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 #ifdef __arm__ 86 int __elfN(fallback_brand) = 9; 87 #else 88 int __elfN(fallback_brand) = -1; 89 #endif 90 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 91 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 92 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 93 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 94 &__elfN(fallback_brand)); 95 96 static int elf_trace = 0; 97 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, ""); 98 99 static int elf_legacy_coredump = 0; 100 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 101 &elf_legacy_coredump, 0, ""); 102 103 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 104 105 int 106 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 107 { 108 int i; 109 110 for (i = 0; i < MAX_BRANDS; i++) { 111 if (elf_brand_list[i] == NULL) { 112 elf_brand_list[i] = entry; 113 break; 114 } 115 } 116 if (i == MAX_BRANDS) 117 return (-1); 118 return (0); 119 } 120 121 int 122 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 123 { 124 int i; 125 126 for (i = 0; i < MAX_BRANDS; i++) { 127 if (elf_brand_list[i] == entry) { 128 elf_brand_list[i] = NULL; 129 break; 130 } 131 } 132 if (i == MAX_BRANDS) 133 return (-1); 134 return (0); 135 } 136 137 int 138 __elfN(brand_inuse)(Elf_Brandinfo *entry) 139 { 140 struct proc *p; 141 int rval = FALSE; 142 143 sx_slock(&allproc_lock); 144 LIST_FOREACH(p, &allproc, p_list) { 145 if (p->p_sysent == entry->sysvec) { 146 rval = TRUE; 147 break; 148 } 149 } 150 sx_sunlock(&allproc_lock); 151 152 return (rval); 153 } 154 155 static Elf_Brandinfo * 156 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp) 157 { 158 Elf_Brandinfo *bi; 159 int i; 160 161 /* 162 * We support three types of branding -- (1) the ELF EI_OSABI field 163 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 164 * branding w/in the ELF header, and (3) path of the `interp_path' 165 * field. We should also look for an ".note.ABI-tag" ELF section now 166 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones. 167 */ 168 169 /* If the executable has a brand, search for it in the brand list. */ 170 for (i = 0; i < MAX_BRANDS; i++) { 171 bi = elf_brand_list[i]; 172 if (bi != NULL && hdr->e_machine == bi->machine && 173 (hdr->e_ident[EI_OSABI] == bi->brand || 174 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 175 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 176 return (bi); 177 } 178 179 /* Lacking a known brand, search for a recognized interpreter. */ 180 if (interp != NULL) { 181 for (i = 0; i < MAX_BRANDS; i++) { 182 bi = elf_brand_list[i]; 183 if (bi != NULL && hdr->e_machine == bi->machine && 184 strcmp(interp, bi->interp_path) == 0) 185 return (bi); 186 } 187 } 188 189 /* Lacking a recognized interpreter, try the default brand */ 190 for (i = 0; i < MAX_BRANDS; i++) { 191 bi = elf_brand_list[i]; 192 if (bi != NULL && hdr->e_machine == bi->machine && 193 __elfN(fallback_brand) == bi->brand) 194 return (bi); 195 } 196 return (NULL); 197 } 198 199 static int 200 __elfN(check_header)(const Elf_Ehdr *hdr) 201 { 202 Elf_Brandinfo *bi; 203 int i; 204 205 if (!IS_ELF(*hdr) || 206 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 207 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 208 hdr->e_ident[EI_VERSION] != EV_CURRENT || 209 hdr->e_phentsize != sizeof(Elf_Phdr) || 210 hdr->e_version != ELF_TARG_VER) 211 return (ENOEXEC); 212 213 /* 214 * Make sure we have at least one brand for this machine. 215 */ 216 217 for (i = 0; i < MAX_BRANDS; i++) { 218 bi = elf_brand_list[i]; 219 if (bi != NULL && bi->machine == hdr->e_machine) 220 break; 221 } 222 if (i == MAX_BRANDS) 223 return (ENOEXEC); 224 225 return (0); 226 } 227 228 static int 229 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 230 vm_offset_t start, vm_offset_t end, vm_prot_t prot, 231 vm_prot_t max) 232 { 233 int error, rv; 234 vm_offset_t off; 235 vm_offset_t data_buf = 0; 236 237 /* 238 * Create the page if it doesn't exist yet. Ignore errors. 