linux_machdep.c revision 166727
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 166727 2007-02-15 00:54:40Z jkim $"); 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 74extern struct sysentvec elf32_freebsd_sysvec; /* defined in i386/i386/elf_machdep.c */ 75 76struct 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 88struct 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 96int 97linux_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 108int 109bsd_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 120int 121linux_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 150struct l_ipc_kludge { 151 struct l_msgbuf *msgp; 152 l_long msgtyp; 153}; 154 155int 156linux_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 273int 274linux_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 297int 298linux_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 336int 337linux_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 385int 386linux_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 doesnt make any problems 417 * because both this flags are set at once usually. 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 relay fully 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 error = fork1(td, ff, 0, &p2); 439 if (error) 440 return (error); 441 442 if (args->flags & (CLONE_PARENT|CLONE_THREAD)) { 443 sx_xlock(&proctree_lock); 444 PROC_LOCK(p2); 445 proc_reparent(p2, td->td_proc->p_pptr); 446 PROC_UNLOCK(p2); 447 sx_xunlock(&proctree_lock); 448 } 449 450 /* create the emuldata */ 451 error = linux_proc_init(td, p2->p_pid, args->flags); 452 /* reference it - no need to check this */ 453 em = em_find(p2, EMUL_DOLOCK); 454 KASSERT(em != NULL, ("clone: emuldata not found.\n")); 455 /* and adjust it */ 456 if (args->flags & CLONE_PARENT_SETTID) { 457 if (args->parent_tidptr == NULL) { 458 EMUL_UNLOCK(&emul_lock); 459 return (EINVAL); 460 } 461 error = copyout(&p2->p_pid, args->parent_tidptr, sizeof(p2->p_pid)); 462 if (error) { 463 EMUL_UNLOCK(&emul_lock); 464 return (error); 465 } 466 } 467 468 if (args->flags & CLONE_THREAD) { 469 /* XXX: linux mangles pgrp and pptr somehow 470 * I think it might be this but I am not sure. 471 */ 472#ifdef notyet 473 PROC_LOCK(p2); 474 p2->p_pgrp = td->td_proc->p_pgrp; 475 PROC_UNLOCK(p2); 476#endif 477 exit_signal = 0; 478 } 479 480 if (args->flags & CLONE_CHILD_SETTID) 481 em->child_set_tid = args->child_tidptr; 482 else 483 em->child_set_tid = NULL; 484 485 if (args->flags & CLONE_CHILD_CLEARTID) 486 em->child_clear_tid = args->child_tidptr; 487 else 488 em->child_clear_tid = NULL; 489 490 EMUL_UNLOCK(&emul_lock); 491 492 PROC_LOCK(p2); 493 p2->p_sigparent = exit_signal; 494 PROC_UNLOCK(p2); 495 td2 = FIRST_THREAD_IN_PROC(p2); 496 /* 497 * in a case of stack = NULL we are supposed to COW calling process stack 498 * this is what normal fork() does so we just keep the tf_esp arg intact 499 */ 500 if (args->stack) 501 td2->td_frame->tf_esp = (unsigned int)args->stack; 502 503 if (args->flags & CLONE_SETTLS) { 504 struct l_user_desc info; 505 int idx; 506 int a[2]; 507 struct segment_descriptor sd; 508 509 error = copyin((void *)td->td_frame->tf_esi, &info, sizeof(struct l_user_desc)); 510 if (error) 511 return (error); 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 return (EINVAL); 521 522 /* this doesnt happen in practice */ 523 if (idx == 6) { 524 /* we might copy out the entry_number as 3 */ 525 info.entry_number = 