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_ioctl.h> 85 #include <compat/linux/linux_mib.h> 86 #include <compat/linux/linux_misc.h> 87 #include <compat/linux/linux_signal.h> 88 #include <compat/linux/linux_util.h> 89 #include <compat/linux/linux_vdso.h> 90 91 MODULE_VERSION(linux, 1); 92 93 const char *linux_kplatform; 94 static int linux_szsigcode; 95 static vm_object_t linux_shared_page_obj; 96 static char *linux_shared_page_mapping; 97 extern char _binary_linux32_locore_o_start; 98 extern char _binary_linux32_locore_o_end; 99 100 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL]; 101 102 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 103 104 static int linux_fixup_elf(uintptr_t *stack_base, 105 struct image_params *iparams); 106 static int linux_copyout_strings(struct image_params *imgp, 107 uintptr_t *stack_base); 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, uintptr_t 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 static void linux32_set_syscall_retval(struct thread *td, int error); 116 117 #define LINUX_T_UNKNOWN 255 118 static int _bsd_to_linux_trapcode[] = { 119 LINUX_T_UNKNOWN, /* 0 */ 120 6, /* 1 T_PRIVINFLT */ 121 LINUX_T_UNKNOWN, /* 2 */ 122 3, /* 3 T_BPTFLT */ 123 LINUX_T_UNKNOWN, /* 4 */ 124 LINUX_T_UNKNOWN, /* 5 */ 125 16, /* 6 T_ARITHTRAP */ 126 254, /* 7 T_ASTFLT */ 127 LINUX_T_UNKNOWN, /* 8 */ 128 13, /* 9 T_PROTFLT */ 129 1, /* 10 T_TRCTRAP */ 130 LINUX_T_UNKNOWN, /* 11 */ 131 14, /* 12 T_PAGEFLT */ 132 LINUX_T_UNKNOWN, /* 13 */ 133 17, /* 14 T_ALIGNFLT */ 134 LINUX_T_UNKNOWN, /* 15 */ 135 LINUX_T_UNKNOWN, /* 16 */ 136 LINUX_T_UNKNOWN, /* 17 */ 137 0, /* 18 T_DIVIDE */ 138 2, /* 19 T_NMI */ 139 4, /* 20 T_OFLOW */ 140 5, /* 21 T_BOUND */ 141 7, /* 22 T_DNA */ 142 8, /* 23 T_DOUBLEFLT */ 143 9, /* 24 T_FPOPFLT */ 144 10, /* 25 T_TSSFLT */ 145 11, /* 26 T_SEGNPFLT */ 146 12, /* 27 T_STKFLT */ 147 18, /* 28 T_MCHK */ 148 19, /* 29 T_XMMFLT */ 149 15 /* 30 T_RESERVED */ 150 }; 151 #define bsd_to_linux_trapcode(code) \ 152 ((code)<nitems(_bsd_to_linux_trapcode)? \ 153 _bsd_to_linux_trapcode[(code)]: \ 154 LINUX_T_UNKNOWN) 155 156 struct linux32_ps_strings { 157 u_int32_t ps_argvstr; /* first of 0 or more argument strings */ 158 u_int ps_nargvstr; /* the number of argument strings */ 159 u_int32_t ps_envstr; /* first of 0 or more environment strings */ 160 u_int ps_nenvstr; /* the number of environment strings */ 161 }; 162 163 LINUX_VDSO_SYM_INTPTR(linux32_sigcode); 164 LINUX_VDSO_SYM_INTPTR(linux32_rt_sigcode); 165 LINUX_VDSO_SYM_INTPTR(linux32_vsyscall); 166 LINUX_VDSO_SYM_CHAR(linux_platform); 167 168 /* 169 * If FreeBSD & Linux have a difference of opinion about what a trap 170 * means, deal with it here. 171 * 172 * MPSAFE 173 */ 174 static int 175 linux_translate_traps(int signal, int trap_code) 176 { 177 if (signal != SIGBUS) 178 return (signal); 179 switch (trap_code) { 180 case T_PROTFLT: 181 case T_TSSFLT: 182 case T_DOUBLEFLT: 183 case T_PAGEFLT: 184 return (SIGSEGV); 185 default: 186 return (signal); 187 } 188 } 189 190 static int 191 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) 192 { 193 Elf32_Auxargs *args; 194 Elf32_Auxinfo *argarray, *pos; 195 int error, issetugid; 196 197 args = (Elf32_Auxargs *)imgp->auxargs; 198 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 199 M_WAITOK | M_ZERO); 200 201 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0; 202 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux32_vsyscall); 203 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, 204 imgp->proc->p_sysent->sv_shared_page_base); 205 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 206 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 207 208 /* 209 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, 210 * as it has appeared in the 2.4.0-rc7 first time. 211 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), 212 * glibc falls back to the hard-coded CLK_TCK value when aux entry 213 * is not present. 