imgact_elf.c revision 43748
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 * $Id: imgact_elf.c,v 1.51 1999/02/05 22:24:26 jdp Exp $ 30 */ 31 32#include "opt_rlimit.h" 33 34#include <sys/param.h> 35#include <sys/acct.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/malloc.h> 42#include <sys/mman.h> 43#include <sys/namei.h> 44#include <sys/pioctl.h> 45#include <sys/proc.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/systm.h> 54#include <sys/vnode.h> 55 56#include <vm/vm.h> 57#include <vm/vm_kern.h> 58#include <vm/vm_param.h> 59#include <vm/pmap.h> 60#include <sys/lock.h> 61#include <vm/vm_map.h> 62#include <vm/vm_object.h> 63#include <vm/vm_prot.h> 64#include <vm/vm_extern.h> 65 66#include <machine/elf.h> 67#include <machine/md_var.h> 68 69__ElfType(Brandinfo); 70__ElfType(Auxargs); 71 72static int elf_check_header __P((const Elf_Ehdr *hdr, int type)); 73static int elf_freebsd_fixup __P((long **stack_base, 74 struct image_params *imgp)); 75static int elf_load_file __P((struct proc *p, char *file, u_long *addr, 76 u_long *entry)); 77static int elf_load_section __P((struct proc *p, 78 struct vmspace *vmspace, struct vnode *vp, 79 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 80 vm_prot_t prot)); 81static int exec_elf_imgact __P((struct image_params *imgp)); 82 83static int elf_trace = 0; 84SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, ""); 85 86/* 87 * XXX Maximum length of an ELF brand (sysctl wants a statically-allocated 88 * buffer). 89 */ 90#define MAXBRANDLEN 16 91 92static struct sysentvec elf_freebsd_sysvec = { 93 SYS_MAXSYSCALL, 94 sysent, 95 0, 96 0, 97 0, 98 0, 99 0, 100 0, 101 elf_freebsd_fixup, 102 sendsig, 103 sigcode, 104 &szsigcode, 105 0, 106 "FreeBSD ELF", 107 elf_coredump 108}; 109 110static Elf_Brandinfo freebsd_brand_info = { 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; 158 159 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 160 if (p->p_sysent == entry->sysvec) 161 return TRUE; 162 } 163 164 return FALSE; 165} 166 167static int 168elf_check_header(const Elf_Ehdr *hdr, int type) 169{ 170 if (!IS_ELF(*hdr) || 171 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 172 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 173 hdr->e_ident[EI_VERSION] != EV_CURRENT) 174 return ENOEXEC; 175 176 if (!ELF_MACHINE_OK(hdr->e_machine)) 177 return ENOEXEC; 178 179 if (hdr->e_type != type || hdr->e_version != ELF_TARG_VER) 180 return ENOEXEC; 181 182 return 0; 183} 184 185static int 186elf_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) 187{ 188 size_t map_len; 189 vm_offset_t map_addr; 190 int error, rv; 191 size_t copy_len; 192 vm_object_t object; 193 vm_offset_t file_addr; 194 vm_offset_t data_buf = 0; 195 196 object = vp->v_object; 197 error = 0; 198 199 map_addr = trunc_page((vm_offset_t)vmaddr); 200 file_addr = trunc_page(offset); 201 202 /* 203 * We have two choices. We can either clear the data in the last page 204 * of an oversized mapping, or we can start the anon mapping a page 205 * early and copy the initialized data into that first page. We 206 * choose the second.. 207 */ 208 if (memsz > filsz) 209 map_len = trunc_page(offset+filsz) - file_addr; 210 else 211 map_len = round_page(offset+filsz) - file_addr; 212 213 if (map_len != 0) { 214 vm_object_reference(object); 215 vm_map_lock(&vmspace->vm_map); 216 rv = vm_map_insert(&vmspace->vm_map, 217 object, 218 file_addr, /* file offset */ 219 map_addr, /* virtual start */ 220 map_addr + map_len,/* virtual end */ 221 prot, 222 VM_PROT_ALL, 223 MAP_COPY_NEEDED | MAP_COPY_ON_WRITE); 224 vm_map_unlock(&vmspace->vm_map); 225 if (rv != KERN_SUCCESS) 226 return EINVAL; 227 228 /* prefault the page tables */ 229 pmap_object_init_pt(&vmspace->vm_pmap, 230 map_addr, 231 object, 232 (vm_pindex_t) OFF_TO_IDX(file_addr), 233 map_len, 234 0); 235 236 /* we can stop now if we've covered it all */ 237 if (memsz == filsz) 238 return 0; 239 } 240 241 242 /* 243 * We have to get the remaining bit of the file into the first part 244 * of the oversized map segment. This is normally because the .data 245 * segment in the file is extended to provide bss. It's a neat idea 246 * to try and save a page, but it's a pain in the behind to implement. 