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 LINUX32_SYS_linux_rt_sendsig 0 109 #define LINUX32_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 linux32_sysent[LINUX32_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)<nitems(_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 int issetugid; 234 235 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; 236 237 KASSERT(curthread->td_proc == imgp->proc, 238 ("unsafe elf_linux_fixup(), should be curproc")); 239 base = (Elf32_Addr *)*stack_base; 240 args = (Elf32_Auxargs *)imgp->auxargs; 241 pos = base + (imgp->args->argc + imgp->args->envc + 2); 242 243 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0; 244 AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO_EHDR, 245 imgp->proc->p_sysent->sv_shared_page_base); 246 AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO, linux32_vsyscall); 247 AUXARGS_ENTRY_32(pos, LINUX_AT_HWCAP, cpu_feature); 248 249 /* 250 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, 251 * as it has appeared in the 2.4.0-rc7 first time. 252 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), 253 * glibc falls back to the hard-coded CLK_TCK value when aux entry 254 * is not present. 255 * Also see linux_times() implementation. 256 */ 257 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) 258 AUXARGS_ENTRY_32(pos, LINUX_AT_CLKTCK, stclohz); 259 AUXARGS_ENTRY_32(pos, AT_PHDR, args->phdr); 260 AUXARGS_ENTRY_32(pos, AT_PHENT, args->phent); 261 AUXARGS_ENTRY_32(pos, AT_PHNUM, args->phnum); 262 AUXARGS_ENTRY_32(pos, AT_PAGESZ, args->pagesz); 263 AUXARGS_ENTRY_32(pos, AT_FLAGS, args->flags); 264 AUXARGS_ENTRY_32(pos, AT_ENTRY, args->entry); 265 AUXARGS_ENTRY_32(pos, AT_BASE, args->base); 266 AUXARGS_ENTRY_32(pos, LINUX_AT_SECURE, issetugid); 267 AUXARGS_ENTRY_32(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 268 AUXARGS_ENTRY_32(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 269 AUXARGS_ENTRY_32(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 270 AUXARGS_ENTRY_32(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 271 AUXARGS_ENTRY_32(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 272 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary)); 273 if (imgp->execpathp != 0) 274 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp)); 275 if (args->execfd != -1) 276 AUXARGS_ENTRY_32(pos, AT_EXECFD, args->execfd); 277 AUXARGS_ENTRY_32(pos, AT_NULL, 0); 278 279 free(imgp->auxargs, M_TEMP); 280 imgp->auxargs = NULL; 281 282 base--; 283 suword32(base, (uint32_t)imgp->args->argc); 284 *stack_base = (register_t *)base; 285 return (0); 286 } 287 288 static void 289 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 290 { 291 struct thread *td = curthread; 292 struct proc *p = td->td_proc; 293 struct sigacts *psp; 294 struct trapframe *regs; 295 struct l_rt_sigframe *fp, frame; 296 int oonstack; 297 int sig; 298 int code; 299 300 sig = ksi->ksi_signo; 301 code = ksi->ksi_code; 302 PROC_LOCK_ASSERT(p, MA_OWNED); 303 psp = p->p_sigacts; 304 mtx_assert(&psp->ps_mtx, MA_OWNED); 305 regs = td->td_frame; 306 oonstack = sigonstack(regs->tf_rsp); 307 308 #ifdef DEBUG 309 if (ldebug(rt_sendsig)) 310 printf(ARGS(rt_sendsig, "%p, %d, %p, %u"), 311 catcher, sig, (void*)mask, code); 312 #endif 313 /* 314 * Allocate space for the signal handler context. 315 */ 316 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 317 SIGISMEMBER(psp->ps_sigonstack, sig)) { 318 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 319 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe)); 320 } else 321 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1; 322 mtx_unlock(&psp->ps_mtx); 323 324 /* 325 * Build the argument list for the signal handler. 