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/capsicum.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/jail.h> 45#include <sys/kernel.h> 46#include <sys/lock.h> 47#include <sys/malloc.h> 48#include <sys/mount.h> 49#include <sys/mman.h> 50#include <sys/namei.h> 51#include <sys/pioctl.h> 52#include <sys/proc.h> 53#include <sys/procfs.h> 54#include <sys/racct.h> 55#include <sys/resourcevar.h> 56#include <sys/rwlock.h> 57#include <sys/sbuf.h> 58#include <sys/sf_buf.h> 59#include <sys/smp.h> 60#include <sys/systm.h> 61#include <sys/signalvar.h> 62#include <sys/stat.h> 63#include <sys/sx.h> 64#include <sys/syscall.h> 65#include <sys/sysctl.h> 66#include <sys/sysent.h> 67#include <sys/vnode.h> 68#include <sys/syslog.h> 69#include <sys/eventhandler.h> 70#include <sys/user.h> 71 72#include <net/zlib.h> 73 74#include <vm/vm.h> 75#include <vm/vm_kern.h> 76#include <vm/vm_param.h> 77#include <vm/pmap.h> 78#include <vm/vm_map.h> 79#include <vm/vm_object.h> 80#include <vm/vm_extern.h> 81 82#include <machine/elf.h> 83#include <machine/md_var.h> 84 85#define ELF_NOTE_ROUNDSIZE 4 86#define OLD_EI_BRAND 8 87 88static int __elfN(check_header)(const Elf_Ehdr *hdr); 89static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 90 const char *interp, int interp_name_len, int32_t *osrel); 91static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 92 u_long *entry, size_t pagesize); 93static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset, 94 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 95 size_t pagesize); 96static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 97static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note, 98 int32_t *osrel); 99static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 100static boolean_t __elfN(check_note)(struct image_params *imgp, 101 Elf_Brandnote *checknote, int32_t *osrel); 102static vm_prot_t __elfN(trans_prot)(Elf_Word); 103static Elf_Word __elfN(untrans_prot)(vm_prot_t); 104 105SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 106 ""); 107 108#ifdef COMPRESS_USER_CORES 109static int compress_core(gzFile, char *, char *, unsigned int, 110 struct thread * td); 111#endif 112#define CORE_BUF_SIZE (16 * 1024) 113 114int __elfN(fallback_brand) = -1; 115SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 116 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 117 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 118TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 119 &__elfN(fallback_brand)); 120 121static int elf_legacy_coredump = 0; 122SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 123 &elf_legacy_coredump, 0, ""); 124 125int __elfN(nxstack) = 126#if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ 127 1; 128#else 129 0; 130#endif 131SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 132 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 133 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack"); 134 135#if __ELF_WORD_SIZE == 32 136#if defined(__amd64__) || defined(__ia64__) 137int i386_read_exec = 0; 138SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 139 "enable execution from readable segments"); 140#endif 141#endif 142 143static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 144 145#define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 146#define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 147#define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a)) 148 149static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; 150 151Elf_Brandnote __elfN(freebsd_brandnote) = { 152 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 153 .hdr.n_descsz = sizeof(int32_t), 154 .hdr.n_type = 1, 155 .vendor = FREEBSD_ABI_VENDOR, 156 .flags = BN_TRANSLATE_OSREL, 157 .trans_osrel = __elfN(freebsd_trans_osrel) 158}; 159 160static boolean_t 161__elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 162{ 163 uintptr_t p; 164 165 p = (uintptr_t)(note + 1); 166 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 167 *osrel = *(const int32_t *)(p); 168 169 return (TRUE); 170} 171 172static const char GNU_ABI_VENDOR[] = "GNU"; 173static int GNU_KFREEBSD_ABI_DESC = 3; 174 175Elf_Brandnote __elfN(kfreebsd_brandnote) = { 176 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 177 .hdr.n_descsz = 16, /* XXX at least 16 */ 178 .hdr.n_type = 1, 179 .vendor = GNU_ABI_VENDOR, 180 .flags = BN_TRANSLATE_OSREL, 181 .trans_osrel = kfreebsd_trans_osrel 182}; 183 184static boolean_t 185kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 186{ 187 const Elf32_Word *desc; 188 uintptr_t p; 189 190 p = (uintptr_t)(note + 1); 191 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 192 193 desc = (const Elf32_Word *)p; 194 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 195 return (FALSE); 196 197 /* 198 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 199 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 200 */ 201 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 202 203 return (TRUE); 204} 205 206int 207__elfN(insert_brand_entry)(Elf_Brandinfo *entry) 208{ 209 int i; 210 211 for (i = 0; i < MAX_BRANDS; i++) { 212 if (elf_brand_list[i] == NULL) { 213 elf_brand_list[i] = entry; 214 break; 215 } 216 } 217 if (i == MAX_BRANDS) { 218 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 219 __func__, entry); 220 return (-1); 221 } 222 return (0); 223} 224 225int 226__elfN(remove_brand_entry)(Elf_Brandinfo *entry) 227{ 228 int i; 229 230 for (i = 0; i < MAX_BRANDS; i++) { 231 if (elf_brand_list[i] == entry) { 232 elf_brand_list[i] = NULL; 233 break; 234 } 235 } 236 if (i == MAX_BRANDS) 237 return (-1); 238 return (0); 239} 240 241int 242__elfN(brand_inuse)(Elf_Brandinfo *entry) 243{ 244 struct proc *p; 245 int rval = FALSE; 246 247 sx_slock(&allproc_lock); 248 FOREACH_PROC_IN_SYSTEM(p) { 249 if (p->p_sysent == entry->sysvec) { 250 rval = TRUE; 251 break; 252 } 253 } 254 sx_sunlock(&allproc_lock); 255 256 return (rval); 257} 258 259static Elf_Brandinfo * 260__elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 261 int interp_name_len, int32_t *osrel) 262{ 263 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 264 Elf_Brandinfo *bi; 265 boolean_t ret; 266 int i; 267 268 /* 269 * We support four types of branding -- (1) the ELF EI_OSABI field 270 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 271 * branding w/in the ELF header, (3) path of the `interp_path' 272 * field, and (4) the ".note.ABI-tag" ELF section. 