1 /*- 2 * Copyright (c) 2000 David O'Brien 3 * Copyright (c) 1995-1996 S�ren Schmidt 4 * Copyright (c) 1996 Peter Wemm 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 * in this position and unchanged. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. The name of the author may not be used to endorse or promote products 17 * derived from this software without specific prior written permission 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 29 * 30 * $FreeBSD$ 31 */ 32 33 #include <sys/param.h> 34 #include <sys/exec.h> 35 #include <sys/fcntl.h> 36 #include <sys/imgact.h> 37 #include <sys/imgact_elf.h> 38 #include <sys/kernel.h> 39 #include <sys/lock.h> 40 #include <sys/malloc.h> 41 #include <sys/mutex.h> 42 #include <sys/mman.h> 43 #include <sys/namei.h> 44 #include <sys/pioctl.h> 45 #include <sys/proc.h> 46 #include <sys/procfs.h> 47 #include <sys/resourcevar.h> 48 #include <sys/systm.h> 49 #include <sys/signalvar.h> 50 #include <sys/stat.h> 51 #include <sys/sx.h> 52 #include <sys/syscall.h> 53 #include <sys/sysctl.h> 54 #include <sys/sysent.h> 55 #include <sys/vnode.h> 56 57 #include <vm/vm.h> 58 #include <vm/vm_kern.h> 59 #include <vm/vm_param.h> 60 #include <vm/pmap.h> 61 #include <vm/vm_map.h> 62 #include <vm/vm_object.h> 63 #include <vm/vm_extern.h> 64 65 #include <machine/elf.h> 66 #include <machine/md_var.h> 67 68 #define OLD_EI_BRAND 8 69 70 __ElfType(Brandinfo); 71 __ElfType(Auxargs); 72 73 static int __elfN(check_header)(const Elf_Ehdr *hdr); 74 static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr, 75 const char *interp); 76 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 77 u_long *entry, size_t pagesize); 78 static int __elfN(load_section)(struct proc *p, 79 struct vmspace *vmspace, struct vnode *vp, vm_object_t object, 80 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 81 vm_prot_t prot, size_t pagesize); 82 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 83 84 static int elf_trace = 0; 85 #if __ELF_WORD_SIZE == 32 86 SYSCTL_INT(_debug, OID_AUTO, elf32_trace, CTLFLAG_RW, &elf_trace, 0, ""); 87 #else 88 SYSCTL_INT(_debug, OID_AUTO, elf64_trace, CTLFLAG_RW, &elf_trace, 0, ""); 89 #endif 90 91 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 92 extern int fallback_elf_brand; 93 94 int 95 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 96 { 97 int i; 98 99 for (i = 0; i < MAX_BRANDS; i++) { 100 if (elf_brand_list[i] == NULL) { 101 elf_brand_list[i] = entry; 102 break; 103 } 104 } 105 if (i == MAX_BRANDS) 106 return (-1); 107 return (0); 108 } 109 110 int 111 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 112 { 113 int i; 114 115 for (i = 0; i < MAX_BRANDS; i++) { 116 if (elf_brand_list[i] == entry) { 117 elf_brand_list[i] = NULL; 118 break; 119 } 120 } 121 if (i == MAX_BRANDS) 122 return (-1); 123 return (0); 124 } 125 126 int 127 __elfN(brand_inuse)(Elf_Brandinfo *entry) 128 { 129 struct proc *p; 130 int rval = FALSE; 131 132 sx_slock(&allproc_lock); 133 LIST_FOREACH(p, &allproc, p_list) { 134 if (p->p_sysent == entry->sysvec) { 135 rval = TRUE; 136 break; 137 } 138 } 139 sx_sunlock(&allproc_lock); 140 141 return (rval); 142 } 143 144 static Elf_Brandinfo * 145 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp) 146 { 147 Elf_Brandinfo *bi; 148 int i; 149 150 /* 151 * We support three types of branding -- (1) the ELF EI_OSABI field 152 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 153 * branding w/in the ELF header, and (3) path of the `interp_path' 154 * field. We should also look for an ".note.ABI-tag" ELF section now 155 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones. 156 */ 157 158 /* If the executable has a brand, search for it in the brand list. */ 159 for (i = 0; i < MAX_BRANDS; i++) { 160 bi = elf_brand_list[i]; 161 if (bi != NULL && hdr->e_machine == bi->machine && 162 (hdr->e_ident[EI_OSABI] == bi->brand || 163 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 164 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 165 return (bi); 166 } 167 168 /* Lacking a known brand, search for a recognized interpreter. */ 169 if (interp != NULL) { 170 for (i = 0; i < MAX_BRANDS; i++) { 171 bi = elf_brand_list[i]; 172 if (bi != NULL && hdr->e_machine == bi->machine && 173 strcmp(interp, bi->interp_path) == 