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/systm.h> 41 #include <sys/exec.h> 42 #include <sys/fcntl.h> 43 #include <sys/imgact.h> 44 #include <sys/imgact_elf.h> 45 #include <sys/kernel.h> 46 #include <sys/ktr.h> 47 #include <sys/lock.h> 48 #include <sys/malloc.h> 49 #include <sys/module.h> 50 #include <sys/mutex.h> 51 #include <sys/proc.h> 52 #include <sys/resourcevar.h> 53 #include <sys/stddef.h> 54 #include <sys/signalvar.h> 55 #include <sys/syscallsubr.h> 56 #include <sys/sysctl.h> 57 #include <sys/sysent.h> 58 #include <sys/sysproto.h> 59 #include <sys/vnode.h> 60 #include <sys/eventhandler.h> 61 62 #include <vm/vm.h> 63 #include <vm/pmap.h> 64 #include <vm/vm_extern.h> 65 #include <vm/vm_map.h> 66 #include <vm/vm_object.h> 67 #include <vm/vm_page.h> 68 #include <vm/vm_param.h> 69 70 #include <machine/cpu.h> 71 #include <machine/md_var.h> 72 #include <machine/pcb.h> 73 #include <machine/specialreg.h> 74 #include <machine/trap.h> 75 76 #include <x86/linux/linux_x86.h> 77 #include <amd64/linux/linux.h> 78 #include <amd64/linux/linux_proto.h> 79 #include <compat/linux/linux_emul.h> 80 #include <compat/linux/linux_fork.h> 81 #include <compat/linux/linux_ioctl.h> 82 #include <compat/linux/linux_mib.h> 83 #include <compat/linux/linux_misc.h> 84 #include <compat/linux/linux_signal.h> 85 #include <compat/linux/linux_sysproto.h> 86 #include <compat/linux/linux_util.h> 87 #include <compat/linux/linux_vdso.h> 88 89 #include <x86/linux/linux_x86_sigframe.h> 90 91 MODULE_VERSION(linux64, 1); 92 93 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2 94 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \ 95 LINUX_VDSOPAGE_SIZE) 96 #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE) 97 /* 98 * PAGE_SIZE - the size 99 * of the native SHAREDPAGE 100 */ 101 #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48 102 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \ 103 sizeof(struct ps_strings)) 104 105 static int linux_szsigcode; 106 static vm_object_t linux_vdso_obj; 107 static char *linux_vdso_mapping; 108 extern char _binary_linux_vdso_so_o_start; 109 extern char _binary_linux_vdso_so_o_end; 110 static vm_offset_t linux_vdso_base; 111 112 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 113 114 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 115 116 static int linux_copyout_strings(struct image_params *imgp, 117 uintptr_t *stack_base); 118 static int linux_fixup_elf(uintptr_t *stack_base, 119 struct image_params *iparams); 120 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 121 static void linux_vdso_install(const void *param); 122 static void linux_vdso_deinstall(const void *param); 123 static void linux_vdso_reloc(char *mapping, Elf_Addr offset); 124 static void linux_set_syscall_retval(struct thread *td, int error); 125 static int linux_fetch_syscall_args(struct thread *td); 126 static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 127 uintptr_t stack); 128 static void linux_exec_sysvec_init(void *param); 129 static int linux_on_exec_vmspace(struct proc *p, 130 struct image_params *imgp); 131 static void linux_set_fork_retval(struct thread *td); 132 static int linux_vsyscall(struct thread *td); 133 134 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 135 LINUX_VDSO_SYM_CHAR(linux_platform); 136 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base); 137 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector); 138 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector); 139 140 static int 141 linux_fetch_syscall_args(struct thread *td) 142 { 143 struct proc *p; 144 struct trapframe *frame; 145 struct syscall_args *sa; 146 147 p = td->td_proc; 148 frame = td->td_frame; 149 sa = &td->td_sa; 150 151 