linux_machdep.c revision 218100
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 218100 2011-01-30 18:17:38Z dchagin $"); 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_misc.h> 67#include <compat/linux/linux_signal.h> 68#include <compat/linux/linux_util.h> 69#include <compat/linux/linux_emul.h> 70 71#include <i386/include/pcb.h> /* needed for pcb definition in linux_set_thread_area */ 72 73#include "opt_posix.h" 74 75extern struct sysentvec elf32_freebsd_sysvec; /* defined in i386/i386/elf_machdep.c */ 76 77struct l_descriptor { 78 l_uint entry_number; 79 l_ulong base_addr; 80 l_uint limit; 81 l_uint seg_32bit:1; 82 l_uint contents:2; 83 l_uint read_exec_only:1; 84 l_uint limit_in_pages:1; 85 l_uint seg_not_present:1; 86 l_uint useable:1; 87}; 88 89struct l_old_select_argv { 90 l_int nfds; 91 l_fd_set *readfds; 92 l_fd_set *writefds; 93 l_fd_set *exceptfds; 94 struct l_timeval *timeout; 95}; 96 97static int linux_mmap_common(struct thread *td, l_uintptr_t addr, 98 l_size_t len, l_int prot, l_int flags, l_int fd, 99 l_loff_t pos); 100 101int 102linux_to_bsd_sigaltstack(int lsa) 103{ 104 int bsa = 0; 105 106 if (lsa & LINUX_SS_DISABLE) 107 bsa |= SS_DISABLE; 108 if (lsa & LINUX_SS_ONSTACK) 109 bsa |= SS_ONSTACK; 110 return (bsa); 111} 112 113int 114bsd_to_linux_sigaltstack(int bsa) 115{ 116 int lsa = 0; 117 118 if (bsa & SS_DISABLE) 119 lsa |= LINUX_SS_DISABLE; 120 if (bsa & SS_ONSTACK) 121 lsa |= LINUX_SS_ONSTACK; 122 return (lsa); 123} 124 125int 126linux_execve(struct thread *td, struct linux_execve_args *args) 127{ 128 int error; 129 char *newpath; 130 struct image_args eargs; 131 132 LCONVPATHEXIST(td, args->path, &newpath); 133 134#ifdef DEBUG 135 if (ldebug(execve)) 136 printf(ARGS(execve, "%s"), newpath); 137#endif 138 139 error = exec_copyin_args(&eargs, newpath, UIO_SYSSPACE, 140 args->argp, args->envp); 141 free(newpath, M_TEMP); 142 if (error == 0) 143 error = kern_execve(td, &eargs, NULL); 144 if (error == 0) 145 /* linux process can exec fbsd one, dont attempt 146 * to create emuldata for such process using 147 * linux_proc_init, this leads to a panic on KASSERT 148 * because such process has p->p_emuldata == NULL 149 */ 150 if (SV_PROC_ABI(td->td_proc) == SV_ABI_LINUX) 151 error = linux_proc_init(td, 0, 0); 152 return (error); 153} 154 155struct l_ipc_kludge { 156 struct l_msgbuf *msgp; 157 l_long msgtyp; 158}; 159 160int 161linux_ipc(struct thread *td, struct linux_ipc_args *args) 162{ 163 164 switch (args->what & 0xFFFF) { 165 case LINUX_SEMOP: { 166 struct linux_semop_args a; 167 168 a.semid = args->arg1; 169 a.tsops = args->ptr; 170 a.nsops = args->arg2; 171 return (linux_semop(td, &a)); 172 } 173 case LINUX_SEMGET: { 174 struct linux_semget_args a; 175 176 a.key = args->arg1; 177 a.nsems = args->arg2; 178 a.semflg = args->arg3; 179 return (linux_semget(td, &a)); 180 } 181 case LINUX_SEMCTL: { 182 struct linux_semctl_args a; 183 int error; 184 185 a.semid = args->arg1; 186 a.semnum = args->arg2; 187 a.cmd = args->arg3; 188 error = copyin(args->ptr, &a.arg, sizeof(a.arg)); 189 if (error) 190 return (error); 191 return (linux_semctl(td, &a)); 192 } 193 case LINUX_MSGSND: { 194 struct linux_msgsnd_args a; 195 196 a.msqid = args->arg1; 197 a.msgp = args->ptr; 198 a.msgsz = args->arg2; 199 a.msgflg = args->arg3; 200 return (linux_msgsnd(td, &a)); 201 } 202 case LINUX_MSGRCV: { 203 struct linux_msgrcv_args a; 204 205 a.msqid = args->arg1; 206 a.msgsz = args->arg2; 207 a.msgflg = args->arg3; 208 if ((args->what >> 16) == 0) { 209 struct l_ipc_kludge tmp; 210 int error; 211 212 if (args->ptr == NULL) 213 return (EINVAL); 214 error = copyin(args->ptr, &tmp, sizeof(tmp)); 215 if (error) 216 return (error); 217 a.msgp = tmp.msgp; 218 a.msgtyp = tmp.msgtyp; 219 } else { 220 a.msgp = args->ptr; 221 a.msgtyp = args->arg5; 222 } 223 return (linux_msgrcv(td, &a)); 224 } 225 case LINUX_MSGGET: { 226 struct linux_msgget_args a; 227 228 a.key = args->arg1; 229 a.msgflg = args->arg2; 230 return (linux_msgget(td, &a)); 231 } 232 case LINUX_MSGCTL: { 233 struct linux_msgctl_args a; 234 235 a.msqid = args->arg1; 236 a.cmd = args->arg2; 237 a.buf = args->ptr; 238 return (linux_msgctl(td, &a)); 239 } 240 case LINUX_SHMAT: { 241 struct linux_shmat_args a; 242 243 a.shmid = args->arg1; 244 a.shmaddr = args->ptr; 245 a.shmflg = args->arg2; 246 a.raddr = (l_ulong *)args->arg3; 247 return (linux_shmat(td, &a)); 248 } 249 case LINUX_SHMDT: { 250 struct linux_shmdt_args a; 251 252 a.shmaddr = args->ptr; 253 return (linux_shmdt(td, &a)); 254 } 255 case LINUX_SHMGET: { 256 struct linux_shmget_args a; 257 258 a.key = args->arg1; 259 a.size = args->arg2; 260 a.shmflg = args->arg3; 261 return (linux_shmget(td, &a)); 262 } 263 case LINUX_SHMCTL: { 264 struct linux_shmctl_args a; 265 266 a.shmid = args->arg1; 267 a.cmd = args->arg2; 268 a.buf = args->ptr; 269 return (linux_shmctl(td, &a)); 270 } 271 default: 272 break; 273 } 274 275 return (EINVAL); 276} 277 278int 279linux_old_select(struct thread *td, struct linux_old_select_args *args) 280{ 281 struct l_old_select_argv linux_args; 282 struct linux_select_args newsel; 283 int error; 284 285#ifdef DEBUG 286 if (ldebug(old_select)) 287 printf(ARGS(old_select, "%p"), args->ptr); 288#endif 289 290 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 291 if (error) 292 return (error); 293 294 newsel.nfds = linux_args.nfds; 295 newsel.readfds = linux_args.readfds; 296 newsel.writefds = linux_args.writefds; 297 newsel.exceptfds = linux_args.exceptfds; 298 newsel.timeout = linux_args.timeout; 299 return (linux_select(td, &newsel)); 300} 301 302int 303linux_fork(struct thread *td, struct linux_fork_args *args) 304{ 305 int error; 306 struct proc *p2; 307 struct thread *td2; 308 309#ifdef DEBUG 310 if (ldebug(fork)) 311 printf(ARGS(fork, "")); 312#endif 313 314 if ((error = fork1(td, RFFDG | RFPROC | RFSTOPPED, 0, &p2)) != 0) 315 return (error); 316 317 if (error == 0) { 318 td->td_retval[0] = p2->p_pid; 319 td->td_retval[1] = 0; 320 } 321 322 if (td->td_retval[1] == 1) 323 td->td_retval[0] = 0; 324 error = linux_proc_init(td, td->td_retval[0], 0); 325 if (error) 326 return (error); 327 328 td2 = FIRST_THREAD_IN_PROC(p2); 329 330 /* 331 * Make this runnable after we are finished with it. 332 */ 333 thread_lock(td2); 334 TD_SET_CAN_RUN(td2); 335 sched_add(td2, SRQ_BORING); 336 thread_unlock(td2); 337 338 return (0); 339} 340 341int 342linux_vfork(struct thread *td, struct linux_vfork_args *args) 343{ 344 int error; 345 struct proc *p2; 346 struct thread *td2; 347 348#ifdef DEBUG 349 if (ldebug(vfork)) 350 printf(ARGS(vfork, "")); 351#endif 352 353 /* exclude RFPPWAIT */ 354 if ((error = fork1(td, RFFDG | RFPROC | RFMEM | RFSTOPPED, 0, &p2)) != 0) 355 return (error); 356 if (error == 0) { 357 td->td_retval[0] = p2->p_pid; 358 td->td_retval[1] = 0; 359 } 360 /* Are we the child? */ 361 if (td->td_retval[1] == 1) 362 td->td_retval[0] = 0; 363 error = linux_proc_init(td, td->td_retval[0], 0); 364 if (error) 365 return (error); 366 367 PROC_LOCK(p2); 368 p2->p_flag |= P_PPWAIT; 369 PROC_UNLOCK(p2); 370 371 td2 = FIRST_THREAD_IN_PROC(p2); 372 373 /* 374 * Make this runnable after we are finished with it. 375 */ 376 thread_lock(td2); 377 TD_SET_CAN_RUN(td2); 378 sched_add(td2, SRQ_BORING); 379 thread_unlock(td2); 380 381 /* wait for the children to exit, ie. emulate vfork */ 382 PROC_LOCK(p2); 383 while (p2->p_flag & P_PPWAIT) 384 cv_wait(&p2->p_pwait, &p2->p_mtx); 385 PROC_UNLOCK(p2); 386 387 return (0); 388} 389 390int 391linux_clone(struct thread *td, struct linux_clone_args *args) 392{ 393 int error, ff = RFPROC | RFSTOPPED; 394 struct proc *p2; 395 struct thread *td2; 396 int exit_signal; 397 struct linux_emuldata *em; 398 399#ifdef DEBUG 400 if (ldebug(clone)) { 401 printf(ARGS(clone, "flags %x, stack %x, parent tid: %x, child tid: %x"), 402 (unsigned int)args->flags, (unsigned int)args->stack, 403 (unsigned int)args->parent_tidptr, (unsigned int)args->child_tidptr); 404 } 405#endif 406 407 exit_signal = args->flags & 0x000000ff; 408 if (LINUX_SIG_VALID(exit_signal)) { 409 if (exit_signal <= LINUX_SIGTBLSZ) 410 exit_signal = 411 linux_to_bsd_signal[_SIG_IDX(exit_signal)]; 412 } else if (exit_signal != 0) 413 return (EINVAL); 414 415 if (args->flags & LINUX_CLONE_VM) 416 ff |= RFMEM; 417 if (args->flags & LINUX_CLONE_SIGHAND) 418 ff |= RFSIGSHARE; 419 /* 420 * XXX: in linux sharing of fs info (chroot/cwd/umask) 421 * and open files is independant. in fbsd its in one 422 * structure but in reality it doesn't cause any problems 423 * because both of these flags are usually set together. 