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 hdr->e_phentsize != sizeof(Elf_Phdr) || 206 hdr->e_version != ELF_TARG_VER) 207 return (ENOEXEC); 208 209 /* 210 * Make sure we have at least one brand for this machine. 211 */ 212 213 for (i = 0; i < MAX_BRANDS; i++) { 214 bi = elf_brand_list[i]; 215 if (bi != NULL && bi->machine == hdr->e_machine) 216 break; 217 } 218 if (i == MAX_BRANDS) 219 return (ENOEXEC); 220 221 return (0); 222 } 223 224 static int 225 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 226 vm_offset_t start, vm_offset_t end, vm_prot_t prot, 227 vm_prot_t max) 228 { 229 int error, rv; 230 vm_offset_t off; 231 vm_offset_t data_buf = 0; 232 233 /* 234 * Create the page if it doesn't exist yet. Ignore errors. 235 */ 236 vm_map_lock(map); 237 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), max, 238 max, 0); 239 vm_map_unlock(map); 240 241 /* 242 * Find the page from the underlying object. 243 */ 244 if (object) { 245 vm_object_reference(object); 246 rv = vm_map_find(exec_map, 247 object, 248 trunc_page(offset), 249 &data_buf, 250 PAGE_SIZE, 251 TRUE, 252 VM_PROT_READ, 253 VM_PROT_ALL, 254 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 255 if (rv != KERN_SUCCESS) { 256 vm_object_deallocate(object); 257 return (rv); 258 } 259 260 off = offset - trunc_page(offset); 261 error = copyout((caddr_t)data_buf + off, (caddr_t)start, 262 end - start); 263 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 264 if (error) { 265 return (KERN_FAILURE); 266 } 267 } 268 269 return (KERN_SUCCESS); 270 } 271 272 static int 273 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 274 vm_offset_t start, vm_offset_t end, vm_prot_t prot, 275 vm_prot_t max, int cow) 276 { 277 vm_offset_t data_buf, off; 278 vm_size_t sz; 279 int error, rv; 280 281 if (start != trunc_page(start)) { 282 rv = __elfN(map_partial)(map, object, offset, start, 283 round_page(start), prot, max); 284 if (rv) 285 return (rv); 286 offset += round_page(start) - start; 287 start = round_page(start); 288 } 289 if (end != round_page(end)) { 290 rv = __elfN(map_partial)(map, object, offset + 291 trunc_page(end) - start, trunc_page(end), end, prot, max); 292 if (rv) 293 return (rv); 294 end = trunc_page(end); 295 } 296 if (end > start) { 297 if (offset & PAGE_MASK) { 298 /* 299 * The mapping is not page aligned. This means we have 300 * to copy the data. Sigh. 301 */ 302 rv = vm_map_find(map, 0, 0, &start, end - start, 303 FALSE, prot, max, 0); 304 if (rv) 305 return (rv); 306 data_buf = 0; 307 while (start < end) { 308 vm_object_reference(object); 309 rv = vm_map_find(exec_map, 310 object, 311 trunc_page(offset), 312 &data_buf, 313 2 * PAGE_SIZE, 314 TRUE, 315 VM_PROT_READ, 316 VM_PROT_ALL, 317 (MAP_COPY_ON_WRITE 318 | MAP_PREFAULT_PARTIAL)); 319 if (rv != KERN_SUCCESS) { 320 vm_object_deallocate(object); 321 return (rv); 322 } 323 off = offset - trunc_page(offset); 324 sz = end - start; 325 if (sz > PAGE_SIZE) 326 sz = PAGE_SIZE; 327 error = copyout((caddr_t)data_buf + off, 328 (caddr_t)start, sz); 329 vm_map_remove(exec_map, data_buf, 330 data_buf + 2 * PAGE_SIZE); 331 if (error) { 332 return (KERN_FAILURE); 333 } 334 start += sz; 335 } 336 rv = KERN_SUCCESS; 337 } else { 338 vm_map_lock(map); 339 rv = vm_map_insert(map, object, offset, start, end, 340 prot, max, cow); 341 vm_map_unlock(map); 342 } 343 return (rv); 344 } else { 345 return (KERN_SUCCESS); 346 } 347 } 348 349 static int 350 __elfN(load_section)(struct proc *p, struct vmspace *vmspace, 351 struct vnode *vp, vm_object_t object, vm_offset_t offset, 352 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 353 size_t pagesize) 354 { 355 size_t map_len; 356 vm_offset_t map_addr; 357 int error, rv, cow; 358 size_t copy_len; 359 vm_offset_t file_addr; 360 vm_offset_t data_buf = 0; 361 362 GIANT_REQUIRED; 363 364 error = 0; 365 366 /* 367 * It's necessary to fail if the filsz + offset taken from the 368 * header is greater than the actual file pager object's size. 369 * If we were to allow this, then the vm_map_find() below would 370 * walk right off the end of the file object and into the ether. 