239 */ 240 vm_map_lock(map); 241 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), max, 242 max, 0); 243 vm_map_unlock(map); 244 245 /* 246 * Find the page from the underlying object. 247 */ 248 if (object) { 249 vm_object_reference(object); 250 rv = vm_map_find(exec_map, 251 object, 252 trunc_page(offset), 253 &data_buf, 254 PAGE_SIZE, 255 TRUE, 256 VM_PROT_READ, 257 VM_PROT_ALL, 258 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 259 if (rv != KERN_SUCCESS) { 260 vm_object_deallocate(object); 261 return (rv); 262 } 263 264 off = offset - trunc_page(offset); 265 error = copyout((caddr_t)data_buf + off, (caddr_t)start, 266 end - start); 267 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 268 if (error) { 269 return (KERN_FAILURE); 270 } 271 } 272 273 return (KERN_SUCCESS); 274 } 275 276 static int 277 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 278 vm_offset_t start, vm_offset_t end, vm_prot_t prot, 279 vm_prot_t max, int cow) 280 { 281 vm_offset_t data_buf, off; 282 vm_size_t sz; 283 int error, rv; 284 285 if (start != trunc_page(start)) { 286 rv = __elfN(map_partial)(map, object, offset, start, 287 round_page(start), prot, max); 288 if (rv) 289 return (rv); 290 offset += round_page(start) - start; 291 start = round_page(start); 292 } 293 if (end != round_page(end)) { 294 rv = __elfN(map_partial)(map, object, offset + 295 trunc_page(end) - start, trunc_page(end), end, prot, max); 296 if (rv) 297 return (rv); 298 end = trunc_page(end); 299 } 300 if (end > start) { 301 if (offset & PAGE_MASK) { 302 /* 303 * The mapping is not page aligned. This means we have 304 * to copy the data. Sigh. 305 */ 306 rv = vm_map_find(map, 0, 0, &start, end - start, 307 FALSE, prot, max, 0); 308 if (rv) 309 return (rv); 310 data_buf = 0; 311 while (start < end) { 312 vm_object_reference(object); 313 rv = vm_map_find(exec_map, 314 object, 315 trunc_page(offset), 316 &data_buf, 317 2 * PAGE_SIZE, 318 TRUE, 319 VM_PROT_READ, 320 VM_PROT_ALL, 321 (MAP_COPY_ON_WRITE 322 | MAP_PREFAULT_PARTIAL)); 323 if (rv != KERN_SUCCESS) { 324 vm_object_deallocate(object); 325 return (rv); 326 } 327 off = offset - trunc_page(offset); 328 sz = end - start; 329 if (sz > PAGE_SIZE) 330 sz = PAGE_SIZE; 331 error = copyout((caddr_t)data_buf + off, 332 (caddr_t)start, sz); 333 vm_map_remove(exec_map, data_buf, 334 data_buf + 2 * PAGE_SIZE); 335 if (error) { 336 return (KERN_FAILURE); 337 } 338 start += sz; 339 } 340 rv = KERN_SUCCESS; 341 } else { 342 vm_map_lock(map); 343 rv = vm_map_insert(map, object, offset, start, end, 344 prot, max, cow); 345 vm_map_unlock(map); 346 } 347 return (rv); 348 } else { 349 return (KERN_SUCCESS); 350 } 351 } 352 353 static int 354 __elfN(load_section)(struct proc *p, struct vmspace *vmspace, 355 struct vnode *vp, vm_object_t object, vm_offset_t offset, 356 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 357 size_t pagesize) 358 { 359 size_t map_len; 360 vm_offset_t map_addr; 361 int error, rv, cow; 362 size_t copy_len; 363 vm_offset_t file_addr; 364 vm_offset_t data_buf = 0; 365 366 error = 0; 367 368 /* 369 * It's necessary to fail if the filsz + offset taken from the 370 * header is greater than the actual file pager object's size. 371 * If we were to allow this, then the vm_map_find() below would 372 * walk right off the end of the file object and into the ether. 373 * 374 * While I'm here, might as well check for something else that 375 * is invalid: filsz cannot be greater than memsz. 376 */ 377 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 378 filsz > memsz) { 379 uprintf("elf_load_section: truncated ELF file\n"); 380 return (ENOEXEC); 381 } 382 383 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 384 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 385 386 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 387 file_addr = trunc_page_ps(offset, pagesize); 388 389 /* 390 * We have two choices. We can either clear the data in the last page 391 * of an oversized mapping, or we can start the anon mapping a page 392 * early and copy the initialized data into that first page. We 393 * choose the second.. 