1 /*- 2 * Copyright (c) 2004 Tim J. Robbins 3 * Copyright (c) 2002 Doug Rabson 4 * Copyright (c) 2000 Marcel Moolenaar 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer 12 * in this position and unchanged. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. The name of the author may not be used to endorse or promote products 17 * derived from this software without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/kernel.h> 36 #include <sys/systm.h> 37 #include <sys/file.h> 38 #include <sys/fcntl.h> 39 #include <sys/clock.h> 40 #include <sys/imgact.h> 41 #include <sys/limits.h> 42 #include <sys/lock.h> 43 #include <sys/malloc.h> 44 #include <sys/mman.h> 45 #include <sys/mutex.h> 46 #include <sys/priv.h> 47 #include <sys/proc.h> 48 #include <sys/resource.h> 49 #include <sys/resourcevar.h> 50 #include <sys/sched.h> 51 #include <sys/syscallsubr.h> 52 #include <sys/sysproto.h> 53 #include <sys/unistd.h> 54 55 #include <machine/frame.h> 56 #include <machine/pcb.h> 57 #include <machine/psl.h> 58 #include <machine/segments.h> 59 #include <machine/specialreg.h> 60 61 #include <vm/vm.h> 62 #include <vm/pmap.h> 63 #include <vm/vm_extern.h> 64 #include <vm/vm_kern.h> 65 #include <vm/vm_map.h> 66 67 #include <amd64/linux32/linux.h> 68 #include <amd64/linux32/linux32_proto.h> 69 #include <compat/linux/linux_ipc.h> 70 #include <compat/linux/linux_signal.h> 71 #include <compat/linux/linux_util.h> 72 #include <compat/linux/linux_emul.h> 73 74 struct l_old_select_argv { 75 l_int nfds; 76 l_uintptr_t readfds; 77 l_uintptr_t writefds; 78 l_uintptr_t exceptfds; 79 l_uintptr_t timeout; 80 } __packed; 81 82 int 83 linux_to_bsd_sigaltstack(int lsa) 84 { 85 int bsa = 0; 86 87 if (lsa & LINUX_SS_DISABLE) 88 bsa |= SS_DISABLE; 89 if (lsa & LINUX_SS_ONSTACK) 90 bsa |= SS_ONSTACK; 91 return (bsa); 92 } 93 94 int 95 bsd_to_linux_sigaltstack(int bsa) 96 { 97 int lsa = 0; 98 99 if (bsa & SS_DISABLE) 100 lsa |= LINUX_SS_DISABLE; 101 if (bsa & SS_ONSTACK) 102 lsa |= LINUX_SS_ONSTACK; 103 return (lsa); 104 } 105 106 /* 107 * Custom version of exec_copyin_args() so that we can translate 108 * the pointers. 109 */ 110 static int 111 linux_exec_copyin_args(struct image_args *args, char *fname, 112 enum uio_seg segflg, char **argv, char **envv) 113 { 114 char *argp, *envp; 115 u_int32_t *p32, arg; 116 size_t length; 117 int error; 118 119 bzero(args, sizeof(*args)); 120 if (argv == NULL) 121 return (EFAULT); 122 123 /* 124 * Allocate temporary demand zeroed space for argument and 125 * environment strings 126 */ 127 args->buf = (char *)kmem_alloc_wait(exec_map, 128 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); 129 if (args->buf == NULL) 130 return (ENOMEM); 131 args->begin_argv = args->buf; 132 args->endp = args->begin_argv; 133 args->stringspace = ARG_MAX; 134 135 args->fname = args->buf + ARG_MAX; 136 137 /* 138 * Copy the file name. 139 */ 140 error = (segflg == UIO_SYSSPACE) ? 141 copystr(fname, args->fname, PATH_MAX, &length) : 142 copyinstr(fname, args->fname, PATH_MAX, &length); 143 if (error != 0) 144 goto err_exit; 145 146 /* 147 * extract arguments first 148 */ 149 p32 = (u_int32_t *)argv; 150 for (;;) { 151 error = copyin(p32++, &arg, sizeof(arg)); 152 if (error) 153 goto err_exit; 154 if (arg == 0) 155 break; 156 argp = PTRIN(arg); 157 error = copyinstr(argp, args->endp, args->stringspace, &length); 158 if (error) { 159 if (error == ENAMETOOLONG) 160 error = E2BIG; 161 162 goto err_exit; 163 } 164 args->stringspace -= length; 165 args->endp += length; 166 args->argc++; 167 } 168 169 args->begin_envv = args->endp; 170 171 /* 172 * extract environment strings 173 */ 174 if (envv) { 175 p32 = (u_int32_t *)envv; 176 for (;;) { 177 error = copyin(p32++, &arg, sizeof(arg)); 178 if (error) 179 goto err_exit; 180 if (arg == 0) 181 break; 182 envp = PTRIN(arg); 183 error = copyinstr(envp, args->endp, args->stringspace, 184 &length); 185 if (error) { 186 if (error == ENAMETOOLONG) 187 error = E2BIG; 188 goto err_exit; 189 } 190 args->stringspace -= length; 191 args->endp += length; 192 args->envc++; 193 } 194 } 195 196 return (0); 197 198 err_exit: 199 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf, 200 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); 201 args->buf = NULL; 202 return (error); 203 } 204 205 int 206 linux_execve(struct thread *td, struct linux_execve_args *args) 207 { 208 struct image_args eargs; 209 char *path; 210 int error; 211 212 LCONVPATHEXIST(td, args->path, &path); 