linux_machdep.c revision 163371
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: head/sys/i386/linux/linux_machdep.c 163371 2006-10-15 13:22:14Z netchild $"); 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/proc.h> 43#include <sys/queue.h> 44#include <sys/resource.h> 45#include <sys/resourcevar.h> 46#include <sys/signalvar.h> 47#include <sys/syscallsubr.h> 48#include <sys/sysproto.h> 49#include <sys/unistd.h> 50#include <sys/wait.h> 51 52#include <machine/frame.h> 53#include <machine/psl.h> 54#include <machine/segments.h> 55#include <machine/sysarch.h> 56 57#include <vm/vm.h> 58#include <vm/pmap.h> 59#include <vm/vm_map.h> 60 61#include <i386/linux/linux.h> 62#include <i386/linux/linux_proto.h> 63#include <compat/linux/linux_ipc.h> 64#include <compat/linux/linux_signal.h> 65#include <compat/linux/linux_util.h> 66#include <compat/linux/linux_emul.h> 67 68#include <i386/include/pcb.h> /* needed for pcb definition in linux_set_thread_area */ 69 70#include "opt_posix.h" 71 72extern struct sysentvec elf32_freebsd_sysvec; /* defined in i386/i386/elf_machdep.c */ 73 74struct l_descriptor { 75 l_uint entry_number; 76 l_ulong base_addr; 77 l_uint limit; 78 l_uint seg_32bit:1; 79 l_uint contents:2; 80 l_uint read_exec_only:1; 81 l_uint limit_in_pages:1; 82 l_uint seg_not_present:1; 83 l_uint useable:1; 84}; 85 86struct l_old_select_argv { 87 l_int nfds; 88 l_fd_set *readfds; 89 l_fd_set *writefds; 90 l_fd_set *exceptfds; 91 struct l_timeval *timeout; 92}; 93 94int 95linux_to_bsd_sigaltstack(int lsa) 96{ 97 int bsa = 0; 98 99 if (lsa & LINUX_SS_DISABLE) 100 bsa |= SS_DISABLE; 101 if (lsa & LINUX_SS_ONSTACK) 102 bsa |= SS_ONSTACK; 103 return (bsa); 104} 105 106int 107bsd_to_linux_sigaltstack(int bsa) 108{ 109 int lsa = 0; 110 111 if (bsa & SS_DISABLE) 112 lsa |= LINUX_SS_DISABLE; 113 if (bsa & SS_ONSTACK) 114 lsa |= LINUX_SS_ONSTACK; 115 return (lsa); 116} 117 118int 119linux_execve(struct thread *td, struct linux_execve_args *args) 120{ 121 int error; 122 char *newpath; 123 struct image_args eargs; 124 125 LCONVPATHEXIST(td, args->path, &newpath); 126 127#ifdef DEBUG 128 if (ldebug(execve)) 129 printf(ARGS(execve, "%s"), newpath); 130#endif 131 132 error = exec_copyin_args(&eargs, newpath, UIO_SYSSPACE, 133 args->argp, args->envp); 134 free(newpath, M_TEMP); 135 if (error == 0) 136 error = kern_execve(td, &eargs, NULL); 137 if (error == 0) 138 /* linux process can exec fbsd one, dont attempt 139 * to create emuldata for such process using 140 * linux_proc_init, this leads to a panic on KASSERT 141 * because such process has p->p_emuldata == NULL 142 */ 143 if (td->td_proc->p_sysent == &elf_linux_sysvec) 144 error = linux_proc_init(td, 0, 0); 145 return (error); 146} 147 148struct l_ipc_kludge { 149 struct l_msgbuf *msgp; 150 l_long msgtyp; 151}; 152 153int 154linux_ipc(struct thread *td, struct linux_ipc_args *args) 155{ 156 157 switch (args->what & 0xFFFF) { 158 case LINUX_SEMOP: { 159 struct linux_semop_args a; 160 161 a.semid = args->arg1; 162 a.tsops = args->ptr; 163 a.nsops = args->arg2; 164 return (linux_semop(td, &a)); 165 } 166 case LINUX_SEMGET: { 167 struct linux_semget_args a; 168 169 a.key = args->arg1; 170 a.nsems = args->arg2; 171 a.semflg = args->arg3; 172 return (linux_semget(td, &a)); 173 } 174 case LINUX_SEMCTL: { 175 struct linux_semctl_args a; 176 int error; 177 178 a.semid = args->arg1; 179 a.semnum = args->arg2; 180 a.cmd = args->arg3; 181 error = copyin(args->ptr, &a.arg, sizeof(a.arg)); 182 if (error) 183 return (error); 184 return (linux_semctl(td, &a)); 185 } 186 case LINUX_MSGSND: { 187 struct linux_msgsnd_args a; 188 189 a.msqid = args->arg1; 190 a.msgp = args->ptr; 191 a.msgsz = args->arg2; 192 a.msgflg = args->arg3; 193 return (linux_msgsnd(td, &a)); 194 } 195 case LINUX_MSGRCV: { 196 struct linux_msgrcv_args a; 197 198 a.msqid = args->arg1; 199 a.msgsz = args->arg2; 200 a.msgflg = args->arg3; 201 if ((args->what >> 16) == 0) { 202 struct l_ipc_kludge tmp; 203 int error; 204 205 if (args->ptr == NULL) 