394 */ 395 if (memsz > filsz) 396 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 397 else 398 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 399 400 if (map_len != 0) { 401 vm_object_reference(object); 402 403 /* cow flags: don't dump readonly sections in core */ 404 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 405 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 406 407 rv = __elfN(map_insert)(&vmspace->vm_map, 408 object, 409 file_addr, /* file offset */ 410 map_addr, /* virtual start */ 411 map_addr + map_len,/* virtual end */ 412 prot, 413 VM_PROT_ALL, 414 cow); 415 if (rv != KERN_SUCCESS) { 416 vm_object_deallocate(object); 417 return (EINVAL); 418 } 419 420 /* we can stop now if we've covered it all */ 421 if (memsz == filsz) { 422 return (0); 423 } 424 } 425 426 427 /* 428 * We have to get the remaining bit of the file into the first part 429 * of the oversized map segment. This is normally because the .data 430 * segment in the file is extended to provide bss. It's a neat idea 431 * to try and save a page, but it's a pain in the behind to implement. 432 */ 433 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 434 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 435 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 436 map_addr; 437 438 /* This had damn well better be true! */ 439 if (map_len != 0) { 440 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr, 441 map_addr + map_len, VM_PROT_ALL, VM_PROT_ALL, 0); 442 if (rv != KERN_SUCCESS) { 443 return (EINVAL); 444 } 445 } 446 447 if (copy_len != 0) { 448 vm_offset_t off; 449 vm_object_reference(object); 450 rv = vm_map_find(exec_map, 451 object, 452 trunc_page(offset + filsz), 453 &data_buf, 454 PAGE_SIZE, 455 TRUE, 456 VM_PROT_READ, 457 VM_PROT_ALL, 458 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 459 if (rv != KERN_SUCCESS) { 460 vm_object_deallocate(object); 461 return (EINVAL); 462 } 463 464 /* send the page fragment to user space */ 465 off = trunc_page_ps(offset + filsz, pagesize) - 466 trunc_page(offset + filsz); 467 error = copyout((caddr_t)data_buf + off, (caddr_t)map_addr, 468 copy_len); 469 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 470 if (error) { 471 return (error); 472 } 473 } 474 475 /* 476 * set it to the specified protection. 477 * XXX had better undo the damage from pasting over the cracks here! 478 */ 479 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr), 480 round_page(map_addr + map_len), prot, FALSE); 481 482 return (error); 483 } 484 485 /* 486 * Load the file "file" into memory. It may be either a shared object 487 * or an executable. 488 * 489 * The "addr" reference parameter is in/out. On entry, it specifies 490 * the address where a shared object should be loaded. If the file is 491 * an executable, this value is ignored. On exit, "addr" specifies 492 * where the file was actually loaded. 493 * 494 * The "entry" reference parameter is out only. On exit, it specifies 495 * the entry point for the loaded file. 496 */ 497 static int 498 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 499 u_long *entry, size_t pagesize) 500 { 501 struct { 502 struct nameidata nd; 503 struct vattr attr; 504 struct image_params image_params; 505 } *tempdata; 506 const Elf_Ehdr *hdr = NULL; 507 const Elf_Phdr *phdr = NULL; 508 struct nameidata *nd; 509 struct vmspace *vmspace = p->p_vmspace; 510 struct vattr *attr; 511 struct image_params *imgp; 512 vm_prot_t prot; 513 u_long rbase; 514 u_long base_addr = 0; 515 int error, i, numsegs; 516 517 if (curthread->td_proc != p) 518 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */ 519 520 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 