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_futex.h> 79 #include <compat/linux/linux_ioctl.h> 80 #include <compat/linux/linux_mib.h> 81 #include <compat/linux/linux_misc.h> 82 #include <compat/linux/linux_signal.h> 83 #include <compat/linux/linux_sysproto.h> 84 #include <compat/linux/linux_util.h> 85 #include <compat/linux/linux_vdso.h> 86 87 MODULE_VERSION(linux64, 1); 88 89 #if defined(DEBUG) 90 SYSCTL_PROC(_compat_linux, OID_AUTO, debug, 91 CTLTYPE_STRING | CTLFLAG_RW, 92 0, 0, linux_sysctl_debug, "A", 93 "Linux 64 debugging control"); 94 #endif 95 96 /* 97 * Allow the sendsig functions to use the ldebug() facility even though they 98 * are not syscalls themselves. Map them to syscall 0. This is slightly less 99 * bogus than using ldebug(sigreturn). 100 */ 101 #define LINUX_SYS_linux_rt_sendsig 0 102 103 const char *linux_kplatform; 104 static int linux_szsigcode; 105 static vm_object_t linux_shared_page_obj; 106 static char *linux_shared_page_mapping; 107 extern char _binary_linux_locore_o_start; 108 extern char _binary_linux_locore_o_end; 109 110 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 111 112 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 113 114 static register_t * linux_copyout_strings(struct image_params *imgp); 115 static int linux_fixup_elf(register_t **stack_base, 116 struct image_params *iparams); 117 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 118 static void linux_vdso_install(void *param); 119 static void linux_vdso_deinstall(void *param); 120 static void linux_set_syscall_retval(struct thread *td, int error); 121 static int linux_fetch_syscall_args(struct thread *td); 122 static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 123 u_long stack); 124 static int linux_vsyscall(struct thread *td); 125 126 #define LINUX_T_UNKNOWN 255 127 static int _bsd_to_linux_trapcode[] = { 128 LINUX_T_UNKNOWN, /* 0 */ 129 6, /* 1 T_PRIVINFLT */ 130 LINUX_T_UNKNOWN, /* 2 */ 131 3, /* 3 T_BPTFLT */ 132 LINUX_T_UNKNOWN, /* 4 */ 133 LINUX_T_UNKNOWN, /* 5 */ 134 16, /* 6 T_ARITHTRAP */ 135 254, /* 7 T_ASTFLT */ 136 LINUX_T_UNKNOWN, /* 8 */ 137 13, /* 9 T_PROTFLT */ 138 1, /* 10 T_TRCTRAP */ 139 LINUX_T_UNKNOWN, /* 11 */ 140 14, /* 12 T_PAGEFLT */ 141 LINUX_T_UNKNOWN, /* 13 */ 142 17, /* 14 T_ALIGNFLT */ 143 LINUX_T_UNKNOWN, /* 15 */ 144 LINUX_T_UNKNOWN, /* 16 */ 145 LINUX_T_UNKNOWN, /* 17 */ 146 0, /* 18 T_DIVIDE */ 147 2, /* 19 T_NMI */ 148 4, /* 20 T_OFLOW */ 149 5, /* 21 T_BOUND */ 150 7, /* 22 T_DNA */ 151 8, /* 23 T_DOUBLEFLT */ 152 9, /* 24 T_FPOPFLT */ 153 10, /* 25 T_TSSFLT */ 154 11, /* 26 T_SEGNPFLT */ 155 12, /* 27 T_STKFLT */ 156 18, /* 28 T_MCHK */ 157 19, /* 29 T_XMMFLT */ 158 15 /* 30 T_RESERVED */ 159 }; 160 #define bsd_to_linux_trapcode(code) \ 161 ((code)<nitems(_bsd_to_linux_trapcode)? \ 162 _bsd_to_linux_trapcode[(code)]: \ 163 LINUX_T_UNKNOWN) 164 165 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 166 LINUX_VDSO_SYM_CHAR(linux_platform); 167 168 /* 169 * If FreeBSD & Linux have a difference of opinion about what a trap 170 * means, deal with it here. 171 * 172 * MPSAFE 173 */ 174 static int 175 linux_translate_traps(int signal, int trap_code) 176 { 177 178 if (signal != SIGBUS) 179 return (signal); 180 switch (trap_code) { 181 case T_PROTFLT: 182 case T_TSSFLT: 183 case T_DOUBLEFLT: 184 case T_PAGEFLT: 185 return (SIGSEGV); 186 default: 187 return (signal); 188 } 189 } 190 191 static int 192 linux_fetch_syscall_args(struct thread *td) 193 { 194 struct proc *p; 195 struct trapframe *frame; 196 struct syscall_args *sa; 197 198 p = td->td_proc; 199 frame = td->td_frame; 200 sa = &td->td_sa; 201 202 sa->args[0] = frame->tf_rdi; 203 sa->args[1] = frame->tf_rsi; 204 sa->args[2] = frame->tf_rdx; 