1 /*- 2 * Copyright (c) 2000 David O'Brien 3 * Copyright (c) 1995-1996 S�ren 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 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include "opt_compat.h" 35 #include "opt_core.h" 36 37 #include <sys/param.h> 38 #include <sys/exec.h> 39 #include <sys/fcntl.h> 40 #include <sys/imgact.h> 41 #include <sys/imgact_elf.h> 42 #include <sys/kernel.h> 43 #include <sys/lock.h> 44 #include <sys/malloc.h> 45 #include <sys/mount.h> 46 #include <sys/mutex.h> 47 #include <sys/mman.h> 48 #include <sys/namei.h> 49 #include <sys/pioctl.h> 50 #include <sys/proc.h> 51 #include <sys/procfs.h> 52 #include <sys/racct.h> 53 #include <sys/resourcevar.h> 54 #include <sys/sf_buf.h> 55 #include <sys/smp.h> 56 #include <sys/systm.h> 57 #include <sys/signalvar.h> 58 #include <sys/stat.h> 59 #include <sys/sx.h> 60 #include <sys/syscall.h> 61 #include <sys/sysctl.h> 62 #include <sys/sysent.h> 63 #include <sys/vnode.h> 64 #include <sys/syslog.h> 65 #include <sys/eventhandler.h> 66 67 #include <net/zlib.h> 68 69 #include <vm/vm.h> 70 #include <vm/vm_kern.h> 71 #include <vm/vm_param.h> 72 #include <vm/pmap.h> 73 #include <vm/vm_map.h> 74 #include <vm/vm_object.h> 75 #include <vm/vm_extern.h> 76 77 #include <machine/elf.h> 78 #include <machine/md_var.h> 79 80 #define OLD_EI_BRAND 8 81 82 static int __elfN(check_header)(const Elf_Ehdr *hdr); 83 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 84 const char *interp, int32_t *osrel); 85 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 86 u_long *entry, size_t pagesize); 87 static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object, 88 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 89 vm_prot_t prot, size_t pagesize); 90 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 91 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note, 92 int32_t *osrel); 93 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 94 static boolean_t __elfN(check_note)(struct image_params *imgp, 95 Elf_Brandnote *checknote, int32_t *osrel); 96 static vm_prot_t __elfN(trans_prot)(Elf_Word); 97 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 98 99 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 100 ""); 101 102 #ifdef COMPRESS_USER_CORES 103 static int compress_core(gzFile, char *, char *, unsigned int, 104 struct thread * td); 105 #define CORE_BUF_SIZE (16 * 1024) 106 #endif 107 108 int __elfN(fallback_brand) = -1; 109 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 110 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 111 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 112 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 113 &__elfN(fallback_brand)); 114 115 static int elf_legacy_coredump = 0; 116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 117 &elf_legacy_coredump, 0, ""); 118 119 static int __elfN(nxstack) = 0; 120 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 121 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 122 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack"); 123 124 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 125 126 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 127 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 128 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a)) 129 130 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; 131 132 Elf_Brandnote __elfN(freebsd_brandnote) = { 133 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 134 .hdr.n_descsz = sizeof(int32_t), 135 .hdr.n_type = 1, 136 .vendor = FREEBSD_ABI_VENDOR, 137 .flags = BN_TRANSLATE_OSREL, 138 .trans_osrel = __elfN(freebsd_trans_osrel) 139 }; 140 141 static boolean_t 142 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 143 { 144 uintptr_t p; 145 146 p = (uintptr_t)(note + 1); 147 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 148 *osrel = *(const int32_t *)(p); 149 150 return (TRUE); 151 } 152 153 static const char GNU_ABI_VENDOR[] = "GNU"; 154 static int GNU_KFREEBSD_ABI_DESC = 3; 155 156 Elf_Brandnote __elfN(kfreebsd_brandnote) = { 157 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 158 .hdr.n_descsz = 16, /* XXX at least 16 */ 159 .hdr.n_type = 1, 160 .vendor = GNU_ABI_VENDOR, 161 .flags = BN_TRANSLATE_OSREL, 162 .trans_osrel = kfreebsd_trans_osrel 163 }; 164 165 static boolean_t 166 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 167 { 168 const Elf32_Word *desc; 169 uintptr_t p; 170 171 p = (uintptr_t)(note + 1); 172 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 173 174 desc = (const Elf32_Word *)p; 175 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 176 return (FALSE); 177 178 /* 179 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 180 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 181 */ 182 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 183 184 return (TRUE); 185 } 186 187 int 188 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 189 { 190 int i; 191 192 for (i = 0; i < MAX_BRANDS; i++) { 193 if (elf_brand_list[i] == NULL) { 194 elf_brand_list[i] = entry; 195 break; 196 } 197 } 198 if (i == MAX_BRANDS) { 199 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 200 __func__, entry); 201 return (-1); 202 } 203 return (0); 204 } 205 206 int 207 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 208 { 209 int i; 210 211 for (i = 0; i < MAX_BRANDS; i++) { 212 if (elf_brand_list[i] == entry) { 213 elf_brand_list[i] = NULL; 214 break; 215 } 216 } 217 if (i == MAX_BRANDS) 218 return (-1); 219 return (0); 220 } 221 222 int 223 __elfN(brand_inuse)(Elf_Brandinfo *entry) 224 { 225 struct proc *p; 226 int rval = FALSE; 227 228 sx_slock(&allproc_lock); 229 FOREACH_PROC_IN_SYSTEM(p) { 230 if (p->p_sysent == entry->sysvec) { 231 rval = TRUE; 232 break; 233 } 234 } 235 sx_sunlock(&allproc_lock); 236 237 return (rval); 238 } 239 240 static Elf_Brandinfo * 241 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 242 int32_t *osrel) 243 { 244 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 245 Elf_Brandinfo *bi; 246 boolean_t ret; 247 int i; 248 249 /* 250 * We support four types of branding -- (1) the ELF EI_OSABI field 251 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 252 * branding w/in the ELF header, (3) path of the `interp_path' 253 * field, and (4) the ".note.ABI-tag" ELF section. 