zfs 0.7 hangs receiving 16M block (reproducible)

[fling-]

System information

Type	Version/Name
Distribution Name	gentoo
Distribution Version	17.1
Linux Kernel	4.15.3
Architecture	amd64
ZFS Version	0.7.6-1
SPL Version	0.7.6-1

Describe the problem you're observing

zfs hangs receiving 16M blocks when I'm trying to migrate data between pools
Looks like I'm stuck in the middle of migration as I can't send without -L because of #6224

Describe how to reproduce the problem

Using https://gist.github.com/fling-/c66bf1e4a082b5cf9cd4d1106fe6e2bc

Include any warning/errors/backtraces from the system logs

# zfs send -LRe studio/gentoo@old-pool | mbuffer -L -m 512M | zfs recv -u new-root/gentoo
in @  0.0 KiB/s, out @  0.0 KiB/s, 2248 MiB total, buffer 100% full^C

recv hangs in D state:

952 pts/11   D+     0:03 zfs recv -u new-root/gentoo

[zrav]

If it is acceptable to re-send all the data using <=128k blocks, as a workaround you could restart sending from scratch without -L. #6224 only occurs when you incrementally send small blocks with large blocks already present on the target.

[behlendorf]

@fling- thank you for the excellent reproducer, that definitely helped in reproducing the issue and identify the root cause. I've opened #7404 with a proposed fix.