205 sa->args[3] = frame->tf_rcx; 206 sa->args[4] = frame->tf_r8; 207 sa->args[5] = frame->tf_r9; 208 sa->code = frame->tf_rax; 209 210 if (sa->code >= p->p_sysent->sv_size) 211 /* nosys */ 212 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 213 else 214 sa->callp = &p->p_sysent->sv_table[sa->code]; 215 sa->narg = sa->callp->sy_narg; 216 217 td->td_retval[0] = 0; 218 return (0); 219 } 220 221 static void 222 linux_set_syscall_retval(struct thread *td, int error) 223 { 224 struct trapframe *frame = td->td_frame; 225 226 /* 227 * On Linux only %rcx and %r11 values are not preserved across 228 * the syscall. So, do not clobber %rdx and %r10. 229 */ 230 td->td_retval[1] = frame->tf_rdx; 231 frame->tf_r10 = frame->tf_rcx; 232 233 cpu_set_syscall_retval(td, error); 234 235 /* Restore all registers. */ 236 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 237 } 238 239 static int 240 linux_fixup_elf(register_t **stack_base, struct image_params *imgp) 241 { 242 Elf_Auxargs *args; 243 Elf_Auxinfo *argarray, *pos; 244 Elf_Addr *auxbase, *base; 245 struct ps_strings *arginfo; 246 struct proc *p; 247 int error, issetugid; 248 249 p = imgp->proc; 250 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 251 252 KASSERT(curthread->td_proc == imgp->proc, 253 ("unsafe linux_fixup_elf(), should be curproc")); 254 base = (Elf64_Addr *)*stack_base; 255 args = (Elf64_Auxargs *)imgp->auxargs; 256 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1; 257 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 258 M_WAITOK | M_ZERO); 259 260 issetugid = p->p_flag & P_SUGID ? 1 : 0; 261 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, 262 imgp->proc->p_sysent->sv_shared_page_base); 263 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 264 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 265 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 266 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 267 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 268 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 269 AUXARGS_ENTRY(pos, AT_BASE, args->base); 270 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 271 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 272 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 273 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 274 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 275 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 276 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 277 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 278 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary); 279 if (imgp->execpathp != 0) 280 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp); 281 if (args->execfd != -1) 282 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 283 AUXARGS_ENTRY(pos, AT_NULL, 0); 284 free(imgp->auxargs, M_TEMP); 285 imgp->auxargs = NULL; 286 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 287 288 error = copyout(argarray, auxbase, sizeof(*argarray) * LINUX_AT_COUNT); 289 free(argarray, M_TEMP); 290 if (error != 0) 291 return (error); 292 293 base--; 294 if (suword(base, (uint64_t)imgp->args->argc) == -1) 295 return (EFAULT); 296 297 *stack_base = (register_t *)base; 298 return (0); 299 } 300 301 /* 302 * Copy strings out to the new process address space, constructing new arg 303 * and env vector tables. Return a pointer to the base so that it can be used 304 * as the initial stack pointer. 