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 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include "opt_capsicum.h" 35 #include "opt_compat.h" 36 #include "opt_core.h" 37 38 #include <sys/param.h> 39 #include <sys/capability.h> 40 #include <sys/exec.h> 41 #include <sys/fcntl.h> 42 #include <sys/imgact.h> 43 #include <sys/imgact_elf.h> 44 #include <sys/kernel.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/mount.h> 48 #include <sys/mman.h> 49 #include <sys/namei.h> 50 #include <sys/pioctl.h> 51 #include <sys/proc.h> 52 #include <sys/procfs.h> 53 #include <sys/racct.h> 54 #include <sys/resourcevar.h> 55 #include <sys/rwlock.h> 56 #include <sys/sbuf.h> 57 #include <sys/sf_buf.h> 58 #include <sys/smp.h> 59 #include <sys/systm.h> 60 #include <sys/signalvar.h> 61 #include <sys/stat.h> 62 #include <sys/sx.h> 63 #include <sys/syscall.h> 64 #include <sys/sysctl.h> 65 #include <sys/sysent.h> 66 #include <sys/vnode.h> 67 #include <sys/syslog.h> 68 #include <sys/eventhandler.h> 69 #include <sys/user.h> 70 71 #include <net/zlib.h> 72 73 #include <vm/vm.h> 74 #include <vm/vm_kern.h> 75 #include <vm/vm_param.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_map.h> 78 #include <vm/vm_object.h> 79 #include <vm/vm_extern.h> 80 81 #include <machine/elf.h> 82 #include <machine/md_var.h> 83 84 #define OLD_EI_BRAND 8 85 86 static int __elfN(check_header)(const Elf_Ehdr *hdr); 87 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 88 const char *interp, int interp_name_len, int32_t *osrel); 89 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 90 u_long *entry, size_t pagesize); 91 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset, 92 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 93 size_t pagesize); 94 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 95 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note, 96 int32_t *osrel); 97 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 98 static boolean_t __elfN(check_note)(struct image_params *imgp, 99 Elf_Brandnote *checknote, int32_t *osrel); 100 static vm_prot_t __elfN(trans_prot)(Elf_Word); 101 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 102 103 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 104 ""); 105 106 #ifdef COMPRESS_USER_CORES 107 static int compress_core(gzFile, char *, char *, unsigned int, 108 struct thread * td); 109 #endif 110 #define CORE_BUF_SIZE (16 * 1024) 111 112 int __elfN(fallback_brand) = -1; 113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 114 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 115 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 116 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 117 &__elfN(fallback_brand)); 118 119 static int elf_legacy_coredump = 0; 120 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 121 &elf_legacy_coredump, 0, ""); 122 123 int __elfN(nxstack) = 124 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ 125 1; 126 #else 127 0; 128 #endif 129 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 130 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 131 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack"); 132 133 #if __ELF_WORD_SIZE == 32 134 #if defined(__amd64__) || defined(__ia64__) 135 int i386_read_exec = 0; 136 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 137 "enable execution from readable segments"); 138 #endif 139 #endif 140 141 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 142 143 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 144 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 145 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a)) 146 147 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; 148 149 Elf_Brandnote __elfN(freebsd_brandnote) = { 150 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 151 .hdr.n_descsz = sizeof(int32_t), 152 .hdr.n_type = 1, 153 .vendor = FREEBSD_ABI_VENDOR, 154 .flags = BN_TRANSLATE_OSREL, 155 .trans_osrel = __elfN(freebsd_trans_osrel) 156 }; 157 158 static boolean_t 159 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 160 { 161 uintptr_t p; 162 163 p = (uintptr_t)(note + 1); 164 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 165 *osrel = *(const int32_t *)(p); 166 167 return (TRUE); 168 } 169 170 static const char GNU_ABI_VENDOR[] = "GNU"; 171 static int GNU_KFREEBSD_ABI_DESC = 3; 172 173 Elf_Brandnote __elfN(kfreebsd_brandnote) = { 174 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 175 .hdr.n_descsz = 16, /* XXX at least 16 */ 176 .hdr.n_type = 1, 177 .vendor = GNU_ABI_VENDOR, 178 .flags = BN_TRANSLATE_OSREL, 179 .trans_osrel = kfreebsd_trans_osrel 180 }; 181 182 static boolean_t 183 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 184 { 185 const Elf32_Word *desc; 186 uintptr_t p; 187 188 p = (uintptr_t)(note + 1); 189 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 190 191 desc = (const Elf32_Word *)p; 192 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 193 return (FALSE); 194 195 /* 196 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 197 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 198 */ 199 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 200 201 return (TRUE); 202 } 203 204 int 205 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 206 { 207 int i; 208 209 for (i = 0; i < MAX_BRANDS; i++) { 210 if (elf_brand_list[i] == NULL) { 211 elf_brand_list[i] = entry; 212 break; 213 } 214 } 215 if (i == MAX_BRANDS) { 216 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 217 __func__, entry); 218 return (-1); 219 } 220 return (0); 221 } 222 223 int 224 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 225 { 226 int i; 227 228 for (i = 0; i < MAX_BRANDS; i++) { 229 if (elf_brand_list[i] == entry) { 230 elf_brand_list[i] = NULL; 231 break; 232 } 233 } 234 if (i == MAX_BRANDS) 235 return (-1); 236 return (0); 237 } 238 239 int 240 __elfN(brand_inuse)(Elf_Brandinfo *entry) 241 { 242 struct proc *p; 243 int rval = FALSE; 244 245 sx_slock(&allproc_lock); 246 FOREACH_PROC_IN_SYSTEM(p) { 247 if (p->p_sysent == entry->sysvec) { 248 rval = TRUE; 249 break; 250 } 251 } 252 sx_sunlock(&allproc_lock); 253 254 return (rval); 255 } 256 257 static Elf_Brandinfo * 258 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 259 int interp_name_len, int32_t *osrel) 260 { 261 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 262 Elf_Brandinfo *bi; 263 boolean_t ret; 264 int i; 265 266 /* 267 * We support four types of branding -- (1) the ELF EI_OSABI field 268 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 269 * branding w/in the ELF header, (3) path of the `interp_path' 270 * field, and (4) the ".note.ABI-tag" ELF section. 