sa->args[0] = frame->tf_rdi; 152 sa->args[1] = frame->tf_rsi; 153 sa->args[2] = frame->tf_rdx; 154 sa->args[3] = frame->tf_rcx; 155 sa->args[4] = frame->tf_r8; 156 sa->args[5] = frame->tf_r9; 157 sa->code = frame->tf_rax; 158 sa->original_code = sa->code; 159 160 if (sa->code >= p->p_sysent->sv_size) 161 /* nosys */ 162 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 163 else 164 sa->callp = &p->p_sysent->sv_table[sa->code]; 165 166 td->td_retval[0] = 0; 167 return (0); 168 } 169 170 static void 171 linux_set_syscall_retval(struct thread *td, int error) 172 { 173 struct trapframe *frame; 174 175 frame = td->td_frame; 176 177 switch (error) { 178 case 0: 179 frame->tf_rax = td->td_retval[0]; 180 frame->tf_r10 = frame->tf_rcx; 181 break; 182 183 case ERESTART: 184 /* 185 * Reconstruct pc, we know that 'syscall' is 2 bytes, 186 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes. 187 * We saved this in tf_err. 188 * 189 */ 190 frame->tf_rip -= frame->tf_err; 191 frame->tf_r10 = frame->tf_rcx; 192 break; 193 194 case EJUSTRETURN: 195 break; 196 197 default: 198 frame->tf_rax = bsd_to_linux_errno(error); 199 frame->tf_r10 = frame->tf_rcx; 200 break; 201 } 202 203 /* 204 * Differently from FreeBSD native ABI, on Linux only %rcx 205 * and %r11 values are not preserved across the syscall. 206 * Require full context restore to get all registers except 207 * those two restored at return to usermode. 208 * 209 * XXX: Would be great to be able to avoid PCB_FULL_IRET 210 * for the error == 0 case. 211 */ 212 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 213 } 214 215 static void 216 linux_set_fork_retval(struct thread *td) 217 { 218 struct trapframe *frame = td->td_frame; 219 220 frame->tf_rax = 0; 221 } 222 223 static int 224 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) 225 { 226 Elf_Auxargs *args; 227 Elf_Auxinfo *argarray, *pos; 228 struct proc *p; 229 int error, issetugid; 230 231 p = imgp->proc; 232 args = (Elf64_Auxargs *)imgp->auxargs; 233 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 234 M_WAITOK | M_ZERO); 235 236 issetugid = p->p_flag & P_SUGID ? 1 : 0; 237 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base); 238 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 239 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 240 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 241 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 242 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 243 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 244 AUXARGS_ENTRY(pos, AT_BASE, args->base); 245 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 246 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 247 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 248 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 249 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 250 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 251 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 252 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); 253 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0); 254 if (imgp->execpathp != 0) 255 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); 256 if (args->execfd != -1) 257 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 258 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 259 AUXARGS_ENTRY(pos, AT_NULL, 0); 260 261 free(imgp->auxargs, M_TEMP); 262 imgp->auxargs = NULL; 263 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 264 265 error = copyout(argarray, (void *)base, 266 sizeof(*argarray) * LINUX_AT_COUNT); 267 free(argarray, M_TEMP); 268 return (error); 269 } 270 271 static int 272 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) 273 { 274 Elf_Addr *base; 275 276 base = (Elf64_Addr *)*stack_base; 277 base--; 278 if (suword(base, (uint64_t)imgp->args->argc) == -1) 279 return (EFAULT); 280 281 *stack_base = (uintptr_t)base; 282 return (0); 283 } 284 285 /* 286 * Copy strings out to the new process address space, constructing new arg 287 * and env vector tables. Return a pointer to the base so that it can be used 288 * as the initial stack pointer. 