213 214 #ifdef DEBUG 215 if (ldebug(execve)) 216 printf(ARGS(execve, "%s"), path); 217 #endif 218 219 error = linux_exec_copyin_args(&eargs, path, UIO_SYSSPACE, args->argp, 220 args->envp); 221 free(path, M_TEMP); 222 if (error == 0) 223 error = kern_execve(td, &eargs, NULL); 224 if (error == 0) 225 /* Linux process can execute FreeBSD one, do not attempt 226 * to create emuldata for such process using 227 * linux_proc_init, this leads to a panic on KASSERT 228 * because such process has p->p_emuldata == NULL. 229 */ 230 if (td->td_proc->p_sysent == &elf_linux_sysvec) 231 error = linux_proc_init(td, 0, 0); 232 return (error); 233 } 234 235 struct iovec32 { 236 u_int32_t iov_base; 237 int iov_len; 238 }; 239 240 CTASSERT(sizeof(struct iovec32) == 8); 241 242 static int 243 linux32_copyinuio(struct iovec32 *iovp, u_int iovcnt, struct uio **uiop) 244 { 245 struct iovec32 iov32; 246 struct iovec *iov; 247 struct uio *uio; 248 u_int iovlen; 249 int error, i; 250 251 *uiop = NULL; 252 if (iovcnt > UIO_MAXIOV) 253 return (EINVAL); 254 iovlen = iovcnt * sizeof(struct iovec); 255 uio = malloc(iovlen + sizeof(*uio), M_IOV, M_WAITOK); 256 iov = (struct iovec *)(uio + 1); 257 for (i = 0; i < iovcnt; i++) { 258 error = copyin(&iovp[i], &iov32, sizeof(struct iovec32)); 259 if (error) { 260 free(uio, M_IOV); 261 return (error); 262 } 263 iov[i].iov_base = PTRIN(iov32.iov_base); 264 iov[i].iov_len = iov32.iov_len; 265 } 266 uio->uio_iov = iov; 267 uio->uio_iovcnt = iovcnt; 268 uio->uio_segflg = UIO_USERSPACE; 269 uio->uio_offset = -1; 270 uio->uio_resid = 0; 271 for (i = 0; i < iovcnt; i++) { 272 if (iov->iov_len > INT_MAX - uio->uio_resid) { 273 free(uio, M_IOV); 274 return (EINVAL); 275 } 276 uio->uio_resid += iov->iov_len; 277 iov++; 278 } 279 *uiop = uio; 280 return (0); 281 } 282 283 int 284 linux_readv(struct thread *td, struct linux_readv_args *uap) 285 { 286 struct uio *auio; 287 int error; 288 289 error = linux32_copyinuio(uap->iovp, uap->iovcnt, &auio); 290 if (error) 291 return (error); 292 error = kern_readv(td, uap->fd, auio); 293 free(auio, M_IOV); 294 return (error); 295 } 296 297 int 298 linux_writev(struct thread *td, struct linux_writev_args *uap) 299 { 300 struct uio *auio; 301 int error; 302 303 error = linux32_copyinuio(uap->iovp, uap->iovcnt, &auio); 304 if (error) 305 return (error); 306 error = kern_writev(td, uap->fd, auio); 307 free(auio, M_IOV); 308 return (error); 309 } 310 311 struct l_ipc_kludge { 312 l_uintptr_t msgp; 313 l_long msgtyp; 314 } __packed; 315 316 int 317 linux_ipc(struct thread *td, struct linux_ipc_args *args) 318 { 319 320 switch (args->what & 0xFFFF) { 321 case LINUX_SEMOP: { 322 struct linux_semop_args a; 323 324 a.semid = args->arg1; 325 a.tsops = args->ptr; 326 a.nsops = args->arg2; 327 return (linux_semop(td, &a)); 328 } 329 case LINUX_SEMGET: { 330 struct linux_semget_args a; 331 332 a.key = args->arg1; 333 a.nsems = args->arg2; 334 a.semflg = args->arg3; 335 return (linux_semget(td, &a)); 336 } 337 case LINUX_SEMCTL: { 338 struct linux_semctl_args a; 339 int error; 340 341 a.semid = args->arg1; 342 a.semnum = args->arg2; 343 a.cmd = args->arg3; 344 error = copyin(args->ptr, &a.arg, sizeof(a.arg)); 345 if (error) 346 return (error); 347 return (linux_semctl(td, &a)); 348 } 349 case LINUX_MSGSND: { 350 struct linux_msgsnd_args a; 351 352 a.msqid = args->arg1; 353 a.msgp = args->ptr; 354 a.msgsz = args->arg2; 355 a.msgflg = args->arg3; 356 return (linux_msgsnd(td, &a)); 357 } 358 case LINUX_MSGRCV: { 359 struct linux_msgrcv_args a; 360 361 a.msqid = args->arg1; 362 a.msgsz = args->arg2; 363 a.msgflg = args->arg3; 364 if ((args->what >> 16) == 0) { 365 struct l_ipc_kludge tmp; 366 int error; 367 368 if (args->ptr == 0) 369 return (EINVAL); 370 error = copyin(args->ptr, &tmp, sizeof(tmp)); 371 if (error) 372 return (error); 373 a.msgp = PTRIN(tmp.msgp); 374 a.msgtyp = tmp.msgtyp; 375 } else { 376 a.msgp = args->ptr; 377 a.msgtyp = args->arg5; 378 } 379 return (linux_msgrcv(td, &a)); 380 } 381 case LINUX_MSGGET: { 382 struct linux_msgget_args a; 383 384 a.key = args->arg1; 385 a.msgflg = args->arg2; 386 return (linux_msgget(td, &a)); 387 } 388 case LINUX_MSGCTL: { 389 struct linux_msgctl_args a; 390 391 a.msqid = args->arg1; 392 a.cmd = args->arg2; 393 a.buf = args->ptr; 394 return (linux_msgctl(td, &a)); 395 } 396 case LINUX_SHMAT: { 397 struct linux_shmat_args a; 398 399 a.shmid = args->arg1; 400 a.shmaddr = args->ptr; 401 a.shmflg = args->arg2; 402 a.raddr = PTRIN((l_uint)args->arg3); 403 return (linux_shmat(td, &a)); 404 } 405 case LINUX_SHMDT: { 406 struct linux_shmdt_args a; 407 408 a.shmaddr = args->ptr; 409 return (linux_shmdt(td, &a)); 410 } 411 case LINUX_SHMGET: { 412 struct linux_shmget_args a; 413 414 a.key = args->arg1; 415 a.size = args->arg2; 416 a.shmflg = args->arg3; 417 return (linux_shmget(td, &a)); 418 } 419 case LINUX_SHMCTL: { 420 struct linux_shmctl_args a; 421 422 a.shmid = args->arg1; 423 a.cmd = args->arg2; 424 a.buf = args->ptr; 425 return (linux_shmctl(td, &a)); 426 } 427 default: 428 break; 429 } 430 431 return (EINVAL); 432 } 433 434 int 435 linux_old_select(struct thread *td, struct linux_old_select_args *args) 436 { 437 struct l_old_select_argv linux_args; 438 struct linux_select_args newsel; 439 int error; 440 441 #ifdef DEBUG 442 if (ldebug(old_select)) 443 printf(ARGS(old_select, "%p"), args->ptr); 444 #endif 445 446 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 447 if (error) 448 return (error); 449 450 newsel.nfds = linux_args.nfds; 451 newsel.readfds = PTRIN(linux_args.readfds); 452 newsel.writefds = PTRIN(linux_args.writefds); 453 newsel.exceptfds = PTRIN(linux_args.exceptfds); 454 newsel.timeout = PTRIN(linux_args.timeout); 455 return (linux_select(td, &newsel)); 456 } 457 458 int 459 linux_fork(struct thread *td, struct linux_fork_args *args) 460 { 461 int error; 462 struct proc *p2; 463 struct thread *td2; 464 465 #ifdef DEBUG 466 if (ldebug(fork)) 467 printf(ARGS(fork, "")); 468 #endif 469 470 if ((error = fork1(td, RFFDG | RFPROC | RFSTOPPED, 0, &p2)) != 0) 471 return (error); 472 473 if (error == 0) { 474 td->td_retval[0] = p2->p_pid; 475 td->td_retval[1] = 0; 476 } 477 478 if (td->td_retval[1] == 1) 479 td->td_retval[0] = 0; 480 error = linux_proc_init(td, td->td_retval[0], 0); 481 if (error) 482 return (error); 483 484 td2 = FIRST_THREAD_IN_PROC(p2); 485 486 /* 487 * Make this runnable after we are finished with it. 488 */ 489 thread_lock(td2); 490 TD_SET_CAN_RUN(td2); 491 sched_add(td2, SRQ_BORING); 492 thread_unlock(td2); 493 494 return (0); 495 } 496 497 int 498 linux_vfork(struct thread *td, struct linux_vfork_args *args) 499 { 500 int error; 501 struct proc *p2; 502 struct thread *td2; 503 504 #ifdef DEBUG 505 if (ldebug(vfork)) 506 printf(ARGS(vfork, "")); 507 #endif 508 509 /* Exclude RFPPWAIT */ 510 if ((error = fork1(td, RFFDG | RFPROC | RFMEM | RFSTOPPED, 0, &p2)) != 0) 511 return (error); 512 if (error == 0) { 513 td->td_retval[0] = p2->p_pid; 514 td->td_retval[1] = 0; 515 } 516 /* Are we the child? */ 517 if (td->td_retval[1] == 1) 518 td->td_retval[0] = 0; 519 error = linux_proc_init(td, td->td_retval[0], 0); 520 if (error) 521 return (error); 522 523 PROC_LOCK(p2); 524 p2->p_flag |= P_PPWAIT; 525 PROC_UNLOCK(p2); 526 527 td2 = FIRST_THREAD_IN_PROC(p2); 528 529 /* 530 * Make this runnable after we are finished with it. 531 */ 532 thread_lock(td2); 533 TD_SET_CAN_RUN(td2); 534 sched_add(td2, SRQ_BORING); 535 thread_unlock(td2); 536 537 /* wait for the children to exit, ie. emulate vfork */ 538 PROC_LOCK(p2); 539 while (p2->p_flag & P_PPWAIT) 540 msleep(td->td_proc, &p2->p_mtx, PWAIT, "ppwait", 0); 541 PROC_UNLOCK(p2); 542 543 return (0); 544 } 545 546 int 547 linux_clone(struct thread *td, struct linux_clone_args *args) 548 { 549 int error, ff = RFPROC | RFSTOPPED; 550 struct proc *p2; 551 struct thread *td2; 552 int exit_signal; 553 struct linux_emuldata *em; 554 555 #ifdef DEBUG 556 if (ldebug(clone)) { 557 printf(ARGS(clone, "flags %x, stack %p, parent tid: %p, " 558 "child tid: %p"), (unsigned)args->flags, 559 args->stack, args->parent_tidptr, args->child_tidptr); 560 } 561 #endif 562 563 exit_signal = args->flags & 0x000000ff; 564 if (LINUX_SIG_VALID(exit_signal)) { 565 if (exit_signal <= LINUX_SIGTBLSZ) 566 exit_signal = 567 linux_to_bsd_signal[_SIG_IDX(exit_signal)]; 568 } else if (exit_signal != 0) 569 return (EINVAL); 570 571 if (args->flags & LINUX_CLONE_VM) 572 ff |= RFMEM; 573 if (args->flags & LINUX_CLONE_SIGHAND) 574 ff |= RFSIGSHARE; 575 /* 576 * XXX: In Linux, sharing of fs info (chroot/cwd/umask) 577 * and open files is independant. In FreeBSD, its in one 578 * structure but in reality it does not cause any problems 579 * because both of these flags are usually set together. 580 */ 581 if (!