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 1.39 2006/05/25 04:17:09 dillon 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 /* 197 * Check if an elf brand is being used anywhere in the system. 198 * 199 * Used by the linux emulatino module unloader. This isn't safe from 200 * races. 201 */ 202 struct elf_brand_inuse_info { 203 int rval; 204 Elf_Brandinfo *entry; 205 }; 206 207 static int elf_brand_inuse_callback(struct proc *p, void *data); 208 209 int 210 elf_brand_inuse(Elf_Brandinfo *entry) 211 { 212 struct elf_brand_inuse_info info; 213 214 info.rval = FALSE; 215 info.entry = entry; 216 allproc_scan(elf_brand_inuse_callback, entry); 217 return (info.rval); 218 } 219 220 static 221 int 222 elf_brand_inuse_callback(struct proc *p, void *data) 223 { 224 struct elf_brand_inuse_info *info = data; 225 226 if (p->p_sysent == info->entry->sysvec) { 227 info->rval = TRUE; 228 return(-1); 229 } 230 return(0); 231 } 232 233 static int 234 elf_check_header(const Elf_Ehdr *hdr) 235 { 236 if (!IS_ELF(*hdr) || 237 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 238 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 239 hdr->e_ident[EI_VERSION] != EV_CURRENT || 240 hdr->e_phentsize != sizeof(Elf_Phdr) || 241 hdr->e_ehsize != sizeof(Elf_Ehdr) || 242 hdr->e_version != ELF_TARG_VER) 243 return ENOEXEC; 244 245 if (!ELF_MACHINE_OK(hdr->e_machine)) 246 return ENOEXEC; 247 248 return 0; 249 } 250 251 static int 252 elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp, 253 vm_offset_t offset, caddr_t vmaddr, size_t memsz, 254 size_t filsz, vm_prot_t prot) 255 { 256 size_t map_len; 257 vm_offset_t map_addr; 258 int error, rv, cow; 259 int count; 260 size_t copy_len; 261 vm_object_t object; 262 vm_offset_t file_addr; 263 vm_offset_t data_buf = 0; 264 265 object = vp->v_object; 266 error = 0; 267 268 /* 269 * It's necessary to fail if the filsz + offset taken from the 270 * header is greater than the actual file pager object's size. 271 * If we were to allow this, then the vm_map_find() below would 272 * walk right off the end of the file object and into the ether. 273 * 274 * While I'm here, might as well check for something else that 275 * is invalid: filsz cannot be greater than memsz. 276 */ 277 if ((off_t)filsz + offset > vp->v_filesize || filsz > memsz) { 278 uprintf("elf_load_section: truncated ELF file\n"); 279 return (ENOEXEC); 280 } 281 282 map_addr = trunc_page((vm_offset_t)vmaddr); 283 file_addr = trunc_page(offset); 284 285 /* 286 * We have two choices. We can either clear the data in the last page 287 * of an oversized mapping, or we can start the anon mapping a page 288 * early and copy the initialized data into that first page. We 289 * choose the second.. 290 */ 291 if (memsz > filsz) 292 map_len = trunc_page(offset+filsz) - file_addr; 293 else 294 map_len = round_page(offset+filsz) - file_addr; 295 296 if (map_len != 0) { 297 vm_object_reference(object); 298 299 /* cow flags: don't dump readonly sections in core */ 300 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 301 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 302 303 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 304 vm_map_lock(&vmspace->vm_map); 305 rv = vm_map_insert(&vmspace->vm_map, &count, 306 object, 307 file_addr, /* file offset */ 308 map_addr, /* virtual start */ 309 map_addr + map_len,/* virtual end */ 310 prot, 311 VM_PROT_ALL, 312 cow); 313 vm_map_unlock(&vmspace->vm_map); 314 vm_map_entry_release(count); 315 if (rv != KERN_SUCCESS) { 316 vm_object_deallocate(object); 317 return EINVAL; 318 } 319 320 /* we can stop now if we've covered it all */ 321 if (memsz == filsz) { 322 return 0; 323 } 324 } 325 326 327 /* 328 * We have to get the remaining bit of the file into the first part 329 * of the oversized map segment. This is normally because the .data 330 * segment in the file is extended to provide bss. It's a neat idea 331 * to try and save a page, but it's a pain in the behind to implement. 332 */ 333 copy_len = (offset + filsz) - trunc_page(offset + filsz); 334 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 335 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 336 337 /* This had damn well better be true! */ 338 if (map_len != 0) { 339 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 340 vm_map_lock(&vmspace->vm_map); 341 rv = vm_map_insert(&vmspace->vm_map, &count, 342 NULL, 0, 343 map_addr, map_addr + map_len, 344 VM_PROT_ALL, VM_PROT_ALL, 0); 345 vm_map_unlock(&vmspace->vm_map); 346 vm_map_entry_release(count); 347 if (rv != KERN_SUCCESS) { 348 return EINVAL; 349 } 350 } 351 352 if (copy_len != 0) { 353 vm_object_reference(object); 354 rv = vm_map_find(exec_map, 355 object, 356 trunc_page(offset + filsz), 357 &data_buf, 358 PAGE_SIZE, 359 TRUE, 360 VM_PROT_READ, 361 VM_PROT_ALL, 362 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 363 if (rv != KERN_SUCCESS) { 364 vm_object_deallocate(object); 365 return EINVAL; 366 } 367 368 /* send the page fragment to user space */ 369 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len); 370 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 371 if (error) { 372 return (error); 373 } 374 } 375 376 /* 377 * set it to the specified protection 378 */ 379 vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len, prot, 380 FALSE); 381 382 return error; 383 } 384 385 /* 386 * Load the file "file" into memory. It may be either a shared object 387 * or an executable. 