289 */ 290 static int 291 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 292 { 293 int argc, envc, error; 294 char **vectp; 295 char *stringp; 296 uintptr_t destp, ustringp; 297 struct ps_strings *arginfo; 298 char canary[LINUX_AT_RANDOM_LEN]; 299 size_t execpath_len; 300 struct proc *p; 301 302 p = imgp->proc; 303 arginfo = (struct ps_strings *)PROC_PS_STRINGS(p); 304 destp = (uintptr_t)arginfo; 305 306 if (imgp->execpath != NULL && imgp->auxargs != NULL) { 307 execpath_len = strlen(imgp->execpath) + 1; 308 destp -= execpath_len; 309 destp = rounddown2(destp, sizeof(void *)); 310 imgp->execpathp = (void *)destp; 311 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 312 if (error != 0) 313 return (error); 314 } 315 316 /* Prepare the canary for SSP. */ 317 arc4rand(canary, sizeof(canary), 0); 318 destp -= roundup(sizeof(canary), sizeof(void *)); 319 imgp->canary = (void *)destp; 320 error = copyout(canary, imgp->canary, sizeof(canary)); 321 if (error != 0) 322 return (error); 323 324 /* Allocate room for the argument and environment strings. */ 325 destp -= ARG_MAX - imgp->args->stringspace; 326 destp = rounddown2(destp, sizeof(void *)); 327 ustringp = destp; 328 329 if (imgp->auxargs) { 330 /* 331 * Allocate room on the stack for the ELF auxargs 332 * array. It has LINUX_AT_COUNT entries. 333 */ 334 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo); 335 destp = rounddown2(destp, sizeof(void *)); 336 } 337 338 vectp = (char **)destp; 339 340 /* 341 * Allocate room for the argv[] and env vectors including the 342 * terminating NULL pointers. 343 */ 344 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 345 346 /* 347 * Starting with 2.24, glibc depends on a 16-byte stack alignment. 348 * One "long argc" will be prepended later. 349 */ 350 vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8); 351 352 /* vectp also becomes our initial stack base. */ 353 *stack_base = (uintptr_t)vectp; 354 355 stringp = imgp->args->begin_argv; 356 argc = imgp->args->argc; 357 envc = imgp->args->envc; 358 359 /* Copy out strings - arguments and environment. */ 360 error = copyout(stringp, (void *)ustringp, 361 ARG_MAX - imgp->args->stringspace); 362 if (error != 0) 363 return (error); 364 365 /* Fill in "ps_strings" struct for ps, w, etc. */ 366 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 367 suword(&arginfo->ps_nargvstr, argc) != 0) 368 return (EFAULT); 369 370 /* Fill in argument portion of vector table. */ 371 for (; argc > 0; --argc) { 372 if (suword(vectp++, ustringp) != 0) 373 return (EFAULT); 374 while (*stringp++ != 0) 375 ustringp++; 376 ustringp++; 377 } 378 379 /* A null vector table pointer separates the argp's from the envp's. */ 380 if (suword(vectp++, 0) != 0) 381 return (EFAULT); 382 383 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 384 suword(&arginfo->ps_nenvstr, envc) != 0) 385 return (EFAULT); 386 387 /* Fill in environment portion of vector table. */ 388 for (; envc > 0; --envc) { 389 if (suword(vectp++, ustringp) != 0) 390 return (EFAULT); 391 while (*stringp++ != 0) 392 ustringp++; 393 ustringp++; 394 } 395 396 /* The end of the vector table is a null pointer. */ 397 if (suword(vectp, 0) != 0) 398 return (EFAULT); 399 400 if (imgp->auxargs) { 401 vectp++; 402 error = imgp->sysent->sv_copyout_auxargs(imgp, 403 (uintptr_t)vectp); 