247 */ 248 copy_len = (offset + filsz) - trunc_page(offset + filsz); 249 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 250 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 251 252 /* This had damn well better be true! */ 253 if (map_len != 0) { 254 vm_map_lock(&vmspace->vm_map); 255 rv = vm_map_insert(&vmspace->vm_map, NULL, 0, 256 map_addr, map_addr + map_len, 257 VM_PROT_ALL, VM_PROT_ALL, 0); 258 vm_map_unlock(&vmspace->vm_map); 259 if (rv != KERN_SUCCESS) 260 return EINVAL; 261 } 262 263 if (copy_len != 0) { 264 vm_object_reference(object); 265 rv = vm_map_find(exec_map, 266 object, 267 trunc_page(offset + filsz), 268 &data_buf, 269 PAGE_SIZE, 270 TRUE, 271 VM_PROT_READ, 272 VM_PROT_ALL, 273 MAP_COPY_ON_WRITE | MAP_COPY_NEEDED); 274 if (rv != KERN_SUCCESS) { 275 vm_object_deallocate(object); 276 return EINVAL; 277 } 278 pmap_object_init_pt(exec_map->pmap, data_buf, object, 279 (vm_pindex_t) OFF_TO_IDX(trunc_page(offset + filsz)), 280 PAGE_SIZE, 1); 281 282 /* send the page fragment to user space */ 283 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len); 284 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE); 285 if (error) 286 return (error); 287 } 288 289 /* 290 * set it to the specified protection 291 */ 292 vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len, prot, 293 FALSE); 294 295 return error; 296} 297 298static int 299elf_load_file(struct proc *p, char *file, u_long *addr, u_long *entry) 300{ 301 const Elf_Ehdr *hdr = NULL; 302 const Elf_Phdr *phdr = NULL; 303 struct nameidata nd; 304 struct vmspace *vmspace = p->p_vmspace; 305 struct vattr attr; 306 struct image_params image_params, *imgp; 307 vm_prot_t prot; 308 unsigned long text_size = 0, data_size = 0; 309 unsigned long text_addr = 0, data_addr = 0; 310 int error, i; 311 312 imgp = &image_params; 313 /* 314 * Initialize part of the common data 315 */ 316 imgp->proc = p; 317 imgp->uap = NULL; 318 imgp->attr = &attr; 319 imgp->firstpage = NULL; 320 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE); 321 322 if (imgp->image_header == NULL) { 323 nd.ni_vp = NULL; 324 error = ENOMEM; 325 goto fail; 326 } 327 328 NDINIT(&nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, p); 329 330 if ((error = namei(&nd)) != 0) { 331 nd.ni_vp = NULL; 332 goto fail; 333 } 334 335 imgp->vp = nd.ni_vp; 336 337 /* 338 * Check permissions, modes, uid, etc on the file, and "open" it. 339 */ 340 error = exec_check_permissions(imgp); 341 if (error) { 342 VOP_UNLOCK(nd.ni_vp, 0, p); 343 goto fail; 344 } 345 346 error = exec_map_first_page(imgp); 347 VOP_UNLOCK(nd.ni_vp, 0, p); 348 if (error) 349 goto fail; 350 351 hdr = (const Elf_Ehdr *)imgp->image_header; 352 if ((error = elf_check_header(hdr, ET_DYN)) != 0) 353 goto fail; 354 355 /* Only support headers that fit within first page for now */ 356 if ((hdr->e_phoff > PAGE_SIZE) || 357 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 358 error = ENOEXEC; 359 goto fail; 360 } 361 362 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 363 364 for (i = 0; i < hdr->e_phnum; i++) { 365 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 366 prot = 0; 367 if (phdr[i].p_flags & PF_X) 368 prot |= VM_PROT_EXECUTE; 369 if (phdr[i].p_flags & PF_W) 370 prot |= VM_PROT_WRITE; 371 if (phdr[i].p_flags & PF_R) 372 prot |= VM_PROT_READ; 373 374 if ((error = elf_load_section(p, vmspace, nd.ni_vp, 375 phdr[i].p_offset, 376 (caddr_t)phdr[i].p_vaddr + 377 (*addr), 378 phdr[i].p_memsz, 379 phdr[i].p_filesz, prot)) != 0) 380 goto fail; 381 382 /* 383 * Is this .text or .data ?? 