1 /*- 2 * Copyright (c) 2004 Tim J. Robbins 3 * Copyright (c) 2003 Peter Wemm 4 * Copyright (c) 2002 Doug Rabson 5 * Copyright (c) 1998-1999 Andrew Gallatin 6 * Copyright (c) 1994-1996 Søren Schmidt 7 * All rights reserved. 8 * Copyright (c) 2013, 2021 Dmitry Chagin <dchagin@FreeBSD.org> 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/stddef.h> 54 #include <sys/signalvar.h> 55 #include <sys/syscallsubr.h> 56 #include <sys/sysctl.h> 57 #include <sys/sysent.h> 58 #include <sys/sysproto.h> 59 #include <sys/vnode.h> 60 #include <sys/eventhandler.h> 61 62 #include <vm/vm.h> 63 #include <vm/pmap.h> 64 #include <vm/vm_extern.h> 65 #include <vm/vm_map.h> 66 #include <vm/vm_object.h> 67 #include <vm/vm_page.h> 68 #include <vm/vm_param.h> 69 70 #include <machine/cpu.h> 71 #include <machine/md_var.h> 72 #include <machine/pcb.h> 73 #include <machine/specialreg.h> 74 #include <machine/trap.h> 75 76 #include <x86/linux/linux_x86.h> 77 #include <amd64/linux/linux.h> 78 #include <amd64/linux/linux_proto.h> 79 #include <compat/linux/linux_emul.h> 80 #include <compat/linux/linux_fork.h> 81 #include <compat/linux/linux_ioctl.h> 82 #include <compat/linux/linux_mib.h> 83 #include <compat/linux/linux_misc.h> 84 #include <compat/linux/linux_signal.h> 85 #include <compat/linux/linux_sysproto.h> 86 #include <compat/linux/linux_util.h> 87 #include <compat/linux/linux_vdso.h> 88 89 MODULE_VERSION(linux64, 1); 90 91 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2 92 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \ 93 LINUX_VDSOPAGE_SIZE) 94 #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE) 95 /* 96 * PAGE_SIZE - the size 97 * of the native SHAREDPAGE 98 */ 99 #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48 100 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \ 101 sizeof(struct ps_strings)) 102 103 static int linux_szsigcode; 104 static vm_object_t linux_vdso_obj; 105 static char *linux_vdso_mapping; 106 extern char _binary_linux_vdso_so_o_start; 107 extern char _binary_linux_vdso_so_o_end; 108 static vm_offset_t linux_vdso_base; 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 int linux_copyout_strings(struct image_params *imgp, 115 uintptr_t *stack_base); 116 static int linux_fixup_elf(uintptr_t *stack_base, 117 struct image_params *iparams); 118 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 119 static void linux_vdso_install(const void *param); 120 static void linux_vdso_deinstall(const void *param); 121 static void linux_vdso_reloc(char *mapping, Elf_Addr offset); 122 static void linux_set_syscall_retval(struct thread *td, int error); 123 static int linux_fetch_syscall_args(struct thread *td); 124 static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 125 uintptr_t stack); 126 static void linux_exec_sysvec_init(void *param); 127 static int linux_on_exec_vmspace(struct proc *p, 128 struct image_params *imgp); 129 static void linux_set_fork_retval(struct thread *td); 130 static int linux_vsyscall(struct thread *td); 131 132 #define LINUX_T_UNKNOWN 255 133 static int _bsd_to_linux_trapcode[] = { 134 LINUX_T_UNKNOWN, /* 0 */ 135 6, /* 1 T_PRIVINFLT */ 136 LINUX_T_UNKNOWN, /* 2 */ 137 3, /* 3 T_BPTFLT */ 138 LINUX_T_UNKNOWN, /* 4 */ 139 LINUX_T_UNKNOWN, /* 5 */ 140 16, /* 6 T_ARITHTRAP */ 141 254, /* 7 T_ASTFLT */ 142 LINUX_T_UNKNOWN, /* 8 */ 143 13, /* 9 T_PROTFLT */ 144 1, /* 10 T_TRCTRAP */ 145 LINUX_T_UNKNOWN, /* 11 */ 146 14, /* 12 T_PAGEFLT */ 147 LINUX_T_UNKNOWN, /* 13 */ 148 17, /* 14 T_ALIGNFLT */ 149 LINUX_T_UNKNOWN, /* 15 */ 150 LINUX_T_UNKNOWN, /* 16 */ 151 LINUX_T_UNKNOWN, /* 17 */ 152 0, /* 18 T_DIVIDE */ 153 2, /* 19 T_NMI */ 154 4, /* 20 T_OFLOW */ 155 5, /* 21 T_BOUND */ 156 7, /* 22 T_DNA */ 157 8, /* 23 T_DOUBLEFLT */ 158 9, /* 24 T_FPOPFLT */ 159 10, /* 25 T_TSSFLT */ 160 11, /* 26 T_SEGNPFLT */ 161 12, /* 27 T_STKFLT */ 162 18, /* 28 T_MCHK */ 163 19, /* 29 T_XMMFLT */ 164 15 /* 30 T_RESERVED */ 165 }; 166 #define bsd_to_linux_trapcode(code) \ 167 ((code)<nitems(_bsd_to_linux_trapcode)? \ 168 _bsd_to_linux_trapcode[(code)]: \ 169 LINUX_T_UNKNOWN) 170 171 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 172 LINUX_VDSO_SYM_CHAR(linux_platform); 173 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base); 174 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector); 175 176 /* 177 * If FreeBSD & Linux have a difference of opinion about what a trap 178 * means, deal with it here. 179 * 180 * MPSAFE 181 */ 182 static int 183 linux_translate_traps(int signal, int trap_code) 184 { 185 186 if (signal != SIGBUS) 187 return (signal); 188 switch (trap_code) { 189 case T_PROTFLT: 190 case T_TSSFLT: 191 case T_DOUBLEFLT: 192 case T_PAGEFLT: 193 return (SIGSEGV); 194 default: 195 return (signal); 196 } 197 } 198 199 static int 200 linux_fetch_syscall_args(struct thread *td) 201 { 202 struct proc *p; 203 struct trapframe *frame; 204 struct syscall_args *sa; 205 206 p = td->td_proc; 207 frame = td->td_frame; 208 sa = &td->td_sa; 209 210 sa->args[0] = frame->tf_rdi; 211 sa->args[1] = frame->tf_rsi; 212 sa->args[2] = frame->tf_rdx; 213 sa->args[3] = frame->tf_rcx; 214 sa->args[4] = frame->tf_r8; 215 sa->args[5] = frame->tf_r9; 216 sa->code = frame->tf_rax; 217 sa->original_code = sa->code; 218 219 if (sa->code >= p->p_sysent->sv_size) 220 /* nosys */ 221 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 222 else 223 sa->callp = &p->p_sysent->sv_table[sa->code]; 224 225 td->td_retval[0] = 0; 226 return (0); 227 } 228 229 static void 230 linux_set_syscall_retval(struct thread *td, int error) 231 { 232 struct trapframe *frame; 233 234 frame = td->td_frame; 235 236 switch (error) { 237 case 0: 238 frame->tf_rax = td->td_retval[0]; 239 frame->tf_r10 = frame->tf_rcx; 240 break; 241 242 case ERESTART: 243 /* 244 * Reconstruct pc, we know that 'syscall' is 2 bytes, 245 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes. 246 * We saved this in tf_err. 247 * 248 */ 249 frame->tf_rip -= frame->tf_err; 250 frame->tf_r10 = frame->tf_rcx; 251 break; 252 253 case EJUSTRETURN: 254 break; 255 256 default: 257 frame->tf_rax = bsd_to_linux_errno(error); 258 frame->tf_r10 = frame->tf_rcx; 259 break; 260 } 261 262 /* 263 * Differently from FreeBSD native ABI, on Linux only %rcx 264 * and %r11 values are not preserved across the syscall. 265 * Require full context restore to get all registers except 266 * those two restored at return to usermode. 267 * 268 * XXX: Would be great to be able to avoid PCB_FULL_IRET 269 * for the error == 0 case. 270 */ 271 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 272 } 273 274 static void 275 linux_set_fork_retval(struct thread *td) 276 { 277 struct trapframe *frame = td->td_frame; 278 279 frame->tf_rax = 0; 280 } 281 282 static int 283 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) 284 { 285 Elf_Auxargs *args; 286 Elf_Auxinfo *argarray, *pos; 287 struct proc *p; 288 int error, issetugid; 289 290 p = imgp->proc; 291 args = (Elf64_Auxargs *)imgp->auxargs; 292 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 293 M_WAITOK | M_ZERO); 294 295 issetugid = p->p_flag & P_SUGID ? 1 : 0; 296 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base); 297 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 298 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 299 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 300 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 301 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 302 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 303 AUXARGS_ENTRY(pos, AT_BASE, args->base); 304 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 305 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 306 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 