273 */ 274 275 /* Look for an ".note.ABI-tag" ELF section */ 276 for (i = 0; i < MAX_BRANDS; i++) { 277 bi = elf_brand_list[i]; 278 if (bi == NULL) 279 continue; 280 if (hdr->e_machine == bi->machine && (bi->flags & 281 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 282 ret = __elfN(check_note)(imgp, bi->brand_note, osrel); 283 if (ret) 284 return (bi); 285 } 286 } 287 288 /* If the executable has a brand, search for it in the brand list. */ 289 for (i = 0; i < MAX_BRANDS; i++) { 290 bi = elf_brand_list[i]; 291 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 292 continue; 293 if (hdr->e_machine == bi->machine && 294 (hdr->e_ident[EI_OSABI] == bi->brand || 295 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 296 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 297 return (bi); 298 } 299 300 /* Lacking a known brand, search for a recognized interpreter. */ 301 if (interp != NULL) { 302 for (i = 0; i < MAX_BRANDS; i++) { 303 bi = elf_brand_list[i]; 304 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 305 continue; 306 if (hdr->e_machine == bi->machine && 307 /* ELF image p_filesz includes terminating zero */ 308 strlen(bi->interp_path) + 1 == interp_name_len && 309 strncmp(interp, bi->interp_path, interp_name_len) 310 == 0) 311 return (bi); 312 } 313 } 314 315 /* Lacking a recognized interpreter, try the default brand */ 316 for (i = 0; i < MAX_BRANDS; i++) { 317 bi = elf_brand_list[i]; 318 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 319 continue; 320 if (hdr->e_machine == bi->machine && 321 __elfN(fallback_brand) == bi->brand) 322 return (bi); 323 } 324 return (NULL); 325} 326 327static int 328__elfN(check_header)(const Elf_Ehdr *hdr) 329{ 330 Elf_Brandinfo *bi; 331 int i; 332 333 if (!IS_ELF(*hdr) || 334 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 335 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 336 hdr->e_ident[EI_VERSION] != EV_CURRENT || 337 hdr->e_phentsize != sizeof(Elf_Phdr) || 338 hdr->e_version != ELF_TARG_VER) 339 return (ENOEXEC); 340 341 /* 342 * Make sure we have at least one brand for this machine. 343 */ 344 345 for (i = 0; i < MAX_BRANDS; i++) { 346 bi = elf_brand_list[i]; 347 if (bi != NULL && bi->machine == hdr->e_machine) 348 break; 349 } 350 if (i == MAX_BRANDS) 351 return (ENOEXEC); 352 353 return (0); 354} 355 356static int 357__elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 358 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 359{ 360 struct sf_buf *sf; 361 int error; 362 vm_offset_t off; 363 364 /* 365 * Create the page if it doesn't exist yet. Ignore errors. 366 */ 367 vm_map_lock(map); 368 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), 369 VM_PROT_ALL, VM_PROT_ALL, 0); 370 vm_map_unlock(map); 371 372 /* 373 * Find the page from the underlying object. 374 */ 375 if (object) { 376 sf = vm_imgact_map_page(object, offset); 377 if (sf == NULL) 378 return (KERN_FAILURE); 379 off = offset - trunc_page(offset); 380 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 381 end - start); 382 vm_imgact_unmap_page(sf); 383 if (error) { 384 return (KERN_FAILURE); 385 } 386 } 387 388 return (KERN_SUCCESS); 389} 390 391static int 392__elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 393 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) 394{ 395 struct sf_buf *sf; 396 vm_offset_t off; 397 vm_size_t sz; 398 int error, rv; 399 400 if (start != trunc_page(start)) { 401 rv = __elfN(map_partial)(map, object, offset, start, 402 round_page(start), prot); 403 if (rv) 404 return (rv); 405 offset += round_page(start) - start; 406 start = round_page(start); 407 } 408 if (end != round_page(end)) { 409 rv = __elfN(map_partial)(map, object, offset + 410 trunc_page(end) - start, trunc_page(end), end, prot); 411 if (rv) 412 return (rv); 413 end = trunc_page(end); 414 } 415 if (end > start) { 416 if (offset & PAGE_MASK) { 417 /* 418 * The mapping is not page aligned. This means we have 419 * to copy the data. Sigh. 420 */ 421 rv = vm_map_find(map, NULL, 0, &start, end - start, 0, 422 VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL, 423 0); 424 if (rv) 425 return (rv); 426 if (object == NULL) 427 return (KERN_SUCCESS); 428 for (; start < end; start += sz) { 429 sf = vm_imgact_map_page(object, offset); 430 if (sf == NULL) 431 return (KERN_FAILURE); 432 off = offset - trunc_page(offset); 433 sz = end - start; 434 if (sz > PAGE_SIZE - off) 435 sz = PAGE_SIZE - off; 436 error = copyout((caddr_t)sf_buf_kva(sf) + off, 437 (caddr_t)start, sz); 438 vm_imgact_unmap_page(sf); 439 if (error) { 440 return (KERN_FAILURE); 441 } 442 offset += sz; 443 } 444 rv = KERN_SUCCESS; 445 } else { 446 vm_object_reference(object); 447 vm_map_lock(map); 448 rv = vm_map_insert(map, object, offset, start, end, 449 prot, VM_PROT_ALL, cow); 450 vm_map_unlock(map); 451 if (rv != KERN_SUCCESS) 452 vm_object_deallocate(object); 453 } 454 return (rv); 455 } else { 456 return (KERN_SUCCESS); 457 } 458} 459 460static int 461__elfN(load_section)(struct image_params *imgp, vm_offset_t offset, 462 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 463 size_t pagesize) 464{ 465 struct sf_buf *sf; 466 size_t map_len; 467 vm_map_t map; 468 vm_object_t object; 469 vm_offset_t map_addr; 470 int error, rv, cow; 471 size_t copy_len; 472 vm_offset_t file_addr; 473 474 /* 475 * It's necessary to fail if the filsz + offset taken from the 476 * header is greater than the actual file pager object's size. 477 * If we were to allow this, then the vm_map_find() below would 478 * walk right off the end of the file object and into the ether. 479 * 480 * While I'm here, might as well check for something else that 481 * is invalid: filsz cannot be greater than memsz. 482 */ 483 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) { 484 uprintf("elf_load_section: truncated ELF file\n"); 485 return (ENOEXEC); 486 } 487 488 object = imgp->object; 489 map = &imgp->proc->p_vmspace->vm_map; 490 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 491 file_addr = trunc_page_ps(offset, pagesize); 492 493 /* 494 * We have two choices. We can either clear the data in the last page 495 * of an oversized mapping, or we can start the anon mapping a page 496 * early and copy the initialized data into that first page. We 497 * choose the second.. 