521 nd = &tempdata->nd; 522 attr = &tempdata->attr; 523 imgp = &tempdata->image_params; 524 525 /* 526 * Initialize part of the common data 527 */ 528 imgp->proc = p; 529 imgp->attr = attr; 530 imgp->firstpage = NULL; 531 imgp->image_header = NULL; 532 imgp->object = NULL; 533 imgp->execlabel = NULL; 534 535 /* XXXKSE */ 536 NDINIT(nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, curthread); 537 538 if ((error = namei(nd)) != 0) { 539 nd->ni_vp = NULL; 540 goto fail; 541 } 542 NDFREE(nd, NDF_ONLY_PNBUF); 543 imgp->vp = nd->ni_vp; 544 545 /* 546 * Check permissions, modes, uid, etc on the file, and "open" it. 547 */ 548 error = exec_check_permissions(imgp); 549 if (error) { 550 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */ 551 goto fail; 552 } 553 554 error = exec_map_first_page(imgp); 555 /* 556 * Also make certain that the interpreter stays the same, so set 557 * its VV_TEXT flag, too. 558 */ 559 if (error == 0) 560 nd->ni_vp->v_vflag |= VV_TEXT; 561 562 imgp->object = nd->ni_vp->v_object; 563 vm_object_reference(imgp->object); 564 565 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */ 566 if (error) 567 goto fail; 568 569 hdr = (const Elf_Ehdr *)imgp->image_header; 570 if ((error = __elfN(check_header)(hdr)) != 0) 571 goto fail; 572 if (hdr->e_type == ET_DYN) 573 rbase = *addr; 574 else if (hdr->e_type == ET_EXEC) 575 rbase = 0; 576 else { 577 error = ENOEXEC; 578 goto fail; 579 } 580 581 /* Only support headers that fit within first page for now */ 582 /* (multiplication of two Elf_Half fields will not overflow) */ 583 if ((hdr->e_phoff > PAGE_SIZE) || 584 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 585 error = ENOEXEC; 586 goto fail; 587 } 588 589 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 590 591 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 592 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 593 prot = 0; 594 if (phdr[i].p_flags & PF_X) 595 prot |= VM_PROT_EXECUTE; 596 if (phdr[i].p_flags & PF_W) 597 prot |= VM_PROT_WRITE; 598 if (phdr[i].p_flags & PF_R) 599 prot |= VM_PROT_READ; 600 601 if ((error = __elfN(load_section)(p, vmspace, 602 nd->ni_vp, imgp->object, phdr[i].p_offset, 603 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 604 phdr[i].p_memsz, phdr[i].p_filesz, prot, 605 pagesize)) != 0) 606 goto fail; 607 /* 608 * Establish the base address if this is the 609 * first segment. 610 */ 611 if (numsegs == 0) 612 base_addr = trunc_page(phdr[i].p_vaddr + 613 rbase); 614 numsegs++; 615 } 616 } 617 *addr = base_addr; 618 *entry = (unsigned long)hdr->e_entry + rbase; 619 620 fail: 621 if (imgp->firstpage) 622 exec_unmap_first_page(imgp); 623 if (imgp->object) 624 vm_object_deallocate(imgp->object); 625 626 if (nd->ni_vp) 627 vrele(nd->ni_vp); 628 629 free(tempdata, M_TEMP); 630 631 return (error); 632 } 633 634 static int 635 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 636 { 637 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 638 const Elf_Phdr *phdr; 639 Elf_Auxargs *elf_auxargs = NULL; 640 struct vmspace *vmspace; 641 vm_prot_t prot; 642 u_long text_size = 0, data_size = 0, total_size = 0; 643 u_long text_addr = 0, data_addr = 0; 644 u_long seg_size, seg_addr; 645 u_long addr, entry = 0, proghdr = 0; 646 int error = 0, i; 647 const char *interp = NULL; 648 Elf_Brandinfo *brand_info; 649 char *path; 650 struct thread *td = curthread; 651 struct sysentvec *sv; 652 653 GIANT_REQUIRED; 654 655 /* 656 * Do we have a valid ELF header ? 657 */ 658 if (__elfN(check_header)(hdr) != 0 || hdr->e_type != ET_EXEC) 659 return (-1); 660 661 /* 662 * From here on down, we return an errno, not -1, as we've 663 * detected an ELF file. 664 */ 665 666 if ((hdr->e_phoff > PAGE_SIZE) || 667 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 668 /* Only support headers in first page for now */ 669 return (ENOEXEC); 670 } 671 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 672 673 /* 674 * From this point on, we may have resources that need to be freed. 