1 /*- 2 * Copyright (c) 2000 Marcel Moolenaar 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer 10 * in this position and unchanged. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. The name of the author may not be used to endorse or promote products 15 * derived from this software without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/systm.h> 34 #include <sys/file.h> 35 #include <sys/fcntl.h> 36 #include <sys/imgact.h> 37 #include <sys/lock.h> 38 #include <sys/malloc.h> 39 #include <sys/mman.h> 40 #include <sys/mutex.h> 41 #include <sys/sx.h> 42 #include <sys/priv.h> 43 #include <sys/proc.h> 44 #include <sys/queue.h> 45 #include <sys/resource.h> 46 #include <sys/resourcevar.h> 47 #include <sys/signalvar.h> 48 #include <sys/syscallsubr.h> 49 #include <sys/sysproto.h> 50 #include <sys/unistd.h> 51 #include <sys/wait.h> 52 #include <sys/sched.h> 53 54 #include <machine/frame.h> 55 #include <machine/psl.h> 56 #include <machine/segments.h> 57 #include <machine/sysarch.h> 58 59 #include <vm/vm.h> 60 #include <vm/pmap.h> 61 #include <vm/vm_map.h> 62 63 #include <i386/linux/linux.h> 64 #include <i386/linux/linux_proto.h> 65 #include <compat/linux/linux_ipc.h> 66 #include <compat/linux/linux_signal.h> 67 #include <compat/linux/linux_util.h> 68 #include <compat/linux/linux_emul.h> 69 70 #include <i386/include/pcb.h> /* needed for pcb definition in linux_set_thread_area */ 71 72 #include "opt_posix.h" 73 74 extern struct sysentvec elf32_freebsd_sysvec; /* defined in i386/i386/elf_machdep.c */ 75 76 struct l_descriptor { 77 l_uint entry_number; 78 l_ulong base_addr; 79 l_uint limit; 80 l_uint seg_32bit:1; 81 l_uint contents:2; 82 l_uint read_exec_only:1; 83 l_uint limit_in_pages:1; 84 l_uint seg_not_present:1; 85 l_uint useable:1; 86 }; 87 88 struct l_old_select_argv { 89 l_int nfds; 90 l_fd_set *readfds; 91 l_fd_set *writefds; 92 l_fd_set *exceptfds; 93 struct l_timeval *timeout; 94 }; 95 96 int 97 linux_to_bsd_sigaltstack(int lsa) 98 { 99 int bsa = 0; 100 101 if (lsa & LINUX_SS_DISABLE) 102 bsa |= SS_DISABLE; 103 if (lsa & LINUX_SS_ONSTACK) 104 bsa |= SS_ONSTACK; 105 return (bsa); 106 } 107 108 int 109 bsd_to_linux_sigaltstack(int bsa) 110 { 111 int lsa = 0; 112 113 if (bsa & SS_DISABLE) 114 lsa |= LINUX_SS_DISABLE; 115 if (bsa & SS_ONSTACK) 116 lsa |= LINUX_SS_ONSTACK; 117 return (lsa); 118 } 119 120 int 121 linux_execve(struct thread *td, struct linux_execve_args *args) 122 { 123 int error; 124 char *newpath; 125 struct image_args eargs; 126 127 LCONVPATHEXIST(td, args->path, &newpath); 128 129 #ifdef DEBUG 130 if (ldebug(execve)) 131 printf(ARGS(execve, "%s"), newpath); 132 #endif 133 134 error = exec_copyin_args(&eargs, newpath, UIO_SYSSPACE, 135 args->argp, args->envp); 136 free(newpath, M_TEMP); 137 if (error == 0) 138 error = kern_execve(td, &eargs, NULL); 139 if (error == 0) 140 /* linux process can exec fbsd one, dont attempt 141 * to create emuldata for such process using 142 * linux_proc_init, this leads to a panic on KASSERT 143 * because such process has p->p_emuldata == NULL 144 */ 145 if (td->td_proc->p_sysent == &elf_linux_sysvec) 146 error = linux_proc_init(td, 0, 0); 147 return (error); 148 } 149 150 struct l_ipc_kludge { 151 struct l_msgbuf *msgp; 152 l_long msgtyp; 153 }; 154 155 int 156 linux_ipc(struct thread *td, struct linux_ipc_args *args) 157 { 158 159 switch (args->what & 0xFFFF) { 160 case LINUX_SEMOP: { 161 struct linux_semop_args a; 162 163 a.semid = args->arg1; 164 a.tsops = args->ptr; 165 a.nsops = args->arg2; 166 return (linux_semop(td, &a)); 167 } 168 case LINUX_SEMGET: { 169 struct linux_semget_args a; 170 171 a.key = args->arg1; 172 a.nsems = args->arg2; 173 a.semflg = args->arg3; 174 return (linux_semget(td, &a)); 175 } 176 case LINUX_SEMCTL: { 177 struct linux_semctl_args a; 178 int error; 179 180 a.semid = args->arg1; 181 a.semnum = args->arg2; 182 a.cmd = args->arg3; 183 error = copyin(args->ptr, &a.arg, sizeof(a.arg)); 184 if (error) 185 return (error); 186 return (linux_semctl(td, &a)); 187 } 188 case LINUX_MSGSND: { 189 struct linux_msgsnd_args a; 190 191 a.msqid = args->arg1; 192 a.msgp = args->ptr; 193 a.msgsz = args->arg2; 194 a.msgflg = args->arg3; 195 return (linux_msgsnd(td, &a)); 196 } 197 case LINUX_MSGRCV: { 198 struct linux_msgrcv_args a; 199 200 a.msqid = args->arg1; 201 a.msgsz = args->arg2; 