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 static u_long pie_base_hint(struct proc *p); 97 98 static int elf_legacy_coredump = 0; 99 static int __elfN(fallback_brand) = -1; 100 static int elf_pie_base_mmap = 0; 101 #if defined(__x86_64__) 102 SYSCTL_NODE(_kern, OID_AUTO, elf64, CTLFLAG_RW, 0, ""); 103 SYSCTL_INT(_debug, OID_AUTO, elf64_legacy_coredump, CTLFLAG_RW, 104 &elf_legacy_coredump, 0, "legacy coredump mode"); 105 SYSCTL_INT(_kern_elf64, OID_AUTO, fallback_brand, CTLFLAG_RW, 106 &elf64_fallback_brand, 0, "ELF64 brand of last resort"); 107 TUNABLE_INT("kern.elf64.fallback_brand", &elf64_fallback_brand); 108 SYSCTL_INT(_kern_elf64, OID_AUTO, pie_base_mmap, CTLFLAG_RW, 109 &elf_pie_base_mmap, 0, 110 "choose a base address for PIE as if it is mapped with mmap()"); 111 TUNABLE_INT("kern.elf64.pie_base_mmap", &elf_pie_base_mmap); 112 #else /* i386 assumed */ 113 SYSCTL_NODE(_kern, OID_AUTO, elf32, CTLFLAG_RW, 0, ""); 114 SYSCTL_INT(_debug, OID_AUTO, elf32_legacy_coredump, CTLFLAG_RW, 115 &elf_legacy_coredump, 0, "legacy coredump mode"); 116 SYSCTL_INT(_kern_elf32, OID_AUTO, fallback_brand, CTLFLAG_RW, 117 &elf32_fallback_brand, 0, "ELF32 brand of last resort"); 118 TUNABLE_INT("kern.elf32.fallback_brand", &elf32_fallback_brand); 119 SYSCTL_INT(_kern_elf32, OID_AUTO, pie_base_mmap, CTLFLAG_RW, 120 &elf_pie_base_mmap, 0, 121 "choose a base address for PIE as if it is mapped with mmap()"); 122 TUNABLE_INT("kern.elf32.pie_base_mmap", &elf_pie_base_mmap); 123 #endif 124 125 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 126 127 static const char DRAGONFLY_ABI_VENDOR[] = "DragonFly"; 128 129 Elf_Brandnote __elfN(dragonfly_brandnote) = { 130 .hdr.n_namesz = sizeof(DRAGONFLY_ABI_VENDOR), 131 .hdr.n_descsz = sizeof(int32_t), 132 .hdr.n_type = 1, 133 .vendor = DRAGONFLY_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, 0); 194 return (info.rval); 195 } 196 197 static 198 int 199 elf_brand_inuse_callback(struct proc *p, void *data) 200 { 201 struct elf_brand_inuse_info *info = data; 202 203 if (p->p_sysent == info->entry->sysvec) { 204 info->rval = TRUE; 205 return (-1); 206 } 207 return (0); 208 } 209 210 static int 211 __elfN(check_header)(const Elf_Ehdr *hdr) 212 { 213 Elf_Brandinfo *bi; 214 int i; 215 216 if (!IS_ELF(*hdr) || 217 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 218 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 219 hdr->e_ident[EI_VERSION] != EV_CURRENT || 220 hdr->e_phentsize != sizeof(Elf_Phdr) || 221 hdr->e_ehsize != sizeof(Elf_Ehdr) || 222 hdr->e_version != ELF_TARG_VER) 223 return (ENOEXEC); 224 225 /* 226 * Make sure we have at least one brand for this machine. 227 */ 228 229 for (i = 0; i < MAX_BRANDS; i++) { 230 bi = elf_brand_list[i]; 231 if (bi != NULL && bi->machine == hdr->e_machine) 232 break; 233 } 234 if (i == MAX_BRANDS) 235 return (ENOEXEC); 236 237 return (0); 238 } 239 240 static int 241 __elfN(load_section)(struct proc *p, struct vmspace *vmspace, struct vnode *vp, 242 vm_offset_t offset, caddr_t vmaddr, size_t memsz, 243 size_t filsz, vm_prot_t prot) 244 { 245 size_t map_len; 246 vm_offset_t map_addr; 247 int error, rv, cow; 248 int count; 249 int shared; 250 size_t copy_len; 251 vm_object_t object; 252 vm_offset_t file_addr; 253 254 object = vp->v_object; 255 error = 0; 256 257 /* 258 * In most cases we will be able to use a shared lock on the 259 * object we are inserting into the map. The lock will be 260 * upgraded in situations where new VM pages must be allocated. 261 */ 262 vm_object_hold_shared(object); 263 shared = 1; 264 265 /* 266 * It's necessary to fail if the filsz + offset taken from the 267 * header is greater than the actual file pager object's size. 268 * If we were to allow this, then the vm_map_find() below would 269 * walk right off the end of the file object and into the ether. 270 * 271 * While I'm here, might as well check for something else that 272 * is invalid: filsz cannot be greater than memsz. 273 */ 274 if ((off_t)filsz + offset > vp->v_filesize || filsz > memsz) { 275 uprintf("elf_load_section: truncated ELF file\n"); 276 vm_object_drop(object); 277 return (ENOEXEC); 278 } 279 280 map_addr = trunc_page((vm_offset_t)vmaddr); 281 file_addr = trunc_page(offset); 282 283 /* 284 * We have two choices. We can either clear the data in the last page 285 * of an oversized mapping, or we can start the anon mapping a page 286 * early and copy the initialized data into that first page. We 287 * choose the second.. 288 */ 289 if (memsz > filsz) 290 map_len = trunc_page(offset+filsz) - file_addr; 291 else 292 map_len = round_page(offset+filsz) - file_addr; 293 294 if (map_len != 0) { 295 vm_object_reference_locked(object); 296 297 /* cow flags: don't dump readonly sections in core */ 298 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT; 299 if ((prot & VM_PROT_WRITE) == 0) 300 cow |= MAP_DISABLE_COREDUMP; 301 if (shared == 0) 302 cow |= MAP_PREFAULT_RELOCK; 303 304 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 305 vm_map_lock(&vmspace->vm_map); 306 rv = vm_map_insert(&vmspace->vm_map, &count, 307 object, NULL, 308 file_addr, /* file offset */ 309 map_addr, /* virtual start */ 310 map_addr + map_len,/* virtual end */ 311 VM_MAPTYPE_NORMAL, 312 VM_SUBSYS_IMGACT, 313 prot, VM_PROT_ALL, cow); 314 vm_map_unlock(&vmspace->vm_map); 315 vm_map_entry_release(count); 316 317 /* 318 * NOTE: Object must have a hold ref when calling 319 * vm_object_deallocate(). 320 */ 321 if (rv != KERN_SUCCESS) { 322 vm_object_drop(object); 323 vm_object_deallocate(object); 324 return (EINVAL); 325 } 326 327 /* we can stop now if we've covered it all */ 328 if (memsz == filsz) { 329 vm_object_drop(object); 330 return (0); 331 } 332 } 333 334 /* 335 * We have to get the remaining bit of the file into the first part 336 * of the oversized map segment. This is normally because the .data 337 * segment in the file is extended to provide bss. It's a neat idea 338 * to try and save a page, but it's a pain in the behind to implement. 