linux_machdep.c revision 166944
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 166944 2007-02-24 16:49:25Z 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/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 doesn't cause any problems 417 * because both of these flags are usually set together. 418 */ 419 if (!(args->flags & (CLONE_FILES | CLONE_FS))) 420 ff |= RFFDG; 421 422 /* 423 * Attempt to detect when linux_clone(2) is used for creating 424 * kernel threads. Unfortunately despite the existence of the 425 * CLONE_THREAD flag, version of linuxthreads package used in 426 * most popular distros as of beginning of 2005 doesn't make 427 * any use of it. Therefore, this detection relies on 428 * empirical observation that linuxthreads sets certain 429 * combination of flags, so that we can make more or less 430 * precise detection and notify the FreeBSD kernel that several 431 * processes are in fact part of the same threading group, so 432 * that special treatment is necessary for signal delivery 433 * between those processes and fd locking. 434 */ 435 if ((args->flags & 0xffffff00) == THREADING_FLAGS) 436 ff |= RFTHREAD; 437 438 if (args->flags & CLONE_PARENT_SETTID) 439 if (args->parent_tidptr == NULL) 440 return (EINVAL); 441 442 error = fork1(td, ff, 0, &p2); 443 if (error) 444 return (error); 445 446 if (args->flags & (CLONE_PARENT|CLONE_THREAD)) { 447 sx_xlock(&proctree_lock); 448 PROC_LOCK(p2); 449 proc_reparent(p2, td->td_proc->p_pptr); 450 PROC_UNLOCK(p2); 451 sx_xunlock(&proctree_lock); 452 } 453 454 /* create the emuldata */ 455 error = linux_proc_init(td, p2->p_pid, args->flags); 456 /* reference it - no need to check this */ 457 em = em_find(p2, EMUL_DOLOCK); 458 KASSERT(em != NULL, ("clone: emuldata not found.\n")); 459 /* and adjust it */ 460 461 if (args->flags & CLONE_THREAD) { 462 /* XXX: linux mangles pgrp and pptr somehow 463 * I think it might be this but I am not sure. 464 */ 465#ifdef notyet 466 PROC_LOCK(p2); 467 p2->p_pgrp = td->td_proc->p_pgrp; 468 PROC_UNLOCK(p2); 469#endif 470 exit_signal = 0; 471 } 472 473 if (args->flags & CLONE_CHILD_SETTID) 474 em->child_set_tid = args->child_tidptr; 475 else 476 em->child_set_tid = NULL; 477 478 if (args->flags & CLONE_CHILD_CLEARTID) 479 em->child_clear_tid = args->child_tidptr; 480 else 481 em->child_clear_tid = NULL; 482 483 EMUL_UNLOCK(&emul_lock); 484 485 if (args->flags & CLONE_PARENT_SETTID) { 486 error = copyout(&p2->p_pid, args->parent_tidptr, sizeof(p2->p_pid)); 487 if (error) 488 printf(LMSG("copyout failed!")); 489 } 490 491 PROC_LOCK(p2); 492 p2->p_sigparent = exit_signal; 493 PROC_UNLOCK(p2); 494 td2 = FIRST_THREAD_IN_PROC(p2); 495 /* 496 * in a case of stack = NULL we are supposed to COW calling process stack 497 * this is what normal fork() does so we just keep the tf_esp arg intact 498 */ 499 if (args->stack) 500 td2->td_frame->tf_esp = (unsigned int)args->stack; 501 502 if (args->flags & CLONE_SETTLS) { 503 struct l_user_desc info; 504 int idx; 505 int a[2]; 506 struct segment_descriptor sd; 507 508 error = copyin((void *)td->td_frame->tf_esi, &info, sizeof(struct l_user_desc)); 