1 /*- 2 * Copyright (c) 1995-1996 S�ren Schmidt 3 * Copyright (c) 1996 Peter Wemm 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer 11 * in this position and unchanged. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. The name of the author may not be used to endorse or promote products 16 * derived from this software withough specific prior written permission 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 * 29 * $FreeBSD: src/sys/kern/imgact_elf.c,v 1.73.2.13 2002/12/28 19:49:41 dillon Exp $ 30 * $DragonFly: src/sys/kern/imgact_elf.c,v 1.32 2005/09/02 07:16:58 hsu Exp $ 31 */ 32 33 #include <sys/param.h> 34 #include <sys/exec.h> 35 #include <sys/fcntl.h> 36 #include <sys/file.h> 37 #include <sys/imgact.h> 38 #include <sys/imgact_elf.h> 39 #include <sys/kernel.h> 40 #include <sys/malloc.h> 41 #include <sys/mman.h> 42 #include <sys/systm.h> 43 #include <sys/proc.h> 44 #include <sys/nlookup.h> 45 #include <sys/pioctl.h> 46 #include <sys/procfs.h> 47 #include <sys/resourcevar.h> 48 #include <sys/signalvar.h> 49 #include <sys/stat.h> 50 #include <sys/syscall.h> 51 #include <sys/sysctl.h> 52 #include <sys/sysent.h> 53 #include <sys/vnode.h> 54 55 #include <vm/vm.h> 56 #include <vm/vm_kern.h> 57 #include <vm/vm_param.h> 58 #include <vm/pmap.h> 59 #include <sys/lock.h> 60 #include <vm/vm_map.h> 61 #include <vm/vm_object.h> 62 #include <vm/vm_extern.h> 63 64 #include <machine/elf.h> 65 #include <machine/md_var.h> 66 #include <sys/mount.h> 67 #include <sys/ckpt.h> 68 #define OLD_EI_BRAND 8 69 70 __ElfType(Brandinfo); 71 __ElfType(Auxargs); 72 73 static int elf_check_header (const Elf_Ehdr *hdr); 74 static int elf_freebsd_fixup (register_t **stack_base, 75 struct image_params *imgp); 76 static int elf_load_file (struct proc *p, const char *file, u_long *addr, 77 u_long *entry); 78 static int elf_load_section (struct proc *p, 79 struct vmspace *vmspace, struct vnode *vp, 80 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 81 vm_prot_t prot); 82 static int exec_elf_imgact (struct image_params *imgp); 83 84 static int elf_trace = 0; 85 SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, ""); 86 static int elf_legacy_coredump = 0; 87 SYSCTL_INT(_debug, OID_AUTO, elf_legacy_coredump, CTLFLAG_RW, 88 &elf_legacy_coredump, 0, ""); 89 90 static int dragonfly_match_abi_note(const Elf_Note *); 91 static int freebsd_match_abi_note(const Elf_Note *); 92 93 static struct sysentvec elf_freebsd_sysvec = { 94 SYS_MAXSYSCALL, 95 sysent, 96 -1, 97 0, 98 0, 99 0, 100 0, 101 0, 102 elf_freebsd_fixup, 103 sendsig, 104 sigcode, 105 &szsigcode, 106 0, 107 "FreeBSD ELF", 108 elf_coredump, 109 NULL, 110 MINSIGSTKSZ 111 }; 112 113 static Elf_Brandinfo freebsd_brand_info = { 114 ELFOSABI_FREEBSD, 115 "FreeBSD", 116 freebsd_match_abi_note, 117 "", 118 "/usr/libexec/ld-elf.so.1", 119 &elf_freebsd_sysvec 120 }; 121 122 static Elf_Brandinfo dragonfly_brand_info = { 123 ELFOSABI_NONE, 124 "DragonFly", 125 dragonfly_match_abi_note, 126 "", 127 "/usr/libexec/ld-elf.so.2", 128 &elf_freebsd_sysvec 129 }; 130 131 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = { 132 &dragonfly_brand_info, 133 &freebsd_brand_info, 134 NULL, NULL, NULL, 135 NULL, NULL, NULL 136 }; 137 138 static int 139 freebsd_match_abi_note(const Elf_Note *abi_note) 140 { 141 const char *abi_name = (const char *) 142 ((const uint8_t *)abi_note + sizeof(*abi_note)); 143 144 if (abi_note->n_namesz != sizeof("FreeBSD")) 145 return(FALSE); 146 if (memcmp(abi_name, "FreeBSD", sizeof("FreeBSD"))) 147 return(FALSE); 148 return(TRUE); 149 } 150 151 static int 152 dragonfly_match_abi_note(const Elf_Note *abi_note) 153 { 154 const char *abi_name = (const char *) 155 ((const uint8_t *)abi_note + sizeof(*abi_note)); 156 157 if (abi_note->n_namesz != sizeof("DragonFly")) 158 return(FALSE); 159 if (memcmp(abi_name, "DragonFly", sizeof("DragonFly"))) 160 return(FALSE); 161 return(TRUE); 162 } 163 164 int 165 elf_insert_brand_entry(Elf_Brandinfo *entry) 166 { 167 int i; 168 169 for (i=1; i<MAX_BRANDS; i++) { 170 if (elf_brand_list[i] == NULL) { 171 elf_brand_list[i] = entry; 172 break; 173 } 174 } 175 if (i == MAX_BRANDS) 176 return -1; 177 return 0; 178 } 179 180 int 181 elf_remove_brand_entry(Elf_Brandinfo *entry) 182 { 183 int i; 184 185 for (i=1; i<MAX_BRANDS; i++) { 186 if (elf_brand_list[i] == entry) { 187 elf_brand_list[i] = NULL; 188 break; 189 } 190 } 191 if (i == MAX_BRANDS) 192 return -1; 193 return 0; 194 } 195 196 int 197 elf_brand_inuse(Elf_Brandinfo *entry) 198 { 199 struct proc *p; 200 int rval = FALSE; 201 202 FOREACH_PROC_IN_SYSTEM(p) { 203 if (p->p_sysent == entry->sysvec) { 204 rval = TRUE; 205 break; 206 } 207 } 208 209 return (rval); 210 } 211 212 static int 213 elf_check_header(const Elf_Ehdr *hdr) 214 { 215 if (!IS_ELF(*hdr) || 216 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 217 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 218 hdr->e_ident[EI_VERSION] != EV_CURRENT || 219 hdr->e_phentsize != sizeof(Elf_Phdr) || 220 hdr->e_ehsize != sizeof(Elf_Ehdr) || 221 hdr->e_version != ELF_TARG_VER) 222 return ENOEXEC; 223 224 if (!ELF_MACHINE_OK(hdr->e_machine)) 225 return ENOEXEC; 226 227 return 0; 228 } 229 230 static int 231 elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp, 232 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 233 vm_prot_t prot) 234 { 235 size_t map_len; 236 vm_offset_t map_addr; 237 int error, rv, cow; 238 int count; 239 size_t copy_len; 240 vm_object_t object; 241 vm_offset_t file_addr; 242 vm_offset_t data_buf = 0; 243 244 VOP_GETVOBJECT(vp, &object); 245 error = 0; 246 247 /* 248 * It's necessary to fail if the filsz + offset taken from the 249 * header is greater than the actual file pager object's size. 250 * If we were to allow this, then the vm_map_find() below would 251 * walk right off the end of the file object and into the ether. 