0) 174 return (bi); 175 } 176 } 177 178 /* Lacking a recognized interpreter, try the default brand */ 179 for (i = 0; i < MAX_BRANDS; i++) { 180 bi = elf_brand_list[i]; 181 if (bi != NULL && hdr->e_machine == bi->machine && 182 fallback_elf_brand == bi->brand) 183 return (bi); 184 } 185 return (NULL); 186 } 187 188 static int 189 __elfN(check_header)(const Elf_Ehdr *hdr) 190 { 191 Elf_Brandinfo *bi; 192 int i; 193 194 if (!IS_ELF(*hdr) || 195 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 196 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 197 hdr->e_ident[EI_VERSION] != EV_CURRENT) 198 return (ENOEXEC); 199 200 /* 201 * Make sure we have at least one brand for this machine. 202 */ 203 204 for (i = 0; i < MAX_BRANDS; i++) { 205 bi = elf_brand_list[i]; 206 if (bi != NULL && bi->machine == hdr->e_machine) 207 break; 208 } 209 if (i == MAX_BRANDS) 210 return (ENOEXEC); 211 212 if (hdr->e_version != ELF_TARG_VER) 213 return (ENOEXEC); 214 215 return (0); 216 } 217 218 static int 219 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 220 vm_offset_t start, vm_offset_t end, vm_prot_t prot, 221 vm_prot_t max) 222 { 223 int error, rv; 224 vm_offset_t off; 225 vm_offset_t data_buf = 0; 226 227 /* 228 * Create the page if it doesn't exist yet. Ignore errors. 229 */ 230 vm_map_lock(map); 231 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), max, 232 max, 0); 233 vm_map_unlock(map); 234 235 /* 236 * Find the page from the underlying object. 237 */ 238 if (object) { 239 vm_object_reference(object); 240 rv = vm_map_find(exec_map, 241 object, 242 trunc_page(offset), 243 &data_buf, 244 PAGE_SIZE, 245 TRUE, 246 VM_PROT_READ, 247 VM_PROT_ALL, 248 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 249 if (rv != KERN_SUCCESS) { 250 vm_object_deallocate(object); 251 return (rv); 252 } 253 254 off = offset - trunc_page(offset); 255 error = copyout((caddr_t)data_buf + off, (caddr_t)start, 256 end - start); 257 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 258 if (error) { 259 return (KERN_FAILURE); 260 } 261 } 262 263 return (KERN_SUCCESS); 264 } 265 266 static int 267 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 268 vm_offset_t start, vm_offset_t end, vm_prot_t prot, 269 vm_prot_t max, int cow) 270 { 271 int rv; 272 273 if (start != trunc_page(start)) { 274 rv = __elfN(map_partial)(map, object, offset, start, 275 round_page(start), prot, max); 276 if (rv) 277 return (rv); 278 offset += round_page(start) - start; 279 start = round_page(start); 280 } 281 if (end != round_page(end)) { 282 rv = __elfN(map_partial)(map, object, offset + 283 trunc_page(end) - start, trunc_page(end), end, prot, max); 284 if (rv) 285 return (rv); 286 end = trunc_page(end); 287 } 288 if (end > start) { 289 if (offset & PAGE_MASK) { 290 vm_offset_t data_buf, off; 291 vm_size_t sz; 292 int error; 293 294 /* 295 * The mapping is not page aligned. This means we have 296 * to copy the data. Sigh. 297 */ 298 rv = vm_map_find(map, 0, 0, &start, end - start, 299 FALSE, prot, max, 0); 300 if (rv) 301 return (rv); 302 while (start < end) { 303 vm_object_reference(object); 304 rv = vm_map_find(exec_map, 305 object, 306 trunc_page(offset), 307 &data_buf, 308 2 * PAGE_SIZE, 309 TRUE, 310 VM_PROT_READ, 311 VM_PROT_ALL, 312 (MAP_COPY_ON_WRITE 313 | MAP_PREFAULT_PARTIAL)); 314 if (rv != KERN_SUCCESS) { 315 vm_object_deallocate(object); 316 return (rv); 317 } 318 off = offset - trunc_page(offset); 319 sz = end - start; 320 if (sz > PAGE_SIZE) 321 sz = PAGE_SIZE; 322 error = copyout((caddr_t)data_buf + off, 323 (caddr_t)start, sz); 324 vm_map_remove(exec_map, data_buf, 325 data_buf + 2 * PAGE_SIZE); 326 if (error) { 327 return (KERN_FAILURE); 328 } 329 start += sz; 330 } 331 rv = KERN_SUCCESS; 332 } else { 333 vm_map_lock(map); 334 rv = vm_map_insert(map, object, offset, start, end, 335 prot, max, cow); 336 vm_map_unlock(map); 337 } 338 return (rv); 339 } else { 340 return (KERN_SUCCESS); 341 } 342 } 343 344 static int 345 __elfN(load_section)(struct proc *p, struct vmspace *vmspace, 346 struct vnode *vp, vm_object_t object, vm_offset_t offset, 347 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 348 size_t pagesize) 349 { 350 size_t map_len; 351 vm_offset_t map_addr; 352 int error, rv; 353 size_t copy_len; 354 vm_offset_t file_addr; 355 vm_offset_t data_buf = 0; 356 357 GIANT_REQUIRED; 358 359 error = 0; 360 361 /* 362 * It's necessary to fail if the filsz + offset taken from the 363 * header is greater than the actual file pager object's size. 364 * If we were to allow this, then the vm_map_find() below would 365 * walk right off the end of the file object and into the ether. 