202 a.msgflg = args->arg3; 203 if ((args->what >> 16) == 0) { 204 struct l_ipc_kludge tmp; 205 int error; 206 207 if (args->ptr == NULL) 208 return (EINVAL); 209 error = copyin(args->ptr, &tmp, sizeof(tmp)); 210 if (error) 211 return (error); 212 a.msgp = tmp.msgp; 213 a.msgtyp = tmp.msgtyp; 214 } else { 215 a.msgp = args->ptr; 216 a.msgtyp = args->arg5; 217 } 218 return (linux_msgrcv(td, &a)); 219 } 220 case LINUX_MSGGET: { 221 struct linux_msgget_args a; 222 223 a.key = args->arg1; 224 a.msgflg = args->arg2; 225 return (linux_msgget(td, &a)); 226 } 227 case LINUX_MSGCTL: { 228 struct linux_msgctl_args a; 229 230 a.msqid = args->arg1; 231 a.cmd = args->arg2; 232 a.buf = args->ptr; 233 return (linux_msgctl(td, &a)); 234 } 235 case LINUX_SHMAT: { 236 struct linux_shmat_args a; 237 238 a.shmid = args->arg1; 239 a.shmaddr = args->ptr; 240 a.shmflg = args->arg2; 241 a.raddr = (l_ulong *)args->arg3; 242 return (linux_shmat(td, &a)); 243 } 244 case LINUX_SHMDT: { 245 struct linux_shmdt_args a; 246 247 a.shmaddr = args->ptr; 248 return (linux_shmdt(td, &a)); 249 } 250 case LINUX_SHMGET: { 251 struct linux_shmget_args a; 252 253 a.key = args->arg1; 254 a.size = args->arg2; 255 a.shmflg = args->arg3; 256 return (linux_shmget(td, &a)); 257 } 258 case LINUX_SHMCTL: { 259 struct linux_shmctl_args a; 260 261 a.shmid = args->arg1; 262 a.cmd = args->arg2; 263 a.buf = args->ptr; 264 return (linux_shmctl(td, &a)); 265 } 266 default: 267 break; 268 } 269 270 return (EINVAL); 271 } 272 273 int 274 linux_old_select(struct thread *td, struct linux_old_select_args *args) 275 { 276 struct l_old_select_argv linux_args; 277 struct linux_select_args newsel; 278 int error; 279 280 #ifdef DEBUG 281 if (ldebug(old_select)) 282 printf(ARGS(old_select, "%p"), args->ptr); 283 #endif 284 285 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 286 if (error) 287 return (error); 288 289 newsel.nfds = linux_args.nfds; 290 newsel.readfds = linux_args.readfds; 291 newsel.writefds = linux_args.writefds; 292 newsel.exceptfds = linux_args.exceptfds; 293 newsel.timeout = linux_args.timeout; 294 return (linux_select(td, &newsel)); 295 } 296 297 int 298 linux_fork(struct thread *td, struct linux_fork_args *args) 299 { 300 int error; 301 struct proc *p2; 302 struct thread *td2; 303 304 #ifdef DEBUG 305 if (ldebug(fork)) 306 printf(ARGS(fork, "")); 307 #endif 308 309 if ((error = fork1(td, RFFDG | RFPROC | RFSTOPPED, 0, &p2)) != 0) 310 return (error); 311 312 if (error == 0) { 313 td->td_retval[0] = p2->p_pid; 314 td->td_retval[1] = 0; 315 } 316 317 if (td->td_retval[1] == 1) 318 td->td_retval[0] = 0; 319 error = linux_proc_init(td, td->td_retval[0], 0); 320 if (error) 321 return (error); 322 323 td2 = FIRST_THREAD_IN_PROC(p2); 324 325 /* 326 * Make this runnable after we are finished with it. 327 */ 328 mtx_lock_spin(&sched_lock); 329 TD_SET_CAN_RUN(td2); 330 sched_add(td2, SRQ_BORING); 331 mtx_unlock_spin(&sched_lock); 332 333 return (0); 334 } 335 336 int 337 linux_vfork(struct thread *td, struct linux_vfork_args *args) 338 { 339 int error; 340 struct proc *p2; 341 struct thread *td2; 342 343 #ifdef DEBUG 344 if (ldebug(vfork)) 345 printf(ARGS(vfork, "")); 346 #endif 347 348 /* exclude RFPPWAIT */ 349 if ((error = fork1(td, RFFDG | RFPROC | RFMEM | RFSTOPPED, 0, &p2)) != 0) 350 return (error); 351 if (error == 0) { 352 td->td_retval[0] = p2->p_pid; 353 td->td_retval[1] = 0; 354 } 355 /* Are we the child? */ 356 if (td->td_retval[1] == 1) 357 td->td_retval[0] = 0; 358 error = linux_proc_init(td, td->td_retval[0], 0); 359 if (error) 360 return (error); 361 362 PROC_LOCK(p2); 363 p2->p_flag |= P_PPWAIT; 364 PROC_UNLOCK(p2); 365 366 td2 = FIRST_THREAD_IN_PROC(p2); 367 368 /* 369 * Make this runnable after we are finished with it. 370 */ 371 mtx_lock_spin(&sched_lock); 372 TD_SET_CAN_RUN(td2); 373 sched_add(td2, SRQ_BORING); 374 mtx_unlock_spin(&sched_lock); 375 376 /* wait for the children to exit, ie. emulate vfork */ 377 PROC_LOCK(p2); 378 while (p2->p_flag & P_PPWAIT) 379 msleep(td->td_proc, &p2->p_mtx, PWAIT, "ppwait", 0); 380 PROC_UNLOCK(p2); 381 382 return (0); 383 } 384 385 int 386 linux_clone(struct thread *td, struct linux_clone_args *args) 387 { 388 int error, ff = RFPROC | RFSTOPPED; 389 struct proc *p2; 390 struct thread *td2; 391 int exit_signal; 392 struct linux_emuldata *em; 