| 1 | //------------------------------------------------------------------------------ |
| 2 | // Copyright (c) 2016 by Lukasz Janyst <lukasz@jany.st> |
| 3 | //------------------------------------------------------------------------------ |
| 4 | // This file is part of thread-bites. |
| 5 | // |
| 6 | // thread-bites is free software: you can redistribute it and/or modify |
| 7 | // it under the terms of the GNU General Public License as published by |
| 8 | // the Free Software Foundation, either version 3 of the License, or |
| 9 | // (at your option) any later version. |
| 10 | // |
| 11 | // thread-bites is distributed in the hope that it will be useful, |
| 12 | // but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | // GNU General Public License for more details. |
| 15 | // |
| 16 | // You should have received a copy of the GNU General Public License |
| 17 | // along with thread-bites. If not, see <http://www.gnu.org/licenses/>. |
| 18 | //------------------------------------------------------------------------------ |
| 19 | |
| 20 | #include <asm/unistd_64.h> // for syscall numbers |
| 21 | |
| 22 | //------------------------------------------------------------------------------ |
| 23 | // All this has been derived from glibc's sysdeps/unix/sysv/linux/x86_64/clone.S |
| 24 | // the Linux kernel sources, and a bunch of websites. |
| 25 | // |
| 26 | // We need to make a function callable from C that calls the clone syscall and |
| 27 | // and starts the user function on the new task. We do this in assembly |
| 28 | // to properly set up the call frame information (CFI) for debugging, |
| 29 | // backtraces, exceptions and such. |
| 30 | // |
| 31 | // The C call has the following format: |
| 32 | // |
| 33 | // int tbclone(int (*fn)(void *arg), void *arg, int flags, void *child_stack |
| 34 | // pid_t *parent_tidptr, pid_t *child_tidptr, void *tls), |
| 35 | // |
| 36 | // This results with the registers having the following values: |
| 37 | // |
| 38 | // rdi: fn |
| 39 | // rsi: arg |
| 40 | // rdx: flags |
| 41 | // rcx: child_stack |
| 42 | // r8: parent TID pointer |
| 43 | // r9: child TID pointer |
| 44 | // %rsp+8: TLS pointer |
| 45 | // |
| 46 | // The syscall has the following interface: |
| 47 | // |
| 48 | // SYSCALL_DEFINE5(clone, |
| 49 | // unsigned long, clone_flags, |
| 50 | // unsigned long, newsp, |
| 51 | // int __user *, parent_tidptr, |
| 52 | // int __user *, child_tidptr, |
| 53 | // unsigned long, tls) |
| 54 | // |
| 55 | // So we need the registers to be: |
| 56 | // |
| 57 | // rax: __NR_clone |
| 58 | // rdi: flags |
| 59 | // rsi: child_stack |
| 60 | // rdx: parent TID pointer |
| 61 | // r10: child TID pointer |
| 62 | // r8: TLS pointer |
| 63 | // |
| 64 | //------------------------------------------------------------------------------ |
| 65 | |
| 66 | .text |
| 67 | |
| 68 | .global tbclone // sets the symbol as externally linkable |
| 69 | .type tbclone,@function // declare this symbol as a function in ELF |
| 70 | .align 16 // place the next instruction at an address |
| 71 | // divisible by 16 in the resulting ELF binary |
| 72 | tbclone: |
| 73 | .cfi_startproc // cfi_* is the stuff for exception frames |
| 74 | subq $16, %rcx // decrement the new stack pointer by 16 because |
| 75 | // we will pop stuff from it later |
| 76 | movq %rsi, 8(%rcx) // store the function argument in the child's stack |
| 77 | movq %rdi, 0(%rcx) // store the function pointer in the child's stack |
| 78 | |
| 79 | movq $__NR_clone, %rax // clone syscall number to rax |
| 80 | movq %rdx, %rdi // flags to rdi |
| 81 | movq %rcx, %rsi // child_stack to rsi |
| 82 | movq %r8, %rdx // parent_tid to rdx |
| 83 | movq %r9, %r10 // child_tid to r10 |
| 84 | movq 8(%rsp), %r8 // TLS pointer to r8 |
| 85 | .cfi_endproc // end FDE now to get proper unwind info in the |
| 86 | // child |
| 87 | syscall // call sys_clone |
| 88 | testq %rax, %rax // compare %rax && %rax to zero |
| 89 | jz .Lstart_thread // 0 == we're in the child |
| 90 | ret // we're in the parent with either an error or |
| 91 | // the child's TID |
| 92 | |
| 93 | .Lstart_thread: // ELF local label |
| 94 | .cfi_startproc |
| 95 | .cfi_undefined rip // previous value of the instruction pointer cannot |
| 96 | // be restored anymore |
| 97 | xorq %rbp, %rbp // clear the frame pointer |
| 98 | |
| 99 | popq %rax // pop the function pointer |
| 100 | popq %rdi // pop the argument |
| 101 | call *%rax // call the function |
| 102 | |
| 103 | movq %rax, %rdi // whatever the function returned goes to rdi |
| 104 | movq $__NR_exit, %rax // exit syscall number to %rax |
| 105 | syscall // call sys_exit |
| 106 | .cfi_endproc |
| 107 | |