[PATCH 0/3] block: Few fixes for throttle and blk_cleanup_queue()

[PATCH 0/3] block: Few fixes for throttle and blk_cleanup_queue()

[Vivek Goyal]

Hi Jens,

These are few fixes for block throtl code and for blk_cleanup_queue(). These
should be a good candidate for 2.6.39.

Please let me know if you have any concerns.

Thanks
Vivek

Vivek Goyal (3):
  block: Initialize ->queue_lock to internal lock at queue allocation
    time
  loop: No need to initialize ->queue_lock explicitly before calling
    blk_cleanup_queue()
  block: Move blk_throtl_exit() call to blk_cleanup_queue()

 block/blk-core.c       |   23 +++++++++++++++++++++--
 block/blk-settings.c   |    7 -------
 block/blk-sysfs.c      |    2 --
 block/blk-throttle.c   |    6 +++---
 drivers/block/loop.c   |    3 ---
 include/linux/blkdev.h |    2 --
 6 files changed, 24 insertions(+), 19 deletions(-)

-- 
1.7.2.3

[PATCH 1/3] block: Initialize ->queue_lock to internal lock at queue allocation time

[Vivek Goyal]

o There does not seem to be a clear convention whether q->queue_lock is
  initialized or not when blk_cleanup_queue() is called. In the past
  it was not necessary but now blk_throtl_exit() takes up queue lock
  by default and needs queue lock to be available.

  In fact elevator_exit() code also has similar requirement just that
  it is less stringent in the sense that elevator_exit() is called only
  if elevator is initialized.

o Two problems have been noticed because of ambiguity about spin lock
  status.

	- If a driver calls blk_alloc_queue() and then soon calls
	  blk_cleanup_queue() almost immediately, (because some other driver
	  structure allocation failed or some other error happened) then
	  blk_throtl_exit() will run into issues as queue lock is not
	  initialized. Loop driver ran into this issue recently and I noticed
	  error paths in md driver too. Similar error paths should exist in
	  other drivers too.

	- If some driver provided external spin lock and zapped the lock
	  before blk_cleanup_queue(), then it can lead to issues.

o So this patch initializes the default queue lock at queue allocation time.
  block throttling code is one of the users of queue lock and it is
  initialized at the queue allocation time, so it makes sense to
  initialize ->queue_lock also to internal lock. A driver can overide that
  lock later. This will take care of the issue where a driver does not have
  to worry about initializing the queue lock to default before calling
  blk_cleanup_queue()

Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
---
 block/blk-core.c     |   16 +++++++++++++++-
 block/blk-settings.c |    7 -------
 2 files changed, 15 insertions(+), 8 deletions(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index 3cc17e6..bc2b7c5 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -446,6 +446,11 @@ void blk_put_queue(struct request_queue *q)
 	kobject_put(&q->kobj);
 }
 
+/*
+ * Note: If a driver supplied the queue lock, it should not zap that lock
+ * unexpectedly as some queue cleanup components like elevator_exit() and
+ * blk_throtl_exit() need queue lock.
+ */
 void blk_cleanup_queue(struct request_queue *q)
 {
 	/*
@@ -540,6 +545,12 @@ struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 	mutex_init(&q->sysfs_lock);
 	spin_lock_init(&q->__queue_lock);
 
+	/*
+	 * By default initialize queue_lock to internal lock and driver can
+	 * override it later if need be.
+	 */
+	q->queue_lock = &q->__queue_lock;
+
 	return q;
 }
 EXPORT_SYMBOL(blk_alloc_queue_node);
@@ -624,7 +635,10 @@ blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
 	q->unprep_rq_fn		= NULL;
 	q->unplug_fn		= generic_unplug_device;
 	q->queue_flags		= QUEUE_FLAG_DEFAULT;
-	q->queue_lock		= lock;
+
+	/* Override internal queue lock with supplied lock pointer */
+	if (lock)
+		q->queue_lock		= lock;
 
