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 mtx_lock_spin(&sched_lock); 490 TD_SET_CAN_RUN(td2); 491 sched_add(td2, SRQ_BORING); 492 mtx_unlock_spin(&sched_lock); 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 /* make it run */ 530 mtx_lock_spin(&sched_lock); 531 TD_SET_CAN_RUN(td2); 532 sched_add(td2, SRQ_BORING); 533 mtx_unlock_spin(&sched_lock); 534 535 /* wait for the children to exit, ie. emulate vfork */ 536 PROC_LOCK(p2); 537 while (p2->p_flag & P_PPWAIT) 538 msleep(td->td_proc, &p2->p_mtx, PWAIT, "ppwait", 0); 539 PROC_UNLOCK(p2); 540 541 return (0); 542 } 543 544 int 545 linux_clone(struct thread *td, struct linux_clone_args *args) 546 { 547 int error, ff = RFPROC | RFSTOPPED; 548 struct proc *p2; 549 struct thread *td2; 550 int exit_signal; 551 struct linux_emuldata *em; 552 553 #ifdef DEBUG 554 if (ldebug(clone)) { 555 printf(ARGS(clone, "flags %x, stack %p, parent tid: %p, " 556 "child tid: %p"), (unsigned)args->flags, 557 args->stack, args->parent_tidptr, args->child_tidptr); 558 } 559 #endif 560 561 exit_signal = args->flags & 0x000000ff; 562 if (!LINUX_SIG_VALID(exit_signal) && exit_signal != 0) 563 return (EINVAL); 564 565 if (exit_signal <= LINUX_SIGTBLSZ) 566 exit_signal = linux_to_bsd_signal[_SIG_IDX(exit_signal)]; 567 568 if (args->flags & LINUX_CLONE_VM) 569 ff |= RFMEM; 570 if (args->flags & LINUX_CLONE_SIGHAND) 571 ff |= RFSIGSHARE; 572 /* 573 * XXX: In Linux, sharing of fs info (chroot/cwd/umask) 574 * and open files is independant. In FreeBSD, its in one 575 * structure but in reality it does not make any problems 576 * because both of these flags are set at once usually. 577 */ 578 if (!(args->flags & (LINUX_CLONE_FILES | LINUX_CLONE_FS))) 579 ff |= RFFDG; 580 581 /* 582 * Attempt to detect when linux_clone(2) is used for creating 583 * kernel threads. Unfortunately despite the existence of the 584 * CLONE_THREAD flag, version of linuxthreads package used in 585 * most popular distros as of beginning of 2005 doesn't make 586 * any use of it. Therefore, this detection relies on 587 * empirical observation that linuxthreads sets certain 588 * combination of flags, so that we can make more or less 589 * precise detection and notify the FreeBSD kernel that several 590 * processes are in fact part of the same threading group, so 591 * that special treatment is necessary for signal delivery 592 * between those processes and fd locking. 593 */ 594 if ((args->flags & 0xffffff00) == LINUX_THREADING_FLAGS) 595 ff |= RFTHREAD; 596 597 if (args->flags & LINUX_CLONE_PARENT_SETTID) 598 if (args->parent_tidptr == NULL) 599 return (EINVAL); 600 601 error = fork1(td, ff, 0, &p2); 602 if (error) 603 return (error); 604 605 if (args->flags & (LINUX_CLONE_PARENT | LINUX_CLONE_THREAD)) { 606 sx_xlock(&proctree_lock); 607 PROC_LOCK(p2); 608 proc_reparent(p2, td->td_proc->p_pptr); 609 PROC_UNLOCK(p2); 610 sx_xunlock(&proctree_lock); 611 } 612 613 /* create the emuldata */ 614 error = linux_proc_init(td, p2->p_pid, args->flags); 615 /* reference it - no need to check this */ 616 em = em_find(p2, EMUL_DOLOCK); 617 KASSERT(em != NULL, ("clone: emuldata not found.