254 */ 255 256 /* Look for an ".note.ABI-tag" ELF section */ 257 for (i = 0; i < MAX_BRANDS; i++) { 258 bi = elf_brand_list[i]; 259 if (bi == NULL) 260 continue; 261 if (hdr->e_machine == bi->machine && (bi->flags & 262 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 263 ret = __elfN(check_note)(imgp, bi->brand_note, osrel); 264 if (ret) 265 return (bi); 266 } 267 } 268 269 /* If the executable has a brand, search for it in the brand list. */ 270 for (i = 0; i < MAX_BRANDS; i++) { 271 bi = elf_brand_list[i]; 272 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 273 continue; 274 if (hdr->e_machine == bi->machine && 275 (hdr->e_ident[EI_OSABI] == bi->brand || 276 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 277 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 278 return (bi); 279 } 280 281 /* Lacking a known brand, search for a recognized interpreter. */ 282 if (interp != NULL) { 283 for (i = 0; i < MAX_BRANDS; i++) { 284 bi = elf_brand_list[i]; 285 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 286 continue; 287 if (hdr->e_machine == bi->machine && 288 strcmp(interp, bi->interp_path) == 0) 289 return (bi); 290 } 291 } 292 293 /* Lacking a recognized interpreter, try the default brand */ 294 for (i = 0; i < MAX_BRANDS; i++) { 295 bi = elf_brand_list[i]; 296 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 297 continue; 298 if (hdr->e_machine == bi->machine && 299 __elfN(fallback_brand) == bi->brand) 300 return (bi); 301 } 302 return (NULL); 303 } 304 305 static int 306 __elfN(check_header)(const Elf_Ehdr *hdr) 307 { 308 Elf_Brandinfo *bi; 309 int i; 310 311 if (!IS_ELF(*hdr) || 312 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 313 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 314 hdr->e_ident[EI_VERSION] != EV_CURRENT || 315 hdr->e_phentsize != sizeof(Elf_Phdr) || 316 hdr->e_version != ELF_TARG_VER) 317 return (ENOEXEC); 318 319 /* 320 * Make sure we have at least one brand for this machine. 321 */ 322 323 for (i = 0; i < MAX_BRANDS; i++) { 324 bi = elf_brand_list[i]; 325 if (bi != NULL && bi->machine == hdr->e_machine) 326 break; 327 } 328 if (i == MAX_BRANDS) 329 return (ENOEXEC); 330 331 return (0); 332 } 333 334 static int 335 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 336 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 337 { 338 struct sf_buf *sf; 339 int error; 340 vm_offset_t off; 341 342 /* 343 * Create the page if it doesn't exist yet. Ignore errors. 344 */ 345 vm_map_lock(map); 346 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), 347 VM_PROT_ALL, VM_PROT_ALL, 0); 348 vm_map_unlock(map); 349 350 /* 351 * Find the page from the underlying object. 352 */ 353 if (object) { 354 sf = vm_imgact_map_page(object, offset); 355 if (sf == NULL) 356 return (KERN_FAILURE); 357 off = offset - trunc_page(offset); 358 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 359 end - start); 360 vm_imgact_unmap_page(sf); 361 if (error) { 362 return (KERN_FAILURE); 363 } 364 } 365 366 return (KERN_SUCCESS); 367 } 368 369 static int 370 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 371 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) 372 { 373 struct sf_buf *sf; 374 vm_offset_t off; 375 vm_size_t sz; 376 int error, rv; 377 378 if (start != trunc_page(start)) { 379 rv = __elfN(map_partial)(map, object, offset, start, 380 round_page(start), prot); 381 if (rv) 382 return (rv); 383 offset += round_page(start) - start; 384 start = round_page(start); 385 } 386 if (end != round_page(end)) { 387 rv = __elfN(map_partial)(map, object, offset + 388 trunc_page(end) - start, trunc_page(end), end, prot); 389 if (rv) 390 return (rv); 391 end = trunc_page(end); 392 } 393 if (end > start) { 394 if (offset & PAGE_MASK) { 395 /* 396 * The mapping is not page aligned. This means we have 397 * to copy the data. Sigh. 398 */ 399 rv = vm_map_find(map, NULL, 0, &start, end - start, 400 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0); 401 if (rv) 402 return (rv); 403 if (object == NULL) 404 return (KERN_SUCCESS); 405 for (; start < end; start += sz) { 406 sf = vm_imgact_map_page(object, offset); 407 if (sf == NULL) 408 return (KERN_FAILURE); 409 off = offset - trunc_page(offset); 410 sz = end - start; 411 if (sz > PAGE_SIZE - off) 412 sz = PAGE_SIZE - off; 413 error = copyout((caddr_t)sf_buf_kva(sf) + off, 414 (caddr_t)start, sz); 415 vm_imgact_unmap_page(sf); 416 if (error) { 417 return (KERN_FAILURE); 418 } 419 offset += sz; 420 } 421 rv = KERN_SUCCESS; 422 } else { 423 vm_object_reference(object); 424 vm_map_lock(map); 425 rv = vm_map_insert(map, object, offset, start, end, 426 prot, VM_PROT_ALL, cow); 427 vm_map_unlock(map); 428 if (rv != KERN_SUCCESS) 429 vm_object_deallocate(object); 430 } 431 return (rv); 432 } else { 433 return (KERN_SUCCESS); 434 } 435 } 436 437 static int 438 __elfN(load_section)(struct vmspace *vmspace, 439 vm_object_t object, vm_offset_t offset, 440 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 441 size_t pagesize) 442 { 443 struct sf_buf *sf; 444 size_t map_len; 445 vm_offset_t map_addr; 446 int error, rv, cow; 447 size_t copy_len; 448 vm_offset_t file_addr; 449 450 /* 451 * It's necessary to fail if the filsz + offset taken from the 452 * header is greater than the actual file pager object's size. 453 * If we were to allow this, then the vm_map_find() below would 454 * walk right off the end of the file object and into the ether. 455 * 456 * While I'm here, might as well check for something else that 457 * is invalid: filsz cannot be greater than memsz. 