Skip to contentSkip to conversation

btrfs-progs: utils: fix btrfs_wipe_existing_sb probe bug

PATCH In review Fox Chen 2 participants 3 messages Mar 25, 2021 – Apr 15, 2021 Btrfs file system

lore
Share & tools
lore

Fox Chen

author
#0 lore raw

PATCH 1/1 btrfs-progs: utils: fix btrfs_wipe_existing_sb probe bug

btrfs_wipe_existing_sb() misses calling blkid_do_fullprobe() to do the real probe. After calling blkid_new_probe() & blkid_probe_set_device() to setup blkid_probe context, it directly calls blkid_probe_lookup_value(). This results in blkid_probe_lookup_value returning -1, because pr->values is empty.

---
 common/device-utils.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
common/device-utils.c +1 -1
index c860b946..f8e2e776 100644
@@ -114,7 +114,7 @@ static int btrfs_wipe_existing_sb(int fd)
 	if (!pr)
 		return -1;
 
-	if (blkid_probe_set_device(pr, fd, 0, 0)) {
+	if (blkid_probe_set_device(pr, fd, 0, 0) || blkid_do_fullprobe(pr)) {
 		ret = -1;
 		goto out;
 	}
Signature · 1 line
2.31.0
#1 lore raw

Hi,

Quoted 26 lines

btrfs_wipe_existing_sb() misses calling blkid_do_fullprobe() to do the real probe. After calling blkid_new_probe() & blkid_probe_set_device() to setup blkid_probe context, it directly calls blkid_probe_lookup_value(). This results in blkid_probe_lookup_value returning -1, because pr->values is empty.

---
 common/device-utils.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
common/device-utils.c +1 -1
index c860b946..f8e2e776 100644
@@ -114,7 +114,7 @@ static int btrfs_wipe_existing_sb(int fd)
 	if (!pr)
 		return -1;
 
-	if (blkid_probe_set_device(pr, fd, 0, 0)) {
+	if (blkid_probe_set_device(pr, fd, 0, 0) || blkid_do_fullprobe(pr)) {
 		ret = -1;
 		goto out;
 	}
Signature · 1 line
2.31.0

With this patch, 'mkfs.btrfs -f /dev/nvme0n1 /dev/sdb' output some error when /dev/nvme0n1 have 2 partitions.
ERROR: cannot wipe superblocks on /dev/nvme0n1

# blkid
/dev/nvme0n1: PTUUID="93a54ce8-04b2-470b-8c05-31bfcef02f28" PTTYPE="gpt"
/dev/sda1: LABEL="OS_USB" UUID="2b7f4cb9-3dac-443f-8c96-a907b9276942" BLOCK_SIZE="512" TYPE="xfs" PARTUUID="ee58e9d3-01"
/dev/nvme0n1p1: UUID="1d94dc2b-abd1-47df-bf39-ab31cf579d29" UUID_SUB="577542b7-91f5-48d4-a54c-98cbd4525c00" BLOCK_SIZE="4096" TYPE="btrfs" PARTLABEL="primary" PARTUUID="efed009f-8ae6-4567-9cd3-80a57cdcf225"
/dev/nvme0n1p2: PARTLABEL="primary" PARTUUID="fcab66cd-daad-457f-a53c-110592d8941f"
...

Without this patch, 'mkfs.btrfs -f /dev/nvme0n1 /dev/sdb' have some issue too when /dev/nvme0n1 and /dev/sdb have some partitions.

some blkid of partition is still left in the output of blkid. 'blockdev --rereadpt' will let them disappear.

Best Regards
Wang Yugui ([email protected])
2021/04/09

author
#2 lore raw
Quoted 22 lines · Wang Yugui

Hi,

Quoted 19 lines

btrfs_wipe_existing_sb() misses calling blkid_do_fullprobe() to do the real probe. After calling blkid_new_probe() & blkid_probe_set_device() to setup blkid_probe context, it directly calls blkid_probe_lookup_value(). This results in blkid_probe_lookup_value returning -1, because pr->values is empty.

---
 common/device-utils.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
common/device-utils.c +0 -0
index c860b946..f8e2e776 100644
@@ -114,7 +114,7 @@ static int btrfs_wipe_existing_sb(int fd)
      if (!pr)
              return -1;
 
Quoted 12 lines
Quoted 7 lines
common/device-utils.c +1 -2
-     if (blkid_probe_set_device(pr, fd, 0, 0)) {
+     if (blkid_probe_set_device(pr, fd, 0, 0) || blkid_do_fullprobe(pr)) {
              ret = -1;
              goto out;
      }
--

2.31.0

With this patch, 'mkfs.btrfs -f /dev/nvme0n1 /dev/sdb' output some error when /dev/nvme0n1 have 2 partitions.
ERROR: cannot wipe superblocks on /dev/nvme0n1

I am not sure this error is caused by my improper fix, or bugs lie somewhere else on mkfs.btrfs path.

But without this patch, I think, btrfs_wipe_existing_sb will not do what is supposed to do -- wiping out existing superblocks. Instead, it does nothing and always returns 1. Since it has no effect and nobody complaints about it, that brings a question, do we really need to call it in btrfs_prepare_device?

Quoted 11 lines
# blkid
/dev/nvme0n1: PTUUID="93a54ce8-04b2-470b-8c05-31bfcef02f28" PTTYPE="gpt"
/dev/sda1: LABEL="OS_USB" UUID="2b7f4cb9-3dac-443f-8c96-a907b9276942" BLOCK_SIZE="512" TYPE="xfs" PARTUUID="ee58e9d3-01"
/dev/nvme0n1p1: UUID="1d94dc2b-abd1-47df-bf39-ab31cf579d29" UUID_SUB="577542b7-91f5-48d4-a54c-98cbd4525c00" BLOCK_SIZE="4096" TYPE="btrfs" PARTLABEL="primary" PARTUUID="efed009f-8ae6-4567-9cd3-80a57cdcf225"
/dev/nvme0n1p2: PARTLABEL="primary" PARTUUID="fcab66cd-daad-457f-a53c-110592d8941f"
...

Without this patch, 'mkfs.btrfs -f /dev/nvme0n1 /dev/sdb' have some issue too when /dev/nvme0n1 and /dev/sdb have some partitions.

Quoted 8 lines

some blkid of partition is still left in the output of blkid. 'blockdev --rereadpt' will let them disappear.

Best Regards
Wang Yugui ([email protected])
2021/04/09

thanks,
fox