1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 1994-1996 Søren Schmidt 5 * All rights reserved. 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/exec.h> 35 #include <sys/fcntl.h> 36 #include <sys/imgact.h> 37 #include <sys/imgact_aout.h> 38 #include <sys/imgact_elf.h> 39 #include <sys/kernel.h> 40 #include <sys/lock.h> 41 #include <sys/malloc.h> 42 #include <sys/module.h> 43 #include <sys/mutex.h> 44 #include <sys/proc.h> 45 #include <sys/signalvar.h> 46 #include <sys/syscallsubr.h> 47 #include <sys/sysctl.h> 48 #include <sys/sysent.h> 49 #include <sys/sysproto.h> 50 #include <sys/vnode.h> 51 #include <sys/eventhandler.h> 52 53 #include <vm/vm.h> 54 #include <vm/pmap.h> 55 #include <vm/vm_extern.h> 56 #include <vm/vm_map.h> 57 #include <vm/vm_object.h> 58 #include <vm/vm_page.h> 59 #include <vm/vm_param.h> 60 61 #include <machine/cpu.h> 62 #include <machine/cputypes.h> 63 #include <machine/md_var.h> 64 #include <machine/pcb.h> 65 #include <machine/trap.h> 66 67 #include <i386/linux/linux.h> 68 #include <i386/linux/linux_proto.h> 69 #include <compat/linux/linux_emul.h> 70 #include <compat/linux/linux_ioctl.h> 71 #include <compat/linux/linux_mib.h> 72 #include <compat/linux/linux_misc.h> 73 #include <compat/linux/linux_signal.h> 74 #include <compat/linux/linux_util.h> 75 #include <compat/linux/linux_vdso.h> 76 77 MODULE_VERSION(linux, 1); 78 79 #define LINUX_PS_STRINGS (LINUX_USRSTACK - sizeof(struct ps_strings)) 80 81 static int linux_szsigcode; 82 static vm_object_t linux_shared_page_obj; 83 static char *linux_shared_page_mapping; 84 extern char _binary_linux_locore_o_start; 85 extern char _binary_linux_locore_o_end; 86 87 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 88 89 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 90 91 static int linux_fixup(uintptr_t *stack_base, 92 struct image_params *iparams); 93 static int linux_fixup_elf(uintptr_t *stack_base, 94 struct image_params *iparams); 95 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask); 96 static void linux_exec_setregs(struct thread *td, 97 struct image_params *imgp, uintptr_t stack); 98 static int linux_copyout_strings(struct image_params *imgp, 99 uintptr_t *stack_base); 100 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 101 static void linux_vdso_install(void *param); 102 static void linux_vdso_deinstall(void *param); 103 104 static int linux_szplatform; 105 const char *linux_kplatform; 106 107 static eventhandler_tag linux_exit_tag; 108 static eventhandler_tag linux_exec_tag; 109 static eventhandler_tag linux_thread_dtor_tag; 110 111 #define LINUX_T_UNKNOWN 255 112 static int _bsd_to_linux_trapcode[] = { 113 LINUX_T_UNKNOWN, /* 0 */ 114 6, /* 1 T_PRIVINFLT */ 115 LINUX_T_UNKNOWN, /* 2 */ 116 3, /* 3 T_BPTFLT */ 117 LINUX_T_UNKNOWN, /* 4 */ 118 LINUX_T_UNKNOWN, /* 5 */ 119 16, /* 6 T_ARITHTRAP */ 120 254, /* 7 T_ASTFLT */ 121 LINUX_T_UNKNOWN, /* 8 */ 122 13, /* 9 T_PROTFLT */ 123 1, /* 10 T_TRCTRAP */ 124 LINUX_T_UNKNOWN, /* 11 */ 125 14, /* 12 T_PAGEFLT */ 126 LINUX_T_UNKNOWN, /* 13 */ 127 17, /* 14 T_ALIGNFLT */ 128 LINUX_T_UNKNOWN, /* 15 */ 129 LINUX_T_UNKNOWN, /* 16 */ 130 LINUX_T_UNKNOWN, /* 17 */ 131 0, /* 18 T_DIVIDE */ 132 2, /* 19 T_NMI */ 133 4, /* 20 T_OFLOW */ 134 5, /* 21 T_BOUND */ 135 7, /* 22 T_DNA */ 136 8, /* 23 T_DOUBLEFLT */ 137 9, /* 24 T_FPOPFLT */ 138 10, /* 25 T_TSSFLT */ 139 11, /* 26 T_SEGNPFLT */ 140 12, /* 27 T_STKFLT */ 141 18, /* 28 T_MCHK */ 142 19, /* 29 T_XMMFLT */ 143 15 /* 30 T_RESERVED */ 144 }; 145 #define bsd_to_linux_trapcode(code) \ 146 ((code)<nitems(_bsd_to_linux_trapcode)? \ 147 _bsd_to_linux_trapcode[(code)]: \ 148 LINUX_T_UNKNOWN) 149 150 LINUX_VDSO_SYM_INTPTR(linux_sigcode); 151 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 152 LINUX_VDSO_SYM_INTPTR(linux_vsyscall); 153 154 /* 155 * If FreeBSD & Linux have a difference of opinion about what a trap 156 * means, deal with it here. 157 * 158 * MPSAFE 159 */ 160 static int 161 linux_translate_traps(int signal, int trap_code) 162 { 163 if (signal != SIGBUS) 164 return (signal); 165 switch (trap_code) { 166 case T_PROTFLT: 167 case T_TSSFLT: 168 