Patch unaligned 64-bit atomic operation panic

[silentmurdock]

Fix unaligned 64-bit atomic operation panic to support 32-bit architectures too.

[anacrolix]

Damn, does it really take that much?!

[anacrolix]

Was moving the stat block to the beginning of the struct not enough? What if it's detached and pointed to instead of embedded?

[silentmurdock]

You're right, I looked through the code a little more and found that if i move the "traversal" block to the beginning of "Announce" struct in "announce.go" that also solves the problem. Working on 32-bit systems.

I am revoking this pull request.
Please modify the code.

[silentmurdock]

A simpler solution is also available.

[anacrolix]

@silentmurdock I commented about this in anacrolix/torrent#483 (comment) but got feedback to the contrary from @compliment . Can you confirm the patch in the linked comment fix this for you on the platforms you're using?

[compliment]

You're right, I looked through the code a little more and found that if i move the "traversal" block to the beginning of "Announce" struct in "announce.go" that also solves the problem. Working on 32-bit systems.

This is fd60ad0
I tried it with with https://github.com/anacrolix/torrent/tree/master/cmd/torrent for linux 386 and got

panic: unaligned 64-bit atomic operation

goroutine 795 [running]:
runtime/internal/atomic.panicUnaligned()
	runtime/internal/atomic/unaligned.go:8 +0x2d
runtime/internal/atomic.Xadd64(0x9e321e4, 0x1, 0x0, 0x0, 0x0)
	runtime/internal/atomic/asm_386.s:107 +0x11
github.com/anacrolix/dht/v2.(*traversal).wrapQuery(0x9e321e4, 0x86c4250, 0x9c753c0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, ...)
	github.com/anacrolix/dht/v2@v2.9.2-0.20210509011734-fd60ad06b482/traversal.go:115 +0x52
github.com/anacrolix/dht/v2.(*traversal).run.func1.1(0x9c0e008)
	github.com/anacrolix/dht/v2@v2.9.2-0.20210509011734-fd60ad06b482/traversal.go:156 +0x35
github.com/anacrolix/dht/v2.(*traversal).beginQuery.func1.1(0x0)
	github.com/anacrolix/dht/v2@v2.9.2-0.20210509011734-fd60ad06b482/traversal.go:179 +0x93
github.com/anacrolix/dht/v2.(*Server).beginQuery.func1.1()
	github.com/anacrolix/dht/v2@v2.9.2-0.20210509011734-fd60ad06b482/server.go:745 +0x23
created by github.com/anacrolix/dht/v2.(*traversal).run
	github.com/anacrolix/dht/v2@v2.9.2-0.20210509011734-fd60ad06b482/traversal.go:164 +0x83 

Are you sure this worked for you and you didn't have your own patch applied?

[anacrolix]

Thanks for chasing this up. I'll follow this up when you guys determine the best way forward.

[silentmurdock]

Are you sure this worked for you and you didn't have your own patch applied?

The fd60ad0 didn’t work for me either, I used my own patch solution first: torrent-issue-483_solution_2.zip

Anacrolix's solution didn't work before, because he modified only the "traversal.go" file, but "announce.go" has an other struct that need to be modified. I extended his solution, so it works now: torrent-issue-483_solution_1.zip

Both versions work, but I think Anacrolix's modified solution is simpler and faster.
Please write over the specified files and test it.

Which one do we use?

[anacrolix]

I think mine is simpler. It would be great if this was checked in CI. It's recently moved to GitHub actions if you're keen/know how to implement that.

[silentmurdock]

Unfortunately, CircleCI does not contain 32-bit images. I don’t feel so confident in using CI. Maybe if we compile the 32-bit code on a 64-bit system image and run the tests there. (Example - run: GOOS = linux GOARCH = arm GOARM = 6)

[silentmurdock]

@anacrolix Can you add the changes because my app doesn't work without it? I don’t want to use a vendor library in my project. I tested it manually on arm, linux, windows 32 bit systems. I will be working on a CI test solution in the future.
Or do I restore the branch, modify, and create a new pull requests?