384 * 385 * We only handle one each of those yet XXX 386 */ 387 if (hdr->e_entry >= phdr[i].p_vaddr && 388 hdr->e_entry <(phdr[i].p_vaddr+phdr[i].p_memsz)) { 389 text_addr = trunc_page(phdr[i].p_vaddr+(*addr)); 390 text_size = round_page(phdr[i].p_memsz + 391 phdr[i].p_vaddr - 392 trunc_page(phdr[i].p_vaddr)); 393 *entry=(unsigned long)hdr->e_entry+(*addr); 394 } else { 395 data_addr = trunc_page(phdr[i].p_vaddr+(*addr)); 396 data_size = round_page(phdr[i].p_memsz + 397 phdr[i].p_vaddr - 398 trunc_page(phdr[i].p_vaddr)); 399 } 400 } 401 } 402 403fail: 404 if (imgp->firstpage) 405 exec_unmap_first_page(imgp); 406 if (imgp->image_header) 407 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header, 408 PAGE_SIZE); 409 if (nd.ni_vp) 410 vrele(nd.ni_vp); 411 412 return error; 413} 414 415static char fallback_elf_brand[MAXBRANDLEN+1] = { "none" }; 416SYSCTL_STRING(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW, 417 fallback_elf_brand, sizeof(fallback_elf_brand), 418 "ELF brand of last resort"); 419 420static int 421exec_elf_imgact(struct image_params *imgp) 422{ 423 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header; 424 const Elf_Phdr *phdr; 425 Elf_Auxargs *elf_auxargs = NULL; 426 struct vmspace *vmspace; 427 vm_prot_t prot; 428 u_long text_size = 0, data_size = 0; 429 u_long text_addr = 0, data_addr = 0; 430 u_long addr, entry = 0, proghdr = 0; 431 int error, i; 432 const char *interp = NULL; 433 Elf_Brandinfo *brand_info; 434 const char *brand; 435 char path[MAXPATHLEN]; 436 437 /* 438 * Do we have a valid ELF header ? 439 */ 440 if (elf_check_header(hdr, ET_EXEC)) 441 return -1; 442 443 /* 444 * From here on down, we return an errno, not -1, as we've 445 * detected an ELF file. 446 */ 447 448 if ((hdr->e_phoff > PAGE_SIZE) || 449 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 450 /* Only support headers in first page for now */ 451 return ENOEXEC; 452 } 453 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff); 454 455 /* 456 * From this point on, we may have resources that need to be freed. 457 */ 458 if ((error = exec_extract_strings(imgp)) != 0) 459 goto fail; 460 461 exec_new_vmspace(imgp); 462 463 vmspace = imgp->proc->p_vmspace; 464 465 for (i = 0; i < hdr->e_phnum; i++) { 466 switch(phdr[i].p_type) { 467 468 case PT_LOAD: /* Loadable segment */ 469 prot = 0; 470 if (phdr[i].p_flags & PF_X) 471 prot |= VM_PROT_EXECUTE; 472 if (phdr[i].p_flags & PF_W) 473 prot |= VM_PROT_WRITE; 474 if (phdr[i].p_flags & PF_R) 475 prot |= VM_PROT_READ; 476 477 if ((error = elf_load_section(imgp->proc, 478 vmspace, imgp->vp, 479 phdr[i].p_offset, 480 (caddr_t)phdr[i].p_vaddr, 481 phdr[i].p_memsz, 482 phdr[i].p_filesz, prot)) != 0) 483 goto fail; 484 485 /* 486 * Is this .text or .data ?? 