1 /*- 2 * Copyright (c) 2013 Dmitry Chagin 3 * Copyright (c) 2004 Tim J. Robbins 4 * Copyright (c) 2003 Peter Wemm 5 * Copyright (c) 2002 Doug Rabson 6 * Copyright (c) 1998-1999 Andrew Gallatin 7 * Copyright (c) 1994-1996 Søren Schmidt 8 * All rights reserved. 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/signalvar.h> 54 #include <sys/syscallsubr.h> 55 #include <sys/sysctl.h> 56 #include <sys/sysent.h> 57 #include <sys/sysproto.h> 58 #include <sys/vnode.h> 59 #include <sys/eventhandler.h> 60 61 #include <vm/vm.h> 62 #include <vm/pmap.h> 63 #include <vm/vm_extern.h> 64 #include <vm/vm_map.h> 65 #include <vm/vm_object.h> 66 #include <vm/vm_page.h> 67 #include <vm/vm_param.h> 68 69 #include <machine/cpu.h> 70 #include <machine/md_var.h> 71 #include <machine/pcb.h> 72 #include <machine/specialreg.h> 73 #include <machine/trap.h> 74 75 #include <amd64/linux/linux.h> 76 #include <amd64/linux/linux_proto.h> 77 #include <compat/linux/linux_emul.h> 78 #include <compat/linux/linux_ioctl.h> 79 #include <compat/linux/linux_mib.h> 80 #include <compat/linux/linux_misc.h> 81 #include <compat/linux/linux_signal.h> 82 #include <compat/linux/linux_sysproto.h> 83 #include <compat/linux/linux_util.h> 84 #include <compat/linux/linux_vdso.h> 85 86 MODULE_VERSION(linux64, 1); 87 88 const char *linux_kplatform; 89 static int linux_szsigcode; 90 static vm_object_t linux_shared_page_obj; 91 static char *linux_shared_page_mapping; 92 extern char _binary_linux_locore_o_start; 93 extern char _binary_linux_locore_o_end; 94 95 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 96 97 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 98 99 static int linux_copyout_strings(struct image_params *imgp, 100 uintptr_t *stack_base); 101 static int linux_fixup_elf(uintptr_t *stack_base, 102 struct image_params *iparams); 103 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 104 static void linux_vdso_install(void *param); 105 static void linux_vdso_deinstall(void *param); 106 static void linux_set_syscall_retval(struct thread *td, int error); 107 static int linux_fetch_syscall_args(struct thread *td); 108 static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 109 uintptr_t stack); 110 static int linux_vsyscall(struct thread *td); 111 112 #define LINUX_T_UNKNOWN 255 113 static int _bsd_to_linux_trapcode[] = { 114 LINUX_T_UNKNOWN, /* 0 */ 115 6, /* 1 T_PRIVINFLT */ 116 LINUX_T_UNKNOWN, /* 2 */ 117 3, /* 3 T_BPTFLT */ 118 LINUX_T_UNKNOWN, /* 4 */ 119 LINUX_T_UNKNOWN, /* 5 */ 120 16, /* 6 T_ARITHTRAP */ 121 254, /* 7 T_ASTFLT */ 122 LINUX_T_UNKNOWN, /* 8 */ 123 13, /* 9 T_PROTFLT */ 124 1, /* 10 T_TRCTRAP */ 125 LINUX_T_UNKNOWN, /* 11 */ 126 14, /* 12 T_PAGEFLT */ 127 LINUX_T_UNKNOWN, /* 13 */ 128 17, /* 14 T_ALIGNFLT */ 129 LINUX_T_UNKNOWN, /* 15 */ 130 LINUX_T_UNKNOWN, /* 16 */ 131 LINUX_T_UNKNOWN, /* 17 */ 132 0, /* 18 T_DIVIDE */ 133 2, /* 19 T_NMI */ 134 4, /* 20 T_OFLOW */ 135 5, /* 21 T_BOUND */ 136 7, /* 22 T_DNA */ 137 8, /* 23 T_DOUBLEFLT */ 138 9, /* 24 T_FPOPFLT */ 139 10, /* 25 T_TSSFLT */ 140 11, /* 26 T_SEGNPFLT */ 141 12, /* 27 T_STKFLT */ 142 18, /* 28 T_MCHK */ 143 19, /* 29 T_XMMFLT */ 144 15 /* 30 T_RESERVED */ 145 }; 146 #define bsd_to_linux_trapcode(code) \ 147 ((code)<nitems(_bsd_to_linux_trapcode)? \ 148 _bsd_to_linux_trapcode[(code)]: \ 149 LINUX_T_UNKNOWN) 150 151 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 152 LINUX_VDSO_SYM_CHAR(linux_platform); 153 154 /* 155 * If FreeBSD & Linux have a difference of opinion about what a trap 156 * means, deal with it here. 157 * 158 * MPSAFE 159 */ 160 static int 161 linux_translate_traps(int signal, int trap_code) 162 { 163 164 if (signal != SIGBUS) 165 return (signal); 166 switch (trap_code) { 167 case T_PROTFLT: 168 case T_TSSFLT: 169 case T_DOUBLEFLT: 170 case T_PAGEFLT: 171 return (SIGSEGV); 172 default: 