1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 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 "opt_posix.h" 30 31 #include <sys/param.h> 32 #include <sys/imgact_aout.h> 33 #include <sys/fcntl.h> 34 #include <sys/lock.h> 35 #include <sys/malloc.h> 36 #include <sys/mman.h> 37 #include <sys/mutex.h> 38 #include <sys/namei.h> 39 #include <sys/priv.h> 40 #include <sys/proc.h> 41 #include <sys/racct.h> 42 #include <sys/resource.h> 43 #include <sys/resourcevar.h> 44 #include <sys/syscallsubr.h> 45 #include <sys/sysproto.h> 46 #include <sys/vnode.h> 47 48 #include <security/audit/audit.h> 49 #include <security/mac/mac_framework.h> 50 51 #include <machine/frame.h> 52 #include <machine/pcb.h> /* needed for pcb definition in linux_set_thread_area */ 53 #include <machine/psl.h> 54 #include <machine/segments.h> 55 #include <machine/sysarch.h> 56 57 #include <vm/pmap.h> 58 #include <vm/vm.h> 59 #include <vm/vm_extern.h> 60 #include <vm/vm_kern.h> 61 #include <vm/vm_map.h> 62 #include <vm/vm_param.h> 63 64 #include <x86/reg.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_fork.h> 70 #include <compat/linux/linux_ipc.h> 71 #include <compat/linux/linux_misc.h> 72 #include <compat/linux/linux_mmap.h> 73 #include <compat/linux/linux_signal.h> 74 #include <compat/linux/linux_util.h> 75 76 77 struct 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 89 struct 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 97 struct l_ipc_kludge { 98 struct l_msgbuf *msgp; 99 l_long msgtyp; 100 }; 101 102 int 103 linux_ipc(struct thread *td, struct linux_ipc_args *args) 104 { 105 106 switch (args->what & 0xFFFF) { 107 case LINUX_SEMOP: { 108 109 return (kern_semop(td, args->arg1, PTRIN(args->ptr), 110 args->arg2, NULL)); 111 } 112 case LINUX_SEMGET: { 113 struct linux_semget_args a; 114 115 a.key = args->arg1; 116 a.nsems = args->arg2; 117 a.semflg = args->arg3; 118 return (linux_semget(td, &a)); 119 } 120 case LINUX_SEMCTL: { 121 struct linux_semctl_args a; 122 int error; 123 124 a.semid = args->arg1; 125 a.semnum = args->arg2; 126 a.cmd = args->arg3; 127 error = copyin(PTRIN(args->ptr), &a.arg, sizeof(a.arg)); 128 if (error) 129 return (error); 130 return (linux_semctl(td, &a)); 131 } 132 case LINUX_SEMTIMEDOP: { 133 struct linux_semtimedop_args a; 134 135 a.semid = args->arg1; 136 a.tsops = PTRIN(args->ptr); 137 a.nsops = args->arg2; 138 a.timeout = PTRIN(args->arg5); 139 return (linux_semtimedop(td, &a)); 140 } 141 case LINUX_MSGSND: { 142 struct linux_msgsnd_args a; 143 144 a.msqid = args->arg1; 145 a.msgp = PTRIN(args->ptr); 146 a.msgsz = args->arg2; 147 a.msgflg = args->arg3; 148 return (linux_msgsnd(td, &a)); 149 } 150 case LINUX_MSGRCV: { 151 struct linux_msgrcv_args a; 152 153 a.msqid = args->arg1; 154 a.msgsz = args->arg2; 155 a.msgflg = args->arg3; 156 if ((args->what >> 16) == 0) { 157 struct l_ipc_kludge tmp; 158 int error; 159 160 if (args->ptr == 0) 161 return (EINVAL); 162 error = copyin(PTRIN(args->ptr), &tmp, sizeof(tmp)); 163 if (error) 164 return (error); 165 a.msgp = PTRIN(tmp.msgp); 166 a.msgtyp = tmp.msgtyp; 167 } else { 168 a.msgp = PTRIN(args->ptr); 169 a.msgtyp = args->arg5; 170 } 171 return (linux_msgrcv(td, &a)); 172 } 173 case LINUX_MSGGET: { 174 struct linux_msgget_args a; 175 176 a.key = args->arg1; 