675 */ 676 677 VOP_UNLOCK(imgp->vp, 0, td); 678 679 for (i = 0; i < hdr->e_phnum; i++) { 680 switch (phdr[i].p_type) { 681 case PT_INTERP: /* Path to interpreter */ 682 if (phdr[i].p_filesz > MAXPATHLEN || 683 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) { 684 error = ENOEXEC; 685 goto fail; 686 } 687 interp = imgp->image_header + phdr[i].p_offset; 688 break; 689 default: 690 break; 691 } 692 } 693 694 brand_info = __elfN(get_brandinfo)(hdr, interp); 695 if (brand_info == NULL) { 696 uprintf("ELF binary type \"%u\" not known.\n", 697 hdr->e_ident[EI_OSABI]); 698 error = ENOEXEC; 699 goto fail; 700 } 701 sv = brand_info->sysvec; 702 if (interp != NULL && brand_info->interp_newpath != NULL) 703 interp = brand_info->interp_newpath; 704 705 exec_new_vmspace(imgp, sv); 706 707 vmspace = imgp->proc->p_vmspace; 708 709 for (i = 0; i < hdr->e_phnum; i++) { 710 switch (phdr[i].p_type) { 711 case PT_LOAD: /* Loadable segment */ 712 prot = 0; 713 if (phdr[i].p_flags & PF_X) 714 prot |= VM_PROT_EXECUTE; 715 if (phdr[i].p_flags & PF_W) 716 prot |= VM_PROT_WRITE; 717 if (phdr[i].p_flags & PF_R) 718 prot |= VM_PROT_READ; 719 720 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER) 721 /* 722 * Some x86 binaries assume read == executable, 723 * notably the M3 runtime and therefore cvsup 724 */ 725 if (prot & VM_PROT_READ) 726 prot |= VM_PROT_EXECUTE; 727 #endif 728 729 if ((error = __elfN(load_section)(imgp->proc, vmspace, 730 imgp->vp, imgp->object, phdr[i].p_offset, 731 (caddr_t)(uintptr_t)phdr[i].p_vaddr, 732 phdr[i].p_memsz, phdr[i].p_filesz, prot, 733 sv->sv_pagesize)) != 0) 734 goto fail; 735 736 /* 737 * If this segment contains the program headers, 738 * remember their virtual address for the AT_PHDR 739 * aux entry. Static binaries don't usually include 740 * a PT_PHDR entry. 741 */ 742 if (phdr[i].p_offset == 0 && 743 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 744 <= phdr[i].p_filesz) 745 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 746 747 seg_addr = trunc_page(phdr[i].p_vaddr); 748 seg_size = round_page(phdr[i].p_memsz + 749 phdr[i].p_vaddr - seg_addr); 750 751 /* 752 * Is this .text or .data? We can't use 753 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 754 * alpha terribly and possibly does other bad 755 * things so we stick to the old way of figuring 756 * it out: If the segment contains the program 757 * entry point, it's a text segment, otherwise it 758 * is a data segment. 759 * 760 * Note that obreak() assumes that data_addr + 761 * data_size == end of data load area, and the ELF 762 * file format expects segments to be sorted by 763 * address. If multiple data segments exist, the 764 * last one will be used. 765 */ 766 if (hdr->e_entry >= phdr[i].p_vaddr && 767 hdr->e_entry < (phdr[i].p_vaddr + 768 phdr[i].p_memsz)) { 769 text_size = seg_size; 770 text_addr = seg_addr; 771 entry = (u_long)hdr->e_entry; 772 } else { 773 data_size = seg_size; 774 data_addr = seg_addr; 775 } 776 total_size += seg_size; 777 break; 778 case PT_PHDR: /* Program header table info */ 779 proghdr = phdr[i].p_vaddr; 780 break; 781 default: 782 break; 783 } 784 } 785 786 if (data_addr == 0 && data_size == 0) { 787 data_addr = text_addr; 788 data_size = text_size; 789 } 790 791 /* 792 * Check limits. It should be safe to check the 793 * limits after loading the segments since we do 794 * not actually fault in all the segments pages. 