252 * 253 * While I'm here, might as well check for something else that 254 * is invalid: filsz cannot be greater than memsz. 255 */ 256 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 257 filsz > memsz) { 258 uprintf("elf_load_section: truncated ELF file\n"); 259 return (ENOEXEC); 260 } 261 262 map_addr = trunc_page((vm_offset_t)vmaddr); 263 file_addr = trunc_page(offset); 264 265 /* 266 * We have two choices. We can either clear the data in the last page 267 * of an oversized mapping, or we can start the anon mapping a page 268 * early and copy the initialized data into that first page. We 269 * choose the second.. 270 */ 271 if (memsz > filsz) 272 map_len = trunc_page(offset+filsz) - file_addr; 273 else 274 map_len = round_page(offset+filsz) - file_addr; 275 276 if (map_len != 0) { 277 vm_object_reference(object); 278 279 /* cow flags: don't dump readonly sections in core */ 280 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 281 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 282 283 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 284 vm_map_lock(&vmspace->vm_map); 285 rv = vm_map_insert(&vmspace->vm_map, &count, 286 object, 287 file_addr, /* file offset */ 288 map_addr, /* virtual start */ 289 map_addr + map_len,/* virtual end */ 290 prot, 291 VM_PROT_ALL, 292 cow); 293 vm_map_unlock(&vmspace->vm_map); 294 vm_map_entry_release(count); 295 if (rv != KERN_SUCCESS) { 296 vm_object_deallocate(object); 297 return EINVAL; 298 } 299 300 /* we can stop now if we've covered it all */ 301 if (memsz == filsz) { 302 return 0; 303 } 304 } 305 306 307 /* 308 * We have to get the remaining bit of the file into the first part 309 * of the oversized map segment. This is normally because the .data 310 * segment in the file is extended to provide bss. It's a neat idea 311 * to try and save a page, but it's a pain in the behind to implement. 312 */ 313 copy_len = (offset + filsz) - trunc_page(offset + filsz); 314 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 315 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 316 317 /* This had damn well better be true! */ 318 if (map_len != 0) { 319 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 320 vm_map_lock(&vmspace->vm_map); 321 rv = vm_map_insert(&vmspace->vm_map, &count, 322 NULL, 0, 323 map_addr, map_addr + map_len, 324 VM_PROT_ALL, VM_PROT_ALL, 0); 325 vm_map_unlock(&vmspace->vm_map); 326 vm_map_entry_release(count); 327 if (rv != KERN_SUCCESS) { 328 return EINVAL; 329 } 330 } 331 332 if (copy_len != 0) { 333 vm_object_reference(object); 334 rv = vm_map_find(exec_map, 335 object, 336 trunc_page(offset + filsz), 337 &data_buf, 338 PAGE_SIZE, 339 TRUE, 340 VM_PROT_READ, 341 VM_PROT_ALL, 342 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 343 if (rv != KERN_SUCCESS) { 344 vm_object_deallocate(object); 345 return EINVAL; 346 } 347 348 /* send the page fragment to user space */ 349 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len); 350 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 351 if (error) { 352 return (error); 353 } 354 } 355 356 /* 357 * set it to the specified protection 358 */ 359 vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len, prot, 360 FALSE); 361 362 return error; 363 } 364 365 /* 366 * Load the file "file" into memory. It may be either a shared object 367 * or an executable. 368 * 369 * The "addr" reference parameter is in/out. On entry, it specifies 370 * the address where a shared object should be loaded. If the file is 371 * an executable, this value is ignored. On exit, "addr" specifies 372 * where the file was actually loaded. 373 * 374 * The "entry" reference parameter is out only. On exit, it specifies 375 * the entry point for the loaded file. 376 */ 377 static int 378 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry) 379 { 380 struct { 381 struct nlookupdata nd; 382 struct vattr attr; 383 struct image_params image_params; 384 } *tempdata; 385 const Elf_Ehdr *hdr = NULL; 386 const Elf_Phdr *phdr = NULL; 387 struct nlookupdata *nd; 388 struct vmspace *vmspace = p->p_vmspace; 389 struct vattr *attr; 390 struct image_params *imgp; 391 vm_prot_t prot; 392 u_long rbase; 393 u_long base_addr = 0; 394 int error, i, numsegs; 395 struct thread *td = p->p_thread; 396 397 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 398 nd = &tempdata->nd; 399 attr = &tempdata->attr; 400 imgp = &tempdata->image_params; 401 402 /* 403 * Initialize part of the common data 404 */ 405 imgp->proc = p; 406 imgp->attr = attr; 407 imgp->firstpage = NULL; 408 imgp->image_header = NULL; 409 imgp->vp = NULL; 410 411 error = nlookup_init(nd, file, UIO_SYSSPACE, NLC_FOLLOW); 