404 if (error != 0) 405 return (error); 406 } 407 408 return (0); 409 } 410 411 /* 412 * Reset registers to default values on exec. 413 */ 414 static void 415 linux_exec_setregs(struct thread *td, struct image_params *imgp, 416 uintptr_t stack) 417 { 418 struct trapframe *regs; 419 struct pcb *pcb; 420 register_t saved_rflags; 421 422 regs = td->td_frame; 423 pcb = td->td_pcb; 424 425 if (td->td_proc->p_md.md_ldt != NULL) 426 user_ldt_free(td); 427 428 pcb->pcb_fsbase = 0; 429 pcb->pcb_gsbase = 0; 430 clear_pcb_flags(pcb, PCB_32BIT); 431 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 432 set_pcb_flags(pcb, PCB_FULL_IRET); 433 434 saved_rflags = regs->tf_rflags & PSL_T; 435 bzero((char *)regs, sizeof(struct trapframe)); 436 regs->tf_rip = imgp->entry_addr; 437 regs->tf_rsp = stack; 438 regs->tf_rflags = PSL_USER | saved_rflags; 439 regs->tf_ss = _udatasel; 440 regs->tf_cs = _ucodesel; 441 regs->tf_ds = _udatasel; 442 regs->tf_es = _udatasel; 443 regs->tf_fs = _ufssel; 444 regs->tf_gs = _ugssel; 445 regs->tf_flags = TF_HASSEGS; 446 447 x86_clear_dbregs(pcb); 448 449 /* 450 * Drop the FP state if we hold it, so that the process gets a 451 * clean FP state if it uses the FPU again. 452 */ 453 fpstate_drop(td); 454 } 455 456 /* 457 * Copied from amd64/amd64/machdep.c 458 * 459 * XXX fpu state need? don't think so 460 */ 461 int 462 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 463 { 464 struct proc *p; 465 struct l_ucontext uc; 466 struct l_sigcontext *context; 467 struct trapframe *regs; 468 unsigned long rflags; 469 sigset_t bmask; 470 int error; 471 ksiginfo_t ksi; 472 473 regs = td->td_frame; 474 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); 475 if (error != 0) 476 return (error); 477 478 p = td->td_proc; 479 context = &uc.uc_mcontext; 480 rflags = context->sc_rflags; 481 482 /* 483 * Don't allow users to change privileged or reserved flags. 484 */ 485 /* 486 * XXX do allow users to change the privileged flag PSL_RF. 487 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 488 * should sometimes set it there too. tf_rflags is kept in 489 * the signal context during signal handling and there is no 490 * other place to remember it, so the PSL_RF bit may be 491 * corrupted by the signal handler without us knowing. 492 * Corruption of the PSL_RF bit at worst causes one more or 493 * one less debugger trap, so allowing it is fairly harmless. 494 */ 495 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 496 uprintf("pid %d comm %s linux mangled rflags %#lx\n", 497 p->p_pid, p->p_comm, rflags); 498 return (EINVAL); 499 } 500 501 /* 502 * Don't allow users to load a valid privileged %cs. Let the 503 * hardware check for invalid selectors, excess privilege in 504 * other selectors, invalid %eip's and invalid %esp's. 505 */ 506 if (!CS_SECURE(context->sc_cs)) { 507 uprintf("pid %d comm %s linux mangled cs %#x\n", 508 p->p_pid, p->p_comm, context->sc_cs); 509 ksiginfo_init_trap(&ksi); 510 ksi.ksi_signo = SIGBUS; 511 ksi.ksi_code = BUS_OBJERR; 512 ksi.ksi_trapno = T_PROTFLT; 513 ksi.ksi_addr = (void *)regs->tf_rip; 514 trapsignal(td, &ksi); 515 return (EINVAL); 516 } 517 518 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); 519 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 520 521 regs->tf_rdi = context->sc_rdi; 522 regs->tf_rsi = context->sc_rsi; 523 regs->tf_rdx = context->sc_rdx; 524 regs->tf_rbp = context->sc_rbp; 525 regs->tf_rbx = context->sc_rbx; 526 regs->tf_rcx = context->sc_rcx; 527 regs->tf_rax = context->sc_rax; 528 regs->tf_rip = context->sc_rip; 529 regs->tf_rsp = context->sc_rsp; 530 regs->tf_r8 = context->sc_r8; 531 regs->tf_r9 = context->sc_r9; 532 regs->tf_r10 = context->sc_r10; 533 regs->tf_r11 = context->sc_r11; 534 regs->tf_r12 = context->sc_r12; 535 regs->tf_r13 = context->sc_r13; 536 regs->tf_r14 = context->sc_r14; 537 regs->tf_r15 = context->sc_r15; 538 regs->tf_cs = context->sc_cs; 539 regs->tf_err = context->sc_err; 540 regs->tf_rflags = rflags; 541 542 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 543 return (EJUSTRETURN); 544 } 545 546 /* 547 * copied from amd64/amd64/machdep.c 548 * 549 * Send an interrupt to process. 