305 */ 306 static register_t * 307 linux_copyout_strings(struct image_params *imgp) 308 { 309 int argc, envc; 310 char **vectp; 311 char *stringp, *destp; 312 register_t *stack_base; 313 struct ps_strings *arginfo; 314 char canary[LINUX_AT_RANDOM_LEN]; 315 size_t execpath_len; 316 struct proc *p; 317 318 /* Calculate string base and vector table pointers. */ 319 if (imgp->execpath != NULL && imgp->auxargs != NULL) 320 execpath_len = strlen(imgp->execpath) + 1; 321 else 322 execpath_len = 0; 323 324 p = imgp->proc; 325 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 326 destp = (caddr_t)arginfo - SPARE_USRSPACE - 327 roundup(sizeof(canary), sizeof(char *)) - 328 roundup(execpath_len, sizeof(char *)) - 329 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *)); 330 331 if (execpath_len != 0) { 332 imgp->execpathp = (uintptr_t)arginfo - execpath_len; 333 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 334 } 335 336 /* Prepare the canary for SSP. */ 337 arc4rand(canary, sizeof(canary), 0); 338 imgp->canary = (uintptr_t)arginfo - 339 roundup(execpath_len, sizeof(char *)) - 340 roundup(sizeof(canary), sizeof(char *)); 341 copyout(canary, (void *)imgp->canary, sizeof(canary)); 342 343 vectp = (char **)destp; 344 if (imgp->auxargs) { 345 /* 346 * Allocate room on the stack for the ELF auxargs 347 * array. It has LINUX_AT_COUNT entries. 348 */ 349 vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf64_Auxinfo), 350 sizeof(*vectp)); 351 } 352 353 /* 354 * Allocate room for the argv[] and env vectors including the 355 * terminating NULL pointers. 356 */ 357 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 358 359 /* vectp also becomes our initial stack base. */ 360 stack_base = (register_t *)vectp; 361 362 stringp = imgp->args->begin_argv; 363 argc = imgp->args->argc; 364 envc = imgp->args->envc; 365 366 /* Copy out strings - arguments and environment. */ 367 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 368 369 /* Fill in "ps_strings" struct for ps, w, etc. */ 370 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 371 suword(&arginfo->ps_nargvstr, argc); 372 373 /* Fill in argument portion of vector table. */ 374 for (; argc > 0; --argc) { 375 suword(vectp++, (long)(intptr_t)destp); 376 while (*stringp++ != 0) 377 destp++; 378 destp++; 379 } 380 381 /* A null vector table pointer separates the argp's from the envp's. */ 382 suword(vectp++, 0); 383 384 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 385 suword(&arginfo->ps_nenvstr, envc); 386 387 /* Fill in environment portion of vector table. */ 388 for (; envc > 0; --envc) { 389 suword(vectp++, (long)(intptr_t)destp); 390 while (*stringp++ != 0) 391 destp++; 392 destp++; 393 } 394 395 /* The end of the vector table is a null pointer. */ 396 suword(vectp, 0); 397 return (stack_base); 398 } 399 400 /* 401 * Reset registers to default values on exec. 402 */ 403 static void 404 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack) 405 { 406 struct trapframe *regs; 407 struct pcb *pcb; 408 register_t saved_rflags; 409 410 regs = td->td_frame; 411 pcb = td->td_pcb; 412 413 if (td->td_proc->p_md.md_ldt != NULL) 414 user_ldt_free(td); 415 416 pcb->pcb_fsbase = 0; 417 pcb->pcb_gsbase = 0; 418 clear_pcb_flags(pcb, PCB_32BIT); 419 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 420 set_pcb_flags(pcb, PCB_FULL_IRET); 421 422 saved_rflags = regs->tf_rflags & PSL_T; 423 bzero((char *)regs, sizeof(struct trapframe)); 424 regs->tf_rip = imgp->entry_addr; 425 regs->tf_rsp = stack; 426 regs->tf_rflags = PSL_USER | saved_rflags; 427 regs->tf_ss = _udatasel; 428 regs->tf_cs = _ucodesel; 429 regs->tf_ds = _udatasel; 430 regs->tf_es = _udatasel; 431 regs->tf_fs = _ufssel; 432 regs->tf_gs = _ugssel; 433 regs->tf_flags = TF_HASSEGS; 434 435 /* 436 * Reset the hardware debug registers if they were in use. 437 * They won't have any meaning for the newly exec'd process. 