 	/*
 	 * This also sets hw/phys segments, boundary and size
diff --git a/block/blk-settings.c b/block/blk-settings.c
index 36c8c1f..df649fa 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -176,13 +176,6 @@ void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
 	blk_queue_max_hw_sectors(q, BLK_SAFE_MAX_SECTORS);
 
 	/*
-	 * If the caller didn't supply a lock, fall back to our embedded
-	 * per-queue locks
-	 */
-	if (!q->queue_lock)
-		q->queue_lock = &q->__queue_lock;
-
-	/*
 	 * by default assume old behaviour and bounce for any highmem page
 	 */
 	blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
-- 
1.7.2.3

[PATCH 2/3] loop: No need to initialize ->queue_lock explicitly before calling blk_cleanup_queue()

[Vivek Goyal]

o Now we initialize ->queue_lock at queue allocation time so driver does
  not have to worry about initializing it before calling blk_cleanup_queue().
---
 drivers/block/loop.c |    3 ---
 1 files changed, 0 insertions(+), 3 deletions(-)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 49e6a54..44e18c0 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -1641,9 +1641,6 @@ out:
 
 static void loop_free(struct loop_device *lo)
 {
-	if (!lo->lo_queue->queue_lock)
-		lo->lo_queue->queue_lock = &lo->lo_queue->__queue_lock;
-
 	blk_cleanup_queue(lo->lo_queue);
 	put_disk(lo->lo_disk);
 	list_del(&lo->lo_list);
-- 
1.7.2.3

[PATCH 3/3] block: Move blk_throtl_exit() call to blk_cleanup_queue()

[Vivek Goyal]

o Move blk_throtl_exit() in blk_cleanup_queue() as blk_throtl_exit() is
  written in such a way that it needs queue lock. In blk_release_queue()
  there is no gurantee that ->queue_lock is still around.

o Initially blk_throtl_exit() was in blk_cleanup_queue() but Ingo reported
  one problem.

  https://lkml.org/lkml/2010/10/23/86

  And a quick fix moved blk_throtl_exit() to blk_release_queue().

        commit 7ad58c028652753814054f4e3ac58f925e7343f4
        Author: Jens Axboe <jaxboe@fusionio.com>
        Date:   Sat Oct 23 20:40:26 2010 +0200

        block: fix use-after-free bug in blk throttle code

o This patch reverts above change and does not try to shutdown the
  throtl work in blk_sync_queue(). By avoiding call to
  throtl_shutdown_timer_wq() from blk_sync_queue(), we should also avoid
  the problem reported by Ingo.

o blk_sync_queue() seems to be used only by md driver and it seems to be
  using it to make sure q->unplug_fn is not called as md registers its
  own unplug functions and it is about to free up the data structures
  used by unplug_fn(). Block throttle does not call back into unplug_fn()
  or into md. So there is no need to cancel blk throttle work.

  In fact I think cancelling block throttle work is bad because it might
  happen that some bios are throttled and scheduled to be dispatched later
  with the help of pending work and if work is cancelled, these bios might
  never be dispatched.

  Block layer also uses blk_sync_queue() during blk_cleanup_queue() and
  blk_release_queue() time. That should be safe as we are also calling
  blk_throtl_exit() which should make sure all the throttling related
  data structures are cleaned up.

Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
---
 block/blk-core.c       |    7 ++++++-
 block/blk-sysfs.c      |    2 --
 block/blk-throttle.c   |    6 +++---
 include/linux/blkdev.h |    2 --
 4 files changed, 9 insertions(+), 8 deletions(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index bc2b7c5..accff29 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -380,13 +380,16 @@ EXPORT_SYMBOL(blk_stop_queue);
  *     that its ->make_request_fn will not re-add plugging prior to calling
  *     this function.
  *
+ *     This function does not cancel any asynchronous activity arising
+ *     out of elevator or throttling code. That would require elevaotor_exit()
+ *     and blk_throtl_exit() to be called with queue lock initialized.
+ *
  */
 void blk_sync_queue(struct request_queue *q)
 {
 	del_timer_sync(&q->unplug_timer);
 	del_timer_sync(&q->timeout);
 	cancel_work_sync(&q->unplug_work);
-	throtl_shutdown_timer_wq(q);
 }
 EXPORT_SYMBOL(blk_sync_queue);
 
