imgact_elf.c revision 46803
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 1.55 1999/02/20 23:52:34 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)); 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) 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_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_pmap(vmspace), 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} 297 298/* 299 * Load the file "file" into memory. It may be either a shared object 300 * or an executable. 301 * 302 * The "addr" reference parameter is in/out. On entry, it specifies 303 * the address where a shared object should be loaded. If the file is 304 * an executable, this value is ignored. On exit, "addr" specifies 305 * where the file was actually loaded. 306 * 307 * The "entry" reference parameter is out only. On exit, it specifies 308 * the entry point for the loaded file. 309 */ 310static int 311elf_load_file(struct proc *p, char *file, u_long *addr, u_long *entry) 312{ 313 const Elf_Ehdr *hdr = NULL; 314 const Elf_Phdr *phdr = NULL; 315 struct nameidata nd; 316 struct vmspace *vmspace = p->p_vmspace; 317 struct vattr attr; 318 struct image_params image_params, *imgp; 319 vm_prot_t prot; 320 u_long rbase; 321 u_long base_addr = 0; 322 int error, i, numsegs; 323 324 imgp = &image_params; 325 /* 326 * Initialize part of the common data 327 */ 328 imgp->proc = p; 329 imgp->uap = NULL; 330 imgp->attr = &attr; 331 imgp->firstpage = NULL; 332 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE); 333 334 if (imgp->image_header == NULL) { 335 nd.ni_vp = NULL; 336 error = ENOMEM; 337 goto fail; 338 } 339 340 NDINIT(&nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, p); 341 342 if ((error = namei(&nd)) != 0) { 343 nd.ni_vp = NULL; 344 goto fail; 345 } 346 347 imgp->vp = nd.ni_vp; 348 349 /* 350 * Check permissions, modes, uid, etc on the file, and "open" it. 351 */ 352 error = exec_check_permissions(imgp); 353 if (error) { 354 VOP_UNLOCK(nd.ni_vp, 0, p); 355 goto fail; 356 } 357 358 error = exec_map_first_page(imgp); 359 VOP_UNLOCK(nd.ni_vp, 0, p); 360 if (error) 361 goto fail; 362 363 hdr = (const Elf_Ehdr *)imgp->image_header; 364 if ((error = elf_check_header(hdr)) != 0) 365 goto fail; 366 if (hdr->e_type == ET_DYN) 367 rbase = *addr; 368 else if (hdr->e_type == ET_EXEC) 369 rbase = 0; 370 else { 371 error = ENOEXEC; 372 goto fail; 373 } 374 375 /* Only support headers that fit within first page for now */ 376 if ((hdr->e_phoff > PAGE_SIZE) || 377 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 378 error = ENOEXEC; 379 goto fail; 380 } 381 382 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 383 384 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 385 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 386 prot = 0; 387 if (phdr[i].p_flags & PF_X) 388 prot |= VM_PROT_EXECUTE; 389 if (phdr[i].p_flags & PF_W) 390 prot |= VM_PROT_WRITE; 391 if (phdr[i].p_flags & PF_R) 392 prot |= VM_PROT_READ; 393 394 if ((error = elf_load_section(p, vmspace, nd.ni_vp, 395 phdr[i].p_offset, 396 (caddr_t)phdr[i].p_vaddr + 397 rbase, 398 phdr[i].p_memsz, 399 phdr[i].p_filesz, prot)) != 0) 400 goto fail; 401 /* 402 * Establish the base address if this is the 403 * first segment. 