393 394 #ifdef DEBUG 395 if (ldebug(clone)) { 396 printf(ARGS(clone, "flags %x, stack %x, parent tid: %x, child tid: %x"), 397 (unsigned int)args->flags, (unsigned int)args->stack, 398 (unsigned int)args->parent_tidptr, (unsigned int)args->child_tidptr); 399 } 400 #endif 401 402 exit_signal = args->flags & 0x000000ff; 403 if (!LINUX_SIG_VALID(exit_signal) && exit_signal != 0) 404 return (EINVAL); 405 406 if (exit_signal <= LINUX_SIGTBLSZ) 407 exit_signal = linux_to_bsd_signal[_SIG_IDX(exit_signal)]; 408 409 if (args->flags & CLONE_VM) 410 ff |= RFMEM; 411 if (args->flags & CLONE_SIGHAND) 412 ff |= RFSIGSHARE; 413 /* 414 * XXX: in linux sharing of fs info (chroot/cwd/umask) 415 * and open files is independant. in fbsd its in one 416 * structure but in reality it doesn't cause any problems 417 * because both of these flags are usually set together. 418 */ 419 if (!(args->flags & (CLONE_FILES | CLONE_FS))) 420 ff |= RFFDG; 421 422 /* 423 * Attempt to detect when linux_clone(2) is used for creating 424 * kernel threads. Unfortunately despite the existence of the 425 * CLONE_THREAD flag, version of linuxthreads package used in 426 * most popular distros as of beginning of 2005 doesn't make 427 * any use of it. Therefore, this detection relies on 428 * empirical observation that linuxthreads sets certain 429 * combination of flags, so that we can make more or less 430 * precise detection and notify the FreeBSD kernel that several 431 * processes are in fact part of the same threading group, so 432 * that special treatment is necessary for signal delivery 433 * between those processes and fd locking. 434 */ 435 if ((args->flags & 0xffffff00) == THREADING_FLAGS) 436 ff |= RFTHREAD; 437 438 if (args->flags & CLONE_PARENT_SETTID) 439 if (args->parent_tidptr == NULL) 440 return (EINVAL); 441 442 error = fork1(td, ff, 0, &p2); 443 if (error) 444 return (error); 445 446 if (args->flags & (CLONE_PARENT|CLONE_THREAD)) { 447 sx_xlock(&proctree_lock); 448 PROC_LOCK(p2); 449 proc_reparent(p2, td->td_proc->p_pptr); 450 PROC_UNLOCK(p2); 451 sx_xunlock(&proctree_lock); 452 } 453 454 /* create the emuldata */ 455 error = linux_proc_init(td, p2->p_pid, args->flags); 456 /* reference it - no need to check this */ 457 em = em_find(p2, EMUL_DOLOCK); 458 KASSERT(em != NULL, ("clone: emuldata not found.\n")); 459 /* and adjust it */ 460 461 if (args->flags & CLONE_THREAD) { 462 /* XXX: linux mangles pgrp and pptr somehow 463 * I think it might be this but I am not sure. 464 */ 465 #ifdef notyet 466 PROC_LOCK(p2); 467 p2->p_pgrp = td->td_proc->p_pgrp; 468 PROC_UNLOCK(p2); 469 #endif 470 exit_signal = 0; 471 } 472 473 if (args->flags & CLONE_CHILD_SETTID) 474 em->child_set_tid = args->child_tidptr; 475 else 476 em->child_set_tid = NULL; 477 478 if (args->flags & CLONE_CHILD_CLEARTID) 479 em->child_clear_tid = args->child_tidptr; 480 else 481 em->child_clear_tid = NULL; 482 483 EMUL_UNLOCK(&emul_lock); 484 485 if (args->flags & CLONE_PARENT_SETTID) { 486 error = copyout(&p2->p_pid, args->parent_tidptr, sizeof(p2->p_pid)); 487 if (error) 488 printf(LMSG("copyout failed!")); 489 } 490 491 PROC_LOCK(p2); 492 p2->p_sigparent = exit_signal; 493 PROC_UNLOCK(p2); 494 td2 = FIRST_THREAD_IN_PROC(p2); 495 /* 496 * in a case of stack = NULL we are supposed to COW calling process stack 497 * this is what normal fork() does so we just keep the tf_esp arg intact 498 */ 499 if (args->stack) 500 td2->td_frame->tf_esp = (unsigned int)args->stack; 501 502 if (args->flags & CLONE_SETTLS) { 503 struct l_user_desc info; 504 int idx; 505 int a[2]; 506 struct segment_descriptor sd; 507 508 error = copyin((void *)td->td_frame->tf_esi, &info, sizeof(struct l_user_desc)); 509 if (error) { 510 printf(LMSG("copyin failed!")); 511 } else { 512 513 idx = info.entry_number; 514 515 /* 516 * looks like we're getting the idx we returned 517 * in the set_thread_area() syscall 518 */ 519 if (idx != 6 && idx != 3) { 520 printf(LMSG("resetting idx!")); 521 idx = 3; 522 } 523 524 /* this doesnt happen in practice */ 525 if (idx == 6) { 526 /* we might copy out the entry_number as 3 */ 527 info.entry_number = 3; 528 error = copyout(&info, (void *) td->td_frame->tf_esi, sizeof(struct l_user_desc)); 529 if (error) 530 printf(LMSG("copyout failed!")); 531 } 532 533 a[0] = LDT_entry_a(&info); 534 a[1] = LDT_entry_b(&info); 