173 return (signal); 174 } 175 } 176 177 static int 178 linux_fetch_syscall_args(struct thread *td) 179 { 180 struct proc *p; 181 struct trapframe *frame; 182 struct syscall_args *sa; 183 184 p = td->td_proc; 185 frame = td->td_frame; 186 sa = &td->td_sa; 187 188 sa->args[0] = frame->tf_rdi; 189 sa->args[1] = frame->tf_rsi; 190 sa->args[2] = frame->tf_rdx; 191 sa->args[3] = frame->tf_rcx; 192 sa->args[4] = frame->tf_r8; 193 sa->args[5] = frame->tf_r9; 194 sa->code = frame->tf_rax; 195 196 if (sa->code >= p->p_sysent->sv_size) 197 /* nosys */ 198 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 199 else 200 sa->callp = &p->p_sysent->sv_table[sa->code]; 201 sa->narg = sa->callp->sy_narg; 202 203 td->td_retval[0] = 0; 204 return (0); 205 } 206 207 static void 208 linux_set_syscall_retval(struct thread *td, int error) 209 { 210 struct trapframe *frame = td->td_frame; 211 212 /* 213 * On Linux only %rcx and %r11 values are not preserved across 214 * the syscall. So, do not clobber %rdx and %r10. 215 */ 216 td->td_retval[1] = frame->tf_rdx; 217 if (error != EJUSTRETURN) 218 frame->tf_r10 = frame->tf_rcx; 219 220 cpu_set_syscall_retval(td, error); 221 222 /* Restore all registers. */ 223 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 224 } 225 226 static int 227 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) 228 { 229 Elf_Auxargs *args; 230 Elf_Auxinfo *argarray, *pos; 231 struct proc *p; 232 int error, issetugid; 233 234 p = imgp->proc; 235 args = (Elf64_Auxargs *)imgp->auxargs; 236 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 237 M_WAITOK | M_ZERO); 238 239 issetugid = p->p_flag & P_SUGID ? 1 : 0; 240 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, 241 imgp->proc->p_sysent->sv_shared_page_base); 242 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 243 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 244 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 245 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 246 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 247 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 248 AUXARGS_ENTRY(pos, AT_BASE, args->base); 249 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 250 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 251 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 252 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 253 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 254 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 255 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 256 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 257 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); 258 if (imgp->execpathp != 0) 259 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); 260 if (args->execfd != -1) 261 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 262 AUXARGS_ENTRY(pos, AT_NULL, 0); 263 264 free(imgp->auxargs, M_TEMP); 265 imgp->auxargs = NULL; 266 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 267 268 error = copyout(argarray, (void *)base, 269 sizeof(*argarray) * LINUX_AT_COUNT); 270 free(argarray, M_TEMP); 271 return (error); 272 } 273 274 static int 275 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) 276 { 277 Elf_Addr *base; 278 279 base = (Elf64_Addr *)*stack_base; 280 base--; 281 if (suword(base, (uint64_t)imgp->args->argc) == -1) 282 return (EFAULT); 283 284 *stack_base = (uintptr_t)base; 285 return (0); 286 } 287 288 /* 289 * Copy strings out to the new process address space, constructing new arg 290 * and env vector tables. Return a pointer to the base so that it can be used 291 * as the initial stack pointer. 