\n")); 618 /* and adjust it */ 619 620 if (args->flags & LINUX_CLONE_THREAD) { 621 #ifdef notyet 622 PROC_LOCK(p2); 623 p2->p_pgrp = td->td_proc->p_pgrp; 624 PROC_UNLOCK(p2); 625 #endif 626 exit_signal = 0; 627 } 628 629 if (args->flags & LINUX_CLONE_CHILD_SETTID) 630 em->child_set_tid = args->child_tidptr; 631 else 632 em->child_set_tid = NULL; 633 634 if (args->flags & LINUX_CLONE_CHILD_CLEARTID) 635 em->child_clear_tid = args->child_tidptr; 636 else 637 em->child_clear_tid = NULL; 638 639 EMUL_UNLOCK(&emul_lock); 640 641 if (args->flags & LINUX_CLONE_PARENT_SETTID) { 642 error = copyout(&p2->p_pid, args->parent_tidptr, 643 sizeof(p2->p_pid)); 644 if (error) 645 printf(LMSG("copyout failed!")); 646 } 647 648 PROC_LOCK(p2); 649 p2->p_sigparent = exit_signal; 650 PROC_UNLOCK(p2); 651 td2 = FIRST_THREAD_IN_PROC(p2); 652 /* 653 * In a case of stack = NULL, we are supposed to COW calling process 654 * stack. This is what normal fork() does, so we just keep tf_rsp arg 655 * intact. 656 */ 657 if (args->stack) 658 td2->td_frame->tf_rsp = PTROUT(args->stack); 659 660 if (args->flags & LINUX_CLONE_SETTLS) { 661 struct user_segment_descriptor sd; 662 struct l_user_desc info; 663 int a[2]; 664 665 error = copyin((void *)td->td_frame->tf_rsi, &info, 666 sizeof(struct l_user_desc)); 667 if (error) { 668 printf(LMSG("copyin failed!")); 669 } else { 670 /* We might copy out the entry_number as GUGS32_SEL. */ 671 info.entry_number = GUGS32_SEL; 672 error = copyout(&info, (void *)td->td_frame->tf_rsi, 673 sizeof(struct l_user_desc)); 674 if (error) 675 printf(LMSG("copyout failed!")); 676 677 a[0] = LINUX_LDT_entry_a(&info); 678 a[1] = LINUX_LDT_entry_b(&info); 679 680 memcpy(&sd, &a, sizeof(a)); 681 #ifdef DEBUG 682 if (ldebug(clone)) 683 printf("Segment created in clone with " 684 "CLONE_SETTLS: lobase: %x, hibase: %x, " 685 "lolimit: %x, hilimit: %x, type: %i, " 686 "dpl: %i, p: %i, xx: %i, long: %i, " 687 "def32: %i, gran: %i\n", sd.sd_lobase, 688 sd.sd_hibase, sd.sd_lolimit, sd.sd_hilimit, 689 sd.sd_type, sd.sd_dpl, sd.sd_p, sd.sd_xx, 690 sd.sd_long, sd.sd_def32, sd.sd_gran); 691 #endif 692 td2->td_pcb->pcb_gsbase = (register_t)info.base_addr; 693 td2->td_pcb->pcb_gs32sd = sd; 694 td2->td_pcb->pcb_gs32p = &gdt[GUGS32_SEL]; 695 td2->td_pcb->pcb_gs = GSEL(GUGS32_SEL, SEL_UPL); 696 td2->td_pcb->pcb_flags |= PCB_32BIT; 697 } 698 } 699 700 #ifdef DEBUG 701 if (ldebug(clone)) 702 printf(LMSG("clone: successful rfork to %d, " 703 "stack %p sig = %d"), (int)p2->p_pid, args->stack, 704 exit_signal); 705 #endif 706 if (args->flags & LINUX_CLONE_VFORK) { 707 PROC_LOCK(p2); 708 p2->p_flag |= P_PPWAIT; 709 PROC_UNLOCK(p2); 710 } 711 712 /* 713 * Make this runnable after we are finished with it. 714 */ 715 mtx_lock_spin(&sched_lock); 716 TD_SET_CAN_RUN(td2); 717 sched_add(td2, SRQ_BORING); 718 mtx_unlock_spin(&sched_lock); 719 720 td->td_retval[0] = p2->p_pid; 721 td->td_retval[1] = 0; 722 723 if (args->flags & LINUX_CLONE_VFORK) { 724 /* wait for the children to exit, ie. emulate vfork */ 725 PROC_LOCK(p2); 726 while (p2->p_flag & P_PPWAIT) 727 msleep(td->td_proc, &p2->p_mtx, PWAIT, "ppwait", 0); 728 PROC_UNLOCK(p2); 729 } 730 731 return (0); 732 } 733 734 #define STACK_SIZE (2 * 1024 * 1024) 735 #define GUARD_SIZE (4 * PAGE_SIZE) 736 737 static int linux_mmap_common(struct thread *, struct l_mmap_argv *); 738 739 int 740 linux_mmap2(struct thread *td, struct linux_mmap2_args *args) 741 { 742 struct l_mmap_argv linux_args; 743 744 #ifdef DEBUG 745 if (ldebug(mmap2)) 746 printf(ARGS(mmap2, "0x%08x, %d, %d, 