498 */ 499 if (memsz > filsz) 500 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 501 else 502 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 503 504 if (map_len != 0) { 505 /* cow flags: don't dump readonly sections in core */ 506 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 507 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 508 509 rv = __elfN(map_insert)(map, 510 object, 511 file_addr, /* file offset */ 512 map_addr, /* virtual start */ 513 map_addr + map_len,/* virtual end */ 514 prot, 515 cow); 516 if (rv != KERN_SUCCESS) 517 return (EINVAL); 518 519 /* we can stop now if we've covered it all */ 520 if (memsz == filsz) { 521 return (0); 522 } 523 } 524 525 526 /* 527 * We have to get the remaining bit of the file into the first part 528 * of the oversized map segment. This is normally because the .data 529 * segment in the file is extended to provide bss. It's a neat idea 530 * to try and save a page, but it's a pain in the behind to implement. 531 */ 532 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 533 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 534 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 535 map_addr; 536 537 /* This had damn well better be true! */ 538 if (map_len != 0) { 539 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr + 540 map_len, VM_PROT_ALL, 0); 541 if (rv != KERN_SUCCESS) { 542 return (EINVAL); 543 } 544 } 545 546 if (copy_len != 0) { 547 vm_offset_t off; 548 549 sf = vm_imgact_map_page(object, offset + filsz); 550 if (sf == NULL) 551 return (EIO); 552 553 /* send the page fragment to user space */ 554 off = trunc_page_ps(offset + filsz, pagesize) - 555 trunc_page(offset + filsz); 556 error = copyout((caddr_t)sf_buf_kva(sf) + off, 557 (caddr_t)map_addr, copy_len); 558 vm_imgact_unmap_page(sf); 559 if (error) { 560 return (error); 561 } 562 } 563 564 /* 565 * set it to the specified protection. 566 * XXX had better undo the damage from pasting over the cracks here! 567 */ 568 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 569 map_len), prot, FALSE); 570 571 return (0); 572} 573 574/* 575 * Load the file "file" into memory. It may be either a shared object 576 * or an executable. 577 * 578 * The "addr" reference parameter is in/out. On entry, it specifies 579 * the address where a shared object should be loaded. If the file is 580 * an executable, this value is ignored. On exit, "addr" specifies 581 * where the file was actually loaded. 582 * 583 * The "entry" reference parameter is out only. On exit, it specifies 584 * the entry point for the loaded file. 585 */ 586static int 587__elfN(load_file)(struct proc *p, const char *file, u_long *addr, 588 u_long *entry, size_t pagesize) 589{ 590 struct { 591 struct nameidata nd; 592 struct vattr attr; 593 struct image_params image_params; 594 } *tempdata; 595 const Elf_Ehdr *hdr = NULL; 596 const Elf_Phdr *phdr = NULL; 597 struct nameidata *nd; 598 struct vattr *attr; 599 struct image_params *imgp; 600 vm_prot_t prot; 601 u_long rbase; 602 u_long base_addr = 0; 603 int error, i, numsegs; 604 605#ifdef CAPABILITY_MODE 606 /* 607 * XXXJA: This check can go away once we are sufficiently confident 608 * that the checks in namei() are correct. 609 */ 610 if (IN_CAPABILITY_MODE(curthread)) 611 return (ECAPMODE); 612#endif 613 614 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 615 nd = &tempdata->nd; 616 attr = &tempdata->attr; 617 imgp = &tempdata->image_params; 618 619 /* 620 * Initialize part of the common data 621 */ 622 imgp->proc = p; 623 imgp->attr = attr; 624 imgp->firstpage = NULL; 625 imgp->image_header = NULL; 626 imgp->object = NULL; 627 imgp->execlabel = NULL; 628 629 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread); 630 if ((error = namei(nd)) != 0) { 631 nd->ni_vp = NULL; 632 goto fail; 633 } 634 NDFREE(nd, NDF_ONLY_PNBUF); 635 imgp->vp = nd->ni_vp; 636 637 /* 638 * Check permissions, modes, uid, etc on the file, and "open" it. 639 */ 640 error = exec_check_permissions(imgp); 641 if (error) 642 goto fail; 643 644 error = exec_map_first_page(imgp); 645 if (error) 646 goto fail; 647 648 /* 649 * Also make certain that the interpreter stays the same, so set 650 * its VV_TEXT flag, too. 651 */ 652 VOP_SET_TEXT(nd->ni_vp); 653 654 imgp->object = nd->ni_vp->v_object; 655 656 hdr = (const Elf_Ehdr *)imgp->image_header; 657 if ((error = __elfN(check_header)(hdr)) != 0) 658 goto fail; 659 if (hdr->e_type == ET_DYN) 660 rbase = *addr; 661 else if (hdr->e_type == ET_EXEC) 662 rbase = 0; 663 else { 664 error = ENOEXEC; 665 goto fail; 666 } 667 668 /* Only support headers that fit within first page for now */ 669 if ((hdr->e_phoff > PAGE_SIZE) || 670 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 671 error = ENOEXEC; 672 goto fail; 673 } 674 675 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 676 if (!aligned(phdr, Elf_Addr)) { 677 error = ENOEXEC; 678 goto fail; 679 } 680 681 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 682 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 683 /* Loadable segment */ 684 prot = __elfN(trans_prot)(phdr[i].p_flags); 685 error = __elfN(load_section)(imgp, phdr[i].p_offset, 686 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 687 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize); 688 if (error != 0) 689 goto fail; 690 /* 691 * Establish the base address if this is the 692 * first segment. 693 */ 694 if (numsegs == 0) 695 base_addr = trunc_page(phdr[i].p_vaddr + 696 rbase); 697 numsegs++; 698 } 699 } 700 *addr = base_addr; 701 *entry = (unsigned long)hdr->e_entry + rbase; 702 703fail: 704 if (imgp->firstpage) 705 exec_unmap_first_page(imgp); 706 707 if (nd->ni_vp) 708 vput(nd->ni_vp); 709 710 free(tempdata, M_TEMP); 711 712 return (error); 713} 714 715static int 716__CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 717{ 718 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 719 const Elf_Phdr *phdr; 720 Elf_Auxargs *elf_auxargs; 721 struct vmspace *vmspace; 722 vm_prot_t prot; 723 u_long text_size = 0, data_size = 0, total_size = 0; 724 u_long text_addr = 0, data_addr = 0; 725 u_long seg_size, seg_addr; 726 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0; 727 int32_t osrel = 0; 728 int error = 0, i, n, interp_name_len = 0; 729 const char *interp = NULL, *newinterp = NULL; 730 Elf_Brandinfo *brand_info; 731 char *path; 732 struct sysentvec *sv; 733 734 /* 735 * Do we have a valid ELF header ? 