(args->flags & (LINUX_CLONE_FILES | LINUX_CLONE_FS))) 582 ff |= RFFDG; 583 584 /* 585 * Attempt to detect when linux_clone(2) is used for creating 586 * kernel threads. Unfortunately despite the existence of the 587 * CLONE_THREAD flag, version of linuxthreads package used in 588 * most popular distros as of beginning of 2005 doesn't make 589 * any use of it. Therefore, this detection relies on 590 * empirical observation that linuxthreads sets certain 591 * combination of flags, so that we can make more or less 592 * precise detection and notify the FreeBSD kernel that several 593 * processes are in fact part of the same threading group, so 594 * that special treatment is necessary for signal delivery 595 * between those processes and fd locking. 596 */ 597 if ((args->flags & 0xffffff00) == LINUX_THREADING_FLAGS) 598 ff |= RFTHREAD; 599 600 if (args->flags & LINUX_CLONE_PARENT_SETTID) 601 if (args->parent_tidptr == NULL) 602 return (EINVAL); 603 604 error = fork1(td, ff, 0, &p2); 605 if (error) 606 return (error); 607 608 if (args->flags & (LINUX_CLONE_PARENT | LINUX_CLONE_THREAD)) { 609 sx_xlock(&proctree_lock); 610 PROC_LOCK(p2); 611 proc_reparent(p2, td->td_proc->p_pptr); 612 PROC_UNLOCK(p2); 613 sx_xunlock(&proctree_lock); 614 } 615 616 /* create the emuldata */ 617 error = linux_proc_init(td, p2->p_pid, args->flags); 618 /* reference it - no need to check this */ 619 em = em_find(p2, EMUL_DOLOCK); 620 KASSERT(em != NULL, ("clone: emuldata not found.\n")); 621 /* and adjust it */ 622 623 if (args->flags & LINUX_CLONE_THREAD) { 624 #ifdef notyet 625 PROC_LOCK(p2); 626 p2->p_pgrp = td->td_proc->p_pgrp; 627 PROC_UNLOCK(p2); 628 #endif 629 exit_signal = 0; 630 } 631 632 if (args->flags & LINUX_CLONE_CHILD_SETTID) 633 em->child_set_tid = args->child_tidptr; 634 else 635 em->child_set_tid = NULL; 636 637 if (args->flags & LINUX_CLONE_CHILD_CLEARTID) 638 em->child_clear_tid = args->child_tidptr; 639 else 640 em->child_clear_tid = NULL; 641 642 EMUL_UNLOCK(&emul_lock); 643 644 if (args->flags & LINUX_CLONE_PARENT_SETTID) { 645 error = copyout(&p2->p_pid, args->parent_tidptr, 646 sizeof(p2->p_pid)); 647 if (error) 648 printf(LMSG("copyout failed!")); 649 } 650 651 PROC_LOCK(p2); 652 p2->p_sigparent = exit_signal; 653 PROC_UNLOCK(p2); 654 td2 = FIRST_THREAD_IN_PROC(p2); 655 /* 656 * In a case of stack = NULL, we are supposed to COW calling process 657 * stack. This is what normal fork() does, so we just keep tf_rsp arg 658 * intact. 659 */ 660 if (args->stack) 661 td2->td_frame->tf_rsp = PTROUT(args->stack); 662 663 if (args->flags & LINUX_CLONE_SETTLS) { 664 struct user_segment_descriptor sd; 665 struct l_user_desc info; 666 int a[2]; 667 668 error = copyin((void *)td->td_frame->tf_rsi, &info, 669 sizeof(struct l_user_desc)); 670 if (error) { 671 printf(LMSG("copyin failed!")); 672 } else { 673 /* We might copy out the entry_number as GUGS32_SEL. */ 674 info.entry_number = GUGS32_SEL; 675 error = copyout(&info, (void *)td->td_frame->tf_rsi, 676 sizeof(struct l_user_desc)); 677 if (error) 678 printf(LMSG("copyout failed!")); 679 680 a[0] = LINUX_LDT_entry_a(&info); 681 a[1] = LINUX_LDT_entry_b(&info); 682 683 memcpy(&sd, &a, sizeof(a)); 684 #ifdef DEBUG 685 if (ldebug(clone)) 686 printf("Segment created in clone with " 687 "CLONE_SETTLS: lobase: %x, hibase: %x, " 688 "lolimit: %x, hilimit: %x, type: %i, " 689 "dpl: %i, p: %i, xx: %i, long: %i, " 690 "def32: %i, gran: %i\n", sd.sd_lobase, 691 sd.sd_hibase, sd.sd_lolimit, sd.sd_hilimit, 692 sd.sd_type, sd.sd_dpl, sd.sd_p, sd.sd_xx, 693 sd.sd_long, sd.sd_def32, sd.sd_gran); 694 #endif 695 td2->td_pcb->pcb_gsbase = (register_t)info.base_addr; 696 td2->td_pcb->pcb_gs32sd = sd; 697 td2->td_pcb->pcb_gs32p = &gdt[GUGS32_SEL]; 698 td2->td_pcb->pcb_gs = GSEL(GUGS32_SEL, SEL_UPL); 699 td2->td_pcb->pcb_flags |= PCB_32BIT; 700 } 701 } 702 703 #ifdef DEBUG 704 if (ldebug(clone)) 705 printf(LMSG("clone: successful rfork to %d, " 706 "stack %p sig = %d"), (int)p2->p_pid, args->stack, 707 exit_signal); 708 #endif 709 if (args->flags & LINUX_CLONE_VFORK) { 710 PROC_LOCK(p2); 711 p2->p_flag |= P_PPWAIT; 712 PROC_UNLOCK(p2); 713 } 714 715 /* 716 * Make this runnable after we are finished with it. 717 */ 718 thread_lock(td2); 719 TD_SET_CAN_RUN(td2); 720 sched_add(td2, SRQ_BORING); 721 thread_unlock(td2); 722 723 td->td_retval[0] = p2->p_pid; 724 td->td_retval[1] = 0; 725 726 if (args->flags & LINUX_CLONE_VFORK) { 727 /* wait for the children to exit, ie. emulate vfork */ 728 PROC_LOCK(p2); 729 while (p2->p_flag & P_PPWAIT) 730 msleep(td->td_proc, &p2->p_mtx, PWAIT, "ppwait", 