Revert of the IO stat fix
Hi Linus,
The fix for cross-partition merges screwing up disk stats turns out to be problematic on various levels. Lets revert this one so we have time to come up with a proper solution for this.
git://git.kernel.dk/linux-2.6-block.git for-linus
Jens Axboe (1):
Revert "block: fix accounting bug on cross partition merges"
block/blk-core.c | 24 ++++++++---------------- block/blk-merge.c | 2 +- block/blk.h | 4 ++++ block/genhd.c | 14 -------------- fs/partitions/check.c | 12 ------------ include/linux/blkdev.h | 1 - include/linux/elevator.h | 2 -- include/linux/genhd.h | 1 - 8 files changed, 13 insertions(+), 47 deletions(-)
Signature · 1 line
Jens Axboe
Quoted 4 lines · Jens Axboe
The fix for cross-partition merges screwing up disk stats turns out to be problematic on various levels. Lets revert this one so we have time to come up with a proper solution for this.
Hmm.. I think the reverted patch looks like it really is the right thing to do, so I hate reverting it this early. What were the problems with it?
Btw, one thing that seems to be missing in the original commit (which is not necessarily the reason for the trouble, of course), is that elv_rq_merge_ok() seems to not check the partition. As far as I can tell, we should have a
if (req->part != bio->bi_bdev->bd_part)
return 0;there, no? And you should _not_ set rq->part in "drive_stat_acct()", you should set it from bio->bi_bdev->bd_part when you create the request.
(And if it is NULL, just don't do partition accounting at all)
Hmm? What am I missing? What were the bugs?
Linus
Quoted 18 lines · Linus Torvalds
Quoted 4 lines · Jens Axboe
The fix for cross-partition merges screwing up disk stats turns out to be problematic on various levels. Lets revert this one so we have time to come up with a proper solution for this.
Hmm.. I think the reverted patch looks like it really is the right thing to do, so I hate reverting it this early. What were the problems with it?
Btw, one thing that seems to be missing in the original commit (which is not necessarily the reason for the trouble, of course), is that
elv_rq_merge_ok()seems to not check the partition. As far as I can tell, we should have aif (req->part != bio->bi_bdev->bd_part) return 0;
Quoted 7 lines
there, no? And you should _not_ set
rq->partin "drive_stat_acct()", you should set it frombio->bi_bdev->bd_partwhen you create the request.(And if it is NULL, just don't do partition accounting at all)
Hmm? What am I missing? What were the bugs?
The patch itself is sound, the problems are around the area of it not really liking non-elevator devices with the elv_quiesce_start/end() parts. I had the below patch for that, but then I could not decide whether we were fully safe on queue free after talking to Vivek about it.
So that shows up as an oops on removal of mspro for instance, and loop throws a fit as well. Coverage of the bug is too large just to let side idle for a few days.
And since I'll be travelling the next few days, I would rather just revert this one and I have time in-flight to fix it for real.
OK?
Signature · 1 line
Jens Axboe
Quoted 25 lines · Jens Axboe
Quoted 24 lines · Linus Torvalds
Quoted 4 lines · Jens Axboe
The fix for cross-partition merges screwing up disk stats turns out to be problematic on various levels. Lets revert this one so we have time to come up with a proper solution for this.
Hmm.. I think the reverted patch looks like it really is the right thing to do, so I hate reverting it this early. What were the problems with it?
Btw, one thing that seems to be missing in the original commit (which is not necessarily the reason for the trouble, of course), is that
elv_rq_merge_ok()seems to not check the partition. As far as I can tell, we should have aif (req->part != bio->bi_bdev->bd_part) return 0;there, no? And you should _not_ set
rq->partin "drive_stat_acct()", you should set it frombio->bi_bdev->bd_partwhen you create the request.(And if it is NULL, just don't do partition accounting at all)
Quoted 7 lines
Quoted 1 line
Hmm? What am I missing? What were the bugs?
The patch itself is sound, the problems are around the area of it not really liking non-elevator devices with the elv_quiesce_start/end() parts. I had the below patch for that, but then I could not decide whether we were fully safe on queue free after talking to Vivek about it.