550 */ 551 static void 552 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 553 { 554 struct l_rt_sigframe sf, *sfp; 555 struct proc *p; 556 struct thread *td; 557 struct sigacts *psp; 558 caddr_t sp; 559 struct trapframe *regs; 560 int sig, code; 561 int oonstack, issiginfo; 562 563 td = curthread; 564 p = td->td_proc; 565 PROC_LOCK_ASSERT(p, MA_OWNED); 566 sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno); 567 psp = p->p_sigacts; 568 issiginfo = SIGISMEMBER(psp->ps_siginfo, sig); 569 code = ksi->ksi_code; 570 mtx_assert(&psp->ps_mtx, MA_OWNED); 571 regs = td->td_frame; 572 oonstack = sigonstack(regs->tf_rsp); 573 574 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 575 catcher, sig, mask, code); 576 577 /* Save user context. */ 578 bzero(&sf, sizeof(sf)); 579 bsd_to_linux_sigset(mask, &sf.sf_uc.uc_sigmask); 580 sf.sf_uc.uc_mcontext.sc_mask = sf.sf_uc.uc_sigmask; 581 582 sf.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 583 sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size; 584 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 585 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 586 587 sf.sf_uc.uc_mcontext.sc_rdi = regs->tf_rdi; 588 sf.sf_uc.uc_mcontext.sc_rsi = regs->tf_rsi; 589 sf.sf_uc.uc_mcontext.sc_rdx = regs->tf_rdx; 590 sf.sf_uc.uc_mcontext.sc_rbp = regs->tf_rbp; 591 sf.sf_uc.uc_mcontext.sc_rbx = regs->tf_rbx; 592 sf.sf_uc.uc_mcontext.sc_rcx = regs->tf_rcx; 593 sf.sf_uc.uc_mcontext.sc_rax = regs->tf_rax; 594 sf.sf_uc.uc_mcontext.sc_rip = regs->tf_rip; 595 sf.sf_uc.uc_mcontext.sc_rsp = regs->tf_rsp; 596 sf.sf_uc.uc_mcontext.sc_r8 = regs->tf_r8; 597 sf.sf_uc.uc_mcontext.sc_r9 = regs->tf_r9; 598 sf.sf_uc.uc_mcontext.sc_r10 = regs->tf_r10; 599 sf.sf_uc.uc_mcontext.sc_r11 = regs->tf_r11; 600 sf.sf_uc.uc_mcontext.sc_r12 = regs->tf_r12; 601 sf.sf_uc.uc_mcontext.sc_r13 = regs->tf_r13; 602 sf.sf_uc.uc_mcontext.sc_r14 = regs->tf_r14; 603 sf.sf_uc.uc_mcontext.sc_r15 = regs->tf_r15; 604 sf.sf_uc.uc_mcontext.sc_cs = regs->tf_cs; 605 sf.sf_uc.uc_mcontext.sc_rflags = regs->tf_rflags; 606 sf.sf_uc.uc_mcontext.sc_err = regs->tf_err; 607 sf.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 608 sf.sf_uc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 609 610 /* Allocate space for the signal handler context. */ 611 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 612 SIGISMEMBER(psp->ps_sigonstack, sig)) { 613 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size; 614 } else 615 sp = (caddr_t)regs->tf_rsp - 128; 616 sp -= sizeof(struct l_rt_sigframe); 617 /* Align to 16 bytes. */ 618 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 619 620 mtx_unlock(&psp->ps_mtx); 621 PROC_UNLOCK(p); 622 623 /* Translate the signal. */ 624 sig = bsd_to_linux_signal(sig); 625 /* Fill in POSIX parts. */ 626 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig); 627 628 /* Copy the sigframe out to the user's stack. */ 629 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 630 uprintf("pid %d comm %s has trashed its stack, killing\n", 631 p->p_pid, p->p_comm); 632 PROC_LOCK(p); 633 sigexit(td, SIGILL); 634 } 635 636 /* Build the argument list for the signal handler. */ 637 regs->tf_rdi = sig; /* arg 1 in %rdi */ 638 regs->tf_rax = 0; 639 if (issiginfo) { 640 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 