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 BYTE_ORDER == LITTLE_ENDIAN 90 #define SHELLMAGIC 0x2123 /* #! */ 91 #else 92 #define SHELLMAGIC 0x2321 93 #endif 94 95 #if defined(DEBUG) 96 SYSCTL_PROC(_compat_linux, OID_AUTO, debug, 97 CTLTYPE_STRING | CTLFLAG_RW, 98 0, 0, linux_sysctl_debug, "A", 99 "Linux 64 debugging control"); 100 #endif 101 102 /* 103 * Allow the sendsig functions to use the ldebug() facility even though they 104 * are not syscalls themselves. Map them to syscall 0. This is slightly less 105 * bogus than using ldebug(sigreturn). 106 */ 107 #define LINUX_SYS_linux_rt_sendsig 0 108 109 const char *linux_kplatform; 110 static int linux_szsigcode; 111 static vm_object_t linux_shared_page_obj; 112 static char *linux_shared_page_mapping; 113 extern char _binary_linux_locore_o_start; 114 extern char _binary_linux_locore_o_end; 115 116 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 117 118 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 119 120 static register_t * linux_copyout_strings(struct image_params *imgp); 121 static int linux_fixup_elf(register_t **stack_base, 122 struct image_params *iparams); 123 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 124 static void linux_vdso_install(void *param); 125 static void linux_vdso_deinstall(void *param); 126 static void linux_set_syscall_retval(struct thread *td, int error); 127 static int linux_fetch_syscall_args(struct thread *td); 128 static int linux_exec_imgact_try(struct image_params *iparams); 129 static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 130 u_long stack); 131 static int linux_vsyscall(struct thread *td); 132 133 #define LINUX_T_UNKNOWN 255 134 static int _bsd_to_linux_trapcode[] = { 135 LINUX_T_UNKNOWN, /* 0 */ 136 6, /* 1 T_PRIVINFLT */ 137 LINUX_T_UNKNOWN, /* 2 */ 138 3, /* 3 T_BPTFLT */ 139 LINUX_T_UNKNOWN, /* 4 */ 140 LINUX_T_UNKNOWN, /* 5 */ 141 16, /* 6 T_ARITHTRAP */ 142 254, /* 7 T_ASTFLT */ 143 LINUX_T_UNKNOWN, /* 8 */ 144 13, /* 9 T_PROTFLT */ 145 1, /* 10 T_TRCTRAP */ 146 LINUX_T_UNKNOWN, /* 11 */ 147 14, /* 12 T_PAGEFLT */ 148 LINUX_T_UNKNOWN, /* 13 */ 149 17, /* 14 T_ALIGNFLT */ 150 LINUX_T_UNKNOWN, /* 15 */ 151 LINUX_T_UNKNOWN, /* 16 */ 152 LINUX_T_UNKNOWN, /* 17 */ 153 0, /* 18 T_DIVIDE */ 154 2, /* 19 T_NMI */ 155 4, /* 20 T_OFLOW */ 156 5, /* 21 T_BOUND */ 157 7, /* 22 T_DNA */ 158 8, /* 23 T_DOUBLEFLT */ 159 9, /* 24 T_FPOPFLT */ 160 10, /* 25 T_TSSFLT */ 161 11, /* 26 T_SEGNPFLT */ 162 12, /* 27 T_STKFLT */ 163 18, /* 28 T_MCHK */ 164 19, /* 29 T_XMMFLT */ 165 15 /* 30 T_RESERVED */ 166 }; 167 #define bsd_to_linux_trapcode(code) \ 168 ((code)<nitems(_bsd_to_linux_trapcode)? \ 169 _bsd_to_linux_trapcode[(code)]: \ 170 LINUX_T_UNKNOWN) 171 172 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 173 LINUX_VDSO_SYM_CHAR(linux_platform); 174 175 /* 176 * If FreeBSD & Linux have a difference of opinion about what a trap 177 * means, deal with it here. 178 * 179 * MPSAFE 180 */ 181 static int 182 linux_translate_traps(int signal, int trap_code) 183 { 184 185 if (signal != SIGBUS) 186 return (signal); 187 switch (trap_code) { 188 case T_PROTFLT: 189 case T_TSSFLT: 190 case T_DOUBLEFLT: 191 case T_PAGEFLT: 192 return (SIGSEGV); 193 default: 194 return (signal); 195 } 196 } 197 198 static int 199 linux_fetch_syscall_args(struct thread *td) 200 { 201 struct proc *p; 202 struct trapframe *frame; 203 struct syscall_args *sa; 204 205 p = td->td_proc; 206 frame = td->td_frame; 207 sa = &td->td_sa; 208 209 sa->args[0] = frame->tf_rdi; 210 sa->args[1] = frame->tf_rsi; 211 sa->args[2] = frame->tf_rdx; 212 sa->args[3] = frame->tf_rcx; 213 sa->args[4] = frame->tf_r8; 214 sa->args[5] = frame->tf_r9; 215 sa->code = frame->tf_rax; 216 217 if (sa->code >= p->p_sysent->sv_size) 218 /* nosys */ 219 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 220 else 221 sa->callp = &p->p_sysent->sv_table[sa->code]; 222 sa->narg = sa->callp->sy_narg; 223 224 td->td_retval[0] = 0; 225 return (0); 226 } 227 228 static void 229 linux_set_syscall_retval(struct thread *td, int error) 230 { 231 struct trapframe *frame = td->td_frame; 232 233 /* 234 * On Linux only %rcx and %r11 values are not preserved across 235 * the syscall. So, do not clobber %rdx and %r10. 