326 */ 327 sig = bsd_to_linux_signal(sig); 328 329 bzero(&frame, sizeof(frame)); 330 331 frame.sf_handler = PTROUT(catcher); 332 frame.sf_sig = sig; 333 frame.sf_siginfo = PTROUT(&fp->sf_si); 334 frame.sf_ucontext = PTROUT(&fp->sf_sc); 335 336 /* Fill in POSIX parts */ 337 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig); 338 339 /* 340 * Build the signal context to be used by sigreturn 341 * and libgcc unwind. 342 */ 343 frame.sf_sc.uc_flags = 0; /* XXX ??? */ 344 frame.sf_sc.uc_link = 0; /* XXX ??? */ 345 346 frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 347 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 348 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 349 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 350 PROC_UNLOCK(p); 351 352 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask); 353 354 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask; 355 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi; 356 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi; 357 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp; 358 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx; 359 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp; 360 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx; 361 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx; 362 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax; 363 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip; 364 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 365 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs; 366 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs; 367 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es; 368 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds; 369 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags; 370 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp; 371 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss; 372 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err; 373 frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 374 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 375 376 #ifdef DEBUG 377 if (ldebug(rt_sendsig)) 378 printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"), 379 frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp, 380 td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask); 381 #endif 382 383 if (copyout(&frame, fp, sizeof(frame)) != 0) { 384 /* 385 * Process has trashed its stack; give it an illegal 386 * instruction to halt it in its tracks. 387 */ 388 #ifdef DEBUG 389 if (ldebug(rt_sendsig)) 390 printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"), 391 fp, oonstack); 392 #endif 393 PROC_LOCK(p); 394 sigexit(td, SIGILL); 395 } 396 397 /* 398 * Build context to run handler in. 399 */ 400 regs->tf_rsp = PTROUT(fp); 401 regs->tf_rip = linux32_rt_sigcode; 402 regs->tf_rflags &= ~(PSL_T | PSL_D); 403 regs->tf_cs = _ucode32sel; 404 regs->tf_ss = _udatasel; 405 regs->tf_ds = _udatasel; 406 regs->tf_es = _udatasel; 407 regs->tf_fs = _ufssel; 408 regs->tf_gs = _ugssel; 409 regs->tf_flags = TF_HASSEGS; 410 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 411 PROC_LOCK(p); 412 mtx_lock(&psp->ps_mtx); 413 } 414 415 416 /* 417 * Send an interrupt to process. 418 * 419 * Stack is set up to allow sigcode stored 420 * in u. to call routine, followed by kcall 421 * to sigreturn routine below. After sigreturn 422 * resets the signal mask, the stack, and the 423 * frame pointer, it returns to the user 424 * specified pc, psl. 425 */ 426 static void 427 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 428 { 429 struct thread *td = curthread; 430 struct proc *p = td->td_proc; 431 struct sigacts *psp; 432 struct trapframe *regs; 433 struct l_sigframe *fp, frame; 434 l_sigset_t lmask; 435 int oonstack; 436 int sig, code; 437 438 sig = ksi->ksi_signo; 439 code = ksi->ksi_code; 440 PROC_LOCK_ASSERT(p, MA_OWNED); 441 psp = p->p_sigacts; 442 mtx_assert(&psp->ps_mtx, MA_OWNED); 443 if (SIGISMEMBER(psp->ps_siginfo, sig)) { 444 /* Signal handler installed with SA_SIGINFO. */ 445 linux_rt_sendsig(catcher, ksi, mask); 446 return; 447 } 448 449 regs = td->td_frame; 450 oonstack = sigonstack(regs->tf_rsp); 451 452 #ifdef DEBUG 453 if (ldebug(sendsig)) 454 printf(ARGS(sendsig, "%p, %d, %p, %u"), 455 catcher, sig, (void*)mask, code); 456 #endif 457 458 /* 459 * Allocate space for the signal handler context. 