271 */ 272 273 /* Look for an ".note.ABI-tag" ELF section */ 274 for (i = 0; i < MAX_BRANDS; i++) { 275 bi = elf_brand_list[i]; 276 if (bi == NULL) 277 continue; 278 if (hdr->e_machine == bi->machine && (bi->flags & 279 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 280 ret = __elfN(check_note)(imgp, bi->brand_note, osrel); 281 if (ret) 282 return (bi); 283 } 284 } 285 286 /* If the executable has a brand, search for it in the brand list. */ 287 for (i = 0; i < MAX_BRANDS; i++) { 288 bi = elf_brand_list[i]; 289 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 290 continue; 291 if (hdr->e_machine == bi->machine && 292 (hdr->e_ident[EI_OSABI] == bi->brand || 293 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 294 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 295 return (bi); 296 } 297 298 /* Lacking a known brand, search for a recognized interpreter. */ 299 if (interp != NULL) { 300 for (i = 0; i < MAX_BRANDS; i++) { 301 bi = elf_brand_list[i]; 302 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 303 continue; 304 if (hdr->e_machine == bi->machine && 305 /* ELF image p_filesz includes terminating zero */ 306 strlen(bi->interp_path) + 1 == interp_name_len && 307 strncmp(interp, bi->interp_path, interp_name_len) 308 == 0) 309 return (bi); 310 } 311 } 312 313 /* Lacking a recognized interpreter, try the default brand */ 314 for (i = 0; i < MAX_BRANDS; i++) { 315 bi = elf_brand_list[i]; 316 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 317 continue; 318 if (hdr->e_machine == bi->machine && 319 __elfN(fallback_brand) == bi->brand) 320 return (bi); 321 } 322 return (NULL); 323 } 324 325 static int 326 __elfN(check_header)(const Elf_Ehdr *hdr) 327 { 328 Elf_Brandinfo *bi; 329 int i; 330 331 if (!IS_ELF(*hdr) || 332 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 333 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 334 hdr->e_ident[EI_VERSION] != EV_CURRENT || 335 hdr->e_phentsize != sizeof(Elf_Phdr) || 336 hdr->e_version != ELF_TARG_VER) 337 return (ENOEXEC); 338 339 /* 340 * Make sure we have at least one brand for this machine. 341 */ 342 343 for (i = 0; i < MAX_BRANDS; i++) { 344 bi = elf_brand_list[i]; 345 if (bi != NULL && bi->machine == hdr->e_machine) 346 break; 347 } 348 if (i == MAX_BRANDS) 349 return (ENOEXEC); 350 351 return (0); 352 } 353 354 static int 355 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 356 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 357 { 358 struct sf_buf *sf; 359 int error; 360 vm_offset_t off; 361 362 /* 363 * Create the page if it doesn't exist yet. Ignore errors. 364 */ 365 vm_map_lock(map); 366 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), 367 VM_PROT_ALL, VM_PROT_ALL, 0); 368 vm_map_unlock(map); 369 370 /* 371 * Find the page from the underlying object. 372 */ 373 if (object) { 374 sf = vm_imgact_map_page(object, offset); 375 if (sf == NULL) 376 return (KERN_FAILURE); 377 off = offset - trunc_page(offset); 378 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 379 end - start); 380 vm_imgact_unmap_page(sf); 381 if (error) { 382 return (KERN_FAILURE); 383 } 384 } 385 386 return (KERN_SUCCESS); 387 } 388 389 static int 390 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 391 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) 392 { 393 struct sf_buf *sf; 394 vm_offset_t off; 395 vm_size_t sz; 396 int error, rv; 397 398 if (start != trunc_page(start)) { 399 rv = __elfN(map_partial)(map, object, offset, start, 400 round_page(start), prot); 401 if (rv) 402 return (rv); 403 offset += round_page(start) - start; 404 start = round_page(start); 405 } 406 if (end != round_page(end)) { 407 rv = __elfN(map_partial)(map, object, offset + 408 trunc_page(end) - start, trunc_page(end), end, prot); 409 if (rv) 410 return (rv); 411 end = trunc_page(end); 412 } 413 if (end > start) { 414 if (offset & PAGE_MASK) { 415 /* 416 * The mapping is not page aligned. This means we have 417 * to copy the data. Sigh. 418 */ 419 rv = vm_map_find(map, NULL, 0, &start, end - start, 420 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0); 421 if (rv) 422 return (rv); 423 if (object == NULL) 424 return (KERN_SUCCESS); 425 for (; start < end; start += sz) { 426 sf = vm_imgact_map_page(object, offset); 427 if (sf == NULL) 428 return (KERN_FAILURE); 429 off = offset - trunc_page(offset); 430 sz = end - start; 431 if (sz > PAGE_SIZE - off) 432 sz = PAGE_SIZE - off; 433 error = copyout((caddr_t)sf_buf_kva(sf) + off, 434 (caddr_t)start, sz); 435 vm_imgact_unmap_page(sf); 436 if (error) { 437 return (KERN_FAILURE); 438 } 439 offset += sz; 440 } 441 rv = KERN_SUCCESS; 442 } else { 443 vm_object_reference(object); 444 vm_map_lock(map); 445 rv = vm_map_insert(map, object, offset, start, end, 446 prot, VM_PROT_ALL, cow); 447 vm_map_unlock(map); 448 if (rv != KERN_SUCCESS) 449 vm_object_deallocate(object); 450 } 451 return (rv); 452 } else { 453 return (KERN_SUCCESS); 454 } 455 } 456 457 static int 458 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset, 459 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 460 size_t pagesize) 461 { 462 struct sf_buf *sf; 463 size_t map_len; 464 vm_map_t map; 465 vm_object_t object; 466 vm_offset_t map_addr; 467 int error, rv, cow; 468 size_t copy_len; 469 vm_offset_t file_addr; 470 471 /* 472 * It's necessary to fail if the filsz + offset taken from the 473 * header is greater than the actual file pager object's size. 474 * If we were to allow this, then the vm_map_find() below would 475 * walk right off the end of the file object and into the ether. 476 * 477 * While I'm here, might as well check for something else that 478 * is invalid: filsz cannot be greater than memsz. 479 */ 480 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) { 481 uprintf("elf_load_section: truncated ELF file\n"); 482 return (ENOEXEC); 483 } 484 485 object = imgp->object; 486 map = &imgp->proc->p_vmspace->vm_map; 487 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 488 file_addr = trunc_page_ps(offset, pagesize); 489 490 /* 491 * We have two choices. We can either clear the data in the last page 492 * of an oversized mapping, or we can start the anon mapping a page 493 * early and copy the initialized data into that first page. We 494 * choose the second.. 495 */ 496 if (memsz > filsz) 497 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 498 else 499 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 500 501 if (map_len != 0) { 502 /* cow flags: don't dump readonly sections in core */ 503 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 504 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 505 506 rv = __elfN(map_insert)(map, 507 object, 508 file_addr, /* file offset */ 509 map_addr, /* virtual start */ 510 map_addr + map_len,/* virtual end */ 511 prot, 512 cow); 513 if (rv != KERN_SUCCESS) 514 return (EINVAL); 515 516 /* we can stop now if we've covered it all */ 517 if (memsz == filsz) { 518 return (0); 519 } 520 } 521 522 523 /* 524 * We have to get the remaining bit of the file into the first part 525 * of the oversized map segment. This is normally because the .data 526 * segment in the file is extended to provide bss. It's a neat idea 527 * to try and save a page, but it's a pain in the behind to implement. 528 */ 529 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 530 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 531 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 532 map_addr; 533 534 /* This had damn well better be true! */ 535 if (map_len != 0) { 536 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr + 537 map_len, VM_PROT_ALL, 0); 538 if (rv != KERN_SUCCESS) { 539 return (EINVAL); 540 } 541 } 542 543 if (copy_len != 0) { 544 vm_offset_t off; 545 546 sf = vm_imgact_map_page(object, offset + filsz); 547 if (sf == NULL) 548 return (EIO); 549 550 /* send the page fragment to user space */ 551 off = trunc_page_ps(offset + filsz, pagesize) - 552 trunc_page(offset + filsz); 553 error = copyout((caddr_t)sf_buf_kva(sf) + off, 554 (caddr_t)map_addr, copy_len); 555 vm_imgact_unmap_page(sf); 556 if (error) { 557 return (error); 558 } 559 } 560 561 /* 562 * set it to the specified protection. 563 * XXX had better undo the damage from pasting over the cracks here! 564 */ 565 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 566 map_len), prot, FALSE); 567 568 return (0); 569 } 570 571 /* 572 * Load the file "file" into memory. It may be either a shared object 573 * or an executable. 574 * 575 * The "addr" reference parameter is in/out. On entry, it specifies 576 * the address where a shared object should be loaded. If the file is 577 * an executable, this value is ignored. On exit, "addr" specifies 578 * where the file was actually loaded. 