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 #include <x86/linux/linux_x86_sigframe.h> 90 91 MODULE_VERSION(linux64, 1); 92 93 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2 94 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \ 95 LINUX_VDSOPAGE_SIZE) 96 #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE) 97 /* 98 * PAGE_SIZE - the size 99 * of the native SHAREDPAGE 100 */ 101 #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48 102 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \ 103 sizeof(struct ps_strings)) 104 105 static int linux_szsigcode; 106 static vm_object_t linux_vdso_obj; 107 static char *linux_vdso_mapping; 108 extern char _binary_linux_vdso_so_o_start; 109 extern char _binary_linux_vdso_so_o_end; 110 static vm_offset_t linux_vdso_base; 111 112 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 113 114 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 115 116 static int linux_copyout_strings(struct image_params *imgp, 117 uintptr_t *stack_base); 118 static int linux_fixup_elf(uintptr_t *stack_base, 119 struct image_params *iparams); 120 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 121 static void linux_vdso_install(const void *param); 122 static void linux_vdso_deinstall(const void *param); 123 static void linux_vdso_reloc(char *mapping, Elf_Addr offset); 124 static void linux_set_syscall_retval(struct thread *td, int error); 125 static int linux_fetch_syscall_args(struct thread *td); 126 static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 127 uintptr_t stack); 128 static void linux_exec_sysvec_init(void *param); 129 static int linux_on_exec_vmspace(struct proc *p, 130 struct image_params *imgp); 131 static void linux_set_fork_retval(struct thread *td); 132 static int linux_vsyscall(struct thread *td); 133 134 #define LINUX_T_UNKNOWN 255 135 static int _bsd_to_linux_trapcode[] = { 136 LINUX_T_UNKNOWN, /* 0 */ 137 6, /* 1 T_PRIVINFLT */ 138 LINUX_T_UNKNOWN, /* 2 */ 139 3, /* 3 T_BPTFLT */ 140 LINUX_T_UNKNOWN, /* 4 */ 141 LINUX_T_UNKNOWN, /* 5 */ 142 16, /* 6 T_ARITHTRAP */ 143 254, /* 7 T_ASTFLT */ 144 LINUX_T_UNKNOWN, /* 8 */ 145 13, /* 9 T_PROTFLT */ 146 1, /* 10 T_TRCTRAP */ 147 LINUX_T_UNKNOWN, /* 11 */ 148 14, /* 12 T_PAGEFLT */ 149 LINUX_T_UNKNOWN, /* 13 */ 150 17, /* 14 T_ALIGNFLT */ 151 LINUX_T_UNKNOWN, /* 15 */ 152 LINUX_T_UNKNOWN, /* 16 */ 153 LINUX_T_UNKNOWN, /* 17 */ 154 0, /* 18 T_DIVIDE */ 155 2, /* 19 T_NMI */ 156 4, /* 20 T_OFLOW */ 157 5, /* 21 T_BOUND */ 158 7, /* 22 T_DNA */ 159 8, /* 23 T_DOUBLEFLT */ 160 9, /* 24 T_FPOPFLT */ 161 10, /* 25 T_TSSFLT */ 162 11, /* 26 T_SEGNPFLT */ 163 12, /* 27 T_STKFLT */ 164 18, /* 28 T_MCHK */ 165 19, /* 29 T_XMMFLT */ 166 15 /* 30 T_RESERVED */ 167 }; 168 #define bsd_to_linux_trapcode(code) \ 169 ((code)<nitems(_bsd_to_linux_trapcode)? \ 170 _bsd_to_linux_trapcode[(code)]: \ 171 LINUX_T_UNKNOWN) 172 173 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 174 LINUX_VDSO_SYM_CHAR(linux_platform); 175 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base); 176 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector); 177 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector); 178 179 /* 180 * If FreeBSD & Linux have a difference of opinion about what a trap 181 * means, deal with it here. 182 * 183 * MPSAFE 184 */ 185 static int 186 linux_translate_traps(int signal, int trap_code) 187 { 188 189 if (signal != SIGBUS) 190 return (signal); 191 switch (trap_code) { 192 case T_PROTFLT: 193 case T_TSSFLT: 194 case T_DOUBLEFLT: 195 case T_PAGEFLT: 196 return (SIGSEGV); 197 default: 198 return (signal); 199 } 200 } 201 202 static int 203 linux_fetch_syscall_args(struct thread *td) 204 { 205 struct proc *p; 