292 */ 293 static int 294 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 295 { 296 int argc, envc, error; 297 char **vectp; 298 char *stringp; 299 uintptr_t destp, ustringp; 300 struct ps_strings *arginfo; 301 char canary[LINUX_AT_RANDOM_LEN]; 302 size_t execpath_len; 303 struct proc *p; 304 305 /* Calculate string base and vector table pointers. */ 306 if (imgp->execpath != NULL && imgp->auxargs != NULL) 307 execpath_len = strlen(imgp->execpath) + 1; 308 else 309 execpath_len = 0; 310 311 p = imgp->proc; 312 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 313 destp = (uintptr_t)arginfo; 314 315 if (execpath_len != 0) { 316 destp -= execpath_len; 317 destp = rounddown2(destp, sizeof(void *)); 318 imgp->execpathp = (void *)destp; 319 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 320 if (error != 0) 321 return (error); 322 } 323 324 /* Prepare the canary for SSP. */ 325 arc4rand(canary, sizeof(canary), 0); 326 destp -= roundup(sizeof(canary), sizeof(void *)); 327 imgp->canary = (void *)destp; 328 error = copyout(canary, imgp->canary, sizeof(canary)); 329 if (error != 0) 330 return (error); 331 332 /* Allocate room for the argument and environment strings. */ 333 destp -= ARG_MAX - imgp->args->stringspace; 334 destp = rounddown2(destp, sizeof(void *)); 335 ustringp = destp; 336 337 if (imgp->auxargs) { 338 /* 339 * Allocate room on the stack for the ELF auxargs 340 * array. It has LINUX_AT_COUNT entries. 341 */ 342 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo); 343 destp = rounddown2(destp, sizeof(void *)); 344 } 345 346 vectp = (char **)destp; 347 348 /* 349 * Allocate room for the argv[] and env vectors including the 350 * terminating NULL pointers. 351 */ 352 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 353 354 /* 355 * Starting with 2.24, glibc depends on a 16-byte stack alignment. 356 * One "long argc" will be prepended later. 357 */ 358 vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8); 359 360 /* vectp also becomes our initial stack base. */ 361 *stack_base = (uintptr_t)vectp; 362 363 stringp = imgp->args->begin_argv; 364 argc = imgp->args->argc; 365 envc = imgp->args->envc; 366 367 /* Copy out strings - arguments and environment. */ 368 error = copyout(stringp, (void *)ustringp, 369 ARG_MAX - imgp->args->stringspace); 370 if (error != 0) 371 return (error); 372 373 /* Fill in "ps_strings" struct for ps, w, etc. */ 374 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 375 suword(&arginfo->ps_nargvstr, argc) != 0) 376 return (EFAULT); 377 378 /* Fill in argument portion of vector table. */ 379 for (; argc > 0; --argc) { 380 if (suword(vectp++, ustringp) != 0) 381 return (EFAULT); 382 while (*stringp++ != 0) 383 ustringp++; 384 ustringp++; 385 } 386 387 /* A null vector table pointer separates the argp's from the envp's. */ 388 if (suword(vectp++, 0) != 0) 389 return (EFAULT); 390 391 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 392 suword(&arginfo->ps_nenvstr, envc) != 0) 393 return (EFAULT); 394 395 /* Fill in environment portion of vector table. */ 396 for (; envc > 0; --envc) { 397 if (suword(vectp++, ustringp) != 0) 398 return (EFAULT); 399 while (*stringp++ != 0) 400 ustringp++; 401 ustringp++; 402 } 403 404 /* The end of the vector table is a null pointer. */ 405 if (suword(vectp, 0) != 0) 406 return (EFAULT); 407 408 if (imgp->auxargs) { 409 vectp++; 410 error = imgp->sysent->sv_copyout_auxargs(imgp, 411 (uintptr_t)vectp); 412 if (error != 0) 413 return (error); 414 } 415 416 return (0); 417 } 418 419 /* 420 * Reset registers to default values on exec. 421 */ 422 static void 423 linux_exec_setregs(struct thread *td, struct image_params *imgp, 424 uintptr_t stack) 425 { 426 struct trapframe *regs; 427 struct pcb *pcb; 428 register_t saved_rflags; 429 430 regs = td->td_frame; 431 pcb = td->td_pcb; 432 433 if (td->td_proc->p_md.md_ldt != NULL) 434 user_ldt_free(td); 435 436 pcb->pcb_fsbase = 0; 437 pcb->pcb_gsbase = 0; 438 clear_pcb_flags(pcb, PCB_32BIT); 439 