458 */ 459 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 460 filsz > memsz) { 461 uprintf("elf_load_section: truncated ELF file\n"); 462 return (ENOEXEC); 463 } 464 465 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 466 file_addr = trunc_page_ps(offset, pagesize); 467 468 /* 469 * We have two choices. We can either clear the data in the last page 470 * of an oversized mapping, or we can start the anon mapping a page 471 * early and copy the initialized data into that first page. We 472 * choose the second.. 473 */ 474 if (memsz > filsz) 475 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 476 else 477 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 478 479 if (map_len != 0) { 480 /* cow flags: don't dump readonly sections in core */ 481 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 482 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 483 484 rv = __elfN(map_insert)(&vmspace->vm_map, 485 object, 486 file_addr, /* file offset */ 487 map_addr, /* virtual start */ 488 map_addr + map_len,/* virtual end */ 489 prot, 490 cow); 491 if (rv != KERN_SUCCESS) 492 return (EINVAL); 493 494 /* we can stop now if we've covered it all */ 495 if (memsz == filsz) { 496 return (0); 497 } 498 } 499 500 501 /* 502 * We have to get the remaining bit of the file into the first part 503 * of the oversized map segment. This is normally because the .data 504 * segment in the file is extended to provide bss. It's a neat idea 505 * to try and save a page, but it's a pain in the behind to implement. 506 */ 507 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 508 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 509 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 510 map_addr; 511 512 /* This had damn well better be true! */ 513 if (map_len != 0) { 514 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr, 515 map_addr + map_len, VM_PROT_ALL, 0); 516 if (rv != KERN_SUCCESS) { 517 return (EINVAL); 518 } 519 } 520 521 if (copy_len != 0) { 522 vm_offset_t off; 523 524 sf = vm_imgact_map_page(object, offset + filsz); 525 if (sf == NULL) 526 return (EIO); 527 528 /* send the page fragment to user space */ 529 off = trunc_page_ps(offset + filsz, pagesize) - 530 trunc_page(offset + filsz); 531 error = copyout((caddr_t)sf_buf_kva(sf) + off, 532 (caddr_t)map_addr, copy_len); 533 vm_imgact_unmap_page(sf); 534 if (error) { 535 return (error); 536 } 537 } 538 539 /* 540 * set it to the specified protection. 541 * XXX had better undo the damage from pasting over the cracks here! 542 */ 543 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr), 544 round_page(map_addr + map_len), prot, FALSE); 545 546 return (0); 547 } 548 549 /* 550 * Load the file "file" into memory. It may be either a shared object 551 * or an executable. 552 * 553 * The "addr" reference parameter is in/out. On entry, it specifies 554 * the address where a shared object should be loaded. If the file is 555 * an executable, this value is ignored. On exit, "addr" specifies 556 * where the file was actually loaded. 557 * 558 * The "entry" reference parameter is out only. On exit, it specifies 559 * the entry point for the loaded file. 560 */ 561 static int 562 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 563 u_long *entry, size_t pagesize) 564 { 565 struct { 566 struct nameidata nd; 567 struct vattr attr; 568 struct image_params image_params; 569 } *tempdata; 570 const Elf_Ehdr *hdr = NULL; 571 const Elf_Phdr *phdr = NULL; 572 struct nameidata *nd; 573 struct vmspace *vmspace = p->p_vmspace; 574 struct vattr *attr; 575 struct image_params *imgp; 576 vm_prot_t prot; 577 u_long rbase; 578 u_long base_addr = 0; 579 int vfslocked, error, i, numsegs; 580 581 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 582 nd = &tempdata->nd; 583 attr = &tempdata->attr; 584 imgp = &tempdata->image_params; 585 586 /* 587 * Initialize part of the common data 588 */ 589 imgp->proc = p; 590 imgp->attr = attr; 591 imgp->firstpage = NULL; 592 imgp->image_header = NULL; 593 imgp->object = NULL; 594 imgp->execlabel = NULL; 595 596 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, 597 curthread); 598 vfslocked = 0; 599 if ((error = namei(nd)) != 0) { 600 nd->ni_vp = NULL; 601 goto fail; 602 } 603 vfslocked = NDHASGIANT(nd); 604 NDFREE(nd, NDF_ONLY_PNBUF); 605 imgp->vp = nd->ni_vp; 606 607 /* 608 * Check permissions, modes, uid, etc on the file, and "open" it. 609 */ 610 error = exec_check_permissions(imgp); 611 if (error) 612 goto fail; 613 614 error = exec_map_first_page(imgp); 615 if (error) 616 goto fail; 617 618 /* 619 * Also make certain that the interpreter stays the same, so set 620 * its VV_TEXT flag, too. 621 */ 622 nd->ni_vp->v_vflag |= VV_TEXT; 623 624 imgp->object = nd->ni_vp->v_object; 625 626 hdr = (const Elf_Ehdr *)imgp->image_header; 627 if ((error = __elfN(check_header)(hdr)) != 0) 628 goto fail; 629 if (hdr->e_type == ET_DYN) 630 rbase = *addr; 631 else if (hdr->e_type == ET_EXEC) 632 rbase = 0; 633 else { 634 error = ENOEXEC; 635 goto fail; 636 } 637 638 /* Only support headers that fit within first page for now */ 639 /* (multiplication of two Elf_Half fields will not overflow) */ 640 if ((hdr->e_phoff > PAGE_SIZE) || 641 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 642 error = ENOEXEC; 643 goto fail; 644 } 645 646 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 647 if (!aligned(phdr, Elf_Addr)) { 648 error = ENOEXEC; 649 goto fail; 650 } 651 652 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 653 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 654 /* Loadable segment */ 655 prot = __elfN(trans_prot)(phdr[i].p_flags); 656 if ((error = __elfN(load_section)(vmspace, 657 imgp->object, phdr[i].p_offset, 658 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 659 phdr[i].p_memsz, phdr[i].p_filesz, prot, 660 pagesize)) != 0) 661 goto fail; 662 /* 663 * Establish the base address if this is the 664 * first segment. 