206 struct trapframe *frame; 207 struct syscall_args *sa; 208 209 p = td->td_proc; 210 frame = td->td_frame; 211 sa = &td->td_sa; 212 213 sa->args[0] = frame->tf_rdi; 214 sa->args[1] = frame->tf_rsi; 215 sa->args[2] = frame->tf_rdx; 216 sa->args[3] = frame->tf_rcx; 217 sa->args[4] = frame->tf_r8; 218 sa->args[5] = frame->tf_r9; 219 sa->code = frame->tf_rax; 220 sa->original_code = sa->code; 221 222 if (sa->code >= p->p_sysent->sv_size) 223 /* nosys */ 224 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 225 else 226 sa->callp = &p->p_sysent->sv_table[sa->code]; 227 228 td->td_retval[0] = 0; 229 return (0); 230 } 231 232 static void 233 linux_set_syscall_retval(struct thread *td, int error) 234 { 235 struct trapframe *frame; 236 237 frame = td->td_frame; 238 239 switch (error) { 240 case 0: 241 frame->tf_rax = td->td_retval[0]; 242 frame->tf_r10 = frame->tf_rcx; 243 break; 244 245 case ERESTART: 246 /* 247 * Reconstruct pc, we know that 'syscall' is 2 bytes, 248 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes. 249 * We saved this in tf_err. 250 * 251 */ 252 frame->tf_rip -= frame->tf_err; 253 frame->tf_r10 = frame->tf_rcx; 254 break; 255 256 case EJUSTRETURN: 257 break; 258 259 default: 260 frame->tf_rax = bsd_to_linux_errno(error); 261 frame->tf_r10 = frame->tf_rcx; 262 break; 263 } 264 265 /* 266 * Differently from FreeBSD native ABI, on Linux only %rcx 267 * and %r11 values are not preserved across the syscall. 268 * Require full context restore to get all registers except 269 * those two restored at return to usermode. 270 * 271 * XXX: Would be great to be able to avoid PCB_FULL_IRET 272 * for the error == 0 case. 273 */ 274 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 275 } 276 277 static void 278 linux_set_fork_retval(struct thread *td) 279 { 280 struct trapframe *frame = td->td_frame; 281 282 frame->tf_rax = 0; 283 } 284 285 static int 286 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) 287 { 288 Elf_Auxargs *args; 289 Elf_Auxinfo *argarray, *pos; 290 struct proc *p; 291 int error, issetugid; 292 293 p = imgp->proc; 294 args = (Elf64_Auxargs *)imgp->auxargs; 295 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 296 M_WAITOK | M_ZERO); 297 298 issetugid = p->p_flag & P_SUGID ? 1 : 0; 299 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base); 300 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 301 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 302 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 303 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 304 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 305 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 306 AUXARGS_ENTRY(pos, AT_BASE, args->base); 307 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 308 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 309 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 310 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 311 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 312 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 313 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 314 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); 315 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0); 316 if (imgp->execpathp != 0) 317 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); 318 if (args->execfd != -1) 319 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 320 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 321 AUXARGS_ENTRY(pos, AT_NULL, 0); 322 323 free(imgp->auxargs, M_TEMP); 324 imgp->auxargs = NULL; 325 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 326 327 error = copyout(argarray, (void *)base, 328 sizeof(*argarray) * LINUX_AT_COUNT); 