Forgot to include it, here it is. I'll be offline from now and 1-2 days forward.
Quoted 1 line
From
96059cec039b666c26d300c2132e24bfd6edacdcMon Sep 17 00:00:00 2001
From: Jens Axboe <[email protected]>
Date: Sun, 24 Oct 2010 08:46:41 +0200
Subject: [PATCH] block: fix partition reload bug with non-elevator devices
The partition reload code was changed to quiesce the block queue so that partition IO stats could safely hold a reference to the partition table. elv_quiesce_{start,end} do not properly work on non-elevator devices. Improve the helper functions so that they don't care, this way we can use the generic interface on partition reload without having to check for queue structures or types.
--- block/elevator.c | 31 +++++++++++++++++++++++-------- block/genhd.c | 10 +--------- fs/partitions/check.c | 5 ----- include/linux/elevator.h | 2 ++ 4 files changed, 26 insertions(+), 22 deletions(-)
block/elevator.c +23 -8
@@ -590,11 +590,8 @@ void elv_drain_elevator(struct request_queue *q)/* * Call with queue lock held, interrupts disabled */ -void elv_quiesce_start (struct request_queue *q) +void __elv_quiesce_start (struct request_queue *q) { -if (!q->elevator) -return ; -queue_flag_set (QUEUE_FLAG_ELVSWITCH , q);/*
@@ -610,11 +607,31 @@ void elv_quiesce_start(struct request_queue *q) } } -void elv_quiesce_end (struct request_queue *q) +void elv_quiesce_start (struct request_queue *q) +{ +if (q->elevator) { +spin_lock_irq (q->queue_lock); +__elv_quiesce_start (q); +spin_unlock_irq (q->queue_lock); + } +} + +void __elv_quiesce_end (struct request_queue *q) {queue_flag_clear (QUEUE_FLAG_ELVSWITCH , q); } +void elv_quiesce_end (struct request_queue *q) +{ +if (q->elevator) { +unsigned long flags; + +spin_lock_irqsave (q->queue_lock, flags); +__elv_quiesce_end (q); +spin_unlock_irqrestore (q->queue_lock, flags); + } +} +void elv_insert (struct request_queue *q,struct request *rq,int where) {int unplug_it =1 ;
@@ -969,7 +986,7 @@ static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)* Turn on BYPASS and drain all requests w/ elevator private data */ spin_lock_irq (q->queue_lock); -elv_quiesce_start (q); +__elv_quiesce_start (q);/* * Remember old elevator.
@@ -995,9 +1012,7 @@ static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)* finally exit old elevator and turn off BYPASS. */ elevator_exit (old_elevator); -spin_lock_irq (q->queue_lock);elv_quiesce_end (q); -spin_unlock_irq (q->queue_lock);blk_add_trace_msg (q,"elv switch: %s" , e->elevator_type->elevator_name);
block/genhd.c +1 -9
@@ -930,14 +930,9 @@ static void disk_free_ptbl_rcu_cb(struct rcu_head *head)struct disk_part_tbl *ptbl = container_of(head, struct disk_part_tbl, rcu_head);struct gendisk *disk = ptbl->disk; -struct request_queue *q = disk->queue; -unsigned long flags;kfree (ptbl); - -spin_lock_irqsave (q->queue_lock, flags); -elv_quiesce_end (q); -spin_unlock_irqrestore (q->queue_lock, flags); +elv_quiesce_end (disk->queue); }/**
@@ -962,10 +957,7 @@ static void disk_replace_part_tbl(struct gendisk *disk,if (old_ptbl) {rcu_assign_pointer (old_ptbl->last_lookup,NULL ); -spin_lock_irq (q->queue_lock);elv_quiesce_start (q); -spin_unlock_irq (q->queue_lock); -call_rcu (&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb); } }
fs/partitions/check.c +0 -5
@@ -367,16 +367,13 @@ static void delete_partition_rcu_cb(struct rcu_head *head) struct hd_struct *part = container_of(head, struct hd_struct, rcu_head);struct gendisk *disk =part_to_disk (part);struct request_queue *q = disk->queue; -unsigned long flags; part->start_sect =0 ; part->nr_sects =0 ;part_stat_set_all (part,0 );put_device (part_to_dev (part)); -spin_lock_irqsave (q->queue_lock, flags);elv_quiesce_end (q); -spin_unlock_irqrestore (q->queue_lock, flags); }void delete_partition (struct gendisk *disk,int partno)