424 */ 425 if (!(args->flags & (LINUX_CLONE_FILES | LINUX_CLONE_FS))) 426 ff |= RFFDG; 427 428 /* 429 * Attempt to detect when linux_clone(2) is used for creating 430 * kernel threads. Unfortunately despite the existence of the 431 * CLONE_THREAD flag, version of linuxthreads package used in 432 * most popular distros as of beginning of 2005 doesn't make 433 * any use of it. Therefore, this detection relies on 434 * empirical observation that linuxthreads sets certain 435 * combination of flags, so that we can make more or less 436 * precise detection and notify the FreeBSD kernel that several 437 * processes are in fact part of the same threading group, so 438 * that special treatment is necessary for signal delivery 439 * between those processes and fd locking. 440 */ 441 if ((args->flags & 0xffffff00) == LINUX_THREADING_FLAGS) 442 ff |= RFTHREAD; 443 444 if (args->flags & LINUX_CLONE_PARENT_SETTID) 445 if (args->parent_tidptr == NULL) 446 return (EINVAL); 447 448 error = fork1(td, ff, 0, &p2); 449 if (error) 450 return (error); 451 452 if (args->flags & (LINUX_CLONE_PARENT | LINUX_CLONE_THREAD)) { 453 sx_xlock(&proctree_lock); 454 PROC_LOCK(p2); 455 proc_reparent(p2, td->td_proc->p_pptr); 456 PROC_UNLOCK(p2); 457 sx_xunlock(&proctree_lock); 458 } 459 460 /* create the emuldata */ 461 error = linux_proc_init(td, p2->p_pid, args->flags); 462 /* reference it - no need to check this */ 463 em = em_find(p2, EMUL_DOLOCK); 464 KASSERT(em != NULL, ("clone: emuldata not found.\n")); 465 /* and adjust it */ 466 467 if (args->flags & LINUX_CLONE_THREAD) { 468 /* XXX: linux mangles pgrp and pptr somehow 469 * I think it might be this but I am not sure. 470 */ 471#ifdef notyet 472 PROC_LOCK(p2); 473 p2->p_pgrp = td->td_proc->p_pgrp; 474 PROC_UNLOCK(p2); 475#endif 476 exit_signal = 0; 477 } 478 479 if (args->flags & LINUX_CLONE_CHILD_SETTID) 480 em->child_set_tid = args->child_tidptr; 481 else 482 em->child_set_tid = NULL; 483 484 if (args->flags & LINUX_CLONE_CHILD_CLEARTID) 485 em->child_clear_tid = args->child_tidptr; 486 else 487 em->child_clear_tid = NULL; 488 489 EMUL_UNLOCK(&emul_lock); 490 491 if (args->flags & LINUX_CLONE_PARENT_SETTID) { 492 error = copyout(&p2->p_pid, args->parent_tidptr, sizeof(p2->p_pid)); 493 if (error) 494 printf(LMSG("copyout failed!")); 495 } 496 497 PROC_LOCK(p2); 498 p2->p_sigparent = exit_signal; 499 PROC_UNLOCK(p2); 500 td2 = FIRST_THREAD_IN_PROC(p2); 501 /* 502 * in a case of stack = NULL we are supposed to COW calling process stack 503 * this is what normal fork() does so we just keep the tf_esp arg intact 504 */ 505 if (args->stack) 506 td2->td_frame->tf_esp = (unsigned int)args->stack; 507 508 if (args->flags & LINUX_CLONE_SETTLS) { 509 struct l_user_desc info; 510 int idx; 511 int a[2]; 512 struct segment_descriptor sd; 513 514 error = copyin((void *)td->td_frame->tf_esi, &info, sizeof(struct l_user_desc)); 515 if (error) { 516 printf(LMSG("copyin failed!")); 517 } else { 518 519 idx = info.entry_number; 520 521 /* 522 * looks like we're getting the idx we returned 523 * in the set_thread_area() syscall 524 */ 525 if (idx != 6 && idx != 3) { 526 printf(LMSG("resetting idx!")); 527 idx = 3; 528 } 529 530 /* this doesnt happen in practice */ 531 if (idx == 6) { 532 /* we might copy out the entry_number as 3 */ 533 info.entry_number = 3; 534 error = copyout(&info, (void *) td->td_frame->tf_esi, sizeof(struct l_user_desc)); 535 if (error) 536 printf(LMSG("copyout failed!")); 537 } 538 539 a[0] = LINUX_LDT_entry_a(&info); 540 a[1] = LINUX_LDT_entry_b(&info); 541 542 memcpy(&sd, &a, sizeof(a)); 543#ifdef DEBUG 544 if (ldebug(clone)) 545 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, 546 sd.sd_hibase, 547 sd.sd_lolimit, 548 sd.sd_hilimit, 549 sd.sd_type, 550 sd.sd_dpl, 551 sd.sd_p, 552 sd.sd_xx, 553 sd.sd_def32, 554 sd.sd_gran); 555#endif 556 557 /* set %gs */ 558 td2->td_pcb->pcb_gsd = sd; 559 td2->td_pcb->pcb_gs = GSEL(GUGS_SEL, SEL_UPL); 560 } 561 } 562 563#ifdef DEBUG 564 if (ldebug(clone)) 565 printf(LMSG("clone: successful rfork to %ld, stack %p sig = %d"), 566 (long)p2->p_pid, args->stack, exit_signal); 567#endif 568 if (args->flags & LINUX_CLONE_VFORK) { 569 PROC_LOCK(p2); 