214 * Also see linux_times() implementation. 215 */ 216 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) 217 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 218 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 219 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 220 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 221 AUXARGS_ENTRY(pos, AT_BASE, args->base); 222 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 223 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 224 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 225 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 226 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 227 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 228 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 229 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary)); 230 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0); 231 if (imgp->execpathp != 0) 232 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp)); 233 if (args->execfd != -1) 234 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 235 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 236 AUXARGS_ENTRY(pos, AT_NULL, 0); 237 238 free(imgp->auxargs, M_TEMP); 239 imgp->auxargs = NULL; 240 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 241 242 error = copyout(argarray, (void *)base, 243 sizeof(*argarray) * LINUX_AT_COUNT); 244 free(argarray, M_TEMP); 245 return (error); 246 } 247 248 static int 249 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) 250 { 251 Elf32_Addr *base; 252 253 base = (Elf32_Addr *)*stack_base; 254 base--; 255 if (suword32(base, (uint32_t)imgp->args->argc) == -1) 256 return (EFAULT); 257 *stack_base = (uintptr_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 * Send an interrupt to process. 366 * 367 * Stack is set up to allow sigcode stored 368 * in u. to call routine, followed by kcall 369 * to sigreturn routine below. After sigreturn 370 * resets the signal mask, the stack, and the 371 * frame pointer, it returns to the user 372 * specified pc, psl. 373 */ 374 static void 375 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 376 { 377 struct thread *td = curthread; 378 struct proc *p = td->td_proc; 379 struct sigacts *psp; 380 struct trapframe *regs; 381 struct l_sigframe *fp, frame; 382 l_sigset_t lmask; 383 int oonstack; 384 int sig, code; 385 386 sig = ksi->ksi_signo; 387 code = ksi->ksi_code; 388 PROC_LOCK_ASSERT(p, MA_OWNED); 389 psp = p->p_sigacts; 390 mtx_assert(&psp->ps_mtx, MA_OWNED); 391 if (SIGISMEMBER(psp->ps_siginfo, sig)) { 392 /* Signal handler installed with SA_SIGINFO. */ 393 linux_rt_sendsig(catcher, ksi, mask); 394 return; 395 } 396 397 regs = td->td_frame; 398 oonstack = sigonstack(regs->tf_rsp); 399 400 /* Allocate space for the signal handler context. */ 401 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 402 SIGISMEMBER(psp->ps_sigonstack, sig)) { 403 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 404 td->td_sigstk.ss_size - sizeof(struct l_sigframe)); 405 } else 406 fp = (struct l_sigframe *)regs->tf_rsp - 1; 407 mtx_unlock(&psp->ps_mtx); 408 PROC_UNLOCK(p); 409 410 /* Build the argument list for the signal handler. */ 411 sig = bsd_to_linux_signal(sig); 412 413 bzero(&frame, sizeof(frame)); 414 415 frame.sf_handler = PTROUT(catcher); 416 frame.sf_sig = sig; 417 418 bsd_to_linux_sigset(mask, &lmask); 419 420 /* Build the signal context to be used by sigreturn. */ 421 frame.sf_sc.sc_mask = lmask.