371 * 372 * While I'm here, might as well check for something else that 373 * is invalid: filsz cannot be greater than memsz. 374 */ 375 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 376 filsz > memsz) { 377 uprintf("elf_load_section: truncated ELF file\n"); 378 return (ENOEXEC); 379 } 380 381 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 382 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 383 384 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 385 file_addr = trunc_page_ps(offset, pagesize); 386 387 /* 388 * We have two choices. We can either clear the data in the last page 389 * of an oversized mapping, or we can start the anon mapping a page 390 * early and copy the initialized data into that first page. We 391 * choose the second.. 392 */ 393 if (memsz > filsz) 394 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 395 else 396 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 397 398 if (map_len != 0) { 399 vm_object_reference(object); 400 401 /* cow flags: don't dump readonly sections in core */ 402 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 403 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 404 405 rv = __elfN(map_insert)(&vmspace->vm_map, 406 object, 407 file_addr, /* file offset */ 408 map_addr, /* virtual start */ 409 map_addr + map_len,/* virtual end */ 410 prot, 411 VM_PROT_ALL, 412 cow); 413 if (rv != KERN_SUCCESS) { 414 vm_object_deallocate(object); 415 return (EINVAL); 416 } 417 418 /* we can stop now if we've covered it all */ 419 if (memsz == filsz) { 420 return (0); 421 } 422 } 423 424 425 /* 426 * We have to get the remaining bit of the file into the first part 427 * of the oversized map segment. This is normally because the .data 428 * segment in the file is extended to provide bss. It's a neat idea 429 * to try and save a page, but it's a pain in the behind to implement. 430 */ 431 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 432 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 433 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 434 map_addr; 435 436 /* This had damn well better be true! */ 437 if (map_len != 0) { 438 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr, 439 map_addr + map_len, VM_PROT_ALL, VM_PROT_ALL, 0); 440 if (rv != KERN_SUCCESS) { 441 return (EINVAL); 442 } 443 } 444 445 if (copy_len != 0) { 446 vm_offset_t off; 447 vm_object_reference(object); 448 rv = vm_map_find(exec_map, 449 object, 450 trunc_page(offset + filsz), 451 &data_buf, 452 PAGE_SIZE, 453 TRUE, 454 VM_PROT_READ, 455 VM_PROT_ALL, 456 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 457 if (rv != KERN_SUCCESS) { 458 vm_object_deallocate(object); 459 return (EINVAL); 460 } 461 462 /* send the page fragment to user space */ 463 off = trunc_page_ps(offset + filsz, pagesize) - 464 trunc_page(offset + filsz); 465 error = copyout((caddr_t)data_buf + off, (caddr_t)map_addr, 466 copy_len); 467 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 468 if (error) { 469 return (error); 470 } 471 } 472 473 /* 474 * set it to the specified protection. 475 * XXX had better undo the damage from pasting over the cracks here! 476 */ 477 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr), 478 round_page(map_addr + map_len), prot, FALSE); 479 480 return (error); 481 } 482 483 /* 484 * Load the file "file" into memory. It may be either a shared object 485 * or an executable. 486 * 487 * The "addr" reference parameter is in/out. On entry, it specifies 488 * the address where a shared object should be loaded. If the file is 489 * an executable, this value is ignored. On exit, "addr" specifies 490 * where the file was actually loaded. 