1 /*- 2 * Copyright (c) 2000 David O'Brien 3 * Copyright (c) 1995-1996 Søren 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 without 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: src/sys/kern/imgact_elf.c,v 1.73.2.13 2002/12/28 19:49:41 dillon Exp $ 31 */ 32 33 #include <sys/param.h> 34 #include <sys/exec.h> 35 #include <sys/fcntl.h> 36 #include <sys/file.h> 37 #include <sys/imgact.h> 38 #include <sys/imgact_elf.h> 39 #include <sys/kernel.h> 40 #include <sys/malloc.h> 41 #include <sys/mman.h> 42 #include <sys/systm.h> 43 #include <sys/proc.h> 44 #include <sys/nlookup.h> 45 #include <sys/pioctl.h> 46 #include <sys/procfs.h> 47 #include <sys/resourcevar.h> 48 #include <sys/signalvar.h> 49 #include <sys/stat.h> 50 #include <sys/syscall.h> 51 #include <sys/sysctl.h> 52 #include <sys/sysent.h> 53 #include <sys/vnode.h> 54 #include <sys/eventhandler.h> 55 56 #include <cpu/lwbuf.h> 57 58 #include <vm/vm.h> 59 #include <vm/vm_kern.h> 60 #include <vm/vm_param.h> 61 #include <vm/pmap.h> 62 #include <sys/lock.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 #include <sys/mount.h> 70 #include <sys/ckpt.h> 71 72 #define OLD_EI_BRAND 8 73 #define truncps(va,ps) ((va) & ~(ps - 1)) 74 #define aligned(a,t) (truncps((u_long)(a), sizeof(t)) == (u_long)(a)) 75 76 static int __elfN(check_header)(const Elf_Ehdr *hdr); 77 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 78 const char *interp, int32_t *osrel); 79 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 80 u_long *entry); 81 static int __elfN(load_section)(struct proc *p, 82 struct vmspace *vmspace, struct vnode *vp, 83 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 84 vm_prot_t prot); 85 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 86 static boolean_t __elfN(bsd_trans_osrel)(const Elf_Note *note, 87 int32_t *osrel); 88 static boolean_t __elfN(check_note)(struct image_params *imgp, 89 Elf_Brandnote *checknote, int32_t *osrel); 90 static vm_prot_t __elfN(trans_prot)(Elf_Word); 91 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 92 static boolean_t check_PT_NOTE(struct image_params *imgp, 93 Elf_Brandnote *checknote, int32_t *osrel, const Elf_Phdr * pnote); 94 static boolean_t extract_interpreter(struct image_params *imgp, 95 const Elf_Phdr *pinterpreter, char *data); 96 97 static int elf_legacy_coredump = 0; 98 static int __elfN(fallback_brand) = -1; 99 #if defined(__x86_64__) 100 SYSCTL_NODE(_kern, OID_AUTO, elf64, CTLFLAG_RW, 0, ""); 101 SYSCTL_INT(_debug, OID_AUTO, elf64_legacy_coredump, CTLFLAG_RW, 102 &elf_legacy_coredump, 0, "legacy coredump mode"); 103 SYSCTL_INT(_kern_elf64, OID_AUTO, fallback_brand, CTLFLAG_RW, 104 &elf64_fallback_brand, 0, "ELF64 brand of last resort"); 105 TUNABLE_INT("kern.elf64.fallback_brand", &elf64_fallback_brand); 106 #else /* i386 assumed */ 107 SYSCTL_NODE(_kern, OID_AUTO, elf32, CTLFLAG_RW, 0, ""); 108 SYSCTL_INT(_debug, OID_AUTO, elf32_legacy_coredump, CTLFLAG_RW, 109 &elf_legacy_coredump, 0, "legacy coredump mode"); 110 SYSCTL_INT(_kern_elf32, OID_AUTO, fallback_brand, CTLFLAG_RW, 111 &elf32_fallback_brand, 0, "ELF32 brand of last resort"); 112 TUNABLE_INT("kern.elf32.fallback_brand", &elf32_fallback_brand); 113 #endif 114 115 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 116 117 static const char DRAGONFLY_ABI_VENDOR[] = "DragonFly"; 118 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; 119 120 Elf_Brandnote __elfN(dragonfly_brandnote) = { 121 .hdr.n_namesz = sizeof(DRAGONFLY_ABI_VENDOR), 122 .hdr.n_descsz = sizeof(int32_t), 123 .hdr.n_type = 1, 124 .vendor = DRAGONFLY_ABI_VENDOR, 125 .flags = BN_TRANSLATE_OSREL, 126 .trans_osrel = __elfN(bsd_trans_osrel), 127 }; 128 129 Elf_Brandnote __elfN(freebsd_brandnote) = { 130 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 131 .hdr.n_descsz = sizeof(int32_t), 132 .hdr.n_type = 1, 133 .vendor = FREEBSD_ABI_VENDOR, 134 .flags = BN_TRANSLATE_OSREL, 135 .trans_osrel = __elfN(bsd_trans_osrel), 136 }; 137 138 int 139 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 140 { 141 int i; 142 143 for (i = 0; i < MAX_BRANDS; i++) { 144 if (elf_brand_list[i] == NULL) { 145 elf_brand_list[i] = entry; 146 break; 147 } 148 } 149 if (i == MAX_BRANDS) { 150 uprintf("WARNING: %s: could not insert brandinfo entry: %p\n", 151 __func__, entry); 152 return (-1); 153 } 154 return (0); 155 } 156 157 int 158 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 159 { 160 int i; 161 162 for (i = 0; i < MAX_BRANDS; i++) { 163 if (elf_brand_list[i] == entry) { 164 elf_brand_list[i] = NULL; 165 break; 166 } 167 } 168 if (i == MAX_BRANDS) 169 return (-1); 170 return (0); 171 } 172 173 /* 174 * Check if an elf brand is being used anywhere in the system. 175 * 176 * Used by the linux emulation module unloader. This isn't safe from 177 * races. 178 */ 179 struct elf_brand_inuse_info { 180 int rval; 181 Elf_Brandinfo *entry; 182 }; 183 184 static int elf_brand_inuse_callback(struct proc *p, void *data); 185 186 int 187 __elfN(brand_inuse)(Elf_Brandinfo *entry) 188 { 189 struct elf_brand_inuse_info info; 190 191 info.rval = FALSE; 192 info.entry = entry; 193 allproc_scan(elf_brand_inuse_callback, &info); 194 return (info.rval); 195 } 196 197 static 198 int 199 elf_brand_inuse_callback(struct proc *p, void *data) 200 { 201 struct elf_brand_inuse_info *info = data; 202 203 if (p->p_sysent == info->entry->sysvec) { 204 info->rval = TRUE; 205 return (-1); 206 } 207 return (0); 208 } 209 210 static int 211 __elfN(check_header)(const Elf_Ehdr *hdr) 212 { 213 Elf_Brandinfo *bi; 214 int i; 215 216 if (!IS_ELF(*hdr) || 217 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 218 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 219 hdr->e_ident[EI_VERSION] != EV_CURRENT || 220 hdr->e_phentsize != sizeof(Elf_Phdr) || 221 hdr->e_ehsize != sizeof(Elf_Ehdr) || 222 hdr->e_version != ELF_TARG_VER) 223 return (ENOEXEC); 224 225 /* 226 * Make sure we have at least one brand for this machine. 227 */ 228 229 for (i = 0; i < MAX_BRANDS; i++) { 230 bi = elf_brand_list[i]; 231 if (bi != NULL && bi->machine == hdr->e_machine) 232 break; 233 } 234 if (i == MAX_BRANDS) 235 return (ENOEXEC); 236 237 return (0); 238 } 239 240 static int 241 __elfN(load_section)(struct proc *p, struct vmspace *vmspace, struct vnode *vp, 242 vm_offset_t offset, caddr_t vmaddr, size_t memsz, 243 size_t filsz, vm_prot_t prot) 244 { 245 size_t map_len; 246 vm_offset_t map_addr; 247 int error, rv, cow; 248 int count; 249 int shared; 250 size_t copy_len; 251 vm_object_t object; 252 vm_offset_t file_addr; 253 254 object = vp->v_object; 255 error = 0; 256 257 /* 258 * In most cases we will be able to use a shared lock on the 259 * object we are inserting into the map. The lock will be 260 * upgraded in situations where new VM pages must be allocated. 