366 * 367 * While I'm here, might as well check for something else that 368 * is invalid: filsz cannot be greater than memsz. 369 */ 370 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 371 filsz > memsz) { 372 uprintf("elf_load_section: truncated ELF file\n"); 373 return (ENOEXEC); 374 } 375 376 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 377 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 378 379 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 380 file_addr = trunc_page_ps(offset, pagesize); 381 382 /* 383 * We have two choices. We can either clear the data in the last page 384 * of an oversized mapping, or we can start the anon mapping a page 385 * early and copy the initialized data into that first page. We 386 * choose the second.. 387 */ 388 if (memsz > filsz) 389 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 390 else 391 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 392 393 if (map_len != 0) { 394 vm_object_reference(object); 395 rv = __elfN(map_insert)(&vmspace->vm_map, 396 object, 397 file_addr, /* file offset */ 398 map_addr, /* virtual start */ 399 map_addr + map_len,/* virtual end */ 400 prot, 401 VM_PROT_ALL, 402 MAP_COPY_ON_WRITE | MAP_PREFAULT); 403 if (rv != KERN_SUCCESS) { 404 vm_object_deallocate(object); 405 return (EINVAL); 406 } 407 408 /* we can stop now if we've covered it all */ 409 if (memsz == filsz) { 410 return (0); 411 } 412 } 413 414 415 /* 416 * We have to get the remaining bit of the file into the first part 417 * of the oversized map segment. This is normally because the .data 418 * segment in the file is extended to provide bss. It's a neat idea 419 * to try and save a page, but it's a pain in the behind to implement. 420 */ 421 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 422 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 423 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 424 map_addr; 425 426 /* This had damn well better be true! */ 427 if (map_len != 0) { 428 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr, 429 map_addr + map_len, VM_PROT_ALL, VM_PROT_ALL, 0); 430 if (rv != KERN_SUCCESS) { 431 return (EINVAL); 432 } 433 } 434 435 if (copy_len != 0) { 436 vm_offset_t off; 437 vm_object_reference(object); 438 rv = vm_map_find(exec_map, 439 object, 440 trunc_page(offset + filsz), 441 &data_buf, 442 PAGE_SIZE, 443 TRUE, 444 VM_PROT_READ, 445 VM_PROT_ALL, 446 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 447 if (rv != KERN_SUCCESS) { 448 vm_object_deallocate(object); 449 return (EINVAL); 450 } 451 452 /* send the page fragment to user space */ 453 off = trunc_page_ps(offset + filsz, pagesize) - 454 trunc_page(offset + filsz); 455 error = copyout((caddr_t)data_buf + off, (caddr_t)map_addr, 456 copy_len); 457 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 458 if (error) { 459 return (error); 460 } 461 } 462 463 /* 464 * set it to the specified protection. 465 * XXX had better undo the damage from pasting over the cracks here! 466 */ 467 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr), 468 round_page(map_addr + map_len), prot, FALSE); 469 470 return (error); 471 } 472 473 /* 474 * Load the file "file" into memory. It may be either a shared object 475 * or an executable. 476 * 477 * The "addr" reference parameter is in/out. On entry, it specifies 478 * the address where a shared object should be loaded. If the file is 479 * an executable, this value is ignored. On exit, "addr" specifies 480 * where the file was actually loaded. 