liowait:

 This patch adds support for the POSIX listio LIO_WAIT mechanism. This works
by adding an IOCB_CMD_CHECKPOINT command.

  As part of listio submission, the user may choose to prepend an empty iocb
to the list with an aio_lio_opcode of IOCB_CMD_CHECKPOINT. All iocbs following
this particular CHECKPOINT iocb are in the same group and sys_io_submit will
block until all iocbs submitted in the group have completed.

 Makefile                |    2 -
 fs/aio.c                |   94 +++++++++++++++++++++++++++++++++++++++++++++---
 include/linux/aio_abi.h |    1
 3 files changed, 91 insertions(+), 6 deletions(-)

Signed-off-by: Sébastien Dugué <sebastien.dugue@bull.net>

Index: linux-2.6.12/fs/aio.c
===================================================================
--- linux-2.6.12.orig/fs/aio.c	2005-07-21 11:30:57.000000000 +0200
+++ linux-2.6.12/fs/aio.c	2005-07-21 11:35:03.000000000 +0200
@@ -954,8 +954,24 @@
 	read_unlock(&tasklist_lock);
 }
 
+static void wait_for_lio(struct kioctx *ctx, struct lio_event *lio)
+{
+	struct task_struct *tsk = current;
+	DECLARE_WAITQUEUE(wait, tsk);
+
+	add_wait_queue(&ctx->wait, &wait);
+	set_task_state(tsk, TASK_UNINTERRUPTIBLE);
+
+	while ( atomic_read(&lio->lio_users) ) {
+		schedule();
+		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
+	}
+	__set_task_state(tsk, TASK_RUNNING);
+	remove_wait_queue(&ctx->wait, &wait);
+}
+
 static inline void check_lio(struct kioctx *ctx, 
-			     struct lio_event *lio)
+			     struct lio_event *lio, int wait)
 {
 	int ret;
 
@@ -979,7 +995,52 @@
 			kfree(lio);
 		}
 
+	} else if (wait) {
+
+		/* if needed, wait end of all AIO of this list */
+
+		wait_for_lio(ctx, lio);
+
+		/* free memory */
+
+		kfree(lio);
+	}
+}
+
+static int create_lio_wait(struct kioctx *ctx, struct lio_event **lio,
+			    int enable)
+{
+	if (*lio != NULL) {
+
+		/* we break an existing sequence (lio exists)
+		 * close the sequence
+		 */
+
+		check_lio(ctx, *lio, 1);
+		*lio = NULL;
+	}
+
+	if (!enable) {
+
+		/* it means we don't want to monitor following commands
+		 * usefull to break current sequence
+		 */
+
+		return 0;
 	}
+
+
+	*lio = kmalloc(sizeof(*lio), GFP_KERNEL);
+	if (!*lio)
+		return -EAGAIN;
+
+	atomic_set(&((*lio)->lio_users), 1);
+	(*lio)->lio_pid = 0;
+	(*lio)->lio_signo = 0;
+	(*lio)->lio_notify = 0;
+	(*lio)->lio_value = 0;
+
+	return 0;
 }
 
 static int create_lio_event(struct kioctx *ctx,struct lio_event **lio,
@@ -1023,7 +1084,7 @@
 	 * close the sequence
 	 */
 
-	check_lio(ctx, *lio);
+	check_lio(ctx, *lio, 1);
 
 	if (signo == 0) {
 
@@ -1153,7 +1214,7 @@
 		__aio_send_signal(iocb->ki_pid, iocb->ki_signo, iocb->ki_notify,
 				  (void*)(unsigned long)iocb->ki_value);
 
-	check_lio(ctx, iocb->ki_lio_event);
+	check_lio(ctx, iocb->ki_lio_event, 0);
 
 	pr_debug("%ld retries: %d of %d\n", iocb->ki_retried,
 		iocb->ki_nbytes - iocb->ki_left, iocb->ki_nbytes);
@@ -1882,7 +1943,30 @@
 			}
 		}
 
-		if (tmp.aio_lio_opcode == IOCB_CMD_EVENT) {
+		if (tmp.aio_lio_opcode == IOCB_CMD_CHECKPOINT) {
+			/* this command means that all following IO commands
+			 * are in the same group.
+			 * io_submit() will wait all commands of the group
+			 * have been processed before exiting.
+			 */
+
+			if (aio_pid) {
+				ret = -EINVAL;
+				break;
+			}
+
+			if ( (aio_signo != 0) && (aio_signo != 1) ) {
+				ret = -EINVAL;
+				break;
+			}
+
+			ret = create_lio_wait(ctx, &lio, aio_signo);
+			if (ret)
+				break;
+
+			continue;
+
+		} else if (tmp.aio_lio_opcode == IOCB_CMD_EVENT) {
 
 			/* this command means that all following IO commands
 			 * are in the same group and when the group becomes
@@ -1921,7 +2005,7 @@
 				break;
 		}
 	}
-	check_lio(ctx, lio);
+	check_lio(ctx, lio, 1);
 
 	put_ioctx(ctx);
 	return i ? i : ret;
Index: linux-2.6.12/include/linux/aio_abi.h
===================================================================
--- linux-2.6.12.orig/include/linux/aio_abi.h	2005-07-21 11:30:03.000000000 +0200
+++ linux-2.6.12/include/linux/aio_abi.h	2005-07-21 11:35:03.000000000 +0200
@@ -42,6 +42,7 @@
 	 */
 	IOCB_CMD_NOOP = 6,
 	IOCB_CMD_EVENT = 7,
+	IOCB_CMD_CHECKPOINT = 8,
 };
 
 /* read() from /dev/aio returns these structures. */
Index: linux-2.6.12/Makefile
===================================================================
--- linux-2.6.12.orig/Makefile	2005-07-21 11:30:03.000000000 +0200
+++ linux-2.6.12/Makefile	2005-07-21 11:35:03.000000000 +0200
@@ -1,7 +1,7 @@
 VERSION = 2
 PATCHLEVEL = 6
 SUBLEVEL = 12
-EXTRAVERSION = .PAIO-lioevent
+EXTRAVERSION = .PAIO-liowait
 NAME=Woozy Numbat
 
 # *DOCUMENTATION*