zdb -ddddddd fails on newly created zpool

[rudis]

System information

Type	Version/Name
Distribution Name	Debian
Distribution Version	Stretch
Linux Kernel	4.9.110
Architecture	amd64
ZFS Version	0.7.11-1~bpo9-1
SPL Version	0.7.11-1~bpo9-1

Describe the problem you're observing

zdb -ddddddd fails on newly created pool

Describe how to reproduce the problem

# zpool create test raidz2 /dev/lvm/zfs{1,2,3,4}
# cd /test
# echo hi > x
# ls -i x
2 x
# zdb -dddddddddddd test 2
Dataset mos [META], ID 0, cr_txg 4, 349K, 43 objects, rootbp DVA[0]=<0:12f000:3000> DVA[1]=<0:260012c000:3000> DVA[2]=<0:4c000d2000:3000> [L0 DMU objset] fletcher4 uncompressed LE contiguous unique triple size=800L/800P birth=9L/9P fill=43 cksum=366b3f29c:647471174cf:5cef7c72f0d45:39743ba7fcabe1d

    Object  lvl   iblk   dblk  dsize  dnsize  lsize   %full  type
zdb: dmu_bonus_hold(2) failed, errno 2
#

I'm also confused by the fact that fletcher4 is displayed, even if I manually select sha512 as hash before creating the file.

[loli10K]

zdb -d $POOLNAME <num> will lookup object <num> in the pool MOS, you want zdb -d $POOLNAME/ <num> to reference objects in the "$POOLNAME" dataset:

root@linux:~# truncate -s 128m /var/tmp/{1,2,3,4}
root@linux:~# zpool create -f test -O checksum=sha512 raidz2 /var/tmp/{1,2,3,4}
root@linux:~# cd /test
root@linux:/test# echo hi > x
root@linux:/test# ls -i x
2 x
root@linux:/test# zdb -dddddddddddd test 2
Dataset mos [META], ID 0, cr_txg 4, 76.2K, 43 objects, rootbp DVA[0]=<0:27c00:600> DVA[1]=<0:6027600:600> DVA[2]=<0:c013e00:600> [L0 DMU objset] fletcher4 lz4 LE contiguous unique triple size=800L/200P birth=7L/7P fill=43 cksum=c5bbe111c:41f4276f87a:b9f3a8cdad3f:17070d262477b3

    Object  lvl   iblk   dblk  dsize  dnsize  lsize   %full  type
zdb: dmu_bonus_hold(2) failed, errno 2
root@linux:/test# zdb -dddddddddddd test/ 2
Dataset test [ZPL], ID 51, cr_txg 1, 33.6K, 7 objects, rootbp DVA[0]=<0:26400:600> DVA[1]=<0:6025e00:600> [L0 DMU objset] sha512 lz4 LE contiguous unique double size=800L/200P birth=7L/7P fill=7 cksum=e91efb4b1f2d8466:eb00002e2c0e9fc7:64981c64563b217f:edb0e231dfcdadbf

    Object  lvl   iblk   dblk  dsize  dnsize  lsize   %full  type
         2    1   128K    512    512     512    512  100.00  ZFS plain file (K=inherit) (Z=inherit)
                                               168   bonus  System attributes
	dnode flags: USED_BYTES USERUSED_ACCOUNTED USEROBJUSED_ACCOUNTED 
	dnode maxblkid: 0
	path	/x
	uid     0
	gid     0
	atime	Thu Oct  4 12:36:12 2018
	mtime	Thu Oct  4 12:36:12 2018
	ctime	Thu Oct  4 12:36:12 2018
	crtime	Thu Oct  4 12:36:12 2018
	gen	7
	mode	100644
	size	3
	parent	34
	links	1
	pflags	40800000004
Indirect blocks:
               0 L0 0:20a00:600 200L/200P F=1 B=7/7

		segment [0000000000000000, 0000000000000200) size   512

root@linux:/test# 

[rudis]

Thank you! pool/ works fine.

Would it be possible to mention this in the man page and improve the error message so the difference between pool and pool/ is clear?

[DeHackEd]

It is.

[rudis]

Ah, sorry. Missed that.

Regarding the error when using the pool instead of the dataset. Is it expected that it returns (to me) arbitrary looking data before dying with this error? I know zdb is just a debug tool, but it would be nice to catch this user error.

[rudis]

Oh, and maybe an example with just the root dataset would be nice. I know it's documented, but I quickly skimmed over the examples and Example 3 looked like I was looking for, except I missed the trailing /.

[rincebrain]

Hm.