261 */ 262 vm_object_hold_shared(object); 263 shared = 1; 264 265 /* 266 * It's necessary to fail if the filsz + offset taken from the 267 * header is greater than the actual file pager object's size. 268 * If we were to allow this, then the vm_map_find() below would 269 * walk right off the end of the file object and into the ether. 270 * 271 * While I'm here, might as well check for something else that 272 * is invalid: filsz cannot be greater than memsz. 273 */ 274 if ((off_t)filsz + offset > vp->v_filesize || filsz > memsz) { 275 uprintf("elf_load_section: truncated ELF file\n"); 276 vm_object_drop(object); 277 return (ENOEXEC); 278 } 279 280 map_addr = trunc_page((vm_offset_t)vmaddr); 281 file_addr = trunc_page(offset); 282 283 /* 284 * We have two choices. We can either clear the data in the last page 285 * of an oversized mapping, or we can start the anon mapping a page 286 * early and copy the initialized data into that first page. We 287 * choose the second.. 288 */ 289 if (memsz > filsz) 290 map_len = trunc_page(offset+filsz) - file_addr; 291 else 292 map_len = round_page(offset+filsz) - file_addr; 293 294 if (map_len != 0) { 295 vm_object_reference_locked(object); 296 297 /* cow flags: don't dump readonly sections in core */ 298 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT; 299 if ((prot & VM_PROT_WRITE) == 0) 300 cow |= MAP_DISABLE_COREDUMP; 301 if (shared == 0) 302 cow |= MAP_PREFAULT_RELOCK; 303 304 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 305 vm_map_lock(&vmspace->vm_map); 306 rv = vm_map_insert(&vmspace->vm_map, &count, 307 object, NULL, 308 file_addr, /* file offset */ 309 map_addr, /* virtual start */ 310 map_addr + map_len,/* virtual end */ 311 VM_MAPTYPE_NORMAL, 312 prot, VM_PROT_ALL, 313 cow); 314 vm_map_unlock(&vmspace->vm_map); 315 vm_map_entry_release(count); 316 317 /* 318 * NOTE: Object must have a hold ref when calling 319 * vm_object_deallocate(). 320 */ 321 if (rv != KERN_SUCCESS) { 322 vm_object_drop(object); 323 vm_object_deallocate(object); 324 return (EINVAL); 325 } 326 327 /* we can stop now if we've covered it all */ 328 if (memsz == filsz) { 329 vm_object_drop(object); 330 return (0); 331 } 332 } 333 334 /* 335 * We have to get the remaining bit of the file into the first part 336 * of the oversized map segment. This is normally because the .data 337 * segment in the file is extended to provide bss. It's a neat idea 338 * to try and save a page, but it's a pain in the behind to implement. 339 */ 340 copy_len = (offset + filsz) - trunc_page(offset + filsz); 341 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 342 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 343 344 /* This had damn well better be true! */ 345 if (map_len != 0) { 346 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 347 vm_map_lock(&vmspace->vm_map); 348 rv = vm_map_insert(&vmspace->vm_map, &count, 349 NULL, NULL, 350 0, 351 map_addr, 352 map_addr + map_len, 353 VM_MAPTYPE_NORMAL, 354 VM_PROT_ALL, VM_PROT_ALL, 355 0); 356 vm_map_unlock(&vmspace->vm_map); 357 vm_map_entry_release(count); 358 if (rv != KERN_SUCCESS) { 359 vm_object_drop(object); 360 return (EINVAL); 361 } 362 } 363 364 if (copy_len != 0) { 365 struct lwbuf *lwb; 366 struct lwbuf lwb_cache; 367 vm_page_t m; 368 369 m = vm_fault_object_page(object, trunc_page(offset + filsz), 370 VM_PROT_READ, 0, &shared, &error); 371 vm_object_drop(object); 372 if (m) { 373 lwb = lwbuf_alloc(m, &lwb_cache); 374 error = copyout((caddr_t)lwbuf_kva(lwb), 375 (caddr_t)map_addr, copy_len); 376 lwbuf_free(lwb); 377 vm_page_unhold(m); 378 } 379 } else { 380 vm_object_drop(object); 381 } 382 383 /* 384 * set it to the specified protection 385 */ 386 if (error == 0) { 387 vm_map_protect(&vmspace->vm_map, 388 map_addr, map_addr + map_len, 389 prot, FALSE); 390 } 391 return (error); 392 } 393 394 /* 395 * Load the file "file" into memory. It may be either a shared object 396 * or an executable. 397 * 398 * The "addr" reference parameter is in/out. On entry, it specifies 399 * the address where a shared object should be loaded. If the file is 400 * an executable, this value is ignored. On exit, "addr" specifies 401 * where the file was actually loaded. 402 * 403 * The "entry" reference parameter is out only. On exit, it specifies 404 * the entry point for the loaded file. 405 */ 406 static int 407 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, u_long *entry) 408 { 409 struct { 410 struct nlookupdata nd; 411 struct vattr attr; 412 struct image_params image_params; 413 } *tempdata; 414 const Elf_Ehdr *hdr = NULL; 415 const Elf_Phdr *phdr = NULL; 416 struct nlookupdata *nd; 417 struct vmspace *vmspace = p->p_vmspace; 418 struct vattr *attr; 419 struct image_params *imgp; 420 struct mount *topmnt; 421 vm_prot_t prot; 422 u_long rbase; 423 u_long base_addr = 0; 424 int error, i, numsegs; 425 426 tempdata = kmalloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 427 nd = &tempdata->nd; 428 attr = &tempdata->attr; 429 imgp = &tempdata->image_params; 430 431 /* 432 * Initialize part of the common data 433 */ 434 imgp->proc = p; 435 imgp->attr = attr; 436 imgp->firstpage = NULL; 437 imgp->image_header = NULL; 438 imgp->vp = NULL; 439 440 error = nlookup_init(nd, file, UIO_SYSSPACE, NLC_FOLLOW); 441 if (error == 0) 442 error = nlookup(nd); 443 if (error == 0) 444 error = cache_vget(&nd->nl_nch, nd->nl_cred, 445 LK_SHARED, &imgp->vp); 446 topmnt = nd->nl_nch.mount; 447 nlookup_done(nd); 448 if (error) 449 goto fail; 450 451 /* 452 * Check permissions, modes, uid, etc on the file, and "open" it. 453 */ 454 error = exec_check_permissions(imgp, topmnt); 455 if (error) { 456 vn_unlock(imgp->vp); 457 goto fail; 458 } 459 460 error = exec_map_first_page(imgp); 461 /* 462 * Also make certain that the interpreter stays the same, so set 463 * its VTEXT flag, too. 464 */ 465 if (error == 0) 466 vsetflags(imgp->vp, VTEXT); 467 vn_unlock(imgp->vp); 468 if (error) 469 goto fail; 470 471 hdr = (const Elf_Ehdr *)imgp->image_header; 472 if ((error = __elfN(check_header)(hdr)) != 0) 473 goto fail; 474 if (hdr->e_type == ET_DYN) 475 rbase = *addr; 476 else if (hdr->e_type == ET_EXEC) 477 rbase = 0; 478 else { 479 error = ENOEXEC; 480 goto fail; 481 } 482 483 /* Only support headers that fit within first page for now */ 484 /* (multiplication of two Elf_Half fields will not overflow) */ 485 if ((hdr->e_phoff > PAGE_SIZE) || 486 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 487 error = ENOEXEC; 488 goto fail; 489 } 490 491 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 492 if (!aligned(phdr, Elf_Addr)) { 493 error = ENOEXEC; 494 goto fail; 495 } 496 497 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 498 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 499 /* Loadable segment */ 500 prot = __elfN(trans_prot)(phdr[i].p_flags); 501 error = __elfN(load_section)( 502 p, vmspace, imgp->vp, 503 phdr[i].p_offset, 504 (caddr_t)phdr[i].p_vaddr + 505 rbase, 506 phdr[i].p_memsz, 507 phdr[i].p_filesz, prot); 508 if (error != 0) 509 goto fail; 510 /* 511 * Establish the base address if this is the 512 * first segment. 