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 440 set_pcb_flags(pcb, PCB_FULL_IRET); 441 442 saved_rflags = regs->tf_rflags & PSL_T; 443 bzero((char *)regs, sizeof(struct trapframe)); 444 regs->tf_rip = imgp->entry_addr; 445 regs->tf_rsp = stack; 446 regs->tf_rflags = PSL_USER | saved_rflags; 447 regs->tf_ss = _udatasel; 448 regs->tf_cs = _ucodesel; 449 regs->tf_ds = _udatasel; 450 regs->tf_es = _udatasel; 451 regs->tf_fs = _ufssel; 452 regs->tf_gs = _ugssel; 453 regs->tf_flags = TF_HASSEGS; 454 455 /* 456 * Reset the hardware debug registers if they were in use. 457 * They won't have any meaning for the newly exec'd process. 458 */ 459 if (pcb->pcb_flags & PCB_DBREGS) { 460 pcb->pcb_dr0 = 0; 461 pcb->pcb_dr1 = 0; 462 pcb->pcb_dr2 = 0; 463 pcb->pcb_dr3 = 0; 464 pcb->pcb_dr6 = 0; 465 pcb->pcb_dr7 = 0; 466 if (pcb == curpcb) { 467 /* 468 * Clear the debug registers on the running 469 * CPU, otherwise they will end up affecting 470 * the next process we switch to. 471 */ 472 reset_dbregs(); 473 } 474 clear_pcb_flags(pcb, PCB_DBREGS); 475 } 476 477 /* 478 * Drop the FP state if we hold it, so that the process gets a 479 * clean FP state if it uses the FPU again. 480 */ 481 fpstate_drop(td); 482 } 483 484 /* 485 * Copied from amd64/amd64/machdep.c 486 * 487 * XXX fpu state need? don't think so 488 */ 489 int 490 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 491 { 492 struct proc *p; 493 struct l_ucontext uc; 494 struct l_sigcontext *context; 495 struct trapframe *regs; 496 unsigned long rflags; 497 int error; 498 ksiginfo_t ksi; 499 500 regs = td->td_frame; 501 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); 502 if (error != 0) 503 return (error); 504 505 p = td->td_proc; 506 context = &uc.uc_mcontext; 507 rflags = context->sc_rflags; 508 509 /* 510 * Don't allow users to change privileged or reserved flags. 511 */ 512 /* 513 * XXX do allow users to change the privileged flag PSL_RF. 514 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 515 * should sometimes set it there too. tf_rflags is kept in 516 * the signal context during signal handling and there is no 517 * other place to remember it, so the PSL_RF bit may be 518 * corrupted by the signal handler without us knowing. 519 * Corruption of the PSL_RF bit at worst causes one more or 520 * one less debugger trap, so allowing it is fairly harmless. 521 */ 522 523 #define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 524 if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 525 printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags); 526 return (EINVAL); 527 } 528 529 /* 530 * Don't allow users to load a valid privileged %cs. Let the 531 * hardware check for invalid selectors, excess privilege in 532 * other selectors, invalid %eip's and invalid %esp's. 533 */ 534 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 535 if (!CS_SECURE(context->sc_cs)) { 536 printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs); 537 ksiginfo_init_trap(&ksi); 538 ksi.ksi_signo = SIGBUS; 539 ksi.ksi_code = BUS_OBJERR; 540 ksi.ksi_trapno = T_PROTFLT; 541 ksi.ksi_addr = (void *)regs->tf_rip; 542 trapsignal(td, &ksi); 543 return (EINVAL); 544 } 545 546 PROC_LOCK(p); 547 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); 548 SIG_CANTMASK(td->td_sigmask); 549 signotify(td); 550 PROC_UNLOCK(p); 551 552 regs->tf_rdi = context->sc_rdi; 553 regs->tf_rsi = context->sc_rsi; 554 regs->tf_rdx = context->sc_rdx; 555 regs->tf_rbp = context->sc_rbp; 556 regs->tf_rbx = context->sc_rbx; 557 regs->tf_rcx = context->sc_rcx; 558 regs->tf_rax = context->sc_rax; 559 regs->tf_rip = context->sc_rip; 560 regs->tf_rsp = context->sc_rsp; 561 regs->tf_r8 = context->sc_r8; 562 regs->tf_r9 = context->sc_r9; 563 regs->tf_r10 = context->sc_r10; 564 regs->tf_r11 = context->sc_r11; 565 regs->tf_r12 = context->sc_r12; 566 regs->tf_r13 = context->sc_r13; 567 regs->tf_r14 = context->sc_r14; 568 regs->tf_r15 = context->sc_r15; 569 regs->tf_cs = context->sc_cs; 570 regs->tf_err = context->sc_err; 571 regs->tf_rflags = rflags; 572 573 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 574 return (EJUSTRETURN); 575 } 576 577 /* 578 * copied from amd64/amd64/machdep.c 579 * 580 * Send an interrupt to process. 