1 /*- 2 * Copyright (c) 2004 Tim J. Robbins 3 * Copyright (c) 2003 Peter Wemm 4 * Copyright (c) 2002 Doug Rabson 5 * Copyright (c) 1998-1999 Andrew Gallatin 6 * Copyright (c) 1994-1996 Søren Schmidt 7 * All rights reserved. 8 * Copyright (c) 2013, 2021 Dmitry Chagin <dchagin@FreeBSD.org> 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer 15 * in this position and unchanged. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. The name of the author may not be used to endorse or promote products 20 * derived from this software without specific prior written permission 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #define __ELF_WORD_SIZE 64 38 39 #include <sys/param.h> 40 #include <sys/exec.h> 41 #include <sys/imgact.h> 42 #include <sys/imgact_elf.h> 43 #include <sys/kernel.h> 44 #include <sys/ktr.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/module.h> 48 #include <sys/mutex.h> 49 #include <sys/proc.h> 50 #include <sys/stddef.h> 51 #include <sys/syscallsubr.h> 52 #include <sys/sysctl.h> 53 #include <sys/sysent.h> 54 55 #include <vm/pmap.h> 56 #include <vm/vm.h> 57 #include <vm/vm_map.h> 58 #include <vm/vm_page.h> 59 60 #include <machine/cpu.h> 61 #include <machine/md_var.h> 62 #include <machine/pcb.h> 63 #include <machine/specialreg.h> 64 #include <machine/trap.h> 65 66 #include <x86/linux/linux_x86.h> 67 #include <amd64/linux/linux.h> 68 #include <amd64/linux/linux_proto.h> 69 #include <compat/linux/linux_elf.h> 70 #include <compat/linux/linux_emul.h> 71 #include <compat/linux/linux_fork.h> 72 #include <compat/linux/linux_ioctl.h> 73 #include <compat/linux/linux_mib.h> 74 #include <compat/linux/linux_misc.h> 75 #include <compat/linux/linux_signal.h> 76 #include <compat/linux/linux_sysproto.h> 77 #include <compat/linux/linux_util.h> 78 #include <compat/linux/linux_vdso.h> 79 80 #include <x86/linux/linux_x86_sigframe.h> 81 82 _Static_assert(sizeof(struct l_fpstate) == 83 sizeof(__typeof(((mcontext_t *)0)->mc_fpstate)), 84 "fxsave area size incorrect"); 85 86 MODULE_VERSION(linux64, 1); 87 88 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2 89 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \ 90 LINUX_VDSOPAGE_SIZE) 91 #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE) 92 /* 93 * PAGE_SIZE - the size 94 * of the native SHAREDPAGE 95 */ 96 #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48 97 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \ 98 sizeof(struct ps_strings)) 99 100 static int linux_szsigcode; 101 static vm_object_t linux_vdso_obj; 102 static char *linux_vdso_mapping; 103 extern char _binary_linux_vdso_so_o_start; 104 extern char _binary_linux_vdso_so_o_end; 105 static vm_offset_t linux_vdso_base; 106 107 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 108 extern const char *linux_syscallnames[]; 109 110 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 111 112 static void linux_vdso_install(const void *param); 113 static void linux_vdso_deinstall(const void *param); 114 static void linux_vdso_reloc(char *mapping, Elf_Addr offset); 115 static void linux_set_syscall_retval(struct thread *td, int error); 116 static int linux_fetch_syscall_args(struct thread *td); 117 static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 118 uintptr_t stack); 119 static void linux_exec_sysvec_init(void *param); 120 static int linux_on_exec_vmspace(struct proc *p, 121 struct image_params *imgp); 122 static void linux_set_fork_retval(struct thread *td); 123 static int linux_vsyscall(struct thread *td); 124 125 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 126 LINUX_VDSO_SYM_CHAR(linux_platform); 127 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base); 