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); 115 116 #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_EHDR, 202 imgp->proc->p_sysent->sv_shared_page_base); 203 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux32_vsyscall); 204 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 205 206 /* 207 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, 208 * as it has appeared in the 2.4.0-rc7 first time. 209 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), 210 * glibc falls back to the hard-coded CLK_TCK value when aux entry 211 * is not present. 212 * Also see linux_times() implementation. 213 */ 214 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) 215 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 216 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 217 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 218 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 219 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 220 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 221 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 222 AUXARGS_ENTRY(pos, AT_BASE, args->base); 223 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 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_PLATFORM, PTROUT(linux_platform)); 229 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary)); 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, AT_NULL, 0); 235 236 free(imgp->auxargs, M_TEMP); 237 imgp->auxargs = NULL; 238 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 239 240 error = copyout(argarray, (void *)base, 241 sizeof(*argarray) * LINUX_AT_COUNT); 242 free(argarray, M_TEMP); 243 return (error); 244 } 245 246 static int 247 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp) 248 { 249 Elf32_Addr *base; 250 251 base = (Elf32_Addr *)*stack_base; 252 base--; 253 if (suword32(base, (uint32_t)imgp->args->argc) == -1) 254 return (EFAULT); 255 *stack_base = (uintptr_t)base; 256 return (0); 257 } 258 259 static void 260 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 261 { 262 struct thread *td = curthread; 263 struct proc *p = td->td_proc; 264 struct sigacts *psp; 265 struct trapframe *regs; 266 struct l_rt_sigframe *fp, frame; 267 int oonstack; 268 int sig; 269 int code; 270 271 sig = ksi->ksi_signo; 272 code = ksi->ksi_code; 273 PROC_LOCK_ASSERT(p, MA_OWNED); 274 psp = p->p_sigacts; 275 mtx_assert(&psp->ps_mtx, MA_OWNED); 276 regs = td->td_frame; 277 oonstack = sigonstack(regs->tf_rsp); 278 279 /* Allocate space for the signal handler context. */ 280 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 281 SIGISMEMBER(psp->ps_sigonstack, sig)) { 282 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 283 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe)); 284 } else 285 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1; 286 mtx_unlock(&psp->ps_mtx); 287 288 /* Build the argument list for the signal handler. */ 289 sig = bsd_to_linux_signal(sig); 290 291 bzero(&frame, sizeof(frame)); 292 293 frame.sf_handler = PTROUT(catcher); 294 frame.sf_sig = sig; 295 frame.sf_siginfo = PTROUT(&fp->sf_si); 296 frame.sf_ucontext = PTROUT(&fp->sf_sc); 297 298 /* Fill in POSIX parts. */ 299 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig); 300 301 /* 302 * Build the signal context to be used by sigreturn and libgcc unwind. 