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 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/exec.h> 36 #include <sys/fcntl.h> 37 #include <sys/imgact.h> 38 #include <sys/imgact_elf.h> 39 #include <sys/kernel.h> 40 #include <sys/lock.h> 41 #include <sys/malloc.h> 42 #include <sys/mutex.h> 43 #include <sys/mman.h> 44 #include <sys/namei.h> 45 #include <sys/pioctl.h> 46 #include <sys/proc.h> 47 #include <sys/procfs.h> 48 #include <sys/resourcevar.h> 49 #include <sys/systm.h> 50 #include <sys/signalvar.h> 51 #include <sys/stat.h> 52 #include <sys/sx.h> 53 #include <sys/syscall.h> 54 #include <sys/sysctl.h> 55 #include <sys/sysent.h> 56 #include <sys/vnode.h> 57 58 #include <vm/vm.h> 59 #include <vm/vm_kern.h> 60 #include <vm/vm_param.h> 61 #include <vm/pmap.h> 62 #include <vm/vm_map.h> 63 #include <vm/vm_object.h> 64 #include <vm/vm_extern.h> 65 66 #include <machine/elf.h> 67 #include <machine/md_var.h> 68 69 #define OLD_EI_BRAND 8 70 71 static int __elfN(check_header)(const Elf_Ehdr *hdr); 72 static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr, 73 const char *interp); 74 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 75 u_long *entry, size_t pagesize); 76 static int __elfN(load_section)(struct proc *p, 77 struct vmspace *vmspace, struct vnode *vp, vm_object_t object, 78 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 79 vm_prot_t prot, size_t pagesize); 80 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 81 82 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 83 ""); 84 85 int __elfN(fallback_brand) = -1; 86 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 87 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 88 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 89 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 90 &__elfN(fallback_brand)); 91 92 static int elf_trace = 0; 93 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, ""); 94 95 static int elf_legacy_coredump = 0; 96 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 97 &elf_legacy_coredump, 0, ""); 98 99 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 100 101 int 102 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 103 { 104 int i; 105 106 for (i = 0; i < MAX_BRANDS; i++) { 107 if (elf_brand_list[i] == NULL) { 108 elf_brand_list[i] = entry; 109 break; 110 } 111 } 112 if (i == MAX_BRANDS) 113 return (-1); 114 return (0); 115 } 116 117 int 118 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 119 { 120 int i; 121 122 for (i = 0; i < MAX_BRANDS; i++) { 123 if (elf_brand_list[i] == entry) { 124 elf_brand_list[i] = NULL; 125 break; 126 } 127 } 128 if (i == MAX_BRANDS) 129 return (-1); 130 return (0); 131 } 132 133 int 134 __elfN(brand_inuse)(Elf_Brandinfo *entry) 135 { 136 struct proc *p; 137 int rval = FALSE; 138 139 sx_slock(&allproc_lock); 140 LIST_FOREACH(p, &allproc, p_list) { 141 if (p->p_sysent == entry->sysvec) { 142 rval = TRUE; 143 break; 144 } 145 } 146 sx_sunlock(&allproc_lock); 147 148 return (rval); 149 } 150 151 static Elf_Brandinfo * 152 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp) 153 { 154 Elf_Brandinfo *bi; 155 int i; 156 157 /* 158 * We support three types of branding -- (1) the ELF EI_OSABI field 159 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 160 * branding w/in the ELF header, and (3) path of the `interp_path' 161 * field. We should also look for an ".note.ABI-tag" ELF section now 162 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones. 163 */ 164 165 /* If the executable has a brand, search for it in the brand list. */ 166 for (i = 0; i < MAX_BRANDS; i++) { 167 bi = elf_brand_list[i]; 168 if (bi != NULL && hdr->e_machine == bi->machine && 169 (hdr->e_ident[EI_OSABI] == bi->brand || 170 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 171 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 172 return (bi); 173 } 174 175 /* Lacking a known brand, search for a recognized interpreter. */ 176 if (interp != NULL) { 177 for (i = 0; i < MAX_BRANDS; i++) { 178 bi = elf_brand_list[i]; 179 if (bi != NULL && hdr->e_machine == bi->machine && 180 strcmp(interp, bi->interp_path) == 0) 181 return (bi); 182 } 183 } 184 185 /* Lacking a recognized interpreter, try the default brand */ 186 for (i = 0; i < MAX_BRANDS; i++) { 187 bi = elf_brand_list[i]; 188 if (bi != NULL && hdr->e_machine == bi->machine && 189 __elfN(fallback_brand) == bi->brand) 190 return (bi); 191 } 192 return (NULL); 193 } 194 195 static int 196 __elfN(check_header)(const Elf_Ehdr *hdr) 197 { 198 Elf_Brandinfo *bi; 199 int i; 200 201 if (!IS_ELF(*hdr) || 202 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 203 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 204 hdr->e_ident[EI_VERSION] != EV_CURRENT) 205 return (ENOEXEC); 206 207 /* 208 * Make sure we have at least one brand for this machine. 