1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 1994-1996 Søren Schmidt 5 * Copyright (c) 2018 Turing Robotic Industries Inc. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/systm.h> 34 #include <sys/cdefs.h> 35 #include <sys/elf.h> 36 #include <sys/exec.h> 37 #include <sys/imgact.h> 38 #include <sys/imgact_elf.h> 39 #include <sys/kernel.h> 40 #include <sys/ktr.h> 41 #include <sys/lock.h> 42 #include <sys/module.h> 43 #include <sys/mutex.h> 44 #include <sys/proc.h> 45 #include <sys/stddef.h> 46 #include <sys/signalvar.h> 47 #include <sys/syscallsubr.h> 48 #include <sys/sysctl.h> 49 #include <sys/sysent.h> 50 51 #include <vm/vm.h> 52 #include <vm/pmap.h> 53 #include <vm/vm_map.h> 54 #include <vm/vm_extern.h> 55 #include <vm/vm_object.h> 56 #include <vm/vm_page.h> 57 #include <vm/vm_param.h> 58 59 #include <arm64/linux/linux.h> 60 #include <arm64/linux/linux_proto.h> 61 #include <compat/linux/linux_dtrace.h> 62 #include <compat/linux/linux_emul.h> 63 #include <compat/linux/linux_fork.h> 64 #include <compat/linux/linux_ioctl.h> 65 #include <compat/linux/linux_mib.h> 66 #include <compat/linux/linux_misc.h> 67 #include <compat/linux/linux_signal.h> 68 #include <compat/linux/linux_util.h> 69 #include <compat/linux/linux_vdso.h> 70 71 #include <arm64/linux/linux_sigframe.h> 72 73 #include <machine/md_var.h> 74 75 #ifdef VFP 76 #include <machine/vfp.h> 77 #endif 78 79 MODULE_VERSION(linux64elf, 1); 80 81 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2 82 #define LINUX_VDSOPAGE (VM_MAXUSER_ADDRESS - \ 83 LINUX_VDSOPAGE_SIZE) 84 #define LINUX_SHAREDPAGE (LINUX_VDSOPAGE - PAGE_SIZE) 85 /* 86 * PAGE_SIZE - the size 87 * of the native SHAREDPAGE 88 */ 89 #define LINUX_USRSTACK LINUX_SHAREDPAGE 90 #define LINUX_PS_STRINGS (LINUX_USRSTACK - \ 91 sizeof(struct ps_strings)) 92 93 static int linux_szsigcode; 94 static vm_object_t linux_vdso_obj; 95 static char *linux_vdso_mapping; 96 extern char _binary_linux_vdso_so_o_start; 97 extern char _binary_linux_vdso_so_o_end; 98 static vm_offset_t linux_vdso_base; 99 100 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 101 102 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 103 104 static int linux_copyout_strings(struct image_params *imgp, 105 uintptr_t *stack_base); 106 static int linux_elf_fixup(uintptr_t *stack_base, 107 struct image_params *iparams); 108 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 109 static void linux_vdso_install(const void *param); 110 static void linux_vdso_deinstall(const void *param); 111 static void linux_vdso_reloc(char *mapping, Elf_Addr offset); 112 static void linux_set_syscall_retval(struct thread *td, int error); 113 static int linux_fetch_syscall_args(struct thread *td); 114 static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 115 uintptr_t stack); 116 static void linux_exec_sysvec_init(void *param); 117 static int linux_on_exec_vmspace(struct proc *p, 118 struct image_params *imgp); 119 120 /* DTrace init */ 121 LIN_SDT_PROVIDER_DECLARE(LINUX_DTRACE); 122 123 /* DTrace probes */ 124 LIN_SDT_PROBE_DEFINE0(sysvec, linux_exec_setregs, todo); 125 LIN_SDT_PROBE_DEFINE0(sysvec, linux_copyout_auxargs, todo); 126 LIN_SDT_PROBE_DEFINE0(sysvec, linux_elf_fixup, todo); 127 128 LINUX_VDSO_SYM_CHAR(linux_platform); 129 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base); 130 LINUX_VDSO_SYM_INTPTR(linux_vdso_sigcode); 