487 * 488 * We only handle one each of those yet XXX 489 */ 490 if (hdr->e_entry >= phdr[i].p_vaddr && 491 hdr->e_entry <(phdr[i].p_vaddr+phdr[i].p_memsz)) { 492 text_addr = trunc_page(phdr[i].p_vaddr); 493 text_size = round_page(phdr[i].p_memsz + 494 phdr[i].p_vaddr - 495 text_addr); 496 entry = (u_long)hdr->e_entry; 497 } else { 498 data_addr = trunc_page(phdr[i].p_vaddr); 499 data_size = round_page(phdr[i].p_memsz + 500 phdr[i].p_vaddr - 501 data_addr); 502 } 503 break; 504 case PT_INTERP: /* Path to interpreter */ 505 if (phdr[i].p_filesz > MAXPATHLEN || 506 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) { 507 error = ENOEXEC; 508 goto fail; 509 } 510 interp = imgp->image_header + phdr[i].p_offset; 511 break; 512 case PT_PHDR: /* Program header table info */ 513 proghdr = phdr[i].p_vaddr; 514 break; 515 default: 516 break; 517 } 518 } 519 520 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 521 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 522 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 523 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 524 525 addr = 2L*MAXDSIZ; /* May depend on OS type XXX */ 526 527 imgp->entry_addr = entry; 528 529 /* If the executable has a brand, search for it in the brand list. */ 530 brand_info = NULL; 531 brand = (const char *)&hdr->e_ident[EI_BRAND]; 532 if (brand[0] != '\0') { 533 for (i = 0; i < MAX_BRANDS; i++) { 534 Elf_Brandinfo *bi = elf_brand_list[i]; 535 536 if (bi != NULL && strcmp(brand, bi->brand) == 0) { 537 brand_info = bi; 538 break; 539 } 540 } 541 } 542 543 /* Lacking a known brand, search for a recognized interpreter. */ 544 if (brand_info == NULL && interp != NULL) { 545 for (i = 0; i < MAX_BRANDS; i++) { 546 Elf_Brandinfo *bi = elf_brand_list[i]; 547 548 if (bi != NULL && 549 strcmp(interp, bi->interp_path) == 0) { 550 brand_info = bi; 551 break; 552 } 553 } 554 } 555 556 /* Lacking a recognized interpreter, try the default brand */ 557 if (brand_info == NULL && fallback_elf_brand[0] != '\0') { 558 for (i = 0; i < MAX_BRANDS; i++) { 559 Elf_Brandinfo *bi = elf_brand_list[i]; 560 561 if (bi != NULL 562 && strcmp(fallback_elf_brand, bi->brand) == 0) { 563 brand_info = bi; 564 break; 565 } 566 } 567 } 568 569#ifdef __alpha__ 570 /* XXX - Assume FreeBSD on the alpha. */ 571 if (brand_info == NULL) 572 brand_info = &freebsd_brand_info; 573#endif 574 575 if (brand_info == NULL) { 576 if (brand[0] == 0) 577 uprintf("ELF binary type not known." 578 " Use \"brandelf\" to brand it.\n"); 579 else 580 uprintf("ELF binary type \"%.*s\" not known.\n", 581 EI_NIDENT - EI_BRAND, brand); 582 error = ENOEXEC; 583 goto fail; 584 } 585 586 imgp->proc->p_sysent = brand_info->sysvec; 587 if (interp != NULL) { 588 snprintf(path, sizeof(path), "%s%s", 589 brand_info->emul_path, interp); 590 if ((error = elf_load_file(imgp->proc, path, &addr, 591 &imgp->entry_addr)) != 0) { 592 uprintf("ELF interpreter %s not found\n", path); 593 goto fail; 594 } 595 } 596 597 /* 598 * Construct auxargs table (used by the fixup routine) 599 */ 600 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 601 elf_auxargs->execfd = -1; 602 elf_auxargs->phdr = proghdr; 603 elf_auxargs->phent = hdr->e_phentsize; 604 elf_auxargs->phnum = hdr->e_phnum; 605 elf_auxargs->pagesz = PAGE_SIZE; 606 elf_auxargs->base = addr; 607 elf_auxargs->flags = 0; 608 elf_auxargs->entry = entry; 609 elf_auxargs->trace = elf_trace; 610 611 imgp->auxargs = elf_auxargs; 612 imgp->interpreted = 0; 613 614 /* don't allow modifying the file while we run it */ 615 imgp->vp->v_flag |= VTEXT; 616 617fail: 618 return error; 619} 620 621static int 622elf_freebsd_fixup(long **stack_base, struct image_params *imgp) 623{ 624 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 625 long *pos; 626 627 pos = *stack_base + (imgp->argc + imgp->envc + 2); 628 629 if (args->trace) { 630 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 631 } 632 if (args->execfd != -1) { 633 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 634 } 635 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 636 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 637 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 638 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 639 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 640 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 641 AUXARGS_ENTRY(pos, AT_BASE, args->base); 642 AUXARGS_ENTRY(pos, AT_NULL, 0); 643 644 free(imgp->auxargs, M_TEMP); 645 imgp->auxargs = NULL; 646 647 (*stack_base)--; 648 suword(*stack_base, (long) imgp->argc); 649 return 0; 650} 651 652/* 653 * Code for generating ELF core dumps. 