imgact_elf.c revision 83222
1/*- 2 * Copyright (c) 2000 David O'Brien 3 * Copyright (c) 1995-1996 S�ren Schmidt 4 * Copyright (c) 1996 Peter Wemm 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer 12 * in this position and unchanged. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. The name of the author may not be used to endorse or promote products 17 * derived from this software withough specific prior written permission 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 29 * 30 * $FreeBSD: head/sys/kern/imgact_elf.c 83222 2001-09-08 20:02:33Z dillon $ 31 */ 32 33#include "opt_rlimit.h" 34 35#include <sys/param.h> 36#include <sys/exec.h> 37#include <sys/fcntl.h> 38#include <sys/imgact.h> 39#include <sys/imgact_elf.h> 40#include <sys/kernel.h> 41#include <sys/lock.h> 42#include <sys/malloc.h> 43#include <sys/mutex.h> 44#include <sys/mman.h> 45#include <sys/namei.h> 46#include <sys/pioctl.h> 47#include <sys/proc.h> 48#include <sys/procfs.h> 49#include <sys/resourcevar.h> 50#include <sys/systm.h> 51#include <sys/signalvar.h> 52#include <sys/stat.h> 53#include <sys/sx.h> 54#include <sys/syscall.h> 55#include <sys/sysctl.h> 56#include <sys/sysent.h> 57#include <sys/vnode.h> 58 59#include <vm/vm.h> 60#include <vm/vm_kern.h> 61#include <vm/vm_param.h> 62#include <vm/pmap.h> 63#include <vm/vm_map.h> 64#include <vm/vm_object.h> 65#include <vm/vm_extern.h> 66 67#include <machine/elf.h> 68#include <machine/md_var.h> 69 70#define OLD_EI_BRAND 8 71 72__ElfType(Brandinfo); 73__ElfType(Auxargs); 74 75static int elf_check_header __P((const Elf_Ehdr *hdr)); 76static int elf_freebsd_fixup __P((register_t **stack_base, 77 struct image_params *imgp)); 78static int elf_load_file __P((struct proc *p, const char *file, u_long *addr, 79 u_long *entry)); 80static int elf_load_section __P((struct proc *p, 81 struct vmspace *vmspace, struct vnode *vp, 82 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 83 vm_prot_t prot)); 84static int exec_elf_imgact __P((struct image_params *imgp)); 85 86static int elf_trace = 0; 87SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, ""); 88 89struct sysentvec elf_freebsd_sysvec = { 90 SYS_MAXSYSCALL, 91 sysent, 92 0, 93 0, 94 0, 95 0, 96 0, 97 0, 98 elf_freebsd_fixup, 99 sendsig, 100 sigcode, 101 &szsigcode, 102 0, 103 "FreeBSD ELF", 104 elf_coredump, 105 NULL, 106 MINSIGSTKSZ 107}; 108 109static Elf_Brandinfo freebsd_brand_info = { 110 ELFOSABI_FREEBSD, 111 "FreeBSD", 112 "", 113 "/usr/libexec/ld-elf.so.1", 114 &elf_freebsd_sysvec 115 }; 116static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = { 117 &freebsd_brand_info, 118 NULL, NULL, NULL, 119 NULL, NULL, NULL, NULL 120 }; 121 122int 123elf_insert_brand_entry(Elf_Brandinfo *entry) 124{ 125 int i; 126 127 for (i=1; i<MAX_BRANDS; i++) { 128 if (elf_brand_list[i] == NULL) { 129 elf_brand_list[i] = entry; 130 break; 131 } 132 } 133 if (i == MAX_BRANDS) 134 return -1; 135 return 0; 136} 137 138int 139elf_remove_brand_entry(Elf_Brandinfo *entry) 140{ 141 int i; 142 143 for (i=1; i<MAX_BRANDS; i++) { 144 if (elf_brand_list[i] == entry) { 145 elf_brand_list[i] = NULL; 146 break; 147 } 148 } 149 if (i == MAX_BRANDS) 150 return -1; 151 return 0; 152} 153 154int 155elf_brand_inuse(Elf_Brandinfo *entry) 156{ 157 struct proc *p; 158 int rval = FALSE; 159 160 sx_slock(&allproc_lock); 161 LIST_FOREACH(p, &allproc, p_list) { 162 if (p->p_sysent == entry->sysvec) { 163 rval = TRUE; 164 break; 165 } 166 } 167 sx_sunlock(&allproc_lock); 168 169 return (rval); 170} 171 172static int 173elf_check_header(const Elf_Ehdr *hdr) 174{ 175 if (!IS_ELF(*hdr) || 176 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 177 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 178 hdr->e_ident[EI_VERSION] != EV_CURRENT) 179 return ENOEXEC; 180 181 if (!ELF_MACHINE_OK(hdr->e_machine)) 182 return ENOEXEC; 183 184 if (hdr->e_version != ELF_TARG_VER) 185 return ENOEXEC; 186 187 return 0; 188} 189 190static int 191elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp, vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot) 192{ 193 size_t map_len; 194 vm_offset_t map_addr; 195 int error, rv; 196 size_t copy_len; 197 vm_object_t object; 198 vm_offset_t file_addr; 199 vm_offset_t data_buf = 0; 200 201 GIANT_REQUIRED; 202 203 VOP_GETVOBJECT(vp, &object); 204 error = 0; 205 206 /* 207 * It's necessary to fail if the filsz + offset taken from the 208 * header is greater than the actual file pager object's size. 209 * If we were to allow this, then the vm_map_find() below would 210 * walk right off the end of the file object and into the ether. 211 * 212 * While I'm here, might as well check for something else that 213 * is invalid: filsz cannot be greater than memsz. 214 */ 215 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 216 filsz > memsz) { 217 uprintf("elf_load_section: truncated ELF file\n"); 218 return (ENOEXEC); 219 } 220 221 map_addr = trunc_page((vm_offset_t)vmaddr); 222 file_addr = trunc_page(offset); 223 224 /* 225 * We have two choices. We can either clear the data in the last page 226 * of an oversized mapping, or we can start the anon mapping a page 227 * early and copy the initialized data into that first page. We 228 * choose the second.. 229 */ 230 if (memsz > filsz) 231 map_len = trunc_page(offset+filsz) - file_addr; 232 else 233 map_len = round_page(offset+filsz) - file_addr; 234 235 if (map_len != 0) { 236 vm_object_reference(object); 237 vm_map_lock(&vmspace->vm_map); 238 rv = vm_map_insert(&vmspace->vm_map, 239 object, 240 file_addr, /* file offset */ 241 map_addr, /* virtual start */ 242 map_addr + map_len,/* virtual end */ 243 prot, 244 VM_PROT_ALL, 245 MAP_COPY_ON_WRITE | MAP_PREFAULT); 246 vm_map_unlock(&vmspace->vm_map); 247 if (rv != KERN_SUCCESS) { 248 vm_object_deallocate(object); 249 return EINVAL; 250 } 251 252 /* we can stop now if we've covered it all */ 253 if (memsz == filsz) { 254 return 0; 255 } 256 } 257 258 259 /* 260 * We have to get the remaining bit of the file into the first part 261 * of the oversized map segment. This is normally because the .data 262 * segment in the file is extended to provide bss. It's a neat idea 263 * to try and save a page, but it's a pain in the behind to implement. 264 */ 265 copy_len = (offset + filsz) - trunc_page(offset + filsz); 266 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 267 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 268 269 /* This had damn well better be true! */ 270 if (map_len != 0) { 271 vm_map_lock(&vmspace->vm_map); 272 rv = vm_map_insert(&vmspace->vm_map, NULL, 0, 273 map_addr, map_addr + map_len, 274 VM_PROT_ALL, VM_PROT_ALL, 0); 275 vm_map_unlock(&vmspace->vm_map); 276 if (rv != KERN_SUCCESS) { 277 return EINVAL; 278 } 279 } 280 281 if (copy_len != 0) { 282 vm_object_reference(object); 283 rv = vm_map_find(exec_map, 284 object, 285 trunc_page(offset + filsz), 286 &data_buf, 287 PAGE_SIZE, 288 TRUE, 289 VM_PROT_READ, 290 VM_PROT_ALL, 291 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL); 292 if (rv != KERN_SUCCESS) { 293 vm_object_deallocate(object); 294 return EINVAL; 295 } 296 297 /* send the page fragment to user space */ 298 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len); 299 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 300 if (error) { 301 return (error); 302 } 303 } 304 305 /* 306 * set it to the specified protection 307 */ 308 vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len, prot, 309 FALSE); 310 311 return error; 312} 313 314/* 315 * Load the file "file" into memory. It may be either a shared object 316 * or an executable. 