236 */ 237 td->td_retval[1] = frame->tf_rdx; 238 frame->tf_r10 = frame->tf_rcx; 239 240 cpu_set_syscall_retval(td, error); 241 242 /* Restore all registers. */ 243 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 244 } 245 246 static int 247 linux_fixup_elf(register_t **stack_base, struct image_params *imgp) 248 { 249 Elf_Auxargs *args; 250 Elf_Addr *base; 251 Elf_Addr *pos; 252 struct ps_strings *arginfo; 253 struct proc *p; 254 int issetugid; 255 256 p = imgp->proc; 257 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 258 259 KASSERT(curthread->td_proc == imgp->proc, 260 ("unsafe linux_fixup_elf(), should be curproc")); 261 base = (Elf64_Addr *)*stack_base; 262 args = (Elf64_Auxargs *)imgp->auxargs; 263 pos = base + (imgp->args->argc + imgp->args->envc + 2); 264 265 issetugid = p->p_flag & P_SUGID ? 1 : 0; 266 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, 267 imgp->proc->p_sysent->sv_shared_page_base); 268 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 269 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 270 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 271 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 272 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 273 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 274 AUXARGS_ENTRY(pos, AT_BASE, args->base); 275 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 276 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 277 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 278 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 279 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 280 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 281 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 282 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 283 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary); 284 if (imgp->execpathp != 0) 285 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp); 286 if (args->execfd != -1) 287 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 288 AUXARGS_ENTRY(pos, AT_NULL, 0); 289 free(imgp->auxargs, M_TEMP); 290 imgp->auxargs = NULL; 291 292 base--; 293 suword(base, (uint64_t)imgp->args->argc); 294 295 *stack_base = (register_t *)base; 296 return (0); 297 } 298 299 /* 300 * Copy strings out to the new process address space, constructing new arg 301 * and env vector tables. Return a pointer to the base so that it can be used 302 * as the initial stack pointer. 303 */ 304 static register_t * 305 linux_copyout_strings(struct image_params *imgp) 306 { 307 int argc, envc; 308 char **vectp; 309 char *stringp, *destp; 310 register_t *stack_base; 311 struct ps_strings *arginfo; 312 char canary[LINUX_AT_RANDOM_LEN]; 313 size_t execpath_len; 314 struct proc *p; 315 316 /* Calculate string base and vector table pointers. */ 317 if (imgp->execpath != NULL && imgp->auxargs != NULL) 318 execpath_len = strlen(imgp->execpath) + 1; 319 else 320 execpath_len = 0; 321 322 p = imgp->proc; 323 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 324 destp = (caddr_t)arginfo - SPARE_USRSPACE - 325 roundup(sizeof(canary), sizeof(char *)) - 326 roundup(execpath_len, sizeof(char *)) - 327 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *)); 328 329 if (execpath_len != 0) { 330 imgp->execpathp = (uintptr_t)arginfo - execpath_len; 331 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 332 } 333 334 /* Prepare the canary for SSP. */ 335 arc4rand(canary, sizeof(canary), 0); 336 imgp->canary = (uintptr_t)arginfo - 337 roundup(execpath_len, sizeof(char *)) - 338 roundup(sizeof(canary), sizeof(char *)); 339 copyout(canary, (void *)imgp->canary, sizeof(canary)); 340 341 /* If we have a valid auxargs ptr, prepare some room on the stack. */ 342 if (imgp->auxargs) { 343 /* 344 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 345 * lower compatibility. 