509 if (error) { 510 printf(LMSG("copyin failed!")); 511 } else { 512 513 idx = info.entry_number; 514 515 /* 516 * looks like we're getting the idx we returned 517 * in the set_thread_area() syscall 518 */ 519 if (idx != 6 && idx != 3) { 520 printf(LMSG("resetting idx!")); 521 idx = 3; 522 } 523 524 /* this doesnt happen in practice */ 525 if (idx == 6) { 526 /* we might copy out the entry_number as 3 */ 527 info.entry_number = 3; 528 error = copyout(&info, (void *) td->td_frame->tf_esi, sizeof(struct l_user_desc)); 529 if (error) 530 printf(LMSG("copyout failed!")); 531 } 532 533 a[0] = LDT_entry_a(&info); 534 a[1] = LDT_entry_b(&info); 535 536 memcpy(&sd, &a, sizeof(a)); 537#ifdef DEBUG 538 if (ldebug(clone)) 539 printf("Segment created in clone with CLONE_SETTLS: lobase: %x, hibase: %x, lolimit: %x, hilimit: %x, type: %i, dpl: %i, p: %i, xx: %i, def32: %i, gran: %i\n", sd.sd_lobase, 540 sd.sd_hibase, 541 sd.sd_lolimit, 542 sd.sd_hilimit, 543 sd.sd_type, 544 sd.sd_dpl, 545 sd.sd_p, 546 sd.sd_xx, 547 sd.sd_def32, 548 sd.sd_gran); 549#endif 550 551 /* set %gs */ 552 td2->td_pcb->pcb_gsd = sd; 553 td2->td_pcb->pcb_gs = GSEL(GUGS_SEL, SEL_UPL); 554 } 555 } 556 557#ifdef DEBUG 558 if (ldebug(clone)) 559 printf(LMSG("clone: successful rfork to %ld, stack %p sig = %d"), 560 (long)p2->p_pid, args->stack, exit_signal); 561#endif 562 if (args->flags & CLONE_VFORK) { 563 PROC_LOCK(p2); 564 p2->p_flag |= P_PPWAIT; 565 PROC_UNLOCK(p2); 566 } 567 568 /* 569 * Make this runnable after we are finished with it. 570 */ 571 mtx_lock_spin(&sched_lock); 572 TD_SET_CAN_RUN(td2); 573 sched_add(td2, SRQ_BORING); 574 mtx_unlock_spin(&sched_lock); 575 576 td->td_retval[0] = p2->p_pid; 577 td->td_retval[1] = 0; 578 579 if (args->flags & CLONE_VFORK) { 580 /* wait for the children to exit, ie. emulate vfork */ 581 PROC_LOCK(p2); 582 while (p2->p_flag & P_PPWAIT) 583 msleep(td->td_proc, &p2->p_mtx, PWAIT, "ppwait", 0); 584 PROC_UNLOCK(p2); 585 } 586 587 return (0); 588} 589 590#define STACK_SIZE (2 * 1024 * 1024) 591#define GUARD_SIZE (4 * PAGE_SIZE) 592 593static int linux_mmap_common(struct thread *, struct l_mmap_argv *); 594 595int 596linux_mmap2(struct thread *td, struct linux_mmap2_args *args) 597{ 598 struct l_mmap_argv linux_args; 599 600#ifdef DEBUG 601 if (ldebug(mmap2)) 602 printf(ARGS(mmap2, "%p, %d, %d, 0x%08x, %d, %d"), 603 (void *)args->addr, args->len, args->prot, 604 args->flags, args->fd, args->pgoff); 605#endif 606 607 linux_args.addr = args->addr; 608 linux_args.len = args->len; 609 linux_args.prot = args->prot; 610 linux_args.flags = args->flags; 611 linux_args.fd = args->fd; 612 linux_args.pgoff = args->pgoff * PAGE_SIZE; 613 614 return (linux_mmap_common(td, &linux_args)); 615} 616 617int 618linux_mmap(struct thread *td, struct linux_mmap_args *args) 619{ 620 int error; 621 struct l_mmap_argv linux_args; 622 623 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 624 if (error) 625 return (error); 626 627#ifdef DEBUG 628 if (ldebug(mmap)) 629 printf(ARGS(mmap, "%p, %d, %d, 0x%08x, %d, %d"), 630 (void *)linux_args.addr, linux_args.len, linux_args.prot, 631 