1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2000 Marcel Moolenaar 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/capsicum.h> 34 #include <sys/fcntl.h> 35 #include <sys/file.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/priv.h> 42 #include <sys/proc.h> 43 #include <sys/queue.h> 44 #include <sys/resource.h> 45 #include <sys/resourcevar.h> 46 #include <sys/sched.h> 47 #include <sys/signalvar.h> 48 #include <sys/syscallsubr.h> 49 #include <sys/sysproto.h> 50 #include <sys/systm.h> 51 #include <sys/sx.h> 52 #include <sys/unistd.h> 53 #include <sys/wait.h> 54 55 #include <machine/frame.h> 56 #include <machine/psl.h> 57 #include <machine/segments.h> 58 #include <machine/sysarch.h> 59 60 #include <vm/pmap.h> 61 #include <vm/vm.h> 62 #include <vm/vm_map.h> 63 64 #include <security/audit/audit.h> 65 66 #include <i386/linux/linux.h> 67 #include <i386/linux/linux_proto.h> 68 #include <compat/linux/linux_emul.h> 69 #include <compat/linux/linux_ipc.h> 70 #include <compat/linux/linux_misc.h> 71 #include <compat/linux/linux_mmap.h> 72 #include <compat/linux/linux_signal.h> 73 #include <compat/linux/linux_util.h> 74 75 #include <i386/include/pcb.h> /* needed for pcb definition in linux_set_thread_area */ 76 77 #include "opt_posix.h" 78 79 struct l_descriptor { 80 l_uint entry_number; 81 l_ulong base_addr; 82 l_uint limit; 83 l_uint seg_32bit:1; 84 l_uint contents:2; 85 l_uint read_exec_only:1; 86 l_uint limit_in_pages:1; 87 l_uint seg_not_present:1; 88 l_uint useable:1; 89 }; 90 91 struct l_old_select_argv { 92 l_int nfds; 93 l_fd_set *readfds; 94 l_fd_set *writefds; 95 l_fd_set *exceptfds; 96 struct l_timeval *timeout; 97 }; 98 99 int 100 linux_execve(struct thread *td, struct linux_execve_args *args) 101 { 102 struct image_args eargs; 103 char *newpath; 104 int error; 105 106 if (!LUSECONVPATH(td)) { 107 error = exec_copyin_args(&eargs, args->path, UIO_USERSPACE, 108 args->argp, args->envp); 109 } else { 110 LCONVPATHEXIST(td, args->path, &newpath); 111 error = exec_copyin_args(&eargs, newpath, UIO_SYSSPACE, 112 args->argp, args->envp); 113 LFREEPATH(newpath); 114 } 115 if (error == 0) 116 error = linux_common_execve(td, &eargs); 117 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 118 return (error); 119 } 120 121 struct l_ipc_kludge { 122 struct l_msgbuf *msgp; 123 l_long msgtyp; 124 }; 125 126 int 127 linux_ipc(struct thread *td, struct linux_ipc_args *args) 128 { 129 130 switch (args->what & 0xFFFF) { 131 case LINUX_SEMOP: { 132 struct linux_semop_args a; 133 134 a.semid = args->arg1; 135 a.tsops = PTRIN(args->ptr); 136 a.nsops = args->arg2; 137 return (linux_semop(td, &a)); 138 } 139 case LINUX_SEMGET: { 140 struct linux_semget_args a; 141 142 a.key = args->arg1; 143 a.nsems = args->arg2; 144 a.semflg = args->arg3; 145 return (linux_semget(td, &a)); 146 } 147 case LINUX_SEMCTL: { 148 struct linux_semctl_args a; 149 int error; 150 151 a.semid = args->arg1; 152 a.semnum = args->arg2; 153 a.cmd = args->arg3; 154 error = copyin(PTRIN(args->ptr), &a.arg, sizeof(a.arg)); 155 if (error) 156 return (error); 157 return (linux_semctl(td, &a)); 158 } 159 case LINUX_MSGSND: { 160 struct linux_msgsnd_args a; 161 162 a.msqid = args->arg1; 163 a.msgp = PTRIN(args->ptr); 164 a.msgsz = args->arg2; 165 a.msgflg = args->arg3; 166 return (linux_msgsnd(td, &a)); 167 } 168 case LINUX_MSGRCV: { 169 struct linux_msgrcv_args a; 170 171 a.msqid = args->arg1; 172 