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