641 regs->tf_rdx = (register_t)&sfp->sf_uc; /* arg 3 in %rdx */ 642 } else { 643 regs->tf_rsi = 0; 644 regs->tf_rdx = 0; 645 } 646 regs->tf_rcx = (register_t)catcher; 647 regs->tf_rsp = (long)sfp; 648 regs->tf_rip = linux_rt_sigcode; 649 regs->tf_rflags &= ~(PSL_T | PSL_D); 650 regs->tf_cs = _ucodesel; 651 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 652 PROC_LOCK(p); 653 mtx_lock(&psp->ps_mtx); 654 } 655 656 #define LINUX_VSYSCALL_START (-10UL << 20) 657 #define LINUX_VSYSCALL_SZ 1024 658 659 const unsigned long linux_vsyscall_vector[] = { 660 LINUX_SYS_gettimeofday, 661 LINUX_SYS_linux_time, 662 LINUX_SYS_linux_getcpu, 663 }; 664 665 static int 666 linux_vsyscall(struct thread *td) 667 { 668 struct trapframe *frame; 669 uint64_t retqaddr; 670 int code, traced; 671 int error; 672 673 frame = td->td_frame; 674 675 /* Check %rip for vsyscall area. */ 676 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 677 return (EINVAL); 678 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 679 return (EINVAL); 680 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 681 if (code >= nitems(linux_vsyscall_vector)) 682 return (EINVAL); 683 684 /* 685 * vsyscall called as callq *(%rax), so we must 686 * use return address from %rsp and also fixup %rsp. 687 */ 688 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 689 if (error) 690 return (error); 691 692 frame->tf_rip = retqaddr; 693 frame->tf_rax = linux_vsyscall_vector[code]; 694 frame->tf_rsp += 8; 695 696 traced = (frame->tf_flags & PSL_T); 697 698 amd64_syscall(td, traced); 699 700 return (0); 701 } 702 703 struct sysentvec elf_linux_sysvec = { 704 .sv_size = LINUX_SYS_MAXSYSCALL, 705 .sv_table = linux_sysent, 706 .sv_fixup = linux_fixup_elf, 707 .sv_sendsig = linux_rt_sendsig, 708 .sv_sigcode = &_binary_linux_vdso_so_o_start, 709 .sv_szsigcode = &linux_szsigcode, 710 .sv_name = "Linux ELF64", 711 .sv_coredump = elf64_coredump, 712 .sv_elf_core_osabi = ELFOSABI_NONE, 713 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR, 714 .sv_elf_core_prepare_notes = linux64_prepare_notes, 715 .sv_imgact_try = linux_exec_imgact_try, 716 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 717 .sv_minuser = VM_MIN_ADDRESS, 718 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48, 719 .sv_usrstack = LINUX_USRSTACK_LA48, 720 .sv_psstrings = LINUX_PS_STRINGS_LA48, 721 .sv_psstringssz = sizeof(struct ps_strings), 722 .sv_stackprot = VM_PROT_ALL, 723 .sv_copyout_auxargs = linux_copyout_auxargs, 724 .sv_copyout_strings = linux_copyout_strings, 725 .sv_setregs = linux_exec_setregs, 726 .sv_fixlimit = NULL, 727 .sv_maxssiz = NULL, 728 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN | 729 SV_SIG_WAITNDQ | SV_TIMEKEEP, 730 .sv_set_syscall_retval = linux_set_syscall_retval, 731 .sv_fetch_syscall_args = linux_fetch_syscall_args, 732 .sv_syscallnames = NULL, 733 .sv_shared_page_base = LINUX_SHAREDPAGE_LA48, 734 .sv_shared_page_len = PAGE_SIZE, 735 .sv_schedtail = linux_schedtail, 736 .sv_thread_detach = linux_thread_detach, 737 .sv_trap = linux_vsyscall, 738 .sv_onexec = linux_on_exec_vmspace, 739 .sv_onexit = linux_on_exit, 740 .sv_ontdexit = linux_thread_dtor, 741 .sv_setid_allowed = &linux_setid_allowed_query, 742 .sv_set_fork_retval = linux_set_fork_retval, 743 }; 744 745 static int 746 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp) 747 { 748 int error; 749 750 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base, 751 LINUX_VDSOPAGE_SIZE, imgp); 752 if (error == 0) 753 linux_on_exec(p, imgp); 754 return (error); 755 } 756 757 /* 758 * linux_vdso_install() and linux_exec_sysvec_init() must be called 759 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY). 