346 */ 347 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 348 (LINUX_AT_COUNT * 2); 349 350 /* 351 * The '+ 2' is for the null pointers at the end of each of 352 * the arg and env vector sets,and imgp->auxarg_size is room 353 * for argument of Runtime loader. 354 */ 355 vectp = (char **)(destp - (imgp->args->argc + 356 imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *)); 357 358 } else { 359 /* 360 * The '+ 2' is for the null pointers at the end of each of 361 * the arg and env vector sets 362 */ 363 vectp = (char **)(destp - (imgp->args->argc + 364 imgp->args->envc + 2) * sizeof(char *)); 365 } 366 367 /* vectp also becomes our initial stack base. */ 368 stack_base = (register_t *)vectp; 369 370 stringp = imgp->args->begin_argv; 371 argc = imgp->args->argc; 372 envc = imgp->args->envc; 373 374 /* Copy out strings - arguments and environment. */ 375 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 376 377 /* Fill in "ps_strings" struct for ps, w, etc. */ 378 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 379 suword(&arginfo->ps_nargvstr, argc); 380 381 /* Fill in argument portion of vector table. */ 382 for (; argc > 0; --argc) { 383 suword(vectp++, (long)(intptr_t)destp); 384 while (*stringp++ != 0) 385 destp++; 386 destp++; 387 } 388 389 /* A null vector table pointer separates the argp's from the envp's. */ 390 suword(vectp++, 0); 391 392 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 393 suword(&arginfo->ps_nenvstr, envc); 394 395 /* Fill in environment portion of vector table. */ 396 for (; envc > 0; --envc) { 397 suword(vectp++, (long)(intptr_t)destp); 398 while (*stringp++ != 0) 399 destp++; 400 destp++; 401 } 402 403 /* The end of the vector table is a null pointer. */ 404 suword(vectp, 0); 405 return (stack_base); 406 } 407 408 /* 409 * Reset registers to default values on exec. 410 */ 411 static void 412 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack) 413 { 414 struct trapframe *regs = td->td_frame; 415 struct pcb *pcb = td->td_pcb; 416 417 if (td->td_proc->p_md.md_ldt != NULL) 418 user_ldt_free(td); 419 420 pcb->pcb_fsbase = 0; 421 pcb->pcb_gsbase = 0; 422 clear_pcb_flags(pcb, PCB_32BIT); 423 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 424 set_pcb_flags(pcb, PCB_FULL_IRET); 425 426 bzero((char *)regs, sizeof(struct trapframe)); 427 regs->tf_rip = imgp->entry_addr; 428 regs->tf_rsp = stack; 429 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T); 430 regs->tf_ss = _udatasel; 431 regs->tf_cs = _ucodesel; 432 regs->tf_ds = _udatasel; 433 regs->tf_es = _udatasel; 434 regs->tf_fs = _ufssel; 435 regs->tf_gs = _ugssel; 436 regs->tf_flags = TF_HASSEGS; 437 438 /* 439 * Reset the hardware debug registers if they were in use. 440 * They won't have any meaning for the newly exec'd process. 441 */ 442 if (pcb->pcb_flags & PCB_DBREGS) { 443 pcb->pcb_dr0 = 0; 444 pcb->pcb_dr1 = 0; 445 pcb->pcb_dr2 = 0; 446 pcb->pcb_dr3 = 0; 447 pcb->pcb_dr6 = 0; 448 pcb->pcb_dr7 = 0; 449 if (pcb == curpcb) { 450 /* 451 * Clear the debug registers on the running 452 * CPU, otherwise they will end up affecting 453 * the next process we switch to. 454 */ 455 reset_dbregs(); 456 } 457 clear_pcb_flags(pcb, PCB_DBREGS); 458 } 459 460 /* 461 * Drop the FP state if we hold it, so that the process gets a 462 * clean FP state if it uses the FPU again. 