329 free(argarray, M_TEMP); 330 return (error); 331 } 332 333 static int 334 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) 335 { 336 Elf_Addr *base; 337 338 base = (Elf64_Addr *)*stack_base; 339 base--; 340 if (suword(base, (uint64_t)imgp->args->argc) == -1) 341 return (EFAULT); 342 343 *stack_base = (uintptr_t)base; 344 return (0); 345 } 346 347 /* 348 * Copy strings out to the new process address space, constructing new arg 349 * and env vector tables. Return a pointer to the base so that it can be used 350 * as the initial stack pointer. 351 */ 352 static int 353 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 354 { 355 int argc, envc, error; 356 char **vectp; 357 char *stringp; 358 uintptr_t destp, ustringp; 359 struct ps_strings *arginfo; 360 char canary[LINUX_AT_RANDOM_LEN]; 361 size_t execpath_len; 362 struct proc *p; 363 364 p = imgp->proc; 365 arginfo = (struct ps_strings *)PROC_PS_STRINGS(p); 366 destp = (uintptr_t)arginfo; 367 368 if (imgp->execpath != NULL && imgp->auxargs != NULL) { 369 execpath_len = strlen(imgp->execpath) + 1; 370 destp -= execpath_len; 371 destp = rounddown2(destp, sizeof(void *)); 372 imgp->execpathp = (void *)destp; 373 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 374 if (error != 0) 375 return (error); 376 } 377 378 /* Prepare the canary for SSP. */ 379 arc4rand(canary, sizeof(canary), 0); 380 destp -= roundup(sizeof(canary), sizeof(void *)); 381 imgp->canary = (void *)destp; 382 error = copyout(canary, imgp->canary, sizeof(canary)); 383 if (error != 0) 384 return (error); 385 386 /* Allocate room for the argument and environment strings. */ 387 destp -= ARG_MAX - imgp->args->stringspace; 388 destp = rounddown2(destp, sizeof(void *)); 389 ustringp = destp; 390 391 if (imgp->auxargs) { 392 /* 393 * Allocate room on the stack for the ELF auxargs 394 * array. It has LINUX_AT_COUNT entries. 395 */ 396 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo); 397 destp = rounddown2(destp, sizeof(void *)); 398 } 399 400 vectp = (char **)destp; 401 402 /* 403 * Allocate room for the argv[] and env vectors including the 404 * terminating NULL pointers. 405 */ 406 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 407 408 /* 409 * Starting with 2.24, glibc depends on a 16-byte stack alignment. 410 * One "long argc" will be prepended later. 411 */ 412 vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8); 413 414 /* vectp also becomes our initial stack base. */ 415 *stack_base = (uintptr_t)vectp; 416 417 stringp = imgp->args->begin_argv; 418 argc = imgp->args->argc; 419 envc = imgp->args->envc; 420 421 /* Copy out strings - arguments and environment. */ 422 error = copyout(stringp, (void *)ustringp, 423 ARG_MAX - imgp->args->stringspace); 424 if (error != 0) 425 return (error); 426 427 /* Fill in "ps_strings" struct for ps, w, etc. */ 428 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 429 suword(&arginfo->ps_nargvstr, argc) != 0) 430 return (EFAULT); 431 432 /* Fill in argument portion of vector table. */ 433 for (; argc > 0; --argc) { 434 if (suword(vectp++, ustringp) != 0) 435 return (EFAULT); 436 while (*stringp++ != 0) 437 ustringp++; 438 ustringp++; 439 } 440 441 /* A null vector table pointer separates the argp's from the envp's. */ 442 if (suword(vectp++, 0) != 0) 443 return (EFAULT); 444 445 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 446 suword(&arginfo->ps_nenvstr, envc) != 0) 447 return (EFAULT); 448 449 /* Fill in environment portion of vector table. */ 450 for (; envc > 0; --envc) { 451 if (suword(vectp++, ustringp) != 0) 452 return (EFAULT); 453 while (*stringp++ != 0) 454 ustringp++; 455 ustringp++; 456 } 457 458 /* The end of the vector