404 */ 405 if (numsegs == 0) 406 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 407 numsegs++; 408 } 409 } 410 *addr = base_addr; 411 *entry=(unsigned long)hdr->e_entry + rbase; 412 413fail: 414 if (imgp->firstpage) 415 exec_unmap_first_page(imgp); 416 if (imgp->image_header) 417 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header, 418 PAGE_SIZE); 419 if (nd.ni_vp) 420 vrele(nd.ni_vp); 421 422 return error; 423} 424 425static char fallback_elf_brand[MAXBRANDLEN+1] = { "none" }; 426SYSCTL_STRING(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW, 427 fallback_elf_brand, sizeof(fallback_elf_brand), 428 "ELF brand of last resort"); 429 430static int 431exec_elf_imgact(struct image_params *imgp) 432{ 433 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header; 434 const Elf_Phdr *phdr; 435 Elf_Auxargs *elf_auxargs = NULL; 436 struct vmspace *vmspace; 437 vm_prot_t prot; 438 u_long text_size = 0, data_size = 0; 439 u_long text_addr = 0, data_addr = 0; 440 u_long addr, entry = 0, proghdr = 0; 441 int error, i; 442 const char *interp = NULL; 443 Elf_Brandinfo *brand_info; 444 const char *brand; 445 char path[MAXPATHLEN]; 446 447 /* 448 * Do we have a valid ELF header ? 449 */ 450 if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC) 451 return -1; 452 453 /* 454 * From here on down, we return an errno, not -1, as we've 455 * detected an ELF file. 456 */ 457 458 if ((hdr->e_phoff > PAGE_SIZE) || 459 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 460 /* Only support headers in first page for now */ 461 return ENOEXEC; 462 } 463 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff); 464 465 /* 466 * From this point on, we may have resources that need to be freed. 467 */ 468 if ((error = exec_extract_strings(imgp)) != 0) 469 goto fail; 470 471 exec_new_vmspace(imgp); 472 473 vmspace = imgp->proc->p_vmspace; 474 475 for (i = 0; i < hdr->e_phnum; i++) { 476 switch(phdr[i].p_type) { 477 478 case PT_LOAD: /* Loadable segment */ 479 prot = 0; 480 if (phdr[i].p_flags & PF_X) 481 prot |= VM_PROT_EXECUTE; 482 if (phdr[i].p_flags & PF_W) 483 prot |= VM_PROT_WRITE; 484 if (phdr[i].p_flags & PF_R) 485 prot |= VM_PROT_READ; 486 487 if ((error = elf_load_section(imgp->proc, 488 vmspace, imgp->vp, 489 phdr[i].p_offset, 490 (caddr_t)phdr[i].p_vaddr, 491 phdr[i].p_memsz, 492 phdr[i].p_filesz, prot)) != 0) 493 goto fail; 494 495 /* 496 * Is this .text or .data ?? 497 * 498 * We only handle one each of those yet XXX 499 */ 500 if (hdr->e_entry >= phdr[i].p_vaddr && 501 hdr->e_entry <(phdr[i].p_vaddr+phdr[i].p_memsz)) { 502 text_addr = trunc_page(phdr[i].p_vaddr); 503 text_size = round_page(phdr[i].p_memsz + 504 phdr[i].p_vaddr - 505 text_addr); 506 entry = (u_long)hdr->e_entry; 507 } else { 508 data_addr = trunc_page(phdr[i].p_vaddr); 509 data_size = round_page(phdr[i].p_memsz + 510 phdr[i].p_vaddr - 511 data_addr); 512 } 513 break; 514 case PT_INTERP: /* Path to interpreter */ 515 if (phdr[i].p_filesz > MAXPATHLEN || 516 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) { 517 error = ENOEXEC; 518 goto fail; 519 } 520 interp = imgp->image_header + phdr[i].p_offset; 521 break; 522 case PT_PHDR: /* Program header table info */ 523 proghdr = phdr[i].p_vaddr; 524 break; 525 default: 526 break; 527 } 528 } 529 530 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 531 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 532 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 533 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 534 535 addr = ELF_RTLD_ADDR(vmspace); 536 537 imgp->entry_addr = entry; 538 539 /* If the executable has a brand, search for it in the brand list. */ 540 brand_info = NULL; 541 brand = (const char *)&hdr->e_ident[EI_BRAND]; 542 if (brand[0] != '\0') { 543 for (i = 0; i < MAX_BRANDS; i++) { 544 Elf_Brandinfo *bi = elf_brand_list[i]; 545 546 if (bi != NULL && strcmp(brand, bi->brand) == 0) { 547 brand_info = bi; 548 break; 549 } 550 } 551 } 552 553 /* Lacking a known brand, search for a recognized interpreter. */ 554 if (brand_info == NULL && interp != NULL) { 555 for (i = 0; i < MAX_BRANDS; i++) { 556 Elf_Brandinfo *bi = elf_brand_list[i]; 557 558 if (bi != NULL && 559 strcmp(interp, bi->interp_path) == 0) { 560 brand_info = bi; 561 break; 562 } 563 } 564 } 565 566 /* Lacking a recognized interpreter, try the default brand */ 567 if (brand_info == NULL && fallback_elf_brand[0] != '\0') { 568 for (i = 0; i < MAX_BRANDS; i++) { 569 Elf_Brandinfo *bi = elf_brand_list[i]; 570 571 if (bi != NULL 572 && strcmp(fallback_elf_brand, bi->brand) == 0) { 573 brand_info = bi; 574 break; 575 } 576 } 577 } 578 579#ifdef __alpha__ 580 /* XXX - Assume FreeBSD on the alpha. */ 581 if (brand_info == NULL) 582 brand_info = &freebsd_brand_info; 583#endif 584 585 if (brand_info == NULL) { 586 if (brand[0] == 0) 587 uprintf("ELF binary type not known." 588 " Use \"brandelf\" to brand it.\n"); 589 else 590 uprintf("ELF binary type \"%.*s\" not known.\n", 591 EI_NIDENT - EI_BRAND, brand); 592 error = ENOEXEC; 593 goto fail; 594 } 595 596 imgp->proc->p_sysent = brand_info->sysvec; 597 if (interp != NULL) { 598 snprintf(path, sizeof(path), "%s%s", 599 brand_info->emul_path, interp); 600 if ((error = elf_load_file(imgp->proc, path, &addr, 601 &imgp->entry_addr)) != 0) { 602 uprintf("ELF interpreter %s not found\n", path); 603 goto fail; 604 } 605 } 606 607 /* 608 * Construct auxargs table (used by the fixup routine) 609 */ 610 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 611 elf_auxargs->execfd = -1; 612 elf_auxargs->phdr = proghdr; 613 elf_auxargs->phent = hdr->e_phentsize; 614 elf_auxargs->phnum = hdr->e_phnum; 615 elf_auxargs->pagesz = PAGE_SIZE; 616 elf_auxargs->base = addr; 617 elf_auxargs->flags = 0; 618 elf_auxargs->entry = entry; 619 elf_auxargs->trace = elf_trace; 620 621 imgp->auxargs = elf_auxargs; 622 imgp->interpreted = 0; 623 624 /* don't allow modifying the file while we run it */ 625 imgp->vp->v_flag |= VTEXT; 626 627fail: 628 return error; 629} 630 631static int 632elf_freebsd_fixup(long **stack_base, struct image_params *imgp) 633{ 634 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 635 long *pos; 636 637 pos = *stack_base + (imgp->argc + imgp->envc + 2); 638 639 if (args->trace) { 640 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 641 } 642 if (args->execfd != -1) { 643 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 644 } 645 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 646 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 647 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 648 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 649 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 650 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 651 AUXARGS_ENTRY(pos, AT_BASE, args->base); 652 AUXARGS_ENTRY(pos, AT_NULL, 0); 653 654 free(imgp->auxargs, M_TEMP); 655 imgp->auxargs = NULL; 656 657 (*stack_base)--; 658 suword(*stack_base, (long) imgp->argc); 659 return 0; 660} 661 662/* 663 * Code for generating ELF core dumps. 