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 "opt_compat.h" 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #ifndef COMPAT_FREEBSD32 41 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!" 42 #endif 43 44 #define __ELF_WORD_SIZE 32 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/exec.h> 49 #include <sys/fcntl.h> 50 #include <sys/imgact.h> 51 #include <sys/imgact_elf.h> 52 #include <sys/kernel.h> 53 #include <sys/lock.h> 54 #include <sys/malloc.h> 55 #include <sys/module.h> 56 #include <sys/mutex.h> 57 #include <sys/proc.h> 58 #include <sys/resourcevar.h> 59 #include <sys/signalvar.h> 60 #include <sys/syscallsubr.h> 61 #include <sys/sysctl.h> 62 #include <sys/sysent.h> 63 #include <sys/sysproto.h> 64 #include <sys/vnode.h> 65 #include <sys/eventhandler.h> 66 67 #include <vm/vm.h> 68 #include <vm/pmap.h> 69 #include <vm/vm_extern.h> 70 #include <vm/vm_map.h> 71 #include <vm/vm_object.h> 72 #include <vm/vm_page.h> 73 #include <vm/vm_param.h> 74 75 #include <machine/cpu.h> 76 #include <machine/md_var.h> 77 #include <machine/pcb.h> 78 #include <machine/specialreg.h> 79 #include <machine/trap.h> 80 81 #include <amd64/linux32/linux.h> 82 #include <amd64/linux32/linux32_proto.h> 83 #include <compat/linux/linux_emul.h> 84 #include <compat/linux/linux_futex.h> 85 #include <compat/linux/linux_ioctl.h> 86 #include <compat/linux/linux_mib.h> 87 #include <compat/linux/linux_misc.h> 88 #include <compat/linux/linux_signal.h> 89 #include <compat/linux/linux_util.h> 90 #include <compat/linux/linux_vdso.h> 91 92 MODULE_VERSION(linux, 1); 93 94 const char *linux_kplatform; 95 static int linux_szsigcode; 96 static vm_object_t linux_shared_page_obj; 97 static char *linux_shared_page_mapping; 98 extern char _binary_linux32_locore_o_start; 99 extern char _binary_linux32_locore_o_end; 100 101 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL]; 102 103 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 104 105 static int linux_fixup_elf(register_t **stack_base, 106 struct image_params *iparams); 107 static register_t *linux_copyout_strings(struct image_params *imgp); 108 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask); 109 static void linux_exec_setregs(struct thread *td, 110 struct image_params *imgp, u_long stack); 111 static void linux32_fixlimit(struct rlimit *rl, int which); 112 static bool linux32_trans_osrel(const Elf_Note *note, int32_t *osrel); 113 static void linux_vdso_install(void *param); 114 static void linux_vdso_deinstall(void *param); 115 116 #define LINUX_T_UNKNOWN 255 117 static int _bsd_to_linux_trapcode[] = { 118 LINUX_T_UNKNOWN, /* 0 */ 119 6, /* 1 T_PRIVINFLT */ 120 LINUX_T_UNKNOWN, /* 2 */ 121 3, /* 3 T_BPTFLT */ 122 LINUX_T_UNKNOWN, /* 4 */ 123 LINUX_T_UNKNOWN, /* 5 */ 124 16, /* 6 T_ARITHTRAP */ 125 254, /* 7 T_ASTFLT */ 126 LINUX_T_UNKNOWN, /* 8 */ 127 13, /* 9 T_PROTFLT */ 128 1, /* 10 T_TRCTRAP */ 129 LINUX_T_UNKNOWN, /* 11 */ 130 14, /* 12 T_PAGEFLT */ 131 LINUX_T_UNKNOWN, /* 13 */ 132 17, /* 14 T_ALIGNFLT */ 133 LINUX_T_UNKNOWN, /* 15 */ 134 LINUX_T_UNKNOWN, /* 16 */ 135 LINUX_T_UNKNOWN, /* 17 */ 136 0, /* 18 T_DIVIDE */ 137 2, /* 19 T_NMI */ 138 4, /* 20 T_OFLOW */ 139 5, /* 21 T_BOUND */ 140 7, /* 22 T_DNA */ 141 8, /* 23 T_DOUBLEFLT */ 142 9, /* 24 T_FPOPFLT */ 143 10, /* 25 T_TSSFLT */ 144 11, /* 26 T_SEGNPFLT */ 145 12, /* 27 T_STKFLT */ 146 18, /* 28 T_MCHK */ 147 19, /* 29 T_XMMFLT */ 148 15 /* 30 T_RESERVED */ 149 }; 150 #define bsd_to_linux_trapcode(code) \ 151 ((code)<nitems(_bsd_to_linux_trapcode)? \ 152 _bsd_to_linux_trapcode[(code)]: \ 153 LINUX_T_UNKNOWN) 154 155 struct linux32_ps_strings { 156 u_int32_t ps_argvstr; /* first of 0 or more argument strings */ 157 u_int ps_nargvstr; /* the number of argument strings */ 158 u_int32_t ps_envstr; /* first of 0 or more environment strings */ 159 u_int ps_nenvstr; /* the number of environment strings */ 160 }; 161 162 LINUX_VDSO_SYM_INTPTR(linux32_sigcode); 163 LINUX_VDSO_SYM_INTPTR(linux32_rt_sigcode); 164 LINUX_VDSO_SYM_INTPTR(linux32_vsyscall); 165 