3; 526 error = copyout(&info, (void *) td->td_frame->tf_esi, sizeof(struct l_user_desc)); 527 if (error) 528 return (error); 529 } 530 531 a[0] = LDT_entry_a(&info); 532 a[1] = LDT_entry_b(&info); 533 534 memcpy(&sd, &a, sizeof(a)); 535#ifdef DEBUG 536 if (ldebug(clone)) 537 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, 538 sd.sd_hibase, 539 sd.sd_lolimit, 540 sd.sd_hilimit, 541 sd.sd_type, 542 sd.sd_dpl, 543 sd.sd_p, 544 sd.sd_xx, 545 sd.sd_def32, 546 sd.sd_gran); 547#endif 548 549 /* set %gs */ 550 td2->td_pcb->pcb_gsd = sd; 551 td2->td_pcb->pcb_gs = GSEL(GUGS_SEL, SEL_UPL); 552 } 553 554#ifdef DEBUG 555 if (ldebug(clone)) 556 printf(LMSG("clone: successful rfork to %ld, stack %p sig = %d"), 557 (long)p2->p_pid, args->stack, exit_signal); 558#endif 559 if (args->flags & CLONE_VFORK) { 560 PROC_LOCK(p2); 561 p2->p_flag |= P_PPWAIT; 562 PROC_UNLOCK(p2); 563 } 564 565 /* 566 * Make this runnable after we are finished with it. 567 */ 568 mtx_lock_spin(&sched_lock); 569 TD_SET_CAN_RUN(td2); 570 sched_add(td2, SRQ_BORING); 571 mtx_unlock_spin(&sched_lock); 572 573 td->td_retval[0] = p2->p_pid; 574 td->td_retval[1] = 0; 575 576 if (args->flags & CLONE_VFORK) { 577 /* wait for the children to exit, ie. emulate vfork */ 578 PROC_LOCK(p2); 579 while (p2->p_flag & P_PPWAIT) 580 msleep(td->td_proc, &p2->p_mtx, PWAIT, "ppwait", 0); 581 PROC_UNLOCK(p2); 582 } 583 584 return (0); 585} 586 587#define STACK_SIZE (2 * 1024 * 1024) 588#define GUARD_SIZE (4 * PAGE_SIZE) 589 590static int linux_mmap_common(struct thread *, struct l_mmap_argv *); 591 592int 593linux_mmap2(struct thread *td, struct linux_mmap2_args *args) 594{ 595 struct l_mmap_argv linux_args; 596 597#ifdef DEBUG 598 if (ldebug(mmap2)) 599 printf(ARGS(mmap2, "%p, %d, %d, 0x%08x, %d, %d"), 600 (void *)args->addr, args->len, args->prot, 601 args->flags, args->fd, args->pgoff); 602#endif 603 604 linux_args.addr = args->addr; 605 linux_args.len = args->len; 606 linux_args.prot = args->prot; 607 linux_args.flags = args->flags; 608 linux_args.fd = args->fd; 609 linux_args.pgoff = args->pgoff * PAGE_SIZE; 610 611 return (linux_mmap_common(td, &linux_args)); 612} 613 614int 615linux_mmap(struct thread *td, struct linux_mmap_args *args) 616{ 617 int error; 618 struct l_mmap_argv linux_args; 619 620 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 621 if (error) 622 return (error); 623 624#ifdef DEBUG 625 if (ldebug(mmap)) 626 printf(ARGS(mmap, "%p, %d, %d, 0x%08x, %d, %d"), 627 (void *)linux_args.addr, linux_args.len, linux_args.prot, 628 linux_args.flags, linux_args.fd, linux_args.pgoff); 629#endif 630 631 return (linux_mmap_common(td, &linux_args)); 632} 633 634static int 635linux_mmap_common(struct thread *td, struct l_mmap_argv *linux_args) 636{ 637 struct proc *p = td->td_proc; 638 struct mmap_args /* { 639 caddr_t addr; 640 size_t len; 641 int prot; 642 int flags; 643 int fd; 644 long pad; 645 off_t pos; 646 } */ bsd_args; 647 int error; 648 struct file *fp; 649 650 error = 0; 651 bsd_args.flags = 0; 652 fp = NULL; 653 654 /* 655 * Linux mmap(2): 656 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE 657 */ 658 if (! ((linux_args->flags & LINUX_MAP_SHARED) ^ 659 (linux_args->flags & LINUX_MAP_PRIVATE))) 660 return (EINVAL); 661 662 if (linux_args->flags & LINUX_MAP_SHARED) 663 bsd_args.flags |= MAP_SHARED; 664 if (linux_args->flags & LINUX_MAP_PRIVATE) 665 bsd_args.flags |= MAP_PRIVATE; 666 if (linux_args->flags & LINUX_MAP_FIXED) 667 bsd_args.flags |= MAP_FIXED; 668 if (linux_args->flags & LINUX_MAP_ANON) 669 bsd_args.flags |= MAP_ANON; 670 else 671 bsd_args.flags |= MAP_NOSYNC; 672 if (linux_args->flags & LINUX_MAP_GROWSDOWN) 673 bsd_args.flags |= MAP_STACK; 674 675 /* 676 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC 677 * on Linux/i386. We do this to ensure maximum compatibility. 