {
+			/* we can't do anything for a non OES_CACHE extent, or
+			 * if there is someone waiting for this extent to be
+			 * flushed, try next one. */
+			ext = next_extent(ext);
+			continue;
+		}
+
+		/* check if they belong to the same rpc slot before trying to
+		 * merge. the extents are not overlapped and contiguous at
+		 * chunk level to get here. */
+		if (ext->oe_max_end != max_end) {
+			/* if they don't belong to the same RPC slot or
+			 * max_pages_per_rpc has ever changed, do not merge. */
+			ext = next_extent(ext);
+			continue;
+		}
+
+		/* it's required that an extent must be contiguous at chunk
+		 * level so that we know the whole extent is covered by grant
+		 * (the pages in the extent are NOT required to be contiguous).
+		 * Otherwise, it will be too much difficult to know which
+		 * chunks have grants allocated. */
+
+		/* try to do front merge - extend ext's start */
+		if (chunk + 1 == ext_chk_start) {
+			/* ext must be chunk size aligned */
+			EASSERT((ext->oe_start & ~chunk_mask) == 0, ext);
+
+			/* pull ext's start back to cover cur */
+			ext->oe_start   = cur->oe_start;
+			ext->oe_grants += chunksize;
+			*grants -= chunksize;
+
+			found = osc_extent_hold(ext);
+		} else if (chunk == ext_chk_end + 1) {
+			/* rear merge */
+			ext->oe_end     = cur->oe_end;
+			ext->oe_grants += chunksize;
+			*grants -= chunksize;
+
+			/* try to merge with the next one because we just fill
+			 * in a gap */
+			if (osc_extent_merge(env, ext, next_extent(ext)) == 0)
+				/* we can save extent tax from next extent */
+				*grants += cli->cl_extent_tax;
+
+			found = osc_extent_hold(ext);
+		}
+		if (found != NULL)
+			break;
+
+		ext = next_extent(ext);
+	}
+
+	osc_extent_tree_dump(D_CACHE, obj);
+	if (found != NULL) {
+		LASSERT(conflict == NULL);
+		if (!IS_ERR(found)) {
+			LASSERT(found->oe_osclock == cur->oe_osclock);
+			OSC_EXTENT_DUMP(D_CACHE, found,
+					"found caching ext for %lu.\n", index);
+		}
+	} else if (conflict == NULL) {
+		/* create a new extent */
+		EASSERT(osc_extent_is_overlapped(obj, cur) == 0, cur);
+		cur->oe_grants = chunksize + cli->cl_extent_tax;
+		*grants -= cur->oe_grants;
+		LASSERT(*grants >= 0);
+
+		cur->oe_state = OES_CACHE;
+		found = osc_extent_hold(cur);
+		osc_extent_insert(obj, cur);
+		OSC_EXTENT_DUMP(D_CACHE, cur, "add into tree %lu/%lu.\n",
+				index, lock->cll_descr.cld_end);
+	}
+	osc_object_unlock(obj);
+
+	if (conflict != NULL) {
+		LASSERT(found == NULL);
+
+		/* waiting for IO to finish. Please notice that it's impossible
+		 * to be an OES_TRUNC extent. */
+		rc = osc_extent_wait(env, conflict, OES_INV);
+		osc_extent_put(env, conflict);
+		conflict = NULL;
+		if (rc < 0)
+			GOTO(out, found = ERR_PTR(rc));
+
+		goto restart;
+	}
+	EXIT;
+
+out:
+	osc_extent_put(env, cur);
+	LASSERT(*grants >= 0);
+	return found;
+}
+
+/**
+ * Called when IO is finished to an extent.
+ */
+int osc_extent_finish(const struct lu_env *env, struct osc_extent *ext,
+		      int sent, int rc)
+{
+	struct client_obd *cli = osc_cli(ext->oe_obj);
+	struct osc_async_page *oap;
+	struct osc_async_page *tmp;
+	struct osc_async_page *last = NULL;
+	int nr_pages = ext->oe_nr_pages;
+	int lost_grant = 0;
+	int blocksize = cli->cl_import->imp_obd->obd_osfs.os_bsize ? : 4096;
+	ENTRY;
+
+	OSC_EXTENT_DUMP(D_CACHE, ext, "extent finished.\n");
+
+	ext->oe_rc = rc ?: ext->oe_nr_pages;
+	EASSERT(ergo(rc == 0, ext->oe_state == OES_RPC), ext);
+	cfs_list_for_each_entry_safe(oap, tmp, &ext->oe_pages,
+				     oap_pending_item) {
+		cfs_list_del_init(&oap->oap_rpc_item);
+		cfs_list_del_init(&oap->oap_pending_item);
+		if (last == NULL || last->oap_obj_off < oap->oap_obj_off)
+			last = oap;
+
+		--ext->oe_nr_pages;
+		osc_ap_completion(env, cli, oap, sent, rc);
+	}
+	EASSERT(ext->oe_nr_pages == 0, ext);
+
+	if (!sent) {
+		lost_grant = ext->oe_grants;
+	} else if (blocksize < CFS_PAGE_SIZE &&
+		   last->oap_count != CFS_PAGE_SIZE) {
+		/* For short writes we shouldn't count parts of pages that
+		 * span a whole chunk on the OST side, or our accounting goes
+		 * wrong.  Should match the code in filter_grant_check. */
+		int offset = oap->oap_page_off & ~CFS_PAGE_MASK;
+		int count = oap->oap_count + (offset & (blocksize - 1));
+		int end = (offset + oap->oap_count) & (blocksize - 1);
+		if (end)
+			count += blocksize - end;
+
+		lost_grant = CFS_PAGE_SIZE - count;
+	}
+	if (ext->oe_grants > 0)
+		osc_free_grant(cli, nr_pages, lost_grant);
+
+	osc_extent_remove(ext);
+	/* put the refcount for RPC */
+	osc_extent_put(env, ext);
+	RETURN(0);
+}
+
+/**
+ * Wait for the extent's state to become @state.
+ */
+static int osc_extent_wait(const struct lu_env *env, struct osc_extent *ext,
+			   int state)
+{
+	struct osc_object *obj = ext->oe_obj;
+	struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
+	int rc = 0;
+	ENTRY;
+
+	osc_object_lock(obj);
+	LASSERT(sanity_check_nolock(ext) == 0);
+	/* `Kick' this extent only if the caller is waiting for it to be
+	 * written out. */
+	if (state == OES_INV && !ext->oe_urgent && !ext->oe_hp) {
+		if (ext->oe_state == OES_ACTIVE) {
+			ext->oe_urgent = 1;
+		} else if (ext->oe_state == OES_CACHE) {
+			ext->oe_urgent = 1;
+			osc_extent_hold(ext);
+			rc = 1;
+		}
+	}
+	osc_object_unlock(obj);
+	if (rc == 1)
+		osc_extent_release(env, ext);
+
+	/* wait for the extent until its state becomes @state */
+	rc = l_wait_event(ext->oe_waitq, ext->oe_state == state, &lwi);
+	if (rc == 0 && ext->oe_rc < 0)
+		rc = ext->oe_rc;
+	RETURN(rc);
+}
+
+/**
+ * Discard pages with index greater than @size. If @ext is overlapped with
+ * @size, then partial truncate happens.
+ */
+static int osc_extent_truncate(struct osc_extent *ext, pgoff_t trunc_index)
+{
+	struct cl_env_nest     nest;
+	struct lu_env         *env;
+	struct cl_io          *io;
+	struct osc_object     *obj = ext->oe_obj;
+	struct client_obd     *cli = osc_cli(obj);
+	struct osc_async_page *oap;
+	struct osc_async_page *tmp;
+	int                    pages_in_chunk = 0;
+	int                    ppc_bits    = cli->cl_chunkbits - CFS_PAGE_SHIFT;
+	__u64                  trunc_chunk = trunc_index >> ppc_bits;
+	int                    grants   = 0;
+	int                    nr_pages = 0;
+	int                    rc       = 0;
+	ENTRY;
+
+	LASSERT(sanity_check(ext) == 0);
+	LASSERT(ext->oe_state == OES_TRUNC);
+	LASSERT(!ext->oe_urgent);
+
+	/* Request new lu_env.
+	 * We can't use that env from osc_cache_truncate_start() because
+	 * it's from lov_io_sub and not fully initialized. */
+	env = cl_env_nested_get(&nest);
+	io  = &osc_env_info(env)->oti_io;
+	io->ci_obj = cl_object_top(osc2cl(obj));
+	rc = cl_io_init(env, io, CIT_MISC, io->ci_obj);
+	if (rc < 0)
+		GOTO(out, rc);
+
+	/* discard all pages with index greater then trunc_index */
+	cfs_list_for_each_entry_safe(oap, tmp, &ext->oe_pages,
+				     oap_pending_item) {
+		struct cl_page  *sub  = oap2cl_page(oap);
+		struct cl_page  *page = cl_page_top(sub);
+
+		LASSERT(cfs_list_empty(&oap->oap_rpc_item));
+
+		/* only discard the pages with their index greater than
+		 * trunc_index, and ... */
+		if (sub->cp_index < trunc_index) {
+			/* accounting how many pages remaining in the chunk
+			 * so that we can calculate grants correctly. */
+			if (sub->cp_index >> ppc_bits == trunc_chunk)
+				++pages_in_chunk;
+			continue;
+		}
+
+		cfs_list_del_init(&oap->oap_pending_item);
+
+		cl_page_get(page);
+		lu_ref_add(&page->cp_reference, "truncate", cfs_current());
+
+		if (cl_page_own(env, io, page) == 0) {
+			cl_page_unmap(env, io, page);
+			cl_page_discard(env, io, page);
+			cl_page_disown(env, io, page);
+		} else {
+			LASSERT(page->cp_state == CPS_FREEING);
+			LASSERT(0);
+		}
+
+		lu_ref_del(&page->cp_reference, "truncate", cfs_current());
+		cl_page_put(env, page);
+
+		--ext->oe_nr_pages;
+		++nr_pages;
+	}
+	EASSERTF(ergo(ext->oe_start >= trunc_index, ext->oe_nr_pages == 0),
+		 ext, "trunc_index %lu\n", trunc_index);
+
+	osc_object_lock(obj);
+	if (ext->oe_nr_pages == 0) {
+		LASSERT(pages_in_chunk == 0);
+		grants = ext->oe_grants;
+		ext->oe_grants = 0;
+	} else { /* calculate how many grants we can free */
+		int     chunks = (ext->oe_end >> ppc_bits) - trunc_chunk;
+		pgoff_t last_index;
+
+
+		/* if there is no pages in this chunk, we can also free grants
+		 * for the last chunk */
+		if (pages_in_chunk == 0) {
+			/* if this is the 1st chunk and no pages in this chunk,
+			 * ext->oe_nr_pages must be zero, so we should be in
+			 * the other if-clause. */
+			LASSERT(trunc_chunk > 0);
+			--trunc_chunk;
+			++chunks;
+		}
+
+		/* this is what we can free from this extent */
+		grants          = chunks << cli->cl_chunkbits;
+		ext->oe_grants -= grants;
+		last_index      = ((trunc_chunk + 1) << ppc_bits) - 1;
+		ext->oe_end     = min(last_index, ext->oe_max_end);
+		LASSERT(ext->oe_end >= ext->oe_start);
+		LASSERT(ext->oe_grants > 0);
+	}
+	osc_object_unlock(obj);
+
+	if (grants > 0 || nr_pages > 0)
+		osc_free_grant(cli, nr_pages, grants);
+
+out:
+	cl_io_fini(env, io);
+	cl_env_nested_put(&nest, env);
+	RETURN(rc);
+}
+
+/**
+ * This function is used to make the extent prepared for transfer.
+ * A race with flusing page - ll_writepage() has to be handled cautiously.
+ */
+static int osc_extent_make_ready(const struct lu_env *env,
+				 struct osc_extent *ext)
+{
+	struct osc_async_page *oap;
+	struct osc_async_page *last = NULL;
+	struct osc_object *obj = ext->oe_obj;
+	int page_count = 0;
+	int rc;
+	ENTRY;
+
+	/* we're going to grab page lock, so object lock must not be taken. */
+	LASSERT(sanity_check(ext) == 0);
+	/* in locking state, any process should not touch this extent. */
+	EASSERT(ext->oe_state == OES_LOCKING, ext);
+	EASSERT(ext->oe_owner != NULL, ext);
+
+	OSC_EXTENT_DUMP(D_CACHE, ext, "make ready\n");
+
+	cfs_list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
+		++page_count;
+		if (last == NULL || last->oap_obj_off < oap->oap_obj_off)
+			last = oap;
+
+		/* checking ASYNC_READY is race safe */
+		if ((oap->oap_async_flags & ASYNC_READY) != 0)
+			continue;
+
+		rc = osc_make_ready(env, oap, OBD_BRW_WRITE);
+		switch (rc) {
+		case 0:
+			cfs_spin_lock(&oap->oap_lock);
+			oap->oap_async_flags |= ASYNC_READY;
+			cfs_spin_unlock(&oap->oap_lock);
+			break;
+		case -EALREADY:
+			LASSERT((oap->oap_async_flags & ASYNC_READY) != 0);
+			break;
+		default:
+			LASSERTF(0, "unknown return code: %d\n", rc);
+		}
+	}
+
+	LASSERT(page_count == ext->oe_nr_pages);
+	LASSERT(last != NULL);
+	/* the last page is the only one we need to refresh its count by
+	 * the size of file. */
+	if (!(last->oap_async_flags & ASYNC_COUNT_STABLE)) {
+		last->oap_count = osc_refresh_count(env, last, OBD_BRW_WRITE);
+		LASSERT(last->oap_count > 0);
+		LASSERT(last->oap_page_off + last->oap_count <= CFS_PAGE_SIZE);
+		last->oap_async_flags |= ASYNC_COUNT_STABLE;
+	}
+
+	/* for the rest of pages, we don't need to call osf_refresh_count()
+	 * because it's known they are not the last page */
+	cfs_list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
+		if (!(oap->oap_async_flags & ASYNC_COUNT_STABLE)) {
+			oap->oap_count = CFS_PAGE_SIZE - oap->oap_page_off;
+			oap->oap_async_flags |= ASYNC_COUNT_STABLE;
+		}
+	}
+
+	osc_object_lock(obj);
+	osc_extent_state_set(ext, OES_RPC);
+	osc_object_unlock(obj);
+	/* get a refcount for RPC. */
+	osc_extent_get(ext);
+
+	RETURN(0);
+}
+
+/**
+ * Quick and simple version of osc_extent_find(). This function is frequently
+ * called to expand the extent for the same IO. To expand the extent, the
+ * page index must be in the same or next chunk of ext->oe_end.
+ */
+static int osc_extent_expand(struct osc_extent *ext, pgoff_t index, int *grants)
+{
+	struct osc_object *obj = ext->oe_obj;
+	struct client_obd *cli = osc_cli(obj);
+	struct osc_extent *next;
+	int ppc_bits = cli->cl_chunkbits - CFS_PAGE_SHIFT;
+	pgoff_t chunk = index >> ppc_bits;
+	pgoff_t end_chunk;
+	pgoff_t end_index;
+	int chunksize = 1 << cli->cl_chunkbits;
+	int rc = 0;
+	ENTRY;
+
+	LASSERT(ext->oe_max_end >= index && ext->oe_start <= index);
+	osc_object_lock(obj);
+	LASSERT(sanity_check_nolock(ext) == 0);
+	end_chunk = ext->oe_end >> ppc_bits;
+	if (chunk > end_chunk + 1)
+		GOTO(out, rc = -ERANGE);
+
+	if (end_chunk >= chunk)
+		GOTO(out, rc = 0);
+
+	LASSERT(end_chunk + 1 == chunk);
+	/* try to expand this extent to cover @index */
+	end_index = min(ext->oe_max_end, ((chunk + 1) << ppc_bits) - 1);
+
+	next = next_extent(ext);
+	if (next != NULL && next->oe_start <= end_index)
+		/* complex mode - overlapped with the next extent,
+		 * this case will be handled by osc_extent_find() */
+		GOTO(out, rc = -EAGAIN);
+
+	ext->oe_end = end_index;
+	ext->oe_grants += chunksize;
+	*grants -= chunksize;
+	LASSERT(*grants >= 0);
+	EASSERTF(osc_extent_is_overlapped(obj, ext) == 0, ext,
+		 "overlapped after expanding for %lu.\n", index);
+	EXIT;
+
+out:
+	osc_object_unlock(obj);
+	RETURN(rc);
+}
+
+static void osc_extent_tree_dump0(int level, struct osc_object *obj,
+				  const char *func, int line)
+{
+	struct osc_extent *ext;
+	int cnt;
+
+	CDEBUG(level, "Dump object %p extents at %s:%d, mppr: %u.\n",
+	       obj, func, line, osc_cli(obj)->cl_max_pages_per_rpc);
+
+	/* osc_object_lock(obj); */
+	cnt = 1;
+	for (ext = first_extent(obj); ext != NULL; ext = next_extent(ext))
+		OSC_EXTENT_DUMP(level, ext, "in tree %d.\n", cnt++);
+
+	cnt = 1;
+	cfs_list_for_each_entry(ext, &obj->oo_hp_exts, oe_link)
+		OSC_EXTENT_DUMP(level, ext, "hp %d.\n", cnt++);
+
+	cnt = 1;
+	cfs_list_for_each_entry(ext, &obj->oo_urgent_exts, oe_link)
+		OSC_EXTENT_DUMP(level, ext, "urgent %d.\n", cnt++);
+
+	cnt = 1;
+	cfs_list_for_each_entry(ext, &obj->oo_reading_exts, oe_link)
+		OSC_EXTENT_DUMP(level, ext, "reading %d.\n", cnt++);
+	/* osc_object_unlock(obj); */
+}
+
+/* ------------------ osc extent end ------------------ */
 
-static inline struct osc_page *oap2osc_page(struct osc_async_page *oap)
+static inline int osc_is_ready(struct osc_object *osc)
 {
-	return (struct osc_page *)container_of(oap, struct osc_page, ops_oap);
+	return !cfs_list_empty(&osc->oo_ready_item) ||
+	       !cfs_list_empty(&osc->oo_hp_ready_item);
 }
 
+#define OSC_IO_DEBUG(OSC, STR, args...)					       \
+	CDEBUG(D_CACHE, "obj %p ready %d|%c|%c wr %d|%c|%c rd %d|%c " STR,     \
+	       (OSC), osc_is_ready(OSC),				       \
+	       list_empty_marker(&(OSC)->oo_hp_ready_item),		       \
+	       list_empty_marker(&(OSC)->oo_ready_item),		       \
+	       cfs_atomic_read(&(OSC)->oo_nr_writes),			       \
+	       list_empty_marker(&(OSC)->oo_hp_exts),			       \
+	       list_empty_marker(&(OSC)->oo_urgent_exts),		       \
+	       cfs_atomic_read(&(OSC)->oo_nr_reads),			       \
+	       list_empty_marker(&(OSC)->oo_reading_exts),		       \
+	       ##args)
+
 static int osc_make_ready(const struct lu_env *env, struct osc_async_page *oap,
 			  int cmd)
 {
 	struct osc_page *opg  = oap2osc_page(oap);
-	struct cl_page  *page = cl_page_top(opg->ops_cl.cpl_page);
+	struct cl_page  *page = cl_page_top(oap2cl_page(oap));
 	int result;
 
 	LASSERT(cmd == OBD_BRW_WRITE); /* no cached reads */
@@ -92,7 +1190,7 @@ static int osc_refresh_count(const struct lu_env *env,
 			     struct osc_async_page *oap, int cmd)
 {
 	struct osc_page  *opg = oap2osc_page(oap);
-	struct cl_page   *page;
+	struct cl_page   *page = oap2cl_page(oap);
 	struct cl_object *obj;
 	struct cl_attr   *attr = &osc_env_info(env)->oti_attr;
 
@@ -102,7 +1200,6 @@ static int osc_refresh_count(const struct lu_env *env,
 	/* readpage queues with _COUNT_STABLE, shouldn't get here. */
 	LASSERT(!(cmd & OBD_BRW_READ));
 	LASSERT(opg != NULL);
-	page = opg->ops_cl.cpl_page;
 	obj = opg->ops_cl.cpl_obj;
 
 	cl_object_attr_lock(obj);
@@ -122,10 +1219,10 @@ static int osc_refresh_count(const struct lu_env *env,
 }
 
 static int osc_completion(const struct lu_env *env, struct osc_async_page *oap,
-			  int cmd, struct obdo *oa, int rc)
+			  int cmd, int rc)
 {
 	struct osc_page   *opg  = oap2osc_page(oap);
-	struct cl_page    *page = cl_page_top(opg->ops_cl.cpl_page);
+	struct cl_page    *page = cl_page_top(oap2cl_page(oap));
 	struct osc_object *obj  = cl2osc(opg->ops_cl.cpl_obj);
 	enum cl_req_type   crt;
 	int srvlock;
@@ -145,11 +1242,6 @@ static int osc_completion(const struct lu_env *env, struct osc_async_page *oap,
 		cl_req_page_done(env, page);
 	LASSERT(page->cp_req == NULL);
 
-	/* As the transfer for this page is being done, clear the flags */
-	cfs_spin_lock(&oap->oap_lock);
-	oap->oap_async_flags = 0;
-	cfs_spin_unlock(&oap->oap_lock);
-
 	crt = cmd == OBD_BRW_READ ? CRT_READ : CRT_WRITE;
 	/* Clear opg->ops_transfer_pinned before VM lock is released. */
 	opg->ops_transfer_pinned = 0;
@@ -158,6 +1250,7 @@ static int osc_completion(const struct lu_env *env, struct osc_async_page *oap,
 	LASSERT(opg->ops_submitter != NULL);
 	LASSERT(!cfs_list_empty(&opg->ops_inflight));
 	cfs_list_del_init(&opg->ops_inflight);
+	opg->ops_submitter = NULL;
 	cfs_spin_unlock(&obj->oo_seatbelt);
 
 	opg->ops_submit_time = 0;
@@ -192,6 +1285,17 @@ static int osc_completion(const struct lu_env *env, struct osc_async_page *oap,
 	RETURN(0);
 }
 
