1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2004 Tim J. Robbins 5 * Copyright (c) 2003 Peter Wemm 6 * Copyright (c) 2002 Doug Rabson 7 * Copyright (c) 1998-1999 Andrew Gallatin 8 * Copyright (c) 1994-1996 Søren Schmidt 9 * All rights reserved. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer 16 * in this position and unchanged. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. The name of the author may not be used to endorse or promote products 21 * derived from this software without specific prior written permission 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 24 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 25 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 26 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 27 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 28 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 32 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 37 #include "opt_compat.h" 38 39 #ifndef COMPAT_FREEBSD32 40 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!" 41 #endif 42 43 #define __ELF_WORD_SIZE 32 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/exec.h> 48 #include <sys/fcntl.h> 49 #include <sys/imgact.h> 50 #include <sys/imgact_elf.h> 51 #include <sys/kernel.h> 52 #include <sys/lock.h> 53 #include <sys/malloc.h> 54 #include <sys/module.h> 55 #include <sys/mutex.h> 56 #include <sys/proc.h> 57 #include <sys/resourcevar.h> 58 #include <sys/signalvar.h> 59 #include <sys/sysctl.h> 60 #include <sys/syscallsubr.h> 61 #include <sys/sysent.h> 62 #include <sys/sysproto.h> 63 #include <sys/vnode.h> 64 #include <sys/eventhandler.h> 65 66 #include <vm/vm.h> 67 #include <vm/pmap.h> 68 #include <vm/vm_extern.h> 69 #include <vm/vm_map.h> 70 #include <vm/vm_object.h> 71 #include <vm/vm_page.h> 72 #include <vm/vm_param.h> 73 74 #include <machine/cpu.h> 75 #include <machine/md_var.h> 76 #include <machine/pcb.h> 77 #include <machine/specialreg.h> 78 79 #include <amd64/linux32/linux.h> 80 #include <amd64/linux32/linux32_proto.h> 81 #include <compat/linux/linux_emul.h> 82 #include <compat/linux/linux_futex.h> 83 #include <compat/linux/linux_ioctl.h> 84 #include <compat/linux/linux_mib.h> 85 #include <compat/linux/linux_misc.h> 86 #include <compat/linux/linux_signal.h> 87 #include <compat/linux/linux_util.h> 88 #include <compat/linux/linux_vdso.h> 89 90 MODULE_VERSION(linux, 1); 91 92 #define AUXARGS_ENTRY_32(pos, id, val) \ 93 do { \ 94 suword32(pos++, id); \ 95 suword32(pos++, val); \ 96 } while (0) 97 98 #if BYTE_ORDER == LITTLE_ENDIAN 99 #define SHELLMAGIC 0x2123 /* #! */ 100 #else 101 #define SHELLMAGIC 0x2321 102 #endif 103 104 /* 105 * Allow the sendsig functions to use the ldebug() facility 106 * even though they are not syscalls themselves. Map them 107 * to syscall 0. This is slightly less bogus than using 108 * ldebug(sigreturn). 109 */ 110 #define LINUX32_SYS_linux_rt_sendsig 0 111 #define LINUX32_SYS_linux_sendsig 0 112 113 const char *linux_kplatform; 114 static int linux_szsigcode; 115 static vm_object_t linux_shared_page_obj; 116 static char *linux_shared_page_mapping; 117 extern char _binary_linux32_locore_o_start; 118 extern char _binary_linux32_locore_o_end; 119 120 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL]; 121 122 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 123 124 static int elf_linux_fixup(register_t **stack_base, 125 struct image_params *iparams); 126 static register_t *linux_copyout_strings(struct image_params *imgp); 127 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask); 128 static void exec_linux_setregs(struct thread *td, 129 struct image_params *imgp, u_long stack); 130 static void linux32_fixlimit(struct rlimit *rl, int which); 131 static boolean_t linux32_trans_osrel(const Elf_Note *note, int32_t *osrel); 132 static void linux_vdso_install(void *param); 133 static void linux_vdso_deinstall(void *param); 134 135 /* 136 * Linux syscalls return negative errno's, we do positive and map them 137 * Reference: 138 * FreeBSD: src/sys/sys/errno.h 139 * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h 140 * linux-2.6.17.8/include/asm-generic/errno.h 141 */ 142 static int bsd_to_linux_errno[ELAST + 1] = { 143 -0, -1, -2, -3, -4, -5, -6, -7, -8, -9, 144 -10, -35, -12, -13, -14, -15, -16, -17, -18, -19, 145 -20, -21, -22, -23, -24, -25, -26, -27, -28, -29, 146 -30, -31, -32, -33, -34, -11,-115,-114, -88, -89, 147 -90, -91, -92, -93, -94, -95, -96, -97, -98, -99, 148 -100,-101,-102,-103,-104,-105,-106,-107,-108,-109, 149 -110,-111, -40, -36,-112,-113, -39, -11, -87,-122, 150 -116, -66, -6, -6, -6, -6, -6, -37, -38, -9, 151 -6, -6, -43, -42, -75,-125, -84, -61, -16, -74, 