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 S��ren 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 86MODULE_VERSION(linux64, 1); 87 88const char *linux_kplatform; 89static int linux_szsigcode; 90static vm_object_t linux_shared_page_obj; 91static char *linux_shared_page_mapping; 92extern char _binary_linux_locore_o_start; 93extern char _binary_linux_locore_o_end; 94 95extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 96 97SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 98 99static int linux_copyout_strings(struct image_params *imgp, 100 uintptr_t *stack_base); 101static int linux_fixup_elf(uintptr_t *stack_base, 102 struct image_params *iparams); 103static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 104static void linux_vdso_install(void *param); 105static void linux_vdso_deinstall(void *param); 106static void linux_set_syscall_retval(struct thread *td, int error); 107static int linux_fetch_syscall_args(struct thread *td); 108static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 109 uintptr_t stack); 110static int linux_vsyscall(struct thread *td); 111 112#define LINUX_T_UNKNOWN 255 113static 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 151LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 152LINUX_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 */ 160static int 161linux_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 177static int 178linux_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]; 201 202 td->td_retval[0] = 0; 203 return (0); 204} 205 206static void 207linux_set_syscall_retval(struct thread *td, int error) 208{ 209 struct trapframe *frame; 210 211 frame = td->td_frame; 212 213 switch (error) { 214 case 0: 215 frame->tf_rax = td->td_retval[0]; 216 frame->tf_r10 = frame->tf_rcx; 217 break; 218 219 case ERESTART: 220 /* 221 * Reconstruct pc, we know that 'syscall' is 2 bytes, 222 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes. 223 * We saved this in tf_err. 224 * 225 */ 226 frame->tf_rip -= frame->tf_err; 227 frame->tf_r10 = frame->tf_rcx; 228 break; 229 230 case EJUSTRETURN: 231 break; 232 233 default: 234 frame->tf_rax = bsd_to_linux_errno(error); 235 frame->tf_r10 = frame->tf_rcx; 236 break; 237 } 238 239 /* 240 * Differently from FreeBSD native ABI, on Linux only %rcx 241 * and %r11 values are not preserved across the syscall. 242 * Require full context restore to get all registers except 243 * those two restored at return to usermode. 244 * 245 * XXX: Would be great to be able to avoid PCB_FULL_IRET 246 * for the error == 0 case. 247 */ 248 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 249} 250 251static int 252linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) 253{ 254 Elf_Auxargs *args; 255 Elf_Auxinfo *argarray, *pos; 256 struct proc *p; 257 int error, issetugid; 258 259 p = imgp->proc; 260 args = (Elf64_Auxargs *)imgp->auxargs; 261 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 262 M_WAITOK | M_ZERO); 263 264 issetugid = p->p_flag & P_SUGID ? 1 : 0; 265 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, 266 imgp->proc->p_sysent->sv_shared_page_base); 267 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 268 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 269 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 270 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 271 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 272 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 273 AUXARGS_ENTRY(pos, AT_BASE, args->base); 274 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 275 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 276 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 277 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 278 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 279 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 280 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 281 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 282 