460 */ 461 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 462 SIGISMEMBER(psp->ps_sigonstack, sig)) { 463 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 464 td->td_sigstk.ss_size - sizeof(struct l_sigframe)); 465 } else 466 fp = (struct l_sigframe *)regs->tf_rsp - 1; 467 mtx_unlock(&psp->ps_mtx); 468 PROC_UNLOCK(p); 469 470 /* 471 * Build the argument list for the signal handler. 472 */ 473 sig = bsd_to_linux_signal(sig); 474 475 bzero(&frame, sizeof(frame)); 476 477 frame.sf_handler = PTROUT(catcher); 478 frame.sf_sig = sig; 479 480 bsd_to_linux_sigset(mask, &lmask); 481 482 /* 483 * Build the signal context to be used by sigreturn. 484 */ 485 frame.sf_sc.sc_mask = lmask.__mask; 486 frame.sf_sc.sc_gs = regs->tf_gs; 487 frame.sf_sc.sc_fs = regs->tf_fs; 488 frame.sf_sc.sc_es = regs->tf_es; 489 frame.sf_sc.sc_ds = regs->tf_ds; 490 frame.sf_sc.sc_edi = regs->tf_rdi; 491 frame.sf_sc.sc_esi = regs->tf_rsi; 492 frame.sf_sc.sc_ebp = regs->tf_rbp; 493 frame.sf_sc.sc_ebx = regs->tf_rbx; 494 frame.sf_sc.sc_esp = regs->tf_rsp; 495 frame.sf_sc.sc_edx = regs->tf_rdx; 496 frame.sf_sc.sc_ecx = regs->tf_rcx; 497 frame.sf_sc.sc_eax = regs->tf_rax; 498 frame.sf_sc.sc_eip = regs->tf_rip; 499 frame.sf_sc.sc_cs = regs->tf_cs; 500 frame.sf_sc.sc_eflags = regs->tf_rflags; 501 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp; 502 frame.sf_sc.sc_ss = regs->tf_ss; 503 frame.sf_sc.sc_err = regs->tf_err; 504 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 505 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code); 506 507 frame.sf_extramask[0] = lmask.__mask; 508 509 if (copyout(&frame, fp, sizeof(frame)) != 0) { 510 /* 511 * Process has trashed its stack; give it an illegal 512 * instruction to halt it in its tracks. 513 */ 514 PROC_LOCK(p); 515 sigexit(td, SIGILL); 516 } 517 518 /* 519 * Build context to run handler in. 520 */ 521 regs->tf_rsp = PTROUT(fp); 522 regs->tf_rip = linux32_sigcode; 523 regs->tf_rflags &= ~(PSL_T | PSL_D); 524 regs->tf_cs = _ucode32sel; 525 regs->tf_ss = _udatasel; 526 regs->tf_ds = _udatasel; 527 regs->tf_es = _udatasel; 528 regs->tf_fs = _ufssel; 529 regs->tf_gs = _ugssel; 530 regs->tf_flags = TF_HASSEGS; 531 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 532 PROC_LOCK(p); 533 mtx_lock(&psp->ps_mtx); 534 } 535 536 /* 537 * System call to cleanup state after a signal 538 * has been taken. Reset signal mask and 539 * stack state from context left by sendsig (above). 540 * Return to previous pc and psl as specified by 541 * context left by sendsig. Check carefully to 542 * make sure that the user has not modified the 543 * psl to gain improper privileges or to cause 544 * a machine fault. 545 */ 546 int 547 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) 548 { 549 struct l_sigframe frame; 550 struct trapframe *regs; 551 sigset_t bmask; 552 l_sigset_t lmask; 553 int eflags; 554 ksiginfo_t ksi; 555 556 regs = td->td_frame; 557 558 #ifdef DEBUG 559 if (ldebug(sigreturn)) 560 printf(ARGS(sigreturn, "%p"), (void *)args->sfp); 561 #endif 562 /* 563 * The trampoline code hands us the sigframe. 564 * It is unsafe to keep track of it ourselves, in the event that a 565 * program jumps out of a signal handler. 566 */ 567 if (copyin(args->sfp, &frame, sizeof(frame)) != 0) 568 return (EFAULT); 569 570 /* 571 * Check for security violations. 