scsi: storvsc: set max_segment_size as UINT_MAX explicitly
Subsystems (3)
PATCH 1/1 scsi: storvsc: set max_segment_size as UINT_MAX explicitly
Set max_segment_size as UINT_MAX explicitly:
- storvrc uses virt_boundary to define segment
- strovrc does not define max_segment_size
So define max_segment_size as UINT_MAX, otherwise __blk_rq_map_sg() takes default 64K max segment size and splits one virtual segment into two parts, then breaks virt_boundary limit.
Before commit ec84ca4025c0 ("scsi: block: Remove now unused queue limits helpers"), max segment size is set as UINT_MAX in case that virt_boundary is defined.
--- drivers/scsi/storvsc_drv.c | 1 + 1 file changed, 1 insertion(+)
drivers/scsi/storvsc_drv.c +1 -0
@@ -1897,6 +1897,7 @@ static struct scsi_host_template scsi_driver = { .no_write_same = 1, .track_queue_depth = 1, .change_queue_depth = storvsc_change_queue_depth, + .max_segment_size = 0xffffffff, };enum {
Signature · 1 line
2.47.0
Quoted 1 line · Ming Lei
- storvrc uses
virt_boundaryto definesegment
^^^^^^^
storvsc?
Quoted 1 line
- strovrc does not define
max_segment_size
^^^^^^^
storvsc?
Otherwise this patch looks good to me.
Thanks,
Bart.
Wei Liu
maintainer+Long for reviewing this patch.
Quoted 6 lines · Ming Lei
Set
max_segment_sizeasUINT_MAXexplicitly:- storvrc uses
virt_boundaryto definesegment- strovrc does not define
max_segment_size
Typo here. It should be storvsc, not storvrc.
Quoted 35 lines
So define
max_segment_sizeasUINT_MAX, otherwise__blk_rq_map_sg()takes default 64K max segment size and splits one virtual segment into two parts, then breaksvirt_boundarylimit.Before commit
ec84ca4025c0("scsi: block: Remove now unused queue limits helpers"), max segment size is set asUINT_MAXin case thatvirt_boundaryis defined.Cc "K. Y. Srinivasan" <[email protected]>Cc Haiyang Zhang <[email protected]>Cc Wei Liu <[email protected]>Cc Christoph Hellwig <[email protected]>Cc Ewan D. Milne <[email protected]>Cc Laurence Oberman <[email protected]>Fixes ec84ca4025c0 ("scsi: block: Remove now unused queue limits helpers")Signed-off-by Ming Lei <[email protected]>--- drivers/scsi/storvsc_drv.c | 1 + 1 file changed, 1 insertion(+)drivers/scsi/storvsc_drv.c +1
-0@@ -1897,6 +1897,7 @@ static struct scsi_host_template scsi_driver = { .no_write_same = 1, .track_queue_depth = 1, .change_queue_depth = storvsc_change_queue_depth, + .max_segment_size = 0xffffffff, };enum {Signature · 1 line
2.47.0
Martin K. Petersen
AppliedQuoted 11 lines · Ming Lei
Set
max_segment_sizeasUINT_MAXexplicitly:- storvrc uses
virt_boundaryto definesegment- strovrc does not define
max_segment_sizeSo define
max_segment_sizeasUINT_MAX, otherwise__blk_rq_map_sg()takes default 64K max segment size and splits one virtual segment into two parts, then breaksvirt_boundarylimit.[...]
Applied to 6.16/scsi-fixes, thanks!
[1/1] scsi: storvsc: set max_segment_size as UINT_MAX explicitly
https://git.kernel.org/mkp/scsi/c/19ec970841ca
Signature · 1 line
Martin K. Petersen Oracle Linux Engineering
Quoted 6 lines · Ming Lei
Set
max_segment_sizeasUINT_MAXexplicitly:- storvrc uses
virt_boundaryto definesegment- strovrc does not define
max_segment_size
Quoted 7 lines
So define
max_segment_sizeasUINT_MAX, otherwise__blk_rq_map_sg()takes default 64K max segment size and splits one virtual segment into two parts, then breaksvirt_boundarylimit.Before commit
ec84ca4025c0("scsi: block: Remove now unused queue limits helpers"), max segment size is set asUINT_MAXin case thatvirt_boundaryis defined.