a.msgsz = args->arg2; 173 a.msgflg = args->arg3; 174 if ((args->what >> 16) == 0) { 175 struct l_ipc_kludge tmp; 176 int error; 177 178 if (args->ptr == 0) 179 return (EINVAL); 180 error = copyin(PTRIN(args->ptr), &tmp, sizeof(tmp)); 181 if (error) 182 return (error); 183 a.msgp = PTRIN(tmp.msgp); 184 a.msgtyp = tmp.msgtyp; 185 } else { 186 a.msgp = PTRIN(args->ptr); 187 a.msgtyp = args->arg5; 188 } 189 return (linux_msgrcv(td, &a)); 190 } 191 case LINUX_MSGGET: { 192 struct linux_msgget_args a; 193 194 a.key = args->arg1; 195 a.msgflg = args->arg2; 196 return (linux_msgget(td, &a)); 197 } 198 case LINUX_MSGCTL: { 199 struct linux_msgctl_args a; 200 201 a.msqid = args->arg1; 202 a.cmd = args->arg2; 203 a.buf = PTRIN(args->ptr); 204 return (linux_msgctl(td, &a)); 205 } 206 case LINUX_SHMAT: { 207 struct linux_shmat_args a; 208 l_uintptr_t addr; 209 int error; 210 211 a.shmid = args->arg1; 212 a.shmaddr = PTRIN(args->ptr); 213 a.shmflg = args->arg2; 214 error = linux_shmat(td, &a); 215 if (error != 0) 216 return (error); 217 addr = td->td_retval[0]; 218 error = copyout(&addr, PTRIN(args->arg3), sizeof(addr)); 219 td->td_retval[0] = 0; 220 return (error); 221 } 222 case LINUX_SHMDT: { 223 struct linux_shmdt_args a; 224 225 a.shmaddr = PTRIN(args->ptr); 226 return (linux_shmdt(td, &a)); 227 } 228 case LINUX_SHMGET: { 229 struct linux_shmget_args a; 230 231 a.key = args->arg1; 232 a.size = args->arg2; 233 a.shmflg = args->arg3; 234 return (linux_shmget(td, &a)); 235 } 236 case LINUX_SHMCTL: { 237 struct linux_shmctl_args a; 238 239 a.shmid = args->arg1; 240 a.cmd = args->arg2; 241 a.buf = PTRIN(args->ptr); 242 return (linux_shmctl(td, &a)); 243 } 244 default: 245 break; 246 } 247 248 return (EINVAL); 249 } 250 251 int 252 linux_old_select(struct thread *td, struct linux_old_select_args *args) 253 { 254 struct l_old_select_argv linux_args; 255 struct linux_select_args newsel; 256 int error; 257 258 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 259 if (error) 260 return (error); 261 262 newsel.nfds = linux_args.nfds; 263 newsel.readfds = linux_args.readfds; 264 newsel.writefds = linux_args.writefds; 265 newsel.exceptfds = linux_args.exceptfds; 266 newsel.timeout = linux_args.timeout; 267 return (linux_select(td, &newsel)); 268 } 269 270 int 271 linux_set_cloned_tls(struct thread *td, void *desc) 272 { 273 struct segment_descriptor sd; 274 struct l_user_desc info; 275 int idx, error; 276 int a[2]; 277 278 error = copyin(desc, &info, sizeof(struct l_user_desc)); 279 if (error) { 280 linux_msg(td, "set_cloned_tls copyin failed!"); 281 } else { 282 idx = info.entry_number; 283 284 /* 285 * looks like we're getting the idx we returned 286 * in the set_thread_area() syscall 287 */ 288 if (idx != 6 && idx != 3) { 289 linux_msg(td, "set_cloned_tls resetting idx!"); 290 idx = 3; 291 } 292 293 /* this doesnt happen in practice */ 294 if (idx == 6) { 295 /* we might copy out the entry_number as 3 */ 296 info.entry_number = 3; 297 error = copyout(&info, desc, sizeof(struct l_user_desc)); 298 if (error) 299 linux_msg(td, "set_cloned_tls copyout failed!"); 300 } 301 302 a[0] = LINUX_LDT_entry_a(&info); 303 a[1] = LINUX_LDT_entry_b(&info); 304 305 memcpy(&sd, &a, sizeof(a)); 306 /* set %gs */ 307 td->td_pcb->pcb_gsd = sd; 308 td->td_pcb->pcb_gs = GSEL(GUGS_SEL, SEL_UPL); 309 } 310 311 return (error); 312 } 313 314 int 315 linux_set_upcall(struct thread *td, register_t stack) 