572 */ 573 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 574 eflags = frame.sf_sc.sc_eflags; 575 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 576 return(EINVAL); 577 578 /* 579 * Don't allow users to load a valid privileged %cs. Let the 580 * hardware check for invalid selectors, excess privilege in 581 * other selectors, invalid %eip's and invalid %esp's. 582 */ 583 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 584 if (!CS_SECURE(frame.sf_sc.sc_cs)) { 585 ksiginfo_init_trap(&ksi); 586 ksi.ksi_signo = SIGBUS; 587 ksi.ksi_code = BUS_OBJERR; 588 ksi.ksi_trapno = T_PROTFLT; 589 ksi.ksi_addr = (void *)regs->tf_rip; 590 trapsignal(td, &ksi); 591 return(EINVAL); 592 } 593 594 lmask.__mask = frame.sf_sc.sc_mask; 595 lmask.__mask = frame.sf_extramask[0]; 596 linux_to_bsd_sigset(&lmask, &bmask); 597 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 598 599 /* 600 * Restore signal context. 601 */ 602 regs->tf_rdi = frame.sf_sc.sc_edi; 603 regs->tf_rsi = frame.sf_sc.sc_esi; 604 regs->tf_rbp = frame.sf_sc.sc_ebp; 605 regs->tf_rbx = frame.sf_sc.sc_ebx; 606 regs->tf_rdx = frame.sf_sc.sc_edx; 607 regs->tf_rcx = frame.sf_sc.sc_ecx; 608 regs->tf_rax = frame.sf_sc.sc_eax; 609 regs->tf_rip = frame.sf_sc.sc_eip; 610 regs->tf_cs = frame.sf_sc.sc_cs; 611 regs->tf_ds = frame.sf_sc.sc_ds; 612 regs->tf_es = frame.sf_sc.sc_es; 613 regs->tf_fs = frame.sf_sc.sc_fs; 614 regs->tf_gs = frame.sf_sc.sc_gs; 615 regs->tf_rflags = eflags; 616 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal; 617 regs->tf_ss = frame.sf_sc.sc_ss; 618 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 619 620 return (EJUSTRETURN); 621 } 622 623 /* 624 * System call to cleanup state after a signal 625 * has been taken. Reset signal mask and 626 * stack state from context left by rt_sendsig (above). 627 * Return to previous pc and psl as specified by 628 * context left by sendsig. Check carefully to 629 * make sure that the user has not modified the 630 * psl to gain improper privileges or to cause 631 * a machine fault. 632 */ 633 int 634 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 635 { 636 struct l_ucontext uc; 637 struct l_sigcontext *context; 638 sigset_t bmask; 639 l_stack_t *lss; 640 stack_t ss; 641 struct trapframe *regs; 642 int eflags; 643 ksiginfo_t ksi; 644 645 regs = td->td_frame; 646 647 #ifdef DEBUG 648 if (ldebug(rt_sigreturn)) 649 printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp); 650 #endif 651 /* 652 * The trampoline code hands us the ucontext. 653 * It is unsafe to keep track of it ourselves, in the event that a 654 * program jumps out of a signal handler. 655 */ 656 if (copyin(args->ucp, &uc, sizeof(uc)) != 0) 657 return (EFAULT); 658 659 context = &uc.uc_mcontext; 660 661 /* 662 * Check for security violations. 663 */ 664 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 665 eflags = context->sc_eflags; 666 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 667 return(EINVAL); 668 669 /* 670 * Don't allow users to load a valid privileged %cs. Let the 671 * hardware check for invalid selectors, excess privilege in 672 * other selectors, invalid %eip's and invalid %esp's. 673 */ 674 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 675 if (!CS_SECURE(context->sc_cs)) { 676 ksiginfo_init_trap(&ksi); 677 ksi.ksi_signo = SIGBUS; 678 ksi.ksi_code = BUS_OBJERR; 679 ksi.ksi_trapno = T_PROTFLT; 680 ksi.ksi_addr = (void *)regs->tf_rip; 681 trapsignal(td, &ksi); 682 return(EINVAL); 683 } 684 685 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); 686 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 687 688 /* 689 * Restore signal context 690 */ 691 regs->tf_gs = context->sc_gs; 692 regs->tf_fs = context->sc_fs; 693 regs->tf_es = context->sc_es; 694 regs->tf_ds = context->sc_ds; 695 regs->tf_rdi = context->sc_edi; 696 regs->tf_rsi = context->sc_esi; 697 regs->tf_rbp = context->sc_ebp; 698 regs->tf_rbx = context->sc_ebx; 699 regs->tf_rdx = context->sc_edx; 700 regs->tf_rcx = context->sc_ecx; 701 regs->tf_rax = context->sc_eax; 702 regs->tf_rip = context->sc_eip; 703 regs->tf_cs = context->sc_cs; 704 regs->tf_rflags = eflags; 705 regs->tf_rsp = context->sc_esp_at_signal; 706 regs->tf_ss = context->sc_ss; 707 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 708 709 /* 710 * call sigaltstack & ignore results.. 