513 */ 514 if (numsegs == 0) 515 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 516 numsegs++; 517 } 518 } 519 *addr = base_addr; 520 *entry = (unsigned long)hdr->e_entry + rbase; 521 522 fail: 523 if (imgp->firstpage) 524 exec_unmap_first_page(imgp); 525 if (imgp->vp) { 526 vrele(imgp->vp); 527 imgp->vp = NULL; 528 } 529 kfree(tempdata, M_TEMP); 530 531 return (error); 532 } 533 534 static Elf_Brandinfo * 535 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 536 int32_t *osrel) 537 { 538 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 539 Elf_Brandinfo *bi; 540 boolean_t ret; 541 int i; 542 543 /* We support four types of branding -- (1) the ELF EI_OSABI field 544 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 545 * branding within the ELF header, (3) path of the `interp_path' field, 546 * and (4) the ".note.ABI-tag" ELF section. 547 */ 548 549 /* Look for an ".note.ABI-tag" ELF section */ 550 for (i = 0; i < MAX_BRANDS; i++) { 551 bi = elf_brand_list[i]; 552 553 if (bi == NULL) 554 continue; 555 if (hdr->e_machine == bi->machine && (bi->flags & 556 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 557 ret = __elfN(check_note)(imgp, bi->brand_note, osrel); 558 if (ret) 559 return (bi); 560 } 561 } 562 563 /* If the executable has a brand, search for it in the brand list. */ 564 for (i = 0; i < MAX_BRANDS; i++) { 565 bi = elf_brand_list[i]; 566 567 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 568 continue; 569 if (hdr->e_machine == bi->machine && 570 (hdr->e_ident[EI_OSABI] == bi->brand || 571 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 572 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 573 return (bi); 574 } 575 576 /* Lacking a known brand, search for a recognized interpreter. */ 577 if (interp != NULL) { 578 for (i = 0; i < MAX_BRANDS; i++) { 579 bi = elf_brand_list[i]; 580 581 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 582 continue; 583 if (hdr->e_machine == bi->machine && 584 strcmp(interp, bi->interp_path) == 0) 585 return (bi); 586 } 587 } 588 589 /* Lacking a recognized interpreter, try the default brand */ 590 for (i = 0; i < MAX_BRANDS; i++) { 591 bi = elf_brand_list[i]; 592 593 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 594 continue; 595 if (hdr->e_machine == bi->machine && 596 __elfN(fallback_brand) == bi->brand) 597 return (bi); 598 } 599 return (NULL); 600 } 601 602 static int 603 __CONCAT(exec_,__elfN(imgact))(struct image_params *imgp) 604 { 605 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header; 606 const Elf_Phdr *phdr; 607 Elf_Auxargs *elf_auxargs; 608 struct vmspace *vmspace; 609 vm_prot_t prot; 610 u_long text_size = 0, data_size = 0, total_size = 0; 611 u_long text_addr = 0, data_addr = 0; 612 u_long seg_size, seg_addr; 613 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0; 614 int32_t osrel = 0; 615 int error = 0, i, n; 616 boolean_t failure; 617 char *interp = NULL; 618 const char *newinterp = NULL; 619 Elf_Brandinfo *brand_info; 620 char *path; 621 622 /* 623 * Do we have a valid ELF header ? 624 * 625 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later if a particular 626 * brand doesn't support it. Both DragonFly platforms do by default. 627 */ 628 if (__elfN(check_header)(hdr) != 0 || 629 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 630 return (-1); 631 632 /* 633 * From here on down, we return an errno, not -1, as we've 634 * detected an ELF file. 635 */ 636 637 if ((hdr->e_phoff > PAGE_SIZE) || 638 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 639 /* Only support headers in first page for now */ 640 return (ENOEXEC); 641 } 642 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 643 if (!aligned(phdr, Elf_Addr)) 644 return (ENOEXEC); 645 n = 0; 646 baddr = 0; 647 for (i = 0; i < hdr->e_phnum; i++) { 648 if (phdr[i].p_type == PT_LOAD) { 649 if (n == 0) 650 baddr = phdr[i].p_vaddr; 651 n++; 652 continue; 653 } 654 if (phdr[i].p_type == PT_INTERP) { 655 /* 656 * If interp is already defined there are more than 657 * one PT_INTERP program headers present. Take only 658 * the first one and ignore the rest. 659 */ 660 if (interp != NULL) 661 continue; 662 663 if (phdr[i].p_filesz == 0 || 664 phdr[i].p_filesz > PAGE_SIZE || 665 phdr[i].p_filesz > MAXPATHLEN) 666 return (ENOEXEC); 667 668 interp = kmalloc(phdr[i].p_filesz, M_TEMP, M_WAITOK); 669 failure = extract_interpreter(imgp, &phdr[i], interp); 670 if (failure) { 671 kfree(interp, M_TEMP); 672 return (ENOEXEC); 673 } 674 continue; 675 } 676 } 677 678 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel); 679 if (brand_info == NULL) { 680 uprintf("ELF binary type \"%u\" not known.\n", 681 hdr->e_ident[EI_OSABI]); 682 if (interp != NULL) 683 kfree(interp, M_TEMP); 684 return (ENOEXEC); 685 } 686 if (hdr->e_type == ET_DYN) { 687 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 688 if (interp != NULL) 689 kfree(interp, M_TEMP); 690 return (ENOEXEC); 691 } 692 /* 693 * Honour the base load address from the dso if it is 694 * non-zero for some reason. 695 */ 696 if (baddr == 0) 697 et_dyn_addr = ET_DYN_LOAD_ADDR; 698 else 699 et_dyn_addr = 0; 700 } else 701 et_dyn_addr = 0; 702 703 if (interp != NULL && brand_info->interp_newpath != NULL) 704 newinterp = brand_info->interp_newpath; 705 706 exec_new_vmspace(imgp, NULL); 707 708 /* 709 * Yeah, I'm paranoid. There is every reason in the world to get 710 * VTEXT now since from here on out, there are places we can have 711 * a context switch. Better safe than sorry; I really don't want 712 * the file to change while it's being loaded. 713 */ 714 vsetflags(imgp->vp, VTEXT); 715 716 vmspace = imgp->proc->p_vmspace; 717 718 for (i = 0; i < hdr->e_phnum; i++) { 719 switch (phdr[i].p_type) { 720 case PT_LOAD: /* Loadable segment */ 721 if (phdr[i].p_memsz == 0) 722 break; 723 prot = __elfN(trans_prot)(phdr[i].p_flags); 724 725 if ((error = __elfN(load_section)( 726 imgp->proc, 727 vmspace, 728 imgp->vp, 729 phdr[i].p_offset, 730 (caddr_t)phdr[i].p_vaddr + et_dyn_addr, 731 phdr[i].p_memsz, 732 phdr[i].p_filesz, 733 prot)) != 0) { 734 if (interp != NULL) 735 kfree (interp, M_TEMP); 736 return (error); 737 } 738 739 /* 740 * If this segment contains the program headers, 741 * remember their virtual address for the AT_PHDR 742 * aux entry. Static binaries don't usually include 743 * a PT_PHDR entry. 