Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1538089 > unrolled thread

[PATCH] libata-scsi: disable SCT Write Same for the moment

Back to article view | Back to linux.kernel

Contents

  [PATCH] libata-scsi: disable SCT Write Same for the moment Nicolai Stange <nicstange@gmail.com> - 2016-12-07 22:30 +0100
    Re: [PATCH] libata-scsi: disable SCT Write Same for the moment Tejun Heo <tj@kernel.org> - 2016-12-07 22:40 +0100

#1538089 — [PATCH] libata-scsi: disable SCT Write Same for the moment

SCT Write Same support had been introduced with
commit 7b2030942859 ("libata: Add support for SCT Write Same")

Some problems, namely excessive userspace segfaults, had been reported at

  http://lkml.kernel.org/r/20160908192736.GA4356@gmail.com

This lead to commit 0ce1b18c42a5 ("libata: Some drives failing on
SCT Write Same") which strived to disable SCT Write Same on !ZAC devices.
Due to the way this was done and to the logic in sd_config_write_same(),
this didn't work for those devices that have
->max_ws_blocks > SD_MAX_WS10_BLOCKS: for these, ->no_write_same and
->max_write_same_sectors would still be non-zero,
but ->ws10 == ->ws16 == 0. This would cause sd_setup_write_same_cmnd() to
demultiplex REQ_OP_WRITE_SAME requests to WRITE_SAME, and these in turn
aren't supported by libata-scsi:

  EXT4-fs (dm-1): Delayed block allocation failed for inode 2625094 at
                  logical offset 2032 with max blocks 2 with error 121
  EXT4-fs (dm-1): This should not happen!! Data will be lost

121 == EREMOTEIO is what scsi_io_completion() asserts in case of
invalid opcodes.

Back to the original problem of userspace segfaults: this can be tracked
down to ata_format_sct_write_same() overwriting the input page. Sometimes,
this page is ZERO_PAGE(0) which ceases to be filled with zeros from that
point on. Since ZERO_PAGE(0) is used for userspace .bss mappings, code of
the following is doomed:

  static char *a = NULL; /* .bss */
  ...
  if (a)
    *a = 'a';

This problem is not solved by disabling SCT Write Same for !ZAC devices
only.

It can certainly be fixed, but the final release is quite close -- so
disable SCT Write Same for all ATA devices rather than introducing some
SCT key buffer allocation schemes at this point.

Fixes: 7b2030942859 ("libata: Add support for SCT Write Same")
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
---
 drivers/ata/libata-scsi.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/ata/libata-scsi.c b/drivers/ata/libata-scsi.c
index 982bcfc..c9abb87 100644
--- a/drivers/ata/libata-scsi.c
+++ b/drivers/ata/libata-scsi.c
@@ -1234,6 +1234,7 @@ static void ata_scsi_sdev_config(struct scsi_device *sdev)
 {
 	sdev->use_10_for_rw = 1;
 	sdev->use_10_for_ms = 1;
+	sdev->no_write_same = 1;
 
 	/* Schedule policy is determined by ->qc_defer() callback and
 	 * it needs to see every deferred qc.  Set dev_blocked to 1 to
-- 
2.10.2

[toc] | [next] | [standalone]

#1538097

On Wed, Dec 07, 2016 at 10:21:33PM +0100, Nicolai Stange wrote:
> It can certainly be fixed, but the final release is quite close -- so
> disable SCT Write Same for all ATA devices rather than introducing some
> SCT key buffer allocation schemes at this point.

Applied to libata/for-4.9-fixes.  Will push out soon.

Thanks.

-- 
tejun

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web