579 * 580 * The "entry" reference parameter is out only. On exit, it specifies 581 * the entry point for the loaded file. 582 */ 583 static int 584 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 585 u_long *entry, size_t pagesize) 586 { 587 struct { 588 struct nameidata nd; 589 struct vattr attr; 590 struct image_params image_params; 591 } *tempdata; 592 const Elf_Ehdr *hdr = NULL; 593 const Elf_Phdr *phdr = NULL; 594 struct nameidata *nd; 595 struct vattr *attr; 596 struct image_params *imgp; 597 vm_prot_t prot; 598 u_long rbase; 599 u_long base_addr = 0; 600 int error, i, numsegs; 601 602 #ifdef CAPABILITY_MODE 603 /* 604 * XXXJA: This check can go away once we are sufficiently confident 605 * that the checks in namei() are correct. 606 */ 607 if (IN_CAPABILITY_MODE(curthread)) 608 return (ECAPMODE); 609 #endif 610 611 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 612 nd = &tempdata->nd; 613 attr = &tempdata->attr; 614 imgp = &tempdata->image_params; 615 616 /* 617 * Initialize part of the common data 618 */ 619 imgp->proc = p; 620 imgp->attr = attr; 621 imgp->firstpage = NULL; 622 imgp->image_header = NULL; 623 imgp->object = NULL; 624 imgp->execlabel = NULL; 625 626 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread); 627 if ((error = namei(nd)) != 0) { 628 nd->ni_vp = NULL; 629 goto fail; 630 } 631 NDFREE(nd, NDF_ONLY_PNBUF); 632 imgp->vp = nd->ni_vp; 633 634 /* 635 * Check permissions, modes, uid, etc on the file, and "open" it. 636 */ 637 error = exec_check_permissions(imgp); 638 if (error) 639 goto fail; 640 641 error = exec_map_first_page(imgp); 642 if (error) 643 goto fail; 644 645 /* 646 * Also make certain that the interpreter stays the same, so set 647 * its VV_TEXT flag, too. 648 */ 649 VOP_SET_TEXT(nd->ni_vp); 650 651 imgp->object = nd->ni_vp->v_object; 652 653 hdr = (const Elf_Ehdr *)imgp->image_header; 654 if ((error = __elfN(check_header)(hdr)) != 0) 655 goto fail; 656 if (hdr->e_type == ET_DYN) 657 rbase = *addr; 658 else if (hdr->e_type == ET_EXEC) 659 rbase = 0; 660 else { 661 error = ENOEXEC; 662 goto fail; 663 } 664 665 /* Only support headers that fit within first page for now */ 666 if ((hdr->e_phoff > PAGE_SIZE) || 667 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 668 error = ENOEXEC; 669 goto fail; 670 } 671 672 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 673 if (!aligned(phdr, Elf_Addr)) { 674 error = ENOEXEC; 675 goto fail; 676 } 677 678 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 679 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 680 /* Loadable segment */ 681 prot = __elfN(trans_prot)(phdr[i].p_flags); 682 error = __elfN(load_section)(imgp, phdr[i].p_offset, 683 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 684 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize); 685 if (error != 0) 686 goto fail; 687 /* 688 * Establish the base address if this is the 689 * first segment. 690 */ 691 if (numsegs == 0) 692 base_addr = trunc_page(phdr[i].p_vaddr + 693 rbase); 694 numsegs++; 695 } 696 } 697 *addr = base_addr; 698 *entry = (unsigned long)hdr->e_entry + rbase; 699 700 fail: 701 if (imgp->firstpage) 702 exec_unmap_first_page(imgp); 703 704 if (nd->ni_vp) 705 vput(nd->ni_vp); 706 707 free(tempdata, M_TEMP); 708 709 return (error); 710 } 711 712 static int 713 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 714 { 715 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 716 const Elf_Phdr *phdr; 717 Elf_Auxargs *elf_auxargs; 718 struct vmspace *vmspace; 719 vm_prot_t prot; 720 u_long text_size = 0, data_size = 0, total_size = 0; 721 u_long text_addr = 0, data_addr = 0; 722 u_long seg_size, seg_addr; 723 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0; 724 int32_t osrel = 0; 725 int error = 0, i, n, interp_name_len = 0; 726 const char *interp = NULL, *newinterp = NULL; 727 Elf_Brandinfo *brand_info; 728 char *path; 729 struct sysentvec *sv; 730 731 /* 732 * Do we have a valid ELF header ? 733 * 734 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 735 * if particular brand doesn't support it. 736 */ 737 if (__elfN(check_header)(hdr) != 0 || 738 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 739 return (-1); 740 741 /* 742 * From here on down, we return an errno, not -1, as we've 743 * detected an ELF file. 744 */ 745 746 if ((hdr->e_phoff > PAGE_SIZE) || 747 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 748 /* Only support headers in first page for now */ 749 return (ENOEXEC); 750 } 751 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 752 if (!aligned(phdr, Elf_Addr)) 753 return (ENOEXEC); 754 n = 0; 755 baddr = 0; 756 for (i = 0; i < hdr->e_phnum; i++) { 757 switch (phdr[i].p_type) { 758 case PT_LOAD: 759 if (n == 0) 760 baddr = phdr[i].p_vaddr; 761 n++; 762 break; 763 case PT_INTERP: 764 /* Path to interpreter */ 765 if (phdr[i].p_filesz > MAXPATHLEN || 766 phdr[i].p_offset > PAGE_SIZE || 767 phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset) 768 return (ENOEXEC); 769 interp = imgp->image_header + phdr[i].p_offset; 770 interp_name_len = phdr[i].p_filesz; 771 break; 772 case PT_GNU_STACK: 773 if (__elfN(nxstack)) 774 imgp->stack_prot = 775 __elfN(trans_prot)(phdr[i].p_flags); 776 break; 777 } 778 } 779 780 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len, 781 &osrel); 782 if (brand_info == NULL) { 783 uprintf("ELF binary type \"%u\" not known.\n", 784 hdr->e_ident[EI_OSABI]); 785 return (ENOEXEC); 786 } 787 if (hdr->e_type == ET_DYN) { 788 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) 789 return (ENOEXEC); 790 /* 791 * Honour the base load address from the dso if it is 792 * non-zero for some reason. 793 */ 794 if (baddr == 0) 795 et_dyn_addr = ET_DYN_LOAD_ADDR; 796 else 797 et_dyn_addr = 0; 798 } else 799 et_dyn_addr = 0; 800 sv = brand_info->sysvec; 801 if (interp != NULL && brand_info->interp_newpath != NULL) 802 newinterp = brand_info->interp_newpath; 803 804 /* 805 * Avoid a possible deadlock if the current address space is destroyed 806 * and that address space maps the locked vnode. In the common case, 807 * the locked vnode's v_usecount is decremented but remains greater 808 * than zero. Consequently, the vnode lock is not needed by vrele(). 809 * However, in cases where the vnode lock is external, such as nullfs, 810 * v_usecount may become zero. 811 * 812 * The VV_TEXT flag prevents modifications to the executable while 813 * the vnode is unlocked. 814 */ 815 VOP_UNLOCK(imgp->vp, 0); 816 817 error = exec_new_vmspace(imgp, sv); 818 imgp->proc->p_sysent = sv; 819 820 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 821 if (error) 822 return (error); 823 824 for (i = 0; i < hdr->e_phnum; i++) { 825 switch (phdr[i].p_type) { 826 case PT_LOAD: /* Loadable segment */ 827 if (phdr[i].p_memsz == 0) 828 break; 829 prot = __elfN(trans_prot)(phdr[i].p_flags); 830 error = __elfN(load_section)(imgp, phdr[i].p_offset, 831 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 832 phdr[i].p_memsz, phdr[i].p_filesz, prot, 833 sv->sv_pagesize); 834 if (error != 0) 835 return (error); 836 837 /* 838 * If this segment contains the program headers, 839 * remember their virtual address for the AT_PHDR 840 * aux entry. Static binaries don't usually include 841 * a PT_PHDR entry. 