128 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector); 129 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector); 130 131 /* 132 * According to the Intel x86 ISA 64-bit syscall 133 * saves %rip to %rcx and rflags to %r11. Registers on syscall entry: 134 * %rax system call number 135 * %rcx return address 136 * %r11 saved rflags 137 * %rdi arg1 138 * %rsi arg2 139 * %rdx arg3 140 * %r10 arg4 141 * %r8 arg5 142 * %r9 arg6 143 * 144 * Then FreeBSD fast_syscall() move registers: 145 * %rcx -> trapframe.tf_rip 146 * %r10 -> trapframe.tf_rcx 147 */ 148 static int 149 linux_fetch_syscall_args(struct thread *td) 150 { 151 struct proc *p; 152 struct trapframe *frame; 153 struct syscall_args *sa; 154 155 p = td->td_proc; 156 frame = td->td_frame; 157 sa = &td->td_sa; 158 159 sa->args[0] = frame->tf_rdi; 160 sa->args[1] = frame->tf_rsi; 161 sa->args[2] = frame->tf_rdx; 162 sa->args[3] = frame->tf_rcx; 163 sa->args[4] = frame->tf_r8; 164 sa->args[5] = frame->tf_r9; 165 sa->code = frame->tf_rax; 166 sa->original_code = sa->code; 167 168 if (sa->code >= p->p_sysent->sv_size) 169 /* nosys */ 170 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 171 else 172 sa->callp = &p->p_sysent->sv_table[sa->code]; 173 174 /* Restore r10 earlier to avoid doing this multiply times. */ 175 frame->tf_r10 = frame->tf_rcx; 176 /* Restore %rcx for machine context. */ 177 frame->tf_rcx = frame->tf_rip; 178 179 td->td_retval[0] = 0; 180 return (0); 181 } 182 183 static void 184 linux_set_syscall_retval(struct thread *td, int error) 185 { 186 struct trapframe *frame; 187 188 frame = td->td_frame; 189 190 switch (error) { 191 case 0: 192 frame->tf_rax = td->td_retval[0]; 193 break; 194 195 case ERESTART: 196 /* 197 * Reconstruct pc, we know that 'syscall' is 2 bytes, 198 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes. 199 * We saved this in tf_err. 200 * 201 */ 202 frame->tf_rip -= frame->tf_err; 203 break; 204 205 case EJUSTRETURN: 206 break; 207 208 default: 209 frame->tf_rax = bsd_to_linux_errno(error); 210 break; 211 } 212 213 /* 214 * Differently from FreeBSD native ABI, on Linux only %rcx 215 * and %r11 values are not preserved across the syscall. 216 * Require full context restore to get all registers except 217 * those two restored at return to usermode. 218 */ 219 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 220 } 221 222 static void 223 linux_set_fork_retval(struct thread *td) 224 { 225 struct trapframe *frame = td->td_frame; 226 227 frame->tf_rax = 0; 228 } 229 230 void 231 linux64_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos) 232 { 233 234 AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base); 235 AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature); 236 AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, linux_x86_elf_hwcap2()); 237 AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform)); 238 } 239 240 /* 241 * Reset registers to default values on exec. 242 */ 243 static void 244 linux_exec_setregs(struct thread *td, struct image_params *imgp, 245 uintptr_t stack) 246 { 247 struct trapframe *regs; 248 struct pcb *pcb; 249 register_t saved_rflags; 250 251 regs = td->td_frame; 252 pcb = td->td_pcb; 253 254 if (td->td_proc->p_md.md_ldt != NULL) 255 user_ldt_free(td); 256 257 pcb->pcb_fsbase = 0; 258 pcb->pcb_gsbase = 0; 259 clear_pcb_flags(pcb, PCB_32BIT); 260 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 261 set_pcb_flags(pcb, PCB_FULL_IRET); 262 263 saved_rflags = regs->tf_rflags & PSL_T; 264 bzero((char *)regs, sizeof(struct trapframe)); 265 regs->tf_rip = imgp->entry_addr; 266 regs->tf_rsp = stack; 267 regs->tf_rflags = PSL_USER | saved_rflags; 268 regs->tf_ss = _udatasel; 269 regs->tf_cs = _ucodesel; 270 regs->tf_ds = _udatasel; 271 regs->tf_es = _udatasel; 272 regs->tf_fs = _ufssel; 273 regs->tf_gs = _ugssel; 274 regs->tf_flags = TF_HASSEGS; 275 276 x86_clear_dbregs(pcb); 277 278 /* 279 * Drop the FP state if we hold it, so that the process gets a 280 * clean FP state if it uses the FPU again. 