303 */ 304 frame.sf_sc.uc_flags = 0; /* XXX ??? */ 305 frame.sf_sc.uc_link = 0; /* XXX ??? */ 306 307 frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 308 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 309 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 310 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 311 PROC_UNLOCK(p); 312 313 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask); 314 315 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask; 316 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi; 317 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi; 318 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp; 319 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx; 320 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp; 321 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx; 322 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx; 323 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax; 324 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip; 325 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 326 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs; 327 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs; 328 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es; 329 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds; 330 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags; 331 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp; 332 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss; 333 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err; 334 frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 335 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 336 337 if (copyout(&frame, fp, sizeof(frame)) != 0) { 338 /* 339 * Process has trashed its stack; give it an illegal 340 * instruction to halt it in its tracks. 341 */ 342 PROC_LOCK(p); 343 sigexit(td, SIGILL); 344 } 345 346 /* Build context to run handler in. */ 347 regs->tf_rsp = PTROUT(fp); 348 regs->tf_rip = linux32_rt_sigcode; 349 regs->tf_rflags &= ~(PSL_T | PSL_D); 350 regs->tf_cs = _ucode32sel; 351 regs->tf_ss = _udatasel; 352 regs->tf_ds = _udatasel; 353 regs->tf_es = _udatasel; 354 regs->tf_fs = _ufssel; 355 regs->tf_gs = _ugssel; 356 regs->tf_flags = TF_HASSEGS; 357 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 358 PROC_LOCK(p); 359 mtx_lock(&psp->ps_mtx); 360 } 361 362 /* 363 * Send an interrupt to process. 364 * 365 * Stack is set up to allow sigcode stored 366 * in u. to call routine, followed by kcall 367 * to sigreturn routine below. After sigreturn 368 * resets the signal mask, the stack, and the 369 * frame pointer, it returns to the user 370 * specified pc, psl. 371 */ 372 static void 373 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 374 { 375 struct thread *td = curthread; 376 struct proc *p = td->td_proc; 377 struct sigacts *psp; 378 struct trapframe *regs; 379 struct l_sigframe *fp, frame; 380 l_sigset_t lmask; 381 int oonstack; 382 int sig, code; 383 384 sig = ksi->ksi_signo; 385 code = ksi->ksi_code; 386 PROC_LOCK_ASSERT(p, MA_OWNED); 387 psp = p->p_sigacts; 388 mtx_assert(&psp->ps_mtx, MA_OWNED); 389 if (SIGISMEMBER(psp->ps_siginfo, sig)) { 390 /* Signal handler installed with SA_SIGINFO. */ 391 linux_rt_sendsig(catcher, ksi, mask); 392 return; 393 } 394 395 regs = td->td_frame; 396 oonstack = sigonstack(regs->tf_rsp); 397 398 /* Allocate space for the signal handler context. */ 399 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && 400 SIGISMEMBER(psp->ps_sigonstack, sig)) { 401 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + 402 td->td_sigstk.ss_size - sizeof(struct l_sigframe)); 403 } else 404 fp = (struct l_sigframe *)regs->tf_rsp - 1; 405 mtx_unlock(&psp->ps_mtx); 406 PROC_UNLOCK(p); 407 408 /* Build the argument list for the signal handler. */ 409 sig = bsd_to_linux_signal(sig); 410 411 bzero(&frame, sizeof(frame)); 412 413 frame.sf_handler = PTROUT(catcher); 414 frame.sf_sig = sig; 415 416 bsd_to_linux_sigset(mask, &lmask); 417 418 /* Build the signal context to be used by sigreturn. */ 419 frame.sf_sc.sc_mask = lmask.