842 */ 843 if (phdr[i].p_offset == 0 && 844 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 845 <= phdr[i].p_filesz) 846 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 847 et_dyn_addr; 848 849 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 850 seg_size = round_page(phdr[i].p_memsz + 851 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 852 853 /* 854 * Make the largest executable segment the official 855 * text segment and all others data. 856 * 857 * Note that obreak() assumes that data_addr + 858 * data_size == end of data load area, and the ELF 859 * file format expects segments to be sorted by 860 * address. If multiple data segments exist, the 861 * last one will be used. 862 */ 863 864 if (phdr[i].p_flags & PF_X && text_size < seg_size) { 865 text_size = seg_size; 866 text_addr = seg_addr; 867 } else { 868 data_size = seg_size; 869 data_addr = seg_addr; 870 } 871 total_size += seg_size; 872 break; 873 case PT_PHDR: /* Program header table info */ 874 proghdr = phdr[i].p_vaddr + et_dyn_addr; 875 break; 876 default: 877 break; 878 } 879 } 880 881 if (data_addr == 0 && data_size == 0) { 882 data_addr = text_addr; 883 data_size = text_size; 884 } 885 886 entry = (u_long)hdr->e_entry + et_dyn_addr; 887 888 /* 889 * Check limits. It should be safe to check the 890 * limits after loading the segments since we do 891 * not actually fault in all the segments pages. 892 */ 893 PROC_LOCK(imgp->proc); 894 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 895 text_size > maxtsiz || 896 total_size > lim_cur(imgp->proc, RLIMIT_VMEM) || 897 racct_set(imgp->proc, RACCT_DATA, data_size) != 0 || 898 racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) { 899 PROC_UNLOCK(imgp->proc); 900 return (ENOMEM); 901 } 902 903 vmspace = imgp->proc->p_vmspace; 904 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 905 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 906 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 907 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 908 909 /* 910 * We load the dynamic linker where a userland call 911 * to mmap(0, ...) would put it. The rationale behind this 912 * calculation is that it leaves room for the heap to grow to 913 * its maximum allowed size. 914 */ 915 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc, 916 RLIMIT_DATA)); 917 PROC_UNLOCK(imgp->proc); 918 919 imgp->entry_addr = entry; 920 921 if (interp != NULL) { 922 int have_interp = FALSE; 923 VOP_UNLOCK(imgp->vp, 0); 924 if (brand_info->emul_path != NULL && 925 brand_info->emul_path[0] != '\0') { 926 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 927 snprintf(path, MAXPATHLEN, "%s%s", 928 brand_info->emul_path, interp); 929 error = __elfN(load_file)(imgp->proc, path, &addr, 930 &imgp->entry_addr, sv->sv_pagesize); 931 free(path, M_TEMP); 932 if (error == 0) 933 have_interp = TRUE; 934 } 935 if (!have_interp && newinterp != NULL) { 936 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 937 &imgp->entry_addr, sv->sv_pagesize); 938 if (error == 0) 939 have_interp = TRUE; 940 } 941 if (!have_interp) { 942 error = __elfN(load_file)(imgp->proc, interp, &addr, 943 &imgp->entry_addr, sv->sv_pagesize); 944 } 945 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 946 if (error != 0) { 947 uprintf("ELF interpreter %s not found\n", interp); 948 return (error); 949 } 950 } else 951 addr = et_dyn_addr; 952 953 /* 954 * Construct auxargs table (used by the fixup routine) 955 */ 956 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 957 elf_auxargs->execfd = -1; 958 elf_auxargs->phdr = proghdr; 959 elf_auxargs->phent = hdr->e_phentsize; 960 elf_auxargs->phnum = hdr->e_phnum; 961 elf_auxargs->pagesz = PAGE_SIZE; 962 elf_auxargs->base = addr; 963 elf_auxargs->flags = 0; 964 elf_auxargs->entry = entry; 965 966 imgp->auxargs = elf_auxargs; 967 imgp->interpreted = 0; 968 imgp->reloc_base = addr; 969 imgp->proc->p_osrel = osrel; 970 971 return (error); 972 } 973 974 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 975 976 int 977 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 978 { 979 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 980 Elf_Addr *base; 981 Elf_Addr *pos; 982 983 base = (Elf_Addr *)*stack_base; 984 pos = base + (imgp->args->argc + imgp->args->envc + 2); 985 986 if (args->execfd != -1) 987 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 988 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 989 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 990 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 991 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 992 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 993 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 994 AUXARGS_ENTRY(pos, AT_BASE, args->base); 995 if (imgp->execpathp != 0) 996 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 997 AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate); 998 if (imgp->canary != 0) { 999 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary); 1000 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1001 } 1002 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1003 if (imgp->pagesizes != 0) { 1004 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes); 1005 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1006 } 1007 if (imgp->sysent->sv_timekeep_base != 0) { 1008 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1009 imgp->sysent->sv_timekeep_base); 1010 } 1011 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1012 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1013 imgp->sysent->sv_stackprot); 1014 AUXARGS_ENTRY(pos, AT_NULL, 0); 1015 1016 free(imgp->auxargs, M_TEMP); 1017 imgp->auxargs = NULL; 1018 1019 base--; 1020 suword(base, (long)imgp->args->argc); 1021 *stack_base = (register_t *)base; 1022 return (0); 1023 } 1024 1025 /* 1026 * Code for generating ELF core dumps. 1027 */ 1028 1029 typedef void (*segment_callback)(vm_map_entry_t, void *); 1030 1031 /* Closure for cb_put_phdr(). */ 1032 struct phdr_closure { 1033 Elf_Phdr *phdr; /* Program header to fill in */ 1034 Elf_Off offset; /* Offset of segment in core file */ 1035 }; 1036 1037 /* Closure for cb_size_segment(). */ 1038 struct sseg_closure { 1039 int count; /* Count of writable segments. */ 1040 size_t size; /* Total size of all writable segments. */ 1041 }; 1042 1043 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *); 1044 1045 struct note_info { 1046 int type; /* Note type. */ 1047 outfunc_t outfunc; /* Output function. */ 1048 void *outarg; /* Argument for the output function. */ 1049 size_t outsize; /* Output size. */ 1050 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1051 }; 1052 1053 TAILQ_HEAD(note_info_list, note_info); 1054 1055 static void cb_put_phdr(vm_map_entry_t, void *); 1056 static void cb_size_segment(vm_map_entry_t, void *); 1057 static void each_writable_segment(struct thread *, segment_callback, void *); 1058 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 1059 int, void *, size_t, struct note_info_list *, size_t, gzFile); 1060 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *, 1061 size_t *); 1062 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t); 1063 static void __elfN(putnote)(struct note_info *, struct sbuf *); 1064 static size_t register_note(struct note_info_list *, int, outfunc_t, void *); 1065 static int sbuf_drain_core_output(void *, const char *, int); 1066 static int sbuf_drain_count(void *arg, const char *data, int len); 1067 1068 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *); 1069 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1070 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *); 1071 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1072 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *); 1073 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1074 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1075 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1076 static void note_procstat_files(void *, struct sbuf *, size_t *); 1077 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1078 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1079 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1080 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1081 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1082 1083 #ifdef COMPRESS_USER_CORES 1084 extern int compress_user_cores; 1085 extern int compress_user_cores_gzlevel; 1086 #endif 1087 1088 static int 1089 core_output(struct vnode *vp, void *base, size_t len, off_t offset, 1090 struct ucred *active_cred, struct ucred *file_cred, 1091 struct thread *td, char *core_buf, gzFile gzfile) { 1092 1093 int error; 1094 if (gzfile) { 1095 #ifdef COMPRESS_USER_CORES 1096 error = compress_core(gzfile, base, core_buf, len, td); 1097 #else 1098 panic("shouldn't be here"); 1099 #endif 1100 } else { 1101 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset, 1102 UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred, 1103 NULL, td); 1104 } 1105 return (error); 1106 } 1107 1108 /* Coredump output parameters for sbuf drain routine. */ 1109 struct sbuf_drain_core_params { 1110 off_t offset; 1111 struct ucred *active_cred; 1112 struct ucred *file_cred; 1113 struct thread *td; 1114 struct vnode *vp; 1115 #ifdef COMPRESS_USER_CORES 1116 gzFile gzfile; 1117 #endif 1118 }; 1119 1120 /* 1121 * Drain into a core file. 1122 */ 1123 static int 1124 sbuf_drain_core_output(void *arg, const char *data, int len) 1125 { 1126 struct sbuf_drain_core_params *p; 1127 int error, locked; 1128 1129 p = (struct sbuf_drain_core_params *)arg; 1130 1131 /* 1132 * Some kern_proc out routines that print to this sbuf may 1133 * call us with the process lock held. Draining with the 1134 * non-sleepable lock held is unsafe. The lock is needed for 1135 * those routines when dumping a live process. In our case we 1136 * can safely release the lock before draining and acquire 1137 * again after. 1138 */ 1139 locked = PROC_LOCKED(p->td->td_proc); 1140 if (locked) 1141 PROC_UNLOCK(p->td->td_proc); 1142 #ifdef COMPRESS_USER_CORES 1143 if (p->gzfile != Z_NULL) 1144 error = compress_core(p->gzfile, NULL, __DECONST(char *, data), 1145 len, p->td); 1146 else 1147 #endif 1148 error = vn_rdwr_inchunks(UIO_WRITE, p->vp, 1149 __DECONST(void *, data), len, p->offset, UIO_SYSSPACE, 1150 IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL, 1151 p->td); 1152 if (locked) 1153 PROC_LOCK(p->td->td_proc); 1154 if (error != 0) 1155 return (-error); 1156 p->offset += len; 1157 return (len); 1158 } 1159 1160 /* 1161 * Drain into a counter. 1162 */ 1163 static int 1164 sbuf_drain_count(void *arg, const char *data __unused, int len) 1165 { 1166 size_t *sizep; 1167 1168 sizep = (size_t *)arg; 1169 *sizep += len; 1170 return (len); 1171 } 1172 1173 int 1174 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1175 { 1176 struct ucred *cred = td->td_ucred; 1177 int error = 0; 1178 struct sseg_closure seginfo; 1179 struct note_info_list notelst; 1180 struct note_info *ninfo; 1181 void *hdr; 1182 size_t hdrsize, notesz, coresize; 1183 1184 gzFile gzfile = Z_NULL; 1185 char *core_buf = NULL; 1186 #ifdef COMPRESS_USER_CORES 1187 char gzopen_flags[8]; 1188 char *p; 1189 int doing_compress = flags & IMGACT_CORE_COMPRESS; 1190 #endif 1191 1192 hdr = NULL; 1193 TAILQ_INIT(¬elst); 1194 1195 #ifdef COMPRESS_USER_CORES 1196 if (doing_compress) { 1197 p = gzopen_flags; 1198 *p++ = 'w'; 1199 if (compress_user_cores_gzlevel >= 0 && 1200 compress_user_cores_gzlevel <= 9) 1201 *p++ = '0' + compress_user_cores_gzlevel; 1202 *p = 0; 1203 gzfile = gz_open("", gzopen_flags, vp); 1204 if (gzfile == Z_NULL) { 1205 error = EFAULT; 1206 goto done; 1207 } 1208 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1209 if (!core_buf) { 1210 error = ENOMEM; 1211 goto done; 1212 } 1213 } 1214 #endif 1215 1216 /* Size the program segments. */ 1217 seginfo.count = 0; 1218 seginfo.size = 0; 1219 each_writable_segment(td, cb_size_segment, &seginfo); 1220 1221 /* 1222 * Collect info about the core file header area. 1223 */ 1224 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1225 __elfN(prepare_notes)(td, ¬elst, ¬esz); 1226 coresize = round_page(hdrsize + notesz) + seginfo.size; 1227 1228 #ifdef RACCT 1229 PROC_LOCK(td->td_proc); 1230 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1231 PROC_UNLOCK(td->td_proc); 1232 if (error != 0) { 1233 error = EFAULT; 1234 goto done; 1235 } 1236 #endif 1237 if (coresize >= limit) { 1238 error = EFAULT; 1239 goto done; 1240 } 1241 1242 /* 1243 * Allocate memory for building the header, fill it up, 1244 * and write it out following the notes. 1245 */ 1246 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1247 if (hdr == NULL) { 1248 error = EINVAL; 1249 goto done; 1250 } 1251 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize, 1252 ¬elst, notesz, gzfile); 1253 1254 /* Write the contents of all of the writable segments. */ 1255 if (error == 0) { 1256 Elf_Phdr *php; 1257 off_t offset; 1258 int i; 1259 1260 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1261 offset = round_page(hdrsize + notesz); 1262 for (i = 0; i < seginfo.count; i++) { 1263 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr, 1264 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile); 1265 if (error != 0) 1266 break; 1267 offset += php->p_filesz; 1268 php++; 1269 } 1270 } 1271 if (error) { 1272 log(LOG_WARNING, 1273 "Failed to write core file for process %s (error %d)\n", 1274 curproc->p_comm, error); 1275 } 1276 1277 done: 1278 #ifdef COMPRESS_USER_CORES 1279 if (core_buf) 1280 free(core_buf, M_TEMP); 1281 if (gzfile) 1282 gzclose(gzfile); 1283 #endif 1284 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1285 TAILQ_REMOVE(¬elst, ninfo, link); 1286 free(ninfo, M_TEMP); 1287 } 1288 if (hdr != NULL) 1289 free(hdr, M_TEMP); 1290 1291 return (error); 1292 } 1293 1294 /* 1295 * A callback for each_writable_segment() to write out the segment's 1296 * program header entry. 1297 */ 1298 static void 1299 cb_put_phdr(entry, closure) 1300 vm_map_entry_t entry; 1301 void *closure; 1302 { 1303 struct phdr_closure *phc = (struct phdr_closure *)closure; 1304 Elf_Phdr *phdr = phc->phdr; 1305 1306 phc->offset = round_page(phc->offset); 1307 1308 phdr->p_type = PT_LOAD; 1309 phdr->p_offset = phc->offset; 1310 phdr->p_vaddr = entry->start; 1311 phdr->p_paddr = 0; 1312 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1313 phdr->p_align = PAGE_SIZE; 1314 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1315 1316 phc->offset += phdr->p_filesz; 1317 phc->phdr++; 1318 } 1319 1320 /* 1321 * A callback for each_writable_segment() to gather information about 1322 * the number of segments and their total size. 1323 */ 1324 static void 1325 cb_size_segment(entry, closure) 1326 vm_map_entry_t entry; 1327 void *closure; 1328 { 1329 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1330 1331 ssc->count++; 1332 ssc->size += entry->end - entry->start; 1333 } 1334 1335 /* 1336 * For each writable segment in the process's memory map, call the given 1337 * function with a pointer to the map entry and some arbitrary 1338 * caller-supplied data. 1339 */ 1340 static void 1341 each_writable_segment(td, func, closure) 1342 struct thread *td; 1343 segment_callback func; 1344 void *closure; 1345 { 1346 struct proc *p = td->td_proc; 1347 vm_map_t map = &p->p_vmspace->vm_map; 1348 vm_map_entry_t entry; 1349 vm_object_t backing_object, object; 1350 boolean_t ignore_entry; 1351 1352 vm_map_lock_read(map); 1353 for (entry = map->header.next; entry != &map->header; 1354 entry = entry->next) { 1355 /* 1356 * Don't dump inaccessible mappings, deal with legacy 1357 * coredump mode. 1358 * 1359 * Note that read-only segments related to the elf binary 1360 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1361 * need to arbitrarily ignore such segments. 