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 static int linux_szsigcode; 94 static vm_object_t linux_shared_page_obj; 95 static char *linux_shared_page_mapping; 96 extern char _binary_linux32_locore_o_start; 97 extern char _binary_linux32_locore_o_end; 98 99 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL]; 100 101 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 102 103 static int linux_fixup_elf(uintptr_t *stack_base, 104 struct image_params *iparams); 105 static int linux_copyout_strings(struct image_params *imgp, 106 uintptr_t *stack_base); 107 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask); 108 static void linux_exec_setregs(struct thread *td, 109 struct image_params *imgp, uintptr_t stack); 110 static void linux32_fixlimit(struct rlimit *rl, int which); 111 static bool linux32_trans_osrel(const Elf_Note *note, int32_t *osrel); 112 static void linux_vdso_install(void *param); 113 static void linux_vdso_deinstall(void *param); 114 static void linux32_set_syscall_retval(struct thread *td, int error); 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_copyout_auxargs(struct image_params *imgp, uintptr_t base) 191 { 192 Elf32_Auxargs *args; 193 Elf32_Auxinfo *argarray, *pos; 194 int error, issetugid; 195 196 args = (Elf32_Auxargs *)imgp->auxargs; 197 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 198 M_WAITOK | M_ZERO); 199 200 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0; 201 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux32_vsyscall); 202 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, 203 imgp->proc->p_sysent->sv_shared_page_base); 204 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 205 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 206 207 /* 208 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, 209 * as it has appeared in the 2.4.0-rc7 first time. 210 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), 211 * glibc falls back to the hard-coded CLK_TCK value when aux entry 212 * is not present. 213 * Also see linux_times() implementation. 214 */ 215 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) 216 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 217 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 218 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 219 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 220 AUXARGS_ENTRY(pos, AT_BASE, args->base); 221 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 222 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 223 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 224 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 225 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 226 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 227 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 228 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary)); 229 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0); 230 if (imgp->execpathp != 0) 231 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp)); 232 if (args->execfd != -1) 233 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 234 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 235 AUXARGS_ENTRY(pos, AT_NULL, 0); 236 237 free(imgp->auxargs, M_TEMP); 238 imgp->auxargs = NULL; 239 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 240 241 error = copyout(argarray, (void *)base, 242 sizeof(*argarray) * LINUX_AT_COUNT); 243 free(argarray, M_TEMP); 244 return (error); 245 } 246 247 static int 248 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) 249 { 250 Elf32_Addr *base; 251 252 base = (Elf32_Addr *)*stack_base; 253 base--; 254 if (suword32(base, (uint32_t)imgp->args->argc) == -1) 255 return (EFAULT); 256 *stack_base = (uintptr_t)base; 257 return (0); 258 } 259 260 static void 261 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 262 { 263 struct thread *td = curthread; 264 struct proc *p = td->td_proc; 265 struct sigacts *psp; 266 struct trapframe *regs; 267 struct l_rt_sigframe *fp, frame; 268 int oonstack; 269 int sig; 270 int code; 271 272 sig = ksi->ksi_signo; 273 code = ksi->ksi_code; 274 PROC_LOCK_ASSERT(p, MA_OWNED); 275 psp = p->p_sigacts; 276 mtx_assert(&psp->ps_mtx, MA_OWNED); 277 regs = td->td_frame; 278 oonstack = sigonstack(regs->tf_rsp); 279 280 /* Allocate space for the signal handler context. */ 281 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 282 SIGISMEMBER(psp->ps_sigonstack, sig)) { 283 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 284 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe)); 285 } else 286 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1; 287 mtx_unlock(&psp->ps_mtx); 288 289 /* Build the argument list for the signal handler. */ 290 sig = bsd_to_linux_signal(sig); 291 292 bzero(&frame, sizeof(frame)); 293 294 frame.sf_handler = PTROUT(catcher); 295 frame.sf_sig = sig; 296 frame.sf_siginfo = PTROUT(&fp->sf_si); 297 frame.sf_ucontext = PTROUT(&fp->sf_sc); 298 299 /* Fill in POSIX parts. */ 300 siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig); 301 302 /* 303 * Build the signal context to be used by sigreturn and libgcc unwind. 304 */ 305 frame.sf_sc.uc_flags = 0; /* XXX ??? */ 306 frame.sf_sc.uc_link = 0; /* XXX ??? */ 307 308 frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 309 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 310 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 311 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 312 PROC_UNLOCK(p); 313 314 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask); 315 316 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask; 317 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi; 318 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi; 319 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp; 320 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx; 321 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp; 322 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx; 323 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx; 324 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax; 325 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip; 326 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 327 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs; 328 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs; 329 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es; 330 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds; 331 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags; 332 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp; 333 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss; 334 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err; 335 frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 336 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 337 338 if (copyout(&frame, fp, sizeof(frame)) != 0) { 339 /* 340 * Process has trashed its stack; give it an illegal 341 * instruction to halt it in its tracks. 342 */ 343 PROC_LOCK(p); 344 sigexit(td, SIGILL); 345 } 346 347 /* Build context to run handler in. */ 348 regs->tf_rsp = PTROUT(fp); 349 regs->tf_rip = linux32_rt_sigcode; 350 regs->tf_rflags &= ~(PSL_T | PSL_D); 351 regs->tf_cs = _ucode32sel; 352 regs->tf_ss = _udatasel; 353 regs->tf_ds = _udatasel; 354 regs->tf_es = _udatasel; 355 regs->tf_fs = _ufssel; 356 regs->tf_gs = _ugssel; 357 regs->tf_flags = TF_HASSEGS; 358 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 359 PROC_LOCK(p); 360 mtx_lock(&psp->ps_mtx); 361 } 362 363 /* 364 * Send an interrupt to process. 