+#define OSC_DUMP_GRANT(cli, fmt, args...) do {				      \
+	struct client_obd *__tmp = (cli);				      \
+	CDEBUG(D_CACHE, "%s: { dirty: %ld/%ld dirty_pages: %d/%d "	      \
+	       "dropped: %ld avail: %ld, reserved: %ld, flight: %d } " fmt,   \
+	       __tmp->cl_import->imp_obd->obd_name,			      \
+	       __tmp->cl_dirty, __tmp->cl_dirty_max,			      \
+	       cfs_atomic_read(&obd_dirty_pages), obd_max_dirty_pages,	      \
+	       __tmp->cl_lost_grant, __tmp->cl_avail_grant,		      \
+	       __tmp->cl_reserved_grant, __tmp->cl_w_in_flight, ##args);      \
+} while (0)
+
 /* caller must hold loi_list_lock */
 static void osc_consume_write_grant(struct client_obd *cli,
 				    struct brw_page *pga)
@@ -200,20 +1304,17 @@ static void osc_consume_write_grant(struct client_obd *cli,
 	LASSERT(!(pga->flag & OBD_BRW_FROM_GRANT));
 	cfs_atomic_inc(&obd_dirty_pages);
 	cli->cl_dirty += CFS_PAGE_SIZE;
-	cli->cl_avail_grant -= CFS_PAGE_SIZE;
 	pga->flag |= OBD_BRW_FROM_GRANT;
 	CDEBUG(D_CACHE, "using %lu grant credits for brw %p page %p\n",
 	       CFS_PAGE_SIZE, pga, pga->pg);
-	LASSERT(cli->cl_avail_grant >= 0);
 	osc_update_next_shrink(cli);
 }
 
 /* the companion to osc_consume_write_grant, called when a brw has completed.
  * must be called with the loi lock held. */
 static void osc_release_write_grant(struct client_obd *cli,
-				    struct brw_page *pga, int sent)
+				    struct brw_page *pga)
 {
-	int blocksize = cli->cl_import->imp_obd->obd_osfs.os_bsize ? : 4096;
 	ENTRY;
 
 	LASSERT_SPIN_LOCKED(&cli->cl_loi_list_lock.lock);
@@ -230,72 +1331,132 @@ static void osc_release_write_grant(struct client_obd *cli,
 		cfs_atomic_dec(&obd_dirty_transit_pages);
 		cli->cl_dirty_transit -= CFS_PAGE_SIZE;
 	}
-	if (!sent) {
-		/* Reclaim grant from truncated pages. This is used to solve
-		 * write-truncate and grant all gone(to lost_grant) problem.
-		 * For a vfs write this problem can be easily solved by a sync
-		 * write, however, this is not an option for page_mkwrite()
-		 * because grant has to be allocated before a page becomes
-		 * dirty. */
-		if (cli->cl_avail_grant < PTLRPC_MAX_BRW_SIZE)
-			cli->cl_avail_grant += CFS_PAGE_SIZE;
-		else
-			cli->cl_lost_grant += CFS_PAGE_SIZE;
-		CDEBUG(D_CACHE, "lost grant: %lu avail grant: %lu dirty: %lu\n",
-		       cli->cl_lost_grant, cli->cl_avail_grant, cli->cl_dirty);
-	} else if (CFS_PAGE_SIZE != blocksize && pga->count != CFS_PAGE_SIZE) {
-		/* For short writes we shouldn't count parts of pages that
-		 * span a whole block on the OST side, or our accounting goes
-		 * wrong.  Should match the code in filter_grant_check. */
-		int offset = pga->off & ~CFS_PAGE_MASK;
-		int count = pga->count + (offset & (blocksize - 1));
-		int end = (offset + pga->count) & (blocksize - 1);
-		if (end)
-			count += blocksize - end;
+	EXIT;
+}
+
+/**
+ * To avoid sleeping with object lock held, it's good for us allocate enough
+ * grants before entering into critical section.
+ *
+ * client_obd_list_lock held by caller
+ */
+static int osc_reserve_grant(struct client_obd *cli, unsigned int bytes)
+{
+	int rc = -EDQUOT;
 
-		cli->cl_lost_grant += CFS_PAGE_SIZE - count;
-		CDEBUG(D_CACHE, "lost %lu grant: %lu avail: %lu dirty: %lu\n",
-		       CFS_PAGE_SIZE - count, cli->cl_lost_grant,
-		       cli->cl_avail_grant, cli->cl_dirty);
+	if (cli->cl_avail_grant >= bytes) {
+		cli->cl_avail_grant    -= bytes;
+		cli->cl_reserved_grant += bytes;
+		rc = 0;
 	}
+	return rc;
+}
 
-	EXIT;
+static void __osc_unreserve_grant(struct client_obd *cli,
+				  unsigned int reserved, unsigned int unused)
+{
+	/* it's quite normal for us to get more grant than reserved.
+	 * Thinking about a case that two extents merged by adding a new
+	 * chunk, we can save one extent tax. If extent tax is greater than
+	 * one chunk, we can save more grant by adding a new chunk */
+	cli->cl_reserved_grant -= reserved;
+	if (unused > reserved) {
+		cli->cl_avail_grant += reserved;
+		cli->cl_lost_grant  += unused - reserved;
+	} else {
+		cli->cl_avail_grant += unused;
+	}
+	if (unused > 0)
+		osc_wake_cache_waiters(cli);
+}
+
+void osc_unreserve_grant(struct client_obd *cli,
+			 unsigned int reserved, unsigned int unused)
+{
+	client_obd_list_lock(&cli->cl_loi_list_lock);
+	__osc_unreserve_grant(cli, reserved, unused);
+	client_obd_list_unlock(&cli->cl_loi_list_lock);
+}
+
+/**
+ * Free grant after IO is finished or canceled.
+ *
+ * @lost_grant is used to remember how many grants we have allocated but not
+ * used, we should return these grants to OST. There're two cases where grants
+ * can be lost:
+ * 1. truncate;
+ * 2. blocksize at OST is less than CFS_PAGE_SIZE and a partial page was
+ *    written. In this case OST may use less chunks to serve this partial
+ *    write. OSTs don't actually know the page size on the client side. so
+ *    clients have to calculate lost grant by the blocksize on the OST.
+ *    See filter_grant_check() for details.
+ */
+static void osc_free_grant(struct client_obd *cli, unsigned int nr_pages,
+			   unsigned int lost_grant)
+{
+	int grant = (1 << cli->cl_chunkbits) + cli->cl_extent_tax;
+
+	client_obd_list_lock(&cli->cl_loi_list_lock);
+	cfs_atomic_sub(nr_pages, &obd_dirty_pages);
+	cli->cl_dirty -= nr_pages << CFS_PAGE_SHIFT;
+	cli->cl_lost_grant += lost_grant;
+	if (cli->cl_avail_grant < grant && cli->cl_lost_grant >= grant) {
+		/* borrow some grant from truncate to avoid the case that
+		 * truncate uses up all avail grant */
+		cli->cl_lost_grant -= grant;
+		cli->cl_avail_grant += grant;
+	}
+	osc_wake_cache_waiters(cli);
+	client_obd_list_unlock(&cli->cl_loi_list_lock);
+	CDEBUG(D_CACHE, "lost %u grant: %lu avail: %lu dirty: %lu\n",
+	       lost_grant, cli->cl_lost_grant,
+	       cli->cl_avail_grant, cli->cl_dirty);
 }
 
 /* The companion to osc_enter_cache(), called when @oap is no longer part of
  * the dirty accounting.  Writeback completes or truncate happens before
  * writing starts.  Must be called with the loi lock held. */
-static void osc_exit_cache(struct client_obd *cli, struct osc_async_page *oap,
-			   int sent)
+static void osc_exit_cache(struct client_obd *cli, struct osc_async_page *oap)
 {
-	osc_release_write_grant(cli, &oap->oap_brw_page, sent);
+	osc_release_write_grant(cli, &oap->oap_brw_page);
 }
 
 /**
  * Non-blocking version of osc_enter_cache() that consumes grant only when it
  * is available.
  */
-static int osc_enter_cache_try(const struct lu_env *env, struct client_obd *cli,
-			       struct osc_async_page *oap, int transient)
+static int osc_enter_cache_try(struct client_obd *cli,
+			       struct osc_async_page *oap,
+			       int bytes, int transient)
 {
-	int has_grant;
+	int rc;
 
-	has_grant = cli->cl_avail_grant >= CFS_PAGE_SIZE;
-	if (has_grant) {
+	OSC_DUMP_GRANT(cli, "need:%d.\n", bytes);
+
+	rc = osc_reserve_grant(cli, bytes);
+	if (rc < 0)
+		return 0;
+
+	if (cli->cl_dirty + CFS_PAGE_SIZE <= cli->cl_dirty_max &&
+	    cfs_atomic_read(&obd_dirty_pages) + 1 <= obd_max_dirty_pages) {
 		osc_consume_write_grant(cli, &oap->oap_brw_page);
 		if (transient) {
 			cli->cl_dirty_transit += CFS_PAGE_SIZE;
 			cfs_atomic_inc(&obd_dirty_transit_pages);
 			oap->oap_brw_flags |= OBD_BRW_NOCACHE;
 		}
+		rc = 1;
+	} else {
+		__osc_unreserve_grant(cli, bytes, bytes);
+		rc = 0;
 	}
-	return has_grant;
+	return rc;
 }
 
 /* Caller must hold loi_list_lock - we drop/regain it if we need to wait for
  * grant or cache space. */
 static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli,
-			   struct osc_async_page *oap)
+			   struct osc_async_page *oap, int bytes)
 {
 	struct osc_object *osc = oap->oap_obj;
 	struct lov_oinfo  *loi = osc->oo_oinfo;
@@ -304,23 +1465,20 @@ static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli,
 	int rc = -EDQUOT;
 	ENTRY;
 
-	CDEBUG(D_CACHE, "dirty: %ld/%d dirty_max: %ld/%d dropped: %lu "
-	       "grant: %lu\n", cli->cl_dirty, cfs_atomic_read(&obd_dirty_pages),
-	       cli->cl_dirty_max, obd_max_dirty_pages,
-	       cli->cl_lost_grant, cli->cl_avail_grant);
+	OSC_DUMP_GRANT(cli, "need:%d.\n", bytes);
+
+	client_obd_list_lock(&cli->cl_loi_list_lock);
 
 	/* force the caller to try sync io.  this can jump the list
 	 * of queued writes and create a discontiguous rpc stream */
 	if (OBD_FAIL_CHECK(OBD_FAIL_OSC_NO_GRANT) ||
 	    cli->cl_dirty_max < CFS_PAGE_SIZE     ||
 	    cli->cl_ar.ar_force_sync || loi->loi_ar.ar_force_sync)
-		RETURN(-EDQUOT);
+		GOTO(out, rc = -EDQUOT);
 
 	/* Hopefully normal case - cache space and write credits available */
-	if (cli->cl_dirty + CFS_PAGE_SIZE <= cli->cl_dirty_max &&
-	    cfs_atomic_read(&obd_dirty_pages) + 1 <= obd_max_dirty_pages &&
-	    osc_enter_cache_try(env, cli, oap, 0))
-		RETURN(0);
+	if (osc_enter_cache_try(cli, oap, bytes, 0))
+		GOTO(out, rc = 0);
 
 	/* We can get here for two reasons: too many dirty pages in cache, or
 	 * run out of grants. In both cases we should write dirty pages out.
@@ -329,19 +1487,20 @@ static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli,
 	 * The exiting condition is no avail grants and no dirty pages caching,
 	 * that really means there is no space on the OST. */
 	cfs_waitq_init(&ocw.ocw_waitq);
-	ocw.ocw_oap = oap;
-	while (cli->cl_dirty > 0) {
+	ocw.ocw_oap   = oap;
+	ocw.ocw_grant = bytes;
+	while (cli->cl_dirty > 0 || cli->cl_w_in_flight > 0) {
 		cfs_list_add_tail(&ocw.ocw_entry, &cli->cl_cache_waiters);
 		ocw.ocw_rc = 0;
+		client_obd_list_unlock(&cli->cl_loi_list_lock);
 
 		osc_io_unplug(env, cli, osc, PDL_POLICY_ROUND);
-		client_obd_list_unlock(&cli->cl_loi_list_lock);
 
 		CDEBUG(D_CACHE, "%s: sleeping for cache space @ %p for %p\n",
 		       cli->cl_import->imp_obd->obd_name, &ocw, oap);
 
-		rc = l_wait_event(ocw.ocw_waitq, cfs_list_empty(&ocw.ocw_entry),
-				  &lwi);
+		rc = l_wait_event(ocw.ocw_waitq,
+				  cfs_list_empty(&ocw.ocw_entry), &lwi);
 
 		client_obd_list_lock(&cli->cl_loi_list_lock);
 		cfs_list_del_init(&ocw.ocw_entry);
@@ -351,8 +1510,16 @@ static int osc_enter_cache(const struct lu_env *env, struct client_obd *cli,
 		rc = ocw.ocw_rc;
 		if (rc != -EDQUOT)
 			break;
+		if (osc_enter_cache_try(cli, oap, bytes, 0)) {
+			rc = 0;
+			break;
+		}
 	}
+	EXIT;
 
+out:
+	client_obd_list_unlock(&cli->cl_loi_list_lock);
+	OSC_DUMP_GRANT(cli, "returned %d.\n", rc);
 	RETURN(rc);
 }
 
@@ -366,8 +1533,8 @@ void osc_wake_cache_waiters(struct client_obd *cli)
 	cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
 		/* if we can't dirty more, we must wait until some is written */
 		if ((cli->cl_dirty + CFS_PAGE_SIZE > cli->cl_dirty_max) ||
-		   (cfs_atomic_read(&obd_dirty_pages) + 1 >
-		    obd_max_dirty_pages)) {
+		    (cfs_atomic_read(&obd_dirty_pages) + 1 >
+		     obd_max_dirty_pages)) {
 			CDEBUG(D_CACHE, "no dirty room: dirty: %ld "
 			       "osc max %ld, sys max %d\n", cli->cl_dirty,
 			       cli->cl_dirty_max, obd_max_dirty_pages);
@@ -384,17 +1551,13 @@ void osc_wake_cache_waiters(struct client_obd *cli)
 
 		ocw = cfs_list_entry(l, struct osc_cache_waiter, ocw_entry);
 		cfs_list_del_init(&ocw->ocw_entry);
-		if (cli->cl_avail_grant < CFS_PAGE_SIZE) {
-			/* no more RPCs in flight to return grant, do sync IO */
+
+		ocw->ocw_rc = 0;
+		if (!osc_enter_cache_try(cli, ocw->ocw_oap, ocw->ocw_grant, 0))
 			ocw->ocw_rc = -EDQUOT;
-			CDEBUG(D_INODE, "wake oap %p for sync\n", ocw->ocw_oap);
-		} else {
-			osc_consume_write_grant(cli,
-						&ocw->ocw_oap->oap_brw_page);
-		}
 
-		CDEBUG(D_CACHE, "wake up %p for oap %p, avail grant %ld\n",
-		       ocw, ocw->ocw_oap, cli->cl_avail_grant);
+		CDEBUG(D_CACHE, "wake up %p for oap %p, avail grant %ld, %d\n",
+		       ocw, ocw->ocw_oap, cli->cl_avail_grant, ocw->ocw_rc);
 
 		cfs_waitq_signal(&ocw->ocw_waitq);
 	}
@@ -404,21 +1567,7 @@ void osc_wake_cache_waiters(struct client_obd *cli)
 
 static int osc_max_rpc_in_flight(struct client_obd *cli, struct osc_object *osc)
 {
-	struct osc_async_page *oap;
-	int hprpc = 0;
-
-	if (!cfs_list_empty(&osc->oo_write_pages.oop_urgent)) {
-		oap = cfs_list_entry(osc->oo_write_pages.oop_urgent.next,
-				     struct osc_async_page, oap_urgent_item);
-		hprpc = !!(oap->oap_async_flags & ASYNC_HP);
-	}
-
-	if (!hprpc && !cfs_list_empty(&osc->oo_read_pages.oop_urgent)) {
-		oap = cfs_list_entry(osc->oo_read_pages.oop_urgent.next,
-				     struct osc_async_page, oap_urgent_item);
-		hprpc = !!(oap->oap_async_flags & ASYNC_HP);
-	}
-
+	int hprpc = !!cfs_list_empty(&osc->oo_hp_exts);
 	return rpcs_in_flight(cli) >= cli->cl_max_rpcs_in_flight + hprpc;
 }
 
@@ -428,35 +1577,31 @@ static int osc_max_rpc_in_flight(struct client_obd *cli, struct osc_object *osc)
 static int osc_makes_rpc(struct client_obd *cli, struct osc_object *osc,
 			 int cmd)
 {
-	struct osc_oap_pages *lop;
+	int invalid_import = 0;
 	ENTRY;
 
-	if (cmd & OBD_BRW_WRITE) {
-		lop = &osc->oo_write_pages;
-	} else {
-		lop = &osc->oo_read_pages;
-	}
-
-	if (lop->oop_num_pending == 0)
-		RETURN(0);
-
 	/* if we have an invalid import we want to drain the queued pages
 	 * by forcing them through rpcs that immediately fail and complete
 	 * the pages.  recovery relies on this to empty the queued pages
 	 * before canceling the locks and evicting down the llite pages */
-	if (cli->cl_import == NULL || cli->cl_import->imp_invalid)
-		RETURN(1);
-
-	/* stream rpcs in queue order as long as as there is an urgent page
-	 * queued.  this is our cheap solution for good batching in the case
-	 * where writepage marks some random page in the middle of the file
-	 * as urgent because of, say, memory pressure */
-	if (!cfs_list_empty(&lop->oop_urgent)) {
-		CDEBUG(D_CACHE, "urgent request forcing RPC\n");
-		RETURN(1);
-	}
+	if ((cli->cl_import == NULL || cli->cl_import->imp_invalid))
+		invalid_import = 1;
 
 	if (cmd & OBD_BRW_WRITE) {
+		if (cfs_atomic_read(&osc->oo_nr_writes) == 0)
+			RETURN(0);
+		if (invalid_import) {
+			CDEBUG(D_CACHE, "invalid import forcing RPC\n");
+			RETURN(1);
+		}
+		if (!cfs_list_empty(&osc->oo_hp_exts)) {
+			CDEBUG(D_CACHE, "high prio request forcing RPC\n");
+			RETURN(1);
+		}
+		if (!cfs_list_empty(&osc->oo_urgent_exts)) {
+			CDEBUG(D_CACHE, "urgent request forcing RPC\n");
+			RETURN(1);
+		}
 		/* trigger a write rpc stream as long as there are dirtiers
 		 * waiting for space.  as they're waiting, they're not going to
 		 * create more pages to coalesce with what's waiting.. */
@@ -464,40 +1609,42 @@ static int osc_makes_rpc(struct client_obd *cli, struct osc_object *osc,
 			CDEBUG(D_CACHE, "cache waiters forcing RPC\n");
 			RETURN(1);
 		}
+		if (cfs_atomic_read(&osc->oo_nr_writes) >=
+		    cli->cl_max_pages_per_rpc)
+			RETURN(1);
+	} else {
+		if (cfs_atomic_read(&osc->oo_nr_reads) == 0)
+			RETURN(0);
+		if (invalid_import) {
+			CDEBUG(D_CACHE, "invalid import forcing RPC\n");
+			RETURN(1);
+		}
+		/* all read are urgent. */
+		if (!cfs_list_empty(&osc->oo_reading_exts))
+			RETURN(1);
 	}
-	if (lop->oop_num_pending >= cli->cl_max_pages_per_rpc)
-		RETURN(1);
 
 	RETURN(0);
 }
 
-static void lop_update_pending(struct client_obd *cli,
-			       struct osc_oap_pages *lop, int cmd, int delta)
+static void osc_update_pending(struct osc_object *obj, int cmd, int delta)
 {
-	lop->oop_num_pending += delta;
-	if (cmd & OBD_BRW_WRITE)
-		cli->cl_pending_w_pages += delta;
-	else
-		cli->cl_pending_r_pages += delta;
+	struct client_obd *cli = osc_cli(obj);
+	if (cmd & OBD_BRW_WRITE) {
+		cfs_atomic_add(delta, &obj->oo_nr_writes);
+		cfs_atomic_add(delta, &cli->cl_pending_w_pages);
+		LASSERT(cfs_atomic_read(&obj->oo_nr_writes) >= 0);
+	} else {
+		cfs_atomic_add(delta, &obj->oo_nr_reads);
+		cfs_atomic_add(delta, &cli->cl_pending_r_pages);
+		LASSERT(cfs_atomic_read(&obj->oo_nr_reads) >= 0);
+	}
+	OSC_IO_DEBUG(obj, "update pending cmd %d delta %d.\n", cmd, delta);
 }
 
-static int osc_makes_hprpc(struct osc_oap_pages *lop)
+static int osc_makes_hprpc(struct osc_object *obj)
 {
-	struct osc_async_page *oap;
-	ENTRY;
-
-	if (cfs_list_empty(&lop->oop_urgent))
-		RETURN(0);
-
-	oap = cfs_list_entry(lop->oop_urgent.next,
-			 struct osc_async_page, oap_urgent_item);
-
-	if (oap->oap_async_flags & ASYNC_HP) {
-		CDEBUG(D_CACHE, "hp request forcing RPC\n");
-		RETURN(1);
-	}
-
-	RETURN(0);
+	return !cfs_list_empty(&obj->oo_hp_exts);
 }
 
 static void on_list(cfs_list_t *item, cfs_list_t *list, int should_be_on)
@@ -510,10 +1657,9 @@ static void on_list(cfs_list_t *item, cfs_list_t *list, int should_be_on)
 
 /* maintain the osc's cli list membership invariants so that osc_send_oap_rpc
  * can find pages to build into rpcs quickly */
-static void osc_list_maint(struct client_obd *cli, struct osc_object *osc)
+static int __osc_list_maint(struct client_obd *cli, struct osc_object *osc)
 {
-	if (osc_makes_hprpc(&osc->oo_write_pages) ||
-	    osc_makes_hprpc(&osc->oo_read_pages)) {
+	if (osc_makes_hprpc(osc)) {
 		/* HP rpc */
 		on_list(&osc->oo_ready_item, &cli->cl_loi_ready_list, 0);
 		on_list(&osc->oo_hp_ready_item, &cli->cl_loi_hp_ready_list, 1);
@@ -525,10 +1671,23 @@ static void osc_list_maint(struct client_obd *cli, struct osc_object *osc)
 	}
 
 	on_list(&osc->oo_write_item, &cli->cl_loi_write_list,
-		osc->oo_write_pages.oop_num_pending);
+		cfs_atomic_read(&osc->oo_nr_writes) > 0);
 
 	on_list(&osc->oo_read_item, &cli->cl_loi_read_list,
-		osc->oo_read_pages.oop_num_pending);
+		cfs_atomic_read(&osc->oo_nr_reads) > 0);
+
+	return osc_is_ready(osc);
+}
+
+static int osc_list_maint(struct client_obd *cli, struct osc_object *osc)
+{
+	int is_ready;
+
+	client_obd_list_lock(&cli->cl_loi_list_lock);
+	is_ready = __osc_list_maint(cli, osc);
+	client_obd_list_unlock(&cli->cl_loi_list_lock);
+
+	return is_ready;
 }
 