206 return (EINVAL); 207 error = copyin(args->ptr, &tmp, sizeof(tmp)); 208 if (error) 209 return (error); 210 a.msgp = tmp.msgp; 211 a.msgtyp = tmp.msgtyp; 212 } else { 213 a.msgp = args->ptr; 214 a.msgtyp = args->arg5; 215 } 216 return (linux_msgrcv(td, &a)); 217 } 218 case LINUX_MSGGET: { 219 struct linux_msgget_args a; 220 221 a.key = args->arg1; 222 a.msgflg = args->arg2; 223 return (linux_msgget(td, &a)); 224 } 225 case LINUX_MSGCTL: { 226 struct linux_msgctl_args a; 227 228 a.msqid = args->arg1; 229 a.cmd = args->arg2; 230 a.buf = args->ptr; 231 return (linux_msgctl(td, &a)); 232 } 233 case LINUX_SHMAT: { 234 struct linux_shmat_args a; 235 236 a.shmid = args->arg1; 237 a.shmaddr = args->ptr; 238 a.shmflg = args->arg2; 239 a.raddr = (l_ulong *)args->arg3; 240 return (linux_shmat(td, &a)); 241 } 242 case LINUX_SHMDT: { 243 struct linux_shmdt_args a; 244 245 a.shmaddr = args->ptr; 246 return (linux_shmdt(td, &a)); 247 } 248 case LINUX_SHMGET: { 249 struct linux_shmget_args a; 250 251 a.key = args->arg1; 252 a.size = args->arg2; 253 a.shmflg = args->arg3; 254 return (linux_shmget(td, &a)); 255 } 256 case LINUX_SHMCTL: { 257 struct linux_shmctl_args a; 258 259 a.shmid = args->arg1; 260 a.cmd = args->arg2; 261 a.buf = args->ptr; 262 return (linux_shmctl(td, &a)); 263 } 264 default: 265 break; 266 } 267 268 return (EINVAL); 269} 270 271int 272linux_old_select(struct thread *td, struct linux_old_select_args *args) 273{ 274 struct l_old_select_argv linux_args; 275 struct linux_select_args newsel; 276 int error; 277 278#ifdef DEBUG 279 if (ldebug(old_select)) 280 printf(ARGS(old_select, "%p"), args->ptr); 281#endif 282 283 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 284 if (error) 285 return (error); 286 287 newsel.nfds = linux_args.nfds; 288 newsel.readfds = linux_args.readfds; 289 newsel.writefds = linux_args.writefds; 290 newsel.exceptfds = linux_args.exceptfds; 291 newsel.timeout = linux_args.timeout; 292 return (linux_select(td, &newsel)); 293} 294 295int 296linux_fork(struct thread *td, struct linux_fork_args *args) 297{ 298 int error; 299 300#ifdef DEBUG 301 if (ldebug(fork)) 302 printf(ARGS(fork, "")); 303#endif 304 305 if ((error = fork(td, (struct fork_args *)args)) != 0) 306 return (error); 307 308 if (td->td_retval[1] == 1) 309 td->td_retval[0] = 0; 310 error = linux_proc_init(td, td->td_retval[0], 0); 311 if (error) 312 return (error); 313 314 return (0); 315} 316 317int 318linux_vfork(struct thread *td, struct linux_vfork_args *args) 319{ 320 int error; 321 struct proc *p2; 322 323#ifdef DEBUG 324 if (ldebug(vfork)) 325 printf(ARGS(vfork, "")); 326#endif 327 328 /* exclude RFPPWAIT */ 329 if ((error = fork1(td, RFFDG | RFPROC | RFMEM, 0, &p2)) != 0) 330 return (error); 331 if (error == 0) { 332 td->td_retval[0] = p2->p_pid; 333 td->td_retval[1] = 0; 334 } 335 /* Are we the child? */ 336 if (td->td_retval[1] == 1) 337 td->td_retval[0] = 0; 338 error = linux_proc_init(td, td->td_retval[0], 0); 339 if (error) 340 return (error); 341 /* wait for the children to exit, ie. emulate vfork */ 342 PROC_LOCK(p2); 343 while (p2->p_flag & P_PPWAIT) 344 msleep(td->td_proc, &p2->p_mtx, PWAIT, "ppwait", 0); 345 PROC_UNLOCK(p2); 346 347 return (0); 348} 349 350int 351linux_clone(struct thread *td, struct linux_clone_args *args) 352{ 353 int error, ff = RFPROC | RFSTOPPED; 354 struct proc *p2; 355 struct thread *td2; 356 int exit_signal; 357 struct linux_emuldata *em; 358 359#ifdef DEBUG 360 if (ldebug(clone)) { 361 printf(ARGS(clone, "flags %x, stack %x, parent tid: %x, child tid: %x"), 362 (unsigned int)args->flags, (unsigned int)args->stack, 363 (unsigned int)args->parent_tidptr, (unsigned int)args->child_tidptr); 364 } 365#endif 366 367 exit_signal = args->flags & 0x000000ff; 368 if (!LINUX_SIG_VALID(exit_signal)) 369 return (EINVAL); 370 371 if (exit_signal <= LINUX_SIGTBLSZ) 372 exit_signal = linux_to_bsd_signal[_SIG_IDX(exit_signal)]; 373 374 if (args->flags & CLONE_VM) 375 ff |= RFMEM; 376 if (args->flags & CLONE_SIGHAND) 377 ff |= RFSIGSHARE; 378 /* 379 * XXX: in linux sharing of fs info (chroot/cwd/umask) 380 * and open files is independant. in fbsd its in one 381 * structure but in reality it doesnt make any problems 382 * because both this flags are set at once usually. 