177 a.msgflg = args->arg2; 178 return (linux_msgget(td, &a)); 179 } 180 case LINUX_MSGCTL: { 181 struct linux_msgctl_args a; 182 183 a.msqid = args->arg1; 184 a.cmd = args->arg2; 185 a.buf = PTRIN(args->ptr); 186 return (linux_msgctl(td, &a)); 187 } 188 case LINUX_SHMAT: { 189 struct linux_shmat_args a; 190 l_uintptr_t addr; 191 int error; 192 193 a.shmid = args->arg1; 194 a.shmaddr = PTRIN(args->ptr); 195 a.shmflg = args->arg2; 196 error = linux_shmat(td, &a); 197 if (error != 0) 198 return (error); 199 addr = td->td_retval[0]; 200 error = copyout(&addr, PTRIN(args->arg3), sizeof(addr)); 201 td->td_retval[0] = 0; 202 return (error); 203 } 204 case LINUX_SHMDT: { 205 struct linux_shmdt_args a; 206 207 a.shmaddr = PTRIN(args->ptr); 208 return (linux_shmdt(td, &a)); 209 } 210 case LINUX_SHMGET: { 211 struct linux_shmget_args a; 212 213 a.key = args->arg1; 214 a.size = args->arg2; 215 a.shmflg = args->arg3; 216 return (linux_shmget(td, &a)); 217 } 218 case LINUX_SHMCTL: { 219 struct linux_shmctl_args a; 220 221 a.shmid = args->arg1; 222 a.cmd = args->arg2; 223 a.buf = PTRIN(args->ptr); 224 return (linux_shmctl(td, &a)); 225 } 226 default: 227 break; 228 } 229 230 return (EINVAL); 231 } 232 233 int 234 linux_old_select(struct thread *td, struct linux_old_select_args *args) 235 { 236 struct l_old_select_argv linux_args; 237 struct linux_select_args newsel; 238 int error; 239 240 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 241 if (error) 242 return (error); 243 244 newsel.nfds = linux_args.nfds; 245 newsel.readfds = linux_args.readfds; 246 newsel.writefds = linux_args.writefds; 247 newsel.exceptfds = linux_args.exceptfds; 248 newsel.timeout = linux_args.timeout; 249 return (linux_select(td, &newsel)); 250 } 251 252 int 253 linux_set_cloned_tls(struct thread *td, void *desc) 254 { 255 struct segment_descriptor sd; 256 struct l_user_desc info; 257 int idx, error; 258 int a[2]; 259 260 error = copyin(desc, &info, sizeof(struct l_user_desc)); 261 if (error) { 262 linux_msg(td, "set_cloned_tls copyin failed!"); 263 } else { 264 idx = info.entry_number; 265 266 /* 267 * looks like we're getting the idx we returned 268 * in the set_thread_area() syscall 269 */ 270 if (idx != 6 && idx != 3) { 271 linux_msg(td, "set_cloned_tls resetting idx!"); 272 idx = 3; 273 } 274 275 /* this doesnt happen in practice */ 276 if (idx == 6) { 277 /* we might copy out the entry_number as 3 */ 278 info.entry_number = 3; 279 error = copyout(&info, desc, sizeof(struct l_user_desc)); 280 if (error) 281 linux_msg(td, "set_cloned_tls copyout failed!"); 282 } 283 284 a[0] = LINUX_LDT_entry_a(&info); 285 a[1] = LINUX_LDT_entry_b(&info); 286 287 memcpy(&sd, &a, sizeof(a)); 288 /* set %gs */ 289 td->td_pcb->pcb_gsd = sd; 290 td->td_pcb->pcb_gs = GSEL(GUGS_SEL, SEL_UPL); 291 } 292 293 return (error); 294 } 295 296 int 297 linux_set_upcall(struct thread *td, register_t stack) 298 { 299 300 if (stack) 301 td->td_frame->tf_esp = stack; 302 303 /* 304 * The newly created Linux thread returns 305 * to the user space by the same path that a parent do. 306 */ 307 td->td_frame->tf_eax = 0; 308 return (0); 309 } 310 311 int 312 linux_mmap2(struct thread *td, struct linux_mmap2_args *args) 313 { 314 315 return (linux_mmap_common(td, args->addr, args->len, args->prot, 