307 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 308 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 309 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 310 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 311 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); 312 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0); 313 if (imgp->execpathp != 0) 314 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); 315 if (args->execfd != -1) 316 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 317 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 318 AUXARGS_ENTRY(pos, AT_NULL, 0); 319 320 free(imgp->auxargs, M_TEMP); 321 imgp->auxargs = NULL; 322 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 323 324 error = copyout(argarray, (void *)base, 325 sizeof(*argarray) * LINUX_AT_COUNT); 326 free(argarray, M_TEMP); 327 return (error); 328 } 329 330 static int 331 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) 332 { 333 Elf_Addr *base; 334 335 base = (Elf64_Addr *)*stack_base; 336 base--; 337 if (suword(base, (uint64_t)imgp->args->argc) == -1) 338 return (EFAULT); 339 340 *stack_base = (uintptr_t)base; 341 return (0); 342 } 343 344 /* 345 * Copy strings out to the new process address space, constructing new arg 346 * and env vector tables. Return a pointer to the base so that it can be used 347 * as the initial stack pointer. 348 */ 349 static int 350 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 351 { 352 int argc, envc, error; 353 char **vectp; 354 char *stringp; 355 uintptr_t destp, ustringp; 356 struct ps_strings *arginfo; 357 char canary[LINUX_AT_RANDOM_LEN]; 358 size_t execpath_len; 359 struct proc *p; 360 361 p = imgp->proc; 362 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 363 destp = (uintptr_t)arginfo; 364 365 if (imgp->execpath != NULL && imgp->auxargs != NULL) { 366 execpath_len = strlen(imgp->execpath) + 1; 367 destp -= execpath_len; 368 destp = rounddown2(destp, sizeof(void *)); 369 imgp->execpathp = (void *)destp; 370 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 371 if (error != 0) 372 return (error); 373 } 374 375 /* Prepare the canary for SSP. */ 376 arc4rand(canary, sizeof(canary), 0); 377 destp -= roundup(sizeof(canary), sizeof(void *)); 378 imgp->canary = (void *)destp; 379 error = copyout(canary, imgp->canary, sizeof(canary)); 380 if (error != 0) 381 return (error); 382 383 /* Allocate room for the argument and environment strings. */ 384 destp -= ARG_MAX - imgp->args->stringspace; 385 destp = rounddown2(destp, sizeof(void *)); 386 ustringp = destp; 387 388 if (imgp->auxargs) { 389 /* 390 * Allocate room on the stack for the ELF auxargs 391 * array. It has LINUX_AT_COUNT entries. 392 */ 393 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo); 394 destp = rounddown2(destp, sizeof(void *)); 395 } 396 397 vectp = (char **)destp; 398 399 /* 400 * Allocate room for the argv[] and env vectors including the 401 * terminating NULL pointers. 402 */ 403 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 404 405 /* 406 * Starting with 2.24, glibc depends on a 16-byte stack alignment. 407 * One "long argc" will be prepended later. 408 */ 409 vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8); 410 411 /* vectp also becomes our initial stack base. */ 412 *stack_base = (uintptr_t)vectp; 413 414 stringp = imgp->args->begin_argv; 415 argc = imgp->args->argc; 416 envc = imgp->args->envc; 417 418 /* Copy out strings - arguments and environment. */ 419 error = copyout(stringp, (void *)ustringp, 420 ARG_MAX - imgp->args->stringspace); 421 if (error != 0) 422 return (error); 423 424 /* Fill in "ps_strings" struct for ps, w, etc. */ 425 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 426 suword(&arginfo->ps_nargvstr, argc) != 0) 427 return (EFAULT); 428 429 /* Fill in argument portion of vector table. */ 