case T_DOUBLEFLT: 169 case T_PAGEFLT: 170 return (SIGSEGV); 171 default: 172 return (signal); 173 } 174 } 175 176 static int 177 linux_fixup(uintptr_t *stack_base, struct image_params *imgp) 178 { 179 register_t *base, *argv, *envp; 180 181 base = (register_t *)*stack_base; 182 argv = base; 183 envp = base + (imgp->args->argc + 1); 184 base--; 185 suword(base, (intptr_t)envp); 186 base--; 187 suword(base, (intptr_t)argv); 188 base--; 189 suword(base, imgp->args->argc); 190 *stack_base = (uintptr_t)base; 191 return (0); 192 } 193 194 static int 195 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) 196 { 197 struct proc *p; 198 Elf32_Auxargs *args; 199 Elf32_Auxinfo *argarray, *pos; 200 Elf32_Addr *uplatform; 201 struct ps_strings *arginfo; 202 int error, issetugid; 203 204 p = imgp->proc; 205 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0; 206 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 207 uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform); 208 args = (Elf32_Auxargs *)imgp->auxargs; 209 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 210 M_WAITOK | M_ZERO); 211 212 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, 213 imgp->proc->p_sysent->sv_shared_page_base); 214 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall); 215 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 216 217 /* 218 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, 219 * as it has appeared in the 2.4.0-rc7 first time. 220 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), 221 * glibc falls back to the hard-coded CLK_TCK value when aux entry 222 * is not present. 223 * Also see linux_times() implementation. 224 */ 225 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) 226 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 227 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 228 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 229 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 230 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 231 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 232 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 233 AUXARGS_ENTRY(pos, AT_BASE, args->base); 234 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 235 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 236 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 237 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 238 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 239 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform)); 240 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); 241 if (imgp->execpathp != 0) 242 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); 243 if (args->execfd != -1) 244 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 245 AUXARGS_ENTRY(pos, AT_NULL, 0); 246 247 free(imgp->auxargs, M_TEMP); 248 imgp->auxargs = NULL; 249 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 250 251 error = copyout(argarray, (void *)base, 252 sizeof(*argarray) * LINUX_AT_COUNT); 253 free(argarray, M_TEMP); 254 return (error); 255 } 256 257 static int 258 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) 259 { 260 register_t *base; 261 262 base = (register_t *)*stack_base; 263 base--; 264 if (suword(base, (register_t)imgp->args->argc) == -1) 265 return (EFAULT); 266 *stack_base = (uintptr_t)base; 267 return (0); 268 } 269 270 /* 271 * Copied from kern/kern_exec.c 272 */ 273 static int 274 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 275 { 276 int argc, envc, error; 277 char **vectp; 278 char *stringp; 279 uintptr_t destp, ustringp; 280 struct ps_strings *arginfo; 281 char canary[LINUX_AT_RANDOM_LEN]; 282 size_t execpath_len; 283 struct proc *p; 284 285 /* Calculate string base and vector table pointers. */ 286 p = imgp->proc; 287 if (imgp->execpath != NULL && imgp->auxargs != NULL) 288 execpath_len = strlen(imgp->execpath) + 1; 289 else 290 execpath_len = 0; 291 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 292 destp = (uintptr_t)arginfo; 293 294 /* Install LINUX_PLATFORM. */ 295 destp -= linux_szplatform; 296 destp = rounddown2(destp, sizeof(void *)); 297 error = copyout(linux_kplatform, (void *)destp, linux_szplatform); 298 if (error != 0) 299 return (error); 300 301 if (execpath_len != 0) { 302 destp -= execpath_len; 303 destp = rounddown2(destp, sizeof(void *)); 304 imgp->execpathp = (void *)destp; 305 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 306 if (error != 0) 307 return (error); 308 } 309 310 /* Prepare the canary for SSP. */ 311 arc4rand(canary, sizeof(canary), 0); 312 destp -= roundup(sizeof(canary), sizeof(void *)); 313 imgp->canary = (void *)destp; 314 error = copyout(canary, imgp->canary, sizeof(canary)); 315 if (error != 0) 316 return (error); 317 318 /* Allocate room for the argument and environment strings. */ 319 destp -= ARG_MAX - imgp->args->stringspace; 320 destp = rounddown2(destp, sizeof(void *)); 321 ustringp = destp; 322 323 if (imgp->auxargs) { 324 /* 325 * Allocate room on the stack for the ELF auxargs 326 * array. It has LINUX_AT_COUNT entries. 327 */ 328 destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo); 329 destp = rounddown2(destp, sizeof(void *)); 330 } 331 332 vectp = (char **)destp; 333 334 /* 335 * Allocate room for the argv[] and env vectors including the 336 * terminating NULL pointers. 337 */ 338 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 339 340 /* vectp also becomes our initial stack base. */ 341 *stack_base = (uintptr_t)vectp; 342 343 stringp = imgp->args->begin_argv; 344 argc = imgp->args->argc; 345 envc = imgp->args->envc; 346 347 /* Copy out strings - arguments and environment. */ 348 error = copyout(stringp, (void *)ustringp, 349 ARG_MAX - imgp->args->stringspace); 350 if (error != 0) 351 return (error); 352 353 /* Fill in "ps_strings" struct for ps, w, etc. */ 354 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 355 suword(&arginfo->ps_nargvstr, argc) != 0) 356 return (EFAULT); 357 358 /* Fill in argument portion of vector table. */ 359 for (; argc > 0; --argc) { 360 if (suword(vectp++, ustringp) != 0) 361 return (EFAULT); 362 while (*stringp++ != 0) 363 ustringp++; 364 ustringp++; 365 } 366 367 /* A null vector table pointer separates the argp's from the envp's. */ 368 if (suword(vectp++, 0) != 0) 369 return (EFAULT); 370 371 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 372 suword(&arginfo->ps_nenvstr, envc) != 0) 373 return (EFAULT); 374 375 /* Fill in environment portion of vector table. */ 376 for (; envc > 0; --envc) { 377 if (suword(vectp++, ustringp) != 0) 378 return (EFAULT); 379 while (*stringp++ != 0) 380 ustringp++; 381 ustringp++; 382 } 383 384 /* The end of the vector table is a null pointer. */ 385 if (suword(vectp, 0) != 0) 386 return (EFAULT); 387 388 if (imgp->auxargs) { 389 vectp++; 390 error = imgp->sysent->sv_copyout_auxargs(imgp, 391 (uintptr_t)vectp); 392 if (error != 0) 393 return (error); 394 } 395 396 return (0); 397 } 398 399 static void 400 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 401 { 402 struct thread *td = curthread; 403 struct proc *p = td->td_proc; 404 struct sigacts *psp; 405 struct trapframe *regs; 406 struct l_rt_sigframe *fp, frame; 407 int sig, code; 408 int oonstack; 409 410 sig = ksi->ksi_signo; 411 code = ksi->ksi_code; 412 PROC_LOCK_ASSERT(p, MA_OWNED); 413 psp = p->p_sigacts; 414 mtx_assert(&psp->ps_mtx, MA_OWNED); 415 regs = td->td_frame; 416 oonstack = sigonstack(regs->tf_esp); 417 418 /* Allocate space for the signal handler context. */ 419 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 420 SIGISMEMBER(psp->ps_sigonstack, sig)) { 421 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 422 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe)); 423 } else 424 fp = (struct l_rt_sigframe *)regs->tf_esp - 1; 425 mtx_unlock(&psp->ps_mtx); 426 427 /* Build the argument list for the signal handler. */ 428 sig = bsd_to_linux_signal(sig); 429 430 bzero(&frame, sizeof(frame)); 431 432 frame.sf_handler = catcher; 433 frame.sf_sig = sig; 434 frame.sf_siginfo = &fp->sf_si; 435 frame.sf_ucontext = &fp->sf_sc; 436 437 /* Fill in POSIX parts. */ 438 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig); 439 440 /* Build the signal