795 */ 796 PROC_LOCK(imgp->proc); 797 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 798 text_size > maxtsiz || 799 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) { 800 PROC_UNLOCK(imgp->proc); 801 error = ENOMEM; 802 goto fail; 803 } 804 805 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 806 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 807 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 808 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 809 810 /* 811 * We load the dynamic linker where a userland call 812 * to mmap(0, ...) would put it. The rationale behind this 813 * calculation is that it leaves room for the heap to grow to 814 * its maximum allowed size. 815 */ 816 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr + 817 lim_max(imgp->proc, RLIMIT_DATA)); 818 PROC_UNLOCK(imgp->proc); 819 820 imgp->entry_addr = entry; 821 822 imgp->proc->p_sysent = sv; 823 if (interp != NULL && brand_info->emul_path != NULL && 824 brand_info->emul_path[0] != '\0') { 825 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 826 snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path, 827 interp); 828 error = __elfN(load_file)(imgp->proc, path, &addr, 829 &imgp->entry_addr, sv->sv_pagesize); 830 free(path, M_TEMP); 831 if (error == 0) 832 interp = NULL; 833 } 834 if (interp != NULL) { 835 error = __elfN(load_file)(imgp->proc, interp, &addr, 836 &imgp->entry_addr, sv->sv_pagesize); 837 if (error != 0) { 838 uprintf("ELF interpreter %s not found\n", interp); 839 goto fail; 840 } 841 } 842 843 /* 844 * Construct auxargs table (used by the fixup routine) 845 */ 846 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 847 elf_auxargs->execfd = -1; 848 elf_auxargs->phdr = proghdr; 849 elf_auxargs->phent = hdr->e_phentsize; 850 elf_auxargs->phnum = hdr->e_phnum; 851 elf_auxargs->pagesz = PAGE_SIZE; 852 elf_auxargs->base = addr; 853 elf_auxargs->flags = 0; 854 elf_auxargs->entry = entry; 855 elf_auxargs->trace = elf_trace; 856 857 imgp->auxargs = elf_auxargs; 858 imgp->interpreted = 0; 859 860 fail: 861 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td); 862 return (error); 863 } 864 865 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 866 867 int 868 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 869 { 870 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 871 Elf_Addr *base; 872 Elf_Addr *pos; 873 874 base = (Elf_Addr *)*stack_base; 875 pos = base + (imgp->args->argc + imgp->args->envc + 2); 876 877 if (args->trace) { 878 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 879 } 880 if (args->execfd != -1) { 881 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 882 } 883 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 884 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 885 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 886 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 887 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 888 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 889 AUXARGS_ENTRY(pos, AT_BASE, args->base); 890 AUXARGS_ENTRY(pos, AT_NULL, 0); 891 892 free(imgp->auxargs, M_TEMP); 893 imgp->auxargs = NULL; 894 895 base--; 896 suword(base, (long)imgp->args->argc); 897 *stack_base = (register_t *)base; 898 return (0); 899 } 900 901 /* 902 * Code for generating ELF core dumps. 903 */ 904 905 typedef void (*segment_callback)(vm_map_entry_t, void *); 906 907 /* Closure for cb_put_phdr(). */ 908 struct phdr_closure { 909 Elf_Phdr *phdr; /* Program header to fill in */ 910 Elf_Off offset; /* Offset of segment in core file */ 911 }; 912 913 /* Closure for cb_size_segment(). */ 914 struct sseg_closure { 915 int count; /* Count of writable segments. */ 916 size_t size; /* Total size of all writable segments. */ 917 }; 918 919 static void cb_put_phdr(vm_map_entry_t, void *); 920 static void cb_size_segment(vm_map_entry_t, void *); 921 static void each_writable_segment(struct thread *, segment_callback, void *); 922 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 923 int, void *, size_t); 924 static void __elfN(puthdr)(struct thread *, void *, size_t *, int); 925 static void __elfN(putnote)(void *, size_t *, const char *, int, 926 const void *, size_t); 927 928 extern int osreldate; 929 930 int 931 __elfN(coredump)(td, vp, limit) 932 struct thread *td; 933 struct vnode *vp; 934 off_t limit; 935 { 936 struct ucred *cred = td->td_ucred; 937 int error = 0; 938 struct sseg_closure seginfo; 939 void *hdr; 940 size_t hdrsize; 941 942 /* Size the program segments. */ 943 seginfo.count = 0; 944 seginfo.size = 0; 945 each_writable_segment(td, cb_size_segment, &seginfo); 946 947 /* 948 * Calculate the size of the core file header area by making 949 * a dry run of generating it. Nothing is written, but the 950 * size is calculated. 