__mask; 422 frame.sf_sc.sc_gs = regs->tf_gs; 423 frame.sf_sc.sc_fs = regs->tf_fs; 424 frame.sf_sc.sc_es = regs->tf_es; 425 frame.sf_sc.sc_ds = regs->tf_ds; 426 frame.sf_sc.sc_edi = regs->tf_rdi; 427 frame.sf_sc.sc_esi = regs->tf_rsi; 428 frame.sf_sc.sc_ebp = regs->tf_rbp; 429 frame.sf_sc.sc_ebx = regs->tf_rbx; 430 frame.sf_sc.sc_esp = regs->tf_rsp; 431 frame.sf_sc.sc_edx = regs->tf_rdx; 432 frame.sf_sc.sc_ecx = regs->tf_rcx; 433 frame.sf_sc.sc_eax = regs->tf_rax; 434 frame.sf_sc.sc_eip = regs->tf_rip; 435 frame.sf_sc.sc_cs = regs->tf_cs; 436 frame.sf_sc.sc_eflags = regs->tf_rflags; 437 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp; 438 frame.sf_sc.sc_ss = regs->tf_ss; 439 frame.sf_sc.sc_err = regs->tf_err; 440 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 441 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code); 442 443 frame.sf_extramask[0] = lmask.__mask; 444 445 if (copyout(&frame, fp, sizeof(frame)) != 0) { 446 /* 447 * Process has trashed its stack; give it an illegal 448 * instruction to halt it in its tracks. 449 */ 450 PROC_LOCK(p); 451 sigexit(td, SIGILL); 452 } 453 454 /* Build context to run handler in. */ 455 regs->tf_rsp = PTROUT(fp); 456 regs->tf_rip = linux32_sigcode; 457 regs->tf_rflags &= ~(PSL_T | PSL_D); 458 regs->tf_cs = _ucode32sel; 459 regs->tf_ss = _udatasel; 460 regs->tf_ds = _udatasel; 461 regs->tf_es = _udatasel; 462 regs->tf_fs = _ufssel; 463 regs->tf_gs = _ugssel; 464 regs->tf_flags = TF_HASSEGS; 465 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 466 PROC_LOCK(p); 467 mtx_lock(&psp->ps_mtx); 468 } 469 470 /* 471 * System call to cleanup state after a signal 472 * has been taken. Reset signal mask and 473 * stack state from context left by sendsig (above). 474 * Return to previous pc and psl as specified by 475 * context left by sendsig. Check carefully to 476 * make sure that the user has not modified the 477 * psl to gain improper privileges or to cause 478 * a machine fault. 479 */ 480 int 481 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) 482 { 483 struct l_sigframe frame; 484 struct trapframe *regs; 485 sigset_t bmask; 486 l_sigset_t lmask; 487 int eflags; 488 ksiginfo_t ksi; 489 490 regs = td->td_frame; 491 492 /* 493 * The trampoline code hands us the sigframe. 494 * It is unsafe to keep track of it ourselves, in the event that a 495 * program jumps out of a signal handler. 496 */ 497 if (copyin(args->sfp, &frame, sizeof(frame)) != 0) 498 return (EFAULT); 499 500 /* Check for security violations. */ 501 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 502 eflags = frame.sf_sc.sc_eflags; 503 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 504 return(EINVAL); 505 506 /* 507 * Don't allow users to load a valid privileged %cs. Let the 508 * hardware check for invalid selectors, excess privilege in 509 * other selectors, invalid %eip's and invalid %esp's. 510 */ 511 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 512 if (!CS_SECURE(frame.sf_sc.sc_cs)) { 513 ksiginfo_init_trap(&ksi); 514 ksi.ksi_signo = SIGBUS; 515 ksi.ksi_code = BUS_OBJERR; 516 ksi.ksi_trapno = T_PROTFLT; 517 ksi.ksi_addr = (void *)regs->tf_rip; 518 trapsignal(td, &ksi); 519 return(EINVAL); 520 } 521 522 lmask.