0x%08x, %d, %d"), 747 args->addr, args->len, args->prot, 748 args->flags, args->fd, args->pgoff); 749 #endif 750 751 linux_args.addr = PTROUT(args->addr); 752 linux_args.len = args->len; 753 linux_args.prot = args->prot; 754 linux_args.flags = args->flags; 755 linux_args.fd = args->fd; 756 linux_args.pgoff = args->pgoff; 757 758 return (linux_mmap_common(td, &linux_args)); 759 } 760 761 int 762 linux_mmap(struct thread *td, struct linux_mmap_args *args) 763 { 764 int error; 765 struct l_mmap_argv linux_args; 766 767 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 768 if (error) 769 return (error); 770 771 #ifdef DEBUG 772 if (ldebug(mmap)) 773 printf(ARGS(mmap, "0x%08x, %d, %d, 0x%08x, %d, %d"), 774 linux_args.addr, linux_args.len, linux_args.prot, 775 linux_args.flags, linux_args.fd, linux_args.pgoff); 776 #endif 777 if ((linux_args.pgoff % PAGE_SIZE) != 0) 778 return (EINVAL); 779 linux_args.pgoff /= PAGE_SIZE; 780 781 return (linux_mmap_common(td, &linux_args)); 782 } 783 784 static int 785 linux_mmap_common(struct thread *td, struct l_mmap_argv *linux_args) 786 { 787 struct proc *p = td->td_proc; 788 struct mmap_args /* { 789 caddr_t addr; 790 size_t len; 791 int prot; 792 int flags; 793 int fd; 794 long pad; 795 off_t pos; 796 } */ bsd_args; 797 int error; 798 struct file *fp; 799 800 error = 0; 801 bsd_args.flags = 0; 802 fp = NULL; 803 804 /* 805 * Linux mmap(2): 806 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE 807 */ 808 if (! ((linux_args->flags & LINUX_MAP_SHARED) ^ 809 (linux_args->flags & LINUX_MAP_PRIVATE))) 810 return (EINVAL); 811 812 if (linux_args->flags & LINUX_MAP_SHARED) 813 bsd_args.flags |= MAP_SHARED; 814 if (linux_args->flags & LINUX_MAP_PRIVATE) 815 bsd_args.flags |= MAP_PRIVATE; 816 if (linux_args->flags & LINUX_MAP_FIXED) 817 bsd_args.flags |= MAP_FIXED; 818 if (linux_args->flags & LINUX_MAP_ANON) 819 bsd_args.flags |= MAP_ANON; 820 else 821 bsd_args.flags |= MAP_NOSYNC; 822 if (linux_args->flags & LINUX_MAP_GROWSDOWN) 823 bsd_args.flags |= MAP_STACK; 824 825 /* 826 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC 827 * on Linux/i386. We do this to ensure maximum compatibility. 828 * Linux/ia64 does the same in i386 emulation mode. 829 */ 830 bsd_args.prot = linux_args->prot; 831 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 832 bsd_args.prot |= PROT_READ | PROT_EXEC; 833 834 /* Linux does not check file descriptor when MAP_ANONYMOUS is set. */ 835 bsd_args.fd = (bsd_args.flags & MAP_ANON) ? -1 : linux_args->fd; 836 if (bsd_args.fd != -1) { 837 /* 838 * Linux follows Solaris mmap(2) description: 839 * The file descriptor fildes is opened with 840 * read permission, regardless of the 841 * protection options specified. 842 */ 843 844 if ((error = fget(td, bsd_args.fd, &fp)) != 0) 845 return (error); 846 if (fp->f_type != DTYPE_VNODE) { 847 fdrop(fp, td); 848 return (EINVAL); 849 } 850 851 /* Linux mmap() just fails for O_WRONLY files */ 852 if (!