LINUX_VDSO_SYM_CHAR(linux_platform); 166 167 /* 168 * If FreeBSD & Linux have a difference of opinion about what a trap 169 * means, deal with it here. 170 * 171 * MPSAFE 172 */ 173 static int 174 linux_translate_traps(int signal, int trap_code) 175 { 176 if (signal != SIGBUS) 177 return (signal); 178 switch (trap_code) { 179 case T_PROTFLT: 180 case T_TSSFLT: 181 case T_DOUBLEFLT: 182 case T_PAGEFLT: 183 return (SIGSEGV); 184 default: 185 return (signal); 186 } 187 } 188 189 static int 190 linux_fixup_elf(register_t **stack_base, struct image_params *imgp) 191 { 192 Elf32_Auxargs *args; 193 Elf32_Auxinfo *argarray, *pos; 194 Elf32_Addr *auxbase, *base; 195 struct linux32_ps_strings *arginfo; 196 int error, issetugid; 197 198 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; 199 200 KASSERT(curthread->td_proc == imgp->proc, 201 ("unsafe linux_fixup_elf(), should be curproc")); 202 base = (Elf32_Addr *)*stack_base; 203 args = (Elf32_Auxargs *)imgp->auxargs; 204 auxbase = base + (imgp->args->argc + 1 + imgp->args->envc + 1); 205 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 206 M_WAITOK | M_ZERO); 207 208 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0; 209 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, 210 imgp->proc->p_sysent->sv_shared_page_base); 211 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux32_vsyscall); 212 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 213 214 /* 215 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, 216 * as it has appeared in the 2.4.0-rc7 first time. 217 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), 218 * glibc falls back to the hard-coded CLK_TCK value when aux entry 219 * is not present. 220 * Also see linux_times() implementation. 221 */ 222 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) 223 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 224 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 225 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 226 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 227 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 228 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 229 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 230 AUXARGS_ENTRY(pos, AT_BASE, args->base); 231 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 232 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 233 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 234 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 235 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 236 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 237 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary)); 238 if (imgp->execpathp != 0) 239 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp)); 240 if (args->execfd != -1) 241 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 242 AUXARGS_ENTRY(pos, AT_NULL, 0); 243 244 free(imgp->auxargs, M_TEMP); 245 imgp->auxargs = NULL; 246 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 247 248 error = copyout(&argarray[0], auxbase, 249 sizeof(*argarray) * LINUX_AT_COUNT); 250 free(argarray, M_TEMP); 251 if (error != 0) 252 return (error); 253 254 base--; 255 if (suword32(base, (uint32_t)imgp->args->argc) == -1) 256 return (EFAULT); 257 *stack_base = (register_t *)base; 258 return (0); 259 } 260 261 static void 262 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 263 { 264 struct thread *td = curthread; 265 struct proc *p = td->td_proc; 266 struct sigacts *psp; 267 struct trapframe *regs; 268 struct l_rt_sigframe *fp, frame; 269 int oonstack; 270 int sig; 271 int code; 272 273 sig = ksi->ksi_signo; 274 code = ksi->ksi_code; 275 PROC_LOCK_ASSERT(p, MA_OWNED); 276 psp = p->p_sigacts; 277 mtx_assert(&psp->ps_mtx, MA_OWNED); 278 regs = td->td_frame; 279 oonstack = sigonstack(regs->tf_rsp); 280 281 /* Allocate space for the signal handler context. */ 282 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 283 SIGISMEMBER(psp->ps_sigonstack, sig)) { 284 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 