context to be used by sigreturn. */ 441 frame.sf_sc.uc_flags = 0; /* XXX ??? */ 442 frame.sf_sc.uc_link = NULL; /* XXX ??? */ 443 444 frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp; 445 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 446 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 447 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 448 PROC_UNLOCK(p); 449 450 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask); 451 452 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask; 453 frame.sf_sc.uc_mcontext.sc_gs = rgs(); 454 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs; 455 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es; 456 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds; 457 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_edi; 458 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_esi; 459 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_ebp; 460 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_ebx; 461 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_esp; 462 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_edx; 463 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_ecx; 464 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_eax; 465 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_eip; 466 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 467 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags; 468 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp; 469 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss; 470 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err; 471 frame.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 472 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 473 474 if (copyout(&frame, fp, sizeof(frame)) != 0) { 475 /* 476 * Process has trashed its stack; give it an illegal 477 * instruction to halt it in its tracks. 478 */ 479 PROC_LOCK(p); 480 sigexit(td, SIGILL); 481 } 482 483 /* Build context to run handler in. */ 484 regs->tf_esp = (int)fp; 485 regs->tf_eip = linux_rt_sigcode; 486 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D); 487 regs->tf_cs = _ucodesel; 488 regs->tf_ds = _udatasel; 489 regs->tf_es = _udatasel; 490 regs->tf_fs = _udatasel; 491 regs->tf_ss = _udatasel; 492 PROC_LOCK(p); 493 mtx_lock(&psp->ps_mtx); 494 } 495 496 /* 497 * Send an interrupt to process. 498 * 499 * Stack is set up to allow sigcode stored 500 * in u. to call routine, followed by kcall 501 * to sigreturn routine below. After sigreturn 502 * resets the signal mask, the stack, and the 503 * frame pointer, it returns to the user 504 * specified pc, psl. 505 */ 506 static void 507 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 508 { 509 struct thread *td = curthread; 510 struct proc *p = td->td_proc; 511 struct sigacts *psp; 512 struct trapframe *regs; 513 struct l_sigframe *fp, frame; 514 l_sigset_t lmask; 515 int sig, code; 516 int oonstack; 517 518 PROC_LOCK_ASSERT(p, MA_OWNED); 519 psp = p->p_sigacts; 520 sig = ksi->ksi_signo; 521 code = ksi->ksi_code; 522 mtx_assert(&psp->ps_mtx, MA_OWNED); 523 if (SIGISMEMBER(psp->ps_siginfo, sig)) { 524 /* Signal handler installed with SA_SIGINFO. */ 525 linux_rt_sendsig(catcher, ksi, mask); 526 return; 527 } 528 regs = td->td_frame; 529 oonstack = sigonstack(regs->tf_esp); 530 531 /* Allocate space for the signal handler context. */ 532 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 533 SIGISMEMBER(psp->ps_sigonstack, sig)) { 534 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 535 td->td_sigstk.ss_size - sizeof(struct l_sigframe)); 536 } else 537 fp = (struct l_sigframe *)regs->tf_esp - 1; 538 mtx_unlock(&psp->ps_mtx); 539 PROC_UNLOCK(p); 540 541 /* Build the argument list for the signal handler. */ 542 sig = bsd_to_linux_signal(sig); 543 544 bzero(&frame, sizeof(frame)); 545 546 frame.sf_handler = catcher; 547 frame.sf_sig = sig; 548 549 bsd_to_linux_sigset(mask, &lmask); 550 551 /* Build the signal context to be used by sigreturn. */ 552 frame.sf_sc.sc_mask = lmask.