317 * 318 * The "addr" reference parameter is in/out. On entry, it specifies 319 * the address where a shared object should be loaded. If the file is 320 * an executable, this value is ignored. On exit, "addr" specifies 321 * where the file was actually loaded. 322 * 323 * The "entry" reference parameter is out only. On exit, it specifies 324 * the entry point for the loaded file. 325 */ 326static int 327elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry) 328{ 329 struct { 330 struct nameidata nd; 331 struct vattr attr; 332 struct image_params image_params; 333 } *tempdata; 334 const Elf_Ehdr *hdr = NULL; 335 const Elf_Phdr *phdr = NULL; 336 struct nameidata *nd; 337 struct vmspace *vmspace = p->p_vmspace; 338 struct vattr *attr; 339 struct image_params *imgp; 340 vm_prot_t prot; 341 u_long rbase; 342 u_long base_addr = 0; 343 int error, i, numsegs; 344 345 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 346 nd = &tempdata->nd; 347 attr = &tempdata->attr; 348 imgp = &tempdata->image_params; 349 350 /* 351 * Initialize part of the common data 352 */ 353 imgp->proc = p; 354 imgp->uap = NULL; 355 imgp->attr = attr; 356 imgp->firstpage = NULL; 357 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE); 358 359 if (imgp->image_header == NULL) { 360 nd->ni_vp = NULL; 361 error = ENOMEM; 362 goto fail; 363 } 364 365 NDINIT(nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, p); 366 367 if ((error = namei(nd)) != 0) { 368 nd->ni_vp = NULL; 369 goto fail; 370 } 371 NDFREE(nd, NDF_ONLY_PNBUF); 372 imgp->vp = nd->ni_vp; 373 374 /* 375 * Check permissions, modes, uid, etc on the file, and "open" it. 376 */ 377 error = exec_check_permissions(imgp); 378 if (error) { 379 VOP_UNLOCK(nd->ni_vp, 0, p); 380 goto fail; 381 } 382 383 error = exec_map_first_page(imgp); 384 /* 385 * Also make certain that the interpreter stays the same, so set 386 * its VTEXT flag, too. 387 */ 388 if (error == 0) 389 nd->ni_vp->v_flag |= VTEXT; 390 VOP_UNLOCK(nd->ni_vp, 0, p); 391 if (error) 392 goto fail; 393 394 hdr = (const Elf_Ehdr *)imgp->image_header; 395 if ((error = elf_check_header(hdr)) != 0) 396 goto fail; 397 if (hdr->e_type == ET_DYN) 398 rbase = *addr; 399 else if (hdr->e_type == ET_EXEC) 400 rbase = 0; 401 else { 402 error = ENOEXEC; 403 goto fail; 404 } 405 406 /* Only support headers that fit within first page for now */ 407 if ((hdr->e_phoff > PAGE_SIZE) || 408 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 409 error = ENOEXEC; 410 goto fail; 411 } 412 413 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 414 415 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 416 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 417 prot = 0; 418 if (phdr[i].p_flags & PF_X) 419 prot |= VM_PROT_EXECUTE; 420 if (phdr[i].p_flags & PF_W) 421 prot |= VM_PROT_WRITE; 422 if (phdr[i].p_flags & PF_R) 423 prot |= VM_PROT_READ; 424 425 if ((error = elf_load_section(p, vmspace, nd->ni_vp, 426 phdr[i].p_offset, 427 (caddr_t)phdr[i].p_vaddr + 428 rbase, 429 phdr[i].p_memsz, 430 phdr[i].p_filesz, prot)) != 0) 431 goto fail; 432 /* 433 * Establish the base address if this is the 434 * first segment. 