744 */ 745 if (phdr[i].p_offset == 0 && 746 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 747 <= phdr[i].p_filesz) 748 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 749 et_dyn_addr; 750 751 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 752 seg_size = round_page(phdr[i].p_memsz + 753 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 754 755 /* 756 * Is this .text or .data? We can't use 757 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 758 * alpha terribly and possibly does other bad 759 * things so we stick to the old way of figuring 760 * it out: If the segment contains the program 761 * entry point, it's a text segment, otherwise it 762 * is a data segment. 763 * 764 * Note that obreak() assumes that data_addr + 765 * data_size == end of data load area, and the ELF 766 * file format expects segments to be sorted by 767 * address. If multiple data segments exist, the 768 * last one will be used. 769 */ 770 if (hdr->e_entry >= phdr[i].p_vaddr && 771 hdr->e_entry < (phdr[i].p_vaddr + 772 phdr[i].p_memsz)) { 773 text_size = seg_size; 774 text_addr = seg_addr; 775 entry = (u_long)hdr->e_entry + et_dyn_addr; 776 } else { 777 data_size = seg_size; 778 data_addr = seg_addr; 779 } 780 total_size += seg_size; 781 782 /* 783 * Check limits. It should be safe to check the 784 * limits after loading the segment since we do 785 * not actually fault in all the segment's pages. 786 */ 787 if (data_size > 788 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur || 789 text_size > maxtsiz || 790 total_size > 791 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) { 792 if (interp != NULL) 793 kfree(interp, M_TEMP); 794 error = ENOMEM; 795 return (error); 796 } 797 break; 798 case PT_PHDR: /* Program header table info */ 799 proghdr = phdr[i].p_vaddr + et_dyn_addr; 800 break; 801 default: 802 break; 803 } 804 } 805 806 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 807 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 808 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 809 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 810 811 addr = ELF_RTLD_ADDR(vmspace); 812 813 imgp->entry_addr = entry; 814 815 imgp->proc->p_sysent = brand_info->sysvec; 816 EVENTHANDLER_INVOKE(process_exec, imgp); 817 818 if (interp != NULL) { 819 int have_interp = FALSE; 820 if (brand_info->emul_path != NULL && 821 brand_info->emul_path[0] != '\0') { 822 path = kmalloc(MAXPATHLEN, M_TEMP, M_WAITOK); 823 ksnprintf(path, MAXPATHLEN, "%s%s", 824 brand_info->emul_path, interp); 825 error = __elfN(load_file)(imgp->proc, path, &addr, 826 &imgp->entry_addr); 827 kfree(path, M_TEMP); 828 if (error == 0) 829 have_interp = TRUE; 830 } 831 if (!have_interp && newinterp != NULL) { 832 error = __elfN(load_file)(imgp->proc, newinterp, 833 &addr, &imgp->entry_addr); 834 if (error == 0) 835 have_interp = TRUE; 836 } 837 if (!have_interp) { 838 error = __elfN(load_file)(imgp->proc, interp, &addr, 839 &imgp->entry_addr); 840 } 841 if (error != 0) { 842 uprintf("ELF interpreter %s not found\n", interp); 843 kfree(interp, M_TEMP); 844 return (error); 845 } 846 kfree(interp, M_TEMP); 847 } else 848 addr = et_dyn_addr; 849 850 /* 851 * Construct auxargs table (used by the fixup routine) 852 */ 853 elf_auxargs = kmalloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 854 elf_auxargs->execfd = -1; 855 elf_auxargs->phdr = proghdr; 856 elf_auxargs->phent = hdr->e_phentsize; 857 elf_auxargs->phnum = hdr->e_phnum; 858 elf_auxargs->pagesz = PAGE_SIZE; 859 elf_auxargs->base = addr; 860 elf_auxargs->flags = 0; 861 elf_auxargs->entry = entry; 862 863 imgp->auxargs = elf_auxargs; 864 imgp->interpreted = 0; 865 imgp->proc->p_osrel = osrel; 866 867 return (error); 868 } 869 870 int 871 __elfN(dragonfly_fixup)(register_t **stack_base, struct image_params *imgp) 872 { 873 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 874 Elf_Addr *base; 875 Elf_Addr *pos; 876 877 base = (Elf_Addr *)*stack_base; 878 pos = base + (imgp->args->argc + imgp->args->envc + 2); 879 880 if (args->execfd != -1) 881 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 882 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 883 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 884 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 885 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 886 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 887 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 888 AUXARGS_ENTRY(pos, AT_BASE, args->base); 889 if (imgp->execpathp != 0) 890 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 891 AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate); 892 AUXARGS_ENTRY(pos, AT_NULL, 0); 893 894 kfree(imgp->auxargs, M_TEMP); 895 imgp->auxargs = NULL; 896 897 base--; 898 suword(base, (long)imgp->args->argc); 899 *stack_base = (register_t *)base; 900 return (0); 901 } 902 903 /* 904 * Code for generating ELF core dumps. 905 */ 906 907 typedef int (*segment_callback)(vm_map_entry_t, void *); 908 909 /* Closure for cb_put_phdr(). */ 910 struct phdr_closure { 911 Elf_Phdr *phdr; /* Program header to fill in (incremented) */ 912 Elf_Phdr *phdr_max; /* Pointer bound for error check */ 913 Elf_Off offset; /* Offset of segment in core file */ 914 }; 915 916 /* Closure for cb_size_segment(). */ 917 struct sseg_closure { 918 int count; /* Count of writable segments. */ 919 size_t vsize; /* Total size of all writable segments. */ 920 }; 921 922 /* Closure for cb_put_fp(). */ 923 struct fp_closure { 924 struct vn_hdr *vnh; 925 struct vn_hdr *vnh_max; 926 int count; 927 struct stat *sb; 928 }; 929 930 typedef struct elf_buf { 931 char *buf; 932 size_t off; 933 size_t off_max; 934 } *elf_buf_t; 935 936 static void *target_reserve(elf_buf_t target, size_t bytes, int *error); 937 938 static int cb_put_phdr (vm_map_entry_t, void *); 939 static int cb_size_segment (vm_map_entry_t, void *); 940 static int cb_fpcount_segment(vm_map_entry_t, void *); 941 static int cb_put_fp(vm_map_entry_t, void *); 942 943 944 static int each_segment (struct proc *, segment_callback, void *, int); 945 static int __elfN(corehdr)(struct lwp *, int, struct file *, struct ucred *, 946 int, elf_buf_t); 947 enum putmode { WRITE, DRYRUN }; 948 static int __elfN(puthdr)(struct lwp *, elf_buf_t, int sig, enum putmode, 949 int, struct file *); 950 static int elf_putallnotes(struct lwp *, elf_buf_t, int, enum putmode); 951 static int __elfN(putnote)(elf_buf_t, const char *, int, const void *, size_t); 952 953 static int elf_putsigs(struct lwp *, elf_buf_t); 954 static int elf_puttextvp(struct proc *, elf_buf_t); 955 static int elf_putfiles(struct proc *, elf_buf_t, struct file *); 956 957 int 958 __elfN(coredump)(struct lwp *lp, int sig, struct vnode *vp, off_t limit) 959 { 960 struct file *fp; 961 int error; 962 963 if ((error = falloc(NULL, &fp, NULL)) != 0) 964 return (error); 965 fsetcred(fp, lp->lwp_proc->p_ucred); 966 967 /* 968 * XXX fixme. 