[PATCH 1/2] scsi: remove bogus use of struct execute_work in sg

[PATCH 1/2] scsi: remove bogus use of struct execute_work in sg

[Tejun Heo]

execute_work usage in sg is bogus.  execute_in_process_context() is
never used and ew is used purely as wrapper around work_struct which
is superflous.  Use work_struct directly.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
---
This hasn't changed from the previous posting.

  http://thread.gmane.org/gmane.linux.scsi/62923

Thanks.

 drivers/scsi/sg.c |   16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

Index: work/drivers/scsi/sg.c
===================================================================
--- work.orig/drivers/scsi/sg.c
+++ work/drivers/scsi/sg.c
@@ -139,7 +139,7 @@ typedef struct sg_request {	/* SG_MAX_QU
 	volatile char done;	/* 0->before bh, 1->before read, 2->read */
 	struct request *rq;
 	struct bio *bio;
-	struct execute_work ew;
+	struct work_struct work;
 } Sg_request;
 
 typedef struct sg_fd {		/* holds the state of a file descriptor */
@@ -162,7 +162,7 @@ typedef struct sg_fd {		/* holds the sta
 	char keep_orphan;	/* 0 -> drop orphan (def), 1 -> keep for read() */
 	char mmap_called;	/* 0 -> mmap() never called on this fd */
 	struct kref f_ref;
-	struct execute_work ew;
+	struct work_struct work;
 } Sg_fd;
 
 typedef struct sg_device { /* holds the state of each scsi generic device */
@@ -1248,7 +1248,7 @@ sg_mmap(struct file *filp, struct vm_are
 
 static void sg_rq_end_io_usercontext(struct work_struct *work)
 {
-	struct sg_request *srp = container_of(work, struct sg_request, ew.work);
+	struct sg_request *srp = container_of(work, struct sg_request, work);
 	struct sg_fd *sfp = srp->parentfp;
 
 	sg_finish_rem_req(srp);
@@ -1335,8 +1335,8 @@ static void sg_rq_end_io(struct request
 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
 		kref_put(&sfp->f_ref, sg_remove_sfp);
 	} else {
-		INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
-		schedule_work(&srp->ew.work);
+		INIT_WORK(&srp->work, sg_rq_end_io_usercontext);
+		schedule_work(&srp->work);
 	}
 }
 
@@ -2089,7 +2089,7 @@ sg_add_sfp(Sg_device * sdp, int dev)
 
 static void sg_remove_sfp_usercontext(struct work_struct *work)
 {
-	struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
+	struct sg_fd *sfp = container_of(work, struct sg_fd, work);
 	struct sg_device *sdp = sfp->parentdp;
 
 	/* Cleanup any responses which were never read(). */
@@ -2126,8 +2126,8 @@ static void sg_remove_sfp(struct kref *k
 	write_unlock_irqrestore(&sg_index_lock, iflags);
 	wake_up_interruptible(&sdp->o_excl_wait);
 
-	INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
-	schedule_work(&sfp->ew.work);
+	INIT_WORK(&sfp->work, sg_remove_sfp_usercontext);
+	schedule_work(&sfp->work);
 }
 
 static int

[PATCH] scsi: don't use execute_in_process_context()

[Tejun Heo]

SCSI is the only subsystem which uses execute_in_process_context() and
its use is racy against module unload.  ie. the reap work is not
properly flushed and could still be running after the scsi module is
unloaded.

Although execute_in_process_context() can be more efficient when the
caller already has a context, in this case, the call paths are quite
cold and the difference is practically meaningless.  With commit
c8efcc25 (workqueue: allow chained queueing during destruction), the
race condition can easily be fixed by using a dedicated workqueue and
destroying it on module unload.

Create and use scsi_wq instead of execute_in_process_context().

* scsi_device->ew is replaced with release_work.  scsi_target->ew is
  replaced with reap_work.

* Both works are initialized with the respective release/reap function
  during device/target init.  scsi_target_reap_usercontext() is moved
  upwards to avoid needing forward declaration.

* scsi_alloc_target() now explicitly flushes the reap_work of the
  found dying target before putting it instead of depending on
  flush_scheduled_work().