430 for (; argc > 0; --argc) { 431 if (suword(vectp++, ustringp) != 0) 432 return (EFAULT); 433 while (*stringp++ != 0) 434 ustringp++; 435 ustringp++; 436 } 437 438 /* A null vector table pointer separates the argp's from the envp's. */ 439 if (suword(vectp++, 0) != 0) 440 return (EFAULT); 441 442 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 443 suword(&arginfo->ps_nenvstr, envc) != 0) 444 return (EFAULT); 445 446 /* Fill in environment portion of vector table. */ 447 for (; envc > 0; --envc) { 448 if (suword(vectp++, ustringp) != 0) 449 return (EFAULT); 450 while (*stringp++ != 0) 451 ustringp++; 452 ustringp++; 453 } 454 455 /* The end of the vector table is a null pointer. */ 456 if (suword(vectp, 0) != 0) 457 return (EFAULT); 458 459 if (imgp->auxargs) { 460 vectp++; 461 error = imgp->sysent->sv_copyout_auxargs(imgp, 462 (uintptr_t)vectp); 463 if (error != 0) 464 return (error); 465 } 466 467 return (0); 468 } 469 470 /* 471 * Reset registers to default values on exec. 472 */ 473 static void 474 linux_exec_setregs(struct thread *td, struct image_params *imgp, 475 uintptr_t stack) 476 { 477 struct trapframe *regs; 478 struct pcb *pcb; 479 register_t saved_rflags; 480 481 regs = td->td_frame; 482 pcb = td->td_pcb; 483 484 if (td->td_proc->p_md.md_ldt != NULL) 485 user_ldt_free(td); 486 487 pcb->pcb_fsbase = 0; 488 pcb->pcb_gsbase = 0; 489 clear_pcb_flags(pcb, PCB_32BIT); 490 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 491 set_pcb_flags(pcb, PCB_FULL_IRET); 492 493 saved_rflags = regs->tf_rflags & PSL_T; 494 bzero((char *)regs, sizeof(struct trapframe)); 495 regs->tf_rip = imgp->entry_addr; 496 regs->tf_rsp = stack; 497 regs->tf_rflags = PSL_USER | saved_rflags; 498 regs->tf_ss = _udatasel; 499 regs->tf_cs = _ucodesel; 500 regs->tf_ds = _udatasel; 501 regs->tf_es = _udatasel; 502 regs->tf_fs = _ufssel; 503 regs->tf_gs = _ugssel; 504 regs->tf_flags = TF_HASSEGS; 505 506 x86_clear_dbregs(pcb); 507 508 /* 509 * Drop the FP state if we hold it, so that the process gets a 510 * clean FP state if it uses the FPU again. 511 */ 512 fpstate_drop(td); 513 } 514 515 /* 516 * Copied from amd64/amd64/machdep.c 517 * 518 * XXX fpu state need? don't think so 519 */ 520 int 521 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 522 { 523 struct proc *p; 524 struct l_ucontext uc; 525 struct l_sigcontext *context; 526 struct trapframe *regs; 527 unsigned long rflags; 528 int error; 529 ksiginfo_t ksi; 530 531 regs = td->td_frame; 532 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); 533 if (error != 0) 534 return (error); 535 536 p = td->td_proc; 537 context = &uc.uc_mcontext; 538 rflags = context->sc_rflags; 539 540 /* 541 * Don't allow users to change privileged or reserved flags. 542 */ 543 /* 544 * XXX do allow users to change the privileged flag PSL_RF. 545 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 546 * should sometimes set it there too. tf_rflags is kept in 547 * the signal context during signal handling and there is no 548 * other place to remember it, so the PSL_RF bit may be 549 * corrupted by the signal handler without us knowing. 550 * Corruption of the PSL_RF bit at worst causes one more or 551 * one less debugger trap, so allowing it is fairly harmless. 552 */ 553 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 554 uprintf("pid %d comm %s linux mangled rflags %#lx\n", 555 p->p_pid, p->p_comm, rflags); 556 return (EINVAL); 557 } 558 559 /* 560 * Don't allow users to load a valid privileged %cs. Let the 561 * hardware check for invalid selectors, excess privilege in 562 * other selectors, invalid %eip's and invalid %esp's. 563 */ 564 if (!CS_SECURE(context->sc_cs)) { 565 uprintf("pid %d comm %s linux mangled cs %#x\n", 566 p->p_pid, p->p_comm, context->sc_cs); 567 ksiginfo_init_trap(&ksi); 568 ksi.ksi_signo = SIGBUS; 569 ksi.ksi_code = BUS_OBJERR; 