__mask = frame.sf_sc.sc_mask; 523 lmask.__mask = frame.sf_extramask[0]; 524 linux_to_bsd_sigset(&lmask, &bmask); 525 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 526 527 /* Restore signal context. */ 528 regs->tf_rdi = frame.sf_sc.sc_edi; 529 regs->tf_rsi = frame.sf_sc.sc_esi; 530 regs->tf_rbp = frame.sf_sc.sc_ebp; 531 regs->tf_rbx = frame.sf_sc.sc_ebx; 532 regs->tf_rdx = frame.sf_sc.sc_edx; 533 regs->tf_rcx = frame.sf_sc.sc_ecx; 534 regs->tf_rax = frame.sf_sc.sc_eax; 535 regs->tf_rip = frame.sf_sc.sc_eip; 536 regs->tf_cs = frame.sf_sc.sc_cs; 537 regs->tf_ds = frame.sf_sc.sc_ds; 538 regs->tf_es = frame.sf_sc.sc_es; 539 regs->tf_fs = frame.sf_sc.sc_fs; 540 regs->tf_gs = frame.sf_sc.sc_gs; 541 regs->tf_rflags = eflags; 542 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal; 543 regs->tf_ss = frame.sf_sc.sc_ss; 544 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 545 546 return (EJUSTRETURN); 547 } 548 549 /* 550 * System call to cleanup state after a signal 551 * has been taken. Reset signal mask and 552 * stack state from context left by rt_sendsig (above). 553 * Return to previous pc and psl as specified by 554 * context left by sendsig. Check carefully to 555 * make sure that the user has not modified the 556 * psl to gain improper privileges or to cause 557 * a machine fault. 558 */ 559 int 560 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 561 { 562 struct l_ucontext uc; 563 struct l_sigcontext *context; 564 sigset_t bmask; 565 l_stack_t *lss; 566 stack_t ss; 567 struct trapframe *regs; 568 int eflags; 569 ksiginfo_t ksi; 570 571 regs = td->td_frame; 572 573 /* 574 * The trampoline code hands us the ucontext. 575 * It is unsafe to keep track of it ourselves, in the event that a 576 * program jumps out of a signal handler. 577 */ 578 if (copyin(args->ucp, &uc, sizeof(uc)) != 0) 579 return (EFAULT); 580 581 context = &uc.uc_mcontext; 582 583 /* Check for security violations. */ 584 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 585 eflags = context->sc_eflags; 586 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 587 return(EINVAL); 588 589 /* 590 * Don't allow users to load a valid privileged %cs. Let the 591 * hardware check for invalid selectors, excess privilege in 592 * other selectors, invalid %eip's and invalid %esp's. 593 */ 594 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 595 if (!CS_SECURE(context->sc_cs)) { 596 ksiginfo_init_trap(&ksi); 597 ksi.ksi_signo = SIGBUS; 598 ksi.ksi_code = BUS_OBJERR; 599 ksi.ksi_trapno = T_PROTFLT; 600 ksi.ksi_addr = (void *)regs->tf_rip; 601 trapsignal(td, &ksi); 602 return(EINVAL); 603 } 604 605 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); 606 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 607 608 /* 609 * Restore signal context 610 */ 611 regs->tf_gs = context->sc_gs; 612 regs->tf_fs = context->sc_fs; 613 regs->tf_es = context->sc_es; 614 regs->tf_ds = context->sc_ds; 615 regs->tf_rdi = context->sc_edi; 616 regs->tf_rsi = context->sc_esi; 617 regs->tf_rbp = context->sc_ebp; 618 regs->tf_rbx = context->sc_ebx; 619 regs->tf_rdx = context->sc_edx; 620 regs->tf_rcx = context->sc_ecx; 621 regs->tf_rax = context->sc_eax; 622 regs->tf_rip = context->sc_eip; 623 regs->tf_cs = context->sc_cs; 624 regs->tf_rflags = eflags; 625 regs->tf_rsp = context->sc_esp_at_signal; 626 regs->tf_ss = context->sc_ss; 627 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 628 629 /* 630 * call sigaltstack & ignore results.. 