481 * 482 * The "entry" reference parameter is out only. On exit, it specifies 483 * the entry point for the loaded file. 484 */ 485 static int 486 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 487 u_long *entry, size_t pagesize) 488 { 489 struct { 490 struct nameidata nd; 491 struct vattr attr; 492 struct image_params image_params; 493 } *tempdata; 494 const Elf_Ehdr *hdr = NULL; 495 const Elf_Phdr *phdr = NULL; 496 struct nameidata *nd; 497 struct vmspace *vmspace = p->p_vmspace; 498 struct vattr *attr; 499 struct image_params *imgp; 500 vm_prot_t prot; 501 u_long rbase; 502 u_long base_addr = 0; 503 int error, i, numsegs; 504 505 if (curthread->td_proc != p) 506 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */ 507 508 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 509 nd = &tempdata->nd; 510 attr = &tempdata->attr; 511 imgp = &tempdata->image_params; 512 513 /* 514 * Initialize part of the common data 515 */ 516 imgp->proc = p; 517 imgp->userspace_argv = NULL; 518 imgp->userspace_envv = NULL; 519 imgp->attr = attr; 520 imgp->firstpage = NULL; 521 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE); 522 imgp->object = NULL; 523 imgp->execlabel = NULL; 524 525 if (imgp->image_header == NULL) { 526 nd->ni_vp = NULL; 527 error = ENOMEM; 528 goto fail; 529 } 530 531 /* XXXKSE */ 532 NDINIT(nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, curthread); 533 534 if ((error = namei(nd)) != 0) { 535 nd->ni_vp = NULL; 536 goto fail; 537 } 538 NDFREE(nd, NDF_ONLY_PNBUF); 539 imgp->vp = nd->ni_vp; 540 541 /* 542 * Check permissions, modes, uid, etc on the file, and "open" it. 543 */ 544 error = exec_check_permissions(imgp); 545 if (error) { 546 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */ 547 goto fail; 548 } 549 550 error = exec_map_first_page(imgp); 551 /* 552 * Also make certain that the interpreter stays the same, so set 553 * its VV_TEXT flag, too. 554 */ 555 if (error == 0) 556 nd->ni_vp->v_vflag |= VV_TEXT; 557 558 VOP_GETVOBJECT(nd->ni_vp, &imgp->object); 559 vm_object_reference(imgp->object); 560 561 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */ 562 if (error) 563 goto fail; 564 565 hdr = (const Elf_Ehdr *)imgp->image_header; 566 if ((error = __elfN(check_header)(hdr)) != 0) 567 goto fail; 568 if (hdr->e_type == ET_DYN) 569 rbase = *addr; 570 else if (hdr->e_type == ET_EXEC) 571 rbase = 0; 572 else { 573 error = ENOEXEC; 574 goto fail; 575 } 576 577 /* Only support headers that fit within first page for now */ 578 if ((hdr->e_phoff > PAGE_SIZE) || 579 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 580 error = ENOEXEC; 581 goto fail; 582 } 583 584 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 585 586 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 587 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 588 prot = 0; 589 if (phdr[i].p_flags & PF_X) 590 prot |= VM_PROT_EXECUTE; 591 if (phdr[i].p_flags & PF_W) 592 prot |= VM_PROT_WRITE; 593 if (phdr[i].p_flags & PF_R) 594 prot |= VM_PROT_READ; 595 596 if ((error = __elfN(load_section)(p, vmspace, 597 nd->ni_vp, imgp->object, phdr[i].p_offset, 598 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 599 phdr[i].p_memsz, phdr[i].p_filesz, prot, 600 pagesize)) != 0) 601 goto fail; 602 /* 603 * Establish the base address if this is the 604 * first segment. 605 */ 606 if (numsegs == 0) 607 base_addr = trunc_page(phdr[i].p_vaddr + 608 rbase); 609 numsegs++; 610 } 611 } 612 *addr = base_addr; 613 *entry = (unsigned long)hdr->e_entry + rbase; 614 615 fail: 616 if (imgp->firstpage) 617 exec_unmap_first_page(imgp); 618 if (imgp->image_header) 619 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header, 620 PAGE_SIZE); 621 if (imgp->object) 622 vm_object_deallocate(imgp->object); 623 624 if (nd->ni_vp) 625 vrele(nd->ni_vp); 626 627 free(tempdata, M_TEMP); 628 629 return (error); 630 } 631 632 static int 633 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 634 { 635 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 636 const Elf_Phdr *phdr; 637 Elf_Auxargs *elf_auxargs = NULL; 638 struct vmspace *vmspace; 639 vm_prot_t prot; 640 u_long text_size = 0, data_size = 0, total_size = 0; 641 u_long text_addr = 0, data_addr = 0; 642 u_long seg_size, seg_addr; 643 u_long addr, entry = 0, proghdr = 0; 644 int error, i; 645 const char *interp = NULL; 646 Elf_Brandinfo *brand_info; 647 char *path; 648 struct thread *td = curthread; 649 struct sysentvec *sv; 650 651 GIANT_REQUIRED; 652 653 /* 654 * Do we have a valid ELF header ? 