388 * 389 * The "addr" reference parameter is in/out. On entry, it specifies 390 * the address where a shared object should be loaded. If the file is 391 * an executable, this value is ignored. On exit, "addr" specifies 392 * where the file was actually loaded. 393 * 394 * The "entry" reference parameter is out only. On exit, it specifies 395 * the entry point for the loaded file. 396 */ 397 static int 398 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry) 399 { 400 struct { 401 struct nlookupdata nd; 402 struct vattr attr; 403 struct image_params image_params; 404 } *tempdata; 405 const Elf_Ehdr *hdr = NULL; 406 const Elf_Phdr *phdr = NULL; 407 struct nlookupdata *nd; 408 struct vmspace *vmspace = p->p_vmspace; 409 struct vattr *attr; 410 struct image_params *imgp; 411 vm_prot_t prot; 412 u_long rbase; 413 u_long base_addr = 0; 414 int error, i, numsegs; 415 416 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 417 nd = &tempdata->nd; 418 attr = &tempdata->attr; 419 imgp = &tempdata->image_params; 420 421 /* 422 * Initialize part of the common data 423 */ 424 imgp->proc = p; 425 imgp->attr = attr; 426 imgp->firstpage = NULL; 427 imgp->image_header = NULL; 428 imgp->vp = NULL; 429 430 error = nlookup_init(nd, file, UIO_SYSSPACE, NLC_FOLLOW); 431 if (error == 0) 432 error = nlookup(nd); 433 if (error == 0) 434 error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 435 nlookup_done(nd); 436 if (error) 437 goto fail; 438 439 /* 440 * Check permissions, modes, uid, etc on the file, and "open" it. 441 */ 442 error = exec_check_permissions(imgp); 443 if (error) { 444 VOP_UNLOCK(imgp->vp, 0); 445 goto fail; 446 } 447 448 error = exec_map_first_page(imgp); 449 /* 450 * Also make certain that the interpreter stays the same, so set 451 * its VTEXT flag, too. 452 */ 453 if (error == 0) 454 imgp->vp->v_flag |= VTEXT; 455 VOP_UNLOCK(imgp->vp, 0); 456 if (error) 457 goto fail; 458 459 hdr = (const Elf_Ehdr *)imgp->image_header; 460 if ((error = elf_check_header(hdr)) != 0) 461 goto fail; 462 if (hdr->e_type == ET_DYN) 463 rbase = *addr; 464 else if (hdr->e_type == ET_EXEC) 465 rbase = 0; 466 else { 467 error = ENOEXEC; 468 goto fail; 469 } 470 471 /* Only support headers that fit within first page for now 472 * (multiplication of two Elf_Half fields will not overflow) */ 473 if ((hdr->e_phoff > PAGE_SIZE) || 474 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 475 error = ENOEXEC; 476 goto fail; 477 } 478 479 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 480 481 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 482 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 483 prot = 0; 484 if (phdr[i].p_flags & PF_X) 485 prot |= VM_PROT_EXECUTE; 486 if (phdr[i].p_flags & PF_W) 487 prot |= VM_PROT_WRITE; 488 if (phdr[i].p_flags & PF_R) 489 prot |= VM_PROT_READ; 490 491 error = elf_load_section( 492 p, vmspace, imgp->vp, 493 phdr[i].p_offset, 494 (caddr_t)phdr[i].p_vaddr + 495 rbase, 496 phdr[i].p_memsz, 497 phdr[i].p_filesz, prot); 498 if (error != 0) 499 goto fail; 500 /* 501 * Establish the base address if this is the 502 * first segment. 503 */ 504 if (numsegs == 0) 505 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 506 numsegs++; 507 } 508 } 509 *addr = base_addr; 510 *entry=(unsigned long)hdr->e_entry + rbase; 511 512 fail: 513 if (imgp->firstpage) 514 exec_unmap_first_page(imgp); 515 if (imgp->vp) { 516 vrele(imgp->vp); 517 imgp->vp = NULL; 518 } 519 free(tempdata, M_TEMP); 520 521 return error; 522 } 523 524 /* 525 * non static, as it can be overridden by start_init() 526 */ 527 int fallback_elf_brand = -1; 528 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW, 529 &fallback_elf_brand, -1, 530 "ELF brand of last resort"); 531 532 static int 533 exec_elf_imgact(struct image_params *imgp) 534 { 535 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header; 536 const Elf_Phdr *phdr; 537 Elf_Auxargs *elf_auxargs = NULL; 538 struct vmspace *vmspace; 539 vm_prot_t prot; 540 u_long text_size = 0, data_size = 0, total_size = 0; 541 u_long text_addr = 0, data_addr = 0; 542 u_long seg_size, seg_addr; 543 u_long addr, entry = 0, proghdr = 0; 544 int error, i; 545 const char *interp = NULL; 546 const Elf_Note *abi_note = NULL; 547 Elf_Brandinfo *brand_info; 548 char *path; 549 550 error = 0; 551 552 /* 553 * Do we have a valid ELF header ? 