438 */ 439 if (pcb->pcb_flags & PCB_DBREGS) { 440 pcb->pcb_dr0 = 0; 441 pcb->pcb_dr1 = 0; 442 pcb->pcb_dr2 = 0; 443 pcb->pcb_dr3 = 0; 444 pcb->pcb_dr6 = 0; 445 pcb->pcb_dr7 = 0; 446 if (pcb == curpcb) { 447 /* 448 * Clear the debug registers on the running 449 * CPU, otherwise they will end up affecting 450 * the next process we switch to. 451 */ 452 reset_dbregs(); 453 } 454 clear_pcb_flags(pcb, PCB_DBREGS); 455 } 456 457 /* 458 * Drop the FP state if we hold it, so that the process gets a 459 * clean FP state if it uses the FPU again. 460 */ 461 fpstate_drop(td); 462 } 463 464 /* 465 * Copied from amd64/amd64/machdep.c 466 * 467 * XXX fpu state need? don't think so 468 */ 469 int 470 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 471 { 472 struct proc *p; 473 struct l_ucontext uc; 474 struct l_sigcontext *context; 475 struct trapframe *regs; 476 unsigned long rflags; 477 int error; 478 ksiginfo_t ksi; 479 480 regs = td->td_frame; 481 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); 482 if (error != 0) 483 return (error); 484 485 p = td->td_proc; 486 context = &uc.uc_mcontext; 487 rflags = context->sc_rflags; 488 489 /* 490 * Don't allow users to change privileged or reserved flags. 491 */ 492 /* 493 * XXX do allow users to change the privileged flag PSL_RF. 494 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 495 * should sometimes set it there too. tf_rflags is kept in 496 * the signal context during signal handling and there is no 497 * other place to remember it, so the PSL_RF bit may be 498 * corrupted by the signal handler without us knowing. 499 * Corruption of the PSL_RF bit at worst causes one more or 500 * one less debugger trap, so allowing it is fairly harmless. 501 */ 502 503 #define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 504 if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 505 printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags); 506 return (EINVAL); 507 } 508 509 /* 510 * Don't allow users to load a valid privileged %cs. Let the 511 * hardware check for invalid selectors, excess privilege in 512 * other selectors, invalid %eip's and invalid %esp's. 513 */ 514 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 515 if (!CS_SECURE(context->sc_cs)) { 516 printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs); 517 ksiginfo_init_trap(&ksi); 518 ksi.ksi_signo = SIGBUS; 519 ksi.ksi_code = BUS_OBJERR; 520 ksi.ksi_trapno = T_PROTFLT; 521 ksi.ksi_addr = (void *)regs->tf_rip; 522 trapsignal(td, &ksi); 523 return (EINVAL); 524 } 525 526 PROC_LOCK(p); 527 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); 528 SIG_CANTMASK(td->td_sigmask); 529 signotify(td); 530 PROC_UNLOCK(p); 531 532 regs->tf_rdi = context->sc_rdi; 533 regs->tf_rsi = context->sc_rsi; 534 regs->tf_rdx = context->sc_rdx; 535 regs->tf_rbp = context->sc_rbp; 536 regs->tf_rbx = context->sc_rbx; 537 regs->tf_rcx = context->sc_rcx; 538 regs->tf_rax = context->sc_rax; 539 regs->tf_rip = context->sc_rip; 540 regs->tf_rsp = context->sc_rsp; 541 regs->tf_r8 = context->sc_r8; 542 regs->tf_r9 = context->sc_r9; 543 regs->tf_r10 = context->sc_r10; 544 regs->tf_r11 = context->sc_r11; 545 regs->tf_r12 = context->sc_r12; 546 regs->tf_r13 = context->sc_r13; 547 regs->tf_r14 = context->sc_r14; 548 regs->tf_r15 = context->sc_r15; 549 regs->tf_cs = context->sc_cs; 550 regs->tf_err = context->sc_err; 551 regs->tf_rflags = rflags; 552 553 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 554 return (EJUSTRETURN); 555 } 556 557 /* 558 * copied from amd64/amd64/machdep.c 559 * 560 * Send an interrupt to process. 