1362 */ 1363 if (elf_legacy_coredump) { 1364 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1365 continue; 1366 } else { 1367 if ((entry->protection & VM_PROT_ALL) == 0) 1368 continue; 1369 } 1370 1371 /* 1372 * Dont include memory segment in the coredump if 1373 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1374 * madvise(2). Do not dump submaps (i.e. parts of the 1375 * kernel map). 1376 */ 1377 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1378 continue; 1379 1380 if ((object = entry->object.vm_object) == NULL) 1381 continue; 1382 1383 /* Ignore memory-mapped devices and such things. */ 1384 VM_OBJECT_RLOCK(object); 1385 while ((backing_object = object->backing_object) != NULL) { 1386 VM_OBJECT_RLOCK(backing_object); 1387 VM_OBJECT_RUNLOCK(object); 1388 object = backing_object; 1389 } 1390 ignore_entry = object->type != OBJT_DEFAULT && 1391 object->type != OBJT_SWAP && object->type != OBJT_VNODE; 1392 VM_OBJECT_RUNLOCK(object); 1393 if (ignore_entry) 1394 continue; 1395 1396 (*func)(entry, closure); 1397 } 1398 vm_map_unlock_read(map); 1399 } 1400 1401 /* 1402 * Write the core file header to the file, including padding up to 1403 * the page boundary. 1404 */ 1405 static int 1406 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred, 1407 int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst, 1408 size_t notesz, gzFile gzfile) 1409 { 1410 struct sbuf_drain_core_params params; 1411 struct note_info *ninfo; 1412 struct sbuf *sb; 1413 int error; 1414 1415 /* Fill in the header. */ 1416 bzero(hdr, hdrsize); 1417 __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz); 1418 1419 params.offset = 0; 1420 params.active_cred = cred; 1421 params.file_cred = NOCRED; 1422 params.td = td; 1423 params.vp = vp; 1424 #ifdef COMPRESS_USER_CORES 1425 params.gzfile = gzfile; 1426 #endif 1427 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1428 sbuf_set_drain(sb, sbuf_drain_core_output, ¶ms); 1429 sbuf_start_section(sb, NULL); 1430 sbuf_bcat(sb, hdr, hdrsize); 1431 TAILQ_FOREACH(ninfo, notelst, link) 1432 __elfN(putnote)(ninfo, sb); 1433 /* Align up to a page boundary for the program segments. */ 1434 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1435 error = sbuf_finish(sb); 1436 sbuf_delete(sb); 1437 1438 return (error); 1439 } 1440 1441 static void 1442 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1443 size_t *sizep) 1444 { 1445 struct proc *p; 1446 struct thread *thr; 1447 size_t size; 1448 1449 p = td->td_proc; 1450 size = 0; 1451 1452 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p); 1453 1454 /* 1455 * To have the debugger select the right thread (LWP) as the initial 1456 * thread, we dump the state of the thread passed to us in td first. 1457 * This is the thread that causes the core dump and thus likely to 1458 * be the right thread one wants to have selected in the debugger. 1459 */ 1460 thr = td; 1461 while (thr != NULL) { 1462 size += register_note(list, NT_PRSTATUS, 1463 __elfN(note_prstatus), thr); 1464 size += register_note(list, NT_FPREGSET, 1465 __elfN(note_fpregset), thr); 1466 size += register_note(list, NT_THRMISC, 1467 __elfN(note_thrmisc), thr); 1468 size += register_note(list, -1, 1469 __elfN(note_threadmd), thr); 1470 1471 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1472 TAILQ_NEXT(thr, td_plist); 1473 if (thr == td) 1474 thr = TAILQ_NEXT(thr, td_plist); 1475 } 1476 1477 size += register_note(list, NT_PROCSTAT_PROC, 1478 __elfN(note_procstat_proc), p); 1479 size += register_note(list, NT_PROCSTAT_FILES, 1480 note_procstat_files, p); 1481 size += register_note(list, NT_PROCSTAT_VMMAP, 1482 note_procstat_vmmap, p); 1483 size += register_note(list, NT_PROCSTAT_GROUPS, 1484 note_procstat_groups, p); 1485 size += register_note(list, NT_PROCSTAT_UMASK, 1486 note_procstat_umask, p); 1487 size += register_note(list, NT_PROCSTAT_RLIMIT, 1488 note_procstat_rlimit, p); 1489 size += register_note(list, NT_PROCSTAT_OSREL, 1490 note_procstat_osrel, p); 1491 size += register_note(list, NT_PROCSTAT_PSSTRINGS, 1492 __elfN(note_procstat_psstrings), p); 1493 size += register_note(list, NT_PROCSTAT_AUXV, 1494 __elfN(note_procstat_auxv), p); 1495 1496 *sizep = size; 1497 } 1498 1499 static void 1500 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1501 size_t notesz) 1502 { 1503 Elf_Ehdr *ehdr; 1504 Elf_Phdr *phdr; 1505 struct phdr_closure phc; 1506 1507 ehdr = (Elf_Ehdr *)hdr; 1508 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)); 1509 1510 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1511 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1512 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1513 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1514 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1515 ehdr->e_ident[EI_DATA] = ELF_DATA; 1516 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1517 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1518 ehdr->e_ident[EI_ABIVERSION] = 0; 1519 ehdr->e_ident[EI_PAD] = 0; 1520 ehdr->e_type = ET_CORE; 1521 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1522 ehdr->e_machine = ELF_ARCH32; 1523 #else 1524 ehdr->e_machine = ELF_ARCH; 1525 #endif 1526 ehdr->e_version = EV_CURRENT; 1527 ehdr->e_entry = 0; 1528 ehdr->e_phoff = sizeof(Elf_Ehdr); 1529 ehdr->e_flags = 0; 1530 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1531 ehdr->e_phentsize = sizeof(Elf_Phdr); 1532 ehdr->e_phnum = numsegs + 1; 1533 ehdr->e_shentsize = sizeof(Elf_Shdr); 1534 ehdr->e_shnum = 0; 1535 ehdr->e_shstrndx = SHN_UNDEF; 1536 1537 /* 1538 * Fill in the program header entries. 1539 */ 1540 1541 /* The note segement. */ 1542 phdr->p_type = PT_NOTE; 1543 phdr->p_offset = hdrsize; 1544 phdr->p_vaddr = 0; 1545 phdr->p_paddr = 0; 1546 phdr->p_filesz = notesz; 1547 phdr->p_memsz = 0; 1548 phdr->p_flags = PF_R; 1549 phdr->p_align = sizeof(Elf32_Size); 1550 phdr++; 1551 1552 /* All the writable segments from the program. */ 1553 phc.phdr = phdr; 1554 phc.offset = round_page(hdrsize + notesz); 1555 each_writable_segment(td, cb_put_phdr, &phc); 1556 } 1557 1558 static size_t 1559 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg) 1560 { 1561 struct note_info *ninfo; 1562 size_t size, notesize; 1563 1564 size = 0; 1565 out(arg, NULL, &size); 1566 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 1567 ninfo->type = type; 1568 ninfo->outfunc = out; 1569 ninfo->outarg = arg; 1570 ninfo->outsize = size; 1571 TAILQ_INSERT_TAIL(list, ninfo, link); 1572 1573 if (type == -1) 1574 return (size); 1575 1576 notesize = sizeof(Elf_Note) + /* note header */ 1577 roundup2(8, sizeof(Elf32_Size)) + /* note name ("FreeBSD") */ 1578 roundup2(size, sizeof(Elf32_Size)); /* note description */ 1579 1580 return (notesize); 1581 } 1582 1583 static void 1584 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb) 1585 { 1586 Elf_Note note; 1587 ssize_t old_len; 1588 1589 if (ninfo->type == -1) { 1590 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1591 return; 1592 } 1593 1594 note.n_namesz = 8; /* strlen("FreeBSD") + 1 */ 1595 note.n_descsz = ninfo->outsize; 1596 note.n_type = ninfo->type; 1597 1598 sbuf_bcat(sb, ¬e, sizeof(note)); 1599 sbuf_start_section(sb, &old_len); 1600 sbuf_bcat(sb, "FreeBSD", note.n_namesz); 1601 sbuf_end_section(sb, old_len, sizeof(Elf32_Size), 0); 1602 if (note.n_descsz == 0) 1603 return; 1604 sbuf_start_section(sb, &old_len); 1605 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1606 sbuf_end_section(sb, old_len, sizeof(Elf32_Size), 0); 1607 } 1608 1609 /* 1610 * Miscellaneous note out functions. 