554 */ 555 if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC) 556 return -1; 557 558 /* 559 * From here on down, we return an errno, not -1, as we've 560 * detected an ELF file. 561 */ 562 563 if ((hdr->e_phoff > PAGE_SIZE) || 564 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 565 /* Only support headers in first page for now */ 566 return ENOEXEC; 567 } 568 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff); 569 570 /* 571 * From this point on, we may have resources that need to be freed. 572 */ 573 574 exec_new_vmspace(imgp, NULL); 575 576 /* 577 * Yeah, I'm paranoid. There is every reason in the world to get 578 * VTEXT now since from here on out, there are places we can have 579 * a context switch. Better safe than sorry; I really don't want 580 * the file to change while it's being loaded. 581 */ 582 vsetflags(imgp->vp, VTEXT); 583 584 vmspace = imgp->proc->p_vmspace; 585 586 for (i = 0; i < hdr->e_phnum; i++) { 587 switch(phdr[i].p_type) { 588 589 case PT_LOAD: /* Loadable segment */ 590 prot = 0; 591 if (phdr[i].p_flags & PF_X) 592 prot |= VM_PROT_EXECUTE; 593 if (phdr[i].p_flags & PF_W) 594 prot |= VM_PROT_WRITE; 595 if (phdr[i].p_flags & PF_R) 596 prot |= VM_PROT_READ; 597 598 if ((error = elf_load_section(imgp->proc, 599 vmspace, imgp->vp, 600 phdr[i].p_offset, 601 (caddr_t)phdr[i].p_vaddr, 602 phdr[i].p_memsz, 603 phdr[i].p_filesz, prot)) != 0) 604 goto fail; 605 606 /* 607 * If this segment contains the program headers, 608 * remember their virtual address for the AT_PHDR 609 * aux entry. Static binaries don't usually include 610 * a PT_PHDR entry. 611 */ 612 if (phdr[i].p_offset == 0 && 613 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 614 <= phdr[i].p_filesz) 615 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 616 617 seg_addr = trunc_page(phdr[i].p_vaddr); 618 seg_size = round_page(phdr[i].p_memsz + 619 phdr[i].p_vaddr - seg_addr); 620 621 /* 622 * Is this .text or .data? We can't use 623 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 624 * alpha terribly and possibly does other bad 625 * things so we stick to the old way of figuring 626 * it out: If the segment contains the program 627 * entry point, it's a text segment, otherwise it 628 * is a data segment. 629 * 630 * Note that obreak() assumes that data_addr + 631 * data_size == end of data load area, and the ELF 632 * file format expects segments to be sorted by 633 * address. If multiple data segments exist, the 634 * last one will be used. 635 */ 636 if (hdr->e_entry >= phdr[i].p_vaddr && 637 hdr->e_entry < (phdr[i].p_vaddr + 638 phdr[i].p_memsz)) { 639 text_size = seg_size; 640 text_addr = seg_addr; 641 entry = (u_long)hdr->e_entry; 642 } else { 643 data_size = seg_size; 644 data_addr = seg_addr; 645 } 646 total_size += seg_size; 647 648 /* 649 * Check limits. It should be safe to check the 650 * limits after loading the segment since we do 651 * not actually fault in all the segment's pages. 652 */ 653 if (data_size > 654 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur || 655 text_size > maxtsiz || 656 total_size > 657 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) { 658 error = ENOMEM; 659 goto fail; 660 } 661 break; 662 case PT_INTERP: /* Path to interpreter */ 663 if (phdr[i].p_filesz > MAXPATHLEN || 664 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) { 665 error = ENOEXEC; 666 goto fail; 667 } 668 interp = imgp->image_header + phdr[i].p_offset; 669 break; 670 case PT_NOTE: /* Check for .note.ABI-tag */ 671 { 672 const Elf_Note *tmp_note; 673 /* XXX handle anything outside the first page */ 674 if (phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) 675 continue; 676 if (phdr[i].p_filesz < sizeof(Elf_Note)) 677 continue; /* ENOEXEC? */ 678 tmp_note = (const Elf_Note *)(imgp->image_header + phdr[i].p_offset); 679 if (tmp_note->n_type != 1) 680 continue; 681 if (tmp_note->n_namesz + sizeof(Elf_Note) + 682 tmp_note->n_descsz > phdr[i].p_filesz) 683 continue; /* ENOEXEC? */ 684 abi_note = tmp_note; 685 } 686 break; 687 case PT_PHDR: /* Program header table info */ 688 proghdr = phdr[i].p_vaddr; 689 break; 690 default: 691 break; 692 } 693 } 694 695 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 696 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 697 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 698 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 699 700 addr = ELF_RTLD_ADDR(vmspace); 701 702 imgp->entry_addr = entry; 703 704 brand_info = NULL; 705 706 /* We support three types of branding -- (1) the ELF EI_OSABI field 707 