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 1.50 2007/02/25 23:17:12 corecode 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 VM_MAPTYPE_NORMAL, 311 prot, 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_MAPTYPE_NORMAL, 345 VM_PROT_ALL, VM_PROT_ALL, 346 0); 347 vm_map_unlock(&vmspace->vm_map); 348 vm_map_entry_release(count); 349 if (rv != KERN_SUCCESS) { 350 return EINVAL; 351 } 352 } 353 354 if (copy_len != 0) { 355 vm_object_reference(object); 356 rv = vm_map_find(&exec_map, 357 object, 358 trunc_page(offset + filsz), 359 &data_buf, 360 PAGE_SIZE, 361 TRUE, 362 VM_MAPTYPE_NORMAL, 363 VM_PROT_READ, VM_PROT_ALL, 364 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 365 if (rv != KERN_SUCCESS) { 366 vm_object_deallocate(object); 367 return EINVAL; 368 } 369 370 /* send the page fragment to user space */ 371 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len); 372 vm_map_remove(&exec_map, data_buf, data_buf + PAGE_SIZE); 373 if (error) { 374 return (error); 375 } 376 } 377 378 /* 379 * set it to the specified protection 380 */ 381 vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len, prot, 382 FALSE); 383 384 return error; 385 } 386 387 /* 388 * Load the file "file" into memory. It may be either a shared object 389 * or an executable. 390 * 391 * The "addr" reference parameter is in/out. On entry, it specifies 392 * the address where a shared object should be loaded. If the file is 393 * an executable, this value is ignored. On exit, "addr" specifies 394 * where the file was actually loaded. 395 * 396 * The "entry" reference parameter is out only. On exit, it specifies 397 * the entry point for the loaded file. 398 */ 399 static int 400 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry) 401 { 402 struct { 403 struct nlookupdata nd; 404 struct vattr attr; 405 struct image_params image_params; 406 } *tempdata; 407 const Elf_Ehdr *hdr = NULL; 408 const Elf_Phdr *phdr = NULL; 409 struct nlookupdata *nd; 410 struct vmspace *vmspace = p->p_vmspace; 411 struct vattr *attr; 412 struct image_params *imgp; 413 vm_prot_t prot; 414 u_long rbase; 415 u_long base_addr = 0; 416 int error, i, numsegs; 417 418 tempdata = kmalloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 419 nd = &tempdata->nd; 420 attr = &tempdata->attr; 421 imgp = &tempdata->image_params; 422 423 /* 424 * Initialize part of the common data 425 */ 426 imgp->proc = p; 427 imgp->attr = attr; 428 imgp->firstpage = NULL; 429 imgp->image_header = NULL; 430 imgp->vp = NULL; 431 432 error = nlookup_init(nd, file, UIO_SYSSPACE, NLC_FOLLOW); 433 if (error == 0) 434 error = nlookup(nd); 435 if (error == 0) 436 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 437 nlookup_done(nd); 438 if (error) 439 goto fail; 440 441 /* 442 * Check permissions, modes, uid, etc on the file, and "open" it. 443 */ 444 error = exec_check_permissions(imgp); 445 if (error) { 446 vn_unlock(imgp->vp); 447 goto fail; 448 } 449 450 error = exec_map_first_page(imgp); 451 /* 452 * Also make certain that the interpreter stays the same, so set 453 * its VTEXT flag, too. 454 */ 455 if (error == 0) 456 imgp->vp->v_flag |= VTEXT; 457 vn_unlock(imgp->vp); 458 if (error) 459 goto fail; 460 461 hdr = (const Elf_Ehdr *)imgp->image_header; 462 if ((error = elf_check_header(hdr)) != 0) 463 goto fail; 464 if (hdr->e_type == ET_DYN) 465 rbase = *addr; 466 else if (hdr->e_type == ET_EXEC) 467 rbase = 0; 468 else { 469 error = ENOEXEC; 470 goto fail; 471 } 472 473 /* Only support headers that fit within first page for now 474 * (multiplication of two Elf_Half fields will not overflow) */ 475 if ((hdr->e_phoff > PAGE_SIZE) || 476 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 477 error = ENOEXEC; 478 goto fail; 479 } 480 481 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 482 483 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 484 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 485 prot = 0; 486 if (phdr[i].p_flags & PF_X) 487 prot |= VM_PROT_EXECUTE; 488 if (phdr[i].p_flags & PF_W) 489 prot |= VM_PROT_WRITE; 490 if (phdr[i].p_flags & PF_R) 491 prot |= VM_PROT_READ; 492 493 error = elf_load_section( 494 p, vmspace, imgp->vp, 495 phdr[i].p_offset, 496 (caddr_t)phdr[i].p_vaddr + 497 rbase, 498 phdr[i].p_memsz, 499 phdr[i].p_filesz, prot); 500 if (error != 0) 501 goto fail; 502 /* 503 * Establish the base address if this is the 504 * first segment. 