561 */ 562 static void 563 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 564 { 565 struct l_rt_sigframe sf, *sfp; 566 struct proc *p; 567 struct thread *td; 568 struct sigacts *psp; 569 caddr_t sp; 570 struct trapframe *regs; 571 int sig, code; 572 int oonstack; 573 574 td = curthread; 575 p = td->td_proc; 576 PROC_LOCK_ASSERT(p, MA_OWNED); 577 sig = ksi->ksi_signo; 578 psp = p->p_sigacts; 579 code = ksi->ksi_code; 580 mtx_assert(&psp->ps_mtx, MA_OWNED); 581 regs = td->td_frame; 582 oonstack = sigonstack(regs->tf_rsp); 583 584 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 585 catcher, sig, mask, code); 586 587 /* Allocate space for the signal handler context. */ 588 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 589 SIGISMEMBER(psp->ps_sigonstack, sig)) { 590 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - 591 sizeof(struct l_rt_sigframe); 592 } else 593 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128; 594 /* Align to 16 bytes. */ 595 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 596 mtx_unlock(&psp->ps_mtx); 597 598 /* Translate the signal. */ 599 sig = bsd_to_linux_signal(sig); 600 601 /* Save user context. */ 602 bzero(&sf, sizeof(sf)); 603 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask); 604 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask); 605 606 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 607 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 608 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 609 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 610 PROC_UNLOCK(p); 611 612 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi; 613 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi; 614 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx; 615 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp; 616 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx; 617 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx; 618 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax; 619 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip; 620 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp; 621 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8; 622 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9; 623 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10; 624 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11; 625 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12; 626 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13; 627 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14; 628 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15; 629 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 630 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags; 631 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err; 632 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 633 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 634 635 /* Build the argument list for the signal handler. */ 636 regs->tf_rdi = sig; /* arg 1 in %rdi */ 637 regs->tf_rax = 0; 638 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 639 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */ 640 641 sf.sf_handler = catcher; 642 /* Fill in POSIX parts. */ 643 ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig); 644 645 /* Copy the sigframe out to the user's stack. */ 646 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 647 #ifdef DEBUG 648 printf("process %ld has trashed its stack\n", (long)p->p_pid); 649 #endif 650 PROC_LOCK(p); 651 sigexit(td, SIGILL); 652 } 653 654 regs->tf_rsp = (long)sfp; 655 regs->tf_rip = linux_rt_sigcode; 656 regs->tf_rflags &= ~(PSL_T | PSL_D); 657 regs->tf_cs = _ucodesel; 658 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 659 PROC_LOCK(p); 660 mtx_lock(&psp->ps_mtx); 661 } 662 663 #define LINUX_VSYSCALL_START (-10UL << 20) 664 #define LINUX_VSYSCALL_SZ 