570 p2->p_flag |= P_PPWAIT; 571 PROC_UNLOCK(p2); 572 } 573 574 /* 575 * Make this runnable after we are finished with it. 576 */ 577 thread_lock(td2); 578 TD_SET_CAN_RUN(td2); 579 sched_add(td2, SRQ_BORING); 580 thread_unlock(td2); 581 582 td->td_retval[0] = p2->p_pid; 583 td->td_retval[1] = 0; 584 585 if (args->flags & LINUX_CLONE_VFORK) { 586 /* wait for the children to exit, ie. emulate vfork */ 587 PROC_LOCK(p2); 588 while (p2->p_flag & P_PPWAIT) 589 cv_wait(&p2->p_pwait, &p2->p_mtx); 590 PROC_UNLOCK(p2); 591 } 592 593 return (0); 594} 595 596#define STACK_SIZE (2 * 1024 * 1024) 597#define GUARD_SIZE (4 * PAGE_SIZE) 598 599int 600linux_mmap2(struct thread *td, struct linux_mmap2_args *args) 601{ 602 603#ifdef DEBUG 604 if (ldebug(mmap2)) 605 printf(ARGS(mmap2, "%p, %d, %d, 0x%08x, %d, %d"), 606 (void *)args->addr, args->len, args->prot, 607 args->flags, args->fd, args->pgoff); 608#endif 609 610 return (linux_mmap_common(td, args->addr, args->len, args->prot, 611 args->flags, args->fd, (uint64_t)(uint32_t)args->pgoff * 612 PAGE_SIZE)); 613} 614 615int 616linux_mmap(struct thread *td, struct linux_mmap_args *args) 617{ 618 int error; 619 struct l_mmap_argv linux_args; 620 621 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 622 if (error) 623 return (error); 624 625#ifdef DEBUG 626 if (ldebug(mmap)) 627 printf(ARGS(mmap, "%p, %d, %d, 0x%08x, %d, %d"), 628 (void *)linux_args.addr, linux_args.len, linux_args.prot, 629 linux_args.flags, linux_args.fd, linux_args.pgoff); 630#endif 631 632 return (linux_mmap_common(td, linux_args.addr, linux_args.len, 633 linux_args.prot, linux_args.flags, linux_args.fd, 634 (uint32_t)linux_args.pgoff)); 635} 636 637static int 638linux_mmap_common(struct thread *td, l_uintptr_t addr, l_size_t len, l_int prot, 639 l_int flags, l_int fd, l_loff_t pos) 640{ 641 struct proc *p = td->td_proc; 642 struct mmap_args /* { 643 caddr_t addr; 644 size_t len; 645 int prot; 646 int flags; 647 int fd; 648 long pad; 649 off_t pos; 650 } */ bsd_args; 651 int error; 652 struct file *fp; 653 654 error = 0; 655 bsd_args.flags = 0; 656 fp = NULL; 657 658 /* 659 * Linux mmap(2): 660 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE 661 */ 662 if (!((flags & LINUX_MAP_SHARED) ^ (flags & LINUX_MAP_PRIVATE))) 663 return (EINVAL); 664 665 if (flags & LINUX_MAP_SHARED) 666 bsd_args.flags |= MAP_SHARED; 667 if (flags & LINUX_MAP_PRIVATE) 668 bsd_args.flags |= MAP_PRIVATE; 669 if (flags & LINUX_MAP_FIXED) 670 bsd_args.flags |= MAP_FIXED; 671 if (flags & LINUX_MAP_ANON) { 672 /* Enforce pos to be on page boundary, then ignore. */ 673 if ((pos & PAGE_MASK) != 0) 674 return (EINVAL); 675 pos = 0; 676 bsd_args.flags |= MAP_ANON; 677 } else 678 bsd_args.flags |= MAP_NOSYNC; 679 if (flags & LINUX_MAP_GROWSDOWN) 680 bsd_args.flags |= MAP_STACK; 681 682 /* 683 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC 684 * on Linux/i386. We do this to ensure maximum compatibility. 685 * Linux/ia64 does the same in i386 emulation mode. 686 */ 687 bsd_args.prot = prot; 688 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 689 bsd_args.prot |= PROT_READ | PROT_EXEC; 690 691 /* Linux does not check file descriptor when MAP_ANONYMOUS is set. */ 692 bsd_args.fd = (bsd_args.flags & MAP_ANON) ? -1 : fd; 693 if (bsd_args.fd != -1) { 694 /* 695 * Linux follows Solaris mmap(2) description: 696 * The file descriptor fildes is opened with 697 * read permission, regardless of the 698 * protection options specified. 699 */ 700 701 if ((error = fget(td, bsd_args.fd, &fp)) != 0) 702 return (error); 703 if (fp->f_type != DTYPE_VNODE) { 704 fdrop(fp, td); 705 return (EINVAL); 706 } 707 708 /* Linux mmap() just fails for O_WRONLY files */ 709 if (!(fp->f_flag & FREAD)) { 710 fdrop(fp, td); 711 return (EACCES); 712 } 713 714 fdrop(fp, td); 715 } 716 717 if (flags & LINUX_MAP_GROWSDOWN) { 718 /* 719 * The Linux MAP_GROWSDOWN option does not limit auto 720 * growth of the region. Linux mmap with this option 721 * takes as addr the inital BOS, and as len, the initial 722 * region size. It can then grow down from addr without 723 * limit. However, linux threads has an implicit internal 724 * limit to stack size of STACK_SIZE. Its just not 725 * enforced explicitly in linux. But, here we impose 726 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack 727 * region, since we can do this with our mmap. 728 * 729 * Our mmap with MAP_STACK takes addr as the maximum 730 * downsize limit on BOS, and as len the max size of 731 * the region. It them maps the top SGROWSIZ bytes, 732 * and auto grows the region down, up to the limit 733 * in addr. 734 * 735 * If we don't use the MAP_STACK option, the effect 736 * of this code is to allocate a stack region of a 737 * fixed size of (STACK_SIZE - GUARD_SIZE). 