463 */ 464 fpstate_drop(td); 465 } 466 467 /* 468 * Copied from amd64/amd64/machdep.c 469 * 470 * XXX fpu state need? don't think so 471 */ 472 int 473 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 474 { 475 struct proc *p; 476 struct l_ucontext uc; 477 struct l_sigcontext *context; 478 struct trapframe *regs; 479 unsigned long rflags; 480 int error; 481 ksiginfo_t ksi; 482 483 regs = td->td_frame; 484 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); 485 if (error != 0) 486 return (error); 487 488 p = td->td_proc; 489 context = &uc.uc_mcontext; 490 rflags = context->sc_rflags; 491 492 /* 493 * Don't allow users to change privileged or reserved flags. 494 */ 495 /* 496 * XXX do allow users to change the privileged flag PSL_RF. 497 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 498 * should sometimes set it there too. tf_rflags is kept in 499 * the signal context during signal handling and there is no 500 * other place to remember it, so the PSL_RF bit may be 501 * corrupted by the signal handler without us knowing. 502 * Corruption of the PSL_RF bit at worst causes one more or 503 * one less debugger trap, so allowing it is fairly harmless. 504 */ 505 506 #define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 507 if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 508 printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags); 509 return (EINVAL); 510 } 511 512 /* 513 * Don't allow users to load a valid privileged %cs. Let the 514 * hardware check for invalid selectors, excess privilege in 515 * other selectors, invalid %eip's and invalid %esp's. 516 */ 517 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 518 if (!CS_SECURE(context->sc_cs)) { 519 printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs); 520 ksiginfo_init_trap(&ksi); 521 ksi.ksi_signo = SIGBUS; 522 ksi.ksi_code = BUS_OBJERR; 523 ksi.ksi_trapno = T_PROTFLT; 524 ksi.ksi_addr = (void *)regs->tf_rip; 525 trapsignal(td, &ksi); 526 return (EINVAL); 527 } 528 529 PROC_LOCK(p); 530 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); 531 SIG_CANTMASK(td->td_sigmask); 532 signotify(td); 533 PROC_UNLOCK(p); 534 535 regs->tf_rdi = context->sc_rdi; 536 regs->tf_rsi = context->sc_rsi; 537 regs->tf_rdx = context->sc_rdx; 538 regs->tf_rbp = context->sc_rbp; 539 regs->tf_rbx = context->sc_rbx; 540 regs->tf_rcx = context->sc_rcx; 541 regs->tf_rax = context->sc_rax; 542 regs->tf_rip = context->sc_rip; 543 regs->tf_rsp = context->sc_rsp; 544 regs->tf_r8 = context->sc_r8; 545 regs->tf_r9 = context->sc_r9; 546 regs->tf_r10 = context->sc_r10; 547 regs->tf_r11 = context->sc_r11; 548 regs->tf_r12 = context->sc_r12; 549 regs->tf_r13 = context->sc_r13; 550 regs->tf_r14 = context->sc_r14; 551 regs->tf_r15 = context->sc_r15; 552 regs->tf_cs = context->sc_cs; 553 regs->tf_err = context->sc_err; 554 regs->tf_rflags = rflags; 555 556 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 557 return (EJUSTRETURN); 558 } 559 560 /* 561 * copied from amd64/amd64/machdep.c 562 * 563 * Send an interrupt to process. 564 */ 565 static void 566 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 567 { 568 struct l_rt_sigframe sf, *sfp; 569 struct proc *p; 570 struct thread *td; 571 struct sigacts *psp; 572 caddr_t sp; 573 struct trapframe *regs; 574 int sig, code; 575 int oonstack; 576 577 td = curthread; 578 p = td->td_proc; 579 PROC_LOCK_ASSERT(p, MA_OWNED); 580 sig = ksi->ksi_signo; 581 psp = p->p_sigacts; 582 code = ksi->ksi_code; 583 mtx_assert(&psp->ps_mtx, MA_OWNED); 584 regs = td->td_frame; 585 oonstack = sigonstack(regs->tf_rsp); 586 587 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 588 catcher, sig, mask, code); 589 590 /* Allocate space for the signal handler context. */ 591 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 592 SIGISMEMBER(psp->ps_sigonstack, sig)) { 593 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - 594 sizeof(struct l_rt_sigframe); 595 } else 596 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128; 597 /* Align to 16 bytes. */ 598 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 599 mtx_unlock(&psp->ps_mtx); 600 601 /* Translate the signal. */ 602 sig = bsd_to_linux_signal(sig); 603 604 /* Save user context. */ 605 bzero(&sf, sizeof(sf)); 606 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask); 607 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask); 608 609 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 610 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 611 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 612 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 613 PROC_UNLOCK(p); 614 615 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi; 616 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi; 617 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx; 618 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp; 619 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx; 620 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx; 621 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax; 622 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip; 623 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp; 624 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8; 625 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9; 626 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10; 627 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11; 628 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12; 629 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13; 630 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14; 631 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15; 632 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 633 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags; 634 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err; 635 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 636 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 637 638 /* Build the argument list for the signal handler. */ 639 regs->tf_rdi = sig; /* arg 1 in %rdi */ 640 regs->tf_rax = 0; 641 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 642 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */ 643 644 sf.sf_handler = catcher; 645 /* Fill in POSIX parts. */ 646 ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig); 647 648 /* Copy the sigframe out to the user's stack. */ 649 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 650 #ifdef DEBUG 651 printf("process %ld has trashed its stack\n", (long)p->p_pid); 652 #endif 653 PROC_LOCK(p); 654 sigexit(td, SIGILL); 655 } 656 657 regs->tf_rsp = (long)sfp; 658 regs->tf_rip = linux_rt_sigcode; 659 regs->tf_rflags &= ~(PSL_T | PSL_D); 660 regs->tf_cs = _ucodesel; 661 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 662 PROC_LOCK(p); 663 mtx_lock(&psp->ps_mtx); 664 } 665 666 /* 667 * If a Linux binary is exec'ing something, try this image activator 668 * first. We override standard shell script execution in order to 669 * be able to modify the interpreter path. We only do this if a Linux 670 * binary is doing the exec, so we do not create an EXEC module for it. 671 */ 672 static int 673 linux_exec_imgact_try(struct image_params *imgp) 674 { 675 const char *head = (const char *)imgp->image_header; 676 char *rpath; 677 int error = -1; 678 679 /* 680 * The interpreter for shell scripts run from a Linux binary needs 681 * to be located in /compat/linux if possible in order to recursively 682 * maintain Linux path emulation. 