316 args->flags, args->fd, (uint64_t)(uint32_t)args->pgoff * 317 PAGE_SIZE)); 318 } 319 320 int 321 linux_mmap(struct thread *td, struct linux_mmap_args *args) 322 { 323 int error; 324 struct l_mmap_argv linux_args; 325 326 error = copyin(args->ptr, &linux_args, sizeof(linux_args)); 327 if (error) 328 return (error); 329 330 return (linux_mmap_common(td, linux_args.addr, linux_args.len, 331 linux_args.prot, linux_args.flags, linux_args.fd, 332 (uint32_t)linux_args.pgoff)); 333 } 334 335 int 336 linux_ioperm(struct thread *td, struct linux_ioperm_args *args) 337 { 338 int error; 339 struct i386_ioperm_args iia; 340 341 iia.start = args->start; 342 iia.length = args->length; 343 iia.enable = args->enable; 344 error = i386_set_ioperm(td, &iia); 345 return (error); 346 } 347 348 int 349 linux_iopl(struct thread *td, struct linux_iopl_args *args) 350 { 351 int error; 352 353 if (args->level < 0 || args->level > 3) 354 return (EINVAL); 355 if ((error = priv_check(td, PRIV_IO)) != 0) 356 return (error); 357 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 358 return (error); 359 td->td_frame->tf_eflags = (td->td_frame->tf_eflags & ~PSL_IOPL) | 360 (args->level * (PSL_IOPL / 3)); 361 return (0); 362 } 363 364 int 365 linux_modify_ldt(struct thread *td, struct linux_modify_ldt_args *uap) 366 { 367 int error; 368 struct i386_ldt_args ldt; 369 struct l_descriptor ld; 370 union descriptor desc; 371 int size, written; 372 373 switch (uap->func) { 374 case 0x00: /* read_ldt */ 375 ldt.start = 0; 376 ldt.descs = uap->ptr; 377 ldt.num = uap->bytecount / sizeof(union descriptor); 378 error = i386_get_ldt(td, &ldt); 379 td->td_retval[0] *= sizeof(union descriptor); 380 break; 381 case 0x02: /* read_default_ldt = 0 */ 382 size = 5*sizeof(struct l_desc_struct); 383 if (size > uap->bytecount) 384 size = uap->bytecount; 385 for (written = error = 0; written < size && error == 0; written++) 386 error = subyte((char *)uap->ptr + written, 0); 387 td->td_retval[0] = written; 388 break; 389 case 0x01: /* write_ldt */ 390 case 0x11: /* write_ldt */ 391 if (uap->bytecount != sizeof(ld)) 392 return (EINVAL); 393 394 error = copyin(uap->ptr, &ld, sizeof(ld)); 395 if (error) 396 return (error); 397 398 ldt.start = ld.entry_number; 399 ldt.descs = &desc; 400 ldt.num = 1; 401 desc.sd.sd_lolimit = (ld.limit & 0x0000ffff); 402 desc.sd.sd_hilimit = (ld.limit & 0x000f0000) >> 16; 403 desc.sd.sd_lobase = (ld.base_addr & 0x00ffffff); 404 desc.sd.sd_hibase = (ld.base_addr & 0xff000000) >> 24; 405 desc.sd.sd_type = SDT_MEMRO | ((ld.read_exec_only ^ 1) << 1) | 406 (ld.contents << 2); 407 desc.sd.sd_dpl = 3; 408 desc.sd.sd_p = (ld.seg_not_present ^ 1); 409 desc.sd.sd_xx = 0; 410 desc.sd.sd_def32 = ld.seg_32bit; 411 desc.sd.sd_gran = ld.limit_in_pages; 412 error = i386_set_ldt(td, &ldt, &desc); 413 break; 414 default: 415 error = ENOSYS; 416 break; 417 } 418 419 if (error == EOPNOTSUPP) { 420 linux_msg(td, "modify_ldt needs kernel option USER_LDT"); 421 error = ENOSYS; 422 } 423 424 return (error); 425 } 426 427 int 428 linux_sigaction(struct thread *td, struct linux_sigaction_args *args) 429 { 430 l_osigaction_t osa; 431 l_sigaction_t act, oact; 432 int error; 433 434 if (args->nsa != NULL) { 435 error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); 436 if (error) 437 return (error); 438 act.lsa_handler = osa.lsa_handler; 439 act.lsa_flags = osa.lsa_flags; 440 act.lsa_restorer = osa.lsa_restorer; 441 LINUX_SIGEMPTYSET(act.lsa_mask); 442 act.lsa_mask.