665 */ 666 if (numsegs == 0) 667 base_addr = trunc_page(phdr[i].p_vaddr + 668 rbase); 669 numsegs++; 670 } 671 } 672 *addr = base_addr; 673 *entry = (unsigned long)hdr->e_entry + rbase; 674 675 fail: 676 if (imgp->firstpage) 677 exec_unmap_first_page(imgp); 678 679 if (nd->ni_vp) 680 vput(nd->ni_vp); 681 682 VFS_UNLOCK_GIANT(vfslocked); 683 free(tempdata, M_TEMP); 684 685 return (error); 686 } 687 688 static int 689 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 690 { 691 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 692 const Elf_Phdr *phdr; 693 Elf_Auxargs *elf_auxargs; 694 struct vmspace *vmspace; 695 vm_prot_t prot; 696 u_long text_size = 0, data_size = 0, total_size = 0; 697 u_long text_addr = 0, data_addr = 0; 698 u_long seg_size, seg_addr; 699 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0; 700 int32_t osrel = 0; 701 int error = 0, i, n; 702 const char *interp = NULL, *newinterp = NULL; 703 Elf_Brandinfo *brand_info; 704 char *path; 705 struct sysentvec *sv; 706 707 /* 708 * Do we have a valid ELF header ? 709 * 710 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 711 * if particular brand doesn't support it. 712 */ 713 if (__elfN(check_header)(hdr) != 0 || 714 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 715 return (-1); 716 717 /* 718 * From here on down, we return an errno, not -1, as we've 719 * detected an ELF file. 720 */ 721 722 if ((hdr->e_phoff > PAGE_SIZE) || 723 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 724 /* Only support headers in first page for now */ 725 return (ENOEXEC); 726 } 727 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 728 if (!aligned(phdr, Elf_Addr)) 729 return (ENOEXEC); 730 n = 0; 731 baddr = 0; 732 for (i = 0; i < hdr->e_phnum; i++) { 733 switch (phdr[i].p_type) { 734 case PT_LOAD: 735 if (n == 0) 736 baddr = phdr[i].p_vaddr; 737 n++; 738 break; 739 case PT_INTERP: 740 /* Path to interpreter */ 741 if (phdr[i].p_filesz > MAXPATHLEN || 742 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) 743 return (ENOEXEC); 744 interp = imgp->image_header + phdr[i].p_offset; 745 break; 746 case PT_GNU_STACK: 747 if (__elfN(nxstack)) 748 imgp->stack_prot = 749 __elfN(trans_prot)(phdr[i].p_flags); 750 break; 751 } 752 } 753 754 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel); 755 if (brand_info == NULL) { 756 uprintf("ELF binary type \"%u\" not known.\n", 757 hdr->e_ident[EI_OSABI]); 758 return (ENOEXEC); 759 } 760 if (hdr->e_type == ET_DYN) { 761 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) 762 return (ENOEXEC); 763 /* 764 * Honour the base load address from the dso if it is 765 * non-zero for some reason. 766 */ 767 if (baddr == 0) 768 et_dyn_addr = ET_DYN_LOAD_ADDR; 769 else 770 et_dyn_addr = 0; 771 } else 772 et_dyn_addr = 0; 773 sv = brand_info->sysvec; 774 if (interp != NULL && brand_info->interp_newpath != NULL) 775 newinterp = brand_info->interp_newpath; 776 777 /* 778 * Avoid a possible deadlock if the current address space is destroyed 779 * and that address space maps the locked vnode. In the common case, 780 * the locked vnode's v_usecount is decremented but remains greater 781 * than zero. Consequently, the vnode lock is not needed by vrele(). 782 * However, in cases where the vnode lock is external, such as nullfs, 783 * v_usecount may become zero. 784 */ 785 VOP_UNLOCK(imgp->vp, 0); 786 787 error = exec_new_vmspace(imgp, sv); 788 imgp->proc->p_sysent = sv; 789 790 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 791 if (error) 792 return (error); 793 794 vmspace = imgp->proc->p_vmspace; 795 796 for (i = 0; i < hdr->e_phnum; i++) { 797 switch (phdr[i].p_type) { 798 case PT_LOAD: /* Loadable segment */ 799 if (phdr[i].p_memsz == 0) 800 break; 801 prot = __elfN(trans_prot)(phdr[i].p_flags); 802 803 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER) 804 /* 805 * Some x86 binaries assume read == executable, 806 * notably the M3 runtime and therefore cvsup 807 */ 808 if (prot & VM_PROT_READ) 809 prot |= VM_PROT_EXECUTE; 810 #endif 811 812 if ((error = __elfN(load_section)(vmspace, 813 imgp->object, phdr[i].p_offset, 814 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 815 phdr[i].p_memsz, phdr[i].p_filesz, prot, 816 sv->sv_pagesize)) != 0) 817 return (error); 818 819 /* 820 * If this segment contains the program headers, 821 * remember their virtual address for the AT_PHDR 822 * aux entry. Static binaries don't usually include 823 * a PT_PHDR entry. 824 */ 825 if (phdr[i].p_offset == 0 && 826 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 827 <= phdr[i].p_filesz) 828 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 829 et_dyn_addr; 830 831 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 832 seg_size = round_page(phdr[i].p_memsz + 833 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 834 835 /* 836 * Make the largest executable segment the official 837 * text segment and all others data. 838 * 839 * Note that obreak() assumes that data_addr + 840 * data_size == end of data load area, and the ELF 841 * file format expects segments to be sorted by 842 * address. If multiple data segments exist, the 843 * last one will be used. 844 */ 845 846 if (phdr[i].p_flags & PF_X && text_size < seg_size) { 847 text_size = seg_size; 848 text_addr = seg_addr; 849 } else { 850 data_size = seg_size; 851 data_addr = seg_addr; 852 } 853 total_size += seg_size; 854 break; 855 case PT_PHDR: /* Program header table info */ 856 proghdr = phdr[i].p_vaddr + et_dyn_addr; 857 break; 858 default: 859 break; 860 } 861 } 862 863 if (data_addr == 0 && data_size == 0) { 864 data_addr = text_addr; 865 data_size = text_size; 866 } 867 868 entry = (u_long)hdr->e_entry + et_dyn_addr; 869 870 /* 871 * Check limits. It should be safe to check the 872 * limits after loading the segments since we do 873 * not actually fault in all the segments pages. 