664 */ 665 666typedef void (*segment_callback) __P((vm_map_entry_t, void *)); 667 668/* Closure for cb_put_phdr(). */ 669struct phdr_closure { 670 Elf_Phdr *phdr; /* Program header to fill in */ 671 Elf_Off offset; /* Offset of segment in core file */ 672}; 673 674/* Closure for cb_size_segment(). */ 675struct sseg_closure { 676 int count; /* Count of writable segments. */ 677 size_t size; /* Total size of all writable segments. */ 678}; 679 680static void cb_put_phdr __P((vm_map_entry_t, void *)); 681static void cb_size_segment __P((vm_map_entry_t, void *)); 682static void each_writable_segment __P((struct proc *, segment_callback, 683 void *)); 684static int elf_corehdr __P((struct proc *, struct vnode *, struct ucred *, 685 int, void *, size_t)); 686static void elf_puthdr __P((struct proc *, void *, size_t *, 687 const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int)); 688static void elf_putnote __P((void *, size_t *, const char *, int, 689 const void *, size_t)); 690 691extern int osreldate; 692 693int 694elf_coredump(p) 695 register struct proc *p; 696{ 697 register struct vnode *vp; 698 register struct ucred *cred = p->p_cred->pc_ucred; 699 struct nameidata nd; 700 struct vattr vattr; 701 int error, error1; 702 char *name; /* name of corefile */ 703 struct sseg_closure seginfo; 704 void *hdr; 705 size_t hdrsize; 706 707 STOPEVENT(p, S_CORE, 0); 708 709 if (sugid_coredump == 0 && p->p_flag & P_SUGID) 710 return (EFAULT); 711 712 /* Size the program segments. */ 713 seginfo.count = 0; 714 seginfo.size = 0; 715 each_writable_segment(p, cb_size_segment, &seginfo); 716 717 /* 718 * Calculate the size of the core file header area by making 719 * a dry run of generating it. Nothing is written, but the 720 * size is calculated. 721 */ 722 hdrsize = 0; 723 elf_puthdr((struct proc *)NULL, (void *)NULL, &hdrsize, 724 (const prstatus_t *)NULL, (const prfpregset_t *)NULL, 725 (const prpsinfo_t *)NULL, seginfo.count); 726 727 if (hdrsize + seginfo.size >= p->p_rlimit[RLIMIT_CORE].rlim_cur) 728 return (EFAULT); 729 name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid); 730 if (name == NULL) 731 return (EFAULT); /* XXX -- not the best error */ 732 733 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, name, p); 734 error = vn_open(&nd, O_CREAT | FWRITE, S_IRUSR | S_IWUSR); 735 free(name, M_TEMP); 736 if (error) 737 return (error); 738 vp = nd.ni_vp; 739 740 /* Don't dump to non-regular files or files with links. */ 741 if (vp->v_type != VREG || 742 VOP_GETATTR(vp, &vattr, cred, p) || vattr.va_nlink != 1) { 743 error = EFAULT; 744 goto out; 745 } 746 VATTR_NULL(&vattr); 747 vattr.va_size = 0; 748 VOP_LEASE(vp, p, cred, LEASE_WRITE); 749 VOP_SETATTR(vp, &vattr, cred, p); 750 p->p_acflag |= ACORE; 751 752 753 /* 754 * Allocate memory for building the header, fill it up, 755 * and write it out. 756 */ 757 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 758 if (hdr == NULL) { 759 error = EINVAL; 760 goto out; 761 } 762 error = elf_corehdr(p, vp, cred, seginfo.count, hdr, hdrsize); 763 764 /* Write the contents of all of the writable segments. */ 765 if (error == 0) { 766 Elf_Phdr *php; 767 off_t offset; 768 int i; 769 770 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 771 offset = hdrsize; 772 for (i = 0; i < seginfo.count; i++) { 773 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)php->p_vaddr, 774 php->p_filesz, offset, UIO_USERSPACE, 775 IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, p); 776 if (error != 0) 777 break; 778 offset += php->p_filesz; 779 php++; 780 } 781 } 782 free(hdr, M_TEMP); 783 784out: 785 VOP_UNLOCK(vp, 0, p); 786 error1 = vn_close(vp, FWRITE, cred, p); 787 if (error == 0) 788 error = error1; 789 return (error); 790} 791 792/* 793 * A callback for each_writable_segment() to write out the segment's 794 * program header entry. 