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 708 * branding w/in the ELF header, and (3) path of the `interp_path' 709 * field. We should also look for an ".note.ABI-tag" ELF section now 710 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones. 711 */ 712 713 /* If the executable has a brand, search for it in the brand list. */ 714 if (brand_info == NULL && hdr->e_ident[EI_OSABI] != ELFOSABI_NONE) { 715 for (i = 0; i < MAX_BRANDS; i++) { 716 Elf_Brandinfo *bi = elf_brand_list[i]; 717 718 if (bi != NULL && 719 (hdr->e_ident[EI_OSABI] == bi->brand 720 || 0 == 721 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 722 bi->compat_3_brand, strlen(bi->compat_3_brand)))) { 723 brand_info = bi; 724 break; 725 } 726 } 727 } 728 729 /* Search for a recognized ABI. */ 730 if (brand_info == NULL && abi_note != NULL) { 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 (*bi->match_abi_note)(abi_note)) { 736 brand_info = bi; 737 break; 738 } 739 } 740 } 741 742 /* 743 * ELFOSABI_NONE == ELFOSABI_SYSV, so a SYSV binary misses all 744 * checks so far, since it is neither branded nor does it have 745 * an ABI note. If the EI_OSABI field is ELFOSABI_NONE, assume 746 * it is svr4 and look for an entry in the elf_brand_list with 747 * match_abi_note == NULL. 748 */ 749 if (brand_info == NULL && hdr->e_ident[EI_OSABI] == ELFOSABI_NONE) { 750 for (i = 0; i < MAX_BRANDS; i++) { 751 Elf_Brandinfo *bi = elf_brand_list[i]; 752 753 if (bi != NULL && bi->match_abi_note == NULL && 754 ELFOSABI_SYSV == bi->brand) { 755 brand_info = bi; 756 break; 757 } 758 } 759 } 760 761 /* Lacking a recognized ABI, search for a recognized interpreter. */ 762 if (brand_info == NULL && interp != NULL) { 763 for (i = 0; i < MAX_BRANDS; i++) { 764 Elf_Brandinfo *bi = elf_brand_list[i]; 765 766 if (bi != NULL && 767 strcmp(interp, bi->interp_path) == 0) { 768 brand_info = bi; 769 break; 770 } 771 } 772 } 773 774 /* Lacking a recognized interpreter, try the default brand */ 775 if (brand_info == NULL) { 776 for (i = 0; i < MAX_BRANDS; i++) { 777 Elf_Brandinfo *bi = elf_brand_list[i]; 778 779 if (bi != NULL && fallback_elf_brand == bi->brand) { 780 brand_info = bi; 781 break; 782 } 783 } 784 } 785 786 if (brand_info == NULL) { 787 uprintf("ELF binary type \"%u\" not known.\n", 788 hdr->e_ident[EI_OSABI]); 789 error = ENOEXEC; 790 goto fail; 791 } 792 793 imgp->proc->p_sysent = brand_info->sysvec; 794 if (interp != NULL) { 795 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 796 snprintf(path, MAXPATHLEN, "%s%s", 797 brand_info->emul_path, interp); 798 if ((error = elf_load_file(imgp->proc, path, &addr, 799 &imgp->entry_addr)) != 0) { 800 if ((error = elf_load_file(imgp->proc, interp, &addr, 801 &imgp->entry_addr)) != 0) { 802 uprintf("ELF interpreter %s not found\n", path); 803 free(path, M_TEMP); 804 goto fail; 805 } 806 } 807 free(path, M_TEMP); 808 } 809 810 /* 811 * Construct auxargs table (used by the fixup routine) 812 */ 813 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 814 elf_auxargs->execfd = -1; 815 elf_auxargs->phdr = proghdr; 816 elf_auxargs->phent = hdr->e_phentsize; 817 elf_auxargs->phnum = hdr->e_phnum; 818 elf_auxargs->pagesz = PAGE_SIZE; 819 elf_auxargs->base = addr; 820 elf_auxargs->flags = 0; 821 elf_auxargs->entry = entry; 822 elf_auxargs->trace = elf_trace; 823 824 imgp->auxargs = elf_auxargs; 825 imgp->interpreted = 0; 826 827 fail: 828 return error; 829 } 830 831 static int 832 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp) 833 { 834 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 835 register_t *pos; 836 837 pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2); 838 839 if (args->trace) { 840 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 841 } 842 if (args->execfd != -1) { 843 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 844 } 845 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 846 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 847 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 848 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 849 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 850 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 851 AUXARGS_ENTRY(pos, AT_BASE, args->base); 852 AUXARGS_ENTRY(pos, AT_NULL, 0); 853 854 free(imgp->auxargs, M_TEMP); 855 imgp->auxargs = NULL; 856 857 (*stack_base)--; 858 suword(*stack_base, (long) imgp->args->argc); 859 return 0; 860 } 861 862 /* 863 * Code for generating ELF core dumps. 