fix panic on debug build

[ikozhukhov]

i have panic on DEBUG build on DilOS with test send_mixed_raw

panic[cpu3]/thread=fffffe0bf62f6c40:
assertion failed: (!copied) is equivalent to (hdr->b_l1hdr.b_freeze_cksum == NULL), file: ../../common/fs/zfs/arc.c, line: 1839


fffffe0012014700 genunix:process_type+1946ed ()
fffffe0012014730 zfs:arc_buf_try_copy_decompressed_data+b7 ()
fffffe00120147c0 zfs:arc_buf_fill+2e3 ()
fffffe0012014810 zfs:arc_untransform+25 ()
fffffe00120148a0 zfs:dbuf_read+4a1 ()
fffffe0012014910 zfs:dmu_buf_hold+80 ()
fffffe00120149f0 zfs:zap_lockdir+5e ()
fffffe0012014aa0 zfs:zap_lookup_norm+6c ()
fffffe0012014b00 zfs:zap_lookup+36 ()
fffffe0012014b60 zfs:zfs_get_zplprop+fc ()
fffffe0012014bc0 zfs:nvl_add_zplprop+36 ()
fffffe0012014c10 zfs:zfs_ioc_objset_zplprops+f5 ()
fffffe0012014cc0 zfs:zfsdev_ioctl+524 ()
fffffe0012014d00 genunix:cdev_ioctl+25 ()
fffffe0012014d50 specfs:spec_ioctl+4d ()
fffffe0012014de0 genunix:fop_ioctl+55 ()
fffffe0012014ef0 genunix:ioctl+a4 ()
fffffe0012014f00 unix:brand_sys_syscall+320 ()

Motivation and Context

Description

based on research and comments, we should to check b_freeze_cksum in loop

How Has This Been Tested?

test send_mixed_raw PASS after changes

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Performance enhancement (non-breaking change which improves efficiency)
• Code cleanup (non-breaking change which makes code smaller or more readable)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation (a change to man pages or other documentation)

Checklist:

• My code follows the ZFS on Linux code style requirements.
• I have updated the documentation accordingly.
• I have read the contributing document.
• I have added tests to cover my changes.
• All new and existing tests passed.
• All commit messages are properly formatted and contain Signed-off-by.

[codecov]

Codecov Report

Merging #8736 into master will decrease coverage by 0.05%.
The diff coverage is 50%.

@@            Coverage Diff             @@
##           master    #8736      +/-   ##
==========================================
- Coverage   78.97%   78.92%   -0.06%     
==========================================
  Files         381      381              
  Lines      117797   117802       +5     
==========================================
- Hits        93034    92978      -56     
- Misses      24763    24824      +61

Flag	Coverage Δ
#kernel	79.4% <25%> (-0.02%)	⬇️
#user	67.9% <50%> (+0.14%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f378f42...919a496. Read the comment docs.

[behlendorf]

@ikozhukhov given that b_l1hdr.b_freeze_cksum can only be set when ZFS_DEBUG_MODIFY is enabled (see arc_cksum_compute(). I think this EQUIV is only valid when ZFS_DEBUG_MODIFY is set, and we should instead wrap it with a conditional that checks for the flag. For the same reason, the ASSERT after the call to arc_buf_try_copy_decompressed_data() in arc_buf_fill() looks wrong to me as well.

My suspicion is we've never seen this in testing because by default for debug builds we enable ZFS_DEBUG_MODIFY. It would be great to get @grwilson's thoughts on this, since I may be misunderstanding the intent of b_freeze_cksum here.

[grwilson]

Yes, b_freeze_cksum should only have data when ZFS_DEBUG_MODIFY is enabled. So if I understand the original problem then wrapping the EQUIV with if (zfs_flags & ZFS_DEBUG_MODIFY) would fix the problem.