For more info on the issues, please read the following thread.

 http://thread.gmane.org/gmane.linux.scsi/62923

Signed-off-by: Tejun Heo <tj@kernel.org>
---
Hello, James.

With commit c8efcc25, workqueue now allows and properly handles
chained queueing during destruction.  ie. destroy_workqueue() won't
whine if a work queues itself or another work on the dying queue and
will make sure all the chained series of works are finished before the
wq is destroyed, which I think is a good idea regardless of ew as
there are other users of such chained works.

With chained flushing, I think this is the simplest way to remove the
above described race condition and deprecation of the ew interface
which, without complicated augmentation, is inherently racy.

Thank you.

 drivers/scsi/scsi.c        |   15 +++++++++++++--
 drivers/scsi/scsi_scan.c   |   26 +++++++++++++-------------
 drivers/scsi/scsi_sysfs.c  |    8 +++++---
 include/scsi/scsi_device.h |    6 ++++--
 4 files changed, 35 insertions(+), 20 deletions(-)

Index: work/drivers/scsi/scsi_scan.c
===================================================================
--- work.orig/drivers/scsi/scsi_scan.c
+++ work/drivers/scsi/scsi_scan.c
@@ -362,6 +362,16 @@ int scsi_is_target_device(const struct d
 }
 EXPORT_SYMBOL(scsi_is_target_device);
 
+static void scsi_target_reap_usercontext(struct work_struct *work)
+{
+	struct scsi_target *starget =
+		container_of(work, struct scsi_target, reap_work);
+
+	transport_remove_device(&starget->dev);
+	device_del(&starget->dev);
+	scsi_target_destroy(starget);
+}
+
 static struct scsi_target *__scsi_find_target(struct device *parent,
 					      int channel, uint id)
 {
@@ -427,6 +437,7 @@ static struct scsi_target *scsi_alloc_ta
 	starget->state = STARGET_CREATED;
 	starget->scsi_level = SCSI_2;
 	starget->max_target_blocked = SCSI_DEFAULT_TARGET_BLOCKED;
+	INIT_WORK(&starget->reap_work, scsi_target_reap_usercontext);
  retry:
 	spin_lock_irqsave(shost->host_lock, flags);
 
@@ -462,21 +473,11 @@ static struct scsi_target *scsi_alloc_ta
 	}
 	/* Unfortunately, we found a dying target; need to
 	 * wait until it's dead before we can get a new one */
+	flush_work(&found_target->reap_work);
 	put_device(&found_target->dev);
-	flush_scheduled_work();
 	goto retry;
 }
 
-static void scsi_target_reap_usercontext(struct work_struct *work)
-{
-	struct scsi_target *starget =
-		container_of(work, struct scsi_target, ew.work);
-
-	transport_remove_device(&starget->dev);
-	device_del(&starget->dev);
-	scsi_target_destroy(starget);
-}
-
 /**
  * scsi_target_reap - check to see if target is in use and destroy if not
  * @starget: target to be checked
@@ -507,8 +508,7 @@ void scsi_target_reap(struct scsi_target
 	if (state == STARGET_CREATED)
 		scsi_target_destroy(starget);
 	else
-		execute_in_process_context(scsi_target_reap_usercontext,
-					   &starget->ew);
+		queue_work(scsi_wq, &starget->reap_work);
 }
 
 /**
Index: work/drivers/scsi/scsi_sysfs.c
===================================================================
--- work.orig/drivers/scsi/scsi_sysfs.c
+++ work/drivers/scsi/scsi_sysfs.c
@@ -300,7 +300,7 @@ static void scsi_device_dev_release_user
 	struct list_head *this, *tmp;
 	unsigned long flags;
 
-	sdev = container_of(work, struct scsi_device, ew.work);
+	sdev = container_of(work, struct scsi_device, release_work);
 
 	parent = sdev->sdev_gendev.parent;
 	starget = to_scsi_target(parent);
@@ -343,8 +343,8 @@ static void scsi_device_dev_release_user
 static void scsi_device_dev_release(struct device *dev)
 {
 	struct scsi_device *sdp = to_scsi_device(dev);
-	execute_in_process_context(scsi_device_dev_release_usercontext,
-				   &sdp->ew);
+
+	queue_work(scsi_wq, &sdp->release_work);
 }
 