339 */ 340 copy_len = (offset + filsz) - trunc_page(offset + filsz); 341 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 342 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 343 344 /* This had damn well better be true! */ 345 if (map_len != 0) { 346 count = vm_map_entry_reserve(MAP_RESERVE_COUNT); 347 vm_map_lock(&vmspace->vm_map); 348 rv = vm_map_insert(&vmspace->vm_map, &count, 349 NULL, NULL, 350 0, 351 map_addr, 352 map_addr + map_len, 353 VM_MAPTYPE_NORMAL, 354 VM_SUBSYS_IMGACT, 355 VM_PROT_ALL, VM_PROT_ALL, 0); 356 vm_map_unlock(&vmspace->vm_map); 357 vm_map_entry_release(count); 358 if (rv != KERN_SUCCESS) { 359 vm_object_drop(object); 360 return (EINVAL); 361 } 362 } 363 364 if (copy_len != 0) { 365 struct lwbuf *lwb; 366 struct lwbuf lwb_cache; 367 vm_page_t m; 368 369 m = vm_fault_object_page(object, trunc_page(offset + filsz), 370 VM_PROT_READ, 0, &shared, &error); 371 vm_object_drop(object); 372 if (m) { 373 lwb = lwbuf_alloc(m, &lwb_cache); 374 error = copyout((caddr_t)lwbuf_kva(lwb), 375 (caddr_t)map_addr, copy_len); 376 lwbuf_free(lwb); 377 vm_page_unhold(m); 378 } 379 } else { 380 vm_object_drop(object); 381 } 382 383 /* 384 * set it to the specified protection 385 */ 386 if (error == 0) { 387 vm_map_protect(&vmspace->vm_map, 388 map_addr, map_addr + map_len, 389 prot, FALSE); 390 } 391 return (error); 392 } 393 394 /* 395 * Load the file "file" into memory. It may be either a shared object 396 * or an executable. 397 * 398 * The "addr" reference parameter is in/out. On entry, it specifies 399 * the address where a shared object should be loaded. If the file is 400 * an executable, this value is ignored. On exit, "addr" specifies 401 * where the file was actually loaded. 402 * 403 * The "entry" reference parameter is out only. On exit, it specifies 404 * the entry point for the loaded file. 405 */ 406 static int 407 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, u_long *entry) 408 { 409 struct { 410 struct nlookupdata nd; 411 struct vattr attr; 412 struct image_params image_params; 413 } *tempdata; 414 const Elf_Ehdr *hdr = NULL; 415 const Elf_Phdr *phdr = NULL; 416 struct nlookupdata *nd; 417 struct vmspace *vmspace = p->p_vmspace; 418 struct vattr *attr; 419 struct image_params *imgp; 420 struct mount *topmnt; 421 vm_prot_t prot; 422 u_long rbase; 423 u_long base_addr = 0; 424 int error, i, numsegs; 425 426 tempdata = kmalloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 427 nd = &tempdata->nd; 428 attr = &tempdata->attr; 429 imgp = &tempdata->image_params; 430 431 /* 432 * Initialize part of the common data 433 */ 434 imgp->proc = p; 435 imgp->attr = attr; 436 imgp->firstpage = NULL; 437 imgp->image_header = NULL; 438 imgp->vp = NULL; 439 440 error = nlookup_init(nd, file, UIO_SYSSPACE, NLC_FOLLOW); 441 if (error == 0) 442 error = nlookup(nd); 443 if (error == 0) 444 error = cache_vget(&nd->nl_nch, nd->nl_cred, 445 LK_SHARED, &imgp->vp); 446 topmnt = nd->nl_nch.mount; 447 nlookup_done(nd); 448 if (error) 449 goto fail; 450 451 /* 452 * Check permissions, modes, uid, etc on the file, and "open" it. 453 */ 454 error = exec_check_permissions(imgp, topmnt); 455 if (error) { 456 vn_unlock(imgp->vp); 457 goto fail; 458 } 459 460 error = exec_map_first_page(imgp); 461 /* 462 * Also make certain that the interpreter stays the same, so set 463 * its VTEXT flag, too. 464 */ 465 if (error == 0) 466 vsetflags(imgp->vp, VTEXT); 467 vn_unlock(imgp->vp); 468 if (error) 469 goto fail; 470 471 hdr = (const Elf_Ehdr *)imgp->image_header; 472 if ((error = __elfN(check_header)(hdr)) != 0) 473 goto fail; 474 if (hdr->e_type == ET_DYN) 475 rbase = *addr; 476 else if (hdr->e_type == ET_EXEC) 477 rbase = 0; 478 else { 479 error = ENOEXEC; 480 goto fail; 481 } 482 483 /* Only support headers that fit within first page for now */ 484 /* (multiplication of two Elf_Half fields will not overflow) */ 485 if ((hdr->e_phoff > PAGE_SIZE) || 486 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 487 error = ENOEXEC; 488 goto fail; 489 } 490 491 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 492 if (!aligned(phdr, Elf_Addr)) { 493 error = ENOEXEC; 494 goto fail; 495 } 496 497 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 498 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 499 /* Loadable segment */ 500 prot = __elfN(trans_prot)(phdr[i].p_flags); 501 error = __elfN(load_section)( 502 p, vmspace, imgp->vp, 503 phdr[i].p_offset, 504 (caddr_t)phdr[i].p_vaddr + 505 rbase, 506 phdr[i].p_memsz, 507 phdr[i].p_filesz, prot); 508 if (error != 0) 509 goto fail; 510 /* 511 * Establish the base address if this is the 512 * first segment. 513 */ 514 if (numsegs == 0) 515 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 516 numsegs++; 517 } 518 } 519 *addr = base_addr; 520 *entry = (unsigned long)hdr->e_entry + rbase; 521 522 fail: 523 if (imgp->firstpage) 524 exec_unmap_first_page(imgp); 525 if (imgp->vp) { 526 vrele(imgp->vp); 527 imgp->vp = NULL; 528 } 529 kfree(tempdata, M_TEMP); 530 531 return (error); 532 } 533 534 static Elf_Brandinfo * 535 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 536 int32_t *osrel) 537 { 538 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 539 Elf_Brandinfo *bi; 540 boolean_t ret; 541 int i; 542 543 /* We support four types of branding -- (1) the ELF EI_OSABI field 544 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 545 * branding within the ELF header, (3) path of the `interp_path' field, 546 * and (4) the ".note.ABI-tag" ELF section. 547 */ 548 549 /* Look for an ".note.ABI-tag" ELF section */ 550 for (i = 0; i < MAX_BRANDS; i++) { 551 bi = elf_brand_list[i]; 552 553 if (bi == NULL) 554 continue; 555 if (hdr->e_machine == bi->machine && (bi->flags & 556 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 557 ret = __elfN(check_note)(imgp, bi->brand_note, osrel); 558 if (ret) 559 return (bi); 560 } 561 } 562 563 /* If the executable has a brand, search for it in the brand list. */ 564 for (i = 0; i < MAX_BRANDS; i++) { 565 bi = elf_brand_list[i]; 566 567 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 568 continue; 569 if (hdr->e_machine == bi->machine && 570 (hdr->e_ident[EI_OSABI] == bi->brand || 571 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 572 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 573 return (bi); 574 } 575 576 /* Lacking a known brand, search for a recognized interpreter. */ 577 if (interp != NULL) { 578 for (i = 0; i < MAX_BRANDS; i++) { 579 bi = elf_brand_list[i]; 580 581 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 582 continue; 583 if (hdr->e_machine == bi->machine && 584 strcmp(interp, bi->interp_path) == 0) 585 return (bi); 586 } 587 } 588 589 /* Lacking a recognized interpreter, try the default brand */ 590 for (i = 0; i < MAX_BRANDS; i++) { 591 bi = elf_brand_list[i]; 592 593 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 594 continue; 595 if (hdr->e_machine == bi->machine && 596 __elfN(fallback_brand) == bi->brand) 597 return (bi); 598 } 599 return (NULL); 600 } 601 602 static int 603 __CONCAT(exec_,__elfN(imgact))(struct image_params *imgp) 604 { 605 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header; 606 const Elf_Phdr *phdr; 607 Elf_Auxargs *elf_auxargs; 608 struct vmspace *vmspace; 609 vm_prot_t prot; 610 u_long text_size = 0, data_size = 0, total_size = 0; 611 u_long text_addr = 0, data_addr = 0; 612 u_long seg_size, seg_addr; 613 u_long addr, baddr, et_dyn_addr = 0, entry = 0, proghdr = 0; 614 int32_t osrel = 0; 615 int error = 0, i, n; 616 boolean_t failure; 617 char *interp = NULL; 618 const char *newinterp = NULL; 619 Elf_Brandinfo *brand_info; 620 char *path; 621 622 /* 623 * Do we have a valid ELF header ? 