874 */ 875 PROC_LOCK(imgp->proc); 876 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 877 text_size > maxtsiz || 878 total_size > lim_cur(imgp->proc, RLIMIT_VMEM) || 879 racct_set(imgp->proc, RACCT_DATA, data_size) != 0 || 880 racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) { 881 PROC_UNLOCK(imgp->proc); 882 return (ENOMEM); 883 } 884 885 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 886 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 887 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 888 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 889 890 /* 891 * We load the dynamic linker where a userland call 892 * to mmap(0, ...) would put it. The rationale behind this 893 * calculation is that it leaves room for the heap to grow to 894 * its maximum allowed size. 895 */ 896 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr + 897 lim_max(imgp->proc, RLIMIT_DATA)); 898 PROC_UNLOCK(imgp->proc); 899 900 imgp->entry_addr = entry; 901 902 if (interp != NULL) { 903 int have_interp = FALSE; 904 VOP_UNLOCK(imgp->vp, 0); 905 if (brand_info->emul_path != NULL && 906 brand_info->emul_path[0] != '\0') { 907 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 908 snprintf(path, MAXPATHLEN, "%s%s", 909 brand_info->emul_path, interp); 910 error = __elfN(load_file)(imgp->proc, path, &addr, 911 &imgp->entry_addr, sv->sv_pagesize); 912 free(path, M_TEMP); 913 if (error == 0) 914 have_interp = TRUE; 915 } 916 if (!have_interp && newinterp != NULL) { 917 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 918 &imgp->entry_addr, sv->sv_pagesize); 919 if (error == 0) 920 have_interp = TRUE; 921 } 922 if (!have_interp) { 923 error = __elfN(load_file)(imgp->proc, interp, &addr, 924 &imgp->entry_addr, sv->sv_pagesize); 925 } 926 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 927 if (error != 0) { 928 uprintf("ELF interpreter %s not found\n", interp); 929 return (error); 930 } 931 } else 932 addr = et_dyn_addr; 933 934 /* 935 * Construct auxargs table (used by the fixup routine) 936 */ 937 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 938 elf_auxargs->execfd = -1; 939 elf_auxargs->phdr = proghdr; 940 elf_auxargs->phent = hdr->e_phentsize; 941 elf_auxargs->phnum = hdr->e_phnum; 942 elf_auxargs->pagesz = PAGE_SIZE; 943 elf_auxargs->base = addr; 944 elf_auxargs->flags = 0; 945 elf_auxargs->entry = entry; 946 947 imgp->auxargs = elf_auxargs; 948 imgp->interpreted = 0; 949 imgp->reloc_base = addr; 950 imgp->proc->p_osrel = osrel; 951 952 return (error); 953 } 954 955 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 956 957 int 958 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 959 { 960 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 961 Elf_Addr *base; 962 Elf_Addr *pos; 963 964 base = (Elf_Addr *)*stack_base; 965 pos = base + (imgp->args->argc + imgp->args->envc + 2); 966 967 if (args->execfd != -1) 968 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 969 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 970 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 971 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 972 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 973 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 974 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 975 AUXARGS_ENTRY(pos, AT_BASE, args->base); 976 if (imgp->execpathp != 0) 977 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 978 AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate); 979 if (imgp->canary != 0) { 980 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary); 981 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 982 } 983 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 984 if (imgp->pagesizes != 0) { 985 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes); 986 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 987 } 988 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 989 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 990 imgp->sysent->sv_stackprot); 991 AUXARGS_ENTRY(pos, AT_NULL, 0); 992 993 free(imgp->auxargs, M_TEMP); 994 imgp->auxargs = NULL; 995 996 base--; 997 suword(base, (long)imgp->args->argc); 998 *stack_base = (register_t *)base; 999 return (0); 1000 } 1001 1002 /* 1003 * Code for generating ELF core dumps. 1004 */ 1005 1006 typedef void (*segment_callback)(vm_map_entry_t, void *); 1007 1008 /* Closure for cb_put_phdr(). */ 1009 struct phdr_closure { 1010 Elf_Phdr *phdr; /* Program header to fill in */ 1011 Elf_Off offset; /* Offset of segment in core file */ 1012 }; 1013 1014 /* Closure for cb_size_segment(). */ 1015 struct sseg_closure { 1016 int count; /* Count of writable segments. */ 1017 size_t size; /* Total size of all writable segments. */ 1018 }; 1019 1020 static void cb_put_phdr(vm_map_entry_t, void *); 1021 static void cb_size_segment(vm_map_entry_t, void *); 1022 static void each_writable_segment(struct thread *, segment_callback, void *); 1023 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 1024 int, void *, size_t, gzFile); 1025 static void __elfN(puthdr)(struct thread *, void *, size_t *, int); 1026 static void __elfN(putnote)(void *, size_t *, const char *, int, 1027 const void *, size_t); 1028 1029 #ifdef COMPRESS_USER_CORES 1030 extern int compress_user_cores; 1031 extern int compress_user_cores_gzlevel; 1032 #endif 1033 1034 static int 1035 core_output(struct vnode *vp, void *base, size_t len, off_t offset, 1036 struct ucred *active_cred, struct ucred *file_cred, 1037 struct thread *td, char *core_buf, gzFile gzfile) { 1038 1039 int error; 1040 if (gzfile) { 1041 #ifdef COMPRESS_USER_CORES 1042 error = compress_core(gzfile, base, core_buf, len, td); 1043 #else 1044 panic("shouldn't be here"); 1045 #endif 1046 } else { 1047 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset, 1048 UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred, 1049 NULL, td); 1050 } 1051 return (error); 1052 } 1053 1054 int 1055 