linux_args.flags, linux_args.fd, linux_args.pgoff); 632#endif 633 634 return (linux_mmap_common(td, &linux_args)); 635} 636 637static int 638linux_mmap_common(struct thread *td, struct l_mmap_argv *linux_args) 639{ 640 struct proc *p = td->td_proc; 641 struct mmap_args /* { 642 caddr_t addr; 643 size_t len; 644 int prot; 645 int flags; 646 int fd; 647 long pad; 648 off_t pos; 649 } */ bsd_args; 650 int error; 651 struct file *fp; 652 653 error = 0; 654 bsd_args.flags = 0; 655 fp = NULL; 656 657 /* 658 * Linux mmap(2): 659 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE 660 */ 661 if (! ((linux_args->flags & LINUX_MAP_SHARED) ^ 662 (linux_args->flags & LINUX_MAP_PRIVATE))) 663 return (EINVAL); 664 665 if (linux_args->flags & LINUX_MAP_SHARED) 666 bsd_args.flags |= MAP_SHARED; 667 if (linux_args->flags & LINUX_MAP_PRIVATE) 668 bsd_args.flags |= MAP_PRIVATE; 669 if (linux_args->flags & LINUX_MAP_FIXED) 670 bsd_args.flags |= MAP_FIXED; 671 if (linux_args->flags & LINUX_MAP_ANON) 672 bsd_args.flags |= MAP_ANON; 673 else 674 bsd_args.flags |= MAP_NOSYNC; 675 if (linux_args->flags & LINUX_MAP_GROWSDOWN) 676 bsd_args.flags |= MAP_STACK; 677 678 /* 679 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC 680 * on Linux/i386. We do this to ensure maximum compatibility. 681 * Linux/ia64 does the same in i386 emulation mode. 682 */ 683 bsd_args.prot = linux_args->prot; 684 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 685 bsd_args.prot |= PROT_READ | PROT_EXEC; 686 687 if (linux_args->fd != -1) { 688 /* 689 * Linux follows Solaris mmap(2) description: 690 * The file descriptor fildes is opened with 691 * read permission, regardless of the 692 * protection options specified. 693 */ 694 695 if ((error = fget(td, linux_args->fd, &fp)) != 0) 696 return (error); 697 if (fp->f_type != DTYPE_VNODE) { 698 fdrop(fp, td); 699 return (EINVAL); 700 } 701 702 /* Linux mmap() just fails for O_WRONLY files */ 703 if (!(fp->f_flag & FREAD)) { 704 fdrop(fp, td); 705 return (EACCES); 706 } 707 708 fdrop(fp, td); 709 } 710 bsd_args.fd = linux_args->fd; 711 712 if (linux_args->flags & LINUX_MAP_GROWSDOWN) { 713 /* 714 * The linux MAP_GROWSDOWN option does not limit auto 715 * growth of the region. Linux mmap with this option 716 * takes as addr the inital BOS, and as len, the initial 717 * region size. It can then grow down from addr without 718 * limit. However, linux threads has an implicit internal 719 * limit to stack size of STACK_SIZE. Its just not 720 * enforced explicitly in linux. But, here we impose 721 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack 722 * region, since we can do this with our mmap. 723 * 724 * Our mmap with MAP_STACK takes addr as the maximum 725 * downsize limit on BOS, and as len the max size of 726 * the region. It them maps the top SGROWSIZ bytes, 727 * and auto grows the region down, up to the limit 728 * in addr. 729 * 730 * If we don't use the MAP_STACK option, the effect 731 * of this code is to allocate a stack region of a 732 * fixed size of (STACK_SIZE - GUARD_SIZE). 