Repeated memory write callbacks

[archercreat]

Hello. While tracking memory writes in heavily packed executable to some memory region I've found that in some cases the callback is triggered once for 8 byte write and at the same time 8 times for 1 byte write.
Unfortunately I cannot reproduce this behavior on simple example, so maybe someone could point on what is the possible error might be.
Example trace:

write size, value, disassembly
size: 8 value: 0xfffffffffffff7ff 0x1400cd26d: mov      qword ptr [r10], r11
size: 1 value: 0x00000000000000ff 0x1400cd26d: mov      qword ptr [r10], r11
size: 1 value: 0x00000000000000f7 0x1400cd26d: mov      qword ptr [r10], r11
size: 1 value: 0x00000000000000ff 0x1400cd26d: mov      qword ptr [r10], r11
size: 1 value: 0x00000000000000ff 0x1400cd26d: mov      qword ptr [r10], r11
size: 1 value: 0x00000000000000ff 0x1400cd26d: mov      qword ptr [r10], r11
size: 1 value: 0x00000000000000ff 0x1400cd26d: mov      qword ptr [r10], r11
size: 1 value: 0x00000000000000ff 0x1400cd26d: mov      qword ptr [r10], r11
size: 1 value: 0x00000000000000ff 0x1400cd26d: mov      qword ptr [r10], r11
size: 8 value: 0xffffffff000165a1 0x140007fad: push     rdi
size: 4 value: 0x00000000fffcb99d 0x140007fae: xor      dword ptr [rsp], eax
size: 8 value: 0x00000001400aae62 0x14009194e: push     rbx

You can see that only size: 8 value: 0xfffffffffffff7ff 0x1400cd26d: mov qword ptr [r10], r11 should be present but for some reason it triggers callback 8 more times.

[wtdcode]

If the address is badly aligned, then a read/write is probably split into multiple aligned read/write. In worst cases on some arch, this means byte by byte read/write.

…

________________________________ From: Pavel ***@***.***> Sent: Wednesday, May 25, 2022 4:17:03 PM To: unicorn-engine/unicorn ***@***.***> Cc: Subscribed ***@***.***> Subject: [unicorn-engine/unicorn] Repeated memory write callbacks (Issue #1627) Hello. While tracking memory writes in heavily packed executable to some memory region I've found that in some cases the callback is triggered once for 8 byte write and at the same time 8 times for 1 byte write. Unfortunately I cannot reproduce this behavior on simple example, so maybe someone could point on what is the possible error might be. Example trace: write size, value, disassembly size: 8 value: 0xfffffffffffff7ff 0x1400cd26d: mov qword ptr [r10], r11 size: 1 value: 0x00000000000000ff 0x1400cd26d: mov qword ptr [r10], r11 size: 1 value: 0x00000000000000f7 0x1400cd26d: mov qword ptr [r10], r11 size: 1 value: 0x00000000000000ff 0x1400cd26d: mov qword ptr [r10], r11 size: 1 value: 0x00000000000000ff 0x1400cd26d: mov qword ptr [r10], r11 size: 1 value: 0x00000000000000ff 0x1400cd26d: mov qword ptr [r10], r11 size: 1 value: 0x00000000000000ff 0x1400cd26d: mov qword ptr [r10], r11 size: 1 value: 0x00000000000000ff 0x1400cd26d: mov qword ptr [r10], r11 size: 1 value: 0x00000000000000ff 0x1400cd26d: mov qword ptr [r10], r11 size: 8 value: 0xffffffff000165a1 0x140007fad: push rdi size: 4 value: 0x00000000fffcb99d 0x140007fae: xor dword ptr [rsp], eax size: 8 value: 0x00000001400aae62 0x14009194e: push rbx You can see that only size: 8 value: 0xfffffffffffff7ff 0x1400cd26d: mov qword ptr [r10], r11 should be present but for some reason it triggers callback 8 more times. ― Reply to this email directly, view it on GitHub<#1627>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AHJULO34FXMC4NQZS4UU37TVLYY57ANCNFSM5W5JNEVA>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[archercreat]

@wtdcode thanks for the response! Indeed, the address is unaligned in all those cases. For example 0x2ede is the write address from example above. I did expect to see 1 write callback for mov qword ptr [mem], reg but I guess I need to handle this case as well.

[rhelmot]

Hi, I ran into this issue as well. The problem isn't that the write gets split up byte-by-byte, the problem is that we see a HOOK_MEM callback for the whole word and then another for each individual byte. This is a problem if you are expecting to see a certain sequence of writes - we can hack around the byte-by-byte behavior, but we can't hack around duplication.

[github-actions]

This issue is stale because it has been open 60 days with no activity. Remove stale label or comment or this will be closed in 15 days.