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1056 { 1057 struct ucred *cred = td->td_ucred; 1058 int error = 0; 1059 struct sseg_closure seginfo; 1060 void *hdr; 1061 size_t hdrsize; 1062 1063 gzFile gzfile = Z_NULL; 1064 char *core_buf = NULL; 1065 #ifdef COMPRESS_USER_CORES 1066 char gzopen_flags[8]; 1067 char *p; 1068 int doing_compress = flags & IMGACT_CORE_COMPRESS; 1069 #endif 1070 1071 hdr = NULL; 1072 1073 #ifdef COMPRESS_USER_CORES 1074 if (doing_compress) { 1075 p = gzopen_flags; 1076 *p++ = 'w'; 1077 if (compress_user_cores_gzlevel >= 0 && 1078 compress_user_cores_gzlevel <= 9) 1079 *p++ = '0' + compress_user_cores_gzlevel; 1080 *p = 0; 1081 gzfile = gz_open("", gzopen_flags, vp); 1082 if (gzfile == Z_NULL) { 1083 error = EFAULT; 1084 goto done; 1085 } 1086 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1087 if (!core_buf) { 1088 error = ENOMEM; 1089 goto done; 1090 } 1091 } 1092 #endif 1093 1094 /* Size the program segments. */ 1095 seginfo.count = 0; 1096 seginfo.size = 0; 1097 each_writable_segment(td, cb_size_segment, &seginfo); 1098 1099 /* 1100 * Calculate the size of the core file header area by making 1101 * a dry run of generating it. Nothing is written, but the 1102 * size is calculated. 1103 */ 1104 hdrsize = 0; 1105 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count); 1106 1107 PROC_LOCK(td->td_proc); 1108 error = racct_add(td->td_proc, RACCT_CORE, hdrsize + seginfo.size); 1109 PROC_UNLOCK(td->td_proc); 1110 if (error != 0) { 1111 error = EFAULT; 1112 goto done; 1113 } 1114 if (hdrsize + seginfo.size >= limit) { 1115 error = EFAULT; 1116 goto done; 1117 } 1118 1119 /* 1120 * Allocate memory for building the header, fill it up, 1121 * and write it out. 1122 */ 1123 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1124 if (hdr == NULL) { 1125 error = EINVAL; 1126 goto done; 1127 } 1128 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize, 1129 gzfile); 1130 1131 /* Write the contents of all of the writable segments. */ 1132 if (error == 0) { 1133 Elf_Phdr *php; 1134 off_t offset; 1135 int i; 1136 1137 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1138 offset = hdrsize; 1139 for (i = 0; i < seginfo.count; i++) { 1140 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr, 1141 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile); 1142 if (error != 0) 1143 break; 1144 offset += php->p_filesz; 1145 php++; 1146 } 1147 } 1148 if (error) { 1149 log(LOG_WARNING, 1150 "Failed to write core file for process %s (error %d)\n", 1151 curproc->p_comm, error); 1152 } 1153 1154 done: 1155 #ifdef COMPRESS_USER_CORES 1156 if (core_buf) 1157 free(core_buf, M_TEMP); 1158 if (gzfile) 1159 gzclose(gzfile); 1160 #endif 1161 1162 free(hdr, M_TEMP); 1163 1164 return (error); 1165 } 1166 1167 /* 1168 * A callback for each_writable_segment() to write out the segment's 1169 * program header entry. 1170 */ 1171 static void 1172 cb_put_phdr(entry, closure) 1173 vm_map_entry_t entry; 1174 void *closure; 1175 { 1176 struct phdr_closure *phc = (struct phdr_closure *)closure; 1177 Elf_Phdr *phdr = phc->phdr; 1178 1179 phc->offset = round_page(phc->offset); 1180 1181 phdr->p_type = PT_LOAD; 1182 phdr->p_offset = phc->offset; 1183 phdr->p_vaddr = entry->start; 1184 phdr->p_paddr = 0; 1185 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1186 phdr->p_align = PAGE_SIZE; 1187 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1188 1189 phc->offset += phdr->p_filesz; 1190 phc->phdr++; 1191 } 1192 1193 /* 1194 * A callback for each_writable_segment() to gather information about 1195 * the number of segments and their total size. 1196 */ 1197 static void 1198 cb_size_segment(entry, closure) 1199 vm_map_entry_t entry; 1200 void *closure; 1201 { 1202 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1203 1204 ssc->count++; 1205 ssc->size += entry->end - entry->start; 1206 } 1207 1208 /* 1209 * For each writable segment in the process's memory map, call the given 1210 * function with a pointer to the map entry and some arbitrary 1211 * caller-supplied data. 1212 */ 1213 static void 1214 each_writable_segment(td, func, closure) 1215 struct thread *td; 1216 segment_callback func; 1217 void *closure; 1218 { 1219 struct proc *p = td->td_proc; 1220 vm_map_t map = &p->p_vmspace->vm_map; 1221 vm_map_entry_t entry; 1222 vm_object_t backing_object, object; 1223 boolean_t ignore_entry; 1224 1225 vm_map_lock_read(map); 1226 for (entry = map->header.next; entry != &map->header; 1227 entry = entry->next) { 1228 /* 1229 * Don't dump inaccessible mappings, deal with legacy 1230 * coredump mode. 1231 * 1232 * Note that read-only segments related to the elf binary 1233 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1234 * need to arbitrarily ignore such segments. 1235 */ 1236 if (elf_legacy_coredump) { 1237 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1238 continue; 1239 } else { 1240 if ((entry->protection & VM_PROT_ALL) == 0) 1241 continue; 1242 } 1243 1244 /* 1245 * Dont include memory segment in the coredump if 1246 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1247 * madvise(2). Do not dump submaps (i.e. parts of the 1248 * kernel map). 1249 */ 1250 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1251 continue; 1252 1253 if ((object = entry->object.vm_object) == NULL) 1254 continue; 1255 1256 /* Ignore memory-mapped devices and such things. */ 1257 VM_OBJECT_LOCK(object); 1258 while ((backing_object = object->backing_object) != NULL) { 1259 VM_OBJECT_LOCK(backing_object); 1260 VM_OBJECT_UNLOCK(object); 1261 object = backing_object; 1262 } 1263 ignore_entry = object->type != OBJT_DEFAULT && 1264 object->type != OBJT_SWAP && object->type != OBJT_VNODE; 1265 VM_OBJECT_UNLOCK(object); 1266 if (ignore_entry) 1267 continue; 1268 1269 (*func)(entry, closure); 1270 } 1271 vm_map_unlock_read(map); 1272 } 1273 1274 /* 1275 * Write the core file header to the file, including padding up to 1276 * the page boundary. 