table is a null pointer. */ 459 if (suword(vectp, 0) != 0) 460 return (EFAULT); 461 462 if (imgp->auxargs) { 463 vectp++; 464 error = imgp->sysent->sv_copyout_auxargs(imgp, 465 (uintptr_t)vectp); 466 if (error != 0) 467 return (error); 468 } 469 470 return (0); 471 } 472 473 /* 474 * Reset registers to default values on exec. 475 */ 476 static void 477 linux_exec_setregs(struct thread *td, struct image_params *imgp, 478 uintptr_t stack) 479 { 480 struct trapframe *regs; 481 struct pcb *pcb; 482 register_t saved_rflags; 483 484 regs = td->td_frame; 485 pcb = td->td_pcb; 486 487 if (td->td_proc->p_md.md_ldt != NULL) 488 user_ldt_free(td); 489 490 pcb->pcb_fsbase = 0; 491 pcb->pcb_gsbase = 0; 492 clear_pcb_flags(pcb, PCB_32BIT); 493 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 494 set_pcb_flags(pcb, PCB_FULL_IRET); 495 496 saved_rflags = regs->tf_rflags & PSL_T; 497 bzero((char *)regs, sizeof(struct trapframe)); 498 regs->tf_rip = imgp->entry_addr; 499 regs->tf_rsp = stack; 500 regs->tf_rflags = PSL_USER | saved_rflags; 501 regs->tf_ss = _udatasel; 502 regs->tf_cs = _ucodesel; 503 regs->tf_ds = _udatasel; 504 regs->tf_es = _udatasel; 505 regs->tf_fs = _ufssel; 506 regs->tf_gs = _ugssel; 507 regs->tf_flags = TF_HASSEGS; 508 509 x86_clear_dbregs(pcb); 510 511 /* 512 * Drop the FP state if we hold it, so that the process gets a 513 * clean FP state if it uses the FPU again. 514 */ 515 fpstate_drop(td); 516 } 517 518 /* 519 * Copied from amd64/amd64/machdep.c 520 * 521 * XXX fpu state need? don't think so 522 */ 523 int 524 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 525 { 526 struct proc *p; 527 struct l_ucontext uc; 528 struct l_sigcontext *context; 529 struct trapframe *regs; 530 unsigned long rflags; 531 int error; 532 ksiginfo_t ksi; 533 534 regs = td->td_frame; 535 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); 536 if (error != 0) 537 return (error); 538 539 p = td->td_proc; 540 context = &uc.uc_mcontext; 541 rflags = context->sc_rflags; 542 543 /* 544 * Don't allow users to change privileged or reserved flags. 545 */ 546 /* 547 * XXX do allow users to change the privileged flag PSL_RF. 548 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 549 * should sometimes set it there too. tf_rflags is kept in 550 * the signal context during signal handling and there is no 551 * other place to remember it, so the PSL_RF bit may be 552 * corrupted by the signal handler without us knowing. 553 * Corruption of the PSL_RF bit at worst causes one more or 554 * one less debugger trap, so allowing it is fairly harmless. 555 */ 556 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 557 uprintf("pid %d comm %s linux mangled rflags %#lx\n", 558 p->p_pid, p->p_comm, rflags); 559 return (EINVAL); 560 } 561 562 /* 563 * Don't allow users to load a valid privileged %cs. Let the 564 * hardware check for invalid selectors, excess privilege in 565 * other selectors, invalid %eip's and invalid %esp's. 566 */ 567 if (!CS_SECURE(context->sc_cs)) { 568 uprintf("pid %d comm %s linux mangled cs %#x\n", 569 p->p_pid, p->p_comm, context->sc_cs); 570 ksiginfo_init_trap(&ksi); 571 ksi.ksi_signo = SIGBUS; 572 ksi.ksi_code = BUS_OBJERR; 573 ksi.ksi_trapno = T_PROTFLT; 574 ksi.ksi_addr = (void *)regs->tf_rip; 575 trapsignal(td, &ksi); 576 return (EINVAL); 577 } 578 579 PROC_LOCK(p); 580 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); 581 SIG_CANTMASK(td->td_sigmask); 582 signotify(td); 583 PROC_UNLOCK(p); 584 585 regs->tf_rdi = context->sc_rdi; 586 regs->tf_rsi = context->sc_rsi; 587 regs->tf_rdx = context->sc_rdx; 588 regs->tf_rbp = context->sc_rbp; 589 regs->tf_rbx = context->sc_rbx; 590 regs->tf_rcx = context->sc_rcx; 591 regs->tf_rax = context->sc_rax; 592 regs->tf_rip = context->sc_rip; 593 regs->tf_rsp = context->sc_rsp; 594 regs->tf_r8 = context->sc_r8; 595 regs->tf_r9 = context->sc_r9; 596 regs->tf_r10 = context->sc_r10; 597 regs->tf_r11 = context->sc_r11; 598 regs->tf_r12 = context->sc_r12; 599 regs->tf_r13 = context->sc_r13; 600 regs->tf_r14 = context->sc_r14; 601 regs->tf_r15 = context->sc_r15; 602 regs->tf_cs = context->sc_cs; 603 regs->tf_err = context->sc_err; 604 regs->tf_rflags = rflags; 605 606 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 607 return (EJUSTRETURN); 608 } 609 610 /* 611 * copied from amd64/amd64/machdep.c 612 * 613 * Send an interrupt to process. 614 */ 615 static void 616 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 617 { 618 struct l_rt_sigframe sf, *sfp; 619 struct proc *p; 620 struct thread *td; 621 struct sigacts *psp; 622 caddr_t sp; 623 struct trapframe *regs; 624 int sig, code; 625 int oonstack; 626 627 td = curthread; 628 p = td->td_proc; 629 PROC_LOCK_ASSERT(p, MA_OWNED); 630 sig = ksi->ksi_signo; 631 psp = p->p_sigacts; 632 code = ksi->ksi_code; 633 mtx_assert(&psp->ps_mtx, MA_OWNED); 634 regs = td->td_frame; 635 oonstack = sigonstack(regs->tf_rsp); 636 637 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 638 catcher, sig, mask, code); 639 640 /* Save user context. */ 641 bzero(&sf, sizeof(sf)); 642 bsd_to_linux_sigset(mask, &sf.sf_uc.uc_sigmask); 643 bsd_to_linux_sigset(mask, &sf.sf_uc.uc_mcontext.sc_mask); 644 645 sf.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 646 sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size; 647 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 648 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 649 650 sf.sf_uc.uc_mcontext.sc_rdi = regs->tf_rdi; 651 sf.sf_uc.uc_mcontext.sc_rsi = regs->tf_rsi; 652 sf.sf_uc.uc_mcontext.sc_rdx = regs->tf_rdx; 653 sf.sf_uc.uc_mcontext.sc_rbp = regs->tf_rbp; 654 sf.sf_uc.uc_mcontext.sc_rbx = regs->tf_rbx; 655 sf.sf_uc.uc_mcontext.sc_rcx = regs->tf_rcx; 656 sf.sf_uc.uc_mcontext.sc_rax = regs->tf_rax; 657 sf.sf_uc.uc_mcontext.sc_rip = regs->tf_rip; 658 sf.sf_uc.uc_mcontext.sc_rsp = regs->tf_rsp; 659 sf.sf_uc.uc_mcontext.sc_r8 = regs->tf_r8; 660 sf.sf_uc.uc_mcontext.sc_r9 = regs->tf_r9; 661 sf.sf_uc.uc_mcontext.sc_r10 = regs->tf_r10; 662 sf.sf_uc.uc_mcontext.sc_r11 = regs->tf_r11; 663 sf.sf_uc.uc_mcontext.sc_r12 = regs->tf_r12; 664 sf.sf_uc.uc_mcontext.sc_r13 = regs->tf_r13; 665 sf.sf_uc.uc_mcontext.sc_r14 = regs->tf_r14; 666 sf.sf_uc.uc_mcontext.sc_r15 = regs->tf_r15; 667 sf.sf_uc.uc_mcontext.sc_cs = regs->tf_cs; 668 sf.sf_uc.uc_mcontext.sc_rflags = regs->tf_rflags; 669 sf.sf_uc.uc_mcontext.sc_err = regs->tf_err; 670 sf.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 671 sf.sf_uc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 672 673 /* Allocate space for the signal handler context. */ 674 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 675 SIGISMEMBER(psp->ps_sigonstack, sig)) { 676 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size; 677 } else 678 sp = (caddr_t)regs->tf_rsp - 128; 679 sp -= sizeof(struct l_rt_sigframe); 680 /* Align to 16 bytes. */ 681 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 682 683 /* Translate the signal. */ 684 sig = bsd_to_linux_signal(sig); 685 686 /* Build the argument list for the signal handler. */ 687 regs->tf_rdi = sig; /* arg 1 in %rdi */ 688 regs->tf_rax = 0; 689 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 690 regs->tf_rdx = (register_t)&sfp->sf_uc; /* arg 3 in %rdx */ 691 regs->tf_rcx = (register_t)catcher; 692 693 /* Fill in POSIX parts. */ 694 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig); 695 696 mtx_unlock(&psp->ps_mtx); 697 