 /* this is trying to propogate async writeback errors back up to the
@@ -553,29 +1712,10 @@ static void osc_process_ar(struct osc_async_rc *ar, __u64 xid,
 		ar->ar_force_sync = 0;
 }
 
-static void osc_oap_to_pending(struct osc_async_page *oap)
-{
-	struct osc_object    *osc = oap->oap_obj;
-	struct osc_oap_pages *lop;
-
-	if (oap->oap_cmd & OBD_BRW_WRITE)
-		lop = &osc->oo_write_pages;
-	else
-		lop = &osc->oo_read_pages;
-
-	if (oap->oap_async_flags & ASYNC_HP)
-		cfs_list_add(&oap->oap_urgent_item, &lop->oop_urgent);
-	else if (oap->oap_async_flags & ASYNC_URGENT)
-		cfs_list_add_tail(&oap->oap_urgent_item, &lop->oop_urgent);
-	cfs_list_add_tail(&oap->oap_pending_item, &lop->oop_pending);
-	lop_update_pending(oap->oap_cli, lop, oap->oap_cmd, 1);
-}
-
 /* this must be called holding the loi list lock to give coverage to exit_cache,
  * async_flag maintenance, and oap_request */
-void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
-		       struct obdo *oa, struct osc_async_page *oap,
-		       int sent, int rc)
+static void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
+			      struct osc_async_page *oap, int sent, int rc)
 {
 	struct osc_object *osc = oap->oap_obj;
 	struct lov_oinfo  *loi = osc->oo_oinfo;
@@ -588,38 +1728,205 @@ void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
 		oap->oap_request = NULL;
 	}
 
-	cfs_spin_lock(&oap->oap_lock);
-	oap->oap_async_flags = 0;
-	cfs_spin_unlock(&oap->oap_lock);
-	oap->oap_interrupted = 0;
+	/* As the transfer for this page is being done, clear the flags */
+	cfs_spin_lock(&oap->oap_lock);
+	oap->oap_async_flags = 0;
+	cfs_spin_unlock(&oap->oap_lock);
+	oap->oap_interrupted = 0;
+
+	if (oap->oap_cmd & OBD_BRW_WRITE && xid > 0) {
+		client_obd_list_lock(&cli->cl_loi_list_lock);
+		osc_process_ar(&cli->cl_ar, xid, rc);
+		osc_process_ar(&loi->loi_ar, xid, rc);
+		client_obd_list_unlock(&cli->cl_loi_list_lock);
+	}
+
+	rc = osc_completion(env, oap, oap->oap_cmd, rc);
+	if (rc)
+		CERROR("completion on oap %p obj %p returns %d.\n",
+		       oap, osc, rc);
+
+	EXIT;
+}
+
+/**
+ * Try to add extent to one RPC. We need to think about the following things:
+ * - # of pages must not be over max_pages_per_rpc
+ * - extent must be compatible with previous ones
+ */
+static int try_to_add_extent_for_io(struct client_obd *cli,
+				    struct osc_extent *ext, cfs_list_t *rpclist,
+				    int *pc, unsigned int *max_pages)
+{
+	struct osc_extent *tmp;
+	ENTRY;
+
+	EASSERT((ext->oe_state == OES_CACHE || ext->oe_state == OES_LOCK_DONE),
+		ext);
+
+	*max_pages = max(ext->oe_mppr, *max_pages);
+	if (*pc + ext->oe_nr_pages > *max_pages)
+		RETURN(0);
+
+	cfs_list_for_each_entry(tmp, rpclist, oe_link) {
+		EASSERT(tmp->oe_owner == cfs_current(), tmp);
+#if 0
+		if (overlapped(tmp, ext)) {
+			OSC_EXTENT_DUMP(D_ERROR, tmp, "overlapped %p.\n", ext);
+			EASSERT(0, ext);
+		}
+#endif
+
+		if (tmp->oe_srvlock != ext->oe_srvlock ||
+		    !tmp->oe_grants != !ext->oe_grants)
+			RETURN(0);
+
+		/* remove break for strict check */
+		break;
+	}
+
+	*pc += ext->oe_nr_pages;
+	cfs_list_move_tail(&ext->oe_link, rpclist);
+	ext->oe_owner = cfs_current();
+	RETURN(1);
+}
+
+/**
+ * In order to prevent multiple ptlrpcd from breaking contiguous extents,
+ * get_write_extent() takes all appropriate extents in atomic.
+ *
+ * The following policy is used to collect extents for IO:
+ * 1. Add as many HP extents as possible;
+ * 2. Add the first urgent extent in urgent extent list and take it out of
+ *    urgent list;
+ * 3. Add subsequent extents of this urgent extent;
+ * 4. If urgent list is not empty, goto 2;
+ * 5. Traverse the extent tree from the 1st extent;
+ * 6. Above steps exit if there is no space in this RPC.
+ */
+static int get_write_extents(struct osc_object *obj, cfs_list_t *rpclist)
+{
+	struct client_obd *cli = osc_cli(obj);
+	struct osc_extent *ext;
+	int page_count = 0;
+	unsigned int max_pages = cli->cl_max_pages_per_rpc;
+
+	LASSERT(osc_object_is_locked(obj));
+	while (!cfs_list_empty(&obj->oo_hp_exts)) {
+		ext = cfs_list_entry(obj->oo_hp_exts.next, struct osc_extent,
+				     oe_link);
+		LASSERT(ext->oe_state == OES_CACHE);
+		if (!try_to_add_extent_for_io(cli, ext, rpclist, &page_count,
+					      &max_pages))
+			return page_count;
+		EASSERT(ext->oe_nr_pages <= max_pages, ext);
+	}
+	if (page_count == max_pages)
+		return page_count;
+
+	while (!cfs_list_empty(&obj->oo_urgent_exts)) {
+		ext = cfs_list_entry(obj->oo_urgent_exts.next,
+				     struct osc_extent, oe_link);
+		if (!try_to_add_extent_for_io(cli, ext, rpclist, &page_count,
+					      &max_pages))
+			return page_count;
+
+		if (!ext->oe_intree)
+			continue;
+
+		while ((ext = next_extent(ext)) != NULL) {
+			if ((ext->oe_state != OES_CACHE) ||
+			    (!cfs_list_empty(&ext->oe_link) &&
+			     ext->oe_owner != NULL))
+				continue;
+
+			if (!try_to_add_extent_for_io(cli, ext, rpclist,
+						      &page_count, &max_pages))
+				return page_count;
+		}
+	}
+	if (page_count == max_pages)
+		return page_count;
+
+	ext = first_extent(obj);
+	while (ext != NULL) {
+		if ((ext->oe_state != OES_CACHE) ||
+		    /* this extent may be already in current rpclist */
+		    (!cfs_list_empty(&ext->oe_link) && ext->oe_owner != NULL)) {
+			ext = next_extent(ext);
+			continue;
+		}
+
+		if (!try_to_add_extent_for_io(cli, ext, rpclist, &page_count,
+					      &max_pages))
+			return page_count;
+
+		ext = next_extent(ext);
+	}
+	return page_count;
+}
+
+static int
+osc_send_write_rpc(const struct lu_env *env, struct client_obd *cli,
+		   struct osc_object *osc, pdl_policy_t pol)
+{
+	CFS_LIST_HEAD(rpclist);
+	struct osc_extent *ext;
+	struct osc_extent *tmp;
+	struct osc_extent *first = NULL;
+	obd_count page_count = 0;
+	int srvlock = 0;
+	int rc = 0;
+	ENTRY;
+
+	LASSERT(osc_object_is_locked(osc));
+
+	page_count = get_write_extents(osc, &rpclist);
+	LASSERT(equi(page_count == 0, cfs_list_empty(&rpclist)));
 
-	if (oap->oap_cmd & OBD_BRW_WRITE) {
-		osc_process_ar(&cli->cl_ar, xid, rc);
-		osc_process_ar(&loi->loi_ar, xid, rc);
+	if (cfs_list_empty(&rpclist))
+		RETURN(0);
+
+	osc_update_pending(osc, OBD_BRW_WRITE, -page_count);
+
+	cfs_list_for_each_entry(ext, &rpclist, oe_link) {
+		LASSERT(ext->oe_state == OES_CACHE ||
+			ext->oe_state == OES_LOCK_DONE);
+		if (ext->oe_state == OES_CACHE)
+			osc_extent_state_set(ext, OES_LOCKING);
+		else
+			osc_extent_state_set(ext, OES_RPC);
 	}
 
-	if (rc == 0 && oa != NULL) {
-		if (oa->o_valid & OBD_MD_FLBLOCKS)
-			loi->loi_lvb.lvb_blocks = oa->o_blocks;
-		if (oa->o_valid & OBD_MD_FLMTIME)
-			loi->loi_lvb.lvb_mtime = oa->o_mtime;
-		if (oa->o_valid & OBD_MD_FLATIME)
-			loi->loi_lvb.lvb_atime = oa->o_atime;
-		if (oa->o_valid & OBD_MD_FLCTIME)
-			loi->loi_lvb.lvb_ctime = oa->o_ctime;
+	/* we're going to grab page lock, so release object lock because
+	 * lock order is page lock -> object lock. */
+	osc_object_unlock(osc);
+
+	cfs_list_for_each_entry_safe(ext, tmp, &rpclist, oe_link) {
+		if (ext->oe_state == OES_LOCKING) {
+			rc = osc_extent_make_ready(env, ext);
+			if (unlikely(rc < 0)) {
+				cfs_list_del_init(&ext->oe_link);
+				osc_extent_finish(env, ext, 0, rc);
+				continue;
+			}
+		}
+		if (first == NULL) {
+			first = ext;
+			srvlock = ext->oe_srvlock;
+		} else {
+			LASSERT(srvlock == ext->oe_srvlock);
+		}
 	}
 
-	rc = osc_completion(env, oap, oap->oap_cmd, oa, rc);
+	if (!cfs_list_empty(&rpclist)) {
+		LASSERT(page_count > 0);
+		rc = osc_build_rpc(env, cli, &rpclist, OBD_BRW_WRITE, pol);
+		LASSERT(cfs_list_empty(&rpclist));
+	}
 
-	/* cl_page_completion() drops PG_locked. so, a new I/O on the page could
-	 * start, but OSC calls it under lock and thus we can add oap back to
-	 * pending safely */
-	if (rc)
-		/* upper layer wants to leave the page on pending queue */
-		osc_oap_to_pending(oap);
-	else
-		osc_exit_cache(cli, oap, sent);
-	EXIT;
+	osc_object_lock(osc);
+	RETURN(rc);
 }
 
 /**
@@ -633,193 +1940,48 @@ void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
  * \return negative on errors.
  */
 static int
-osc_send_oap_rpc(const struct lu_env *env, struct client_obd *cli,
-		 struct osc_object *osc, int cmd,
-		 struct osc_oap_pages *lop, pdl_policy_t pol)
+osc_send_read_rpc(const struct lu_env *env, struct client_obd *cli,
+		  struct osc_object *osc, pdl_policy_t pol)
 {
-	obd_count page_count = 0;
-	struct osc_async_page *oap = NULL, *tmp;
-	CFS_LIST_HEAD(rpc_list);
-	int srvlock = 0, mem_tight = 0;
-	obd_off starting_offset = OBD_OBJECT_EOF;
-	unsigned int ending_offset;
-	int starting_page_off = 0;
-	int rc;
+	struct osc_extent *ext;
+	struct osc_extent *next;
+	CFS_LIST_HEAD(rpclist);
+	int page_count = 0;
+	unsigned int max_pages = cli->cl_max_pages_per_rpc;
+	int rc = 0;
 	ENTRY;
 
-	/* ASYNC_HP pages first. At present, when the lock the pages is
-	 * to be canceled, the pages covered by the lock will be sent out
-	 * with ASYNC_HP. We have to send out them as soon as possible. */
-	cfs_list_for_each_entry_safe(oap, tmp, &lop->oop_urgent, oap_urgent_item) {
-		if (oap->oap_async_flags & ASYNC_HP)
-			cfs_list_move(&oap->oap_pending_item, &rpc_list);
-		else if (!(oap->oap_brw_flags & OBD_BRW_SYNC))
-			/* only do this for writeback pages. */
-			cfs_list_move_tail(&oap->oap_pending_item, &rpc_list);
-		if (++page_count >= cli->cl_max_pages_per_rpc)
-			break;
-	}
-	cfs_list_splice_init(&rpc_list, &lop->oop_pending);
-	page_count = 0;
-
-	/* first we find the pages we're allowed to work with */
-	cfs_list_for_each_entry_safe(oap, tmp, &lop->oop_pending,
-				     oap_pending_item) {
-		LASSERTF(oap->oap_magic == OAP_MAGIC, "Bad oap magic: oap %p, "
-			 "magic 0x%x\n", oap, oap->oap_magic);
-
-		if (page_count != 0 &&
-		    srvlock != !!(oap->oap_brw_flags & OBD_BRW_SRVLOCK)) {
-			CDEBUG(D_PAGE, "SRVLOCK flag mismatch,"
-			       " oap %p, page %p, srvlock %u\n",
-			       oap, oap->oap_brw_page.pg, (unsigned)!srvlock);
-			break;
-		}
-
-		/* If there is a gap at the start of this page, it can't merge
-		 * with any previous page, so we'll hand the network a
-		 * "fragmented" page array that it can't transfer in 1 RDMA */
-		if (oap->oap_obj_off < starting_offset) {
-			if (starting_page_off != 0)
-				break;
-
-			starting_page_off = oap->oap_page_off;
-			starting_offset = oap->oap_obj_off + starting_page_off;
-		} else if (oap->oap_page_off != 0)
-			break;
-
-		/* in llite being 'ready' equates to the page being locked
-		 * until completion unlocks it.  commit_write submits a page
-		 * as not ready because its unlock will happen unconditionally
-		 * as the call returns.  if we race with commit_write giving
-		 * us that page we don't want to create a hole in the page
-		 * stream, so we stop and leave the rpc to be fired by
-		 * another dirtier or kupdated interval (the not ready page
-		 * will still be on the dirty list).  we could call in
-		 * at the end of ll_file_write to process the queue again. */
-		if (!(oap->oap_async_flags & ASYNC_READY)) {
-			int rc = osc_make_ready(env, oap, cmd);
-			if (rc < 0)
-				CDEBUG(D_INODE, "oap %p page %p returned %d "
-						"instead of ready\n", oap,
-						oap->oap_page, rc);
-			switch (rc) {
-			case -EAGAIN:
-				/* llite is telling us that the page is still
-				 * in commit_write and that we should try
-				 * and put it in an rpc again later.  we
-				 * break out of the loop so we don't create
-				 * a hole in the sequence of pages in the rpc
-				 * stream.*/
-				oap = NULL;
-				break;
-			case -EINTR:
-				/* the io isn't needed.. tell the checks
-				 * below to complete the rpc with EINTR */
-				cfs_spin_lock(&oap->oap_lock);
-				oap->oap_async_flags |= ASYNC_COUNT_STABLE;
-				cfs_spin_unlock(&oap->oap_lock);
-				oap->oap_count = -EINTR;
-				break;
-			case 0:
-				cfs_spin_lock(&oap->oap_lock);
-				oap->oap_async_flags |= ASYNC_READY;
-				cfs_spin_unlock(&oap->oap_lock);
-				break;
-			default:
-				LASSERTF(0, "oap %p page %p returned %d "
-					    "from make_ready\n", oap,
-					    oap->oap_page, rc);
-				break;
-			}
-		}
-		if (oap == NULL)
-			break;
-
-		/* take the page out of our book-keeping */
-		cfs_list_del_init(&oap->oap_pending_item);
-		lop_update_pending(cli, lop, cmd, -1);
-		cfs_list_del_init(&oap->oap_urgent_item);
-
-		/* ask the caller for the size of the io as the rpc leaves. */
-		if (!(oap->oap_async_flags & ASYNC_COUNT_STABLE)) {
-			oap->oap_count = osc_refresh_count(env, oap, cmd);
-			LASSERT(oap->oap_page_off + oap->oap_count <= CFS_PAGE_SIZE);
-		}
-		if (oap->oap_count <= 0) {
-			CDEBUG(D_CACHE, "oap %p count %d, completing\n", oap,
-			       oap->oap_count);
-			osc_ap_completion(env, cli, NULL,
-					  oap, 0, oap->oap_count);
-			continue;
-		}
-
-		/* now put the page back in our accounting */
-		cfs_list_add_tail(&oap->oap_rpc_item, &rpc_list);
-		if (page_count++ == 0)
-			srvlock = !!(oap->oap_brw_flags & OBD_BRW_SRVLOCK);
-
-		if (oap->oap_brw_flags & OBD_BRW_MEMALLOC)
-			mem_tight = 1;
-
-		/* End on a PTLRPC_MAX_BRW_SIZE boundary.  We want full-sized
-		 * RPCs aligned on PTLRPC_MAX_BRW_SIZE boundaries to help reads
-		 * have the same alignment as the initial writes that allocated
-		 * extents on the server. */
-		ending_offset = oap->oap_obj_off + oap->oap_page_off +
-				oap->oap_count;
-		if (!(ending_offset & (PTLRPC_MAX_BRW_SIZE - 1)))
-			break;
-
-		if (page_count >= cli->cl_max_pages_per_rpc)
-			break;
-
-		/* If there is a gap at the end of this page, it can't merge
-		 * with any subsequent pages, so we'll hand the network a
-		 * "fragmented" page array that it can't transfer in 1 RDMA */
-		if (oap->oap_page_off + oap->oap_count < CFS_PAGE_SIZE)
+	LASSERT(osc_object_is_locked(osc));
+	cfs_list_for_each_entry_safe(ext, next,
+				     &osc->oo_reading_exts, oe_link) {
+		EASSERT(ext->oe_state == OES_LOCK_DONE, ext);
+		if (!try_to_add_extent_for_io(cli, ext, &rpclist, &page_count,
+					      &max_pages))
 			break;
+		osc_extent_state_set(ext, OES_RPC);
+		EASSERT(ext->oe_nr_pages <= max_pages, ext);
 	}
+	LASSERT(page_count <= max_pages);
 
-	osc_list_maint(cli, osc);
-
-	client_obd_list_unlock(&cli->cl_loi_list_lock);
+	osc_update_pending(osc, OBD_BRW_READ, -page_count);
 
-	if (page_count == 0) {
-		client_obd_list_lock(&cli->cl_loi_list_lock);
-		RETURN(0);
-	}
+	if (!cfs_list_empty(&rpclist)) {
+		osc_object_unlock(osc);
 
-	if (mem_tight)
-		cmd |= OBD_BRW_MEMALLOC;
-	rc = osc_build_rpc(env, cli, &rpc_list, page_count, cmd, pol);
-	if (rc != 0) {
-		LASSERT(cfs_list_empty(&rpc_list));
-		osc_list_maint(cli, osc);
-		RETURN(rc);
-	}
+		LASSERT(page_count > 0);
+		rc = osc_build_rpc(env, cli, &rpclist, OBD_BRW_READ, pol);
+		LASSERT(cfs_list_empty(&rpclist));
 
-	starting_offset &= PTLRPC_MAX_BRW_SIZE - 1;
-	if (cmd == OBD_BRW_READ) {
-		cli->cl_r_in_flight++;
-		lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
-		lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
-		lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
-				      (starting_offset >> CFS_PAGE_SHIFT) + 1);
-	} else {
-		cli->cl_w_in_flight++;
-		lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
-		lprocfs_oh_tally(&cli->cl_write_rpc_hist,
-				 cli->cl_w_in_flight);
-		lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
-				      (starting_offset >> CFS_PAGE_SHIFT) + 1);
+		osc_object_lock(osc);
 	}
-
-	RETURN(1);
+	RETURN(rc);
 }
 
-#define list_to_obj(list, item) \
-	cfs_list_entry((list)->next, struct osc_object, oo_##item)
+#define list_to_obj(list, item) ({					      \
+	cfs_list_t *__tmp = (list)->next;				      \
+	cfs_list_del_init(__tmp);					      \
+	cfs_list_entry(__tmp, struct osc_object, oo_##item);		      \
+})
 
 /* This is called by osc_check_rpcs() to find which objects have pages that
  * we could be sending.  These lists are maintained by osc_makes_rpc(). */
@@ -861,14 +2023,23 @@ static void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli,
 			   pdl_policy_t pol)
 {
 	struct osc_object *osc;
-	int rc = 0, race_counter = 0;
+	int rc = 0;
 	ENTRY;
 
 	while ((osc = osc_next_obj(cli)) != NULL) {
+		struct cl_object *obj = osc2cl(osc);
+		struct lu_ref_link *link;
+
 		OSC_IO_DEBUG(osc, "%lu in flight\n", rpcs_in_flight(cli));
 
-		if (osc_max_rpc_in_flight(cli, osc))
+		if (osc_max_rpc_in_flight(cli, osc)) {
+			__osc_list_maint(cli, osc);
 			break;
+		}
+
+		cl_object_get(obj);
+		client_obd_list_unlock(&cli->cl_loi_list_lock);
+		link = lu_object_ref_add(&obj->co_lu, "check", cfs_current());
 
 		/* attempt some read/write balancing by alternating between
 		 * reads and writes in an object.  The makes_rpc checks here
@@ -876,13 +2047,13 @@ static void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli,
 		 * instead of objects.  we don't want send_oap_rpc to drain a
 		 * partial read pending queue when we're given this object to
 		 * do io on writes while there are cache waiters */
+		osc_object_lock(osc);
 		if (osc_makes_rpc(cli, osc, OBD_BRW_WRITE)) {
-			rc = osc_send_oap_rpc(env, cli, osc, OBD_BRW_WRITE,
-					      &osc->oo_write_pages, pol);
+			rc = osc_send_write_rpc(env, cli, osc, pol);
 			if (rc < 0) {
 				CERROR("Write request failed with %d\n", rc);
 