535 536 memcpy(&sd, &a, sizeof(a)); 537 #ifdef DEBUG 538 if (ldebug(clone)) 539 printf("Segment created in clone with CLONE_SETTLS: lobase: %x, hibase: %x, lolimit: %x, hilimit: %x, type: %i, dpl: %i, p: %i, xx: %i, def32: %i, gran: %i\n", sd.sd_lobase, 540 sd.sd_hibase, 541 sd.sd_lolimit, 542 sd.sd_hilimit, 543 sd.sd_type, 544 sd.sd_dpl, 545 sd.sd_p, 546 sd.sd_xx, 547 sd.sd_def32, 548 sd.sd_gran); 549 #endif 550 551 /* set %gs */ 552 td2->td_pcb->pcb_gsd = sd; 553 td2->td_pcb->pcb_gs = GSEL(GUGS_SEL, SEL_UPL); 554 } 555 } 556 557 #ifdef DEBUG 558 if (ldebug(clone)) 559 printf(LMSG("clone: successful rfork to %ld, stack %p sig = %d"), 560 (long)p2->p_pid, args->stack, exit_signal); 561 #endif 562 if (args->flags & CLONE_VFORK) { 563 PROC_LOCK(p2); 564 p2->p_flag |= P_PPWAIT; 565 PROC_UNLOCK(p2); 566 } 567 568 /* 569 * Make this runnable after we are finished with it. 570 */ 571 mtx_lock_spin(&sched_lock); 572 TD_SET_CAN_RUN(td2); 573 sched_add(td2, SRQ_BORING); 574 mtx_unlock_spin(&sched_lock); 575 576 td->td_retval[0] = p2->p_pid; 577 td->td_retval[1] = 0; 578 579 if (args->flags & CLONE_VFORK) { 580 /* wait for the children to exit, ie. emulate vfork */ 581 PROC_LOCK(p2); 582 while (p2->p_flag & P_PPWAIT) 583 msleep(td->td_proc, &p2->p_mtx, PWAIT, "ppwait", 0); 584 PROC_UNLOCK(p2); 585 } 586 587 return (0); 588 } 589 590 #define STACK_SIZE (2 * 1024 * 1024) 591 #define GUARD_SIZE (4 * PAGE_SIZE) 592 593 static int linux_mmap_common(struct thread *, struct l_mmap_argv *); 594 595 int 596 linux_mmap2(struct thread *td, struct linux_mmap2_args *args) 597 { 598 struct l_mmap_argv linux_args; 599 600 #ifdef DEBUG 601 if (ldebug(mmap2)) 602 printf(ARGS(mmap2, "%p, %d, %d, 0x%08x, %d, %d"), 603 (void *)args->addr, args->len, args->prot, 604 args->flags, args->fd, args->pgoff); 605 #endif 606 607 linux_args.addr = args->addr; 608 linux_args.len = args->len; 609 linux_args.prot = args->prot; 610 linux_args.flags = args->flags; 611 linux_args.fd = args->fd; 612 linux_args.pgoff = args->pgoff * PAGE_SIZE; 613 614 return (linux_mmap_common(td, &linux_args)); 615 } 616 617 int 618 linux_mmap(struct thread *td, struct linux_mmap_args *args) 619 { 620 int error; 621 struct l_mmap_argv linux_args; 622 623 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 624 if (error) 625 return (error); 626 627 #ifdef DEBUG 628 if (ldebug(mmap)) 629 printf(ARGS(mmap, "%p, %d, %d, 0x%08x, %d, %d"), 630 (void *)linux_args.addr, linux_args.len, linux_args.prot, 631 linux_args.flags, linux_args.fd, linux_args.pgoff); 632 #endif 633 634 return (linux_mmap_common(td, &linux_args)); 635 } 636 637 static int 638 linux_mmap_common(struct thread *td, struct l_mmap_argv *linux_args) 639 { 640 struct proc *p = td->td_proc; 641 struct mmap_args /* { 642 caddr_t addr; 643 size_t len; 644 int prot; 645 int flags; 646 int fd; 647 long pad; 648 off_t pos; 649 } */ bsd_args; 650 int error; 651 struct file *fp; 652 653 error = 0; 654 bsd_args.flags = 0; 655 fp = NULL; 656 657 /* 658 * Linux mmap(2): 659 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE 660 */ 661 if (! ((linux_args->flags & LINUX_MAP_SHARED) ^ 662 (linux_args->flags & LINUX_MAP_PRIVATE))) 663 return (EINVAL); 664 665 if (linux_args->flags & LINUX_MAP_SHARED) 666 bsd_args.flags |= MAP_SHARED; 667 if (linux_args->flags & LINUX_MAP_PRIVATE) 668 bsd_args.flags |= MAP_PRIVATE; 669 if (linux_args->flags & LINUX_MAP_FIXED) 670 bsd_args.flags |= MAP_FIXED; 671 if (linux_args->flags & LINUX_MAP_ANON) 672 bsd_args.flags |= MAP_ANON; 673 else 674 bsd_args.flags |= MAP_NOSYNC; 675 if (linux_args->flags & LINUX_MAP_GROWSDOWN) 676 bsd_args.flags |= MAP_STACK; 677 678 /* 679 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC 680 * on Linux/i386. We do this to ensure maximum compatibility. 681 * Linux/ia64 does the same in i386 emulation mode. 682 */ 683 bsd_args.prot = linux_args->prot; 684 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 685 bsd_args.prot |= PROT_READ | PROT_EXEC; 686 687 /* Linux does not check file descriptor when MAP_ANONYMOUS is set. */ 688 bsd_args.fd = (bsd_args.flags & MAP_ANON) ? -1 : linux_args->fd; 689 if (bsd_args.fd != -1) { 690 /* 691 * Linux follows Solaris mmap(2) description: 692 * The file descriptor fildes is opened with 693 * read permission, regardless of the 694 * protection options specified. 