654 */ 655 656typedef void (*segment_callback) __P((vm_map_entry_t, void *)); 657 658/* Closure for cb_put_phdr(). */ 659struct phdr_closure { 660 Elf_Phdr *phdr; /* Program header to fill in */ 661 Elf_Off offset; /* Offset of segment in core file */ 662}; 663 664/* Closure for cb_size_segment(). */ 665struct sseg_closure { 666 int count; /* Count of writable segments. */ 667 size_t size; /* Total size of all writable segments. */ 668}; 669 670static void cb_put_phdr __P((vm_map_entry_t, void *)); 671static void cb_size_segment __P((vm_map_entry_t, void *)); 672static void each_writable_segment __P((struct proc *, segment_callback, 673 void *)); 674static int elf_corehdr __P((struct proc *, struct vnode *, struct ucred *, 675 int, void *, size_t)); 676static void elf_puthdr __P((struct proc *, void *, size_t *, 677 const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int)); 678static void elf_putnote __P((void *, size_t *, const char *, int, 679 const void *, size_t)); 680 681extern int osreldate; 682 683int 684elf_coredump(p) 685 register struct proc *p; 686{ 687 register struct vnode *vp; 688 register struct ucred *cred = p->p_cred->pc_ucred; 689 struct nameidata nd; 690 struct vattr vattr; 691 int error, error1; 692 char *name; /* name of corefile */ 693 struct sseg_closure seginfo; 694 void *hdr; 695 size_t hdrsize; 696 697 STOPEVENT(p, S_CORE, 0); 698 699 if (sugid_coredump == 0 && p->p_flag & P_SUGID) 700 return (EFAULT); 701 702 /* Size the program segments. */ 703 seginfo.count = 0; 704 seginfo.size = 0; 705 each_writable_segment(p, cb_size_segment, &seginfo); 706 707 /* 708 * Calculate the size of the core file header area by making 709 * a dry run of generating it. Nothing is written, but the 710 * size is calculated. 711 */ 712 hdrsize = 0; 713 elf_puthdr((struct proc *)NULL, (void *)NULL, &hdrsize, 714 (const prstatus_t *)NULL, (const prfpregset_t *)NULL, 715 (const prpsinfo_t *)NULL, seginfo.count); 716 717 if (hdrsize + seginfo.size >= p->p_rlimit[RLIMIT_CORE].rlim_cur) 718 return (EFAULT); 719 name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid); 720 if (name == NULL) 721 return (EFAULT); /* XXX -- not the best error */ 722 723 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, name, p); 724 error = vn_open(&nd, O_CREAT | FWRITE, S_IRUSR | S_IWUSR); 725 free(name, M_TEMP); 726 if (error) 727 return (error); 728 vp = nd.ni_vp; 729 730 /* Don't dump to non-regular files or files with links. */ 731 if (vp->v_type != VREG || 732 VOP_GETATTR(vp, &vattr, cred, p) || vattr.va_nlink != 1) { 733 error = EFAULT; 734 goto out; 735 } 736 VATTR_NULL(&vattr); 737 vattr.va_size = 0; 738 VOP_LEASE(vp, p, cred, LEASE_WRITE); 739 VOP_SETATTR(vp, &vattr, cred, p); 740 p->p_acflag |= ACORE; 741 742 743 /* 744 * Allocate memory for building the header, fill it up, 745 * and write it out. 746 */ 747 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 748 if (hdr == NULL) { 749 error = EINVAL; 750 goto out; 751 } 752 error = elf_corehdr(p, vp, cred, seginfo.count, hdr, hdrsize); 753 754 /* Write the contents of all of the writable segments. */ 755 if (error == 0) { 756 Elf_Phdr *php; 757 off_t offset; 758 int i; 759 760 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 761 offset = hdrsize; 762 for (i = 0; i < seginfo.count; i++) { 763 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)php->p_vaddr, 764 php->p_filesz, offset, UIO_USERSPACE, 765 IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, p); 766 if (error != 0) 767 break; 768 offset += php->p_filesz; 769 php++; 770 } 771 } 772 free(hdr, M_TEMP); 773 774out: 775 VOP_UNLOCK(vp, 0, p); 776 error1 = vn_close(vp, FWRITE, cred, p); 777 if (error == 0) 778 error = error1; 779 return (error); 780} 781 782/* 783 * A callback for each_writable_segment() to write out the segment's 784 * program header entry. 