281 */ 282 fpstate_drop(td); 283 } 284 285 static int 286 linux_fxrstor(struct thread *td, mcontext_t *mcp, struct l_sigcontext *sc) 287 { 288 struct savefpu *fp = (struct savefpu *)&mcp->mc_fpstate[0]; 289 int error; 290 291 error = copyin(PTRIN(sc->sc_fpstate), fp, sizeof(mcp->mc_fpstate)); 292 if (error != 0) 293 return (error); 294 bzero(&fp->sv_pad[0], sizeof(fp->sv_pad)); 295 return (set_fpcontext(td, mcp, NULL, 0)); 296 } 297 298 static int 299 linux_xrstor(struct thread *td, mcontext_t *mcp, struct l_sigcontext *sc) 300 { 301 struct savefpu *fp = (struct savefpu *)&mcp->mc_fpstate[0]; 302 char *xfpustate; 303 struct proc *p; 304 uint32_t magic2; 305 int error; 306 307 p = td->td_proc; 308 mcp->mc_xfpustate_len = cpu_max_ext_state_size - sizeof(struct savefpu); 309 310 /* Legacy region of an xsave area. */ 311 error = copyin(PTRIN(sc->sc_fpstate), fp, sizeof(mcp->mc_fpstate)); 312 if (error != 0) 313 return (error); 314 bzero(&fp->sv_pad[0], sizeof(fp->sv_pad)); 315 316 /* Extended region of an xsave area. */ 317 sc->sc_fpstate += sizeof(mcp->mc_fpstate); 318 xfpustate = (char *)fpu_save_area_alloc(); 319 error = copyin(PTRIN(sc->sc_fpstate), xfpustate, mcp->mc_xfpustate_len); 320 if (error != 0) { 321 fpu_save_area_free((struct savefpu *)xfpustate); 322 uprintf("pid %d (%s): linux xrstor failed\n", p->p_pid, 323 td->td_name); 324 return (error); 325 } 326 327 /* Linux specific end of xsave area marker. */ 328 sc->sc_fpstate += mcp->mc_xfpustate_len; 329 error = copyin(PTRIN(sc->sc_fpstate), &magic2, LINUX_FP_XSTATE_MAGIC2_SIZE); 330 if (error != 0 || magic2 != LINUX_FP_XSTATE_MAGIC2) { 331 fpu_save_area_free((struct savefpu *)xfpustate); 332 uprintf("pid %d (%s): sigreturn magic2 0x%x error %d\n", 333 p->p_pid, td->td_name, magic2, error); 334 return (error); 335 } 336 337 error = set_fpcontext(td, mcp, xfpustate, mcp->mc_xfpustate_len); 338 fpu_save_area_free((struct savefpu *)xfpustate); 339 if (error != 0) { 340 uprintf("pid %d (%s): sigreturn set_fpcontext error %d\n", 341 p->p_pid, td->td_name, error); 342 } 343 return (error); 344 } 345 346 static int 347 linux_copyin_fpstate(struct thread *td, struct l_ucontext *uc) 348 { 349 mcontext_t mc; 350 351 bzero(&mc, sizeof(mc)); 352 mc.mc_ownedfp = _MC_FPOWNED_FPU; 353 mc.mc_fpformat = _MC_FPFMT_XMM; 354 355 if ((uc->uc_flags & LINUX_UC_FP_XSTATE) != 0) 356 return (linux_xrstor(td, &mc, &uc->uc_mcontext)); 357 else 358 return (linux_fxrstor(td, &mc, &uc->uc_mcontext)); 359 } 360 361 /* 362 * Copied from amd64/amd64/machdep.c 363 */ 364 int 365 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 366 { 367 struct proc *p; 368 struct l_rt_sigframe sf; 369 struct l_sigcontext *context; 370 struct trapframe *regs; 371 unsigned long rflags; 372 sigset_t bmask; 373 int error; 374 ksiginfo_t ksi; 375 376 regs = td->td_frame; 377 error = copyin((void *)regs->tf_rbx, &sf, sizeof(sf)); 378 if (error != 0) 379 return (error); 380 381 p = td->td_proc; 382 context = &sf.sf_uc.uc_mcontext; 383 rflags = context->sc_rflags; 384 385 /* 386 * Don't allow users to change privileged or reserved flags. 