__mask = osa.lsa_mask; 443 } 444 445 error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, 446 args->osa ? &oact : NULL); 447 448 if (args->osa != NULL && !error) { 449 osa.lsa_handler = oact.lsa_handler; 450 osa.lsa_flags = oact.lsa_flags; 451 osa.lsa_restorer = oact.lsa_restorer; 452 osa.lsa_mask = oact.lsa_mask.__mask; 453 error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); 454 } 455 456 return (error); 457 } 458 459 /* 460 * Linux has two extra args, restart and oldmask. We dont use these, 461 * but it seems that "restart" is actually a context pointer that 462 * enables the signal to happen with a different register set. 463 */ 464 int 465 linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) 466 { 467 sigset_t sigmask; 468 l_sigset_t mask; 469 470 LINUX_SIGEMPTYSET(mask); 471 mask.__mask = args->mask; 472 linux_to_bsd_sigset(&mask, &sigmask); 473 return (kern_sigsuspend(td, sigmask)); 474 } 475 476 int 477 linux_pause(struct thread *td, struct linux_pause_args *args) 478 { 479 struct proc *p = td->td_proc; 480 sigset_t sigmask; 481 482 PROC_LOCK(p); 483 sigmask = td->td_sigmask; 484 PROC_UNLOCK(p); 485 return (kern_sigsuspend(td, sigmask)); 486 } 487 488 int 489 linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args) 490 { 491 struct l_user_desc info; 492 int error; 493 int idx; 494 int a[2]; 495 struct segment_descriptor sd; 496 497 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 498 if (error) 499 return (error); 500 501 idx = info.entry_number; 502 /* 503 * Semantics of Linux version: every thread in the system has array of 504 * 3 tls descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. This 505 * syscall loads one of the selected tls descriptors with a value and 506 * also loads GDT descriptors 6, 7 and 8 with the content of the 507 * per-thread descriptors. 508 * 509 * Semantics of FreeBSD version: I think we can ignore that Linux has 3 510 * per-thread descriptors and use just the 1st one. The tls_array[] 511 * is used only in set/get-thread_area() syscalls and for loading the 512 * GDT descriptors. In FreeBSD we use just one GDT descriptor for TLS 513 * so we will load just one. 514 * 515 * XXX: this doesn't work when a user space process tries to use more 516 * than 1 TLS segment. Comment in the Linux sources says wine might do 517 * this. 518 */ 519 520 /* 521 * we support just GLIBC TLS now 522 * we should let 3 proceed as well because we use this segment so 523 * if code does two subsequent calls it should succeed 524 */ 525 if (idx != 6 && idx != -1 && idx != 3) 526 return (EINVAL); 527 528 /* 529 * we have to copy out the GDT entry we use 530 * FreeBSD uses GDT entry #3 for storing %gs so load that 531 * 532 * XXX: what if a user space program doesn't check this value and tries 533 * to use 6, 7 or 8? 