695 */ 696 697 if ((error = fget(td, bsd_args.fd, &fp)) != 0) 698 return (error); 699 if (fp->f_type != DTYPE_VNODE) { 700 fdrop(fp, td); 701 return (EINVAL); 702 } 703 704 /* Linux mmap() just fails for O_WRONLY files */ 705 if (!(fp->f_flag & FREAD)) { 706 fdrop(fp, td); 707 return (EACCES); 708 } 709 710 fdrop(fp, td); 711 } 712 713 if (linux_args->flags & LINUX_MAP_GROWSDOWN) { 714 /* 715 * The linux MAP_GROWSDOWN option does not limit auto 716 * growth of the region. Linux mmap with this option 717 * takes as addr the inital BOS, and as len, the initial 718 * region size. It can then grow down from addr without 719 * limit. However, linux threads has an implicit internal 720 * limit to stack size of STACK_SIZE. Its just not 721 * enforced explicitly in linux. But, here we impose 722 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack 723 * region, since we can do this with our mmap. 724 * 725 * Our mmap with MAP_STACK takes addr as the maximum 726 * downsize limit on BOS, and as len the max size of 727 * the region. It them maps the top SGROWSIZ bytes, 728 * and auto grows the region down, up to the limit 729 * in addr. 730 * 731 * If we don't use the MAP_STACK option, the effect 732 * of this code is to allocate a stack region of a 733 * fixed size of (STACK_SIZE - GUARD_SIZE). 734 */ 735 736 if ((caddr_t)PTRIN(linux_args->addr) + linux_args->len > 737 p->p_vmspace->vm_maxsaddr) { 738 /* 739 * Some linux apps will attempt to mmap 740 * thread stacks near the top of their 741 * address space. If their TOS is greater 742 * than vm_maxsaddr, vm_map_growstack() 743 * will confuse the thread stack with the 744 * process stack and deliver a SEGV if they 745 * attempt to grow the thread stack past their 746 * current stacksize rlimit. To avoid this, 747 * adjust vm_maxsaddr upwards to reflect 748 * the current stacksize rlimit rather 749 * than the maximum possible stacksize. 750 * It would be better to adjust the 751 * mmap'ed region, but some apps do not check 752 * mmap's return value. 753 */ 754 PROC_LOCK(p); 755 p->p_vmspace->vm_maxsaddr = (char *)USRSTACK - 756 lim_cur(p, RLIMIT_STACK); 757 PROC_UNLOCK(p); 758 } 759 760 /* This gives us our maximum stack size */ 761 if (linux_args->len > STACK_SIZE - GUARD_SIZE) 762 bsd_args.len = linux_args->len; 763 else 764 bsd_args.len = STACK_SIZE - GUARD_SIZE; 765 766 /* 767 * This gives us a new BOS. If we're using VM_STACK, then 768 * mmap will just map the top SGROWSIZ bytes, and let 769 * the stack grow down to the limit at BOS. If we're 770 * not using VM_STACK we map the full stack, since we 771 * don't have a way to autogrow it. 772 */ 773 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr) - 774 bsd_args.len; 775 } else { 776 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr); 777 bsd_args.len = linux_args->len; 778 } 779 bsd_args.pos = linux_args->pgoff; 780 bsd_args.pad = 0; 781 782 #ifdef DEBUG 783 if (ldebug(mmap)) 784 printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n", 785 __func__, 786 (void *)bsd_args.addr, bsd_args.len, bsd_args.prot, 787 bsd_args.flags, bsd_args.fd, (int)bsd_args.pos); 788 #endif 789 error = mmap(td, &bsd_args); 790 #ifdef DEBUG 791 if (ldebug(mmap)) 792 printf("-> %s() return: 0x%x (0x%08x)\n", 793 __func__, error, (u_int)td->td_retval[0]); 794 #endif 795 return (error); 796 } 797 798 int 799 linux_mprotect(struct thread *td, struct linux_mprotect_args *uap) 800 { 801 struct mprotect_args bsd_args; 802 803 bsd_args.addr = uap->addr; 804 bsd_args.len = uap->len; 805 bsd_args.prot = uap->prot; 806 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 807 bsd_args.prot |= PROT_READ | PROT_EXEC; 808 return (mprotect(td, &bsd_args)); 809 } 810 811 int 812 linux_pipe(struct thread *td, struct linux_pipe_args *args) 813 { 814 int error; 815 int reg_edx; 816 817 #ifdef DEBUG 818 if (ldebug(pipe)) 819 printf(ARGS(pipe, "*")); 820 #endif 821 822 reg_edx = td->td_retval[1]; 823 error = pipe(td, 0); 824 if (error) { 825 td->td_retval[1] = reg_edx; 826 return (error); 827 } 828 829 error = copyout(td->td_retval, args->pipefds, 2*sizeof(int)); 830 if (error) { 831 td->td_retval[1] = reg_edx; 832 return (error); 833 } 834 835 td->td_retval[1] = reg_edx; 836 td->td_retval[0] = 0; 837 return (0); 838 } 839 840 int 841 linux_ioperm(struct thread *td, struct