387 */ 388 /* 389 * XXX do allow users to change the privileged flag PSL_RF. 390 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 391 * should sometimes set it there too. tf_rflags is kept in 392 * the signal context during signal handling and there is no 393 * other place to remember it, so the PSL_RF bit may be 394 * corrupted by the signal handler without us knowing. 395 * Corruption of the PSL_RF bit at worst causes one more or 396 * one less debugger trap, so allowing it is fairly harmless. 397 */ 398 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 399 uprintf("pid %d comm %s linux mangled rflags %#lx\n", 400 p->p_pid, p->p_comm, rflags); 401 return (EINVAL); 402 } 403 404 /* 405 * Don't allow users to load a valid privileged %cs. Let the 406 * hardware check for invalid selectors, excess privilege in 407 * other selectors, invalid %eip's and invalid %esp's. 408 */ 409 if (!CS_SECURE(context->sc_cs)) { 410 uprintf("pid %d comm %s linux mangled cs %#x\n", 411 p->p_pid, p->p_comm, context->sc_cs); 412 ksiginfo_init_trap(&ksi); 413 ksi.ksi_signo = SIGBUS; 414 ksi.ksi_code = BUS_OBJERR; 415 ksi.ksi_trapno = T_PROTFLT; 416 ksi.ksi_addr = (void *)regs->tf_rip; 417 trapsignal(td, &ksi); 418 return (EINVAL); 419 } 420 421 linux_to_bsd_sigset(&sf.sf_uc.uc_sigmask, &bmask); 422 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 423 424 regs->tf_rdi = context->sc_rdi; 425 regs->tf_rsi = context->sc_rsi; 426 regs->tf_rdx = context->sc_rdx; 427 regs->tf_rbp = context->sc_rbp; 428 regs->tf_rbx = context->sc_rbx; 429 regs->tf_rcx = context->sc_rcx; 430 regs->tf_rax = context->sc_rax; 431 regs->tf_rip = context->sc_rip; 432 regs->tf_rsp = context->sc_rsp; 433 regs->tf_r8 = context->sc_r8; 434 regs->tf_r9 = context->sc_r9; 435 regs->tf_r10 = context->sc_r10; 436 regs->tf_r11 = context->sc_r11; 437 regs->tf_r12 = context->sc_r12; 438 regs->tf_r13 = context->sc_r13; 439 regs->tf_r14 = context->sc_r14; 440 regs->tf_r15 = context->sc_r15; 441 regs->tf_cs = context->sc_cs; 442 regs->tf_err = context->sc_err; 443 regs->tf_rflags = rflags; 444 445 error = linux_copyin_fpstate(td, &sf.sf_uc); 446 if (error != 0) { 447 uprintf("pid %d comm %s linux can't restore fpu state %d\n", 448 p->p_pid, p->p_comm, error); 449 return (error); 450 } 451 452 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 453 return (EJUSTRETURN); 454 } 455 456 static int 457 linux_fxsave(mcontext_t *mcp, void *ufp) 458 { 459 struct l_fpstate *fx = (struct l_fpstate *)&mcp->mc_fpstate[0]; 460 461 bzero(&fx->reserved2[0], sizeof(fx->reserved2)); 462 return (copyout(fx, ufp, sizeof(*fx))); 463 } 464 465 static int 466 linux_xsave(mcontext_t *mcp, char *xfpusave, char *ufp) 467 { 468 struct l_fpstate *fx = (struct l_fpstate *)&mcp->mc_fpstate[0]; 469 uint32_t magic2; 470 int error; 471 472 /* Legacy region of an xsave area. */ 473 fx->sw_reserved.magic1 = LINUX_FP_XSTATE_MAGIC1; 474 fx->sw_reserved.xstate_size = mcp->mc_xfpustate_len + sizeof(*fx); 475 fx->sw_reserved.extended_size = fx->sw_reserved.xstate_size + 476 LINUX_FP_XSTATE_MAGIC2_SIZE; 477 fx->sw_reserved.xfeatures = xsave_mask; 478 479 error = copyout(fx, ufp, sizeof(*fx)); 480 if (error != 0) 481 return (error); 482 ufp += sizeof(*fx); 483 484 /* Extended region of an xsave area. */ 485 error = copyout(xfpusave, ufp, mcp->mc_xfpustate_len); 486 if (error != 0) 487 return (error); 488 489 /* Linux specific end of xsave area marker. */ 490 ufp += mcp->mc_xfpustate_len; 491 magic2 = LINUX_FP_XSTATE_MAGIC2; 492 return (copyout(&magic2, ufp, LINUX_FP_XSTATE_MAGIC2_SIZE)); 493 } 494 495 static int 496 linux_copyout_fpstate(struct thread *td, struct l_ucontext *uc, char **sp) 497 { 498 size_t xfpusave_len; 499 char *xfpusave; 500 mcontext_t mc; 501 char *ufp = *sp; 502 503 get_fpcontext(td, &mc, &xfpusave, &xfpusave_len); 504 