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