linux_ioperm_args *args) 842 { 843 int error; 844 struct i386_ioperm_args iia; 845 846 iia.start = args->start; 847 iia.length = args->length; 848 iia.enable = args->enable; 849 mtx_lock(&Giant); 850 error = i386_set_ioperm(td, &iia); 851 mtx_unlock(&Giant); 852 return (error); 853 } 854 855 int 856 linux_iopl(struct thread *td, struct linux_iopl_args *args) 857 { 858 int error; 859 860 if (args->level < 0 || args->level > 3) 861 return (EINVAL); 862 if ((error = priv_check(td, PRIV_IO)) != 0) 863 return (error); 864 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 865 return (error); 866 td->td_frame->tf_eflags = (td->td_frame->tf_eflags & ~PSL_IOPL) | 867 (args->level * (PSL_IOPL / 3)); 868 return (0); 869 } 870 871 int 872 linux_modify_ldt(struct thread *td, struct linux_modify_ldt_args *uap) 873 { 874 int error; 875 struct i386_ldt_args ldt; 876 struct l_descriptor ld; 877 union descriptor desc; 878 879 if (uap->ptr == NULL) 880 return (EINVAL); 881 882 switch (uap->func) { 883 case 0x00: /* read_ldt */ 884 ldt.start = 0; 885 ldt.descs = uap->ptr; 886 ldt.num = uap->bytecount / sizeof(union descriptor); 887 mtx_lock(&Giant); 888 error = i386_get_ldt(td, &ldt); 889 td->td_retval[0] *= sizeof(union descriptor); 890 mtx_unlock(&Giant); 891 break; 892 case 0x01: /* write_ldt */ 893 case 0x11: /* write_ldt */ 894 if (uap->bytecount != sizeof(ld)) 895 return (EINVAL); 896 897 error = copyin(uap->ptr, &ld, sizeof(ld)); 898 if (error) 899 return (error); 900 901 ldt.start = ld.entry_number; 902 ldt.descs = &desc; 903 ldt.num = 1; 904 desc.sd.sd_lolimit = (ld.limit & 0x0000ffff); 905 desc.sd.sd_hilimit = (ld.limit & 0x000f0000) >> 16; 906 desc.sd.sd_lobase = (ld.base_addr & 0x00ffffff); 907 desc.sd.sd_hibase = (ld.base_addr & 0xff000000) >> 24; 908 desc.sd.sd_type = SDT_MEMRO | ((ld.read_exec_only ^ 1) << 1) | 909 (ld.contents << 2); 910 desc.sd.sd_dpl = 3; 911 desc.sd.sd_p = (ld.seg_not_present ^ 1); 912 desc.sd.sd_xx = 0; 913 desc.sd.sd_def32 = ld.seg_32bit; 914 desc.sd.sd_gran = ld.limit_in_pages; 915 mtx_lock(&Giant); 916 error = i386_set_ldt(td, &ldt, &desc); 917 mtx_unlock(&Giant); 918 break; 919 default: 920 error = EINVAL; 921 break; 922 } 923 924 if (error == EOPNOTSUPP) { 925 printf("linux: modify_ldt needs kernel option USER_LDT\n"); 926 error = ENOSYS; 927 } 928 929 return (error); 930 } 931 932 int 933 linux_sigaction(struct thread *td, struct linux_sigaction_args *args) 934 { 935 l_osigaction_t osa; 936 l_sigaction_t act, oact; 937 int error; 938 939 #ifdef DEBUG 940 if (ldebug(sigaction)) 941 printf(ARGS(sigaction, "%d, %p, %p"), 942 args->sig, (void *)args->nsa, (void *)args->osa); 943 #endif 944 945 if (args->nsa != NULL) { 946 error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); 947 if (error) 948 return (error); 949 act.lsa_handler = osa.lsa_handler; 950 act.lsa_flags = osa.lsa_flags; 951 act.lsa_restorer = osa.lsa_restorer; 952 LINUX_SIGEMPTYSET(act.lsa_mask); 953 act.lsa_mask.__bits[0] = osa.lsa_mask; 954 } 955 956 error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, 957 args->osa ? &oact : NULL); 958 959 if (args->osa != NULL && !error) { 960 osa.lsa_handler = oact.lsa_handler; 961 osa.lsa_flags = oact.lsa_flags; 962 osa.lsa_restorer = oact.lsa_restorer; 963 osa.lsa_mask = oact.lsa_mask.__bits[0]; 964 error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); 965 } 966 967 return (error); 968 } 969 970 /* 971 * Linux has two extra args, restart and oldmask. We dont use these, 972 * but it seems that "restart" is actually a context pointer that 973 * enables the signal to happen with a different register set. 974 */ 975 int 976 linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) 977 { 978 sigset_t sigmask; 979 l_sigset_t mask; 980 981 #ifdef DEBUG 982 if (ldebug(sigsuspend)) 983 printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask); 984 #endif 985 986 LINUX_SIGEMPTYSET(mask); 987 mask.