365 * 366 * Stack is set up to allow sigcode stored 367 * in u. to call routine, followed by kcall 368 * to sigreturn routine below. After sigreturn 369 * resets the signal mask, the stack, and the 370 * frame pointer, it returns to the user 371 * specified pc, psl. 372 */ 373 static void 374 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 375 { 376 struct thread *td = curthread; 377 struct proc *p = td->td_proc; 378 struct sigacts *psp; 379 struct trapframe *regs; 380 struct l_sigframe *fp, frame; 381 l_sigset_t lmask; 382 int oonstack; 383 int sig, code; 384 385 sig = ksi->ksi_signo; 386 code = ksi->ksi_code; 387 PROC_LOCK_ASSERT(p, MA_OWNED); 388 psp = p->p_sigacts; 389 mtx_assert(&psp->ps_mtx, MA_OWNED); 390 if (SIGISMEMBER(psp->ps_siginfo, sig)) { 391 /* Signal handler installed with SA_SIGINFO. */ 392 linux_rt_sendsig(catcher, ksi, mask); 393 return; 394 } 395 396 regs = td->td_frame; 397 oonstack = sigonstack(regs->tf_rsp); 398 399 /* Allocate space for the signal handler context. */ 400 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 401 SIGISMEMBER(psp->ps_sigonstack, sig)) { 402 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 403 td->td_sigstk.ss_size - sizeof(struct l_sigframe)); 404 } else 405 fp = (struct l_sigframe *)regs->tf_rsp - 1; 406 mtx_unlock(&psp->ps_mtx); 407 PROC_UNLOCK(p); 408 409 /* Build the argument list for the signal handler. */ 410 sig = bsd_to_linux_signal(sig); 411 412 bzero(&frame, sizeof(frame)); 413 414 frame.sf_handler = PTROUT(catcher); 415 frame.sf_sig = sig; 416 417 bsd_to_linux_sigset(mask, &lmask); 418 419 /* Build the signal context to be used by sigreturn. */ 420 frame.sf_sc.sc_mask = lmask.__mask; 421 frame.sf_sc.sc_gs = regs->tf_gs; 422 frame.sf_sc.sc_fs = regs->tf_fs; 423 frame.sf_sc.sc_es = regs->tf_es; 424 frame.sf_sc.sc_ds = regs->tf_ds; 425 frame.sf_sc.sc_edi = regs->tf_rdi; 426 frame.sf_sc.sc_esi = regs->tf_rsi; 427 frame.sf_sc.sc_ebp = regs->tf_rbp; 428 frame.sf_sc.sc_ebx = regs->tf_rbx; 429 frame.sf_sc.sc_esp = regs->tf_rsp; 430 frame.sf_sc.sc_edx = regs->tf_rdx; 431 frame.sf_sc.sc_ecx = regs->tf_rcx; 432 frame.sf_sc.sc_eax = regs->tf_rax; 433 frame.sf_sc.sc_eip = regs->tf_rip; 434 frame.sf_sc.sc_cs = regs->tf_cs; 435 frame.sf_sc.sc_eflags = regs->tf_rflags; 436 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp; 437 frame.sf_sc.sc_ss = regs->tf_ss; 438 frame.sf_sc.sc_err = regs->tf_err; 439 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 440 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code); 441 442 frame.sf_extramask[0] = lmask.__mask; 443 444 if (copyout(&frame, fp, sizeof(frame)) != 0) { 445 /* 446 * Process has trashed its stack; give it an illegal 447 * instruction to halt it in its tracks. 448 */ 449 PROC_LOCK(p); 450 sigexit(td, SIGILL); 451 } 452 453 /* Build context to run handler in. */ 454 regs->tf_rsp = PTROUT(fp); 455 regs->tf_rip = linux32_sigcode; 456 regs->tf_rflags &= ~(PSL_T | PSL_D); 457 regs->tf_cs = _ucode32sel; 458 regs->tf_ss = _udatasel; 459 regs->tf_ds = _udatasel; 460 regs->tf_es = _udatasel; 461 regs->tf_fs = _ufssel; 462 regs->tf_gs = _ugssel; 463 regs->tf_flags = TF_HASSEGS; 464 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 465 PROC_LOCK(p); 466 mtx_lock(&psp->ps_mtx); 467 } 468 469 /* 470 * System call to cleanup state after a signal 471 * has been taken. Reset signal mask and 472 * stack state from context left by sendsig (above). 