655 */ 656 if (__elfN(check_header)(hdr) != 0 || hdr->e_type != ET_EXEC) 657 return (-1); 658 659 /* 660 * From here on down, we return an errno, not -1, as we've 661 * detected an ELF file. 662 */ 663 664 if ((hdr->e_phoff > PAGE_SIZE) || 665 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 666 /* Only support headers in first page for now */ 667 return (ENOEXEC); 668 } 669 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 670 671 /* 672 * From this point on, we may have resources that need to be freed. 673 */ 674 675 VOP_UNLOCK(imgp->vp, 0, td); 676 677 for (i = 0; i < hdr->e_phnum; i++) { 678 switch (phdr[i].p_type) { 679 case PT_INTERP: /* Path to interpreter */ 680 if (phdr[i].p_filesz > MAXPATHLEN || 681 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) { 682 error = ENOEXEC; 683 goto fail; 684 } 685 interp = imgp->image_header + phdr[i].p_offset; 686 break; 687 default: 688 break; 689 } 690 } 691 692 brand_info = __elfN(get_brandinfo)(hdr, interp); 693 if (brand_info == NULL) { 694 uprintf("ELF binary type \"%u\" not known.\n", 695 hdr->e_ident[EI_OSABI]); 696 error = ENOEXEC; 697 goto fail; 698 } 699 sv = brand_info->sysvec; 700 701 if ((error = exec_extract_strings(imgp)) != 0) 702 goto fail; 703 704 exec_new_vmspace(imgp, sv); 705 706 vmspace = imgp->proc->p_vmspace; 707 708 for (i = 0; i < hdr->e_phnum; i++) { 709 switch (phdr[i].p_type) { 710 case PT_LOAD: /* Loadable segment */ 711 prot = 0; 712 if (phdr[i].p_flags & PF_X) 713 prot |= VM_PROT_EXECUTE; 714 if (phdr[i].p_flags & PF_W) 715 prot |= VM_PROT_WRITE; 716 if (phdr[i].p_flags & PF_R) 717 prot |= VM_PROT_READ; 718 719 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER) 720 /* 721 * Some x86 binaries assume read == executable, 722 * notably the M3 runtime and therefore cvsup 723 */ 724 if (prot & VM_PROT_READ) 725 prot |= VM_PROT_EXECUTE; 726 #endif 727 728 if ((error = __elfN(load_section)(imgp->proc, vmspace, 729 imgp->vp, imgp->object, phdr[i].p_offset, 730 (caddr_t)(uintptr_t)phdr[i].p_vaddr, 731 phdr[i].p_memsz, phdr[i].p_filesz, prot, 732 sv->sv_pagesize)) != 0) 733 goto fail; 734 735 seg_addr = trunc_page(phdr[i].p_vaddr); 736 seg_size = round_page(phdr[i].p_memsz + 737 phdr[i].p_vaddr - seg_addr); 738 739 /* 740 * Is this .text or .data? We can't use 741 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 742 * alpha terribly and possibly does other bad 743 * things so we stick to the old way of figuring 744 * it out: If the segment contains the program 745 * entry point, it's a text segment, otherwise it 746 * is a data segment. 747 * 748 * Note that obreak() assumes that data_addr + 749 * data_size == end of data load area, and the ELF 750 * file format expects segments to be sorted by 751 * address. If multiple data segments exist, the 752 * last one will be used. 753 */ 754 if (hdr->e_entry >= phdr[i].p_vaddr && 755 hdr->e_entry < (phdr[i].p_vaddr + 756 phdr[i].p_memsz)) { 757 text_size = seg_size; 758 text_addr = seg_addr; 759 entry = (u_long)hdr->e_entry; 760 } else { 761 data_size = seg_size; 762 data_addr = seg_addr; 763 } 764 total_size += seg_size; 765 break; 766 case PT_PHDR: /* Program header table info */ 767 proghdr = phdr[i].p_vaddr; 768 break; 769 default: 770 break; 771 } 772 } 773 774 if (data_addr == 0 && data_size == 0) { 775 data_addr = text_addr; 776 data_size = text_size; 777 } 778 779 /* 780 * Check limits. It should be safe to check the 781 * limits after loading the segments since we do 782 * not actually fault in all the segments pages. 783 */ 784 if (data_size > 785 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur || 786 text_size > maxtsiz || 787 total_size > 788 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) { 789 error = ENOMEM; 790 goto fail; 791 } 792 793 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 794 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 795 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 796 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 797 798 addr = ELF_RTLD_ADDR(vmspace); 799 800 imgp->entry_addr = entry; 801 802 imgp->proc->p_sysent = sv; 803 if (interp != NULL) { 804 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 805 snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path, 806 interp); 807 if ((error = __elfN(load_file)(imgp->proc, path, &addr, 808 &imgp->entry_addr, sv->sv_pagesize)) != 0) { 809 if ((error = __elfN(load_file)(imgp->proc, interp, 810 &addr, &imgp->entry_addr, sv->sv_pagesize)) != 0) { 811 uprintf("ELF interpreter %s not found\n", 812 path); 813 free(path, M_TEMP); 814 goto fail; 815 } 816 } 817 free(path, M_TEMP); 818 } 819 820 /* 821 * Construct auxargs table (used by the fixup routine) 822 */ 823 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 824 elf_auxargs->execfd = -1; 825 elf_auxargs->phdr = proghdr; 826 elf_auxargs->phent = hdr->e_phentsize; 827 elf_auxargs->phnum = hdr->e_phnum; 828 elf_auxargs->pagesz = PAGE_SIZE; 829 elf_auxargs->base = addr; 830 elf_auxargs->flags = 0; 831 elf_auxargs->entry = entry; 832 elf_auxargs->trace = elf_trace; 833 834 imgp->auxargs = elf_auxargs; 835 imgp->interpreted = 0; 836 837 fail: 838 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td); 839 return (error); 840 } 841 842 #if __ELF_WORD_SIZE == 32 843 #define suword suword32 844 #define stacktype u_int32_t 845 #else 846 #define suword suword64 847 #define stacktype u_int64_t 848 #endif 849 850 int 851 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 852 { 853 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 854 stacktype *base; 855 stacktype *pos; 856 857 base = (stacktype *)*stack_base; 858 pos = base + (imgp->argc + imgp->envc + 2); 859 860 if (args->trace) { 861 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 862 } 863 if (args->execfd != -1) { 864 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 865 } 866 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 867 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 868 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 869 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 870 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 871 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 872 AUXARGS_ENTRY(pos, AT_BASE, args->base); 873 AUXARGS_ENTRY(pos, AT_NULL, 0); 874 875 free(imgp->auxargs, M_TEMP); 876 imgp->auxargs = NULL; 877 878 base--; 879 suword(base, (long)imgp->argc); 880 *stack_base = (register_t *)base; 881 return (0); 882 } 883 884 /* 885 * Code for generating ELF core dumps. 886 */ 887 888 typedef void (*segment_callback)(vm_map_entry_t, void *); 889 890 /* Closure for cb_put_phdr(). */ 891 struct phdr_closure { 892 Elf_Phdr *phdr; /* Program header to fill in */ 893 Elf_Off offset; /* Offset of segment in core file */ 894 }; 895 896 /* Closure for cb_size_segment(). */ 897 struct sseg_closure { 898 int count; /* Count of writable segments. */ 899 size_t size; /* Total size of all writable segments. */ 900 }; 901 902 static void cb_put_phdr(vm_map_entry_t, void *); 903 static void cb_size_segment(vm_map_entry_t, void *); 904 static void each_writable_segment(struct proc *, segment_callback, void *); 905 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 906 int, void *, size_t); 907 static void __elfN(puthdr)(struct proc *, void *, size_t *, 908 const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int); 909 static void __elfN(putnote)(void *, size_t *, const char *, int, 910 const void *, size_t); 911 912 extern int osreldate; 913 914 int 915 __elfN(coredump)(td, vp, limit) 916 struct thread *td; 917 register struct vnode *vp; 918 off_t limit; 919 { 920 register struct proc *p = td->td_proc; 921 register struct ucred *cred = td->td_ucred; 922 int error = 0; 923 struct sseg_closure seginfo; 924 void *hdr; 925 size_t hdrsize; 926 927 /* Size the program segments. */ 928 seginfo.count = 0; 929 seginfo.size = 0; 930 each_writable_segment(p, cb_size_segment, &seginfo); 931 932 /* 933 * Calculate the size of the core file header area by making 934 * a dry run of generating it. Nothing is written, but the 935 * size is calculated. 