When enabling DIX T10-DIF-TYPE1-IP protection you can hit the
bip_vec full condition which fails to attach the integrity metadata
and returns 0 back to bio_integrity_prep()
[ 3.837493] sd 0:0:17:1089486880: [sda] Enabling DIX T10-DIF-TYPE1-IP protection
[ 3.839089] sd 0:0:17:1089486880: [sda] Attached SCSI disk
[ 3.841309] bio_integrity_add_page: bip_vec full
<debug>
[ 439.180928] bio_integrity_prep: sectors 8 len 64
[ 439.180929] bio_integrity_prep: nr_pages 1 end 706577 start 706576
[ 439.180930] bio_integrity_alloc: bs 00000000b2c87880 nr_vecs 1
[ 439.180931] bio_integrity_alloc: nr_vecs 1 inline_vecs 4
[ 439.180932] bio_integrity_alloc: bip->bip_vec 00000000aeaf9158 bip->bip_slab 15
[ 439.180933] bio_integrity_prep: offset 704 nr_pages 1
[ 439.180934] bio_integrity_add_page: len 64 offset 704
[ 439.180935] bio_integrity_add_page: bip_vec full bip_vcnt 0 bvec_nr_vecs 0 bip_slab 15
<debug>
Ultimately you can BUG_ON(!nents) in scsi_alloc_sgtable()
scsi_init_io
blk_integrity_rq
scsi_alloc_sgtable
With attached patch device functions normally
<dmesg>
[ 198.481838] sd 0:0:24:1089486880: [sda] Enabling DIF Type 1 protection
[ 198.481845] sd 0:0:24:1089486880: [sda] 209715200 512-byte logical blocks: (107 GB/100 GiB)
[ 198.482552] sd 0:0:24:1089486880: [sda] Write Protect is off
[ 198.482556] sd 0:0:24:1089486880: [sda] Mode Sense: ed 00 00 08
[ 198.482889] sd 0:0:24:1089486880: [sda] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA
[ 198.485850] sda: sda1
[ 198.485926] sd 0:0:24:1089486880: [sda] Enabling DIX T10-DIF-TYPE1-IP protection
[ 198.487652] sd 0:0:24:1089486880: [sda] Attached SCSI disk
[ 276.566682] XFS (sda1): Mounting Filesystem
[ 276.576558] XFS (sda1): Ending clean mount
<dmesg>
Signed-off-by: David Milburn <[email protected]>
---
fs/bio-integrity.c | 3 ++-
1 files changed, 2 insertions(+), 1 deletions(-)
diff --git a/fs/bio-integrity.c b/fs/bio-integrity.c
index 45e944f..9d36a09 100644
--- a/fs/bio-integrity.c
+++ b/fs/bio-integrity.c
@@ -129,7 +129,8 @@ int bio_integrity_add_page(struct bio *bio, struct page *page,
struct bio_integrity_payload *bip = bio->bi_integrity;
struct bio_vec *iv;
- if (bip->bip_vcnt >= bvec_nr_vecs(bip->bip_slab)) {
+ if (bip->bip_slab != BIO_POOL_NONE &&
+ bip->bip_vcnt >= bvec_nr_vecs(bip->bip_slab)) {
printk(KERN_ERR "%s: bip_vec full\n", __func__);
return 0;
}
>>>>> "David" == David Milburn <[email protected]> writes:
David> When enabling DIX T10-DIF-TYPE1-IP protection you can hit the
David> bip_vec full condition which fails to attach the integrity
David> metadata and returns 0 back to bio_integrity_prep()
Looks like Kent accidentally broke this when he changed the bvec pool
setup.
David> - if (bip->bip_vcnt >= bvec_nr_vecs(bip->bip_slab)) {
David> + if (bip->bip_slab != BIO_POOL_NONE &&
David> + bip->bip_vcnt >= bvec_nr_vecs(bip->bip_slab)) {
David> printk(KERN_ERR "%s: bip_vec full\n", __func__);
David> return 0;
David> }
We still need to check that the page will actually fit, though:
block: Fix nr_vecs for inline integrity vectors
Commit 9f060e2231ca changed the way we handle allocations for the
integrity vectors. When the vectors are inline there is no associated
slab and consequently bvec_nr_vecs() returns 0. Ensure that we check
against BIP_INLINE_VECS in that case.