152 -72, -67, -71 153 }; 154 155 #define LINUX_T_UNKNOWN 255 156 static int _bsd_to_linux_trapcode[] = { 157 LINUX_T_UNKNOWN, /* 0 */ 158 6, /* 1 T_PRIVINFLT */ 159 LINUX_T_UNKNOWN, /* 2 */ 160 3, /* 3 T_BPTFLT */ 161 LINUX_T_UNKNOWN, /* 4 */ 162 LINUX_T_UNKNOWN, /* 5 */ 163 16, /* 6 T_ARITHTRAP */ 164 254, /* 7 T_ASTFLT */ 165 LINUX_T_UNKNOWN, /* 8 */ 166 13, /* 9 T_PROTFLT */ 167 1, /* 10 T_TRCTRAP */ 168 LINUX_T_UNKNOWN, /* 11 */ 169 14, /* 12 T_PAGEFLT */ 170 LINUX_T_UNKNOWN, /* 13 */ 171 17, /* 14 T_ALIGNFLT */ 172 LINUX_T_UNKNOWN, /* 15 */ 173 LINUX_T_UNKNOWN, /* 16 */ 174 LINUX_T_UNKNOWN, /* 17 */ 175 0, /* 18 T_DIVIDE */ 176 2, /* 19 T_NMI */ 177 4, /* 20 T_OFLOW */ 178 5, /* 21 T_BOUND */ 179 7, /* 22 T_DNA */ 180 8, /* 23 T_DOUBLEFLT */ 181 9, /* 24 T_FPOPFLT */ 182 10, /* 25 T_TSSFLT */ 183 11, /* 26 T_SEGNPFLT */ 184 12, /* 27 T_STKFLT */ 185 18, /* 28 T_MCHK */ 186 19, /* 29 T_XMMFLT */ 187 15 /* 30 T_RESERVED */ 188 }; 189 #define bsd_to_linux_trapcode(code) \ 190 ((code)<nitems(_bsd_to_linux_trapcode)? \ 191 _bsd_to_linux_trapcode[(code)]: \ 192 LINUX_T_UNKNOWN) 193 194 struct linux32_ps_strings { 195 u_int32_t ps_argvstr; /* first of 0 or more argument strings */ 196 u_int ps_nargvstr; /* the number of argument strings */ 197 u_int32_t ps_envstr; /* first of 0 or more environment strings */ 198 u_int ps_nenvstr; /* the number of environment strings */ 199 }; 200 201 LINUX_VDSO_SYM_INTPTR(linux32_sigcode); 202 LINUX_VDSO_SYM_INTPTR(linux32_rt_sigcode); 203 LINUX_VDSO_SYM_INTPTR(linux32_vsyscall); 204 LINUX_VDSO_SYM_CHAR(linux_platform); 205 206 /* 207 * If FreeBSD & Linux have a difference of opinion about what a trap 208 * means, deal with it here. 209 * 210 * MPSAFE 211 */ 212 static int 213 translate_traps(int signal, int trap_code) 214 { 215 if (signal != SIGBUS) 216 return signal; 217 switch (trap_code) { 218 case T_PROTFLT: 219 case T_TSSFLT: 220 case T_DOUBLEFLT: 221 case T_PAGEFLT: 222 return SIGSEGV; 223 default: 224 return signal; 225 } 226 } 227 228 static int 229 elf_linux_fixup(register_t **stack_base, struct image_params *imgp) 230 { 231 Elf32_Auxargs *args; 232 Elf32_Addr *base; 233 Elf32_Addr *pos; 234 struct linux32_ps_strings *arginfo; 235 int issetugid; 236 237 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; 238 239 KASSERT(curthread->td_proc == imgp->proc, 240 ("unsafe elf_linux_fixup(), should be curproc")); 241 base = (Elf32_Addr *)*stack_base; 242 args = (Elf32_Auxargs *)imgp->auxargs; 243 pos = base + (imgp->args->argc + imgp->args->envc + 2); 244 245 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0; 246 AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO_EHDR, 247 imgp->proc->p_sysent->sv_shared_page_base); 248 AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO, linux32_vsyscall); 249 AUXARGS_ENTRY_32(pos, LINUX_AT_HWCAP, cpu_feature); 250 251 /* 252 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, 253 * as it has appeared in the 2.4.0-rc7 first time. 254 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), 255 * glibc falls back to the hard-coded CLK_TCK value when aux entry 256 * is not present. 257 * Also see linux_times() implementation. 258 */ 259 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) 260 AUXARGS_ENTRY_32(pos, LINUX_AT_CLKTCK, stclohz); 261 AUXARGS_ENTRY_32(pos, AT_PHDR, args->phdr); 262 AUXARGS_ENTRY_32(pos, AT_PHENT, args->phent); 263 AUXARGS_ENTRY_32(pos, AT_PHNUM, args->phnum); 264 AUXARGS_ENTRY_32(pos, AT_PAGESZ, args->pagesz); 265 AUXARGS_ENTRY_32(pos, AT_FLAGS, args->flags); 266 AUXARGS_ENTRY_32(pos, AT_ENTRY, args->entry); 267 AUXARGS_ENTRY_32(pos, AT_BASE, args->base); 268 AUXARGS_ENTRY_32(pos, LINUX_AT_SECURE, issetugid); 269 AUXARGS_ENTRY_32(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 270 AUXARGS_ENTRY_32(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 271 AUXARGS_ENTRY_32(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 272 AUXARGS_ENTRY_32(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 273 AUXARGS_ENTRY_32(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 274 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary)); 275 if (imgp->execpathp != 0) 276 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp)); 277 if (args->execfd != -1) 278 AUXARGS_ENTRY_32(pos, AT_EXECFD, args->execfd); 279 AUXARGS_ENTRY_32(pos, AT_NULL, 0); 280 281 free(imgp->auxargs, M_TEMP); 282 imgp->auxargs = NULL; 283 284 base--; 285 suword32(base, (uint32_t)imgp->args->argc); 286 *stack_base = (register_t *)base; 287 return (0); 288 } 289 290 static void 291 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 292 { 293 struct thread *td = curthread; 294 struct proc *p = td->td_proc; 295 struct sigacts *psp; 296 struct trapframe *regs; 297 struct l_rt_sigframe *fp, frame; 298 int oonstack; 299 int sig; 300 int code; 301 302 sig = ksi->ksi_signo; 303 code = ksi->ksi_code; 304 PROC_LOCK_ASSERT(p, MA_OWNED); 305 psp = p->p_sigacts; 306 mtx_assert(&psp->ps_mtx, MA_OWNED); 307 regs = td->td_frame; 308 oonstack = sigonstack(regs->tf_rsp); 309 310 #ifdef DEBUG 311 if (ldebug(rt_sendsig)) 312 printf(ARGS(rt_sendsig, "%p, %d, %p, %u"), 313 catcher, sig, (void*)mask, code); 314 #endif 315 /* 316 * Allocate space for the signal handler context. 