711 */ 712 lss = &uc.uc_stack; 713 ss.ss_sp = PTRIN(lss->ss_sp); 714 ss.ss_size = lss->ss_size; 715 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags); 716 717 #ifdef DEBUG 718 if (ldebug(rt_sigreturn)) 719 printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"), 720 ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask); 721 #endif 722 (void)kern_sigaltstack(td, &ss, NULL); 723 724 return (EJUSTRETURN); 725 } 726 727 static int 728 linux32_fetch_syscall_args(struct thread *td) 729 { 730 struct proc *p; 731 struct trapframe *frame; 732 struct syscall_args *sa; 733 734 p = td->td_proc; 735 frame = td->td_frame; 736 sa = &td->td_sa; 737 738 sa->args[0] = frame->tf_rbx; 739 sa->args[1] = frame->tf_rcx; 740 sa->args[2] = frame->tf_rdx; 741 sa->args[3] = frame->tf_rsi; 742 sa->args[4] = frame->tf_rdi; 743 sa->args[5] = frame->tf_rbp; /* Unconfirmed */ 744 sa->code = frame->tf_rax; 745 746 if (sa->code >= p->p_sysent->sv_size) 747 /* nosys */ 748 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 749 else 750 sa->callp = &p->p_sysent->sv_table[sa->code]; 751 sa->narg = sa->callp->sy_narg; 752 753 td->td_retval[0] = 0; 754 td->td_retval[1] = frame->tf_rdx; 755 756 return (0); 757 } 758 759 /* 760 * If a linux binary is exec'ing something, try this image activator 761 * first. We override standard shell script execution in order to 762 * be able to modify the interpreter path. We only do this if a linux 763 * binary is doing the exec, so we do not create an EXEC module for it. 764 */ 765 static int exec_linux_imgact_try(struct image_params *iparams); 766 767 static int 768 exec_linux_imgact_try(struct image_params *imgp) 769 { 770 const char *head = (const char *)imgp->image_header; 771 char *rpath; 772 int error = -1; 773 774 /* 775 * The interpreter for shell scripts run from a linux binary needs 776 * to be located in /compat/linux if possible in order to recursively 777 * maintain linux path emulation. 778 */ 779 if (((const short *)head)[0] == SHELLMAGIC) { 780 /* 781 * Run our normal shell image activator. If it succeeds attempt 782 * to use the alternate path for the interpreter. If an 783 * alternate * path is found, use our stringspace to store it. 784 */ 785 if ((error = exec_shell_imgact(imgp)) == 0) { 786 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc), 787 imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, 788 AT_FDCWD); 789 if (rpath != NULL) 790 imgp->args->fname_buf = 791 imgp->interpreter_name = rpath; 792 } 793 } 794 return (error); 795 } 796 797 /* 798 * Clear registers on exec 799 * XXX copied from ia32_signal.c. 800 */ 801 static void 802 exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack) 803 { 804 struct trapframe *regs = td->td_frame; 805 struct pcb *pcb = td->td_pcb; 806 807 mtx_lock(&dt_lock); 808 if (td->td_proc->p_md.md_ldt != NULL) 809 user_ldt_free(td); 810 else 811 mtx_unlock(&dt_lock); 812 813 critical_enter(); 814 wrmsr(MSR_FSBASE, 0); 815 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */ 816 pcb->pcb_fsbase = 0; 817 pcb->pcb_gsbase = 0; 818 critical_exit(); 819 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 820 821 bzero((char *)regs, sizeof(struct trapframe)); 822 regs->tf_rip = imgp->entry_addr; 823 regs->tf_rsp = stack; 824 