505 */ 506 if (numsegs == 0) 507 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 508 numsegs++; 509 } 510 } 511 *addr = base_addr; 512 *entry=(unsigned long)hdr->e_entry + rbase; 513 514 fail: 515 if (imgp->firstpage) 516 exec_unmap_first_page(imgp); 517 if (imgp->vp) { 518 vrele(imgp->vp); 519 imgp->vp = NULL; 520 } 521 kfree(tempdata, M_TEMP); 522 523 return error; 524 } 525 526 /* 527 * non static, as it can be overridden by start_init() 528 */ 529 int fallback_elf_brand = -1; 530 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW, 531 &fallback_elf_brand, -1, 532 "ELF brand of last resort"); 533 534 static int 535 exec_elf_imgact(struct image_params *imgp) 536 { 537 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header; 538 const Elf_Phdr *phdr; 539 Elf_Auxargs *elf_auxargs = NULL; 540 struct vmspace *vmspace; 541 vm_prot_t prot; 542 u_long text_size = 0, data_size = 0, total_size = 0; 543 u_long text_addr = 0, data_addr = 0; 544 u_long seg_size, seg_addr; 545 u_long addr, entry = 0, proghdr = 0; 546 int error, i; 547 const char *interp = NULL; 548 const Elf_Note *abi_note = NULL; 549 Elf_Brandinfo *brand_info; 550 char *path; 551 552 error = 0; 553 554 /* 555 * Do we have a valid ELF header ? 556 */ 557 if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC) 558 return -1; 559 560 /* 561 * From here on down, we return an errno, not -1, as we've 562 * detected an ELF file. 563 */ 564 565 if ((hdr->e_phoff > PAGE_SIZE) || 566 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 567 /* Only support headers in first page for now */ 568 return ENOEXEC; 569 } 570 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff); 571 572 /* 573 * From this point on, we may have resources that need to be freed. 574 */ 575 576 exec_new_vmspace(imgp, NULL); 577 578 /* 579 * Yeah, I'm paranoid. There is every reason in the world to get 580 * VTEXT now since from here on out, there are places we can have 581 * a context switch. Better safe than sorry; I really don't want 582 * the file to change while it's being loaded. 583 */ 584 vsetflags(imgp->vp, VTEXT); 585 586 vmspace = imgp->proc->p_vmspace; 587 588 for (i = 0; i < hdr->e_phnum; i++) { 589 switch(phdr[i].p_type) { 590 591 case PT_LOAD: /* Loadable segment */ 592 prot = 0; 593 if (phdr[i].p_flags & PF_X) 594 prot |= VM_PROT_EXECUTE; 595 if (phdr[i].p_flags & PF_W) 596 prot |= VM_PROT_WRITE; 597 if (phdr[i].p_flags & PF_R) 598 prot |= VM_PROT_READ; 599 600 if ((error = elf_load_section(imgp->proc, 601 vmspace, imgp->vp, 602 phdr[i].p_offset, 603 (caddr_t)phdr[i].p_vaddr, 604 phdr[i].p_memsz, 605 phdr[i].p_filesz, prot)) != 0) 606 goto fail; 607 608 /* 609 * If this segment contains the program headers, 610 * remember their virtual address for the AT_PHDR 611 * aux entry. Static binaries don't usually include 612 * a PT_PHDR entry. 613 */ 614 if (phdr[i].p_offset == 0 && 615 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 616 <= phdr[i].p_filesz) 617 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 618 619 seg_addr = trunc_page(phdr[i].p_vaddr); 620 seg_size = round_page(phdr[i].p_memsz + 621 phdr[i].p_vaddr - seg_addr); 622 623 /* 624 * Is this .text or .data? We can't use 625 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 626 * alpha terribly and possibly does other bad 627 * things so we stick to the old way of figuring 628 * it out: If the segment contains the program 629 * entry point, it's a text segment, otherwise it 630 * is a data segment. 631 * 632 * Note that obreak() assumes that data_addr + 633 * data_size == end of data load area, and the ELF 634 * file format expects segments to be sorted by 635 * address. If multiple data segments exist, the 636 * last one will be used. 637 */ 638 if (hdr->e_entry >= phdr[i].p_vaddr && 639 hdr->e_entry < (phdr[i].p_vaddr + 640 phdr[i].p_memsz)) { 641 text_size = seg_size; 642 text_addr = seg_addr; 643 entry = (u_long)hdr->e_entry; 644 } else { 645 data_size = seg_size; 646 data_addr = seg_addr; 647 } 648 total_size += seg_size; 649 650 /* 651 * Check limits. It should be safe to check the 652 * limits after loading the segment since we do 653 * not actually fault in all the segment's pages. 