435 */ 436 if (numsegs == 0) 437 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 438 numsegs++; 439 } 440 } 441 *addr = base_addr; 442 *entry=(unsigned long)hdr->e_entry + rbase; 443 444fail: 445 if (imgp->firstpage) 446 exec_unmap_first_page(imgp); 447 if (imgp->image_header) 448 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header, 449 PAGE_SIZE); 450 if (nd->ni_vp) 451 vrele(nd->ni_vp); 452 453 free(tempdata, M_TEMP); 454 455 return error; 456} 457 458/* 459 * non static, as it can be overridden by start_init() 460 */ 461#ifdef __ia64__ 462int fallback_elf_brand = ELFOSABI_FREEBSD; 463#else 464int fallback_elf_brand = -1; 465#endif 466SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW, 467 &fallback_elf_brand, -1, 468 "ELF brand of last resort"); 469 470static int 471exec_elf_imgact(struct image_params *imgp) 472{ 473 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header; 474 const Elf_Phdr *phdr; 475 Elf_Auxargs *elf_auxargs = NULL; 476 struct vmspace *vmspace; 477 vm_prot_t prot; 478 u_long text_size = 0, data_size = 0; 479 u_long text_addr = 0, data_addr = 0; 480 u_long addr, entry = 0, proghdr = 0; 481 int error, i; 482 const char *interp = NULL; 483 Elf_Brandinfo *brand_info; 484 char *path; 485 486 GIANT_REQUIRED; 487 488 /* 489 * Do we have a valid ELF header ? 490 */ 491 if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC) 492 return -1; 493 494 /* 495 * From here on down, we return an errno, not -1, as we've 496 * detected an ELF file. 497 */ 498 499 if ((hdr->e_phoff > PAGE_SIZE) || 500 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 501 /* Only support headers in first page for now */ 502 return ENOEXEC; 503 } 504 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff); 505 506 /* 507 * From this point on, we may have resources that need to be freed. 508 */ 509 510 /* 511 * Yeah, I'm paranoid. There is every reason in the world to get 512 * VTEXT now since from here on out, there are places we can have 513 * a context switch. Better safe than sorry; I really don't want 514 * the file to change while it's being loaded. 515 */ 516 mtx_lock(&imgp->vp->v_interlock); 517 imgp->vp->v_flag |= VTEXT; 518 mtx_unlock(&imgp->vp->v_interlock); 519 520 if ((error = exec_extract_strings(imgp)) != 0) 521 goto fail; 522 523 exec_new_vmspace(imgp); 524 525 vmspace = imgp->proc->p_vmspace; 526 527 for (i = 0; i < hdr->e_phnum; i++) { 528 switch(phdr[i].p_type) { 529 530 case PT_LOAD: /* Loadable segment */ 531 prot = 0; 532 if (phdr[i].p_flags & PF_X) 533 prot |= VM_PROT_EXECUTE; 534 if (phdr[i].p_flags & PF_W) 535 prot |= VM_PROT_WRITE; 536 if (phdr[i].p_flags & PF_R) 537 prot |= VM_PROT_READ; 538 539 if ((error = elf_load_section(imgp->proc, 540 vmspace, imgp->vp, 541 phdr[i].p_offset, 542 (caddr_t)phdr[i].p_vaddr, 543 phdr[i].p_memsz, 544 phdr[i].p_filesz, prot)) != 0) 545 goto fail; 546 547 /* 548 * Is this .text or .data ?? 549 * 550 * We only handle one each of those yet XXX 551 */ 552 if (hdr->e_entry >= phdr[i].p_vaddr && 553 hdr->e_entry <(phdr[i].p_vaddr+phdr[i].p_memsz)) { 554 text_addr = trunc_page(phdr[i].p_vaddr); 555 text_size = round_page(phdr[i].p_memsz + 556 phdr[i].p_vaddr - 557 text_addr); 558 entry = (u_long)hdr->e_entry; 559 } else { 560 data_addr = trunc_page(phdr[i].p_vaddr); 561 data_size = round_page(phdr[i].p_memsz + 562 phdr[i].p_vaddr - 563 data_addr); 564 } 565 break; 566 case PT_INTERP: /* Path to interpreter */ 567 if (phdr[i].p_filesz > MAXPATHLEN || 568 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) { 569 error = ENOEXEC; 570 goto fail; 571 } 572 interp = imgp->image_header + phdr[i].p_offset; 573 break; 574 case PT_PHDR: /* Program header table info */ 575 proghdr = phdr[i].p_vaddr; 576 break; 577 default: 578 break; 579 } 580 } 581 582 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 583 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 584 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 585 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 586 587 addr = ELF_RTLD_ADDR(vmspace); 588 589 imgp->entry_addr = entry; 590 591 brand_info = NULL; 592 593 /* We support three types of branding -- (1) the ELF EI_OSABI field 594 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 595 * branding w/in the ELF header, and (3) path of the `interp_path' 596 * field. We should also look for an ".note.ABI-tag" ELF section now 597 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones. 598 */ 599 600 /* If the executable has a brand, search for it in the brand list. */ 601 if (brand_info == NULL) { 602 for (i = 0; i < MAX_BRANDS; i++) { 603 Elf_Brandinfo *bi = elf_brand_list[i]; 604 605 if (bi != NULL && 606 (hdr->e_ident[EI_OSABI] == bi->brand 607 || 0 == 608 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 609 bi->compat_3_brand, strlen(bi->compat_3_brand)))) { 610 brand_info = bi; 611 break; 612 } 613 } 614 } 615 616 /* Lacking a known brand, search for a recognized interpreter. */ 617 if (brand_info == NULL && interp != NULL) { 618 for (i = 0; i < MAX_BRANDS; i++) { 619 Elf_Brandinfo *bi = elf_brand_list[i]; 620 621 if (bi != NULL && 622 strcmp(interp, bi->interp_path) == 0) { 623 brand_info = bi; 624 break; 625 } 626 } 627 } 628 629 /* Lacking a recognized interpreter, try the default brand */ 630 if (brand_info == NULL) { 631 for (i = 0; i < MAX_BRANDS; i++) { 632 Elf_Brandinfo *bi = elf_brand_list[i]; 633 634 if (bi != NULL && fallback_elf_brand == bi->brand) { 635 brand_info = bi; 636 break; 637 } 638 } 639 } 640 641 if (brand_info == NULL) { 642 uprintf("ELF binary type \"%u\" not known.\n", 643 hdr->e_ident[EI_OSABI]); 644 error = ENOEXEC; 645 goto fail; 646 } 647 648 imgp->proc->p_sysent = brand_info->sysvec; 649 if (interp != NULL) { 650 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 651 snprintf(path, MAXPATHLEN, "%s%s", 652 brand_info->emul_path, interp); 653 if ((error = elf_load_file(imgp->proc, path, &addr, 654 &imgp->entry_addr)) != 0) { 655 if ((error = elf_load_file(imgp->proc, interp, &addr, 656 &imgp->entry_addr)) != 0) { 657 uprintf("ELF interpreter %s not found\n", path); 658 free(path, M_TEMP); 659 goto fail; 660 } 661 } 662 free(path, M_TEMP); 663 } 664 665 /* 666 * Construct auxargs table (used by the fixup routine) 667 */ 668 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 669 elf_auxargs->execfd = -1; 670 elf_auxargs->phdr = proghdr; 671 elf_auxargs->phent = hdr->e_phentsize; 672 elf_auxargs->phnum = hdr->e_phnum; 673 elf_auxargs->pagesz = PAGE_SIZE; 674 elf_auxargs->base = addr; 675 elf_auxargs->flags = 0; 676 elf_auxargs->entry = entry; 677 elf_auxargs->trace = elf_trace; 678 679 imgp->auxargs = elf_auxargs; 680 imgp->interpreted = 0; 681 682fail: 683 return error; 684} 685 686static int 687elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp) 688{ 689 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 690 register_t *pos; 691 692 pos = *stack_base + (imgp->argc + imgp->envc + 2); 693 694 if (args->trace) { 695 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 696 } 697 if (args->execfd != -1) { 698 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 699 } 700 