When I tried this while discussing it last night, I used zdb -vvvvvvv testpool 2 and got the same error.

(Also on ZoL 0.7.11 from Debian stretch backports.)

Many hours later:

# zdb -vvvvvvv 'testpool/' 2
Dataset testpool [ZPL], ID 51, cr_txg 1, 37.4K, 9 objects, rootbp DVA[0]=<0:17a600:600> DVA[1]=<0:60179400:600> [L0 DMU objset] fletcher4 lz4 LE contiguous unique double size=800L/200P birth=97L/97P fill=9 cksum=c7b732a36:4cd8c0903c0:f4be20942ca7:2190a8426be727

    Object  lvl   iblk   dblk  dsize  dnsize  lsize   %full  type
zdb: dmu_bonus_hold(2) failed, errno 2

ZFS_DBGMSG(zdb):
#

(I too had never known that particular bit of esoterica involving the root dataset, but I also almost never store things there, so...)

[Lady-Galadriel]

Perhaps Example 3 should be for the "root" dataset with a trailing / (slash);

  Example 3 Display basic information about object 2 in rpool top level dataset

        # zdb -d rpool/ 2

And all the other examples renumbered.

Further, it would be nicer if the error message returned made more sense for the actual issue. For example;

zdb: object not found, errno X

(Add me to learning something new about ZFS today...)
Edit: Corrected spelling and English syntax.

[loli10K]

When I tried this while discussing it last night, I used zdb -vvvvvvv testpool 2 and got the same error.
(Also on ZoL 0.7.11 from Debian stretch backports.)

@rincebrain if you are able to reproduce this on newly created pools could you please create a sample on a small file-vdev pool (like in my example) and upload it here?

[rincebrain]

Sure, here you are.

(Regardless of which file, both contain 4 sparse files of 2GB each that have about 11 MB of nonzero data between them. zdb -vvvvvvv testpool/ 2 reproducibly does the above for me even after export+import. The directory on the pool isn't important, I was just trying to see if I could make it stop reproducing.)

7986_pool_ex.zip

And if that doesn't work for some reason, you can find a .tgz here, since Github claims it won't let me upload a tgz.

[loli10K]

@rincebrain unless i am missing something obvious your reproducer (zip-compressed pool files) have no file with inum=2 so that error is expected:

root@linux:~# ls -li /testpool/
total 3
128 drwxr-xr-x 2 root root 2 Oct  4 09:52 foo
256 -rw-r--r-- 1 root root 3 Oct  4 09:56 monkeymadness
  3 -rw-r--r-- 1 root root 3 Oct  4 09:50 test
root@linux:~# zpool get guid testpool
NAME      PROPERTY  VALUE  SOURCE
testpool  guid      11682838301589054648  -
root@linux:~# zdb -dddddd testpool/ 3
Dataset testpool [ZPL], ID 51, cr_txg 1, 37.4K, 9 objects, rootbp DVA[0]=<0:803c600:600> DVA[1]=<0:6803c600:600> [L0 DMU objset] fletcher4 lz4 LE contiguous unique double size=800L/200P birth=15704L/15704P fill=9 cksum=c329c21a1:4b4478a7c9d:f09638a4fbb1:21277810630dc5

    Object  lvl   iblk   dblk  dsize  dnsize  lsize   %full  type
         3    1   128K    512    512     512    512  100.00  ZFS plain file (K=inherit) (Z=inherit)
                                               168   bonus  System attributes
	dnode flags: USED_BYTES USERUSED_ACCOUNTED USEROBJUSED_ACCOUNTED 
	dnode maxblkid: 0
	path	/test
	uid     0
	gid     0
	atime	Thu Oct  4 09:50:20 2018
	mtime	Thu Oct  4 09:50:20 2018
	ctime	Thu Oct  4 09:50:20 2018
	crtime	Thu Oct  4 09:50:20 2018
	gen	28
	mode	100644
	size	3
	parent	34
	links	1
	pflags	40800000004
Indirect blocks:
               0 L0 0:12c600:600 200L/200P F=1 B=28/28

		segment [0000000000000000, 0000000000000200) size   512

root@linux:~# 

[rincebrain]

Nice, I am going to assume I tried making a new file and deleted the old one.

Feel free to close it and subject me to ribbing for trying to look at bugs on too little sleep; I'll probably go file a separate enhancement request complaining about the error being opaque.

[stale]

This issue has been automatically marked as "stale" because it has not had any activity for a while. It will be closed in 90 days if no further activity occurs. Thank you for your contributions.