654 */ 655 if (data_size > 656 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur || 657 text_size > maxtsiz || 658 total_size > 659 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) { 660 error = ENOMEM; 661 goto fail; 662 } 663 break; 664 case PT_INTERP: /* Path to interpreter */ 665 if (phdr[i].p_filesz > MAXPATHLEN || 666 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) { 667 error = ENOEXEC; 668 goto fail; 669 } 670 interp = imgp->image_header + phdr[i].p_offset; 671 break; 672 case PT_NOTE: /* Check for .note.ABI-tag */ 673 { 674 const Elf_Note *tmp_note; 675 /* XXX handle anything outside the first page */ 676 if (phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) 677 continue; 678 if (phdr[i].p_filesz < sizeof(Elf_Note)) 679 continue; /* ENOEXEC? */ 680 tmp_note = (const Elf_Note *)(imgp->image_header + phdr[i].p_offset); 681 if (tmp_note->n_type != 1) 682 continue; 683 if (tmp_note->n_namesz + sizeof(Elf_Note) + 684 tmp_note->n_descsz > phdr[i].p_filesz) 685 continue; /* ENOEXEC? */ 686 abi_note = tmp_note; 687 } 688 break; 689 case PT_PHDR: /* Program header table info */ 690 proghdr = phdr[i].p_vaddr; 691 break; 692 default: 693 break; 694 } 695 } 696 697 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 698 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 699 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 700 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 701 702 addr = ELF_RTLD_ADDR(vmspace); 703 704 imgp->entry_addr = entry; 705 706 brand_info = NULL; 707 708 /* We support three types of branding -- (1) the ELF EI_OSABI field 709 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 710 * branding w/in the ELF header, and (3) path of the `interp_path' 711 * field. We should also look for an ".note.ABI-tag" ELF section now 712 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones. 713 */ 714 715 /* If the executable has a brand, search for it in the brand list. */ 716 if (brand_info == NULL && hdr->e_ident[EI_OSABI] != ELFOSABI_NONE) { 717 for (i = 0; i < MAX_BRANDS; i++) { 718 Elf_Brandinfo *bi = elf_brand_list[i]; 719 720 if (bi != NULL && 721 (hdr->e_ident[EI_OSABI] == bi->brand 722 || 0 == 723 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 724 bi->compat_3_brand, strlen(bi->compat_3_brand)))) { 725 brand_info = bi; 726 break; 727 } 728 } 729 } 730 731 /* Search for a recognized ABI. */ 732 if (brand_info == NULL && abi_note != NULL) { 733 for (i = 0; i < MAX_BRANDS; i++) { 734 Elf_Brandinfo *bi = elf_brand_list[i]; 735 736 if (bi != NULL && bi->match_abi_note != NULL && 737 (*bi->match_abi_note)(abi_note)) { 738 brand_info = bi; 739 break; 740 } 741 } 742 } 743 744 /* 745 * ELFOSABI_NONE == ELFOSABI_SYSV, so a SYSV binary misses all 746 * checks so far, since it is neither branded nor does it have 747 * an ABI note. If the EI_OSABI field is ELFOSABI_NONE, assume 748 * it is svr4 and look for an entry in the elf_brand_list with 749 * match_abi_note == NULL. 750 */ 751 if (brand_info == NULL && hdr->e_ident[EI_OSABI] == ELFOSABI_NONE) { 752 for (i = 0; i < MAX_BRANDS; i++) { 753 Elf_Brandinfo *bi = elf_brand_list[i]; 754 755 if (bi != NULL && bi->match_abi_note == NULL && 756 ELFOSABI_SYSV == bi->brand) { 757 brand_info = bi; 758 break; 759 } 760 } 761 } 762 763 /* Lacking a recognized ABI, search for a recognized interpreter. */ 764 if (brand_info == NULL && interp != NULL) { 765 for (i = 0; i < MAX_BRANDS; i++) { 766 Elf_Brandinfo *bi = elf_brand_list[i]; 767 768 if (bi != NULL && 769 strcmp(interp, bi->interp_path) == 0) { 770 brand_info = bi; 771 break; 772 } 773 } 774 } 775 776 /* Lacking a recognized interpreter, try the default brand */ 777 if (brand_info == NULL) { 778 for (i = 0; i < MAX_BRANDS; i++) { 779 Elf_Brandinfo *bi = elf_brand_list[i]; 780 781 if (bi != NULL && fallback_elf_brand == bi->brand) { 782 brand_info = bi; 783 break; 784 } 785 } 786 } 787 788 if (brand_info == NULL) { 789 uprintf("ELF binary type \"%u\" not known.