951 */ 952 hdrsize = 0; 953 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count); 954 955 if (hdrsize + seginfo.size >= limit) 956 return (EFAULT); 957 958 /* 959 * Allocate memory for building the header, fill it up, 960 * and write it out. 961 */ 962 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 963 if (hdr == NULL) { 964 return (EINVAL); 965 } 966 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize); 967 968 /* Write the contents of all of the writable segments. */ 969 if (error == 0) { 970 Elf_Phdr *php; 971 off_t offset; 972 int i; 973 974 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 975 offset = hdrsize; 976 for (i = 0; i < seginfo.count; i++) { 977 error = vn_rdwr_inchunks(UIO_WRITE, vp, 978 (caddr_t)(uintptr_t)php->p_vaddr, 979 php->p_filesz, offset, UIO_USERSPACE, 980 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 981 curthread); /* XXXKSE */ 982 if (error != 0) 983 break; 984 offset += php->p_filesz; 985 php++; 986 } 987 } 988 free(hdr, M_TEMP); 989 990 return (error); 991 } 992 993 /* 994 * A callback for each_writable_segment() to write out the segment's 995 * program header entry. 996 */ 997 static void 998 cb_put_phdr(entry, closure) 999 vm_map_entry_t entry; 1000 void *closure; 1001 { 1002 struct phdr_closure *phc = (struct phdr_closure *)closure; 1003 Elf_Phdr *phdr = phc->phdr; 1004 1005 phc->offset = round_page(phc->offset); 1006 1007 phdr->p_type = PT_LOAD; 1008 phdr->p_offset = phc->offset; 1009 phdr->p_vaddr = entry->start; 1010 phdr->p_paddr = 0; 1011 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1012 phdr->p_align = PAGE_SIZE; 1013 phdr->p_flags = 0; 1014 if (entry->protection & VM_PROT_READ) 1015 phdr->p_flags |= PF_R; 1016 if (entry->protection & VM_PROT_WRITE) 1017 phdr->p_flags |= PF_W; 1018 if (entry->protection & VM_PROT_EXECUTE) 1019 phdr->p_flags |= PF_X; 1020 1021 phc->offset += phdr->p_filesz; 1022 phc->phdr++; 1023 } 1024 1025 /* 1026 * A callback for each_writable_segment() to gather information about 1027 * the number of segments and their total size. 1028 */ 1029 static void 1030 cb_size_segment(entry, closure) 1031 vm_map_entry_t entry; 1032 void *closure; 1033 { 1034 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1035 1036 ssc->count++; 1037 ssc->size += entry->end - entry->start; 1038 } 1039 1040 /* 1041 * For each writable segment in the process's memory map, call the given 1042 * function with a pointer to the map entry and some arbitrary 1043 * caller-supplied data. 1044 */ 1045 static void 1046 each_writable_segment(td, func, closure) 1047 struct thread *td; 1048 segment_callback func; 1049 void *closure; 1050 { 1051 struct proc *p = td->td_proc; 1052 vm_map_t map = &p->p_vmspace->vm_map; 1053 vm_map_entry_t entry; 1054 1055 for (entry = map->header.next; entry != &map->header; 1056 entry = entry->next) { 1057 vm_object_t obj; 1058 1059 /* 1060 * Don't dump inaccessible mappings, deal with legacy 1061 * coredump mode. 1062 * 1063 * Note that read-only segments related to the elf binary 1064 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1065 * need to arbitrarily ignore such segments. 1066 */ 1067 if (elf_legacy_coredump) { 1068 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1069 continue; 1070 } else { 1071 if ((entry->protection & VM_PROT_ALL) == 0) 1072 continue; 1073 } 1074 1075 /* 1076 * Dont include memory segment in the coredump if 1077 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1078 * madvise(2). Do not dump submaps (i.e. parts of the 1079 * kernel map). 