785 */ 786static void 787cb_put_phdr(entry, closure) 788 vm_map_entry_t entry; 789 void *closure; 790{ 791 struct phdr_closure *phc = (struct phdr_closure *)closure; 792 Elf_Phdr *phdr = phc->phdr; 793 794 phc->offset = round_page(phc->offset); 795 796 phdr->p_type = PT_LOAD; 797 phdr->p_offset = phc->offset; 798 phdr->p_vaddr = entry->start; 799 phdr->p_paddr = 0; 800 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 801 phdr->p_align = PAGE_SIZE; 802 phdr->p_flags = 0; 803 if (entry->protection & VM_PROT_READ) 804 phdr->p_flags |= PF_R; 805 if (entry->protection & VM_PROT_WRITE) 806 phdr->p_flags |= PF_W; 807 if (entry->protection & VM_PROT_EXECUTE) 808 phdr->p_flags |= PF_X; 809 810 phc->offset += phdr->p_filesz; 811 phc->phdr++; 812} 813 814/* 815 * A callback for each_writable_segment() to gather information about 816 * the number of segments and their total size. 817 */ 818static void 819cb_size_segment(entry, closure) 820 vm_map_entry_t entry; 821 void *closure; 822{ 823 struct sseg_closure *ssc = (struct sseg_closure *)closure; 824 825 ssc->count++; 826 ssc->size += entry->end - entry->start; 827} 828 829/* 830 * For each writable segment in the process's memory map, call the given 831 * function with a pointer to the map entry and some arbitrary 832 * caller-supplied data. 833 */ 834static void 835each_writable_segment(p, func, closure) 836 struct proc *p; 837 segment_callback func; 838 void *closure; 839{ 840 vm_map_t map = &p->p_vmspace->vm_map; 841 vm_map_entry_t entry; 842 843 for (entry = map->header.next; entry != &map->header; 844 entry = entry->next) { 845 vm_object_t obj; 846 847 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) || 848 (entry->protection & (VM_PROT_READ|VM_PROT_WRITE)) != 849 (VM_PROT_READ|VM_PROT_WRITE)) 850 continue; 851 852 if ((obj = entry->object.vm_object) == NULL) 853 continue; 854 855 /* Find the deepest backing object. */ 856 while (obj->backing_object != NULL) 857 obj = obj->backing_object; 858 859 /* Ignore memory-mapped devices and such things. */ 860 if (obj->type != OBJT_DEFAULT && 861 obj->type != OBJT_SWAP && 862 obj->type != OBJT_VNODE) 863 continue; 864 865 (*func)(entry, closure); 866 } 867} 868 869/* 870 * Write the core file header to the file, including padding up to 871 * the page boundary. 872 */ 873static int 874elf_corehdr(p, vp, cred, numsegs, hdr, hdrsize) 875 struct proc *p; 876 struct vnode *vp; 877 struct ucred *cred; 878 int numsegs; 879 size_t hdrsize; 880 void *hdr; 881{ 882 size_t off; 883 prstatus_t status; 884 prfpregset_t fpregset; 885 prpsinfo_t psinfo; 886 887 /* Gather the information for the header. */ 888 bzero(&status, sizeof status); 889 status.pr_version = PRSTATUS_VERSION; 890 status.pr_statussz = sizeof(prstatus_t); 891 status.pr_gregsetsz = sizeof(gregset_t); 892 status.pr_fpregsetsz = sizeof(fpregset_t); 893 status.pr_osreldate = osreldate; 894 status.pr_cursig = p->p_sig; 895 status.pr_pid = p->p_pid; 896 fill_regs(p, &status.pr_reg); 897 898 fill_fpregs(p, &fpregset); 899 900 bzero(&psinfo, sizeof psinfo); 901 psinfo.pr_version = PRPSINFO_VERSION; 902 psinfo.pr_psinfosz = sizeof(prpsinfo_t); 