@@ -398,9 +395,7 @@ void delete_partition(struct gendisk *disk, int partno)kobject_put (part->holder_dir);device_del (part_to_dev (part)); -spin_lock_irq (q->queue_lock);elv_quiesce_start (q); -spin_unlock_irq (q->queue_lock);call_rcu (&part->rcu_head, delete_partition_rcu_cb); }
include/linux/elevator.h +2 -0
@@ -122,7 +122,9 @@ extern void elv_completed_request(struct request_queue *, struct request *);extern int elv_set_request (struct request_queue *,struct request *,gfp_t );extern void elv_put_request (struct request_queue *,struct request *);extern void elv_drain_elevator (struct request_queue *); +extern void __elv_quiesce_start (struct request_queue *);extern void elv_quiesce_start (struct request_queue *); +extern void __elv_quiesce_end (struct request_queue *);extern void elv_quiesce_end (struct request_queue *);/*
Signature · 5 lines
1.7.3 -- Jens Axboe
Quoted 9 lines · Jens Axboe
Quoted 6 lines
The patch itself is sound, the problems are around the area of it not really liking non-elevator devices with the elv_quiesce_start/end() parts. I had the below patch for that, but then I could not decide whether we were fully safe on queue free after talking to Vivek about it.
Forgot to include it, here it is. I'll be offline from now and 1-2 days forward.
Hmm. I actually like this patch more than the revert. Except that I think the __elv_quiesce_start/end() functions should be static, and not exposed in the header file. There don't seem to be any other files that would want to use them anyway, and the whole change looks very sane since all the outside users used to have to take the lock explicitly (and thus moving it inside the elv_quiesce_start/end functions seems to be a nice cleanup).
So if this patch is supposed to fix things and doesn't have any *known* problems, but you're just not entirely sure about it and we're talking just a few days of you being gone, I'd almost rather take the risk. If worst comes to worst, I can revert both this patch and the original one. We _are_ in the middle of the merge window after all, so I'm much more open to risks than I would be later in the -rc series..
Linus
Quoted 26 lines · Jens Axboe
Quoted 25 lines · Jens Axboe
Quoted 24 lines · Linus Torvalds
Quoted 4 lines · Jens Axboe
The fix for cross-partition merges screwing up disk stats turns out to be problematic on various levels. Lets revert this one so we have time to come up with a proper solution for this.
Hmm.. I think the reverted patch looks like it really is the right thing to do, so I hate reverting it this early. What were the problems with it?
Btw, one thing that seems to be missing in the original commit (which is not necessarily the reason for the trouble, of course), is that
elv_rq_merge_ok()seems to not check the partition. As far as I can tell, we should have aif (req->part != bio->bi_bdev->bd_part) return 0;there, no? And you should _not_ set
rq->partin "drive_stat_acct()", you should set it frombio->bi_bdev->bd_partwhen you create the request.(And if it is NULL, just don't do partition accounting at all)
Quoted 7 lines
Quoted 7 lines
Quoted 1 line
Hmm? What am I missing? What were the bugs?
The patch itself is sound, the problems are around the area of it not really liking non-elevator devices with the elv_quiesce_start/end() parts. I had the below patch for that, but then I could not decide whether we were fully safe on queue free after talking to Vivek about it.
Hi Jens,
What's the status of this ? Has there been any progress on the issue ?
Thanks,
Jerome
Quoted 172 lines
Forgot to include it, here it is. I'll be offline from now and 1-2 days forward.