(fp->f_flag & FREAD)) { 853 fdrop(fp, td); 854 return (EACCES); 855 } 856 857 fdrop(fp, td); 858 } 859 860 if (linux_args->flags & LINUX_MAP_GROWSDOWN) { 861 /* 862 * The Linux MAP_GROWSDOWN option does not limit auto 863 * growth of the region. Linux mmap with this option 864 * takes as addr the inital BOS, and as len, the initial 865 * region size. It can then grow down from addr without 866 * limit. However, Linux threads has an implicit internal 867 * limit to stack size of STACK_SIZE. Its just not 868 * enforced explicitly in Linux. But, here we impose 869 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack 870 * region, since we can do this with our mmap. 871 * 872 * Our mmap with MAP_STACK takes addr as the maximum 873 * downsize limit on BOS, and as len the max size of 874 * the region. It them maps the top SGROWSIZ bytes, 875 * and auto grows the region down, up to the limit 876 * in addr. 877 * 878 * If we don't use the MAP_STACK option, the effect 879 * of this code is to allocate a stack region of a 880 * fixed size of (STACK_SIZE - GUARD_SIZE). 881 */ 882 883 if ((caddr_t)PTRIN(linux_args->addr) + linux_args->len > 884 p->p_vmspace->vm_maxsaddr) { 885 /* 886 * Some Linux apps will attempt to mmap 887 * thread stacks near the top of their 888 * address space. If their TOS is greater 889 * than vm_maxsaddr, vm_map_growstack() 890 * will confuse the thread stack with the 891 * process stack and deliver a SEGV if they 892 * attempt to grow the thread stack past their 893 * current stacksize rlimit. To avoid this, 894 * adjust vm_maxsaddr upwards to reflect 895 * the current stacksize rlimit rather 896 * than the maximum possible stacksize. 897 * It would be better to adjust the 898 * mmap'ed region, but some apps do not check 899 * mmap's return value. 900 */ 901 PROC_LOCK(p); 902 p->p_vmspace->vm_maxsaddr = (char *)LINUX32_USRSTACK - 903 lim_cur(p, RLIMIT_STACK); 904 PROC_UNLOCK(p); 905 } 906 907 /* This gives us our maximum stack size */ 908 if (linux_args->len > STACK_SIZE - GUARD_SIZE) 909 bsd_args.len = linux_args->len; 910 else 911 bsd_args.len = STACK_SIZE - GUARD_SIZE; 912 913 /* 914 * This gives us a new BOS. If we're using VM_STACK, then 915 * mmap will just map the top SGROWSIZ bytes, and let 916 * the stack grow down to the limit at BOS. If we're 917 * not using VM_STACK we map the full stack, since we 918 * don't have a way to autogrow it. 919 */ 920 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr) - 921 bsd_args.len; 922 } else { 923 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr); 924 bsd_args.len = linux_args->len; 925 } 926 bsd_args.pos = (off_t)linux_args->pgoff * PAGE_SIZE; 927 bsd_args.pad = 0; 928 929 #ifdef DEBUG 930 if (ldebug(mmap)) 931 printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n", 932 __func__, 933 (void *)bsd_args.addr, (int)bsd_args.len, bsd_args.prot, 934 bsd_args.flags, bsd_args.fd, (int)bsd_args.pos); 935 #endif 936 error = mmap(td, &bsd_args); 937 #ifdef DEBUG 938 if (ldebug(mmap)) 939 printf("-> %s() return: 0x%x (0x%08x)\n", 940 __func__, error, (u_int)td->td_retval[0]); 941 #endif 942 return (error); 943 } 944 945 int 946 linux_mprotect(struct thread *td, struct linux_mprotect_args *uap) 947 { 948 struct mprotect_args bsd_args; 949 950 bsd_args.addr = uap->addr; 951 bsd_args.len = uap->len; 952 bsd_args.prot = uap->prot; 953 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 954 bsd_args.prot |= PROT_READ | PROT_EXEC; 955 return (mprotect(td, &bsd_args)); 956 } 957 958 int 959 linux_iopl(struct thread *td, struct linux_iopl_args *args) 960 { 961 int error; 962 963 if (args->level < 0 || args->level > 3) 964 return (EINVAL); 965 if ((error = priv_check(td, PRIV_IO)) != 0) 966 