678 * Linux/ia64 does the same in i386 emulation mode. 679 */ 680 bsd_args.prot = linux_args->prot; 681 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 682 bsd_args.prot |= PROT_READ | PROT_EXEC; 683 684 if (linux_args->fd != -1) { 685 /* 686 * Linux follows Solaris mmap(2) description: 687 * The file descriptor fildes is opened with 688 * read permission, regardless of the 689 * protection options specified. 690 */ 691 692 if ((error = fget(td, linux_args->fd, &fp)) != 0) 693 return (error); 694 if (fp->f_type != DTYPE_VNODE) { 695 fdrop(fp, td); 696 return (EINVAL); 697 } 698 699 /* Linux mmap() just fails for O_WRONLY files */ 700 if (!(fp->f_flag & FREAD)) { 701 fdrop(fp, td); 702 return (EACCES); 703 } 704 705 fdrop(fp, td); 706 } 707 bsd_args.fd = linux_args->fd; 708 709 if (linux_args->flags & LINUX_MAP_GROWSDOWN) { 710 /* 711 * The linux MAP_GROWSDOWN option does not limit auto 712 * growth of the region. Linux mmap with this option 713 * takes as addr the inital BOS, and as len, the initial 714 * region size. It can then grow down from addr without 715 * limit. However, linux threads has an implicit internal 716 * limit to stack size of STACK_SIZE. Its just not 717 * enforced explicitly in linux. But, here we impose 718 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack 719 * region, since we can do this with our mmap. 720 * 721 * Our mmap with MAP_STACK takes addr as the maximum 722 * downsize limit on BOS, and as len the max size of 723 * the region. It them maps the top SGROWSIZ bytes, 724 * and autgrows the region down, up to the limit 725 * in addr. 726 * 727 * If we don't use the MAP_STACK option, the effect 728 * of this code is to allocate a stack region of a 729 * fixed size of (STACK_SIZE - GUARD_SIZE). 730 */ 731 732 if ((caddr_t)PTRIN(linux_args->addr) + linux_args->len > 733 p->p_vmspace->vm_maxsaddr) { 734 /* 735 * Some linux apps will attempt to mmap 736 * thread stacks near the top of their 737 * address space. If their TOS is greater 738 * than vm_maxsaddr, vm_map_growstack() 739 * will confuse the thread stack with the 740 * process stack and deliver a SEGV if they 741 * attempt to grow the thread stack past their 742 * current stacksize rlimit. To avoid this, 743 * adjust vm_maxsaddr upwards to reflect 744 * the current stacksize rlimit rather 745 * than the maximum possible stacksize. 746 * It would be better to adjust the 747 * mmap'ed region, but some apps do not check 748 * mmap's return value. 749 */ 750 PROC_LOCK(p); 751 p->p_vmspace->vm_maxsaddr = (char *)USRSTACK - 752 lim_cur(p, RLIMIT_STACK); 753 PROC_UNLOCK(p); 754 } 755 756 /* This gives us our maximum stack size */ 757 if (linux_args->len > STACK_SIZE - GUARD_SIZE) 758 bsd_args.len = linux_args->len; 759 else 760 bsd_args.len = STACK_SIZE - GUARD_SIZE; 761 762 /* 763 * This gives us a new BOS. If we're using VM_STACK, then 764 * mmap will just map the top SGROWSIZ bytes, and let 765 * the stack grow down to the limit at BOS. If we're 766 * not using VM_STACK we map the full stack, since we 767 * don't have a way to autogrow it. 