1277 */ 1278 static int 1279 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize, gzfile) 1280 struct thread *td; 1281 struct vnode *vp; 1282 struct ucred *cred; 1283 int numsegs; 1284 size_t hdrsize; 1285 void *hdr; 1286 gzFile gzfile; 1287 { 1288 size_t off; 1289 1290 /* Fill in the header. */ 1291 bzero(hdr, hdrsize); 1292 off = 0; 1293 __elfN(puthdr)(td, hdr, &off, numsegs); 1294 1295 if (!gzfile) { 1296 /* Write it to the core file. */ 1297 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1298 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1299 td)); 1300 } else { 1301 #ifdef COMPRESS_USER_CORES 1302 if (gzwrite(gzfile, hdr, hdrsize) != hdrsize) { 1303 log(LOG_WARNING, 1304 "Failed to compress core file header for process" 1305 " %s.\n", curproc->p_comm); 1306 return (EFAULT); 1307 } 1308 else { 1309 return (0); 1310 } 1311 #else 1312 panic("shouldn't be here"); 1313 #endif 1314 } 1315 } 1316 1317 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1318 #include <compat/freebsd32/freebsd32.h> 1319 1320 typedef struct prstatus32 elf_prstatus_t; 1321 typedef struct prpsinfo32 elf_prpsinfo_t; 1322 typedef struct fpreg32 elf_prfpregset_t; 1323 typedef struct fpreg32 elf_fpregset_t; 1324 typedef struct reg32 elf_gregset_t; 1325 typedef struct thrmisc32 elf_thrmisc_t; 1326 #else 1327 typedef prstatus_t elf_prstatus_t; 1328 typedef prpsinfo_t elf_prpsinfo_t; 1329 typedef prfpregset_t elf_prfpregset_t; 1330 typedef prfpregset_t elf_fpregset_t; 1331 typedef gregset_t elf_gregset_t; 1332 typedef thrmisc_t elf_thrmisc_t; 1333 #endif 1334 1335 static void 1336 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs) 1337 { 1338 struct { 1339 elf_prstatus_t status; 1340 elf_prfpregset_t fpregset; 1341 elf_prpsinfo_t psinfo; 1342 elf_thrmisc_t thrmisc; 1343 } *tempdata; 1344 elf_prstatus_t *status; 1345 elf_prfpregset_t *fpregset; 1346 elf_prpsinfo_t *psinfo; 1347 elf_thrmisc_t *thrmisc; 1348 struct proc *p; 1349 struct thread *thr; 1350 size_t ehoff, noteoff, notesz, phoff; 1351 1352 p = td->td_proc; 1353 1354 ehoff = *off; 1355 *off += sizeof(Elf_Ehdr); 1356 1357 phoff = *off; 1358 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1359 1360 noteoff = *off; 1361 /* 1362 * Don't allocate space for the notes if we're just calculating 1363 * the size of the header. We also don't collect the data. 1364 */ 1365 if (dst != NULL) { 1366 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK); 1367 status = &tempdata->status; 1368 fpregset = &tempdata->fpregset; 1369 psinfo = &tempdata->psinfo; 1370 thrmisc = &tempdata->thrmisc; 1371 } else { 1372 tempdata = NULL; 1373 status = NULL; 1374 fpregset = NULL; 1375 psinfo = NULL; 1376 thrmisc = NULL; 1377 } 1378 1379 if (dst != NULL) { 1380 psinfo->pr_version = PRPSINFO_VERSION; 1381 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1382 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1383 /* 1384 * XXX - We don't fill in the command line arguments properly 1385 * yet. 1386 */ 1387 strlcpy(psinfo->pr_psargs, p->p_comm, 1388 sizeof(psinfo->pr_psargs)); 1389 } 1390 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1391 sizeof *psinfo); 1392 1393 /* 1394 * To have the debugger select the right thread (LWP) as the initial 1395 * thread, we dump the state of the thread passed to us in td first. 1396 * This is the thread that causes the core dump and thus likely to 1397 * be the right thread one wants to have selected in the debugger. 1398 */ 1399 thr = td; 1400 while (thr != NULL) { 1401 if (dst != NULL) { 1402 status->pr_version = PRSTATUS_VERSION; 1403 status->pr_statussz = sizeof(elf_prstatus_t); 1404 status->pr_gregsetsz = sizeof(elf_gregset_t); 1405 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1406 status->pr_osreldate = osreldate; 1407 status->pr_cursig = p->p_sig; 1408 status->pr_pid = thr->td_tid; 1409 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1410 fill_regs32(thr, &status->pr_reg); 1411 fill_fpregs32(thr, fpregset); 1412 #else 1413 fill_regs(thr, &status->pr_reg); 1414 fill_fpregs(thr, fpregset); 1415 #endif 1416 memset(&thrmisc->_pad, 0, sizeof (thrmisc->_pad)); 1417 strcpy(thrmisc->pr_tname, thr->td_name); 1418 } 1419 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1420 sizeof *status); 1421 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1422 sizeof *fpregset); 1423 __elfN(putnote)(dst, off, "FreeBSD", NT_THRMISC, thrmisc, 1424 sizeof *thrmisc); 1425 /* 1426 * Allow for MD specific notes, as well as any MD 1427 * specific preparations for writing MI notes. 1428 */ 1429 __elfN(dump_thread)(thr, dst, off); 1430 1431 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1432 TAILQ_NEXT(thr, td_plist); 1433 if (thr == td) 1434 thr = TAILQ_NEXT(thr, td_plist); 1435 } 1436 1437 notesz = *off - noteoff; 1438 1439 if (dst != NULL) 1440 free(tempdata, M_TEMP); 1441 1442 /* Align up to a page boundary for the program segments. */ 1443 *off = round_page(*off); 1444 1445 if (dst != NULL) { 1446 Elf_Ehdr *ehdr; 1447 Elf_Phdr *phdr; 1448 struct phdr_closure phc; 1449 1450 /* 1451 * Fill in the ELF header. 1452 */ 1453 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1454 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1455 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1456 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1457 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1458 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1459 ehdr->e_ident[EI_DATA] = ELF_DATA; 1460 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1461 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1462 ehdr->e_ident[EI_ABIVERSION] = 0; 1463 ehdr->e_ident[EI_PAD] = 0; 1464 ehdr->e_type = ET_CORE; 1465 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1466 ehdr->e_machine = ELF_ARCH32; 1467 #else 1468 ehdr->e_machine = ELF_ARCH; 1469 #endif 1470 ehdr->e_version = EV_CURRENT; 1471 ehdr->e_entry = 0; 1472 ehdr->e_phoff = phoff; 1473 ehdr->e_flags = 0; 1474 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1475 ehdr->e_phentsize = sizeof(Elf_Phdr); 1476 ehdr->e_phnum = numsegs + 1; 1477 ehdr->e_shentsize = sizeof(Elf_Shdr); 1478 ehdr->e_shnum = 0; 1479 ehdr->e_shstrndx = SHN_UNDEF; 1480 1481 /* 1482 * Fill in the program header entries. 