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T); 825 regs->tf_gs = _ugssel; 826 regs->tf_fs = _ufssel; 827 regs->tf_es = _udatasel; 828 regs->tf_ds = _udatasel; 829 regs->tf_ss = _udatasel; 830 regs->tf_flags = TF_HASSEGS; 831 regs->tf_cs = _ucode32sel; 832 regs->tf_rbx = imgp->ps_strings; 833 834 fpstate_drop(td); 835 836 /* Do full restore on return so that we can change to a different %cs */ 837 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET); 838 td->td_retval[1] = 0; 839 } 840 841 /* 842 * XXX copied from ia32_sysvec.c. 843 */ 844 static register_t * 845 linux_copyout_strings(struct image_params *imgp) 846 { 847 int argc, envc; 848 u_int32_t *vectp; 849 char *stringp, *destp; 850 u_int32_t *stack_base; 851 struct linux32_ps_strings *arginfo; 852 char canary[LINUX_AT_RANDOM_LEN]; 853 size_t execpath_len; 854 855 /* 856 * Calculate string base and vector table pointers. 857 */ 858 if (imgp->execpath != NULL && imgp->auxargs != NULL) 859 execpath_len = strlen(imgp->execpath) + 1; 860 else 861 execpath_len = 0; 862 863 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; 864 destp = (caddr_t)arginfo - SPARE_USRSPACE - 865 roundup(sizeof(canary), sizeof(char *)) - 866 roundup(execpath_len, sizeof(char *)) - 867 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *)); 868 869 if (execpath_len != 0) { 870 imgp->execpathp = (uintptr_t)arginfo - execpath_len; 871 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 872 } 873 874 /* 875 * Prepare the canary for SSP. 876 */ 877 arc4rand(canary, sizeof(canary), 0); 878 imgp->canary = (uintptr_t)arginfo - 879 roundup(execpath_len, sizeof(char *)) - 880 roundup(sizeof(canary), sizeof(char *)); 881 copyout(canary, (void *)imgp->canary, sizeof(canary)); 882 883 /* 884 * If we have a valid auxargs ptr, prepare some room 885 * on the stack. 886 */ 887 if (imgp->auxargs) { 888 /* 889 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 890 * lower compatibility. 891 */ 892 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 893 (LINUX_AT_COUNT * 2); 894 /* 895 * The '+ 2' is for the null pointers at the end of each of 896 * the arg and env vector sets,and imgp->auxarg_size is room 897 * for argument of Runtime loader. 898 */ 899 vectp = (u_int32_t *) (destp - (imgp->args->argc + 900 imgp->args->envc + 2 + imgp->auxarg_size) * 901 sizeof(u_int32_t)); 902 903 } else 904 /* 905 * The '+ 2' is for the null pointers at the end of each of 906 * the arg and env vector sets 907 */ 908 vectp = (u_int32_t *)(destp - (imgp->args->argc + 909 imgp->args->envc + 2) * sizeof(u_int32_t)); 910 911 /* 912 * vectp also becomes our initial stack base 913 */ 914 stack_base = vectp; 915 916 stringp = imgp->args->begin_argv; 917 argc = imgp->args->argc; 918 envc = imgp->args->envc; 919 /* 920 * Copy out strings - arguments and environment. 921 */ 922 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 923 924 /* 925 * Fill in "ps_strings" struct for ps, w, etc. 926 */ 927 suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp); 928 suword32(&arginfo->ps_nargvstr, argc); 929 930 /* 931 * Fill in argument portion of vector table. 932 */ 933 for (; argc > 0; --argc) { 934 suword32(vectp++, (uint32_t)(intptr_t)destp); 935 while (*stringp++ != 0) 936 destp++; 937 destp++; 938 } 939 940 /* a null vector table pointer separates the argp's from the envp's */ 941 suword32(vectp++, 0); 942 943 suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp); 944 suword32(&arginfo->ps_nenvstr, envc); 945 946 /* 947 * Fill in environment portion of vector table. 