__bits[0] = args->mask; 988 linux_to_bsd_sigset(&mask, &sigmask); 989 return (kern_sigsuspend(td, sigmask)); 990 } 991 992 int 993 linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap) 994 { 995 l_sigset_t lmask; 996 sigset_t sigmask; 997 int error; 998 999 #ifdef DEBUG 1000 if (ldebug(rt_sigsuspend)) 1001 printf(ARGS(rt_sigsuspend, "%p, %d"), 1002 (void *)uap->newset, uap->sigsetsize); 1003 #endif 1004 1005 if (uap->sigsetsize != sizeof(l_sigset_t)) 1006 return (EINVAL); 1007 1008 error = copyin(uap->newset, &lmask, sizeof(l_sigset_t)); 1009 if (error) 1010 return (error); 1011 1012 linux_to_bsd_sigset(&lmask, &sigmask); 1013 return (kern_sigsuspend(td, sigmask)); 1014 } 1015 1016 int 1017 linux_pause(struct thread *td, struct linux_pause_args *args) 1018 { 1019 struct proc *p = td->td_proc; 1020 sigset_t sigmask; 1021 1022 #ifdef DEBUG 1023 if (ldebug(pause)) 1024 printf(ARGS(pause, "")); 1025 #endif 1026 1027 PROC_LOCK(p); 1028 sigmask = td->td_sigmask; 1029 PROC_UNLOCK(p); 1030 return (kern_sigsuspend(td, sigmask)); 1031 } 1032 1033 int 1034 linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap) 1035 { 1036 stack_t ss, oss; 1037 l_stack_t lss; 1038 int error; 1039 1040 #ifdef DEBUG 1041 if (ldebug(sigaltstack)) 1042 printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss); 1043 #endif 1044 1045 if (uap->uss != NULL) { 1046 error = copyin(uap->uss, &lss, sizeof(l_stack_t)); 1047 if (error) 1048 return (error); 1049 1050 ss.ss_sp = lss.ss_sp; 1051 ss.ss_size = lss.ss_size; 1052 ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags); 1053 } 1054 error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL, 1055 (uap->uoss != NULL) ? &oss : NULL); 1056 if (!error && uap->uoss != NULL) { 1057 lss.ss_sp = oss.ss_sp; 1058 lss.ss_size = oss.ss_size; 1059 lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags); 1060 error = copyout(&lss, uap->uoss, sizeof(l_stack_t)); 1061 } 1062 1063 return (error); 1064 } 1065 1066 int 1067 linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args) 1068 { 1069 struct ftruncate_args sa; 1070 1071 #ifdef DEBUG 1072 if (ldebug(ftruncate64)) 1073 printf(ARGS(ftruncate64, "%u, %jd"), args->fd, 1074 (intmax_t)args->length); 1075 #endif 1076 1077 sa.fd = args->fd; 1078 sa.pad = 0; 1079 sa.length = args->length; 1080 return ftruncate(td, &sa); 1081 } 1082 1083 int 1084 linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args) 1085 { 1086 struct l_user_desc info; 1087 int error; 1088 int idx; 1089 int a[2]; 1090 struct segment_descriptor sd; 1091 1092 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1093 if (error) 1094 return (error); 1095 1096 #ifdef DEBUG 1097 if (ldebug(set_thread_area)) 1098 printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, %i, %i, %i\n"), 1099 info.entry_number, 1100 info.base_addr, 1101 info.limit, 1102 info.seg_32bit, 1103 info.contents, 1104 info.read_exec_only, 1105 info.limit_in_pages, 1106 info.seg_not_present, 1107 info.useable); 1108 #endif 1109 1110 idx = info.entry_number; 1111 /* 1112 * Semantics of linux version: every thread in the system has array of 1113 * 3 tls descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. This 1114 * syscall loads one of the selected tls decriptors with a value and 1115 * also loads GDT descriptors 6, 7 and 8 with the content of the 1116 * per-thread descriptors. 1117 * 1118 * Semantics of fbsd version: I think we can ignore that linux has 3 1119 * per-thread descriptors and use just the 1st one. The tls_array[] 1120 * is used only in set/get-thread_area() syscalls and for loading the 1121 * GDT descriptors. In fbsd we use just one GDT descriptor for TLS so 1122 * we will load just one. 1123 * 1124 * XXX: this doesn't work when a user space process tries to use more 1125 * than 1 TLS segment. Comment in the linux sources says wine might do 1126 * this. 1127 */ 1128 1129 /* 1130 * we support just GLIBC TLS now 1131 * we should let 3 proceed as well because we use this segment so 1132 * if code does two subsequent calls it should succeed 1133 */ 1134 if (idx != 6 && idx != -1 && idx != 3) 1135 return (EINVAL); 1136 1137 /* 1138 * we have to copy out the GDT entry we use 1139 * FreeBSD uses GDT entry #3 for storing %gs so load that 1140 * 1141 * XXX: what if a user space program doesn't check this value and tries 1142 * to use 6, 7 or 8? 