-				/* osc_send_oap_rpc failed, mostly because of
+				/* osc_send_write_rpc failed, mostly because of
 				 * memory pressure.
 				 *
 				 * It can't break here, because if:
@@ -900,53 +2071,56 @@ static void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli,
 				 * Anyway, continue to drain pages. */
 				/* break; */
 			}
-
-			if (rc > 0)
-				race_counter = 0;
-			else if (rc == 0)
-				race_counter++;
 		}
 		if (osc_makes_rpc(cli, osc, OBD_BRW_READ)) {
-			rc = osc_send_oap_rpc(env, cli, osc, OBD_BRW_READ,
-					      &osc->oo_read_pages, pol);
+			rc = osc_send_read_rpc(env, cli, osc, pol);
 			if (rc < 0)
 				CERROR("Read request failed with %d\n", rc);
-
-			if (rc > 0)
-				race_counter = 0;
-			else if (rc == 0)
-				race_counter++;
 		}
-
-		/* attempt some inter-object balancing by issuing rpcs
-		 * for each object in turn */
-		if (!cfs_list_empty(&osc->oo_hp_ready_item))
-			cfs_list_del_init(&osc->oo_hp_ready_item);
-		if (!cfs_list_empty(&osc->oo_ready_item))
-			cfs_list_del_init(&osc->oo_ready_item);
-		if (!cfs_list_empty(&osc->oo_write_item))
-			cfs_list_del_init(&osc->oo_write_item);
-		if (!cfs_list_empty(&osc->oo_read_item))
-			cfs_list_del_init(&osc->oo_read_item);
+		osc_object_unlock(osc);
 
 		osc_list_maint(cli, osc);
+		lu_object_ref_del_at(&obj->co_lu, link, "check", cfs_current());
+		cl_object_put(env, obj);
 
-		/* send_oap_rpc fails with 0 when make_ready tells it to
-		 * back off.  llite's make_ready does this when it tries
-		 * to lock a page queued for write that is already locked.
-		 * we want to try sending rpcs from many objects, but we
-		 * don't want to spin failing with 0.  */
-		if (race_counter == 10)
-			break;
+		client_obd_list_lock(&cli->cl_loi_list_lock);
+	}
+}
+
+static int osc_io_unplug0(const struct lu_env *env, struct client_obd *cli,
+			  struct osc_object *osc, pdl_policy_t pol, int async)
+{
+	int has_rpcs = 1;
+	int rc = 0;
+
+	client_obd_list_lock(&cli->cl_loi_list_lock);
+	if (osc != NULL)
+		has_rpcs = __osc_list_maint(cli, osc);
+	if (has_rpcs) {
+		if (!async) {
+			osc_check_rpcs(env, cli, pol);
+		} else {
+			CDEBUG(D_CACHE, "Queue writeback work for client %p.\n",
+			       cli);
+			LASSERT(cli->cl_writeback_work != NULL);
+			rc = ptlrpcd_queue_work(cli->cl_writeback_work);
+		}
 	}
+	client_obd_list_unlock(&cli->cl_loi_list_lock);
+	return rc;
+}
+
+static int osc_io_unplug_async(const struct lu_env *env,
+				struct client_obd *cli, struct osc_object *osc)
+{
+	/* XXX: policy is no use actually. */
+	return osc_io_unplug0(env, cli, osc, PDL_POLICY_ROUND, 1);
 }
 
 void osc_io_unplug(const struct lu_env *env, struct client_obd *cli,
 		   struct osc_object *osc, pdl_policy_t pol)
 {
-	if (osc)
-		osc_list_maint(cli, osc);
-	osc_check_rpcs(env, cli, pol);
+	(void)osc_io_unplug0(env, cli, osc, pol, 0);
 }
 
 int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
@@ -971,24 +2145,28 @@ int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
 		oap->oap_brw_flags = OBD_BRW_NOQUOTA;
 
 	CFS_INIT_LIST_HEAD(&oap->oap_pending_item);
-	CFS_INIT_LIST_HEAD(&oap->oap_urgent_item);
 	CFS_INIT_LIST_HEAD(&oap->oap_rpc_item);
 
 	cfs_spin_lock_init(&oap->oap_lock);
-	CDEBUG(D_CACHE, "oap %p page %p obj off "LPU64"\n",
+	CDEBUG(D_INFO, "oap %p page %p obj off "LPU64"\n",
 	       oap, page, oap->oap_obj_off);
 	RETURN(0);
 }
 
-int osc_queue_async_io(const struct lu_env *env, struct osc_page *ops)
+int osc_queue_async_io(const struct lu_env *env, struct cl_io *io,
+		       struct osc_page *ops)
 {
+	struct osc_io *oio = osc_env_io(env);
+	struct osc_extent     *ext = NULL;
 	struct osc_async_page *oap = &ops->ops_oap;
 	struct client_obd     *cli = oap->oap_cli;
 	struct osc_object     *osc = oap->oap_obj;
-	struct obd_export     *exp = osc_export(osc);
-	int brw_flags = OBD_BRW_ASYNC;
-	int cmd = OBD_BRW_WRITE;
-	int rc = 0;
+	pgoff_t index;
+	int    grants = 0;
+	int    brw_flags = OBD_BRW_ASYNC;
+	int    cmd = OBD_BRW_WRITE;
+	int    need_release = 0;
+	int    rc = 0;
 	ENTRY;
 
 	if (oap->oap_magic != OAP_MAGIC)
@@ -998,13 +2176,13 @@ int osc_queue_async_io(const struct lu_env *env, struct osc_page *ops)
 		RETURN(-EIO);
 
 	if (!cfs_list_empty(&oap->oap_pending_item) ||
-	    !cfs_list_empty(&oap->oap_urgent_item) ||
 	    !cfs_list_empty(&oap->oap_rpc_item))
 		RETURN(-EBUSY);
 
 	/* Set the OBD_BRW_SRVLOCK before the page is queued. */
 	brw_flags |= ops->ops_srvlock ? OBD_BRW_SRVLOCK : 0;
-	if (!client_is_remote(exp) && cfs_capable(CFS_CAP_SYS_RESOURCE)) {
+	if (!client_is_remote(osc_export(osc)) &&
+	    cfs_capable(CFS_CAP_SYS_RESOURCE)) {
 		brw_flags |= OBD_BRW_NOQUOTA;
 		cmd |= OBD_BRW_NOQUOTA;
 	}
@@ -1024,132 +2202,228 @@ int osc_queue_async_io(const struct lu_env *env, struct osc_page *ops)
 
 		qid[USRQUOTA] = attr->cat_uid;
 		qid[GRPQUOTA] = attr->cat_gid;
-		if (rc == 0 &&
-		    osc_quota_chkdq(cli, qid) == NO_QUOTA)
+		if (rc == 0 && osc_quota_chkdq(cli, qid) == NO_QUOTA)
 			rc = -EDQUOT;
 		if (rc)
 			RETURN(rc);
 	}
 
-	client_obd_list_lock(&cli->cl_loi_list_lock);
-
 	oap->oap_cmd = cmd;
 	oap->oap_page_off = ops->ops_from;
 	oap->oap_count = ops->ops_to - ops->ops_from;
 	oap->oap_async_flags = 0;
 	oap->oap_brw_flags = brw_flags;
-	/* Give a hint to OST that requests are coming from kswapd - bug19529 */
-	if (cfs_memory_pressure_get())
-		oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
 
-	rc = osc_enter_cache(env, cli, oap);
-	if (rc) {
+	OSC_IO_DEBUG(osc, "oap %p page %p added for cmd %d\n",
+		     oap, oap->oap_page, oap->oap_cmd & OBD_BRW_RWMASK);
+
+	index = oap2cl_page(oap)->cp_index;
+
+	/* Add this page into extent by the following steps:
+	 * 1. if there exists an active extent for this IO, mostly this page
+	 *    can be added to the active extent and sometimes we need to
+	 *    expand extent to accomodate this page;
+	 * 2. otherwise, a new extent will be allocated. */
+
+	ext = oio->oi_active;
+	if (ext != NULL && ext->oe_start <= index && ext->oe_max_end >= index) {
+		/* one chunk plus extent overhead must be enough to write this
+		 * page */
+		grants = (1 << cli->cl_chunkbits) + cli->cl_extent_tax;
+		if (ext->oe_end >= index)
+			grants = 0;
+
+		/* it doesn't need any grant to dirty this page */
+		client_obd_list_lock(&cli->cl_loi_list_lock);
+		rc = osc_enter_cache_try(cli, oap, grants, 0);
 		client_obd_list_unlock(&cli->cl_loi_list_lock);
-		RETURN(rc);
+		if (rc == 0) { /* try failed */
+			grants = 0;
+			need_release = 1;
+		} else if (ext->oe_end < index) {
+			int tmp = grants;
+			/* try to expand this extent */
+			rc = osc_extent_expand(ext, index, &tmp);
+			if (rc < 0) {
+				need_release = 1;
+				/* don't free reserved grant */
+			} else {
+				OSC_EXTENT_DUMP(D_CACHE, ext,
+						"expanded for %lu.\n", index);
+				osc_unreserve_grant(cli, grants, tmp);
+				grants = 0;
+			}
+		}
+		rc = 0;
+	} else if (ext != NULL) {
+		/* index is located outside of active extent */
+		need_release = 1;
+	}
+	if (need_release) {
+		osc_extent_release(env, ext);
+		oio->oi_active = NULL;
+		ext = NULL;
 	}
 
-	OSC_IO_DEBUG(osc, "oap %p page %p added for cmd %d\n",
-		     oap, oap->oap_page, cmd);
-
-	osc_oap_to_pending(oap);
-	osc_list_maint(cli, osc);
-	if (!osc_max_rpc_in_flight(cli, osc) &&
-	    osc_makes_rpc(cli, osc, OBD_BRW_WRITE)) {
-		LASSERT(cli->cl_writeback_work != NULL);
-		rc = ptlrpcd_queue_work(cli->cl_writeback_work);
+	if (ext == NULL) {
+		int tmp = (1 << cli->cl_chunkbits) + cli->cl_extent_tax;
+
+		/* try to find new extent to cover this page */
+		LASSERT(oio->oi_active == NULL);
+		/* we may have allocated grant for this page if we failed
+		 * to expand the previous active extent. */
+		LASSERT(ergo(grants > 0, grants >= tmp));
+
+		rc = 0;
+		if (grants == 0) {
+			/* we haven't allocated grant for this page. */
+			rc = osc_enter_cache(env, cli, oap, tmp);
+			if (rc == 0)
+				grants = tmp;
+		}
 
-		CDEBUG(D_CACHE, "Queued writeback work for client obd %p/%d.\n",
-		       cli, rc);
+		tmp = grants;
+		if (rc == 0) {
+			ext = osc_extent_find(env, osc, index, &tmp);
+			if (IS_ERR(ext)) {
+				LASSERT(tmp == grants);
+				osc_exit_cache(cli, oap);
+				rc = PTR_ERR(ext);
+				ext = NULL;
+			} else {
+				oio->oi_active = ext;
+			}
+		}
+		if (grants > 0)
+			osc_unreserve_grant(cli, grants, tmp);
 	}
-	client_obd_list_unlock(&cli->cl_loi_list_lock);
 
-	RETURN(0);
+	LASSERT(ergo(rc == 0, ext != NULL));
+	if (ext != NULL) {
+		EASSERTF(ext->oe_end >= index && ext->oe_start <= index,
+			 ext, "index = %lu.\n", index);
+		LASSERT((oap->oap_brw_flags & OBD_BRW_FROM_GRANT) != 0);
+
+		osc_object_lock(osc);
+		if (ext->oe_nr_pages == 0)
+			ext->oe_srvlock = ops->ops_srvlock;
+		else
+			LASSERT(ext->oe_srvlock == ops->ops_srvlock);
+		++ext->oe_nr_pages;
+		cfs_list_add_tail(&oap->oap_pending_item, &ext->oe_pages);
+		osc_object_unlock(osc);
+	}
+	RETURN(rc);
 }
 
-int osc_teardown_async_page(struct osc_object *obj, struct osc_page *ops)
+int osc_teardown_async_page(const struct lu_env *env,
+			    struct osc_object *obj, struct osc_page *ops)
 {
 	struct osc_async_page *oap = &ops->ops_oap;
-	struct client_obd     *cli = oap->oap_cli;
-	struct osc_oap_pages  *lop;
+	struct osc_extent     *ext = NULL;
 	int rc = 0;
 	ENTRY;
 
-	if (oap->oap_magic != OAP_MAGIC)
-		RETURN(-EINVAL);
-
-	if (oap->oap_cmd & OBD_BRW_WRITE) {
-		lop = &obj->oo_write_pages;
-	} else {
-		lop = &obj->oo_read_pages;
-	}
-
-	client_obd_list_lock(&cli->cl_loi_list_lock);
-
-	if (!cfs_list_empty(&oap->oap_rpc_item))
-		GOTO(out, rc = -EBUSY);
-
-	osc_exit_cache(cli, oap, 0);
-	osc_wake_cache_waiters(cli);
-
-	if (!cfs_list_empty(&oap->oap_urgent_item)) {
-		cfs_list_del_init(&oap->oap_urgent_item);
-		cfs_spin_lock(&oap->oap_lock);
-		oap->oap_async_flags &= ~(ASYNC_URGENT | ASYNC_HP);
-		cfs_spin_unlock(&oap->oap_lock);
-	}
-	if (!cfs_list_empty(&oap->oap_pending_item)) {
-		cfs_list_del_init(&oap->oap_pending_item);
-		lop_update_pending(cli, lop, oap->oap_cmd, -1);
+	LASSERT(oap->oap_magic == OAP_MAGIC);
+
+	CDEBUG(D_INFO, "teardown oap %p page %p at index %lu.\n",
+	       oap, ops, oap2cl_page(oap)->cp_index);
+
+	osc_object_lock(obj);
+	if (!cfs_list_empty(&oap->oap_rpc_item)) {
+		CDEBUG(D_CACHE, "oap %p is not in cache.\n", oap);
+		rc = -EBUSY;
+	} else if (!cfs_list_empty(&oap->oap_pending_item)) {
+		ext = osc_extent_lookup(obj, oap2cl_page(oap)->cp_index);
+		/* only truncated pages are allowed to be taken out.
+		 * See osc_extent_truncate() and osc_cache_truncate_start()
+		 * for details. */
+		if (ext != NULL && ext->oe_state != OES_TRUNC) {
+			OSC_EXTENT_DUMP(D_ERROR, ext, "trunc at %lu.\n",
+					oap2cl_page(oap)->cp_index);
+			rc = -EBUSY;
+		}
 	}
-	osc_list_maint(cli, obj);
-	OSC_IO_DEBUG(obj, "oap %p page %p torn down\n", oap, oap->oap_page);
-out:
-	client_obd_list_unlock(&cli->cl_loi_list_lock);
+	osc_object_unlock(obj);
+	if (ext != NULL)
+		osc_extent_put(env, ext);
 	RETURN(rc);
 }
 
-/* aka (~was & now & flag), but this is more clear :) */
-#define SETTING(was, now, flag) (!(was & flag) && (now & flag))
-
-int osc_set_async_flags(struct osc_object *obj, struct osc_page *opg,
-			obd_flag async_flags)
+/**
+ * This is called when a page is picked up by kernel to write out.
+ *
+ * We should find out the corresponding extent and add the whole extent
+ * into urgent list. The extent may be being truncated or used, handle it
+ * carefully.
+ */
+int osc_flush_async_page(const struct lu_env *env, struct cl_io *io,
+			 struct osc_page *ops)
 {
-	struct osc_async_page *oap = &opg->ops_oap;
-	struct osc_oap_pages *lop;
-	int flags = 0;
+	struct osc_extent *ext   = NULL;
+	struct osc_object *obj   = cl2osc(ops->ops_cl.cpl_obj);
+	struct cl_page    *cp    = ops->ops_cl.cpl_page;
+	pgoff_t            index = cp->cp_index;
+	struct osc_async_page *oap = &ops->ops_oap;
+	int unplug = 0;
+	int rc = 0;
 	ENTRY;
 
-	LASSERT(!cfs_list_empty(&oap->oap_pending_item));
-
-	if (oap->oap_cmd & OBD_BRW_WRITE) {
-		lop = &obj->oo_write_pages;
-	} else {
-		lop = &obj->oo_read_pages;
+	osc_object_lock(obj);
+	ext = osc_extent_lookup(obj, index);
+	if (ext == NULL) {
+		osc_extent_tree_dump(D_ERROR, obj);
+		LASSERTF(0, "page index %lu is NOT covered.\n", index);
 	}
 
-	if ((oap->oap_async_flags & async_flags) == async_flags)
-		RETURN(0);
+	switch (ext->oe_state) {
+	case OES_RPC:
+	case OES_LOCK_DONE:
+		CL_PAGE_DEBUG(D_ERROR, env, cl_page_top(cp),
+			      "flush an in-rpc page?\n");
+		LASSERT(0);
+		break;
+	case OES_LOCKING:
+		/* If we know this extent is being written out, we should abort
+		 * so that the writer can make this page ready. Otherwise, there
+		 * exists a deadlock problem because other process can wait for
+		 * page writeback bit holding page lock; and meanwhile in
+		 * vvp_page_make_ready(), we need to grab page lock before
+		 * really sending the RPC. */
+	case OES_TRUNC:
+		/* race with truncate, page will be redirtied */
+		GOTO(out, rc = -EAGAIN);
+	default:
+		break;
+	}
 
-	if (SETTING(oap->oap_async_flags, async_flags, ASYNC_READY))
-		flags |= ASYNC_READY;
+	rc = cl_page_prep(env, io, cl_page_top(cp), CRT_WRITE);
+	if (rc)
+		GOTO(out, rc);
 
-	if (SETTING(oap->oap_async_flags, async_flags, ASYNC_URGENT) &&
-	    cfs_list_empty(&oap->oap_rpc_item)) {
-		if (oap->oap_async_flags & ASYNC_HP)
-			cfs_list_add(&oap->oap_urgent_item, &lop->oop_urgent);
-		else
-			cfs_list_add_tail(&oap->oap_urgent_item,
-					  &lop->oop_urgent);
-		flags |= ASYNC_URGENT;
-		osc_list_maint(oap->oap_cli, obj);
-	}
 	cfs_spin_lock(&oap->oap_lock);
-	oap->oap_async_flags |= flags;
+	oap->oap_async_flags |= ASYNC_READY|ASYNC_URGENT;
 	cfs_spin_unlock(&oap->oap_lock);
 
-	OSC_IO_DEBUG(obj, "oap %p page %p has flags %x\n", oap,
-		     oap->oap_page, oap->oap_async_flags);
-	RETURN(0);
+	if (cfs_memory_pressure_get())
+		ext->oe_memalloc = 1;
+
+	ext->oe_urgent = 1;
+	if (ext->oe_state == OES_CACHE && cfs_list_empty(&ext->oe_link)) {
+		OSC_EXTENT_DUMP(D_CACHE, ext,
+				"flush page %p make it urgent.\n", oap);
+		cfs_list_add_tail(&ext->oe_link, &obj->oo_urgent_exts);
+		unplug = 1;
+	}
+	rc = 0;
+	EXIT;
+
+out:
+	osc_object_unlock(obj);
+	osc_extent_put(env, ext);
+	if (unplug)
+		osc_io_unplug_async(env, osc_cli(obj), obj);
+	return rc;
 }
 
 /**
@@ -1162,75 +2436,422 @@ int osc_set_async_flags(struct osc_object *obj, struct osc_page *opg,
 int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops)
 {
 	struct osc_async_page *oap = &ops->ops_oap;
-	int rc = -EBUSY;
+	struct osc_object     *obj = oap->oap_obj;
+	struct client_obd     *cli = osc_cli(obj);
+	struct osc_extent     *ext;
+	struct osc_extent     *found = NULL;
+	cfs_list_t            *plist;
+	pgoff_t index = oap2cl_page(oap)->cp_index;
+	int     rc = -EBUSY;
+	int     cmd;
 	ENTRY;
 
 	LASSERT(!oap->oap_interrupted);
 	oap->oap_interrupted = 1;
 
-	/* ok, it's been put in an rpc. only one oap gets a request reference */
-	if (oap->oap_request != NULL) {
-		ptlrpc_mark_interrupted(oap->oap_request);
-		ptlrpcd_wake(oap->oap_request);
-		ptlrpc_req_finished(oap->oap_request);
-		oap->oap_request = NULL;
+	/* Find out the caching extent */
+	osc_object_lock(obj);
+	if (oap->oap_cmd & OBD_BRW_WRITE) {
+		plist = &obj->oo_urgent_exts;
+		cmd   = OBD_BRW_WRITE;
+	} else {
+		plist = &obj->oo_reading_exts;
+		cmd   = OBD_BRW_READ;
+	}
+	cfs_list_for_each_entry(ext, plist, oe_link) {
+		if (ext->oe_start <= index && ext->oe_end >= index) {
+			LASSERT(ext->oe_state == OES_LOCK_DONE);
+			/* For OES_LOCK_DONE state extent, it has already held
+			 * a refcount for RPC. */
+			found = osc_extent_get(ext);
+			break;
+		}
+	}
+	if (found != NULL) {
+		cfs_list_del_init(&found->oe_link);
+		osc_update_pending(obj, cmd, -found->oe_nr_pages);
+		osc_object_unlock(obj);
+
+		osc_extent_finish(env, found, 0, -EINTR);
+		osc_extent_put(env, found);
+		rc = 0;
+	} else {
+		osc_object_unlock(obj);
+		/* ok, it's been put in an rpc. only one oap gets a request
+		 * reference */
+		if (oap->oap_request != NULL) {
+			ptlrpc_mark_interrupted(oap->oap_request);
+			ptlrpcd_wake(oap->oap_request);
+			ptlrpc_req_finished(oap->oap_request);
+			oap->oap_request = NULL;
+		}
 	}
 
-	/*
-	 * page completion may be called only if ->cpo_prep() method was
-	 * executed by osc_io_submit(), that also adds page the to pending list
-	 */
-	if (!cfs_list_empty(&oap->oap_pending_item)) {
-		struct osc_oap_pages *lop;
-		struct osc_object *osc = oap->oap_obj;
+	osc_list_maint(cli, obj);
+	RETURN(rc);
+}
 
-		cfs_list_del_init(&oap->oap_pending_item);
-		cfs_list_del_init(&oap->oap_urgent_item);
+int osc_queue_sync_pages(const struct lu_env *env, struct osc_object *obj,
+			 cfs_list_t *list, int cmd, int brw_flags)
+{
+	struct client_obd     *cli = osc_cli(obj);
+	struct osc_extent     *ext;
+	struct osc_async_page *oap;
+	int     page_count = 0;
+	int     mppr       = cli->cl_max_pages_per_rpc;
+	pgoff_t start      = CL_PAGE_EOF;
+	pgoff_t end        = 0;
+	ENTRY;
 