316 { 317 318 if (stack) 319 td->td_frame->tf_esp = stack; 320 321 /* 322 * The newly created Linux thread returns 323 * to the user space by the same path that a parent do. 324 */ 325 td->td_frame->tf_eax = 0; 326 return (0); 327 } 328 329 int 330 linux_mmap2(struct thread *td, struct linux_mmap2_args *args) 331 { 332 333 return (linux_mmap_common(td, args->addr, args->len, args->prot, 334 args->flags, args->fd, (uint64_t)(uint32_t)args->pgoff * 335 PAGE_SIZE)); 336 } 337 338 int 339 linux_mmap(struct thread *td, struct linux_mmap_args *args) 340 { 341 int error; 342 struct l_mmap_argv linux_args; 343 344 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 345 if (error) 346 return (error); 347 348 return (linux_mmap_common(td, linux_args.addr, linux_args.len, 349 linux_args.prot, linux_args.flags, linux_args.fd, 350 (uint32_t)linux_args.pgoff)); 351 } 352 353 int 354 linux_mprotect(struct thread *td, struct linux_mprotect_args *uap) 355 { 356 357 return (linux_mprotect_common(td, PTROUT(uap->addr), uap->len, uap->prot)); 358 } 359 360 int 361 linux_madvise(struct thread *td, struct linux_madvise_args *uap) 362 { 363 364 return (linux_madvise_common(td, PTROUT(uap->addr), uap->len, uap->behav)); 365 } 366 367 int 368 linux_ioperm(struct thread *td, struct linux_ioperm_args *args) 369 { 370 int error; 371 struct i386_ioperm_args iia; 372 373 iia.start = args->start; 374 iia.length = args->length; 375 iia.enable = args->enable; 376 error = i386_set_ioperm(td, &iia); 377 return (error); 378 } 379 380 int 381 linux_iopl(struct thread *td, struct linux_iopl_args *args) 382 { 383 int error; 384 385 if (args->level < 0 || args->level > 3) 386 return (EINVAL); 387 if ((error = priv_check(td, PRIV_IO)) != 0) 388 return (error); 389 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 390 return (error); 391 td->td_frame->tf_eflags = (td->td_frame->tf_eflags & ~PSL_IOPL) | 392 (args->level * (PSL_IOPL / 3)); 393 return (0); 394 } 395 396 int 397 linux_modify_ldt(struct thread *td, struct linux_modify_ldt_args *uap) 398 { 399 int error; 400 struct i386_ldt_args ldt; 401 struct l_descriptor ld; 402 union descriptor desc; 403 int size, written; 404 405 switch (uap->func) { 406 case 0x00: /* read_ldt */ 407 ldt.start = 0; 408 ldt.descs = uap->ptr; 409 ldt.num = uap->bytecount / sizeof(union descriptor); 410 error = i386_get_ldt(td, &ldt); 411 td->td_retval[0] *= sizeof(union descriptor); 412 break; 413 case 0x02: /* read_default_ldt = 0 */ 414 size = 5*sizeof(struct l_desc_struct); 415 if (size > uap->bytecount) 416 size = uap->bytecount; 417 for (written = error = 0; written < size && error == 0; written++) 418 error = subyte((char *)uap->ptr + written, 0); 419 td->td_retval[0] = written; 420 break; 421 case 0x01: /* write_ldt */ 422 case 0x11: /* write_ldt */ 423 if (uap->bytecount != sizeof(ld)) 424 return (EINVAL); 425 426 error = copyin(uap->ptr, &ld, sizeof(ld)); 427 if (error) 428 return (error); 429 430 ldt.start = ld.entry_number; 431 ldt.descs = &desc; 432 ldt.num = 1; 433 desc.sd.sd_lolimit = (ld.limit & 0x0000ffff); 434 desc.sd.sd_hilimit = (ld.limit & 0x000f0000) >> 16; 435 desc.sd.sd_lobase = (ld.base_addr & 0x00ffffff); 436 desc.sd.sd_hibase = (ld.base_addr & 0xff000000) >> 24; 437 desc.sd.sd_type = SDT_MEMRO | ((ld.read_exec_only ^ 1) << 1) | 438 (ld.contents << 2); 439 desc.sd.sd_dpl = 3; 440 desc.sd.sd_p = (ld.seg_not_present ^ 1); 441 desc.sd.sd_xx = 