624 * 625 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later if a particular 626 * brand doesn't support it. Both DragonFly platforms do by default. 627 */ 628 if (__elfN(check_header)(hdr) != 0 || 629 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 630 return (-1); 631 632 /* 633 * From here on down, we return an errno, not -1, as we've 634 * detected an ELF file. 635 */ 636 637 if ((hdr->e_phoff > PAGE_SIZE) || 638 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 639 /* Only support headers in first page for now */ 640 return (ENOEXEC); 641 } 642 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 643 if (!aligned(phdr, Elf_Addr)) 644 return (ENOEXEC); 645 n = 0; 646 baddr = 0; 647 for (i = 0; i < hdr->e_phnum; i++) { 648 if (phdr[i].p_type == PT_LOAD) { 649 if (n == 0) 650 baddr = phdr[i].p_vaddr; 651 n++; 652 continue; 653 } 654 if (phdr[i].p_type == PT_INTERP) { 655 /* 656 * If interp is already defined there are more than 657 * one PT_INTERP program headers present. Take only 658 * the first one and ignore the rest. 659 */ 660 if (interp != NULL) 661 continue; 662 663 if (phdr[i].p_filesz == 0 || 664 phdr[i].p_filesz > PAGE_SIZE || 665 phdr[i].p_filesz > MAXPATHLEN) 666 return (ENOEXEC); 667 668 interp = kmalloc(phdr[i].p_filesz, M_TEMP, M_WAITOK); 669 failure = extract_interpreter(imgp, &phdr[i], interp); 670 if (failure) { 671 kfree(interp, M_TEMP); 672 return (ENOEXEC); 673 } 674 continue; 675 } 676 } 677 678 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel); 679 if (brand_info == NULL) { 680 uprintf("ELF binary type \"%u\" not known.\n", 681 hdr->e_ident[EI_OSABI]); 682 if (interp != NULL) 683 kfree(interp, M_TEMP); 684 return (ENOEXEC); 685 } 686 if (hdr->e_type == ET_DYN) { 687 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 688 if (interp != NULL) 689 kfree(interp, M_TEMP); 690 return (ENOEXEC); 691 } 692 /* 693 * If p_vaddr field of PT_LOAD program header is zero and type of an 694 * executale is ET_DYN, then it must be a position independent 695 * executable (PIE). In this case the system needs to pick a base 696 * address for us. Set et_dyn_addr to non-zero and choose the actual 697 * address when we are ready. 698 */ 699 if (baddr == 0) 700 et_dyn_addr = 1; 701 } 702 703 if (interp != NULL && brand_info->interp_newpath != NULL) 704 newinterp = brand_info->interp_newpath; 705 706 exec_new_vmspace(imgp, NULL); 707 708 /* 709 * Yeah, I'm paranoid. There is every reason in the world to get 710 * VTEXT now since from here on out, there are places we can have 711 * a context switch. Better safe than sorry; I really don't want 712 * the file to change while it's being loaded. 713 */ 714 vsetflags(imgp->vp, VTEXT); 715 716 vmspace = imgp->proc->p_vmspace; 717 /* Choose the base address for dynamic executables if we need to. */ 718 if (et_dyn_addr) 719 et_dyn_addr = pie_base_hint(imgp->proc); 720 721 for (i = 0; i < hdr->e_phnum; i++) { 722 switch (phdr[i].p_type) { 723 case PT_LOAD: /* Loadable segment */ 724 if (phdr[i].p_memsz == 0) 725 break; 726 prot = __elfN(trans_prot)(phdr[i].p_flags); 727 728 if ((error = __elfN(load_section)( 729 imgp->proc, 730 vmspace, 731 imgp->vp, 732 phdr[i].p_offset, 733 (caddr_t)phdr[i].p_vaddr + et_dyn_addr, 734 phdr[i].p_memsz, 735 phdr[i].p_filesz, 736 prot)) != 0) { 737 if (interp != NULL) 738 kfree (interp, M_TEMP); 739 return (error); 740 } 741 742 /* 743 * If this segment contains the program headers, 744 * remember their virtual address for the AT_PHDR 745 * aux entry. Static binaries don't usually include 746 * a PT_PHDR entry. 747 */ 748 if (phdr[i].p_offset == 0 && 749 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 750 <= phdr[i].p_filesz) 751 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 752 et_dyn_addr; 753 754 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 755 seg_size = round_page(phdr[i].p_memsz + 756 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 757 758 /* 759 * Is this .text or .data? We can't use 760 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 761 * alpha terribly and possibly does other bad 762 * things so we stick to the old way of figuring 763 * it out: If the segment contains the program 764 * entry point, it's a text segment, otherwise it 765 * is a data segment. 766 * 767 * Note that obreak() assumes that data_addr + 768 * data_size == end of data load area, and the ELF 769 * file format expects segments to be sorted by 770 * address. If multiple data segments exist, the 771 * last one will be used. 772 */ 773 if (hdr->e_entry >= phdr[i].p_vaddr && 774 hdr->e_entry < (phdr[i].p_vaddr + 775 phdr[i].p_memsz)) { 776 text_size = seg_size; 777 text_addr = seg_addr; 778 entry = (u_long)hdr->e_entry + et_dyn_addr; 779 } else { 780 data_size = seg_size; 781 data_addr = seg_addr; 782 } 783 total_size += seg_size; 784 785 /* 786 * Check limits. It should be safe to check the 787 * limits after loading the segment since we do 788 * not actually fault in all the segment's pages. 