864 */ 865 866 typedef int (*segment_callback) (vm_map_entry_t, void *); 867 868 /* Closure for cb_put_phdr(). */ 869 struct phdr_closure { 870 Elf_Phdr *phdr; /* Program header to fill in (incremented) */ 871 Elf_Phdr *phdr_max; /* Pointer bound for error check */ 872 Elf_Off offset; /* Offset of segment in core file */ 873 }; 874 875 /* Closure for cb_size_segment(). */ 876 struct sseg_closure { 877 int count; /* Count of writable segments. */ 878 size_t vsize; /* Total size of all writable segments. */ 879 }; 880 881 /* Closure for cb_put_fp(). */ 882 struct fp_closure { 883 struct vn_hdr *vnh; 884 struct vn_hdr *vnh_max; 885 int count; 886 struct stat *sb; 887 }; 888 889 typedef struct elf_buf { 890 char *buf; 891 size_t off; 892 size_t off_max; 893 } *elf_buf_t; 894 895 static void *target_reserve(elf_buf_t target, size_t bytes, int *error); 896 897 static int cb_put_phdr (vm_map_entry_t, void *); 898 static int cb_size_segment (vm_map_entry_t, void *); 899 static int cb_fpcount_segment(vm_map_entry_t, void *); 900 static int cb_put_fp(vm_map_entry_t, void *); 901 902 903 static int each_segment (struct proc *, segment_callback, void *, int); 904 static int elf_corehdr (struct proc *, struct file *, struct ucred *, 905 int, elf_buf_t); 906 static int elf_puthdr (struct proc *, elf_buf_t, const prstatus_t *, 907 const prfpregset_t *, const prpsinfo_t *, int); 908 static int elf_putnote (elf_buf_t, const char *, int, const void *, size_t); 909 910 static int elf_putsigs(struct proc *, elf_buf_t); 911 static int elf_puttextvp(struct proc *, elf_buf_t); 912 static int elf_putfiles(struct proc *, elf_buf_t); 913 914 extern int osreldate; 915 916 int 917 elf_coredump(struct proc *p, struct vnode *vp, off_t limit) 918 { 919 struct file *fp; 920 int error; 921 922 if ((error = falloc(NULL, &fp, NULL)) != 0) 923 return (error); 924 fsetcred(fp, p->p_ucred); 925 926 /* 927 * XXX fixme. 928 */ 929 fp->f_type = DTYPE_VNODE; 930 fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW; 931 fp->f_ops = &vnode_fileops; 932 fp->f_data = vp; 933 VOP_UNLOCK(vp, 0); 934 935 error = generic_elf_coredump(p, fp, limit); 936 937 fp->f_type = 0; 938 fp->f_flag = 0; 939 fp->f_ops = &badfileops; 940 fp->f_data = NULL; 941 fdrop(fp); 942 return (error); 943 } 944 945 int 946 generic_elf_coredump(struct proc *p, struct file *fp, off_t limit) 947 { 948 struct ucred *cred = p->p_ucred; 949 int error = 0; 950 struct sseg_closure seginfo; 951 struct elf_buf target; 952 953 if (!fp) 954 printf("can't dump core - null fp\n"); 955 956 /* 957 * Size the program segments 958 */ 959 seginfo.count = 0; 960 seginfo.vsize = 0; 961 each_segment(p, cb_size_segment, &seginfo, 1); 962 963 /* 964 * Calculate the size of the core file header area by making 965 * a dry run of generating it. Nothing is written, but the 966 * size is calculated. 967 */ 968 bzero(&target, sizeof(target)); 969 elf_puthdr(p, &target, NULL, NULL, NULL, seginfo.count); 970 971 if (target.off + seginfo.vsize >= limit) 972 return (EFAULT); 973 974 /* 975 * Allocate memory for building the header, fill it up, 976 * and write it out. 977 */ 978 target.off_max = target.off; 979 target.off = 0; 980 target.buf = malloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO); 981 982 if (target.buf == NULL) 983 return EINVAL; 984 error = elf_corehdr(p, fp, cred, seginfo.count, &target); 985 986 /* Write the contents of all of the writable segments. */ 987 if (error == 0) { 988 Elf_Phdr *php; 989 int i; 990 int nbytes; 991 992 php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1; 993 for (i = 0; i < seginfo.count; i++) { 994 error = fp_write(fp, (caddr_t)php->p_vaddr, 995 php->p_filesz, &nbytes); 996 if (error != 0) 997 break; 998 php++; 999 } 1000 } 1001 free(target.buf, M_TEMP); 1002 1003 return error; 1004 } 1005 1006 /* 1007 * A callback for each_segment() to write out the segment's 1008 * program header entry. 1009 */ 1010 static int 1011 cb_put_phdr(vm_map_entry_t entry, void *closure) 1012 { 1013 struct phdr_closure *phc = closure; 1014 Elf_Phdr *phdr = phc->phdr; 1015 1016 if (phc->phdr == phc->phdr_max) 1017 return EINVAL; 1018 1019 phc->offset = round_page(phc->offset); 1020 1021 phdr->p_type = PT_LOAD; 1022 phdr->p_offset = phc->offset; 1023 phdr->p_vaddr = entry->start; 1024 phdr->p_paddr = 0; 1025 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1026 phdr->p_align = PAGE_SIZE; 1027 phdr->p_flags = 0; 1028 if (entry->protection & VM_PROT_READ) 1029 phdr->p_flags |= PF_R; 1030 if (entry->protection & VM_PROT_WRITE) 1031 phdr->p_flags |= PF_W; 1032 if (entry->protection & VM_PROT_EXECUTE) 1033 phdr->p_flags |= PF_X; 1034 1035 phc->offset += phdr->p_filesz; 1036 ++phc->phdr; 1037 return 0; 1038 } 1039 1040 /* 1041 * A callback for each_writable_segment() to gather information about 1042 * the number of segments and their total size. 1043 */ 1044 static int 1045 cb_size_segment(vm_map_entry_t entry, void *closure) 1046 { 1047 struct sseg_closure *ssc = closure; 1048 1049 ++ssc->count; 1050 ssc->vsize += entry->end - entry->start; 1051 return 0; 1052 } 1053 1054 /* 1055 * A callback for each_segment() to gather information about 1056 * the number of text segments. 