285 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe)); 286 } else 287 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1; 288 mtx_unlock(&psp->ps_mtx); 289 290 /* Build the argument list for the signal handler. */ 291 sig = bsd_to_linux_signal(sig); 292 293 bzero(&frame, sizeof(frame)); 294 295 frame.sf_handler = PTROUT(catcher); 296 frame.sf_sig = sig; 297 frame.sf_siginfo = PTROUT(&fp->sf_si); 298 frame.sf_ucontext = PTROUT(&fp->sf_sc); 299 300 /* Fill in POSIX parts. */ 301 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig); 302 303 /* 304 * Build the signal context to be used by sigreturn and libgcc unwind. 305 */ 306 frame.sf_sc.uc_flags = 0; /* XXX ??? */ 307 frame.sf_sc.uc_link = 0; /* XXX ??? */ 308 309 frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 310 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 311 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 312 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 313 PROC_UNLOCK(p); 314 315 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask); 316 317 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask; 318 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi; 319 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi; 320 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp; 321 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx; 322 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp; 323 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx; 324 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx; 325 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax; 326 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip; 327 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 328 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs; 329 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs; 330 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es; 331 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds; 332 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags; 333 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp; 334 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss; 335 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err; 336 frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 337 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 338 339 if (copyout(&frame, fp, sizeof(frame)) != 0) { 340 /* 341 * Process has trashed its stack; give it an illegal 342 * instruction to halt it in its tracks. 343 */ 344 PROC_LOCK(p); 345 sigexit(td, SIGILL); 346 } 347 348 /* Build context to run handler in. */ 349 regs->tf_rsp = PTROUT(fp); 350 regs->tf_rip = linux32_rt_sigcode; 351 regs->tf_rflags &= ~(PSL_T | PSL_D); 352 regs->tf_cs = _ucode32sel; 353 regs->tf_ss = _udatasel; 354 regs->tf_ds = _udatasel; 355 regs->tf_es = _udatasel; 356 regs->tf_fs = _ufssel; 357 regs->tf_gs = _ugssel; 358 regs->tf_flags = TF_HASSEGS; 359 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 360 PROC_LOCK(p); 361 mtx_lock(&psp->ps_mtx); 362 } 363 364 365 /* 366 * Send an interrupt to process. 367 * 368 * Stack is set up to allow sigcode stored 369 * in u. to call routine, followed by kcall 370 * to sigreturn routine below. After sigreturn 371 * resets the signal mask, the stack, and the 372 * frame pointer, it returns to the user 373 * specified pc, psl. 374 */ 375 static void 376 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 377 { 378 struct thread *td = curthread; 379 struct proc *p = td->td_proc; 380 struct sigacts *psp; 381 struct trapframe *regs; 382 struct l_sigframe *fp, frame; 383 l_sigset_t lmask; 384 int oonstack; 385 int sig, code; 386 387 sig = ksi->ksi_signo; 388 code = ksi->ksi_code; 389 PROC_LOCK_ASSERT(p, MA_OWNED); 390 psp = p->p_sigacts; 391 mtx_assert(&psp->ps_mtx, MA_OWNED); 392 if (SIGISMEMBER(psp->ps_siginfo, sig)) { 393 /* Signal handler installed with SA_SIGINFO. */ 394 linux_rt_sendsig(catcher, ksi, mask); 395 return; 396 } 397 398 regs = td->td_frame; 399 oonstack = sigonstack(regs->tf_rsp); 400 401 /* Allocate space for the signal handler context. */ 402 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 403 SIGISMEMBER(psp->ps_sigonstack, sig)) { 404 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 405 td->td_sigstk.ss_size - sizeof(struct l_sigframe)); 406 } else 407 fp = (struct l_sigframe *)regs->tf_rsp - 1; 408 mtx_unlock(&psp->ps_mtx); 409 PROC_UNLOCK(p); 410 411 /* Build the argument list for the signal handler. */ 412 sig = bsd_to_linux_signal(sig); 413 414 bzero(&frame, sizeof(frame)); 415 416 frame.sf_handler = PTROUT(catcher); 417 frame.sf_sig = sig; 418 419 bsd_to_linux_sigset(mask, &lmask); 420 421 /* Build the signal context to be used by sigreturn. */ 422 frame.sf_sc.sc_mask = lmask.__mask; 423 frame.sf_sc.sc_gs = regs->tf_gs; 424 frame.sf_sc.sc_fs = regs->tf_fs; 425 frame.sf_sc.sc_es = regs->tf_es; 426 frame.sf_sc.sc_ds = regs->tf_ds; 427 frame.sf_sc.sc_edi = regs->tf_rdi; 428 frame.sf_sc.sc_esi = regs->tf_rsi; 429 frame.sf_sc.sc_ebp = regs->tf_rbp; 430 frame.sf_sc.sc_ebx = regs->tf_rbx; 431 frame.sf_sc.sc_esp = regs->tf_rsp; 432 frame.sf_sc.sc_edx = regs->tf_rdx; 433 frame.sf_sc.sc_ecx = regs->tf_rcx; 434 frame.sf_sc.sc_eax = regs->tf_rax; 435 frame.sf_sc.sc_eip = regs->tf_rip; 436 frame.sf_sc.sc_cs = regs->tf_cs; 437 frame.sf_sc.sc_eflags = regs->tf_rflags; 438 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp; 439 frame.sf_sc.sc_ss = regs->tf_ss; 440 frame.sf_sc.sc_err = regs->tf_err; 441 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 442 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code); 443 444 frame.sf_extramask[0] = lmask.__mask; 445 446 if (copyout(&frame, fp, sizeof(frame)) != 0) { 447 /* 448 * Process has trashed its stack; give it an illegal 449 * instruction to halt it in its tracks. 450 */ 451 PROC_LOCK(p); 452 sigexit(td, SIGILL); 453 } 454 455 /* Build context to run handler in. */ 456 regs->tf_rsp = PTROUT(fp); 457 regs->tf_rip = linux32_sigcode; 458 regs->tf_rflags &= ~(PSL_T | PSL_D); 459 regs->tf_cs = _ucode32sel; 460 regs->tf_ss = _udatasel; 461 regs->tf_ds = _udatasel; 462 regs->tf_es = _udatasel; 463 regs->tf_fs = _ufssel; 464 regs->tf_gs = _ugssel; 465 regs->tf_flags = TF_HASSEGS; 466 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 467 PROC_LOCK(p); 468 mtx_lock(&psp->ps_mtx); 469 } 470 471 /* 472 * System call to cleanup state after a signal 473 * has been taken. Reset signal mask and 474 * stack state from context left by sendsig (above). 475 * Return to previous pc and psl as specified by 476 * context left by sendsig. Check carefully to 477 * make sure that the user has not modified the 478 * psl to gain improper privileges or to cause 479 * a machine fault. 480 */ 481 int 482 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) 483 { 484 struct l_sigframe frame; 485 struct trapframe *regs; 486 sigset_t bmask; 487 l_sigset_t lmask; 488 int eflags; 489 ksiginfo_t ksi; 490 491 regs = td->td_frame; 492 493 /* 494 * The trampoline code hands us the sigframe. 495 * It is unsafe to keep track of it ourselves, in the event that a 496 * program jumps out of a signal handler. 497 */ 498 if (copyin(args->sfp, &frame, sizeof(frame)) != 0) 499 return (EFAULT); 500 501 /* Check for security violations. */ 502 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 503 eflags = frame.sf_sc.sc_eflags; 504 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 505 return(EINVAL); 506 507 /* 508 * Don't allow users to load a valid privileged %cs. Let the 509 * hardware check for invalid selectors, excess privilege in 510 * other selectors, invalid %eip's and invalid %esp's. 511 */ 512 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 513 if (!CS_SECURE(frame.sf_sc.sc_cs)) { 514 ksiginfo_init_trap(&ksi); 515 ksi.ksi_signo = SIGBUS; 516 ksi.ksi_code = BUS_OBJERR; 517 ksi.ksi_trapno = T_PROTFLT; 518 ksi.ksi_addr = (void *)regs->tf_rip; 519 trapsignal(td, &ksi); 520 return(EINVAL); 521 } 522 523 lmask.__mask = frame.sf_sc.sc_mask; 524 lmask.