412 if (error == 0) 413 error = nlookup(nd); 414 if (error == 0) 415 error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 416 nlookup_done(nd); 417 if (error) 418 goto fail; 419 420 /* 421 * Check permissions, modes, uid, etc on the file, and "open" it. 422 */ 423 error = exec_check_permissions(imgp); 424 if (error) { 425 VOP_UNLOCK(imgp->vp, 0, td); 426 goto fail; 427 } 428 429 error = exec_map_first_page(imgp); 430 /* 431 * Also make certain that the interpreter stays the same, so set 432 * its VTEXT flag, too. 433 */ 434 if (error == 0) 435 imgp->vp->v_flag |= VTEXT; 436 VOP_UNLOCK(imgp->vp, 0, td); 437 if (error) 438 goto fail; 439 440 hdr = (const Elf_Ehdr *)imgp->image_header; 441 if ((error = elf_check_header(hdr)) != 0) 442 goto fail; 443 if (hdr->e_type == ET_DYN) 444 rbase = *addr; 445 else if (hdr->e_type == ET_EXEC) 446 rbase = 0; 447 else { 448 error = ENOEXEC; 449 goto fail; 450 } 451 452 /* Only support headers that fit within first page for now 453 * (multiplication of two Elf_Half fields will not overflow) */ 454 if ((hdr->e_phoff > PAGE_SIZE) || 455 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 456 error = ENOEXEC; 457 goto fail; 458 } 459 460 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 461 462 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 463 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 464 prot = 0; 465 if (phdr[i].p_flags & PF_X) 466 prot |= VM_PROT_EXECUTE; 467 if (phdr[i].p_flags & PF_W) 468 prot |= VM_PROT_WRITE; 469 if (phdr[i].p_flags & PF_R) 470 prot |= VM_PROT_READ; 471 472 error = elf_load_section( 473 p, vmspace, imgp->vp, 474 phdr[i].p_offset, 475 (caddr_t)phdr[i].p_vaddr + 476 rbase, 477 phdr[i].p_memsz, 478 phdr[i].p_filesz, prot); 479 if (error != 0) 480 goto fail; 481 /* 482 * Establish the base address if this is the 483 * first segment. 484 */ 485 if (numsegs == 0) 486 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 487 numsegs++; 488 } 489 } 490 *addr = base_addr; 491 *entry=(unsigned long)hdr->e_entry + rbase; 492 493 fail: 494 if (imgp->firstpage) 495 exec_unmap_first_page(imgp); 496 if (imgp->vp) { 497 vrele(imgp->vp); 498 imgp->vp = NULL; 499 } 500 free(tempdata, M_TEMP); 501 502 return error; 503 } 504 505 /* 506 * non static, as it can be overridden by start_init() 507 */ 508 int fallback_elf_brand = -1; 509 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW, 510 &fallback_elf_brand, -1, 511 "ELF brand of last resort"); 512 513 static int 514 exec_elf_imgact(struct image_params *imgp) 515 { 516 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header; 517 const Elf_Phdr *phdr; 518 Elf_Auxargs *elf_auxargs = NULL; 519 struct vmspace *vmspace; 520 vm_prot_t prot; 521 u_long text_size = 0, data_size = 0, total_size = 0; 522 u_long text_addr = 0, data_addr = 0; 523 u_long seg_size, seg_addr; 524 u_long addr, entry = 0, proghdr = 0; 525 int error, i; 526 const char *interp = NULL; 527 const Elf_Note *abi_note = NULL; 528 Elf_Brandinfo *brand_info; 529 char *path; 530 531 error = 0; 532 533 /* 534 * Do we have a valid ELF header ? 535 */ 536 if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC) 537 return -1; 538 539 /* 540 * From here on down, we return an errno, not -1, as we've 541 * detected an ELF file. 542 */ 543 544 if ((hdr->e_phoff > PAGE_SIZE) || 545 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 546 /* Only support headers in first page for now */ 547 return ENOEXEC; 548 } 549 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff); 550 551 /* 552 * From this point on, we may have resources that need to be freed. 553 */ 554 555 exec_new_vmspace(imgp, NULL); 556 557 /* 558 * Yeah, I'm paranoid. There is every reason in the world to get 559 * VTEXT now since from here on out, there are places we can have 560 * a context switch. Better safe than sorry; I really don't want 561 * the file to change while it's being loaded. 562 */ 563 vsetflags(imgp->vp, VTEXT); 564 565 vmspace = imgp->proc->p_vmspace; 566 567 for (i = 0; i < hdr->e_phnum; i++) { 568 switch(phdr[i].p_type) { 569 570 case PT_LOAD: /* Loadable segment */ 571 prot = 0; 572 if (phdr[i].p_flags & PF_X) 573 prot |= VM_PROT_EXECUTE; 574 if (phdr[i].p_flags & PF_W) 575 prot |= VM_PROT_WRITE; 576 if (phdr[i].p_flags & PF_R) 577 prot |= VM_PROT_READ; 578 579 if ((error = elf_load_section(imgp->proc, 580 vmspace, imgp->vp, 581 phdr[i].p_offset, 582 (caddr_t)phdr[i].p_vaddr, 583 phdr[i].p_memsz, 584 phdr[i].p_filesz, prot)) != 0) 585 goto fail; 586 587 /* 588 * If this segment contains the program headers, 589 * remember their virtual address for the AT_PHDR 590 * aux entry. Static binaries don't usually include 591 * a PT_PHDR entry. 592 */ 593 if (phdr[i].p_offset == 0 && 594 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 595 <= phdr[i].p_filesz) 596 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 597 598 seg_addr = trunc_page(phdr[i].p_vaddr); 599 seg_size = round_page(phdr[i].p_memsz + 600 phdr[i].p_vaddr - seg_addr); 601 602 /* 603 * Is this .text or .data? We can't use 604 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 605 * alpha terribly and possibly does other bad 606 * things so we stick to the old way of figuring 607 * it out: If the segment contains the program 608 * entry point, it's a text segment, otherwise it 609 * is a data segment. 