768 */ 769 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr) - 770 bsd_args.len; 771 } else { 772 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr); 773 bsd_args.len = linux_args->len; 774 } 775 bsd_args.pos = linux_args->pgoff; 776 bsd_args.pad = 0; 777 778#ifdef DEBUG 779 if (ldebug(mmap)) 780 printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n", 781 __func__, 782 (void *)bsd_args.addr, bsd_args.len, bsd_args.prot, 783 bsd_args.flags, bsd_args.fd, (int)bsd_args.pos); 784#endif 785 error = mmap(td, &bsd_args); 786#ifdef DEBUG 787 if (ldebug(mmap)) 788 printf("-> %s() return: 0x%x (0x%08x)\n", 789 __func__, error, (u_int)td->td_retval[0]); 790#endif 791 return (error); 792} 793 794int 795linux_mprotect(struct thread *td, struct linux_mprotect_args *uap) 796{ 797 struct mprotect_args bsd_args; 798 799 bsd_args.addr = uap->addr; 800 bsd_args.len = uap->len; 801 bsd_args.prot = uap->prot; 802 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 803 bsd_args.prot |= PROT_READ | PROT_EXEC; 804 return (mprotect(td, &bsd_args)); 805} 806 807int 808linux_pipe(struct thread *td, struct linux_pipe_args *args) 809{ 810 int error; 811 int reg_edx; 812 813#ifdef DEBUG 814 if (ldebug(pipe)) 815 printf(ARGS(pipe, "*")); 816#endif 817 818 reg_edx = td->td_retval[1]; 819 error = pipe(td, 0); 820 if (error) { 821 td->td_retval[1] = reg_edx; 822 return (error); 823 } 824 825 error = copyout(td->td_retval, args->pipefds, 2*sizeof(int)); 826 if (error) { 827 td->td_retval[1] = reg_edx; 828 return (error); 829 } 830 831 td->td_retval[1] = reg_edx; 832 td->td_retval[0] = 0; 833 return (0); 834} 835 836int 837linux_ioperm(struct thread *td, struct linux_ioperm_args *args) 838{ 839 int error; 840 struct i386_ioperm_args iia; 841 842 iia.start = args->start; 843 iia.length = args->length; 844 iia.enable = args->enable; 845 mtx_lock(&Giant); 846 error = i386_set_ioperm(td, &iia); 847 mtx_unlock(&Giant); 848 return (error); 849} 850 851int 852linux_iopl(struct thread *td, struct linux_iopl_args *args) 853{ 854 int error; 855 856 if (args->level < 0 || args->level > 3) 857 return (EINVAL); 858 if ((error = priv_check(td, PRIV_IO)) != 0) 859 return (error); 860 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 861 return (error); 862 td->td_frame->tf_eflags = (td->td_frame->tf_eflags & ~PSL_IOPL) | 863 (args->level * (PSL_IOPL / 3)); 864 return (0); 865} 866 867int 868linux_modify_ldt(struct thread *td, struct linux_modify_ldt_args *uap) 869{ 870 int error; 871 struct i386_ldt_args ldt; 872 struct l_descriptor ld; 873 union descriptor desc; 874 875 if (uap->ptr == NULL) 876 return (EINVAL); 877 878 switch (uap->func) { 879 case 0x00: /* read_ldt */ 880 ldt.start = 0; 881 ldt.descs = uap->ptr; 882 ldt.num = uap->bytecount / sizeof(union descriptor); 883 mtx_lock(&Giant); 884 error = i386_get_ldt(td, &ldt); 885 td->td_retval[0] *= sizeof(union descriptor); 886 mtx_unlock(&Giant); 887 break; 888 case 0x01: /* write_ldt */ 889 case 0x11: /* write_ldt */ 890 if (uap->bytecount != sizeof(ld)) 891 return (EINVAL); 892 893 error = copyin(uap->ptr, &ld, sizeof(ld)); 894 if (error) 895 return (error); 896 897 ldt.start = ld.entry_number; 898 ldt.descs = &desc; 899 ldt.num = 1; 900 desc.sd.sd_lolimit = (ld.limit & 0x0000ffff); 901 desc.sd.sd_hilimit = (ld.limit & 0x000f0000) >> 16; 902 desc.sd.sd_lobase = (ld.base_addr & 0x00ffffff); 903 desc.sd.sd_hibase = (ld.base_addr & 0xff000000) >> 24; 904 desc.sd.sd_type = SDT_MEMRO | ((ld.read_exec_only ^ 1) << 1) | 905 (ld.contents << 2); 906 desc.sd.sd_dpl = 3; 907 desc.sd.sd_p = (ld.seg_not_present ^ 1); 908 desc.sd.sd_xx = 0; 909 desc.sd.sd_def32 = ld.seg_32bit; 910 desc.sd.sd_gran = ld.limit_in_pages; 911 mtx_lock(&Giant); 912 error = i386_set_ldt(td, &ldt, &desc); 913 mtx_unlock(&Giant); 914 break; 915 default: 916 error = EINVAL; 917 break; 918 } 919 920 if (error == EOPNOTSUPP) { 921 printf("linux: modify_ldt needs kernel option USER_LDT\n"); 922 error = ENOSYS; 923 } 924 925 return (error); 926} 927 928int 929linux_sigaction(struct thread *td, struct linux_sigaction_args *args) 930{ 931 l_osigaction_t osa; 932 l_sigaction_t act, oact; 933 int error; 934 935#ifdef DEBUG 936 if (ldebug(sigaction)) 937 printf(ARGS(sigaction, "%d, %p, %p"), 938 args->sig, (void *)args->nsa, (void *)args->osa); 939#endif 940 941 if (args->nsa != NULL) { 942 error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); 943 if (error) 944 return (error); 945 act.lsa_handler = osa.lsa_handler; 946 act.lsa_flags = osa.lsa_flags; 947 act.lsa_restorer = osa.lsa_restorer; 948 LINUX_SIGEMPTYSET(act.lsa_mask); 949 act.lsa_mask.__bits[0] = osa.lsa_mask; 950 } 951 952 error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, 953 args->osa ? &oact : NULL); 954 955 if (args->osa != NULL && !error) { 956 osa.lsa_handler = oact.lsa_handler; 957 osa.lsa_flags = oact.lsa_flags; 958 osa.lsa_restorer = oact.lsa_restorer; 959 osa.lsa_mask = oact.lsa_mask.__bits[0]; 960 error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); 961 } 962 963 return (error); 964} 965 966/* 967 * Linux has two extra args, restart and oldmask. We dont use these, 968 * but it seems that "restart" is actually a context pointer that 969 * enables the signal to happen with a different register set. 970 */ 971int 972linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) 973{ 974 sigset_t sigmask; 975 l_sigset_t mask; 976 977#ifdef DEBUG 978 if (ldebug(sigsuspend)) 979 printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask); 980#endif 981 982 LINUX_SIGEMPTYSET(mask); 983 mask.__bits[0] = args->mask; 984 linux_to_bsd_sigset(&mask, &sigmask); 985 return (kern_sigsuspend(td, sigmask)); 986} 987 988int 989linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap) 990{ 991 l_sigset_t lmask; 992 sigset_t sigmask; 993 int error; 994 995#ifdef DEBUG 996 if (ldebug(rt_sigsuspend)) 997 printf(ARGS(rt_sigsuspend, "%p, %d"), 998 (void *)uap->newset, uap->sigsetsize); 999#endif 1000 1001 if (uap->sigsetsize != sizeof(l_sigset_t)) 1002 return (EINVAL); 1003 1004 error = copyin(uap->newset, &lmask, sizeof(l_sigset_t)); 1005 if (error) 1006 return (error); 1007 1008 linux_to_bsd_sigset(&lmask, &sigmask); 1009 return (kern_sigsuspend(td, sigmask)); 1010} 1011 1012int 1013linux_pause(struct thread *td, struct linux_pause_args *args) 1014{ 1015 struct proc *p = td->td_proc; 1016 sigset_t sigmask; 1017 1018#ifdef DEBUG 1019 if (ldebug(pause)) 1020 printf(ARGS(pause, "")); 1021#endif 1022 1023 PROC_LOCK(p); 1024 sigmask = td->td_sigmask; 1025 PROC_UNLOCK(p); 1026 return (kern_sigsuspend(td, sigmask)); 1027} 1028 1029int 1030linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap) 1031{ 1032 stack_t ss, oss; 1033 l_stack_t lss; 1034 int error; 1035 1036#ifdef DEBUG 1037 if (ldebug(sigaltstack)) 1038 printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss); 1039#endif 1040 1041 if (uap->uss != NULL) { 1042 error = copyin(uap->uss, &lss, sizeof(l_stack_t)); 1043 if (error) 1044 return (error); 1045 1046 ss.ss_sp = lss.ss_sp; 1047 ss.ss_size = lss.ss_size; 1048 ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags); 1049 } 1050 error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL, 1051 (uap->uoss != NULL) ? &oss : NULL); 1052 if (!error && uap->uoss != NULL) { 1053 lss.ss_sp = oss.ss_sp; 1054 lss.ss_size = oss.ss_size; 1055 lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags); 1056 error = copyout(&lss, uap->uoss, sizeof(l_stack_t)); 1057 } 1058 1059 return (error); 1060} 1061 1062int 1063linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args) 1064{ 1065 struct ftruncate_args sa; 1066 1067#ifdef DEBUG 1068 if (ldebug(ftruncate64)) 1069 printf(ARGS(ftruncate64, "%u, %jd"), args->fd, 1070 (intmax_t)args->length); 1071#endif 1072 1073 sa.fd = args->fd; 1074 sa.pad = 0; 1075 sa.length = args->length; 1076 return ftruncate(td, &sa); 1077} 1078 1079int 1080linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args) 1081{ 1082 struct l_user_desc info; 1083 int error; 1084 int idx; 1085 int a[2]; 1086 struct segment_descriptor sd; 1087 1088 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1089 if (error) 1090 return (error); 1091 1092#ifdef DEBUG 1093 if (ldebug(set_thread_area)) 1094 printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, %i, %i, %i\n"), 1095 info.entry_number, 1096 info.base_addr, 1097 info.limit, 1098 info.seg_32bit, 1099 info.contents, 1100 info.read_exec_only, 1101 info.limit_in_pages, 1102 info.seg_not_present, 1103 info.useable); 1104#endif 1105 1106 idx = info.entry_number; 1107 /* 1108 * Semantics of linux version: every thread in the system has array 1109 * of 3 tls descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. This 1110 * syscall loads one of the selected tls decriptors with a value 1111 * and also loads GDT descriptors 6, 7 and 8 with the content of the per-thread 1112 * descriptors. 1113 * 1114 * Semantics of fbsd version: I think we can ignore that linux has 3 per-thread 1115 * descriptors and use just the 1st one. The tls_array[] is used only in 1116 * set/get-thread_area() syscalls and for loading the GDT descriptors. In fbsd 1117 * we use just one GDT descriptor for TLS so we will load just one. 1118 * XXX: this doesnt work when user-space process tries to use more then 1 TLS segment 1119 * comment in the linux sources says wine might do that. 1120 */ 1121 1122 /* 1123 * we support just GLIBC TLS now 1124 * we should let 3 proceed as well because we use this segment so 1125 * if code does two subsequent calls it should succeed 1126 */ 1127 if (idx != 6 && idx != -1 && idx != 3) 1128 return (EINVAL); 1129 1130 /* 1131 * we have to copy out the GDT entry we use 1132 * FreeBSD uses GDT entry #3 for storing %gs so load that 1133 * XXX: what if userspace program doesnt 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 (LDT_empty(&info)) { 1142 a[0] = 0; 1143 a[1] = 0; 1144 } else { 1145 a[0] = LDT_entry_a(&info); 1146 a[1] = 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 = GET_BASE(&desc); 1209 info.limit = GET_LIMIT(&desc); 1210 info.seg_32bit = GET_32BIT(&desc); 1211 info.contents = GET_CONTENTS(&desc); 1212 info.read_exec_only = !GET_WRITABLE(&desc); 1213 info.limit_in_pages = GET_LIMIT_PAGES(&desc); 1214 info.seg_not_present = !GET_PRESENT(&desc); 1215 info.useable = 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 1316