PROC_UNLOCK(p); 698 699 /* Copy the sigframe out to the user's stack. */ 700 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 701 uprintf("pid %d comm %s has trashed its stack, killing\n", 702 p->p_pid, p->p_comm); 703 PROC_LOCK(p); 704 sigexit(td, SIGILL); 705 } 706 707 regs->tf_rsp = (long)sfp; 708 regs->tf_rip = linux_rt_sigcode; 709 regs->tf_rflags &= ~(PSL_T | PSL_D); 710 regs->tf_cs = _ucodesel; 711 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 712 PROC_LOCK(p); 713 mtx_lock(&psp->ps_mtx); 714 } 715 716 #define LINUX_VSYSCALL_START (-10UL << 20) 717 #define LINUX_VSYSCALL_SZ 1024 718 719 const unsigned long linux_vsyscall_vector[] = { 720 LINUX_SYS_gettimeofday, 721 LINUX_SYS_linux_time, 722 LINUX_SYS_linux_getcpu, 723 }; 724 725 static int 726 linux_vsyscall(struct thread *td) 727 { 728 struct trapframe *frame; 729 uint64_t retqaddr; 730 int code, traced; 731 int error; 732 733 frame = td->td_frame; 734 735 /* Check %rip for vsyscall area. */ 736 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 737 return (EINVAL); 738 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 739 return (EINVAL); 740 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 741 if (code >= nitems(linux_vsyscall_vector)) 742 return (EINVAL); 743 744 /* 745 * vsyscall called as callq *(%rax), so we must 746 * use return address from %rsp and also fixup %rsp. 747 */ 748 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 749 if (error) 750 return (error); 751 752 frame->tf_rip = retqaddr; 753 frame->tf_rax = linux_vsyscall_vector[code]; 754 frame->tf_rsp += 8; 755 756 traced = (frame->tf_flags & PSL_T); 757 758 amd64_syscall(td, traced); 759 760 return (0); 761 } 762 763 struct sysentvec elf_linux_sysvec = { 764 .sv_size = LINUX_SYS_MAXSYSCALL, 765 .sv_table = linux_sysent, 766 .sv_transtrap = linux_translate_traps, 767 .sv_fixup = linux_fixup_elf, 768 .sv_sendsig = linux_rt_sendsig, 769 .sv_sigcode = &_binary_linux_vdso_so_o_start, 770 .sv_szsigcode = &linux_szsigcode, 771 .sv_name = "Linux ELF64", 772 .sv_coredump = elf64_coredump, 773 .sv_elf_core_osabi = ELFOSABI_NONE, 774 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR, 775 .sv_elf_core_prepare_notes = linux64_prepare_notes, 776 .sv_imgact_try = linux_exec_imgact_try, 777 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 778 .sv_minuser = VM_MIN_ADDRESS, 779 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48, 780 .sv_usrstack = LINUX_USRSTACK_LA48, 781 .sv_psstrings = LINUX_PS_STRINGS_LA48, 782 .sv_psstringssz = sizeof(struct ps_strings), 783 .sv_stackprot = VM_PROT_ALL, 784 .sv_copyout_auxargs = linux_copyout_auxargs, 785 .sv_copyout_strings = linux_copyout_strings, 786 .sv_setregs = linux_exec_setregs, 787 .sv_fixlimit = NULL, 788 .sv_maxssiz = NULL, 789 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN | 790 SV_SIG_WAITNDQ | SV_TIMEKEEP, 791 .sv_set_syscall_retval = linux_set_syscall_retval, 792 .sv_fetch_syscall_args = linux_fetch_syscall_args, 793 .sv_syscallnames = NULL, 794 .sv_shared_page_base = LINUX_SHAREDPAGE_LA48, 795 .sv_shared_page_len = PAGE_SIZE, 796 .sv_schedtail = linux_schedtail, 797 .sv_thread_detach = linux_thread_detach, 798 .sv_trap = linux_vsyscall, 799 .sv_onexec = linux_on_exec_vmspace, 800 .sv_onexit = linux_on_exit, 801 .sv_ontdexit = linux_thread_dtor, 802 .sv_setid_allowed = &linux_setid_allowed_query, 803 .sv_set_fork_retval = linux_set_fork_retval, 804 }; 805 806 static int 807 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp) 808 { 809 int error; 810 811 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base, 812 LINUX_VDSOPAGE_SIZE, imgp); 813 if (error == 0) 814 linux_on_exec(p, imgp); 815 return (error); 816 } 817 818 /* 819 * linux_vdso_install() and linux_exec_sysvec_init() must be called 820 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY). 