789 */ 790 if (data_size > 791 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur || 792 text_size > maxtsiz || 793 total_size > 794 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) { 795 if (interp != NULL) 796 kfree(interp, M_TEMP); 797 error = ENOMEM; 798 return (error); 799 } 800 break; 801 case PT_PHDR: /* Program header table info */ 802 proghdr = phdr[i].p_vaddr + et_dyn_addr; 803 break; 804 default: 805 break; 806 } 807 } 808 809 vmspace->vm_tsize = text_size; /* in bytes */ 810 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 811 vmspace->vm_dsize = data_size; /* in bytes */ 812 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 813 814 addr = ELF_RTLD_ADDR(vmspace); 815 816 imgp->entry_addr = entry; 817 818 imgp->proc->p_sysent = brand_info->sysvec; 819 820 if (interp != NULL) { 821 int have_interp = FALSE; 822 if (brand_info->emul_path != NULL && 823 brand_info->emul_path[0] != '\0') { 824 path = kmalloc(MAXPATHLEN, M_TEMP, M_WAITOK); 825 ksnprintf(path, MAXPATHLEN, "%s%s", 826 brand_info->emul_path, interp); 827 error = __elfN(load_file)(imgp->proc, path, &addr, 828 &imgp->entry_addr); 829 kfree(path, M_TEMP); 830 if (error == 0) 831 have_interp = TRUE; 832 } 833 if (!have_interp && newinterp != NULL) { 834 error = __elfN(load_file)(imgp->proc, newinterp, 835 &addr, &imgp->entry_addr); 836 if (error == 0) 837 have_interp = TRUE; 838 } 839 if (!have_interp) { 840 error = __elfN(load_file)(imgp->proc, interp, &addr, 841 &imgp->entry_addr); 842 } 843 if (error != 0) { 844 uprintf("ELF interpreter %s not found\n", interp); 845 kfree(interp, M_TEMP); 846 return (error); 847 } 848 kfree(interp, M_TEMP); 849 } else 850 addr = et_dyn_addr; 851 852 /* 853 * Construct auxargs table (used by the fixup routine) 854 */ 855 elf_auxargs = kmalloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 856 elf_auxargs->execfd = -1; 857 elf_auxargs->phdr = proghdr; 858 elf_auxargs->phent = hdr->e_phentsize; 859 elf_auxargs->phnum = hdr->e_phnum; 860 elf_auxargs->pagesz = PAGE_SIZE; 861 elf_auxargs->base = addr; 862 elf_auxargs->flags = 0; 863 elf_auxargs->entry = entry; 864 865 imgp->auxargs = elf_auxargs; 866 imgp->interpreted = 0; 867 imgp->proc->p_osrel = osrel; 868 869 return (error); 870 } 871 872 int 873 __elfN(dragonfly_fixup)(register_t **stack_base, struct image_params *imgp) 874 { 875 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 876 Elf_Addr *base; 877 Elf_Addr *pos; 878 879 base = (Elf_Addr *)*stack_base; 880 pos = base + (imgp->args->argc + imgp->args->envc + 2); 881 882 if (args->execfd != -1) 883 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 884 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 885 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 886 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 887 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 888 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 889 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 890 AUXARGS_ENTRY(pos, AT_BASE, args->base); 891 if (imgp->execpathp != 0) 892 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 893 AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate); 894 AUXARGS_ENTRY(pos, AT_NULL, 0); 895 896 kfree(imgp->auxargs, M_TEMP); 897 imgp->auxargs = NULL; 898 899 base--; 900 suword64(base, (long)imgp->args->argc); 901 *stack_base = (register_t *)base; 902 return (0); 903 } 904 905 /* 906 * Code for generating ELF core dumps. 907 */ 908 909 typedef int (*segment_callback)(vm_map_entry_t, void *); 910 911 /* Closure for cb_put_phdr(). */ 912 struct phdr_closure { 913 Elf_Phdr *phdr; /* Program header to fill in (incremented) */ 914 Elf_Phdr *phdr_max; /* Pointer bound for error check */ 915 Elf_Off offset; /* Offset of segment in core file */ 916 }; 917 918 /* Closure for cb_size_segment(). */ 919 struct sseg_closure { 920 int count; /* Count of writable segments. */ 921 size_t vsize; /* Total size of all writable segments. */ 922 }; 923 924 /* Closure for cb_put_fp(). */ 925 struct fp_closure { 926 struct vn_hdr *vnh; 927 struct vn_hdr *vnh_max; 928 int count; 929 struct stat *sb; 930 }; 931 932 typedef struct elf_buf { 933 char *buf; 934 size_t off; 935 size_t off_max; 936 } *elf_buf_t; 937 938 static void *target_reserve(elf_buf_t target, size_t bytes, int *error); 939 940 static int cb_put_phdr (vm_map_entry_t, void *); 941 static int cb_size_segment (vm_map_entry_t, void *); 942 static int cb_fpcount_segment(vm_map_entry_t, void *); 943 static int cb_put_fp(vm_map_entry_t, void *); 944 945 946 static int each_segment (struct proc *, segment_callback, void *, int); 947 static int __elfN(corehdr)(struct lwp *, int, struct file *, struct ucred *, 948 int, elf_buf_t); 949 enum putmode { WRITE, DRYRUN }; 950 static int __elfN(puthdr)(struct lwp *, elf_buf_t, int sig, enum putmode, 951 int, struct file *); 952 static int elf_putallnotes(struct lwp *, elf_buf_t, int, enum putmode); 953 static int __elfN(putnote)(elf_buf_t, const char *, int, const void *, size_t); 954 955 static int elf_putsigs(struct lwp *, elf_buf_t); 956 static int elf_puttextvp(struct proc *, elf_buf_t); 957 static int elf_putfiles(struct proc *, elf_buf_t, struct file *); 958 959 int 960 __elfN(coredump)(struct lwp *lp, int sig, struct vnode *vp, off_t limit) 961 { 962 struct file *fp; 963 int error; 964 965 if ((error = falloc(NULL, &fp, NULL)) != 0) 966 return (error); 967 fsetcred(fp, lp->lwp_proc->p_ucred); 968 969 /* 970 * XXX fixme. 971 */ 972 fp->f_type = DTYPE_VNODE; 973 fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW; 974 fp->f_ops = &vnode_fileops; 975 fp->f_data = vp; 976 977 error = generic_elf_coredump(lp, sig, fp, limit); 978 979 fp->f_type = 0; 980 fp->f_flag = 0; 981 fp->f_ops = &badfileops; 982 fp->f_data = NULL; 983 fdrop(fp); 984 return (error); 985 } 986 987 int 988 generic_elf_coredump(struct lwp *lp, int sig, struct file *fp, off_t limit) 989 { 990 struct proc *p = lp->lwp_proc; 991 struct ucred *cred = p->p_ucred; 992 int error = 0; 993 struct sseg_closure seginfo; 994 struct elf_buf target; 995 996 if (!fp) 997 kprintf("can't dump core - null fp\n"); 998 999 /* 1000 * Size the program segments 1001 */ 1002 seginfo.count = 0; 1003 seginfo.vsize = 0; 1004 each_segment(p, cb_size_segment, &seginfo, 1); 1005 1006 /* 1007 * Calculate the size of the core file header area by making 1008 * a dry run of generating it. Nothing is written, but the 1009 * size is calculated. 1010 */ 1011 bzero(&target, sizeof(target)); 1012 __elfN(puthdr)(lp, &target, sig, DRYRUN, seginfo.count, fp); 1013 1014 if (target.off + seginfo.vsize >= limit) 1015 return (EFAULT); 1016 1017 /* 1018 * Allocate memory for building the header, fill it up, 1019 * and write it out. 1020 */ 1021 target.off_max = target.off; 1022 target.off = 0; 1023 target.buf = kmalloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO); 1024 1025 error = __elfN(corehdr)(lp, sig, fp, cred, seginfo.count, &target); 1026 1027 /* Write the contents of all of the writable segments. */ 1028 if (error == 0) { 1029 Elf_Phdr *php; 1030 int i; 1031 ssize_t nbytes; 1032 1033 php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1; 1034 for (i = 0; i < seginfo.count; i++) { 1035 error = fp_write(fp, (caddr_t)php->p_vaddr, 1036 php->p_filesz, &nbytes, UIO_USERSPACE); 1037 if (error != 0) 1038 break; 1039 php++; 1040 } 1041 } 1042 kfree(target.buf, M_TEMP); 1043 1044 return (error); 1045 } 1046 1047 /* 1048 * A callback for each_segment() to write out the segment's 1049 * program header entry. 