733 */ 734 735 if ((caddr_t)PTRIN(linux_args->addr) + linux_args->len > 736 p->p_vmspace->vm_maxsaddr) { 737 /* 738 * Some linux apps will attempt to mmap 739 * thread stacks near the top of their 740 * address space. If their TOS is greater 741 * than vm_maxsaddr, vm_map_growstack() 742 * will confuse the thread stack with the 743 * process stack and deliver a SEGV if they 744 * attempt to grow the thread stack past their 745 * current stacksize rlimit. To avoid this, 746 * adjust vm_maxsaddr upwards to reflect 747 * the current stacksize rlimit rather 748 * than the maximum possible stacksize. 749 * It would be better to adjust the 750 * mmap'ed region, but some apps do not check 751 * mmap's return value. 752 */ 753 PROC_LOCK(p); 754 p->p_vmspace->vm_maxsaddr = (char *)USRSTACK - 755 lim_cur(p, RLIMIT_STACK); 756 PROC_UNLOCK(p); 757 } 758 759 /* This gives us our maximum stack size */ 760 if (linux_args->len > STACK_SIZE - GUARD_SIZE) 761 bsd_args.len = linux_args->len; 762 else 763 bsd_args.len = STACK_SIZE - GUARD_SIZE; 764 765 /* 766 * This gives us a new BOS. If we're using VM_STACK, then 767 * mmap will just map the top SGROWSIZ bytes, and let 768 * the stack grow down to the limit at BOS. If we're 769 * not using VM_STACK we map the full stack, since we 770 * don't have a way to autogrow it. 771 */ 772 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr) - 773 bsd_args.len; 774 } else { 775 bsd_args.addr = (caddr_t)PTRIN(linux_args->addr); 776 bsd_args.len = linux_args->len; 777 } 778 bsd_args.pos = linux_args->pgoff; 779 bsd_args.pad = 0; 780 781#ifdef DEBUG 782 if (ldebug(mmap)) 783 printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n", 784 __func__, 785 (void *)bsd_args.addr, bsd_args.len, bsd_args.prot, 786 bsd_args.flags, bsd_args.fd, (int)bsd_args.pos); 787#endif 788 error = mmap(td, &bsd_args); 789#ifdef DEBUG 790 if (ldebug(mmap)) 791 printf("-> %s() return: 0x%x (0x%08x)\n", 792 __func__, error, (u_int)td->td_retval[0]); 793#endif 794 return (error); 795} 796 797int 798linux_mprotect(struct thread *td, struct linux_mprotect_args *uap) 799{ 800 struct mprotect_args bsd_args; 801 802 bsd_args.addr = uap->addr; 803 bsd_args.len = uap->len; 804 bsd_args.prot = uap->prot; 805 if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) 806 bsd_args.prot |= PROT_READ | PROT_EXEC; 807 return (mprotect(td, &bsd_args)); 808} 809 810int 811linux_pipe(struct thread *td, struct linux_pipe_args *args) 812{ 813 int error; 814 int reg_edx; 815 816#ifdef DEBUG 817 if (ldebug(pipe)) 818 printf(ARGS(pipe, "*")); 819#endif 820 821 reg_edx = td->td_retval[1]; 822 error = pipe(td, 0); 823 if (error) { 824 td->td_retval[1] = reg_edx; 825 return (error); 826 } 827 828 error = copyout(td->td_retval, args->pipefds, 2*sizeof(int)); 829 if (error) { 830 td->td_retval[1] = reg_edx; 831 return (error); 832 } 833 834 td->td_retval[1] = reg_edx; 835 td->td_retval[0] = 0; 836 return (0); 837} 838 839int 840linux_ioperm(struct thread *td, struct linux_ioperm_args *args) 841{ 842 int error; 843 struct i386_ioperm_args iia; 844 845 iia.start = args->start; 846 iia.length = args->length; 847 iia.enable = args->enable; 848 mtx_lock(&Giant); 849 error = i386_set_ioperm(td, &iia); 