131 132 static int 133 linux_fetch_syscall_args(struct thread *td) 134 { 135 struct proc *p; 136 struct syscall_args *sa; 137 register_t *ap; 138 139 p = td->td_proc; 140 ap = td->td_frame->tf_x; 141 sa = &td->td_sa; 142 143 sa->code = td->td_frame->tf_x[8]; 144 sa->original_code = sa->code; 145 /* LINUXTODO: generic syscall? */ 146 if (sa->code >= p->p_sysent->sv_size) 147 sa->callp = &p->p_sysent->sv_table[0]; 148 else 149 sa->callp = &p->p_sysent->sv_table[sa->code]; 150 151 if (sa->callp->sy_narg > nitems(sa->args)) 152 panic("ARM64TODO: Could we have more than %zu args?", 153 nitems(sa->args)); 154 memcpy(sa->args, ap, nitems(sa->args) * sizeof(register_t)); 155 156 td->td_retval[0] = 0; 157 return (0); 158 } 159 160 static void 161 linux_set_syscall_retval(struct thread *td, int error) 162 { 163 164 td->td_retval[1] = td->td_frame->tf_x[1]; 165 cpu_set_syscall_retval(td, error); 166 167 if (__predict_false(error != 0)) { 168 if (error != ERESTART && error != EJUSTRETURN) 169 td->td_frame->tf_x[0] = bsd_to_linux_errno(error); 170 } 171 } 172 173 static int 174 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) 175 { 176 Elf_Auxargs *args; 177 Elf_Auxinfo *argarray, *pos; 178 struct proc *p; 179 int error, issetugid; 180 181 LIN_SDT_PROBE0(sysvec, linux_copyout_auxargs, todo); 182 p = imgp->proc; 183 184 args = (Elf64_Auxargs *)imgp->auxargs; 185 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 186 M_WAITOK | M_ZERO); 187 188 issetugid = p->p_flag & P_SUGID ? 1 : 0; 189 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base); 190 AUXARGS_ENTRY(pos, LINUX_AT_MINSIGSTKSZ, LINUX_MINSIGSTKSZ); 191 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, *imgp->sysent->sv_hwcap); 192 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 193 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 194 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 195 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 196 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 197 AUXARGS_ENTRY(pos, AT_BASE, args->base); 198 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 199 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 200 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 201 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 202 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 203 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 204 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 205 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); 206 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, *imgp->sysent->sv_hwcap2); 207 if (imgp->execpathp != 0) 208 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); 209 if (args->execfd != -1) 210 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 211 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 212 AUXARGS_ENTRY(pos, AT_NULL, 0); 213 214 free(imgp->auxargs, M_TEMP); 215 imgp->auxargs = NULL; 216 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 217 218 error = copyout(argarray, (void *)base, 219 sizeof(*argarray) * LINUX_AT_COUNT); 220 free(argarray, M_TEMP); 221 return (error); 222 } 223 224 static int 225 linux_elf_fixup(uintptr_t *stack_base, struct image_params *imgp) 226 { 227 228 LIN_SDT_PROBE0(sysvec, linux_elf_fixup, todo); 229 230 return (0); 231 } 232 233 /* 234 * Copy strings out to the new process address space, constructing new arg 235 * and env vector tables. Return a pointer to the base so that it can be used 236 * as the initial stack pointer. 