969 */ 970 fp->f_type = DTYPE_VNODE; 971 fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW; 972 fp->f_ops = &vnode_fileops; 973 fp->f_data = vp; 974 975 error = generic_elf_coredump(lp, sig, fp, limit); 976 977 fp->f_type = 0; 978 fp->f_flag = 0; 979 fp->f_ops = &badfileops; 980 fp->f_data = NULL; 981 fdrop(fp); 982 return (error); 983 } 984 985 int 986 generic_elf_coredump(struct lwp *lp, int sig, struct file *fp, off_t limit) 987 { 988 struct proc *p = lp->lwp_proc; 989 struct ucred *cred = p->p_ucred; 990 int error = 0; 991 struct sseg_closure seginfo; 992 struct elf_buf target; 993 994 if (!fp) 995 kprintf("can't dump core - null fp\n"); 996 997 /* 998 * Size the program segments 999 */ 1000 seginfo.count = 0; 1001 seginfo.vsize = 0; 1002 each_segment(p, cb_size_segment, &seginfo, 1); 1003 1004 /* 1005 * Calculate the size of the core file header area by making 1006 * a dry run of generating it. Nothing is written, but the 1007 * size is calculated. 1008 */ 1009 bzero(&target, sizeof(target)); 1010 __elfN(puthdr)(lp, &target, sig, DRYRUN, seginfo.count, fp); 1011 1012 if (target.off + seginfo.vsize >= limit) 1013 return (EFAULT); 1014 1015 /* 1016 * Allocate memory for building the header, fill it up, 1017 * and write it out. 1018 */ 1019 target.off_max = target.off; 1020 target.off = 0; 1021 target.buf = kmalloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO); 1022 1023 error = __elfN(corehdr)(lp, sig, fp, cred, seginfo.count, &target); 1024 1025 /* Write the contents of all of the writable segments. */ 1026 if (error == 0) { 1027 Elf_Phdr *php; 1028 int i; 1029 ssize_t nbytes; 1030 1031 php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1; 1032 for (i = 0; i < seginfo.count; i++) { 1033 error = fp_write(fp, (caddr_t)php->p_vaddr, 1034 php->p_filesz, &nbytes, UIO_USERSPACE); 1035 if (error != 0) 1036 break; 1037 php++; 1038 } 1039 } 1040 kfree(target.buf, M_TEMP); 1041 1042 return (error); 1043 } 1044 1045 /* 1046 * A callback for each_segment() to write out the segment's 1047 * program header entry. 1048 */ 1049 static int 1050 cb_put_phdr(vm_map_entry_t entry, void *closure) 1051 { 1052 struct phdr_closure *phc = closure; 1053 Elf_Phdr *phdr = phc->phdr; 1054 1055 if (phc->phdr == phc->phdr_max) 1056 return (EINVAL); 1057 1058 phc->offset = round_page(phc->offset); 1059 1060 phdr->p_type = PT_LOAD; 1061 phdr->p_offset = phc->offset; 1062 phdr->p_vaddr = entry->start; 1063 phdr->p_paddr = 0; 1064 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1065 phdr->p_align = PAGE_SIZE; 1066 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1067 1068 phc->offset += phdr->p_filesz; 1069 ++phc->phdr; 1070 return (0); 1071 } 1072 1073 /* 1074 * A callback for each_writable_segment() to gather information about 1075 * the number of segments and their total size. 1076 */ 1077 static int 1078 cb_size_segment(vm_map_entry_t entry, void *closure) 1079 { 1080 struct sseg_closure *ssc = closure; 1081 1082 ++ssc->count; 1083 ssc->vsize += entry->end - entry->start; 1084 return (0); 1085 } 1086 1087 /* 1088 * A callback for each_segment() to gather information about 1089 * the number of text segments. 1090 */ 1091 static int 1092 cb_fpcount_segment(vm_map_entry_t entry, void *closure) 1093 { 1094 int *count = closure; 1095 struct vnode *vp; 1096 1097 if (entry->object.vm_object->type == OBJT_VNODE) { 1098 vp = (struct vnode *)entry->object.vm_object->handle; 1099 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp) 1100 return (0); 1101 ++*count; 1102 } 1103 return (0); 1104 } 1105 1106 static int 1107 cb_put_fp(vm_map_entry_t entry, void *closure) 1108 { 1109 struct fp_closure *fpc = closure; 1110 struct vn_hdr *vnh = fpc->vnh; 1111 Elf_Phdr *phdr = &vnh->vnh_phdr; 1112 struct vnode *vp; 1113 int error; 1114 1115 /* 1116 * If an entry represents a vnode then write out a file handle. 1117 * 1118 * If we are checkpointing a checkpoint-restored program we do 1119 * NOT record the filehandle for the old checkpoint vnode (which 1120 * is mapped all over the place). Instead we rely on the fact 1121 * that a checkpoint-restored program does not mmap() the checkpt 1122 * vnode NOCORE, so its contents will be written out to the 1123 * new checkpoint file. This is necessary because the 'old' 1124 * checkpoint file is typically destroyed when a new one is created 1125 * and thus cannot be used to restore the new checkpoint. 1126 * 1127 * Theoretically we could create a chain of checkpoint files and 1128 * operate the checkpointing operation kinda like an incremental 1129 * checkpoint, but a checkpoint restore would then likely wind up 1130 * referencing many prior checkpoint files and that is a bit over 1131 * the top for the purpose of the checkpoint API. 1132 */ 1133 if (entry->object.vm_object->type == OBJT_VNODE) { 1134 vp = (struct vnode *)entry->object.vm_object->handle; 1135 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp) 1136 return (0); 1137 if (vnh == fpc->vnh_max) 1138 return (EINVAL); 1139 1140 if (vp->v_mount) 1141 vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1142 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid); 1143 if (error) { 1144 char *freepath, *fullpath; 1145 1146 if (vn_fullpath(curproc, vp, &fullpath, &freepath, 0)) { 1147 kprintf("Warning: coredump, error %d: cannot store file handle for vnode %p\n", error, vp); 1148 } else { 1149 kprintf("Warning: coredump, error %d: cannot store file handle for %s\n", error, fullpath); 1150 kfree(freepath, M_TEMP); 1151 } 1152 error = 0; 1153 } 1154 1155 phdr->p_type = PT_LOAD; 1156 phdr->p_offset = 0; /* not written to core */ 1157 phdr->p_vaddr = entry->start; 1158 phdr->p_paddr = 0; 1159 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1160 phdr->p_align = PAGE_SIZE; 1161 phdr->p_flags = 0; 1162 if (entry->protection & VM_PROT_READ) 1163 phdr->p_flags |= PF_R; 1164 if (entry->protection & VM_PROT_WRITE) 1165 phdr->p_flags |= PF_W; 1166 if (entry->protection & VM_PROT_EXECUTE) 1167 phdr->p_flags |= PF_X; 1168 ++fpc->vnh; 1169 ++fpc->count; 1170 } 1171 return (0); 1172 } 1173 1174 /* 1175 * For each writable segment in the process's memory map, call the given 1176 * function with a pointer to the map entry and some arbitrary 1177 * caller-supplied data. 1178 */ 1179 static int 1180 each_segment(struct proc *p, segment_callback func, void *closure, int writable) 1181 { 1182 int error = 0; 1183 vm_map_t map = &p->p_vmspace->vm_map; 1184 vm_map_entry_t entry; 1185 1186 for (entry = map->header.next; error == 0 && entry != &map->header; 1187 entry = entry->next) { 1188 vm_object_t obj; 1189 vm_object_t lobj; 1190 vm_object_t tobj; 1191 1192 /* 1193 * Don't dump inaccessible mappings, deal with legacy 1194 * coredump mode. 1195 * 1196 * Note that read-only segments related to the elf binary 1197 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1198 * need to arbitrarily ignore such segments. 