534 */ 535 idx = info.entry_number = 3; 536 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 537 if (error) 538 return (error); 539 540 if (LINUX_LDT_empty(&info)) { 541 a[0] = 0; 542 a[1] = 0; 543 } else { 544 a[0] = LINUX_LDT_entry_a(&info); 545 a[1] = LINUX_LDT_entry_b(&info); 546 } 547 548 memcpy(&sd, &a, sizeof(a)); 549 /* this is taken from i386 version of cpu_set_user_tls() */ 550 critical_enter(); 551 /* set %gs */ 552 td->td_pcb->pcb_gsd = sd; 553 PCPU_GET(fsgs_gdt)[1] = sd; 554 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 555 critical_exit(); 556 557 return (0); 558 } 559 560 int 561 linux_get_thread_area(struct thread *td, struct linux_get_thread_area_args *args) 562 { 563 564 struct l_user_desc info; 565 int error; 566 int idx; 567 struct l_desc_struct desc; 568 struct segment_descriptor sd; 569 570 error = copyin(args->desc, &info, sizeof(struct l_user_desc)); 571 if (error) 572 return (error); 573 574 idx = info.entry_number; 575 /* XXX: I am not sure if we want 3 to be allowed too. */ 576 if (idx != 6 && idx != 3) 577 return (EINVAL); 578 579 idx = 3; 580 581 memset(&info, 0, sizeof(info)); 582 583 sd = PCPU_GET(fsgs_gdt)[1]; 584 585 memcpy(&desc, &sd, sizeof(desc)); 586 587 info.entry_number = idx; 588 info.base_addr = LINUX_GET_BASE(&desc); 589 info.limit = LINUX_GET_LIMIT(&desc); 590 info.seg_32bit = LINUX_GET_32BIT(&desc); 591 info.contents = LINUX_GET_CONTENTS(&desc); 592 info.read_exec_only = !LINUX_GET_WRITABLE(&desc); 593 info.limit_in_pages = LINUX_GET_LIMIT_PAGES(&desc); 594 info.seg_not_present = !LINUX_GET_PRESENT(&desc); 595 info.useable = LINUX_GET_USEABLE(&desc); 596 597 error = copyout(&info, args->desc, sizeof(struct l_user_desc)); 598 if (error) 599 return (EFAULT); 600 601 return (0); 602 } 603 604 /* XXX: this wont work with module - convert it */ 605 int 606 linux_mq_open(struct thread *td, struct linux_mq_open_args *args) 607 { 608 #ifdef P1003_1B_MQUEUE 609 return (sys_kmq_open(td, (struct kmq_open_args *)args)); 610 #else 611 return (ENOSYS); 612 #endif 613 } 614 615 int 616 linux_mq_unlink(struct thread *td, struct linux_mq_unlink_args *args) 617 { 618 #ifdef P1003_1B_MQUEUE 619 return (sys_kmq_unlink(td, (struct kmq_unlink_args *)args)); 620 #else 621 return (ENOSYS); 622 #endif 623 } 624 625 int 626 linux_mq_timedsend(struct thread *td, struct linux_mq_timedsend_args *args) 627 { 628 #ifdef P1003_1B_MQUEUE 629 return (sys_kmq_timedsend(td, (struct kmq_timedsend_args *)args)); 630 #else 631 return (ENOSYS); 632 #endif 633 } 634 635 int 636 linux_mq_timedreceive(struct thread *td, struct linux_mq_timedreceive_args *args) 637 { 638 #ifdef P1003_1B_MQUEUE 639 return (sys_kmq_timedreceive(td, (struct kmq_timedreceive_args *)args)); 640 #else 641 return (ENOSYS); 642 #endif 643 } 644 645 int 646 linux_mq_notify(struct thread *td, struct linux_mq_notify_args *args) 647 { 648 #ifdef P1003_1B_MQUEUE 649 return (sys_kmq_notify(td, (struct kmq_notify_args *)args)); 650 #else 651 return (ENOSYS); 652 #endif 653 } 654 655 int 656 linux_mq_getsetattr(struct thread *td, struct linux_mq_getsetattr_args *args) 657 { 658 #ifdef P1003_1B_MQUEUE 659 return (sys_kmq_setattr(td, (struct kmq_setattr_args *)args)); 660 #else 661 return (ENOSYS); 662 #endif 663 } 664 665 void 666 bsd_to_linux_regset(const struct reg *b_reg, 667 struct linux_pt_regset *l_regset) 668 { 669 670 l_regset->ebx = b_reg->r_ebx; 671 l_regset->ecx = b_reg->r_ecx; 672 l_regset->edx = b_reg->r_edx; 673 l_regset->esi = b_reg->r_esi; 674 l_regset->edi = b_reg->r_edi; 675 l_regset->ebp = b_reg->r_ebp; 676 l_regset->eax = b_reg->r_eax; 677 l_regset->ds = b_reg->r_ds; 678 l_regset->es = b_reg->r_es; 679 l_regset->fs = b_reg->r_fs; 680 l_regset->gs = b_reg->r_gs; 681 l_regset->orig_eax = b_reg->r_eax; 682 l_regset->eip = b_reg->r_eip; 683 l_regset->cs = b_reg->r_cs; 684 l_regset->eflags = b_reg->r_eflags; 685 l_regset->esp = b_reg->r_esp; 686 l_regset->ss = b_reg->r_ss; 687 } 688 689 int 690 linux_uselib(struct thread *td, struct linux_uselib_args *args) 691 { 692 struct nameidata ni; 693 struct vnode *vp; 694 struct exec *a_out; 695 vm_map_t map; 696 vm_map_entry_t entry; 697 struct vattr attr; 698 vm_offset_t vmaddr; 699 unsigned long file_offset; 700 unsigned long bss_size; 701 ssize_t aresid; 702 int error; 703 bool locked, opened, textset; 704 705 a_out = NULL; 706 vp = NULL; 707 locked = false; 708 textset = false; 709 opened = false; 710 711 NDINIT(&ni, LOOKUP, ISOPEN | FOLLOW | LOCKLEAF | AUDITVNODE1, 712 UIO_USERSPACE, args->library); 713 error = namei(&ni); 714 if (error) 715 goto cleanup; 716 717 vp = ni.ni_vp; 718 NDFREE_PNBUF(&ni); 719 720 /* 721 * From here on down, we have a locked vnode that must be unlocked. 722 * XXX: The code below largely duplicates exec_check_permissions(). 723 */ 724 locked = true; 725 726 /* Executable? */ 727 error = VOP_GETATTR(vp, &attr, td->td_ucred); 728 if (error) 729 goto cleanup; 730 731 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 732 ((attr.va_mode & 0111) == 0) || (attr.va_type != VREG)) { 733 /* EACCESS is what exec(2) returns. */ 734 error = ENOEXEC; 735 goto cleanup; 736 } 737 738 /* Sensible size? */ 739 if (attr.va_size == 0) { 740 error = ENOEXEC; 741 goto cleanup; 742 } 743 744 /* Can we access it? */ 745 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 746 if (error) 747 goto cleanup; 748 749 /* 750 * XXX: This should use vn_open() so that it is properly authorized, 751 * and to reduce code redundancy all over the place here. 752 * XXX: Not really, it duplicates far more of exec_check_permissions() 753 * than vn_open(). 754 */ 755 #ifdef MAC 756 error = mac_vnode_check_open(td->td_ucred, vp, VREAD); 757 if (error) 758 goto cleanup; 759 #endif 760 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 761 if (error) 762 goto cleanup; 763 opened = true; 764 765 /* Pull in executable header into exec_map */ 766 error = vm_mmap(exec_map, (vm_offset_t *)&a_out, PAGE_SIZE, 767 VM_PROT_READ, VM_PROT_READ, 0, OBJT_VNODE, vp, 0); 768 if (error) 769 goto cleanup; 770 771 /* Is it a Linux binary ? */ 772 if (((a_out->a_magic >> 16) & 0xff) != 0x64) { 773 error = ENOEXEC; 774 goto cleanup; 775 } 776 777 /* 778 * While we are here, we should REALLY do some more checks 779 */ 780 781 /* Set file/virtual offset based on a.out variant. */ 782 switch ((int)(a_out->a_magic & 0xffff)) { 783 case 0413: /* ZMAGIC */ 784 file_offset = 1024; 785 break; 786 case 0314: /* QMAGIC */ 787 file_offset = 0; 788 break; 789 default: 790 error = ENOEXEC; 791 goto cleanup; 792 } 793 794 bss_size = round_page(a_out->a_bss); 795 796 /* Check various fields in header for validity/bounds. */ 797 if (a_out->a_text & PAGE_MASK || a_out->a_data & PAGE_MASK) { 798 error = ENOEXEC; 799 goto cleanup; 800 } 801 802 /* text + data can't exceed file size */ 803 if (a_out->a_data + a_out->a_text > attr.va_size) { 804 error = EFAULT; 805 goto cleanup; 806 } 807 808 /* 809 * text/data/bss must not exceed limits 810 * XXX - this is not complete. it should check current usage PLUS 811 * the resources needed by this library. 