__mask; 553 frame.sf_sc.sc_gs = rgs(); 554 frame.sf_sc.sc_fs = regs->tf_fs; 555 frame.sf_sc.sc_es = regs->tf_es; 556 frame.sf_sc.sc_ds = regs->tf_ds; 557 frame.sf_sc.sc_edi = regs->tf_edi; 558 frame.sf_sc.sc_esi = regs->tf_esi; 559 frame.sf_sc.sc_ebp = regs->tf_ebp; 560 frame.sf_sc.sc_ebx = regs->tf_ebx; 561 frame.sf_sc.sc_esp = regs->tf_esp; 562 frame.sf_sc.sc_edx = regs->tf_edx; 563 frame.sf_sc.sc_ecx = regs->tf_ecx; 564 frame.sf_sc.sc_eax = regs->tf_eax; 565 frame.sf_sc.sc_eip = regs->tf_eip; 566 frame.sf_sc.sc_cs = regs->tf_cs; 567 frame.sf_sc.sc_eflags = regs->tf_eflags; 568 frame.sf_sc.sc_esp_at_signal = regs->tf_esp; 569 frame.sf_sc.sc_ss = regs->tf_ss; 570 frame.sf_sc.sc_err = regs->tf_err; 571 frame.sf_sc.sc_cr2 = (register_t)ksi->ksi_addr; 572 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno); 573 574 frame.sf_extramask[0] = lmask.__mask; 575 576 if (copyout(&frame, fp, sizeof(frame)) != 0) { 577 /* 578 * Process has trashed its stack; give it an illegal 579 * instruction to halt it in its tracks. 580 */ 581 PROC_LOCK(p); 582 sigexit(td, SIGILL); 583 } 584 585 /* Build context to run handler in. */ 586 regs->tf_esp = (int)fp; 587 regs->tf_eip = linux_sigcode; 588 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D); 589 regs->tf_cs = _ucodesel; 590 regs->tf_ds = _udatasel; 591 regs->tf_es = _udatasel; 592 regs->tf_fs = _udatasel; 593 regs->tf_ss = _udatasel; 594 PROC_LOCK(p); 595 mtx_lock(&psp->ps_mtx); 596 } 597 598 /* 599 * System call to cleanup state after a signal 600 * has been taken. Reset signal mask and 601 * stack state from context left by sendsig (above). 602 * Return to previous pc and psl as specified by 603 * context left by sendsig. Check carefully to 604 * make sure that the user has not modified the 605 * psl to gain improper privileges or to cause 606 * a machine fault. 607 */ 608 int 609 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) 610 { 611 struct l_sigframe frame; 612 struct trapframe *regs; 613 l_sigset_t lmask; 614 sigset_t bmask; 615 int eflags; 616 ksiginfo_t ksi; 617 618 regs = td->td_frame; 619 620 /* 621 * The trampoline code hands us the sigframe. 622 * It is unsafe to keep track of it ourselves, in the event that a 623 * program jumps out of a signal handler. 624 */ 625 if (copyin(args->sfp, &frame, sizeof(frame)) != 0) 626 return (EFAULT); 627 628 /* Check for security violations. */ 629 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 630 eflags = frame.sf_sc.sc_eflags; 631 if (!EFLAGS_SECURE(eflags, regs->tf_eflags)) 632 return (EINVAL); 633 634 /* 635 * Don't allow users to load a valid privileged %cs. Let the 636 * hardware check for invalid selectors, excess privilege in 637 * other selectors, invalid %eip's and invalid %esp's. 638 */ 639 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 640 if (!CS_SECURE(frame.sf_sc.sc_cs)) { 641 ksiginfo_init_trap(&ksi); 642 ksi.ksi_signo = SIGBUS; 643 ksi.ksi_code = BUS_OBJERR; 644 ksi.ksi_trapno = T_PROTFLT; 645 ksi.ksi_addr = (void *)regs->tf_eip; 646 trapsignal(td, &ksi); 647 return (EINVAL); 648 } 649 650 lmask.__mask = frame.sf_sc.sc_mask; 651 linux_to_bsd_sigset(&lmask, &bmask); 652 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 653 654 /* Restore signal context. */ 655 /* %gs was restored by the trampoline. */ 656 regs->tf_fs = frame.sf_sc.sc_fs; 657 regs->tf_es = frame.sf_sc.sc_es; 658 regs->tf_ds = frame.sf_sc.sc_ds; 659 regs->tf_edi = frame.sf_sc.sc_edi; 660 regs->tf_esi = frame.sf_sc.sc_esi; 661 regs->tf_ebp = frame.sf_sc.sc_ebp; 662 regs->tf_ebx = frame.sf_sc.sc_ebx; 663 regs->tf_edx = frame.sf_sc.sc_edx; 664 regs->tf_ecx = frame.sf_sc.sc_ecx; 665 regs->tf_eax = frame.sf_sc.sc_eax; 666 regs->tf_eip = frame.sf_sc.sc_eip; 667 regs->tf_cs = frame.sf_sc.sc_cs; 668 regs->tf_eflags = eflags; 669 regs->tf_esp = frame.sf_sc.sc_esp_at_signal; 670 regs->tf_ss = frame.sf_sc.sc_ss; 671 672 return (EJUSTRETURN); 673 } 674 675 /* 676 * System call to cleanup state after a signal 677 * has been taken. Reset signal mask and 678 * stack state from context left by rt_sendsig (above). 679 * Return to previous pc and psl as specified by 680 * context left by sendsig. Check carefully to 681 * make sure that the user has not modified the 682 * psl to gain improper privileges or to cause 683 * a machine fault. 