0; 442 desc.sd.sd_def32 = ld.seg_32bit; 443 desc.sd.sd_gran = ld.limit_in_pages; 444 error = i386_set_ldt(td, &ldt, &desc); 445 break; 446 default: 447 error = ENOSYS; 448 break; 449 } 450 451 if (error == EOPNOTSUPP) { 452 linux_msg(td, "modify_ldt needs kernel option USER_LDT"); 453 error = ENOSYS; 454 } 455 456 return (error); 457 } 458 459 int 460 linux_sigaction(struct thread *td, struct linux_sigaction_args *args) 461 { 462 l_osigaction_t osa; 463 l_sigaction_t act, oact; 464 int error; 465 466 if (args->nsa != NULL) { 467 error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); 468 if (error) 469 return (error); 470 act.lsa_handler = osa.lsa_handler; 471 act.lsa_flags = osa.lsa_flags; 472 act.lsa_restorer = osa.lsa_restorer; 473 LINUX_SIGEMPTYSET(act.lsa_mask); 474 act.lsa_mask.__mask = osa.lsa_mask; 475 } 476 477 error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, 478 args->osa ? &oact : NULL); 479 480 if (args->osa != NULL && !error) { 481 osa.lsa_handler = oact.lsa_handler; 482 osa.lsa_flags = oact.lsa_flags; 483 osa.lsa_restorer = oact.lsa_restorer; 484 osa.lsa_mask = oact.lsa_mask.__mask; 485 error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); 486 } 487 488 return (error); 489 } 490 491 /* 492 * Linux has two extra args, restart and oldmask. We dont use these, 493 * but it seems that "restart" is actually a context pointer that 494 * enables the signal to happen with a different register set. 495 */ 496 int 497 linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) 498 { 499 sigset_t sigmask; 500 l_sigset_t mask; 501 502 LINUX_SIGEMPTYSET(mask); 503 mask.__mask = args->mask; 504 linux_to_bsd_sigset(&mask, &sigmask); 505 return (kern_sigsuspend(td, sigmask)); 506 } 507 508 int 509 linux_pause(struct thread *td, struct linux_pause_args *args) 510 { 511 struct proc *p = td->td_proc; 512 sigset_t sigmask; 513 514 PROC_LOCK(p); 515 sigmask = td->td_sigmask; 516 PROC_UNLOCK(p); 517 return (kern_sigsuspend(td, sigmask)); 518 } 519 520 int 521 linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args) 522 { 523 struct l_user_desc info; 524 int error; 525 int idx; 526 int a[2]; 527 struct segment_descriptor sd; 528 529 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 530 if (error) 531 return (error); 532 533 idx = info.entry_number; 534 /* 535 * Semantics of Linux version: every thread in the system has array of 536 * 3 tls descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. This 537 * syscall loads one of the selected tls decriptors with a value and 538 * also loads GDT descriptors 6, 7 and 8 with the content of the 539 * per-thread descriptors. 540 * 541 * Semantics of FreeBSD version: I think we can ignore that Linux has 3 542 * per-thread descriptors and use just the 1st one. The tls_array[] 543 * is used only in set/get-thread_area() syscalls and for loading the 544 * GDT descriptors. In FreeBSD we use just one GDT descriptor for TLS 545 * so we will load just one. 546 * 547 * XXX: this doesn't work when a user space process tries to use more 548 * than 1 TLS segment. Comment in the Linux sources says wine might do 549 * this. 550 */ 551 552 /* 553 * we support just GLIBC TLS now 554 * we should let 3 proceed as well because we use this segment so 555 * if code does two subsequent calls it should succeed 556 */ 557 if (idx != 6 && idx != -1 && idx != 3) 558 return (EINVAL); 559 560 /* 561 * we have to copy out the GDT entry we use 562 * FreeBSD uses GDT entry #3 for storing %gs so load that 563 * 564 * XXX: what if a user space program doesn't check this value and tries 565 * to use 6, 7 or 8? 