383 */ 384 if (!(args->flags & (CLONE_FILES | CLONE_FS))) 385 ff |= RFFDG; 386 387 /* 388 * Attempt to detect when linux_clone(2) is used for creating 389 * kernel threads. Unfortunately despite the existence of the 390 * CLONE_THREAD flag, version of linuxthreads package used in 391 * most popular distros as of beginning of 2005 doesn't make 392 * any use of it. Therefore, this detection relay fully on 393 * empirical observation that linuxthreads sets certain 394 * combination of flags, so that we can make more or less 395 * precise detection and notify the FreeBSD kernel that several 396 * processes are in fact part of the same threading group, so 397 * that special treatment is necessary for signal delivery 398 * between those processes and fd locking. 399 */ 400 if ((args->flags & 0xffffff00) == THREADING_FLAGS) 401 ff |= RFTHREAD; 402 403 error = fork1(td, ff, 0, &p2); 404 if (error) 405 return (error); 406 407 /* create the emuldata */ 408 error = linux_proc_init(td, p2->p_pid, args->flags); 409 /* reference it - no need to check this */ 410 em = em_find(p2, EMUL_UNLOCKED); 411 KASSERT(em != NULL, ("clone: emuldata not found.\n")); 412 /* and adjust it */ 413 if (args->flags & CLONE_PARENT_SETTID) { 414 if (args->parent_tidptr == NULL) { 415 EMUL_UNLOCK(&emul_lock); 416 return (EINVAL); 417 } 418 error = copyout(&p2->p_pid, args->parent_tidptr, sizeof(p2->p_pid)); 419 if (error) { 420 EMUL_UNLOCK(&emul_lock); 421 return (error); 422 } 423 } 424 425 if (args->flags & (CLONE_PARENT|CLONE_THREAD)) { 426 sx_xlock(&proctree_lock); 427 PROC_LOCK(p2); 428 proc_reparent(p2, td->td_proc->p_pptr); 429 PROC_UNLOCK(p2); 430 sx_xunlock(&proctree_lock); 431 } 432 433 if (args->flags & CLONE_THREAD) { 434 /* XXX: linux mangles pgrp and pptr somehow 435 * I think it might be this but I am not sure. 436 */ 437#ifdef notyet 438 PROC_LOCK(p2); 439 p2->p_pgrp = td->td_proc->p_pgrp; 440 PROC_UNLOCK(p2); 441#endif 442 exit_signal = 0; 443 } 444 445 if (args->flags & CLONE_CHILD_SETTID) 446 em->child_set_tid = args->child_tidptr; 447 else 448 em->child_set_tid = NULL; 449 450 if (args->flags & CLONE_CHILD_CLEARTID) 451 em->child_clear_tid = args->child_tidptr; 452 else 453 em->child_clear_tid = NULL; 454 455 EMUL_UNLOCK(&emul_lock); 456 457 PROC_LOCK(p2); 458 p2->p_sigparent = exit_signal; 459 PROC_UNLOCK(p2); 460 td2 = FIRST_THREAD_IN_PROC(p2); 461 /* 462 * in a case of stack = NULL we are supposed to COW calling process stack 463 * this is what normal fork() does so we just keep the tf_esp arg intact 464 */ 465 if (args->stack) 466 td2->td_frame->tf_esp = (unsigned int)args->stack; 467 468 if (args->flags & CLONE_SETTLS) { 469 struct l_user_desc info; 470 int idx; 471 int a[2]; 472 struct segment_descriptor sd; 473 474 error = copyin((void *)td->td_frame->tf_esi, &info, sizeof(struct l_user_desc)); 475 if (error) 476 return (error); 477 478 idx = info.entry_number; 479 480 /* 481 * looks like we're getting the idx we returned 482 * in the set_thread_area() syscall 483 */ 484 if (idx != 6 && idx != 3) 485 return (EINVAL); 486 487 /* this doesnt happen in practice */ 488 if (idx == 6) { 489 /* we might copy out the entry_number as 3 */ 490 info.entry_number = 3; 491 error = copyout(&info, (void *) td->td_frame->tf_esi, sizeof(struct l_user_desc)); 492 if (error) 493 return (error); 494 } 495 496 a[0] = LDT_entry_a(&info); 497 a[1] = LDT_entry_b(&info); 498 499 memcpy(&sd, &a, sizeof(a)); 500#ifdef DEBUG 501 if (ldebug(clone)) 502 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, 503 sd.sd_hibase, 504 sd.sd_lolimit, 505 sd.sd_hilimit, 506 sd.sd_type, 507 sd.sd_dpl, 508 sd.sd_p, 509 sd.sd_xx, 510 sd.sd_def32, 511 sd.sd_gran); 512#endif 513 514 /* set %gs */ 515 td2->td_pcb->pcb_gsd = sd; 516 