317 */ 318 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 319 SIGISMEMBER(psp->ps_sigonstack, sig)) { 320 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 321 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe)); 322 } else 323 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1; 324 mtx_unlock(&psp->ps_mtx); 325 326 /* 327 * Build the argument list for the signal handler. 328 */ 329 sig = bsd_to_linux_signal(sig); 330 331 bzero(&frame, sizeof(frame)); 332 333 frame.sf_handler = PTROUT(catcher); 334 frame.sf_sig = sig; 335 frame.sf_siginfo = PTROUT(&fp->sf_si); 336 frame.sf_ucontext = PTROUT(&fp->sf_sc); 337 338 /* Fill in POSIX parts */ 339 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig); 340 341 /* 342 * Build the signal context to be used by sigreturn 343 * and libgcc unwind. 344 */ 345 frame.sf_sc.uc_flags = 0; /* XXX ??? */ 346 frame.sf_sc.uc_link = 0; /* XXX ??? */ 347 348 frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 349 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 350 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 351 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 352 PROC_UNLOCK(p); 353 354 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask); 355 356 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask; 357 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi; 358 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi; 359 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp; 360 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx; 361 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp; 362 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx; 363 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx; 364 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax; 365 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip; 366 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 367 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs; 368 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs; 369 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es; 370 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds; 371 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags; 372 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp; 373 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss; 374 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err; 375 frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 376 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 377 378 #ifdef DEBUG 379 if (ldebug(rt_sendsig)) 380 printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"), 381 frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp, 382 td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask); 383 #endif 384 385 if (copyout(&frame, fp, sizeof(frame)) != 0) { 386 /* 387 * Process has trashed its stack; give it an illegal 388 * instruction to halt it in its tracks. 389 */ 390 #ifdef DEBUG 391 if (ldebug(rt_sendsig)) 392 printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"), 393 fp, oonstack); 394 #endif 395 PROC_LOCK(p); 396 sigexit(td, SIGILL); 397 } 398 399 /* 400 * Build context to run handler in. 401 */ 402 regs->tf_rsp = PTROUT(fp); 403 regs->tf_rip = linux32_rt_sigcode; 404 regs->tf_rflags &= ~(PSL_T | PSL_D); 405 regs->tf_cs = _ucode32sel; 406 regs->tf_ss = _udatasel; 407 regs->tf_ds = _udatasel; 408 regs->tf_es = _udatasel; 409 regs->tf_fs = _ufssel; 410 regs->tf_gs = _ugssel; 411 regs->tf_flags = TF_HASSEGS; 412 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 413 PROC_LOCK(p); 414 mtx_lock(&psp->ps_mtx); 415 } 416 417 418 /* 419 * Send an interrupt to process. 