237 * LINUXTODO: deduplicate against other linuxulator archs 238 */ 239 static int 240 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 241 { 242 char **vectp; 243 char *stringp; 244 uintptr_t destp, ustringp; 245 struct ps_strings *arginfo; 246 char canary[LINUX_AT_RANDOM_LEN]; 247 size_t execpath_len; 248 struct proc *p; 249 int argc, envc, error; 250 251 p = imgp->proc; 252 arginfo = (struct ps_strings *)PROC_PS_STRINGS(p); 253 destp = (uintptr_t)arginfo; 254 255 if (imgp->execpath != NULL && imgp->auxargs != NULL) { 256 execpath_len = strlen(imgp->execpath) + 1; 257 destp -= execpath_len; 258 destp = rounddown2(destp, sizeof(void *)); 259 imgp->execpathp = (void *)destp; 260 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 261 if (error != 0) 262 return (error); 263 } 264 265 /* Prepare the canary for SSP. */ 266 arc4rand(canary, sizeof(canary), 0); 267 destp -= roundup(sizeof(canary), sizeof(void *)); 268 imgp->canary = (void *)destp; 269 error = copyout(canary, imgp->canary, sizeof(canary)); 270 if (error != 0) 271 return (error); 272 273 /* Allocate room for the argument and environment strings. */ 274 destp -= ARG_MAX - imgp->args->stringspace; 275 destp = rounddown2(destp, sizeof(void *)); 276 ustringp = destp; 277 278 if (imgp->auxargs) { 279 /* 280 * Allocate room on the stack for the ELF auxargs 281 * array. It has up to LINUX_AT_COUNT entries. 282 */ 283 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo); 284 destp = rounddown2(destp, sizeof(void *)); 285 } 286 287 vectp = (char **)destp; 288 289 /* 290 * Allocate room for argc and the argv[] and env vectors including the 291 * terminating NULL pointers. 292 */ 293 vectp -= 1 + imgp->args->argc + 1 + imgp->args->envc + 1; 294 vectp = (char **)STACKALIGN(vectp); 295 296 /* vectp also becomes our initial stack base. */ 297 *stack_base = (uintptr_t)vectp; 298 299 stringp = imgp->args->begin_argv; 300 argc = imgp->args->argc; 301 envc = imgp->args->envc; 302 303 /* Copy out strings - arguments and environment. */ 304 error = copyout(stringp, (void *)ustringp, 305 ARG_MAX - imgp->args->stringspace); 306 if (error != 0) 307 return (error); 308 309 /* Fill in "ps_strings" struct for ps, w, etc. */ 310 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 311 suword(&arginfo->ps_nargvstr, argc) != 0) 312 return (EFAULT); 313 314 if (suword(vectp++, argc) != 0) 315 return (EFAULT); 316 317 /* Fill in argument portion of vector table. */ 318 for (; argc > 0; --argc) { 319 if (suword(vectp++, ustringp) != 0) 320 return (EFAULT); 321 while (*stringp++ != 0) 322 ustringp++; 323 ustringp++; 324 } 325 326 /* A null vector table pointer separates the argp's from the envp's. */ 327 if (suword(vectp++, 0) != 0) 328 return (EFAULT); 329 330 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 331 suword(&arginfo->ps_nenvstr, envc) != 0) 332 return (EFAULT); 333 334 /* Fill in environment portion of vector table. */ 335 for (; envc > 0; --envc) { 336 if (suword(vectp++, ustringp) != 0) 337 return (EFAULT); 338 while (*stringp++ != 0) 339 ustringp++; 340 ustringp++; 341 } 342 343 /* The end of the vector table is a null pointer. */ 344 if (suword(vectp, 0) != 0) 345 return (EFAULT); 346 347 