570 ksi.ksi_trapno = T_PROTFLT; 571 ksi.ksi_addr = (void *)regs->tf_rip; 572 trapsignal(td, &ksi); 573 return (EINVAL); 574 } 575 576 PROC_LOCK(p); 577 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); 578 SIG_CANTMASK(td->td_sigmask); 579 signotify(td); 580 PROC_UNLOCK(p); 581 582 regs->tf_rdi = context->sc_rdi; 583 regs->tf_rsi = context->sc_rsi; 584 regs->tf_rdx = context->sc_rdx; 585 regs->tf_rbp = context->sc_rbp; 586 regs->tf_rbx = context->sc_rbx; 587 regs->tf_rcx = context->sc_rcx; 588 regs->tf_rax = context->sc_rax; 589 regs->tf_rip = context->sc_rip; 590 regs->tf_rsp = context->sc_rsp; 591 regs->tf_r8 = context->sc_r8; 592 regs->tf_r9 = context->sc_r9; 593 regs->tf_r10 = context->sc_r10; 594 regs->tf_r11 = context->sc_r11; 595 regs->tf_r12 = context->sc_r12; 596 regs->tf_r13 = context->sc_r13; 597 regs->tf_r14 = context->sc_r14; 598 regs->tf_r15 = context->sc_r15; 599 regs->tf_cs = context->sc_cs; 600 regs->tf_err = context->sc_err; 601 regs->tf_rflags = rflags; 602 603 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 604 return (EJUSTRETURN); 605 } 606 607 /* 608 * copied from amd64/amd64/machdep.c 609 * 610 * Send an interrupt to process. 611 */ 612 static void 613 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 614 { 615 struct l_rt_sigframe sf, *sfp; 616 struct proc *p; 617 struct thread *td; 618 struct sigacts *psp; 619 caddr_t sp; 620 struct trapframe *regs; 621 int sig, code; 622 int oonstack; 623 624 td = curthread; 625 p = td->td_proc; 626 PROC_LOCK_ASSERT(p, MA_OWNED); 627 sig = ksi->ksi_signo; 628 psp = p->p_sigacts; 629 code = ksi->ksi_code; 630 mtx_assert(&psp->ps_mtx, MA_OWNED); 631 regs = td->td_frame; 632 oonstack = sigonstack(regs->tf_rsp); 633 634 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 635 catcher, sig, mask, code); 636 637 /* Save user context. */ 638 bzero(&sf, sizeof(sf)); 639 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask); 640 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask); 641 642 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 643 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 644 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 645 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 646 647 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi; 648 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi; 649 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx; 650 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp; 651 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx; 652 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx; 653 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax; 654 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip; 655 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp; 656 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8; 657 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9; 658 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10; 659 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11; 660 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12; 661 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13; 662 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14; 663 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15; 664 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 665 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags; 666 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err; 667 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 668 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 669 670 /* Allocate space for the signal handler context. */ 671 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 672 SIGISMEMBER(psp->ps_sigonstack, sig)) { 673 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - 674 sizeof(struct l_rt_sigframe); 675 } else 676 