821 */ 822 static void 823 linux_exec_sysvec_init(void *param) 824 { 825 l_uintptr_t *ktimekeep_base, *ktsc_selector; 826 struct sysentvec *sv; 827 ptrdiff_t tkoff; 828 829 sv = param; 830 amd64_lower_shared_page(sv); 831 /* Fill timekeep_base */ 832 exec_sysvec_init(sv); 833 834 tkoff = kern_timekeep_base - linux_vdso_base; 835 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 836 *ktimekeep_base = sv->sv_timekeep_base; 837 838 tkoff = kern_tsc_selector - linux_vdso_base; 839 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 840 *ktsc_selector = linux_vdso_tsc_selector_idx(); 841 if (bootverbose) 842 printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector); 843 844 tkoff = kern_cpu_selector - linux_vdso_base; 845 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 846 *ktsc_selector = linux_vdso_cpu_selector_idx(); 847 if (bootverbose) 848 printf("Linux x86-64 vDSO cpu_selector: %lu\n", *ktsc_selector); 849 } 850 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY, 851 linux_exec_sysvec_init, &elf_linux_sysvec); 852 853 static void 854 linux_vdso_install(const void *param) 855 { 856 char *vdso_start = &_binary_linux_vdso_so_o_start; 857 char *vdso_end = &_binary_linux_vdso_so_o_end; 858 859 linux_szsigcode = vdso_end - vdso_start; 860 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE); 861 862 linux_vdso_base = LINUX_VDSOPAGE_LA48; 863 if (hw_lower_amd64_sharedpage != 0) 864 linux_vdso_base -= PAGE_SIZE; 865 866 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base); 867 868 linux_vdso_obj = __elfN(linux_shared_page_init) 869 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 870 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode); 871 872 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base); 873 } 874 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST, 875 linux_vdso_install, NULL); 876 877 static void 878 linux_vdso_deinstall(const void *param) 879 { 880 881 __elfN(linux_shared_page_fini)(linux_vdso_obj, 882 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 883 } 884 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 885 linux_vdso_deinstall, NULL); 886 887 static void 888 linux_vdso_reloc(char *mapping, Elf_Addr offset) 889 { 890 const Elf_Ehdr *ehdr; 891 const Elf_Shdr *shdr; 892 Elf64_Addr *where, val; 893 Elf_Size rtype, symidx; 894 const Elf_Rela *rela; 895 Elf_Addr addr, addend; 896 int relacnt; 897 int i, j; 898 899 MPASS(offset != 0); 900 901 relacnt = 0; 902 ehdr = (const Elf_Ehdr *)mapping; 903 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff); 904 for (i = 0; i < ehdr->e_shnum; i++) 905 { 906 switch (shdr[i].sh_type) { 907 case SHT_REL: 908 printf("Linux x86_64 vDSO: unexpected Rel section\n"); 909 break; 910 case SHT_RELA: 911 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset); 912 relacnt = shdr[i].sh_size / sizeof(*rela); 913 } 914 } 915 916 for (j = 0; j < relacnt; j++, rela++) { 917 where = (Elf_Addr *)(mapping + rela->r_offset); 918 addend = rela->r_addend; 919 rtype = ELF_R_TYPE(rela->r_info); 920 symidx = ELF_R_SYM(rela->r_info); 921 922 switch (rtype) { 923 case R_X86_64_NONE: /* none */ 924 break; 925 926 case R_X86_64_RELATIVE: /* B + A */ 927 addr = (Elf_Addr)(offset + addend); 928 val = addr; 929 if (*where != val) 930 *where = val; 931 break; 932 case R_X86_64_IRELATIVE: 933 printf("Linux x86_64 vDSO: unexpected ifunc relocation, " 934 "symbol index %ld\n", symidx); 935 break; 936 default: 937 printf("Linux x86_64 vDSO: unexpected relocation type %ld, " 938 "symbol index %ld\n", rtype, symidx); 939 } 940 } 941 } 942 