if (imgp->auxargs) { 348 vectp++; 349 error = imgp->sysent->sv_copyout_auxargs(imgp, 350 (uintptr_t)vectp); 351 if (error != 0) 352 return (error); 353 } 354 355 return (0); 356 } 357 358 /* 359 * Reset registers to default values on exec. 360 */ 361 static void 362 linux_exec_setregs(struct thread *td, struct image_params *imgp, 363 uintptr_t stack) 364 { 365 struct trapframe *regs = td->td_frame; 366 struct pcb *pcb = td->td_pcb; 367 368 /* LINUXTODO: validate */ 369 LIN_SDT_PROBE0(sysvec, linux_exec_setregs, todo); 370 371 memset(regs, 0, sizeof(*regs)); 372 /* glibc start.S registers function pointer in x0 with atexit. */ 373 regs->tf_sp = stack; 374 #if 0 /* LINUXTODO: See if this is used. */ 375 regs->tf_lr = imgp->entry_addr; 376 #else 377 regs->tf_lr = 0xffffffffffffffff; 378 #endif 379 regs->tf_elr = imgp->entry_addr; 380 381 pcb->pcb_tpidr_el0 = 0; 382 pcb->pcb_tpidrro_el0 = 0; 383 WRITE_SPECIALREG(tpidrro_el0, 0); 384 WRITE_SPECIALREG(tpidr_el0, 0); 385 386 #ifdef VFP 387 vfp_reset_state(td, pcb); 388 #endif 389 390 /* 391 * Clear debug register state. It is not applicable to the new process. 392 */ 393 bzero(&pcb->pcb_dbg_regs, sizeof(pcb->pcb_dbg_regs)); 394 } 395 396 int 397 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 398 { 399 struct l_sigframe *frame; 400 ucontext_t uc; 401 struct trapframe *tf; 402 int error; 403 404 tf = td->td_frame; 405 frame = (struct l_sigframe *)tf->tf_sp; 406 407 if (copyin((void *)&frame->uc, &uc, sizeof(uc))) 408 return (EFAULT); 409 410 error = set_mcontext(td, &uc.uc_mcontext); 411 if (error != 0) 412 return (error); 413 414 /* Restore signal mask. */ 415 kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0); 416 417 return (EJUSTRETURN); 418 } 419 420 static void 421 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 422 { 423 struct thread *td; 424 struct proc *p; 425 struct trapframe *tf; 426 struct l_sigframe *fp, *frame; 427 struct l_fpsimd_context *fpsimd; 428 struct l_esr_context *esr; 429 l_stack_t uc_stack; 430 ucontext_t uc; 431 uint8_t *scr; 432 struct sigacts *psp; 433 int onstack, sig, issiginfo; 434 435 td = curthread; 436 p = td->td_proc; 437 PROC_LOCK_ASSERT(p, MA_OWNED); 438 439 sig = ksi->ksi_signo; 440 psp = p->p_sigacts; 441 mtx_assert(&psp->ps_mtx, MA_OWNED); 442 443 tf = td->td_frame; 444 onstack = sigonstack(tf->tf_sp); 445 issiginfo = SIGISMEMBER(psp->ps_siginfo, sig); 446 447 CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm, 448 catcher, sig); 449 450 /* Allocate and validate space for the signal handler context. */ 451 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !onstack && 452 SIGISMEMBER(psp->ps_sigonstack, sig)) { 453 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 454 td->td_sigstk.ss_size); 455 #if defined(COMPAT_43) 456 td->td_sigstk.ss_flags |= SS_ONSTACK; 457 #endif 458 } else { 459 fp = (struct l_sigframe *)td->td_frame->tf_sp; 460 } 461 462 /* Make room, keeping the stack aligned */ 463 fp--; 464 fp = (struct l_sigframe *)STACKALIGN(fp); 465 466 get_mcontext(td, &uc.uc_mcontext, 0); 467 uc.uc_sigmask = *mask; 468 469 uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 470 uc_stack.ss_size = td->td_sigstk.ss_size; 471 uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) != 0 ? 