1024 665 666 const unsigned long linux_vsyscall_vector[] = { 667 LINUX_SYS_gettimeofday, 668 LINUX_SYS_linux_time, 669 /* getcpu not implemented */ 670 }; 671 672 static int 673 linux_vsyscall(struct thread *td) 674 { 675 struct trapframe *frame; 676 uint64_t retqaddr; 677 int code, traced; 678 int error; 679 680 frame = td->td_frame; 681 682 /* Check %rip for vsyscall area. */ 683 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 684 return (EINVAL); 685 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 686 return (EINVAL); 687 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 688 if (code >= nitems(linux_vsyscall_vector)) 689 return (EINVAL); 690 691 /* 692 * vsyscall called as callq *(%rax), so we must 693 * use return address from %rsp and also fixup %rsp. 694 */ 695 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 696 if (error) 697 return (error); 698 699 frame->tf_rip = retqaddr; 700 frame->tf_rax = linux_vsyscall_vector[code]; 701 frame->tf_rsp += 8; 702 703 traced = (frame->tf_flags & PSL_T); 704 705 amd64_syscall(td, traced); 706 707 return (0); 708 } 709 710 struct sysentvec elf_linux_sysvec = { 711 .sv_size = LINUX_SYS_MAXSYSCALL, 712 .sv_table = linux_sysent, 713 .sv_mask = 0, 714 .sv_errsize = ELAST + 1, 715 .sv_errtbl = linux_errtbl, 716 .sv_transtrap = linux_translate_traps, 717 .sv_fixup = linux_fixup_elf, 718 .sv_sendsig = linux_rt_sendsig, 719 .sv_sigcode = &_binary_linux_locore_o_start, 720 .sv_szsigcode = &linux_szsigcode, 721 .sv_name = "Linux ELF64", 722 .sv_coredump = elf64_coredump, 723 .sv_imgact_try = linux_exec_imgact_try, 724 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 725 .sv_pagesize = PAGE_SIZE, 726 .sv_minuser = VM_MIN_ADDRESS, 727 .sv_maxuser = VM_MAXUSER_ADDRESS, 728 .sv_usrstack = USRSTACK, 729 .sv_psstrings = PS_STRINGS, 730 .sv_stackprot = VM_PROT_ALL, 731 .sv_copyout_strings = linux_copyout_strings, 732 .sv_setregs = linux_exec_setregs, 733 .sv_fixlimit = NULL, 734 .sv_maxssiz = NULL, 735 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP, 736 .sv_set_syscall_retval = linux_set_syscall_retval, 737 .sv_fetch_syscall_args = linux_fetch_syscall_args, 738 .sv_syscallnames = NULL, 739 .sv_shared_page_base = SHAREDPAGE, 740 .sv_shared_page_len = PAGE_SIZE, 741 .sv_schedtail = linux_schedtail, 742 .sv_thread_detach = linux_thread_detach, 743 .sv_trap = linux_vsyscall, 744 }; 745 746 static void 747 linux_vdso_install(void *param) 748 { 749 750 amd64_lower_shared_page(&elf_linux_sysvec); 751 752 linux_szsigcode = (&_binary_linux_locore_o_end - 753 &_binary_linux_locore_o_start); 754 755 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 756 panic("Linux invalid vdso size\n"); 757 758 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 759 760 linux_shared_page_obj = __elfN(linux_shared_page_init) 761 (&linux_shared_page_mapping); 762 763 __elfN(linux_vdso_reloc)(&elf_linux_sysvec); 764 765 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 766 linux_szsigcode); 767 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 768 769 linux_kplatform = linux_shared_page_mapping + 770 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base); 771 } 772 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 773 linux_vdso_install, NULL); 774 775 static void 776 linux_vdso_deinstall(void *param) 777 { 778 779 __elfN(linux_shared_page_fini)(linux_shared_page_obj); 780 } 781 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 782 linux_vdso_deinstall, NULL); 783 784 static char GNULINUX_ABI_VENDOR[] = "GNU"; 785 static int GNULINUX_ABI_DESC = 0; 786 787 static bool 788 linux_trans_osrel(const Elf_Note *note, int32_t *osrel) 789 { 790 const Elf32_Word *desc; 791 uintptr_t p; 792 793 p = (uintptr_t)(note + 1); 794 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 795 796 desc = (const Elf32_Word *)p; 797 if (desc[0] != GNULINUX_ABI_DESC) 798 return (false); 799 800 /* 801 * For Linux we encode osrel as follows (see linux_mib.c): 802 * VVVMMMIII (version, major, minor), see linux_mib.c. 803 */ 804 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3]; 805 806 return (true); 807 } 808 809 static Elf_Brandnote linux64_brandnote = { 810 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), 811 .hdr.n_descsz = 16, 812 .hdr.n_type = 1, 813 .vendor = GNULINUX_ABI_VENDOR, 814 .flags = BN_TRANSLATE_OSREL, 815 .trans_osrel = linux_trans_osrel 816 }; 817 818 static Elf64_Brandinfo linux_glibc2brand = { 819 .brand = ELFOSABI_LINUX, 820 .machine = EM_X86_64, 821 .compat_3_brand = "Linux", 822 .emul_path = "/compat/linux", 823 .interp_path = "/lib64/ld-linux-x86-64.so.2", 824 .sysvec = &elf_linux_sysvec, 825 .interp_newpath = NULL, 826 .brand_note = &linux64_brandnote, 827 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 828 }; 829 830 static Elf64_Brandinfo linux_glibc2brandshort = { 831 .brand = ELFOSABI_LINUX, 832 .machine = EM_X86_64, 833 .compat_3_brand = "Linux", 834 .emul_path = "/compat/linux", 835 .interp_path = "/lib64/ld-linux.so.2", 836 .sysvec = &elf_linux_sysvec, 837 .interp_newpath = NULL, 838 .brand_note = &linux64_brandnote, 839 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 840 }; 841 842 static Elf64_Brandinfo linux_muslbrand = { 843 .brand = ELFOSABI_LINUX, 844 .machine = EM_X86_64, 845 .compat_3_brand = "Linux", 846 .emul_path = "/compat/linux", 847 .interp_path = "/lib/ld-musl-x86_64.so.1", 848 .sysvec = &elf_linux_sysvec, 849 .interp_newpath = NULL, 850 .brand_note = &linux64_brandnote, 851 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 852 }; 853 854 Elf64_Brandinfo *linux_brandlist[] = { 855 &linux_glibc2brand, 856 &linux_glibc2brandshort, 857 &linux_muslbrand, 858 NULL 859 }; 860 861 static int 862 linux64_elf_modevent(module_t mod, int type, void *data) 863 { 864 Elf64_Brandinfo **brandinfo; 865 int error; 866 struct linux_ioctl_handler **lihp; 867 868 error = 0; 869 870 switch(type) { 871 case MOD_LOAD: 872 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 873 ++brandinfo) 874 if (elf64_insert_brand_entry(*brandinfo) < 0) 875 error = EINVAL; 876 if (error == 0) { 877 SET_FOREACH(lihp, linux_ioctl_handler_set) 878 linux_ioctl_register_handler(*lihp); 879 LIST_INIT(&futex_list); 880 mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF); 881 stclohz = (stathz ? stathz : hz); 882 if (bootverbose) 883 printf("Linux x86-64 ELF exec handler installed\n"); 884 } else 885 printf("cannot insert Linux x86-64 ELF brand handler\n"); 886 break; 887 case MOD_UNLOAD: 888 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 889 ++brandinfo) 890 if (elf64_brand_inuse(*brandinfo)) 891 error = EBUSY; 892 if (error == 0) { 893 for (brandinfo = &linux_brandlist[0]; 894 *brandinfo != NULL; ++brandinfo) 895 if (elf64_remove_brand_entry(*brandinfo) < 0) 896 error = EINVAL; 897 } 898 if (error == 0) { 899 SET_FOREACH(lihp, linux_ioctl_handler_set) 900 linux_ioctl_unregister_handler(*lihp); 901 mtx_destroy(&futex_mtx); 902 if (bootverbose) 903 printf("Linux ELF exec handler removed\n"); 904 } else 905 printf("Could not deinstall ELF interpreter entry\n"); 906 break; 907 default: 908 return (EOPNOTSUPP); 909 } 910 return (error); 911 } 912 913 static moduledata_t linux64_elf_mod = { 914 "linux64elf", 915 linux64_elf_modevent, 916 0 917 }; 918 919 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 920 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 921 FEATURE(linux64, "Linux 64bit support"); 922