812 */ 813 PROC_LOCK(td->td_proc); 814 if (a_out->a_text > maxtsiz || 815 a_out->a_data + bss_size > lim_cur_proc(td->td_proc, RLIMIT_DATA) || 816 racct_set(td->td_proc, RACCT_DATA, a_out->a_data + 817 bss_size) != 0) { 818 PROC_UNLOCK(td->td_proc); 819 error = ENOMEM; 820 goto cleanup; 821 } 822 PROC_UNLOCK(td->td_proc); 823 824 /* 825 * Prevent more writers. 826 */ 827 error = VOP_SET_TEXT(vp); 828 if (error != 0) 829 goto cleanup; 830 textset = true; 831 832 /* 833 * Lock no longer needed 834 */ 835 locked = false; 836 VOP_UNLOCK(vp); 837 838 /* 839 * Check if file_offset page aligned. Currently we cannot handle 840 * misalinged file offsets, and so we read in the entire image 841 * (what a waste). 842 */ 843 if (file_offset & PAGE_MASK) { 844 /* Map text+data read/write/execute */ 845 846 /* a_entry is the load address and is page aligned */ 847 vmaddr = trunc_page(a_out->a_entry); 848 849 /* get anon user mapping, read+write+execute */ 850 error = vm_map_find(&td->td_proc->p_vmspace->vm_map, NULL, 0, 851 &vmaddr, a_out->a_text + a_out->a_data, 0, VMFS_NO_SPACE, 852 VM_PROT_ALL, VM_PROT_ALL, 0); 853 if (error) 854 goto cleanup; 855 856 error = vn_rdwr(UIO_READ, vp, (void *)vmaddr, file_offset, 857 a_out->a_text + a_out->a_data, UIO_USERSPACE, 0, 858 td->td_ucred, NOCRED, &aresid, td); 859 if (error != 0) 860 goto cleanup; 861 if (aresid != 0) { 862 error = ENOEXEC; 863 goto cleanup; 864 } 865 } else { 866 /* 867 * for QMAGIC, a_entry is 20 bytes beyond the load address 868 * to skip the executable header 869 */ 870 vmaddr = trunc_page(a_out->a_entry); 871 872 /* 873 * Map it all into the process's space as a single 874 * copy-on-write "data" segment. 875 */ 876 map = &td->td_proc->p_vmspace->vm_map; 877 error = vm_mmap(map, &vmaddr, 878 a_out->a_text + a_out->a_data, VM_PROT_ALL, VM_PROT_ALL, 879 MAP_PRIVATE | MAP_FIXED, OBJT_VNODE, vp, file_offset); 880 if (error) 881 goto cleanup; 882 vm_map_lock(map); 883 if (!vm_map_lookup_entry(map, vmaddr, &entry)) { 884 vm_map_unlock(map); 885 error = EDOOFUS; 886 goto cleanup; 887 } 888 entry->eflags |= MAP_ENTRY_VN_EXEC; 889 vm_map_unlock(map); 890 textset = false; 891 } 892 893 if (bss_size != 0) { 894 /* Calculate BSS start address */ 895 vmaddr = trunc_page(a_out->a_entry) + a_out->a_text + 896 a_out->a_data; 897 898 /* allocate some 'anon' space */ 899 error = vm_map_find(&td->td_proc->p_vmspace->vm_map, NULL, 0, 900 &vmaddr, bss_size, 0, VMFS_NO_SPACE, VM_PROT_ALL, 901 VM_PROT_ALL, 0); 902 if (error) 903 goto cleanup; 904 } 905 906 cleanup: 907 if (opened) { 908 if (locked) 909 VOP_UNLOCK(vp); 910 locked = false; 911 VOP_CLOSE(vp, FREAD, td->td_ucred, td); 912 } 913 if (textset) { 914 if (!locked) { 915 locked = true; 916 VOP_LOCK(vp, LK_SHARED | LK_RETRY); 917 } 918 VOP_UNSET_TEXT_CHECKED(vp); 919 } 920 if (locked) 921 VOP_UNLOCK(vp); 922 923 /* Release the temporary mapping. */ 924 if (a_out) 925 kmap_free_wakeup(exec_map, (vm_offset_t)a_out, PAGE_SIZE); 926 927 return (error); 928 } 929