Reported-by: David Milburn <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>
diff --git a/fs/bio-integrity.c b/fs/bio-integrity.c
index fc60b31453ee..6dea2b90b4d5 100644
--- a/fs/bio-integrity.c
+++ b/fs/bio-integrity.c
@@ -114,6 +114,14 @@ void bio_integrity_free(struct bio *bio)
}
EXPORT_SYMBOL(bio_integrity_free);
+static inline unsigned int bip_integrity_vecs(struct bio_integrity_payload *bip)
+{
+ if (bip->bip_slab == BIO_POOL_NONE)
+ return BIP_INLINE_VECS;
+
+ return bvec_nr_vecs(bip->bip_slab);
+}
+
/**
* bio_integrity_add_page - Attach integrity metadata
* @bio: bio to update
@@ -129,7 +137,7 @@ int bio_integrity_add_page(struct bio *bio, struct page *page,
struct bio_integrity_payload *bip = bio->bi_integrity;
struct bio_vec *iv;
- if (bip->bip_vcnt >= bvec_nr_vecs(bip->bip_slab)) {
+ if (bip->bip_vcnt >= bip_integrity_vecs(bip)) {
printk(KERN_ERR "%s: bip_vec full\n", __func__);
return 0;
}
On 02/03/2014 09:55 PM, Martin K. Petersen wrote:
>>>>>> "David" == David Milburn <[email protected]> writes:
>
> David> When enabling DIX T10-DIF-TYPE1-IP protection you can hit the
> David> bip_vec full condition which fails to attach the integrity
> David> metadata and returns 0 back to bio_integrity_prep()
>
> Looks like Kent accidentally broke this when he changed the bvec pool
> setup.
>
> David> - if (bip->bip_vcnt >= bvec_nr_vecs(bip->bip_slab)) {
> David> + if (bip->bip_slab != BIO_POOL_NONE &&
> David> + bip->bip_vcnt >= bvec_nr_vecs(bip->bip_slab)) {
> David> printk(KERN_ERR "%s: bip_vec full\n", __func__);
> David> return 0;
> David> }
>
> We still need to check that the page will actually fit, though:
>
>
> block: Fix nr_vecs for inline integrity vectors
>
> Commit 9f060e2231ca changed the way we handle allocations for the
> integrity vectors. When the vectors are inline there is no associated
> slab and consequently bvec_nr_vecs() returns 0. Ensure that we check
> against BIP_INLINE_VECS in that case.
>
> Reported-by: David Milburn <[email protected]>
> Signed-off-by: Martin K. Petersen <[email protected]>
>
> diff --git a/fs/bio-integrity.c b/fs/bio-integrity.c
> index fc60b31453ee..6dea2b90b4d5 100644
> --- a/fs/bio-integrity.c
> +++ b/fs/bio-integrity.c
> @@ -114,6 +114,14 @@ void bio_integrity_free(struct bio *bio)
> }
> EXPORT_SYMBOL(bio_integrity_free);
>
> +static inline unsigned int bip_integrity_vecs(struct bio_integrity_payload *bip)
> +{
> + if (bip->bip_slab == BIO_POOL_NONE)
> + return BIP_INLINE_VECS;
> +
> + return bvec_nr_vecs(bip->bip_slab);
> +}
> +
> /**
> * bio_integrity_add_page - Attach integrity metadata
> * @bio: bio to update
> @@ -129,7 +137,7 @@ int bio_integrity_add_page(struct bio *bio, struct page *page,
> struct bio_integrity_payload *bip = bio->bi_integrity;
> struct bio_vec *iv;
>
> - if (bip->bip_vcnt >= bvec_nr_vecs(bip->bip_slab)) {
> + if (bip->bip_vcnt >= bip_integrity_vecs(bip)) {
> printk(KERN_ERR "%s: bip_vec full\n", __func__);
> return 0;
> }
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
Hi Martin,
Your patch has been tested successfully.
Thanks for your help,
David