610 * 611 * Note that obreak() assumes that data_addr + 612 * data_size == end of data load area, and the ELF 613 * file format expects segments to be sorted by 614 * address. If multiple data segments exist, the 615 * last one will be used. 616 */ 617 if (hdr->e_entry >= phdr[i].p_vaddr && 618 hdr->e_entry < (phdr[i].p_vaddr + 619 phdr[i].p_memsz)) { 620 text_size = seg_size; 621 text_addr = seg_addr; 622 entry = (u_long)hdr->e_entry; 623 } else { 624 data_size = seg_size; 625 data_addr = seg_addr; 626 } 627 total_size += seg_size; 628 629 /* 630 * Check limits. It should be safe to check the 631 * limits after loading the segment since we do 632 * not actually fault in all the segment's pages. 633 */ 634 if (data_size > 635 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur || 636 text_size > maxtsiz || 637 total_size > 638 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) { 639 error = ENOMEM; 640 goto fail; 641 } 642 break; 643 case PT_INTERP: /* Path to interpreter */ 644 if (phdr[i].p_filesz > MAXPATHLEN || 645 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) { 646 error = ENOEXEC; 647 goto fail; 648 } 649 interp = imgp->image_header + phdr[i].p_offset; 650 break; 651 case PT_NOTE: /* Check for .note.ABI-tag */ 652 { 653 const Elf_Note *tmp_note; 654 /* XXX handle anything outside the first page */ 655 if (phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) 656 continue; 657 if (phdr[i].p_filesz < sizeof(Elf_Note)) 658 continue; /* ENOEXEC? */ 659 tmp_note = (const Elf_Note *)(imgp->image_header + phdr[i].p_offset); 660 if (tmp_note->n_type != 1) 661 continue; 662 if (tmp_note->n_namesz + sizeof(Elf_Note) + 663 tmp_note->n_descsz > phdr[i].p_filesz) 664 continue; /* ENOEXEC? */ 665 abi_note = tmp_note; 666 } 667 break; 668 case PT_PHDR: /* Program header table info */ 669 proghdr = phdr[i].p_vaddr; 670 break; 671 default: 672 break; 673 } 674 } 675 676 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 677 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 678 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 679 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 680 681 addr = ELF_RTLD_ADDR(vmspace); 682 683 imgp->entry_addr = entry; 684 685 brand_info = NULL; 686 687 /* We support three types of branding -- (1) the ELF EI_OSABI field 688 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 689 * branding w/in the ELF header, and (3) path of the `interp_path' 690 * field. We should also look for an ".note.ABI-tag" ELF section now 691 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones. 692 */ 693 694 /* If the executable has a brand, search for it in the brand list. */ 695 if (brand_info == NULL && hdr->e_ident[EI_OSABI] != ELFOSABI_NONE) { 696 for (i = 0; i < MAX_BRANDS; i++) { 697 Elf_Brandinfo *bi = elf_brand_list[i]; 698 699 if (bi != NULL && 700 (hdr->e_ident[EI_OSABI] == bi->brand 701 || 0 == 702 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 703 bi->compat_3_brand, strlen(bi->compat_3_brand)))) { 704 brand_info = bi; 705 break; 706 } 707 } 708 } 709 710 /* Search for a recognized ABI. */ 711 if (brand_info == NULL && abi_note != NULL) { 712 for (i = 0; i < MAX_BRANDS; i++) { 713 Elf_Brandinfo *bi = elf_brand_list[i]; 714 715 if (bi != NULL && bi->match_abi_note != NULL && 716 (*bi->match_abi_note)(abi_note)) { 717 brand_info = bi; 718 break; 719 } 720 } 721 } 722 723 /* 724 * ELFOSABI_NONE == ELFOSABI_SYSV, so a SYSV binary misses all 725 * checks so far, since it is neither branded nor does it have 726 * an ABI note. If the EI_OSABI field is ELFOSABI_NONE, assume 727 * it is svr4 and look for an entry in the elf_brand_list with 728 * match_abi_note == NULL. 729 */ 730 if (brand_info == NULL && hdr->e_ident[EI_OSABI] == ELFOSABI_NONE) { 731 for (i = 0; i < MAX_BRANDS; i++) { 732 Elf_Brandinfo *bi = elf_brand_list[i]; 733 734 if (bi != NULL && bi->match_abi_note == NULL && 735 ELFOSABI_SYSV == bi->brand) { 736 brand_info = bi; 737 break; 738 } 739 } 740 } 741 742 /* Lacking a recognized ABI, search for a recognized interpreter. */ 743 if (brand_info == NULL && interp != NULL) { 744 for (i = 0; i < MAX_BRANDS; i++) { 745 Elf_Brandinfo *bi = elf_brand_list[i]; 746 747 if (bi != NULL && 748 strcmp(interp, bi->interp_path) == 0) { 749 brand_info = bi; 750 break; 751 } 752 } 753 } 754 755 /* Lacking a recognized interpreter, try the default brand */ 756 if (brand_info == NULL) { 757 for (i = 0; i < MAX_BRANDS; i++) { 758 Elf_Brandinfo *bi = elf_brand_list[i]; 759 760 if (bi != NULL && fallback_elf_brand == bi->brand) { 761 brand_info = bi; 762 break; 763 } 764 } 765 } 766 767 if (brand_info == NULL) { 768 uprintf("ELF binary type \"%u\" not known.