420 * 421 * Stack is set up to allow sigcode stored 422 * in u. to call routine, followed by kcall 423 * to sigreturn routine below. After sigreturn 424 * resets the signal mask, the stack, and the 425 * frame pointer, it returns to the user 426 * specified pc, psl. 427 */ 428 static void 429 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 430 { 431 struct thread *td = curthread; 432 struct proc *p = td->td_proc; 433 struct sigacts *psp; 434 struct trapframe *regs; 435 struct l_sigframe *fp, frame; 436 l_sigset_t lmask; 437 int oonstack; 438 int sig, code; 439 440 sig = ksi->ksi_signo; 441 code = ksi->ksi_code; 442 PROC_LOCK_ASSERT(p, MA_OWNED); 443 psp = p->p_sigacts; 444 mtx_assert(&psp->ps_mtx, MA_OWNED); 445 if (SIGISMEMBER(psp->ps_siginfo, sig)) { 446 /* Signal handler installed with SA_SIGINFO. */ 447 linux_rt_sendsig(catcher, ksi, mask); 448 return; 449 } 450 451 regs = td->td_frame; 452 oonstack = sigonstack(regs->tf_rsp); 453 454 #ifdef DEBUG 455 if (ldebug(sendsig)) 456 printf(ARGS(sendsig, "%p, %d, %p, %u"), 457 catcher, sig, (void*)mask, code); 458 #endif 459 460 /* 461 * Allocate space for the signal handler context. 462 */ 463 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 464 SIGISMEMBER(psp->ps_sigonstack, sig)) { 465 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 466 td->td_sigstk.ss_size - sizeof(struct l_sigframe)); 467 } else 468 fp = (struct l_sigframe *)regs->tf_rsp - 1; 469 mtx_unlock(&psp->ps_mtx); 470 PROC_UNLOCK(p); 471 472 /* 473 * Build the argument list for the signal handler. 474 */ 475 sig = bsd_to_linux_signal(sig); 476 477 bzero(&frame, sizeof(frame)); 478 479 frame.sf_handler = PTROUT(catcher); 480 frame.sf_sig = sig; 481 482 bsd_to_linux_sigset(mask, &lmask); 483 484 /* 485 * Build the signal context to be used by sigreturn. 486 */ 487 frame.sf_sc.sc_mask = lmask.__mask; 488 frame.sf_sc.sc_gs = regs->tf_gs; 489 frame.sf_sc.sc_fs = regs->tf_fs; 490 frame.sf_sc.sc_es = regs->tf_es; 491 frame.sf_sc.sc_ds = regs->tf_ds; 492 frame.sf_sc.sc_edi = regs->tf_rdi; 493 frame.sf_sc.sc_esi = regs->tf_rsi; 494 frame.sf_sc.sc_ebp = regs->tf_rbp; 495 frame.sf_sc.sc_ebx = regs->tf_rbx; 496 frame.sf_sc.sc_esp = regs->tf_rsp; 497 frame.sf_sc.sc_edx = regs->tf_rdx; 498 frame.sf_sc.sc_ecx = regs->tf_rcx; 499 frame.sf_sc.sc_eax = regs->tf_rax; 500 frame.sf_sc.sc_eip = regs->tf_rip; 501 frame.sf_sc.sc_cs = regs->tf_cs; 502 frame.sf_sc.sc_eflags = regs->tf_rflags; 503 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp; 504 frame.sf_sc.sc_ss = regs->tf_ss; 505 frame.sf_sc.sc_err = regs->tf_err; 506 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 507 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code); 508 509 frame.sf_extramask[0] = lmask.__mask; 510 511 if (copyout(&frame, fp, sizeof(frame)) != 0) { 512 /* 513 * Process has trashed its stack; give it an illegal 514 * instruction to halt it in its tracks. 515 */ 516 PROC_LOCK(p); 517 sigexit(td, SIGILL); 518 } 519 520 /* 521 * Build context to run handler in. 522 */ 523 regs->tf_rsp = PTROUT(fp); 524 regs->tf_rip = linux32_sigcode; 525 regs->tf_rflags &= ~(PSL_T | PSL_D); 526 regs->tf_cs = _ucode32sel; 527 regs->tf_ss = _udatasel; 528 regs->tf_ds = _udatasel; 529 regs->tf_es = _udatasel; 530 regs->tf_fs = _ufssel; 531 regs->tf_gs = _ugssel; 532 regs->tf_flags = TF_HASSEGS; 533 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 534 PROC_LOCK(p); 535 mtx_lock(&psp->ps_mtx); 536 } 537 538 /* 539 * System call to cleanup state after a signal 540 * has been taken. Reset signal mask and 541 * stack state from context left by sendsig (above). 542 * Return to previous pc and psl as specified by 543 * context left by sendsig. Check carefully to 544 * make sure that the user has not modified the 545 * psl to gain improper privileges or to cause 546 * a machine fault. 547 */ 548 int 549 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) 550 { 551 struct l_sigframe frame; 552 struct trapframe *regs; 553 sigset_t bmask; 554 l_sigset_t lmask; 555 int eflags; 556 ksiginfo_t ksi; 557 558 regs = td->td_frame; 559 560 #ifdef DEBUG 561 if (ldebug(sigreturn)) 562 printf(ARGS(sigreturn, "%p"), (void *)args->sfp); 563 #endif 564 /* 565 * The trampoline code hands us the sigframe. 566 * It is unsafe to keep track of it ourselves, in the event that a 567 * program jumps out of a signal handler. 568 */ 569 if (copyin(args->sfp, &frame, sizeof(frame)) != 0) 570 return (EFAULT); 571 572 /* 573 * Check for security violations. 574 */ 575 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 576 eflags = frame.sf_sc.sc_eflags; 577 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 578 return(EINVAL); 579 580 /* 581 * Don't allow users to load a valid privileged %cs. Let the 582 * hardware check for invalid selectors, excess privilege in 583 * other selectors, invalid %eip's and invalid %esp's. 584 */ 585 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 586 if (!CS_SECURE(frame.sf_sc.sc_cs)) { 587 ksiginfo_init_trap(&ksi); 588 ksi.ksi_signo = SIGBUS; 589 ksi.ksi_code = BUS_OBJERR; 590 ksi.ksi_trapno = T_PROTFLT; 591 ksi.ksi_addr = (void *)regs->tf_rip; 592 trapsignal(td, &ksi); 593 return(EINVAL); 594 } 595 596 lmask.