-		lop = (oap->oap_cmd & OBD_BRW_WRITE) ?
-			&osc->oo_write_pages : &osc->oo_read_pages;
-		lop_update_pending(oap->oap_cli, lop, oap->oap_cmd, -1);
-		osc_list_maint(oap->oap_cli, osc);
-		rc = osc_completion(env, oap, oap->oap_cmd, NULL, -EINTR);
+	cfs_list_for_each_entry(oap, list, oap_pending_item) {
+		struct cl_page *cp = oap2cl_page(oap);
+		if (cp->cp_index > end)
+			end = cp->cp_index;
+		if (cp->cp_index < start)
+			start = cp->cp_index;
+		++page_count;
+		mppr <<= (page_count > mppr);
 	}
 
-	RETURN(rc);
+	ext = osc_extent_alloc(obj);
+	if (ext == NULL) {
+		cfs_list_for_each_entry(oap, list, oap_pending_item) {
+			cfs_list_del_init(&oap->oap_pending_item);
+			osc_ap_completion(env, cli, oap, 0, -ENOMEM);
+		}
+		RETURN(-ENOMEM);
+	}
+
+	ext->oe_rw = !!(cmd & OBD_BRW_READ);
+	ext->oe_urgent = 1;
+	ext->oe_start = start;
+	ext->oe_end = ext->oe_max_end = end;
+	ext->oe_obj = obj;
+	ext->oe_srvlock = !!(brw_flags & OBD_BRW_SRVLOCK);
+	ext->oe_nr_pages = page_count;
+	ext->oe_mppr = mppr;
+	cfs_list_splice_init(list, &ext->oe_pages);
+
+	osc_object_lock(obj);
+	/* Reuse the initial refcount for RPC, don't drop it */
+	osc_extent_state_set(ext, OES_LOCK_DONE);
+	if (cmd & OBD_BRW_WRITE) {
+		cfs_list_add_tail(&ext->oe_link, &obj->oo_urgent_exts);
+		osc_update_pending(obj, OBD_BRW_WRITE, page_count);
+	} else {
+		cfs_list_add_tail(&ext->oe_link, &obj->oo_reading_exts);
+		osc_update_pending(obj, OBD_BRW_READ, page_count);
+	}
+	osc_object_unlock(obj);
+
+	osc_io_unplug(env, cli, obj, PDL_POLICY_ROUND);
+	RETURN(0);
 }
 
-int osc_queue_sync_page(const struct lu_env *env, struct osc_page *opg,
-			int cmd, int brw_flags)
+/**
+ * Called by osc_io_setattr_start() to freeze and destroy covering extents.
+ */
+int osc_cache_truncate_start(const struct lu_env *env, struct osc_io *oio,
+			     struct osc_object *obj, __u64 size)
 {
-	struct osc_async_page *oap = &opg->ops_oap;
-	struct client_obd     *cli = oap->oap_cli;
-	int flags = 0;
+	struct client_obd *cli = osc_cli(obj);
+	struct osc_extent *ext;
+	struct osc_extent *waiting = NULL;
+	pgoff_t index;
+	CFS_LIST_HEAD(list);
+	int result = 0;
 	ENTRY;
 
-	oap->oap_cmd       = cmd;
-	oap->oap_page_off  = opg->ops_from;
-	oap->oap_count     = opg->ops_to - opg->ops_from;
-	oap->oap_brw_flags = OBD_BRW_SYNC | brw_flags;
+	/* pages with index greater or equal to index will be truncated. */
+	index = cl_index(osc2cl(obj), size + CFS_PAGE_SIZE - 1);
+
+again:
+	osc_object_lock(obj);
+	ext = osc_extent_search(obj, index);
+	if (ext == NULL)
+		ext = first_extent(obj);
+	else if (ext->oe_end < index)
+		ext = next_extent(ext);
+	while (ext != NULL) {
+		EASSERT(ext->oe_state != OES_TRUNC, ext);
+
+		if (ext->oe_state > OES_CACHE || ext->oe_urgent) {
+			/* if ext is in urgent state, it means there must exist
+			 * a page already having been flushed by write_page().
+			 * We have to wait for this extent because we can't
+			 * truncate that page. */
+			LASSERT(!ext->oe_hp);
+			OSC_EXTENT_DUMP(D_CACHE, ext,
+					"waiting for busy extent\n");
+			waiting = osc_extent_get(ext);
+			break;
+		}
 
-	/* Give a hint to OST that requests are coming from kswapd - bug19529 */
-	if (cfs_memory_pressure_get())
-		oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
+		osc_extent_get(ext);
+		if (ext->oe_state == OES_ACTIVE) {
+			/* though we grab inode mutex for write path, but we
+			 * release it before releasing extent(in osc_io_end()),
+			 * so there is a race window that an extent is still
+			 * in OES_ACTIVE when truncate starts. */
+			LASSERT(!ext->oe_trunc_pending);
+			ext->oe_trunc_pending = 1;
+		} else {
+			EASSERT(ext->oe_state == OES_CACHE, ext);
+			osc_extent_state_set(ext, OES_TRUNC);
+			osc_update_pending(obj, OBD_BRW_WRITE,
+					   -ext->oe_nr_pages);
+		}
+		EASSERT(cfs_list_empty(&ext->oe_link), ext);
+		cfs_list_add_tail(&ext->oe_link, &list);
 
-	if (!client_is_remote(osc_export(cl2osc(opg->ops_cl.cpl_obj))) &&
-	    cfs_capable(CFS_CAP_SYS_RESOURCE)) {
-		oap->oap_brw_flags |= OBD_BRW_NOQUOTA;
-		oap->oap_cmd |= OBD_BRW_NOQUOTA;
+		ext = next_extent(ext);
 	}
+	osc_object_unlock(obj);
+
+	osc_list_maint(cli, obj);
 
-	if (oap->oap_cmd & OBD_BRW_READ)
-		flags = ASYNC_COUNT_STABLE;
-	else if (!(oap->oap_brw_page.flag & OBD_BRW_FROM_GRANT))
-		osc_enter_cache_try(env, cli, oap, 1);
+	while (!cfs_list_empty(&list)) {
+		int rc;
 
-	cfs_spin_lock(&oap->oap_lock);
-	oap->oap_async_flags |= OSC_FLAGS | flags;
-	cfs_spin_unlock(&oap->oap_lock);
+		ext = cfs_list_entry(list.next, struct osc_extent, oe_link);
+		cfs_list_del_init(&ext->oe_link);
 
-	osc_oap_to_pending(oap);
-	RETURN(0);
+		/* extent may be in OES_ACTIVE state because inode mutex
+		 * is released before osc_io_end() in file write case */
+		if (ext->oe_state != OES_TRUNC)
+			osc_extent_wait(env, ext, OES_TRUNC);
+
+		rc = osc_extent_truncate(ext, index);
+		if (rc < 0) {
+			if (result == 0)
+				result = rc;
+
+			OSC_EXTENT_DUMP(D_ERROR, ext,
+					"truncate error %d\n", rc);
+		} else if (ext->oe_nr_pages == 0) {
+			osc_extent_remove(ext);
+		} else {
+			/* this must be an overlapped extent which means only
+			 * part of pages in this extent have been truncated.
+			 */
+			EASSERTF(ext->oe_start < index, ext,
+				 "trunc index = %lu.\n", index);
+			/* fix index to skip this partially truncated extent */
+			index = ext->oe_end + 1;
+
+			/* we need to hold this extent in OES_TRUNC state so
+			 * that no writeback will happen. This is to avoid
+			 * BUG 17397. */
+			LASSERT(oio->oi_trunc == NULL);
+			oio->oi_trunc = osc_extent_get(ext);
+			OSC_EXTENT_DUMP(D_CACHE, ext,
+					"trunc at "LPU64"\n", size);
+		}
+		osc_extent_put(env, ext);
+	}
+	if (waiting != NULL) {
+		if (result == 0)
+			result = osc_extent_wait(env, waiting, OES_INV);
+
+		osc_extent_put(env, waiting);
+		waiting = NULL;
+		if (result == 0)
+			goto again;
+	}
+	RETURN(result);
+}
+
+/**
+ * Called after osc_io_setattr_end to add oio->oi_trunc back to cache.
+ */
+void osc_cache_truncate_end(const struct lu_env *env, struct osc_io *oio,
+			    struct osc_object *obj)
+{
+	struct osc_extent *ext = oio->oi_trunc;
+
+	oio->oi_trunc = NULL;
+	if (ext != NULL) {
+		EASSERT(ext->oe_nr_pages > 0, ext);
+		EASSERT(ext->oe_state == OES_TRUNC, ext);
+		EASSERT(!ext->oe_urgent, ext);
+
+		OSC_EXTENT_DUMP(D_CACHE, ext, "trunc -> cache.\n");
+		osc_object_lock(obj);
+		osc_extent_state_set(ext, OES_CACHE);
+		osc_update_pending(obj, OBD_BRW_WRITE, ext->oe_nr_pages);
+		osc_object_unlock(obj);
+		osc_extent_put(env, ext);
+
+		osc_list_maint(osc_cli(obj), obj);
+	}
+}
+
+/**
+ * Wait for extents in a specific range to be written out.
+ * The caller must have called osc_cache_writeback_range() to issue IO
+ * otherwise it will take a long time for this function to finish.
+ *
+ * Caller must hold inode_mutex and i_alloc_sem, or cancel exclusive
+ * dlm lock so that nobody else can dirty this range of file while we're
+ * waiting for extents to be written.
+ */
+int osc_cache_wait_range(const struct lu_env *env, struct osc_object *obj,
+			 pgoff_t start, pgoff_t end)
+{
+	struct osc_extent *ext;
+	pgoff_t index = start;
+	int     result = 0;
+	ENTRY;
+
+again:
+	osc_object_lock(obj);
+	ext = osc_extent_search(obj, index);
+	if (ext == NULL)
+		ext = first_extent(obj);
+	else if (ext->oe_end < index)
+		ext = next_extent(ext);
+	while (ext != NULL) {
+		int rc;
+
+		if (ext->oe_start > end)
+			break;
+
+		if (!ext->oe_fsync_wait) {
+			ext = next_extent(ext);
+			continue;
+		}
+
+		EASSERT(ergo(ext->oe_state == OES_CACHE,
+			     ext->oe_hp || ext->oe_urgent), ext);
+		EASSERT(ergo(ext->oe_state == OES_ACTIVE,
+			     !ext->oe_hp && ext->oe_urgent), ext);
+
+		index = ext->oe_end + 1;
+		osc_extent_get(ext);
+		osc_object_unlock(obj);
+
+		rc = osc_extent_wait(env, ext, OES_INV);
+		if (result == 0)
+			result = rc;
+		osc_extent_put(env, ext);
+		goto again;
+	}
+	osc_object_unlock(obj);
+
+	OSC_IO_DEBUG(obj, "sync file range.\n");
+	RETURN(result);
+}
+
+/**
+ * Called to write out a range of osc object.
+ *
+ * @hp     : should be set this is caused by lock cancel;
+ * @discard: is set if dirty pages should be dropped - file will be deleted or
+ *	   truncated, this implies there is no partially discarding extents.
+ *
+ * Return how many pages will be issued, or error code if error occurred.
+ */
+int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj,
+			      pgoff_t start, pgoff_t end, int hp, int discard)
+{
+	struct osc_extent *ext;
+	CFS_LIST_HEAD(discard_list);
+	int unplug = 0;
+	int result = 0;
+	ENTRY;
+
+	osc_object_lock(obj);
+	ext = osc_extent_search(obj, start);
+	if (ext == NULL)
+		ext = first_extent(obj);
+	else if (ext->oe_end < start)
+		ext = next_extent(ext);
+	while (ext != NULL) {
+		if (ext->oe_start > end)
+			break;
+
+		ext->oe_fsync_wait = 1;
+		switch (ext->oe_state) {
+		case OES_CACHE:
+			result += ext->oe_nr_pages;
+			if (!discard) {
+				cfs_list_t *list = NULL;
+				if (hp) {
+					EASSERT(!ext->oe_hp, ext);
+					ext->oe_hp = 1;
+					list = &obj->oo_hp_exts;
+				} else if (!ext->oe_urgent) {
+					ext->oe_urgent = 1;
+					list = &obj->oo_urgent_exts;
+				}
+				if (list != NULL) {
+					cfs_list_move_tail(&ext->oe_link, list);
+					unplug = 1;
+				}
+			} else {
+				/* the only discarder is lock cancelling, so
+				 * [start, end] must contain this extent */
+				EASSERT(ext->oe_start >= start &&
+					ext->oe_max_end <= end, ext);
+				osc_extent_state_set(ext, OES_LOCKING);
+				ext->oe_owner = cfs_current();
+				cfs_list_move_tail(&ext->oe_link,
+						   &discard_list);
+				osc_update_pending(obj, OBD_BRW_WRITE,
+						   -ext->oe_nr_pages);
+			}
+			break;
+		case OES_ACTIVE:
+			/* It's pretty bad to wait for ACTIVE extents, because
+			 * we don't know how long we will wait for it to be
+			 * flushed since it may be blocked at awaiting more
+			 * grants. We do this for the correctness of fsync. */
+			LASSERT(hp == 0 && discard == 0);
+			ext->oe_urgent = 1;
+		default:
+			break;
+		}
+		ext = next_extent(ext);
+	}
+	osc_object_unlock(obj);
+
+	LASSERT(ergo(!discard, cfs_list_empty(&discard_list)));
+	if (!cfs_list_empty(&discard_list)) {
+		struct osc_extent *tmp;
+		int rc;
+
+		osc_list_maint(osc_cli(obj), obj);
+		cfs_list_for_each_entry_safe(ext, tmp, &discard_list, oe_link) {
+			cfs_list_del_init(&ext->oe_link);
+			EASSERT(ext->oe_state == OES_LOCKING, ext);
+
+			/* Discard caching pages. We don't actually write this
+			 * extent out but we complete it as if we did. */
+			rc = osc_extent_make_ready(env, ext);
+			if (unlikely(rc < 0)) {
+				OSC_EXTENT_DUMP(D_ERROR, ext,
+						"make_ready returned %d\n", rc);
+				if (result >= 0)
+					result = rc;
+			}
+
+			/* finish the extent as if the pages were sent */
+			osc_extent_finish(env, ext, 0, 0);
+		}
+	}
+
+	if (unplug)
+		osc_io_unplug(env, osc_cli(obj), obj, PDL_POLICY_ROUND);
+
+	if (hp || discard) {
+		int rc;
+		rc = osc_cache_wait_range(env, obj, start, end);
+		if (result >= 0 && rc < 0)
+			result = rc;
+	}
+
+	OSC_IO_DEBUG(obj, "cache page out.\n");
+	RETURN(result);
 }
 
 /** @} osc */
diff --git a/lustre/osc/osc_cl_internal.h b/lustre/osc/osc_cl_internal.h
index 3094ca2..54ded82 100644
--- a/lustre/osc/osc_cl_internal.h
+++ b/lustre/osc/osc_cl_internal.h
@@ -40,6 +40,7 @@
  * Internal interfaces of OSC layer.
  *
  *   Author: Nikita Danilov <nikita.danilov@sun.com>
+ *   Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
  */
 
 #ifndef OSC_CL_INTERNAL_H
@@ -61,6 +62,8 @@
  *  @{
  */
 
+struct osc_extent;
+
 /**
  * State maintained by osc layer for each IO context.
  */
@@ -69,6 +72,12 @@ struct osc_io {
         struct cl_io_slice oi_cl;
         /** true if this io is lockless. */
         int                oi_lockless;
+	/** active extents, we know how many bytes is going to be written,
+	 * so having an active extent will prevent it from being fragmented */
+	struct osc_extent *oi_active;
+	/** partially truncated extent, we need to hold this extent to prevent
+	 * page writeback from happening. */
+	struct osc_extent *oi_trunc;
 
 	struct obd_info    oi_info;
 	struct obdo        oi_oa;
@@ -99,24 +108,9 @@ struct osc_thread_info {
         struct cl_attr          oti_attr;
         struct lustre_handle    oti_handle;
         struct cl_page_list     oti_plist;
+	struct cl_io	    oti_io;
 };
 
-/**
- * Manage osc_async_page
- */
-struct osc_oap_pages {
-	cfs_list_t      oop_pending;
-	cfs_list_t      oop_urgent;
-	int             oop_num_pending;
-};
-
-static inline void osc_oap_pages_init(struct osc_oap_pages *list)
-{
-	CFS_INIT_LIST_HEAD(&list->oop_pending);
-	CFS_INIT_LIST_HEAD(&list->oop_urgent);
-	list->oop_num_pending = 0;
-}
-
 struct osc_object {
         struct cl_object   oo_cl;
         struct lov_oinfo  *oo_oinfo;
@@ -144,16 +138,55 @@ struct osc_object {
         cfs_spinlock_t     oo_seatbelt;
 
 	/**
-	 * used by the osc to keep track of what objects to build into rpcs
+	 * used by the osc to keep track of what objects to build into rpcs.
+	 * Protected by client_obd->cli_loi_list_lock.
+	 */
+	cfs_list_t	   oo_ready_item;
+	cfs_list_t	   oo_hp_ready_item;
+	cfs_list_t	   oo_write_item;
+	cfs_list_t	   oo_read_item;
+
+	/**
+	 * extent is a red black tree to manage (async) dirty pages.
 	 */
-	struct osc_oap_pages oo_read_pages;
-	struct osc_oap_pages oo_write_pages;
-	cfs_list_t           oo_ready_item;
-	cfs_list_t           oo_hp_ready_item;
-	cfs_list_t           oo_write_item;
-	cfs_list_t           oo_read_item;
+	struct rb_root       oo_root;
+	/**
+	 * Manage write(dirty) extents.
+	 */
+	cfs_list_t	   oo_hp_exts; /* list of hp extents */
+	cfs_list_t	   oo_urgent_exts; /* list of writeback extents */
+	cfs_list_t	   oo_rpc_exts;
+
+	cfs_list_t	   oo_reading_exts;
+
+	cfs_atomic_t	 oo_nr_reads;
+	cfs_atomic_t	 oo_nr_writes;
+
+	/** Protect extent tree. Will be used to protect
+	 * oo_{read|write}_pages soon. */
+	cfs_spinlock_t       oo_lock;
 };
 
+static inline void osc_object_lock(struct osc_object *obj)
+{
+	cfs_spin_lock(&obj->oo_lock);
+}
+
+static inline int osc_object_trylock(struct osc_object *obj)
+{
+	return cfs_spin_trylock(&obj->oo_lock);
+}
+
+static inline void osc_object_unlock(struct osc_object *obj)
+{
+	cfs_spin_unlock(&obj->oo_lock);
+}
+
+static inline int osc_object_is_locked(struct osc_object *obj)
+{
+	return cfs_spin_is_locked(&obj->oo_lock);
+}
+
 /*
  * Lock "micro-states" for osc layer.
  */
@@ -361,6 +394,7 @@ extern cfs_mem_cache_t *osc_object_kmem;
 extern cfs_mem_cache_t *osc_thread_kmem;
 extern cfs_mem_cache_t *osc_session_kmem;
 extern cfs_mem_cache_t *osc_req_kmem;
+extern cfs_mem_cache_t *osc_extent_kmem;
 
 extern struct lu_device_type osc_device_type;
 extern struct lu_context_key osc_key;
@@ -389,22 +423,29 @@ void osc_index2policy  (ldlm_policy_data_t *policy, const struct cl_object *obj,
 int  osc_lvb_print     (const struct lu_env *env, void *cookie,
                         lu_printer_t p, const struct ost_lvb *lvb);
 
-void osc_io_submit_page(const struct lu_env *env,
-                        struct osc_io *oio, struct osc_page *opg,
-                        enum cl_req_type crt);
-void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
-		       struct obdo *oa, struct osc_async_page *oap,
-		       int sent, int rc);
+void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
+		     enum cl_req_type crt, int brw_flags);
 int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops);
 int osc_set_async_flags(struct osc_object *obj, struct osc_page *opg,
 			obd_flag async_flags);
 int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
 			cfs_page_t *page, loff_t offset);
-int osc_queue_async_io(const struct lu_env *env, struct osc_page *ops);
-int osc_teardown_async_page(struct osc_object *obj,
+int osc_queue_async_io(const struct lu_env *env, struct cl_io *io,
+		       struct osc_page *ops);
+int osc_teardown_async_page(const struct lu_env *env, struct osc_object *obj,
 			    struct osc_page *ops);
-int osc_queue_sync_page(const struct lu_env *env, struct osc_page *ops,
-			int cmd, int brw_flags);
+int osc_flush_async_page(const struct lu_env *env, struct cl_io *io,
+			 struct osc_page *ops);
+int osc_queue_sync_pages(const struct lu_env *env, struct osc_object *obj,
+			 cfs_list_t *list, int cmd, int brw_flags);
+int osc_cache_truncate_start(const struct lu_env *env, struct osc_io *oio,
+			     struct osc_object *obj, __u64 size);
+void osc_cache_truncate_end(const struct lu_env *env, struct osc_io *oio,
+			    struct osc_object *obj);
+int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj,
+			      pgoff_t start, pgoff_t end, int hp, int discard);
+int osc_cache_wait_range(const struct lu_env *env, struct osc_object *obj,
+			 pgoff_t start, pgoff_t end);
 void osc_io_unplug(const struct lu_env *env, struct client_obd *cli,
 		   struct osc_object *osc, pdl_policy_t pol);
 
@@ -459,12 +500,22 @@ static inline struct obd_export *osc_export(const struct osc_object *obj)
         return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->od_exp;
 }
 
+static inline struct client_obd *osc_cli(const struct osc_object *obj)
+{
+	return &osc_export(obj)->exp_obd->u.cli;
+}
+
 static inline struct osc_object *cl2osc(const struct cl_object *obj)
 {
         LINVRNT(osc_is_object(&obj->co_lu));
         return container_of0(obj, struct osc_object, oo_cl);
 }
 
+static inline struct cl_object *osc2cl(const struct osc_object *obj)
+{
+	return (struct cl_object *)&obj->oo_cl;
+}
+
 static inline ldlm_mode_t osc_cl_lock2ldlm(enum cl_lock_mode mode)
 {
         LASSERT(mode == CLM_READ || mode == CLM_WRITE || mode == CLM_GROUP);
@@ -493,6 +544,21 @@ static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
         return container_of0(slice, struct osc_page, ops_cl);
 }
 