795 */ 796static void 797cb_put_phdr(entry, closure) 798 vm_map_entry_t entry; 799 void *closure; 800{ 801 struct phdr_closure *phc = (struct phdr_closure *)closure; 802 Elf_Phdr *phdr = phc->phdr; 803 804 phc->offset = round_page(phc->offset); 805 806 phdr->p_type = PT_LOAD; 807 phdr->p_offset = phc->offset; 808 phdr->p_vaddr = entry->start; 809 phdr->p_paddr = 0; 810 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 811 phdr->p_align = PAGE_SIZE; 812 phdr->p_flags = 0; 813 if (entry->protection & VM_PROT_READ) 814 phdr->p_flags |= PF_R; 815 if (entry->protection & VM_PROT_WRITE) 816 phdr->p_flags |= PF_W; 817 if (entry->protection & VM_PROT_EXECUTE) 818 phdr->p_flags |= PF_X; 819 820 phc->offset += phdr->p_filesz; 821 phc->phdr++; 822} 823 824/* 825 * A callback for each_writable_segment() to gather information about 826 * the number of segments and their total size. 827 */ 828static void 829cb_size_segment(entry, closure) 830 vm_map_entry_t entry; 831 void *closure; 832{ 833 struct sseg_closure *ssc = (struct sseg_closure *)closure; 834 835 ssc->count++; 836 ssc->size += entry->end - entry->start; 837} 838 839/* 840 * For each writable segment in the process's memory map, call the given 841 * function with a pointer to the map entry and some arbitrary 842 * caller-supplied data. 843 */ 844static void 845each_writable_segment(p, func, closure) 846 struct proc *p; 847 segment_callback func; 848 void *closure; 849{ 850 vm_map_t map = &p->p_vmspace->vm_map; 851 vm_map_entry_t entry; 852 853 for (entry = map->header.next; entry != &map->header; 854 entry = entry->next) { 855 vm_object_t obj; 856 857 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) || 858 (entry->protection & (VM_PROT_READ|VM_PROT_WRITE)) != 859 (VM_PROT_READ|VM_PROT_WRITE)) 860 continue; 861 862 if ((obj = entry->object.vm_object) == NULL) 863 continue; 864 865 /* Find the deepest backing object. */ 866 while (obj->backing_object != NULL) 867 obj = obj->backing_object; 868 869 /* Ignore memory-mapped devices and such things. */ 870 if (obj->type != OBJT_DEFAULT && 871 obj->type != OBJT_SWAP && 872 obj->type != OBJT_VNODE) 873 continue; 874 875 (*func)(entry, closure); 876 } 877} 878 879/* 880 * Write the core file header to the file, including padding up to 881 * the page boundary. 882 */ 883static int 884elf_corehdr(p, vp, cred, numsegs, hdr, hdrsize) 885 struct proc *p; 886 struct vnode *vp; 887 struct ucred *cred; 888 int numsegs; 889 size_t hdrsize; 890 void *hdr; 891{ 892 size_t off; 893 prstatus_t status; 894 prfpregset_t fpregset; 895 prpsinfo_t psinfo; 896 897 /* Gather the information for the header. */ 898 bzero(&status, sizeof status); 899 status.pr_version = PRSTATUS_VERSION; 900 status.pr_statussz = sizeof(prstatus_t); 901 status.pr_gregsetsz = sizeof(gregset_t); 902 status.pr_fpregsetsz = sizeof(fpregset_t); 903 status.pr_osreldate = osreldate; 904 status.pr_cursig = p->p_sig; 905 status.pr_pid = p->p_pid; 906 fill_regs(p, &status.pr_reg); 907 908 fill_fpregs(p, &fpregset); 909 910 bzero(&psinfo, sizeof psinfo); 911 psinfo.pr_version = PRPSINFO_VERSION; 912 psinfo.pr_psinfosz = sizeof(prpsinfo_t); 913 strncpy(psinfo.pr_fname, p->p_comm, MAXCOMLEN); 914 /* XXX - We don't fill in the command line arguments properly