__mask = frame.sf_sc.sc_mask; 597 lmask.__mask = frame.sf_extramask[0]; 598 linux_to_bsd_sigset(&lmask, &bmask); 599 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 600 601 /* 602 * Restore signal context. 603 */ 604 regs->tf_rdi = frame.sf_sc.sc_edi; 605 regs->tf_rsi = frame.sf_sc.sc_esi; 606 regs->tf_rbp = frame.sf_sc.sc_ebp; 607 regs->tf_rbx = frame.sf_sc.sc_ebx; 608 regs->tf_rdx = frame.sf_sc.sc_edx; 609 regs->tf_rcx = frame.sf_sc.sc_ecx; 610 regs->tf_rax = frame.sf_sc.sc_eax; 611 regs->tf_rip = frame.sf_sc.sc_eip; 612 regs->tf_cs = frame.sf_sc.sc_cs; 613 regs->tf_ds = frame.sf_sc.sc_ds; 614 regs->tf_es = frame.sf_sc.sc_es; 615 regs->tf_fs = frame.sf_sc.sc_fs; 616 regs->tf_gs = frame.sf_sc.sc_gs; 617 regs->tf_rflags = eflags; 618 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal; 619 regs->tf_ss = frame.sf_sc.sc_ss; 620 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 621 622 return (EJUSTRETURN); 623 } 624 625 /* 626 * System call to cleanup state after a signal 627 * has been taken. Reset signal mask and 628 * stack state from context left by rt_sendsig (above). 629 * Return to previous pc and psl as specified by 630 * context left by sendsig. Check carefully to 631 * make sure that the user has not modified the 632 * psl to gain improper privileges or to cause 633 * a machine fault. 634 */ 635 int 636 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 637 { 638 struct l_ucontext uc; 639 struct l_sigcontext *context; 640 sigset_t bmask; 641 l_stack_t *lss; 642 stack_t ss; 643 struct trapframe *regs; 644 int eflags; 645 ksiginfo_t ksi; 646 647 regs = td->td_frame; 648 649 #ifdef DEBUG 650 if (ldebug(rt_sigreturn)) 651 printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp); 652 #endif 653 /* 654 * The trampoline code hands us the ucontext. 655 * It is unsafe to keep track of it ourselves, in the event that a 656 * program jumps out of a signal handler. 657 */ 658 if (copyin(args->ucp, &uc, sizeof(uc)) != 0) 659 return (EFAULT); 660 661 context = &uc.uc_mcontext; 662 663 /* 664 * Check for security violations. 665 */ 666 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 667 eflags = context->sc_eflags; 668 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 669 return(EINVAL); 670 671 /* 672 * Don't allow users to load a valid privileged %cs. Let the 673 * hardware check for invalid selectors, excess privilege in 674 * other selectors, invalid %eip's and invalid %esp's. 675 */ 676 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 677 if (!CS_SECURE(context->sc_cs)) { 678 ksiginfo_init_trap(&ksi); 679 ksi.ksi_signo = SIGBUS; 680 ksi.ksi_code = BUS_OBJERR; 681 ksi.ksi_trapno = T_PROTFLT; 682 ksi.ksi_addr = (void *)regs->tf_rip; 683 trapsignal(td, &ksi); 684 return(EINVAL); 685 } 686 687 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); 688 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 689 690 /* 691 * Restore signal context 692 */ 693 regs->tf_gs = context->sc_gs; 694 regs->tf_fs = context->sc_fs; 695 regs->tf_es = context->sc_es; 696 regs->tf_ds = context->sc_ds; 697 regs->tf_rdi = context->sc_edi; 698 regs->tf_rsi = context->sc_esi; 699 regs->tf_rbp = context->sc_ebp; 700 regs->tf_rbx = context->sc_ebx; 701 regs->tf_rdx = context->sc_edx; 702 regs->tf_rcx = context->sc_ecx; 703 regs->tf_rax = context->sc_eax; 704 regs->tf_rip = context->sc_eip; 705 regs->tf_cs = context->sc_cs; 706 regs->tf_rflags = eflags; 707 regs->tf_rsp = context->sc_esp_at_signal; 708 regs->tf_ss = context->sc_ss; 709 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 710 711 /* 712 * call sigaltstack & ignore results.. 713 */ 714 lss = &uc.uc_stack; 715 ss.ss_sp = PTRIN(lss->ss_sp); 716 ss.ss_size = lss->ss_size; 717 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags); 718 719 #ifdef DEBUG 720 if (ldebug(rt_sigreturn)) 721 printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"), 722 ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask); 723 #endif 724 (void)kern_sigaltstack(td, &ss, NULL); 725 726 return (EJUSTRETURN); 727 } 728 729 static int 730 linux32_fetch_syscall_args(struct thread *td) 731 { 732 struct proc *p; 733 struct trapframe *frame; 734 struct syscall_args *sa; 735 736 p = td->td_proc; 737 frame = td->td_frame; 738 sa = &td->td_sa; 739 740 sa->args[0] = frame->tf_rbx; 741 sa->args[1] = frame->tf_rcx; 742 sa->args[2] = frame->tf_rdx; 743 sa->args[3] = frame->tf_rsi; 744 sa->args[4] = frame->tf_rdi; 745 sa->args[5] = frame->tf_rbp; /* Unconfirmed */ 746 sa->code = frame->tf_rax; 747 748 if (sa->code >= p->p_sysent->sv_size) 749 /* nosys */ 750 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 