\n", 790 hdr->e_ident[EI_OSABI]); 791 error = ENOEXEC; 792 goto fail; 793 } 794 795 imgp->proc->p_sysent = brand_info->sysvec; 796 if (interp != NULL) { 797 path = kmalloc(MAXPATHLEN, M_TEMP, M_WAITOK); 798 ksnprintf(path, MAXPATHLEN, "%s%s", 799 brand_info->emul_path, interp); 800 if ((error = elf_load_file(imgp->proc, path, &addr, 801 &imgp->entry_addr)) != 0) { 802 if ((error = elf_load_file(imgp->proc, interp, &addr, 803 &imgp->entry_addr)) != 0) { 804 uprintf("ELF interpreter %s not found\n", path); 805 kfree(path, M_TEMP); 806 goto fail; 807 } 808 } 809 kfree(path, M_TEMP); 810 } 811 812 /* 813 * Construct auxargs table (used by the fixup routine) 814 */ 815 elf_auxargs = kmalloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 816 elf_auxargs->execfd = -1; 817 elf_auxargs->phdr = proghdr; 818 elf_auxargs->phent = hdr->e_phentsize; 819 elf_auxargs->phnum = hdr->e_phnum; 820 elf_auxargs->pagesz = PAGE_SIZE; 821 elf_auxargs->base = addr; 822 elf_auxargs->flags = 0; 823 elf_auxargs->entry = entry; 824 elf_auxargs->trace = elf_trace; 825 826 imgp->auxargs = elf_auxargs; 827 imgp->interpreted = 0; 828 829 fail: 830 return error; 831 } 832 833 static int 834 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp) 835 { 836 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 837 register_t *pos; 838 839 pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2); 840 841 if (args->trace) { 842 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 843 } 844 if (args->execfd != -1) { 845 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 846 } 847 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 848 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 849 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 850 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 851 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 852 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 853 AUXARGS_ENTRY(pos, AT_BASE, args->base); 854 AUXARGS_ENTRY(pos, AT_NULL, 0); 855 856 kfree(imgp->auxargs, M_TEMP); 857 imgp->auxargs = NULL; 858 859 (*stack_base)--; 860 suword(*stack_base, (long) imgp->args->argc); 861 return 0; 862 } 863 864 /* 865 * Code for generating ELF core dumps. 866 */ 867 868 typedef int (*segment_callback) (vm_map_entry_t, void *); 869 870 /* Closure for cb_put_phdr(). */ 871 struct phdr_closure { 872 Elf_Phdr *phdr; /* Program header to fill in (incremented) */ 873 Elf_Phdr *phdr_max; /* Pointer bound for error check */ 874 Elf_Off offset; /* Offset of segment in core file */ 875 }; 876 877 /* Closure for cb_size_segment(). */ 878 struct sseg_closure { 879 int count; /* Count of writable segments. */ 880 size_t vsize; /* Total size of all writable segments. */ 881 }; 882 883 /* Closure for cb_put_fp(). */ 884 struct fp_closure { 885 struct vn_hdr *vnh; 886 struct vn_hdr *vnh_max; 887 int count; 888 struct stat *sb; 889 }; 890 891 typedef struct elf_buf { 892 char *buf; 893 size_t off; 894 size_t off_max; 895 } *elf_buf_t; 896 897 static void *target_reserve(elf_buf_t target, size_t bytes, int *error); 898 899 static int cb_put_phdr (vm_map_entry_t, void *); 900 static int cb_size_segment (vm_map_entry_t, void *); 901 static int cb_fpcount_segment(vm_map_entry_t, void *); 902 static int cb_put_fp(vm_map_entry_t, void *); 903 904 905 static int each_segment (struct proc *, segment_callback, void *, int); 906 static int elf_corehdr (struct lwp *, int, struct file *, struct ucred *, 907 int, elf_buf_t); 908 static int elf_puthdr (struct lwp *, elf_buf_t, const prstatus_t *, 909 const prfpregset_t *, const prpsinfo_t *, int); 910 static int elf_putnote (elf_buf_t, const char *, int, const void *, size_t); 911 912 static int elf_putsigs(struct lwp *, elf_buf_t); 913 static int elf_puttextvp(struct proc *, elf_buf_t); 914 static int elf_putfiles(struct proc *, elf_buf_t); 915 916 extern int osreldate; 917 918 int 919 elf_coredump(struct lwp *lp, int sig, struct vnode *vp, off_t limit) 920 { 921 struct file *fp; 922 int error; 923 924 if ((error = falloc(NULL, &fp, NULL)) != 0) 925 return (error); 926 fsetcred(fp, lp->lwp_proc->p_ucred); 927 928 /* 929 * XXX fixme. 930 */ 931 fp->f_type = DTYPE_VNODE; 932 fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW; 933 fp->f_ops = &vnode_fileops; 934 fp->f_data = vp; 935 vn_unlock(vp); 936 937 error = generic_elf_coredump(lp, sig, fp, limit); 938 939 fp->f_type = 0; 940 fp->f_flag = 0; 941 fp->f_ops = &badfileops; 942 fp->f_data = NULL; 943 fdrop(fp); 944 return (error); 945 } 946 947 int 948 generic_elf_coredump(struct lwp *lp, int sig, struct file *fp, off_t limit) 949 { 950 struct proc *p = lp->lwp_proc; 951 struct ucred *cred = p->p_ucred; 952 int error = 0; 953 struct sseg_closure seginfo; 954 struct elf_buf target; 955 956 if (!fp) 957 kprintf("can't dump core - null fp\n"); 958 959 /* 960 * Size the program segments 961 */ 962 seginfo.count = 0; 963 seginfo.vsize = 0; 964 each_segment(p, cb_size_segment, &seginfo, 1); 965 966 /* 967 * Calculate the size of the core file header area by making 968 * a dry run of generating it. Nothing is written, but the 969 * size is calculated. 