\n", 769 hdr->e_ident[EI_OSABI]); 770 error = ENOEXEC; 771 goto fail; 772 } 773 774 imgp->proc->p_sysent = brand_info->sysvec; 775 if (interp != NULL) { 776 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 777 snprintf(path, MAXPATHLEN, "%s%s", 778 brand_info->emul_path, interp); 779 if ((error = elf_load_file(imgp->proc, path, &addr, 780 &imgp->entry_addr)) != 0) { 781 if ((error = elf_load_file(imgp->proc, interp, &addr, 782 &imgp->entry_addr)) != 0) { 783 uprintf("ELF interpreter %s not found\n", path); 784 free(path, M_TEMP); 785 goto fail; 786 } 787 } 788 free(path, M_TEMP); 789 } 790 791 /* 792 * Construct auxargs table (used by the fixup routine) 793 */ 794 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 795 elf_auxargs->execfd = -1; 796 elf_auxargs->phdr = proghdr; 797 elf_auxargs->phent = hdr->e_phentsize; 798 elf_auxargs->phnum = hdr->e_phnum; 799 elf_auxargs->pagesz = PAGE_SIZE; 800 elf_auxargs->base = addr; 801 elf_auxargs->flags = 0; 802 elf_auxargs->entry = entry; 803 elf_auxargs->trace = elf_trace; 804 805 imgp->auxargs = elf_auxargs; 806 imgp->interpreted = 0; 807 808 fail: 809 return error; 810 } 811 812 static int 813 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp) 814 { 815 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 816 register_t *pos; 817 818 pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2); 819 820 if (args->trace) { 821 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 822 } 823 if (args->execfd != -1) { 824 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 825 } 826 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 827 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 828 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 829 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 830 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 831 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 832 AUXARGS_ENTRY(pos, AT_BASE, args->base); 833 AUXARGS_ENTRY(pos, AT_NULL, 0); 834 835 free(imgp->auxargs, M_TEMP); 836 imgp->auxargs = NULL; 837 838 (*stack_base)--; 839 suword(*stack_base, (long) imgp->args->argc); 840 return 0; 841 } 842 843 /* 844 * Code for generating ELF core dumps. 845 */ 846 847 typedef int (*segment_callback) (vm_map_entry_t, void *); 848 849 /* Closure for cb_put_phdr(). */ 850 struct phdr_closure { 851 Elf_Phdr *phdr; /* Program header to fill in (incremented) */ 852 Elf_Phdr *phdr_max; /* Pointer bound for error check */ 853 Elf_Off offset; /* Offset of segment in core file */ 854 }; 855 856 /* Closure for cb_size_segment(). */ 857 struct sseg_closure { 858 int count; /* Count of writable segments. */ 859 size_t vsize; /* Total size of all writable segments. */ 860 }; 861 862 /* Closure for cb_put_fp(). */ 863 struct fp_closure { 864 struct vn_hdr *vnh; 865 struct vn_hdr *vnh_max; 866 int count; 867 struct stat *sb; 868 }; 869 870 typedef struct elf_buf { 871 char *buf; 872 size_t off; 873 size_t off_max; 874 } *elf_buf_t; 875 876 static void *target_reserve(elf_buf_t target, size_t bytes, int *error); 877 878 static int cb_put_phdr (vm_map_entry_t, void *); 879 static int cb_size_segment (vm_map_entry_t, void *); 880 static int cb_fpcount_segment(vm_map_entry_t, void *); 881 static int cb_put_fp(vm_map_entry_t, void *); 882 883 884 static int each_segment (struct proc *, segment_callback, void *, int); 885 static int elf_corehdr (struct proc *, struct file *, struct ucred *, 886 int, elf_buf_t); 887 static int elf_puthdr (struct proc *, elf_buf_t, const prstatus_t *, 888 const prfpregset_t *, const prpsinfo_t *, int); 889 static int elf_putnote (elf_buf_t, const char *, int, const void *, size_t); 890 891 static int elf_putsigs(struct proc *, elf_buf_t); 892 static int elf_puttextvp(struct proc *, elf_buf_t); 893 static int elf_putfiles(struct proc *, elf_buf_t); 894 895 extern int osreldate; 896 897 int 898 elf_coredump(struct proc *p, struct vnode *vp, off_t limit) 899 { 900 struct file *fp; 901 int error; 902 903 if ((error = falloc(NULL, &fp, NULL)) != 0) 904 return (error); 905 fsetcred(fp, p->p_ucred); 906 907 /* 908 * XXX fixme. 909 */ 910 fp->f_type = DTYPE_VNODE; 911 fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW; 912 fp->f_ops = &vnode_fileops; 913 fp->f_data = vp; 914 VOP_UNLOCK(vp, 0, p->p_thread); 915 916 error = generic_elf_coredump(p, fp, limit); 917 918 fp->f_type = 0; 919 fp->f_flag = 0; 920 fp->f_ops = &badfileops; 921 fp->f_data = NULL; 922 fdrop(fp, p->p_thread); 923 return (error); 924 } 925 926 int 927 generic_elf_coredump(struct proc *p, struct file *fp, off_t limit) 928 { 929 struct ucred *cred = p->p_ucred; 930 int error = 0; 931 struct sseg_closure seginfo; 932 struct elf_buf target; 933 934 if (!fp) 935 printf("can't dump core - null fp\n"); 936 937 /* 938 * Size the program segments 939 */ 940 seginfo.count = 0; 941 seginfo.vsize = 0; 942 each_segment(p, cb_size_segment, &seginfo, 1); 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 bzero(&target, sizeof(target)); 950 elf_puthdr(p, &target, NULL, NULL, NULL, seginfo.count); 951 952 if (target.off + seginfo.vsize >= limit) 953 return (EFAULT); 954 955 /* 956 * Allocate memory for building the header, fill it up, 957 * and write it out. 958 */ 959 target.off_max = target.off; 960 target.off = 0; 961 target.buf = malloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO); 962 963 if (target.buf == NULL) 964 return EINVAL; 965 error = elf_corehdr(p, fp, cred, seginfo.count, &target); 966 967 /* Write the contents of all of the writable segments. */ 968 if (error == 0) { 969 Elf_Phdr *php; 970 int i; 971 int nbytes; 972 973 php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1; 974 for (i = 0; i < seginfo.count; i++) { 975 error = fp_write(fp, (caddr_t)php->p_vaddr, 976 php->p_filesz, &nbytes); 977 if (error != 0) 978 break; 979 php++; 980 } 981 } 982 free(target.buf, M_TEMP); 983 984 return error; 985 } 986 987 /* 988 * A callback for each_segment() to write out the segment's 989 * program header entry. 