736 * 737 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 738 * if particular brand doesn't support it. 739 */ 740 if (__elfN(check_header)(hdr) != 0 || 741 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 742 return (-1); 743 744 /* 745 * From here on down, we return an errno, not -1, as we've 746 * detected an ELF file. 747 */ 748 749 if ((hdr->e_phoff > PAGE_SIZE) || 750 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 751 /* Only support headers in first page for now */ 752 return (ENOEXEC); 753 } 754 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 755 if (!aligned(phdr, Elf_Addr)) 756 return (ENOEXEC); 757 n = 0; 758 baddr = 0; 759 for (i = 0; i < hdr->e_phnum; i++) { 760 switch (phdr[i].p_type) { 761 case PT_LOAD: 762 if (n == 0) 763 baddr = phdr[i].p_vaddr; 764 n++; 765 break; 766 case PT_INTERP: 767 /* Path to interpreter */ 768 if (phdr[i].p_filesz > MAXPATHLEN || 769 phdr[i].p_offset > PAGE_SIZE || 770 phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset) 771 return (ENOEXEC); 772 interp = imgp->image_header + phdr[i].p_offset; 773 interp_name_len = phdr[i].p_filesz; 774 break; 775 case PT_GNU_STACK: 776 if (__elfN(nxstack)) 777 imgp->stack_prot = 778 __elfN(trans_prot)(phdr[i].p_flags); 779 imgp->stack_sz = phdr[i].p_memsz; 780 break; 781 } 782 } 783 784 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len, 785 &osrel); 786 if (brand_info == NULL) { 787 uprintf("ELF binary type \"%u\" not known.\n", 788 hdr->e_ident[EI_OSABI]); 789 return (ENOEXEC); 790 } 791 if (hdr->e_type == ET_DYN) { 792 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) 793 return (ENOEXEC); 794 /* 795 * Honour the base load address from the dso if it is 796 * non-zero for some reason. 797 */ 798 if (baddr == 0) 799 et_dyn_addr = ET_DYN_LOAD_ADDR; 800 else 801 et_dyn_addr = 0; 802 } else 803 et_dyn_addr = 0; 804 sv = brand_info->sysvec; 805 if (interp != NULL && brand_info->interp_newpath != NULL) 806 newinterp = brand_info->interp_newpath; 807 808 /* 809 * Avoid a possible deadlock if the current address space is destroyed 810 * and that address space maps the locked vnode. In the common case, 811 * the locked vnode's v_usecount is decremented but remains greater 812 * than zero. Consequently, the vnode lock is not needed by vrele(). 813 * However, in cases where the vnode lock is external, such as nullfs, 814 * v_usecount may become zero. 815 * 816 * The VV_TEXT flag prevents modifications to the executable while 817 * the vnode is unlocked. 818 */ 819 VOP_UNLOCK(imgp->vp, 0); 820 821 error = exec_new_vmspace(imgp, sv); 822 imgp->proc->p_sysent = sv; 823 824 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 825 if (error) 826 return (error); 827 828 for (i = 0; i < hdr->e_phnum; i++) { 829 switch (phdr[i].p_type) { 830 case PT_LOAD: /* Loadable segment */ 831 if (phdr[i].p_memsz == 0) 832 break; 833 prot = __elfN(trans_prot)(phdr[i].p_flags); 834 error = __elfN(load_section)(imgp, phdr[i].p_offset, 835 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 836 phdr[i].p_memsz, phdr[i].p_filesz, prot, 837 sv->sv_pagesize); 838 if (error != 0) 839 return (error); 840 841 /* 842 * If this segment contains the program headers, 843 * remember their virtual address for the AT_PHDR 844 * aux entry. Static binaries don't usually include 845 * a PT_PHDR entry. 846 */ 847 if (phdr[i].p_offset == 0 && 848 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 849 <= phdr[i].p_filesz) 850 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 851 et_dyn_addr; 852 853 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 854 seg_size = round_page(phdr[i].p_memsz + 855 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 856 857 /* 858 * Make the largest executable segment the official 859 * text segment and all others data. 860 * 861 * Note that obreak() assumes that data_addr + 862 * data_size == end of data load area, and the ELF 863 * file format expects segments to be sorted by 864 * address. If multiple data segments exist, the 865 * last one will be used. 866 */ 867 868 if (phdr[i].p_flags & PF_X && text_size < seg_size) { 869 text_size = seg_size; 870 text_addr = seg_addr; 871 } else { 872 data_size = seg_size; 873 data_addr = seg_addr; 874 } 875 total_size += seg_size; 876 break; 877 case PT_PHDR: /* Program header table info */ 878 proghdr = phdr[i].p_vaddr + et_dyn_addr; 879 break; 880 default: 881 break; 882 } 883 } 884 885 if (data_addr == 0 && data_size == 0) { 886 data_addr = text_addr; 887 data_size = text_size; 888 } 889 890 entry = (u_long)hdr->e_entry + et_dyn_addr; 891 892 /* 893 * Check limits. It should be safe to check the 894 * limits after loading the segments since we do 895 * not actually fault in all the segments pages. 896 */ 897 PROC_LOCK(imgp->proc); 898 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 899 text_size > maxtsiz || 900 total_size > lim_cur(imgp->proc, RLIMIT_VMEM) || 901 racct_set(imgp->proc, RACCT_DATA, data_size) != 0 || 902 racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) { 903 PROC_UNLOCK(imgp->proc); 904 return (ENOMEM); 905 } 906 907 vmspace = imgp->proc->p_vmspace; 908 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 909 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 910 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 911 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 912 913 /* 914 * We load the dynamic linker where a userland call 915 * to mmap(0, ...) would put it. The rationale behind this 916 * calculation is that it leaves room for the heap to grow to 917 * its maximum allowed size. 918 */ 919 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc, 920 RLIMIT_DATA)); 921 PROC_UNLOCK(imgp->proc); 922 923 imgp->entry_addr = entry; 924 925 if (interp != NULL) { 926 int have_interp = FALSE; 927 VOP_UNLOCK(imgp->vp, 0); 928 if (brand_info->emul_path != NULL && 929 brand_info->emul_path[0] != '\0') { 930 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 931 snprintf(path, MAXPATHLEN, "%s%s", 932 brand_info->emul_path, interp); 933 error = __elfN(load_file)(imgp->proc, path, &addr, 934 &imgp->entry_addr, sv->sv_pagesize); 935 free(path, M_TEMP); 936 if (error == 0) 937 have_interp = TRUE; 938 } 939 if (!have_interp && newinterp != NULL) { 940 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 941 &imgp->entry_addr, sv->sv_pagesize); 942 if (error == 0) 943 have_interp = TRUE; 944 } 945 if (!have_interp) { 946 error = __elfN(load_file)(imgp->proc, interp, &addr, 947 &imgp->entry_addr, sv->sv_pagesize); 948 } 949 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 950 if (error != 0) { 951 uprintf("ELF interpreter %s not found\n", interp); 952 return (error); 953 } 954 } else 955 addr = et_dyn_addr; 956 957 /* 958 * Construct auxargs table (used by the fixup routine) 959 */ 960 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 961 elf_auxargs->execfd = -1; 962 elf_auxargs->phdr = proghdr; 963 elf_auxargs->phent = hdr->e_phentsize; 964 elf_auxargs->phnum = hdr->e_phnum; 965 elf_auxargs->pagesz = PAGE_SIZE; 