0); 731 PROC_UNLOCK(p2); 732 } 733 734 return (0); 735 } 736 737 #define STACK_SIZE (2 * 1024 * 1024) 738 #define GUARD_SIZE (4 * PAGE_SIZE) 739 740 static int linux_mmap_common(struct thread *, struct l_mmap_argv *); 741 742 int 743 linux_mmap2(struct thread *td, struct linux_mmap2_args *args) 744 { 745 struct l_mmap_argv linux_args; 746 747 #ifdef DEBUG 748 if (ldebug(mmap2)) 749 printf(ARGS(mmap2, "0x%08x, %d, %d, 0x%08x, %d, %d"), 750 args->addr, args->len, args->prot, 751 args->flags, args->fd, args->pgoff); 752 #endif 753 754 linux_args.addr = PTROUT(args->addr); 755 linux_args.len = args->len; 756 linux_args.prot = args->prot; 757 linux_args.flags = args->flags; 758 linux_args.fd = args->fd; 759 linux_args.pgoff = args->pgoff; 760 761 return (linux_mmap_common(td, &linux_args)); 762 } 763 764 int 765 linux_mmap(struct thread *td, struct linux_mmap_args *args) 766 { 767 int error; 768 struct l_mmap_argv linux_args; 769 770 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 771 if (error) 772 return (error); 773 774 #ifdef DEBUG 775 if (ldebug(mmap)) 776 printf(ARGS(mmap, "0x%08x, %d, %d, 0x%08x, %d, %d"), 777 linux_args.addr, linux_args.len, linux_args.prot, 778 linux_args.flags, linux_args.fd, linux_args.pgoff); 779 #endif 780 if ((linux_args.pgoff % PAGE_SIZE) != 0) 781 return (EINVAL); 782 linux_args.pgoff /= PAGE_SIZE; 783 784 return (linux_mmap_common(td, &linux_args)); 785 } 786 787 static int 788 linux_mmap_common(struct thread *td, struct l_mmap_argv *linux_args) 789 { 790 struct proc *p = td->td_proc; 791 struct mmap_args /* { 792 caddr_t addr; 793 size_t len; 794 int prot; 795 int flags; 796 int fd; 797 long pad; 798 off_t pos; 799 } */ bsd_args; 800 int error; 801 struct file *fp; 802 803 error = 0; 804 bsd_args.flags = 0; 805 fp = NULL; 806 807 /* 808 * Linux mmap(2): 809 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE 810 */ 811 if (! ((linux_args->flags & LINUX_MAP_SHARED) ^ 812 (linux_args->flags & LINUX_MAP_PRIVATE))) 813 return (EINVAL); 814 815 if (linux_args->flags & LINUX_MAP_SHARED) 816 bsd_args.flags |= MAP_SHARED; 817 if (linux_args->flags & LINUX_MAP_PRIVATE) 818 bsd_args.flags |= MAP_PRIVATE; 819 if (linux_args->flags & LINUX_MAP_FIXED) 820 bsd_args.flags |= MAP_FIXED; 821 if (linux_args->flags & LINUX_MAP_ANON) 822 bsd_args.flags |= MAP_ANON; 823 else 824 bsd_args.flags |= MAP_NOSYNC; 825 if (linux_args->flags & LINUX_MAP_GROWSDOWN) 826 bsd_args.flags |= MAP_STACK; 827 828 /* 829 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC 830 * on Linux/i386. We do this to ensure maximum compatibility. 831 * Linux/ia64 does the same in i386 emulation mode. 832 */ 833 bsd_args.prot = linux_args->prot; 834 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 835 bsd_args.prot |= PROT_READ | PROT_EXEC; 836 837 /* Linux does not check file descriptor when MAP_ANONYMOUS is set. */ 838 bsd_args.fd = (bsd_args.flags & MAP_ANON) ? -1 : linux_args->fd; 839 if (bsd_args.fd != -1) { 840 /* 841 * Linux follows Solaris mmap(2) description: 842 * The file descriptor fildes is opened with 843 * read permission, regardless of the 844 * protection options specified. 845 */ 846 847 if ((error = fget(td, bsd_args.fd, &fp)) != 0) 848 return (error); 849 if (fp->f_type != DTYPE_VNODE) { 850 fdrop(fp, td); 851 return (EINVAL); 852 } 853 854 /* Linux mmap() just fails for O_WRONLY files */ 855 if (!(fp->f_flag & FREAD)) { 856 fdrop(fp, td); 857 return (EACCES); 858 } 859 860 fdrop(fp, td); 861 } 862 863 if (linux_args->flags & LINUX_MAP_GROWSDOWN) { 864 /* 865 * The Linux MAP_GROWSDOWN option does not limit auto 866 * growth of the region. Linux mmap with this option 867 * takes as addr the inital BOS, and as len, the initial 868 * region size. It can then grow down from addr without 869 * limit. However, Linux threads has an implicit internal 870 * limit to stack size of STACK_SIZE. Its just not 871 * enforced explicitly in Linux. But, here we impose 872 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack 873 * region, since we can do this with our mmap. 874 * 875 * Our mmap with MAP_STACK takes addr as the maximum 876 * downsize limit on BOS, and as len the max size of 877 * the region. It then maps the top SGROWSIZ bytes, 878 * and auto grows the region down, up to the limit 879 * in addr. 880 * 881 * If we don't use the MAP_STACK option, the effect 882 * of this code is to allocate a stack region of a 883 * fixed size of (STACK_SIZE - GUARD_SIZE). 