473 * Return to previous pc and psl as specified by 474 * context left by sendsig. Check carefully to 475 * make sure that the user has not modified the 476 * psl to gain improper privileges or to cause 477 * a machine fault. 478 */ 479 int 480 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) 481 { 482 struct l_sigframe frame; 483 struct trapframe *regs; 484 sigset_t bmask; 485 l_sigset_t lmask; 486 int eflags; 487 ksiginfo_t ksi; 488 489 regs = td->td_frame; 490 491 /* 492 * The trampoline code hands us the sigframe. 493 * It is unsafe to keep track of it ourselves, in the event that a 494 * program jumps out of a signal handler. 495 */ 496 if (copyin(args->sfp, &frame, sizeof(frame)) != 0) 497 return (EFAULT); 498 499 /* Check for security violations. */ 500 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 501 eflags = frame.sf_sc.sc_eflags; 502 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 503 return(EINVAL); 504 505 /* 506 * Don't allow users to load a valid privileged %cs. Let the 507 * hardware check for invalid selectors, excess privilege in 508 * other selectors, invalid %eip's and invalid %esp's. 509 */ 510 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 511 if (!CS_SECURE(frame.sf_sc.sc_cs)) { 512 ksiginfo_init_trap(&ksi); 513 ksi.ksi_signo = SIGBUS; 514 ksi.ksi_code = BUS_OBJERR; 515 ksi.ksi_trapno = T_PROTFLT; 516 ksi.ksi_addr = (void *)regs->tf_rip; 517 trapsignal(td, &ksi); 518 return(EINVAL); 519 } 520 521 lmask.__mask = frame.sf_sc.sc_mask; 522 lmask.__mask = frame.sf_extramask[0]; 523 linux_to_bsd_sigset(&lmask, &bmask); 524 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 525 526 /* Restore signal context. */ 527 regs->tf_rdi = frame.sf_sc.sc_edi; 528 regs->tf_rsi = frame.sf_sc.sc_esi; 529 regs->tf_rbp = frame.sf_sc.sc_ebp; 530 regs->tf_rbx = frame.sf_sc.sc_ebx; 531 regs->tf_rdx = frame.sf_sc.sc_edx; 532 regs->tf_rcx = frame.sf_sc.sc_ecx; 533 regs->tf_rax = frame.sf_sc.sc_eax; 534 regs->tf_rip = frame.sf_sc.sc_eip; 535 regs->tf_cs = frame.sf_sc.sc_cs; 536 regs->tf_ds = frame.sf_sc.sc_ds; 537 regs->tf_es = frame.sf_sc.sc_es; 538 regs->tf_fs = frame.sf_sc.sc_fs; 539 regs->tf_gs = frame.sf_sc.sc_gs; 540 regs->tf_rflags = eflags; 541 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal; 542 regs->tf_ss = frame.sf_sc.sc_ss; 543 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 544 545 return (EJUSTRETURN); 546 } 547 548 /* 549 * System call to cleanup state after a signal 550 * has been taken. Reset signal mask and 551 * stack state from context left by rt_sendsig (above). 552 * Return to previous pc and psl as specified by 553 * context left by sendsig. Check carefully to 554 * make sure that the user has not modified the 555 * psl to gain improper privileges or to cause 556 * a machine fault. 557 */ 558 int 559 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 560 { 561 struct l_ucontext uc; 562 struct l_sigcontext *context; 563 sigset_t bmask; 564 l_stack_t *lss; 565 stack_t ss; 566 struct trapframe *regs; 567 int eflags; 568 ksiginfo_t ksi; 569 570 regs = td->td_frame; 571 572 /* 573 * The trampoline code hands us the ucontext. 574 * It is unsafe to keep track of it ourselves, in the event that a 575 * program jumps out of a signal handler. 