491 * 492 * The "entry" reference parameter is out only. On exit, it specifies 493 * the entry point for the loaded file. 494 */ 495 static int 496 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 497 u_long *entry, size_t pagesize) 498 { 499 struct { 500 struct nameidata nd; 501 struct vattr attr; 502 struct image_params image_params; 503 } *tempdata; 504 const Elf_Ehdr *hdr = NULL; 505 const Elf_Phdr *phdr = NULL; 506 struct nameidata *nd; 507 struct vmspace *vmspace = p->p_vmspace; 508 struct vattr *attr; 509 struct image_params *imgp; 510 vm_prot_t prot; 511 u_long rbase; 512 u_long base_addr = 0; 513 int error, i, numsegs; 514 515 if (curthread->td_proc != p) 516 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */ 517 518 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 519 nd = &tempdata->nd; 520 attr = &tempdata->attr; 521 imgp = &tempdata->image_params; 522 523 /* 524 * Initialize part of the common data 525 */ 526 imgp->proc = p; 527 imgp->userspace_argv = NULL; 528 imgp->userspace_envv = NULL; 529 imgp->attr = attr; 530 imgp->firstpage = NULL; 531 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE); 532 imgp->object = NULL; 533 imgp->execlabel = NULL; 534 535 if (imgp->image_header == NULL) { 536 nd->ni_vp = NULL; 537 error = ENOMEM; 538 goto fail; 539 } 540 541 /* XXXKSE */ 542 NDINIT(nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, curthread); 543 544 if ((error = namei(nd)) != 0) { 545 nd->ni_vp = NULL; 546 goto fail; 547 } 548 NDFREE(nd, NDF_ONLY_PNBUF); 549 imgp->vp = nd->ni_vp; 550 551 /* 552 * Check permissions, modes, uid, etc on the file, and "open" it. 553 */ 554 error = exec_check_permissions(imgp); 555 if (error) { 556 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */ 557 goto fail; 558 } 559 560 error = exec_map_first_page(imgp); 561 /* 562 * Also make certain that the interpreter stays the same, so set 563 * its VV_TEXT flag, too. 564 */ 565 if (error == 0) 566 nd->ni_vp->v_vflag |= VV_TEXT; 567 568 VOP_GETVOBJECT(nd->ni_vp, &imgp->object); 569 vm_object_reference(imgp->object); 570 571 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */ 572 if (error) 573 goto fail; 574 575 hdr = (const Elf_Ehdr *)imgp->image_header; 576 if ((error = __elfN(check_header)(hdr)) != 0) 577 goto fail; 578 if (hdr->e_type == ET_DYN) 579 rbase = *addr; 580 else if (hdr->e_type == ET_EXEC) 581 rbase = 0; 582 else { 583 error = ENOEXEC; 584 goto fail; 585 } 586 587 /* Only support headers that fit within first page for now */ 588 /* (multiplication of two Elf_Half fields will not overflow) */ 589 if ((hdr->e_phoff > PAGE_SIZE) || 590 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 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 if (interp != NULL && brand_info->interp_newpath != NULL) 712 interp = brand_info->interp_newpath; 713 714 if ((error = exec_extract_strings(imgp)) != 0) 715 goto fail; 716 717 exec_new_vmspace(imgp, sv); 718 719 vmspace = imgp->proc->p_vmspace; 720 721 for (i = 0; i < hdr->e_phnum; i++) { 722 switch (phdr[i].p_type) { 723 case PT_LOAD: /* Loadable segment */ 724 prot = 0; 725 if (phdr[i].p_flags & PF_X) 726 prot |= VM_PROT_EXECUTE; 727 if (phdr[i].p_flags & PF_W) 728 prot |= VM_PROT_WRITE; 729 if (phdr[i].p_flags & PF_R) 730 prot |= VM_PROT_READ; 731 732 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER) 733 /* 734 * Some x86 binaries assume read == executable, 735 * notably the M3 runtime and therefore cvsup 736 */ 737 if (prot & VM_PROT_READ) 738 prot |= VM_PROT_EXECUTE; 739 #endif 740 741 if ((error = __elfN(load_section)(imgp->proc, vmspace, 742 imgp->vp, imgp->object, phdr[i].p_offset, 743 (caddr_t)(uintptr_t)phdr[i].p_vaddr, 744 phdr[i].p_memsz, phdr[i].p_filesz, prot, 745 sv->sv_pagesize)) != 0) 746 goto fail; 747 748 seg_addr = trunc_page(phdr[i].p_vaddr); 749 seg_size = round_page(phdr[i].p_memsz + 750 phdr[i].p_vaddr - seg_addr); 751 752 /* 753 * Is this .text or .data? We can't use 754 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 755 * alpha terribly and possibly does other bad 756 * things so we stick to the old way of figuring 757 * it out: If the segment contains the program 758 * entry point, it's a text segment, otherwise it 759 * is a data segment. 