KASSERT(mc.mc_fpformat != _MC_FPFMT_NODEV, ("fpu not present")); 505 506 /* Room for fxsave area. */ 507 ufp -= sizeof(struct l_fpstate); 508 if (xfpusave != NULL) { 509 /* Room for xsave area. */ 510 ufp -= (xfpusave_len + LINUX_FP_XSTATE_MAGIC2_SIZE); 511 uc->uc_flags |= LINUX_UC_FP_XSTATE; 512 } 513 *sp = ufp = (char *)((unsigned long)ufp & ~0x3Ful); 514 515 if (xfpusave != NULL) 516 return (linux_xsave(&mc, xfpusave, ufp)); 517 else 518 return (linux_fxsave(&mc, ufp)); 519 } 520 521 /* 522 * copied from amd64/amd64/machdep.c 523 * 524 * Send an interrupt to process. 525 */ 526 static void 527 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 528 { 529 struct l_rt_sigframe sf, *sfp; 530 struct proc *p; 531 struct thread *td; 532 struct sigacts *psp; 533 char *sp; 534 struct trapframe *regs; 535 int sig, code; 536 int oonstack, issiginfo; 537 538 td = curthread; 539 p = td->td_proc; 540 PROC_LOCK_ASSERT(p, MA_OWNED); 541 sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno); 542 psp = p->p_sigacts; 543 issiginfo = SIGISMEMBER(psp->ps_siginfo, sig); 544 code = ksi->ksi_code; 545 mtx_assert(&psp->ps_mtx, MA_OWNED); 546 regs = td->td_frame; 547 oonstack = sigonstack(regs->tf_rsp); 548 549 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 550 catcher, sig, mask, code); 551 552 bzero(&sf, sizeof(sf)); 553 sf.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 554 sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size; 555 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 556 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 557 558 /* Allocate space for the signal handler context. */ 559 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 560 SIGISMEMBER(psp->ps_sigonstack, sig)) { 561 sp = (char *)td->td_sigstk.ss_sp + td->td_sigstk.ss_size; 562 } else 563 sp = (char *)regs->tf_rsp - 128; 564 565 mtx_unlock(&psp->ps_mtx); 566 PROC_UNLOCK(p); 567 568 if (linux_copyout_fpstate(td, &sf.sf_uc, &sp) != 0) { 569 uprintf("pid %d comm %s linux can't save fpu state, killing\n", 570 p->p_pid, p->p_comm); 571 PROC_LOCK(p); 572 sigexit(td, SIGILL); 573 } 574 sf.sf_uc.uc_mcontext.sc_fpstate = (register_t)sp; 575 576 /* Make room, keeping the stack aligned. */ 577 sp -= sizeof(struct l_rt_sigframe); 578 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 579 580 /* Save user context. */ 581 bsd_to_linux_sigset(mask, &sf.sf_uc.uc_sigmask); 582 sf.sf_uc.uc_mcontext.sc_mask = sf.sf_uc.uc_sigmask; 583 sf.sf_uc.uc_mcontext.sc_rdi = regs->tf_rdi; 584 sf.sf_uc.uc_mcontext.sc_rsi = regs->tf_rsi; 585 sf.sf_uc.uc_mcontext.sc_rdx = regs->tf_rdx; 586 sf.sf_uc.uc_mcontext.sc_rbp = regs->tf_rbp; 587 sf.sf_uc.uc_mcontext.sc_rbx = regs->tf_rbx; 588 sf.sf_uc.uc_mcontext.sc_rcx = regs->tf_rcx; 589 sf.sf_uc.uc_mcontext.sc_rax = regs->tf_rax; 590 sf.sf_uc.uc_mcontext.sc_rip = regs->tf_rip; 591 sf.sf_uc.uc_mcontext.sc_rsp = regs->tf_rsp; 592 sf.sf_uc.uc_mcontext.sc_r8 = regs->tf_r8; 593 sf.sf_uc.uc_mcontext.sc_r9 = regs->tf_r9; 594 sf.sf_uc.uc_mcontext.sc_r10 = regs->tf_r10; 595 sf.sf_uc.uc_mcontext.sc_r11 = regs->tf_r11; 596 sf.sf_uc.uc_mcontext.sc_r12 = regs->tf_r12; 597 sf.sf_uc.uc_mcontext.sc_r13 = regs->tf_r13; 598 sf.sf_uc.uc_mcontext.sc_r14 = regs->tf_r14; 599 sf.sf_uc.uc_mcontext.sc_r15 = regs->tf_r15; 600 sf.sf_uc.uc_mcontext.sc_cs = regs->tf_cs; 601 sf.sf_uc.uc_mcontext.sc_rflags = regs->tf_rflags; 602 sf.sf_uc.uc_mcontext.sc_err = regs->tf_err; 603 sf.