990 */ 991 static int 992 cb_put_phdr(vm_map_entry_t entry, void *closure) 993 { 994 struct phdr_closure *phc = closure; 995 Elf_Phdr *phdr = phc->phdr; 996 997 if (phc->phdr == phc->phdr_max) 998 return EINVAL; 999 1000 phc->offset = round_page(phc->offset); 1001 1002 phdr->p_type = PT_LOAD; 1003 phdr->p_offset = phc->offset; 1004 phdr->p_vaddr = entry->start; 1005 phdr->p_paddr = 0; 1006 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1007 phdr->p_align = PAGE_SIZE; 1008 phdr->p_flags = 0; 1009 if (entry->protection & VM_PROT_READ) 1010 phdr->p_flags |= PF_R; 1011 if (entry->protection & VM_PROT_WRITE) 1012 phdr->p_flags |= PF_W; 1013 if (entry->protection & VM_PROT_EXECUTE) 1014 phdr->p_flags |= PF_X; 1015 1016 phc->offset += phdr->p_filesz; 1017 ++phc->phdr; 1018 return 0; 1019 } 1020 1021 /* 1022 * A callback for each_writable_segment() to gather information about 1023 * the number of segments and their total size. 1024 */ 1025 static int 1026 cb_size_segment(vm_map_entry_t entry, void *closure) 1027 { 1028 struct sseg_closure *ssc = closure; 1029 1030 ++ssc->count; 1031 ssc->vsize += entry->end - entry->start; 1032 return 0; 1033 } 1034 1035 /* 1036 * A callback for each_segment() to gather information about 1037 * the number of text segments. 1038 */ 1039 static int 1040 cb_fpcount_segment(vm_map_entry_t entry, void *closure) 1041 { 1042 int *count = closure; 1043 struct vnode *vp; 1044 1045 if (entry->object.vm_object->type == OBJT_VNODE) { 1046 vp = (struct vnode *)entry->object.vm_object->handle; 1047 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp) 1048 return 0; 1049 ++*count; 1050 } 1051 return 0; 1052 } 1053 1054 static int 1055 cb_put_fp(vm_map_entry_t entry, void *closure) 1056 { 1057 struct fp_closure *fpc = closure; 1058 struct vn_hdr *vnh = fpc->vnh; 1059 Elf_Phdr *phdr = &vnh->vnh_phdr; 1060 struct vnode *vp; 1061 int error; 1062 1063 /* 1064 * If an entry represents a vnode then write out a file handle. 1065 * 1066 * If we are checkpointing a checkpoint-restored program we do 1067 * NOT record the filehandle for the old checkpoint vnode (which 1068 * is mapped all over the place). Instead we rely on the fact 1069 * that a checkpoint-restored program does not mmap() the checkpt 1070 * vnode NOCORE, so its contents will be written out to the 1071 * new checkpoint file. This is necessary because the 'old' 1072 * checkpoint file is typically destroyed when a new one is created 1073 * and thus cannot be used to restore the new checkpoint. 1074 * 1075 * Theoretically we could create a chain of checkpoint files and 1076 * operate the checkpointing operation kinda like an incremental 1077 * checkpoint, but a checkpoint restore would then likely wind up 1078 * referencing many prior checkpoint files and that is a bit over 1079 * the top for the purpose of the checkpoint API. 1080 */ 1081 if (entry->object.vm_object->type == OBJT_VNODE) { 1082 vp = (struct vnode *)entry->object.vm_object->handle; 1083 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp) 1084 return 0; 1085 if (vnh == fpc->vnh_max) 1086 return EINVAL; 1087 1088 if (vp->v_mount) 1089 vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1090 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid); 1091 if (error) { 1092 char *freepath, *fullpath; 1093 1094 if (vn_fullpath(curproc, vp, &fullpath, &freepath)) { 1095 printf("Warning: coredump, error %d: cannot store file handle for vnode %p\n", error, vp); 1096 } else { 1097 printf("Warning: coredump, error %d: cannot store file handle for %s\n", error, fullpath); 1098 free(freepath, M_TEMP); 1099 } 1100 error = 0; 1101 } 1102 1103 phdr->p_type = PT_LOAD; 1104 phdr->p_offset = 0; /* not written to core */ 1105 phdr->p_vaddr = entry->start; 1106 phdr->p_paddr = 0; 1107 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1108 phdr->p_align = PAGE_SIZE; 1109 phdr->p_flags = 0; 1110 if (entry->protection & VM_PROT_READ) 1111 phdr->p_flags |= PF_R; 1112 if (entry->protection & VM_PROT_WRITE) 1113 phdr->p_flags |= PF_W; 1114 if (entry->protection & VM_PROT_EXECUTE) 1115 phdr->p_flags |= PF_X; 1116 ++fpc->vnh; 1117 ++fpc->count; 1118 } 1119 return 0; 1120 } 1121 1122 /* 1123 * For each writable segment in the process's memory map, call the given 1124 * function with a pointer to the map entry and some arbitrary 1125 * caller-supplied data. 1126 */ 1127 static int 1128 each_segment(struct proc *p, segment_callback func, void *closure, int writable) 1129 { 1130 int error = 0; 1131 vm_map_t map = &p->p_vmspace->vm_map; 1132 vm_map_entry_t entry; 1133 1134 for (entry = map->header.next; error == 0 && entry != &map->header; 1135 entry = entry->next) { 1136 vm_object_t obj; 1137 1138 /* 1139 * Don't dump inaccessible mappings, deal with legacy 1140 * coredump mode. 1141 * 1142 * Note that read-only segments related to the elf binary 1143 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1144 * need to arbitrarily ignore such segments. 