751 else 752 sa->callp = &p->p_sysent->sv_table[sa->code]; 753 sa->narg = sa->callp->sy_narg; 754 755 td->td_retval[0] = 0; 756 td->td_retval[1] = frame->tf_rdx; 757 758 return (0); 759 } 760 761 /* 762 * If a linux binary is exec'ing something, try this image activator 763 * first. We override standard shell script execution in order to 764 * be able to modify the interpreter path. We only do this if a linux 765 * binary is doing the exec, so we do not create an EXEC module for it. 766 */ 767 static int exec_linux_imgact_try(struct image_params *iparams); 768 769 static int 770 exec_linux_imgact_try(struct image_params *imgp) 771 { 772 const char *head = (const char *)imgp->image_header; 773 char *rpath; 774 int error = -1; 775 776 /* 777 * The interpreter for shell scripts run from a linux binary needs 778 * to be located in /compat/linux if possible in order to recursively 779 * maintain linux path emulation. 780 */ 781 if (((const short *)head)[0] == SHELLMAGIC) { 782 /* 783 * Run our normal shell image activator. If it succeeds attempt 784 * to use the alternate path for the interpreter. If an 785 * alternate * path is found, use our stringspace to store it. 786 */ 787 if ((error = exec_shell_imgact(imgp)) == 0) { 788 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc), 789 imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, 790 AT_FDCWD); 791 if (rpath != NULL) 792 imgp->args->fname_buf = 793 imgp->interpreter_name = rpath; 794 } 795 } 796 return (error); 797 } 798 799 /* 800 * Clear registers on exec 801 * XXX copied from ia32_signal.c. 802 */ 803 static void 804 exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack) 805 { 806 struct trapframe *regs = td->td_frame; 807 struct pcb *pcb = td->td_pcb; 808 809 if (td->td_proc->p_md.md_ldt != NULL) 810 user_ldt_free(td); 811 812 critical_enter(); 813 wrmsr(MSR_FSBASE, 0); 814 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */ 815 pcb->pcb_fsbase = 0; 816 pcb->pcb_gsbase = 0; 817 critical_exit(); 818 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 819 820 bzero((char *)regs, sizeof(struct trapframe)); 821 regs->tf_rip = imgp->entry_addr; 822 regs->tf_rsp = stack; 823 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T); 824 regs->tf_gs = _ugssel; 825 regs->tf_fs = _ufssel; 826 regs->tf_es = _udatasel; 827 regs->tf_ds = _udatasel; 828 regs->tf_ss = _udatasel; 829 regs->tf_flags = TF_HASSEGS; 830 regs->tf_cs = _ucode32sel; 831 regs->tf_rbx = imgp->ps_strings; 832 833 fpstate_drop(td); 834 835 /* Do full restore on return so that we can change to a different %cs */ 836 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET); 837 } 838 839 /* 840 * XXX copied from ia32_sysvec.c. 841 */ 842 static register_t * 843 linux_copyout_strings(struct image_params *imgp) 844 { 845 int argc, envc; 846 u_int32_t *vectp; 847 char *stringp, *destp; 848 u_int32_t *stack_base; 849 struct linux32_ps_strings *arginfo; 850 char canary[LINUX_AT_RANDOM_LEN]; 851 size_t execpath_len; 852 853 /* 854 * Calculate string base and vector table pointers. 855 */ 856 if (imgp->execpath != NULL && imgp->auxargs != NULL) 857 execpath_len = strlen(imgp->execpath) + 1; 858 else 859 execpath_len = 0; 860 861 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; 862 destp = (caddr_t)arginfo - SPARE_USRSPACE - 863 roundup(sizeof(canary), sizeof(char *)) - 864 roundup(execpath_len, sizeof(char *)) - 865 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *)); 866 867 if (execpath_len != 0) { 868 imgp->execpathp = (uintptr_t)arginfo - execpath_len; 869 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 870 } 871 872 /* 873 * Prepare the canary for SSP. 874 */ 875 arc4rand(canary, sizeof(canary), 0); 876 imgp->canary = (uintptr_t)arginfo - 877 roundup(execpath_len, sizeof(char *)) - 878 roundup(sizeof(canary), sizeof(char *)); 879 copyout(canary, (void *)imgp->canary, sizeof(canary)); 880 881 /* 882 * If we have a valid auxargs ptr, prepare some room 883 * on the stack. 884 */ 885 if (imgp->auxargs) { 886 /* 887 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 888 * lower compatibility. 889 */ 890 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 891 (LINUX_AT_COUNT * 2); 892 /* 893 * The '+ 2' is for the null pointers at the end of each of 894 * the arg and env vector sets,and imgp->auxarg_size is room 895 * for argument of Runtime loader. 