A FreeBSD user reported a problem with doing a force receive of a full stream
(non-incremental) while using a debug build:
panic: solaris assert: dbuf_is_metadata(db) == arc_is_metadata(buf) (0x0 == 0x1)
assfail3()
dbuf_assign_arcbuf()
dmu_assign_arcbuf()
receive_writer_thread()
Here is a corresponding write record header:
header = {
drr_type = DRR_WRITE,
drr_payloadlen = 0,
drr_u = {
...
drr_write = {
drr_object = 357,
drr_type = DMU_OT_DIRECTORY_CONTENTS,
drr_pad = 0,
drr_offset = 0,
drr_logical_size = 512,
drr_toguid = 6461692233939371501,
drr_checksumtype = 2 '\002',
drr_checksumflags = 0 '\000',
drr_compressiontype = 0 '\000',
drr_pad2 = "\000\000\000\000",
drr_key = {
ddk_cksum = {
zc_word = {0, 0, 0, 0}
},
ddk_prop = 0
},
drr_compressed_size = 0
},
Here is the ARC header:
$6 = {
b_dva = {
dva_word = {0, 0}
},
b_birth = 0,
b_type = ARC_BUFC_METADATA,
b_hash_next = 0x0 <avl_balance2child>,
b_flags = (ARC_FLAG_BUFC_METADATA | ARC_FLAG_HAS_L1HDR |
ARC_FLAG_COMPRESSED_ARC | ARC_FLAG_SHARED_DATA | ARC_FLAG_COMPRESS_1),
b_psize = 1,
b_lsize = 1,
b_spa = 7546455437792279428,
...
Here is the dnode:
dn_objset = 0xfffff800815f1800,
dn_object = 357,
dn_dbuf = 0xfffff800826da6b8,
dn_handle = 0xfffff800815ca208,
dn_phys = 0xfffffe0002d1aa00,
dn_type = DMU_OT_PLAIN_FILE_CONTENTS,
dn_bonuslen = 204,
dn_bonustype = 44 ',',
dn_nblkptr = 1 '\001',
dn_checksum = 0 '\000',
dn_compress = 0 '\000',
dn_nlevels = 1 '\001',
dn_indblkshift = 17 '\021',
dn_datablkshift = 9 '\t',
dn_moved = 0 '\000',
dn_datablkszsec = 1,
dn_datablksz = 512,
dn_maxblkid = 0,
dn_next_type = "\000\000\000",
dn_next_nblkptr = "\000\000\000",
dn_next_nlevels = "\000\000\000",
dn_next_indblkshift = "\000\000\000\021",
dn_next_bonustype = "\000\000\000,",
dn_rm_spillblk = "\000\000\000",
dn_next_bonuslen = {0, 0, 0, 204},
dn_next_blksz = {0, 0, 0, 512},
dn_dbufs_count = 2,
dn_allocated_txg = 199719,
dn_free_txg = 0,
dn_assigned_txg = 199719,
dn_dirtyctx = DN_DIRTY_OPEN,
tx_txg = 199719 as well.
The problem started happening after the compressed send / receive feature was
introduced. But the feature was not used when the crash happened.
I think that that being a force receive we are probably changing the object type
to a different one (from a plain file to a directory), but something goes wrong
with that. Alternatively, maybe there is an issue with setting
drr_write.drr_type on the sender side.
I think that previously we didn't verify that drr_write.drr_type matches actual
dn_type, so the problem could be obscured.
Do you have any suggestions?
Thank you!
--
Andriy Gapon
------------------------------------------
openzfs-developer
Archives: https://openzfs.topicbox.com/groups/developer/
Powered by Topicbox: https://topicbox.com