@@ -469,6 +472,8 @@ void blk_cleanup_queue(struct request_queue *q)
 	if (q->elevator)
 		elevator_exit(q->elevator);
 
+	blk_throtl_exit(q);
+
 	blk_put_queue(q);
 }
 EXPORT_SYMBOL(blk_cleanup_queue);
diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 41fb691..261c75c 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -471,8 +471,6 @@ static void blk_release_queue(struct kobject *kobj)
 
 	blk_sync_queue(q);
 
-	blk_throtl_exit(q);
-
 	if (rl->rq_pool)
 		mempool_destroy(rl->rq_pool);
 
diff --git a/block/blk-throttle.c b/block/blk-throttle.c
index a89043a..c0f6237 100644
--- a/block/blk-throttle.c
+++ b/block/blk-throttle.c
@@ -965,7 +965,7 @@ static void throtl_update_blkio_group_write_iops(void *key,
 	throtl_schedule_delayed_work(td->queue, 0);
 }
 
-void throtl_shutdown_timer_wq(struct request_queue *q)
+static void throtl_shutdown_wq(struct request_queue *q)
 {
 	struct throtl_data *td = q->td;
 
@@ -1099,7 +1099,7 @@ void blk_throtl_exit(struct request_queue *q)
 
 	BUG_ON(!td);
 
-	throtl_shutdown_timer_wq(q);
+	throtl_shutdown_wq(q);
 
 	spin_lock_irq(q->queue_lock);
 	throtl_release_tgs(td);
@@ -1129,7 +1129,7 @@ void blk_throtl_exit(struct request_queue *q)
 	 * update limits through cgroup and another work got queued, cancel
 	 * it.
 	 */
-	throtl_shutdown_timer_wq(q);
+	throtl_shutdown_wq(q);
 	throtl_td_free(td);
 }
 
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index e3ee74f..23fb925 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -1144,7 +1144,6 @@ extern int blk_throtl_init(struct request_queue *q);
 extern void blk_throtl_exit(struct request_queue *q);
 extern int blk_throtl_bio(struct request_queue *q, struct bio **bio);
 extern void throtl_schedule_delayed_work(struct request_queue *q, unsigned long delay);
-extern void throtl_shutdown_timer_wq(struct request_queue *q);
 #else /* CONFIG_BLK_DEV_THROTTLING */
 static inline int blk_throtl_bio(struct request_queue *q, struct bio **bio)
 {
@@ -1154,7 +1153,6 @@ static inline int blk_throtl_bio(struct request_queue *q, struct bio **bio)
 static inline int blk_throtl_init(struct request_queue *q) { return 0; }
 static inline int blk_throtl_exit(struct request_queue *q) { return 0; }
 static inline void throtl_schedule_delayed_work(struct request_queue *q, unsigned long delay) {}
-static inline void throtl_shutdown_timer_wq(struct request_queue *q) {}
 #endif /* CONFIG_BLK_DEV_THROTTLING */
 
 #define MODULE_ALIAS_BLOCKDEV(major,minor) \
-- 
1.7.2.3

Re: [PATCH 0/3] block: Few fixes for throttle and blk_cleanup_queue()

[Jens Axboe]

On 2011-02-28 14:25, Vivek Goyal wrote:
> Hi Jens,
> 
> These are few fixes for block throtl code and for blk_cleanup_queue(). These
> should be a good candidate for 2.6.39.

They look good, thanks.

A small please to revisit your changelog style. Lets drop the bullet
points, and it's best to keep at eg 72 chars so that the output is
readable in git log easily. I sometimes change these for you, would be
nice if they followed normal style though.