209 */ 210 211 for (i = 0; i < MAX_BRANDS; i++) { 212 bi = elf_brand_list[i]; 213 if (bi != NULL && bi->machine == hdr->e_machine) 214 break; 215 } 216 if (i == MAX_BRANDS) 217 return (ENOEXEC); 218 219 if (hdr->e_version != ELF_TARG_VER) 220 return (ENOEXEC); 221 222 return (0); 223 } 224 225 static int 226 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 227 vm_offset_t start, vm_offset_t end, vm_prot_t prot, 228 vm_prot_t max) 229 { 230 int error, rv; 231 vm_offset_t off; 232 vm_offset_t data_buf = 0; 233 234 /* 235 * Create the page if it doesn't exist yet. Ignore errors. 236 */ 237 vm_map_lock(map); 238 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), max, 239 max, 0); 240 vm_map_unlock(map); 241 242 /* 243 * Find the page from the underlying object. 244 */ 245 if (object) { 246 vm_object_reference(object); 247 rv = vm_map_find(exec_map, 248 object, 249 trunc_page(offset), 250 &data_buf, 251 PAGE_SIZE, 252 TRUE, 253 VM_PROT_READ, 254 VM_PROT_ALL, 255 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 256 if (rv != KERN_SUCCESS) { 257 vm_object_deallocate(object); 258 return (rv); 259 } 260 261 off = offset - trunc_page(offset); 262 error = copyout((caddr_t)data_buf + off, (caddr_t)start, 263 end - start); 264 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 265 if (error) { 266 return (KERN_FAILURE); 267 } 268 } 269 270 return (KERN_SUCCESS); 271 } 272 273 static int 274 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 275 vm_offset_t start, vm_offset_t end, vm_prot_t prot, 276 vm_prot_t max, int cow) 277 { 278 vm_offset_t data_buf, off; 279 vm_size_t sz; 280 int error, rv; 281 282 if (start != trunc_page(start)) { 283 rv = __elfN(map_partial)(map, object, offset, start, 284 round_page(start), prot, max); 285 if (rv) 286 return (rv); 287 offset += round_page(start) - start; 288 start = round_page(start); 289 } 290 if (end != round_page(end)) { 291 rv = __elfN(map_partial)(map, object, offset + 292 trunc_page(end) - start, trunc_page(end), end, prot, max); 293 if (rv) 294 return (rv); 295 end = trunc_page(end); 296 } 297 if (end > start) { 298 if (offset & PAGE_MASK) { 299 /* 300 * The mapping is not page aligned. This means we have 301 * to copy the data. Sigh. 302 */ 303 rv = vm_map_find(map, 0, 0, &start, end - start, 304 FALSE, prot, max, 0); 305 if (rv) 306 return (rv); 307 data_buf = 0; 308 while (start < end) { 309 vm_object_reference(object); 310 rv = vm_map_find(exec_map, 311 object, 312 trunc_page(offset), 313 &data_buf, 314 2 * PAGE_SIZE, 315 TRUE, 316 VM_PROT_READ, 317 VM_PROT_ALL, 318 (MAP_COPY_ON_WRITE 319 | MAP_PREFAULT_PARTIAL)); 320 if (rv != KERN_SUCCESS) { 321 vm_object_deallocate(object); 322 return (rv); 323 } 324 off = offset - trunc_page(offset); 325 sz = end - start; 326 if (sz > PAGE_SIZE) 327 sz = PAGE_SIZE; 328 error = copyout((caddr_t)data_buf + off, 329 (caddr_t)start, sz); 330 vm_map_remove(exec_map, data_buf, 331 data_buf + 2 * PAGE_SIZE); 332 if (error) { 333 return (KERN_FAILURE); 334 } 335 start += sz; 336 } 337 rv = KERN_SUCCESS; 338 } else { 339 vm_map_lock(map); 340 rv = vm_map_insert(map, object, offset, start, end, 341 prot, max, cow); 342 vm_map_unlock(map); 343 } 344 return (rv); 345 } else { 346 return (KERN_SUCCESS); 347 } 348 } 349 350 static int 351 __elfN(load_section)(struct proc *p, struct vmspace *vmspace, 352 struct vnode *vp, vm_object_t object, vm_offset_t offset, 353 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 354 size_t pagesize) 355 { 356 size_t map_len; 357 vm_offset_t map_addr; 358 int error, rv, cow; 359 size_t copy_len; 360 vm_offset_t file_addr; 361 vm_offset_t data_buf = 0; 362 363 GIANT_REQUIRED; 364 365 error = 0; 366 367 /* 368 * It's necessary to fail if the filsz + offset taken from the 369 * header is greater than the actual file pager object's size. 370 * If we were to allow this, then the vm_map_find() below would 371 * walk right off the end of the file object and into the ether. 