__mask; 420 frame.sf_sc.sc_gs = regs->tf_gs; 421 frame.sf_sc.sc_fs = regs->tf_fs; 422 frame.sf_sc.sc_es = regs->tf_es; 423 frame.sf_sc.sc_ds = regs->tf_ds; 424 frame.sf_sc.sc_edi = regs->tf_rdi; 425 frame.sf_sc.sc_esi = regs->tf_rsi; 426 frame.sf_sc.sc_ebp = regs->tf_rbp; 427 frame.sf_sc.sc_ebx = regs->tf_rbx; 428 frame.sf_sc.sc_esp = regs->tf_rsp; 429 frame.sf_sc.sc_edx = regs->tf_rdx; 430 frame.sf_sc.sc_ecx = regs->tf_rcx; 431 frame.sf_sc.sc_eax = regs->tf_rax; 432 frame.sf_sc.sc_eip = regs->tf_rip; 433 frame.sf_sc.sc_cs = regs->tf_cs; 434 frame.sf_sc.sc_eflags = regs->tf_rflags; 435 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp; 436 frame.sf_sc.sc_ss = regs->tf_ss; 437 frame.sf_sc.sc_err = regs->tf_err; 438 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; 439 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code); 440 441 frame.sf_extramask[0] = lmask.__mask; 442 443 if (copyout(&frame, fp, sizeof(frame)) != 0) { 444 /* 445 * Process has trashed its stack; give it an illegal 446 * instruction to halt it in its tracks. 447 */ 448 PROC_LOCK(p); 449 sigexit(td, SIGILL); 450 } 451 452 /* Build context to run handler in. */ 453 regs->tf_rsp = PTROUT(fp); 454 regs->tf_rip = linux32_sigcode; 455 regs->tf_rflags &= ~(PSL_T | PSL_D); 456 regs->tf_cs = _ucode32sel; 457 regs->tf_ss = _udatasel; 458 regs->tf_ds = _udatasel; 459 regs->tf_es = _udatasel; 460 regs->tf_fs = _ufssel; 461 regs->tf_gs = _ugssel; 462 regs->tf_flags = TF_HASSEGS; 463 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 464 PROC_LOCK(p); 465 mtx_lock(&psp->ps_mtx); 466 } 467 468 /* 469 * System call to cleanup state after a signal 470 * has been taken. Reset signal mask and 471 * stack state from context left by sendsig (above). 472 * Return to previous pc and psl as specified by 473 * context left by sendsig. Check carefully to 474 * make sure that the user has not modified the 475 * psl to gain improper privileges or to cause 476 * a machine fault. 477 */ 478 int 479 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) 480 { 481 struct l_sigframe frame; 482 struct trapframe *regs; 483 sigset_t bmask; 484 l_sigset_t lmask; 485 int eflags; 486 ksiginfo_t ksi; 487 488 regs = td->td_frame; 489 490 /* 491 * The trampoline code hands us the sigframe. 492 * It is unsafe to keep track of it ourselves, in the event that a 493 * program jumps out of a signal handler. 494 */ 495 if (copyin(args->sfp, &frame, sizeof(frame)) != 0) 496 return (EFAULT); 497 498 /* Check for security violations. */ 499 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 500 eflags = frame.sf_sc.sc_eflags; 501 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 502 return(EINVAL); 503 504 /* 505 * Don't allow users to load a valid privileged %cs. Let the 506 * hardware check for invalid selectors, excess privilege in 507 * other selectors, invalid %eip's and invalid %esp's. 508 */ 509 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 510 if (!CS_SECURE(frame.sf_sc.sc_cs)) { 511 ksiginfo_init_trap(&ksi); 512 ksi.ksi_signo = SIGBUS; 513 ksi.ksi_code = BUS_OBJERR; 514 ksi.ksi_trapno = T_PROTFLT; 515 ksi.ksi_addr = (void *)regs->tf_rip; 516 trapsignal(td, &ksi); 517 return(EINVAL); 518 } 519 520 lmask.__mask = frame.sf_sc.sc_mask; 521 lmask.__mask = frame.sf_extramask[0]; 522 linux_to_bsd_sigset(&lmask, &bmask); 523 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 524 525 /* Restore signal context. */ 526 regs->tf_rdi = frame.sf_sc.sc_edi; 527 regs->tf_rsi = frame.sf_sc.sc_esi; 528 regs->tf_rbp = frame.sf_sc.sc_ebp; 529 regs->tf_rbx = frame.sf_sc.sc_ebx; 530 regs->tf_rdx = frame.sf_sc.sc_edx; 531 regs->tf_rcx = frame.sf_sc.sc_ecx; 532 regs->tf_rax = frame.sf_sc.sc_eax; 533 regs->tf_rip = frame.sf_sc.sc_eip; 534 regs->tf_cs = frame.sf_sc.sc_cs; 535 regs->tf_ds = frame.sf_sc.sc_ds; 536 regs->tf_es = frame.sf_sc.sc_es; 537 regs->tf_fs = frame.sf_sc.sc_fs; 538 regs->tf_gs = frame.sf_sc.sc_gs; 539 regs->tf_rflags = eflags; 540 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal; 541 regs->tf_ss = frame.sf_sc.sc_ss; 542 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 543 544 return (EJUSTRETURN); 545 } 546 547 /* 548 * System call to cleanup state after a signal 549 * has been taken. Reset signal mask and 550 * stack state from context left by rt_sendsig (above). 