1199 */ 1200 if (elf_legacy_coredump) { 1201 if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW) 1202 continue; 1203 } else { 1204 if (writable && (entry->protection & VM_PROT_ALL) == 0) 1205 continue; 1206 } 1207 1208 /* 1209 * Dont include memory segment in the coredump if 1210 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1211 * madvise(2). 1212 * 1213 * Currently we only dump normal VM object maps. We do 1214 * not dump submaps or virtual page tables. 1215 */ 1216 if (writable && (entry->eflags & MAP_ENTRY_NOCOREDUMP)) 1217 continue; 1218 if (entry->maptype != VM_MAPTYPE_NORMAL) 1219 continue; 1220 if ((obj = entry->object.vm_object) == NULL) 1221 continue; 1222 1223 /* 1224 * Find the bottom-most object, leaving the base object 1225 * and the bottom-most object held (but only one hold 1226 * if they happen to be the same). 1227 */ 1228 vm_object_hold_shared(obj); 1229 1230 lobj = obj; 1231 while (lobj && (tobj = lobj->backing_object) != NULL) { 1232 KKASSERT(tobj != obj); 1233 vm_object_hold_shared(tobj); 1234 if (tobj == lobj->backing_object) { 1235 if (lobj != obj) { 1236 vm_object_lock_swap(); 1237 vm_object_drop(lobj); 1238 } 1239 lobj = tobj; 1240 } else { 1241 vm_object_drop(tobj); 1242 } 1243 } 1244 1245 /* 1246 * The callback only applies to default, swap, or vnode 1247 * objects. Other types of objects such as memory-mapped 1248 * devices are ignored. 1249 */ 1250 if (lobj->type == OBJT_DEFAULT || lobj->type == OBJT_SWAP || 1251 lobj->type == OBJT_VNODE) { 1252 error = (*func)(entry, closure); 1253 } 1254 if (lobj != obj) 1255 vm_object_drop(lobj); 1256 vm_object_drop(obj); 1257 } 1258 return (error); 1259 } 1260 1261 static 1262 void * 1263 target_reserve(elf_buf_t target, size_t bytes, int *error) 1264 { 1265 void *res = NULL; 1266 1267 if (target->buf) { 1268 if (target->off + bytes > target->off_max) 1269 *error = EINVAL; 1270 else 1271 res = target->buf + target->off; 1272 } 1273 target->off += bytes; 1274 return (res); 1275 } 1276 1277 /* 1278 * Write the core file header to the file, including padding up to 1279 * the page boundary. 1280 */ 1281 static int 1282 __elfN(corehdr)(struct lwp *lp, int sig, struct file *fp, struct ucred *cred, 1283 int numsegs, elf_buf_t target) 1284 { 1285 int error; 1286 ssize_t nbytes; 1287 1288 /* 1289 * Fill in the header. The fp is passed so we can detect and flag 1290 * a checkpoint file pointer within the core file itself, because 1291 * it may not be restored from the same file handle. 1292 */ 1293 error = __elfN(puthdr)(lp, target, sig, WRITE, numsegs, fp); 1294 1295 /* Write it to the core file. */ 1296 if (error == 0) { 1297 error = fp_write(fp, target->buf, target->off, &nbytes, 1298 UIO_SYSSPACE); 1299 } 1300 return (error); 1301 } 1302 1303 static int 1304 __elfN(puthdr)(struct lwp *lp, elf_buf_t target, int sig, enum putmode mode, 1305 int numsegs, struct file *fp) 1306 { 1307 struct proc *p = lp->lwp_proc; 1308 int error = 0; 1309 size_t phoff; 1310 size_t noteoff; 1311 size_t notesz; 1312 Elf_Ehdr *ehdr; 1313 Elf_Phdr *phdr; 1314 1315 ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error); 1316 1317 phoff = target->off; 1318 phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error); 1319 1320 noteoff = target->off; 1321 if (error == 0) 1322 elf_putallnotes(lp, target, sig, mode); 1323 notesz = target->off - noteoff; 1324 1325 /* 1326 * put extra cruft for dumping process state here 1327 * - we really want it be before all the program 1328 * mappings 1329 * - we just need to update the offset accordingly 1330 * and GDB will be none the wiser. 1331 */ 1332 if (error == 0) 1333 error = elf_puttextvp(p, target); 1334 if (error == 0) 1335 error = elf_putsigs(lp, target); 1336 if (error == 0) 1337 error = elf_putfiles(p, target, fp); 1338 1339 /* 1340 * Align up to a page boundary for the program segments. The 1341 * actual data will be written to the outptu file, not to elf_buf_t, 1342 * so we do not have to do any further bounds checking. 1343 */ 1344 target->off = round_page(target->off); 1345 if (error == 0 && ehdr != NULL) { 1346 /* 1347 * Fill in the ELF header. 1348 */ 1349 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1350 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1351 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1352 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1353 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1354 ehdr->e_ident[EI_DATA] = ELF_DATA; 1355 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1356 ehdr->e_ident[EI_OSABI] = ELFOSABI_NONE; 1357 ehdr->e_ident[EI_ABIVERSION] = 0; 1358 ehdr->e_ident[EI_PAD] = 0; 1359 ehdr->e_type = ET_CORE; 1360 ehdr->e_machine = ELF_ARCH; 1361 ehdr->e_version = EV_CURRENT; 1362 ehdr->e_entry = 0; 1363 ehdr->e_phoff = phoff; 1364 ehdr->e_flags = 0; 1365 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1366 ehdr->e_phentsize = sizeof(Elf_Phdr); 1367 ehdr->e_phnum = numsegs + 1; 1368 ehdr->e_shentsize = sizeof(Elf_Shdr); 1369 ehdr->e_shnum = 0; 1370 ehdr->e_shstrndx = SHN_UNDEF; 1371 } 1372 if (error == 0 && phdr != NULL) { 1373 /* 1374 * Fill in the program header entries. 1375 */ 1376 struct phdr_closure phc; 1377 1378 /* The note segement. */ 1379 phdr->p_type = PT_NOTE; 1380 phdr->p_offset = noteoff; 1381 phdr->p_vaddr = 0; 1382 phdr->p_paddr = 0; 1383 phdr->p_filesz = notesz; 1384 phdr->p_memsz = 0; 1385 phdr->p_flags = 0; 1386 phdr->p_align = 0; 1387 ++phdr; 1388 1389 /* All the writable segments from the program. */ 1390 phc.phdr = phdr; 1391 phc.phdr_max = phdr + numsegs; 1392 phc.offset = target->off; 1393 each_segment(p, cb_put_phdr, &phc, 1); 1394 } 1395 return (error); 1396 } 1397 1398 /* 1399 * Append core dump notes to target ELF buffer or simply update target size 1400 * if dryrun selected. 1401 */ 1402 static int 1403 elf_putallnotes(struct lwp *corelp, elf_buf_t target, int sig, 1404 enum putmode mode) 1405 { 1406 struct proc *p = corelp->lwp_proc; 1407 int error; 1408 struct { 1409 prstatus_t status; 1410 prfpregset_t fpregs; 1411 prpsinfo_t psinfo; 1412 } *tmpdata; 1413 prstatus_t *status; 1414 prfpregset_t *fpregs; 1415 prpsinfo_t *psinfo; 1416 struct lwp *lp; 1417 1418 /* 1419 * Allocate temporary storage for notes on heap to avoid stack overflow. 1420 */ 1421 if (mode != DRYRUN) { 1422 tmpdata = kmalloc(sizeof(*tmpdata), M_TEMP, M_ZERO | M_WAITOK); 1423 status = &tmpdata->status; 1424 fpregs = &tmpdata->fpregs; 1425 psinfo = &tmpdata->psinfo; 1426 } else { 1427 tmpdata = NULL; 1428 status = NULL; 1429 fpregs = NULL; 1430 psinfo = NULL; 1431 } 1432 1433 /* 1434 * Append LWP-agnostic note. 1435 */ 1436 if (mode != DRYRUN) { 1437 psinfo->pr_version = PRPSINFO_VERSION; 1438 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 1439 strlcpy(psinfo->pr_fname, p->p_comm, 1440 sizeof(psinfo->pr_fname)); 1441 /* 1442 * XXX - We don't fill in the command line arguments 1443 * properly yet. 1444 */ 1445 strlcpy(psinfo->pr_psargs, p->p_comm, 1446 sizeof(psinfo->pr_psargs)); 1447 } 1448 error = 1449 __elfN(putnote)(target, "CORE", NT_PRPSINFO, psinfo, sizeof *psinfo); 1450 if (error) 1451 goto exit; 1452 1453 /* 1454 * Append first note for LWP that triggered core so that it is 1455 * the selected one when the debugger starts. 