From
96059cec039b666c26d300c2132e24bfd6edacdcMon Sep 17 00:00:00 2001
From: Jens Axboe <[email protected]>
Date: Sun, 24 Oct 2010 08:46:41 +0200
Subject: [PATCH] block: fix partition reload bug with non-elevator devicesThe partition reload code was changed to quiesce the block queue so that partition IO stats could safely hold a reference to the partition table. elv_quiesce_{start,end} do not properly work on non-elevator devices. Improve the helper functions so that they don't care, this way we can use the generic interface on partition reload without having to check for queue structures or types.
Reported-by Eric Dumazet <[email protected]>Signed-off-by Jens Axboe <[email protected]>--- block/elevator.c | 31 +++++++++++++++++++++++-------- block/genhd.c | 10 +--------- fs/partitions/check.c | 5 ----- include/linux/elevator.h | 2 ++ 4 files changed, 26 insertions(+), 22 deletions(-)block/elevator.c +23
-8@@ -590,11 +590,8 @@ void elv_drain_elevator(struct request_queue *q)/* * Call with queue lock held, interrupts disabled */ -void elv_quiesce_start (struct request_queue *q) +void __elv_quiesce_start (struct request_queue *q) { -if (!q->elevator) -return ; -queue_flag_set (QUEUE_FLAG_ELVSWITCH , q);/* @@ -610,11 +607,31 @@ void elv_quiesce_start(struct request_queue *q) } } -void elv_quiesce_end (struct request_queue *q) +void elv_quiesce_start (struct request_queue *q) +{ +if (q->elevator) { +spin_lock_irq (q->queue_lock); +__elv_quiesce_start (q); +spin_unlock_irq (q->queue_lock); + } +} + +void __elv_quiesce_end (struct request_queue *q) {queue_flag_clear (QUEUE_FLAG_ELVSWITCH , q); } +void elv_quiesce_end (struct request_queue *q) +{ +if (q->elevator) { +unsigned long flags; + +spin_lock_irqsave (q->queue_lock, flags); +__elv_quiesce_end (q); +spin_unlock_irqrestore (q->queue_lock, flags); + } +} +void elv_insert (struct request_queue *q,struct request *rq,int where) {int unplug_it =1 ;@@ -969,7 +986,7 @@ static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)* Turn on BYPASS and drain all requests w/ elevator private data */ spin_lock_irq (q->queue_lock); -elv_quiesce_start (q); +__elv_quiesce_start (q);/* * Remember old elevator. @@ -995,9 +1012,7 @@ static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)* finally exit old elevator and turn off BYPASS. */ elevator_exit (old_elevator); -spin_lock_irq (q->queue_lock);elv_quiesce_end (q); -spin_unlock_irq (q->queue_lock);blk_add_trace_msg (q,"elv switch: %s" , e->elevator_type->elevator_name);block/genhd.c +1
-9@@ -930,14 +930,9 @@ static void disk_free_ptbl_rcu_cb(struct rcu_head *head)struct disk_part_tbl *ptbl = container_of(head, struct disk_part_tbl, rcu_head);struct gendisk *disk = ptbl->disk; -struct request_queue *q = disk->queue; -unsigned long flags;kfree (ptbl); - -spin_lock_irqsave (q->queue_lock, flags); -elv_quiesce_end (q); -spin_unlock_irqrestore (q->queue_lock, flags); +elv_quiesce_end (disk->queue); }/** @@ -962,10 +957,7 @@ static void disk_replace_part_tbl(struct gendisk *disk,if (old_ptbl) {rcu_assign_pointer (old_ptbl->last_lookup,NULL ); -spin_lock_irq (q->queue_lock);elv_quiesce_start (q); -spin_unlock_irq (q->queue_lock); -call_rcu (&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb); } }fs/partitions/check.c +0
-5@@ -367,16 +367,13 @@ static void delete_partition_rcu_cb(struct rcu_head *head) struct hd_struct *part = container_of(head, struct hd_struct, rcu_head);struct gendisk *disk =part_to_disk (part);struct request_queue *q = disk->queue; -unsigned long flags; part->start_sect =0 ; part->nr_sects =0 ;part_stat_set_all (part,0 );put_device (part_to_dev (part)); -spin_lock_irqsave (q->queue_lock, flags);elv_quiesce_end (q); -spin_unlock_irqrestore (q->queue_lock, flags); }void delete_partition (struct gendisk *disk,int partno)@@ -398,9 +395,7 @@ void delete_partition(struct gendisk *disk, int partno)kobject_put (part->holder_dir);device_del (part_to_dev (part)); -spin_lock_irq (q->queue_lock);elv_quiesce_start (q); -spin_unlock_irq (q->queue_lock);call_rcu (&part->rcu_head, delete_partition_rcu_cb); }include/linux/elevator.h +2