760 * 761 * Note that obreak() assumes that data_addr + 762 * data_size == end of data load area, and the ELF 763 * file format expects segments to be sorted by 764 * address. If multiple data segments exist, the 765 * last one will be used. 766 */ 767 if (hdr->e_entry >= phdr[i].p_vaddr && 768 hdr->e_entry < (phdr[i].p_vaddr + 769 phdr[i].p_memsz)) { 770 text_size = seg_size; 771 text_addr = seg_addr; 772 entry = (u_long)hdr->e_entry; 773 } else { 774 data_size = seg_size; 775 data_addr = seg_addr; 776 } 777 total_size += seg_size; 778 break; 779 case PT_PHDR: /* Program header table info */ 780 proghdr = phdr[i].p_vaddr; 781 break; 782 default: 783 break; 784 } 785 } 786 787 if (data_addr == 0 && data_size == 0) { 788 data_addr = text_addr; 789 data_size = text_size; 790 } 791 792 /* 793 * Check limits. It should be safe to check the 794 * limits after loading the segments since we do 795 * not actually fault in all the segments pages. 796 */ 797 PROC_LOCK(imgp->proc); 798 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 799 text_size > maxtsiz || 800 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) { 801 PROC_UNLOCK(imgp->proc); 802 error = ENOMEM; 803 goto fail; 804 } 805 806 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 807 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 808 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 809 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 810 811 /* 812 * We load the dynamic linker where a userland call 813 * to mmap(0, ...) would put it. The rationale behind this 814 * calculation is that it leaves room for the heap to grow to 815 * its maximum allowed size. 816 */ 817 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr + 818 lim_max(imgp->proc, RLIMIT_DATA)); 819 PROC_UNLOCK(imgp->proc); 820 821 imgp->entry_addr = entry; 822 823 imgp->proc->p_sysent = sv; 824 if (interp != NULL && brand_info->emul_path != NULL && 825 brand_info->emul_path[0] != '\0') { 826 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 827 snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path, 828 interp); 829 error = __elfN(load_file)(imgp->proc, path, &addr, 830 &imgp->entry_addr, sv->sv_pagesize); 831 free(path, M_TEMP); 832 if (error == 0) 833 interp = NULL; 834 } 835 if (interp != NULL) { 836 error = __elfN(load_file)(imgp->proc, interp, &addr, 837 &imgp->entry_addr, sv->sv_pagesize); 838 if (error != 0) { 839 uprintf("ELF interpreter %s not found\n", interp); 840 goto fail; 841 } 842 } 843 844 /* 845 * Construct auxargs table (used by the fixup routine) 846 */ 847 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 848 elf_auxargs->execfd = -1; 849 elf_auxargs->phdr = proghdr; 850 elf_auxargs->phent = hdr->e_phentsize; 851 elf_auxargs->phnum = hdr->e_phnum; 852 elf_auxargs->pagesz = PAGE_SIZE; 853 elf_auxargs->base = addr; 854 elf_auxargs->flags = 0; 855 elf_auxargs->entry = entry; 856 elf_auxargs->trace = elf_trace; 857 858 imgp->auxargs = elf_auxargs; 859 imgp->interpreted = 0; 860 861 fail: 862 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td); 863 return (error); 864 } 865 866 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 867 868 int 869 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 870 { 871 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 872 Elf_Addr *base; 873 Elf_Addr *pos; 874 875 base = (Elf_Addr *)*stack_base; 876 pos = base + (imgp->argc + imgp->envc + 2); 877 878 if (args->trace) { 879 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 880 } 881 if (args->execfd != -1) { 882 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 883 } 884 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 885 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 886 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 887 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 888 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 889 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 890 AUXARGS_ENTRY(pos, AT_BASE, args->base); 891 AUXARGS_ENTRY(pos, AT_NULL, 0); 892 893 free(imgp->auxargs, M_TEMP); 894 imgp->auxargs = NULL; 895 896 base--; 897 suword(base, (long)imgp->argc); 898 *stack_base = (register_t *)base; 899 return (0); 900 } 901 902 /* 903 * Code for generating ELF core dumps. 