896 */ 897 vectp = (u_int32_t *) (destp - (imgp->args->argc + 898 imgp->args->envc + 2 + imgp->auxarg_size) * 899 sizeof(u_int32_t)); 900 901 } else 902 /* 903 * The '+ 2' is for the null pointers at the end of each of 904 * the arg and env vector sets 905 */ 906 vectp = (u_int32_t *)(destp - (imgp->args->argc + 907 imgp->args->envc + 2) * sizeof(u_int32_t)); 908 909 /* 910 * vectp also becomes our initial stack base 911 */ 912 stack_base = vectp; 913 914 stringp = imgp->args->begin_argv; 915 argc = imgp->args->argc; 916 envc = imgp->args->envc; 917 /* 918 * Copy out strings - arguments and environment. 919 */ 920 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 921 922 /* 923 * Fill in "ps_strings" struct for ps, w, etc. 924 */ 925 suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp); 926 suword32(&arginfo->ps_nargvstr, argc); 927 928 /* 929 * Fill in argument portion of vector table. 930 */ 931 for (; argc > 0; --argc) { 932 suword32(vectp++, (uint32_t)(intptr_t)destp); 933 while (*stringp++ != 0) 934 destp++; 935 destp++; 936 } 937 938 /* a null vector table pointer separates the argp's from the envp's */ 939 suword32(vectp++, 0); 940 941 suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp); 942 suword32(&arginfo->ps_nenvstr, envc); 943 944 /* 945 * Fill in environment portion of vector table. 946 */ 947 for (; envc > 0; --envc) { 948 suword32(vectp++, (uint32_t)(intptr_t)destp); 949 while (*stringp++ != 0) 950 destp++; 951 destp++; 952 } 953 954 /* end of vector table is a null pointer */ 955 suword32(vectp, 0); 956 957 return ((register_t *)stack_base); 958 } 959 960 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 0, 961 "32-bit Linux emulation"); 962 963 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ; 964 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW, 965 &linux32_maxdsiz, 0, ""); 966 static u_long linux32_maxssiz = LINUX32_MAXSSIZ; 967 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW, 968 &linux32_maxssiz, 0, ""); 969 static u_long linux32_maxvmem = LINUX32_MAXVMEM; 970 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW, 971 &linux32_maxvmem, 0, ""); 972 973 #if defined(DEBUG) 974 SYSCTL_PROC(_compat_linux32, OID_AUTO, debug, 975 CTLTYPE_STRING | CTLFLAG_RW, 976 0, 0, linux_sysctl_debug, "A", 977 "Linux debugging control"); 978 #endif 979 980 static void 981 linux32_fixlimit(struct rlimit *rl, int which) 982 { 983 984 switch (which) { 985 case RLIMIT_DATA: 986 if (linux32_maxdsiz != 0) { 987 if (rl->rlim_cur > linux32_maxdsiz) 988 rl->rlim_cur = linux32_maxdsiz; 989 if (rl->rlim_max > linux32_maxdsiz) 990 rl->rlim_max = linux32_maxdsiz; 991 } 992 break; 993 case RLIMIT_STACK: 994 if (linux32_maxssiz != 0) { 995 if (rl->rlim_cur > linux32_maxssiz) 996 rl->rlim_cur = linux32_maxssiz; 997 if (rl->rlim_max > linux32_maxssiz) 998 rl->rlim_max = linux32_maxssiz; 999 } 1000 break; 1001 case RLIMIT_VMEM: 1002 if (linux32_maxvmem != 0) { 1003 if (rl->rlim_cur > linux32_maxvmem) 1004 rl->rlim_cur = linux32_maxvmem; 1005 if (rl->rlim_max > linux32_maxvmem) 1006 rl->rlim_max = linux32_maxvmem; 1007 } 1008 break; 1009 } 1010 } 1011 1012 struct sysentvec elf_linux_sysvec = { 1013 .sv_size = LINUX32_SYS_MAXSYSCALL, 1014 .sv_table = linux32_sysent, 1015 .sv_mask = 0, 1016 .sv_errsize = ELAST + 1, 1017 .sv_errtbl = bsd_to_linux_errno, 1018 .sv_transtrap = translate_traps, 1019 .sv_fixup = elf_linux_fixup, 1020 .sv_sendsig = linux_sendsig, 1021 .sv_sigcode = &_binary_linux32_locore_o_start, 1022 .sv_szsigcode = &linux_szsigcode, 1023 .sv_name = "Linux ELF32", 1024 .sv_coredump = elf32_coredump, 1025 .sv_imgact_try = exec_linux_imgact_try, 1026 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 1027 .sv_pagesize = PAGE_SIZE, 1028 .sv_minuser = VM_MIN_ADDRESS, 1029 .sv_maxuser = LINUX32_MAXUSER, 1030 .sv_usrstack = LINUX32_USRSTACK, 1031 .sv_psstrings = LINUX32_PS_STRINGS, 1032 .sv_stackprot = VM_PROT_ALL, 1033 .sv_copyout_strings = linux_copyout_strings, 1034 .sv_setregs = exec_linux_setregs, 1035 .sv_fixlimit = linux32_fixlimit, 1036 .sv_maxssiz = &linux32_maxssiz, 1037 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP, 1038 .sv_set_syscall_retval = cpu_set_syscall_retval, 1039 .sv_fetch_syscall_args = linux32_fetch_syscall_args, 1040 .sv_syscallnames = NULL, 1041 .sv_shared_page_base = LINUX32_SHAREDPAGE, 1042 .sv_shared_page_len = PAGE_SIZE, 1043 .sv_schedtail = linux_schedtail, 1044 .sv_thread_detach = linux_thread_detach, 1045 .sv_trap = NULL, 1046 }; 1047 1048 static void 1049 linux_vdso_install(void *param) 1050 { 1051 1052 linux_szsigcode = (&_binary_linux32_locore_o_end - 1053 &_binary_linux32_locore_o_start); 1054 1055 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 1056 panic("Linux invalid vdso size\n"); 1057 1058 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 1059 1060 linux_shared_page_obj = __elfN(linux_shared_page_init) 1061 (&linux_shared_page_mapping); 1062 1063 __elfN(linux_vdso_reloc)(&elf_linux_sysvec); 