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 701 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 702 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 703 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 704 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 705 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 706 AUXARGS_ENTRY(pos, AT_BASE, args->base); 707 AUXARGS_ENTRY(pos, AT_NULL, 0); 708 709 free(imgp->auxargs, M_TEMP); 710 imgp->auxargs = NULL; 711 712 (*stack_base)--; 713 suword(*stack_base, (long) imgp->argc); 714 return 0; 715} 716 717/* 718 * Code for generating ELF core dumps. 719 */ 720 721typedef void (*segment_callback) __P((vm_map_entry_t, void *)); 722 723/* Closure for cb_put_phdr(). */ 724struct phdr_closure { 725 Elf_Phdr *phdr; /* Program header to fill in */ 726 Elf_Off offset; /* Offset of segment in core file */ 727}; 728 729/* Closure for cb_size_segment(). */ 730struct sseg_closure { 731 int count; /* Count of writable segments. */ 732 size_t size; /* Total size of all writable segments. */ 733}; 734 735static void cb_put_phdr __P((vm_map_entry_t, void *)); 736static void cb_size_segment __P((vm_map_entry_t, void *)); 737static void each_writable_segment __P((struct proc *, segment_callback, 738 void *)); 739static int elf_corehdr __P((struct proc *, struct vnode *, struct ucred *, 740 int, void *, size_t)); 741static void elf_puthdr __P((struct proc *, void *, size_t *, 742 const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int)); 743static void elf_putnote __P((void *, size_t *, const char *, int, 744 const void *, size_t)); 745 746extern int osreldate; 747 748int 749elf_coredump(p, vp, limit) 750 register struct proc *p; 751 register struct vnode *vp; 752 off_t limit; 753{ 754 register struct ucred *cred = p->p_ucred; 755 int error = 0; 756 struct sseg_closure seginfo; 757 void *hdr; 758 size_t hdrsize; 759 760 /* Size the program segments. */ 761 seginfo.count = 0; 762 seginfo.size = 0; 763 each_writable_segment(p, cb_size_segment, &seginfo); 764 765 /* 766 * Calculate the size of the core file header area by making 767 * a dry run of generating it. Nothing is written, but the 768 * size is calculated. 769 */ 770 hdrsize = 0; 771 elf_puthdr((struct proc *)NULL, (void *)NULL, &hdrsize, 772 (const prstatus_t *)NULL, (const prfpregset_t *)NULL, 773 (const prpsinfo_t *)NULL, seginfo.count); 774 775 if (hdrsize + seginfo.size >= limit) 776 return (EFAULT); 777 778 /* 779 * Allocate memory for building the header, fill it up, 780 * and write it out. 781 */ 782 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 783 if (hdr == NULL) { 784 return EINVAL; 785 } 786 error = elf_corehdr(p, vp, cred, seginfo.count, hdr, hdrsize); 787 788 /* Write the contents of all of the writable segments. */ 789 if (error == 0) { 790 Elf_Phdr *php; 791 off_t offset; 792 int i; 793 794 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 795 offset = hdrsize; 796 for (i = 0; i < seginfo.count; i++) { 797 error = vn_rdwr_inchunks(UIO_WRITE, vp, 798 (caddr_t)php->p_vaddr, 799 php->p_filesz, offset, UIO_USERSPACE, 800 IO_UNIT, cred, (int *)NULL, p); 801 if (error != 0) 802 break; 803 offset += php->p_filesz; 804 php++; 805 } 806 } 807 free(hdr, M_TEMP); 808 809 return error; 810} 811 812/* 813 * A callback for each_writable_segment() to write out the segment's 814 * program header entry. 