738 */ 739 740 if ((caddr_t)PTRIN(addr) + len > p->p_vmspace->vm_maxsaddr) { 741 /* 742 * Some linux apps will attempt to mmap 743 * thread stacks near the top of their 744 * address space. If their TOS is greater 745 * than vm_maxsaddr, vm_map_growstack() 746 * will confuse the thread stack with the 747 * process stack and deliver a SEGV if they 748 * attempt to grow the thread stack past their 749 * current stacksize rlimit. To avoid this, 750 * adjust vm_maxsaddr upwards to reflect 751 * the current stacksize rlimit rather 752 * than the maximum possible stacksize. 753 * It would be better to adjust the 754 * mmap'ed region, but some apps do not check 755 * mmap's return value. 756 */ 757 PROC_LOCK(p); 758 p->p_vmspace->vm_maxsaddr = (char *)USRSTACK - 759 lim_cur(p, RLIMIT_STACK); 760 PROC_UNLOCK(p); 761 } 762 763 /* 764 * This gives us our maximum stack size and a new BOS. 765 * If we're using VM_STACK, then mmap will just map 766 * the top SGROWSIZ bytes, and let the stack grow down 767 * to the limit at BOS. If we're not using VM_STACK 768 * we map the full stack, since we don't have a way 769 * to autogrow it. 770 */ 771 if (len > STACK_SIZE - GUARD_SIZE) { 772 bsd_args.addr = (caddr_t)PTRIN(addr); 773 bsd_args.len = len; 774 } else { 775 bsd_args.addr = (caddr_t)PTRIN(addr) - 776 (STACK_SIZE - GUARD_SIZE - len); 777 bsd_args.len = STACK_SIZE - GUARD_SIZE; 778 } 779 } else { 780 bsd_args.addr = (caddr_t)PTRIN(addr); 781 bsd_args.len = len; 782 } 783 bsd_args.pos = pos; 784 785#ifdef DEBUG 786 if (ldebug(mmap)) 787 printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n", 788 __func__, 789 (void *)bsd_args.addr, bsd_args.len, bsd_args.prot, 790 bsd_args.flags, bsd_args.fd, (int)bsd_args.pos); 791#endif 792 error = mmap(td, &bsd_args); 793#ifdef DEBUG 794 if (ldebug(mmap)) 795 printf("-> %s() return: 0x%x (0x%08x)\n", 796 __func__, error, (u_int)td->td_retval[0]); 797#endif 798 return (error); 799} 800 801int 802linux_mprotect(struct thread *td, struct linux_mprotect_args *uap) 803{ 804 struct mprotect_args bsd_args; 805 806 bsd_args.addr = uap->addr; 807 bsd_args.len = uap->len; 808 bsd_args.prot = uap->prot; 809 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 810 bsd_args.prot |= PROT_READ | PROT_EXEC; 811 return (mprotect(td, &bsd_args)); 812} 813 814int 815linux_pipe(struct thread *td, struct linux_pipe_args *args) 816{ 817 int error; 818 int fildes[2]; 819 820#ifdef DEBUG 821 if (ldebug(pipe)) 822 printf(ARGS(pipe, "*")); 823#endif 824 825 error = kern_pipe(td, fildes); 826 if (error) 827 return (error); 828 829 /* XXX: Close descriptors on error. */ 830 return (copyout(fildes, args->pipefds, sizeof fildes)); 831} 832 833int 834linux_ioperm(struct thread *td, struct linux_ioperm_args *args) 835{ 836 int error; 837 struct i386_ioperm_args iia; 838 839 iia.start = args->start; 840 iia.length = args->length; 841 iia.enable = args->enable; 842 error = i386_set_ioperm(td, &iia); 843 return (error); 844} 845 846int 847linux_iopl(struct thread *td, struct linux_iopl_args *args) 848{ 849 int error; 850 851 if (args->level < 0 || args->level > 3) 852 return (EINVAL); 853 if ((error = priv_check(td, PRIV_IO)) != 0) 854 return (error); 855 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 856 return (error); 857 td->td_frame->tf_eflags = (td->td_frame->tf_eflags & ~PSL_IOPL) | 858 (args->level * (PSL_IOPL / 3)); 859 return (0); 860} 861 862int 863linux_modify_ldt(struct thread *td, struct linux_modify_ldt_args *uap) 864{ 865 int error; 866 struct i386_ldt_args ldt; 867 struct l_descriptor ld; 868 union descriptor desc; 869 int size, written; 870 871 switch (uap->func) { 872 case 0x00: /* read_ldt */ 873 ldt.start = 0; 874 ldt.descs = uap->ptr; 875 