__mask = frame.sf_extramask[0]; 525 linux_to_bsd_sigset(&lmask, &bmask); 526 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 527 528 /* Restore signal context. */ 529 regs->tf_rdi = frame.sf_sc.sc_edi; 530 regs->tf_rsi = frame.sf_sc.sc_esi; 531 regs->tf_rbp = frame.sf_sc.sc_ebp; 532 regs->tf_rbx = frame.sf_sc.sc_ebx; 533 regs->tf_rdx = frame.sf_sc.sc_edx; 534 regs->tf_rcx = frame.sf_sc.sc_ecx; 535 regs->tf_rax = frame.sf_sc.sc_eax; 536 regs->tf_rip = frame.sf_sc.sc_eip; 537 regs->tf_cs = frame.sf_sc.sc_cs; 538 regs->tf_ds = frame.sf_sc.sc_ds; 539 regs->tf_es = frame.sf_sc.sc_es; 540 regs->tf_fs = frame.sf_sc.sc_fs; 541 regs->tf_gs = frame.sf_sc.sc_gs; 542 regs->tf_rflags = eflags; 543 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal; 544 regs->tf_ss = frame.sf_sc.sc_ss; 545 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 546 547 return (EJUSTRETURN); 548 } 549 550 /* 551 * System call to cleanup state after a signal 552 * has been taken. Reset signal mask and 553 * stack state from context left by rt_sendsig (above). 554 * Return to previous pc and psl as specified by 555 * context left by sendsig. Check carefully to 556 * make sure that the user has not modified the 557 * psl to gain improper privileges or to cause 558 * a machine fault. 559 */ 560 int 561 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 562 { 563 struct l_ucontext uc; 564 struct l_sigcontext *context; 565 sigset_t bmask; 566 l_stack_t *lss; 567 stack_t ss; 568 struct trapframe *regs; 569 int eflags; 570 ksiginfo_t ksi; 571 572 regs = td->td_frame; 573 574 /* 575 * The trampoline code hands us the ucontext. 576 * It is unsafe to keep track of it ourselves, in the event that a 577 * program jumps out of a signal handler. 578 */ 579 if (copyin(args->ucp, &uc, sizeof(uc)) != 0) 580 return (EFAULT); 581 582 context = &uc.uc_mcontext; 583 584 /* Check for security violations. */ 585 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 586 eflags = context->sc_eflags; 587 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 588 return(EINVAL); 589 590 /* 591 * Don't allow users to load a valid privileged %cs. Let the 592 * hardware check for invalid selectors, excess privilege in 593 * other selectors, invalid %eip's and invalid %esp's. 594 */ 595 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 596 if (!CS_SECURE(context->sc_cs)) { 597 ksiginfo_init_trap(&ksi); 598 ksi.ksi_signo = SIGBUS; 599 ksi.ksi_code = BUS_OBJERR; 600 ksi.ksi_trapno = T_PROTFLT; 601 ksi.ksi_addr = (void *)regs->tf_rip; 602 trapsignal(td, &ksi); 603 return(EINVAL); 604 } 605 606 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); 607 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 608 609 /* 610 * Restore signal context 611 */ 612 regs->tf_gs = context->sc_gs; 613 regs->tf_fs = context->sc_fs; 614 regs->tf_es = context->sc_es; 615 regs->tf_ds = context->sc_ds; 616 regs->tf_rdi = context->sc_edi; 617 regs->tf_rsi = context->sc_esi; 618 regs->tf_rbp = context->sc_ebp; 619 regs->tf_rbx = context->sc_ebx; 620 regs->tf_rdx = context->sc_edx; 621 regs->tf_rcx = context->sc_ecx; 622 regs->tf_rax = context->sc_eax; 623 regs->tf_rip = context->sc_eip; 624 regs->tf_cs = context->sc_cs; 625 regs->tf_rflags = eflags; 626 regs->tf_rsp = context->sc_esp_at_signal; 627 regs->tf_ss = context->sc_ss; 628 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 629 630 /* 631 * call sigaltstack & ignore results.. 632 */ 633 lss = &uc.uc_stack; 634 ss.ss_sp = PTRIN(lss->ss_sp); 635 ss.ss_size = lss->ss_size; 636 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags); 637 638 (void)kern_sigaltstack(td, &ss, NULL); 639 640 return (EJUSTRETURN); 641 } 642 643 static int 644 linux32_fetch_syscall_args(struct thread *td) 645 { 646 struct proc *p; 647 struct trapframe *frame; 648 struct syscall_args *sa; 649 650 p = td->td_proc; 651 frame = td->td_frame; 652 sa = &td->td_sa; 653 654 sa->args[0] = frame->tf_rbx; 655 sa->args[1] = frame->tf_rcx; 656 sa->args[2] = frame->tf_rdx; 657 sa->args[3] = frame->tf_rsi; 658 sa->args[4] = frame->tf_rdi; 659 sa->args[5] = frame->tf_rbp; /* Unconfirmed */ 660 sa->code = frame->tf_rax; 661 662 if (sa->code >= p->p_sysent->sv_size) 663 /* nosys */ 664 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 665 else 666 sa->callp = &p->p_sysent->sv_table[sa->code]; 667 sa->narg = sa->callp->sy_narg; 668 669 td->td_retval[0] = 0; 670 td->td_retval[1] = frame->tf_rdx; 671 672 return (0); 673 } 674 675 /* 676 * Clear registers on exec 677 * XXX copied from ia32_signal.c. 678 */ 679 static void 680 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack) 681 { 682 struct trapframe *regs = td->td_frame; 683 struct pcb *pcb = td->td_pcb; 684 register_t saved_rflags; 685 686 regs = td->td_frame; 687 pcb = td->td_pcb; 688 689 if (td->td_proc->p_md.md_ldt != NULL) 690 user_ldt_free(td); 691 692 critical_enter(); 693 wrmsr(MSR_FSBASE, 0); 694 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */ 695 pcb->pcb_fsbase = 0; 696 pcb->pcb_gsbase = 0; 697 critical_exit(); 698 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 699 700 saved_rflags = regs->tf_rflags & PSL_T; 701 bzero((char *)regs, sizeof(struct trapframe)); 702 regs->tf_rip = imgp->entry_addr; 703 regs->tf_rsp = stack; 704 regs->tf_rflags = PSL_USER | saved_rflags; 705 regs->tf_gs = _ugssel; 706 regs->tf_fs = _ufssel; 707 regs->tf_es = _udatasel; 708 regs->tf_ds = _udatasel; 709 regs->tf_ss = _udatasel; 710 regs->tf_flags = TF_HASSEGS; 711 regs->tf_cs = _ucode32sel; 712 regs->tf_rbx = imgp->ps_strings; 713 714 fpstate_drop(td); 715 716 /* Do full restore on return so that we can change to a different %cs */ 717 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET); 718 } 719 720 /* 721 * XXX copied from ia32_sysvec.c. 722 */ 723 static register_t * 724 linux_copyout_strings(struct image_params *imgp) 725 { 726 int argc, envc; 727 u_int32_t *vectp; 728 char *stringp, *destp; 729 u_int32_t *stack_base; 730 struct linux32_ps_strings *arginfo; 731 char canary[LINUX_AT_RANDOM_LEN]; 732 size_t execpath_len; 733 734 /* Calculate string base and vector table pointers. */ 735 if (imgp->execpath != NULL && imgp->auxargs != NULL) 736 execpath_len = strlen(imgp->execpath) + 1; 737 else 738 execpath_len = 0; 739 740 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; 741 destp = (caddr_t)arginfo - SPARE_USRSPACE - 742 roundup(sizeof(canary), sizeof(char *)) - 743 roundup(execpath_len, sizeof(char *)) - 744 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *)); 745 746 if (execpath_len != 0) { 747 imgp->execpathp = (uintptr_t)arginfo - execpath_len; 748 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 749 } 750 751 /* Prepare the canary for SSP. */ 752 arc4rand(canary, sizeof(canary), 0); 753 imgp->canary = (uintptr_t)arginfo - 754 roundup(execpath_len, sizeof(char *)) - 755 roundup(sizeof(canary), sizeof(char *)); 756 copyout(canary, (void *)imgp->canary, sizeof(canary)); 757 758 vectp = (uint32_t *)destp; 759 if (imgp->auxargs) { 760 /* 761 * Allocate room on the stack for the ELF auxargs 762 * array. It has LINUX_AT_COUNT entries. 763 */ 764 vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf32_Auxinfo), 765 sizeof(*vectp)); 766 } 767 768 /* 769 * Allocate room for the argv[] and env vectors including the 770 * terminating NULL pointers. 771 */ 772 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 773 774 /* vectp also becomes our initial stack base. */ 775 stack_base = vectp; 776 777 stringp = imgp->args->begin_argv; 778 argc = imgp->args->argc; 779 envc = imgp->args->envc; 780 /* Copy out strings - arguments and environment. */ 781 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 782 783 /* Fill in "ps_strings" struct for ps, w, etc. */ 784 suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp); 785 suword32(&arginfo->ps_nargvstr, argc); 786 787 /* Fill in argument portion of vector table. */ 788 for (; argc > 0; --argc) { 789 suword32(vectp++, (uint32_t)(intptr_t)destp); 790 while (*stringp++ != 0) 791 destp++; 792 destp++; 793 } 794 795 /* A null vector table pointer separates the argp's from the envp's. */ 796 suword32(vectp++, 