884 */ 885 886 if ((caddr_t)PTRIN(linux_args->addr) + linux_args->len > 887 p->p_vmspace->vm_maxsaddr) { 888 /* 889 * Some Linux apps will attempt to mmap 890 * thread stacks near the top of their 891 * address space. If their TOS is greater 892 * than vm_maxsaddr, vm_map_growstack() 893 * will confuse the thread stack with the 894 * process stack and deliver a SEGV if they 895 * attempt to grow the thread stack past their 896 * current stacksize rlimit. To avoid this, 897 * adjust vm_maxsaddr upwards to reflect 898 * the current stacksize rlimit rather 899 * than the maximum possible stacksize. 900 * It would be better to adjust the 901 * mmap'ed region, but some apps do not check 902 * mmap's return value. 903 */ 904 PROC_LOCK(p); 905 p->p_vmspace->vm_maxsaddr = (char *)LINUX32_USRSTACK - 906 lim_cur(p, RLIMIT_STACK); 907 PROC_UNLOCK(p); 908 } 909 910 /* 911 * This gives us our maximum stack size and a new BOS. 912 * If we're using VM_STACK, then mmap will just map 913 * the top SGROWSIZ bytes, and let the stack grow down 914 * to the limit at BOS. If we're not using VM_STACK 915 * we map the full stack, since we don't have a way 916 * to autogrow it. 917 */ 918 if (linux_args->len > STACK_SIZE - GUARD_SIZE) { 919 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr); 920 bsd_args.len = linux_args->len; 921 } else { 922 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr) - 923 (STACK_SIZE - GUARD_SIZE - linux_args->len); 924 bsd_args.len = STACK_SIZE - GUARD_SIZE; 925 } 926 } else { 927 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr); 928 bsd_args.len = linux_args->len; 929 } 930 bsd_args.pos = (off_t)linux_args->pgoff * PAGE_SIZE; 931 932 #ifdef DEBUG 933 if (ldebug(mmap)) 934 printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n", 935 __func__, 936 (void *)bsd_args.addr, (int)bsd_args.len, bsd_args.prot, 937 bsd_args.flags, bsd_args.fd, (int)bsd_args.pos); 938 #endif 939 error = mmap(td, &bsd_args); 940 #ifdef DEBUG 941 if (ldebug(mmap)) 942 printf("-> %s() return: 0x%x (0x%08x)\n", 943 __func__, error, (u_int)td->td_retval[0]); 944 #endif 945 return (error); 946 } 947 948 int 949 linux_mprotect(struct thread *td, struct linux_mprotect_args *uap) 950 { 951 struct mprotect_args bsd_args; 952 953 bsd_args.addr = uap->addr; 954 bsd_args.len = uap->len; 955 bsd_args.prot = uap->prot; 956 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 957 bsd_args.prot |= PROT_READ | PROT_EXEC; 958 return (mprotect(td, &bsd_args)); 959 } 960 961 int 962 linux_iopl(struct thread *td, struct linux_iopl_args *args) 963 { 964 int error; 965 966 if (args->level < 0 || args->level > 3) 967 return (EINVAL); 968 if ((error = priv_check(td, PRIV_IO)) != 0) 969 return (error); 970 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 971 return (error); 972 td->td_frame->tf_rflags = (td->td_frame->tf_rflags & ~PSL_IOPL) | 973 (args->level * (PSL_IOPL / 3)); 974 975 return (0); 976 } 977 978 int 979 linux_pipe(struct thread *td, struct linux_pipe_args *args) 980 { 981 int pip[2]; 982 int error; 983 register_t reg_rdx; 984 985 #ifdef DEBUG 986 if (ldebug(pipe)) 987 printf(ARGS(pipe, "*")); 988 #endif 989 990 reg_rdx = td->td_retval[1]; 991 error = pipe(td, 0); 992 if (error) { 993 td->td_retval[1] = reg_rdx; 994 return (error); 995 } 996 997 pip[0] = td->td_retval[0]; 998 pip[1] = td->td_retval[1]; 999 error = copyout(pip, args->pipefds, 2 * sizeof(int)); 1000 if (error) { 1001 td->td_retval[1] = reg_rdx; 1002 return (error); 1003 } 1004 1005 td->td_retval[1] = reg_rdx; 1006 td->td_retval[0] = 0; 1007 return (0); 1008 } 1009 1010 int 1011 linux_sigaction(struct thread *td, struct linux_sigaction_args *args) 1012 { 1013 l_osigaction_t osa; 1014 l_sigaction_t act, oact; 1015 int error; 1016 1017 #ifdef DEBUG 1018 if (ldebug(sigaction)) 1019 printf(ARGS(sigaction, "%d, %p, %p"), 1020 args->sig, (void *)args->nsa, (void *)args->osa); 1021 #endif 1022 1023 if (args->nsa != NULL) { 1024 error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); 1025 if (error) 1026 return (error); 1027 act.lsa_handler = osa.lsa_handler; 1028 act.lsa_flags = osa.lsa_flags; 1029 act.lsa_restorer = osa.lsa_restorer; 1030 LINUX_SIGEMPTYSET(act.lsa_mask); 1031 act.lsa_mask.__bits[0] = osa.lsa_mask; 1032 } 1033 1034 error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, 1035 args->osa ? &oact : NULL); 1036 1037 if (args->osa != NULL && !error) { 1038 osa.lsa_handler = oact.lsa_handler; 1039 osa.lsa_flags = oact.lsa_flags; 1040 osa.lsa_restorer = oact.lsa_restorer; 1041 osa.lsa_mask = oact.lsa_mask.