943 static char GNULINUX_ABI_VENDOR[] = "GNU"; 944 static int GNULINUX_ABI_DESC = 0; 945 946 static bool 947 linux_trans_osrel(const Elf_Note *note, int32_t *osrel) 948 { 949 const Elf32_Word *desc; 950 uintptr_t p; 951 952 p = (uintptr_t)(note + 1); 953 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 954 955 desc = (const Elf32_Word *)p; 956 if (desc[0] != GNULINUX_ABI_DESC) 957 return (false); 958 959 /* 960 * For Linux we encode osrel using the Linux convention of 961 * (version << 16) | (major << 8) | (minor) 962 * See macro in linux_mib.h 963 */ 964 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 965 966 return (true); 967 } 968 969 static Elf_Brandnote linux64_brandnote = { 970 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), 971 .hdr.n_descsz = 16, 972 .hdr.n_type = 1, 973 .vendor = GNULINUX_ABI_VENDOR, 974 .flags = BN_TRANSLATE_OSREL, 975 .trans_osrel = linux_trans_osrel 976 }; 977 978 static Elf64_Brandinfo linux_glibc2brand = { 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-x86-64.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_glibc2brandshort = { 991 .brand = ELFOSABI_LINUX, 992 .machine = EM_X86_64, 993 .compat_3_brand = "Linux", 994 .emul_path = linux_emul_path, 995 .interp_path = "/lib64/ld-linux.so.2", 996 .sysvec = &elf_linux_sysvec, 997 .interp_newpath = NULL, 998 .brand_note = &linux64_brandnote, 999 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1000 }; 1001 1002 static Elf64_Brandinfo linux_muslbrand = { 1003 .brand = ELFOSABI_LINUX, 1004 .machine = EM_X86_64, 1005 .compat_3_brand = "Linux", 1006 .emul_path = linux_emul_path, 1007 .interp_path = "/lib/ld-musl-x86_64.so.1", 1008 .sysvec = &elf_linux_sysvec, 1009 .interp_newpath = NULL, 1010 .brand_note = &linux64_brandnote, 1011 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE | 1012 LINUX_BI_FUTEX_REQUEUE 1013 }; 1014 1015 Elf64_Brandinfo *linux_brandlist[] = { 1016 &linux_glibc2brand, 1017 &linux_glibc2brandshort, 1018 &linux_muslbrand, 1019 NULL 1020 }; 1021 1022 static int 1023 linux64_elf_modevent(module_t mod, int type, void *data) 1024 { 1025 Elf64_Brandinfo **brandinfo; 1026 int error; 1027 struct linux_ioctl_handler **lihp; 1028 1029 error = 0; 1030 1031 switch(type) { 1032 case MOD_LOAD: 1033 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1034 ++brandinfo) 1035 if (elf64_insert_brand_entry(*brandinfo) < 0) 1036 error = EINVAL; 1037 if (error == 0) { 1038 SET_FOREACH(lihp, linux_ioctl_handler_set) 1039 linux_ioctl_register_handler(*lihp); 1040 stclohz = (stathz ? stathz : hz); 1041 if (bootverbose) 1042 printf("Linux x86-64 ELF exec handler installed\n"); 1043 } else 1044 printf("cannot insert Linux x86-64 ELF brand handler\n"); 1045 break; 1046 case MOD_UNLOAD: 1047 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1048 ++brandinfo) 1049 if (elf64_brand_inuse(*brandinfo)) 1050 error = EBUSY; 1051 if (error == 0) { 1052 for (brandinfo = &linux_brandlist[0]; 1053 *brandinfo != NULL; ++brandinfo) 1054 if (elf64_remove_brand_entry(*brandinfo) < 0) 1055 error = EINVAL; 1056 } 1057 if (error == 0) { 1058 SET_FOREACH(lihp, linux_ioctl_handler_set) 1059 linux_ioctl_unregister_handler(*lihp); 1060 if (bootverbose) 1061 printf("Linux x86_64 ELF exec handler removed\n"); 1062 } else 1063 printf("Could not deinstall Linux x86_64 ELF interpreter entry\n"); 1064 break; 1065 default: 1066 return (EOPNOTSUPP); 1067 } 1068 return (error); 1069 } 1070 1071 static moduledata_t linux64_elf_mod = { 1072 "linux64elf", 1073 linux64_elf_modevent, 1074 0 1075 }; 1076 1077 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 1078 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 1079 FEATURE(linux64, "Linux 64bit support"); 1080