372 * 373 * While I'm here, might as well check for something else that 374 * is invalid: filsz cannot be greater than memsz. 375 */ 376 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 377 filsz > memsz) { 378 uprintf("elf_load_section: truncated ELF file\n"); 379 return (ENOEXEC); 380 } 381 382 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 383 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 384 385 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 386 file_addr = trunc_page_ps(offset, pagesize); 387 388 /* 389 * We have two choices. We can either clear the data in the last page 390 * of an oversized mapping, or we can start the anon mapping a page 391 * early and copy the initialized data into that first page. We 392 * choose the second.. 393 */ 394 if (memsz > filsz) 395 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 396 else 397 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 398 399 if (map_len != 0) { 400 vm_object_reference(object); 401 402 /* cow flags: don't dump readonly sections in core */ 403 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 404 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 405 406 rv = __elfN(map_insert)(&vmspace->vm_map, 407 object, 408 file_addr, /* file offset */ 409 map_addr, /* virtual start */ 410 map_addr + map_len,/* virtual end */ 411 prot, 412 VM_PROT_ALL, 413 cow); 414 if (rv != KERN_SUCCESS) { 415 vm_object_deallocate(object); 416 return (EINVAL); 417 } 418 419 /* we can stop now if we've covered it all */ 420 if (memsz == filsz) { 421 return (0); 422 } 423 } 424 425 426 /* 427 * We have to get the remaining bit of the file into the first part 428 * of the oversized map segment. This is normally because the .data 429 * segment in the file is extended to provide bss. It's a neat idea 430 * to try and save a page, but it's a pain in the behind to implement. 431 */ 432 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 433 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 434 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 435 map_addr; 436 437 /* This had damn well better be true! */ 438 if (map_len != 0) { 439 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr, 440 map_addr + map_len, VM_PROT_ALL, VM_PROT_ALL, 0); 441 if (rv != KERN_SUCCESS) { 442 return (EINVAL); 443 } 444 } 445 446 if (copy_len != 0) { 447 vm_offset_t off; 448 vm_object_reference(object); 449 rv = vm_map_find(exec_map, 450 object, 451 trunc_page(offset + filsz), 452 &data_buf, 453 PAGE_SIZE, 454 TRUE, 455 VM_PROT_READ, 456 VM_PROT_ALL, 457 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 458 if (rv != KERN_SUCCESS) { 459 vm_object_deallocate(object); 460 return (EINVAL); 461 } 462 463 /* send the page fragment to user space */ 464 off = trunc_page_ps(offset + filsz, pagesize) - 465 trunc_page(offset + filsz); 466 error = copyout((caddr_t)data_buf + off, (caddr_t)map_addr, 467 copy_len); 468 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 469 if (error) { 470 return (error); 471 } 472 } 473 474 /* 475 * set it to the specified protection. 476 * XXX had better undo the damage from pasting over the cracks here! 477 */ 478 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr), 479 round_page(map_addr + map_len), prot, FALSE); 480 481 return (error); 482 } 483 484 /* 485 * Load the file "file" into memory. It may be either a shared object 486 * or an executable. 487 * 488 * The "addr" reference parameter is in/out. On entry, it specifies 489 * the address where a shared object should be loaded. If the file is 490 * an executable, this value is ignored. On exit, "addr" specifies 491 * where the file was actually loaded. 