Design of !epthook
Design of !epthook command
The most interesting feature in classic hypervisors is EPT hidden hooks.
The classic EPT Hook uses the Execute Only feature in Intel Extended Page Table (EPT).
This hook copies all the target page contents (containing the breakpoint address(es)) to a new location then puts the breakpoint on the new page.
After that, it unsets the Read and Write bits of that page but sets the Execute bit and change the physical address of that entry to the new location.
Now, if someone tries to Read or Write on that virtual address, then EPT Violation occurs, and we have a chance to bring the original physical address back to the page entry and set both Read and Write bits of that page but unset the Execute bit, so the page is not executable, we also set VMCS's Monitor Trap Flag (a.k.a. MTF) and thus after a read or write (which causes the EPT Violation), we change everything back to the executable but not readable and writable page.
This way, everything that tries to read or write on that page can conclude that it's not modified, and meanwhile, we have our breakpoints there.
Whenever breakpoint triggers, we set the Exception Bitmap of breakpoint to cause vm-exit. If the vm-exit was because of our breakpoint, then we pass it to the debugger; otherwise, we inject a #BP back to the guest. Using this way, we're sure that even the breakpoint itself is transparent from the Operating System and the target application.
You should keep in mind that exactly like classic debugger's breakpoints, you have to put breakpoints (0xcc) to the addresses that are aligned by instruction length, or in other words, you should put a breakpoint on the first byte of the instructions; otherwise, it changes the instruction to a new instruction and causes errors.
This feature works on vmx root-mode.