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128; 677 /* Align to 16 bytes. */ 678 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 679 680 /* Translate the signal. */ 681 sig = bsd_to_linux_signal(sig); 682 683 /* Build the argument list for the signal handler. */ 684 regs->tf_rdi = sig; /* arg 1 in %rdi */ 685 regs->tf_rax = 0; 686 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 687 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */ 688 689 /* Fill in POSIX parts. */ 690 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig); 691 sf.sf_handler = catcher; 692 693 mtx_unlock(&psp->ps_mtx); 694 PROC_UNLOCK(p); 695 696 /* Copy the sigframe out to the user's stack. */ 697 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 698 PROC_LOCK(p); 699 sigexit(td, SIGILL); 700 } 701 702 regs->tf_rsp = (long)sfp; 703 regs->tf_rip = linux_rt_sigcode; 704 regs->tf_rflags &= ~(PSL_T | PSL_D); 705 regs->tf_cs = _ucodesel; 706 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 707 PROC_LOCK(p); 708 mtx_lock(&psp->ps_mtx); 709 } 710 711 #define LINUX_VSYSCALL_START (-10UL << 20) 712 #define LINUX_VSYSCALL_SZ 1024 713 714 const unsigned long linux_vsyscall_vector[] = { 715 LINUX_SYS_gettimeofday, 716 LINUX_SYS_linux_time, 717 LINUX_SYS_linux_getcpu, 718 }; 719 720 static int 721 linux_vsyscall(struct thread *td) 722 { 723 struct trapframe *frame; 724 uint64_t retqaddr; 725 int code, traced; 726 int error; 727 728 frame = td->td_frame; 729 730 /* Check %rip for vsyscall area. */ 731 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 732 return (EINVAL); 733 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 734 return (EINVAL); 735 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 736 if (code >= nitems(linux_vsyscall_vector)) 737 return (EINVAL); 738 739 /* 740 * vsyscall called as callq *(%rax), so we must 741 * use return address from %rsp and also fixup %rsp. 742 */ 743 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 744 if (error) 745 return (error); 746 747 frame->tf_rip = retqaddr; 748 frame->tf_rax = linux_vsyscall_vector[code]; 749 frame->tf_rsp += 8; 750 751 traced = (frame->tf_flags & PSL_T); 752 753 amd64_syscall(td, traced); 754 755 return (0); 756 } 757 758 struct sysentvec elf_linux_sysvec = { 759 .sv_size = LINUX_SYS_MAXSYSCALL, 760 .sv_table = linux_sysent, 761 .sv_transtrap = linux_translate_traps, 762 .sv_fixup = linux_fixup_elf, 763 .sv_sendsig = linux_rt_sendsig, 764 .sv_sigcode = &_binary_linux_vdso_so_o_start, 765 .sv_szsigcode = &linux_szsigcode, 766 .sv_name = "Linux ELF64", 767 .sv_coredump = elf64_coredump, 768 .sv_elf_core_osabi = ELFOSABI_NONE, 769 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR, 770 .sv_elf_core_prepare_notes = linux64_prepare_notes, 771 .sv_imgact_try = linux_exec_imgact_try, 772 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 773 .sv_minuser = VM_MIN_ADDRESS, 774 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48, 775 .sv_usrstack = LINUX_USRSTACK_LA48, 776 .sv_psstrings = LINUX_PS_STRINGS_LA48, 777 .sv_stackprot = VM_PROT_ALL, 778 .sv_copyout_auxargs = linux_copyout_auxargs, 779 .sv_copyout_strings = linux_copyout_strings, 780 .sv_setregs = linux_exec_setregs, 781 .sv_fixlimit = NULL, 782 .sv_maxssiz = NULL, 783 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN | 784 SV_SIG_WAITNDQ | SV_TIMEKEEP, 785 .sv_set_syscall_retval = linux_set_syscall_retval, 786 .sv_fetch_syscall_args = linux_fetch_syscall_args, 787 .sv_syscallnames = NULL, 788 .sv_shared_page_base = LINUX_SHAREDPAGE_LA48, 789 .sv_shared_page_len = PAGE_SIZE, 790 .sv_schedtail = linux_schedtail, 791 .sv_thread_detach = linux_thread_detach, 792 .sv_trap = linux_vsyscall, 793 .sv_onexec = linux_on_exec_vmspace, 794 .sv_onexit = linux_on_exit, 795 .sv_ontdexit = linux_thread_dtor, 796 .sv_setid_allowed = &linux_setid_allowed_query, 797 .sv_set_fork_retval = linux_set_fork_retval, 798 }; 799 800 static int 801 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp) 802 { 803 int error; 804 805 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base, 806 LINUX_VDSOPAGE_SIZE, imgp); 807 if (error == 0) 808 linux_on_exec(p, imgp); 809 return (error); 810 } 811 812 /* 813 * linux_vdso_install() and linux_exec_sysvec_init() must be called 814 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY). 