492 * 493 * The "entry" reference parameter is out only. On exit, it specifies 494 * the entry point for the loaded file. 495 */ 496 static int 497 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 498 u_long *entry, size_t pagesize) 499 { 500 struct { 501 struct nameidata nd; 502 struct vattr attr; 503 struct image_params image_params; 504 } *tempdata; 505 const Elf_Ehdr *hdr = NULL; 506 const Elf_Phdr *phdr = NULL; 507 struct nameidata *nd; 508 struct vmspace *vmspace = p->p_vmspace; 509 struct vattr *attr; 510 struct image_params *imgp; 511 vm_prot_t prot; 512 u_long rbase; 513 u_long base_addr = 0; 514 int error, i, numsegs; 515 516 if (curthread->td_proc != p) 517 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */ 518 519 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 520 nd = &tempdata->nd; 521 attr = &tempdata->attr; 522 imgp = &tempdata->image_params; 523 524 /* 525 * Initialize part of the common data 526 */ 527 imgp->proc = p; 528 imgp->userspace_argv = NULL; 529 imgp->userspace_envv = NULL; 530 imgp->attr = attr; 531 imgp->firstpage = NULL; 532 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE); 533 imgp->object = NULL; 534 imgp->execlabel = NULL; 535 536 if (imgp->image_header == NULL) { 537 nd->ni_vp = NULL; 538 error = ENOMEM; 539 goto fail; 540 } 541 542 /* XXXKSE */ 543 NDINIT(nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, curthread); 544 545 if ((error = namei(nd)) != 0) { 546 nd->ni_vp = NULL; 547 goto fail; 548 } 549 NDFREE(nd, NDF_ONLY_PNBUF); 550 imgp->vp = nd->ni_vp; 551 552 /* 553 * Check permissions, modes, uid, etc on the file, and "open" it. 554 */ 555 error = exec_check_permissions(imgp); 556 if (error) { 557 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */ 558 goto fail; 559 } 560 561 error = exec_map_first_page(imgp); 562 /* 563 * Also make certain that the interpreter stays the same, so set 564 * its VV_TEXT flag, too. 565 */ 566 if (error == 0) 567 nd->ni_vp->v_vflag |= VV_TEXT; 568 569 VOP_GETVOBJECT(nd->ni_vp, &imgp->object); 570 vm_object_reference(imgp->object); 571 572 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */ 573 if (error) 574 goto fail; 575 576 hdr = (const Elf_Ehdr *)imgp->image_header; 577 if ((error = __elfN(check_header)(hdr)) != 0) 578 goto fail; 579 if (hdr->e_type == ET_DYN) 580 rbase = *addr; 581 else if (hdr->e_type == ET_EXEC) 582 rbase = 0; 583 else { 584 error = ENOEXEC; 585 goto fail; 586 } 587 588 /* Only support headers that fit within first page for now */ 589 if ((hdr->e_phoff > PAGE_SIZE) || 590 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 591 error = ENOEXEC; 592 goto fail; 593 } 594 595 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 596 597 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 598 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 599 prot = 0; 600 if (phdr[i].p_flags & PF_X) 601 prot |= VM_PROT_EXECUTE; 602 if (phdr[i].p_flags & PF_W) 603 prot |= VM_PROT_WRITE; 604 if (phdr[i].p_flags & PF_R) 605 prot |= VM_PROT_READ; 606 607 if ((error = __elfN(load_section)(p, vmspace, 608 nd->ni_vp, imgp->object, phdr[i].p_offset, 609 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 610 phdr[i].p_memsz, phdr[i].p_filesz, prot, 611 pagesize)) != 0) 612 goto fail; 613 /* 614 * Establish the base address if this is the 615 * first segment. 616 */ 617 if (numsegs == 0) 618 base_addr = trunc_page(phdr[i].p_vaddr + 619 rbase); 620 numsegs++; 621 } 622 } 623 *addr = base_addr; 624 *entry = (unsigned long)hdr->e_entry + rbase; 625 626 fail: 627 if (imgp->firstpage) 628 exec_unmap_first_page(imgp); 629 if (imgp->image_header) 630 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header, 631 PAGE_SIZE); 632 if (imgp->object) 633 vm_object_deallocate(imgp->object); 634 635 if (nd->ni_vp) 636 vrele(nd->ni_vp); 637 638 free(tempdata, M_TEMP); 639 640 return (error); 641 } 642 643 static int 644 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 645 { 646 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 647 const Elf_Phdr *phdr; 648 Elf_Auxargs *elf_auxargs = NULL; 649 struct vmspace *vmspace; 650 vm_prot_t prot; 651 u_long text_size = 0, data_size = 0, total_size = 0; 652 u_long text_addr = 0, data_addr = 0; 653 u_long seg_size, seg_addr; 654 u_long addr, entry = 0, proghdr = 0; 655 int error, i; 656 const char *interp = NULL; 657 Elf_Brandinfo *brand_info; 658 char *path; 659 struct thread *td = curthread; 660 struct sysentvec *sv; 661 662 GIANT_REQUIRED; 663 664 /* 665 * Do we have a valid ELF header ? 666 */ 667 if (__elfN(check_header)(hdr) != 0 || hdr->e_type != ET_EXEC) 668 return (-1); 669 670 /* 671 * From here on down, we return an errno, not -1, as we've 672 * detected an ELF file. 673 */ 674 675 if ((hdr->e_phoff > PAGE_SIZE) || 676 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 677 /* Only support headers in first page for now */ 678 return (ENOEXEC); 679 } 680 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 681 682 /* 683 * From this point on, we may have resources that need to be freed. 