-0@@ -122,7 +122,9 @@ extern void elv_completed_request(struct request_queue *, struct request *);extern int elv_set_request (struct request_queue *,struct request *,gfp_t );extern void elv_put_request (struct request_queue *,struct request *);extern void elv_drain_elevator (struct request_queue *); +extern void __elv_quiesce_start (struct request_queue *);extern void elv_quiesce_start (struct request_queue *); +extern void __elv_quiesce_end (struct request_queue *);extern void elv_quiesce_end (struct request_queue *);/*
Quoted 7 lines · Jens Axboe
block/genhd.c +0
-0@@ -930,14 +930,9 @@ static void disk_free_ptbl_rcu_cb(struct rcu_head *head)struct disk_part_tbl *ptbl = container_of(head, struct disk_part_tbl, rcu_head);
block/genhd.c static void disk_free_ptbl_rcu_cb(struct rcu_head *head) · line 935
Quoted 10 lines
block/genhd.c +1
-6struct gendisk *disk = ptbl->disk; -struct request_queue *q = disk->queue; -unsigned long flags;kfree (ptbl); - -spin_lock_irqsave (q->queue_lock, flags); -elv_quiesce_end (q); -spin_unlock_irqrestore (q->queue_lock, flags); +elv_quiesce_end (disk->queue);
Here, the queue may be already gone. We should check it is still here and alive:
static void disk_free_ptbl_rcu_cb(struct rcu_head *head)
{
struct disk_part_tbl *ptbl =
container_of(head, struct disk_part_tbl, rcu_head);
struct gendisk *disk = ptbl->disk;
struct request_queue *q = disk->queue;
kfree(ptbl);
if (q && !test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))
elv_quiesce_end(q);
}block/genhd.c static void disk_replace_part_tbl(struct gendisk *disk, · line 960
Quoted 9 lines
block/genhd.c +0
-1}/** @@ -962,10 +957,7 @@ static void disk_replace_part_tbl(struct gendisk *disk,if (old_ptbl) {rcu_assign_pointer (old_ptbl->last_lookup,NULL ); -spin_lock_irq (q->queue_lock);elv_quiesce_start (q);
Same as above.
I'm not sure that is enough, but on my test machine these changes fix the crash at USB key device removal.
Regards,
Jerome
Quoted 5 lines
block/genhd.c +0
-2-spin_unlock_irq (q->queue_lock); -call_rcu (&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb); } }
Quoted 33 lines · Jens Axboe
Quoted 18 lines · Linus Torvalds
Quoted 4 lines · Jens Axboe
The fix for cross-partition merges screwing up disk stats turns out to be problematic on various levels. Lets revert this one so we have time to come up with a proper solution for this.
Hmm.. I think the reverted patch looks like it really is the right thing to do, so I hate reverting it this early. What were the problems with it?
Btw, one thing that seems to be missing in the original commit (which is not necessarily the reason for the trouble, of course), is that
elv_rq_merge_ok()seems to not check the partition. As far as I can tell, we should have aif (req->part != bio->bi_bdev->bd_part) return 0;Quoted 7 lines
there, no? And you should _not_ set
rq->partin "drive_stat_acct()", you should set it frombio->bi_bdev->bd_partwhen you create the request.(And if it is NULL, just don't do partition accounting at all)
Hmm? What am I missing? What were the bugs?
The patch itself is sound, the problems are around the area of it not really liking non-elevator devices with the elv_quiesce_start/end() parts. I had the below patch for that, but then I could not decide whether we were fully safe on queue free after talking to Vivek about it.
Quoted 3 lines
So that shows up as an oops on removal of mspro for instance, and loop throws a fit as well. Coverage of the bug is too large just to let side idle for a few days.