815 */ 816 static void 817 linux_exec_sysvec_init(void *param) 818 { 819 l_uintptr_t *ktimekeep_base, *ktsc_selector; 820 struct sysentvec *sv; 821 ptrdiff_t tkoff; 822 823 sv = param; 824 amd64_lower_shared_page(sv); 825 /* Fill timekeep_base */ 826 exec_sysvec_init(sv); 827 828 tkoff = kern_timekeep_base - linux_vdso_base; 829 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 830 *ktimekeep_base = sv->sv_timekeep_base; 831 832 tkoff = kern_tsc_selector - linux_vdso_base; 833 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 834 *ktsc_selector = linux_vdso_tsc_selector_idx(); 835 if (bootverbose) 836 printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector); 837 } 838 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY, 839 linux_exec_sysvec_init, &elf_linux_sysvec); 840 841 static void 842 linux_vdso_install(const void *param) 843 { 844 char *vdso_start = &_binary_linux_vdso_so_o_start; 845 char *vdso_end = &_binary_linux_vdso_so_o_end; 846 847 linux_szsigcode = vdso_end - vdso_start; 848 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE); 849 850 linux_vdso_base = LINUX_VDSOPAGE_LA48; 851 if (hw_lower_amd64_sharedpage != 0) 852 linux_vdso_base -= PAGE_SIZE; 853 854 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base); 855 856 linux_vdso_obj = __elfN(linux_shared_page_init) 857 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 858 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode); 859 860 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base); 861 } 862 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST, 863 linux_vdso_install, NULL); 864 865 static void 866 linux_vdso_deinstall(const void *param) 867 { 868 869 __elfN(linux_shared_page_fini)(linux_vdso_obj, 870 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 871 } 872 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 873 linux_vdso_deinstall, NULL); 874 875 static void 876 linux_vdso_reloc(char *mapping, Elf_Addr offset) 877 { 878 const Elf_Ehdr *ehdr; 879 const Elf_Shdr *shdr; 880 Elf64_Addr *where, val; 881 Elf_Size rtype, symidx; 882 const Elf_Rela *rela; 883 Elf_Addr addr, addend; 884 int relacnt; 885 int i, j; 886 887 MPASS(offset != 0); 888 889 relacnt = 0; 890 ehdr = (const Elf_Ehdr *)mapping; 891 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff); 892 for (i = 0; i < ehdr->e_shnum; i++) 893 { 894 switch (shdr[i].sh_type) { 895 case SHT_REL: 896 printf("Linux x86_64 vDSO: unexpected Rel section\n"); 897 break; 898 case SHT_RELA: 899 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset); 900 relacnt = shdr[i].sh_size / sizeof(*rela); 901 } 902 } 903 904 for (j = 0; j < relacnt; j++, rela++) { 905 where = (Elf_Addr *)(mapping + rela->r_offset); 906 addend = rela->r_addend; 907 rtype = ELF_R_TYPE(rela->r_info); 908 symidx = ELF_R_SYM(rela->r_info); 909 910 switch (rtype) { 911 case R_X86_64_NONE: /* none */ 912 break; 913 914 case R_X86_64_RELATIVE: /* B + A */ 915 addr = (Elf_Addr)(offset + addend); 916 val = addr; 917 if (*where != val) 918 *where = val; 919 break; 920 case R_X86_64_IRELATIVE: 921 printf("Linux x86_64 vDSO: unexpected ifunc relocation, " 922 "symbol index %ld\n", symidx); 923 break; 924 default: 925 printf("Linux x86_64 vDSO: unexpected relocation type %ld, " 926 "symbol index %ld\n", rtype, symidx); 927 } 928 } 929 } 930 931 static char GNULINUX_ABI_VENDOR[] = "GNU"; 932 static int GNULINUX_ABI_DESC = 0; 933 934 