551 * Return to previous pc and psl as specified by 552 * context left by sendsig. Check carefully to 553 * make sure that the user has not modified the 554 * psl to gain improper privileges or to cause 555 * a machine fault. 556 */ 557 int 558 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 559 { 560 struct l_ucontext uc; 561 struct l_sigcontext *context; 562 sigset_t bmask; 563 l_stack_t *lss; 564 stack_t ss; 565 struct trapframe *regs; 566 int eflags; 567 ksiginfo_t ksi; 568 569 regs = td->td_frame; 570 571 /* 572 * The trampoline code hands us the ucontext. 573 * It is unsafe to keep track of it ourselves, in the event that a 574 * program jumps out of a signal handler. 575 */ 576 if (copyin(args->ucp, &uc, sizeof(uc)) != 0) 577 return (EFAULT); 578 579 context = &uc.uc_mcontext; 580 581 /* Check for security violations. */ 582 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 583 eflags = context->sc_eflags; 584 if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) 585 return(EINVAL); 586 587 /* 588 * Don't allow users to load a valid privileged %cs. Let the 589 * hardware check for invalid selectors, excess privilege in 590 * other selectors, invalid %eip's and invalid %esp's. 591 */ 592 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 593 if (!CS_SECURE(context->sc_cs)) { 594 ksiginfo_init_trap(&ksi); 595 ksi.ksi_signo = SIGBUS; 596 ksi.ksi_code = BUS_OBJERR; 597 ksi.ksi_trapno = T_PROTFLT; 598 ksi.ksi_addr = (void *)regs->tf_rip; 599 trapsignal(td, &ksi); 600 return(EINVAL); 601 } 602 603 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); 604 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 605 606 /* 607 * Restore signal context 608 */ 609 regs->tf_gs = context->sc_gs; 610 regs->tf_fs = context->sc_fs; 611 regs->tf_es = context->sc_es; 612 regs->tf_ds = context->sc_ds; 613 regs->tf_rdi = context->sc_edi; 614 regs->tf_rsi = context->sc_esi; 615 regs->tf_rbp = context->sc_ebp; 616 regs->tf_rbx = context->sc_ebx; 617 regs->tf_rdx = context->sc_edx; 618 regs->tf_rcx = context->sc_ecx; 619 regs->tf_rax = context->sc_eax; 620 regs->tf_rip = context->sc_eip; 621 regs->tf_cs = context->sc_cs; 622 regs->tf_rflags = eflags; 623 regs->tf_rsp = context->sc_esp_at_signal; 624 regs->tf_ss = context->sc_ss; 625 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 626 627 /* 628 * call sigaltstack & ignore results.. 629 */ 630 lss = &uc.uc_stack; 631 ss.ss_sp = PTRIN(lss->ss_sp); 632 ss.ss_size = lss->ss_size; 633 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags); 634 635 (void)kern_sigaltstack(td, &ss, NULL); 636 637 return (EJUSTRETURN); 638 } 639 640 static int 641 linux32_fetch_syscall_args(struct thread *td) 642 { 643 struct proc *p; 644 struct trapframe *frame; 645 struct syscall_args *sa; 646 647 p = td->td_proc; 648 frame = td->td_frame; 649 sa = &td->td_sa; 650 651 sa->args[0] = frame->tf_rbx; 652 sa->args[1] = frame->tf_rcx; 653 sa->args[2] = frame->tf_rdx; 654 sa->args[3] = frame->tf_rsi; 655 sa->args[4] = frame->tf_rdi; 656 sa->args[5] = frame->tf_rbp; /* Unconfirmed */ 657 sa->code = frame->tf_rax; 658 659 if (sa->code >= p->p_sysent->sv_size) 660 /* nosys */ 661 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 662 else 663 sa->callp = &p->p_sysent->sv_table[sa->code]; 664 sa->narg = sa->callp->sy_narg; 665 666 td->td_retval[0] = 0; 667 td->td_retval[1] = frame->tf_rdx; 668 669 return (0); 670 } 671 672 /* 673 * Clear registers on exec 674 * XXX copied from ia32_signal.c. 675 */ 676 static void 677 linux_exec_setregs(struct thread *td, struct image_params *imgp, 678 uintptr_t stack) 679 { 680 struct trapframe *regs = td->td_frame; 681 struct pcb *pcb = td->td_pcb; 682 register_t saved_rflags; 683 684 regs = td->td_frame; 685 pcb = td->td_pcb; 686 687 if (td->td_proc->p_md.md_ldt != NULL) 688 user_ldt_free(td); 689 690 critical_enter(); 691 wrmsr(MSR_FSBASE, 0); 692 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */ 693 pcb->pcb_fsbase = 0; 694 pcb->pcb_gsbase = 0; 695 critical_exit(); 696 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 697 698 saved_rflags = regs->tf_rflags & PSL_T; 699 bzero((char *)regs, sizeof(struct trapframe)); 700 regs->tf_rip = imgp->entry_addr; 701 regs->tf_rsp = stack; 702 regs->tf_rflags = PSL_USER | saved_rflags; 703 regs->tf_gs = _ugssel; 704 regs->tf_fs = _ufssel; 705 regs->tf_es = _udatasel; 706 regs->tf_ds = _udatasel; 707 regs->tf_ss = _udatasel; 708 regs->tf_flags = TF_HASSEGS; 709 regs->tf_cs = _ucode32sel; 710 regs->tf_rbx = (register_t)imgp->ps_strings; 711 712 fpstate_drop(td); 713 714 /* Do full restore on return so that we can change to a different %cs */ 715 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET); 716 } 717 718 /* 719 * XXX copied from ia32_sysvec.c. 720 */ 721 static int 722 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 723 { 724 int argc, envc, error; 725 u_int32_t *vectp; 726 char *stringp; 727 uintptr_t destp, ustringp; 728 struct linux32_ps_strings *arginfo; 729 char canary[LINUX_AT_RANDOM_LEN]; 730 size_t execpath_len; 731 732 /* Calculate string base and vector table pointers. */ 733 if (imgp->execpath != NULL && imgp->auxargs != NULL) 734 execpath_len = strlen(imgp->execpath) + 1; 735 else 736 execpath_len = 0; 737 738 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; 739 destp = (uintptr_t)arginfo; 740 741 if (execpath_len != 0) { 742 destp -= execpath_len; 743 destp = rounddown2(destp, sizeof(uint32_t)); 744 imgp->execpathp = (void *)destp; 745 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 746 if (error != 0) 747 return (error); 748 } 749 750 /* Prepare the canary for SSP. */ 751 arc4rand(canary, sizeof(canary), 0); 752 destp -= roundup(sizeof(canary), sizeof(uint32_t)); 753 imgp->canary = (void *)destp; 754 error = copyout(canary, imgp->canary, sizeof(canary)); 755 if (error != 0) 756 return (error); 757 758 /* Allocate room for the argument and environment strings. */ 759 destp -= ARG_MAX - imgp->args->stringspace; 760 destp = rounddown2(destp, sizeof(uint32_t)); 761 ustringp = destp; 762 763 if (imgp->auxargs) { 764 /* 765 * Allocate room on the stack for the ELF auxargs 766 * array. It has LINUX_AT_COUNT entries. 767 */ 768 destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo); 769 destp = rounddown2(destp, sizeof(uint32_t)); 770 } 771 772 vectp = (uint32_t *)destp; 773 774 /* 775 * Allocate room for the argv[] and env vectors including the 776 * terminating NULL pointers. 777 */ 778 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 779 780 /* vectp also becomes our initial stack base. */ 781 *stack_base = (uintptr_t)vectp; 782 783 stringp = imgp->args->begin_argv; 784 argc = imgp->args->argc; 785 envc = imgp->args->envc; 786 787 /* Copy out strings - arguments and environment. */ 788 error = copyout(stringp, (void *)ustringp, 789 ARG_MAX - imgp->args->stringspace); 790 if (error != 0) 791 return (error); 792 793 /* Fill in "ps_strings" struct for ps, w, etc. */ 794 if (suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp) != 0 || 795 suword32(&arginfo->ps_nargvstr, argc) != 0) 796 return (EFAULT); 797 798 /* Fill in argument portion of vector table. */ 799 for (; argc > 0; --argc) { 800 if (suword32(vectp++, ustringp) != 0) 801 return (EFAULT); 802 while (*stringp++ != 0) 803 ustringp++; 804 ustringp++; 805 } 806 807 /* A null vector table pointer separates the argp's from the envp's. */ 808 if (suword32(vectp++, 0) != 0) 809 return (EFAULT); 810 811 if (suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp) != 0 || 812 suword32(&arginfo->ps_nenvstr, envc) != 0) 813 return (EFAULT); 814 815 /* Fill in environment portion of vector table. */ 816 for (; envc > 0; --envc) { 817 if (suword32(vectp++, ustringp) != 0) 818 return (EFAULT); 819 while (*stringp++ != 0) 820 ustringp++; 821 ustringp++; 822 } 823 824 /* The end of the vector