970 */ 971 bzero(&target, sizeof(target)); 972 elf_puthdr(lp, &target, NULL, NULL, NULL, seginfo.count); 973 974 if (target.off + seginfo.vsize >= limit) 975 return (EFAULT); 976 977 /* 978 * Allocate memory for building the header, fill it up, 979 * and write it out. 980 */ 981 target.off_max = target.off; 982 target.off = 0; 983 target.buf = kmalloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO); 984 985 if (target.buf == NULL) 986 return EINVAL; 987 error = elf_corehdr(lp, sig, fp, cred, seginfo.count, &target); 988 989 /* Write the contents of all of the writable segments. */ 990 if (error == 0) { 991 Elf_Phdr *php; 992 int i; 993 int nbytes; 994 995 php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1; 996 for (i = 0; i < seginfo.count; i++) { 997 error = fp_write(fp, (caddr_t)php->p_vaddr, 998 php->p_filesz, &nbytes, UIO_USERSPACE); 999 if (error != 0) 1000 break; 1001 php++; 1002 } 1003 } 1004 kfree(target.buf, M_TEMP); 1005 1006 return error; 1007 } 1008 1009 /* 1010 * A callback for each_segment() to write out the segment's 1011 * program header entry. 1012 */ 1013 static int 1014 cb_put_phdr(vm_map_entry_t entry, void *closure) 1015 { 1016 struct phdr_closure *phc = closure; 1017 Elf_Phdr *phdr = phc->phdr; 1018 1019 if (phc->phdr == phc->phdr_max) 1020 return EINVAL; 1021 1022 phc->offset = round_page(phc->offset); 1023 1024 phdr->p_type = PT_LOAD; 1025 phdr->p_offset = phc->offset; 1026 phdr->p_vaddr = entry->start; 1027 phdr->p_paddr = 0; 1028 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1029 phdr->p_align = PAGE_SIZE; 1030 phdr->p_flags = 0; 1031 if (entry->protection & VM_PROT_READ) 1032 phdr->p_flags |= PF_R; 1033 if (entry->protection & VM_PROT_WRITE) 1034 phdr->p_flags |= PF_W; 1035 if (entry->protection & VM_PROT_EXECUTE) 1036 phdr->p_flags |= PF_X; 1037 1038 phc->offset += phdr->p_filesz; 1039 ++phc->phdr; 1040 return 0; 1041 } 1042 1043 /* 1044 * A callback for each_writable_segment() to gather information about 1045 * the number of segments and their total size. 1046 */ 1047 static int 1048 cb_size_segment(vm_map_entry_t entry, void *closure) 1049 { 1050 struct sseg_closure *ssc = closure; 1051 1052 ++ssc->count; 1053 ssc->vsize += entry->end - entry->start; 1054 return 0; 1055 } 1056 1057 /* 1058 * A callback for each_segment() to gather information about 1059 * the number of text segments. 1060 */ 1061 static int 1062 cb_fpcount_segment(vm_map_entry_t entry, void *closure) 1063 { 1064 int *count = closure; 1065 struct vnode *vp; 1066 1067 if (entry->object.vm_object->type == OBJT_VNODE) { 1068 vp = (struct vnode *)entry->object.vm_object->handle; 1069 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp) 1070 return 0; 1071 ++*count; 1072 } 1073 return 0; 1074 } 1075 1076 static int 1077 cb_put_fp(vm_map_entry_t entry, void *closure) 1078 { 1079 struct fp_closure *fpc = closure; 1080 struct vn_hdr *vnh = fpc->vnh; 1081 Elf_Phdr *phdr = &vnh->vnh_phdr; 1082 struct vnode *vp; 1083 int error; 1084 1085 /* 1086 * If an entry represents a vnode then write out a file handle. 1087 * 1088 * If we are checkpointing a checkpoint-restored program we do 1089 * NOT record the filehandle for the old checkpoint vnode (which 1090 * is mapped all over the place). Instead we rely on the fact 1091 * that a checkpoint-restored program does not mmap() the checkpt 1092 * vnode NOCORE, so its contents will be written out to the 1093 * new checkpoint file. This is necessary because the 'old' 1094 * checkpoint file is typically destroyed when a new one is created 1095 * and thus cannot be used to restore the new checkpoint. 