static bool 935 linux_trans_osrel(const Elf_Note *note, int32_t *osrel) 936 { 937 const Elf32_Word *desc; 938 uintptr_t p; 939 940 p = (uintptr_t)(note + 1); 941 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 942 943 desc = (const Elf32_Word *)p; 944 if (desc[0] != GNULINUX_ABI_DESC) 945 return (false); 946 947 /* 948 * For Linux we encode osrel using the Linux convention of 949 * (version << 16) | (major << 8) | (minor) 950 * See macro in linux_mib.h 951 */ 952 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 953 954 return (true); 955 } 956 957 static Elf_Brandnote linux64_brandnote = { 958 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), 959 .hdr.n_descsz = 16, 960 .hdr.n_type = 1, 961 .vendor = GNULINUX_ABI_VENDOR, 962 .flags = BN_TRANSLATE_OSREL, 963 .trans_osrel = linux_trans_osrel 964 }; 965 966 static Elf64_Brandinfo linux_glibc2brand = { 967 .brand = ELFOSABI_LINUX, 968 .machine = EM_X86_64, 969 .compat_3_brand = "Linux", 970 .emul_path = linux_emul_path, 971 .interp_path = "/lib64/ld-linux-x86-64.so.2", 972 .sysvec = &elf_linux_sysvec, 973 .interp_newpath = NULL, 974 .brand_note = &linux64_brandnote, 975 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 976 }; 977 978 static Elf64_Brandinfo linux_glibc2brandshort = { 979 .brand = ELFOSABI_LINUX, 980 .machine = EM_X86_64, 981 .compat_3_brand = "Linux", 982 .emul_path = linux_emul_path, 983 .interp_path = "/lib64/ld-linux.so.2", 984 .sysvec = &elf_linux_sysvec, 985 .interp_newpath = NULL, 986 .brand_note = &linux64_brandnote, 987 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 988 }; 989 990 static Elf64_Brandinfo linux_muslbrand = { 991 .brand = ELFOSABI_LINUX, 992 .machine = EM_X86_64, 993 .compat_3_brand = "Linux", 994 .emul_path = linux_emul_path, 995 .interp_path = "/lib/ld-musl-x86_64.so.1", 996 .sysvec = &elf_linux_sysvec, 997 .interp_newpath = NULL, 998 .brand_note = &linux64_brandnote, 999 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE | 1000 LINUX_BI_FUTEX_REQUEUE 1001 }; 1002 1003 Elf64_Brandinfo *linux_brandlist[] = { 1004 &linux_glibc2brand, 1005 &linux_glibc2brandshort, 1006 &linux_muslbrand, 1007 NULL 1008 }; 1009 1010 static int 1011 linux64_elf_modevent(module_t mod, int type, void *data) 1012 { 1013 Elf64_Brandinfo **brandinfo; 1014 int error; 1015 struct linux_ioctl_handler **lihp; 1016 1017 error = 0; 1018 1019 switch(type) { 1020 case MOD_LOAD: 1021 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1022 ++brandinfo) 1023 if (elf64_insert_brand_entry(*brandinfo) < 0) 1024 error = EINVAL; 1025 if (error == 0) { 1026 SET_FOREACH(lihp, linux_ioctl_handler_set) 1027 linux_ioctl_register_handler(*lihp); 1028 stclohz = (stathz ? stathz : hz); 1029 if (bootverbose) 1030 printf("Linux x86-64 ELF exec handler installed\n"); 1031 } else 1032 printf("cannot insert Linux x86-64 ELF brand handler\n"); 1033 break; 1034 case MOD_UNLOAD: 1035 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1036 ++brandinfo) 1037 if (elf64_brand_inuse(*brandinfo)) 1038 error = EBUSY; 1039 if (error == 0) { 1040 for (brandinfo = &linux_brandlist[0]; 1041 *brandinfo != NULL; ++brandinfo) 1042 if (elf64_remove_brand_entry(*brandinfo) < 0) 1043 error = EINVAL; 1044 } 1045 if (error == 0) { 1046 SET_FOREACH(lihp, linux_ioctl_handler_set) 1047 linux_ioctl_unregister_handler(*lihp); 1048 if (bootverbose) 1049 printf("Linux x86_64 ELF exec handler removed\n"); 1050 } else 1051 printf("Could not deinstall Linux x86_64 ELF interpreter entry\n"); 1052 break; 1053 default: 1054 return (EOPNOTSUPP); 1055 } 1056 return (error); 1057 } 1058 1059 static moduledata_t linux64_elf_mod = { 1060 "linux64elf", 1061 linux64_elf_modevent, 1062 0 1063 }; 1064 1065 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 1066 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 1067 FEATURE(linux64, "Linux 64bit support"); 1068