850 mtx_unlock(&Giant); 851 return (error); 852} 853 854int 855linux_iopl(struct thread *td, struct linux_iopl_args *args) 856{ 857 int error; 858 859 if (args->level < 0 || args->level > 3) 860 return (EINVAL); 861 if ((error = priv_check(td, PRIV_IO)) != 0) 862 return (error); 863 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 864 return (error); 865 td->td_frame->tf_eflags = (td->td_frame->tf_eflags & ~PSL_IOPL) | 866 (args->level * (PSL_IOPL / 3)); 867 return (0); 868} 869 870int 871linux_modify_ldt(struct thread *td, struct linux_modify_ldt_args *uap) 872{ 873 int error; 874 struct i386_ldt_args ldt; 875 struct l_descriptor ld; 876 union descriptor desc; 877 878 if (uap->ptr == NULL) 879 return (EINVAL); 880 881 switch (uap->func) { 882 case 0x00: /* read_ldt */ 883 ldt.start = 0; 884 ldt.descs = uap->ptr; 885 ldt.num = uap->bytecount / sizeof(union descriptor); 886 mtx_lock(&Giant); 887 error = i386_get_ldt(td, &ldt); 888 td->td_retval[0] *= sizeof(union descriptor); 889 mtx_unlock(&Giant); 890 break; 891 case 0x01: /* write_ldt */ 892 case 0x11: /* write_ldt */ 893 if (uap->bytecount != sizeof(ld)) 894 return (EINVAL); 895 896 error = copyin(uap->ptr, &ld, sizeof(ld)); 897 if (error) 898 return (error); 899 900 ldt.start = ld.entry_number; 901 ldt.descs = &desc; 902 ldt.num = 1; 903 desc.sd.sd_lolimit = (ld.limit & 0x0000ffff); 904 desc.sd.sd_hilimit = (ld.limit & 0x000f0000) >> 16; 905 desc.sd.sd_lobase = (ld.base_addr & 0x00ffffff); 906 desc.sd.sd_hibase = (ld.base_addr & 0xff000000) >> 24; 907 desc.sd.sd_type = SDT_MEMRO | ((ld.read_exec_only ^ 1) << 1) | 908 (ld.contents << 2); 909 desc.sd.sd_dpl = 3; 910 desc.sd.sd_p = (ld.seg_not_present ^ 1); 911 desc.sd.sd_xx = 0; 912 desc.sd.sd_def32 = ld.seg_32bit; 913 desc.sd.sd_gran = ld.limit_in_pages; 914 mtx_lock(&Giant); 915 error = i386_set_ldt(td, &ldt, &desc); 916 mtx_unlock(&Giant); 917 break; 918 default: 919 error = EINVAL; 920 break; 921 } 922 923 if (error == EOPNOTSUPP) { 924 printf("linux: modify_ldt needs kernel option USER_LDT\n"); 925 error = ENOSYS; 926 } 927 928 return (error); 929} 930 931int 932linux_sigaction(struct thread *td, struct linux_sigaction_args *args) 933{ 934 l_osigaction_t osa; 935 l_sigaction_t act, oact; 936 int error; 937 938#ifdef DEBUG 939 if (ldebug(sigaction)) 940 printf(ARGS(sigaction, "%d, %p, %p"), 941 args->sig, (void *)args->nsa, (void *)args->osa); 942#endif 943 944 if (args->nsa != NULL) { 945 error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); 946 if (error) 947 return (error); 948 act.lsa_handler = osa.lsa_handler; 949 act.lsa_flags = osa.lsa_flags; 950 act.lsa_restorer = osa.lsa_restorer; 951 LINUX_SIGEMPTYSET(act.lsa_mask); 952 act.lsa_mask.__bits[0] = osa.lsa_mask; 953 } 954 955 error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, 956 args->osa ? &oact : NULL); 957 958 if (args->osa != NULL && !error) { 959 osa.lsa_handler = oact.lsa_handler; 960 osa.lsa_flags = oact.lsa_flags; 961 osa.lsa_restorer = oact.lsa_restorer; 962 osa.lsa_mask = oact.lsa_mask.