1064 1065 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 1066 linux_szsigcode); 1067 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 1068 1069 linux_kplatform = linux_shared_page_mapping + 1070 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base); 1071 } 1072 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 1073 (sysinit_cfunc_t)linux_vdso_install, NULL); 1074 1075 static void 1076 linux_vdso_deinstall(void *param) 1077 { 1078 1079 __elfN(linux_shared_page_fini)(linux_shared_page_obj); 1080 }; 1081 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 1082 (sysinit_cfunc_t)linux_vdso_deinstall, NULL); 1083 1084 static char GNU_ABI_VENDOR[] = "GNU"; 1085 static int GNULINUX_ABI_DESC = 0; 1086 1087 static boolean_t 1088 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel) 1089 { 1090 const Elf32_Word *desc; 1091 uintptr_t p; 1092 1093 p = (uintptr_t)(note + 1); 1094 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 1095 1096 desc = (const Elf32_Word *)p; 1097 if (desc[0] != GNULINUX_ABI_DESC) 1098 return (FALSE); 1099 1100 /* 1101 * For linux we encode osrel as follows (see linux_mib.c): 1102 * VVVMMMIII (version, major, minor), see linux_mib.c. 1103 */ 1104 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3]; 1105 1106 return (TRUE); 1107 } 1108 1109 static Elf_Brandnote linux32_brandnote = { 1110 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 1111 .hdr.n_descsz = 16, /* XXX at least 16 */ 1112 .hdr.n_type = 1, 1113 .vendor = GNU_ABI_VENDOR, 1114 .flags = BN_TRANSLATE_OSREL, 1115 .trans_osrel = linux32_trans_osrel 1116 }; 1117 1118 static Elf32_Brandinfo linux_brand = { 1119 .brand = ELFOSABI_LINUX, 1120 .machine = EM_386, 1121 .compat_3_brand = "Linux", 1122 .emul_path = "/compat/linux", 1123 .interp_path = "/lib/ld-linux.so.1", 1124 .sysvec = &elf_linux_sysvec, 1125 .interp_newpath = NULL, 1126 .brand_note = &linux32_brandnote, 1127 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1128 }; 1129 1130 static Elf32_Brandinfo linux_glibc2brand = { 1131 .brand = ELFOSABI_LINUX, 1132 .machine = EM_386, 1133 .compat_3_brand = "Linux", 1134 .emul_path = "/compat/linux", 1135 .interp_path = "/lib/ld-linux.so.2", 1136 .sysvec = &elf_linux_sysvec, 1137 .interp_newpath = NULL, 1138 .brand_note = &linux32_brandnote, 1139 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1140 }; 1141 1142 static Elf32_Brandinfo linux_muslbrand = { 1143 .brand = ELFOSABI_LINUX, 1144 .machine = EM_386, 1145 .compat_3_brand = "Linux", 1146 .emul_path = "/compat/linux", 1147 .interp_path = "/lib/ld-musl-i386.so.1", 1148 .sysvec = &elf_linux_sysvec, 1149 .interp_newpath = NULL, 1150 .brand_note = &linux32_brandnote, 1151 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1152 }; 1153 1154 Elf32_Brandinfo *linux_brandlist[] = { 1155 &linux_brand, 1156 &linux_glibc2brand, 1157 &linux_muslbrand, 1158 NULL 1159 }; 1160 1161 static int 1162 linux_elf_modevent(module_t mod, int type, void *data) 1163 { 1164 Elf32_Brandinfo **brandinfo; 1165 int error; 1166 struct linux_ioctl_handler **lihp; 1167 1168 error = 0; 1169 1170 switch(type) { 1171 case MOD_LOAD: 1172 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1173 ++brandinfo) 1174 if (elf32_insert_brand_entry(*brandinfo) < 0) 1175 error = EINVAL; 1176 if (error == 0) { 1177 SET_FOREACH(lihp, linux_ioctl_handler_set) 1178 linux_ioctl_register_handler(*lihp); 1179 LIST_INIT(&futex_list); 1180 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF); 1181 stclohz = (stathz ? stathz : hz); 1182 if (bootverbose) 1183 printf("Linux ELF exec handler installed\n"); 1184 } else 1185 printf("cannot insert Linux ELF brand handler\n"); 1186 break; 1187 case MOD_UNLOAD: 1188 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1189 ++brandinfo) 1190 if (elf32_brand_inuse(*brandinfo)) 1191 error = EBUSY; 1192 if (error == 0) { 1193 for (brandinfo = &linux_brandlist[0]; 1194 *brandinfo != NULL; ++brandinfo) 1195 if (elf32_remove_brand_entry(*brandinfo) < 0) 1196 error = EINVAL; 1197 } 1198 if (error == 0) { 1199 SET_FOREACH(lihp, linux_ioctl_handler_set) 1200 linux_ioctl_unregister_handler(*lihp); 1201 mtx_destroy(&futex_mtx); 1202 if (bootverbose) 1203 printf("Linux ELF exec handler removed\n"); 1204 } else 1205 printf("Could not deinstall ELF interpreter entry\n"); 1206 break; 1207 default: 1208 return (EOPNOTSUPP); 1209 } 1210 return (error); 1211 } 1212 1213 static moduledata_t linux_elf_mod = { 1214 "linuxelf", 1215 linux_elf_modevent, 1216 0 1217 }; 1218 1219 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 1220 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1); 1221 FEATURE(linux, "Linux 32bit support"); 1222