1456 */ 1457 if (mode != DRYRUN) { 1458 status->pr_version = PRSTATUS_VERSION; 1459 status->pr_statussz = sizeof(prstatus_t); 1460 status->pr_gregsetsz = sizeof(gregset_t); 1461 status->pr_fpregsetsz = sizeof(fpregset_t); 1462 status->pr_osreldate = osreldate; 1463 status->pr_cursig = sig; 1464 /* 1465 * XXX GDB needs unique pr_pid for each LWP and does not 1466 * not support pr_pid==0 but lwp_tid can be 0, so hack unique 1467 * value. 1468 */ 1469 status->pr_pid = corelp->lwp_tid; 1470 fill_regs(corelp, &status->pr_reg); 1471 fill_fpregs(corelp, fpregs); 1472 } 1473 error = 1474 __elfN(putnote)(target, "CORE", NT_PRSTATUS, status, sizeof *status); 1475 if (error) 1476 goto exit; 1477 error = 1478 __elfN(putnote)(target, "CORE", NT_FPREGSET, fpregs, sizeof *fpregs); 1479 if (error) 1480 goto exit; 1481 1482 /* 1483 * Then append notes for other LWPs. 1484 */ 1485 FOREACH_LWP_IN_PROC(lp, p) { 1486 if (lp == corelp) 1487 continue; 1488 /* skip lwps being created */ 1489 if (lp->lwp_thread == NULL) 1490 continue; 1491 if (mode != DRYRUN) { 1492 status->pr_pid = lp->lwp_tid; 1493 fill_regs(lp, &status->pr_reg); 1494 fill_fpregs(lp, fpregs); 1495 } 1496 error = __elfN(putnote)(target, "CORE", NT_PRSTATUS, 1497 status, sizeof *status); 1498 if (error) 1499 goto exit; 1500 error = __elfN(putnote)(target, "CORE", NT_FPREGSET, 1501 fpregs, sizeof *fpregs); 1502 if (error) 1503 goto exit; 1504 } 1505 1506 exit: 1507 if (tmpdata != NULL) 1508 kfree(tmpdata, M_TEMP); 1509 return (error); 1510 } 1511 1512 /* 1513 * Generate a note sub-structure. 1514 * 1515 * NOTE: 4-byte alignment. 1516 */ 1517 static int 1518 __elfN(putnote)(elf_buf_t target, const char *name, int type, 1519 const void *desc, size_t descsz) 1520 { 1521 int error = 0; 1522 char *dst; 1523 Elf_Note note; 1524 1525 note.n_namesz = strlen(name) + 1; 1526 note.n_descsz = descsz; 1527 note.n_type = type; 1528 dst = target_reserve(target, sizeof(note), &error); 1529 if (dst != NULL) 1530 bcopy(¬e, dst, sizeof note); 1531 dst = target_reserve(target, note.n_namesz, &error); 1532 if (dst != NULL) 1533 bcopy(name, dst, note.n_namesz); 1534 target->off = roundup2(target->off, sizeof(Elf_Word)); 1535 dst = target_reserve(target, note.n_descsz, &error); 1536 if (dst != NULL) 1537 bcopy(desc, dst, note.n_descsz); 1538 target->off = roundup2(target->off, sizeof(Elf_Word)); 1539 return (error); 1540 } 1541 1542 1543 static int 1544 elf_putsigs(struct lwp *lp, elf_buf_t target) 1545 { 1546 /* XXX lwp handle more than one lwp */ 1547 struct proc *p = lp->lwp_proc; 1548 int error = 0; 1549 struct ckpt_siginfo *csi; 1550 1551 csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error); 1552 if (csi) { 1553 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo); 1554 bcopy(p->p_sigacts, &csi->csi_sigacts, sizeof(*p->p_sigacts)); 1555 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval)); 1556 bcopy(&lp->lwp_sigmask, &csi->csi_sigmask, 1557 sizeof(sigset_t)); 1558 csi->csi_sigparent = p->p_sigparent; 1559 } 1560 return (error); 1561 } 1562 1563 static int 1564 elf_putfiles(struct proc *p, elf_buf_t target, struct file *ckfp) 1565 { 1566 int error = 0; 1567 int i; 1568 struct ckpt_filehdr *cfh = NULL; 1569 struct ckpt_fileinfo *cfi; 1570 struct file *fp; 1571 struct vnode *vp; 1572 /* 1573 * the duplicated loop is gross, but it was the only way 1574 * to eliminate uninitialized variable warnings 1575 */ 1576 cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error); 1577 if (cfh) { 1578 cfh->cfh_nfiles = 0; 1579 } 1580 1581 /* 1582 * ignore STDIN/STDERR/STDOUT. 1583 */ 1584 for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) { 1585 fp = holdfp(p->p_fd, i, -1); 1586 if (fp == NULL) 1587 continue; 1588 /* 1589 * XXX Only checkpoint vnodes for now. 1590 */ 1591 if (fp->f_type != DTYPE_VNODE) { 1592 fdrop(fp); 1593 continue; 1594 } 1595 cfi = target_reserve(target, sizeof(struct ckpt_fileinfo), 1596 &error); 1597 if (cfi == NULL) { 1598 fdrop(fp); 1599 continue; 1600 } 1601 cfi->cfi_index = -1; 1602 cfi->cfi_type = fp->f_type; 1603 cfi->cfi_flags = fp->f_flag; 1604 cfi->cfi_offset = fp->f_offset; 1605 cfi->cfi_ckflags = 0; 1606 1607 if (fp == ckfp) 1608 cfi->cfi_ckflags |= CKFIF_ISCKPTFD; 1609 /* f_count and f_msgcount should not be saved/restored */ 1610 /* XXX save cred info */ 1611 1612 switch(fp->f_type) { 1613 case DTYPE_VNODE: 1614 vp = (struct vnode *)fp->f_data; 1615 /* 1616 * it looks like a bug in ptrace is marking 1617 * a non-vnode as a vnode - until we find the 1618 * root cause this will at least prevent 1619 * further panics from truss 1620 */ 1621 if (vp == NULL || vp->v_mount == NULL) 1622 break; 1623 cfh->cfh_nfiles++; 1624 cfi->cfi_index = i; 1625 cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1626 error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid); 1627 break; 1628 default: 1629 break; 1630 } 1631 fdrop(fp); 1632 } 1633 return (error); 1634 } 1635 1636 static int 1637 elf_puttextvp(struct proc *p, elf_buf_t target) 1638 { 1639 int error = 0; 1640 int *vn_count; 1641 struct fp_closure fpc; 1642 struct ckpt_vminfo *vminfo; 1643 1644 vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error); 1645 if (vminfo != NULL) { 1646 vminfo->cvm_dsize = p->p_vmspace->vm_dsize; 1647 vminfo->cvm_tsize = p->p_vmspace->vm_tsize; 1648 vminfo->cvm_daddr = p->p_vmspace->vm_daddr; 1649 vminfo->cvm_taddr = p->p_vmspace->vm_taddr; 1650 } 1651 1652 fpc.count = 0; 1653 vn_count = target_reserve(target, sizeof(int), &error); 1654 if (target->buf != NULL) { 1655 fpc.vnh = (struct vn_hdr *)(target->buf + target->off); 1656 fpc.vnh_max = fpc.vnh + 1657 (target->off_max - target->off) / sizeof(struct vn_hdr); 1658 error = each_segment(p, cb_put_fp, &fpc, 0); 1659 if (vn_count) 1660 *vn_count = fpc.count; 1661 } else { 1662 error = each_segment(p, cb_fpcount_segment, &fpc.count, 0); 1663 } 1664 target->off += fpc.count * sizeof(struct vn_hdr); 1665 return (error); 1666 } 1667 1668 /* 1669 * Try to find the appropriate ABI-note section for checknote, 1670 * The entire image is searched if necessary, not only the first page. 1671 */ 1672 static boolean_t 1673 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 1674 int32_t *osrel) 1675 { 1676 boolean_t valid_note_found; 1677 const Elf_Phdr *phdr, *pnote; 1678 const Elf_Ehdr *hdr; 1679 int i; 1680 1681 valid_note_found = FALSE; 1682 hdr = (const Elf_Ehdr *)imgp->image_header; 1683 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1684 1685 for (i = 0; i < hdr->e_phnum; i++) { 1686 if (phdr[i].p_type == PT_NOTE) { 1687 pnote = &phdr[i]; 1688 valid_note_found = check_PT_NOTE (imgp, checknote, 1689 osrel, pnote); 1690 if (valid_note_found) 1691 break; 1692 } 1693 } 1694 return valid_note_found; 1695 } 1696 1697 /* 1698 * Be careful not to create new overflow conditions when checking 1699 * for overflow. 