815 */ 816static void 817cb_put_phdr(entry, closure) 818 vm_map_entry_t entry; 819 void *closure; 820{ 821 struct phdr_closure *phc = (struct phdr_closure *)closure; 822 Elf_Phdr *phdr = phc->phdr; 823 824 phc->offset = round_page(phc->offset); 825 826 phdr->p_type = PT_LOAD; 827 phdr->p_offset = phc->offset; 828 phdr->p_vaddr = entry->start; 829 phdr->p_paddr = 0; 830 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 831 phdr->p_align = PAGE_SIZE; 832 phdr->p_flags = 0; 833 if (entry->protection & VM_PROT_READ) 834 phdr->p_flags |= PF_R; 835 if (entry->protection & VM_PROT_WRITE) 836 phdr->p_flags |= PF_W; 837 if (entry->protection & VM_PROT_EXECUTE) 838 phdr->p_flags |= PF_X; 839 840 phc->offset += phdr->p_filesz; 841 phc->phdr++; 842} 843 844/* 845 * A callback for each_writable_segment() to gather information about 846 * the number of segments and their total size. 847 */ 848static void 849cb_size_segment(entry, closure) 850 vm_map_entry_t entry; 851 void *closure; 852{ 853 struct sseg_closure *ssc = (struct sseg_closure *)closure; 854 855 ssc->count++; 856 ssc->size += entry->end - entry->start; 857} 858 859/* 860 * For each writable segment in the process's memory map, call the given 861 * function with a pointer to the map entry and some arbitrary 862 * caller-supplied data. 863 */ 864static void 865each_writable_segment(p, func, closure) 866 struct proc *p; 867 segment_callback func; 868 void *closure; 869{ 870 vm_map_t map = &p->p_vmspace->vm_map; 871 vm_map_entry_t entry; 872 873 for (entry = map->header.next; entry != &map->header; 874 entry = entry->next) { 875 vm_object_t obj; 876 877 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) || 878 (entry->protection & (VM_PROT_READ|VM_PROT_WRITE)) != 879 (VM_PROT_READ|VM_PROT_WRITE)) 880 continue; 881 882 /* 883 ** Dont include memory segment in the coredump if 884 ** MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 885 ** madvise(2). 886 */ 887 if (entry->eflags & MAP_ENTRY_NOCOREDUMP) 888 continue; 889 890 if ((obj = entry->object.vm_object) == NULL) 891 continue; 892 893 /* Find the deepest backing object. */ 894 while (obj->backing_object != NULL) 895 obj = obj->backing_object; 896 897 /* Ignore memory-mapped devices and such things. */ 898 if (obj->type != OBJT_DEFAULT && 899 obj->type != OBJT_SWAP && 900 obj->type != OBJT_VNODE) 901 continue; 902 903 (*func)(entry, closure); 904 } 905} 906 907/* 908 * Write the core file header to the file, including padding up to 909 * the page boundary. 910 */ 911static int 912elf_corehdr(p, vp, cred, numsegs, hdr, hdrsize) 913 struct proc *p; 914 struct vnode *vp; 915 struct ucred *cred; 916 int numsegs; 917 size_t hdrsize; 918 void *hdr; 919{ 920 struct { 921 prstatus_t status; 922 prfpregset_t fpregset; 923 prpsinfo_t psinfo; 924 } *tempdata; 925 size_t off; 926 prstatus_t *status; 927 prfpregset_t *fpregset; 928 prpsinfo_t *psinfo; 929 930 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK); 931 status = &tempdata->status; 932 fpregset = &tempdata->fpregset; 933 psinfo = &tempdata->psinfo; 934 935 /* Gather the information for the header. */ 936 status->pr_version = PRSTATUS_VERSION; 937 status->pr_statussz = sizeof(prstatus_t); 938 status->pr_gregsetsz = sizeof(gregset_t); 939 status->pr_fpregsetsz = sizeof(fpregset_t); 940 status->pr_osreldate = osreldate; 941 status->pr_cursig = p->p_sig; 942 status->pr_pid = p->p_pid; 943 fill_regs(p, &status->pr_reg); 944 945 fill_fpregs(p, fpregset); 946 947 psinfo->pr_version = PRPSINFO_VERSION; 948 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 949 strncpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname) - 1); 950 951 /* XXX - We don't fill in the command line arguments properly yet. */ 952 strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ); 