684 */ 685 int 686 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 687 { 688 struct l_ucontext uc; 689 struct l_sigcontext *context; 690 sigset_t bmask; 691 l_stack_t *lss; 692 stack_t ss; 693 struct trapframe *regs; 694 int eflags; 695 ksiginfo_t ksi; 696 697 regs = td->td_frame; 698 699 /* 700 * The trampoline code hands us the ucontext. 701 * It is unsafe to keep track of it ourselves, in the event that a 702 * program jumps out of a signal handler. 703 */ 704 if (copyin(args->ucp, &uc, sizeof(uc)) != 0) 705 return (EFAULT); 706 707 context = &uc.uc_mcontext; 708 709 /* Check for security violations. */ 710 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 711 eflags = context->sc_eflags; 712 if (!EFLAGS_SECURE(eflags, regs->tf_eflags)) 713 return (EINVAL); 714 715 /* 716 * Don't allow users to load a valid privileged %cs. Let the 717 * hardware check for invalid selectors, excess privilege in 718 * other selectors, invalid %eip's and invalid %esp's. 719 */ 720 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 721 if (!CS_SECURE(context->sc_cs)) { 722 ksiginfo_init_trap(&ksi); 723 ksi.ksi_signo = SIGBUS; 724 ksi.ksi_code = BUS_OBJERR; 725 ksi.ksi_trapno = T_PROTFLT; 726 ksi.ksi_addr = (void *)regs->tf_eip; 727 trapsignal(td, &ksi); 728 return (EINVAL); 729 } 730 731 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); 732 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 733 734 /* Restore signal context. */ 735 /* %gs was restored by the trampoline. */ 736 regs->tf_fs = context->sc_fs; 737 regs->tf_es = context->sc_es; 738 regs->tf_ds = context->sc_ds; 739 regs->tf_edi = context->sc_edi; 740 regs->tf_esi = context->sc_esi; 741 regs->tf_ebp = context->sc_ebp; 742 regs->tf_ebx = context->sc_ebx; 743 regs->tf_edx = context->sc_edx; 744 regs->tf_ecx = context->sc_ecx; 745 regs->tf_eax = context->sc_eax; 746 regs->tf_eip = context->sc_eip; 747 regs->tf_cs = context->sc_cs; 748 regs->tf_eflags = eflags; 749 regs->tf_esp = context->sc_esp_at_signal; 750 regs->tf_ss = context->sc_ss; 751 752 /* Call sigaltstack & ignore results. */ 753 lss = &uc.uc_stack; 754 ss.ss_sp = lss->ss_sp; 755 ss.ss_size = lss->ss_size; 756 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags); 757 758 (void)kern_sigaltstack(td, &ss, NULL); 759 760 return (EJUSTRETURN); 761 } 762 763 static int 764 linux_fetch_syscall_args(struct thread *td) 765 { 766 struct proc *p; 767 struct trapframe *frame; 768 struct syscall_args *sa; 769 770 p = td->td_proc; 771 frame = td->td_frame; 772 sa = &td->td_sa; 773 774 sa->code = frame->tf_eax; 775 sa->args[0] = frame->tf_ebx; 776 sa->args[1] = frame->tf_ecx; 777 sa->args[2] = frame->tf_edx; 778 sa->args[3] = frame->tf_esi; 779 sa->args[4] = frame->tf_edi; 780 sa->args[5] = frame->tf_ebp; /* Unconfirmed */ 781 782 if (sa->code >= p->p_sysent->sv_size) 783 /* nosys */ 784 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 785 else 786 sa->callp = &p->p_sysent->sv_table[sa->code]; 787 788 td->td_retval[0] = 0; 789 td->td_retval[1] = frame->tf_edx; 790 791 return (0); 792 } 793 794 static void 795 linux_set_syscall_retval(struct thread *td, int error) 796 { 797 struct trapframe *frame = td->td_frame; 798 799 cpu_set_syscall_retval(td, error); 800 801 if (__predict_false(error != 0)) { 802 if (error != ERESTART && error != EJUSTRETURN) 803 frame->tf_eax = bsd_to_linux_errno(error); 804 } 805 } 806 807 /* 808 * exec_setregs may initialize some registers differently than Linux 809 * does, thus potentially confusing Linux binaries. If necessary, we 810 * override the exec_setregs default(s) here. 811 */ 812 static void 813 linux_exec_setregs(struct thread *td, struct image_params *imgp, 814 uintptr_t stack) 815 { 816 struct pcb *pcb = td->td_pcb; 817 818 exec_setregs(td, imgp, stack); 819 820 /* Linux sets %gs to 0, we default to _udatasel. */ 821 pcb->pcb_gs = 0; 822 load_gs(0); 823 824 pcb->pcb_initial_npxcw = __LINUX_NPXCW__; 825 } 826 827 static void 828 linux_get_machine(const char **dst) 829 { 830 831 switch (cpu_class) { 832 