td2->td_pcb->pcb_gs = GSEL(GUGS_SEL, SEL_UPL); 517 } 518 519#ifdef DEBUG 520 if (ldebug(clone)) 521 printf(LMSG("clone: successful rfork to %ld, stack %p sig = %d"), 522 (long)p2->p_pid, args->stack, exit_signal); 523#endif 524 525 /* 526 * Make this runnable after we are finished with it. 527 */ 528 mtx_lock_spin(&sched_lock); 529 TD_SET_CAN_RUN(td2); 530 setrunqueue(td2, SRQ_BORING); 531 mtx_unlock_spin(&sched_lock); 532 533 td->td_retval[0] = p2->p_pid; 534 td->td_retval[1] = 0; 535 return (0); 536} 537 538/* XXX move */ 539struct l_mmap_argv { 540 l_caddr_t addr; 541 l_int len; 542 l_int prot; 543 l_int flags; 544 l_int fd; 545 l_int pos; 546}; 547 548#define STACK_SIZE (2 * 1024 * 1024) 549#define GUARD_SIZE (4 * PAGE_SIZE) 550 551static int linux_mmap_common(struct thread *, struct l_mmap_argv *); 552 553int 554linux_mmap2(struct thread *td, struct linux_mmap2_args *args) 555{ 556 struct l_mmap_argv linux_args; 557 558#ifdef DEBUG 559 if (ldebug(mmap2)) 560 printf(ARGS(mmap2, "%p, %d, %d, 0x%08x, %d, %d"), 561 (void *)args->addr, args->len, args->prot, 562 args->flags, args->fd, args->pgoff); 563#endif 564 565 linux_args.addr = (l_caddr_t)args->addr; 566 linux_args.len = args->len; 567 linux_args.prot = args->prot; 568 linux_args.flags = args->flags; 569 linux_args.fd = args->fd; 570 linux_args.pos = args->pgoff * PAGE_SIZE; 571 572 return (linux_mmap_common(td, &linux_args)); 573} 574 575int 576linux_mmap(struct thread *td, struct linux_mmap_args *args) 577{ 578 int error; 579 struct l_mmap_argv linux_args; 580 581 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 582 if (error) 583 return (error); 584 585#ifdef DEBUG 586 if (ldebug(mmap)) 587 printf(ARGS(mmap, "%p, %d, %d, 0x%08x, %d, %d"), 588 (void *)linux_args.addr, linux_args.len, linux_args.prot, 589 linux_args.flags, linux_args.fd, linux_args.pos); 590#endif 591 592 return (linux_mmap_common(td, &linux_args)); 593} 594 595static int 596linux_mmap_common(struct thread *td, struct l_mmap_argv *linux_args) 597{ 598 struct proc *p = td->td_proc; 599 struct mmap_args /* { 600 caddr_t addr; 601 size_t len; 602 int prot; 603 int flags; 604 int fd; 605 long pad; 606 off_t pos; 607 } */ bsd_args; 608 int error; 609 struct file *fp; 610 611 error = 0; 612 bsd_args.flags = 0; 613 fp = NULL; 614 615 /* 616 * Linux mmap(2): 617 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE 618 */ 619 if (! ((linux_args->flags & LINUX_MAP_SHARED) ^ 620 (linux_args->flags & LINUX_MAP_PRIVATE))) 621 return (EINVAL); 622 623 if (linux_args->flags & LINUX_MAP_SHARED) 624 bsd_args.flags |= MAP_SHARED; 625 if (linux_args->flags & LINUX_MAP_PRIVATE) 626 bsd_args.flags |= MAP_PRIVATE; 627 if (linux_args->flags & LINUX_MAP_FIXED) 628 bsd_args.flags |= MAP_FIXED; 629 if (linux_args->flags & LINUX_MAP_ANON) 630 bsd_args.flags |= MAP_ANON; 631 else 632 bsd_args.flags |= MAP_NOSYNC; 633 if (linux_args->flags & LINUX_MAP_GROWSDOWN) { 634 bsd_args.flags |= MAP_STACK; 635 636 /* 637 * The linux MAP_GROWSDOWN option does not limit auto 638 * growth of the region. Linux mmap with this option 639 * takes as addr the inital BOS, and as len, the initial 640 * region size. It can then grow down from addr without 641 * limit. However, linux threads has an implicit internal 642 * limit to stack size of STACK_SIZE. Its just not 643 * enforced explicitly in linux. But, here we impose 644 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack 645 * region, since we can do this with our mmap. 646 * 647 * Our mmap with MAP_STACK takes addr as the maximum 648 * downsize limit on BOS, and as len the max size of 649 * the region. It them maps the top SGROWSIZ bytes, 650 * and autgrows the region down, up to the limit 651 * in addr. 652 * 653 * If we don't use the MAP_STACK option, the effect 654 * of this code is to allocate a stack region of a 655 * fixed size of (STACK_SIZE - GUARD_SIZE). 