 static struct class sdev_class = {
@@ -1069,6 +1069,8 @@ void scsi_sysfs_device_initialize(struct
 	dev_set_name(&sdev->sdev_dev, "%d:%d:%d:%d",
 		     sdev->host->host_no, sdev->channel, sdev->id, sdev->lun);
 	sdev->scsi_level = starget->scsi_level;
+	INIT_WORK(&sdev->release_work, scsi_device_dev_release_usercontext);
+
 	transport_setup_device(&sdev->sdev_gendev);
 	spin_lock_irqsave(shost->host_lock, flags);
 	list_add_tail(&sdev->same_target_siblings, &starget->devices);
Index: work/include/scsi/scsi_device.h
===================================================================
--- work.orig/include/scsi/scsi_device.h
+++ work/include/scsi/scsi_device.h
@@ -168,7 +168,7 @@ struct scsi_device {
 	struct device		sdev_gendev,
 				sdev_dev;
 
-	struct execute_work	ew; /* used to get process context on put */
+	struct work_struct	release_work; /* for process context on put */
 
 	struct scsi_dh_data	*scsi_dh_data;
 	enum scsi_device_state sdev_state;
@@ -258,7 +258,7 @@ struct scsi_target {
 #define SCSI_DEFAULT_TARGET_BLOCKED	3
 
 	char			scsi_level;
-	struct execute_work	ew;
+	struct work_struct	reap_work;
 	enum scsi_target_state	state;
 	void 			*hostdata; /* available to low-level driver */
 	unsigned long		starget_data[0]; /* for the transport */
@@ -276,6 +276,8 @@ static inline struct scsi_target *scsi_t
 #define starget_printk(prefix, starget, fmt, a...)	\
 	dev_printk(prefix, &(starget)->dev, fmt, ##a)
 
+extern struct workqueue_struct *scsi_wq;
+
 extern struct scsi_device *__scsi_add_device(struct Scsi_Host *,
 		uint, uint, uint, void *hostdata);
 extern int scsi_add_device(struct Scsi_Host *host, uint channel,
Index: work/drivers/scsi/scsi.c
===================================================================
--- work.orig/drivers/scsi/scsi.c
+++ work/drivers/scsi/scsi.c
@@ -70,6 +70,11 @@
 #define CREATE_TRACE_POINTS
 #include <trace/events/scsi.h>
 
+/*
+ * Utility multithreaded workqueue for SCSI.
+ */
+struct workqueue_struct *scsi_wq;
+
 static void scsi_done(struct scsi_cmnd *cmd);
 
 /*
@@ -1306,11 +1311,14 @@ MODULE_PARM_DESC(scsi_logging_level, "a
 
 static int __init init_scsi(void)
 {
-	int error;
+	int error = -ENOMEM;
 
+	scsi_wq = alloc_workqueue("scsi", 0, 0);
+	if (!scsi_wq)
+		return error;
 	error = scsi_init_queue();
 	if (error)
-		return error;
+		goto cleanup_wq;
 	error = scsi_init_procfs();
 	if (error)
 		goto cleanup_queue;
@@ -1342,6 +1350,8 @@ cleanup_procfs:
 	scsi_exit_procfs();
 cleanup_queue:
 	scsi_exit_queue();
+cleanup_wq:
+	destroy_workqueue(scsi_wq);
 	printk(KERN_ERR "SCSI subsystem failed to initialize, error = %d\n",
 	       -error);
 	return error;
@@ -1356,6 +1366,7 @@ static void __exit exit_scsi(void)
 	scsi_exit_devinfo();
 	scsi_exit_procfs();
 	scsi_exit_queue();
+	destroy_workqueue(scsi_wq);
 }
 
 subsys_initcall(init_scsi);

Re: [PATCH 1/2] scsi: remove bogus use of struct execute_work in sg

[Tejun Heo]

On Mon, Jan 24, 2011 at 02:31:05PM +0100, Tejun Heo wrote:
> execute_work usage in sg is bogus.  execute_in_process_context() is
> never used and ew is used purely as wrapper around work_struct which
> is superflous.  Use work_struct directly.
> 
> Signed-off-by: Tejun Heo <tj@kernel.org>
> Acked-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
> Acked-by: Douglas Gilbert <dgilbert@interlog.com>
> ---
> This hasn't changed from the previous posting.
> 
>   http://thread.gmane.org/gmane.linux.scsi/62923

James, are you gonna take these two?