-- 
Jens Axboe

Re: [PATCH 0/3] block: Few fixes for throttle and blk_cleanup_queue()

[Vivek Goyal]

On Wed, Mar 02, 2011 at 06:57:37PM -0500, Jens Axboe wrote:
> On 2011-02-28 14:25, Vivek Goyal wrote:
> > Hi Jens,
> > 
> > These are few fixes for block throtl code and for blk_cleanup_queue(). These
> > should be a good candidate for 2.6.39.
> 
> They look good, thanks.
> 
> A small please to revisit your changelog style. Lets drop the bullet
> points, and it's best to keep at eg 72 chars so that the output is
> readable in git log easily. I sometimes change these for you, would be
> nice if they followed normal style though.

Sure. I will drop bullet style and also limit line width to 72 characters
in future postings.

Do you want me to repost this patch series with changelogs corrected?

Thanks
Vivek

Re: [PATCH 0/3] block: Few fixes for throttle and blk_cleanup_queue()

[Jens Axboe]

On 2011-03-02 19:09, Vivek Goyal wrote:
> On Wed, Mar 02, 2011 at 06:57:37PM -0500, Jens Axboe wrote:
>> On 2011-02-28 14:25, Vivek Goyal wrote:
>>> Hi Jens,
>>>
>>> These are few fixes for block throtl code and for blk_cleanup_queue(). These
>>> should be a good candidate for 2.6.39.
>>
>> They look good, thanks.
>>
>> A small please to revisit your changelog style. Lets drop the bullet
>> points, and it's best to keep at eg 72 chars so that the output is
>> readable in git log easily. I sometimes change these for you, would be
>> nice if they followed normal style though.
> 
> Sure. I will drop bullet style and also limit line width to 72 characters
> in future postings.

Thanks!

> Do you want me to repost this patch series with changelogs corrected?

No, I just fixed these up. It's in for-2.6.39 now.

-- 
Jens Axboe

[PATCH 0/3] [RFC] block: Enforce that ->queue_lock is initialized during call to blk_cleanup_queue()

[Vivek Goyal]

Hi All,

Currently it is not clear what's the status of ->queue_lock when
blk_cleanup_queue() is called. Now block throttle code (blk_throtl_exit())
requires queue lock to be initialized as it takes the spin lock, so we need
some kind of clear convention that when is it safe to assume that queue lock
is initiliazed and blk_throtle_exit() can be called safely.

We recently ran into issues with loop driver which was trying to free
up queue almost immediately after block_alloc_queue() due to some internal
errors. Also NeilBrown complained that in current form, md provides its
own spinlock and zaps that before blk_cleanup_queue() call and that will
have issues with throttling code. Neil has not fixed the issue with
md driver and queued up a patch in this tree.

So there is a need to make sure blk_throtl_exit() can be called safely
and to also catch any errors if some drivers are not doing so. This
patch series puts a WARN_ON(!q->queue_lock) in blk_cleanup_queue() and
moves the queue lock initialization in queue allocation code. Also it
move blk_throtl_exit() from release_queue() to cleanup_queue().

These pathes are only boot tested on one machine. Before I do more
extensive testing, I wanted to throw it out there and figure out
if it makes sense or not.

Thanks
Vivek

Vivek Goyal (3):
  block: Initialize ->queue_lock to internal lock at queue allocation
    time
  loop: No need to initialize ->queue_lock explicitly before calling
    blk_cleanup_queue()
  block: Move blk_throtl_exit() call to blk_cleanup_queue()

 block/blk-core.c       |   26 ++++++++++++++++++++++++--
 block/blk-settings.c   |    7 -------
 block/blk-sysfs.c      |    2 --
 block/blk-throttle.c   |    2 +-
 drivers/block/loop.c   |    3 ---
 include/linux/blkdev.h |    2 --
 6 files changed, 25 insertions(+), 17 deletions(-)

[PATCH 1/3] block: Initialize ->queue_lock to internal lock at queue allocation time

[Vivek Goyal]

o There does not seem to be a clear convention whether q->queue_lock is
  initialized or not when blk_cleanup_queue() is called. In the past
  it was not necessary but now blk_throtl_exit() takes up queue lock
  by default and needs queue lock to be available.