+static inline struct osc_page *oap2osc(struct osc_async_page *oap)
+{
+	return container_of0(oap, struct osc_page, ops_oap);
+}
+
+static inline struct cl_page *oap2cl_page(struct osc_async_page *oap)
+{
+	return oap2osc(oap)->ops_cl.cpl_page;
+}
+
+static inline struct osc_page *oap2osc_page(struct osc_async_page *oap)
+{
+	return (struct osc_page *)container_of(oap, struct osc_page, ops_oap);
+}
+
 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
 {
         LINVRNT(osc_is_object(&slice->cls_obj->co_lu));
@@ -509,6 +575,106 @@ static inline int osc_io_srvlock(struct osc_io *oio)
         return (oio->oi_lockless && !oio->oi_cl.cis_io->ci_no_srvlock);
 }
 
+enum osc_extent_state {
+	OES_INV       = 0, /** extent is just initialized or destroyed */
+	OES_ACTIVE    = 1, /** process is using this extent */
+	OES_CACHE     = 2, /** extent is ready for IO */
+	OES_LOCKING   = 3, /** locking page to prepare IO */
+	OES_LOCK_DONE = 4, /** locking finished, ready to send */
+	OES_RPC       = 5, /** in RPC */
+	OES_TRUNC     = 6, /** being truncated */
+	OES_STATE_MAX
+};
+#define OES_STRINGS { "inv", "active", "cache", "locking", "lockdone", "rpc", \
+		      "trunc", NULL }
+
+/**
+ * osc_extent data to manage dirty pages.
+ * osc_extent has the following attributes:
+ * 1. all pages in the same must be in one RPC in write back;
+ * 2. # of pages must be less than max_pages_per_rpc - implied by 1;
+ * 3. must be covered by only 1 osc_lock;
+ * 4. exclusive. It's impossible to have overlapped osc_extent.
+ *
+ * The lifetime of an extent is from when the 1st page is dirtied to when
+ * all pages inside it are written out.
+ *
+ * LOCKING ORDER
+ * =============
+ * page lock -> client_obd_list_lock -> object lock(osc_object::oo_lock)
+ */
+struct osc_extent {
+	/** red-black tree node */
+	struct rb_node     oe_node;
+	/** osc_object of this extent */
+	struct osc_object *oe_obj;
+	/** refcount, removed from red-black tree if reaches zero. */
+	cfs_atomic_t       oe_refc;
+	/** busy if non-zero */
+	cfs_atomic_t       oe_users;
+	/** link list of osc_object's oo_{hp|urgent|locking}_exts. */
+	cfs_list_t	 oe_link;
+	/** state of this extent */
+	unsigned int       oe_state;
+	/** flags for this extent. */
+	unsigned int       oe_intree:1,
+	/** 0 is write, 1 is read */
+			   oe_rw:1,
+			   oe_srvlock:1,
+			   oe_memalloc:1,
+	/** an ACTIVE extent is going to be truncated, so when this extent
+	 * is released, it will turn into TRUNC state instead of CACHE. */
+			   oe_trunc_pending:1,
+	/** this extent should be written asap and someone may wait for the
+	 * write to finish. This bit is usually set along with urgent if
+	 * the extent was CACHE state.
+	 * fsync_wait extent can't be merged because new extent region may
+	 * exceed fsync range. */
+			   oe_fsync_wait:1,
+	/** covering lock is being canceled */
+			   oe_hp:1,
+	/** this extent should be written back asap. set if one of pages is
+	 * called by page WB daemon, or sync write or reading requests. */
+			   oe_urgent:1;
+	/** how many grants allocated for this extent.
+	 *  Grant allocated for this extent. There is no grant allocated
+	 *  for reading extents and sync write extents. */
+	unsigned int       oe_grants;
+	/** # of dirty pages in this extent */
+	unsigned int       oe_nr_pages;
+	/** list of pending oap pages. Pages in this list are NOT sorted. */
+	cfs_list_t         oe_pages;
+	/** Since an extent has to be written out in atomic, this is used to
+	 * remember the next page need to be locked to write this extent out.
+	 * Not used right now.
+	 */
+	struct osc_page   *oe_next_page;
+	/** start and end index of this extent, include start and end
+	 * themselves. Page offset here is the page index of osc_pages.
+	 * oe_start is used as keyword for red-black tree. */
+	pgoff_t            oe_start;
+	pgoff_t            oe_end;
+	/** maximum ending index of this extent, this is limited by
+	 * max_pages_per_rpc, lock extent and chunk size. */
+	pgoff_t            oe_max_end;
+	/** waitqueue - for those who want to be notified if this extent's
+	 * state has changed. */
+	cfs_waitq_t        oe_waitq;
+	/** lock covering this extent */
+	struct cl_lock    *oe_osclock;
+	/** terminator of this extent. Must be true if this extent is in IO. */
+	cfs_task_t        *oe_owner;
+	/** return value of writeback. If somebody is waiting for this extent,
+	 * this value can be known by outside world. */
+	int                oe_rc;
+	/** max pages per rpc when this extent was created */
+	unsigned int       oe_mppr;
+};
+
+int osc_extent_finish(const struct lu_env *env, struct osc_extent *ext,
+		      int sent, int rc);
+int osc_extent_release(const struct lu_env *env, struct osc_extent *ext);
+
 /** @} osc */
 
 #endif /* OSC_CL_INTERNAL_H */
diff --git a/lustre/osc/osc_dev.c b/lustre/osc/osc_dev.c
index 7f5174a..281ea61 100644
--- a/lustre/osc/osc_dev.c
+++ b/lustre/osc/osc_dev.c
@@ -53,6 +53,7 @@ cfs_mem_cache_t *osc_object_kmem;
 cfs_mem_cache_t *osc_thread_kmem;
 cfs_mem_cache_t *osc_session_kmem;
 cfs_mem_cache_t *osc_req_kmem;
+cfs_mem_cache_t *osc_extent_kmem;
 
 struct lu_kmem_descr osc_caches[] = {
         {
@@ -86,6 +87,11 @@ struct lu_kmem_descr osc_caches[] = {
                 .ckd_size  = sizeof (struct osc_req)
         },
         {
+		.ckd_cache = &osc_extent_kmem,
+		.ckd_name  = "osc_extent_kmem",
+		.ckd_size  = sizeof (struct osc_extent)
+	},
+	{
                 .ckd_cache = NULL
         }
 };
diff --git a/lustre/osc/osc_internal.h b/lustre/osc/osc_internal.h
index 3f81c7b..d3ebac1 100644
--- a/lustre/osc/osc_internal.h
+++ b/lustre/osc/osc_internal.h
@@ -57,7 +57,6 @@ struct osc_async_page {
         unsigned short          oap_interrupted:1;
 
         cfs_list_t              oap_pending_item;
-        cfs_list_t              oap_urgent_item;
         cfs_list_t              oap_rpc_item;
 
         obd_off                 oap_obj_off;
@@ -82,6 +81,7 @@ struct osc_cache_waiter {
         cfs_list_t              ocw_entry;
         cfs_waitq_t             ocw_waitq;
         struct osc_async_page  *ocw_oap;
+	int                     ocw_grant;
         int                     ocw_rc;
 };
 
@@ -142,10 +142,8 @@ int osc_sync_base(struct obd_export *exp, struct obd_info *oinfo,
 
 int osc_process_config_base(struct obd_device *obd, struct lustre_cfg *cfg);
 int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
-		  cfs_list_t *rpc_list, int page_count, int cmd,
-		  pdl_policy_t p);
+		  cfs_list_t *ext_list, int cmd, pdl_policy_t p);
 
-struct cl_page *osc_oap2cl_page(struct osc_async_page *oap);
 extern cfs_spinlock_t osc_ast_guard;
 
 int osc_cleanup(struct obd_device *obd);
diff --git a/lustre/osc/osc_io.c b/lustre/osc/osc_io.c
index 5818b8d..a94ac61 100644
--- a/lustre/osc/osc_io.c
+++ b/lustre/osc/osc_io.c
@@ -88,11 +88,6 @@ static void osc_io_fini(const struct lu_env *env, const struct cl_io_slice *io)
 {
 }
 
-struct cl_page *osc_oap2cl_page(struct osc_async_page *oap)
-{
-        return container_of(oap, struct osc_page, ops_oap)->ops_cl.cpl_page;
-}
-
 /**
  * An implementation of cl_io_operations::cio_io_submit() method for osc
  * layer. Iterates over pages in the in-queue, prepares each for io by calling
@@ -102,32 +97,43 @@ struct cl_page *osc_oap2cl_page(struct osc_async_page *oap)
  */
 static int osc_io_submit(const struct lu_env *env,
                          const struct cl_io_slice *ios,
-                         enum cl_req_type crt, struct cl_2queue *queue,
-                         enum cl_req_priority priority)
+			 enum cl_req_type crt, struct cl_2queue *queue)
 {
+	struct obd_export *exp;
         struct cl_page    *page;
         struct cl_page    *tmp;
-        struct osc_object *osc0 = NULL;
         struct client_obd *cli  = NULL;
         struct osc_object *osc  = NULL; /* to keep gcc happy */
         struct osc_page   *opg;
         struct cl_io      *io;
+	CFS_LIST_HEAD     (list);
 
-        struct cl_page_list *qin      = &queue->c2_qin;
-        struct cl_page_list *qout     = &queue->c2_qout;
-        int queued = 0;
-        int result = 0;
+	struct cl_page_list *qin      = &queue->c2_qin;
+	struct cl_page_list *qout     = &queue->c2_qout;
+	int queued = 0;
+	int result = 0;
+	int cmd;
+	int brw_flags;
+	int max_pages;
 
-        LASSERT(qin->pl_nr > 0);
+	LASSERT(qin->pl_nr > 0);
+
+	CDEBUG(D_CACHE, "%d %d\n", qin->pl_nr, crt);
+
+	osc = cl2osc(ios->cis_obj);
+	exp = osc_export(osc);
+	cli = osc_cli(osc);
+	max_pages = cli->cl_max_pages_per_rpc;
+
+	cmd = crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
+	brw_flags = osc_io_srvlock(cl2osc_io(env, ios)) ? OBD_BRW_SRVLOCK : 0;
 
-        CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, crt);
         /*
          * NOTE: here @page is a top-level page. This is done to avoid
          *       creation of sub-page-list.
          */
         cl_page_list_for_each_safe(page, tmp, qin) {
                 struct osc_async_page *oap;
-                struct obd_export     *exp;
 
                 /* Top level IO. */
                 io = page->cp_owner;
@@ -135,52 +141,18 @@ static int osc_io_submit(const struct lu_env *env,
 
                 opg = osc_cl_page_osc(page);
                 oap = &opg->ops_oap;
-                osc = cl2osc(opg->ops_cl.cpl_obj);
-                exp = osc_export(osc);
+		LASSERT(osc == oap->oap_obj);
 
-                if (priority > CRP_NORMAL) {
-                        cfs_spin_lock(&oap->oap_lock);
-                        oap->oap_async_flags |= ASYNC_HP;
-                        cfs_spin_unlock(&oap->oap_lock);
-                }
-
-                if (osc0 == NULL) { /* first iteration */
-                        cli = &exp->exp_obd->u.cli;
-                        osc0 = osc;
-                        client_obd_list_lock(&cli->cl_loi_list_lock);
-                } else /* check that all pages are against the same object
-                        * (for now) */
-                        LASSERT(osc == osc0);
-
-                if (!cfs_list_empty(&oap->oap_urgent_item) ||
-                    !cfs_list_empty(&oap->oap_rpc_item)) {
+		if (!cfs_list_empty(&oap->oap_pending_item) ||
+		    !cfs_list_empty(&oap->oap_rpc_item)) {
+			CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
+			       oap, opg);
                         result = -EBUSY;
                         break;
                 }
 
                 result = cl_page_prep(env, io, page, crt);
-                if (result == 0) {
-                        ++queued;
-                        cl_page_list_move(qout, qin, page);
-                        if (cfs_list_empty(&oap->oap_pending_item)) {
-                                osc_io_submit_page(env, cl2osc_io(env, ios),
-                                                   opg, crt);
-                        } else {
-				result = osc_set_async_flags(osc, opg,
-							     OSC_FLAGS);
-				/*
-				 * bug 18881: we can't just break out here when
-				 * error occurs after cl_page_prep has been
-				 * called against the page. The correct
-				 * way is to call page's completion routine,
-				 * as in osc_oap_interrupted.  For simplicity,
-				 * we just force osc_set_async_flags() to
-				 * not return error.
-				 */
-                                LASSERT(result == 0);
-                        }
-                        opg->ops_submit_time = cfs_time_current();
-                } else {
+		if (result != 0) {
                         LASSERT(result < 0);
                         if (result != -EALREADY)
                                 break;
@@ -190,30 +162,29 @@ static int osc_io_submit(const struct lu_env *env,
                          * is not dirty.
                          */
                         result = 0;
+			continue;
                 }
 
-                /*
-                 * We might hold client_obd_list_lock() for too long and cause
-                 * soft-lockups (see bug 16651). But on the other hand, pages
-                 * are queued here with ASYNC_URGENT flag, thus will be sent
-                 * out immediately once osc_io_unplug() be called, possibly
-                 * resulting sub-optimal RPCs.
-                 *
-                 * We think creating optimal-sized RPCs is more important than
-                 * avoiding the transient soft-lockups, plus I believe the
-                 * soft-locks only happen in full debug testing.
-                 */
-        }
-
-        LASSERT(ergo(result == 0, cli != NULL));
-        LASSERT(ergo(result == 0, osc == osc0));
-
-        if (queued > 0)
-		osc_io_unplug(env, cli, osc, PDL_POLICY_ROUND);
-        if (osc0)
-                client_obd_list_unlock(&cli->cl_loi_list_lock);
-        CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
-        return qout->pl_nr > 0 ? 0 : result;
+		cl_page_list_move(qout, qin, page);
+		oap->oap_async_flags = ASYNC_URGENT|ASYNC_READY;
+		oap->oap_async_flags |= ASYNC_COUNT_STABLE;
+
+		osc_page_submit(env, opg, crt, brw_flags);
+		cfs_list_add_tail(&oap->oap_pending_item, &list);
+		if (++queued == max_pages) {
+			queued = 0;
+			result = osc_queue_sync_pages(env, osc, &list, cmd,
+						      brw_flags);
+			if (result < 0)
+				break;
+		}
+	}
+
+	if (queued > 0)
+		result = osc_queue_sync_pages(env, osc, &list, cmd, brw_flags);
+
+	CDEBUG(D_INFO, "%d/%d %d\n", qin->pl_nr, qout->pl_nr, result);
+	return qout->pl_nr > 0 ? 0 : result;
 }
 
 static void osc_page_touch_at(const struct lu_env *env,
@@ -372,64 +343,64 @@ static int osc_async_upcall(void *a, int rc)
         return 0;
 }
 
-/* Disable osc_trunc_check() because it is naturally race between read and
- * truncate. See bug 20645 for details.
- */
-#if 0 && defined(__KERNEL__)
+#if defined(__KERNEL__)
 /**
  * Checks that there are no pages being written in the extent being truncated.
  */
+static int trunc_check_cb(const struct lu_env *env, struct cl_io *io,
+			  struct cl_page *page, void *cbdata)
+{
+	const struct cl_page_slice *slice;
+	struct osc_page *ops;
+	struct osc_async_page *oap;
+	__u64 start = *(__u64 *)cbdata;
+
+	slice = cl_page_at(page, &osc_device_type);
+	LASSERT(slice != NULL);
+	ops = cl2osc_page(slice);
+	oap = &ops->ops_oap;
+
+	if (oap->oap_cmd & OBD_BRW_WRITE &&
+	    !cfs_list_empty(&oap->oap_pending_item))
+		CL_PAGE_DEBUG(D_ERROR, env, page, "exists " LPU64 "/%s.\n",
+				start, current->comm);
+
+#ifdef __linux__
+	{
+		cfs_page_t *vmpage = cl_page_vmpage(env, page);
+		if (PageLocked(vmpage))
+			CDEBUG(D_CACHE, "page %p index %lu locked for %d.\n",
+			       ops, page->cp_index,
+			       (oap->oap_cmd & OBD_BRW_RWMASK));
+	}
+#endif
+
+	return CLP_GANG_OKAY;
+}
+
 static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
-                            struct osc_io *oio, size_t size)
+			    struct osc_io *oio, __u64 size)
 {
-        struct osc_page     *cp;
-        struct osc_object   *obj;
-        struct cl_object    *clob;
-        struct cl_page      *page;
-        struct cl_page_list *list;
-        int                  partial;
-        pgoff_t              start;
+	struct cl_object *clob;
+	int     partial;
+	pgoff_t start;
 
         clob    = oio->oi_cl.cis_obj;
-        obj     = cl2osc(clob);
         start   = cl_index(clob, size);
         partial = cl_offset(clob, start) < size;
-        list    = &osc_env_info(env)->oti_plist;
 
         /*
          * Complain if there are pages in the truncated region.
-         *
-         * XXX this is quite expensive check.
          */
-        cl_page_list_init(list);
-        cl_page_gang_lookup(env, clob, io, start + partial, CL_PAGE_EOF, list);
-
-        cl_page_list_for_each(page, list)
-                CL_PAGE_DEBUG(D_ERROR, env, page, "exists %lu\n", start);
-
-        cl_page_list_disown(env, io, list);
-        cl_page_list_fini(env, list);
-
-        cfs_spin_lock(&obj->oo_seatbelt);
-        cfs_list_for_each_entry(cp, &obj->oo_inflight[CRT_WRITE],
-                                ops_inflight) {
-                page = cp->ops_cl.cpl_page;
-                if (page->cp_index >= start + partial) {
-                        cfs_task_t *submitter;
-
-                        submitter = cp->ops_submitter;
-                        /*
-                         * XXX Linux specific debugging stuff.
-                         */
-                        CL_PAGE_DEBUG(D_ERROR, env, page, "%s/%d %lu\n",
-                                      submitter->comm, submitter->pid, start);
-                        libcfs_debug_dumpstack(submitter);
-                }
-        }
-        cfs_spin_unlock(&obj->oo_seatbelt);
+	cl_page_gang_lookup(env, clob, io, start + partial, CL_PAGE_EOF,
+			    trunc_check_cb, (void *)&size);
 }
 #else /* __KERNEL__ */
-# define osc_trunc_check(env, io, oio, size) do {;} while (0)
+static void osc_trunc_check(const struct lu_env *env, struct cl_io *io,
+			    struct osc_io *oio, __u64 size)
+{
+	return;
+}
 #endif
 
 static int osc_io_setattr_start(const struct lu_env *env,
@@ -442,34 +413,36 @@ static int osc_io_setattr_start(const struct lu_env *env,
         struct cl_attr          *attr   = &osc_env_info(env)->oti_attr;
         struct obdo             *oa     = &oio->oi_oa;
 	struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
-        loff_t                   size   = io->u.ci_setattr.sa_attr.lvb_size;
-        unsigned int             ia_valid = io->u.ci_setattr.sa_valid;
-        int                      result = 0;
-        struct obd_info          oinfo = { { { 0 } } };
-
-        if (ia_valid & ATTR_SIZE)
-                osc_trunc_check(env, io, oio, size);
-
-        if (oio->oi_lockless == 0) {
-                cl_object_attr_lock(obj);
-                result = cl_object_attr_get(env, obj, attr);
-                if (result == 0) {
-                        unsigned int cl_valid = 0;
-
-                        if (ia_valid & ATTR_SIZE) {
-                                attr->cat_size = attr->cat_kms = size;
-                                cl_valid = (CAT_SIZE | CAT_KMS);
-                        }
-                        if (ia_valid & ATTR_MTIME_SET) {
-                                attr->cat_mtime = io->u.ci_setattr.sa_attr.lvb_mtime;
-                                cl_valid |= CAT_MTIME;
-                        }
-                        if (ia_valid & ATTR_ATIME_SET) {
-                                attr->cat_atime = io->u.ci_setattr.sa_attr.lvb_atime;
-                                cl_valid |= CAT_ATIME;
-                        }
-                        if (ia_valid & ATTR_CTIME_SET) {
-                                attr->cat_ctime = io->u.ci_setattr.sa_attr.lvb_ctime;
+	__u64                    size   = io->u.ci_setattr.sa_attr.lvb_size;
+	unsigned int             ia_valid = io->u.ci_setattr.sa_valid;
+	int                      result = 0;
+	struct obd_info          oinfo = { { { 0 } } };
+
+	/* truncate cache dirty pages first */
+	if (cl_io_is_trunc(io))
+		result = osc_cache_truncate_start(env, oio, cl2osc(obj), size);
+
+	if (result == 0 && oio->oi_lockless == 0) {
+		cl_object_attr_lock(obj);
+		result = cl_object_attr_get(env, obj, attr);
+		if (result == 0) {
+			struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
+			unsigned int cl_valid = 0;
+
+			if (ia_valid & ATTR_SIZE) {
+				attr->cat_size = attr->cat_kms = size;
+				cl_valid = (CAT_SIZE | CAT_KMS);
+			}
+			if (ia_valid & ATTR_MTIME_SET) {
+				attr->cat_mtime = lvb->lvb_mtime;
+				cl_valid |= CAT_MTIME;
+			}
+			if (ia_valid & ATTR_ATIME_SET) {
+				attr->cat_atime = lvb->lvb_atime;
+				cl_valid |= CAT_ATIME;
+			}
+			if (ia_valid & ATTR_CTIME_SET) {
+				attr->cat_ctime = lvb->lvb_ctime;
                                 cl_valid |= CAT_CTIME;
                         }
                         result = cl_object_attr_set(env, obj, attr, cl_valid);
@@ -518,8 +491,9 @@ static int osc_io_setattr_start(const struct lu_env *env,
 static void osc_io_setattr_end(const struct lu_env *env,
                                const struct cl_io_slice *slice)
 {
-        struct cl_io            *io     = slice->cis_io;
-        struct osc_io           *oio    = cl2osc_io(env, slice);
+	struct cl_io     *io  = slice->cis_io;
+	struct osc_io    *oio = cl2osc_io(env, slice);
+	struct cl_object *obj = slice->cis_obj;
 	struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
         int result;
 