1096 * 1097 * Theoretically we could create a chain of checkpoint files and 1098 * operate the checkpointing operation kinda like an incremental 1099 * checkpoint, but a checkpoint restore would then likely wind up 1100 * referencing many prior checkpoint files and that is a bit over 1101 * the top for the purpose of the checkpoint API. 1102 */ 1103 if (entry->object.vm_object->type == OBJT_VNODE) { 1104 vp = (struct vnode *)entry->object.vm_object->handle; 1105 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp) 1106 return 0; 1107 if (vnh == fpc->vnh_max) 1108 return EINVAL; 1109 1110 if (vp->v_mount) 1111 vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1112 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid); 1113 if (error) { 1114 char *freepath, *fullpath; 1115 1116 if (vn_fullpath(curproc, vp, &fullpath, &freepath)) { 1117 kprintf("Warning: coredump, error %d: cannot store file handle for vnode %p\n", error, vp); 1118 } else { 1119 kprintf("Warning: coredump, error %d: cannot store file handle for %s\n", error, fullpath); 1120 kfree(freepath, M_TEMP); 1121 } 1122 error = 0; 1123 } 1124 1125 phdr->p_type = PT_LOAD; 1126 phdr->p_offset = 0; /* not written to core */ 1127 phdr->p_vaddr = entry->start; 1128 phdr->p_paddr = 0; 1129 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1130 phdr->p_align = PAGE_SIZE; 1131 phdr->p_flags = 0; 1132 if (entry->protection & VM_PROT_READ) 1133 phdr->p_flags |= PF_R; 1134 if (entry->protection & VM_PROT_WRITE) 1135 phdr->p_flags |= PF_W; 1136 if (entry->protection & VM_PROT_EXECUTE) 1137 phdr->p_flags |= PF_X; 1138 ++fpc->vnh; 1139 ++fpc->count; 1140 } 1141 return 0; 1142 } 1143 1144 /* 1145 * For each writable segment in the process's memory map, call the given 1146 * function with a pointer to the map entry and some arbitrary 1147 * caller-supplied data. 1148 */ 1149 static int 1150 each_segment(struct proc *p, segment_callback func, void *closure, int writable) 1151 { 1152 int error = 0; 1153 vm_map_t map = &p->p_vmspace->vm_map; 1154 vm_map_entry_t entry; 1155 1156 for (entry = map->header.next; error == 0 && entry != &map->header; 1157 entry = entry->next) { 1158 vm_object_t obj; 1159 1160 /* 1161 * Don't dump inaccessible mappings, deal with legacy 1162 * coredump mode. 1163 * 1164 * Note that read-only segments related to the elf binary 1165 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1166 * need to arbitrarily ignore such segments. 1167 */ 1168 if (elf_legacy_coredump) { 1169 if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW) 1170 continue; 1171 } else { 1172 if (writable && (entry->protection & VM_PROT_ALL) == 0) 1173 continue; 1174 } 1175 1176 /* 1177 * Dont include memory segment in the coredump if 1178 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1179 * madvise(2). 1180 * 1181 * Currently we only dump normal VM object maps. We do 1182 * not dump submaps or virtual page tables. 1183 */ 1184 if (writable && (entry->eflags & MAP_ENTRY_NOCOREDUMP)) 1185 continue; 1186 if (entry->maptype != VM_MAPTYPE_NORMAL) 1187 continue; 1188 if ((obj = entry->object.vm_object) == NULL) 1189 continue; 1190 1191 /* Find the deepest backing object. */ 1192 while (obj->backing_object != NULL) 1193 obj = obj->backing_object; 1194 1195 /* Ignore memory-mapped devices and such things. */ 1196 if (obj->type != OBJT_DEFAULT && 1197 obj->type != OBJT_SWAP && 1198 obj->type != OBJT_VNODE) 1199 continue; 1200 1201 error = (*func)(entry, closure); 1202 } 1203 return error; 1204 } 1205 1206 static 1207 void * 1208 target_reserve(elf_buf_t target, size_t bytes, int *error) 1209 { 1210 void *res = NULL; 1211 1212 if (target->buf) { 1213 if (target->off + bytes > target->off_max) 1214 *error = EINVAL; 1215 else 1216 res = target->buf + target->off; 1217 } 1218 target->off += bytes; 1219 return (res); 1220 } 1221 1222 /* 1223 * Write the core file header to the file, including padding up to 1224 * the page boundary. 1225 */ 1226 static int 1227 elf_corehdr(struct lwp *lp, int sig, struct file *fp, struct ucred *cred, int numsegs, 1228 elf_buf_t target) 1229 { 1230 /* XXX lwp handle more than one lwp */ 1231 struct proc *p = lp->lwp_proc; 1232 struct { 1233 prstatus_t status; 1234 prfpregset_t fpregset; 1235 prpsinfo_t psinfo; 1236 } *tempdata; 1237 int error; 1238 prstatus_t *status; 1239 prfpregset_t *fpregset; 1240 prpsinfo_t *psinfo; 1241 int nbytes; 1242 1243 tempdata = kmalloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK); 1244 status = &tempdata->status; 1245 fpregset = &tempdata->fpregset; 1246 psinfo = &tempdata->psinfo; 1247 1248 /* Gather the information for the header. */ 1249 status->pr_version = PRSTATUS_VERSION; 1250 status->pr_statussz = sizeof(prstatus_t); 1251 status->pr_gregsetsz = sizeof(gregset_t); 1252 status->pr_fpregsetsz = sizeof(fpregset_t); 1253 status->pr_osreldate = osreldate; 1254 status->pr_cursig = sig; 1255 status->pr_pid = p->p_pid; 1256 fill_regs(lp, &status->pr_reg); 1257 1258 fill_fpregs(lp, fpregset); 1259 1260 psinfo->pr_version = PRPSINFO_VERSION; 1261 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 1262 strncpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname) - 1); 1263 1264 /* XXX - We don't fill in the command line arguments properly yet. */ 1265 strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ); 1266 1267 /* Fill in the header. */ 1268 error = elf_puthdr(lp, target, status, fpregset, psinfo, numsegs); 1269 1270 kfree(tempdata, M_TEMP); 1271 1272 /* Write it to the core file. */ 1273 if (error == 0) { 1274 error = fp_write(fp, target->buf, target->off, &nbytes, 1275 UIO_SYSSPACE); 1276 } 1277 return error; 1278 } 1279 1280 static int 1281 elf_puthdr(struct lwp *lp, elf_buf_t target, const prstatus_t *status, 1282 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs) 1283 { 1284 struct proc *p = lp->lwp_proc; 1285 int error = 0; 1286 size_t phoff; 1287 size_t noteoff; 1288 size_t notesz; 1289 Elf_Ehdr *ehdr; 1290 Elf_Phdr *phdr; 1291 1292 ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error); 1293 1294 phoff = target->off; 1295 phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error); 1296 1297 noteoff = target->off; 1298 if (error == 0) { 1299 error = elf_putnote(target, "FreeBSD", NT_PRSTATUS, 1300 status, sizeof *status); 1301 } 1302 if (error == 0) { 1303 error = elf_putnote(target, "FreeBSD", NT_FPREGSET, 1304 fpregset, sizeof *fpregset); 1305 } 1306 if (error == 0) { 1307 error = elf_putnote(target, "FreeBSD", NT_PRPSINFO, 1308 psinfo, sizeof *psinfo); 1309 } 1310 notesz = target->off - noteoff; 1311 1312 /* 1313 * put extra cruft for dumping process state here 1314 * - we really want it be before all the program 1315 * mappings 1316 * - we just need to update the offset accordingly 1317 * and GDB will be none the wiser. 1318 */ 1319 if (error == 0) 1320 error = elf_puttextvp(p, target); 1321 if (error == 0) 1322 error = elf_putsigs(lp, target); 1323 if (error == 0) 1324 error = elf_putfiles(p, target); 1325 1326 /* 1327 * Align up to a page boundary for the program segments. The 1328 * actual data will be written to the outptu file, not to elf_buf_t, 1329 * so we do not have to do any further bounds checking. 1330 */ 1331 target->off = round_page(target->off); 1332 if (error == 0 && ehdr != NULL) { 1333 /* 1334 * Fill in the ELF header. 1335 */ 1336 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1337 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1338 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1339 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1340 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1341 ehdr->e_ident[EI_DATA] = ELF_DATA; 1342 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1343 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1344 ehdr->e_ident[EI_ABIVERSION] = 0; 1345 ehdr->e_ident[EI_PAD] = 0; 1346 ehdr->e_type = ET_CORE; 1347 ehdr->e_machine = ELF_ARCH; 1348 ehdr->e_version = EV_CURRENT; 1349 ehdr->e_entry = 0; 1350 ehdr->e_phoff = phoff; 1351 ehdr->e_flags = 0; 1352 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1353 ehdr->e_phentsize = sizeof(Elf_Phdr); 1354 ehdr->e_phnum = numsegs + 1; 1355 ehdr->e_shentsize = sizeof(Elf_Shdr); 1356 ehdr->e_shnum = 0; 1357 ehdr->e_shstrndx = SHN_UNDEF; 1358 } 1359 if (error == 0 && phdr != NULL) { 1360 /* 1361 * Fill in the program header entries. 