return (error); 967 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 968 return (error); 969 td->td_frame->tf_rflags = (td->td_frame->tf_rflags & ~PSL_IOPL) | 970 (args->level * (PSL_IOPL / 3)); 971 972 return (0); 973 } 974 975 int 976 linux_pipe(struct thread *td, struct linux_pipe_args *args) 977 { 978 int pip[2]; 979 int error; 980 register_t reg_rdx; 981 982 #ifdef DEBUG 983 if (ldebug(pipe)) 984 printf(ARGS(pipe, "*")); 985 #endif 986 987 reg_rdx = td->td_retval[1]; 988 error = pipe(td, 0); 989 if (error) { 990 td->td_retval[1] = reg_rdx; 991 return (error); 992 } 993 994 pip[0] = td->td_retval[0]; 995 pip[1] = td->td_retval[1]; 996 error = copyout(pip, args->pipefds, 2 * sizeof(int)); 997 if (error) { 998 td->td_retval[1] = reg_rdx; 999 return (error); 1000 } 1001 1002 td->td_retval[1] = reg_rdx; 1003 td->td_retval[0] = 0; 1004 return (0); 1005 } 1006 1007 int 1008 linux_sigaction(struct thread *td, struct linux_sigaction_args *args) 1009 { 1010 l_osigaction_t osa; 1011 l_sigaction_t act, oact; 1012 int error; 1013 1014 #ifdef DEBUG 1015 if (ldebug(sigaction)) 1016 printf(ARGS(sigaction, "%d, %p, %p"), 1017 args->sig, (void *)args->nsa, (void *)args->osa); 1018 #endif 1019 1020 if (args->nsa != NULL) { 1021 error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); 1022 if (error) 1023 return (error); 1024 act.lsa_handler = osa.lsa_handler; 1025 act.lsa_flags = osa.lsa_flags; 1026 act.lsa_restorer = osa.lsa_restorer; 1027 LINUX_SIGEMPTYSET(act.lsa_mask); 1028 act.lsa_mask.__bits[0] = osa.lsa_mask; 1029 } 1030 1031 error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, 1032 args->osa ? &oact : NULL); 1033 1034 if (args->osa != NULL && !error) { 1035 osa.lsa_handler = oact.lsa_handler; 1036 osa.lsa_flags = oact.lsa_flags; 1037 osa.lsa_restorer = oact.lsa_restorer; 1038 osa.lsa_mask = oact.lsa_mask.__bits[0]; 1039 error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); 1040 } 1041 1042 return (error); 1043 } 1044 1045 /* 1046 * Linux has two extra args, restart and oldmask. We don't use these, 1047 * but it seems that "restart" is actually a context pointer that 1048 * enables the signal to happen with a different register set. 1049 */ 1050 int 1051 linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) 1052 { 1053 sigset_t sigmask; 1054 l_sigset_t mask; 1055 1056 #ifdef DEBUG 1057 if (ldebug(sigsuspend)) 1058 printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask); 1059 #endif 1060 1061 LINUX_SIGEMPTYSET(mask); 1062 mask.__bits[0] = args->mask; 1063 linux_to_bsd_sigset(&mask, &sigmask); 1064 return (kern_sigsuspend(td, sigmask)); 1065 } 1066 1067 int 1068 linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap) 1069 { 1070 l_sigset_t lmask; 1071 sigset_t sigmask; 1072 int error; 1073 1074 #ifdef DEBUG 1075 if (ldebug(rt_sigsuspend)) 1076 printf(ARGS(rt_sigsuspend, "%p, %d"), 1077 (void *)uap->newset, uap->sigsetsize); 1078 #endif 1079 1080 if (uap->sigsetsize != sizeof(l_sigset_t)) 1081 return (EINVAL); 1082 1083 error = copyin(uap->newset, &lmask, sizeof(l_sigset_t)); 1084 if (error) 1085 return (error); 1086 1087 linux_to_bsd_sigset(&lmask, &sigmask); 1088 return (kern_sigsuspend(td, sigmask)); 1089 } 1090 1091 int 1092 linux_pause(struct thread *td, struct linux_pause_args *args) 1093 { 1094 struct proc *p = td->td_proc; 1095 sigset_t sigmask; 