760 */ 761 static void 762 linux_exec_sysvec_init(void *param) 763 { 764 l_uintptr_t *ktimekeep_base, *ktsc_selector; 765 struct sysentvec *sv; 766 ptrdiff_t tkoff; 767 768 sv = param; 769 amd64_lower_shared_page(sv); 770 /* Fill timekeep_base */ 771 exec_sysvec_init(sv); 772 773 tkoff = kern_timekeep_base - linux_vdso_base; 774 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 775 *ktimekeep_base = sv->sv_timekeep_base; 776 777 tkoff = kern_tsc_selector - linux_vdso_base; 778 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 779 *ktsc_selector = linux_vdso_tsc_selector_idx(); 780 if (bootverbose) 781 printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector); 782 783 tkoff = kern_cpu_selector - linux_vdso_base; 784 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 785 *ktsc_selector = linux_vdso_cpu_selector_idx(); 786 if (bootverbose) 787 printf("Linux x86-64 vDSO cpu_selector: %lu\n", *ktsc_selector); 788 } 789 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY, 790 linux_exec_sysvec_init, &elf_linux_sysvec); 791 792 static void 793 linux_vdso_install(const void *param) 794 { 795 char *vdso_start = &_binary_linux_vdso_so_o_start; 796 char *vdso_end = &_binary_linux_vdso_so_o_end; 797 798 linux_szsigcode = vdso_end - vdso_start; 799 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE); 800 801 linux_vdso_base = LINUX_VDSOPAGE_LA48; 802 if (hw_lower_amd64_sharedpage != 0) 803 linux_vdso_base -= PAGE_SIZE; 804 805 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base); 806 807 linux_vdso_obj = __elfN(linux_shared_page_init) 808 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 809 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode); 810 811 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base); 812 } 813 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST, 814 linux_vdso_install, NULL); 815 816 static void 817 linux_vdso_deinstall(const void *param) 818 { 819 820 __elfN(linux_shared_page_fini)(linux_vdso_obj, 821 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 822 } 823 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 824 linux_vdso_deinstall, NULL); 825 826 static void 827 linux_vdso_reloc(char *mapping, Elf_Addr offset) 828 { 829 const Elf_Ehdr *ehdr; 830 const Elf_Shdr *shdr; 831 Elf64_Addr *where, val; 832 Elf_Size rtype, symidx; 833 const Elf_Rela *rela; 834 Elf_Addr addr, addend; 835 int relacnt; 836 int i, j; 837 838 MPASS(offset != 0); 839 840 relacnt = 0; 841 ehdr = (const Elf_Ehdr *)mapping; 842 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff); 843 for (i = 0; i < ehdr->e_shnum; i++) 844 { 845 switch (shdr[i].sh_type) { 846 case SHT_REL: 847 printf("Linux x86_64 vDSO: unexpected Rel section\n"); 848 break; 849 case SHT_RELA: 850 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset); 851 relacnt = shdr[i].sh_size / sizeof(*rela); 852 } 853 } 854 855 for (j = 0; j < relacnt; j++, rela++) { 856 where = (Elf_Addr *)(mapping + rela->r_offset); 857 addend = rela->r_addend; 858 rtype = ELF_R_TYPE(rela->r_info); 859 symidx = ELF_R_SYM(rela->r_info); 860 861 switch (rtype) { 862 case R_X86_64_NONE: /* none */ 863 break; 864 865 case R_X86_64_RELATIVE: /* B + A */ 866 addr = (Elf_Addr)(offset + addend); 867 val = addr; 868 if (*where != val) 869 *where = val; 870 break; 871 case R_X86_64_IRELATIVE: 872 printf("Linux x86_64 vDSO: unexpected ifunc relocation, " 873 "symbol index %ld\n", symidx); 874 break; 875 default: 876 printf("Linux x86_64 vDSO: unexpected relocation type %ld, " 877 "symbol index %ld\n", rtype, symidx); 878 } 879 } 880 } 881 882 static char GNULINUX_ABI_VENDOR[] = "GNU"; 883 static int GNULINUX_ABI_DESC = 0; 884 885 static bool 886 linux_trans_osrel(const Elf_Note *note, int32_t *osrel) 887 { 888 