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