656 */ 657 658 /* This gives us TOS */ 659 bsd_args.addr = linux_args->addr + linux_args->len; 660 661 if (bsd_args.addr > p->p_vmspace->vm_maxsaddr) { 662 /* 663 * Some linux apps will attempt to mmap 664 * thread stacks near the top of their 665 * address space. If their TOS is greater 666 * than vm_maxsaddr, vm_map_growstack() 667 * will confuse the thread stack with the 668 * process stack and deliver a SEGV if they 669 * attempt to grow the thread stack past their 670 * current stacksize rlimit. To avoid this, 671 * adjust vm_maxsaddr upwards to reflect 672 * the current stacksize rlimit rather 673 * than the maximum possible stacksize. 674 * It would be better to adjust the 675 * mmap'ed region, but some apps do not check 676 * mmap's return value. 677 */ 678 PROC_LOCK(p); 679 p->p_vmspace->vm_maxsaddr = (char *)USRSTACK - 680 lim_cur(p, RLIMIT_STACK); 681 PROC_UNLOCK(p); 682 } 683 684 /* This gives us our maximum stack size */ 685 if (linux_args->len > STACK_SIZE - GUARD_SIZE) 686 bsd_args.len = linux_args->len; 687 else 688 bsd_args.len = STACK_SIZE - GUARD_SIZE; 689 690 /* 691 * This gives us a new BOS. If we're using VM_STACK, then 692 * mmap will just map the top SGROWSIZ bytes, and let 693 * the stack grow down to the limit at BOS. If we're 694 * not using VM_STACK we map the full stack, since we 695 * don't have a way to autogrow it. 696 */ 697 bsd_args.addr -= bsd_args.len; 698 } else { 699 bsd_args.addr = linux_args->addr; 700 bsd_args.len = linux_args->len; 701 } 702 703 bsd_args.prot = linux_args->prot; 704 if (linux_args->flags & LINUX_MAP_ANON) 705 bsd_args.fd = -1; 706 else { 707 /* 708 * Linux follows Solaris mmap(2) description: 709 * The file descriptor fildes is opened with 710 * read permission, regardless of the 711 * protection options specified. 712 * If PROT_WRITE is specified, the application 713 * must have opened the file descriptor 714 * fildes with write permission unless 715 * MAP_PRIVATE is specified in the flag 716 * argument as described below. 717 */ 718 719 if ((error = fget(td, linux_args->fd, &fp)) != 0) 720 return (error); 721 if (fp->f_type != DTYPE_VNODE) { 722 fdrop(fp, td); 723 return (EINVAL); 724 } 725 726 /* Linux mmap() just fails for O_WRONLY files */ 727 if (! (fp->f_flag & FREAD)) { 728 fdrop(fp, td); 729 return (EACCES); 730 } 731 732 bsd_args.fd = linux_args->fd; 733 fdrop(fp, td); 734 } 735 bsd_args.pos = linux_args->pos; 736 bsd_args.pad = 0; 737 738#ifdef DEBUG 739 if (ldebug(mmap)) 740 printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n", 741 __func__, 742 (void *)bsd_args.addr, bsd_args.len, bsd_args.prot, 743 bsd_args.flags, bsd_args.fd, (int)bsd_args.pos); 744#endif 745 error = mmap(td, &bsd_args); 746#ifdef DEBUG 747 if (ldebug(mmap)) 748 printf("-> %s() return: 0x%x (0x%08x)\n", 749 __func__, error, (u_int)td->td_retval[0]); 750#endif 751 return (error); 752} 753 754int 755linux_pipe(struct thread *td, struct linux_pipe_args *args) 756{ 757 int error; 758 int reg_edx; 759 760#ifdef DEBUG 761 if (ldebug(pipe)) 762 printf(ARGS(pipe, "*")); 763#endif 764 765 reg_edx = td->td_retval[1]; 766 error = pipe(td, 0); 767 if (error) { 768 td->td_retval[1] = reg_edx; 769 return (error); 770 } 771 772 error = copyout(td->td_retval, args->pipefds, 2*sizeof(int)); 773 if (error) { 774 td->td_retval[1] = reg_edx; 775 return (error); 776 } 777 778 td->td_retval[1] = reg_edx; 779 td->td_retval[0] = 0; 780 return (0); 781} 782 783int 784linux_ioperm(struct thread *td, struct linux_ioperm_args *args) 785{ 786 int error; 787 struct i386_ioperm_args iia; 788 789 iia.start = args->start; 790 iia.length = args->length; 791 iia.enable = args->enable; 792 mtx_lock(&Giant); 793 error = i386_set_ioperm(td, &iia); 794 mtx_unlock(&Giant); 795 return (error); 796} 797 798int 799linux_iopl(struct thread *td, struct linux_iopl_args *args) 800{ 801 int error; 802 803 if (args->level < 0 || args->level > 3) 804 return (EINVAL); 805 if ((error = suser(td)) != 0) 806 return (error); 807 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 808 return (error); 809 td->td_frame->tf_eflags = (td->td_frame->tf_eflags & ~PSL_IOPL) | 810 (args->level * (PSL_IOPL / 3)); 811 return (0); 812} 813 814int 815linux_modify_ldt(struct thread *td, struct linux_modify_ldt_args *uap) 816{ 817 int error; 818 struct i386_ldt_args ldt; 819 struct l_descriptor ld; 820 union descriptor desc; 821 822 if (uap->ptr == NULL) 823 return (EINVAL); 824 825 switch (uap->func) { 826 case 0x00: /* read_ldt */ 827 ldt.start = 0; 828 ldt.descs = uap->ptr; 829 ldt.num = uap->bytecount / sizeof(union descriptor); 830 mtx_lock(&Giant); 831 error = i386_get_ldt(td, &ldt); 832 td->td_retval[0] *= sizeof(union descriptor); 833 mtx_unlock(&Giant); 834 break; 835 case 0x01: /* write_ldt */ 836 case 0x11: /* write_ldt */ 837 if (uap->bytecount != sizeof(ld)) 838 return (EINVAL); 839 840 error = copyin(uap->ptr, &ld, sizeof(ld)); 841 if (error) 842 return (error); 843 844 ldt.start = ld.entry_number; 845 ldt.descs = &desc; 846 ldt.num = 1; 847 desc.sd.sd_lolimit = (ld.limit & 0x0000ffff); 848 desc.sd.sd_hilimit = (ld.limit & 0x000f0000) >> 16; 849 desc.sd.sd_lobase = (ld.base_addr & 0x00ffffff); 850 desc.sd.sd_hibase = (ld.base_addr & 0xff000000) >> 24; 851 desc.sd.sd_type = SDT_MEMRO | ((ld.read_exec_only ^ 1) << 1) | 852 (ld.contents << 2); 853 desc.sd.sd_dpl = 3; 854 desc.sd.sd_p = (ld.seg_not_present ^ 1); 855 desc.sd.sd_xx = 0; 856 desc.sd.sd_def32 = ld.seg_32bit; 857 desc.sd.sd_gran = ld.limit_in_pages; 858 mtx_lock(&Giant); 859 error = i386_set_ldt(td, &ldt, &desc); 860 mtx_unlock(&Giant); 861 break; 862 default: 863 error = EINVAL; 864 break; 865 } 866 867 if (error == EOPNOTSUPP) { 868 printf("linux: modify_ldt needs kernel option USER_LDT\n"); 869 error = ENOSYS; 870 } 871 872 return (error); 873} 874 875int 876linux_sigaction(struct thread *td, struct linux_sigaction_args *args) 877{ 878 l_osigaction_t osa; 879 l_sigaction_t act, oact; 880 int error; 881 882#ifdef DEBUG 883 if (ldebug(sigaction)) 884 printf(ARGS(sigaction, "%d, %p, %p"), 885 args->sig, (void *)args->nsa, (void *)args->osa); 886#endif 887 888 if (args->nsa != NULL) { 889 error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); 890 if (error) 891 return (error); 892 act.lsa_handler = osa.lsa_handler; 893 act.lsa_flags = osa.lsa_flags; 894 act.lsa_restorer = osa.lsa_restorer; 895 LINUX_SIGEMPTYSET(act.lsa_mask); 896 act.lsa_mask.__bits[0] = osa.lsa_mask; 897 } 898 899 error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, 900 args->osa ? &oact : NULL); 901 902 if (args->osa != NULL && !error) { 903 osa.lsa_handler = oact.lsa_handler; 904 osa.lsa_flags = oact.lsa_flags; 905 osa.lsa_restorer = oact.lsa_restorer; 906 osa.lsa_mask = oact.lsa_mask.__bits[0]; 907 error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); 908 } 909 910 return (error); 911} 912 913/* 914 * Linux has two extra args, restart and oldmask. We dont use these, 915 * but it seems that "restart" is actually a context pointer that 916 * enables the signal to happen with a different register set. 917 */ 918int 919linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) 920{ 921 sigset_t sigmask; 922 l_sigset_t mask; 923 924#ifdef DEBUG 925 if (ldebug(sigsuspend)) 926 printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask); 927#endif 928 929 LINUX_SIGEMPTYSET(mask); 930 mask.__bits[0] = args->mask; 931 linux_to_bsd_sigset(&mask, &sigmask); 932 return (kern_sigsuspend(td, sigmask)); 933} 934 935int 936linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap) 937{ 938 l_sigset_t lmask; 939 sigset_t sigmask; 940 int error; 941 942#ifdef DEBUG 943 if (ldebug(rt_sigsuspend)) 944 printf(ARGS(rt_sigsuspend, "%p, %d"), 945 (void *)uap->newset, uap->sigsetsize); 946#endif 947 948 if (uap->sigsetsize != sizeof(l_sigset_t)) 949 return (EINVAL); 950 951 error = copyin(uap->newset, &lmask, sizeof(l_sigset_t)); 952 if (error) 953 return (error); 954 955 linux_to_bsd_sigset(&lmask, &sigmask); 956 return (kern_sigsuspend(td, sigmask)); 957} 958 959int 960linux_pause(struct thread *td, struct linux_pause_args *args) 961{ 962 struct proc *p = td->td_proc; 963 sigset_t sigmask; 964 965#ifdef DEBUG 966 if (ldebug(pause)) 967 printf(ARGS(pause, "")); 968#endif 969 970 PROC_LOCK(p); 971 sigmask = td->td_sigmask; 972 PROC_UNLOCK(p); 973 return (kern_sigsuspend(td, sigmask)); 974} 975 976int 977linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap) 978{ 979 stack_t ss, oss; 980 l_stack_t lss; 981 int error; 982 983#ifdef DEBUG 984 if (ldebug(sigaltstack)) 985 printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss); 986#endif 987 988 if (uap->uss != NULL) { 989 error = copyin(uap->uss, &lss, sizeof(l_stack_t)); 990 if (error) 991 return (error); 992 993 ss.ss_sp = lss.ss_sp; 994 ss.ss_size = lss.ss_size; 995 ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags); 996 } 997 error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL, 998 (uap->uoss != NULL) ? &oss : NULL); 999 if (!error && uap->uoss != NULL) { 1000 lss.ss_sp = oss.ss_sp; 1001 lss.ss_size = oss.ss_size; 1002 lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags); 1003 error = copyout(&lss, uap->uoss, sizeof(l_stack_t)); 1004 } 1005 1006 return (error); 1007} 1008 1009int 1010linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args) 1011{ 1012 struct ftruncate_args sa; 1013 1014#ifdef DEBUG 1015 if (ldebug(ftruncate64)) 1016 printf(ARGS(ftruncate64, "%u, %jd"), args->fd, 1017 (intmax_t)args->length); 1018#endif 1019 1020 sa.fd = args->fd; 1021 sa.pad = 0; 1022 sa.length = args->length; 1023 return ftruncate(td, &sa); 1024} 1025 1026int 1027linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args) 1028{ 1029 struct l_user_desc info; 1030 int error; 1031 int idx; 1032 int a[2]; 1033 struct segment_descriptor sd; 1034 1035 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1036 if (error) 1037 return (error); 1038 1039#ifdef DEBUG 1040 if (ldebug(set_thread_area)) 1041 printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, %i, %i, %i\n"), 1042 info.entry_number, 1043 info.base_addr, 1044 info.limit, 1045 info.seg_32bit, 1046 info.contents, 1047 info.read_exec_only, 1048 info.limit_in_pages, 1049 info.seg_not_present, 1050 info.useable); 1051#endif 1052 1053 idx = info.entry_number; 1054 /* 1055 * Semantics of linux version: every thread in the system has array 1056 * of 3 tls descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. This 1057 * syscall loads one of the selected tls decriptors with a value 1058 * and also loads GDT descriptors 6, 7 and 8 with the content of the per-thread 1059 * descriptors. 1060 * 1061 * Semantics of fbsd version: I think we can ignore that linux has 3 per-thread 1062 * descriptors and use just the 1st one. The tls_array[] is used only in 1063 * set/get-thread_area() syscalls and for loading the GDT descriptors. In fbsd 1064 * we use just one GDT descriptor for TLS so we will load just one. 1065 * XXX: this doesnt work when user-space process tries to use more then 1 TLS segment 1066 * comment in the linux sources says wine might do that. 1067 */ 1068 1069 /* 1070 * we support just GLIBC TLS now 1071 * we should let 3 proceed as well because we use this segment so 1072 * if code does two subsequent calls it should succeed 1073 */ 1074 if (idx != 6 && idx != -1 && idx != 3) 1075 return (EINVAL); 1076 1077 /* 1078 * we have to copy out the GDT entry we use 1079 * FreeBSD uses GDT entry #3 for storing %gs so load that 1080 * XXX: what if userspace program doesnt check this value and tries 1081 * to use 6, 7 or 8? 