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 604 sf.sf_uc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 605 606 /* Translate the signal. */ 607 sig = bsd_to_linux_signal(sig); 608 /* Fill in POSIX parts. */ 609 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig); 610 611 /* Copy the sigframe out to the user's stack. */ 612 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 613 uprintf("pid %d comm %s has trashed its stack, killing\n", 614 p->p_pid, p->p_comm); 615 PROC_LOCK(p); 616 sigexit(td, SIGILL); 617 } 618 619 fpstate_drop(td); 620 /* Build the argument list for the signal handler. */ 621 regs->tf_rdi = sig; /* arg 1 in %rdi */ 622 regs->tf_rax = 0; 623 if (issiginfo) { 624 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 625 regs->tf_rdx = (register_t)&sfp->sf_uc; /* arg 3 in %rdx */ 626 } else { 627 regs->tf_rsi = 0; 628 regs->tf_rdx = 0; 629 } 630 regs->tf_rcx = (register_t)catcher; 631 regs->tf_rsp = (long)sfp; 632 regs->tf_rip = linux_rt_sigcode; 633 regs->tf_rflags &= ~(PSL_T | PSL_D); 634 regs->tf_cs = _ucodesel; 635 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 636 PROC_LOCK(p); 637 mtx_lock(&psp->ps_mtx); 638 } 639 640 #define LINUX_VSYSCALL_START (-10UL << 20) 641 #define LINUX_VSYSCALL_SZ 1024 642 643 const unsigned long linux_vsyscall_vector[] = { 644 LINUX_SYS_gettimeofday, 645 LINUX_SYS_linux_time, 646 LINUX_SYS_linux_getcpu, 647 }; 648 649 static int 650 linux_vsyscall(struct thread *td) 651 { 652 struct trapframe *frame; 653 uint64_t retqaddr; 654 int code, traced; 655 int error; 656 657 frame = td->td_frame; 658 659 /* Check %rip for vsyscall area. */ 660 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 661 return (EINVAL); 662 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 663 return (EINVAL); 664 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 665 if (code >= nitems(linux_vsyscall_vector)) 666 return (EINVAL); 667 668 /* 669 * vsyscall called as callq *(%rax), so we must 670 * use return address from %rsp and also fixup %rsp. 671 */ 672 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 673 if (error) 674 return (error); 675 676 frame->tf_rip = retqaddr; 677 frame->tf_rax = linux_vsyscall_vector[code]; 678 frame->tf_rsp += 8; 679 680 traced = (frame->tf_flags & PSL_T); 681 682 amd64_syscall(td, traced); 683 684 return (0); 685 } 686 687 struct sysentvec elf_linux_sysvec = { 688 .sv_size = LINUX_SYS_MAXSYSCALL, 689 .sv_table = linux_sysent, 690 .sv_fixup = __elfN(freebsd_fixup), 691 .sv_sendsig = linux_rt_sendsig, 692 .sv_sigcode = &_binary_linux_vdso_so_o_start, 693 .sv_szsigcode = &linux_szsigcode, 694 .sv_name = "Linux ELF64", 695 .sv_coredump = elf64_coredump, 696 .sv_elf_core_osabi = ELFOSABI_NONE, 697 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR, 698 .sv_elf_core_prepare_notes = linux64_prepare_notes, 699 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 700 .sv_minuser = VM_MIN_ADDRESS, 701 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48, 702 .sv_usrstack = LINUX_USRSTACK_LA48, 703 .sv_psstrings = LINUX_PS_STRINGS_LA48, 704 .sv_psstringssz = sizeof(struct ps_strings), 705 .sv_stackprot = VM_PROT_ALL, 706 .sv_copyout_auxargs = __linuxN(copyout_auxargs), 707 .sv_copyout_strings = __linuxN(copyout_strings), 708 .sv_setregs = linux_exec_setregs, 709 .sv_fixlimit = NULL, 710 .sv_maxssiz = NULL, 711 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN | 712 SV_SIG_WAITNDQ | SV_TIMEKEEP, 713 .sv_set_syscall_retval = linux_set_syscall_retval, 714 .sv_fetch_syscall_args = linux_fetch_syscall_args, 715 .sv_syscallnames = linux_syscallnames, 716 .sv_shared_page_base = LINUX_SHAREDPAGE_LA48, 717 .sv_shared_page_len = PAGE_SIZE, 718 .sv_schedtail = linux_schedtail, 719 .sv_thread_detach = linux_thread_detach, 720 .sv_trap = linux_vsyscall, 721 .sv_hwcap = NULL, 722 .sv_hwcap2 = NULL, 723 .sv_onexec = linux_on_exec_vmspace, 724 .sv_onexit = linux_on_exit, 725 .sv_ontdexit = linux_thread_dtor, 726 .sv_setid_allowed = &linux_setid_allowed_query, 727 .sv_set_fork_retval = linux_set_fork_retval, 728 }; 729 730 static int 731 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp) 732 { 733 int error; 734 735 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base, 736 LINUX_VDSOPAGE_SIZE, imgp); 737 if (error == 0) 738 error = linux_on_exec(p, imgp); 739 return (error); 740 } 741 742 /* 743 * linux_vdso_install() and linux_exec_sysvec_init() must be called 744 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY). 745 */ 746 static void 747 linux_exec_sysvec_init(void *param) 748 { 749 l_uintptr_t *ktimekeep_base, *ktsc_selector; 750 struct sysentvec *sv; 751 ptrdiff_t tkoff; 752 753 sv = param; 754 amd64_lower_shared_page(sv); 755 /* Fill timekeep_base */ 756 exec_sysvec_init(sv); 757 758 tkoff = kern_timekeep_base - linux_vdso_base; 759 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 760 *ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset; 761 762 tkoff = kern_tsc_selector - linux_vdso_base; 763 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 764 *ktsc_selector = linux_vdso_tsc_selector_idx(); 765 if (bootverbose) 766 printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector); 767 768 tkoff = kern_cpu_selector - linux_vdso_base; 769 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 770 *ktsc_selector = linux_vdso_cpu_selector_idx(); 771 if (bootverbose) 772 printf("Linux x86-64 vDSO cpu_selector: %lu\n", *ktsc_selector); 773 } 774 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY, 775 linux_exec_sysvec_init, &elf_linux_sysvec); 776 777 static void 778 linux_vdso_install(const void *param) 779 { 780 char *vdso_start = &_binary_linux_vdso_so_o_start; 781 char *vdso_end = &_binary_linux_vdso_so_o_end; 782 783 linux_szsigcode = vdso_end - vdso_start; 784 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE); 785 786 linux_vdso_base = LINUX_VDSOPAGE_LA48; 787 if (hw_lower_amd64_sharedpage != 0) 788 linux_vdso_base -= PAGE_SIZE; 789 790 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base); 791 792 linux_vdso_obj = __elfN(linux_shared_page_init) 793 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 794 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode); 795 796 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base); 797 } 798 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST, 799 linux_vdso_install, NULL); 800 801 static void 802 linux_vdso_deinstall(const void *param) 803 { 804 805 __elfN(linux_shared_page_fini)(linux_vdso_obj, 806 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 807 } 808 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 809 linux_vdso_deinstall, NULL); 810 811 static void 812 linux_vdso_reloc(char *mapping, Elf_Addr offset) 813 { 814 const Elf_Ehdr *ehdr; 815 const Elf_Shdr *shdr; 816 Elf64_Addr *where, val; 817 Elf_Size rtype, symidx; 818 const Elf_Rela *rela; 819 Elf_Addr addr, addend; 820 int relacnt; 821 int i, j; 822 823 MPASS(offset != 0); 824 825 relacnt = 0; 826 ehdr = (const Elf_Ehdr *)mapping; 827 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff); 828 for (i = 0; i < ehdr->e_shnum; i++) 829 { 830 switch (shdr[i].sh_type) { 831 case SHT_REL: 832 printf("Linux x86_64 vDSO: unexpected Rel section\n"); 833 break; 834 case SHT_RELA: 835 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset); 836 relacnt = shdr[i].sh_size / sizeof(*rela); 837 } 838 } 839 840 for (j = 0; j < relacnt; j++, rela++) { 841 where = (Elf_Addr *)(mapping + rela->r_offset); 842 addend = rela->r_addend; 843 rtype = ELF_R_TYPE(rela->r_info); 844 