576 */ 577 if (copyin(args->ucp, &uc, sizeof(uc)) != 0) 578 return (EFAULT); 579 580 context = &uc.uc_mcontext; 581 582 /* Check for security violations. */ 583 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 584 eflags = context->sc_eflags; 585 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 586 return(EINVAL); 587 588 /* 589 * Don't allow users to load a valid privileged %cs. Let the 590 * hardware check for invalid selectors, excess privilege in 591 * other selectors, invalid %eip's and invalid %esp's. 592 */ 593 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 594 if (!CS_SECURE(context->sc_cs)) { 595 ksiginfo_init_trap(&ksi); 596 ksi.ksi_signo = SIGBUS; 597 ksi.ksi_code = BUS_OBJERR; 598 ksi.ksi_trapno = T_PROTFLT; 599 ksi.ksi_addr = (void *)regs->tf_rip; 600 trapsignal(td, &ksi); 601 return(EINVAL); 602 } 603 604 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); 605 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 606 607 /* 608 * Restore signal context 609 */ 610 regs->tf_gs = context->sc_gs; 611 regs->tf_fs = context->sc_fs; 612 regs->tf_es = context->sc_es; 613 regs->tf_ds = context->sc_ds; 614 regs->tf_rdi = context->sc_edi; 615 regs->tf_rsi = context->sc_esi; 616 regs->tf_rbp = context->sc_ebp; 617 regs->tf_rbx = context->sc_ebx; 618 regs->tf_rdx = context->sc_edx; 619 regs->tf_rcx = context->sc_ecx; 620 regs->tf_rax = context->sc_eax; 621 regs->tf_rip = context->sc_eip; 622 regs->tf_cs = context->sc_cs; 623 regs->tf_rflags = eflags; 624 regs->tf_rsp = context->sc_esp_at_signal; 625 regs->tf_ss = context->sc_ss; 626 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 627 628 /* 629 * call sigaltstack & ignore results.. 630 */ 631 lss = &uc.uc_stack; 632 ss.ss_sp = PTRIN(lss->ss_sp); 633 ss.ss_size = lss->ss_size; 634 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags); 635 636 (void)kern_sigaltstack(td, &ss, NULL); 637 638 return (EJUSTRETURN); 639 } 640 641 static int 642 linux32_fetch_syscall_args(struct thread *td) 643 { 644 struct proc *p; 645 struct trapframe *frame; 646 struct syscall_args *sa; 647 648 p = td->td_proc; 649 frame = td->td_frame; 650 sa = &td->td_sa; 651 652 sa->args[0] = frame->tf_rbx; 653 sa->args[1] = frame->tf_rcx; 654 sa->args[2] = frame->tf_rdx; 655 sa->args[3] = frame->tf_rsi; 656 sa->args[4] = frame->tf_rdi; 657 sa->args[5] = frame->tf_rbp; /* Unconfirmed */ 658 sa->code = frame->tf_rax; 659 660 if (sa->code >= p->p_sysent->sv_size) 661 /* nosys */ 662 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 663 else 664 sa->callp = &p->p_sysent->sv_table[sa->code]; 665 666 td->td_retval[0] = 0; 667 td->td_retval[1] = frame->tf_rdx; 668 669 return (0); 670 } 671 672 static void 673 linux32_set_syscall_retval(struct thread *td, int error) 674 { 675 struct trapframe *frame = td->td_frame; 676 677 cpu_set_syscall_retval(td, error); 678 679 if (__predict_false(error != 0)) { 680 if (error != ERESTART && error != EJUSTRETURN) 681 frame->tf_rax = bsd_to_linux_errno(error); 682 } 683 } 684 685 /* 686 * Clear registers on exec 687 * XXX copied from ia32_signal.c. 688 */ 689 static void 690 linux_exec_setregs(struct thread *td, struct image_params *imgp, 691 uintptr_t stack) 692 { 693 struct trapframe *regs = td->td_frame; 694 struct pcb *pcb = td->td_pcb; 695 register_t saved_rflags; 696 697 regs = td->td_frame; 698 pcb = td->td_pcb; 699 700 if (td->td_proc->p_md.md_ldt != NULL) 701 user_ldt_free(td); 702 703 critical_enter(); 704 wrmsr(MSR_FSBASE, 0); 705 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */ 706 pcb->pcb_fsbase = 0; 707 pcb->pcb_gsbase = 0; 708 critical_exit(); 709 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 710 711 saved_rflags = regs->tf_rflags & PSL_T; 712 bzero((char *)regs, sizeof(struct trapframe)); 713 regs->tf_rip = imgp->entry_addr; 714 regs->tf_rsp = stack; 715 regs->tf_rflags = PSL_USER | saved_rflags; 716 regs->tf_gs = _ugssel; 717 regs->tf_fs = _ufssel; 718 regs->tf_es = _udatasel; 719 regs->tf_ds = _udatasel; 720 regs->tf_ss = _udatasel; 721 regs->tf_flags = TF_HASSEGS; 722 regs->tf_cs = _ucode32sel; 723 regs->tf_rbx = (register_t)imgp->ps_strings; 724 725 x86_clear_dbregs(pcb); 726 727 fpstate_drop(td); 728 729 /* Do full restore on return so that we can change to a different %cs */ 730 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET); 731 } 732 733 /* 734 * XXX copied from ia32_sysvec.c. 735 */ 736 static int 737 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 738 { 739 int argc, envc, error; 740 u_int32_t *vectp; 741 char *stringp; 742 uintptr_t destp, ustringp; 743 struct linux32_ps_strings *arginfo; 744 char canary[LINUX_AT_RANDOM_LEN]; 745 size_t execpath_len; 746 747 /* Calculate string base and vector table pointers. */ 748 if (imgp->execpath != NULL && imgp->auxargs != NULL) 749 execpath_len = strlen(imgp->execpath) + 1; 750 else 751 execpath_len = 0; 752 753 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; 754 destp = (uintptr_t)arginfo; 755 756 if (execpath_len != 0) { 757 destp -= execpath_len; 758 destp = rounddown2(destp, sizeof(uint32_t)); 759 imgp->execpathp = (void *)destp; 760 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 761 if (error != 0) 762 return (error); 763 } 764 765 /* Prepare the canary for SSP. */ 766 arc4rand(canary, sizeof(canary), 0); 767 destp -= roundup(sizeof(canary), sizeof(uint32_t)); 768 imgp->canary = (void *)destp; 769 error = copyout(canary, imgp->canary, sizeof(canary)); 770 if (error != 0) 771 return (error); 772 773 /* Allocate room for the argument and environment strings. */ 774 destp -= ARG_MAX - imgp->args->stringspace; 775 destp = rounddown2(destp, sizeof(uint32_t)); 776 ustringp = destp; 777 778 if (imgp->auxargs) { 779 /* 780 * Allocate room on the stack for the ELF auxargs 781 * array. It has LINUX_AT_COUNT entries. 782 */ 783 destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo); 784 destp = rounddown2(destp, sizeof(uint32_t)); 785 } 786 787 vectp = (uint32_t *)destp; 788 789 /* 790 * Allocate room for the argv[] and env vectors including the 791 * terminating NULL pointers. 792 */ 793 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 794 795 /* vectp also becomes our initial stack base. */ 796 *stack_base = (uintptr_t)vectp; 797 798 stringp = imgp->args->begin_argv; 799 argc = imgp->args->argc; 800 envc = imgp->args->envc; 801 802 /* Copy out strings - arguments and environment. */ 803 error = copyout(stringp, (void *)ustringp, 804 ARG_MAX - imgp->args->stringspace); 805 if (error != 0) 806 return (error); 807 808 /* Fill in "ps_strings" struct for ps, w, etc. */ 809 if (suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp) != 0 || 810 suword32(&arginfo->ps_nargvstr, argc) != 0) 811 return (EFAULT); 812 813 /* Fill in argument portion of vector table. */ 814 for (; argc > 0; --argc) { 815 if (suword32(vectp++, ustringp) != 0) 816 return (EFAULT); 817 while (*stringp++ != 0) 818 ustringp++; 819 ustringp++; 820 } 821 822 /* A null vector table pointer separates the argp's from the envp's. */ 823 if (suword32(vectp++, 0) != 0) 824 return (EFAULT); 825 826 if (suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp) != 0 || 827 suword32(&arginfo->ps_nenvstr, envc) != 0) 828 return (EFAULT); 829 830 /* Fill in environment portion of vector table. */ 831 for (; envc > 0; --envc) { 832 if (suword32(vectp++, ustringp) != 0) 833 return (EFAULT); 834 while (*stringp++ != 0) 835 ustringp++; 836 ustringp++; 837 } 838 839 /* The end of the vector table is a null pointer. */ 840 if (suword32(vectp, 0) != 0) 841 return (EFAULT); 842 843 if (imgp->auxargs) { 844 vectp++; 845 error = imgp->sysent->sv_copyout_auxargs(imgp, 846 (uintptr_t)vectp); 847 if (error != 0) 848 return (error); 849 } 850 851 return (0); 852 } 853 854 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 855 "32-bit Linux emulation"); 856 857 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ; 858 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW, 859 &linux32_maxdsiz, 0, ""); 860 static u_long linux32_maxssiz = LINUX32_MAXSSIZ; 861 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW, 862 &linux32_maxssiz, 0, ""); 863 static u_long linux32_maxvmem = LINUX32_MAXVMEM; 864 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW, 865 &linux32_maxvmem, 0, ""); 866 867 static void 868 linux32_fixlimit(struct rlimit *rl, int which) 869 { 870 871 switch (which) { 872 case RLIMIT_DATA: 873 if (linux32_maxdsiz != 0) { 874 if (rl->rlim_cur > linux32_maxdsiz) 875 rl->rlim_cur = linux32_maxdsiz; 876 if (rl->rlim_max > linux32_maxdsiz) 877 rl->rlim_max = linux32_maxdsiz; 878 } 879 break; 880 case RLIMIT_STACK: 881 if (linux32_maxssiz != 0) { 882 if (rl->rlim_cur > linux32_maxssiz) 883 rl->rlim_cur = linux32_maxssiz; 884 if (rl->rlim_max > linux32_maxssiz) 885 rl->rlim_max = linux32_maxssiz; 886 } 887 break; 888 case RLIMIT_VMEM: 889 if (linux32_maxvmem != 0) { 890 if (rl->rlim_cur > linux32_maxvmem) 891 rl->rlim_cur = linux32_maxvmem; 892 if (rl->rlim_max > linux32_maxvmem) 893 rl->rlim_max = linux32_maxvmem; 894 } 895 break; 896 } 897 } 898 899 struct sysentvec elf_linux_sysvec = { 900 .sv_size = LINUX32_SYS_MAXSYSCALL, 901 .sv_table = linux32_sysent, 902 .sv_transtrap = linux_translate_traps, 903 .sv_fixup = linux_fixup_elf, 904 .sv_sendsig = linux_sendsig, 905 .sv_sigcode = &_binary_linux32_locore_o_start, 906 .sv_szsigcode = &linux_szsigcode, 907 .sv_name = "Linux ELF32", 908 .sv_coredump = elf32_coredump, 909 .sv_imgact_try = linux_exec_imgact_try, 910 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 911 .sv_minuser = VM_MIN_ADDRESS, 912 .sv_maxuser = LINUX32_MAXUSER, 913 .sv_usrstack = LINUX32_USRSTACK, 914 .sv_psstrings = LINUX32_PS_STRINGS, 915 .sv_stackprot = VM_PROT_ALL, 916 .sv_copyout_auxargs = linux_copyout_auxargs, 917 .sv_copyout_strings = linux_copyout_strings, 918 .sv_setregs = linux_exec_setregs, 919 .sv_fixlimit = linux32_fixlimit, 920 .sv_maxssiz = &linux32_maxssiz, 921 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP | 922 SV_SIG_DISCIGN | SV_SIG_WAITNDQ, 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 .sv_setid_allowed = &linux_setid_allowed_query, 935 }; 936 937 static void 938 linux_vdso_install(void *param) 939 { 940 941 linux_szsigcode = (&_binary_linux32_locore_o_end - 942 &_binary_linux32_locore_o_start); 943 944 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 945 panic("Linux invalid vdso size\n"); 946 947 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 948 949 linux_shared_page_obj = __elfN(linux_shared_page_init) 950 (&linux_shared_page_mapping); 951 952 __elfN(linux_vdso_reloc)(&elf_linux_sysvec); 953 954 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 955 linux_szsigcode); 956 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 957 } 958 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 959 linux_vdso_install, NULL); 960 961 static void 962 linux_vdso_deinstall(void *param) 963 { 964 965 __elfN(linux_shared_page_fini)(linux_shared_page_obj, 966 linux_shared_page_mapping); 967 } 968 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 969 linux_vdso_deinstall, NULL); 970 971 static char GNU_ABI_VENDOR[] = "GNU"; 972 static int GNULINUX_ABI_DESC = 0; 973 974 static bool 975 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel) 976 { 977 const Elf32_Word *desc; 978 uintptr_t p; 979 980 p = (uintptr_t)(note + 1); 981 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 982 983 desc = (const Elf32_Word *)p; 984 if (desc[0] != GNULINUX_ABI_DESC) 985 return (false); 986 987 /* 988 * For Linux we encode osrel using the Linux convention of 989 * (version << 16) | (major << 8) | (minor) 990 * See macro in linux_mib.h 991 */ 992 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 993 994 return (true); 995 } 996 997 static Elf_Brandnote linux32_brandnote = { 998 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 999 .hdr.n_descsz = 16, /* XXX at least 16 */ 1000 .hdr.n_type = 1, 1001 .vendor = GNU_ABI_VENDOR, 1002 .flags = BN_TRANSLATE_OSREL, 1003 .trans_osrel = linux32_trans_osrel 1004 }; 1005 1006 static Elf32_Brandinfo linux_brand = { 1007 .brand = ELFOSABI_LINUX, 1008 .machine = EM_386, 1009 .compat_3_brand = "Linux", 1010 .emul_path = linux_emul_path, 1011 .interp_path = "/lib/ld-linux.so.1", 1012 .sysvec = &elf_linux_sysvec, 1013 .interp_newpath = NULL, 1014 .brand_note = &linux32_brandnote, 1015 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1016 }; 1017 1018 static Elf32_Brandinfo linux_glibc2brand = { 1019 .brand = ELFOSABI_LINUX, 1020 .machine = EM_386, 1021 .compat_3_brand = "Linux", 1022 .emul_path = linux_emul_path, 1023 .interp_path = "/lib/ld-linux.so.2", 1024 .sysvec = &elf_linux_sysvec, 1025 .interp_newpath = NULL, 1026 .brand_note = &linux32_brandnote, 1027 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1028 }; 1029 1030 static Elf32_Brandinfo linux_muslbrand = { 1031 .brand = ELFOSABI_LINUX, 1032 .machine = EM_386, 1033 .compat_3_brand = "Linux", 1034 .emul_path = linux_emul_path, 1035 .interp_path = "/lib/ld-musl-i386.so.1", 1036 .sysvec = &elf_linux_sysvec, 1037 .interp_newpath = NULL, 1038 .brand_note = &linux32_brandnote, 1039 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1040 }; 1041 1042 Elf32_Brandinfo *linux_brandlist[] = { 1043 &linux_brand, 1044 &linux_glibc2brand, 1045 &linux_muslbrand, 1046 NULL 1047 }; 1048 1049 static int 1050 linux_elf_modevent(module_t mod, int type, void *data) 1051 { 1052 Elf32_Brandinfo **brandinfo; 1053 int error; 1054 struct linux_ioctl_handler **lihp; 1055 1056 error = 0; 1057 1058 switch(type) { 1059 case MOD_LOAD: 1060 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1061 ++brandinfo) 1062 if (elf32_insert_brand_entry(*brandinfo) < 0) 1063 error = EINVAL; 1064 if (error == 0) { 1065 SET_FOREACH(lihp, linux_ioctl_handler_set) 1066 linux32_ioctl_register_handler(*lihp); 1067 stclohz = (stathz ? stathz : hz); 1068 if (bootverbose) 1069 printf("Linux ELF exec handler installed\n"); 1070 } else 1071 printf("cannot insert Linux ELF brand handler\n"); 1072 break; 1073 case MOD_UNLOAD: 1074 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1075 ++brandinfo) 1076 if (elf32_brand_inuse(*brandinfo)) 1077 error = EBUSY; 1078 if (error == 0) { 1079 for (brandinfo = &linux_brandlist[0]; 1080 *brandinfo != NULL; ++brandinfo) 1081 if (elf32_remove_brand_entry(*brandinfo) < 0) 1082 error = EINVAL; 1083 } 1084 if (error == 0) { 1085 SET_FOREACH(lihp, linux_ioctl_handler_set) 1086 linux32_ioctl_unregister_handler(*lihp); 1087 if (bootverbose) 1088 printf("Linux ELF exec handler removed\n"); 1089 } else 1090 printf("Could not deinstall ELF interpreter entry\n"); 1091 break; 1092 default: 1093 return (EOPNOTSUPP); 1094 } 1095 return (error); 1096 } 1097 1098 static moduledata_t linux_elf_mod = { 1099 "linuxelf", 1100 linux_elf_modevent, 1101 0 1102 }; 1103 1104 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 1105 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1); 1106 FEATURE(linux, "Linux 32bit support"); 1107