@@ -527,7 +501,6 @@ static void osc_io_setattr_end(const struct lu_env *env,
 
         result = io->ci_result = cbargs->opc_rc;
         if (result == 0) {
-                struct cl_object *obj = slice->cis_obj;
                 if (oio->oi_lockless) {
                         /* lockless truncate */
                         struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
@@ -537,6 +510,15 @@ static void osc_io_setattr_end(const struct lu_env *env,
                         osd->od_stats.os_lockless_truncates++;
                 }
         }
+
+	if (cl_io_is_trunc(io)) {
+		__u64 size = io->u.ci_setattr.sa_attr.lvb_size;
+		osc_trunc_check(env, io, oio, size);
+		if (oio->oi_trunc != NULL) {
+			osc_cache_truncate_end(env, oio, cl2osc(obj));
+			oio->oi_trunc = NULL;
+		}
+	}
 }
 
 static int osc_io_read_start(const struct lu_env *env,
@@ -584,17 +566,15 @@ static int osc_io_write_start(const struct lu_env *env,
         RETURN(result);
 }
 
-static int osc_io_fsync_start(const struct lu_env *env,
-			      const struct cl_io_slice *slice)
+static int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
+			 struct cl_fsync_io *fio)
 {
-	struct cl_io     *io    = slice->cis_io;
-	struct osc_io    *oio   = cl2osc_io(env, slice);
+	struct osc_io    *oio   = osc_env_io(env);
 	struct obdo      *oa    = &oio->oi_oa;
 	struct obd_info  *oinfo = &oio->oi_info;
+	struct lov_oinfo *loi   = obj->oo_oinfo;
 	struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
-	struct cl_object *obj   = slice->cis_obj;
-	struct lov_oinfo *loi   = cl2osc(obj)->oo_oinfo;
-	int result = 0;
+	int rc = 0;
 	ENTRY;
 
 	memset(oa, 0, sizeof(*oa));
@@ -602,32 +582,94 @@ static int osc_io_fsync_start(const struct lu_env *env,
 	oa->o_seq = loi->loi_seq;
 	oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
 
-	/* reload size and blocks for start and end of sync range */
-	oa->o_size = io->u.ci_fsync.fi_start;
-	oa->o_blocks = io->u.ci_fsync.fi_end;
+	/* reload size abd blocks for start and end of sync range */
+	oa->o_size = fio->fi_start;
+	oa->o_blocks = fio->fi_end;
 	oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 
-	obdo_set_parent_fid(oa, io->u.ci_fsync.fi_fid);
+	obdo_set_parent_fid(oa, fio->fi_fid);
 
 	memset(oinfo, 0, sizeof(*oinfo));
 	oinfo->oi_oa = oa;
-	oinfo->oi_capa = io->u.ci_fsync.fi_capa;
+	oinfo->oi_capa = fio->fi_capa;
 	cfs_init_completion(&cbargs->opc_sync);
 
-	result = osc_sync_base(osc_export(cl2osc(obj)), oinfo,
-			       osc_async_upcall, cbargs, PTLRPCD_SET);
+	rc = osc_sync_base(osc_export(obj), oinfo, osc_async_upcall, cbargs,
+			   PTLRPCD_SET);
+	RETURN(rc);
+}
+
+static int osc_io_fsync_start(const struct lu_env *env,
+			      const struct cl_io_slice *slice)
+{
+	struct cl_io       *io  = slice->cis_io;
+	struct cl_fsync_io *fio = &io->u.ci_fsync;
+	struct cl_object   *obj = slice->cis_obj;
+	struct osc_object  *osc = cl2osc(obj);
+	pgoff_t start  = cl_index(obj, fio->fi_start);
+	pgoff_t end    = cl_index(obj, fio->fi_end);
+	int     result = 0;
+	ENTRY;
+
+	if (fio->fi_end == OBD_OBJECT_EOF)
+		end = CL_PAGE_EOF;
+
+	result = osc_cache_writeback_range(env, osc, start, end, 0,
+					   fio->fi_mode == CL_FSYNC_DISCARD);
+	if (result > 0) {
+		fio->fi_nr_written += result;
+		result = 0;
+	}
+	if (fio->fi_mode == CL_FSYNC_ALL) {
+		int rc;
+
+		/* we have to wait for writeback to finish before we can
+		 * send OST_SYNC RPC. This is bad because it causes extents
+		 * to be written osc by osc. However, we usually start
+		 * writeback before CL_FSYNC_ALL so this won't have any real
+		 * problem. */
+		rc = osc_cache_wait_range(env, osc, start, end);
+		if (result == 0)
+			result = rc;
+		rc = osc_fsync_ost(env, osc, fio);
+		if (result == 0)
+			result = rc;
+	}
+
 	RETURN(result);
 }
 
 static void osc_io_fsync_end(const struct lu_env *env,
 			     const struct cl_io_slice *slice)
 {
-	struct cl_io  *io  = slice->cis_io;
+	struct cl_fsync_io *fio = &slice->cis_io->u.ci_fsync;
+	struct cl_object   *obj = slice->cis_obj;
+	pgoff_t start = cl_index(obj, fio->fi_start);
+	pgoff_t end   = cl_index(obj, fio->fi_end);
+	int result = 0;
+
+	if (fio->fi_mode == CL_FSYNC_LOCAL) {
+		result = osc_cache_wait_range(env, cl2osc(obj), start, end);
+	} else if (fio->fi_mode == CL_FSYNC_ALL) {
+		struct osc_io           *oio    = cl2osc_io(env, slice);
+		struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
+
+		cfs_wait_for_completion(&cbargs->opc_sync);
+		if (result == 0)
+			result = cbargs->opc_rc;
+	}
+	slice->cis_io->ci_result = result;
+}
+
+static void osc_io_end(const struct lu_env *env,
+		       const struct cl_io_slice *slice)
+{
 	struct osc_io *oio = cl2osc_io(env, slice);
-	struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
 
-	cfs_wait_for_completion(&cbargs->opc_sync);
-	io->ci_result = cbargs->opc_rc;
+	if (oio->oi_active) {
+		osc_extent_release(env, oio->oi_active);
+		oio->oi_active = NULL;
+	}
 }
 
 static const struct cl_io_operations osc_io_ops = {
@@ -638,6 +680,7 @@ static const struct cl_io_operations osc_io_ops = {
                 },
                 [CIT_WRITE] = {
                         .cio_start  = osc_io_write_start,
+			.cio_end    = osc_io_end,
                         .cio_fini   = osc_io_fini
                 },
                 [CIT_SETATTR] = {
@@ -645,9 +688,10 @@ static const struct cl_io_operations osc_io_ops = {
                         .cio_end    = osc_io_setattr_end
                 },
                 [CIT_FAULT] = {
-                        .cio_fini   = osc_io_fini,
-                        .cio_start  = osc_io_fault_start
-                },
+			.cio_start  = osc_io_fault_start,
+			.cio_end    = osc_io_end,
+			.cio_fini   = osc_io_fini
+		},
 		[CIT_FSYNC] = {
 			.cio_start  = osc_io_fsync_start,
 			.cio_end    = osc_io_fsync_end,
diff --git a/lustre/osc/osc_lock.c b/lustre/osc/osc_lock.c
index de06556..c673e9f 100644
--- a/lustre/osc/osc_lock.c
+++ b/lustre/osc/osc_lock.c
@@ -1330,14 +1330,32 @@ static int osc_lock_use(const struct lu_env *env,
 
 static int osc_lock_flush(struct osc_lock *ols, int discard)
 {
-        struct cl_lock       *lock  = ols->ols_cl.cls_lock;
-        struct cl_env_nest    nest;
-        struct lu_env        *env;
-        int result = 0;
+	struct cl_lock       *lock  = ols->ols_cl.cls_lock;
+	struct cl_env_nest    nest;
+	struct lu_env        *env;
+	int result = 0;
+	ENTRY;
+
+	env = cl_env_nested_get(&nest);
+	if (!IS_ERR(env)) {
+		struct osc_object    *obj   = cl2osc(ols->ols_cl.cls_obj);
+		struct cl_lock_descr *descr = &lock->cll_descr;
+		int rc = 0;
+
+		if (descr->cld_mode >= CLM_WRITE) {
+			result = osc_cache_writeback_range(env, obj,
+					descr->cld_start, descr->cld_end,
+					1, discard);
+			CDEBUG(D_DLMTRACE, "write out %d pages for lock %p.\n",
+			       result, lock);
+			if (result > 0)
+				result = 0;
+		}
+
+		rc = cl_lock_discard_pages(env, lock);
+		if (result == 0 && rc < 0)
+			result = rc;
 
-        env = cl_env_nested_get(&nest);
-        if (!IS_ERR(env)) {
-                result = cl_lock_page_out(env, lock, discard);
                 cl_env_nested_put(&nest, env);
         } else
                 result = PTR_ERR(env);
@@ -1345,7 +1363,7 @@ static int osc_lock_flush(struct osc_lock *ols, int discard)
                 ols->ols_flush = 1;
                 LINVRNT(!osc_lock_has_pages(ols));
         }
-        return result;
+	RETURN(result);
 }
 
 /**
diff --git a/lustre/osc/osc_object.c b/lustre/osc/osc_object.c
index 47ebe5f..5f577ea 100644
--- a/lustre/osc/osc_object.c
+++ b/lustre/osc/osc_object.c
@@ -84,26 +84,46 @@ static int osc_object_init(const struct lu_env *env, struct lu_object *obj,
         for (i = 0; i < CRT_NR; ++i)
                 CFS_INIT_LIST_HEAD(&osc->oo_inflight[i]);
 
-	osc_oap_pages_init(&osc->oo_read_pages);
-	osc_oap_pages_init(&osc->oo_write_pages);
 	CFS_INIT_LIST_HEAD(&osc->oo_ready_item);
 	CFS_INIT_LIST_HEAD(&osc->oo_hp_ready_item);
 	CFS_INIT_LIST_HEAD(&osc->oo_write_item);
 	CFS_INIT_LIST_HEAD(&osc->oo_read_item);
 
+	osc->oo_root.rb_node = NULL;
+	CFS_INIT_LIST_HEAD(&osc->oo_hp_exts);
+	CFS_INIT_LIST_HEAD(&osc->oo_urgent_exts);
+	CFS_INIT_LIST_HEAD(&osc->oo_rpc_exts);
+	CFS_INIT_LIST_HEAD(&osc->oo_reading_exts);
+	cfs_atomic_set(&osc->oo_nr_reads, 0);
+	cfs_atomic_set(&osc->oo_nr_writes, 0);
+	cfs_spin_lock_init(&osc->oo_lock);
+
 	return 0;
 }
 
 static void osc_object_free(const struct lu_env *env, struct lu_object *obj)
 {
-        struct osc_object *osc = lu2osc(obj);
-        int i;
-
-        for (i = 0; i < CRT_NR; ++i)
-                LASSERT(cfs_list_empty(&osc->oo_inflight[i]));
-
-        lu_object_fini(obj);
-        OBD_SLAB_FREE_PTR(osc, osc_object_kmem);
+	struct osc_object *osc = lu2osc(obj);
+	int i;
+
+	for (i = 0; i < CRT_NR; ++i)
+		LASSERT(cfs_list_empty(&osc->oo_inflight[i]));
+
+	LASSERT(cfs_list_empty(&osc->oo_ready_item));
+	LASSERT(cfs_list_empty(&osc->oo_hp_ready_item));
+	LASSERT(cfs_list_empty(&osc->oo_write_item));
+	LASSERT(cfs_list_empty(&osc->oo_read_item));
+
+	LASSERT(osc->oo_root.rb_node == NULL);
+	LASSERT(cfs_list_empty(&osc->oo_hp_exts));
+	LASSERT(cfs_list_empty(&osc->oo_urgent_exts));
+	LASSERT(cfs_list_empty(&osc->oo_rpc_exts));
+	LASSERT(cfs_list_empty(&osc->oo_reading_exts));
+	LASSERT(cfs_atomic_read(&osc->oo_nr_reads) == 0);
+	LASSERT(cfs_atomic_read(&osc->oo_nr_writes) == 0);
+
+	lu_object_fini(obj);
+	OBD_SLAB_FREE_PTR(osc, osc_object_kmem);
 }
 
 int osc_lvb_print(const struct lu_env *env, void *cookie,
diff --git a/lustre/osc/osc_page.c b/lustre/osc/osc_page.c
index 32fb5f4..bc234e2 100644
--- a/lustre/osc/osc_page.c
+++ b/lustre/osc/osc_page.c
@@ -206,9 +206,10 @@ static void osc_page_transfer_add(const struct lu_env *env,
 }
 
 static int osc_page_cache_add(const struct lu_env *env,
-                              const struct cl_page_slice *slice,
-                              struct cl_io *unused)
+			      const struct cl_page_slice *slice,
+			      struct cl_io *io)
 {
+	struct osc_io   *oio = osc_env_io(env);
 	struct osc_page *opg = cl2osc_page(slice);
 	int result;
 	ENTRY;
@@ -216,11 +217,22 @@ static int osc_page_cache_add(const struct lu_env *env,
 	LINVRNT(osc_page_protected(env, opg, CLM_WRITE, 0));
 
 	osc_page_transfer_get(opg, "transfer\0cache");
-	result = osc_queue_async_io(env, opg);
+	result = osc_queue_async_io(env, io, opg);
 	if (result != 0)
 		osc_page_transfer_put(env, opg);
 	else
 		osc_page_transfer_add(env, opg, CRT_WRITE);
+
+	/* for sync write, kernel will wait for this page to be flushed before
+	 * osc_io_end() is called, so release it earlier.
+	 * for mkwrite(), it's known there is no further pages. */
+	if (cl_io_is_sync_write(io) || cl_io_is_mkwrite(io)) {
+		if (oio->oi_active != NULL) {
+			osc_extent_release(env, oio->oi_active);
+			oio->oi_active = NULL;
+		}
+	}
+
 	RETURN(result);
 }
 
@@ -341,17 +353,16 @@ static int osc_page_print(const struct lu_env *env,
         struct client_obd     *cli = &osc_export(obj)->exp_obd->u.cli;
 
         return (*printer)(env, cookie, LUSTRE_OSC_NAME"-page@%p: "
-                          "1< %#x %d %u %s %s %s > "
+			  "1< %#x %d %u %s %s > "
 			  "2< "LPU64" %u %u %#x %#x | %p %p %p > "
-                          "3< %s %p %d %lu %d > "
-                          "4< %d %d %d %lu %s | %s %s %s %s > "
-                          "5< %s %s %s %s | %d %s %s | %d %s %s>\n",
+			  "3< %s %p %d %lu %d > "
+			  "4< %d %d %d %lu %s | %s %s %s %s > "
+			  "5< %s %s %s %s | %d %s | %d %s %s>\n",
                           opg,
                           /* 1 */
                           oap->oap_magic, oap->oap_cmd,
                           oap->oap_interrupted,
                           osc_list(&oap->oap_pending_item),
-                          osc_list(&oap->oap_urgent_item),
                           osc_list(&oap->oap_rpc_item),
                           /* 2 */
                           oap->oap_obj_off, oap->oap_page_off, oap->oap_count,
@@ -375,12 +386,11 @@ static int osc_page_print(const struct lu_env *env,
 			  osc_list(&obj->oo_hp_ready_item),
 			  osc_list(&obj->oo_write_item),
 			  osc_list(&obj->oo_read_item),
-			  obj->oo_read_pages.oop_num_pending,
-			  osc_list(&obj->oo_read_pages.oop_pending),
-			  osc_list(&obj->oo_read_pages.oop_urgent),
-			  obj->oo_write_pages.oop_num_pending,
-			  osc_list(&obj->oo_write_pages.oop_pending),
-			  osc_list(&obj->oo_write_pages.oop_urgent));
+			  cfs_atomic_read(&obj->oo_nr_reads),
+			  osc_list(&obj->oo_reading_exts),
+			  cfs_atomic_read(&obj->oo_nr_writes),
+			  osc_list(&obj->oo_hp_exts),
+			  osc_list(&obj->oo_urgent_exts));
 }
 
 static void osc_page_delete(const struct lu_env *env,
@@ -395,7 +405,7 @@ static void osc_page_delete(const struct lu_env *env,
         ENTRY;
         CDEBUG(D_TRACE, "%p\n", opg);
         osc_page_transfer_put(env, opg);
-	rc = osc_teardown_async_page(obj, opg);
+	rc = osc_teardown_async_page(env, obj, opg);
         if (rc) {
                 CL_PAGE_DEBUG(D_ERROR, env, cl_page_top(slice->cpl_page),
                               "Trying to teardown failed: %d\n", rc);
@@ -425,23 +435,31 @@ void osc_page_clip(const struct lu_env *env, const struct cl_page_slice *slice,
 static int osc_page_cancel(const struct lu_env *env,
                            const struct cl_page_slice *slice)
 {
-        struct osc_page *opg       = cl2osc_page(slice);
-        struct osc_async_page *oap = &opg->ops_oap;
+	struct osc_page *opg = cl2osc_page(slice);
         int rc = 0;
 
         LINVRNT(osc_page_protected(env, opg, CLM_READ, 0));
 
-        client_obd_list_lock(&oap->oap_cli->cl_loi_list_lock);
         /* Check if the transferring against this page
          * is completed, or not even queued. */
         if (opg->ops_transfer_pinned)
                 /* FIXME: may not be interrupted.. */
 		rc = osc_cancel_async_page(env, opg);
         LASSERT(ergo(rc == 0, opg->ops_transfer_pinned == 0));
-        client_obd_list_unlock(&oap->oap_cli->cl_loi_list_lock);
         return rc;
 }
 
+static int osc_page_flush(const struct lu_env *env,
+			  const struct cl_page_slice *slice,
+			  struct cl_io *io)
+{
+	struct osc_page *opg = cl2osc_page(slice);
+	int rc = 0;
+	ENTRY;
+	rc = osc_flush_async_page(env, io, opg);
+	RETURN(rc);
+}
+
 static const struct cl_page_operations osc_page_ops = {
         .cpo_fini          = osc_page_fini,
         .cpo_print         = osc_page_print,
@@ -458,7 +476,8 @@ static const struct cl_page_operations osc_page_ops = {
                 }
         },
         .cpo_clip           = osc_page_clip,
-        .cpo_cancel         = osc_page_cancel
+	.cpo_cancel         = osc_page_cancel,
+	.cpo_flush          = osc_page_flush
 };
 
 struct cl_page *osc_page_init(const struct lu_env *env,
@@ -499,19 +518,34 @@ struct cl_page *osc_page_init(const struct lu_env *env,
  * Helper function called by osc_io_submit() for every page in an immediate
  * transfer (i.e., transferred synchronously).
  */
-void osc_io_submit_page(const struct lu_env *env,
-                        struct osc_io *oio, struct osc_page *opg,
-                        enum cl_req_type crt)
+void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
+		     enum cl_req_type crt, int brw_flags)
 {
-        LINVRNT(osc_page_protected(env, opg,
-                                   crt == CRT_WRITE ? CLM_WRITE : CLM_READ, 1));
-
-	osc_queue_sync_page(env, opg,
-			    crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ,
-			    osc_io_srvlock(oio) ? OBD_BRW_SRVLOCK : 0);
-
-        osc_page_transfer_get(opg, "transfer\0imm");
-        osc_page_transfer_add(env, opg, crt);
+	struct osc_async_page *oap = &opg->ops_oap;
+	struct osc_object     *obj = oap->oap_obj;
+
+	LINVRNT(osc_page_protected(env, opg,
+				   crt == CRT_WRITE ? CLM_WRITE : CLM_READ, 1));
+
+	LASSERTF(oap->oap_magic == OAP_MAGIC, "Bad oap magic: oap %p, "
+		 "magic 0x%x\n", oap, oap->oap_magic);
+	LASSERT(oap->oap_async_flags & ASYNC_READY);
+	LASSERT(oap->oap_async_flags & ASYNC_COUNT_STABLE);
+
+	oap->oap_cmd       = crt == CRT_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ;
+	oap->oap_page_off  = opg->ops_from;
+	oap->oap_count     = opg->ops_to - opg->ops_from;
+	oap->oap_brw_flags = OBD_BRW_SYNC | brw_flags;
+
+	if (!client_is_remote(osc_export(obj)) &&
+			cfs_capable(CFS_CAP_SYS_RESOURCE)) {
+		oap->oap_brw_flags |= OBD_BRW_NOQUOTA;
+		oap->oap_cmd |= OBD_BRW_NOQUOTA;
+	}
+
+	opg->ops_submit_time = cfs_time_current();
+	osc_page_transfer_get(opg, "transfer\0imm");
+	osc_page_transfer_add(env, opg, crt);
 }
 
 /** @} osc */
diff --git a/lustre/osc/osc_request.c b/lustre/osc/osc_request.c
index 3aaf175..8e86162 100644
--- a/lustre/osc/osc_request.c
+++ b/lustre/osc/osc_request.c
@@ -815,7 +815,7 @@ static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
                                 (cli->cl_max_rpcs_in_flight + 1);
                 oa->o_undirty = max(cli->cl_dirty_max, max_in_flight);
         }
-        oa->o_grant = cli->cl_avail_grant;
+	oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
         oa->o_dropped = cli->cl_lost_grant;
         cli->cl_lost_grant = 0;
         client_obd_list_unlock(&cli->cl_loi_list_lock);
@@ -1030,15 +1030,17 @@ static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
                 cli->cl_avail_grant = ocd->ocd_grant;
         }
 
-        client_obd_list_unlock(&cli->cl_loi_list_lock);
+	/* determine the appropriate chunk size used by osc_extent. */
+	cli->cl_chunkbits = max_t(int, CFS_PAGE_SHIFT, ocd->ocd_blocksize);
+	client_obd_list_unlock(&cli->cl_loi_list_lock);
 
-        CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld \n",
-               cli->cl_import->imp_obd->obd_name,
-               cli->cl_avail_grant, cli->cl_lost_grant);
+	CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld."
+		"chunk bits: %d.\n", cli->cl_import->imp_obd->obd_name,
+		cli->cl_avail_grant, cli->cl_lost_grant, cli->cl_chunkbits);
 
-        if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
-            cfs_list_empty(&cli->cl_grant_shrink_list))
-                osc_add_shrink_grant(cli);
+	if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
+	    cfs_list_empty(&cli->cl_grant_shrink_list))
+		osc_add_shrink_grant(cli);
 }
 