966 elf_auxargs->base = addr; 967 elf_auxargs->flags = 0; 968 elf_auxargs->entry = entry; 969 970 imgp->auxargs = elf_auxargs; 971 imgp->interpreted = 0; 972 imgp->reloc_base = addr; 973 imgp->proc->p_osrel = osrel; 974 975 return (error); 976} 977 978#define suword __CONCAT(suword, __ELF_WORD_SIZE) 979 980int 981__elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 982{ 983 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 984 Elf_Addr *base; 985 Elf_Addr *pos; 986 987 base = (Elf_Addr *)*stack_base; 988 pos = base + (imgp->args->argc + imgp->args->envc + 2); 989 990 if (args->execfd != -1) 991 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 992 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 993 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 994 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 995 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 996 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 997 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 998 AUXARGS_ENTRY(pos, AT_BASE, args->base); 999 if (imgp->execpathp != 0) 1000 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 1001 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1002 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1003 if (imgp->canary != 0) { 1004 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary); 1005 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1006 } 1007 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1008 if (imgp->pagesizes != 0) { 1009 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes); 1010 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1011 } 1012 if (imgp->sysent->sv_timekeep_base != 0) { 1013 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1014 imgp->sysent->sv_timekeep_base); 1015 } 1016 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1017 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1018 imgp->sysent->sv_stackprot); 1019 AUXARGS_ENTRY(pos, AT_NULL, 0); 1020 1021 free(imgp->auxargs, M_TEMP); 1022 imgp->auxargs = NULL; 1023 1024 base--; 1025 suword(base, (long)imgp->args->argc); 1026 *stack_base = (register_t *)base; 1027 return (0); 1028} 1029 1030/* 1031 * Code for generating ELF core dumps. 1032 */ 1033 1034typedef void (*segment_callback)(vm_map_entry_t, void *); 1035 1036/* Closure for cb_put_phdr(). */ 1037struct phdr_closure { 1038 Elf_Phdr *phdr; /* Program header to fill in */ 1039 Elf_Off offset; /* Offset of segment in core file */ 1040}; 1041 1042/* Closure for cb_size_segment(). */ 1043struct sseg_closure { 1044 int count; /* Count of writable segments. */ 1045 size_t size; /* Total size of all writable segments. */ 1046}; 1047 1048typedef void (*outfunc_t)(void *, struct sbuf *, size_t *); 1049 1050struct note_info { 1051 int type; /* Note type. */ 1052 outfunc_t outfunc; /* Output function. */ 1053 void *outarg; /* Argument for the output function. */ 1054 size_t outsize; /* Output size. */ 1055 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1056}; 1057 1058TAILQ_HEAD(note_info_list, note_info); 1059 1060static void cb_put_phdr(vm_map_entry_t, void *); 1061static void cb_size_segment(vm_map_entry_t, void *); 1062static void each_writable_segment(struct thread *, segment_callback, void *); 1063static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 1064 int, void *, size_t, struct note_info_list *, size_t, gzFile); 1065static void __elfN(prepare_notes)(struct thread *, struct note_info_list *, 1066 size_t *); 1067static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t); 1068static void __elfN(putnote)(struct note_info *, struct sbuf *); 1069static size_t register_note(struct note_info_list *, int, outfunc_t, void *); 1070static int sbuf_drain_core_output(void *, const char *, int); 1071static int sbuf_drain_count(void *arg, const char *data, int len); 1072 1073static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *); 1074static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1075static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *); 1076static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1077static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *); 1078static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1079static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1080static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1081static void note_procstat_files(void *, struct sbuf *, size_t *); 1082static void note_procstat_groups(void *, struct sbuf *, size_t *); 1083static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1084static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1085static void note_procstat_umask(void *, struct sbuf *, size_t *); 1086static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1087 1088#ifdef COMPRESS_USER_CORES 1089extern int compress_user_cores; 1090extern int compress_user_cores_gzlevel; 1091#endif 1092 1093static int 1094core_output(struct vnode *vp, void *base, size_t len, off_t offset, 1095 struct ucred *active_cred, struct ucred *file_cred, 1096 struct thread *td, char *core_buf, gzFile gzfile) { 1097 1098 int error; 1099 if (gzfile) { 1100#ifdef COMPRESS_USER_CORES 1101 error = compress_core(gzfile, base, core_buf, len, td); 1102#else 1103 panic("shouldn't be here"); 1104#endif 1105 } else { 1106 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset, 1107 UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred, 1108 NULL, td); 1109 } 1110 return (error); 1111} 1112 1113/* Coredump output parameters for sbuf drain routine. */ 1114struct sbuf_drain_core_params { 1115 off_t offset; 1116 struct ucred *active_cred; 1117 struct ucred *file_cred; 1118 struct thread *td; 1119 struct vnode *vp; 1120#ifdef COMPRESS_USER_CORES 1121 gzFile gzfile; 1122#endif 1123}; 1124 1125/* 1126 * Drain into a core file. 1127 */ 1128static int 1129sbuf_drain_core_output(void *arg, const char *data, int len) 1130{ 1131 struct sbuf_drain_core_params *p; 1132 int error, locked; 1133 1134 p = (struct sbuf_drain_core_params *)arg; 1135 1136 /* 1137 * Some kern_proc out routines that print to this sbuf may 1138 * call us with the process lock held. Draining with the 1139 * non-sleepable lock held is unsafe. The lock is needed for 1140 * those routines when dumping a live process. In our case we 1141 * can safely release the lock before draining and acquire 1142 * again after. 