Patchwork [v2,0/9] fix max discard sectors limit

login
register
mail settings
Submitter James Bottomley
Date April 20, 2013, 7:50 p.m.
Message ID <1366487439.1993.9.camel@dabdike>
Download mbox | patch
Permalink /patch/238188/
State New
Headers show

Comments

James Bottomley - April 20, 2013, 7:50 p.m.
On Sat, 2013-04-20 at 01:40 +0900, Namjae Jeon wrote:
> From: Namjae Jeon <namjae.jeon@samsung.com>
> 
> linux-v3.8-rc1 and later support for plug for blkdev_issue_discard with
> commit 0cfbcafcae8b7364b5fa96c2b26ccde7a3a296a9 
> (block: add plug for blkdev_issue_discard )
> 
> For example,
> 1) DISCARD rq-1 with size size 4GB
> 2) DISCARD rq-2 with size size 1GB
> 
> If these 2 discard requests get merged, final request size will be 5GB.
> 
> In this case, request's __data_len field may overflow as it can store
> max 4GB(unsigned int).
> 
> This issue was observed while doing mkfs.f2fs on 5GB SD card:
> https://lkml.org/lkml/2013/4/1/292
> 
> # mkfs.f2fs /dev/mmcblk0p3
> Info: sector size = 512
> Info: total sectors = 11370496 (in 512bytes)
> Info: zone aligned segment0 blkaddr: 512
> [  257.789764] blk_update_request: bio idx 0 >= vcnt 0
> 
> mkfs process gets stuck in D state and I see the following in the dmesg:
> 
> [  257.789733] __end_that: dev mmcblk0: type=1, flags=122c8081
> [  257.789764]   sector 4194304, nr/cnr 2981888/4294959104
> [  257.789764]   bio df3840c0, biotail df3848c0, buffer   (null), len 1526726656
> [  257.789764] blk_update_request: bio idx 0 >= vcnt 0
> [  257.794921] request botched: dev mmcblk0: type=1, flags=122c8081
> [  257.794921]   sector 4194304, nr/cnr 2981888/4294959104
> [  257.794921]   bio df3840c0, biotail df3848c0, buffer   (null), len 1526726656
> 
> Few drivers(e.g. mmc, mtd..) set q->limits.max_discard_sectors
> more than UINT_MAX >> 9 sectors which is incorrect and it may lead to overflow
> of request's __data_len field if merged discard request's size exceeds 4GB.
> 
> This patchset fixes this issue by updating helper function
> blk_queue_max_discard_sectors which is used to set max_discard_sectors limit.
> 
> This patchset also replaces "q->limits.max_discard_sector = max_discard_sectors"
> with blk_queue_max_discard_sectors call in other drivers like mmc, mtd etc.

I really don't understand this explanation.  How can you be affected by
the incorrect setting of q->limits.max_discard sectors when  n the
blkdev_issue_discard() code you see:

	max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >>
9);

?

The problem is not that we issue discards bigger than __data_len can
allow, the problem is that we merge them larger than __data_len will
allow.  That means the merge code needs fixing to pay attention to
max_discard_sectors, so isn't this the correct fix:

James

---
NamJae Jeon - April 21, 2013, 1:37 a.m.
2013/4/21 James Bottomley <James.Bottomley@hansenpartnership.com>:
> On Sat, 2013-04-20 at 01:40 +0900, Namjae Jeon wrote:
>> From: Namjae Jeon <namjae.jeon@samsung.com>
>>
>> linux-v3.8-rc1 and later support for plug for blkdev_issue_discard with
>> commit 0cfbcafcae8b7364b5fa96c2b26ccde7a3a296a9
>> (block: add plug for blkdev_issue_discard )
>>
>> For example,
>> 1) DISCARD rq-1 with size size 4GB
>> 2) DISCARD rq-2 with size size 1GB
>>
>> If these 2 discard requests get merged, final request size will be 5GB.
>>
>> In this case, request's __data_len field may overflow as it can store
>> max 4GB(unsigned int).
>>
>> This issue was observed while doing mkfs.f2fs on 5GB SD card:
>> https://lkml.org/lkml/2013/4/1/292
>>
>> # mkfs.f2fs /dev/mmcblk0p3
>> Info: sector size = 512
>> Info: total sectors = 11370496 (in 512bytes)
>> Info: zone aligned segment0 blkaddr: 512
>> [  257.789764] blk_update_request: bio idx 0 >= vcnt 0
>>
>> mkfs process gets stuck in D state and I see the following in the dmesg:
>>
>> [  257.789733] __end_that: dev mmcblk0: type=1, flags=122c8081
>> [  257.789764]   sector 4194304, nr/cnr 2981888/4294959104
>> [  257.789764]   bio df3840c0, biotail df3848c0, buffer   (null), len 1526726656
>> [  257.789764] blk_update_request: bio idx 0 >= vcnt 0
>> [  257.794921] request botched: dev mmcblk0: type=1, flags=122c8081
>> [  257.794921]   sector 4194304, nr/cnr 2981888/4294959104
>> [  257.794921]   bio df3840c0, biotail df3848c0, buffer   (null), len 1526726656
>>
>> Few drivers(e.g. mmc, mtd..) set q->limits.max_discard_sectors
>> more than UINT_MAX >> 9 sectors which is incorrect and it may lead to overflow
>> of request's __data_len field if merged discard request's size exceeds 4GB.
>>
>> This patchset fixes this issue by updating helper function
>> blk_queue_max_discard_sectors which is used to set max_discard_sectors limit.
>>
>> This patchset also replaces "q->limits.max_discard_sector = max_discard_sectors"
>> with blk_queue_max_discard_sectors call in other drivers like mmc, mtd etc.
>
Hi. James.
> I really don't understand this explanation.  How can you be affected by
> the incorrect setting of q->limits.max_discard sectors when  n the
> blkdev_issue_discard() code you see:
>
>         max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >>
> 9);
>
> ?
>
> The problem is not that we issue discards bigger than __data_len can
> allow, the problem is that we merge them larger than __data_len will
> allow.  That means the merge code needs fixing to pay attention to
> max_discard_sectors, so isn't this the correct fix:
Yes, I agree. And the below patch looks good to fix this issue.
Thanks for your comment.