631 */ 632 lss = &uc.uc_stack; 633 ss.ss_sp = PTRIN(lss->ss_sp); 634 ss.ss_size = lss->ss_size; 635 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags); 636 637 (void)kern_sigaltstack(td, &ss, NULL); 638 639 return (EJUSTRETURN); 640 } 641 642 static int 643 linux32_fetch_syscall_args(struct thread *td) 644 { 645 struct proc *p; 646 struct trapframe *frame; 647 struct syscall_args *sa; 648 649 p = td->td_proc; 650 frame = td->td_frame; 651 sa = &td->td_sa; 652 653 sa->args[0] = frame->tf_rbx; 654 sa->args[1] = frame->tf_rcx; 655 sa->args[2] = frame->tf_rdx; 656 sa->args[3] = frame->tf_rsi; 657 sa->args[4] = frame->tf_rdi; 658 sa->args[5] = frame->tf_rbp; /* Unconfirmed */ 659 sa->code = frame->tf_rax; 660 661 if (sa->code >= p->p_sysent->sv_size) 662 /* nosys */ 663 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 664 else 665 sa->callp = &p->p_sysent->sv_table[sa->code]; 666 667 td->td_retval[0] = 0; 668 td->td_retval[1] = frame->tf_rdx; 669 670 return (0); 671 } 672 673 static void 674 linux32_set_syscall_retval(struct thread *td, int error) 675 { 676 struct trapframe *frame = td->td_frame; 677 678 cpu_set_syscall_retval(td, error); 679 680 if (__predict_false(error != 0)) { 681 if (error != ERESTART && error != EJUSTRETURN) 682 frame->tf_rax = bsd_to_linux_errno(error); 683 } 684 } 685 686 /* 687 * Clear registers on exec 688 * XXX copied from ia32_signal.c. 689 */ 690 static void 691 linux_exec_setregs(struct thread *td, struct image_params *imgp, 692 uintptr_t stack) 693 { 694 struct trapframe *regs = td->td_frame; 695 struct pcb *pcb = td->td_pcb; 696 register_t saved_rflags; 697 698 regs = td->td_frame; 699 pcb = td->td_pcb; 700 701 if (td->td_proc->p_md.md_ldt != NULL) 702 user_ldt_free(td); 703 704 critical_enter(); 705 wrmsr(MSR_FSBASE, 0); 706 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */ 707 pcb->pcb_fsbase = 0; 708 pcb->pcb_gsbase = 0; 709 critical_exit(); 710 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 711 712 saved_rflags = regs->tf_rflags & PSL_T; 713 bzero((char *)regs, sizeof(struct trapframe)); 714 regs->tf_rip = imgp->entry_addr; 715 regs->tf_rsp = stack; 716 regs->tf_rflags = PSL_USER | saved_rflags; 717 regs->tf_gs = _ugssel; 718 regs->tf_fs = _ufssel; 719 regs->tf_es = _udatasel; 720 regs->tf_ds = _udatasel; 721 regs->tf_ss = _udatasel; 722 regs->tf_flags = TF_HASSEGS; 723 regs->tf_cs = _ucode32sel; 724 regs->tf_rbx = (register_t)imgp->ps_strings; 725 726 x86_clear_dbregs(pcb); 727 728 fpstate_drop(td); 729 730 /* Do full restore on return so that we can change to a different %cs */ 731 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET); 732 } 733 734 /* 735 * XXX copied from ia32_sysvec.c. 736 */ 737 static int 738 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 739 { 740 int argc, envc, error; 741 u_int32_t *vectp; 742 char *stringp; 743 uintptr_t destp, ustringp; 744 struct linux32_ps_strings *arginfo; 745 char canary[LINUX_AT_RANDOM_LEN]; 746 size_t execpath_len; 747 748 /* Calculate string base and vector table pointers. */ 749 if (imgp->execpath != NULL && imgp->auxargs != NULL) 750 execpath_len = strlen(imgp->execpath) + 1; 751 else 752 execpath_len = 0; 753 754 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; 755 destp = (uintptr_t)arginfo; 