symidx = ELF_R_SYM(rela->r_info); 845 846 switch (rtype) { 847 case R_X86_64_NONE: /* none */ 848 break; 849 850 case R_X86_64_RELATIVE: /* B + A */ 851 addr = (Elf_Addr)(offset + addend); 852 val = addr; 853 if (*where != val) 854 *where = val; 855 break; 856 case R_X86_64_IRELATIVE: 857 printf("Linux x86_64 vDSO: unexpected ifunc relocation, " 858 "symbol index %ld\n", symidx); 859 break; 860 default: 861 printf("Linux x86_64 vDSO: unexpected relocation type %ld, " 862 "symbol index %ld\n", rtype, symidx); 863 } 864 } 865 } 866 867 static Elf_Brandnote linux64_brandnote = { 868 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 869 .hdr.n_descsz = 16, 870 .hdr.n_type = 1, 871 .vendor = GNU_ABI_VENDOR, 872 .flags = BN_TRANSLATE_OSREL, 873 .trans_osrel = linux_trans_osrel 874 }; 875 876 static Elf64_Brandinfo linux_glibc2brand = { 877 .brand = ELFOSABI_LINUX, 878 .machine = EM_X86_64, 879 .compat_3_brand = "Linux", 880 .interp_path = "/lib64/ld-linux-x86-64.so.2", 881 .sysvec = &elf_linux_sysvec, 882 .interp_newpath = NULL, 883 .brand_note = &linux64_brandnote, 884 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 885 }; 886 887 static Elf64_Brandinfo linux_glibc2brandshort = { 888 .brand = ELFOSABI_LINUX, 889 .machine = EM_X86_64, 890 .compat_3_brand = "Linux", 891 .interp_path = "/lib64/ld-linux.so.2", 892 .sysvec = &elf_linux_sysvec, 893 .interp_newpath = NULL, 894 .brand_note = &linux64_brandnote, 895 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 896 }; 897 898 static Elf64_Brandinfo linux_muslbrand = { 899 .brand = ELFOSABI_LINUX, 900 .machine = EM_X86_64, 901 .compat_3_brand = "Linux", 902 .interp_path = "/lib/ld-musl-x86_64.so.1", 903 .sysvec = &elf_linux_sysvec, 904 .interp_newpath = NULL, 905 .brand_note = &linux64_brandnote, 906 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE | 907 LINUX_BI_FUTEX_REQUEUE 908 }; 909 910 Elf64_Brandinfo *linux_brandlist[] = { 911 &linux_glibc2brand, 912 &linux_glibc2brandshort, 913 &linux_muslbrand, 914 NULL 915 }; 916 917 static int 918 linux64_elf_modevent(module_t mod, int type, void *data) 919 { 920 Elf64_Brandinfo **brandinfo; 921 int error; 922 struct linux_ioctl_handler **lihp; 923 924 error = 0; 925 926 switch(type) { 927 case MOD_LOAD: 928 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 929 ++brandinfo) 930 if (elf64_insert_brand_entry(*brandinfo) < 0) 931 error = EINVAL; 932 if (error == 0) { 933 SET_FOREACH(lihp, linux_ioctl_handler_set) 934 linux_ioctl_register_handler(*lihp); 935 stclohz = (stathz ? stathz : hz); 936 if (bootverbose) 937 printf("Linux x86-64 ELF exec handler installed\n"); 938 } else 939 printf("cannot insert Linux x86-64 ELF brand handler\n"); 940 break; 941 case MOD_UNLOAD: 942 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 943 ++brandinfo) 944 if (elf64_brand_inuse(*brandinfo)) 945 error = EBUSY; 946 if (error == 0) { 947 for (brandinfo = &linux_brandlist[0]; 948 *brandinfo != NULL; ++brandinfo) 949 if (elf64_remove_brand_entry(*brandinfo) < 0) 950 error = EINVAL; 951 } 952 if (error == 0) { 953 SET_FOREACH(lihp, linux_ioctl_handler_set) 954 linux_ioctl_unregister_handler(*lihp); 955 if (bootverbose) 956 printf("Linux x86_64 ELF exec handler removed\n"); 957 } else 958 printf("Could not deinstall Linux x86_64 ELF interpreter entry\n"); 959 break; 960 default: 961 return (EOPNOTSUPP); 962 } 963 return (error); 964 } 965 966 static moduledata_t linux64_elf_mod = { 967 "linux64elf", 968 linux64_elf_modevent, 969 0 970 }; 971 972 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 973 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 974 FEATURE(linux64, "Linux 64bit support"); 975