>
> James
>
> ---
>
> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> index 78feda9..33f358f 100644
> --- a/include/linux/blkdev.h
> +++ b/include/linux/blkdev.h
> @@ -838,7 +838,7 @@ static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
>                                                      unsigned int cmd_flags)
>  {
>         if (unlikely(cmd_flags & REQ_DISCARD))
> -               return q->limits.max_discard_sectors;
> +               return min(q->limits.max_discard_sectors, UINT_MAX >> 9);
>
>         if (unlikely(cmd_flags & REQ_WRITE_SAME))
>                 return q->limits.max_write_same_sectors;
>
>
Hannes Reinecke - April 26, 2013, 7:33 a.m.
On 04/20/2013 09:50 PM, James Bottomley wrote:
> On Sat, 2013-04-20 at 01:40 +0900, Namjae Jeon wrote:
>> From: Namjae Jeon <namjae.jeon@samsung.com>
>>
>> linux-v3.8-rc1 and later support for plug for blkdev_issue_discard with
>> commit 0cfbcafcae8b7364b5fa96c2b26ccde7a3a296a9 
>> (block: add plug for blkdev_issue_discard )
>>
>> For example,
>> 1) DISCARD rq-1 with size size 4GB
>> 2) DISCARD rq-2 with size size 1GB
>>
>> If these 2 discard requests get merged, final request size will be 5GB.
>>
>> In this case, request's __data_len field may overflow as it can store
>> max 4GB(unsigned int).
>>
>> This issue was observed while doing mkfs.f2fs on 5GB SD card:
>> https://lkml.org/lkml/2013/4/1/292
>>
>> # mkfs.f2fs /dev/mmcblk0p3
>> Info: sector size = 512
>> Info: total sectors = 11370496 (in 512bytes)
>> Info: zone aligned segment0 blkaddr: 512
>> [  257.789764] blk_update_request: bio idx 0 >= vcnt 0
>>
>> mkfs process gets stuck in D state and I see the following in the dmesg:
>>
>> [  257.789733] __end_that: dev mmcblk0: type=1, flags=122c8081
>> [  257.789764]   sector 4194304, nr/cnr 2981888/4294959104
>> [  257.789764]   bio df3840c0, biotail df3848c0, buffer   (null), len 1526726656
>> [  257.789764] blk_update_request: bio idx 0 >= vcnt 0
>> [  257.794921] request botched: dev mmcblk0: type=1, flags=122c8081
>> [  257.794921]   sector 4194304, nr/cnr 2981888/4294959104
>> [  257.794921]   bio df3840c0, biotail df3848c0, buffer   (null), len 1526726656
>>
>> Few drivers(e.g. mmc, mtd..) set q->limits.max_discard_sectors
>> more than UINT_MAX >> 9 sectors which is incorrect and it may lead to overflow
>> of request's __data_len field if merged discard request's size exceeds 4GB.
>>
>> This patchset fixes this issue by updating helper function
>> blk_queue_max_discard_sectors which is used to set max_discard_sectors limit.
>>
>> This patchset also replaces "q->limits.max_discard_sector = max_discard_sectors"
>> with blk_queue_max_discard_sectors call in other drivers like mmc, mtd etc.
> 
> I really don't understand this explanation.  How can you be affected by
> the incorrect setting of q->limits.max_discard sectors when  n the
> blkdev_issue_discard() code you see:
> 
> 	max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >>
> 9);
> 
> ?
> 
> The problem is not that we issue discards bigger than __data_len can
> allow, the problem is that we merge them larger than __data_len will
> allow.  That means the merge code needs fixing to pay attention to
> max_discard_sectors, so isn't this the correct fix:
> 
> James
> 
> ---
> 
> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> index 78feda9..33f358f 100644
> --- a/include/linux/blkdev.h
> +++ b/include/linux/blkdev.h
> @@ -838,7 +838,7 @@ static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
>  						     unsigned int cmd_flags)
>  {
>  	if (unlikely(cmd_flags & REQ_DISCARD))
> -		return q->limits.max_discard_sectors;
> +		return min(q->limits.max_discard_sectors, UINT_MAX >> 9);
>  
>  	if (unlikely(cmd_flags & REQ_WRITE_SAME))
>  		return q->limits.max_write_same_sectors;
> 
> 
Patch works, and fixes the discard failing issue I've been pestering
mkp with at LSF. Please apply.

Cheers,

Hannes

Patch

diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 78feda9..33f358f 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -838,7 +838,7 @@  static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
 						     unsigned int cmd_flags)
 {
 	if (unlikely(cmd_flags & REQ_DISCARD))
-		return q->limits.max_discard_sectors;
+		return min(q->limits.max_discard_sectors, UINT_MAX >> 9);
 
 	if (unlikely(cmd_flags & REQ_WRITE_SAME))
 		return q->limits.max_write_same_sectors;