0); 797 798 suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp); 799 suword32(&arginfo->ps_nenvstr, envc); 800 801 /* Fill in environment portion of vector table. */ 802 for (; envc > 0; --envc) { 803 suword32(vectp++, (uint32_t)(intptr_t)destp); 804 while (*stringp++ != 0) 805 destp++; 806 destp++; 807 } 808 809 /* The end of the vector table is a null pointer. */ 810 suword32(vectp, 0); 811 812 return ((register_t *)stack_base); 813 } 814 815 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 0, 816 "32-bit Linux emulation"); 817 818 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ; 819 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW, 820 &linux32_maxdsiz, 0, ""); 821 static u_long linux32_maxssiz = LINUX32_MAXSSIZ; 822 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW, 823 &linux32_maxssiz, 0, ""); 824 static u_long linux32_maxvmem = LINUX32_MAXVMEM; 825 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW, 826 &linux32_maxvmem, 0, ""); 827 828 static void 829 linux32_fixlimit(struct rlimit *rl, int which) 830 { 831 832 switch (which) { 833 case RLIMIT_DATA: 834 if (linux32_maxdsiz != 0) { 835 if (rl->rlim_cur > linux32_maxdsiz) 836 rl->rlim_cur = linux32_maxdsiz; 837 if (rl->rlim_max > linux32_maxdsiz) 838 rl->rlim_max = linux32_maxdsiz; 839 } 840 break; 841 case RLIMIT_STACK: 842 if (linux32_maxssiz != 0) { 843 if (rl->rlim_cur > linux32_maxssiz) 844 rl->rlim_cur = linux32_maxssiz; 845 if (rl->rlim_max > linux32_maxssiz) 846 rl->rlim_max = linux32_maxssiz; 847 } 848 break; 849 case RLIMIT_VMEM: 850 if (linux32_maxvmem != 0) { 851 if (rl->rlim_cur > linux32_maxvmem) 852 rl->rlim_cur = linux32_maxvmem; 853 if (rl->rlim_max > linux32_maxvmem) 854 rl->rlim_max = linux32_maxvmem; 855 } 856 break; 857 } 858 } 859 860 struct sysentvec elf_linux_sysvec = { 861 .sv_size = LINUX32_SYS_MAXSYSCALL, 862 .sv_table = linux32_sysent, 863 .sv_errsize = ELAST + 1, 864 .sv_errtbl = linux_errtbl, 865 .sv_transtrap = linux_translate_traps, 866 .sv_fixup = linux_fixup_elf, 867 .sv_sendsig = linux_sendsig, 868 .sv_sigcode = &_binary_linux32_locore_o_start, 869 .sv_szsigcode = &linux_szsigcode, 870 .sv_name = "Linux ELF32", 871 .sv_coredump = elf32_coredump, 872 .sv_imgact_try = linux_exec_imgact_try, 873 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 874 .sv_minuser = VM_MIN_ADDRESS, 875 .sv_maxuser = LINUX32_MAXUSER, 876 .sv_usrstack = LINUX32_USRSTACK, 877 .sv_psstrings = LINUX32_PS_STRINGS, 878 .sv_stackprot = VM_PROT_ALL, 879 .sv_copyout_strings = linux_copyout_strings, 880 .sv_setregs = linux_exec_setregs, 881 .sv_fixlimit = linux32_fixlimit, 882 .sv_maxssiz = &linux32_maxssiz, 883 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP, 884 .sv_set_syscall_retval = cpu_set_syscall_retval, 885 .sv_fetch_syscall_args = linux32_fetch_syscall_args, 886 .sv_syscallnames = NULL, 887 .sv_shared_page_base = LINUX32_SHAREDPAGE, 888 .sv_shared_page_len = PAGE_SIZE, 889 .sv_schedtail = linux_schedtail, 890 .sv_thread_detach = linux_thread_detach, 891 .sv_trap = NULL, 892 }; 893 894 static void 895 linux_vdso_install(void *param) 896 { 897 898 linux_szsigcode = (&_binary_linux32_locore_o_end - 899 &_binary_linux32_locore_o_start); 900 901 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 902 panic("Linux invalid vdso size\n"); 903 904 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 905 906 linux_shared_page_obj = __elfN(linux_shared_page_init) 907 (&linux_shared_page_mapping); 908 909 __elfN(linux_vdso_reloc)(&elf_linux_sysvec); 910 911 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 912 linux_szsigcode); 913 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 914 915 linux_kplatform = linux_shared_page_mapping + 916 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base); 917 } 918 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 919 linux_vdso_install, NULL); 920 921 static void 922 linux_vdso_deinstall(void *param) 923 { 924 925 __elfN(linux_shared_page_fini)(linux_shared_page_obj); 926 } 927 