yet. */ 915 strncpy(psinfo.pr_psargs, p->p_comm, PRARGSZ); 916 917 /* Fill in the header. */ 918 bzero(hdr, hdrsize); 919 off = 0; 920 elf_puthdr(p, hdr, &off, &status, &fpregset, &psinfo, numsegs); 921 922 /* Write it to the core file. */ 923 return vn_rdwr(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 924 UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, NULL, p); 925} 926 927static void 928elf_puthdr(struct proc *p, void *dst, size_t *off, const prstatus_t *status, 929 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs) 930{ 931 size_t ehoff; 932 size_t phoff; 933 size_t noteoff; 934 size_t notesz; 935 936 ehoff = *off; 937 *off += sizeof(Elf_Ehdr); 938 939 phoff = *off; 940 *off += (numsegs + 1) * sizeof(Elf_Phdr); 941 942 noteoff = *off; 943 elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status, 944 sizeof *status); 945 elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 946 sizeof *fpregset); 947 elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 948 sizeof *psinfo); 949 notesz = *off - noteoff; 950 951 /* Align up to a page boundary for the program segments. */ 952 *off = round_page(*off); 953 954 if (dst != NULL) { 955 Elf_Ehdr *ehdr; 956 Elf_Phdr *phdr; 957 struct phdr_closure phc; 958 959 /* 960 * Fill in the ELF header. 961 */ 962 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 963 ehdr->e_ident[EI_MAG0] = ELFMAG0; 964 ehdr->e_ident[EI_MAG1] = ELFMAG1; 965 ehdr->e_ident[EI_MAG2] = ELFMAG2; 966 ehdr->e_ident[EI_MAG3] = ELFMAG3; 967 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 968 ehdr->e_ident[EI_DATA] = ELF_DATA; 969 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 970 ehdr->e_ident[EI_PAD] = 0; 971 strncpy(ehdr->e_ident + EI_BRAND, "FreeBSD", 972 EI_NIDENT - EI_BRAND); 973 ehdr->e_type = ET_CORE; 974 ehdr->e_machine = ELF_ARCH; 975 ehdr->e_version = EV_CURRENT; 976 ehdr->e_entry = 0; 977 ehdr->e_phoff = phoff; 978 ehdr->e_flags = 0; 979 ehdr->e_ehsize = sizeof(Elf_Ehdr); 980 ehdr->e_phentsize = sizeof(Elf_Phdr); 981 ehdr->e_phnum = numsegs + 1; 982 ehdr->e_shentsize = sizeof(Elf_Shdr); 983 ehdr->e_shnum = 0; 984 ehdr->e_shstrndx = SHN_UNDEF; 985 986 /* 987 * Fill in the program header entries. 988 */ 989 phdr = (Elf_Phdr *)((char *)dst + phoff); 990 991 /* The note segement. */ 992 phdr->p_type = PT_NOTE; 993 phdr->p_offset = noteoff; 994 phdr->p_vaddr = 0; 995 phdr->p_paddr = 0; 996 phdr->p_filesz = notesz; 997 phdr->p_memsz = 0; 998 phdr->p_flags = 0; 999 phdr->p_align = 0; 1000 phdr++; 1001 1002 /* All the writable segments from the program. */ 1003 phc.phdr = phdr; 1004 phc.offset = *off; 1005 each_writable_segment(p, cb_put_phdr, &phc); 1006 } 1007} 1008 1009static void 1010elf_putnote(void *dst, size_t *off, const char *name, int type, 1011 const void *desc, size_t descsz) 1012{ 1013 Elf_Note note; 1014 1015 note.n_namesz = strlen(name) + 1; 1016 note.n_descsz = descsz; 1017 note.n_type = type; 1018 if (dst != NULL) 1019 bcopy(¬e, (char *)dst + *off, sizeof note); 1020 *off += sizeof note; 1021 if (dst != NULL) 1022 bcopy(name, (char *)dst + *off, note.n_namesz); 1023 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1024 if (dst != NULL) 1025 bcopy(desc, (char *)dst + *off, note.n_descsz); 1026 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1027} 1028 1029/* 1030 * Tell kern_execve.c about it, with a little help from the linker. 1031 */ 1032static struct execsw elf_execsw = {exec_elf_imgact, "ELF"}; 1033EXEC_SET(elf, elf_execsw); 1034