__bits[0]; 1042 error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); 1043 } 1044 1045 return (error); 1046 } 1047 1048 /* 1049 * Linux has two extra args, restart and oldmask. We don't use these, 1050 * but it seems that "restart" is actually a context pointer that 1051 * enables the signal to happen with a different register set. 1052 */ 1053 int 1054 linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) 1055 { 1056 sigset_t sigmask; 1057 l_sigset_t mask; 1058 1059 #ifdef DEBUG 1060 if (ldebug(sigsuspend)) 1061 printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask); 1062 #endif 1063 1064 LINUX_SIGEMPTYSET(mask); 1065 mask.__bits[0] = args->mask; 1066 linux_to_bsd_sigset(&mask, &sigmask); 1067 return (kern_sigsuspend(td, sigmask)); 1068 } 1069 1070 int 1071 linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap) 1072 { 1073 l_sigset_t lmask; 1074 sigset_t sigmask; 1075 int error; 1076 1077 #ifdef DEBUG 1078 if (ldebug(rt_sigsuspend)) 1079 printf(ARGS(rt_sigsuspend, "%p, %d"), 1080 (void *)uap->newset, uap->sigsetsize); 1081 #endif 1082 1083 if (uap->sigsetsize != sizeof(l_sigset_t)) 1084 return (EINVAL); 1085 1086 error = copyin(uap->newset, &lmask, sizeof(l_sigset_t)); 1087 if (error) 1088 return (error); 1089 1090 linux_to_bsd_sigset(&lmask, &sigmask); 1091 return (kern_sigsuspend(td, sigmask)); 1092 } 1093 1094 int 1095 linux_pause(struct thread *td, struct linux_pause_args *args) 1096 { 1097 struct proc *p = td->td_proc; 1098 sigset_t sigmask; 1099 1100 #ifdef DEBUG 1101 if (ldebug(pause)) 1102 printf(ARGS(pause, "")); 1103 #endif 1104 1105 PROC_LOCK(p); 1106 sigmask = td->td_sigmask; 1107 PROC_UNLOCK(p); 1108 return (kern_sigsuspend(td, sigmask)); 1109 } 1110 1111 int 1112 linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap) 1113 { 1114 stack_t ss, oss; 1115 l_stack_t lss; 1116 int error; 1117 1118 #ifdef DEBUG 1119 if (ldebug(sigaltstack)) 1120 printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss); 1121 #endif 1122 1123 if (uap->uss != NULL) { 1124 error = copyin(uap->uss, &lss, sizeof(l_stack_t)); 1125 if (error) 1126 return (error); 1127 1128 ss.ss_sp = PTRIN(lss.ss_sp); 1129 ss.ss_size = lss.ss_size; 1130 ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags); 1131 } 1132 error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL, 1133 (uap->uoss != NULL) ? &oss : NULL); 1134 if (!error && uap->uoss != NULL) { 1135 lss.ss_sp = PTROUT(oss.ss_sp); 1136 lss.ss_size = oss.ss_size; 1137 lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags); 1138 error = copyout(&lss, uap->uoss, sizeof(l_stack_t)); 1139 } 1140 1141 return (error); 1142 } 1143 1144 int 1145 linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args) 1146 { 1147 struct ftruncate_args sa; 1148 1149 #ifdef DEBUG 1150 if (ldebug(ftruncate64)) 1151 printf(ARGS(ftruncate64, "%u, %jd"), args->fd, 1152 (intmax_t)args->length); 1153 #endif 1154 1155 sa.fd = args->fd; 1156 sa.length = args->length; 1157 return ftruncate(td, &sa); 1158 } 1159 1160 int 1161 linux_gettimeofday(struct thread *td, struct linux_gettimeofday_args *uap) 1162 { 1163 struct timeval atv; 1164 l_timeval atv32; 1165 struct timezone rtz; 1166 int error = 0; 1167 1168 if (uap->tp) { 1169 microtime(&atv); 1170 atv32.tv_sec = atv.tv_sec; 1171 atv32.tv_usec = atv.tv_usec; 1172 error = copyout(&atv32, uap->tp, sizeof(atv32)); 1173 } 1174 if (error == 0 && uap->tzp != NULL) { 1175 rtz.tz_minuteswest = tz_minuteswest; 1176 rtz.tz_dsttime = tz_dsttime; 1177 error = copyout(&rtz, uap->tzp, sizeof(rtz)); 1178 } 1179 return (error); 1180 } 1181 1182 int 1183 linux_settimeofday(struct thread *td, struct linux_settimeofday_args *uap) 1184 { 1185 l_timeval atv32; 1186 struct timeval atv, *tvp; 1187 struct timezone atz, *tzp; 1188 int error; 1189 1190 if (uap->tp) { 1191 error = copyin(uap->tp, &atv32, sizeof(atv32)); 1192 if (error) 1193 return (error); 1194 atv.tv_sec = atv32.tv_sec; 1195 atv.tv_usec = atv32.tv_usec; 1196 tvp = &atv; 1197 } else 1198 tvp = NULL; 1199 if (uap->tzp) { 1200 error = copyin(uap->tzp, &atz, sizeof(atz)); 1201 if (error) 1202 return (error); 1203 tzp = &atz; 1204 } else 1205 tzp = NULL; 1206 return (kern_settimeofday(td, tvp, tzp)); 1207 } 1208 1209 int 1210 linux_getrusage(struct thread *td, struct linux_getrusage_args *uap) 1211 { 1212 struct l_rusage s32; 1213 struct rusage s; 1214 int