1050 */ 1051 static int 1052 cb_put_phdr(vm_map_entry_t entry, void *closure) 1053 { 1054 struct phdr_closure *phc = closure; 1055 Elf_Phdr *phdr = phc->phdr; 1056 1057 if (phc->phdr == phc->phdr_max) 1058 return (EINVAL); 1059 1060 phc->offset = round_page(phc->offset); 1061 1062 phdr->p_type = PT_LOAD; 1063 phdr->p_offset = phc->offset; 1064 phdr->p_vaddr = entry->start; 1065 phdr->p_paddr = 0; 1066 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1067 phdr->p_align = PAGE_SIZE; 1068 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1069 1070 phc->offset += phdr->p_filesz; 1071 ++phc->phdr; 1072 return (0); 1073 } 1074 1075 /* 1076 * A callback for each_writable_segment() to gather information about 1077 * the number of segments and their total size. 1078 */ 1079 static int 1080 cb_size_segment(vm_map_entry_t entry, void *closure) 1081 { 1082 struct sseg_closure *ssc = closure; 1083 1084 ++ssc->count; 1085 ssc->vsize += entry->end - entry->start; 1086 return (0); 1087 } 1088 1089 /* 1090 * A callback for each_segment() to gather information about 1091 * the number of text segments. 1092 */ 1093 static int 1094 cb_fpcount_segment(vm_map_entry_t entry, void *closure) 1095 { 1096 int *count = closure; 1097 struct vnode *vp; 1098 1099 if (entry->object.vm_object->type == OBJT_VNODE) { 1100 vp = (struct vnode *)entry->object.vm_object->handle; 1101 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp) 1102 return (0); 1103 ++*count; 1104 } 1105 return (0); 1106 } 1107 1108 static int 1109 cb_put_fp(vm_map_entry_t entry, void *closure) 1110 { 1111 struct fp_closure *fpc = closure; 1112 struct vn_hdr *vnh = fpc->vnh; 1113 Elf_Phdr *phdr = &vnh->vnh_phdr; 1114 struct vnode *vp; 1115 int error; 1116 1117 /* 1118 * If an entry represents a vnode then write out a file handle. 1119 * 1120 * If we are checkpointing a checkpoint-restored program we do 1121 * NOT record the filehandle for the old checkpoint vnode (which 1122 * is mapped all over the place). Instead we rely on the fact 1123 * that a checkpoint-restored program does not mmap() the checkpt 1124 * vnode NOCORE, so its contents will be written out to the 1125 * new checkpoint file. This is necessary because the 'old' 1126 * checkpoint file is typically destroyed when a new one is created 1127 * and thus cannot be used to restore the new checkpoint. 1128 * 1129 * Theoretically we could create a chain of checkpoint files and 1130 * operate the checkpointing operation kinda like an incremental 1131 * checkpoint, but a checkpoint restore would then likely wind up 1132 * referencing many prior checkpoint files and that is a bit over 1133 * the top for the purpose of the checkpoint API. 1134 */ 1135 if (entry->object.vm_object->type == OBJT_VNODE) { 1136 vp = (struct vnode *)entry->object.vm_object->handle; 1137 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp) 1138 return (0); 1139 if (vnh == fpc->vnh_max) 1140 return (EINVAL); 1141 1142 if (vp->v_mount) 1143 vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1144 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid); 1145 if (error) { 1146 char *freepath, *fullpath; 1147 1148 /* 1149 * This is actually a relatively common occurance, 1150 * so don't spew on the console by default. 1151 */ 1152 if (vn_fullpath(curproc, vp, &fullpath, &freepath, 0)) { 1153 if (bootverbose) 1154 kprintf("Warning: coredump, error %d: cannot store file handle for vnode %p\n", error, vp); 1155 } else { 1156 if (bootverbose) 1157 kprintf("Warning: coredump, error %d: cannot store file handle for %s\n", error, fullpath); 1158 kfree(freepath, M_TEMP); 1159 } 1160 error = 0; 1161 } 1162 1163 phdr->p_type = PT_LOAD; 1164 phdr->p_offset = 0; /* not written to core */ 1165 phdr->p_vaddr = entry->start; 1166 phdr->p_paddr = 0; 1167 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1168 phdr->p_align = PAGE_SIZE; 1169 phdr->p_flags = 0; 1170 if (entry->protection & VM_PROT_READ) 1171 phdr->p_flags |= PF_R; 1172 if (entry->protection & VM_PROT_WRITE) 1173 phdr->p_flags |= PF_W; 1174 if (entry->protection & VM_PROT_EXECUTE) 1175 phdr->p_flags |= PF_X; 1176 ++fpc->vnh; 1177 ++fpc->count; 1178 } 1179 return (0); 1180 } 1181 1182 /* 1183 * For each writable segment in the process's memory map, call the given 1184 * function with a pointer to the map entry and some arbitrary 1185 * caller-supplied data. 1186 */ 1187 static int 1188 each_segment(struct proc *p, segment_callback func, void *closure, int writable) 1189 { 1190 int error = 0; 1191 vm_map_t map = &p->p_vmspace->vm_map; 1192 vm_map_entry_t entry; 1193 1194 RB_FOREACH(entry, vm_map_rb_tree, &map->rb_root) { 1195 vm_object_t obj; 1196 vm_object_t lobj; 1197 vm_object_t tobj; 1198 1199 /* 1200 * Don't dump inaccessible mappings, deal with legacy 1201 * coredump mode. 1202 * 1203 * Note that read-only segments related to the elf binary 1204 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1205 * need to arbitrarily ignore such segments. 1206 */ 1207 if (elf_legacy_coredump) { 1208 if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW) 1209 continue; 1210 } else { 1211 if (writable && (entry->protection & VM_PROT_ALL) == 0) 1212 continue; 1213 } 1214 1215 /* 1216 * Dont include memory segment in the coredump if 1217 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1218 * madvise(2). 1219 * 1220 * Currently we only dump normal VM object maps. We do 1221 * not dump submaps or virtual page tables. 1222 */ 1223 if (writable && (entry->eflags & MAP_ENTRY_NOCOREDUMP)) 1224 continue; 1225 if (entry->maptype != VM_MAPTYPE_NORMAL) 1226 continue; 1227 if ((obj = entry->object.vm_object) == NULL) 1228 continue; 1229 1230 /* 1231 * Find the bottom-most object, leaving the base object 1232 * and the bottom-most object held (but only one hold 1233 * if they happen to be the same). 1234 */ 1235 vm_object_hold_shared(obj); 1236 1237 lobj = obj; 1238 while (lobj && (tobj = lobj->backing_object) != NULL) { 1239 KKASSERT(tobj != obj); 1240 vm_object_hold_shared(tobj); 1241 if (tobj == lobj->backing_object) { 1242 if (lobj != obj) { 1243 vm_object_lock_swap(); 1244 vm_object_drop(lobj); 1245 } 1246 lobj = tobj; 1247 } else { 1248 vm_object_drop(tobj); 1249 } 1250 } 1251 1252 /* 1253 * The callback only applies to default, swap, or vnode 1254 * objects. Other types of objects such as memory-mapped 1255 * devices are ignored. 