904 */ 905 906 typedef void (*segment_callback)(vm_map_entry_t, void *); 907 908 /* Closure for cb_put_phdr(). */ 909 struct phdr_closure { 910 Elf_Phdr *phdr; /* Program header to fill in */ 911 Elf_Off offset; /* Offset of segment in core file */ 912 }; 913 914 /* Closure for cb_size_segment(). */ 915 struct sseg_closure { 916 int count; /* Count of writable segments. */ 917 size_t size; /* Total size of all writable segments. */ 918 }; 919 920 static void cb_put_phdr(vm_map_entry_t, void *); 921 static void cb_size_segment(vm_map_entry_t, void *); 922 static void each_writable_segment(struct proc *, segment_callback, void *); 923 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 924 int, void *, size_t); 925 static void __elfN(puthdr)(struct proc *, void *, size_t *, int); 926 static void __elfN(putnote)(void *, size_t *, const char *, int, 927 const void *, size_t); 928 929 extern int osreldate; 930 931 int 932 __elfN(coredump)(td, vp, limit) 933 struct thread *td; 934 register struct vnode *vp; 935 off_t limit; 936 { 937 register struct proc *p = td->td_proc; 938 register struct ucred *cred = td->td_ucred; 939 int error = 0; 940 struct sseg_closure seginfo; 941 void *hdr; 942 size_t hdrsize; 943 944 /* Size the program segments. */ 945 seginfo.count = 0; 946 seginfo.size = 0; 947 each_writable_segment(p, cb_size_segment, &seginfo); 948 949 /* 950 * Calculate the size of the core file header area by making 951 * a dry run of generating it. Nothing is written, but the 952 * size is calculated. 953 */ 954 hdrsize = 0; 955 __elfN(puthdr)(p, (void *)NULL, &hdrsize, seginfo.count); 956 957 if (hdrsize + seginfo.size >= limit) 958 return (EFAULT); 959 960 /* 961 * Allocate memory for building the header, fill it up, 962 * and write it out. 963 */ 964 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 965 if (hdr == NULL) { 966 return (EINVAL); 967 } 968 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize); 969 970 /* Write the contents of all of the writable segments. */ 971 if (error == 0) { 972 Elf_Phdr *php; 973 off_t offset; 974 int i; 975 976 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 977 offset = hdrsize; 978 for (i = 0; i < seginfo.count; i++) { 979 error = vn_rdwr_inchunks(UIO_WRITE, vp, 980 (caddr_t)(uintptr_t)php->p_vaddr, 981 php->p_filesz, offset, UIO_USERSPACE, 982 IO_UNIT | IO_DIRECT, cred, NOCRED, (int *)NULL, 983 curthread); /* XXXKSE */ 984 if (error != 0) 985 break; 986 offset += php->p_filesz; 987 php++; 988 } 989 } 990 free(hdr, M_TEMP); 991 992 return (error); 993 } 994 995 /* 996 * A callback for each_writable_segment() to write out the segment's 997 * program header entry. 998 */ 999 static void 1000 cb_put_phdr(entry, closure) 1001 vm_map_entry_t entry; 1002 void *closure; 1003 { 1004 struct phdr_closure *phc = (struct phdr_closure *)closure; 1005 Elf_Phdr *phdr = phc->phdr; 1006 1007 phc->offset = round_page(phc->offset); 1008 1009 phdr->p_type = PT_LOAD; 1010 phdr->p_offset = phc->offset; 1011 phdr->p_vaddr = entry->start; 1012 phdr->p_paddr = 0; 1013 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1014 phdr->p_align = PAGE_SIZE; 1015 phdr->p_flags = 0; 1016 if (entry->protection & VM_PROT_READ) 1017 phdr->p_flags |= PF_R; 1018 if (entry->protection & VM_PROT_WRITE) 1019 phdr->p_flags |= PF_W; 1020 if (entry->protection & VM_PROT_EXECUTE) 1021 phdr->p_flags |= PF_X; 1022 1023 phc->offset += phdr->p_filesz; 1024 phc->phdr++; 1025 } 1026 1027 /* 1028 * A callback for each_writable_segment() to gather information about 1029 * the number of segments and their total size. 1030 */ 1031 static void 1032 cb_size_segment(entry, closure) 1033 vm_map_entry_t entry; 1034 void *closure; 1035 { 1036 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1037 1038 ssc->count++; 1039 ssc->size += entry->end - entry->start; 1040 } 1041 1042 /* 1043 * For each writable segment in the process's memory map, call the given 1044 * function with a pointer to the map entry and some arbitrary 1045 * caller-supplied data. 1046 */ 1047 static void 1048 each_writable_segment(p, func, closure) 1049 struct proc *p; 1050 segment_callback func; 1051 void *closure; 1052 { 1053 vm_map_t map = &p->p_vmspace->vm_map; 1054 vm_map_entry_t entry; 1055 1056 for (entry = map->header.next; entry != &map->header; 1057 entry = entry->next) { 1058 vm_object_t obj; 1059 1060 /* 1061 * Don't dump inaccessible mappings, deal with legacy 1062 * coredump mode. 1063 * 1064 * Note that read-only segments related to the elf binary 1065 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1066 * need to arbitrarily ignore such segments. 1067 */ 1068 if (elf_legacy_coredump) { 1069 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1070 continue; 1071 } else { 1072 if ((entry->protection & VM_PROT_ALL) == 0) 1073 continue; 1074 } 1075 1076 /* 1077 * Dont include memory segment in the coredump if 1078 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1079 * madvise(2). Do not dump submaps (i.e. parts of the 1080 * kernel map). 1081 */ 1082 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1083 continue; 1084 1085 if ((obj = entry->object.vm_object) == NULL) 1086 continue; 1087 1088 /* Find the deepest backing object. */ 1089 while (obj->backing_object != NULL) 1090 obj = obj->backing_object; 1091 1092 /* Ignore memory-mapped devices and such things. */ 1093 if (obj->type != OBJT_DEFAULT && 1094 obj->type != OBJT_SWAP && 1095 obj->type != OBJT_VNODE) 1096 continue; 1097 1098 (*func)(entry, closure); 1099 } 1100 } 1101 1102 /* 1103 * Write the core file header to the file, including padding up to 1104 * the page boundary. 1105 */ 1106 static int 1107 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize) 1108 struct thread *td; 1109 struct vnode *vp; 1110 struct ucred *cred; 1111 int numsegs; 1112 size_t hdrsize; 1113 void *hdr; 1114 { 1115 struct proc *p = td->td_proc; 1116 size_t off; 1117 1118 /* Fill in the header. */ 1119 bzero(hdr, hdrsize); 1120 off = 0; 1121 __elfN(puthdr)(p, hdr, &off, numsegs); 1122 1123 /* Write it to the core file. */ 1124 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1125 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1126 td)); /* XXXKSE */ 1127 } 1128 1129 static void 1130 __elfN(puthdr)(struct proc *p, void *dst, size_t *off, int numsegs) 1131 { 1132 struct { 1133 prstatus_t status; 1134 prfpregset_t fpregset; 1135 prpsinfo_t psinfo; 1136 } *tempdata; 1137 prstatus_t *status; 1138 prfpregset_t *fpregset; 1139 prpsinfo_t *psinfo; 1140 struct thread *first, *thr; 1141 size_t ehoff, noteoff, notesz, phoff; 1142 1143 ehoff = *off; 1144 *off += sizeof(Elf_Ehdr); 1145 1146 phoff = *off; 1147 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1148 1149 noteoff = *off; 1150 /* 1151 * Don't allocate space for the notes if we're just calculating 1152 * the size of the header. We also don't collect the data. 1153 */ 1154 if (dst != NULL) { 1155 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK); 1156 status = &tempdata->status; 1157 fpregset = &tempdata->fpregset; 1158 psinfo = &tempdata->psinfo; 1159 } else { 1160 tempdata = NULL; 1161 status = NULL; 1162 fpregset = NULL; 1163 psinfo = NULL; 1164 } 1165 1166 if (dst != NULL) { 1167 psinfo->pr_version = PRPSINFO_VERSION; 1168 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 1169 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1170 /* 1171 * XXX - We don't fill in the command line arguments properly 1172 * yet. 1173 */ 1174 strlcpy(psinfo->pr_psargs, p->p_comm, 1175 sizeof(psinfo->pr_psargs)); 1176 } 1177 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1178 sizeof *psinfo); 1179 1180 /* 1181 * We want to start with the registers of the first thread in the 1182 * process so that the .reg and .reg2 pseudo-sections created by bfd 1183 * will be identical to the .reg/$PID and .reg2/$PID pseudo-sections. 