 /* We assume that the reason this OSC got a short read is because it read
@@ -1652,7 +1654,6 @@ int osc_brw_redo_request(struct ptlrpc_request *request,
                          struct osc_brw_async_args *aa)
 {
         struct ptlrpc_request *new_req;
-        struct ptlrpc_request_set *set = request->rq_set;
         struct osc_brw_async_args *new_aa;
         struct osc_async_page *oap;
         int rc = 0;
@@ -1669,15 +1670,12 @@ int osc_brw_redo_request(struct ptlrpc_request *request,
         if (rc)
                 RETURN(rc);
 
-        client_obd_list_lock(&aa->aa_cli->cl_loi_list_lock);
-
         cfs_list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
                 if (oap->oap_request != NULL) {
                         LASSERTF(request == oap->oap_request,
                                  "request %p != oap_request %p\n",
                                  request, oap->oap_request);
                         if (oap->oap_interrupted) {
-                                client_obd_list_unlock(&aa->aa_cli->cl_loi_list_lock);
                                 ptlrpc_req_finished(new_req);
                                 RETURN(-EINTR);
                         }
@@ -1695,8 +1693,9 @@ int osc_brw_redo_request(struct ptlrpc_request *request,
         new_aa = ptlrpc_req_async_args(new_req);
 
         CFS_INIT_LIST_HEAD(&new_aa->aa_oaps);
-        cfs_list_splice(&aa->aa_oaps, &new_aa->aa_oaps);
-        CFS_INIT_LIST_HEAD(&aa->aa_oaps);
+	cfs_list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
+	CFS_INIT_LIST_HEAD(&new_aa->aa_exts);
+	cfs_list_splice_init(&aa->aa_exts, &new_aa->aa_exts);
 
         cfs_list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
                 if (oap->oap_request) {
@@ -1708,16 +1707,14 @@ int osc_brw_redo_request(struct ptlrpc_request *request,
         new_aa->aa_ocapa = aa->aa_ocapa;
         aa->aa_ocapa = NULL;
 
-        /* use ptlrpc_set_add_req is safe because interpret functions work
-         * in check_set context. only one way exist with access to request
-         * from different thread got -EINTR - this way protected with
-         * cl_loi_list_lock */
-        ptlrpc_set_add_req(set, new_req);
+	/* XXX: This code will run into problem if we're going to support
+	 * to add a series of BRW RPCs into a self-defined ptlrpc_request_set
+	 * and wait for all of them to be finished. We should inherit request
+	 * set from old request. */
+	ptlrpcd_add_req(new_req, PDL_POLICY_SAME, -1);
 
-        client_obd_list_unlock(&aa->aa_cli->cl_loi_list_lock);
-
-        DEBUG_REQ(D_INFO, new_req, "new request");
-        RETURN(0);
+	DEBUG_REQ(D_INFO, new_req, "new request");
+	RETURN(0);
 }
 
 /*
@@ -1874,9 +1871,11 @@ out:
 static int brw_interpret(const struct lu_env *env,
                          struct ptlrpc_request *req, void *data, int rc)
 {
-        struct osc_brw_async_args *aa = data;
-	struct osc_async_page *oap, *tmp;
-        struct client_obd *cli;
+	struct osc_brw_async_args *aa = data;
+	struct osc_extent *ext;
+	struct osc_extent *tmp;
+	struct cl_object  *obj = NULL;
+	struct client_obd *cli = aa->aa_cli;
         ENTRY;
 
         rc = osc_brw_fini_request(req, rc);
@@ -1911,46 +1910,80 @@ static int brw_interpret(const struct lu_env *env,
                 aa->aa_ocapa = NULL;
         }
 
-        cli = aa->aa_cli;
-        client_obd_list_lock(&cli->cl_loi_list_lock);
+	cfs_list_for_each_entry_safe(ext, tmp, &aa->aa_exts, oe_link) {
+		if (obj == NULL && rc == 0) {
+			obj = osc2cl(ext->oe_obj);
+			cl_object_get(obj);
+		}
 
-        /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
-         * is called so we know whether to go to sync BRWs or wait for more
-         * RPCs to complete */
-        if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
-                cli->cl_w_in_flight--;
-        else
-                cli->cl_r_in_flight--;
-
-	/* the caller may re-use the oap after the completion call so
-	 * we need to clean it up a little */
-	cfs_list_for_each_entry_safe(oap, tmp, &aa->aa_oaps,
-			oap_rpc_item) {
-		cfs_list_del_init(&oap->oap_rpc_item);
-		osc_ap_completion(env, cli, aa->aa_oa, oap, 1, rc);
+		cfs_list_del_init(&ext->oe_link);
+		osc_extent_finish(env, ext, 1, rc);
+	}
+	LASSERT(cfs_list_empty(&aa->aa_exts));
+	LASSERT(cfs_list_empty(&aa->aa_oaps));
+
+	if (obj != NULL) {
+		struct obdo *oa = aa->aa_oa;
+		struct cl_attr *attr  = &osc_env_info(env)->oti_attr;
+		unsigned long valid = 0;
+
+		LASSERT(rc == 0);
+		if (oa->o_valid & OBD_MD_FLBLOCKS) {
+			attr->cat_blocks = oa->o_blocks;
+			valid |= CAT_BLOCKS;
+		}
+		if (oa->o_valid & OBD_MD_FLMTIME) {
+			attr->cat_mtime = oa->o_mtime;
+			valid |= CAT_MTIME;
+		}
+		if (oa->o_valid & OBD_MD_FLATIME) {
+			attr->cat_atime = oa->o_atime;
+			valid |= CAT_ATIME;
+		}
+		if (oa->o_valid & OBD_MD_FLCTIME) {
+			attr->cat_ctime = oa->o_ctime;
+			valid |= CAT_CTIME;
+		}
+		if (valid != 0) {
+			cl_object_attr_lock(obj);
+			cl_object_attr_set(env, obj, attr, valid);
+			cl_object_attr_unlock(obj);
+		}
+		cl_object_put(env, obj);
 	}
 	OBDO_FREE(aa->aa_oa);
 
-	osc_wake_cache_waiters(cli);
-	osc_io_unplug(env, cli, NULL, PDL_POLICY_SAME);
-	client_obd_list_unlock(&cli->cl_loi_list_lock);
-
 	cl_req_completion(env, aa->aa_clerq, rc < 0 ? rc :
 			  req->rq_bulk->bd_nob_transferred);
 	osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
 	ptlrpc_lprocfs_brw(req, req->rq_bulk->bd_nob_transferred);
 
+	client_obd_list_lock(&cli->cl_loi_list_lock);
+	/* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
+	 * is called so we know whether to go to sync BRWs or wait for more
+	 * RPCs to complete */
+	if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
+		cli->cl_w_in_flight--;
+	else
+		cli->cl_r_in_flight--;
+	osc_wake_cache_waiters(cli);
+	client_obd_list_unlock(&cli->cl_loi_list_lock);
+
+	osc_io_unplug(env, cli, NULL, PDL_POLICY_SAME);
 	RETURN(rc);
 }
 
-/* The most tricky part of this function is that it will return with
- * cli->cli_loi_list_lock held.
+/**
+ * Build an RPC by the list of extent @ext_list. The caller must ensure
+ * that the total pages in this list are NOT over max pages per RPC.
+ * Extents in the list must be in OES_RPC state.
  */
 int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
-		  cfs_list_t *rpc_list, int page_count, int cmd,
-		  pdl_policy_t pol)
+		  cfs_list_t *ext_list, int cmd, pdl_policy_t pol)
 {
 	struct ptlrpc_request *req = NULL;
+	struct osc_extent *ext;
+	CFS_LIST_HEAD(rpc_list);
 	struct brw_page **pga = NULL;
 	struct osc_brw_async_args *aa = NULL;
         struct obdo *oa = NULL;
@@ -1960,17 +1993,39 @@ int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
         enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE : CRT_READ;
         struct ldlm_lock *lock = NULL;
         struct cl_req_attr crattr;
-        int i, rc, mpflag = 0;
+	obd_off starting_offset = OBD_OBJECT_EOF;
+	obd_off ending_offset = 0;
+	int i, rc, mpflag = 0, mem_tight = 0, page_count = 0;
 
-        ENTRY;
-        LASSERT(!cfs_list_empty(rpc_list));
+	ENTRY;
+	LASSERT(!cfs_list_empty(ext_list));
+
+	/* add pages into rpc_list to build BRW rpc */
+	cfs_list_for_each_entry(ext, ext_list, oe_link) {
+		LASSERT(ext->oe_state == OES_RPC);
+		mem_tight |= ext->oe_memalloc;
+		cfs_list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
+			++page_count;
+			cfs_list_add_tail(&oap->oap_rpc_item, &rpc_list);
+			if (starting_offset > oap->oap_obj_off)
+				starting_offset = oap->oap_obj_off;
+			else
+				LASSERT(oap->oap_page_off == 0);
+			if (ending_offset < oap->oap_obj_off + oap->oap_count)
+				ending_offset = oap->oap_obj_off +
+						oap->oap_count;
+			else
+				LASSERT(oap->oap_page_off + oap->oap_count ==
+					CFS_PAGE_SIZE);
+		}
+	}
 
-        if (cmd & OBD_BRW_MEMALLOC)
-                mpflag = cfs_memory_pressure_get_and_set();
+	if (mem_tight)
+		mpflag = cfs_memory_pressure_get_and_set();
 
-        memset(&crattr, 0, sizeof crattr);
-        OBD_ALLOC(pga, sizeof(*pga) * page_count);
-        if (pga == NULL)
+	memset(&crattr, 0, sizeof crattr);
+	OBD_ALLOC(pga, sizeof(*pga) * page_count);
+	if (pga == NULL)
 		GOTO(out, rc = -ENOMEM);
 
 	OBDO_ALLOC(oa);
@@ -1978,16 +2033,18 @@ int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
 		GOTO(out, rc = -ENOMEM);
 
 	i = 0;
-	cfs_list_for_each_entry(oap, rpc_list, oap_rpc_item) {
-		struct cl_page *page = osc_oap2cl_page(oap);
+	cfs_list_for_each_entry(oap, &rpc_list, oap_rpc_item) {
+		struct cl_page *page = oap2cl_page(oap);
 		if (clerq == NULL) {
 			clerq = cl_req_alloc(env, page, crt,
 					     1 /* only 1-object rpcs for
 						* now */);
 			if (IS_ERR(clerq))
 				GOTO(out, rc = PTR_ERR(clerq));
-                        lock = oap->oap_ldlm_lock;
-                }
+			lock = oap->oap_ldlm_lock;
+		}
+		if (mem_tight)
+			oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
                 pga[i] = &oap->oap_brw_page;
                 pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
                 CDEBUG(0, "put page %p index %lu oap %p flg %x to pga\n",
@@ -1998,8 +2055,8 @@ int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
 
         /* always get the data for the obdo for the rpc */
 	LASSERT(clerq != NULL);
-        crattr.cra_oa = oa;
-        crattr.cra_capa = NULL;
+	crattr.cra_oa = oa;
+	crattr.cra_capa = NULL;
 	memset(crattr.cra_jobid, 0, JOBSTATS_JOBID_SIZE);
         cl_req_attr_set(env, clerq, &crattr, ~0ULL);
         if (lock) {
@@ -2011,18 +2068,18 @@ int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
         if (rc != 0) {
                 CERROR("cl_req_prep failed: %d\n", rc);
 		GOTO(out, rc);
-        }
+	}
 
-        sort_brw_pages(pga, page_count);
-        rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count,
-                                  pga, &req, crattr.cra_capa, 1, 0);
-        if (rc != 0) {
-                CERROR("prep_req failed: %d\n", rc);
+	sort_brw_pages(pga, page_count);
+	rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count,
+			pga, &req, crattr.cra_capa, 1, 0);
+	if (rc != 0) {
+		CERROR("prep_req failed: %d\n", rc);
 		GOTO(out, rc);
 	}
 
 	req->rq_interpret_reply = brw_interpret;
-        if (cmd & OBD_BRW_MEMALLOC)
+	if (mem_tight != 0)
                 req->rq_memalloc = 1;
 
         /* Need to update the timestamps after the request is built in case
@@ -2035,17 +2092,72 @@ int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
 
 	lustre_msg_set_jobid(req->rq_reqmsg, crattr.cra_jobid);
 
-        CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
-        aa = ptlrpc_req_async_args(req);
-        CFS_INIT_LIST_HEAD(&aa->aa_oaps);
-        cfs_list_splice(rpc_list, &aa->aa_oaps);
-        CFS_INIT_LIST_HEAD(rpc_list);
-        aa->aa_clerq = clerq;
+	CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
+	aa = ptlrpc_req_async_args(req);
+	CFS_INIT_LIST_HEAD(&aa->aa_oaps);
+	cfs_list_splice_init(&rpc_list, &aa->aa_oaps);
+	CFS_INIT_LIST_HEAD(&aa->aa_exts);
+	cfs_list_splice_init(ext_list, &aa->aa_exts);
+	aa->aa_clerq = clerq;
+
+	/* queued sync pages can be torn down while the pages
+	 * were between the pending list and the rpc */
+	tmp = NULL;
+	cfs_list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
+		/* only one oap gets a request reference */
+		if (tmp == NULL)
+			tmp = oap;
+		if (oap->oap_interrupted && !req->rq_intr) {
+			CDEBUG(D_INODE, "oap %p in req %p interrupted\n",
+					oap, req);
+			ptlrpc_mark_interrupted(req);
+		}
+	}
+	if (tmp != NULL)
+		tmp->oap_request = ptlrpc_request_addref(req);
+
+	client_obd_list_lock(&cli->cl_loi_list_lock);
+	starting_offset >>= CFS_PAGE_SHIFT;
+	if (cmd == OBD_BRW_READ) {
+		cli->cl_r_in_flight++;
+		lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
+		lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
+		lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
+				      starting_offset + 1);
+	} else {
+		cli->cl_w_in_flight++;
+		lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
+		lprocfs_oh_tally(&cli->cl_write_rpc_hist, cli->cl_w_in_flight);
+		lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
+				      starting_offset + 1);
+	}
+	client_obd_list_unlock(&cli->cl_loi_list_lock);
+
+	DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %dr/%dw in flight",
+		  page_count, aa, cli->cl_r_in_flight,
+		  cli->cl_w_in_flight);
+
+	/* XXX: Maybe the caller can check the RPC bulk descriptor to
+	 * see which CPU/NUMA node the majority of pages were allocated
+	 * on, and try to assign the async RPC to the CPU core
+	 * (PDL_POLICY_PREFERRED) to reduce cross-CPU memory traffic.
+	 *
+	 * But on the other hand, we expect that multiple ptlrpcd
+	 * threads and the initial write sponsor can run in parallel,
+	 * especially when data checksum is enabled, which is CPU-bound
+	 * operation and single ptlrpcd thread cannot process in time.
+	 * So more ptlrpcd threads sharing BRW load
+	 * (with PDL_POLICY_ROUND) seems better.
+	 */
+	ptlrpcd_add_req(req, pol, -1);
+	rc = 0;
+	EXIT;
+
 out:
-        if (cmd & OBD_BRW_MEMALLOC)
-                cfs_memory_pressure_restore(mpflag);
+	if (mem_tight != 0)
+		cfs_memory_pressure_restore(mpflag);
 
-        capa_put(crattr.cra_capa);
+	capa_put(crattr.cra_capa);
 	if (rc != 0) {
 		LASSERT(req == NULL);
 
@@ -2055,59 +2167,14 @@ out:
                         OBD_FREE(pga, sizeof(*pga) * page_count);
                 /* this should happen rarely and is pretty bad, it makes the
                  * pending list not follow the dirty order */
-                client_obd_list_lock(&cli->cl_loi_list_lock);
-                cfs_list_for_each_entry_safe(oap, tmp, rpc_list, oap_rpc_item) {
-                        cfs_list_del_init(&oap->oap_rpc_item);
-
-                        /* queued sync pages can be torn down while the pages
-                         * were between the pending list and the rpc */
-                        if (oap->oap_interrupted) {
-                                CDEBUG(D_INODE, "oap %p interrupted\n", oap);
-                                osc_ap_completion(env, cli, NULL, oap, 0,
-                                                  oap->oap_count);
-                                continue;
-                        }
-			osc_ap_completion(env, cli, NULL, oap, 0, rc);
+		while (!cfs_list_empty(ext_list)) {
+			ext = cfs_list_entry(ext_list->next, struct osc_extent,
+					     oe_link);
+			cfs_list_del_init(&ext->oe_link);
+			osc_extent_finish(env, ext, 0, rc);
 		}
 		if (clerq && !IS_ERR(clerq))
 			cl_req_completion(env, clerq, rc);
-	} else {
-		struct osc_async_page *tmp = NULL;
-
-		/* queued sync pages can be torn down while the pages
-		 * were between the pending list and the rpc */
-		LASSERT(aa != NULL);
-		client_obd_list_lock(&cli->cl_loi_list_lock);
-		cfs_list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
-			/* only one oap gets a request reference */
-			if (tmp == NULL)
-				tmp = oap;
-			if (oap->oap_interrupted && !req->rq_intr) {
-				CDEBUG(D_INODE, "oap %p in req %p interrupted\n",
-						oap, req);
-				ptlrpc_mark_interrupted(req);
-			}
-		}
-		if (tmp != NULL)
-			tmp->oap_request = ptlrpc_request_addref(req);
-
-		DEBUG_REQ(D_INODE,req, "%d pages, aa %p. now %dr/%dw in flight",
-			  page_count, aa, cli->cl_r_in_flight,
-			  cli->cl_w_in_flight);
-
-		/* XXX: Maybe the caller can check the RPC bulk descriptor to
-		 * see which CPU/NUMA node the majority of pages were allocated
-		 * on, and try to assign the async RPC to the CPU core
-		 * (PDL_POLICY_PREFERRED) to reduce cross-CPU memory traffic.
-		 *
-		 * But on the other hand, we expect that multiple ptlrpcd
-		 * threads and the initial write sponsor can run in parallel,
-		 * especially when data checksum is enabled, which is CPU-bound
-		 * operation and single ptlrpcd thread cannot process in time.
-		 * So more ptlrpcd threads sharing BRW load
-		 * (with PDL_POLICY_ROUND) seems better.
-		 */
-		ptlrpcd_add_req(req, pol, -1);
 	}
 	RETURN(rc);
 }
@@ -3281,15 +3348,12 @@ static int osc_reconnect(const struct lu_env *env,
                 cli->cl_lost_grant = 0;
                 client_obd_list_unlock(&cli->cl_loi_list_lock);
 
-                CDEBUG(D_CACHE, "request ocd_grant: %d cl_avail_grant: %ld "
-                       "cl_dirty: %ld cl_lost_grant: %ld\n", data->ocd_grant,
-                       cli->cl_avail_grant, cli->cl_dirty, lost_grant);
                 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d"
-                       " ocd_grant: %d\n", data->ocd_connect_flags,
-                       data->ocd_version, data->ocd_grant);
-        }
+		       " ocd_grant: %d, lost: %ld.\n", data->ocd_connect_flags,
+		       data->ocd_version, data->ocd_grant, lost_grant);
+	}
 
-        RETURN(0);
+	RETURN(0);
 }
 
 static int osc_disconnect(struct obd_export *exp)
@@ -3374,11 +3438,9 @@ static int osc_import_event(struct obd_device *obd,
                 if (!IS_ERR(env)) {
                         /* Reset grants */
                         cli = &obd->u.cli;
-                        client_obd_list_lock(&cli->cl_loi_list_lock);
                         /* all pages go to failing rpcs due to the invalid
                          * import */
 			osc_io_unplug(env, cli, NULL, PDL_POLICY_ROUND);
-                        client_obd_list_unlock(&cli->cl_loi_list_lock);
 
                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
                         cl_env_put(env, &refcheck);
@@ -3459,9 +3521,7 @@ static int brw_queue_work(const struct lu_env *env, void *data)
 
 	CDEBUG(D_CACHE, "Run writeback work for client obd %p.\n", cli);
 
-	client_obd_list_lock(&cli->cl_loi_list_lock);
 	osc_io_unplug(env, cli, NULL, PDL_POLICY_SAME);
-	client_obd_list_unlock(&cli->cl_loi_list_lock);
 	RETURN(0);
 }
 
diff --git a/lustre/ptlrpc/client.c b/lustre/ptlrpc/client.c
index 261e50d..140b342 100644
--- a/lustre/ptlrpc/client.c
+++ b/lustre/ptlrpc/client.c
@@ -2776,7 +2776,7 @@ void ptlrpc_abort_inflight(struct obd_import *imp)
                 cfs_spin_lock (&req->rq_lock);
                 if (req->rq_import_generation < imp->imp_generation) {
                         req->rq_err = 1;
-                        req->rq_status = -EINTR;
+			req->rq_status = -EIO;
                         ptlrpc_client_wake_req(req);
                 }
                 cfs_spin_unlock (&req->rq_lock);
@@ -2791,7 +2791,7 @@ void ptlrpc_abort_inflight(struct obd_import *imp)
                 cfs_spin_lock (&req->rq_lock);
                 if (req->rq_import_generation < imp->imp_generation) {
                         req->rq_err = 1;
-                        req->rq_status = -EINTR;
+			req->rq_status = -EIO;
                         ptlrpc_client_wake_req(req);
                 }
                 cfs_spin_unlock (&req->rq_lock);
diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index 0a667b7..a008b0b 100644
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -3032,6 +3032,29 @@ test_48e() { # bug 4134
 }
 run_test 48e "Access to recreated parent subdir (should return errors)"
 
+test_49() { # LU-1030
+	# get ost1 size - lustre-OST0000
+	ost1_size=$(do_facet ost1 lfs df |grep ${ost1_svc} |awk '{print $4}')
+	# write 800M at maximum
+	[ $ost1_size -gt 819200 ] && ost1_size=819200
+
+	lfs setstripe -c 1 -i 0 $DIR/$tfile
+	dd if=/dev/zero of=$DIR/$tfile bs=4k count=$((ost1_size >> 2)) &
+	local dd_pid=$!
+
+	# change max_pages_per_rpc while writing the file
+	local osc1_mppc=osc.$(get_osc_import_name client ost1).max_pages_per_rpc
+	local orig_mppc=`$LCTL get_param -n $osc1_mppc`
+	# loop until dd process exits
+	while ps ax -opid | grep -q $dd_pid; do
+		$LCTL set_param $osc1_mppc=$((RANDOM % 256 + 1))
+		sleep $((RANDOM % 5 + 1))
+	done
+	# restore original max_pages_per_rpc
+	$LCTL set_param $osc1_mppc=$orig_mppc
+}
+run_test 49 "Change max_pages_per_rpc won't break osc extent"
+
 test_50() {
 	# bug 1485
 	mkdir $DIR/d50