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); 283 if (imgp->execpathp != 0) 284 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); 285 if (args->execfd != -1) 286 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 287 AUXARGS_ENTRY(pos, AT_NULL, 0); 288 289 free(imgp->auxargs, M_TEMP); 290 imgp->auxargs = NULL; 291 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 292 293 error = copyout(argarray, (void *)base, 294 sizeof(*argarray) * LINUX_AT_COUNT); 295 free(argarray, M_TEMP); 296 return (error); 297} 298 299static int 300linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) 301{ 302 Elf_Addr *base; 303 304 base = (Elf64_Addr *)*stack_base; 305 base--; 306 if (suword(base, (uint64_t)imgp->args->argc) == -1) 307 return (EFAULT); 308 309 *stack_base = (uintptr_t)base; 310 return (0); 311} 312 313/* 314 * Copy strings out to the new process address space, constructing new arg 315 * and env vector tables. Return a pointer to the base so that it can be used 316 * as the initial stack pointer. 317 */ 318static int 319linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 320{ 321 int argc, envc, error; 322 char **vectp; 323 char *stringp; 324 uintptr_t destp, ustringp; 325 struct ps_strings *arginfo; 326 char canary[LINUX_AT_RANDOM_LEN]; 327 size_t execpath_len; 328 struct proc *p; 329 330 /* Calculate string base and vector table pointers. */ 331 if (imgp->execpath != NULL && imgp->auxargs != NULL) 332 execpath_len = strlen(imgp->execpath) + 1; 333 else 334 execpath_len = 0; 335 336 p = imgp->proc; 337 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 338 destp = (uintptr_t)arginfo; 339 340 if (execpath_len != 0) { 341 destp -= execpath_len; 342 destp = rounddown2(destp, sizeof(void *)); 343 imgp->execpathp = (void *)destp; 344 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 345 if (error != 0) 346 return (error); 347 } 348 349 /* Prepare the canary for SSP. */ 350 arc4rand(canary, sizeof(canary), 0); 351 destp -= roundup(sizeof(canary), sizeof(void *)); 352 imgp->canary = (void *)destp; 353 error = copyout(canary, imgp->canary, sizeof(canary)); 354 if (error != 0) 355 return (error); 356 357 /* Allocate room for the argument and environment strings. */ 358 destp -= ARG_MAX - imgp->args->stringspace; 359 destp = rounddown2(destp, sizeof(void *)); 360 ustringp = destp; 361 362 if (imgp->auxargs) { 363 /* 364 * Allocate room on the stack for the ELF auxargs 365 * array. It has LINUX_AT_COUNT entries. 366 */ 367 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo); 368 destp = rounddown2(destp, sizeof(void *)); 369 } 370 371 vectp = (char **)destp; 372 373 /* 374 * Allocate room for the argv[] and env vectors including the 375 * terminating NULL pointers. 376 */ 377 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 378 379 /* 380 * Starting with 2.24, glibc depends on a 16-byte stack alignment. 381 * One "long argc" will be prepended later. 382 */ 383 vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8); 384 385 /* vectp also becomes our initial stack base. */ 386 *stack_base = (uintptr_t)vectp; 387 388 stringp = imgp->args->begin_argv; 389 argc = imgp->args->argc; 390 envc = imgp->args->envc; 391 392 /* Copy out strings - arguments and environment. */ 393 error = copyout(stringp, (void *)ustringp, 394 ARG_MAX - imgp->args->stringspace); 395 if (error != 0) 396 return (error); 397 398 /* Fill in "ps_strings" struct for ps, w, etc. */ 399 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 400 suword(&arginfo->ps_nargvstr, argc) != 0) 401 return (EFAULT); 402 403 /* Fill in argument portion of vector table. */ 404 for (; argc > 0; --argc) { 405 if (suword(vectp++, ustringp) != 0) 406 return (EFAULT); 407 while (*stringp++ != 0) 408 ustringp++; 409 ustringp++; 410 } 411 412 /* A null vector table pointer separates the argp's from the envp's. */ 413 if (suword(vectp++, 0) != 0) 414 return (EFAULT); 415 416 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 417 