1057 */ 1058 static int 1059 cb_fpcount_segment(vm_map_entry_t entry, void *closure) 1060 { 1061 int *count = closure; 1062 struct vnode *vp; 1063 1064 if (entry->object.vm_object->type == OBJT_VNODE) { 1065 vp = (struct vnode *)entry->object.vm_object->handle; 1066 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp) 1067 return 0; 1068 ++*count; 1069 } 1070 return 0; 1071 } 1072 1073 static int 1074 cb_put_fp(vm_map_entry_t entry, void *closure) 1075 { 1076 struct fp_closure *fpc = closure; 1077 struct vn_hdr *vnh = fpc->vnh; 1078 Elf_Phdr *phdr = &vnh->vnh_phdr; 1079 struct vnode *vp; 1080 int error; 1081 1082 /* 1083 * If an entry represents a vnode then write out a file handle. 1084 * 1085 * If we are checkpointing a checkpoint-restored program we do 1086 * NOT record the filehandle for the old checkpoint vnode (which 1087 * is mapped all over the place). Instead we rely on the fact 1088 * that a checkpoint-restored program does not mmap() the checkpt 1089 * vnode NOCORE, so its contents will be written out to the 1090 * new checkpoint file. This is necessary because the 'old' 1091 * checkpoint file is typically destroyed when a new one is created 1092 * and thus cannot be used to restore the new checkpoint. 1093 * 1094 * Theoretically we could create a chain of checkpoint files and 1095 * operate the checkpointing operation kinda like an incremental 1096 * checkpoint, but a checkpoint restore would then likely wind up 1097 * referencing many prior checkpoint files and that is a bit over 1098 * the top for the purpose of the checkpoint API. 1099 */ 1100 if (entry->object.vm_object->type == OBJT_VNODE) { 1101 vp = (struct vnode *)entry->object.vm_object->handle; 1102 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp) 1103 return 0; 1104 if (vnh == fpc->vnh_max) 1105 return EINVAL; 1106 1107 if (vp->v_mount) 1108 vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1109 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid); 1110 if (error) { 1111 char *freepath, *fullpath; 1112 1113 if (vn_fullpath(curproc, vp, &fullpath, &freepath)) { 1114 printf("Warning: coredump, error %d: cannot store file handle for vnode %p\n", error, vp); 1115 } else { 1116 printf("Warning: coredump, error %d: cannot store file handle for %s\n", error, fullpath); 1117 free(freepath, M_TEMP); 1118 } 1119 error = 0; 1120 } 1121 1122 phdr->p_type = PT_LOAD; 1123 phdr->p_offset = 0; /* not written to core */ 1124 phdr->p_vaddr = entry->start; 1125 phdr->p_paddr = 0; 1126 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1127 phdr->p_align = PAGE_SIZE; 1128 phdr->p_flags = 0; 1129 if (entry->protection & VM_PROT_READ) 1130 phdr->p_flags |= PF_R; 1131 if (entry->protection & VM_PROT_WRITE) 1132 phdr->p_flags |= PF_W; 1133 if (entry->protection & VM_PROT_EXECUTE) 1134 phdr->p_flags |= PF_X; 1135 ++fpc->vnh; 1136 ++fpc->count; 1137 } 1138 return 0; 1139 } 1140 1141 /* 1142 * For each writable segment in the process's memory map, call the given 1143 * function with a pointer to the map entry and some arbitrary 1144 * caller-supplied data. 1145 */ 1146 static int 1147 each_segment(struct proc *p, segment_callback func, void *closure, int writable) 1148 { 1149 int error = 0; 1150 vm_map_t map = &p->p_vmspace->vm_map; 1151 vm_map_entry_t entry; 1152 1153 for (entry = map->header.next; error == 0 && entry != &map->header; 1154 entry = entry->next) { 1155 vm_object_t obj; 1156 1157 /* 1158 * Don't dump inaccessible mappings, deal with legacy 1159 * coredump mode. 1160 * 1161 * Note that read-only segments related to the elf binary 1162 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1163 * need to arbitrarily ignore such segments. 1164 */ 1165 if (elf_legacy_coredump) { 1166 if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW) 1167 continue; 1168 } else { 1169 if (writable && (entry->protection & VM_PROT_ALL) == 0) 1170 continue; 1171 } 1172 1173 /* 1174 * Dont include memory segment in the coredump if 1175 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1176 * madvise(2). Do not dump submaps (i.e. parts of the 1177 * kernel map). 1178 */ 1179 if (writable && entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1180 continue; 1181 1182 if ((obj = entry->object.vm_object) == NULL) 1183 continue; 1184 1185 /* Find the deepest backing object. */ 1186 while (obj->backing_object != NULL) 1187 obj = obj->backing_object; 1188 1189 /* Ignore memory-mapped devices and such things. */ 1190 if (obj->type != OBJT_DEFAULT && 1191 obj->type != OBJT_SWAP && 1192 obj->type != OBJT_VNODE) 1193 continue; 1194 1195 error = (*func)(entry, closure); 1196 } 1197 return error; 1198 } 1199 1200 static 1201 void * 1202 target_reserve(elf_buf_t target, size_t bytes, int *error) 1203 { 1204 void *res = NULL; 1205 1206 if (target->buf) { 1207 if (target->off + bytes > target->off_max) 1208 *error = EINVAL; 1209 else 1210 res = target->buf + target->off; 1211 } 1212 target->off += bytes; 1213 return (res); 1214 } 1215 1216 /* 1217 * Write the core file header to the file, including padding up to 1218 * the page boundary. 