581 */ 582 static void 583 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 584 { 585 struct l_rt_sigframe sf, *sfp; 586 struct proc *p; 587 struct thread *td; 588 struct sigacts *psp; 589 caddr_t sp; 590 struct trapframe *regs; 591 int sig, code; 592 int oonstack; 593 594 td = curthread; 595 p = td->td_proc; 596 PROC_LOCK_ASSERT(p, MA_OWNED); 597 sig = ksi->ksi_signo; 598 psp = p->p_sigacts; 599 code = ksi->ksi_code; 600 mtx_assert(&psp->ps_mtx, MA_OWNED); 601 regs = td->td_frame; 602 oonstack = sigonstack(regs->tf_rsp); 603 604 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 605 catcher, sig, mask, code); 606 607 /* Allocate space for the signal handler context. */ 608 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 609 SIGISMEMBER(psp->ps_sigonstack, sig)) { 610 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - 611 sizeof(struct l_rt_sigframe); 612 } else 613 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128; 614 /* Align to 16 bytes. */ 615 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 616 mtx_unlock(&psp->ps_mtx); 617 618 /* Translate the signal. */ 619 sig = bsd_to_linux_signal(sig); 620 621 /* Save user context. */ 622 bzero(&sf, sizeof(sf)); 623 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask); 624 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask); 625 626 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 627 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 628 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 629 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 630 PROC_UNLOCK(p); 631 632 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi; 633 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi; 634 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx; 635 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp; 636 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx; 637 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx; 638 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax; 639 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip; 640 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp; 641 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8; 642 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9; 643 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10; 644 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11; 645 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12; 646 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13; 647 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14; 648 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15; 649 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 650 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags; 651 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err; 652 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 653 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 654 655 /* Build the argument list for the signal handler. */ 656 regs->tf_rdi = sig; /* arg 1 in %rdi */ 657 regs->tf_rax = 0; 658 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 659 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */ 660 661 sf.sf_handler = catcher; 662 /* Fill in POSIX parts. */ 663 ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig); 664 665 /* Copy the sigframe out to the user's stack. */ 666 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 667 PROC_LOCK(p); 668 sigexit(td, SIGILL); 