1082 */ 1083 idx = info.entry_number = 3; 1084 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 1085 if (error) 1086 return (error); 1087 1088 if (LDT_empty(&info)) { 1089 a[0] = 0; 1090 a[1] = 0; 1091 } else { 1092 a[0] = LDT_entry_a(&info); 1093 a[1] = LDT_entry_b(&info); 1094 } 1095 1096 memcpy(&sd, &a, sizeof(a)); 1097#ifdef DEBUG 1098 if (ldebug(set_thread_area)) 1099 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, 1100 sd.sd_hibase, 1101 sd.sd_lolimit, 1102 sd.sd_hilimit, 1103 sd.sd_type, 1104 sd.sd_dpl, 1105 sd.sd_p, 1106 sd.sd_xx, 1107 sd.sd_def32, 1108 sd.sd_gran); 1109#endif 1110 1111 /* this is taken from i386 version of cpu_set_user_tls() */ 1112 critical_enter(); 1113 /* set %gs */ 1114 td->td_pcb->pcb_gsd = sd; 1115 PCPU_GET(fsgs_gdt)[1] = sd; 1116 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 1117 critical_exit(); 1118 1119 return (0); 1120} 1121 1122int 1123linux_get_thread_area(struct thread *td, struct linux_get_thread_area_args *args) 1124{ 1125 1126 struct l_user_desc info; 1127 int error; 1128 int idx; 1129 struct l_desc_struct desc; 1130 struct segment_descriptor sd; 1131 1132#ifdef DEBUG 1133 if (ldebug(get_thread_area)) 1134 printf(ARGS(get_thread_area, "%p"), args->desc); 1135#endif 1136 1137 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1138 if (error) 1139 return (error); 1140 1141 idx = info.entry_number; 1142 /* XXX: I am not sure if we want 3 to be allowed too. */ 1143 if (idx != 6 && idx != 3) 1144 return (EINVAL); 1145 1146 idx = 3; 1147 1148 memset(&info, 0, sizeof(info)); 1149 1150 sd = PCPU_GET(fsgs_gdt)[1]; 1151 1152 memcpy(&desc, &sd, sizeof(desc)); 1153 1154 info.entry_number = idx; 1155 info.base_addr = GET_BASE(&desc); 1156 info.limit = GET_LIMIT(&desc); 1157 info.seg_32bit = GET_32BIT(&desc); 1158 info.contents = GET_CONTENTS(&desc); 1159 info.read_exec_only = !GET_WRITABLE(&desc); 1160 info.limit_in_pages = GET_LIMIT_PAGES(&desc); 1161 info.seg_not_present = !GET_PRESENT(&desc); 1162 info.useable = GET_USEABLE(&desc); 1163 1164 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 1165 if (error) 1166 return (EFAULT); 1167 1168 return (0); 1169} 1170 1171/* copied from kern/kern_time.c */ 1172int 1173linux_timer_create(struct thread *td, struct linux_timer_create_args *args) 1174{ 1175 return ktimer_create(td, (struct ktimer_create_args *) args); 1176} 1177 1178int 1179linux_timer_settime(struct thread *td, struct linux_timer_settime_args *args) 1180{ 1181 return ktimer_settime(td, (struct ktimer_settime_args *) args); 1182} 1183 1184int 1185linux_timer_gettime(struct thread *td, struct linux_timer_gettime_args *args) 1186{ 1187 return ktimer_gettime(td, (struct ktimer_gettime_args *) args); 1188} 1189 1190int 1191linux_timer_getoverrun(struct thread *td, struct linux_timer_getoverrun_args *args) 1192{ 1193 return ktimer_getoverrun(td, (struct ktimer_getoverrun_args *) args); 1194} 1195 1196int 1197linux_timer_delete(struct thread *td, struct linux_timer_delete_args *args) 1198{ 1199 return ktimer_delete(td, (struct ktimer_delete_args *) args); 1200} 1201 1202/* XXX: this wont work with module - convert it */ 1203int 1204linux_mq_open(struct thread *td, struct linux_mq_open_args *args) 1205{ 1206#ifdef P1003_1B_MQUEUE 1207 return kmq_open(td, (struct kmq_open_args *) args); 1208#else 1209 return (ENOSYS); 1210#endif 1211} 1212 1213int 1214linux_mq_unlink(struct thread *td, struct linux_mq_unlink_args *args) 1215{ 1216#ifdef P1003_1B_MQUEUE 1217 return kmq_unlink(td, (struct kmq_unlink_args *) args); 1218#else 1219 return (ENOSYS); 1220#endif 1221} 1222 1223int 1224linux_mq_timedsend(struct thread *td, struct linux_mq_timedsend_args *args) 1225{ 1226#ifdef P1003_1B_MQUEUE 1227 return kmq_timedsend(td, (struct kmq_timedsend_args *) args); 1228#else 1229 return (ENOSYS); 1230#endif 1231} 1232 1233int 1234linux_mq_timedreceive(struct thread *td, struct linux_mq_timedreceive_args *args) 1235{ 1236#ifdef P1003_1B_MQUEUE 1237 return kmq_timedreceive(td, (struct kmq_timedreceive_args *) args); 1238#else 1239 return (ENOSYS); 1240#endif 1241} 1242 1243int 1244linux_mq_notify(struct thread *td, struct linux_mq_notify_args *args) 1245{ 1246#ifdef P1003_1B_MQUEUE 1247 return kmq_notify(td, (struct kmq_notify_args *) args); 1248#else 1249 return (ENOSYS); 1250#endif 1251} 1252 1253int 1254linux_mq_getsetattr(struct thread *td, struct linux_mq_getsetattr_args *args) 1255{ 1256#ifdef P1003_1B_MQUEUE 1257 return kmq_setattr(td, (struct kmq_setattr_args *) args); 1258#else 1259 return (ENOSYS); 1260#endif 1261} 1262 1263