table is a null pointer. */ 825 if (suword32(vectp, 0) != 0) 826 return (EFAULT); 827 828 if (imgp->auxargs) { 829 vectp++; 830 error = imgp->sysent->sv_copyout_auxargs(imgp, 831 (uintptr_t)vectp); 832 if (error != 0) 833 return (error); 834 } 835 836 return (0); 837 } 838 839 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 840 "32-bit Linux emulation"); 841 842 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ; 843 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW, 844 &linux32_maxdsiz, 0, ""); 845 static u_long linux32_maxssiz = LINUX32_MAXSSIZ; 846 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW, 847 &linux32_maxssiz, 0, ""); 848 static u_long linux32_maxvmem = LINUX32_MAXVMEM; 849 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW, 850 &linux32_maxvmem, 0, ""); 851 852 static void 853 linux32_fixlimit(struct rlimit *rl, int which) 854 { 855 856 switch (which) { 857 case RLIMIT_DATA: 858 if (linux32_maxdsiz != 0) { 859 if (rl->rlim_cur > linux32_maxdsiz) 860 rl->rlim_cur = linux32_maxdsiz; 861 if (rl->rlim_max > linux32_maxdsiz) 862 rl->rlim_max = linux32_maxdsiz; 863 } 864 break; 865 case RLIMIT_STACK: 866 if (linux32_maxssiz != 0) { 867 if (rl->rlim_cur > linux32_maxssiz) 868 rl->rlim_cur = linux32_maxssiz; 869 if (rl->rlim_max > linux32_maxssiz) 870 rl->rlim_max = linux32_maxssiz; 871 } 872 break; 873 case RLIMIT_VMEM: 874 if (linux32_maxvmem != 0) { 875 if (rl->rlim_cur > linux32_maxvmem) 876 rl->rlim_cur = linux32_maxvmem; 877 if (rl->rlim_max > linux32_maxvmem) 878 rl->rlim_max = linux32_maxvmem; 879 } 880 break; 881 } 882 } 883 884 struct sysentvec elf_linux_sysvec = { 885 .sv_size = LINUX32_SYS_MAXSYSCALL, 886 .sv_table = linux32_sysent, 887 .sv_errsize = ELAST + 1, 888 .sv_errtbl = linux_errtbl, 889 .sv_transtrap = linux_translate_traps, 890 .sv_fixup = linux_fixup_elf, 891 .sv_sendsig = linux_sendsig, 892 .sv_sigcode = &_binary_linux32_locore_o_start, 893 .sv_szsigcode = &linux_szsigcode, 894 .sv_name = "Linux ELF32", 895 .sv_coredump = elf32_coredump, 896 .sv_imgact_try = linux_exec_imgact_try, 897 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 898 .sv_minuser = VM_MIN_ADDRESS, 899 .sv_maxuser = LINUX32_MAXUSER, 900 .sv_usrstack = LINUX32_USRSTACK, 901 .sv_psstrings = LINUX32_PS_STRINGS, 902 .sv_stackprot = VM_PROT_ALL, 903 .sv_copyout_auxargs = linux_copyout_auxargs, 904 .sv_copyout_strings = linux_copyout_strings, 905 .sv_setregs = linux_exec_setregs, 906 .sv_fixlimit = linux32_fixlimit, 907 .sv_maxssiz = &linux32_maxssiz, 908 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP, 909 .sv_set_syscall_retval = cpu_set_syscall_retval, 910 .sv_fetch_syscall_args = linux32_fetch_syscall_args, 911 .sv_syscallnames = NULL, 912 .sv_shared_page_base = LINUX32_SHAREDPAGE, 913 .sv_shared_page_len = PAGE_SIZE, 914 .sv_schedtail = linux_schedtail, 915 .sv_thread_detach = linux_thread_detach, 916 .sv_trap = NULL, 917 }; 918 919 static void 920 linux_vdso_install(void *param) 921 { 922 923 linux_szsigcode = (&_binary_linux32_locore_o_end - 924 &_binary_linux32_locore_o_start); 925 926 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 927 panic("Linux invalid vdso size\n"); 928 929 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 930 931 linux_shared_page_obj = __elfN(linux_shared_page_init) 932 (&linux_shared_page_mapping); 933 934 __elfN(linux_vdso_reloc)(&elf_linux_sysvec); 935 936 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 937 linux_szsigcode); 938 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 939 940 linux_kplatform = linux_shared_page_mapping + 941 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base); 942 } 943 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 944 linux_vdso_install, NULL); 945 946 static void 947 linux_vdso_deinstall(void *param) 948 { 949 950 __elfN(linux_shared_page_fini)(linux_shared_page_obj); 951 } 952 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 