suword(&arginfo->ps_nenvstr, envc) != 0) 418 return (EFAULT); 419 420 /* Fill in environment portion of vector table. */ 421 for (; envc > 0; --envc) { 422 if (suword(vectp++, ustringp) != 0) 423 return (EFAULT); 424 while (*stringp++ != 0) 425 ustringp++; 426 ustringp++; 427 } 428 429 /* The end of the vector table is a null pointer. */ 430 if (suword(vectp, 0) != 0) 431 return (EFAULT); 432 433 if (imgp->auxargs) { 434 vectp++; 435 error = imgp->sysent->sv_copyout_auxargs(imgp, 436 (uintptr_t)vectp); 437 if (error != 0) 438 return (error); 439 } 440 441 return (0); 442} 443 444/* 445 * Reset registers to default values on exec. 446 */ 447static void 448linux_exec_setregs(struct thread *td, struct image_params *imgp, 449 uintptr_t stack) 450{ 451 struct trapframe *regs; 452 struct pcb *pcb; 453 register_t saved_rflags; 454 455 regs = td->td_frame; 456 pcb = td->td_pcb; 457 458 if (td->td_proc->p_md.md_ldt != NULL) 459 user_ldt_free(td); 460 461 pcb->pcb_fsbase = 0; 462 pcb->pcb_gsbase = 0; 463 clear_pcb_flags(pcb, PCB_32BIT); 464 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 465 set_pcb_flags(pcb, PCB_FULL_IRET); 466 467 saved_rflags = regs->tf_rflags & PSL_T; 468 bzero((char *)regs, sizeof(struct trapframe)); 469 regs->tf_rip = imgp->entry_addr; 470 regs->tf_rsp = stack; 471 regs->tf_rflags = PSL_USER | saved_rflags; 472 regs->tf_ss = _udatasel; 473 regs->tf_cs = _ucodesel; 474 regs->tf_ds = _udatasel; 475 regs->tf_es = _udatasel; 476 regs->tf_fs = _ufssel; 477 regs->tf_gs = _ugssel; 478 regs->tf_flags = TF_HASSEGS; 479 480 x86_clear_dbregs(pcb); 481 482 /* 483 * Drop the FP state if we hold it, so that the process gets a 484 * clean FP state if it uses the FPU again. 485 */ 486 fpstate_drop(td); 487} 488 489/* 490 * Copied from amd64/amd64/machdep.c 491 * 492 * XXX fpu state need? don't think so 493 */ 494int 495linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 496{ 497 struct proc *p; 498 struct l_ucontext uc; 499 struct l_sigcontext *context; 500 struct trapframe *regs; 501 unsigned long rflags; 502 int error; 503 ksiginfo_t ksi; 504 505 regs = td->td_frame; 506 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); 507 if (error != 0) 508 return (error); 509 510 p = td->td_proc; 511 context = &uc.uc_mcontext; 512 rflags = context->sc_rflags; 513 514 /* 515 * Don't allow users to change privileged or reserved flags. 516 */ 517 /* 518 * XXX do allow users to change the privileged flag PSL_RF. 519 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 520 * should sometimes set it there too. tf_rflags is kept in 521 * the signal context during signal handling and there is no 522 * other place to remember it, so the PSL_RF bit may be 523 * corrupted by the signal handler without us knowing. 524 * Corruption of the PSL_RF bit at worst causes one more or 525 * one less debugger trap, so allowing it is fairly harmless. 526 */ 527 528#define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 529 if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 530 printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags); 531 return (EINVAL); 532 } 533 534 /* 535 * Don't allow users to load a valid privileged %cs. Let the 536 * hardware check for invalid selectors, excess privilege in 537 * other selectors, invalid %eip's and invalid %esp's. 538 */ 539#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 540 if (!CS_SECURE(context->sc_cs)) { 541 printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs); 542 ksiginfo_init_trap(&ksi); 543 ksi.ksi_signo = SIGBUS; 544 ksi.ksi_code = BUS_OBJERR; 545 ksi.ksi_trapno = T_PROTFLT; 546 ksi.ksi_addr = (void *)regs->tf_rip; 547 trapsignal(td, &ksi); 548 return (EINVAL); 549 } 550 551 PROC_LOCK(p); 552 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); 553 SIG_CANTMASK(td->td_sigmask); 554 signotify(td); 555 PROC_UNLOCK(p); 556 557 regs->tf_rdi = context->sc_rdi; 558 regs->tf_rsi = context->sc_rsi; 559 regs->tf_rdx = context->sc_rdx; 560 regs->tf_rbp = context->sc_rbp; 561 regs->tf_rbx = context->sc_rbx; 562 regs->tf_rcx = context->sc_rcx; 563 regs->tf_rax = context->sc_rax; 564 regs->tf_rip = context->sc_rip; 565 regs->tf_rsp = context->sc_rsp; 566 regs->tf_r8 = context->sc_r8; 567 regs->tf_r9 = context->sc_r9; 568 regs->tf_r10 = context->sc_r10; 569 regs->tf_r11 = context->sc_r11; 570 regs->tf_r12 = context->sc_r12; 571 regs->tf_r13 = context->sc_r13; 572 regs->tf_r14 = context->sc_r14; 573 regs->tf_r15 = context->sc_r15; 574 regs->tf_cs = context->sc_cs; 575 regs->tf_err = context->sc_err; 576 regs->tf_rflags = rflags; 577 578 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 579 return (EJUSTRETURN); 580} 581 582/* 583 * copied from amd64/amd64/machdep.c 584 * 585 * Send an interrupt to process. 586 */ 587static void 588linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 589{ 590 struct l_rt_sigframe sf, *sfp; 591 struct proc *p; 592 struct thread *td; 593 struct sigacts *psp; 594 caddr_t sp; 595 struct trapframe *regs; 596 int sig, code; 597 int oonstack; 598 599 td = curthread; 600 p = td->td_proc; 601 PROC_LOCK_ASSERT(p, MA_OWNED); 602 sig = ksi->ksi_signo; 603 psp = p->p_sigacts; 604 code = ksi->ksi_code; 605 mtx_assert(&psp->ps_mtx, MA_OWNED); 606 regs = td->td_frame; 607 oonstack = sigonstack(regs->tf_rsp); 608 609 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 610 catcher, sig, mask, code); 611 612 /* Allocate space for the signal handler context. */ 613 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 614 SIGISMEMBER(psp->ps_sigonstack, sig)) { 615 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - 616 sizeof(struct l_rt_sigframe); 617 } else 618 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128; 619 /* Align to 16 bytes. */ 620 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 621 mtx_unlock(&psp->ps_mtx); 622 623 /* Translate the signal. */ 624 sig = bsd_to_linux_signal(sig); 625 626 /* Save user context. */ 627 bzero(&sf, sizeof(sf)); 628 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask); 629 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask); 630 631 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 632 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 633 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 634 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 635 PROC_UNLOCK(p); 636 637 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi; 638 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi; 639 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx; 640 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp; 641 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx; 642 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx; 643 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax; 644 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip; 645 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp; 646 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8; 647 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9; 648 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10; 649 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11; 650 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12; 651 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13; 652 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14; 653 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15; 654 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 655 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags; 656 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err; 657 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 658 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 659 660 /* Build the argument list for the signal handler. */ 661 regs->tf_rdi = sig; /* arg 1 in %rdi */ 662 regs->tf_rax = 0; 663 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 664 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */ 665 666 sf.sf_handler = catcher; 667 /* Fill in POSIX parts. */ 668 ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig); 669 670 /* Copy the sigframe out to the user's stack. */ 671 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 672 PROC_LOCK(p); 673 sigexit(td, SIGILL); 674 } 675 676 regs->tf_rsp = (long)sfp; 677 regs->tf_rip = linux_rt_sigcode; 678 regs->tf_rflags &= ~(PSL_T | PSL_D); 679 regs->tf_cs = _ucodesel; 680 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 681 PROC_LOCK(p); 682 mtx_lock(&psp->ps_mtx); 683} 684 685#define LINUX_VSYSCALL_START (-10UL << 20) 686#define LINUX_VSYSCALL_SZ 1024 687 688const unsigned long linux_vsyscall_vector[] = { 689 LINUX_SYS_gettimeofday, 690 LINUX_SYS_linux_time, 691 LINUX_SYS_linux_getcpu, 692}; 693 694static int 695linux_vsyscall(struct thread *td) 696{ 697 struct trapframe *frame; 698 uint64_t retqaddr; 699 int code, traced; 700 int error; 701 702 frame = td->td_frame; 703 704 /* Check %rip for vsyscall area. */ 705 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 706 return (EINVAL); 707 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 708 return (EINVAL); 709 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 710 if (code >= nitems(linux_vsyscall_vector)) 711 return (EINVAL); 712 713 /* 714 * vsyscall called as callq *(%rax), so we must 715 * use return address from %rsp and also fixup %rsp. 716 */ 717 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 718 if (error) 719 return (error); 720 721 frame->tf_rip = retqaddr; 722 frame->tf_rax = linux_vsyscall_vector[code]; 723 frame->tf_rsp += 8; 724 725 traced = (frame->tf_flags & PSL_T); 726 727 amd64_syscall(td, traced); 728 729 return (0); 730} 731 732struct sysentvec elf_linux_sysvec = { 733 .sv_size = LINUX_SYS_MAXSYSCALL, 734 .sv_table = linux_sysent, 735 .sv_transtrap = linux_translate_traps, 736 .sv_fixup = linux_fixup_elf, 737 .sv_sendsig = linux_rt_sendsig, 738 .sv_sigcode = &_binary_linux_locore_o_start, 739 .sv_szsigcode = &linux_szsigcode, 740 .sv_name = "Linux ELF64", 741 .sv_coredump = elf64_coredump, 742 .sv_imgact_try = linux_exec_imgact_try, 743 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 744 .sv_minuser = VM_MIN_ADDRESS, 745 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48, 746 .sv_usrstack = USRSTACK_LA48, 747 .sv_psstrings = PS_STRINGS_LA48, 748 .sv_stackprot = VM_PROT_ALL, 749 .sv_copyout_auxargs = linux_copyout_auxargs, 750 .sv_copyout_strings = linux_copyout_strings, 751 .sv_setregs = linux_exec_setregs, 752 .sv_fixlimit = NULL, 753 .sv_maxssiz = NULL, 754 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN | 755 SV_SIG_WAITNDQ, 756 .sv_set_syscall_retval = linux_set_syscall_retval, 757 .sv_fetch_syscall_args = linux_fetch_syscall_args, 758 .sv_syscallnames = NULL, 759 .sv_shared_page_base = SHAREDPAGE_LA48, 760 .sv_shared_page_len = PAGE_SIZE, 761 .sv_schedtail = linux_schedtail, 762 .sv_thread_detach = linux_thread_detach, 763 .sv_trap = linux_vsyscall, 764 .sv_onexec = linux_on_exec, 765 .sv_onexit = linux_on_exit, 766 .sv_ontdexit = linux_thread_dtor, 767 .sv_setid_allowed = &linux_setid_allowed_query, 768}; 769 770static void 771linux_vdso_install(void *param) 772{ 773 774 amd64_lower_shared_page(&elf_linux_sysvec); 775 776 linux_szsigcode = (&_binary_linux_locore_o_end - 777 &_binary_linux_locore_o_start); 778 779 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 780 panic("Linux invalid vdso size\n"); 781 782 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 783 784 linux_shared_page_obj = __elfN(linux_shared_page_init) 785 (&linux_shared_page_mapping); 786 787 __elfN(linux_vdso_reloc)(&elf_linux_sysvec); 788 789 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 790 linux_szsigcode); 791 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 792 793 linux_kplatform = linux_shared_page_mapping + 794 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base); 795} 796SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 797 linux_vdso_install, NULL); 798 799static void 800linux_vdso_deinstall(void *param) 801{ 802 803 __elfN(linux_shared_page_fini)(linux_shared_page_obj, 804 linux_shared_page_mapping); 805} 806SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 807 linux_vdso_deinstall, NULL); 808 809static char GNULINUX_ABI_VENDOR[] = "GNU"; 810static int GNULINUX_ABI_DESC = 0; 811 812static bool 813linux_trans_osrel(const Elf_Note *note, int32_t *osrel) 814{ 815 const Elf32_Word *desc; 816 uintptr_t p; 817 818 p = (uintptr_t)(note + 1); 819 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 820 821 desc = (const Elf32_Word *)p; 822 if (desc[0] != GNULINUX_ABI_DESC) 823 return (false); 824 825 /* 826 * For Linux we encode osrel using the Linux convention of 827 * (version << 16) | (major << 8) | (minor) 828 * See macro in linux_mib.h 829 */ 830 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 831 832 return (true); 833} 834 835static Elf_Brandnote linux64_brandnote = { 836 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), 837 .hdr.n_descsz = 16, 838 .hdr.n_type = 1, 839 .vendor = GNULINUX_ABI_VENDOR, 840 .flags = BN_TRANSLATE_OSREL, 841 .trans_osrel = linux_trans_osrel 842}; 843 844static Elf64_Brandinfo linux_glibc2brand = { 845 .brand = ELFOSABI_LINUX, 846 .machine = EM_X86_64, 847 .compat_3_brand = "Linux", 848 .emul_path = linux_emul_path, 849 .interp_path = "/lib64/ld-linux-x86-64.so.2", 850 .sysvec = &elf_linux_sysvec, 851 .interp_newpath = NULL, 852 .brand_note = &linux64_brandnote, 853 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 854}; 855 856static Elf64_Brandinfo linux_glibc2brandshort = { 857 .brand = ELFOSABI_LINUX, 858 .machine = EM_X86_64, 859 .compat_3_brand = "Linux", 860 .emul_path = linux_emul_path, 861 .interp_path = "/lib64/ld-linux.so.2", 862 .sysvec = &elf_linux_sysvec, 863 .interp_newpath = NULL, 864 .brand_note = &linux64_brandnote, 865 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 866}; 867 868static Elf64_Brandinfo linux_muslbrand = { 869 .brand = ELFOSABI_LINUX, 870 .machine = EM_X86_64, 871 .compat_3_brand = "Linux", 872 .emul_path = linux_emul_path, 873 .interp_path = "/lib/ld-musl-x86_64.so.1", 874 .sysvec = &elf_linux_sysvec, 875 .interp_newpath = NULL, 876 .brand_note = &linux64_brandnote, 877 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 878}; 879 880Elf64_Brandinfo *linux_brandlist[] = { 881 &linux_glibc2brand, 882 &linux_glibc2brandshort, 883 &linux_muslbrand, 884 NULL 885}; 886 887static int 888linux64_elf_modevent(module_t mod, int type, void *data) 889{ 890 Elf64_Brandinfo **brandinfo; 891 int error; 892 struct linux_ioctl_handler **lihp; 893 894 error = 0; 895 896 switch(type) { 897 case MOD_LOAD: 898 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 899 ++brandinfo) 900 if (elf64_insert_brand_entry(*brandinfo) < 0) 901 error = EINVAL; 902 if (error == 0) { 903 SET_FOREACH(lihp, linux_ioctl_handler_set) 904 linux_ioctl_register_handler(*lihp); 905 stclohz = (stathz ? stathz : hz); 906 if (bootverbose) 907 printf("Linux x86-64 ELF exec handler installed\n"); 908 } else 909 printf("cannot insert Linux x86-64 ELF brand handler\n"); 910 break; 911 case MOD_UNLOAD: 912 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 913 ++brandinfo) 914 if (elf64_brand_inuse(*brandinfo)) 915 error = EBUSY; 916 if (error == 0) { 917 for (brandinfo = &linux_brandlist[0]; 918 *brandinfo != NULL; ++brandinfo) 919 if (elf64_remove_brand_entry(*brandinfo) < 0) 920 error = EINVAL; 921 } 922 if (error == 0) { 923 SET_FOREACH(lihp, linux_ioctl_handler_set) 924 linux_ioctl_unregister_handler(*lihp); 925 if (bootverbose) 926 printf("Linux ELF exec handler removed\n"); 927 } else 928 printf("Could not deinstall ELF interpreter entry\n"); 929 break; 930 default: 931 return (EOPNOTSUPP); 932 } 933 return (error); 934} 935 936static moduledata_t linux64_elf_mod = { 937 "linux64elf", 938 linux64_elf_modevent, 939 0 940}; 941 942DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 943MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 944FEATURE(linux64, "Linux 64bit support"); 945