1143 */ 1144 locked = PROC_LOCKED(p->td->td_proc); 1145 if (locked) 1146 PROC_UNLOCK(p->td->td_proc); 1147#ifdef COMPRESS_USER_CORES 1148 if (p->gzfile != Z_NULL) 1149 error = compress_core(p->gzfile, NULL, __DECONST(char *, data), 1150 len, p->td); 1151 else 1152#endif 1153 error = vn_rdwr_inchunks(UIO_WRITE, p->vp, 1154 __DECONST(void *, data), len, p->offset, UIO_SYSSPACE, 1155 IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL, 1156 p->td); 1157 if (locked) 1158 PROC_LOCK(p->td->td_proc); 1159 if (error != 0) 1160 return (-error); 1161 p->offset += len; 1162 return (len); 1163} 1164 1165/* 1166 * Drain into a counter. 1167 */ 1168static int 1169sbuf_drain_count(void *arg, const char *data __unused, int len) 1170{ 1171 size_t *sizep; 1172 1173 sizep = (size_t *)arg; 1174 *sizep += len; 1175 return (len); 1176} 1177 1178int 1179__elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1180{ 1181 struct ucred *cred = td->td_ucred; 1182 int error = 0; 1183 struct sseg_closure seginfo; 1184 struct note_info_list notelst; 1185 struct note_info *ninfo; 1186 void *hdr; 1187 size_t hdrsize, notesz, coresize; 1188 1189 gzFile gzfile = Z_NULL; 1190 char *core_buf = NULL; 1191#ifdef COMPRESS_USER_CORES 1192 char gzopen_flags[8]; 1193 char *p; 1194 int doing_compress = flags & IMGACT_CORE_COMPRESS; 1195#endif 1196 1197 hdr = NULL; 1198 TAILQ_INIT(¬elst); 1199 1200#ifdef COMPRESS_USER_CORES 1201 if (doing_compress) { 1202 p = gzopen_flags; 1203 *p++ = 'w'; 1204 if (compress_user_cores_gzlevel >= 0 && 1205 compress_user_cores_gzlevel <= 9) 1206 *p++ = '0' + compress_user_cores_gzlevel; 1207 *p = 0; 1208 gzfile = gz_open("", gzopen_flags, vp); 1209 if (gzfile == Z_NULL) { 1210 error = EFAULT; 1211 goto done; 1212 } 1213 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1214 if (!core_buf) { 1215 error = ENOMEM; 1216 goto done; 1217 } 1218 } 1219#endif 1220 1221 /* Size the program segments. */ 1222 seginfo.count = 0; 1223 seginfo.size = 0; 1224 each_writable_segment(td, cb_size_segment, &seginfo); 1225 1226 /* 1227 * Collect info about the core file header area. 1228 */ 1229 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1230 __elfN(prepare_notes)(td, ¬elst, ¬esz); 1231 coresize = round_page(hdrsize + notesz) + seginfo.size; 1232 1233#ifdef RACCT
|
1242 } 1243#endif 1244 if (coresize >= limit) { 1245 error = EFAULT; 1246 goto done; 1247 } 1248 1249 /* 1250 * Allocate memory for building the header, fill it up, 1251 * and write it out following the notes. 1252 */ 1253 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1254 if (hdr == NULL) { 1255 error = EINVAL; 1256 goto done; 1257 } 1258 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize, 1259 ¬elst, notesz, gzfile); 1260 1261 /* Write the contents of all of the writable segments. */ 1262 if (error == 0) { 1263 Elf_Phdr *php; 1264 off_t offset; 1265 int i; 1266 1267 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1268 offset = round_page(hdrsize + notesz); 1269 for (i = 0; i < seginfo.count; i++) { 1270 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr, 1271 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile); 1272 if (error != 0) 1273 break; 1274 offset += php->p_filesz; 1275 php++; 1276 } 1277 } 1278 if (error) { 1279 log(LOG_WARNING, 1280 "Failed to write core file for process %s (error %d)\n", 1281 curproc->p_comm, error); 1282 } 1283 1284done: 1285#ifdef COMPRESS_USER_CORES 1286 if (core_buf) 1287 free(core_buf, M_TEMP); 1288 if (gzfile) 1289 gzclose(gzfile); 1290#endif 1291 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1292 TAILQ_REMOVE(¬elst, ninfo, link); 1293 free(ninfo, M_TEMP); 1294 } 1295 if (hdr != NULL) 1296 free(hdr, M_TEMP); 1297 1298 return (error); 1299} 1300 1301/* 1302 * A callback for each_writable_segment() to write out the segment's 1303 * program header entry. 1304 */ 1305static void 1306cb_put_phdr(entry, closure) 1307 vm_map_entry_t entry; 1308 void *closure; 1309{ 1310 struct phdr_closure *phc = (struct phdr_closure *)closure; 1311 Elf_Phdr *phdr = phc->phdr; 1312 1313 phc->offset = round_page(phc->offset); 1314 1315 phdr->p_type = PT_LOAD; 1316 phdr->p_offset = phc->offset; 1317 phdr->p_vaddr = entry->start; 1318 phdr->p_paddr = 0; 1319 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1320 phdr->p_align = PAGE_SIZE; 1321 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1322 1323 phc->offset += phdr->p_filesz; 1324 phc->phdr++; 1325} 1326 1327/* 1328 * A callback for each_writable_segment() to gather information about 1329 * the number of segments and their total size. 1330 */ 1331static void 1332cb_size_segment(entry, closure) 1333 vm_map_entry_t entry; 1334 void *closure; 1335{ 1336 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1337 1338 ssc->count++; 1339 ssc->size += entry->end - entry->start; 1340} 1341 1342/* 1343 * For each writable segment in the process's memory map, call the given 1344 * function with a pointer to the map entry and some arbitrary 1345 * caller-supplied data. 1346 */ 1347static void 1348each_writable_segment(td, func, closure) 1349 struct thread *td; 1350 segment_callback func; 1351 void *closure; 1352{ 1353 struct proc *p = td->td_proc; 1354 vm_map_t map = &p->p_vmspace->vm_map; 1355 vm_map_entry_t entry; 1356 vm_object_t backing_object, object; 1357 boolean_t ignore_entry; 1358 1359 vm_map_lock_read(map); 1360 for (entry = map->header.next; entry != &map->header; 1361 entry = entry->next) { 1362 /* 1363 * Don't dump inaccessible mappings, deal with legacy 1364 * coredump mode. 1365 * 1366 * Note that read-only segments related to the elf binary 1367 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1368 * need to arbitrarily ignore such segments. 1369 */ 1370 if (elf_legacy_coredump) { 1371 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1372 continue; 1373 } else { 1374 if ((entry->protection & VM_PROT_ALL) == 0) 1375 continue; 1376 } 1377 1378 /* 1379 * Dont include memory segment in the coredump if 1380 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1381 * madvise(2). Do not dump submaps (i.e. parts of the 1382 * kernel map). 1383 */ 1384 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1385 continue; 1386 1387 if ((object = entry->object.vm_object) == NULL) 1388 continue; 1389 1390 /* Ignore memory-mapped devices and such things. */ 1391 VM_OBJECT_RLOCK(object); 1392 while ((backing_object = object->backing_object) != NULL) { 1393 VM_OBJECT_RLOCK(backing_object); 1394 VM_OBJECT_RUNLOCK(object); 1395 object = backing_object; 1396 } 1397 ignore_entry = object->type != OBJT_DEFAULT && 1398 object->type != OBJT_SWAP && object->type != OBJT_VNODE && 1399 object->type != OBJT_PHYS; 1400 VM_OBJECT_RUNLOCK(object); 1401 if (ignore_entry) 1402 continue; 1403 1404 (*func)(entry, closure); 1405 } 1406 vm_map_unlock_read(map); 1407} 1408 1409/* 1410 * Write the core file header to the file, including padding up to 1411 * the page boundary. 