  In fact elevator_exit() code also has similar requirement just that
  it is less stringent in the sense that elevator_exit() is called only
  if elevator is initialized.

o Two problems have been noticed because of ambiguity about spin lock
  status.

	- If a driver calls blk_alloc_queue() and then soon calls
	  blk_cleanup_queue() almost immediately, (because some other driver
	  structure allocation failed or some other error happened) then
	  blk_throtl_exit() will run into issues as queue lock is not
	  initialized. Loop driver ran into this issue recently and I noticed
	  error paths in md driver too. Similar error paths should exist in
	  other drivers too.

	- If some driver provided external spin lock and zapped the lock
	  before blk_cleanup_queue(), then it can lead to issues.

o So this patch does two things.

	- Initialize the default queue lock at queue allocation time.
	  block throttling code is one of the users of queue lock and
	  it is initialized at the queue allocation time, so it makes
	  sense to initialize ->queue_lock also to internal lock. A
	  driver can overide that lock later. This will take care of
	  first issue where a driver does not have to worry about
	  initializing the queue lock to default before calling
	  blk_cleanup_queue()

	- Put a WARN_ON() in blk_cleanup_queue() to make sure ->queue_lock
	  is initialized and it will catch the cases if there is a bad
	  driver which provides an external queue lock but zaps it
	  unexpectedly. So this WARN_ON() will catch that and we can
	  fix the driver.

	  Recently NeilBrown noted that md is doing something similar and
	  he has queued a fix in this tree. This patch relies on that
	  patch otherwise it might WARN and then crash in blk_throtl_exit().

Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
---
 block/blk-core.c     |   19 ++++++++++++++++++-
 block/blk-settings.c |    7 -------
 2 files changed, 18 insertions(+), 8 deletions(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index 2f4002f..9a8e256 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -459,6 +459,14 @@ void blk_put_queue(struct request_queue *q)
 void blk_cleanup_queue(struct request_queue *q)
 {
 	/*
+	 * If a driver supplied the queue lock, it should not zap that
+	 * unexpectedly as some queue cleanup components like
+	 * elevator_exit() and blk_throtl_exit() need queue lock. Hence
+	 * warn if queue lock is NULL
+	 */
+	WARN_ON(!q->queue_lock);
+
+	/*
 	 * We know we have process context here, so we can be a little
 	 * cautious and ensure that pending block actions on this device
 	 * are done before moving on. Going into this function, we should
@@ -548,6 +556,12 @@ struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 	mutex_init(&q->sysfs_lock);
 	spin_lock_init(&q->__queue_lock);
 
+	/*
+	 * By default initialize queue_lock to internal lock and driver can
+	 * override it later if need be.
+	 */
+	q->queue_lock = &q->__queue_lock;
+
 	return q;
 }
 EXPORT_SYMBOL(blk_alloc_queue_node);
@@ -632,7 +646,10 @@ blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
 	q->unprep_rq_fn		= NULL;
 	q->unplug_fn		= generic_unplug_device;
 	q->queue_flags		= QUEUE_FLAG_DEFAULT;
-	q->queue_lock		= lock;
+
+	/* Override internal queue lock with supplied lock pointer */
+	if (lock)
+		q->queue_lock		= lock;
 
 	/*
 	 * This also sets hw/phys segments, boundary and size
diff --git a/block/blk-settings.c b/block/blk-settings.c
index 36c8c1f..df649fa 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -176,13 +176,6 @@ void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
 	blk_queue_max_hw_sectors(q, BLK_SAFE_MAX_SECTORS);
 
 	/*
-	 * If the caller didn't supply a lock, fall back to our embedded
-	 * per-queue locks
-	 */
-	if (!q->queue_lock)
-		q->queue_lock = &q->__queue_lock;
-
-	/*
 	 * by default assume old behaviour and bounce for any highmem page
 	 */
 	blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
-- 
1.7.1