Drivers should not have done this. If you need this someone (probably me) broke the block layer code ensuring it's not needed, and that affects all drivers using virt_boundary.
Please try this proper fix instead:
drivers/scsi/hosts.c +3 -4
@@ -473,7 +473,9 @@ struct Scsi_Host *scsi_host_alloc(const struct scsi_host_template *sht, int privelse shost->max_sectors =SCSI_DEFAULT_MAX_SECTORS ; -if (sht->max_segment_size) +if (sht->virt_boundary_mask) + shost->virt_boundary_mask = sht->virt_boundary_mask; +else if (sht->max_segment_size) shost->max_segment_size = sht->max_segment_size;else shost->max_segment_size =BLK_MAX_SEGMENT_SIZE ;
@@ -492,9 +494,6 @@ struct Scsi_Host *scsi_host_alloc(const struct scsi_host_template *sht, int privelse shost->dma_boundary =0xffffffff ; -if (sht->virt_boundary_mask) - shost->virt_boundary_mask = sht->virt_boundary_mask; -device_initialize (&shost->shost_gendev);dev_set_name (&shost->shost_gendev,"host%d" , shost->host_no); shost->shost_gendev.bus = &scsi_bus_type;
Quoted 7 lines · Christoph Hellwig
Please try this proper fix instead:
drivers/scsi/hosts.c +0
-0
drivers/scsi/hosts.c struct Scsi_Host *scsi_host_alloc(const struct scsi_host_template *sht, int priv · line 481
Quoted 10 lines
drivers/scsi/hosts.c +3
-1@@ -473,7 +473,9 @@ struct Scsi_Host *scsi_host_alloc(const struct scsi_host_template *sht, int privelse shost->max_sectors =SCSI_DEFAULT_MAX_SECTORS ; -if (sht->max_segment_size) +if (sht->virt_boundary_mask) + shost->virt_boundary_mask = sht->virt_boundary_mask; +else if (sht->max_segment_size) shost->max_segment_size = sht->max_segment_size;else shost->max_segment_size =BLK_MAX_SEGMENT_SIZE ;
This way works, but I prefer to set it explicitly in driver, instead of making block layer more fragile to deal with def ->max_segment_size if ->virt_boundary_mask is defined
- for low level driver, if ->virt_boundary_mask is defined, ->max_segment_size should be UINT_MAX obviously since it implies single virt segment. Setting UINT_MAX in driver has document benefit too.
- for logical block device(md, dm, ...), both ->virt_boundary_mask and ->max_segment_size may be set, and it is fine since logical block device driver needn't to deal with sg
Thanks,
Ming
Quoted 4 lines · Ming Lei
Quoted 4 lines
drivers/scsi/hosts.c +0
-0@@ -473,7 +473,9 @@ struct Scsi_Host *scsi_host_alloc(const struct scsi_host_template *sht, int privelse shost->max_sectors =SCSI_DEFAULT_MAX_SECTORS ;
Quoted 11 lines
drivers/scsi/hosts.c struct Scsi_Host *scsi_host_alloc(const struct scsi_host_template *sht, int priv · line 481
Quoted 7 lines
drivers/scsi/hosts.c +3
-1-if (sht->max_segment_size) +if (sht->virt_boundary_mask) + shost->virt_boundary_mask = sht->virt_boundary_mask; +else if (sht->max_segment_size) shost->max_segment_size = sht->max_segment_size;else shost->max_segment_size =BLK_MAX_SEGMENT_SIZE ;This way works, but I prefer to set it explicitly in driver, instead of making block layer more fragile to deal with def ->max_segment_size if ->virt_boundary_mask is defined
The block layer already enforces this as it is a requirement. It is just the SCSI wrapper that broke it. Without this proper fix iser is still broken, and srp might or might not.
Quoted 3 lines
- for low level driver, if ->virt_boundary_mask is defined, ->max_segment_size should be
UINT_MAXobviously since it implies singlevirt segment. SettingUINT_MAXin driver has document benefit too.