756 757 if (execpath_len != 0) { 758 destp -= execpath_len; 759 destp = rounddown2(destp, sizeof(uint32_t)); 760 imgp->execpathp = (void *)destp; 761 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 762 if (error != 0) 763 return (error); 764 } 765 766 /* Prepare the canary for SSP. */ 767 arc4rand(canary, sizeof(canary), 0); 768 destp -= roundup(sizeof(canary), sizeof(uint32_t)); 769 imgp->canary = (void *)destp; 770 error = copyout(canary, imgp->canary, sizeof(canary)); 771 if (error != 0) 772 return (error); 773 774 /* Allocate room for the argument and environment strings. */ 775 destp -= ARG_MAX - imgp->args->stringspace; 776 destp = rounddown2(destp, sizeof(uint32_t)); 777 ustringp = destp; 778 779 if (imgp->auxargs) { 780 /* 781 * Allocate room on the stack for the ELF auxargs 782 * array. It has LINUX_AT_COUNT entries. 783 */ 784 destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo); 785 destp = rounddown2(destp, sizeof(uint32_t)); 786 } 787 788 vectp = (uint32_t *)destp; 789 790 /* 791 * Allocate room for the argv[] and env vectors including the 792 * terminating NULL pointers. 793 */ 794 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 795 796 /* vectp also becomes our initial stack base. */ 797 *stack_base = (uintptr_t)vectp; 798 799 stringp = imgp->args->begin_argv; 800 argc = imgp->args->argc; 801 envc = imgp->args->envc; 802 803 /* Copy out strings - arguments and environment. */ 804 error = copyout(stringp, (void *)ustringp, 805 ARG_MAX - imgp->args->stringspace); 806 if (error != 0) 807 return (error); 808 809 /* Fill in "ps_strings" struct for ps, w, etc. */ 810 if (suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp) != 0 || 811 suword32(&arginfo->ps_nargvstr, argc) != 0) 812 return (EFAULT); 813 814 /* Fill in argument portion of vector table. */ 815 for (; argc > 0; --argc) { 816 if (suword32(vectp++, ustringp) != 0) 817 return (EFAULT); 818 while (*stringp++ != 0) 819 ustringp++; 820 ustringp++; 821 } 822 823 /* A null vector table pointer separates the argp's from the envp's. */ 824 if (suword32(vectp++, 0) != 0) 825 return (EFAULT); 826 827 if (suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp) != 0 || 828 suword32(&arginfo->ps_nenvstr, envc) != 0) 829 return (EFAULT); 830 831 /* Fill in environment portion of vector table. */ 832 for (; envc > 0; --envc) { 833 if (suword32(vectp++, ustringp) != 0) 834 return (EFAULT); 835 while (*stringp++ != 0) 836 ustringp++; 837 ustringp++; 838 } 839 840 /* The end of the vector table is a null pointer. */ 841 if (suword32(vectp, 0) != 0) 842 return (EFAULT); 843 844 if (imgp->auxargs) { 845 vectp++; 846 error = imgp->sysent->sv_copyout_auxargs(imgp, 847 (uintptr_t)vectp); 848 if (error != 0) 849 return (error); 850 } 851 852 return (0); 853 } 854 855 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 856 "32-bit Linux emulation"); 857 858 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ; 859 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW, 860 &linux32_maxdsiz, 0, ""); 861 static u_long linux32_maxssiz = LINUX32_MAXSSIZ; 862 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW, 863 &linux32_maxssiz, 0, ""); 864 static u_long linux32_maxvmem = LINUX32_MAXVMEM; 865 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW, 866 &linux32_maxvmem, 0, ""); 867 868 static void 869 linux32_fixlimit(struct rlimit *rl, int which) 870 { 871 872 switch (which) { 873 case RLIMIT_DATA: 874 if (linux32_maxdsiz != 0) { 875 if (rl->rlim_cur > linux32_maxdsiz) 876 rl->rlim_cur = linux32_maxdsiz; 877 if (rl->rlim_max > linux32_maxdsiz) 878 rl->rlim_max = linux32_maxdsiz; 879 } 880 break; 881 case RLIMIT_STACK: 882 if (linux32_maxssiz != 0) { 883 if (rl->rlim_cur > linux32_maxssiz) 884 rl->rlim_cur = linux32_maxssiz; 885 if (rl->rlim_max > linux32_maxssiz) 886 rl->rlim_max = linux32_maxssiz; 887 } 888 break; 889 case RLIMIT_VMEM: 890 if (linux32_maxvmem != 0) { 891 if (rl->rlim_cur > linux32_maxvmem) 892 rl->rlim_cur = linux32_maxvmem; 893 if (rl->rlim_max > linux32_maxvmem) 894 rl->rlim_max = linux32_maxvmem; 895 } 896 break; 897 } 898 } 899 900 struct sysentvec elf_linux_sysvec = { 901 .sv_size = LINUX32_SYS_MAXSYSCALL, 902 .sv_table = linux32_sysent, 903 .sv_transtrap = linux_translate_traps, 904 .sv_fixup = linux_fixup_elf, 905 .sv_sendsig = linux_sendsig, 906 .sv_sigcode = &_binary_linux32_locore_o_start, 907 .sv_szsigcode = &linux_szsigcode, 908 .sv_name = "Linux ELF32", 909 .sv_coredump = elf32_coredump, 910 .sv_imgact_try = linux_exec_imgact_try, 911 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 912 .sv_minuser = VM_MIN_ADDRESS, 913 .sv_maxuser = LINUX32_MAXUSER, 914 .sv_usrstack = LINUX32_USRSTACK, 915 .sv_psstrings = LINUX32_PS_STRINGS, 916 .sv_stackprot = VM_PROT_ALL, 917 .sv_copyout_auxargs = linux_copyout_auxargs, 918 .sv_copyout_strings = linux_copyout_strings, 919 .sv_setregs = linux_exec_setregs, 920 .sv_fixlimit = linux32_fixlimit, 921 .sv_maxssiz = &linux32_maxssiz, 922 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP, 923 .sv_set_syscall_retval = linux32_set_syscall_retval, 924 .sv_fetch_syscall_args = linux32_fetch_syscall_args, 925 .sv_syscallnames = NULL, 926 .sv_shared_page_base = LINUX32_SHAREDPAGE, 927 .sv_shared_page_len = PAGE_SIZE, 928 .sv_schedtail = linux_schedtail, 929 .sv_thread_detach = linux_thread_detach, 930 .sv_trap = NULL, 931 .sv_onexec = linux_on_exec, 932 .sv_onexit = linux_on_exit, 933 .sv_ontdexit = linux_thread_dtor, 934 }; 935 936 static void 937 linux_vdso_install(void *param) 938 { 939 940 linux_szsigcode = (&_binary_linux32_locore_o_end - 941 &_binary_linux32_locore_o_start); 942 943 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 944 panic("Linux invalid vdso size\n"); 945 946 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 947 948 linux_shared_page_obj = __elfN(linux_shared_page_init) 949 (&linux_shared_page_mapping); 950 951 __elfN(linux_vdso_reloc)(&elf_linux_sysvec); 952 953 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 954 linux_szsigcode); 955 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 956 957 linux_kplatform = linux_shared_page_mapping + 958 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base); 959 } 960 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 961 linux_vdso_install, NULL); 962 963 static void 964 linux_vdso_deinstall(void *param) 965 { 966 967 __elfN(linux_shared_page_fini)(linux_shared_page_obj, 968 linux_shared_page_mapping); 969 } 970 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 971 linux_vdso_deinstall, NULL); 972 973 static char GNU_ABI_VENDOR[] = "GNU"; 974 static int GNULINUX_ABI_DESC = 0; 975 976 static bool 977 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel) 978 { 979 const Elf32_Word *desc; 980 uintptr_t p; 981 982 p = (uintptr_t)(note + 1); 983 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 984 985 desc = (const Elf32_Word *)p; 986 if (desc[0] != GNULINUX_ABI_DESC) 987 return (false); 988 989 /* 990 * For Linux we encode osrel using the Linux convention of 991 * (version << 16) | (major << 8) | (minor) 992 * See macro in linux_mib.h 993 */ 994 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 995 996 return (true); 997 } 998 999 static Elf_Brandnote linux32_brandnote = { 1000 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 1001 .hdr.n_descsz = 16, /* XXX at least 16 */ 1002 .hdr.n_type = 1, 1003 .vendor = GNU_ABI_VENDOR, 1004 .flags = BN_TRANSLATE_OSREL, 1005 .trans_osrel = linux32_trans_osrel 1006 }; 1007 1008 static Elf32_Brandinfo linux_brand = { 1009 .brand = ELFOSABI_LINUX, 1010 .machine = EM_386, 1011 .compat_3_brand = "Linux", 1012 .emul_path = linux_emul_path, 1013 .interp_path = "/lib/ld-linux.so.1", 1014 .sysvec = &elf_linux_sysvec, 1015 .interp_newpath = NULL, 1016 .brand_note = &linux32_brandnote, 1017 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1018 }; 1019 1020 static Elf32_Brandinfo linux_glibc2brand = { 1021 .brand = ELFOSABI_LINUX, 1022 .machine = EM_386, 1023 .compat_3_brand = "Linux", 1024 .emul_path = linux_emul_path, 1025 .interp_path = "/lib/ld-linux.so.2", 1026 .sysvec = &elf_linux_sysvec, 1027 .interp_newpath = NULL, 1028 .brand_note = &linux32_brandnote, 1029 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1030 }; 1031 1032 static Elf32_Brandinfo linux_muslbrand = { 1033 .brand = ELFOSABI_LINUX, 1034 .machine = EM_386, 1035 .compat_3_brand = "Linux", 1036 .emul_path = linux_emul_path, 1037 .interp_path = "/lib/ld-musl-i386.so.1", 1038 .sysvec = &elf_linux_sysvec, 1039 .interp_newpath = NULL, 1040 .brand_note = &linux32_brandnote, 1041 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1042 }; 1043 1044 Elf32_Brandinfo *linux_brandlist[] = { 1045 &linux_brand, 1046 &linux_glibc2brand, 1047 &linux_muslbrand, 1048 NULL 1049 }; 1050 1051 static int 1052 linux_elf_modevent(module_t mod, int type, void *data) 1053 { 1054 Elf32_Brandinfo **brandinfo; 1055 int error; 1056 struct linux_ioctl_handler **lihp; 1057 1058 error = 0; 1059 1060 switch(type) { 1061 case MOD_LOAD: 1062 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1063 ++brandinfo) 1064 if (elf32_insert_brand_entry(*brandinfo) < 0) 1065 error = EINVAL; 1066 if (error == 0) { 1067 SET_FOREACH(lihp, linux_ioctl_handler_set) 1068 linux32_ioctl_register_handler(*lihp); 1069 stclohz = (stathz ? stathz : hz); 1070 if (bootverbose) 1071 printf("Linux ELF exec handler installed\n"); 1072 } else 1073 printf("cannot insert Linux ELF brand handler\n"); 1074 break; 1075 case MOD_UNLOAD: 1076 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1077 ++brandinfo) 1078 if (elf32_brand_inuse(*brandinfo)) 1079 error = EBUSY; 1080 if (error == 0) { 1081 for (brandinfo = &linux_brandlist[0]; 1082 *brandinfo != NULL; ++brandinfo) 1083 if (elf32_remove_brand_entry(*brandinfo) < 0) 1084 error = EINVAL; 1085 } 1086 if (error == 0) { 1087 SET_FOREACH(lihp, linux_ioctl_handler_set) 1088 linux32_ioctl_unregister_handler(*lihp); 1089 if (bootverbose) 1090 printf("Linux ELF exec handler removed\n"); 1091 } else 1092 printf("Could not deinstall ELF interpreter entry\n"); 1093 break; 1094 default: 1095 return (EOPNOTSUPP); 1096 } 1097 return (error); 1098 } 1099 1100 static moduledata_t linux_elf_mod = { 1101 "linuxelf", 1102 linux_elf_modevent, 1103 0 1104 }; 1105 1106 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 1107 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1); 1108 FEATURE(linux, "Linux 32bit support"); 1109