error; 1215 1216 error = kern_getrusage(td, uap->who, &s); 1217 if (error != 0) 1218 return (error); 1219 if (uap->rusage != NULL) { 1220 s32.ru_utime.tv_sec = s.ru_utime.tv_sec; 1221 s32.ru_utime.tv_usec = s.ru_utime.tv_usec; 1222 s32.ru_stime.tv_sec = s.ru_stime.tv_sec; 1223 s32.ru_stime.tv_usec = s.ru_stime.tv_usec; 1224 s32.ru_maxrss = s.ru_maxrss; 1225 s32.ru_ixrss = s.ru_ixrss; 1226 s32.ru_idrss = s.ru_idrss; 1227 s32.ru_isrss = s.ru_isrss; 1228 s32.ru_minflt = s.ru_minflt; 1229 s32.ru_majflt = s.ru_majflt; 1230 s32.ru_nswap = s.ru_nswap; 1231 s32.ru_inblock = s.ru_inblock; 1232 s32.ru_oublock = s.ru_oublock; 1233 s32.ru_msgsnd = s.ru_msgsnd; 1234 s32.ru_msgrcv = s.ru_msgrcv; 1235 s32.ru_nsignals = s.ru_nsignals; 1236 s32.ru_nvcsw = s.ru_nvcsw; 1237 s32.ru_nivcsw = s.ru_nivcsw; 1238 error = copyout(&s32, uap->rusage, sizeof(s32)); 1239 } 1240 return (error); 1241 } 1242 1243 int 1244 linux_sched_rr_get_interval(struct thread *td, 1245 struct linux_sched_rr_get_interval_args *uap) 1246 { 1247 struct timespec ts; 1248 struct l_timespec ts32; 1249 int error; 1250 1251 error = kern_sched_rr_get_interval(td, uap->pid, &ts); 1252 if (error != 0) 1253 return (error); 1254 ts32.tv_sec = ts.tv_sec; 1255 ts32.tv_nsec = ts.tv_nsec; 1256 return (copyout(&ts32, uap->interval, sizeof(ts32))); 1257 } 1258 1259 int 1260 linux_set_thread_area(struct thread *td, 1261 struct linux_set_thread_area_args *args) 1262 { 1263 struct l_user_desc info; 1264 struct user_segment_descriptor sd; 1265 int a[2]; 1266 int error; 1267 1268 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1269 if (error) 1270 return (error); 1271 1272 #ifdef DEBUG 1273 if (ldebug(set_thread_area)) 1274 printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, " 1275 "%i, %i, %i"), info.entry_number, info.base_addr, 1276 info.limit, info.seg_32bit, info.contents, 1277 info.read_exec_only, info.limit_in_pages, 1278 info.seg_not_present, info.useable); 1279 #endif 1280 1281 /* 1282 * Semantics of Linux version: every thread in the system has array 1283 * of three TLS descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. 1284 * This syscall loads one of the selected TLS decriptors with a value 1285 * and also loads GDT descriptors 6, 7 and 8 with the content of 1286 * the per-thread descriptors. 1287 * 1288 * Semantics of FreeBSD version: I think we can ignore that Linux has 1289 * three per-thread descriptors and use just the first one. 1290 * The tls_array[] is used only in [gs]et_thread_area() syscalls and 1291 * for loading the GDT descriptors. We use just one GDT descriptor 1292 * for TLS, so we will load just one. 1293 * 1294 * XXX: This doesn't work when a user space process tries to use more 1295 * than one TLS segment. Comment in the Linux source says wine might 1296 * do this. 1297 */ 1298 1299 /* 1300 * GLIBC reads current %gs and call set_thread_area() with it. 1301 * We should let GUDATA_SEL and GUGS32_SEL proceed as well because 1302 * we use these segments. 1303 */ 1304 switch (info.entry_number) { 1305 case GUGS32_SEL: 1306 case GUDATA_SEL: 1307 case 6: 1308 case -1: 1309 info.entry_number = GUGS32_SEL; 1310 break; 1311 default: 1312 return (EINVAL); 1313 } 1314 1315 /* 1316 * We have to copy out the GDT entry we use. 1317 * 1318 * XXX: What if a user space program does not check the return value 1319 * and tries to use 6, 7 or 8? 1320 */ 1321 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 1322 if (error) 1323 return (error); 1324 1325 if (LINUX_LDT_empty(&info)) { 1326 a[0] = 0; 1327 a[1] = 0; 1328 } else { 1329 a[0] = LINUX_LDT_entry_a(&info); 1330 a[1] = LINUX_LDT_entry_b(&info); 1331 } 1332 1333 memcpy(&sd, &a, sizeof(a)); 1334 #ifdef DEBUG 1335 if (ldebug(set_thread_area)) 1336 printf("Segment created in set_thread_area: " 1337 "lobase: %x, hibase: %x, lolimit: %x, hilimit: %x, " 1338 "type: %i, dpl: %i, p: %i, xx: %i, long: %i, " 1339 "def32: %i, gran: %i\n", 1340 sd.sd_lobase, 1341 sd.sd_hibase, 1342 sd.sd_lolimit, 1343 sd.sd_hilimit, 1344 sd.sd_type, 1345 sd.sd_dpl, 1346 sd.sd_p, 1347 sd.sd_xx, 1348 sd.sd_long, 1349 sd.sd_def32, 1350 sd.sd_gran); 1351 #endif 1352 1353 critical_enter(); 1354 td->td_pcb->pcb_gsbase = (register_t)info.base_addr; 1355 td->td_pcb->pcb_gs32sd = gdt[GUGS32_SEL] = sd; 1356 td->td_pcb->pcb_gs32p = &gdt[GUGS32_SEL]; 1357 td->td_pcb->pcb_flags |= PCB_32BIT; 1358 wrmsr(MSR_KGSBASE, td->td_pcb->pcb_gsbase); 1359 critical_exit(); 1360 1361 return (0); 1362 } 1363