1145 */ 1146 if (elf_legacy_coredump) { 1147 if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW) 1148 continue; 1149 } else { 1150 if (writable && (entry->protection & VM_PROT_ALL) == 0) 1151 continue; 1152 } 1153 1154 /* 1155 * Dont include memory segment in the coredump if 1156 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1157 * madvise(2). Do not dump submaps (i.e. parts of the 1158 * kernel map). 1159 */ 1160 if (writable && entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1161 continue; 1162 1163 if ((obj = entry->object.vm_object) == NULL) 1164 continue; 1165 1166 /* Find the deepest backing object. */ 1167 while (obj->backing_object != NULL) 1168 obj = obj->backing_object; 1169 1170 /* Ignore memory-mapped devices and such things. */ 1171 if (obj->type != OBJT_DEFAULT && 1172 obj->type != OBJT_SWAP && 1173 obj->type != OBJT_VNODE) 1174 continue; 1175 1176 error = (*func)(entry, closure); 1177 } 1178 return error; 1179 } 1180 1181 static 1182 void * 1183 target_reserve(elf_buf_t target, size_t bytes, int *error) 1184 { 1185 void *res = NULL; 1186 1187 if (target->buf) { 1188 if (target->off + bytes > target->off_max) 1189 *error = EINVAL; 1190 else 1191 res = target->buf + target->off; 1192 } 1193 target->off += bytes; 1194 return (res); 1195 } 1196 1197 /* 1198 * Write the core file header to the file, including padding up to 1199 * the page boundary. 1200 */ 1201 static int 1202 elf_corehdr(struct proc *p, struct file *fp, struct ucred *cred, int numsegs, 1203 elf_buf_t target) 1204 { 1205 struct { 1206 prstatus_t status; 1207 prfpregset_t fpregset; 1208 prpsinfo_t psinfo; 1209 } *tempdata; 1210 int error; 1211 prstatus_t *status; 1212 prfpregset_t *fpregset; 1213 prpsinfo_t *psinfo; 1214 int nbytes; 1215 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK); 1216 status = &tempdata->status; 1217 fpregset = &tempdata->fpregset; 1218 psinfo = &tempdata->psinfo; 1219 1220 /* Gather the information for the header. */ 1221 status->pr_version = PRSTATUS_VERSION; 1222 status->pr_statussz = sizeof(prstatus_t); 1223 status->pr_gregsetsz = sizeof(gregset_t); 1224 status->pr_fpregsetsz = sizeof(fpregset_t); 1225 status->pr_osreldate = osreldate; 1226 status->pr_cursig = p->p_sig; 1227 status->pr_pid = p->p_pid; 1228 fill_regs(p, &status->pr_reg); 1229 1230 fill_fpregs(p, fpregset); 1231 1232 psinfo->pr_version = PRPSINFO_VERSION; 1233 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 1234 strncpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname) - 1); 1235 1236 /* XXX - We don't fill in the command line arguments properly yet. */ 1237 strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ); 1238 1239 /* Fill in the header. */ 1240 error = elf_puthdr(p, target, status, fpregset, psinfo, numsegs); 1241 1242 free(tempdata, M_TEMP); 1243 1244 /* Write it to the core file. */ 1245 if (error == 0) 1246 error = fp_write(fp, target->buf, target->off, &nbytes); 1247 return error; 1248 } 1249 1250 static int 1251 elf_puthdr(struct proc *p, elf_buf_t target, const prstatus_t *status, 1252 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs) 1253 { 1254 int error = 0; 1255 size_t phoff; 1256 size_t noteoff; 1257 size_t notesz; 1258 Elf_Ehdr *ehdr; 1259 Elf_Phdr *phdr; 1260 1261 ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error); 1262 1263 phoff = target->off; 1264 phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error); 1265 1266 noteoff = target->off; 1267 if (error == 0) { 1268 error = elf_putnote(target, "FreeBSD", NT_PRSTATUS, 1269 status, sizeof *status); 1270 } 1271 if (error == 0) { 1272 error = elf_putnote(target, "FreeBSD", NT_FPREGSET, 1273 fpregset, sizeof *fpregset); 1274 } 1275 if (error == 0) { 1276 error = elf_putnote(target, "FreeBSD", NT_PRPSINFO, 1277 psinfo, sizeof *psinfo); 1278 } 1279 notesz = target->off - noteoff; 1280 1281 /* 1282 * put extra cruft for dumping process state here 1283 * - we really want it be before all the program 1284 * mappings 1285 * - we just need to update the offset accordingly 1286 * and GDB will be none the wiser. 1287 */ 1288 if (error == 0) 1289 error = elf_puttextvp(p, target); 1290 if (error == 0) 1291 error = elf_putsigs(p, target); 1292 if (error == 0) 1293 error = elf_putfiles(p, target); 1294 1295 /* 1296 * Align up to a page boundary for the program segments. The 1297 * actual data will be written to the outptu file, not to elf_buf_t, 1298 * so we do not have to do any further bounds checking. 1299 */ 1300 target->off = round_page(target->off); 1301 if (error == 0 && ehdr != NULL) { 1302 /* 1303 * Fill in the ELF header. 1304 */ 1305 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1306 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1307 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1308 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1309 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1310 ehdr->e_ident[EI_DATA] = ELF_DATA; 1311 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1312 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1313 ehdr->e_ident[EI_ABIVERSION] = 0; 1314 ehdr->e_ident[EI_PAD] = 0; 1315 ehdr->e_type = ET_CORE; 1316 ehdr->e_machine = ELF_ARCH; 1317 ehdr->e_version = EV_CURRENT; 1318 ehdr->e_entry = 0; 1319 ehdr->e_phoff = phoff; 1320 ehdr->e_flags = 0; 1321 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1322 ehdr->e_phentsize = sizeof(Elf_Phdr); 1323 ehdr->e_phnum = numsegs + 1; 1324 ehdr->e_shentsize = sizeof(Elf_Shdr); 1325 ehdr->e_shnum = 0; 1326 ehdr->e_shstrndx = SHN_UNDEF; 1327 } 1328 if (error == 0 && phdr != NULL) { 1329 /* 1330 * Fill in the program header entries. 