1080 */ 1081 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1082 continue; 1083 1084 if ((obj = entry->object.vm_object) == NULL) 1085 continue; 1086 1087 /* Find the deepest backing object. */ 1088 while (obj->backing_object != NULL) 1089 obj = obj->backing_object; 1090 1091 /* Ignore memory-mapped devices and such things. */ 1092 if (obj->type != OBJT_DEFAULT && 1093 obj->type != OBJT_SWAP && 1094 obj->type != OBJT_VNODE) 1095 continue; 1096 1097 (*func)(entry, closure); 1098 } 1099 } 1100 1101 /* 1102 * Write the core file header to the file, including padding up to 1103 * the page boundary. 1104 */ 1105 static int 1106 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize) 1107 struct thread *td; 1108 struct vnode *vp; 1109 struct ucred *cred; 1110 int numsegs; 1111 size_t hdrsize; 1112 void *hdr; 1113 { 1114 size_t off; 1115 1116 /* Fill in the header. */ 1117 bzero(hdr, hdrsize); 1118 off = 0; 1119 __elfN(puthdr)(td, hdr, &off, numsegs); 1120 1121 /* Write it to the core file. */ 1122 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1123 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1124 td)); /* XXXKSE */ 1125 } 1126 1127 static void 1128 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs) 1129 { 1130 struct { 1131 prstatus_t status; 1132 prfpregset_t fpregset; 1133 prpsinfo_t psinfo; 1134 } *tempdata; 1135 prstatus_t *status; 1136 prfpregset_t *fpregset; 1137 prpsinfo_t *psinfo; 1138 struct proc *p; 1139 struct thread *thr; 1140 size_t ehoff, noteoff, notesz, phoff; 1141 1142 p = td->td_proc; 1143 1144 ehoff = *off; 1145 *off += sizeof(Elf_Ehdr); 1146 1147 phoff = *off; 1148 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1149 1150 noteoff = *off; 1151 /* 1152 * Don't allocate space for the notes if we're just calculating 1153 * the size of the header. We also don't collect the data. 1154 */ 1155 if (dst != NULL) { 1156 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK); 1157 status = &tempdata->status; 1158 fpregset = &tempdata->fpregset; 1159 psinfo = &tempdata->psinfo; 1160 } else { 1161 tempdata = NULL; 1162 status = NULL; 1163 fpregset = NULL; 1164 psinfo = NULL; 1165 } 1166 1167 if (dst != NULL) { 1168 psinfo->pr_version = PRPSINFO_VERSION; 1169 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 1170 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1171 /* 1172 * XXX - We don't fill in the command line arguments properly 1173 * yet. 1174 */ 1175 strlcpy(psinfo->pr_psargs, p->p_comm, 1176 sizeof(psinfo->pr_psargs)); 1177 } 1178 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1179 sizeof *psinfo); 1180 1181 /* 1182 * To have the debugger select the right thread (LWP) as the initial 1183 * thread, we dump the state of the thread passed to us in td first. 1184 * This is the thread that causes the core dump and thus likely to 1185 * be the right thread one wants to have selected in the debugger. 1186 */ 1187 thr = td; 1188 while (thr != NULL) { 1189 if (dst != NULL) { 1190 status->pr_version = PRSTATUS_VERSION; 1191 status->pr_statussz = sizeof(prstatus_t); 1192 status->pr_gregsetsz = sizeof(gregset_t); 1193 status->pr_fpregsetsz = sizeof(fpregset_t); 1194 status->pr_osreldate = osreldate; 1195 status->pr_cursig = p->p_sig; 1196 status->pr_pid = thr->td_tid; 1197 fill_regs(thr, &status->pr_reg); 1198 fill_fpregs(thr, fpregset); 1199 } 1200 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1201 sizeof *status); 1202 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1203 sizeof *fpregset); 1204 /* 1205 * Allow for MD specific notes, as well as any MD 1206 * specific preparations for writing MI notes. 1207 */ 1208 __elfN(dump_thread)(thr, dst, off); 1209 1210 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1211 TAILQ_NEXT(thr, td_plist); 1212 if (thr == td) 1213 thr = TAILQ_NEXT(thr, td_plist); 1214 } 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(td, 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