1219 */ 1220 static int 1221 elf_corehdr(struct proc *p, struct file *fp, struct ucred *cred, int numsegs, 1222 elf_buf_t target) 1223 { 1224 struct { 1225 prstatus_t status; 1226 prfpregset_t fpregset; 1227 prpsinfo_t psinfo; 1228 } *tempdata; 1229 int error; 1230 prstatus_t *status; 1231 prfpregset_t *fpregset; 1232 prpsinfo_t *psinfo; 1233 int nbytes; 1234 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK); 1235 status = &tempdata->status; 1236 fpregset = &tempdata->fpregset; 1237 psinfo = &tempdata->psinfo; 1238 1239 /* Gather the information for the header. */ 1240 status->pr_version = PRSTATUS_VERSION; 1241 status->pr_statussz = sizeof(prstatus_t); 1242 status->pr_gregsetsz = sizeof(gregset_t); 1243 status->pr_fpregsetsz = sizeof(fpregset_t); 1244 status->pr_osreldate = osreldate; 1245 status->pr_cursig = p->p_sig; 1246 status->pr_pid = p->p_pid; 1247 fill_regs(&p->p_lwp, &status->pr_reg); 1248 1249 fill_fpregs(&p->p_lwp, fpregset); 1250 1251 psinfo->pr_version = PRPSINFO_VERSION; 1252 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 1253 strncpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname) - 1); 1254 1255 /* XXX - We don't fill in the command line arguments properly yet. */ 1256 strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ); 1257 1258 /* Fill in the header. */ 1259 error = elf_puthdr(p, target, status, fpregset, psinfo, numsegs); 1260 1261 free(tempdata, M_TEMP); 1262 1263 /* Write it to the core file. */ 1264 if (error == 0) 1265 error = fp_write(fp, target->buf, target->off, &nbytes); 1266 return error; 1267 } 1268 1269 static int 1270 elf_puthdr(struct proc *p, elf_buf_t target, const prstatus_t *status, 1271 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs) 1272 { 1273 int error = 0; 1274 size_t phoff; 1275 size_t noteoff; 1276 size_t notesz; 1277 Elf_Ehdr *ehdr; 1278 Elf_Phdr *phdr; 1279 1280 ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error); 1281 1282 phoff = target->off; 1283 phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error); 1284 1285 noteoff = target->off; 1286 if (error == 0) { 1287 error = elf_putnote(target, "FreeBSD", NT_PRSTATUS, 1288 status, sizeof *status); 1289 } 1290 if (error == 0) { 1291 error = elf_putnote(target, "FreeBSD", NT_FPREGSET, 1292 fpregset, sizeof *fpregset); 1293 } 1294 if (error == 0) { 1295 error = elf_putnote(target, "FreeBSD", NT_PRPSINFO, 1296 psinfo, sizeof *psinfo); 1297 } 1298 notesz = target->off - noteoff; 1299 1300 /* 1301 * put extra cruft for dumping process state here 1302 * - we really want it be before all the program 1303 * mappings 1304 * - we just need to update the offset accordingly 1305 * and GDB will be none the wiser. 1306 */ 1307 if (error == 0) 1308 error = elf_puttextvp(p, target); 1309 if (error == 0) 1310 error = elf_putsigs(p, target); 1311 if (error == 0) 1312 error = elf_putfiles(p, target); 1313 1314 /* 1315 * Align up to a page boundary for the program segments. The 1316 * actual data will be written to the outptu file, not to elf_buf_t, 1317 * so we do not have to do any further bounds checking. 1318 */ 1319 target->off = round_page(target->off); 1320 if (error == 0 && ehdr != NULL) { 1321 /* 1322 * Fill in the ELF header. 1323 */ 1324 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1325 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1326 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1327 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1328 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1329 ehdr->e_ident[EI_DATA] = ELF_DATA; 1330 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1331 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1332 ehdr->e_ident[EI_ABIVERSION] = 0; 1333 ehdr->e_ident[EI_PAD] = 0; 1334 ehdr->e_type = ET_CORE; 1335 ehdr->e_machine = ELF_ARCH; 1336 ehdr->e_version = EV_CURRENT; 1337 ehdr->e_entry = 0; 1338 ehdr->e_phoff = phoff; 1339 ehdr->e_flags = 0; 1340 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1341 ehdr->e_phentsize = sizeof(Elf_Phdr); 1342 ehdr->e_phnum = numsegs + 1; 1343 ehdr->e_shentsize = sizeof(Elf_Shdr); 1344 ehdr->e_shnum = 0; 1345 ehdr->e_shstrndx = SHN_UNDEF; 1346 } 1347 if (error == 0 && phdr != NULL) { 1348 /* 1349 * Fill in the program header entries. 