948 */ 949 for (; envc > 0; --envc) { 950 suword32(vectp++, (uint32_t)(intptr_t)destp); 951 while (*stringp++ != 0) 952 destp++; 953 destp++; 954 } 955 956 /* end of vector table is a null pointer */ 957 suword32(vectp, 0); 958 959 return ((register_t *)stack_base); 960 } 961 962 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 0, 963 "32-bit Linux emulation"); 964 965 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ; 966 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW, 967 &linux32_maxdsiz, 0, ""); 968 static u_long linux32_maxssiz = LINUX32_MAXSSIZ; 969 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW, 970 &linux32_maxssiz, 0, ""); 971 static u_long linux32_maxvmem = LINUX32_MAXVMEM; 972 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW, 973 &linux32_maxvmem, 0, ""); 974 975 #if defined(DEBUG) 976 SYSCTL_PROC(_compat_linux32, OID_AUTO, debug, 977 CTLTYPE_STRING | CTLFLAG_RW, 978 0, 0, linux_sysctl_debug, "A", 979 "Linux debugging control"); 980 #endif 981 982 static void 983 linux32_fixlimit(struct rlimit *rl, int which) 984 { 985 986 switch (which) { 987 case RLIMIT_DATA: 988 if (linux32_maxdsiz != 0) { 989 if (rl->rlim_cur > linux32_maxdsiz) 990 rl->rlim_cur = linux32_maxdsiz; 991 if (rl->rlim_max > linux32_maxdsiz) 992 rl->rlim_max = linux32_maxdsiz; 993 } 994 break; 995 case RLIMIT_STACK: 996 if (linux32_maxssiz != 0) { 997 if (rl->rlim_cur > linux32_maxssiz) 998 rl->rlim_cur = linux32_maxssiz; 999 if (rl->rlim_max > linux32_maxssiz) 1000 rl->rlim_max = linux32_maxssiz; 1001 } 1002 break; 1003 case RLIMIT_VMEM: 1004 if (linux32_maxvmem != 0) { 1005 if (rl->rlim_cur > linux32_maxvmem) 1006 rl->rlim_cur = linux32_maxvmem; 1007 if (rl->rlim_max > linux32_maxvmem) 1008 rl->rlim_max = linux32_maxvmem; 1009 } 1010 break; 1011 } 1012 } 1013 1014 struct sysentvec elf_linux_sysvec = { 1015 .sv_size = LINUX32_SYS_MAXSYSCALL, 1016 .sv_table = linux32_sysent, 1017 .sv_mask = 0, 1018 .sv_errsize = ELAST + 1, 1019 .sv_errtbl = bsd_to_linux_errno, 1020 .sv_transtrap = translate_traps, 1021 .sv_fixup = elf_linux_fixup, 1022 .sv_sendsig = linux_sendsig, 1023 .sv_sigcode = &_binary_linux32_locore_o_start, 1024 .sv_szsigcode = &linux_szsigcode, 1025 .sv_name = "Linux ELF32", 1026 .sv_coredump = elf32_coredump, 1027 .sv_imgact_try = exec_linux_imgact_try, 1028 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 1029 .sv_pagesize = PAGE_SIZE, 1030 .sv_minuser = VM_MIN_ADDRESS, 1031 .sv_maxuser = LINUX32_MAXUSER, 1032 .sv_usrstack = LINUX32_USRSTACK, 1033 .sv_psstrings = LINUX32_PS_STRINGS, 1034 .sv_stackprot = VM_PROT_ALL, 1035 .sv_copyout_strings = linux_copyout_strings, 1036 .sv_setregs = exec_linux_setregs, 1037 .sv_fixlimit = linux32_fixlimit, 1038 .sv_maxssiz = &linux32_maxssiz, 1039 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP, 1040 .sv_set_syscall_retval = cpu_set_syscall_retval, 1041 .sv_fetch_syscall_args = linux32_fetch_syscall_args, 1042 .sv_syscallnames = NULL, 1043 .sv_shared_page_base = LINUX32_SHAREDPAGE, 1044 .sv_shared_page_len = PAGE_SIZE, 1045 .sv_schedtail = linux_schedtail, 1046 .sv_thread_detach = linux_thread_detach, 1047 .sv_trap = NULL, 1048 }; 1049 1050 static void 1051 linux_vdso_install(void *param) 1052 { 1053 1054 linux_szsigcode = (&_binary_linux32_locore_o_end - 1055 &_binary_linux32_locore_o_start); 1056 1057 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 1058 panic("Linux invalid vdso size\n"); 1059 1060 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 1061 1062 linux_shared_page_obj = __elfN(linux_shared_page_init) 1063 (&linux_shared_page_mapping); 1064 1065 __elfN(linux_vdso_reloc)(&elf_linux_sysvec, LINUX32_SHAREDPAGE); 1066 1067 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 1068 linux_szsigcode); 1069 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 1070 1071 linux_kplatform = linux_shared_page_mapping + 1072 (linux_platform - (caddr_t)LINUX32_SHAREDPAGE); 1073 } 1074 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 1075 (sysinit_cfunc_t)linux_vdso_install, NULL); 1076 1077 static void 1078 linux_vdso_deinstall(void *param) 1079 { 1080 1081 __elfN(linux_shared_page_fini)(linux_shared_page_obj); 1082 }; 1083 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 1084 (sysinit_cfunc_t)linux_vdso_deinstall, NULL); 1085 1086 static char GNU_ABI_VENDOR[] = "GNU"; 1087 static int GNULINUX_ABI_DESC = 0; 1088 1089 static boolean_t 1090 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel) 1091 { 1092 const Elf32_Word *desc; 1093 uintptr_t p; 1094 1095 p = (uintptr_t)(note + 1); 1096 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 1097 1098 desc = (const Elf32_Word *)p; 1099 if (desc[0] != GNULINUX_ABI_DESC) 1100 return (FALSE); 1101 1102 /* 1103 * For linux we encode osrel as follows (see linux_mib.c): 1104 * VVVMMMIII (version, major, minor), see linux_mib.c. 1105 */ 1106 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3]; 1107 1108 return (TRUE); 1109 } 1110 1111 static Elf_Brandnote linux32_brandnote = { 1112 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 1113 .hdr.n_descsz = 16, /* XXX at least 16 */ 1114 .hdr.n_type = 1, 1115 .vendor = GNU_ABI_VENDOR, 1116 .flags = BN_TRANSLATE_OSREL, 1117 .trans_osrel = linux32_trans_osrel 1118 }; 1119 1120 static Elf32_Brandinfo linux_brand = { 1121 .brand = ELFOSABI_LINUX, 1122 .machine = EM_386, 1123 .compat_3_brand = "Linux", 1124 .emul_path = "/compat/linux", 1125 .interp_path = "/lib/ld-linux.so.1", 1126 .sysvec = &elf_linux_sysvec, 1127 .interp_newpath = NULL, 1128 .brand_note = &linux32_brandnote, 1129 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1130 }; 1131 1132 static Elf32_Brandinfo linux_glibc2brand = { 1133 .brand = ELFOSABI_LINUX, 1134 .machine = EM_386, 1135 .compat_3_brand = "Linux", 1136 .emul_path = "/compat/linux", 1137 .interp_path = "/lib/ld-linux.so.2", 1138 .sysvec = &elf_linux_sysvec, 1139 .interp_newpath = NULL, 1140 .brand_note = &linux32_brandnote, 1141 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1142 }; 1143 1144 Elf32_Brandinfo *linux_brandlist[] = { 1145 &linux_brand, 1146 &linux_glibc2brand, 1147 NULL 1148 }; 1149 1150 static int 1151 linux_elf_modevent(module_t mod, int type, void *data) 1152 { 1153 Elf32_Brandinfo **brandinfo; 1154 int error; 1155 struct linux_ioctl_handler **lihp; 1156 1157 error = 0; 1158 1159 switch(type) { 1160 case MOD_LOAD: 1161 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1162 ++brandinfo) 1163 if (elf32_insert_brand_entry(*brandinfo) < 0) 1164 error = EINVAL; 1165 if (error == 0) { 1166 SET_FOREACH(lihp, linux_ioctl_handler_set) 1167 linux_ioctl_register_handler(*lihp); 1168 LIST_INIT(&futex_list); 1169 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF); 1170 stclohz = (stathz ? stathz : hz); 1171 if (bootverbose) 1172 printf("Linux ELF exec handler installed\n"); 1173 } else 1174 printf("cannot insert Linux ELF brand handler\n"); 1175 break; 1176 case MOD_UNLOAD: 1177 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1178 ++brandinfo) 1179 if (elf32_brand_inuse(*brandinfo)) 1180 error = EBUSY; 1181 if (error == 0) { 1182 for (brandinfo = &linux_brandlist[0]; 1183 *brandinfo != NULL; ++brandinfo) 1184 if (elf32_remove_brand_entry(*brandinfo) < 0) 1185 error = EINVAL; 1186 } 1187 if (error == 0) { 1188 SET_FOREACH(lihp, linux_ioctl_handler_set) 1189 linux_ioctl_unregister_handler(*lihp); 1190 mtx_destroy(&futex_mtx); 1191 if (bootverbose) 1192 printf("Linux ELF exec handler removed\n"); 1193 } else 1194 printf("Could not deinstall ELF interpreter entry\n"); 1195 break; 1196 default: 1197 return (EOPNOTSUPP); 1198 } 1199 return (error); 1200 } 1201 1202 static moduledata_t linux_elf_mod = { 1203 "linuxelf", 1204 linux_elf_modevent, 1205 0 1206 }; 1207 1208 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 1209 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1); 1210 FEATURE(linux, "Linux 32bit support"); 1211