No, it doesn't. It means you need to cargo cult copy and paste code instead of solving the problem in the proper place.
Quoted 3 lines
- for logical block
device(md, dm, ...), both ->virt_boundary_mask and ->max_segment_size may be set, and it is fine since logical block device driver needn't to deal with sg
Stacked devices should not inherit the hardware limits at all because we split below them, but that's a separate story. Either way this is not relevant here as we don't have stacking drivers that use the SCSI layer.
Laurence Oberman
Tested-byQuoted 21 lines · Christoph Hellwig
Please try this proper fix instead:
drivers/scsi/hosts.c +0
-0@@ -473,7 +473,9 @@ struct Scsi_Host *scsi_host_alloc(const struct
scsi_host_template*sht, int priv
else
shost->max_sectors =SCSI_DEFAULT_MAX_SECTORS;
- if (sht->max_segment_size)
+ if (sht->virt_boundary_mask)
+ shost->virt_boundary_mask =sht->virt_boundary_mask; + else if (sht->max_segment_size)shost->max_segment_size=sht->max_segment_size; else
shost->max_segment_size =BLK_MAX_SEGMENT_SIZE;drivers/scsi/hosts.c +0
-0@@ -492,9 +494,6 @@ struct Scsi_Host *scsi_host_alloc(const struct
scsi_host_template*sht, int priv
else
Quoted 9 lines
shost->dma_boundary =
0xffffffff;
- if (sht->virt_boundary_mask)
- shost->virt_boundary_mask =sht->virt_boundary_mask; -
device_initialize(&shost->shost_gendev);dev_set_name(&shost->shost_gendev, "host%d", shost->host_no);shost->shost_gendev.bus=&scsi_bus_type;
Hello
Not sure what you folks will decide as the final fix but Christoph's patch prevents the REDO corruption as well.
Ran it long enough to know its clean from corruption with this patch above
Thanks vey mauch as always folks
Regards
Laurence
Quoted 41 lines · Ming Lei
Set
max_segment_sizeasUINT_MAXexplicitly:- storvrc uses
virt_boundaryto definesegment- strovrc does not define
max_segment_sizeSo define
max_segment_sizeasUINT_MAX, otherwise__blk_rq_map_sg()takes
default 64K max segment size and splits one virtual segment into two parts,
then breaksvirt_boundarylimit.Before commit
ec84ca4025c0("scsi: block: Remove now unused queue limits helpers"),
max segment size is set asUINT_MAXin case thatvirt_boundaryis defined.Cc "K. Y. Srinivasan" <[email protected]>Cc Haiyang Zhang <[email protected]>Cc Wei Liu <[email protected]>Cc Christoph Hellwig <[email protected]>Cc Ewan D. Milne <[email protected]>Cc Laurence Oberman <[email protected]>Fixes ec84ca4025c0 ("scsi: block: Remove now unused queue limitshelpers")
Signed-off-by Ming Lei <[email protected]>---
drivers/scsi/storvsc_drv.c | 1 +1 file changed, 1 insertion(+)drivers/scsi/storvsc_drv.c +1
-0@@ -1897,6 +1897,7 @@ static struct scsi_host_template scsi_driver = { .no_write_same = 1, .track_queue_depth = 1, .change_queue_depth = storvsc_change_queue_depth, + .max_segment_size = 0xffffffff, };
Quoted 2 lines
enum {
Hello
For what it is worth, I tested Ming's patch in our lab and at our customers and it fixed a very serious corruption in Oracle REDO logs.
I will test what Christoph share dbut our initial way to deal with this in RHEL will be the point fix in storvsc as its a critical issue needing an urgent fix.
Thanks
Laurence
Quoted 1 line · Ming Lei
Fixes ec84ca4025c0 ("scsi: block: Remove now unused queue limits helpers")
That's not the correct fixes tag. This got broken by:
commit b561ea56a26415bf44ce8ca6a8e625c7c390f1ea
Author: Ming Lei <[email protected]> Date: Sun Apr 7 21:19:31 2024 +0800
block: allow device to have both virt_boundary_mask and max segment size
for which I explicitly warned about this kind of breakage.
So please, let's fix this properly instead of piling crap over crap.