Skip to contentSkip to conversation

nvmet-file: clamp-down file namespace lba_shift

PATCH New Sagi Grimberg 4 participants 7 messages Mar 5, 2019 – Mar 8, 2019

Subsystems (2)
lore
Share & tools
lore

Sagi Grimberg

author maintainer
#0 lore raw

PATCH 1/1 nvmet-file: clamp-down file namespace lba_shift

When the backing file is a tempfile for example, the inode i_blkbits can be 1M in size which causes problems for hosts to support as the disk block size.

Signed-off-by Sagi Grimberg <sagi at grimberg.me>

---
This issue was observed in I/O errors in the log running test nvme/009.

 drivers/nvme/target/io-cmd-file.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/nvme/target/io-cmd-file.c +1 -1
index 517522305e5c..5008826cbd6a 100644
@@ -49,7 +49,7 @@ int nvmet_file_ns_enable(struct nvmet_ns *ns)
 		goto err;
 
 	ns->size = stat.size;
-	ns->blksize_shift = file_inode(ns->file)->i_blkbits;
+	ns->blksize_shift = min_t(u8, file_inode(ns->file)->i_blkbits, PAGE_SHIFT);
 
 	ns->bvec_cache = kmem_cache_create("nvmet-bvec",
 			NVMET_MAX_MPOOL_BVEC * sizeof(struct bio_vec),
Signature · 1 line
2.17.1
#1 lore raw

PATCH 1/1 nvmet-file: clamp-down file namespace lba_shift

Quoted 8 lines · Sagi Grimberg

When the backing file is a tempfile for example, the inode i_blkbits can be 1M in size which causes problems for hosts to support as the disk block size.

Signed-off-by Sagi Grimberg <sagi at grimberg.me>

---
This issue was observed in I/O errors in the log running test nvme/009.

We are not doing any I/O in nvme/009 explicitly using dd/fio, is this due to internal commands we send as a part of connect and disconnect ? Or test number is different ?

Quoted 17 lines
  drivers/nvme/target/io-cmd-file.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/nvme/target/io-cmd-file.c +1 -1
index 517522305e5c..5008826cbd6a 100644
@@ -49,7 +49,7 @@ int nvmet_file_ns_enable(struct nvmet_ns *ns)
  		goto err;
 
  	ns->size = stat.size;
-	ns->blksize_shift = file_inode(ns->file)->i_blkbits;
+	ns->blksize_shift = min_t(u8, file_inode(ns->file)->i_blkbits, PAGE_SHIFT);
 
  	ns->bvec_cache = kmem_cache_create("nvmet-bvec",
  			NVMET_MAX_MPOOL_BVEC * sizeof(struct bio_vec),
author maintainer
#2 lore raw

PATCH 1/1 nvmet-file: clamp-down file namespace lba_shift

Quoted 5 lines
Quoted 5 lines

When the backing file is a tempfile for example, the inode i_blkbits can be 1M in size which causes problems for hosts to support as the disk block size.

Quoted 7 lines
Quoted 4 lines
Signed-off-by Sagi Grimberg <sagi at grimberg.me>

---
This issue was observed in I/O errors in the log running test nvme/009.

We are not doing any I/O in nvme/009 explicitly using dd/fio, is this due to internal commands we send as a part of connect and disconnect ? Or test number is different ?

The former.

Also, I just now hit a divide by zero BUG when running test 009 with multipath enabled.

Quick look shows that this is coming from blk_queue_stack_limits for the ns_head.

Looks like nvme_update_disk_info() is reading a bs=0:
--

         unsigned short bs = 1 << ns->lba_shift;

--

bs is 16 bits and lba_shift is 20
(coming directly from the tempfile file_inode(ns->file)->i_blkbits)

So this fixes that one as well. Regardless, I think bs should be unsigned int?

#3 lore raw

PATCH 1/1 nvmet-file: clamp-down file namespace lba_shift

Quoted 23 lines · Sagi Grimberg
Quoted 5 lines
Quoted 5 lines

When the backing file is a tempfile for example, the inode i_blkbits can be 1M in size which causes problems for hosts to support as the disk block size.

Quoted 7 lines
Quoted 4 lines
Signed-off-by Sagi Grimberg <sagi at grimberg.me>

---
This issue was observed in I/O errors in the log running test nvme/009.

We are not doing any I/O in nvme/009 explicitly using dd/fio, is this due to internal commands we send as a part of connect and disconnect ? Or test number is different ?

The former.

Also, I just now hit a divide by zero BUG when running test 009 with multipath enabled.

Quick look shows that this is coming from blk_queue_stack_limits for the ns_head.

Looks like nvme_update_disk_info() is reading a bs=0:
--

Quoted 8 lines
          unsigned short bs = 1 << ns->lba_shift;

--

bs is 16 bits and lba_shift is 20
(coming directly from the tempfile file_inode(ns->file)->i_blkbits)

So this fixes that one as well. Regardless, I think bs should be unsigned int?

Yes.

Quoted 6 lines

_______________________________________________
Linux-nvme mailing list
Linux-nvme at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

maintainer
#4 lore raw

PATCH 1/1 nvmet-file: clamp-down file namespace lba_shift

Quoted 9 lines · Chaitanya Kulkarni
Quoted 8 lines · Sagi Grimberg

Quick look shows that this is coming from blk_queue_stack_limits for the ns_head.

Looks like nvme_update_disk_info() is reading a bs=0:
--

          unsigned short bs = 1 << ns->lba_shift;

--

Quoted 7 lines
Quoted 5 lines

bs is 16 bits and lba_shift is 20
(coming directly from the tempfile file_inode(ns->file)->i_blkbits)

So this fixes that one as well. Regardless, I think bs should be unsigned int?

Yes.

blk_queue_logical_block_size() takes an 'unsigned short', so be sure to check 'bs' is less than USHRT_MAX.

author maintainer
#5 lore raw

PATCH 1/1 nvmet-file: clamp-down file namespace lba_shift

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines

Quick look shows that this is coming from blk_queue_stack_limits for the ns_head.

Looks like nvme_update_disk_info() is reading a bs=0:
--

           unsigned short bs = 1 << ns->lba_shift;

--

bs is 16 bits and lba_shift is 20
(coming directly from the tempfile file_inode(ns->file)->i_blkbits)

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

So this fixes that one as well. Regardless, I think bs should be unsigned int?

Yes.

blk_queue_logical_block_size() takes an 'unsigned short', so be sure to check 'bs' is less than USHRT_MAX.

So this means that we should fail ns allocation if lba_shift is more than fits in a short.

#6 lore raw

PATCH 1/1 nvmet-file: clamp-down file namespace lba_shift

In reply to Sagi Grimberg

drivers/nvme/target/io-cmd-file.c int nvmet_file_ns_enable(struct nvmet_ns *ns) · line 52

Quoted 2 lines
drivers/nvme/target/io-cmd-file.c +1 -1
-	ns->blksize_shift = file_inode(ns->file)->i_blkbits;
+	ns->blksize_shift = min_t(u8, file_inode(ns->file)->i_blkbits, PAGE_SHIFT);

I think we want to limit to 4k as the universally acceptable upper bound. Also this should not use u8 for the case but probably an unsigned int.

Last but not least this introduced a > 80 char line.