1362 */ 1363 struct phdr_closure phc; 1364 1365 /* The note segement. */ 1366 phdr->p_type = PT_NOTE; 1367 phdr->p_offset = noteoff; 1368 phdr->p_vaddr = 0; 1369 phdr->p_paddr = 0; 1370 phdr->p_filesz = notesz; 1371 phdr->p_memsz = 0; 1372 phdr->p_flags = 0; 1373 phdr->p_align = 0; 1374 ++phdr; 1375 1376 /* All the writable segments from the program. */ 1377 phc.phdr = phdr; 1378 phc.phdr_max = phdr + numsegs; 1379 phc.offset = target->off; 1380 each_segment(p, cb_put_phdr, &phc, 1); 1381 } 1382 return (error); 1383 } 1384 1385 static int 1386 elf_putnote(elf_buf_t target, const char *name, int type, 1387 const void *desc, size_t descsz) 1388 { 1389 int error = 0; 1390 char *dst; 1391 Elf_Note note; 1392 1393 note.n_namesz = strlen(name) + 1; 1394 note.n_descsz = descsz; 1395 note.n_type = type; 1396 dst = target_reserve(target, sizeof(note), &error); 1397 if (dst != NULL) 1398 bcopy(¬e, dst, sizeof note); 1399 dst = target_reserve(target, note.n_namesz, &error); 1400 if (dst != NULL) 1401 bcopy(name, dst, note.n_namesz); 1402 target->off = roundup2(target->off, sizeof(Elf_Size)); 1403 dst = target_reserve(target, note.n_descsz, &error); 1404 if (dst != NULL) 1405 bcopy(desc, dst, note.n_descsz); 1406 target->off = roundup2(target->off, sizeof(Elf_Size)); 1407 return(error); 1408 } 1409 1410 1411 static int 1412 elf_putsigs(struct lwp *lp, elf_buf_t target) 1413 { 1414 /* XXX lwp handle more than one lwp */ 1415 struct proc *p = lp->lwp_proc; 1416 int error = 0; 1417 struct ckpt_siginfo *csi; 1418 1419 csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error); 1420 if (csi) { 1421 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo); 1422 bcopy(p->p_sigacts, &csi->csi_sigacts, sizeof(*p->p_sigacts)); 1423 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval)); 1424 bcopy(&lp->lwp_sigmask, &csi->csi_sigmask, 1425 sizeof(sigset_t)); 1426 csi->csi_sigparent = p->p_sigparent; 1427 } 1428 return(error); 1429 } 1430 1431 static int 1432 elf_putfiles(struct proc *p, elf_buf_t target) 1433 { 1434 int error = 0; 1435 int i; 1436 struct ckpt_filehdr *cfh = NULL; 1437 struct ckpt_fileinfo *cfi; 1438 struct file *fp; 1439 struct vnode *vp; 1440 /* 1441 * the duplicated loop is gross, but it was the only way 1442 * to eliminate uninitialized variable warnings 1443 */ 1444 cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error); 1445 if (cfh) { 1446 cfh->cfh_nfiles = 0; 1447 } 1448 1449 /* 1450 * ignore STDIN/STDERR/STDOUT. 1451 */ 1452 for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) { 1453 fp = holdfp(p->p_fd, i, -1); 1454 if (fp == NULL) 1455 continue; 1456 /* 1457 * XXX Only checkpoint vnodes for now. 1458 */ 1459 if (fp->f_type != DTYPE_VNODE) { 1460 fdrop(fp); 1461 continue; 1462 } 1463 cfi = target_reserve(target, sizeof(struct ckpt_fileinfo), 1464 &error); 1465 if (cfi == NULL) { 1466 fdrop(fp); 1467 continue; 1468 } 1469 cfi->cfi_index = -1; 1470 cfi->cfi_type = fp->f_type; 1471 cfi->cfi_flags = fp->f_flag; 1472 cfi->cfi_offset = fp->f_offset; 1473 /* f_count and f_msgcount should not be saved/restored */ 1474 /* XXX save cred info */ 1475 1476 switch(fp->f_type) { 1477 case DTYPE_VNODE: 1478 vp = (struct vnode *)fp->f_data; 1479 /* 1480 * it looks like a bug in ptrace is marking 1481 * a non-vnode as a vnode - until we find the 1482 * root cause this will at least prevent 1483 * further panics from truss 1484 */ 1485 if (vp == NULL || vp->v_mount == NULL) 1486 break; 1487 cfh->cfh_nfiles++; 1488 cfi->cfi_index = i; 1489 cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1490 error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid); 1491 break; 1492 default: 1493 break; 1494 } 1495 fdrop(fp); 1496 } 1497 return(error); 1498 } 1499 1500 static int 1501 elf_puttextvp(struct proc *p, elf_buf_t target) 1502 { 1503 int error = 0; 1504 int *vn_count; 1505 struct fp_closure fpc; 1506 struct ckpt_vminfo *vminfo; 1507 1508 vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error); 1509 if (vminfo != NULL) { 1510 vminfo->cvm_dsize = p->p_vmspace->vm_dsize; 1511 vminfo->cvm_tsize = p->p_vmspace->vm_tsize; 1512 vminfo->cvm_daddr = p->p_vmspace->vm_daddr; 1513 vminfo->cvm_taddr = p->p_vmspace->vm_taddr; 1514 } 1515 1516 fpc.count = 0; 1517 vn_count = target_reserve(target, sizeof(int), &error); 1518 if (target->buf != NULL) { 1519 fpc.vnh = (struct vn_hdr *)(target->buf + target->off); 1520 fpc.vnh_max = fpc.vnh + 1521 (target->off_max - target->off) / sizeof(struct vn_hdr); 1522 error = each_segment(p, cb_put_fp, &fpc, 0); 1523 if (vn_count) 1524 *vn_count = fpc.count; 1525 } else { 1526 error = each_segment(p, cb_fpcount_segment, &fpc.count, 0); 1527 } 1528 target->off += fpc.count * sizeof(struct vn_hdr); 1529 return(error); 1530 } 1531 1532 1533 /* 1534 * Tell kern_execve.c about it, with a little help from the linker. 1535 */ 1536 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"}; 1537 EXEC_SET_ORDERED(elf, elf_execsw, SI_ORDER_FIRST); 1538