472 (onstack ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 473 mtx_unlock(&psp->ps_mtx); 474 PROC_UNLOCK(td->td_proc); 475 476 /* Fill in the frame to copy out */ 477 frame = malloc(sizeof(*frame), M_LINUX, M_WAITOK | M_ZERO); 478 479 memcpy(&frame->sf.sf_uc.uc_sc.regs, tf->tf_x, sizeof(tf->tf_x)); 480 frame->sf.sf_uc.uc_sc.regs[30] = tf->tf_lr; 481 frame->sf.sf_uc.uc_sc.sp = tf->tf_sp; 482 frame->sf.sf_uc.uc_sc.pc = tf->tf_lr; 483 frame->sf.sf_uc.uc_sc.pstate = tf->tf_spsr; 484 frame->sf.sf_uc.uc_sc.fault_address = (register_t)ksi->ksi_addr; 485 486 /* Stack frame for unwinding */ 487 frame->fp = tf->tf_x[29]; 488 frame->lr = tf->tf_lr; 489 490 /* Translate the signal. */ 491 sig = bsd_to_linux_signal(sig); 492 siginfo_to_lsiginfo(&ksi->ksi_info, &frame->sf.sf_si, sig); 493 bsd_to_linux_sigset(mask, &frame->sf.sf_uc.uc_sigmask); 494 495 /* 496 * Prepare fpsimd & esr. Does not check sizes, as 497 * __reserved is big enougth. 498 */ 499 scr = (uint8_t *)&frame->sf.sf_uc.uc_sc.__reserved; 500 #ifdef VFP 501 fpsimd = (struct l_fpsimd_context *) scr; 502 fpsimd->head.magic = L_FPSIMD_MAGIC; 503 fpsimd->head.size = sizeof(struct l_fpsimd_context); 504 fpsimd->fpsr = uc.uc_mcontext.mc_fpregs.fp_sr; 505 fpsimd->fpcr = uc.uc_mcontext.mc_fpregs.fp_cr; 506 507 memcpy(fpsimd->vregs, &uc.uc_mcontext.mc_fpregs.fp_q, 508 sizeof(uc.uc_mcontext.mc_fpregs.fp_q)); 509 scr += roundup(sizeof(struct l_fpsimd_context), 16); 510 #endif 511 if (ksi->ksi_addr != 0) { 512 esr = (struct l_esr_context *) scr; 513 esr->head.magic = L_ESR_MAGIC; 514 esr->head.size = sizeof(struct l_esr_context); 515 esr->esr = tf->tf_esr; 516 } 517 518 memcpy(&frame->sf.sf_uc.uc_stack, &uc_stack, sizeof(uc_stack)); 519 memcpy(&frame->uc, &uc, sizeof(uc)); 520 521 /* Copy the sigframe out to the user's stack. */ 522 if (copyout(frame, fp, sizeof(*fp)) != 0) { 523 /* Process has trashed its stack. Kill it. */ 524 free(frame, M_LINUX); 525 CTR2(KTR_SIG, "sendsig: sigexit td=%p fp=%p", td, fp); 526 PROC_LOCK(p); 527 sigexit(td, SIGILL); 528 } 529 free(frame, M_LINUX); 530 531 tf->tf_x[0]= sig; 532 if (issiginfo) { 533 tf->tf_x[1] = (register_t)&fp->sf.sf_si; 534 tf->tf_x[2] = (register_t)&fp->sf.sf_uc; 535 } else { 536 tf->tf_x[1] = 0; 537 tf->tf_x[2] = 0; 538 } 539 tf->tf_x[8] = (register_t)catcher; 540 tf->tf_sp = (register_t)fp; 541 tf->tf_elr = (register_t)linux_vdso_sigcode; 542 543 CTR3(KTR_SIG, "sendsig: return td=%p pc=%#x sp=%#x", td, tf->tf_elr, 544 tf->tf_sp); 545 546 PROC_LOCK(p); 547 mtx_lock(&psp->ps_mtx); 548 } 549 550 struct sysentvec elf_linux_sysvec = { 551 .sv_size = LINUX_SYS_MAXSYSCALL, 552 .sv_table = linux_sysent, 553 .sv_fixup = linux_elf_fixup, 554 .sv_sendsig = linux_rt_sendsig, 555 .sv_sigcode = &_binary_linux_vdso_so_o_start, 556 .sv_szsigcode = &linux_szsigcode, 557 .sv_name = "Linux ELF64", 558 .sv_coredump = elf64_coredump, 559 .sv_elf_core_osabi = ELFOSABI_NONE, 560 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR, 561 .sv_elf_core_prepare_notes = linux64_prepare_notes, 562 .sv_imgact_try = linux_exec_imgact_try, 563 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 564 .sv_minuser = VM_MIN_ADDRESS, 565 .sv_maxuser = VM_MAXUSER_ADDRESS, 566 .sv_usrstack = LINUX_USRSTACK, 567 .sv_psstrings = LINUX_PS_STRINGS, 568 .sv_psstringssz = sizeof(struct ps_strings), 569 .sv_stackprot = VM_PROT_READ | VM_PROT_WRITE, 570 .sv_copyout_auxargs = linux_copyout_auxargs, 571 .sv_copyout_strings = linux_copyout_strings, 572 .sv_setregs = linux_exec_setregs, 573 .sv_fixlimit = NULL, 574 .sv_maxssiz = NULL, 575 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN | 576 SV_SIG_WAITNDQ | SV_TIMEKEEP, 577 .sv_set_syscall_retval = linux_set_syscall_retval, 578 .sv_fetch_syscall_args = linux_fetch_syscall_args, 579 .sv_syscallnames = NULL, 580 .sv_shared_page_base = LINUX_SHAREDPAGE, 581 .sv_shared_page_len = PAGE_SIZE, 582 .sv_schedtail = linux_schedtail, 583 .sv_thread_detach = linux_thread_detach, 584 .sv_trap = NULL, 585 .sv_hwcap = &elf_hwcap, 586 .sv_hwcap2 = &elf_hwcap2, 587 .sv_onexec = linux_on_exec_vmspace, 588 .sv_onexit = linux_on_exit, 589 .sv_ontdexit = linux_thread_dtor, 590 .sv_setid_allowed = &linux_setid_allowed_query, 591 }; 592 593 static int 594 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp) 595 { 596 int error; 597 598 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base, 599 LINUX_VDSOPAGE_SIZE, imgp); 600 if (error == 0) 601 linux_on_exec(p, imgp); 602 return (error); 603 } 604 605 /* 606 * linux_vdso_install() and linux_exec_sysvec_init() must be called 607 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY). 608 */ 609 static void 610 linux_exec_sysvec_init(void *param) 611 { 612 l_uintptr_t *ktimekeep_base; 613 struct sysentvec *sv; 614 ptrdiff_t tkoff; 615 616 sv = param; 617 /* Fill timekeep_base */ 618 exec_sysvec_init(sv); 619 620 tkoff = kern_timekeep_base - linux_vdso_base; 621 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 622 *ktimekeep_base = sv->sv_timekeep_base; 623 } 624 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY, 625 linux_exec_sysvec_init, &elf_linux_sysvec); 626 627 static void 628 linux_vdso_install(const void *param) 629 { 630 char *vdso_start = &_binary_linux_vdso_so_o_start; 631 char *vdso_end = &_binary_linux_vdso_so_o_end; 632 633 linux_szsigcode = vdso_end - vdso_start; 634 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE); 635 636 linux_vdso_base = LINUX_VDSOPAGE; 637 638 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base); 639 640 linux_vdso_obj = __elfN(linux_shared_page_init) 641 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 642 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode); 643 644 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base); 645 } 646 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST, 647 linux_vdso_install, NULL); 648 649 static void 650 linux_vdso_deinstall(const void *param) 651 { 652 653 __elfN(linux_shared_page_fini)(linux_vdso_obj, 654 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 655 } 656 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 657 linux_vdso_deinstall, NULL); 658 659 static void 660 linux_vdso_reloc(char *mapping, Elf_Addr offset) 661 { 662 Elf_Size rtype, symidx; 663 const Elf_Rela *rela; 664 const Elf_Shdr *shdr; 665 const Elf_Ehdr *ehdr; 666 Elf_Addr *where; 667 Elf_Addr addr, addend; 668 int i, relacnt; 669 670 MPASS(offset != 0); 671 672 relacnt = 0; 673 ehdr = (const Elf_Ehdr *)mapping; 674 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff); 675 for (i = 0; i < ehdr->e_shnum; i++) 676 { 677 switch (shdr[i].sh_type) { 678 case SHT_REL: 679 printf("Linux Aarch64 vDSO: unexpected Rel section\n"); 680 break; 681 case SHT_RELA: 682 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset); 683 relacnt = shdr[i].sh_size / sizeof(*rela); 684 } 685 } 686 687 for (i = 