1096 1097 #ifdef DEBUG 1098 if (ldebug(pause)) 1099 printf(ARGS(pause, "")); 1100 #endif 1101 1102 PROC_LOCK(p); 1103 sigmask = td->td_sigmask; 1104 PROC_UNLOCK(p); 1105 return (kern_sigsuspend(td, sigmask)); 1106 } 1107 1108 int 1109 linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap) 1110 { 1111 stack_t ss, oss; 1112 l_stack_t lss; 1113 int error; 1114 1115 #ifdef DEBUG 1116 if (ldebug(sigaltstack)) 1117 printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss); 1118 #endif 1119 1120 if (uap->uss != NULL) { 1121 error = copyin(uap->uss, &lss, sizeof(l_stack_t)); 1122 if (error) 1123 return (error); 1124 1125 ss.ss_sp = PTRIN(lss.ss_sp); 1126 ss.ss_size = lss.ss_size; 1127 ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags); 1128 } 1129 error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL, 1130 (uap->uoss != NULL) ? &oss : NULL); 1131 if (!error && uap->uoss != NULL) { 1132 lss.ss_sp = PTROUT(oss.ss_sp); 1133 lss.ss_size = oss.ss_size; 1134 lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags); 1135 error = copyout(&lss, uap->uoss, sizeof(l_stack_t)); 1136 } 1137 1138 return (error); 1139 } 1140 1141 int 1142 linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args) 1143 { 1144 struct ftruncate_args sa; 1145 1146 #ifdef DEBUG 1147 if (ldebug(ftruncate64)) 1148 printf(ARGS(ftruncate64, "%u, %jd"), args->fd, 1149 (intmax_t)args->length); 1150 #endif 1151 1152 sa.fd = args->fd; 1153 sa.pad = 0; 1154 sa.length = args->length; 1155 return ftruncate(td, &sa); 1156 } 1157 1158 int 1159 linux_gettimeofday(struct thread *td, struct linux_gettimeofday_args *uap) 1160 { 1161 struct timeval atv; 1162 l_timeval atv32; 1163 struct timezone rtz; 1164 int error = 0; 1165 1166 if (uap->tp) { 1167 microtime(&atv); 1168 atv32.tv_sec = atv.tv_sec; 1169 atv32.tv_usec = atv.tv_usec; 1170 error = copyout(&atv32, uap->tp, sizeof (atv32)); 1171 } 1172 if (error == 0 && uap->tzp != NULL) { 1173 rtz.tz_minuteswest = tz_minuteswest; 1174 rtz.tz_dsttime = tz_dsttime; 1175 error = copyout(&rtz, uap->tzp, sizeof (rtz)); 1176 } 1177 return (error); 1178 } 1179 1180 int 1181 linux_getrusage(struct thread *td, struct linux_getrusage_args *uap) 1182 { 1183 struct l_rusage s32; 1184 struct rusage s; 1185 int error; 1186 1187 error = kern_getrusage(td, uap->who, &s); 1188 if (error != 0) 1189 return (error); 1190 if (uap->rusage != NULL) { 1191 s32.ru_utime.tv_sec = s.ru_utime.tv_sec; 1192 s32.ru_utime.tv_usec = s.ru_utime.tv_usec; 1193 s32.ru_stime.tv_sec = s.ru_stime.tv_sec; 1194 s32.ru_stime.tv_usec = s.ru_stime.tv_usec; 1195 s32.ru_maxrss = s.ru_maxrss; 1196 s32.ru_ixrss = s.ru_ixrss; 1197 s32.ru_idrss = s.ru_idrss; 1198 s32.ru_isrss = s.ru_isrss; 1199 s32.ru_minflt = s.ru_minflt; 1200 s32.ru_majflt = s.ru_majflt; 1201 s32.ru_nswap = s.ru_nswap; 1202 s32.ru_inblock = s.ru_inblock; 1203 s32.ru_oublock = s.ru_oublock; 1204 s32.ru_msgsnd = s.ru_msgsnd; 1205 s32.ru_msgrcv = s.ru_msgrcv; 1206 s32.ru_nsignals = s.ru_nsignals; 1207 s32.ru_nvcsw = s.ru_nvcsw; 1208 s32.ru_nivcsw = s.ru_nivcsw; 1209 error = copyout(&s32, uap->rusage, sizeof(s32)); 1210 } 1211 return (error); 1212 } 1213 1214 int 1215 linux_sched_rr_get_interval(struct thread *td, 1216 struct linux_sched_rr_get_interval_args *uap) 1217 { 1218 struct timespec ts; 1219 struct l_timespec ts32; 1220 int error; 1221 1222 error = kern_sched_rr_get_interval(td, uap->pid, &ts); 1223 if (error != 0) 1224 return (error); 1225 ts32.tv_sec = ts.tv_sec; 1226 ts32.tv_nsec = ts.tv_nsec; 1227 return (copyout(&ts32, uap->interval, sizeof(ts32))); 1228 } 1229 1230 int 1231 linux_set_thread_area(struct thread *td, 1232 struct linux_set_thread_area_args *args) 1233 { 1234 struct l_user_desc info; 1235 struct user_segment_descriptor sd; 1236 int a[2]; 1237 int error; 1238 1239 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1240 if (error) 1241 return (error); 1242 1243 #ifdef DEBUG 1244 if (ldebug(set_thread_area)) 1245 printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, " 1246 "%i, %i, %i"), info.entry_number, info.base_addr, 1247 info.limit, info.seg_32bit, info.contents, 1248 info.read_exec_only, info.limit_in_pages, 1249 info.seg_not_present, info.useable); 1250 #endif 1251 1252 /* 1253 * Semantics of Linux version: every thread in the system has array 1254 * of three TLS descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. 1255 * This syscall loads one of the selected TLS decriptors with a value 1256 * and also loads GDT descriptors 6, 7 and 8 with the content of 1257 * the per-thread descriptors. 1258 * 1259 * Semantics of FreeBSD version: I think we can ignore that Linux has 1260 * three per-thread descriptors and use just the first one. 1261 * The tls_array[] is used only in [gs]et_thread_area() syscalls and 1262 * for loading the GDT descriptors. We use just one GDT descriptor 1263 * for TLS, so we will load just one. 1264 * XXX: This doesnt work when user-space process tries to use more 1265 * than one TLS segment. Comment in the Linux source says wine might 1266 * do that. 1267 */ 1268 1269 /* 1270 * GLIBC reads current %gs and call set_thread_area() with it. 1271 * We should let GUDATA_SEL and GUGS32_SEL proceed as well because 1272 * we use these segments. 1273 */ 1274 switch (info.entry_number) { 1275 case GUGS32_SEL: 1276 case GUDATA_SEL: 1277 case 6: 1278 case -1: 1279 info.entry_number = GUGS32_SEL; 1280 break; 1281 default: 1282 return (EINVAL); 1283 } 1284 1285 /* 1286 * We have to copy out the GDT entry we use. 1287 * XXX: What if userspace program does not check return value and 1288 * tries to use 6, 7 or 8? 1289 */ 1290 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 1291 if (error) 1292 return (error); 1293 1294 if (LINUX_LDT_empty(&info)) { 1295 a[0] = 0; 1296 a[1] = 0; 1297 } else { 1298 a[0] = LINUX_LDT_entry_a(&info); 1299 a[1] = LINUX_LDT_entry_b(&info); 1300 } 1301 1302 memcpy(&sd, &a, sizeof(a)); 1303 #ifdef DEBUG 1304 if (ldebug(set_thread_area)) 1305 printf("Segment created in set_thread_area: " 1306 "lobase: %x, hibase: %x, lolimit: %x, hilimit: %x, " 1307 "type: %i, dpl: %i, p: %i, xx: %i, long: %i, " 1308 "def32: %i, gran: %i\n", 1309 sd.sd_lobase, 1310 sd.sd_hibase, 1311 sd.sd_lolimit, 1312 sd.sd_hilimit, 1313 sd.sd_type, 1314 sd.sd_dpl, 1315 sd.sd_p, 1316 sd.sd_xx, 1317 sd.sd_long, 1318 sd.sd_def32, 1319 sd.sd_gran); 1320 #endif 1321 1322 critical_enter(); 1323 td->td_pcb->pcb_gsbase = (register_t)info.base_addr; 1324 td->td_pcb->pcb_gs32sd = gdt[GUGS32_SEL] = sd; 1325 td->td_pcb->pcb_gs32p = &gdt[GUGS32_SEL]; 1326 td->td_pcb->pcb_flags |= PCB_32BIT; 1327 wrmsr(MSR_KGSBASE, td->td_pcb->pcb_gsbase); 1328 critical_exit(); 1329 1330 return (0); 1331 } 1332