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