936 */ 937 hdrsize = 0; 938 __elfN(puthdr)((struct proc *)NULL, (void *)NULL, &hdrsize, 939 (const prstatus_t *)NULL, (const prfpregset_t *)NULL, 940 (const prpsinfo_t *)NULL, seginfo.count); 941 942 if (hdrsize + seginfo.size >= limit) 943 return (EFAULT); 944 945 /* 946 * Allocate memory for building the header, fill it up, 947 * and write it out. 948 */ 949 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 950 if (hdr == NULL) { 951 return (EINVAL); 952 } 953 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize); 954 955 /* Write the contents of all of the writable segments. */ 956 if (error == 0) { 957 Elf_Phdr *php; 958 off_t offset; 959 int i; 960 961 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 962 offset = hdrsize; 963 for (i = 0; i < seginfo.count; i++) { 964 error = vn_rdwr_inchunks(UIO_WRITE, vp, 965 (caddr_t)(uintptr_t)php->p_vaddr, 966 php->p_filesz, offset, UIO_USERSPACE, 967 IO_UNIT | IO_DIRECT, cred, NOCRED, (int *)NULL, 968 curthread); /* XXXKSE */ 969 if (error != 0) 970 break; 971 offset += php->p_filesz; 972 php++; 973 } 974 } 975 free(hdr, M_TEMP); 976 977 return (error); 978 } 979 980 /* 981 * A callback for each_writable_segment() to write out the segment's 982 * program header entry. 983 */ 984 static void 985 cb_put_phdr(entry, closure) 986 vm_map_entry_t entry; 987 void *closure; 988 { 989 struct phdr_closure *phc = (struct phdr_closure *)closure; 990 Elf_Phdr *phdr = phc->phdr; 991 992 phc->offset = round_page(phc->offset); 993 994 phdr->p_type = PT_LOAD; 995 phdr->p_offset = phc->offset; 996 phdr->p_vaddr = entry->start; 997 phdr->p_paddr = 0; 998 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 999 phdr->p_align = PAGE_SIZE; 1000 phdr->p_flags = 0; 1001 if (entry->protection & VM_PROT_READ) 1002 phdr->p_flags |= PF_R; 1003 if (entry->protection & VM_PROT_WRITE) 1004 phdr->p_flags |= PF_W; 1005 if (entry->protection & VM_PROT_EXECUTE) 1006 phdr->p_flags |= PF_X; 1007 1008 phc->offset += phdr->p_filesz; 1009 phc->phdr++; 1010 } 1011 1012 /* 1013 * A callback for each_writable_segment() to gather information about 1014 * the number of segments and their total size. 1015 */ 1016 static void 1017 cb_size_segment(entry, closure) 1018 vm_map_entry_t entry; 1019 void *closure; 1020 { 1021 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1022 1023 ssc->count++; 1024 ssc->size += entry->end - entry->start; 1025 } 1026 1027 /* 1028 * For each writable segment in the process's memory map, call the given 1029 * function with a pointer to the map entry and some arbitrary 1030 * caller-supplied data. 1031 */ 1032 static void 1033 each_writable_segment(p, func, closure) 1034 struct proc *p; 1035 segment_callback func; 1036 void *closure; 1037 { 1038 vm_map_t map = &p->p_vmspace->vm_map; 1039 vm_map_entry_t entry; 1040 1041 for (entry = map->header.next; entry != &map->header; 1042 entry = entry->next) { 1043 vm_object_t obj; 1044 1045 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) || 1046 (entry->protection & (VM_PROT_READ|VM_PROT_WRITE)) != 1047 (VM_PROT_READ|VM_PROT_WRITE)) 1048 continue; 1049 1050 /* 1051 ** Dont include memory segment in the coredump if 1052 ** MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1053 ** madvise(2). 1054 */ 1055 if (entry->eflags & MAP_ENTRY_NOCOREDUMP) 1056 continue; 1057 1058 if ((obj = entry->object.vm_object) == NULL) 1059 continue; 1060 1061 /* Find the deepest backing object. */ 1062 while (obj->backing_object != NULL) 1063 obj = obj->backing_object; 1064 1065 /* Ignore memory-mapped devices and such things. */ 1066 if (obj->type != OBJT_DEFAULT && 1067 obj->type != OBJT_SWAP && 1068 obj->type != OBJT_VNODE) 1069 continue; 1070 1071 (*func)(entry, closure); 1072 } 1073 } 1074 1075 /* 1076 * Write the core file header to the file, including padding up to 1077 * the page boundary. 1078 */ 1079 static int 1080 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize) 1081 struct thread *td; 1082 struct vnode *vp; 1083 struct ucred *cred; 1084 int numsegs; 1085 size_t hdrsize; 1086 void *hdr; 1087 { 1088 struct { 1089 prstatus_t status; 1090 prfpregset_t fpregset; 1091 prpsinfo_t psinfo; 1092 } *tempdata; 1093 struct proc *p = td->td_proc; 1094 size_t off; 1095 prstatus_t *status; 1096 prfpregset_t *fpregset; 1097 prpsinfo_t *psinfo; 1098 1099 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK); 1100 status = &tempdata->status; 1101 fpregset = &tempdata->fpregset; 1102 psinfo = &tempdata->psinfo; 1103 1104 /* Gather the information for the header. */ 1105 status->pr_version = PRSTATUS_VERSION; 1106 status->pr_statussz = sizeof(prstatus_t); 1107 status->pr_gregsetsz = sizeof(gregset_t); 1108 status->pr_fpregsetsz = sizeof(fpregset_t); 1109 status->pr_osreldate = osreldate; 1110 status->pr_cursig = p->p_sig; 1111 status->pr_pid = p->p_pid; 1112 fill_regs(td, &status->pr_reg); 1113 1114 fill_fpregs(td, fpregset); 1115 1116 psinfo->pr_version = PRPSINFO_VERSION; 1117 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 1118 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1119 1120 /* XXX - We don't fill in the command line arguments properly yet. */ 1121 strlcpy(psinfo->pr_psargs, p->p_comm, sizeof(psinfo->pr_psargs)); 1122 1123 /* Fill in the header. */ 1124 bzero(hdr, hdrsize); 1125 off = 0; 1126 __elfN(puthdr)(p, hdr, &off, status, fpregset, psinfo, numsegs); 1127 1128 free(tempdata, M_TEMP); 1129 1130 /* Write it to the core file. */ 1131 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1132 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1133 td)); /* XXXKSE */ 1134 } 1135 1136 static void 1137 __elfN(puthdr)(struct proc *p, void *dst, size_t *off, const prstatus_t *status, 1138 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs) 1139 { 1140 size_t ehoff; 1141 size_t phoff; 1142 size_t noteoff; 1143 size_t notesz; 1144 1145 ehoff = *off; 1146 *off += sizeof(Elf_Ehdr); 1147 1148 phoff = *off; 1149 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1150 1151 noteoff = *off; 1152 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1153 sizeof *status); 1154 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1155 sizeof *fpregset); 1156 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1157 sizeof *psinfo); 1158 notesz = *off - noteoff; 1159 1160 /* Align up to a page boundary for the program segments. */ 1161 *off = round_page(*off); 1162 1163 if (dst != NULL) { 1164 Elf_Ehdr *ehdr; 1165 Elf_Phdr *phdr; 1166 struct phdr_closure phc; 1167 1168 /* 1169 * Fill in the ELF header. 1170 */ 1171 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1172 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1173 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1174 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1175 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1176 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1177 ehdr->e_ident[EI_DATA] = ELF_DATA; 1178 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1179 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1180 ehdr->e_ident[EI_ABIVERSION] = 0; 1181 ehdr->e_ident[EI_PAD] = 0; 1182 ehdr->e_type = ET_CORE; 1183 ehdr->e_machine = ELF_ARCH; 1184 ehdr->e_version = EV_CURRENT; 1185 ehdr->e_entry = 0; 1186 ehdr->e_phoff = phoff; 1187 ehdr->e_flags = 0; 1188 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1189 ehdr->e_phentsize = sizeof(Elf_Phdr); 1190 ehdr->e_phnum = numsegs + 1; 1191 ehdr->e_shentsize = sizeof(Elf_Shdr); 1192 ehdr->e_shnum = 0; 1193 ehdr->e_shstrndx = SHN_UNDEF; 1194 1195 /* 1196 * Fill in the program header entries. 1197 */ 1198 phdr = (Elf_Phdr *)((char *)dst + phoff); 1199 1200 /* The note segement. */ 1201 phdr->p_type = PT_NOTE; 1202 phdr->p_offset = noteoff; 1203 phdr->p_vaddr = 0; 1204 phdr->p_paddr = 0; 1205 phdr->p_filesz = notesz; 1206 phdr->p_memsz = 0; 1207 phdr->p_flags = 0; 1208 phdr->p_align = 0; 1209 phdr++; 1210 1211 /* All the writable segments from the program. */ 1212 phc.phdr = phdr; 1213 phc.offset = *off; 1214 each_writable_segment(p, cb_put_phdr, &phc); 1215 } 1216 } 1217 1218 static void 1219 __elfN(putnote)(void *dst, size_t *off, const char *name, int type, 1220 const void *desc, size_t descsz) 1221 { 1222 Elf_Note note; 1223 1224 note.n_namesz = strlen(name) + 1; 1225 note.n_descsz = descsz; 1226 note.n_type = type; 1227 if (dst != NULL) 1228 bcopy(¬e, (char *)dst + *off, sizeof note); 1229 *off += sizeof note; 1230 if (dst != NULL) 1231 bcopy(name, (char *)dst + *off, note.n_namesz); 1232 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1233 if (dst != NULL) 1234 bcopy(desc, (char *)dst + *off, note.n_descsz); 1235 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1236 } 1237 1238 /* 1239 * Tell kern_execve.c about it, with a little help from the linker. 1240 */ 1241 #if __ELF_WORD_SIZE == 32 1242 static struct execsw elf_execsw = {exec_elf32_imgact, "ELF32"}; 1243 EXEC_SET(elf32, elf_execsw); 1244 #else 1245 static struct execsw elf_execsw = {exec_elf64_imgact, "ELF64"}; 1246 EXEC_SET(elf64, elf_execsw); 1247 #endif 1248