__bits[0]; 963 error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); 964 } 965 966 return (error); 967} 968 969/* 970 * Linux has two extra args, restart and oldmask. We dont use these, 971 * but it seems that "restart" is actually a context pointer that 972 * enables the signal to happen with a different register set. 973 */ 974int 975linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) 976{ 977 sigset_t sigmask; 978 l_sigset_t mask; 979 980#ifdef DEBUG 981 if (ldebug(sigsuspend)) 982 printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask); 983#endif 984 985 LINUX_SIGEMPTYSET(mask); 986 mask.__bits[0] = args->mask; 987 linux_to_bsd_sigset(&mask, &sigmask); 988 return (kern_sigsuspend(td, sigmask)); 989} 990 991int 992linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap) 993{ 994 l_sigset_t lmask; 995 sigset_t sigmask; 996 int error; 997 998#ifdef DEBUG 999 if (ldebug(rt_sigsuspend)) 1000 printf(ARGS(rt_sigsuspend, "%p, %d"), 1001 (void *)uap->newset, uap->sigsetsize); 1002#endif 1003 1004 if (uap->sigsetsize != sizeof(l_sigset_t)) 1005 return (EINVAL); 1006 1007 error = copyin(uap->newset, &lmask, sizeof(l_sigset_t)); 1008 if (error) 1009 return (error); 1010 1011 linux_to_bsd_sigset(&lmask, &sigmask); 1012 return (kern_sigsuspend(td, sigmask)); 1013} 1014 1015int 1016linux_pause(struct thread *td, struct linux_pause_args *args) 1017{ 1018 struct proc *p = td->td_proc; 1019 sigset_t sigmask; 1020 1021#ifdef DEBUG 1022 if (ldebug(pause)) 1023 printf(ARGS(pause, "")); 1024#endif 1025 1026 PROC_LOCK(p); 1027 sigmask = td->td_sigmask; 1028 PROC_UNLOCK(p); 1029 return (kern_sigsuspend(td, sigmask)); 1030} 1031 1032int 1033linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap) 1034{ 1035 stack_t ss, oss; 1036 l_stack_t lss; 1037 int error; 1038 1039#ifdef DEBUG 1040 if (ldebug(sigaltstack)) 1041 printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss); 1042#endif 1043 1044 if (uap->uss != NULL) { 1045 error = copyin(uap->uss, &lss, sizeof(l_stack_t)); 1046 if (error) 1047 return (error); 1048 1049 ss.ss_sp = lss.ss_sp; 1050 ss.ss_size = lss.ss_size; 1051 ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags); 1052 } 1053 error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL, 1054 (uap->uoss != NULL) ? &oss : NULL); 1055 if (!error && uap->uoss != NULL) { 1056 lss.ss_sp = oss.ss_sp; 1057 lss.ss_size = oss.ss_size; 1058 lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags); 1059 error = copyout(&lss, uap->uoss, sizeof(l_stack_t)); 1060 } 1061 1062 return (error); 1063} 1064 1065int 1066linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args) 1067{ 1068 struct ftruncate_args sa; 1069 1070#ifdef DEBUG 1071 if (ldebug(ftruncate64)) 1072 printf(ARGS(ftruncate64, "%u, %jd"), args->fd, 1073 (intmax_t)args->length); 1074#endif 1075 1076 sa.fd = args->fd; 1077 sa.pad = 0; 1078 sa.length = args->length; 1079 return ftruncate(td, &sa); 1080} 1081 1082int 1083linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args) 1084{ 1085 struct l_user_desc info; 1086 int error; 1087 int idx; 1088 int a[2]; 1089 struct segment_descriptor sd; 1090 1091 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1092 if (error) 1093 return (error); 1094 