Thanks.

-- 
tejun

[PATCH 1/2] scsi: remove bogus use of struct execute_work in sg

[Tejun Heo]

execute_work usage in sg is bogus.  execute_in_process_context() is
never used and ew is used purely as wrapper around work_struct which
is superflous.  Use work_struct directly.

Signed-off-by: Tejun Heo <tj@kernel.org>
---
 drivers/scsi/sg.c |   16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

Index: work/drivers/scsi/sg.c
===================================================================
--- work.orig/drivers/scsi/sg.c
+++ work/drivers/scsi/sg.c
@@ -137,7 +137,7 @@ typedef struct sg_request {	/* SG_MAX_QU
 	volatile char done;	/* 0->before bh, 1->before read, 2->read */
 	struct request *rq;
 	struct bio *bio;
-	struct execute_work ew;
+	struct work_struct work;
 } Sg_request;

 typedef struct sg_fd {		/* holds the state of a file descriptor */
@@ -160,7 +160,7 @@ typedef struct sg_fd {		/* holds the sta
 	char keep_orphan;	/* 0 -> drop orphan (def), 1 -> keep for read() */
 	char mmap_called;	/* 0 -> mmap() never called on this fd */
 	struct kref f_ref;
-	struct execute_work ew;
+	struct work_struct work;
 } Sg_fd;

 typedef struct sg_device { /* holds the state of each scsi generic device */
@@ -1246,7 +1246,7 @@ sg_mmap(struct file *filp, struct vm_are

 static void sg_rq_end_io_usercontext(struct work_struct *work)
 {
-	struct sg_request *srp = container_of(work, struct sg_request, ew.work);
+	struct sg_request *srp = container_of(work, struct sg_request, work);
 	struct sg_fd *sfp = srp->parentfp;

 	sg_finish_rem_req(srp);
@@ -1333,8 +1333,8 @@ static void sg_rq_end_io(struct request
 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
 		kref_put(&sfp->f_ref, sg_remove_sfp);
 	} else {
-		INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
-		schedule_work(&srp->ew.work);
+		INIT_WORK(&srp->work, sg_rq_end_io_usercontext);
+		schedule_work(&srp->work);
 	}
 }

@@ -2086,7 +2086,7 @@ sg_add_sfp(Sg_device * sdp, int dev)

 static void sg_remove_sfp_usercontext(struct work_struct *work)
 {
-	struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
+	struct sg_fd *sfp = container_of(work, struct sg_fd, work);
 	struct sg_device *sdp = sfp->parentdp;

 	/* Cleanup any responses which were never read(). */
@@ -2123,8 +2123,8 @@ static void sg_remove_sfp(struct kref *k
 	write_unlock_irqrestore(&sg_index_lock, iflags);
 	wake_up_interruptible(&sdp->o_excl_wait);

-	INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
-	schedule_work(&sfp->ew.work);
+	INIT_WORK(&sfp->work, sg_remove_sfp_usercontext);
+	schedule_work(&sfp->work);
 }

 static int

Re: [PATCH 1/2] scsi: remove bogus use of struct execute_work in sg

[FUJITA Tomonori]

On Tue, 19 Oct 2010 14:57:36 +0200
Tejun Heo <tj@kernel.org> wrote:

> execute_work usage in sg is bogus.  execute_in_process_context() is
> never used and ew is used purely as wrapper around work_struct which
> is superflous.  Use work_struct directly.
> 
> Signed-off-by: Tejun Heo <tj@kernel.org>
> ---
>  drivers/scsi/sg.c |   16 ++++++++--------
>  1 file changed, 8 insertions(+), 8 deletions(-)

Acked-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>

Re: [PATCH 1/2] scsi: remove bogus use of struct execute_work in sg

[Douglas Gilbert]

On 10-10-19 02:57 PM, Tejun Heo wrote:
> execute_work usage in sg is bogus.  execute_in_process_context() is
> never used and ew is used purely as wrapper around work_struct which
> is superflous.  Use work_struct directly.
>
> Signed-off-by: Tejun Heo<tj@kernel.org>

Acked-by: Douglas Gilbert <dgilbert@interlog.com>