1611 */ 1612 1613 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1614 #include <compat/freebsd32/freebsd32.h> 1615 1616 typedef struct prstatus32 elf_prstatus_t; 1617 typedef struct prpsinfo32 elf_prpsinfo_t; 1618 typedef struct fpreg32 elf_prfpregset_t; 1619 typedef struct fpreg32 elf_fpregset_t; 1620 typedef struct reg32 elf_gregset_t; 1621 typedef struct thrmisc32 elf_thrmisc_t; 1622 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 1623 typedef struct kinfo_proc32 elf_kinfo_proc_t; 1624 typedef uint32_t elf_ps_strings_t; 1625 #else 1626 typedef prstatus_t elf_prstatus_t; 1627 typedef prpsinfo_t elf_prpsinfo_t; 1628 typedef prfpregset_t elf_prfpregset_t; 1629 typedef prfpregset_t elf_fpregset_t; 1630 typedef gregset_t elf_gregset_t; 1631 typedef thrmisc_t elf_thrmisc_t; 1632 #define ELF_KERN_PROC_MASK 0 1633 typedef struct kinfo_proc elf_kinfo_proc_t; 1634 typedef vm_offset_t elf_ps_strings_t; 1635 #endif 1636 1637 static void 1638 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 1639 { 1640 struct proc *p; 1641 elf_prpsinfo_t *psinfo; 1642 1643 p = (struct proc *)arg; 1644 if (sb != NULL) { 1645 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 1646 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 1647 psinfo->pr_version = PRPSINFO_VERSION; 1648 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1649 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1650 /* 1651 * XXX - We don't fill in the command line arguments properly 1652 * yet. 1653 */ 1654 strlcpy(psinfo->pr_psargs, p->p_comm, 1655 sizeof(psinfo->pr_psargs)); 1656 1657 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 1658 free(psinfo, M_TEMP); 1659 } 1660 *sizep = sizeof(*psinfo); 1661 } 1662 1663 static void 1664 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep) 1665 { 1666 struct thread *td; 1667 elf_prstatus_t *status; 1668 1669 td = (struct thread *)arg; 1670 if (sb != NULL) { 1671 KASSERT(*sizep == sizeof(*status), ("invalid size")); 1672 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK); 1673 status->pr_version = PRSTATUS_VERSION; 1674 status->pr_statussz = sizeof(elf_prstatus_t); 1675 status->pr_gregsetsz = sizeof(elf_gregset_t); 1676 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1677 status->pr_osreldate = osreldate; 1678 status->pr_cursig = td->td_proc->p_sig; 1679 status->pr_pid = td->td_tid; 1680 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1681 fill_regs32(td, &status->pr_reg); 1682 #else 1683 fill_regs(td, &status->pr_reg); 1684 #endif 1685 sbuf_bcat(sb, status, sizeof(*status)); 1686 free(status, M_TEMP); 1687 } 1688 *sizep = sizeof(*status); 1689 } 1690 1691 static void 1692 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep) 1693 { 1694 struct thread *td; 1695 elf_prfpregset_t *fpregset; 1696 1697 td = (struct thread *)arg; 1698 if (sb != NULL) { 1699 KASSERT(*sizep == sizeof(*fpregset), ("invalid size")); 1700 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK); 1701 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1702 fill_fpregs32(td, fpregset); 1703 #else 1704 fill_fpregs(td, fpregset); 1705 #endif 1706 sbuf_bcat(sb, fpregset, sizeof(*fpregset)); 1707 free(fpregset, M_TEMP); 1708 } 1709 *sizep = sizeof(*fpregset); 1710 } 1711 1712 static void 1713 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep) 1714 { 1715 struct thread *td; 1716 elf_thrmisc_t thrmisc; 1717 1718 td = (struct thread *)arg; 1719 if (sb != NULL) { 1720 KASSERT(*sizep == sizeof(thrmisc), ("invalid size")); 1721 bzero(&thrmisc._pad, sizeof(thrmisc._pad)); 1722 strcpy(thrmisc.pr_tname, td->td_name); 1723 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc)); 1724 } 1725 *sizep = sizeof(thrmisc); 1726 } 1727 1728 /* 1729 * Allow for MD specific notes, as well as any MD 1730 * specific preparations for writing MI notes. 1731 */ 1732 static void 1733 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 1734 { 1735 struct thread *td; 1736 void *buf; 1737 size_t size; 1738 1739 td = (struct thread *)arg; 1740 size = *sizep; 1741 buf = NULL; 1742 if (size != 0 && sb != NULL) 1743 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 1744 size = 0; 1745 __elfN(dump_thread)(td, buf, &size); 1746 KASSERT(*sizep == size, ("invalid size")); 1747 if (size != 0 && sb != NULL) 1748 sbuf_bcat(sb, buf, size); 1749 *sizep = size; 1750 } 1751 1752 #ifdef KINFO_PROC_SIZE 1753 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 1754 #endif 1755 1756 static void 1757 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 1758 { 1759 struct proc *p; 1760 size_t size; 1761 int structsize; 1762 1763 p = (struct proc *)arg; 1764 size = sizeof(structsize) + p->p_numthreads * 1765 sizeof(elf_kinfo_proc_t); 1766 1767 if (sb != NULL) { 1768 KASSERT(*sizep == size, ("invalid size")); 1769 structsize = sizeof(elf_kinfo_proc_t); 1770 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1771 PROC_LOCK(p); 1772 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 1773 } 1774 *sizep = size; 1775 } 1776 1777 #ifdef KINFO_FILE_SIZE 1778 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 1779 #endif 1780 1781 static void 1782 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 1783 { 1784 struct proc *p; 1785 size_t size; 1786 int structsize; 1787 1788 p = (struct proc *)arg; 1789 if (sb == NULL) { 1790 size = 0; 1791 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 1792 sbuf_set_drain(sb, sbuf_drain_count, &size); 1793 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1794 PROC_LOCK(p); 1795 kern_proc_filedesc_out(p, sb, -1); 1796 sbuf_finish(sb); 1797 sbuf_delete(sb); 1798 *sizep = size; 1799 } else { 1800 structsize = sizeof(struct kinfo_file); 1801 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1802 PROC_LOCK(p); 1803 kern_proc_filedesc_out(p, sb, -1); 1804 } 1805 } 1806 1807 #ifdef KINFO_VMENTRY_SIZE 1808 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 1809 #endif 1810 1811 static void 1812 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 1813 { 1814 struct proc *p; 1815 size_t size; 1816 int structsize; 1817 1818 p = (struct proc *)arg; 1819 if (sb == NULL) { 1820 size = 0; 1821 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 1822 sbuf_set_drain(sb, sbuf_drain_count, &size); 1823 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1824 PROC_LOCK(p); 1825 kern_proc_vmmap_out(p, sb); 1826 sbuf_finish(sb); 1827 sbuf_delete(sb); 1828 *sizep = size; 1829 } else { 1830 structsize = sizeof(struct kinfo_vmentry); 1831 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1832 PROC_LOCK(p); 1833 kern_proc_vmmap_out(p, sb); 1834 } 1835 } 1836 1837 static void 1838 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 1839 { 1840 struct proc *p; 1841 size_t size; 1842 int structsize; 1843 1844 p = (struct proc *)arg; 1845 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 1846 if (sb != NULL) { 1847 KASSERT(*sizep == size, ("invalid size")); 1848 structsize = sizeof(gid_t); 1849 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1850 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 1851 sizeof(gid_t)); 1852 } 1853 *sizep = size; 1854 } 1855 1856 static void 1857 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 1858 { 1859 struct proc *p; 1860 size_t size; 1861 int structsize; 1862 1863 p = (struct proc *)arg; 1864 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask); 1865 if (sb != NULL) { 1866 KASSERT(*sizep == size, ("invalid size")); 1867 structsize = sizeof(p->p_fd->fd_cmask); 1868 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1869 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask)); 1870 } 1871 *sizep = size; 1872 } 1873 