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