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 928 linux_vdso_deinstall, NULL); 929 930 static char GNU_ABI_VENDOR[] = "GNU"; 931 static int GNULINUX_ABI_DESC = 0; 932 933 static bool 934 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel) 935 { 936 const Elf32_Word *desc; 937 uintptr_t p; 938 939 p = (uintptr_t)(note + 1); 940 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 941 942 desc = (const Elf32_Word *)p; 943 if (desc[0] != GNULINUX_ABI_DESC) 944 return (false); 945 946 /* 947 * For Linux we encode osrel using the Linux convention of 948 * (version << 16) | (major << 8) | (minor) 949 * See macro in linux_mib.h 950 */ 951 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 952 953 return (true); 954 } 955 956 static Elf_Brandnote linux32_brandnote = { 957 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 958 .hdr.n_descsz = 16, /* XXX at least 16 */ 959 .hdr.n_type = 1, 960 .vendor = GNU_ABI_VENDOR, 961 .flags = BN_TRANSLATE_OSREL, 962 .trans_osrel = linux32_trans_osrel 963 }; 964 965 static Elf32_Brandinfo linux_brand = { 966 .brand = ELFOSABI_LINUX, 967 .machine = EM_386, 968 .compat_3_brand = "Linux", 969 .emul_path = "/compat/linux", 970 .interp_path = "/lib/ld-linux.so.1", 971 .sysvec = &elf_linux_sysvec, 972 .interp_newpath = NULL, 973 .brand_note = &linux32_brandnote, 974 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 975 }; 976 977 static Elf32_Brandinfo linux_glibc2brand = { 978 .brand = ELFOSABI_LINUX, 979 .machine = EM_386, 980 .compat_3_brand = "Linux", 981 .emul_path = "/compat/linux", 982 .interp_path = "/lib/ld-linux.so.2", 983 .sysvec = &elf_linux_sysvec, 984 .interp_newpath = NULL, 985 .brand_note = &linux32_brandnote, 986 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 987 }; 988 989 static Elf32_Brandinfo linux_muslbrand = { 990 .brand = ELFOSABI_LINUX, 991 .machine = EM_386, 992 .compat_3_brand = "Linux", 993 .emul_path = "/compat/linux", 994 .interp_path = "/lib/ld-musl-i386.so.1", 995 .sysvec = &elf_linux_sysvec, 996 .interp_newpath = NULL, 997 .brand_note = &linux32_brandnote, 998 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 999 }; 1000 1001 Elf32_Brandinfo *linux_brandlist[] = { 1002 &linux_brand, 1003 &linux_glibc2brand, 1004 &linux_muslbrand, 1005 NULL 1006 }; 1007 1008 static int 1009 linux_elf_modevent(module_t mod, int type, void *data) 1010 { 1011 Elf32_Brandinfo **brandinfo; 1012 int error; 1013 struct linux_ioctl_handler **lihp; 1014 1015 error = 0; 1016 1017 switch(type) { 1018 case MOD_LOAD: 1019 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1020 ++brandinfo) 1021 if (elf32_insert_brand_entry(*brandinfo) < 0) 1022 error = EINVAL; 1023 if (error == 0) { 1024 SET_FOREACH(lihp, linux_ioctl_handler_set) 1025 linux32_ioctl_register_handler(*lihp); 1026 LIST_INIT(&futex_list); 1027 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF); 1028 stclohz = (stathz ? stathz : hz); 1029 if (bootverbose) 1030 printf("Linux ELF exec handler installed\n"); 1031 } else 1032 printf("cannot insert Linux ELF brand handler\n"); 1033 break; 1034 case MOD_UNLOAD: 1035 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1036 ++brandinfo) 1037 if (elf32_brand_inuse(*brandinfo)) 1038 error = EBUSY; 1039 if (error == 0) { 1040 for (brandinfo = &linux_brandlist[0]; 1041 *brandinfo != NULL; ++brandinfo) 1042 if (elf32_remove_brand_entry(*brandinfo) < 0) 1043 error = EINVAL; 1044 } 1045 if (error == 0) { 1046 SET_FOREACH(lihp, linux_ioctl_handler_set) 1047 linux32_ioctl_unregister_handler(*lihp); 1048 mtx_destroy(&futex_mtx); 1049 if (bootverbose) 1050 printf("Linux ELF exec handler removed\n"); 1051 } else 1052 printf("Could not deinstall ELF interpreter entry\n"); 1053 break; 1054 default: 1055 return (EOPNOTSUPP); 1056 } 1057 return (error); 1058 } 1059 1060 static moduledata_t linux_elf_mod = { 1061 "linuxelf", 1062 linux_elf_modevent, 1063 0 1064 }; 1065 1066 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 1067 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1); 1068 FEATURE(linux, "Linux 32bit support"); 1069