1350 */ 1351 struct phdr_closure phc; 1352 1353 /* The note segement. */ 1354 phdr->p_type = PT_NOTE; 1355 phdr->p_offset = noteoff; 1356 phdr->p_vaddr = 0; 1357 phdr->p_paddr = 0; 1358 phdr->p_filesz = notesz; 1359 phdr->p_memsz = 0; 1360 phdr->p_flags = 0; 1361 phdr->p_align = 0; 1362 ++phdr; 1363 1364 /* All the writable segments from the program. */ 1365 phc.phdr = phdr; 1366 phc.phdr_max = phdr + numsegs; 1367 phc.offset = target->off; 1368 each_segment(p, cb_put_phdr, &phc, 1); 1369 } 1370 return (error); 1371 } 1372 1373 static int 1374 elf_putnote(elf_buf_t target, const char *name, int type, 1375 const void *desc, size_t descsz) 1376 { 1377 int error = 0; 1378 char *dst; 1379 Elf_Note note; 1380 1381 note.n_namesz = strlen(name) + 1; 1382 note.n_descsz = descsz; 1383 note.n_type = type; 1384 dst = target_reserve(target, sizeof(note), &error); 1385 if (dst != NULL) 1386 bcopy(¬e, dst, sizeof note); 1387 dst = target_reserve(target, note.n_namesz, &error); 1388 if (dst != NULL) 1389 bcopy(name, dst, note.n_namesz); 1390 target->off = roundup2(target->off, sizeof(Elf_Size)); 1391 dst = target_reserve(target, note.n_descsz, &error); 1392 if (dst != NULL) 1393 bcopy(desc, dst, note.n_descsz); 1394 target->off = roundup2(target->off, sizeof(Elf_Size)); 1395 return(error); 1396 } 1397 1398 1399 static int 1400 elf_putsigs(struct proc *p, elf_buf_t target) 1401 { 1402 int error = 0; 1403 struct ckpt_siginfo *csi; 1404 1405 csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error); 1406 if (csi) { 1407 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo); 1408 bcopy(p->p_procsig, &csi->csi_procsig, sizeof(struct procsig)); 1409 bcopy(p->p_procsig->ps_sigacts, &csi->csi_sigacts, sizeof(struct sigacts)); 1410 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval)); 1411 bcopy(&p->p_sigmask, &csi->csi_sigmask, sizeof(sigset_t)); 1412 csi->csi_sigparent = p->p_sigparent; 1413 } 1414 return(error); 1415 } 1416 1417 static int 1418 elf_putfiles(struct proc *p, elf_buf_t target) 1419 { 1420 int error = 0; 1421 int i; 1422 struct ckpt_filehdr *cfh = NULL; 1423 struct ckpt_fileinfo *cfi; 1424 struct file *fp; 1425 struct vnode *vp; 1426 /* 1427 * the duplicated loop is gross, but it was the only way 1428 * to eliminate uninitialized variable warnings 1429 */ 1430 cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error); 1431 if (cfh) { 1432 cfh->cfh_nfiles = 0; 1433 } 1434 1435 /* 1436 * ignore STDIN/STDERR/STDOUT. 1437 */ 1438 for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) { 1439 fp = holdfp(p->p_fd, i, -1); 1440 if (fp == NULL) 1441 continue; 1442 /* 1443 * XXX Only checkpoint vnodes for now. 1444 */ 1445 if (fp->f_type != DTYPE_VNODE) { 1446 fdrop(fp); 1447 continue; 1448 } 1449 cfi = target_reserve(target, sizeof(struct ckpt_fileinfo), 1450 &error); 1451 if (cfi == NULL) { 1452 fdrop(fp); 1453 continue; 1454 } 1455 cfi->cfi_index = -1; 1456 cfi->cfi_type = fp->f_type; 1457 cfi->cfi_flags = fp->f_flag; 1458 cfi->cfi_offset = fp->f_offset; 1459 /* f_count and f_msgcount should not be saved/restored */ 1460 /* XXX save cred info */ 1461 1462 switch(fp->f_type) { 1463 case DTYPE_VNODE: 1464 vp = (struct vnode *)fp->f_data; 1465 /* 1466 * it looks like a bug in ptrace is marking 1467 * a non-vnode as a vnode - until we find the 1468 * root cause this will at least prevent 1469 * further panics from truss 1470 */ 1471 if (vp == NULL || vp->v_mount == NULL) 1472 break; 1473 cfh->cfh_nfiles++; 1474 cfi->cfi_index = i; 1475 cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1476 error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid); 1477 break; 1478 default: 1479 break; 1480 } 1481 fdrop(fp); 1482 } 1483 return(error); 1484 } 1485 1486 static int 1487 elf_puttextvp(struct proc *p, elf_buf_t target) 1488 { 1489 int error = 0; 1490 int *vn_count; 1491 struct fp_closure fpc; 1492 struct ckpt_vminfo *vminfo; 1493 1494 vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error); 1495 if (vminfo != NULL) { 1496 vminfo->cvm_dsize = p->p_vmspace->vm_dsize; 1497 vminfo->cvm_tsize = p->p_vmspace->vm_tsize; 1498 vminfo->cvm_daddr = p->p_vmspace->vm_daddr; 1499 vminfo->cvm_taddr = p->p_vmspace->vm_taddr; 1500 } 1501 1502 fpc.count = 0; 1503 vn_count = target_reserve(target, sizeof(int), &error); 1504 if (target->buf != NULL) { 1505 fpc.vnh = (struct vn_hdr *)(target->buf + target->off); 1506 fpc.vnh_max = fpc.vnh + 1507 (target->off_max - target->off) / sizeof(struct vn_hdr); 1508 error = each_segment(p, cb_put_fp, &fpc, 0); 1509 if (vn_count) 1510 *vn_count = fpc.count; 1511 } else { 1512 error = each_segment(p, cb_fpcount_segment, &fpc.count, 0); 1513 } 1514 target->off += fpc.count * sizeof(struct vn_hdr); 1515 return(error); 1516 } 1517 1518 1519 /* 1520 * Tell kern_execve.c about it, with a little help from the linker. 1521 */ 1522 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"}; 1523 EXEC_SET_ORDERED(elf, elf_execsw, SI_ORDER_FIRST); 1524