669 } 670 671 regs->tf_rsp = (long)sfp; 672 regs->tf_rip = linux_rt_sigcode; 673 regs->tf_rflags &= ~(PSL_T | PSL_D); 674 regs->tf_cs = _ucodesel; 675 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 676 PROC_LOCK(p); 677 mtx_lock(&psp->ps_mtx); 678 } 679 680 #define LINUX_VSYSCALL_START (-10UL << 20) 681 #define LINUX_VSYSCALL_SZ 1024 682 683 const unsigned long linux_vsyscall_vector[] = { 684 LINUX_SYS_gettimeofday, 685 LINUX_SYS_linux_time, 686 /* getcpu not implemented */ 687 }; 688 689 static int 690 linux_vsyscall(struct thread *td) 691 { 692 struct trapframe *frame; 693 uint64_t retqaddr; 694 int code, traced; 695 int error; 696 697 frame = td->td_frame; 698 699 /* Check %rip for vsyscall area. */ 700 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 701 return (EINVAL); 702 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 703 return (EINVAL); 704 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 705 if (code >= nitems(linux_vsyscall_vector)) 706 return (EINVAL); 707 708 /* 709 * vsyscall called as callq *(%rax), so we must 710 * use return address from %rsp and also fixup %rsp. 711 */ 712 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 713 if (error) 714 return (error); 715 716 frame->tf_rip = retqaddr; 717 frame->tf_rax = linux_vsyscall_vector[code]; 718 frame->tf_rsp += 8; 719 720 traced = (frame->tf_flags & PSL_T); 721 722 amd64_syscall(td, traced); 723 724 return (0); 725 } 726 727 struct sysentvec elf_linux_sysvec = { 728 .sv_size = LINUX_SYS_MAXSYSCALL, 729 .sv_table = linux_sysent, 730 .sv_errsize = ELAST + 1, 731 .sv_errtbl = linux_errtbl, 732 .sv_transtrap = linux_translate_traps, 733 .sv_fixup = linux_fixup_elf, 734 .sv_sendsig = linux_rt_sendsig, 735 .sv_sigcode = &_binary_linux_locore_o_start, 736 .sv_szsigcode = &linux_szsigcode, 737 .sv_name = "Linux ELF64", 738 .sv_coredump = elf64_coredump, 739 .sv_imgact_try = linux_exec_imgact_try, 740 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 741 .sv_minuser = VM_MIN_ADDRESS, 742 .sv_maxuser = VM_MAXUSER_ADDRESS, 743 .sv_usrstack = USRSTACK, 744 .sv_psstrings = PS_STRINGS, 745 .sv_stackprot = VM_PROT_ALL, 746 .sv_copyout_auxargs = linux_copyout_auxargs, 747 .sv_copyout_strings = linux_copyout_strings, 748 .sv_setregs = linux_exec_setregs, 749 .sv_fixlimit = NULL, 750 .sv_maxssiz = NULL, 751 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP, 752 .sv_set_syscall_retval = linux_set_syscall_retval, 753 .sv_fetch_syscall_args = linux_fetch_syscall_args, 754 .sv_syscallnames = NULL, 755 .sv_shared_page_base = SHAREDPAGE, 756 .sv_shared_page_len = PAGE_SIZE, 757 .sv_schedtail = linux_schedtail, 758 .sv_thread_detach = linux_thread_detach, 759 .sv_trap = linux_vsyscall, 760 }; 761 762 static void 763 linux_vdso_install(void *param) 764 { 765 766 amd64_lower_shared_page(&elf_linux_sysvec); 767 768 linux_szsigcode = (&_binary_linux_locore_o_end - 769 &_binary_linux_locore_o_start); 770 771 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 772 panic("Linux invalid vdso size\n"); 773 774 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 775 776 linux_shared_page_obj = __elfN(linux_shared_page_init) 777 (&linux_shared_page_mapping); 778 779 __elfN(linux_vdso_reloc)(&elf_linux_sysvec); 780 781 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 782 linux_szsigcode); 783 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 784 785 linux_kplatform = linux_shared_page_mapping + 786 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base); 787 } 788 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 789 linux_vdso_install, NULL); 790 791 static void 792 linux_vdso_deinstall(void *param) 793 { 794 795 __elfN(linux_shared_page_fini)(linux_shared_page_obj); 796 } 797 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 798 linux_vdso_deinstall, NULL); 799 800 