1412 */ 1413static int 1414__elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred, 1415 int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst, 1416 size_t notesz, gzFile gzfile) 1417{ 1418 struct sbuf_drain_core_params params; 1419 struct note_info *ninfo; 1420 struct sbuf *sb; 1421 int error; 1422 1423 /* Fill in the header. */ 1424 bzero(hdr, hdrsize); 1425 __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz); 1426 1427 params.offset = 0; 1428 params.active_cred = cred; 1429 params.file_cred = NOCRED; 1430 params.td = td; 1431 params.vp = vp; 1432#ifdef COMPRESS_USER_CORES 1433 params.gzfile = gzfile; 1434#endif 1435 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1436 sbuf_set_drain(sb, sbuf_drain_core_output, ¶ms); 1437 sbuf_start_section(sb, NULL); 1438 sbuf_bcat(sb, hdr, hdrsize); 1439 TAILQ_FOREACH(ninfo, notelst, link) 1440 __elfN(putnote)(ninfo, sb); 1441 /* Align up to a page boundary for the program segments. */ 1442 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1443 error = sbuf_finish(sb); 1444 sbuf_delete(sb); 1445 1446 return (error); 1447} 1448 1449static void 1450__elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1451 size_t *sizep) 1452{ 1453 struct proc *p; 1454 struct thread *thr; 1455 size_t size; 1456 1457 p = td->td_proc; 1458 size = 0; 1459 1460 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p); 1461 1462 /* 1463 * To have the debugger select the right thread (LWP) as the initial 1464 * thread, we dump the state of the thread passed to us in td first. 1465 * This is the thread that causes the core dump and thus likely to 1466 * be the right thread one wants to have selected in the debugger. 1467 */ 1468 thr = td; 1469 while (thr != NULL) { 1470 size += register_note(list, NT_PRSTATUS, 1471 __elfN(note_prstatus), thr); 1472 size += register_note(list, NT_FPREGSET, 1473 __elfN(note_fpregset), thr); 1474 size += register_note(list, NT_THRMISC, 1475 __elfN(note_thrmisc), thr); 1476 size += register_note(list, -1, 1477 __elfN(note_threadmd), thr); 1478 1479 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1480 TAILQ_NEXT(thr, td_plist); 1481 if (thr == td) 1482 thr = TAILQ_NEXT(thr, td_plist); 1483 } 1484 1485 size += register_note(list, NT_PROCSTAT_PROC, 1486 __elfN(note_procstat_proc), p); 1487 size += register_note(list, NT_PROCSTAT_FILES, 1488 note_procstat_files, p); 1489 size += register_note(list, NT_PROCSTAT_VMMAP, 1490 note_procstat_vmmap, p); 1491 size += register_note(list, NT_PROCSTAT_GROUPS, 1492 note_procstat_groups, p); 1493 size += register_note(list, NT_PROCSTAT_UMASK, 1494 note_procstat_umask, p); 1495 size += register_note(list, NT_PROCSTAT_RLIMIT, 1496 note_procstat_rlimit, p); 1497 size += register_note(list, NT_PROCSTAT_OSREL, 1498 note_procstat_osrel, p); 1499 size += register_note(list, NT_PROCSTAT_PSSTRINGS, 1500 __elfN(note_procstat_psstrings), p); 1501 size += register_note(list, NT_PROCSTAT_AUXV, 1502 __elfN(note_procstat_auxv), p); 1503 1504 *sizep = size; 1505} 1506 1507static void 1508__elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1509 size_t notesz) 1510{ 1511 Elf_Ehdr *ehdr; 1512 Elf_Phdr *phdr; 1513 struct phdr_closure phc; 1514 1515 ehdr = (Elf_Ehdr *)hdr; 1516 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)); 1517 1518 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1519 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1520 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1521 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1522 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1523 ehdr->e_ident[EI_DATA] = ELF_DATA; 1524 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1525 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1526 ehdr->e_ident[EI_ABIVERSION] = 0; 1527 ehdr->e_ident[EI_PAD] = 0; 1528 ehdr->e_type = ET_CORE; 1529#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1530 ehdr->e_machine = ELF_ARCH32; 1531#else 1532 ehdr->e_machine = ELF_ARCH; 1533#endif 1534 ehdr->e_version = EV_CURRENT; 1535 ehdr->e_entry = 0; 1536 ehdr->e_phoff = sizeof(Elf_Ehdr); 1537 ehdr->e_flags = 0; 1538 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1539 ehdr->e_phentsize = sizeof(Elf_Phdr); 1540 ehdr->e_phnum = numsegs + 1; 1541 ehdr->e_shentsize = sizeof(Elf_Shdr); 1542 ehdr->e_shnum = 0; 1543 ehdr->e_shstrndx = SHN_UNDEF; 1544 1545 /* 1546 * Fill in the program header entries. 1547 */ 1548 1549 /* The note segement. */ 1550 phdr->p_type = PT_NOTE; 1551 phdr->p_offset = hdrsize; 1552 phdr->p_vaddr = 0; 1553 phdr->p_paddr = 0; 1554 phdr->p_filesz = notesz; 1555 phdr->p_memsz = 0; 1556 phdr->p_flags = PF_R; 1557 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1558 phdr++; 1559 1560 /* All the writable segments from the program. */ 1561 phc.phdr = phdr; 1562 phc.offset = round_page(hdrsize + notesz); 1563 each_writable_segment(td, cb_put_phdr, &phc); 1564} 1565 1566static size_t 1567register_note(struct note_info_list *list, int type, outfunc_t out, void *arg) 1568{ 1569 struct note_info *ninfo; 1570 size_t size, notesize; 1571 1572 size = 0; 1573 out(arg, NULL, &size); 1574 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 1575 ninfo->type = type; 1576 ninfo->outfunc = out; 1577 ninfo->outarg = arg; 1578 ninfo->outsize = size; 1579 TAILQ_INSERT_TAIL(list, ninfo, link); 1580 1581 if (type == -1) 1582 return (size); 1583 1584 notesize = sizeof(Elf_Note) + /* note header */ 1585 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1586 /* note name */ 1587 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1588 1589 return (notesize); 1590} 1591 1592static size_t 1593append_note_data(const void *src, void *dst, size_t len) 1594{ 1595 size_t padded_len; 1596 1597 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 1598 if (dst != NULL) { 1599 bcopy(src, dst, len); 1600 bzero((char *)dst + len, padded_len - len); 1601 } 1602 return (padded_len); 1603} 1604 1605size_t 1606__elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 1607{ 1608 Elf_Note *note; 1609 char *buf; 1610 size_t notesize; 1611 1612 buf = dst; 1613 if (buf != NULL) { 1614 note = (Elf_Note *)buf; 1615 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1616 note->n_descsz = size; 1617 note->n_type = type; 1618 buf += sizeof(*note); 1619 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 1620 sizeof(FREEBSD_ABI_VENDOR)); 1621 