1256 */ 1257 if (lobj->type == OBJT_DEFAULT || lobj->type == OBJT_SWAP || 1258 lobj->type == OBJT_VNODE) { 1259 error = (*func)(entry, closure); 1260 } 1261 if (lobj != obj) 1262 vm_object_drop(lobj); 1263 vm_object_drop(obj); 1264 } 1265 return (error); 1266 } 1267 1268 static 1269 void * 1270 target_reserve(elf_buf_t target, size_t bytes, int *error) 1271 { 1272 void *res = NULL; 1273 1274 if (target->buf) { 1275 if (target->off + bytes > target->off_max) 1276 *error = EINVAL; 1277 else 1278 res = target->buf + target->off; 1279 } 1280 target->off += bytes; 1281 return (res); 1282 } 1283 1284 /* 1285 * Write the core file header to the file, including padding up to 1286 * the page boundary. 1287 */ 1288 static int 1289 __elfN(corehdr)(struct lwp *lp, int sig, struct file *fp, struct ucred *cred, 1290 int numsegs, elf_buf_t target) 1291 { 1292 int error; 1293 ssize_t nbytes; 1294 1295 /* 1296 * Fill in the header. The fp is passed so we can detect and flag 1297 * a checkpoint file pointer within the core file itself, because 1298 * it may not be restored from the same file handle. 1299 */ 1300 error = __elfN(puthdr)(lp, target, sig, WRITE, numsegs, fp); 1301 1302 /* Write it to the core file. */ 1303 if (error == 0) { 1304 error = fp_write(fp, target->buf, target->off, &nbytes, 1305 UIO_SYSSPACE); 1306 } 1307 return (error); 1308 } 1309 1310 static int 1311 __elfN(puthdr)(struct lwp *lp, elf_buf_t target, int sig, enum putmode mode, 1312 int numsegs, struct file *fp) 1313 { 1314 struct proc *p = lp->lwp_proc; 1315 int error = 0; 1316 size_t phoff; 1317 size_t noteoff; 1318 size_t notesz; 1319 Elf_Ehdr *ehdr; 1320 Elf_Phdr *phdr; 1321 1322 ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error); 1323 1324 phoff = target->off; 1325 phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error); 1326 1327 noteoff = target->off; 1328 if (error == 0) 1329 elf_putallnotes(lp, target, sig, mode); 1330 notesz = target->off - noteoff; 1331 1332 /* 1333 * put extra cruft for dumping process state here 1334 * - we really want it be before all the program 1335 * mappings 1336 * - we just need to update the offset accordingly 1337 * and GDB will be none the wiser. 1338 */ 1339 if (error == 0) 1340 error = elf_puttextvp(p, target); 1341 if (error == 0) 1342 error = elf_putsigs(lp, target); 1343 if (error == 0) 1344 error = elf_putfiles(p, target, fp); 1345 1346 /* 1347 * Align up to a page boundary for the program segments. The 1348 * actual data will be written to the outptu file, not to elf_buf_t, 1349 * so we do not have to do any further bounds checking. 1350 */ 1351 target->off = round_page(target->off); 1352 if (error == 0 && ehdr != NULL) { 1353 /* 1354 * Fill in the ELF header. 1355 */ 1356 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1357 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1358 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1359 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1360 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1361 ehdr->e_ident[EI_DATA] = ELF_DATA; 1362 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1363 ehdr->e_ident[EI_OSABI] = ELFOSABI_NONE; 1364 ehdr->e_ident[EI_ABIVERSION] = 0; 1365 ehdr->e_ident[EI_PAD] = 0; 1366 ehdr->e_type = ET_CORE; 1367 ehdr->e_machine = ELF_ARCH; 1368 ehdr->e_version = EV_CURRENT; 1369 ehdr->e_entry = 0; 1370 ehdr->e_phoff = phoff; 1371 ehdr->e_flags = 0; 1372 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1373 ehdr->e_phentsize = sizeof(Elf_Phdr); 1374 ehdr->e_phnum = numsegs + 1; 1375 ehdr->e_shentsize = sizeof(Elf_Shdr); 1376 ehdr->e_shnum = 0; 1377 ehdr->e_shstrndx = SHN_UNDEF; 1378 } 1379 if (error == 0 && phdr != NULL) { 1380 /* 1381 * Fill in the program header entries. 1382 */ 1383 struct phdr_closure phc; 1384 1385 /* The note segement. */ 1386 phdr->p_type = PT_NOTE; 1387 phdr->p_offset = noteoff; 1388 phdr->p_vaddr = 0; 1389 phdr->p_paddr = 0; 1390 phdr->p_filesz = notesz; 1391 phdr->p_memsz = 0; 1392 phdr->p_flags = 0; 1393 phdr->p_align = 0; 1394 ++phdr; 1395 1396 /* All the writable segments from the program. */ 1397 phc.phdr = phdr; 1398 phc.phdr_max = phdr + numsegs; 1399 phc.offset = target->off; 1400 each_segment(p, cb_put_phdr, &phc, 1); 1401 } 1402 return (error); 1403 } 1404 1405 /* 1406 * Append core dump notes to target ELF buffer or simply update target size 1407 * if dryrun selected. 1408 */ 1409 static int 1410 elf_putallnotes(struct lwp *corelp, elf_buf_t target, int sig, 1411 enum putmode mode) 1412 { 1413 struct proc *p = corelp->lwp_proc; 1414 int error; 1415 struct { 1416 prstatus_t status; 1417 prfpregset_t fpregs; 1418 prpsinfo_t psinfo; 1419 } *tmpdata; 1420 prstatus_t *status; 1421 prfpregset_t *fpregs; 1422 prpsinfo_t *psinfo; 1423 struct lwp *lp; 1424 1425 /* 1426 * Allocate temporary storage for notes on heap to avoid stack overflow. 1427 */ 1428 if (mode != DRYRUN) { 1429 tmpdata = kmalloc(sizeof(*tmpdata), M_TEMP, M_ZERO | M_WAITOK); 1430 status = &tmpdata->status; 1431 fpregs = &tmpdata->fpregs; 1432 psinfo = &tmpdata->psinfo; 1433 } else { 1434 tmpdata = NULL; 1435 status = NULL; 1436 fpregs = NULL; 1437 psinfo = NULL; 1438 } 1439 1440 /* 1441 * Append LWP-agnostic note. 1442 */ 1443 if (mode != DRYRUN) { 1444 psinfo->pr_version = PRPSINFO_VERSION; 1445 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 1446 strlcpy(psinfo->pr_fname, p->p_comm, 1447 sizeof(psinfo->pr_fname)); 1448 /* 1449 * XXX - We don't fill in the command line arguments 1450 * properly yet. 1451 */ 1452 strlcpy(psinfo->pr_psargs, p->p_comm, 1453 sizeof(psinfo->pr_psargs)); 1454 } 1455 error = 1456 __elfN(putnote)(target, "CORE", NT_PRPSINFO, psinfo, sizeof *psinfo); 1457 if (error) 1458 goto exit; 1459 1460 /* 1461 * Append first note for LWP that triggered core so that it is 1462 * the selected one when the debugger starts. 