953 linux_vdso_deinstall, NULL); 954 955 static char GNU_ABI_VENDOR[] = "GNU"; 956 static int GNULINUX_ABI_DESC = 0; 957 958 static bool 959 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel) 960 { 961 const Elf32_Word *desc; 962 uintptr_t p; 963 964 p = (uintptr_t)(note + 1); 965 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 966 967 desc = (const Elf32_Word *)p; 968 if (desc[0] != GNULINUX_ABI_DESC) 969 return (false); 970 971 /* 972 * For Linux we encode osrel using the Linux convention of 973 * (version << 16) | (major << 8) | (minor) 974 * See macro in linux_mib.h 975 */ 976 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]); 977 978 return (true); 979 } 980 981 static Elf_Brandnote linux32_brandnote = { 982 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 983 .hdr.n_descsz = 16, /* XXX at least 16 */ 984 .hdr.n_type = 1, 985 .vendor = GNU_ABI_VENDOR, 986 .flags = BN_TRANSLATE_OSREL, 987 .trans_osrel = linux32_trans_osrel 988 }; 989 990 static Elf32_Brandinfo linux_brand = { 991 .brand = ELFOSABI_LINUX, 992 .machine = EM_386, 993 .compat_3_brand = "Linux", 994 .emul_path = linux_emul_path, 995 .interp_path = "/lib/ld-linux.so.1", 996 .sysvec = &elf_linux_sysvec, 997 .interp_newpath = NULL, 998 .brand_note = &linux32_brandnote, 999 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1000 }; 1001 1002 static Elf32_Brandinfo linux_glibc2brand = { 1003 .brand = ELFOSABI_LINUX, 1004 .machine = EM_386, 1005 .compat_3_brand = "Linux", 1006 .emul_path = linux_emul_path, 1007 .interp_path = "/lib/ld-linux.so.2", 1008 .sysvec = &elf_linux_sysvec, 1009 .interp_newpath = NULL, 1010 .brand_note = &linux32_brandnote, 1011 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1012 }; 1013 1014 static Elf32_Brandinfo linux_muslbrand = { 1015 .brand = ELFOSABI_LINUX, 1016 .machine = EM_386, 1017 .compat_3_brand = "Linux", 1018 .emul_path = linux_emul_path, 1019 .interp_path = "/lib/ld-musl-i386.so.1", 1020 .sysvec = &elf_linux_sysvec, 1021 .interp_newpath = NULL, 1022 .brand_note = &linux32_brandnote, 1023 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 1024 }; 1025 1026 Elf32_Brandinfo *linux_brandlist[] = { 1027 &linux_brand, 1028 &linux_glibc2brand, 1029 &linux_muslbrand, 1030 NULL 1031 }; 1032 1033 static int 1034 linux_elf_modevent(module_t mod, int type, void *data) 1035 { 1036 Elf32_Brandinfo **brandinfo; 1037 int error; 1038 struct linux_ioctl_handler **lihp; 1039 1040 error = 0; 1041 1042 switch(type) { 1043 case MOD_LOAD: 1044 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1045 ++brandinfo) 1046 if (elf32_insert_brand_entry(*brandinfo) < 0) 1047 error = EINVAL; 1048 if (error == 0) { 1049 SET_FOREACH(lihp, linux_ioctl_handler_set) 1050 linux32_ioctl_register_handler(*lihp); 1051 stclohz = (stathz ? stathz : hz); 1052 if (bootverbose) 1053 printf("Linux ELF exec handler installed\n"); 1054 } else 1055 printf("cannot insert Linux ELF brand handler\n"); 1056 break; 1057 case MOD_UNLOAD: 1058 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 1059 ++brandinfo) 1060 if (elf32_brand_inuse(*brandinfo)) 1061 error = EBUSY; 1062 if (error == 0) { 1063 for (brandinfo = &linux_brandlist[0]; 1064 *brandinfo != NULL; ++brandinfo) 1065 if (elf32_remove_brand_entry(*brandinfo) < 0) 1066 error = EINVAL; 1067 } 1068 if (error == 0) { 1069 SET_FOREACH(lihp, linux_ioctl_handler_set) 1070 linux32_ioctl_unregister_handler(*lihp); 1071 if (bootverbose) 1072 printf("Linux ELF exec handler removed\n"); 1073 } else 1074 printf("Could not deinstall ELF interpreter entry\n"); 1075 break; 1076 default: 1077 return (EOPNOTSUPP); 1078 } 1079 return (error); 1080 } 1081 1082 static moduledata_t linux_elf_mod = { 1083 "linuxelf", 1084 linux_elf_modevent, 1085 0 1086 }; 1087 1088 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 1089 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1); 1090 FEATURE(linux, "Linux 32bit support"); 1091