1095#ifdef DEBUG 1096 if (ldebug(set_thread_area)) 1097 printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, %i, %i, %i\n"), 1098 info.entry_number, 1099 info.base_addr, 1100 info.limit, 1101 info.seg_32bit, 1102 info.contents, 1103 info.read_exec_only, 1104 info.limit_in_pages, 1105 info.seg_not_present, 1106 info.useable); 1107#endif 1108 1109 idx = info.entry_number; 1110 /* 1111 * Semantics of linux version: every thread in the system has array of 1112 * 3 tls descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. This 1113 * syscall loads one of the selected tls decriptors with a value and 1114 * also loads GDT descriptors 6, 7 and 8 with the content of the 1115 * per-thread descriptors. 1116 * 1117 * Semantics of fbsd version: I think we can ignore that linux has 3 1118 * per-thread descriptors and use just the 1st one. The tls_array[] 1119 * is used only in set/get-thread_area() syscalls and for loading the 1120 * GDT descriptors. In fbsd we use just one GDT descriptor for TLS so 1121 * we will load just one. 1122 * 1123 * XXX: this doesn't work when a user space process tries to use more 1124 * than 1 TLS segment. Comment in the linux sources says wine might do 1125 * this. 1126 */ 1127 1128 /* 1129 * we support just GLIBC TLS now 1130 * we should let 3 proceed as well because we use this segment so 1131 * if code does two subsequent calls it should succeed 1132 */ 1133 if (idx != 6 && idx != -1 && idx != 3) 1134 return (EINVAL); 1135 1136 /* 1137 * we have to copy out the GDT entry we use 1138 * FreeBSD uses GDT entry #3 for storing %gs so load that 1139 * 1140 * XXX: what if a user space program doesn't check this value and tries 1141 * to use 6, 7 or 8? 1142 */ 1143 idx = info.entry_number = 3; 1144 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 1145 if (error) 1146 return (error); 1147 1148 if (LDT_empty(&info)) { 1149 a[0] = 0; 1150 a[1] = 0; 1151 } else { 1152 a[0] = LDT_entry_a(&info); 1153 a[1] = LDT_entry_b(&info); 1154 } 1155 1156 memcpy(&sd, &a, sizeof(a)); 1157#ifdef DEBUG 1158 if (ldebug(set_thread_area)) 1159 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, 1160 sd.sd_hibase, 1161 sd.sd_lolimit, 1162 sd.sd_hilimit, 1163 sd.sd_type, 1164 sd.sd_dpl, 1165 sd.sd_p, 1166 sd.sd_xx, 1167 sd.sd_def32, 1168 sd.sd_gran); 1169#endif 1170 1171 /* this is taken from i386 version of cpu_set_user_tls() */ 1172 critical_enter(); 1173 /* set %gs */ 1174 td->td_pcb->pcb_gsd = sd; 1175 PCPU_GET(fsgs_gdt)[1] = sd; 1176 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 1177 critical_exit(); 1178 1179 return (0); 1180} 1181 1182int 1183linux_get_thread_area(struct thread *td, struct linux_get_thread_area_args *args) 1184{ 1185 1186 struct l_user_desc info; 1187 int error; 1188 int idx; 1189 struct l_desc_struct desc; 1190 struct segment_descriptor sd; 1191 1192#ifdef DEBUG 1193 if (ldebug(get_thread_area)) 1194 printf(ARGS(get_thread_area, "%p"), args->desc); 1195#endif 1196 1197 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 1198 if (error) 1199 return (error); 1200 1201 idx = info.entry_number; 1202 /* XXX: I am not sure if we want 3 to