ldt.num = uap->bytecount / sizeof(union descriptor); 876 error = i386_get_ldt(td, &ldt); 877 td->td_retval[0] *= sizeof(union descriptor); 878 break; 879 case 0x02: /* read_default_ldt = 0 */ 880 size = 5*sizeof(struct l_desc_struct); 881 if (size > uap->bytecount) 882 size = uap->bytecount; 883 for (written = error = 0; written < size && error == 0; written++) 884 error = subyte((char *)uap->ptr + written, 0); 885 td->td_retval[0] = written; 886 break; 887 case 0x01: /* write_ldt */ 888 case 0x11: /* write_ldt */ 889 if (uap->bytecount != sizeof(ld)) 890 return (EINVAL); 891 892 error = copyin(uap->ptr, &ld, sizeof(ld)); 893 if (error) 894 return (error); 895 896 ldt.start = ld.entry_number; 897 ldt.descs = &desc; 898 ldt.num = 1; 899 desc.sd.sd_lolimit = (ld.limit & 0x0000ffff); 900 desc.sd.sd_hilimit = (ld.limit & 0x000f0000) >> 16; 901 desc.sd.sd_lobase = (ld.base_addr & 0x00ffffff); 902 desc.sd.sd_hibase = (ld.base_addr & 0xff000000) >> 24; 903 desc.sd.sd_type = SDT_MEMRO | ((ld.read_exec_only ^ 1) << 1) | 904 (ld.contents << 2); 905 desc.sd.sd_dpl = 3; 906 desc.sd.sd_p = (ld.seg_not_present ^ 1); 907 desc.sd.sd_xx = 0; 908 desc.sd.sd_def32 = ld.seg_32bit; 909 desc.sd.sd_gran = ld.limit_in_pages; 910 error = i386_set_ldt(td, &ldt, &desc); 911 break; 912 default: 913 error = ENOSYS; 914 break; 915 } 916 917 if (error == EOPNOTSUPP) { 918 printf("linux: modify_ldt needs kernel option USER_LDT\n"); 919 error = ENOSYS; 920 } 921 922 return (error); 923} 924 925int 926linux_sigaction(struct thread *td, struct linux_sigaction_args *args) 927{ 928 l_osigaction_t osa; 929 l_sigaction_t act, oact; 930 int error; 931 932#ifdef DEBUG 933 if (ldebug(sigaction)) 934 printf(ARGS(sigaction, "%d, %p, %p"), 935 args->sig, (void *)args->nsa, (void *)args->osa); 936#endif 937 938 if (args->nsa != NULL) { 939 error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); 940 if (error) 941 return (error); 942 act.lsa_handler = osa.lsa_handler; 943 act.lsa_flags = osa.lsa_flags; 944 act.lsa_restorer = osa.lsa_restorer; 945 LINUX_SIGEMPTYSET(act.lsa_mask); 946 act.lsa_mask.__bits[0] = osa.lsa_mask; 947 } 948 949 error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, 950 args->osa ? &oact : NULL); 951 952 if (args->osa != NULL && !error) { 953 osa.lsa_handler = oact.lsa_handler; 954 osa.lsa_flags = oact.lsa_flags; 955 osa.lsa_restorer = oact.lsa_restorer; 956 osa.lsa_mask = oact.lsa_mask.__bits[0]; 957 error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); 958 } 959 960 return (error); 961} 962 963/* 964 * Linux has two extra args, restart and oldmask. We dont use these, 965 * but it seems that "restart" is actually a context pointer that 966 * enables the signal to happen with a different register set. 967 */ 968int 969linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) 970{ 971 sigset_t sigmask; 972 l_sigset_t mask; 973 974#ifdef DEBUG 975 if (ldebug(sigsuspend)) 976 printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask); 977#endif 978 979 LINUX_SIGEMPTYSET(mask); 980 mask.__bits[0] = args->mask; 981 linux_to_bsd_sigset(&mask, &sigmask); 982 return (kern_sigsuspend(td, sigmask)); 983} 984 985int 986linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap) 987{ 988 l_sigset_t lmask; 989 sigset_t sigmask; 990 int error; 991 992#ifdef DEBUG 993 if (ldebug(rt_sigsuspend)) 994 printf(ARGS(rt_sigsuspend, "%p, %d"), 995 (void *)uap->newset, uap->sigsetsize); 996#endif 997 998 if (uap->sigsetsize != sizeof(l_sigset_t)) 999 return (EINVAL); 1000 1001 error = copyin(uap->newset, &lmask, sizeof(l_sigset_t)); 1002 if (error) 1003 return (error); 1004 1005 linux_to_bsd_sigset(&lmask, &sigmask); 1006 return (kern_sigsuspend(td, sigmask)); 1007} 1008 1009int 1010linux_pause(struct thread *td, struct linux_pause_args *args) 1011{ 1012 struct proc *p = td->td_proc; 1013 sigset_t sigmask; 1014 1015#ifdef DEBUG 1016 if (ldebug(pause)) 1017 printf(ARGS(pause, "")); 1018#endif 1019 1020 PROC_LOCK(p); 1021 sigmask = td->td_sigmask; 1022 PROC_UNLOCK(p); 1023 return (kern_sigsuspend(td, sigmask)); 1024} 1025 1026int 1027linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap) 1028{ 1029 stack_t ss, oss; 1030 l_stack_t lss; 1031 int error; 1032 1033#ifdef DEBUG 1034 if (ldebug(sigaltstack)) 