684 */ 685 686 VOP_UNLOCK(imgp->vp, 0, td); 687 688 for (i = 0; i < hdr->e_phnum; i++) { 689 switch (phdr[i].p_type) { 690 case PT_INTERP: /* Path to interpreter */ 691 if (phdr[i].p_filesz > MAXPATHLEN || 692 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) { 693 error = ENOEXEC; 694 goto fail; 695 } 696 interp = imgp->image_header + phdr[i].p_offset; 697 break; 698 default: 699 break; 700 } 701 } 702 703 brand_info = __elfN(get_brandinfo)(hdr, interp); 704 if (brand_info == NULL) { 705 uprintf("ELF binary type \"%u\" not known.\n", 706 hdr->e_ident[EI_OSABI]); 707 error = ENOEXEC; 708 goto fail; 709 } 710 sv = brand_info->sysvec; 711 712 if ((error = exec_extract_strings(imgp)) != 0) 713 goto fail; 714 715 exec_new_vmspace(imgp, sv); 716 717 vmspace = imgp->proc->p_vmspace; 718 719 for (i = 0; i < hdr->e_phnum; i++) { 720 switch (phdr[i].p_type) { 721 case PT_LOAD: /* Loadable segment */ 722 prot = 0; 723 if (phdr[i].p_flags & PF_X) 724 prot |= VM_PROT_EXECUTE; 725 if (phdr[i].p_flags & PF_W) 726 prot |= VM_PROT_WRITE; 727 if (phdr[i].p_flags & PF_R) 728 prot |= VM_PROT_READ; 729 730 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER) 731 /* 732 * Some x86 binaries assume read == executable, 733 * notably the M3 runtime and therefore cvsup 734 */ 735 if (prot & VM_PROT_READ) 736 prot |= VM_PROT_EXECUTE; 737 #endif 738 739 if ((error = __elfN(load_section)(imgp->proc, vmspace, 740 imgp->vp, imgp->object, phdr[i].p_offset, 741 (caddr_t)(uintptr_t)phdr[i].p_vaddr, 742 phdr[i].p_memsz, phdr[i].p_filesz, prot, 743 sv->sv_pagesize)) != 0) 744 goto fail; 745 746 seg_addr = trunc_page(phdr[i].p_vaddr); 747 seg_size = round_page(phdr[i].p_memsz + 748 phdr[i].p_vaddr - seg_addr); 749 750 /* 751 * Is this .text or .data? We can't use 752 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 753 * alpha terribly and possibly does other bad 754 * things so we stick to the old way of figuring 755 * it out: If the segment contains the program 756 * entry point, it's a text segment, otherwise it 757 * is a data segment. 758 * 759 * Note that obreak() assumes that data_addr + 760 * data_size == end of data load area, and the ELF 761 * file format expects segments to be sorted by 762 * address. If multiple data segments exist, the 763 * last one will be used. 764 */ 765 if (hdr->e_entry >= phdr[i].p_vaddr && 766 hdr->e_entry < (phdr[i].p_vaddr + 767 phdr[i].p_memsz)) { 768 text_size = seg_size; 769 text_addr = seg_addr; 770 entry = (u_long)hdr->e_entry; 771 } else { 772 data_size = seg_size; 773 data_addr = seg_addr; 774 } 775 total_size += seg_size; 776 break; 777 case PT_PHDR: /* Program header table info */ 778 proghdr = phdr[i].p_vaddr; 779 break; 780 default: 781 break; 782 } 783 } 784 785 if (data_addr == 0 && data_size == 0) { 786 data_addr = text_addr; 787 data_size = text_size; 788 } 789 790 /* 791 * Check limits. It should be safe to check the 792 * limits after loading the segments since we do 793 * not actually fault in all the segments pages. 794 */ 795 if (data_size > 796 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur || 797 text_size > maxtsiz || 798 total_size > 799 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) { 800 error = ENOMEM; 801 goto fail; 802 } 803 804 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 805 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 806 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 807 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 808 809 /* 810 * We load the dynamic linker where a userland call 811 * to mmap(0, ...) would put it. The rationale behind this 812 * calculation is that it leaves room for the heap to grow to 813 * its maximum allowed size. 814 */ 815 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr + 816 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_max); 817 818 imgp->entry_addr = entry; 819 820 imgp->proc->p_sysent = sv; 821 if (interp != NULL) { 822 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 823 snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path, 824 interp); 825 if ((error = __elfN(load_file)(imgp->proc, path, &addr, 826 &imgp->entry_addr, sv->sv_pagesize)) != 0) { 827 if ((error = __elfN(load_file)(imgp->proc, interp, 828 &addr, &imgp->entry_addr, sv->sv_pagesize)) != 0) { 829 uprintf("ELF interpreter %s not found\n", 830 path); 831 free(path, M_TEMP); 832 goto fail; 833 } 834 } 835 free(path, M_TEMP); 836 } 837 838 /* 839 * Construct auxargs table (used by the fixup routine) 840 */ 841 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 842 elf_auxargs->execfd = -1; 843 elf_auxargs->phdr = proghdr; 844 elf_auxargs->phent = hdr->e_phentsize; 845 elf_auxargs->phnum = hdr->e_phnum; 846 elf_auxargs->pagesz = PAGE_SIZE; 847 elf_auxargs->base = addr; 848 elf_auxargs->flags = 0; 849 elf_auxargs->entry = entry; 850 elf_auxargs->trace = elf_trace; 851 852 imgp->auxargs = elf_auxargs; 853 imgp->interpreted = 0; 854 855 fail: 856 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td); 857 return (error); 858 } 859 860 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 861 862 int 863 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 864 { 865 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 866 Elf_Addr *base; 867 Elf_Addr *pos; 868 869 base = (Elf_Addr *)*stack_base; 870 pos = base + (imgp->argc + imgp->envc + 2); 871 872 if (args->trace) { 873 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 874 } 875 if (args->execfd != -1) { 876 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 877 } 878 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 879 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 880 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 881 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 882 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 883 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 884 AUXARGS_ENTRY(pos, AT_BASE, args->base); 885 AUXARGS_ENTRY(pos, AT_NULL, 0); 886 887 free(imgp->auxargs, M_TEMP); 888 imgp->auxargs = NULL; 889 890 base--; 891 suword(base, (long)imgp->argc); 892 *stack_base = (register_t *)base; 893 return (0); 894 } 895 896 /* 897 * Code for generating ELF core dumps. 898 */ 899 900 typedef void (*segment_callback)(vm_map_entry_t, void *); 901 902 /* Closure for cb_put_phdr(). */ 903 struct phdr_closure { 904 Elf_Phdr *phdr; /* Program header to fill in */ 905 Elf_Off offset; /* Offset of segment in core file */ 906 }; 907 908 /* Closure for cb_size_segment(). */ 909 struct sseg_closure { 910 int count; /* Count of writable segments. */ 911 size_t size; /* Total size of all writable segments. */ 912 }; 913 914 static void cb_put_phdr(vm_map_entry_t, void *); 915 static void cb_size_segment(vm_map_entry_t, void *); 916 static void each_writable_segment(struct proc *, segment_callback, void *); 917 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 918 int, void *, size_t); 919 static void __elfN(puthdr)(struct proc *, void *, size_t *, 920 const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int); 921 static void __elfN(putnote)(void *, size_t *, const char *, int, 922 const void *, size_t); 923 924 extern int osreldate; 925 926 int 927 __elfN(coredump)(td, vp, limit) 928 struct thread *td; 929 register struct vnode *vp; 930 off_t limit; 931 { 932 register struct proc *p = td->td_proc; 933 register struct ucred *cred = td->td_ucred; 934 int error = 0; 935 struct sseg_closure seginfo; 936 void *hdr; 937 size_t hdrsize; 938 939 /* Size the program segments. */ 940 seginfo.count = 0; 941 seginfo.size = 0; 942 each_writable_segment(p, cb_size_segment, &seginfo); 943 944 /* 945 * Calculate the size of the core file header area by making 946 * a dry run of generating it. Nothing is written, but the 947 * size is calculated. 948 */ 949 hdrsize = 0; 950 __elfN(puthdr)((struct proc *)NULL, (void *)NULL, &hdrsize, 951 (const prstatus_t *)NULL, (const prfpregset_t *)NULL, 952 (const prpsinfo_t *)NULL, seginfo.count); 953 954 if (hdrsize + seginfo.size >= limit) 955 return (EFAULT); 956 957 /* 958 * Allocate memory for building the header, fill it up, 959 * and write it out. 960 */ 961 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 962 if (hdr == NULL) { 963 return (EINVAL); 964 } 965 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize); 966 967 /* Write the contents of all of the writable segments. */ 968 if (error == 0) { 969 Elf_Phdr *php; 970 off_t offset; 971 int i; 972 973 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 974 offset = hdrsize; 975 for (i = 0; i < seginfo.count; i++) { 976 error = vn_rdwr_inchunks(UIO_WRITE, vp, 977 (caddr_t)(uintptr_t)php->p_vaddr, 978 php->p_filesz, offset, UIO_USERSPACE, 979 IO_UNIT | IO_DIRECT, cred, NOCRED, (int *)NULL, 980 curthread); /* XXXKSE */ 981 if (error != 0) 982 break; 983 offset += php->p_filesz; 984 php++; 985 } 986 } 987 free(hdr, M_TEMP); 988 989 return (error); 990 } 991 992 /* 993 * A callback for each_writable_segment() to write out the segment's 994 * program header entry. 