1331 */ 1332 struct phdr_closure phc; 1333 1334 /* The note segement. */ 1335 phdr->p_type = PT_NOTE; 1336 phdr->p_offset = noteoff; 1337 phdr->p_vaddr = 0; 1338 phdr->p_paddr = 0; 1339 phdr->p_filesz = notesz; 1340 phdr->p_memsz = 0; 1341 phdr->p_flags = 0; 1342 phdr->p_align = 0; 1343 ++phdr; 1344 1345 /* All the writable segments from the program. */ 1346 phc.phdr = phdr; 1347 phc.phdr_max = phdr + numsegs; 1348 phc.offset = target->off; 1349 each_segment(p, cb_put_phdr, &phc, 1); 1350 } 1351 return (error); 1352 } 1353 1354 static int 1355 elf_putnote(elf_buf_t target, const char *name, int type, 1356 const void *desc, size_t descsz) 1357 { 1358 int error = 0; 1359 char *dst; 1360 Elf_Note note; 1361 1362 note.n_namesz = strlen(name) + 1; 1363 note.n_descsz = descsz; 1364 note.n_type = type; 1365 dst = target_reserve(target, sizeof(note), &error); 1366 if (dst != NULL) 1367 bcopy(¬e, dst, sizeof note); 1368 dst = target_reserve(target, note.n_namesz, &error); 1369 if (dst != NULL) 1370 bcopy(name, dst, note.n_namesz); 1371 target->off = roundup2(target->off, sizeof(Elf_Size)); 1372 dst = target_reserve(target, note.n_descsz, &error); 1373 if (dst != NULL) 1374 bcopy(desc, dst, note.n_descsz); 1375 target->off = roundup2(target->off, sizeof(Elf_Size)); 1376 return(error); 1377 } 1378 1379 1380 static int 1381 elf_putsigs(struct proc *p, elf_buf_t target) 1382 { 1383 int error = 0; 1384 struct ckpt_siginfo *csi; 1385 1386 csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error); 1387 if (csi) { 1388 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo); 1389 bcopy(p->p_procsig, &csi->csi_procsig, sizeof(struct procsig)); 1390 bcopy(p->p_procsig->ps_sigacts, &csi->csi_sigacts, sizeof(struct sigacts)); 1391 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval)); 1392 bcopy(&p->p_sigmask, &csi->csi_sigmask, sizeof(sigset_t)); 1393 csi->csi_sigparent = p->p_sigparent; 1394 } 1395 return(error); 1396 } 1397 1398 static int 1399 elf_putfiles(struct proc *p, elf_buf_t target) 1400 { 1401 int error = 0; 1402 int i; 1403 struct ckpt_filehdr *cfh = NULL; 1404 struct ckpt_fileinfo *cfi; 1405 struct file *fp; 1406 struct vnode *vp; 1407 /* 1408 * the duplicated loop is gross, but it was the only way 1409 * to eliminate uninitialized variable warnings 1410 */ 1411 cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error); 1412 if (cfh) { 1413 cfh->cfh_nfiles = 0; 1414 } 1415 1416 /* 1417 * ignore STDIN/STDERR/STDOUT. 1418 */ 1419 for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) { 1420 if ((fp = p->p_fd->fd_files[i].fp) == NULL) 1421 continue; 1422 /* 1423 * XXX Only checkpoint vnodes for now. 1424 */ 1425 if (fp->f_type != DTYPE_VNODE) 1426 continue; 1427 cfi = target_reserve(target, sizeof(struct ckpt_fileinfo), 1428 &error); 1429 if (cfi == NULL) 1430 continue; 1431 cfi->cfi_index = -1; 1432 cfi->cfi_type = fp->f_type; 1433 cfi->cfi_flags = fp->f_flag; 1434 cfi->cfi_offset = fp->f_offset; 1435 /* f_count and f_msgcount should not be saved/restored */ 1436 /* XXX save cred info */ 1437 1438 switch(fp->f_type) { 1439 case DTYPE_VNODE: 1440 vp = (struct vnode *)fp->f_data; 1441 /* 1442 * it looks like a bug in ptrace is marking 1443 * a non-vnode as a vnode - until we find the 1444 * root cause this will at least prevent 1445 * further panics from truss 1446 */ 1447 if (vp == NULL || vp->v_mount == NULL) 1448 break; 1449 cfh->cfh_nfiles++; 1450 cfi->cfi_index = i; 1451 cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1452 error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid); 1453 break; 1454 default: 1455 break; 1456 } 1457 } 1458 return(error); 1459 } 1460 1461 static int 1462 elf_puttextvp(struct proc *p, elf_buf_t target) 1463 { 1464 int error = 0; 1465 int *vn_count; 1466 struct fp_closure fpc; 1467 struct ckpt_vminfo *vminfo; 1468 1469 vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error); 1470 if (vminfo != NULL) { 1471 vminfo->cvm_dsize = p->p_vmspace->vm_dsize; 1472 vminfo->cvm_tsize = p->p_vmspace->vm_tsize; 1473 vminfo->cvm_daddr = p->p_vmspace->vm_daddr; 1474 vminfo->cvm_taddr = p->p_vmspace->vm_taddr; 1475 } 1476 1477 fpc.count = 0; 1478 vn_count = target_reserve(target, sizeof(int), &error); 1479 if (target->buf != NULL) { 1480 fpc.vnh = (struct vn_hdr *)(target->buf + target->off); 1481 fpc.vnh_max = fpc.vnh + 1482 (target->off_max - target->off) / sizeof(struct vn_hdr); 1483 error = each_segment(p, cb_put_fp, &fpc, 0); 1484 if (vn_count) 1485 *vn_count = fpc.count; 1486 } else { 1487 error = each_segment(p, cb_fpcount_segment, &fpc.count, 0); 1488 } 1489 target->off += fpc.count * sizeof(struct vn_hdr); 1490 return(error); 1491 } 1492 1493 1494 /* 1495 * Tell kern_execve.c about it, with a little help from the linker. 1496 */ 1497 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"}; 1498 EXEC_SET_ORDERED(elf, elf_execsw, SI_ORDER_FIRST); 1499