1874 static void 1875 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 1876 { 1877 struct proc *p; 1878 struct rlimit rlim[RLIM_NLIMITS]; 1879 size_t size; 1880 int structsize, i; 1881 1882 p = (struct proc *)arg; 1883 size = sizeof(structsize) + sizeof(rlim); 1884 if (sb != NULL) { 1885 KASSERT(*sizep == size, ("invalid size")); 1886 structsize = sizeof(rlim); 1887 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1888 PROC_LOCK(p); 1889 for (i = 0; i < RLIM_NLIMITS; i++) 1890 lim_rlimit(p, i, &rlim[i]); 1891 PROC_UNLOCK(p); 1892 sbuf_bcat(sb, rlim, sizeof(rlim)); 1893 } 1894 *sizep = size; 1895 } 1896 1897 static void 1898 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 1899 { 1900 struct proc *p; 1901 size_t size; 1902 int structsize; 1903 1904 p = (struct proc *)arg; 1905 size = sizeof(structsize) + sizeof(p->p_osrel); 1906 if (sb != NULL) { 1907 KASSERT(*sizep == size, ("invalid size")); 1908 structsize = sizeof(p->p_osrel); 1909 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1910 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 1911 } 1912 *sizep = size; 1913 } 1914 1915 static void 1916 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 1917 { 1918 struct proc *p; 1919 elf_ps_strings_t ps_strings; 1920 size_t size; 1921 int structsize; 1922 1923 p = (struct proc *)arg; 1924 size = sizeof(structsize) + sizeof(ps_strings); 1925 if (sb != NULL) { 1926 KASSERT(*sizep == size, ("invalid size")); 1927 structsize = sizeof(ps_strings); 1928 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1929 ps_strings = PTROUT(p->p_sysent->sv_psstrings); 1930 #else 1931 ps_strings = p->p_sysent->sv_psstrings; 1932 #endif 1933 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1934 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 1935 } 1936 *sizep = size; 1937 } 1938 1939 static void 1940 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 1941 { 1942 struct proc *p; 1943 size_t size; 1944 int structsize; 1945 1946 p = (struct proc *)arg; 1947 if (sb == NULL) { 1948 size = 0; 1949 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 1950 sbuf_set_drain(sb, sbuf_drain_count, &size); 1951 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1952 PHOLD(p); 1953 proc_getauxv(curthread, p, sb); 1954 PRELE(p); 1955 sbuf_finish(sb); 1956 sbuf_delete(sb); 1957 *sizep = size; 1958 } else { 1959 structsize = sizeof(Elf_Auxinfo); 1960 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1961 PHOLD(p); 1962 proc_getauxv(curthread, p, sb); 1963 PRELE(p); 1964 } 1965 } 1966 1967 static boolean_t 1968 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote, 1969 int32_t *osrel, const Elf_Phdr *pnote) 1970 { 1971 const Elf_Note *note, *note0, *note_end; 1972 const char *note_name; 1973 int i; 1974 1975 if (pnote == NULL || pnote->p_offset > PAGE_SIZE || 1976 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) 1977 return (FALSE); 1978 1979 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset); 1980 note_end = (const Elf_Note *)(imgp->image_header + 1981 pnote->p_offset + pnote->p_filesz); 1982 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 1983 if (!aligned(note, Elf32_Addr) || (const char *)note_end - 1984 (const char *)note < sizeof(Elf_Note)) 1985 return (FALSE); 1986 if (note->n_namesz != checknote->hdr.n_namesz || 1987 note->n_descsz != checknote->hdr.n_descsz || 1988 note->n_type != checknote->hdr.n_type) 1989 goto nextnote; 1990 note_name = (const char *)(note + 1); 1991 if (note_name + checknote->hdr.n_namesz >= 1992 (const char *)note_end || strncmp(checknote->vendor, 1993 note_name, checknote->hdr.n_namesz) != 0) 1994 goto nextnote; 1995 1996 /* 1997 * Fetch the osreldate for binary 1998 * from the ELF OSABI-note if necessary. 1999 */ 2000 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 2001 checknote->trans_osrel != NULL) 2002 return (checknote->trans_osrel(note, osrel)); 2003 return (TRUE); 2004 2005 nextnote: 2006 note = (const Elf_Note *)((const char *)(note + 1) + 2007 roundup2(note->n_namesz, sizeof(Elf32_Addr)) + 2008 roundup2(note->n_descsz, sizeof(Elf32_Addr))); 2009 } 2010 2011 return (FALSE); 2012 } 2013 2014 /* 2015 * Try to find the appropriate ABI-note section for checknote, 2016 * fetch the osreldate for binary from the ELF OSABI-note. Only the 2017 * first page of the image is searched, the same as for headers. 2018 */ 2019 static boolean_t 2020 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 2021 int32_t *osrel) 2022 { 2023 const Elf_Phdr *phdr; 2024 const Elf_Ehdr *hdr; 2025 int i; 2026 2027 hdr = (const Elf_Ehdr *)imgp->image_header; 2028 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2029 2030 for (i = 0; i < hdr->e_phnum; i++) { 2031 if (phdr[i].p_type == PT_NOTE && 2032 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i])) 2033 return (TRUE); 2034 } 2035 return (FALSE); 2036 2037 } 2038 2039 /* 2040 * Tell kern_execve.c about it, with a little help from the linker. 2041 */ 2042 static struct execsw __elfN(execsw) = { 2043 __CONCAT(exec_, __elfN(imgact)), 2044 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2045 }; 2046 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2047 2048 #ifdef COMPRESS_USER_CORES 2049 /* 2050 * Compress and write out a core segment for a user process. 2051 * 2052 * 'inbuf' is the starting address of a VM segment in the process' address 2053 * space that is to be compressed and written out to the core file. 'dest_buf' 2054 * is a buffer in the kernel's address space. The segment is copied from 2055 * 'inbuf' to 'dest_buf' first before being processed by the compression 2056 * routine gzwrite(). This copying is necessary because the content of the VM 2057 * segment may change between the compression pass and the crc-computation pass 2058 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel. 2059 * 2060 * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'. 2061 */ 2062 static int 2063 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len, 2064 struct thread *td) 2065 { 2066 int len_compressed; 2067 int error = 0; 2068 unsigned int chunk_len; 2069 2070 while (len) { 2071 if (inbuf != NULL) { 2072 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len; 2073 copyin(inbuf, dest_buf, chunk_len); 2074 inbuf += chunk_len; 2075 } else { 2076 chunk_len = len; 2077 } 2078 len_compressed = gzwrite(file, dest_buf, chunk_len); 2079 2080 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed); 2081 2082 if ((unsigned int)len_compressed != chunk_len) { 2083 log(LOG_WARNING, 2084 "compress_core: length mismatch (0x%x returned, " 2085 "0x%x expected)\n", len_compressed, chunk_len); 2086 EVENTHANDLER_INVOKE(app_coredump_error, td, 2087 "compress_core: length mismatch %x -> %x", 2088 chunk_len, len_compressed); 2089 error = EFAULT; 2090 break; 2091 } 2092 len -= chunk_len; 2093 maybe_yield(); 2094 } 2095 2096 return (error); 2097 } 2098 #endif /* COMPRESS_USER_CORES */ 2099 2100 static vm_prot_t 2101 __elfN(trans_prot)(Elf_Word flags) 2102 { 2103 vm_prot_t prot; 2104 2105 prot = 0; 2106 if (flags & PF_X) 2107 prot |= VM_PROT_EXECUTE; 2108 if (flags & PF_W) 2109 prot |= VM_PROT_WRITE; 2110 if (flags & PF_R) 2111 prot |= VM_PROT_READ; 2112 #if __ELF_WORD_SIZE == 32 2113 #if defined(__amd64__) || defined(__ia64__) 2114 if (i386_read_exec && (flags & PF_R)) 2115 prot |= VM_PROT_EXECUTE; 2116 #endif 2117 #endif 2118 return (prot); 2119 } 2120 2121 static Elf_Word 2122 __elfN(untrans_prot)(vm_prot_t prot) 2123 { 2124 Elf_Word flags; 2125 2126 flags = 0; 2127 if (prot & VM_PROT_EXECUTE) 2128 flags |= PF_X; 2129 if (prot & VM_PROT_READ) 2130 flags |= PF_R; 2131 if (prot & VM_PROT_WRITE) 2132 flags |= PF_W; 2133 return (flags); 2134 } 2135