683 */ 684 if (((const short *)head)[0] == SHELLMAGIC) { 685 /* 686 * Run our normal shell image activator. If it succeeds then 687 * attempt to use the alternate path for the interpreter. If 688 * an alternate path is found, use our stringspace to store it. 689 */ 690 if ((error = exec_shell_imgact(imgp)) == 0) { 691 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc), 692 imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, 693 AT_FDCWD); 694 if (rpath != NULL) 695 imgp->args->fname_buf = 696 imgp->interpreter_name = rpath; 697 } 698 } 699 return (error); 700 } 701 702 #define LINUX_VSYSCALL_START (-10UL << 20) 703 #define LINUX_VSYSCALL_SZ 1024 704 705 const unsigned long linux_vsyscall_vector[] = { 706 LINUX_SYS_gettimeofday, 707 LINUX_SYS_linux_time, 708 /* getcpu not implemented */ 709 }; 710 711 static int 712 linux_vsyscall(struct thread *td) 713 { 714 struct trapframe *frame; 715 uint64_t retqaddr; 716 int code, traced; 717 int error; 718 719 frame = td->td_frame; 720 721 /* Check %rip for vsyscall area. */ 722 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 723 return (EINVAL); 724 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 725 return (EINVAL); 726 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 727 if (code >= nitems(linux_vsyscall_vector)) 728 return (EINVAL); 729 730 /* 731 * vsyscall called as callq *(%rax), so we must 732 * use return address from %rsp and also fixup %rsp. 733 */ 734 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 735 if (error) 736 return (error); 737 738 frame->tf_rip = retqaddr; 739 frame->tf_rax = linux_vsyscall_vector[code]; 740 frame->tf_rsp += 8; 741 742 traced = (frame->tf_flags & PSL_T); 743 744 amd64_syscall(td, traced); 745 746 return (0); 747 } 748 749 struct sysentvec elf_linux_sysvec = { 750 .sv_size = LINUX_SYS_MAXSYSCALL, 751 .sv_table = linux_sysent, 752 .sv_mask = 0, 753 .sv_errsize = ELAST + 1, 754 .sv_errtbl = linux_errtbl, 755 .sv_transtrap = linux_translate_traps, 756 .sv_fixup = linux_fixup_elf, 757 .sv_sendsig = linux_rt_sendsig, 758 .sv_sigcode = &_binary_linux_locore_o_start, 759 .sv_szsigcode = &linux_szsigcode, 760 .sv_name = "Linux ELF64", 761 .sv_coredump = elf64_coredump, 762 .sv_imgact_try = linux_exec_imgact_try, 763 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 764 .sv_pagesize = PAGE_SIZE, 765 .sv_minuser = VM_MIN_ADDRESS, 766 .sv_maxuser = VM_MAXUSER_ADDRESS, 767 .sv_usrstack = USRSTACK, 768 .sv_psstrings = PS_STRINGS, 769 .sv_stackprot = VM_PROT_ALL, 770 .sv_copyout_strings = linux_copyout_strings, 771 .sv_setregs = linux_exec_setregs, 772 .sv_fixlimit = NULL, 773 .sv_maxssiz = NULL, 774 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP, 775 .sv_set_syscall_retval = linux_set_syscall_retval, 776 .sv_fetch_syscall_args = linux_fetch_syscall_args, 777 .sv_syscallnames = NULL, 778 .sv_shared_page_base = SHAREDPAGE, 779 .sv_shared_page_len = PAGE_SIZE, 780 .sv_schedtail = linux_schedtail, 781 .sv_thread_detach = linux_thread_detach, 782 .sv_trap = linux_vsyscall, 783 }; 784 785 static void 786 linux_vdso_install(void *param) 787 { 788 789 amd64_lower_shared_page(&elf_linux_sysvec); 790 791 linux_szsigcode = (&_binary_linux_locore_o_end - 792 &_binary_linux_locore_o_start); 793 794 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 795 panic("Linux invalid vdso size\n"); 796 797 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 798 799 linux_shared_page_obj = __elfN(linux_shared_page_init) 800 (&linux_shared_page_mapping); 801 802 __elfN(linux_vdso_reloc)(&elf_linux_sysvec); 803 804 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 805 linux_szsigcode); 806 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 807 808 linux_kplatform = linux_shared_page_mapping + 809 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base); 810 } 811 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 812 linux_vdso_install, NULL); 813 814 static void 815 linux_vdso_deinstall(void *param) 816 { 817 818 __elfN(linux_shared_page_fini)(linux_shared_page_obj); 819 } 820 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 821 linux_vdso_deinstall, NULL); 822 823 static char GNULINUX_ABI_VENDOR[] = "GNU"; 824 static int GNULINUX_ABI_DESC = 0; 825 826 static bool 827 linux_trans_osrel(const Elf_Note *note, int32_t *osrel) 828 { 829 const Elf32_Word *desc; 830 uintptr_t p; 831 832 p = (uintptr_t)(note + 1); 833 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 834 835 desc = (const Elf32_Word *)p; 836 if (desc[0] != GNULINUX_ABI_DESC) 837 return (false); 838 839 /* 840 * For Linux we encode osrel as follows (see linux_mib.c): 841 * VVVMMMIII (version, major, minor), see linux_mib.c. 842 */ 843 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3]; 844 845 return (true); 846 } 847 848 static Elf_Brandnote linux64_brandnote = { 849 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), 850 .hdr.n_descsz = 16, 851 .hdr.n_type = 1, 852 .vendor = GNULINUX_ABI_VENDOR, 853 .flags = BN_TRANSLATE_OSREL, 854 .trans_osrel = linux_trans_osrel 855 }; 856 857 static Elf64_Brandinfo linux_glibc2brand = { 858 .brand = ELFOSABI_LINUX, 859 .machine = EM_X86_64, 860 .compat_3_brand = "Linux", 861 .emul_path = "/compat/linux", 862 .interp_path = "/lib64/ld-linux-x86-64.so.2", 863 .sysvec = &elf_linux_sysvec, 864 .interp_newpath = NULL, 865 .brand_note = &linux64_brandnote, 866 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 867 }; 868 869 static Elf64_Brandinfo linux_glibc2brandshort = { 870 .brand = ELFOSABI_LINUX, 871 .machine = EM_X86_64, 872 .compat_3_brand = "Linux", 873 .emul_path = "/compat/linux", 874 .interp_path = "/lib64/ld-linux.so.2", 875 .sysvec = &elf_linux_sysvec, 876 .interp_newpath = NULL, 877 .brand_note = &linux64_brandnote, 878 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 879 }; 880 881 static Elf64_Brandinfo linux_muslbrand = { 882 .brand = ELFOSABI_LINUX, 883 .machine = EM_X86_64, 884 .compat_3_brand = "Linux", 885 .emul_path = "/compat/linux", 886 .interp_path = "/lib/ld-musl-x86_64.so.1", 887 .sysvec = &elf_linux_sysvec, 888 .interp_newpath = NULL, 889 .brand_note = &linux64_brandnote, 890 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 891 }; 892 893 Elf64_Brandinfo *linux_brandlist[] = { 894 &linux_glibc2brand, 895 &linux_glibc2brandshort, 896 &linux_muslbrand, 897 NULL 898 }; 899 900 static int 901 linux64_elf_modevent(module_t mod, int type, void *data) 902 { 903 Elf64_Brandinfo **brandinfo; 904 int error; 905 struct linux_ioctl_handler **lihp; 906 907 error = 0; 908 909 switch(type) { 910 case MOD_LOAD: 911 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 912 ++brandinfo) 913 if (elf64_insert_brand_entry(*brandinfo) < 0) 914 error = EINVAL; 915 if (error == 0) { 916 SET_FOREACH(lihp, linux_ioctl_handler_set) 917 linux_ioctl_register_handler(*lihp); 918 LIST_INIT(&futex_list); 919 mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF); 920 stclohz = (stathz ? stathz : hz); 921 if (bootverbose) 922 printf("Linux x86-64 ELF exec handler installed\n"); 923 } else 924 printf("cannot insert Linux x86-64 ELF brand handler\n"); 925 break; 926 case MOD_UNLOAD: 927 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 928 ++brandinfo) 929 if (elf64_brand_inuse(*brandinfo)) 930 error = EBUSY; 931 if (error == 0) { 932 for (brandinfo = &linux_brandlist[0]; 933 *brandinfo != NULL; ++brandinfo) 934 if (elf64_remove_brand_entry(*brandinfo) < 0) 935 error = EINVAL; 936 } 937 if (error == 0) { 938 SET_FOREACH(lihp, linux_ioctl_handler_set) 939 linux_ioctl_unregister_handler(*lihp); 940 mtx_destroy(&futex_mtx); 941 if (bootverbose) 942 printf("Linux ELF exec handler removed\n"); 943 } else 944 printf("Could not deinstall ELF interpreter entry\n"); 945 break; 946 default: 947 return (EOPNOTSUPP); 948 } 949 return (error); 950 } 951 952 static moduledata_t linux64_elf_mod = { 953 "linux64elf", 954 linux64_elf_modevent, 955 0 956 }; 957 958 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 959 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 960 FEATURE(linux64, "Linux 64bit support"); 961