1143 */ 1144 idx = info.entry_number = 3; 1145 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 1146 if (error) 1147 return (error); 1148 1149 if (LDT_empty(&info)) { 1150 a[0] = 0; 1151 a[1] = 0; 1152 } else { 1153 a[0] = LDT_entry_a(&info); 1154 a[1] = LDT_entry_b(&info); 1155 } 1156 1157 memcpy(&sd, &a, sizeof(a)); 1158 #ifdef DEBUG 1159 if (ldebug(set_thread_area)) 1160 printf("Segment created in set_thread_area: lobase: %x, hibase: %x, lolimit: %x, hilimit: %x, type: %i, dpl: %i, p: %i, xx: %i, def32: %i, gran: %i\n", sd.sd_lobase, 1161 sd.sd_hibase, 1162 sd.sd_lolimit, 1163 sd.sd_hilimit, 1164 sd.sd_type, 1165 sd.sd_dpl, 1166 sd.sd_p, 1167 sd.sd_xx, 1168 sd.sd_def32, 1169 sd.sd_gran); 1170 #endif 1171 1172 /* this is taken from i386 version of cpu_set_user_tls() */ 1173 critical_enter(); 1174 /* set %gs */ 1175 td->td_pcb->pcb_gsd = sd; 1176 PCPU_GET(fsgs_gdt)[1] = sd; 1177 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 1178 critical_exit(); 1179 1180 return (0); 1181 } 1182 1183 int 1184 linux_get_thread_area(struct thread *td, struct linux_get_thread_area_args *args) 1185 { 1186 1187 struct l_user_desc info; 1188 int error; 1189 int idx; 1190 struct l_desc_struct desc; 1191 struct segment_descriptor sd; 1192 1193 #ifdef DEBUG 1194 if (ldebug(get_thread_area)) 1195 printf(ARGS(get_thread_area, "%p"), args->desc); 1196 #endif 1197 1198 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1199 if (error) 1200 return (error); 1201 1202 idx = info.entry_number; 1203 /* XXX: I am not sure if we want 3 to be allowed too. */ 1204 if (idx != 6 && idx != 3) 1205 return (EINVAL); 1206 1207 idx = 3; 1208 1209 memset(&info, 0, sizeof(info)); 1210 1211 sd = PCPU_GET(fsgs_gdt)[1]; 1212 1213 memcpy(&desc, &sd, sizeof(desc)); 1214 1215 info.entry_number = idx; 1216 info.base_addr = GET_BASE(&desc); 1217 info.limit = GET_LIMIT(&desc); 1218 info.seg_32bit = GET_32BIT(&desc); 1219 info.contents = GET_CONTENTS(&desc); 1220 info.read_exec_only = !GET_WRITABLE(&desc); 1221 info.limit_in_pages = GET_LIMIT_PAGES(&desc); 1222 info.seg_not_present = !GET_PRESENT(&desc); 1223 info.useable = GET_USEABLE(&desc); 1224 1225 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 1226 if (error) 1227 return (EFAULT); 1228 1229 return (0); 1230 } 1231 1232 /* copied from kern/kern_time.c */ 1233 int 1234 linux_timer_create(struct thread *td, struct linux_timer_create_args *args) 1235 { 1236 return ktimer_create(td, (struct ktimer_create_args *) args); 1237 } 1238 1239 int 1240 linux_timer_settime(struct thread *td, struct linux_timer_settime_args *args) 1241 { 1242 return ktimer_settime(td, (struct ktimer_settime_args *) args); 1243 } 1244 1245 int 1246 linux_timer_gettime(struct thread *td, struct linux_timer_gettime_args *args) 1247 { 1248 return ktimer_gettime(td, (struct ktimer_gettime_args *) args); 1249 } 1250 1251 int 1252 linux_timer_getoverrun(struct thread *td, struct linux_timer_getoverrun_args *args) 1253 { 1254 return ktimer_getoverrun(td, (struct ktimer_getoverrun_args *) args); 1255 } 1256 1257 int 1258 linux_timer_delete(struct thread *td, struct linux_timer_delete_args *args) 1259 { 1260 return ktimer_delete(td, (struct ktimer_delete_args *) args); 1261 } 1262 1263 /* XXX: this wont work with module - convert it */ 1264 int 1265 linux_mq_open(struct thread *td, struct linux_mq_open_args *args) 1266 { 1267 #ifdef P1003_1B_MQUEUE 1268 return kmq_open(td, (struct kmq_open_args *) args); 1269 #else 1270 return (ENOSYS); 1271 #endif 1272 } 1273 1274 int 1275 linux_mq_unlink(struct thread *td, struct linux_mq_unlink_args *args) 1276 { 1277 #ifdef P1003_1B_MQUEUE 1278 return kmq_unlink(td, (struct kmq_unlink_args *) args); 1279 #else 1280 return (ENOSYS); 1281 #endif 1282 } 1283 1284 int 1285 linux_mq_timedsend(struct thread *td, struct linux_mq_timedsend_args *args) 1286 { 1287 #ifdef P1003_1B_MQUEUE 1288 return kmq_timedsend(td, (struct kmq_timedsend_args *) args); 1289 #else 1290 return (ENOSYS); 1291 #endif 1292 } 1293 1294 int 1295 linux_mq_timedreceive(struct thread *td, struct linux_mq_timedreceive_args *args) 1296 { 1297 #ifdef P1003_1B_MQUEUE 1298 return kmq_timedreceive(td, (struct kmq_timedreceive_args *) args); 1299 #else 1300 return (ENOSYS); 1301 #endif 1302 } 1303 1304 int 1305 linux_mq_notify(struct thread *td, struct linux_mq_notify_args *args) 1306 { 1307 #ifdef P1003_1B_MQUEUE 1308 return kmq_notify(td, (struct kmq_notify_args *) args); 1309 #else 1310 return (ENOSYS); 1311 #endif 1312 } 1313 1314 int 1315 linux_mq_getsetattr(struct thread *td, struct linux_mq_getsetattr_args *args) 1316 { 1317 #ifdef P1003_1B_MQUEUE 1318 return kmq_setattr(td, (struct kmq_setattr_args *) args); 1319 #else 1320 return (ENOSYS); 1321 #endif 1322 } 1323 1324