const Elf32_Word *desc; 889 uintptr_t p; 890 891 p = (uintptr_t)(note + 1); 892 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 893 894 desc = (const Elf32_Word *)p; 895 if (desc[0] != GNULINUX_ABI_DESC) 896 return (false); 897 898 /* 899 * For Linux we encode osrel using the Linux convention of 900 * (version << 16) | (major << 8) | (minor) 901 * See macro in linux_mib.h 902 */ 903 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 904 905 return (true); 906 } 907 908 static Elf_Brandnote linux64_brandnote = { 909 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), 910 .hdr.n_descsz = 16, 911 .hdr.n_type = 1, 912 .vendor = GNULINUX_ABI_VENDOR, 913 .flags = BN_TRANSLATE_OSREL, 914 .trans_osrel = linux_trans_osrel 915 }; 916 917 static Elf64_Brandinfo linux_glibc2brand = { 918 .brand = ELFOSABI_LINUX, 919 .machine = EM_X86_64, 920 .compat_3_brand = "Linux", 921 .emul_path = linux_emul_path, 922 .interp_path = "/lib64/ld-linux-x86-64.so.2", 923 .sysvec = &elf_linux_sysvec, 924 .interp_newpath = NULL, 925 .brand_note = &linux64_brandnote, 926 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 927 }; 928 929 static Elf64_Brandinfo linux_glibc2brandshort = { 930 .brand = ELFOSABI_LINUX, 931 .machine = EM_X86_64, 932 .compat_3_brand = "Linux", 933 .emul_path = linux_emul_path, 934 .interp_path = "/lib64/ld-linux.so.2", 935 .sysvec = &elf_linux_sysvec, 936 .interp_newpath = NULL, 937 .brand_note = &linux64_brandnote, 938 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 939 }; 940 941 static Elf64_Brandinfo linux_muslbrand = { 942 .brand = ELFOSABI_LINUX, 943 .machine = EM_X86_64, 944 .compat_3_brand = "Linux", 945 .emul_path = linux_emul_path, 946 .interp_path = "/lib/ld-musl-x86_64.so.1", 947 .sysvec = &elf_linux_sysvec, 948 .interp_newpath = NULL, 949 .brand_note = &linux64_brandnote, 950 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE | 951 LINUX_BI_FUTEX_REQUEUE 952 }; 953 954 Elf64_Brandinfo *linux_brandlist[] = { 955 &linux_glibc2brand, 956 &linux_glibc2brandshort, 957 &linux_muslbrand, 958 NULL 959 }; 960 961 static int 962 linux64_elf_modevent(module_t mod, int type, void *data) 963 { 964 Elf64_Brandinfo **brandinfo; 965 int error; 966 struct linux_ioctl_handler **lihp; 967 968 error = 0; 969 970 switch(type) { 971 case MOD_LOAD: 972 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 973 ++brandinfo) 974 if (elf64_insert_brand_entry(*brandinfo) < 0) 975 error = EINVAL; 976 if (error == 0) { 977 SET_FOREACH(lihp, linux_ioctl_handler_set) 978 linux_ioctl_register_handler(*lihp); 979 stclohz = (stathz ? stathz : hz); 980 if (bootverbose) 981 printf("Linux x86-64 ELF exec handler installed\n"); 982 } else 983 printf("cannot insert Linux x86-64 ELF brand handler\n"); 984 break; 985 case MOD_UNLOAD: 986 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 987 ++brandinfo) 988 if (elf64_brand_inuse(*brandinfo)) 989 error = EBUSY; 990 if (error == 0) { 991 for (brandinfo = &linux_brandlist[0]; 992 *brandinfo != NULL; ++brandinfo) 993 if (elf64_remove_brand_entry(*brandinfo) < 0) 994 error = EINVAL; 995 } 996 if (error == 0) { 997 SET_FOREACH(lihp, linux_ioctl_handler_set) 998 linux_ioctl_unregister_handler(*lihp); 999 if (bootverbose) 1000 printf("Linux x86_64 ELF exec handler removed\n"); 1001 } else 1002 printf("Could not deinstall Linux x86_64 ELF interpreter entry\n"); 1003 break; 1004 default: 1005 return (EOPNOTSUPP); 1006 } 1007 return (error); 1008 } 1009 1010 static moduledata_t linux64_elf_mod = { 1011 "linux64elf", 1012 linux64_elf_modevent, 1013 0 1014 }; 1015 1016 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 1017 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 1018 FEATURE(linux64, "Linux 64bit support"); 1019