case CPUCLASS_686: 833 *dst = "i686"; 834 break; 835 case CPUCLASS_586: 836 *dst = "i586"; 837 break; 838 case CPUCLASS_486: 839 *dst = "i486"; 840 break; 841 default: 842 *dst = "i386"; 843 } 844 } 845 846 struct sysentvec linux_sysvec = { 847 .sv_size = LINUX_SYS_MAXSYSCALL, 848 .sv_table = linux_sysent, 849 .sv_transtrap = linux_translate_traps, 850 .sv_fixup = linux_fixup, 851 .sv_sendsig = linux_sendsig, 852 .sv_sigcode = &_binary_linux_locore_o_start, 853 .sv_szsigcode = &linux_szsigcode, 854 .sv_name = "Linux a.out", 855 .sv_coredump = NULL, 856 .sv_imgact_try = linux_exec_imgact_try, 857 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 858 .sv_minuser = VM_MIN_ADDRESS, 859 .sv_maxuser = VM_MAXUSER_ADDRESS, 860 .sv_usrstack = LINUX_USRSTACK, 861 .sv_psstrings = PS_STRINGS, 862 .sv_stackprot = VM_PROT_ALL, 863 .sv_copyout_strings = exec_copyout_strings, 864 .sv_setregs = linux_exec_setregs, 865 .sv_fixlimit = NULL, 866 .sv_maxssiz = NULL, 867 .sv_flags = SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32, 868 .sv_set_syscall_retval = linux_set_syscall_retval, 869 .sv_fetch_syscall_args = linux_fetch_syscall_args, 870 .sv_syscallnames = NULL, 871 .sv_shared_page_base = LINUX_SHAREDPAGE, 872 .sv_shared_page_len = PAGE_SIZE, 873 .sv_schedtail = linux_schedtail, 874 .sv_thread_detach = linux_thread_detach, 875 .sv_trap = NULL, 876 }; 877 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec); 878 879 struct sysentvec elf_linux_sysvec = { 880 .sv_size = LINUX_SYS_MAXSYSCALL, 881 .sv_table = linux_sysent, 882 .sv_transtrap = linux_translate_traps, 883 .sv_fixup = linux_fixup_elf, 884 .sv_sendsig = linux_sendsig, 885 .sv_sigcode = &_binary_linux_locore_o_start, 886 .sv_szsigcode = &linux_szsigcode, 887 .sv_name = "Linux ELF", 888 .sv_coredump = elf32_coredump, 889 .sv_imgact_try = linux_exec_imgact_try, 890 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 891 .sv_minuser = VM_MIN_ADDRESS, 892 .sv_maxuser = VM_MAXUSER_ADDRESS, 893 .sv_usrstack = LINUX_USRSTACK, 894 .sv_psstrings = LINUX_PS_STRINGS, 895 .sv_stackprot = VM_PROT_ALL, 896 .sv_copyout_auxargs = linux_copyout_auxargs, 897 .sv_copyout_strings = linux_copyout_strings, 898 .sv_setregs = linux_exec_setregs, 899 .sv_fixlimit = NULL, 900 .sv_maxssiz = NULL, 901 .sv_flags = SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP, 902 .sv_set_syscall_retval = linux_set_syscall_retval, 903 .sv_fetch_syscall_args = linux_fetch_syscall_args, 904 .sv_syscallnames = NULL, 905 .sv_shared_page_base = LINUX_SHAREDPAGE, 906 .sv_shared_page_len = PAGE_SIZE, 907 .sv_schedtail = linux_schedtail, 908 .sv_thread_detach = linux_thread_detach, 909 .sv_trap = NULL, 910 }; 911 912 static void 913 linux_vdso_install(void *param) 914 { 915 916 linux_szsigcode = (&_binary_linux_locore_o_end - 917 &_binary_linux_locore_o_start); 918 919 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 920 panic("Linux invalid vdso size\n"); 921 922 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 923 924 linux_shared_page_obj = __elfN(linux_shared_page_init) 925 (&linux_shared_page_mapping); 926 927 __elfN(linux_vdso_reloc)(&elf_linux_sysvec); 928 929 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 930 linux_szsigcode); 931 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 932 } 933 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 934 linux_vdso_install, NULL); 935 936 static void 937 linux_vdso_deinstall(void *param) 938 { 939 940 __elfN(linux_shared_page_fini)(linux_shared_page_obj); 941 } 942 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 943 linux_vdso_deinstall, NULL); 944 945 static char GNU_ABI_VENDOR[] = "GNU"; 946 static int GNULINUX_ABI_DESC = 0; 947 948 static bool 949 linux_trans_osrel(const Elf_Note *note, int32_t *osrel) 950 { 951 const Elf32_Word *desc; 952 uintptr_t p; 953 954 p = (uintptr_t)(note + 1); 955 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 956 957 desc = (const Elf32_Word *)p; 958 if (desc[0] != GNULINUX_ABI_DESC) 959 return (false); 960 961 /* 962 * For Linux we encode osrel using the Linux convention of 963 * (version << 16) | (major << 8) | (minor) 