0; i < relacnt; i++, rela++) { 688 where = (Elf_Addr *)(mapping + rela->r_offset); 689 addend = rela->r_addend; 690 rtype = ELF_R_TYPE(rela->r_info); 691 symidx = ELF_R_SYM(rela->r_info); 692 693 switch (rtype) { 694 case R_AARCH64_NONE: /* none */ 695 break; 696 697 case R_AARCH64_RELATIVE: /* B + A */ 698 addr = (Elf_Addr)(mapping + addend); 699 if (*where != addr) 700 *where = addr; 701 break; 702 default: 703 printf("Linux Aarch64 vDSO: unexpected relocation type %ld, " 704 "symbol index %ld\n", rtype, symidx); 705 } 706 } 707 } 708 709 static char GNU_ABI_VENDOR[] = "GNU"; 710 static int GNU_ABI_LINUX = 0; 711 712 /* LINUXTODO: deduplicate */ 713 static bool 714 linux_trans_osrel(const Elf_Note *note, int32_t *osrel) 715 { 716 const Elf32_Word *desc; 717 uintptr_t p; 718 719 p = (uintptr_t)(note + 1); 720 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 721 722 desc = (const Elf32_Word *)p; 723 if (desc[0] != GNU_ABI_LINUX) 724 return (false); 725 726 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 727 return (true); 728 } 729 730 static Elf_Brandnote linux64_brandnote = { 731 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 732 .hdr.n_descsz = 16, 733 .hdr.n_type = 1, 734 .vendor = GNU_ABI_VENDOR, 735 .flags = BN_TRANSLATE_OSREL, 736 .trans_osrel = linux_trans_osrel 737 }; 738 739 static Elf64_Brandinfo linux_glibc2brand = { 740 .brand = ELFOSABI_LINUX, 741 .machine = EM_AARCH64, 742 .compat_3_brand = "Linux", 743 .emul_path = linux_emul_path, 744 .interp_path = "/lib64/ld-linux-x86-64.so.2", 745 .sysvec = &elf_linux_sysvec, 746 .interp_newpath = NULL, 747 .brand_note = &linux64_brandnote, 748 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 749 }; 750 751 Elf64_Brandinfo *linux_brandlist[] = { 752 &linux_glibc2brand, 753 NULL 754 }; 755 756 static int 757 linux64_elf_modevent(module_t mod, int type, void *data) 758 { 759 Elf64_Brandinfo **brandinfo; 760 struct linux_ioctl_handler**lihp; 761 int error; 762 763 error = 0; 764 switch(type) { 765 case MOD_LOAD: 766 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 767 ++brandinfo) 768 if (elf64_insert_brand_entry(*brandinfo) < 0) 769 error = EINVAL; 770 if (error == 0) { 771 SET_FOREACH(lihp, linux_ioctl_handler_set) 772 linux_ioctl_register_handler(*lihp); 773 stclohz = (stathz ? stathz : hz); 774 if (bootverbose) 775 printf("Linux arm64 ELF exec handler installed\n"); 776 } 777 break; 778 case MOD_UNLOAD: 779 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 780 ++brandinfo) 781 if (elf64_brand_inuse(*brandinfo)) 782 error = EBUSY; 783 if (error == 0) { 784 for (brandinfo = &linux_brandlist[0]; 785 *brandinfo != NULL; ++brandinfo) 786 if (elf64_remove_brand_entry(*brandinfo) < 0) 787 error = EINVAL; 788 } 789 if (error == 0) { 790 SET_FOREACH(lihp, linux_ioctl_handler_set) 791 linux_ioctl_unregister_handler(*lihp); 792 if (bootverbose) 793 printf("Linux arm64 ELF exec handler removed\n"); 794 } else 795 printf("Could not deinstall Linux arm64 ELF interpreter entry\n"); 796 break; 797 default: 798 return (EOPNOTSUPP); 799 } 800 return (error); 801 } 802 803 static moduledata_t linux64_elf_mod = { 804 "linux64elf", 805 linux64_elf_modevent, 806 0 807 }; 808 809 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 810 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 811 FEATURE(linux64, "AArch64 Linux 64bit support"); 812