be allowed too. */ 1203 if (idx != 6 && idx != 3) 1204 return (EINVAL); 1205 1206 idx = 3; 1207 1208 memset(&info, 0, sizeof(info)); 1209 1210 sd = PCPU_GET(fsgs_gdt)[1]; 1211 1212 memcpy(&desc, &sd, sizeof(desc)); 1213 1214 info.entry_number = idx; 1215 info.base_addr = GET_BASE(&desc); 1216 info.limit = GET_LIMIT(&desc); 1217 info.seg_32bit = GET_32BIT(&desc); 1218 info.contents = GET_CONTENTS(&desc); 1219 info.read_exec_only = !GET_WRITABLE(&desc); 1220 info.limit_in_pages = GET_LIMIT_PAGES(&desc); 1221 info.seg_not_present = !GET_PRESENT(&desc); 1222 info.useable = GET_USEABLE(&desc); 1223 1224 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 1225 if (error) 1226 return (EFAULT); 1227 1228 return (0); 1229} 1230 1231/* copied from kern/kern_time.c */ 1232int 1233linux_timer_create(struct thread *td, struct linux_timer_create_args *args) 1234{ 1235 return ktimer_create(td, (struct ktimer_create_args *) args); 1236} 1237 1238int 1239linux_timer_settime(struct thread *td, struct linux_timer_settime_args *args) 1240{ 1241 return ktimer_settime(td, (struct ktimer_settime_args *) args); 1242} 1243 1244int 1245linux_timer_gettime(struct thread *td, struct linux_timer_gettime_args *args) 1246{ 1247 return ktimer_gettime(td, (struct ktimer_gettime_args *) args); 1248} 1249 1250int 1251linux_timer_getoverrun(struct thread *td, struct linux_timer_getoverrun_args *args) 1252{ 1253 return ktimer_getoverrun(td, (struct ktimer_getoverrun_args *) args); 1254} 1255 1256int 1257linux_timer_delete(struct thread *td, struct linux_timer_delete_args *args) 1258{ 1259 return ktimer_delete(td, (struct ktimer_delete_args *) args); 1260} 1261 1262/* XXX: this wont work with module - convert it */ 1263int 1264linux_mq_open(struct thread *td, struct linux_mq_open_args *args) 1265{ 1266#ifdef P1003_1B_MQUEUE 1267 return kmq_open(td, (struct kmq_open_args *) args); 1268#else 1269 return (ENOSYS); 1270#endif 1271} 1272 1273int 1274linux_mq_unlink(struct thread *td, struct linux_mq_unlink_args *args) 1275{ 1276#ifdef P1003_1B_MQUEUE 1277 return kmq_unlink(td, (struct kmq_unlink_args *) args); 1278#else 1279 return (ENOSYS); 1280#endif 1281} 1282 1283int 1284linux_mq_timedsend(struct thread *td, struct linux_mq_timedsend_args *args) 1285{ 1286#ifdef P1003_1B_MQUEUE 1287 return kmq_timedsend(td, (struct kmq_timedsend_args *) args); 1288#else 1289 return (ENOSYS); 1290#endif 1291} 1292 1293int 1294linux_mq_timedreceive(struct thread *td, struct linux_mq_timedreceive_args *args) 1295{ 1296#ifdef P1003_1B_MQUEUE 1297 return kmq_timedreceive(td, (struct kmq_timedreceive_args *) args); 1298#else 1299 return (ENOSYS); 1300#endif 1301} 1302 1303int 1304linux_mq_notify(struct thread *td, struct linux_mq_notify_args *args) 1305{ 1306#ifdef P1003_1B_MQUEUE 1307 return kmq_notify(td, (struct kmq_notify_args *) args); 1308#else 1309 return (ENOSYS); 1310#endif 1311} 1312 1313int 1314linux_mq_getsetattr(struct thread *td, struct linux_mq_getsetattr_args *args) 1315{ 1316#ifdef P1003_1B_MQUEUE 1317 return kmq_setattr(td, (struct kmq_setattr_args *) args); 1318#else 1319 return (ENOSYS); 1320#endif 1321} 1322 1323