566 */ 567 idx = info.entry_number = 3; 568 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 569 if (error) 570 return (error); 571 572 if (LINUX_LDT_empty(&info)) { 573 a[0] = 0; 574 a[1] = 0; 575 } else { 576 a[0] = LINUX_LDT_entry_a(&info); 577 a[1] = LINUX_LDT_entry_b(&info); 578 } 579 580 memcpy(&sd, &a, sizeof(a)); 581 /* this is taken from i386 version of cpu_set_user_tls() */ 582 critical_enter(); 583 /* set %gs */ 584 td->td_pcb->pcb_gsd = sd; 585 PCPU_GET(fsgs_gdt)[1] = sd; 586 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 587 critical_exit(); 588 589 return (0); 590 } 591 592 int 593 linux_get_thread_area(struct thread *td, struct linux_get_thread_area_args *args) 594 { 595 596 struct l_user_desc info; 597 int error; 598 int idx; 599 struct l_desc_struct desc; 600 struct segment_descriptor sd; 601 602 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 603 if (error) 604 return (error); 605 606 idx = info.entry_number; 607 /* XXX: I am not sure if we want 3 to be allowed too. */ 608 if (idx != 6 && idx != 3) 609 return (EINVAL); 610 611 idx = 3; 612 613 memset(&info, 0, sizeof(info)); 614 615 sd = PCPU_GET(fsgs_gdt)[1]; 616 617 memcpy(&desc, &sd, sizeof(desc)); 618 619 info.entry_number = idx; 620 info.base_addr = LINUX_GET_BASE(&desc); 621 info.limit = LINUX_GET_LIMIT(&desc); 622 info.seg_32bit = LINUX_GET_32BIT(&desc); 623 info.contents = LINUX_GET_CONTENTS(&desc); 624 info.read_exec_only = !LINUX_GET_WRITABLE(&desc); 625 info.limit_in_pages = LINUX_GET_LIMIT_PAGES(&desc); 626 info.seg_not_present = !LINUX_GET_PRESENT(&desc); 627 info.useable = LINUX_GET_USEABLE(&desc); 628 629 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 630 if (error) 631 return (EFAULT); 632 633 return (0); 634 } 635 636 /* XXX: this wont work with module - convert it */ 637 int 638 linux_mq_open(struct thread *td, struct linux_mq_open_args *args) 639 { 640 #ifdef P1003_1B_MQUEUE 641 return (sys_kmq_open(td, (struct kmq_open_args *)args)); 642 #else 643 return (ENOSYS); 644 #endif 645 } 646 647 int 648 linux_mq_unlink(struct thread *td, struct linux_mq_unlink_args *args) 649 { 650 #ifdef P1003_1B_MQUEUE 651 return (sys_kmq_unlink(td, (struct kmq_unlink_args *)args)); 652 #else 653 return (ENOSYS); 654 #endif 655 } 656 657 int 658 linux_mq_timedsend(struct thread *td, struct linux_mq_timedsend_args *args) 659 { 660 #ifdef P1003_1B_MQUEUE 661 return (sys_kmq_timedsend(td, (struct kmq_timedsend_args *)args)); 662 #else 663 return (ENOSYS); 664 #endif 665 } 666 667 int 668 linux_mq_timedreceive(struct thread *td, struct linux_mq_timedreceive_args *args) 669 { 670 #ifdef P1003_1B_MQUEUE 671 return (sys_kmq_timedreceive(td, (struct kmq_timedreceive_args *)args)); 672 #else 673 return (ENOSYS); 674 #endif 675 } 676 677 int 678 linux_mq_notify(struct thread *td, struct linux_mq_notify_args *args) 679 { 680 #ifdef P1003_1B_MQUEUE 681 return (sys_kmq_notify(td, (struct kmq_notify_args *)args)); 682 #else 683 return (ENOSYS); 684 #endif 685 } 686 687 int 688 linux_mq_getsetattr(struct thread *td, struct linux_mq_getsetattr_args *args) 689 { 690 #ifdef P1003_1B_MQUEUE 691 return (sys_kmq_setattr(td, (struct kmq_setattr_args *)args)); 692 #else 693 return (ENOSYS); 694 #endif 695 } 696