static char GNULINUX_ABI_VENDOR[] = "GNU"; 801 static int GNULINUX_ABI_DESC = 0; 802 803 static bool 804 linux_trans_osrel(const Elf_Note *note, int32_t *osrel) 805 { 806 const Elf32_Word *desc; 807 uintptr_t p; 808 809 p = (uintptr_t)(note + 1); 810 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 811 812 desc = (const Elf32_Word *)p; 813 if (desc[0] != GNULINUX_ABI_DESC) 814 return (false); 815 816 /* 817 * For Linux we encode osrel using the Linux convention of 818 * (version << 16) | (major << 8) | (minor) 819 * See macro in linux_mib.h 820 */ 821 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 822 823 return (true); 824 } 825 826 static Elf_Brandnote linux64_brandnote = { 827 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), 828 .hdr.n_descsz = 16, 829 .hdr.n_type = 1, 830 .vendor = GNULINUX_ABI_VENDOR, 831 .flags = BN_TRANSLATE_OSREL, 832 .trans_osrel = linux_trans_osrel 833 }; 834 835 static Elf64_Brandinfo linux_glibc2brand = { 836 .brand = ELFOSABI_LINUX, 837 .machine = EM_X86_64, 838 .compat_3_brand = "Linux", 839 .emul_path = linux_emul_path, 840 .interp_path = "/lib64/ld-linux-x86-64.so.2", 841 .sysvec = &elf_linux_sysvec, 842 .interp_newpath = NULL, 843 .brand_note = &linux64_brandnote, 844 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 845 }; 846 847 static Elf64_Brandinfo linux_glibc2brandshort = { 848 .brand = ELFOSABI_LINUX, 849 .machine = EM_X86_64, 850 .compat_3_brand = "Linux", 851 .emul_path = linux_emul_path, 852 .interp_path = "/lib64/ld-linux.so.2", 853 .sysvec = &elf_linux_sysvec, 854 .interp_newpath = NULL, 855 .brand_note = &linux64_brandnote, 856 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 857 }; 858 859 static Elf64_Brandinfo linux_muslbrand = { 860 .brand = ELFOSABI_LINUX, 861 .machine = EM_X86_64, 862 .compat_3_brand = "Linux", 863 .emul_path = linux_emul_path, 864 .interp_path = "/lib/ld-musl-x86_64.so.1", 865 .sysvec = &elf_linux_sysvec, 866 .interp_newpath = NULL, 867 .brand_note = &linux64_brandnote, 868 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 869 }; 870 871 Elf64_Brandinfo *linux_brandlist[] = { 872 &linux_glibc2brand, 873 &linux_glibc2brandshort, 874 &linux_muslbrand, 875 NULL 876 }; 877 878 static int 879 linux64_elf_modevent(module_t mod, int type, void *data) 880 { 881 Elf64_Brandinfo **brandinfo; 882 int error; 883 struct linux_ioctl_handler **lihp; 884 885 error = 0; 886 887 switch(type) { 888 case MOD_LOAD: 889 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 890 ++brandinfo) 891 if (elf64_insert_brand_entry(*brandinfo) < 0) 892 error = EINVAL; 893 if (error == 0) { 894 SET_FOREACH(lihp, linux_ioctl_handler_set) 895 linux_ioctl_register_handler(*lihp); 896 stclohz = (stathz ? stathz : hz); 897 if (bootverbose) 898 printf("Linux x86-64 ELF exec handler installed\n"); 899 } else 900 printf("cannot insert Linux x86-64 ELF brand handler\n"); 901 break; 902 case MOD_UNLOAD: 903 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 904 ++brandinfo) 905 if (elf64_brand_inuse(*brandinfo)) 906 error = EBUSY; 907 if (error == 0) { 908 for (brandinfo = &linux_brandlist[0]; 909 *brandinfo != NULL; ++brandinfo) 910 if (elf64_remove_brand_entry(*brandinfo) < 0) 911 error = EINVAL; 912 } 913 if (error == 0) { 914 SET_FOREACH(lihp, linux_ioctl_handler_set) 915 linux_ioctl_unregister_handler(*lihp); 916 if (bootverbose) 917 printf("Linux ELF exec handler removed\n"); 918 } else 919 printf("Could not deinstall ELF interpreter entry\n"); 920 break; 921 default: 922 return (EOPNOTSUPP); 923 } 924 return (error); 925 } 926 927 static moduledata_t linux64_elf_mod = { 928 "linux64elf", 929 linux64_elf_modevent, 930 0 931 }; 932 933 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 934 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 935 FEATURE(linux64, "Linux 64bit support"); 936