903 strncpy(psinfo.pr_fname, p->p_comm, MAXCOMLEN); 904 /* XXX - We don't fill in the command line arguments properly yet. */ 905 strncpy(psinfo.pr_psargs, p->p_comm, PRARGSZ); 906 907 /* Fill in the header. */ 908 bzero(hdr, hdrsize); 909 off = 0; 910 elf_puthdr(p, hdr, &off, &status, &fpregset, &psinfo, numsegs); 911 912 /* Write it to the core file. */ 913 return vn_rdwr(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 914 UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, NULL, p); 915} 916 917static void 918elf_puthdr(struct proc *p, void *dst, size_t *off, const prstatus_t *status, 919 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs) 920{ 921 size_t ehoff; 922 size_t phoff; 923 size_t noteoff; 924 size_t notesz; 925 926 ehoff = *off; 927 *off += sizeof(Elf_Ehdr); 928 929 phoff = *off; 930 *off += (numsegs + 1) * sizeof(Elf_Phdr); 931 932 noteoff = *off; 933 elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status, 934 sizeof *status); 935 elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 936 sizeof *fpregset); 937 elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 938 sizeof *psinfo); 939 notesz = *off - noteoff; 940 941 /* Align up to a page boundary for the program segments. */ 942 *off = round_page(*off); 943 944 if (dst != NULL) { 945 Elf_Ehdr *ehdr; 946 Elf_Phdr *phdr; 947 struct phdr_closure phc; 948 949 /* 950 * Fill in the ELF header. 951 */ 952 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 953 ehdr->e_ident[EI_MAG0] = ELFMAG0; 954 ehdr->e_ident[EI_MAG1] = ELFMAG1; 955 ehdr->e_ident[EI_MAG2] = ELFMAG2; 956 ehdr->e_ident[EI_MAG3] = ELFMAG3; 957 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 958 ehdr->e_ident[EI_DATA] = ELF_DATA; 959 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 960 ehdr->e_ident[EI_PAD] = 0; 961 strncpy(ehdr->e_ident + EI_BRAND, "FreeBSD", 962 EI_NIDENT - EI_BRAND); 963 ehdr->e_type = ET_CORE; 964 ehdr->e_machine = ELF_ARCH; 965 ehdr->e_version = EV_CURRENT; 966 ehdr->e_entry = 0; 967 ehdr->e_phoff = phoff; 968 ehdr->e_flags = 0; 969 ehdr->e_ehsize = sizeof(Elf_Ehdr); 970 ehdr->e_phentsize = sizeof(Elf_Phdr); 971 ehdr->e_phnum = numsegs + 1; 972 ehdr->e_shentsize = sizeof(Elf_Shdr); 973 ehdr->e_shnum = 0; 974 ehdr->e_shstrndx = SHN_UNDEF; 975 976 /* 977 * Fill in the program header entries. 978 */ 979 phdr = (Elf_Phdr *)((char *)dst + phoff); 980 981 /* The note segement. */ 982 phdr->p_type = PT_NOTE; 983 phdr->p_offset = noteoff; 984 phdr->p_vaddr = 0; 985 phdr->p_paddr = 0; 986 phdr->p_filesz = notesz; 987 phdr->p_memsz = 0; 988 phdr->p_flags = 0; 989 phdr->p_align = 0; 990 phdr++; 991 992 /* All the writable segments from the program. */ 993 phc.phdr = phdr; 994 phc.offset = *off; 995 each_writable_segment(p, cb_put_phdr, &phc); 996 } 997} 998 999static void 1000elf_putnote(void *dst, size_t *off, const char *name, int type, 1001 const void *desc, size_t descsz) 1002{ 1003 Elf_Note note; 1004 1005 note.n_namesz = strlen(name) + 1; 1006 note.n_descsz = descsz; 1007 note.n_type = type; 1008 if (dst != NULL) 1009 bcopy(¬e, (char *)dst + *off, sizeof note); 1010 *off += sizeof note; 1011 if (dst != NULL) 1012 bcopy(name, (char *)dst + *off, note.n_namesz); 1013 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1014 if (dst != NULL) 1015 bcopy(desc, (char *)dst + *off, note.n_descsz); 1016 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1017} 1018 1019/* 1020 * Tell kern_execve.c about it, with a little help from the linker. 1021 * Since `const' objects end up in the text segment, TEXT_SET is the 1022 * correct directive to use. 1023 */ 1024static struct execsw elf_execsw = {exec_elf_imgact, "ELF"}; 1025EXEC_SET(elf, elf_execsw); 1026