1700 */ 1701 static boolean_t 1702 note_overflow(const Elf_Note *note, size_t maxsize) 1703 { 1704 if (sizeof(*note) > maxsize) 1705 return TRUE; 1706 if (note->n_namesz > maxsize - sizeof(*note)) 1707 return TRUE; 1708 return FALSE; 1709 } 1710 1711 static boolean_t 1712 hdr_overflow(__ElfN(Off) off_beg, __ElfN(Size) size) 1713 { 1714 __ElfN(Off) off_end; 1715 1716 off_end = off_beg + size; 1717 if (off_end < off_beg) 1718 return TRUE; 1719 return FALSE; 1720 } 1721 1722 static boolean_t 1723 check_PT_NOTE(struct image_params *imgp, Elf_Brandnote *checknote, 1724 int32_t *osrel, const Elf_Phdr * pnote) 1725 { 1726 boolean_t limited_to_first_page; 1727 boolean_t found = FALSE; 1728 const Elf_Note *note, *note0, *note_end; 1729 const char *note_name; 1730 __ElfN(Off) noteloc, firstloc; 1731 __ElfN(Size) notesz, firstlen, endbyte; 1732 struct lwbuf *lwb; 1733 struct lwbuf lwb_cache; 1734 const char *page; 1735 char *data = NULL; 1736 int n; 1737 1738 if (hdr_overflow(pnote->p_offset, pnote->p_filesz)) 1739 return (FALSE); 1740 notesz = pnote->p_filesz; 1741 noteloc = pnote->p_offset; 1742 endbyte = noteloc + notesz; 1743 limited_to_first_page = noteloc < PAGE_SIZE && endbyte < PAGE_SIZE; 1744 1745 if (limited_to_first_page) { 1746 note = (const Elf_Note *)(imgp->image_header + noteloc); 1747 note_end = (const Elf_Note *)(imgp->image_header + endbyte); 1748 note0 = note; 1749 } else { 1750 firstloc = noteloc & PAGE_MASK; 1751 firstlen = PAGE_SIZE - firstloc; 1752 if (notesz < sizeof(Elf_Note) || notesz > PAGE_SIZE) 1753 return (FALSE); 1754 1755 lwb = &lwb_cache; 1756 if (exec_map_page(imgp, noteloc >> PAGE_SHIFT, &lwb, &page)) 1757 return (FALSE); 1758 if (firstlen < notesz) { /* crosses page boundary */ 1759 data = kmalloc(notesz, M_TEMP, M_WAITOK); 1760 bcopy(page + firstloc, data, firstlen); 1761 1762 exec_unmap_page(lwb); 1763 lwb = &lwb_cache; 1764 if (exec_map_page(imgp, (noteloc >> PAGE_SHIFT) + 1, 1765 &lwb, &page)) { 1766 kfree(data, M_TEMP); 1767 return (FALSE); 1768 } 1769 bcopy(page, data + firstlen, notesz - firstlen); 1770 note = note0 = (const Elf_Note *)(data); 1771 note_end = (const Elf_Note *)(data + notesz); 1772 } else { 1773 note = note0 = (const Elf_Note *)(page + firstloc); 1774 note_end = (const Elf_Note *)(page + firstloc + 1775 firstlen); 1776 } 1777 } 1778 1779 for (n = 0; n < 100 && note >= note0 && note < note_end; n++) { 1780 if (!aligned(note, Elf32_Addr)) 1781 break; 1782 if (note_overflow(note, (const char *)note_end - 1783 (const char *)note)) { 1784 break; 1785 } 1786 note_name = (const char *)(note + 1); 1787 1788 if (note->n_namesz == checknote->hdr.n_namesz 1789 && note->n_descsz == checknote->hdr.n_descsz 1790 && note->n_type == checknote->hdr.n_type 1791 && (strncmp(checknote->vendor, note_name, 1792 checknote->hdr.n_namesz) == 0)) { 1793 /* Fetch osreldata from ABI.note-tag */ 1794 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 1795 checknote->trans_osrel != NULL) 1796 checknote->trans_osrel(note, osrel); 1797 found = TRUE; 1798 break; 1799 } 1800 note = (const Elf_Note *)((const char *)(note + 1) + 1801 roundup2(note->n_namesz, sizeof(Elf32_Addr)) + 1802 roundup2(note->n_descsz, sizeof(Elf32_Addr))); 1803 } 1804 1805 if (!limited_to_first_page) { 1806 if (data != NULL) 1807 kfree(data, M_TEMP); 1808 exec_unmap_page(lwb); 1809 } 1810 return (found); 1811 } 1812 1813 /* 1814 * The interpreter program header may be located beyond the first page, so 1815 * regardless of its location, a copy of the interpreter path is created so 1816 * that it may be safely referenced by the calling function in all case. The 1817 * memory is allocated by calling function, and the copying is done here. 1818 */ 1819 static boolean_t 1820 extract_interpreter(struct image_params *imgp, const Elf_Phdr *pinterpreter, 1821 char *data) 1822 { 1823 boolean_t limited_to_first_page; 1824 const boolean_t result_success = FALSE; 1825 const boolean_t result_failure = TRUE; 1826 __ElfN(Off) pathloc, firstloc; 1827 __ElfN(Size) pathsz, firstlen, endbyte; 1828 struct lwbuf *lwb; 1829 struct lwbuf lwb_cache; 1830 const char *page; 1831 1832 if (hdr_overflow(pinterpreter->p_offset, pinterpreter->p_filesz)) 1833 return (result_failure); 1834 pathsz = pinterpreter->p_filesz; 1835 pathloc = pinterpreter->p_offset; 1836 endbyte = pathloc + pathsz; 1837 1838 limited_to_first_page = pathloc < PAGE_SIZE && endbyte < PAGE_SIZE; 1839 if (limited_to_first_page) { 1840 bcopy(imgp->image_header + pathloc, data, pathsz); 1841 return (result_success); 1842 } 1843 1844 firstloc = pathloc & PAGE_MASK; 1845 firstlen = PAGE_SIZE - firstloc; 1846 1847 lwb = &lwb_cache; 1848 if (exec_map_page(imgp, pathloc >> PAGE_SHIFT, &lwb, &page)) 1849 return (result_failure); 1850 1851 if (firstlen < pathsz) { /* crosses page boundary */ 1852 bcopy(page + firstloc, data, firstlen); 1853 1854 exec_unmap_page(lwb); 1855 lwb = &lwb_cache; 1856 if (exec_map_page(imgp, (pathloc >> PAGE_SHIFT) + 1, &lwb, 1857 &page)) 1858 return (result_failure); 1859 bcopy(page, data + firstlen, pathsz - firstlen); 1860 } else 1861 bcopy(page + firstloc, data, pathsz); 1862 1863 exec_unmap_page(lwb); 1864 return (result_success); 1865 } 1866 1867 static boolean_t 1868 __elfN(bsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 1869 { 1870 uintptr_t p; 1871 1872 p = (uintptr_t)(note + 1); 1873 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 1874 *osrel = *(const int32_t *)(p); 1875 1876 return (TRUE); 1877 } 1878 1879 /* 1880 * Tell kern_execve.c about it, with a little help from the linker. 1881 */ 1882 #if defined(__x86_64__) 1883 static struct execsw elf_execsw = {exec_elf64_imgact, "ELF64"}; 1884 EXEC_SET_ORDERED(elf64, elf_execsw, SI_ORDER_FIRST); 1885 #else /* i386 assumed */ 1886 static struct execsw elf_execsw = {exec_elf32_imgact, "ELF32"}; 1887 EXEC_SET_ORDERED(elf32, elf_execsw, SI_ORDER_FIRST); 1888 #endif 1889 1890 static vm_prot_t 1891 __elfN(trans_prot)(Elf_Word flags) 1892 { 1893 vm_prot_t prot; 1894 1895 prot = 0; 1896 if (flags & PF_X) 1897 prot |= VM_PROT_EXECUTE; 1898 if (flags & PF_W) 1899 prot |= VM_PROT_WRITE; 1900 if (flags & PF_R) 1901 prot |= VM_PROT_READ; 1902 return (prot); 1903 } 1904 1905 static Elf_Word 1906 __elfN(untrans_prot)(vm_prot_t prot) 1907 { 1908 Elf_Word flags; 1909 1910 flags = 0; 1911 if (prot & VM_PROT_EXECUTE) 1912 flags |= PF_X; 1913 if (prot & VM_PROT_READ) 1914 flags |= PF_R; 1915 if (prot & VM_PROT_WRITE) 1916 flags |= PF_W; 1917 return (flags); 1918 } 1919