1035 printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss); 1036#endif 1037 1038 if (uap->uss != NULL) { 1039 error = copyin(uap->uss, &lss, sizeof(l_stack_t)); 1040 if (error) 1041 return (error); 1042 1043 ss.ss_sp = lss.ss_sp; 1044 ss.ss_size = lss.ss_size; 1045 ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags); 1046 } 1047 error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL, 1048 (uap->uoss != NULL) ? &oss : NULL); 1049 if (!error && uap->uoss != NULL) { 1050 lss.ss_sp = oss.ss_sp; 1051 lss.ss_size = oss.ss_size; 1052 lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags); 1053 error = copyout(&lss, uap->uoss, sizeof(l_stack_t)); 1054 } 1055 1056 return (error); 1057} 1058 1059int 1060linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args) 1061{ 1062 struct ftruncate_args sa; 1063 1064#ifdef DEBUG 1065 if (ldebug(ftruncate64)) 1066 printf(ARGS(ftruncate64, "%u, %jd"), args->fd, 1067 (intmax_t)args->length); 1068#endif 1069 1070 sa.fd = args->fd; 1071 sa.length = args->length; 1072 return ftruncate(td, &sa); 1073} 1074 1075int 1076linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args) 1077{ 1078 struct l_user_desc info; 1079 int error; 1080 int idx; 1081 int a[2]; 1082 struct segment_descriptor sd; 1083 1084 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1085 if (error) 1086 return (error); 1087 1088#ifdef DEBUG 1089 if (ldebug(set_thread_area)) 1090 printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, %i, %i, %i\n"), 1091 info.entry_number, 1092 info.base_addr, 1093 info.limit, 1094 info.seg_32bit, 1095 info.contents, 1096 info.read_exec_only, 1097 info.limit_in_pages, 1098 info.seg_not_present, 1099 info.useable); 1100#endif 1101 1102 idx = info.entry_number; 1103 /* 1104 * Semantics of linux version: every thread in the system has array of 1105 * 3 tls descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. This 1106 * syscall loads one of the selected tls decriptors with a value and 1107 * also loads GDT descriptors 6, 7 and 8 with the content of the 1108 * per-thread descriptors. 1109 * 1110 * Semantics of fbsd version: I think we can ignore that linux has 3 1111 * per-thread descriptors and use just the 1st one. The tls_array[] 1112 * is used only in set/get-thread_area() syscalls and for loading the 1113 * GDT descriptors. In fbsd we use just one GDT descriptor for TLS so 1114 * we will load just one. 1115 * 1116 * XXX: this doesn't work when a user space process tries to use more 1117 * than 1 TLS segment. Comment in the linux sources says wine might do 1118 * this. 1119 */ 1120 1121 /* 1122 * we support just GLIBC TLS now 1123 * we should let 3 proceed as well because we use this segment so 1124 * if code does two subsequent calls it should succeed 1125 */ 1126 if (idx != 6 && idx != -1 && idx != 3) 1127 return (EINVAL); 1128 1129 /* 1130 * we have to copy out the GDT entry we use 1131 * FreeBSD uses GDT entry #3 for storing %gs so load that 1132 * 1133 * XXX: what if a user space program doesn't check this value and tries 1134 * to use 6, 7 or 8? 1135 */ 1136 idx = info.entry_number = 3; 1137 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 1138 if (error) 1139 return (error); 1140 1141 if (LINUX_LDT_empty(&info)) { 1142 a[0] = 0; 1143 a[1] = 0; 1144 } else { 1145 a[0] = LINUX_LDT_entry_a(&info); 1146 a[1] = LINUX_LDT_entry_b(&info); 1147 } 1148 1149 memcpy(&sd, &a, sizeof(a)); 1150#ifdef DEBUG 1151 if (ldebug(set_thread_area)) 1152 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, 1153 sd.sd_hibase, 1154 sd.sd_lolimit, 1155 sd.sd_hilimit, 1156 sd.sd_type, 1157 sd.sd_dpl, 1158 sd.sd_p, 1159 sd.sd_xx, 1160 sd.sd_def32, 1161 sd.sd_gran); 1162#endif 1163 1164 /* this is taken from i386 version of cpu_set_user_tls() */ 1165 critical_enter(); 1166 /* set %gs */ 1167 td->td_pcb->pcb_gsd = sd; 1168 PCPU_GET(fsgs_gdt)[1] = sd; 1169 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 1170 critical_exit(); 1171 1172 return (0); 1173} 1174 1175int 1176linux_get_thread_area(struct thread *td, struct linux_get_thread_area_args *args) 1177{ 1178 1179 struct l_user_desc info; 1180 int error; 1181 int idx; 1182 struct l_desc_struct desc; 1183 struct segment_descriptor sd; 1184 1185#ifdef DEBUG 1186 if (ldebug(get_thread_area)) 1187 printf(ARGS(get_thread_area, "%p"), args->desc); 1188#endif 1189 1190 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1191 if (error) 1192 return (error); 1193 1194 idx = info.entry_number; 1195 /* XXX: I am not sure if we want 3 to be allowed too. */ 1196 if (idx != 6 && idx != 3) 1197 return (EINVAL); 1198 1199 idx = 3; 1200 1201 memset(&info, 0, sizeof(info)); 1202 1203 sd = PCPU_GET(fsgs_gdt)[1]; 1204 1205 memcpy(&desc, &sd, sizeof(desc)); 1206 1207 info.entry_number = idx; 1208 info.base_addr = LINUX_GET_BASE(&desc); 1209 info.limit = LINUX_GET_LIMIT(&desc); 1210 info.seg_32bit = LINUX_GET_32BIT(&desc); 1211 info.contents = LINUX_GET_CONTENTS(&desc); 1212 info.read_exec_only = !LINUX_GET_WRITABLE(&desc); 1213 info.limit_in_pages = LINUX_GET_LIMIT_PAGES(&desc); 1214 info.seg_not_present = !LINUX_GET_PRESENT(&desc); 1215 info.useable = LINUX_GET_USEABLE(&desc); 1216 1217 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 1218 if (error) 1219 return (EFAULT); 1220 1221 return (0); 1222} 1223 1224/* copied from kern/kern_time.c */ 1225int 1226linux_timer_create(struct thread *td, struct linux_timer_create_args *args) 1227{ 1228 return ktimer_create(td, (struct ktimer_create_args *) args); 1229} 1230 1231int 1232linux_timer_settime(struct thread *td, struct linux_timer_settime_args *args) 1233{ 1234 return ktimer_settime(td, (struct ktimer_settime_args *) args); 1235} 1236 1237int 1238linux_timer_gettime(struct thread *td, struct linux_timer_gettime_args *args) 1239{ 1240 return ktimer_gettime(td, (struct ktimer_gettime_args *) args); 1241} 1242 1243int 1244linux_timer_getoverrun(struct thread *td, struct linux_timer_getoverrun_args *args) 1245{ 1246 return ktimer_getoverrun(td, (struct ktimer_getoverrun_args *) args); 1247} 1248 1249int 1250linux_timer_delete(struct thread *td, struct linux_timer_delete_args *args) 1251{ 1252 return ktimer_delete(td, (struct ktimer_delete_args *) args); 1253} 1254 1255/* XXX: this wont work with module - convert it */ 1256int 1257linux_mq_open(struct thread *td, struct linux_mq_open_args *args) 1258{ 1259#ifdef P1003_1B_MQUEUE 1260 return kmq_open(td, (struct kmq_open_args *) args); 1261#else 1262 return (ENOSYS); 1263#endif 1264} 1265 1266int 1267linux_mq_unlink(struct thread *td, struct linux_mq_unlink_args *args) 1268{ 1269#ifdef P1003_1B_MQUEUE 1270 return kmq_unlink(td, (struct kmq_unlink_args *) args); 1271#else 1272 return (ENOSYS); 1273#endif 1274} 1275 1276int 1277linux_mq_timedsend(struct thread *td, struct linux_mq_timedsend_args *args) 1278{ 1279#ifdef P1003_1B_MQUEUE 1280 return kmq_timedsend(td, (struct kmq_timedsend_args *) args); 1281#else 1282 return (ENOSYS); 1283#endif 1284} 1285 1286int 1287linux_mq_timedreceive(struct thread *td, struct linux_mq_timedreceive_args *args) 1288{ 1289#ifdef P1003_1B_MQUEUE 1290 return kmq_timedreceive(td, (struct kmq_timedreceive_args *) args); 1291#else 1292 return (ENOSYS); 1293#endif 1294} 1295 1296int 1297linux_mq_notify(struct thread *td, struct linux_mq_notify_args *args) 1298{ 1299#ifdef P1003_1B_MQUEUE 1300 return kmq_notify(td, (struct kmq_notify_args *) args); 1301#else 1302 return (ENOSYS); 1303#endif 1304} 1305 1306int 1307linux_mq_getsetattr(struct thread *td, struct linux_mq_getsetattr_args *args) 1308{ 1309#ifdef P1003_1B_MQUEUE 1310 return kmq_setattr(td, (struct kmq_setattr_args *) args); 1311#else 1312 return (ENOSYS); 1313#endif 1314} 1315 1316int 1317linux_wait4(struct thread *td, struct linux_wait4_args *args) 1318{ 1319 int error, options; 1320 struct rusage ru, *rup; 1321 1322#ifdef DEBUG 1323 if (ldebug(wait4)) 1324 printf(ARGS(wait4, "%d, %p, %d, %p"), 1325 args->pid, (void *)args->status, args->options, 1326 (void *)args->rusage); 1327#endif 1328 1329 options = (args->options & (WNOHANG | WUNTRACED)); 1330 /* WLINUXCLONE should be equal to __WCLONE, but we make sure */ 1331 if (args->options & __WCLONE) 1332 options |= WLINUXCLONE; 1333 1334 if (args->rusage != NULL) 1335 rup = &ru; 1336 else 1337 rup = NULL; 1338 error = linux_common_wait(td, args->pid, args->status, options, rup); 1339 if (error) 1340 return (error); 1341 if (args->rusage != NULL) 1342 error = copyout(&ru, args->rusage, sizeof(ru)); 1343 1344 return (error); 1345} 1346