995 */ 996 static void 997 cb_put_phdr(entry, closure) 998 vm_map_entry_t entry; 999 void *closure; 1000 { 1001 struct phdr_closure *phc = (struct phdr_closure *)closure; 1002 Elf_Phdr *phdr = phc->phdr; 1003 1004 phc->offset = round_page(phc->offset); 1005 1006 phdr->p_type = PT_LOAD; 1007 phdr->p_offset = phc->offset; 1008 phdr->p_vaddr = entry->start; 1009 phdr->p_paddr = 0; 1010 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1011 phdr->p_align = PAGE_SIZE; 1012 phdr->p_flags = 0; 1013 if (entry->protection & VM_PROT_READ) 1014 phdr->p_flags |= PF_R; 1015 if (entry->protection & VM_PROT_WRITE) 1016 phdr->p_flags |= PF_W; 1017 if (entry->protection & VM_PROT_EXECUTE) 1018 phdr->p_flags |= PF_X; 1019 1020 phc->offset += phdr->p_filesz; 1021 phc->phdr++; 1022 } 1023 1024 /* 1025 * A callback for each_writable_segment() to gather information about 1026 * the number of segments and their total size. 1027 */ 1028 static void 1029 cb_size_segment(entry, closure) 1030 vm_map_entry_t entry; 1031 void *closure; 1032 { 1033 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1034 1035 ssc->count++; 1036 ssc->size += entry->end - entry->start; 1037 } 1038 1039 /* 1040 * For each writable segment in the process's memory map, call the given 1041 * function with a pointer to the map entry and some arbitrary 1042 * caller-supplied data. 1043 */ 1044 static void 1045 each_writable_segment(p, func, closure) 1046 struct proc *p; 1047 segment_callback func; 1048 void *closure; 1049 { 1050 vm_map_t map = &p->p_vmspace->vm_map; 1051 vm_map_entry_t entry; 1052 1053 for (entry = map->header.next; entry != &map->header; 1054 entry = entry->next) { 1055 vm_object_t obj; 1056 1057 /* 1058 * Don't dump inaccessible mappings, deal with legacy 1059 * coredump mode. 1060 * 1061 * Note that read-only segments related to the elf binary 1062 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1063 * need to arbitrarily ignore such segments. 1064 */ 1065 if (elf_legacy_coredump) { 1066 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1067 continue; 1068 } else { 1069 if ((entry->protection & VM_PROT_ALL) == 0) 1070 continue; 1071 } 1072 1073 /* 1074 * Dont include memory segment in the coredump if 1075 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1076 * madvise(2). Do not dump submaps (i.e. parts of the 1077 * kernel map). 1078 */ 1079 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1080 continue; 1081 1082 if ((obj = entry->object.vm_object) == NULL) 1083 continue; 1084 1085 /* Find the deepest backing object. */ 1086 while (obj->backing_object != NULL) 1087 obj = obj->backing_object; 1088 1089 /* Ignore memory-mapped devices and such things. */ 1090 if (obj->type != OBJT_DEFAULT && 1091 obj->type != OBJT_SWAP && 1092 obj->type != OBJT_VNODE) 1093 continue; 1094 1095 (*func)(entry, closure); 1096 } 1097 } 1098 1099 /* 1100 * Write the core file header to the file, including padding up to 1101 * the page boundary. 1102 */ 1103 static int 1104 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize) 1105 struct thread *td; 1106 struct vnode *vp; 1107 struct ucred *cred; 1108 int numsegs; 1109 size_t hdrsize; 1110 void *hdr; 1111 { 1112 struct { 1113 prstatus_t status; 1114 prfpregset_t fpregset; 1115 prpsinfo_t psinfo; 1116 } *tempdata; 1117 struct proc *p = td->td_proc; 1118 size_t off; 1119 prstatus_t *status; 1120 prfpregset_t *fpregset; 1121 prpsinfo_t *psinfo; 1122 1123 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK); 1124 status = &tempdata->status; 1125 fpregset = &tempdata->fpregset; 1126 psinfo = &tempdata->psinfo; 1127 1128 /* Gather the information for the header. */ 1129 status->pr_version = PRSTATUS_VERSION; 1130 status->pr_statussz = sizeof(prstatus_t); 1131 status->pr_gregsetsz = sizeof(gregset_t); 1132 status->pr_fpregsetsz = sizeof(fpregset_t); 1133 status->pr_osreldate = osreldate; 1134 status->pr_cursig = p->p_sig; 1135 status->pr_pid = p->p_pid; 1136 fill_regs(td, &status->pr_reg); 1137 1138 fill_fpregs(td, fpregset); 1139 1140 psinfo->pr_version = PRPSINFO_VERSION; 1141 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 1142 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1143 1144 /* XXX - We don't fill in the command line arguments properly yet. */ 1145 strlcpy(psinfo->pr_psargs, p->p_comm, sizeof(psinfo->pr_psargs)); 1146 1147 /* Fill in the header. */ 1148 bzero(hdr, hdrsize); 1149 off = 0; 1150 __elfN(puthdr)(p, hdr, &off, status, fpregset, psinfo, numsegs); 1151 1152 free(tempdata, M_TEMP); 1153 1154 /* Write it to the core file. */ 1155 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1156 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1157 td)); /* XXXKSE */ 1158 } 1159 1160 static void 1161 __elfN(puthdr)(struct proc *p, void *dst, size_t *off, const prstatus_t *status, 1162 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs) 1163 { 1164 size_t ehoff; 1165 size_t phoff; 1166 size_t noteoff; 1167 size_t notesz; 1168 1169 ehoff = *off; 1170 *off += sizeof(Elf_Ehdr); 1171 1172 phoff = *off; 1173 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1174 1175 noteoff = *off; 1176 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1177 sizeof *status); 1178 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1179 sizeof *fpregset); 1180 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1181 sizeof *psinfo); 1182 notesz = *off - noteoff; 1183 1184 /* Align up to a page boundary for the program segments. */ 1185 *off = round_page(*off); 1186 1187 if (dst != NULL) { 1188 Elf_Ehdr *ehdr; 1189 Elf_Phdr *phdr; 1190 struct phdr_closure phc; 1191 1192 /* 1193 * Fill in the ELF header. 1194 */ 1195 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1196 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1197 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1198 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1199 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1200 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1201 ehdr->e_ident[EI_DATA] = ELF_DATA; 1202 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1203 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1204 ehdr->e_ident[EI_ABIVERSION] = 0; 1205 ehdr->e_ident[EI_PAD] = 0; 1206 ehdr->e_type = ET_CORE; 1207 ehdr->e_machine = ELF_ARCH; 1208 ehdr->e_version = EV_CURRENT; 1209 ehdr->e_entry = 0; 1210 ehdr->e_phoff = phoff; 1211 ehdr->e_flags = 0; 1212 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1213 ehdr->e_phentsize = sizeof(Elf_Phdr); 1214 ehdr->e_phnum = numsegs + 1; 1215 ehdr->e_shentsize = sizeof(Elf_Shdr); 1216 ehdr->e_shnum = 0; 1217 ehdr->e_shstrndx = SHN_UNDEF; 1218 1219 /* 1220 * Fill in the program header entries. 1221 */ 1222 phdr = (Elf_Phdr *)((char *)dst + phoff); 1223 1224 /* The note segement. */ 1225 phdr->p_type = PT_NOTE; 1226 phdr->p_offset = noteoff; 1227 phdr->p_vaddr = 0; 1228 phdr->p_paddr = 0; 1229 phdr->p_filesz = notesz; 1230 phdr->p_memsz = 0; 1231 phdr->p_flags = 0; 1232 phdr->p_align = 0; 1233 phdr++; 1234 1235 /* All the writable segments from the program. */ 1236 phc.phdr = phdr; 1237 phc.offset = *off; 1238 each_writable_segment(p, cb_put_phdr, &phc); 1239 } 1240 } 1241 1242 static void 1243 __elfN(putnote)(void *dst, size_t *off, const char *name, int type, 1244 const void *desc, size_t descsz) 1245 { 1246 Elf_Note note; 1247 1248 note.n_namesz = strlen(name) + 1; 1249 note.n_descsz = descsz; 1250 note.n_type = type; 1251 if (dst != NULL) 1252 bcopy(¬e, (char *)dst + *off, sizeof note); 1253 *off += sizeof note; 1254 if (dst != NULL) 1255 bcopy(name, (char *)dst + *off, note.n_namesz); 1256 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1257 if (dst != NULL) 1258 bcopy(desc, (char *)dst + *off, note.n_descsz); 1259 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1260 } 1261 1262 /* 1263 * Tell kern_execve.c about it, with a little help from the linker. 1264 */ 1265 static struct execsw __elfN(execsw) = { 1266 __CONCAT(exec_, __elfN(imgact)), 1267 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 1268 }; 1269 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 1270