964 * See macro in linux_mib.h 965 */ 966 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 967 968 return (true); 969 } 970 971 static Elf_Brandnote linux_brandnote = { 972 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 973 .hdr.n_descsz = 16, /* XXX at least 16 */ 974 .hdr.n_type = 1, 975 .vendor = GNU_ABI_VENDOR, 976 .flags = BN_TRANSLATE_OSREL, 977 .trans_osrel = linux_trans_osrel 978 }; 979 980 static Elf32_Brandinfo linux_brand = { 981 .brand = ELFOSABI_LINUX, 982 .machine = EM_386, 983 .compat_3_brand = "Linux", 984 .emul_path = linux_emul_path, 985 .interp_path = "/lib/ld-linux.so.1", 986 .sysvec = &elf_linux_sysvec, 987 .interp_newpath = NULL, 988 .brand_note = &linux_brandnote, 989 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 990 }; 991 992 static Elf32_Brandinfo linux_glibc2brand = { 993 .brand = ELFOSABI_LINUX, 994 .machine = EM_386, 995 .compat_3_brand = "Linux", 996 .emul_path = linux_emul_path, 997 .interp_path = "/lib/ld-linux.so.2", 998 .sysvec = &elf_linux_sysvec, 999 .interp_newpath = NULL, 1000 .brand_note = &linux_brandnote, 1001 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1002 }; 1003 1004 static Elf32_Brandinfo linux_muslbrand = { 1005 .brand = ELFOSABI_LINUX, 1006 .machine = EM_386, 1007 .compat_3_brand = "Linux", 1008 .emul_path = linux_emul_path, 1009 .interp_path = "/lib/ld-musl-i386.so.1", 1010 .sysvec = &elf_linux_sysvec, 1011 .interp_newpath = NULL, 1012 .brand_note = &linux_brandnote, 1013 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1014 }; 1015 1016 Elf32_Brandinfo *linux_brandlist[] = { 1017 &linux_brand, 1018 &linux_glibc2brand, 1019 &linux_muslbrand, 1020 NULL 1021 }; 1022 1023 static int 1024 linux_elf_modevent(module_t mod, int type, void *data) 1025 { 1026 Elf32_Brandinfo **brandinfo; 1027 int error; 1028 struct linux_ioctl_handler **lihp; 1029 1030 error = 0; 1031 1032 switch(type) { 1033 case MOD_LOAD: 1034 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1035 ++brandinfo) 1036 if (elf32_insert_brand_entry(*brandinfo) < 0) 1037 error = EINVAL; 1038 if (error == 0) { 1039 SET_FOREACH(lihp, linux_ioctl_handler_set) 1040 linux_ioctl_register_handler(*lihp); 1041 LIST_INIT(&futex_list); 1042 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF); 1043 linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit, 1044 NULL, 1000); 1045 linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec, 1046 NULL, 1000); 1047 linux_thread_dtor_tag = EVENTHANDLER_REGISTER(thread_dtor, 1048 linux_thread_dtor, NULL, EVENTHANDLER_PRI_ANY); 1049 linux_get_machine(&linux_kplatform); 1050 linux_szplatform = roundup(strlen(linux_kplatform) + 1, 1051 sizeof(char *)); 1052 linux_dev_shm_create(); 1053 linux_osd_jail_register(); 1054 stclohz = (stathz ? stathz : hz); 1055 if (bootverbose) 1056 printf("Linux ELF exec handler installed\n"); 1057 } else 1058 printf("cannot insert Linux ELF brand handler\n"); 1059 break; 1060 case MOD_UNLOAD: 1061 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1062 ++brandinfo) 1063 if (elf32_brand_inuse(*brandinfo)) 1064 error = EBUSY; 1065 if (error == 0) { 1066 for (brandinfo = &linux_brandlist[0]; 1067 *brandinfo != NULL; ++brandinfo) 1068 if (elf32_remove_brand_entry(*brandinfo) < 0) 1069 error = EINVAL; 1070 } 1071 if (error == 0) { 1072 SET_FOREACH(lihp, linux_ioctl_handler_set) 1073 linux_ioctl_unregister_handler(*lihp); 1074 mtx_destroy(&futex_mtx); 1075 EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag); 1076 EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag); 1077 EVENTHANDLER_DEREGISTER(thread_dtor, linux_thread_dtor_tag); 1078 linux_dev_shm_destroy(); 1079 linux_osd_jail_deregister(); 1080 if (bootverbose) 1081 printf("Linux ELF exec handler removed\n"); 1082 } else 1083 printf("Could not deinstall ELF interpreter entry\n"); 1084 break; 1085 default: 1086 return (EOPNOTSUPP); 1087 } 1088 return (error); 1089 } 1090 1091 static moduledata_t linux_elf_mod = { 1092 "linuxelf", 1093 linux_elf_modevent, 1094 0 1095 }; 1096 1097 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 1098 FEATURE(linux, "Linux 32bit support"); 1099