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