953 954 /* Fill in the header. */ 955 bzero(hdr, hdrsize); 956 off = 0; 957 elf_puthdr(p, hdr, &off, status, fpregset, psinfo, numsegs); 958 959 free(tempdata, M_TEMP); 960 961 /* Write it to the core file. */ 962 return vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 963 UIO_SYSSPACE, IO_UNIT, cred, NULL, p); 964} 965 966static void 967elf_puthdr(struct proc *p, void *dst, size_t *off, const prstatus_t *status, 968 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs) 969{ 970 size_t ehoff; 971 size_t phoff; 972 size_t noteoff; 973 size_t notesz; 974 975 ehoff = *off; 976 *off += sizeof(Elf_Ehdr); 977 978 phoff = *off; 979 *off += (numsegs + 1) * sizeof(Elf_Phdr); 980 981 noteoff = *off; 982 elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status, 983 sizeof *status); 984 elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 985 sizeof *fpregset); 986 elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 987 sizeof *psinfo); 988 notesz = *off - noteoff; 989 990 /* Align up to a page boundary for the program segments. */ 991 *off = round_page(*off); 992 993 if (dst != NULL) { 994 Elf_Ehdr *ehdr; 995 Elf_Phdr *phdr; 996 struct phdr_closure phc; 997 998 /* 999 * Fill in the ELF header. 1000 */ 1001 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1002 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1003 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1004 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1005 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1006 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1007 ehdr->e_ident[EI_DATA] = ELF_DATA; 1008 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1009 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1010 ehdr->e_ident[EI_ABIVERSION] = 0; 1011 ehdr->e_ident[EI_PAD] = 0; 1012 ehdr->e_type = ET_CORE; 1013 ehdr->e_machine = ELF_ARCH; 1014 ehdr->e_version = EV_CURRENT; 1015 ehdr->e_entry = 0; 1016 ehdr->e_phoff = phoff; 1017 ehdr->e_flags = 0; 1018 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1019 ehdr->e_phentsize = sizeof(Elf_Phdr); 1020 ehdr->e_phnum = numsegs + 1; 1021 ehdr->e_shentsize = sizeof(Elf_Shdr); 1022 ehdr->e_shnum = 0; 1023 ehdr->e_shstrndx = SHN_UNDEF; 1024 1025 /* 1026 * Fill in the program header entries. 1027 */ 1028 phdr = (Elf_Phdr *)((char *)dst + phoff); 1029 1030 /* The note segement. */ 1031 phdr->p_type = PT_NOTE; 1032 phdr->p_offset = noteoff; 1033 phdr->p_vaddr = 0; 1034 phdr->p_paddr = 0; 1035 phdr->p_filesz = notesz; 1036 phdr->p_memsz = 0; 1037 phdr->p_flags = 0; 1038 phdr->p_align = 0; 1039 phdr++; 1040 1041 /* All the writable segments from the program. */ 1042 phc.phdr = phdr; 1043 phc.offset = *off; 1044 each_writable_segment(p, cb_put_phdr, &phc); 1045 } 1046} 1047 1048static void 1049elf_putnote(void *dst, size_t *off, const char *name, int type, 1050 const void *desc, size_t descsz) 1051{ 1052 Elf_Note note; 1053 1054 note.n_namesz = strlen(name) + 1; 1055 note.n_descsz = descsz; 1056 note.n_type = type; 1057 if (dst != NULL) 1058 bcopy(¬e, (char *)dst + *off, sizeof note); 1059 *off += sizeof note; 1060 if (dst != NULL) 1061 bcopy(name, (char *)dst + *off, note.n_namesz); 1062 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1063 if (dst != NULL) 1064 bcopy(desc, (char *)dst + *off, note.n_descsz); 1065 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1066} 1067 1068/* 1069 * Tell kern_execve.c about it, with a little help from the linker. 1070 */ 1071static struct execsw elf_execsw = {exec_elf_imgact, "ELF"}; 1072EXEC_SET(elf, elf_execsw); 1073