Btw I had that bug reproduced 100% for all removeable devices I have. (xD, ms, usb).
Attached logs for each.
Best regards,
Maxim Levitsky
Quoted 3 lines · Maxim Levitsky
Btw I had that bug reproduced 100% for all removeable devices I have. (xD, ms, usb).
Does Jens' suggested patch (the non-revert one) fix your oops? Or only the revert?
IOW, I would want to know whether I could try just merging the fix instead of backing out the patch that looks like a real improvement..
Linus
Quoted 7 lines · Linus Torvalds
Quoted 3 lines · Maxim Levitsky
Btw I had that bug reproduced 100% for all removeable devices I have. (xD, ms, usb).
Does Jens' suggested patch (the non-revert one) fix your oops? Or only the revert?
The suggested patch doesn't fix the oops.
Testing revert now.
Best regards,
Maxim Levitsky
Quoted 5 lines
IOW, I would want to know whether I could try just merging the fix instead of backing out the patch that looks like a real improvement..
Linus
Quoted 10 lines · Maxim Levitsky
Quoted 7 lines · Linus Torvalds
Quoted 3 lines · Maxim Levitsky
Btw I had that bug reproduced 100% for all removeable devices I have. (xD, ms, usb).
Does Jens' suggested patch (the non-revert one) fix your oops? Or only the revert?
The suggested patch doesn't fix the oops.
Testing revert now.
Yes, revert works.
Tested xD, MS, USB.
Best regards,
Maxim Levitsky
Quoted 8 lines
Quoted 6 lines
IOW, I would want to know whether I could try just merging the fix instead of backing out the patch that looks like a real improvement..
Linus
Quoted 10 lines · Linus Torvalds
Quoted 4 lines · Jens Axboe
The fix for cross-partition merges screwing up disk stats turns out to be problematic on various levels. Lets revert this one so we have time to come up with a proper solution for this.
Hmm.. I think the reverted patch looks like it really is the right thing to do, so I hate reverting it this early. What were the problems with it?
Quoted 7 lines
Btw, one thing that seems to be missing in the original commit (which is not necessarily the reason for the trouble, of course), is that
elv_rq_merge_ok()seems to not check the partition. As far as I can tell, we should have aif (req->part != bio->bi_bdev->bd_part) return 0;
Ok, not allowing requests of different partitions to merge will atleast solve part of the problem for Yasuaki Ishimatsu. (in-flight accounting going out of sync due to request merging across partitions).
Quoted 4 lines
there, no? And you should _not_ set
rq->partin "drive_stat_acct()", you should set it frombio->bi_bdev->bd_partwhen you create the request.
Not allowing request merging across partitions, I guess should remove the need of storing this info in request and allow us to remove req->part as introduced by commit 7681bfeeccff5efa9eb29bf09249a3c400b15327.
Quoted 4 lines
(And if it is NULL, just don't do partition accounting at all)
Hmm? What am I missing? What were the bugs?
I think atleast one of the bugs was that we don't know whether request queue pointer obtained from disk->queue is still valid or not.
Looking at the backtrace of crash reported by Maxim Levitsky here.
http://lkml.org/lkml/2010/10/23/165
mspro_block_remove() calls blk_cleanup_queue() to put its reference on the queue and which in turn would call kmem_cache_free(blk_requestq_cachep, q) and free up the queue.
Now driver calls mspro_block_disk_release() and which in turn ends up accessing request queue and trying to make sure there are no inflight requests. But request queue might be gone by now and we might end up accessing freed object.
So one of the quick solutions probably is to set disk->queue = NULL after del_gendisk() and before blk_cleanup_queue(). IIUC, blk_cleanup_queue() driver's reference to the request queue and driver should in-turn cleanup the disk->queue, atleast in theory.
Later disk_replace_part_tbl() can do
q = disk->queue;
if (q)
quiesce_elevator;
I am not sure if it is completely race free or not from the other call path of disk_expand_part_tbl(). What kind of locking is there to make sure once disk->queue is fetched, some other cpu does not do blk_cleanup_queue().
Thanks
Vivek