1483 */ 1484 phdr = (Elf_Phdr *)((char *)dst + phoff); 1485 1486 /* The note segement. */ 1487 phdr->p_type = PT_NOTE; 1488 phdr->p_offset = noteoff; 1489 phdr->p_vaddr = 0; 1490 phdr->p_paddr = 0; 1491 phdr->p_filesz = notesz; 1492 phdr->p_memsz = 0; 1493 phdr->p_flags = 0; 1494 phdr->p_align = 0; 1495 phdr++; 1496 1497 /* All the writable segments from the program. */ 1498 phc.phdr = phdr; 1499 phc.offset = *off; 1500 each_writable_segment(td, cb_put_phdr, &phc); 1501 } 1502 } 1503 1504 static void 1505 __elfN(putnote)(void *dst, size_t *off, const char *name, int type, 1506 const void *desc, size_t descsz) 1507 { 1508 Elf_Note note; 1509 1510 note.n_namesz = strlen(name) + 1; 1511 note.n_descsz = descsz; 1512 note.n_type = type; 1513 if (dst != NULL) 1514 bcopy(¬e, (char *)dst + *off, sizeof note); 1515 *off += sizeof note; 1516 if (dst != NULL) 1517 bcopy(name, (char *)dst + *off, note.n_namesz); 1518 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1519 if (dst != NULL) 1520 bcopy(desc, (char *)dst + *off, note.n_descsz); 1521 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1522 } 1523 1524 /* 1525 * Try to find the appropriate ABI-note section for checknote, 1526 * fetch the osreldate for binary from the ELF OSABI-note. Only the 1527 * first page of the image is searched, the same as for headers. 1528 */ 1529 static boolean_t 1530 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 1531 int32_t *osrel) 1532 { 1533 const Elf_Note *note, *note0, *note_end; 1534 const Elf_Phdr *phdr, *pnote; 1535 const Elf_Ehdr *hdr; 1536 const char *note_name; 1537 int i; 1538 1539 pnote = NULL; 1540 hdr = (const Elf_Ehdr *)imgp->image_header; 1541 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1542 1543 for (i = 0; i < hdr->e_phnum; i++) { 1544 if (phdr[i].p_type == PT_NOTE) { 1545 pnote = &phdr[i]; 1546 break; 1547 } 1548 } 1549 1550 if (pnote == NULL || pnote->p_offset >= PAGE_SIZE || 1551 pnote->p_offset + pnote->p_filesz >= PAGE_SIZE) 1552 return (FALSE); 1553 1554 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset); 1555 note_end = (const Elf_Note *)(imgp->image_header + 1556 pnote->p_offset + pnote->p_filesz); 1557 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 1558 if (!aligned(note, Elf32_Addr)) 1559 return (FALSE); 1560 if (note->n_namesz != checknote->hdr.n_namesz || 1561 note->n_descsz != checknote->hdr.n_descsz || 1562 note->n_type != checknote->hdr.n_type) 1563 goto nextnote; 1564 note_name = (const char *)(note + 1); 1565 if (strncmp(checknote->vendor, note_name, 1566 checknote->hdr.n_namesz) != 0) 1567 goto nextnote; 1568 1569 /* 1570 * Fetch the osreldate for binary 1571 * from the ELF OSABI-note if necessary. 1572 */ 1573 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 1574 checknote->trans_osrel != NULL) 1575 return (checknote->trans_osrel(note, osrel)); 1576 return (TRUE); 1577 1578 nextnote: 1579 note = (const Elf_Note *)((const char *)(note + 1) + 1580 roundup2(note->n_namesz, sizeof(Elf32_Addr)) + 1581 roundup2(note->n_descsz, sizeof(Elf32_Addr))); 1582 } 1583 1584 return (FALSE); 1585 } 1586 1587 /* 1588 * Tell kern_execve.c about it, with a little help from the linker. 1589 */ 1590 static struct execsw __elfN(execsw) = { 1591 __CONCAT(exec_, __elfN(imgact)), 1592 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 1593 }; 1594 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 1595 1596 #ifdef COMPRESS_USER_CORES 1597 /* 1598 * Compress and write out a core segment for a user process. 1599 * 1600 * 'inbuf' is the starting address of a VM segment in the process' address 1601 * space that is to be compressed and written out to the core file. 'dest_buf' 1602 * is a buffer in the kernel's address space. The segment is copied from 1603 * 'inbuf' to 'dest_buf' first before being processed by the compression 1604 * routine gzwrite(). This copying is necessary because the content of the VM 1605 * segment may change between the compression pass and the crc-computation pass 1606 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel. 1607 */ 1608 static int 1609 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len, 1610 struct thread *td) 1611 { 1612 int len_compressed; 1613 int error = 0; 1614 unsigned int chunk_len; 1615 1616 while (len) { 1617 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len; 1618 copyin(inbuf, dest_buf, chunk_len); 1619 len_compressed = gzwrite(file, dest_buf, chunk_len); 1620 1621 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed); 1622 1623 if ((unsigned int)len_compressed != chunk_len) { 1624 log(LOG_WARNING, 1625 "compress_core: length mismatch (0x%x returned, " 1626 "0x%x expected)\n", len_compressed, chunk_len); 1627 EVENTHANDLER_INVOKE(app_coredump_error, td, 1628 "compress_core: length mismatch %x -> %x", 1629 chunk_len, len_compressed); 1630 error = EFAULT; 1631 break; 1632 } 1633 inbuf += chunk_len; 1634 len -= chunk_len; 1635 maybe_yield(); 1636 } 1637 1638 return (error); 1639 } 1640 #endif /* COMPRESS_USER_CORES */ 1641 1642 static vm_prot_t 1643 __elfN(trans_prot)(Elf_Word flags) 1644 { 1645 vm_prot_t prot; 1646 1647 prot = 0; 1648 if (flags & PF_X) 1649 prot |= VM_PROT_EXECUTE; 1650 if (flags & PF_W) 1651 prot |= VM_PROT_WRITE; 1652 if (flags & PF_R) 1653 prot |= VM_PROT_READ; 1654 return (prot); 1655 } 1656 1657 static Elf_Word 1658 __elfN(untrans_prot)(vm_prot_t prot) 1659 { 1660 Elf_Word flags; 1661 1662 flags = 0; 1663 if (prot & VM_PROT_EXECUTE) 1664 flags |= PF_X; 1665 if (prot & VM_PROT_READ) 1666 flags |= PF_R; 1667 if (prot & VM_PROT_WRITE) 1668 flags |= PF_W; 1669 return (flags); 1670 } 1671