1463 */ 1464 if (mode != DRYRUN) { 1465 status->pr_version = PRSTATUS_VERSION; 1466 status->pr_statussz = sizeof(prstatus_t); 1467 status->pr_gregsetsz = sizeof(gregset_t); 1468 status->pr_fpregsetsz = sizeof(fpregset_t); 1469 status->pr_osreldate = osreldate; 1470 status->pr_cursig = sig; 1471 status->pr_pid = corelp->lwp_tid; 1472 fill_regs(corelp, &status->pr_reg); 1473 fill_fpregs(corelp, fpregs); 1474 } 1475 error = 1476 __elfN(putnote)(target, "CORE", NT_PRSTATUS, status, sizeof *status); 1477 if (error) 1478 goto exit; 1479 error = 1480 __elfN(putnote)(target, "CORE", NT_FPREGSET, fpregs, sizeof *fpregs); 1481 if (error) 1482 goto exit; 1483 1484 /* 1485 * Then append notes for other LWPs. 1486 */ 1487 FOREACH_LWP_IN_PROC(lp, p) { 1488 if (lp == corelp) 1489 continue; 1490 /* skip lwps being created */ 1491 if (lp->lwp_thread == NULL) 1492 continue; 1493 if (mode != DRYRUN) { 1494 status->pr_pid = lp->lwp_tid; 1495 fill_regs(lp, &status->pr_reg); 1496 fill_fpregs(lp, fpregs); 1497 } 1498 error = __elfN(putnote)(target, "CORE", NT_PRSTATUS, 1499 status, sizeof *status); 1500 if (error) 1501 goto exit; 1502 error = __elfN(putnote)(target, "CORE", NT_FPREGSET, 1503 fpregs, sizeof *fpregs); 1504 if (error) 1505 goto exit; 1506 } 1507 1508 exit: 1509 if (tmpdata != NULL) 1510 kfree(tmpdata, M_TEMP); 1511 return (error); 1512 } 1513 1514 /* 1515 * Generate a note sub-structure. 1516 * 1517 * NOTE: 4-byte alignment. 1518 */ 1519 static int 1520 __elfN(putnote)(elf_buf_t target, const char *name, int type, 1521 const void *desc, size_t descsz) 1522 { 1523 int error = 0; 1524 char *dst; 1525 Elf_Note note; 1526 1527 note.n_namesz = strlen(name) + 1; 1528 note.n_descsz = descsz; 1529 note.n_type = type; 1530 dst = target_reserve(target, sizeof(note), &error); 1531 if (dst != NULL) 1532 bcopy(¬e, dst, sizeof note); 1533 dst = target_reserve(target, note.n_namesz, &error); 1534 if (dst != NULL) 1535 bcopy(name, dst, note.n_namesz); 1536 target->off = roundup2(target->off, sizeof(Elf_Word)); 1537 dst = target_reserve(target, note.n_descsz, &error); 1538 if (dst != NULL) 1539 bcopy(desc, dst, note.n_descsz); 1540 target->off = roundup2(target->off, sizeof(Elf_Word)); 1541 return (error); 1542 } 1543 1544 1545 static int 1546 elf_putsigs(struct lwp *lp, elf_buf_t target) 1547 { 1548 /* XXX lwp handle more than one lwp */ 1549 struct proc *p = lp->lwp_proc; 1550 int error = 0; 1551 struct ckpt_siginfo *csi; 1552 1553 csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error); 1554 if (csi) { 1555 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo); 1556 bcopy(p->p_sigacts, &csi->csi_sigacts, sizeof(*p->p_sigacts)); 1557 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval)); 1558 bcopy(&lp->lwp_sigmask, &csi->csi_sigmask, 1559 sizeof(sigset_t)); 1560 csi->csi_sigparent = p->p_sigparent; 1561 } 1562 return (error); 1563 } 1564 1565 static int 1566 elf_putfiles(struct proc *p, elf_buf_t target, struct file *ckfp) 1567 { 1568 thread_t td = curthread; 1569 int error = 0; 1570 int i; 1571 struct ckpt_filehdr *cfh = NULL; 1572 struct ckpt_fileinfo *cfi; 1573 struct file *fp; 1574 struct vnode *vp; 1575 1576 /* 1577 * the duplicated loop is gross, but it was the only way 1578 * to eliminate uninitialized variable warnings 1579 */ 1580 cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error); 1581 if (cfh) { 1582 cfh->cfh_nfiles = 0; 1583 } 1584 1585 /* 1586 * ignore STDIN/STDERR/STDOUT. 1587 */ 1588 KKASSERT(td->td_proc == p); 1589 for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) { 1590 fp = holdfp(td, i, -1); 1591 if (fp == NULL) 1592 continue; 1593 /* 1594 * XXX Only checkpoint vnodes for now. 1595 */ 1596 if (fp->f_type != DTYPE_VNODE) { 1597 fdrop(fp); 1598 continue; 1599 } 1600 cfi = target_reserve(target, sizeof(struct ckpt_fileinfo), 1601 &error); 1602 if (cfi == NULL) { 1603 fdrop(fp); 1604 continue; 1605 } 1606 cfi->cfi_index = -1; 1607 cfi->cfi_type = fp->f_type; 1608 cfi->cfi_flags = fp->f_flag; 1609 cfi->cfi_offset = fp->f_offset; 1610 cfi->cfi_ckflags = 0; 1611 1612 if (fp == ckfp) 1613 cfi->cfi_ckflags |= CKFIF_ISCKPTFD; 1614 /* f_count and f_msgcount should not be saved/restored */ 1615 /* XXX save cred info */ 1616 1617 switch(fp->f_type) { 1618 case DTYPE_VNODE: 1619 vp = (struct vnode *)fp->f_data; 1620 /* 1621 * it looks like a bug in ptrace is marking 1622 * a non-vnode as a vnode - until we find the 1623 * root cause this will at least prevent 1624 * further panics from truss 1625 */ 1626 if (vp == NULL || vp->v_mount == NULL) 1627 break; 1628 cfh->cfh_nfiles++; 1629 cfi->cfi_index = i; 1630 cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1631 error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid); 1632 break; 1633 default: 1634 break; 1635 } 1636 fdrop(fp); 1637 } 1638 return (error); 1639 } 1640 1641 static int 1642 elf_puttextvp(struct proc *p, elf_buf_t target) 1643 { 1644 int error = 0; 1645 int *vn_count; 1646 struct fp_closure fpc; 1647 struct ckpt_vminfo *vminfo; 1648 1649 vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error); 1650 if (vminfo != NULL) { 1651 vminfo->cvm_dsize = btoc(p->p_vmspace->vm_dsize); /* pages */ 1652 vminfo->cvm_tsize = btoc(p->p_vmspace->vm_tsize); /* pages */ 1653 vminfo->cvm_daddr = p->p_vmspace->vm_daddr; 1654 vminfo->cvm_taddr = p->p_vmspace->vm_taddr; 1655 } 1656 1657 fpc.count = 0; 1658 vn_count = target_reserve(target, sizeof(int), &error); 1659 if (target->buf != NULL) { 1660 fpc.vnh = (struct vn_hdr *)(target->buf + target->off); 1661 fpc.vnh_max = fpc.vnh + 1662 (target->off_max - target->off) / sizeof(struct vn_hdr); 1663 error = each_segment(p, cb_put_fp, &fpc, 0); 1664 if (vn_count) 1665 *vn_count = fpc.count; 1666 } else { 1667 error = each_segment(p, cb_fpcount_segment, &fpc.count, 0); 1668 } 1669 target->off += fpc.count * sizeof(struct vn_hdr); 1670 return (error); 1671 } 1672 1673 /* 1674 * Try to find the appropriate ABI-note section for checknote, 1675 * The entire image is searched if necessary, not only the first page. 1676 */ 1677 static boolean_t 1678 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 1679 int32_t *osrel) 1680 { 1681 boolean_t valid_note_found; 1682 const Elf_Phdr *phdr, *pnote; 1683 const Elf_Ehdr *hdr; 1684 int i; 1685 1686 valid_note_found = FALSE; 1687 hdr = (const Elf_Ehdr *)imgp->image_header; 1688 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1689 1690 for (i = 0; i < hdr->e_phnum; i++) { 1691 if (phdr[i].p_type == PT_NOTE) { 1692 pnote = &phdr[i]; 1693 valid_note_found = check_PT_NOTE (imgp, checknote, 1694 osrel, pnote); 1695 if (valid_note_found) 1696 break; 1697 } 1698 } 1699 return valid_note_found; 1700 } 1701 1702 /* 1703 * Be careful not to create new overflow conditions when checking 1704 * for overflow. 