1184 * This makes sure that any tool that only looks for .reg and .reg2 1185 * and not for .reg/$PID and .reg2/$PID will behave the same as 1186 * before. The first thread is the thread with an ID equal to the 1187 * process' ID. 1188 */ 1189 first = TAILQ_FIRST(&p->p_threads); 1190 while (first->td_tid > PID_MAX) 1191 first = TAILQ_NEXT(first, td_plist); 1192 thr = first; 1193 do { 1194 if (dst != NULL) { 1195 status->pr_version = PRSTATUS_VERSION; 1196 status->pr_statussz = sizeof(prstatus_t); 1197 status->pr_gregsetsz = sizeof(gregset_t); 1198 status->pr_fpregsetsz = sizeof(fpregset_t); 1199 status->pr_osreldate = osreldate; 1200 status->pr_cursig = p->p_sig; 1201 status->pr_pid = thr->td_tid; 1202 fill_regs(thr, &status->pr_reg); 1203 fill_fpregs(thr, fpregset); 1204 } 1205 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1206 sizeof *status); 1207 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1208 sizeof *fpregset); 1209 /* XXX allow for MD specific notes. */ 1210 thr = (thr == first) ? TAILQ_FIRST(&p->p_threads) : 1211 TAILQ_NEXT(thr, td_plist); 1212 if (thr == first) 1213 thr = TAILQ_NEXT(thr, td_plist); 1214 } while (thr != NULL); 1215 1216 notesz = *off - noteoff; 1217 1218 if (dst != NULL) 1219 free(tempdata, M_TEMP); 1220 1221 /* Align up to a page boundary for the program segments. */ 1222 *off = round_page(*off); 1223 1224 if (dst != NULL) { 1225 Elf_Ehdr *ehdr; 1226 Elf_Phdr *phdr; 1227 struct phdr_closure phc; 1228 1229 /* 1230 * Fill in the ELF header. 1231 */ 1232 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1233 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1234 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1235 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1236 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1237 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1238 ehdr->e_ident[EI_DATA] = ELF_DATA; 1239 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1240 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1241 ehdr->e_ident[EI_ABIVERSION] = 0; 1242 ehdr->e_ident[EI_PAD] = 0; 1243 ehdr->e_type = ET_CORE; 1244 ehdr->e_machine = ELF_ARCH; 1245 ehdr->e_version = EV_CURRENT; 1246 ehdr->e_entry = 0; 1247 ehdr->e_phoff = phoff; 1248 ehdr->e_flags = 0; 1249 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1250 ehdr->e_phentsize = sizeof(Elf_Phdr); 1251 ehdr->e_phnum = numsegs + 1; 1252 ehdr->e_shentsize = sizeof(Elf_Shdr); 1253 ehdr->e_shnum = 0; 1254 ehdr->e_shstrndx = SHN_UNDEF; 1255 1256 /* 1257 * Fill in the program header entries. 1258 */ 1259 phdr = (Elf_Phdr *)((char *)dst + phoff); 1260 1261 /* The note segement. */ 1262 phdr->p_type = PT_NOTE; 1263 phdr->p_offset = noteoff; 1264 phdr->p_vaddr = 0; 1265 phdr->p_paddr = 0; 1266 phdr->p_filesz = notesz; 1267 phdr->p_memsz = 0; 1268 phdr->p_flags = 0; 1269 phdr->p_align = 0; 1270 phdr++; 1271 1272 /* All the writable segments from the program. */ 1273 phc.phdr = phdr; 1274 phc.offset = *off; 1275 each_writable_segment(p, cb_put_phdr, &phc); 1276 } 1277 } 1278 1279 static void 1280 __elfN(putnote)(void *dst, size_t *off, const char *name, int type, 1281 const void *desc, size_t descsz) 1282 { 1283 Elf_Note note; 1284 1285 note.n_namesz = strlen(name) + 1; 1286 note.n_descsz = descsz; 1287 note.n_type = type; 1288 if (dst != NULL) 1289 bcopy(¬e, (char *)dst + *off, sizeof note); 1290 *off += sizeof note; 1291 if (dst != NULL) 1292 bcopy(name, (char *)dst + *off, note.n_namesz); 1293 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1294 if (dst != NULL) 1295 bcopy(desc, (char *)dst + *off, note.n_descsz); 1296 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1297 } 1298 1299 /* 1300 * Tell kern_execve.c about it, with a little help from the linker. 1301 */ 1302 static struct execsw __elfN(execsw) = { 1303 __CONCAT(exec_, __elfN(imgact)), 1304 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 1305 }; 1306 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 1307