1705 */ 1706 static boolean_t 1707 note_overflow(const Elf_Note *note, size_t maxsize) 1708 { 1709 if (sizeof(*note) > maxsize) 1710 return TRUE; 1711 if (note->n_namesz > maxsize - sizeof(*note)) 1712 return TRUE; 1713 return FALSE; 1714 } 1715 1716 static boolean_t 1717 hdr_overflow(__ElfN(Off) off_beg, __ElfN(Size) size) 1718 { 1719 __ElfN(Off) off_end; 1720 1721 off_end = off_beg + size; 1722 if (off_end < off_beg) 1723 return TRUE; 1724 return FALSE; 1725 } 1726 1727 static boolean_t 1728 check_PT_NOTE(struct image_params *imgp, Elf_Brandnote *checknote, 1729 int32_t *osrel, const Elf_Phdr * pnote) 1730 { 1731 boolean_t limited_to_first_page; 1732 boolean_t found = FALSE; 1733 const Elf_Note *note, *note0, *note_end; 1734 const char *note_name; 1735 __ElfN(Off) noteloc, firstloc; 1736 __ElfN(Size) notesz, firstlen, endbyte; 1737 struct lwbuf *lwb; 1738 struct lwbuf lwb_cache; 1739 const char *page; 1740 char *data = NULL; 1741 int n; 1742 1743 if (hdr_overflow(pnote->p_offset, pnote->p_filesz)) 1744 return (FALSE); 1745 notesz = pnote->p_filesz; 1746 noteloc = pnote->p_offset; 1747 endbyte = noteloc + notesz; 1748 limited_to_first_page = noteloc < PAGE_SIZE && endbyte < PAGE_SIZE; 1749 1750 if (limited_to_first_page) { 1751 note = (const Elf_Note *)(imgp->image_header + noteloc); 1752 note_end = (const Elf_Note *)(imgp->image_header + endbyte); 1753 note0 = note; 1754 } else { 1755 firstloc = noteloc & PAGE_MASK; 1756 firstlen = PAGE_SIZE - firstloc; 1757 if (notesz < sizeof(Elf_Note) || notesz > PAGE_SIZE) 1758 return (FALSE); 1759 1760 lwb = &lwb_cache; 1761 if (exec_map_page(imgp, noteloc >> PAGE_SHIFT, &lwb, &page)) 1762 return (FALSE); 1763 if (firstlen < notesz) { /* crosses page boundary */ 1764 data = kmalloc(notesz, M_TEMP, M_WAITOK); 1765 bcopy(page + firstloc, data, firstlen); 1766 1767 exec_unmap_page(lwb); 1768 lwb = &lwb_cache; 1769 if (exec_map_page(imgp, (noteloc >> PAGE_SHIFT) + 1, 1770 &lwb, &page)) { 1771 kfree(data, M_TEMP); 1772 return (FALSE); 1773 } 1774 bcopy(page, data + firstlen, notesz - firstlen); 1775 note = note0 = (const Elf_Note *)(data); 1776 note_end = (const Elf_Note *)(data + notesz); 1777 } else { 1778 note = note0 = (const Elf_Note *)(page + firstloc); 1779 note_end = (const Elf_Note *)(page + firstloc + 1780 firstlen); 1781 } 1782 } 1783 1784 for (n = 0; n < 100 && note >= note0 && note < note_end; n++) { 1785 if (!aligned(note, Elf32_Addr)) 1786 break; 1787 if (note_overflow(note, (const char *)note_end - 1788 (const char *)note)) { 1789 break; 1790 } 1791 note_name = (const char *)(note + 1); 1792 1793 if (note->n_namesz == checknote->hdr.n_namesz 1794 && note->n_descsz == checknote->hdr.n_descsz 1795 && note->n_type == checknote->hdr.n_type 1796 && (strncmp(checknote->vendor, note_name, 1797 checknote->hdr.n_namesz) == 0)) { 1798 /* Fetch osreldata from ABI.note-tag */ 1799 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 1800 checknote->trans_osrel != NULL) 1801 checknote->trans_osrel(note, osrel); 1802 found = TRUE; 1803 break; 1804 } 1805 note = (const Elf_Note *)((const char *)(note + 1) + 1806 roundup2(note->n_namesz, sizeof(Elf32_Addr)) + 1807 roundup2(note->n_descsz, sizeof(Elf32_Addr))); 1808 } 1809 1810 if (!limited_to_first_page) { 1811 if (data != NULL) 1812 kfree(data, M_TEMP); 1813 exec_unmap_page(lwb); 1814 } 1815 return (found); 1816 } 1817 1818 /* 1819 * The interpreter program header may be located beyond the first page, so 1820 * regardless of its location, a copy of the interpreter path is created so 1821 * that it may be safely referenced by the calling function in all case. The 1822 * memory is allocated by calling function, and the copying is done here. 1823 */ 1824 static boolean_t 1825 extract_interpreter(struct image_params *imgp, const Elf_Phdr *pinterpreter, 1826 char *data) 1827 { 1828 boolean_t limited_to_first_page; 1829 const boolean_t result_success = FALSE; 1830 const boolean_t result_failure = TRUE; 1831 __ElfN(Off) pathloc, firstloc; 1832 __ElfN(Size) pathsz, firstlen, endbyte; 1833 struct lwbuf *lwb; 1834 struct lwbuf lwb_cache; 1835 const char *page; 1836 1837 if (hdr_overflow(pinterpreter->p_offset, pinterpreter->p_filesz)) 1838 return (result_failure); 1839 pathsz = pinterpreter->p_filesz; 1840 pathloc = pinterpreter->p_offset; 1841 endbyte = pathloc + pathsz; 1842 1843 limited_to_first_page = pathloc < PAGE_SIZE && endbyte < PAGE_SIZE; 1844 if (limited_to_first_page) { 1845 bcopy(imgp->image_header + pathloc, data, pathsz); 1846 return (result_success); 1847 } 1848 1849 firstloc = pathloc & PAGE_MASK; 1850 firstlen = PAGE_SIZE - firstloc; 1851 1852 lwb = &lwb_cache; 1853 if (exec_map_page(imgp, pathloc >> PAGE_SHIFT, &lwb, &page)) 1854 return (result_failure); 1855 1856 if (firstlen < pathsz) { /* crosses page boundary */ 1857 bcopy(page + firstloc, data, firstlen); 1858 1859 exec_unmap_page(lwb); 1860 lwb = &lwb_cache; 1861 if (exec_map_page(imgp, (pathloc >> PAGE_SHIFT) + 1, &lwb, 1862 &page)) 1863 return (result_failure); 1864 bcopy(page, data + firstlen, pathsz - firstlen); 1865 } else 1866 bcopy(page + firstloc, data, pathsz); 1867 1868 exec_unmap_page(lwb); 1869 return (result_success); 1870 } 1871 1872 static boolean_t 1873 __elfN(bsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 1874 { 1875 uintptr_t p; 1876 1877 p = (uintptr_t)(note + 1); 1878 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 1879 *osrel = *(const int32_t *)(p); 1880 1881 return (TRUE); 1882 } 1883 1884 /* 1885 * Tell kern_execve.c about it, with a little help from the linker. 1886 */ 1887 #if defined(__x86_64__) 1888 static struct execsw elf_execsw = {exec_elf64_imgact, "ELF64"}; 1889 EXEC_SET_ORDERED(elf64, elf_execsw, SI_ORDER_FIRST); 1890 #else /* i386 assumed */ 1891 static struct execsw elf_execsw = {exec_elf32_imgact, "ELF32"}; 1892 EXEC_SET_ORDERED(elf32, elf_execsw, SI_ORDER_FIRST); 1893 #endif 1894 1895 static vm_prot_t 1896 __elfN(trans_prot)(Elf_Word flags) 1897 { 1898 vm_prot_t prot; 1899 1900 prot = 0; 1901 if (flags & PF_X) 1902 prot |= VM_PROT_EXECUTE; 1903 if (flags & PF_W) 1904 prot |= VM_PROT_WRITE; 1905 if (flags & PF_R) 1906 prot |= VM_PROT_READ; 1907 return (prot); 1908 } 1909 1910 static Elf_Word 1911 __elfN(untrans_prot)(vm_prot_t prot) 1912 { 1913 Elf_Word flags; 1914 1915 flags = 0; 1916 if (prot & VM_PROT_EXECUTE) 1917 flags |= PF_X; 1918 if (prot & VM_PROT_READ) 1919 flags |= PF_R; 1920 if (prot & VM_PROT_WRITE) 1921 flags |= PF_W; 1922 return (flags); 1923 } 1924 1925 static u_long 1926 pie_base_hint(struct proc *p) 1927 { 1928 u_long base; 1929 1930 if (elf_pie_base_mmap) 1931 base = vm_map_hint(p, 0, VM_PROT_READ | VM_PROT_EXECUTE); 1932 else 1933 base = ET_DYN_LOAD_ADDR; 1934 return base; 1935 } 1936