append_note_data(src, buf, size); 1622 if (descp != NULL) 1623 *descp = buf; 1624 } 1625 1626 notesize = sizeof(Elf_Note) + /* note header */ 1627 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1628 /* note name */ 1629 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1630 1631 return (notesize); 1632} 1633 1634static void 1635__elfN(putnote)(struct note_info *ninfo, struct sbuf *sb) 1636{ 1637 Elf_Note note; 1638 ssize_t old_len; 1639 1640 if (ninfo->type == -1) { 1641 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1642 return; 1643 } 1644 1645 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1646 note.n_descsz = ninfo->outsize; 1647 note.n_type = ninfo->type; 1648 1649 sbuf_bcat(sb, ¬e, sizeof(note)); 1650 sbuf_start_section(sb, &old_len); 1651 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR)); 1652 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1653 if (note.n_descsz == 0) 1654 return; 1655 sbuf_start_section(sb, &old_len); 1656 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1657 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1658} 1659 1660/* 1661 * Miscellaneous note out functions. 1662 */ 1663 1664#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1665#include <compat/freebsd32/freebsd32.h> 1666 1667typedef struct prstatus32 elf_prstatus_t; 1668typedef struct prpsinfo32 elf_prpsinfo_t; 1669typedef struct fpreg32 elf_prfpregset_t; 1670typedef struct fpreg32 elf_fpregset_t; 1671typedef struct reg32 elf_gregset_t; 1672typedef struct thrmisc32 elf_thrmisc_t; 1673#define ELF_KERN_PROC_MASK KERN_PROC_MASK32 1674typedef struct kinfo_proc32 elf_kinfo_proc_t; 1675typedef uint32_t elf_ps_strings_t; 1676#else 1677typedef prstatus_t elf_prstatus_t; 1678typedef prpsinfo_t elf_prpsinfo_t; 1679typedef prfpregset_t elf_prfpregset_t; 1680typedef prfpregset_t elf_fpregset_t; 1681typedef gregset_t elf_gregset_t; 1682typedef thrmisc_t elf_thrmisc_t; 1683#define ELF_KERN_PROC_MASK 0 1684typedef struct kinfo_proc elf_kinfo_proc_t; 1685typedef vm_offset_t elf_ps_strings_t; 1686#endif 1687 1688static void 1689__elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 1690{ 1691 struct proc *p; 1692 elf_prpsinfo_t *psinfo; 1693 1694 p = (struct proc *)arg; 1695 if (sb != NULL) { 1696 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 1697 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 1698 psinfo->pr_version = PRPSINFO_VERSION; 1699 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1700 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1701 /* 1702 * XXX - We don't fill in the command line arguments properly 1703 * yet. 1704 */ 1705 strlcpy(psinfo->pr_psargs, p->p_comm, 1706 sizeof(psinfo->pr_psargs)); 1707 1708 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 1709 free(psinfo, M_TEMP); 1710 } 1711 *sizep = sizeof(*psinfo); 1712} 1713 1714static void 1715__elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep) 1716{ 1717 struct thread *td; 1718 elf_prstatus_t *status; 1719 1720 td = (struct thread *)arg; 1721 if (sb != NULL) { 1722 KASSERT(*sizep == sizeof(*status), ("invalid size")); 1723 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK); 1724 status->pr_version = PRSTATUS_VERSION; 1725 status->pr_statussz = sizeof(elf_prstatus_t); 1726 status->pr_gregsetsz = sizeof(elf_gregset_t); 1727 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1728 status->pr_osreldate = osreldate; 1729 status->pr_cursig = td->td_proc->p_sig; 1730 status->pr_pid = td->td_tid; 1731#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1732 fill_regs32(td, &status->pr_reg); 1733#else 1734 fill_regs(td, &status->pr_reg); 1735#endif 1736 sbuf_bcat(sb, status, sizeof(*status)); 1737 free(status, M_TEMP); 1738 } 1739 *sizep = sizeof(*status); 1740} 1741 1742static void 1743__elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep) 1744{ 1745 struct thread *td; 1746 elf_prfpregset_t *fpregset; 1747 1748 td = (struct thread *)arg; 1749 if (sb != NULL) { 1750 KASSERT(*sizep == sizeof(*fpregset), ("invalid size")); 1751 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK); 1752#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1753 fill_fpregs32(td, fpregset); 1754#else 1755 fill_fpregs(td, fpregset); 1756#endif 1757 sbuf_bcat(sb, fpregset, sizeof(*fpregset)); 1758 free(fpregset, M_TEMP); 1759 } 1760 *sizep = sizeof(*fpregset); 1761} 1762 1763static void 1764__elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep) 1765{ 1766 struct thread *td; 1767 elf_thrmisc_t thrmisc; 1768 1769 td = (struct thread *)arg; 1770 if (sb != NULL) { 1771 KASSERT(*sizep == sizeof(thrmisc), ("invalid size")); 1772 bzero(&thrmisc._pad, sizeof(thrmisc._pad)); 1773 strcpy(thrmisc.pr_tname, td->td_name); 1774 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc)); 1775 } 1776 *sizep = sizeof(thrmisc); 1777} 1778 1779/* 1780 * Allow for MD specific notes, as well as any MD 1781 * specific preparations for writing MI notes. 1782 */ 1783static void 1784__elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 1785{ 1786 struct thread *td; 1787 void *buf; 1788 size_t size; 1789 1790 td = (struct thread *)arg; 1791 size = *sizep; 1792 if (size != 0 && sb != NULL) 1793 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 1794 else 1795 buf = NULL; 1796 size = 0; 1797 __elfN(dump_thread)(td, buf, &size); 1798 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 1799 if (size != 0 && sb != NULL) 1800 sbuf_bcat(sb, buf, size); 1801 free(buf, M_TEMP); 1802 *sizep = size; 1803} 1804 1805#ifdef KINFO_PROC_SIZE 1806CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 1807#endif 1808 1809static void 1810__elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 1811{ 1812 struct proc *p; 1813 size_t size; 1814 int structsize; 1815 1816 p = (struct proc *)arg; 1817 size = sizeof(structsize) + p->p_numthreads * 1818 sizeof(elf_kinfo_proc_t); 1819 1820 if (sb != NULL) { 1821 KASSERT(*sizep == size, ("invalid size")); 1822 structsize = sizeof(elf_kinfo_proc_t); 1823 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1824 PROC_LOCK(p); 1825 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 1826 } 1827 *sizep = size; 1828} 1829 1830#ifdef KINFO_FILE_SIZE 1831CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 1832#endif 1833 1834static void 1835note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 1836{ 1837 struct proc *p; 1838 size_t size; 1839 int structsize; 1840 1841 p = (struct proc *)arg; 1842 if (sb == NULL) { 1843 size = 0; 1844 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 1845 sbuf_set_drain(sb, sbuf_drain_count, &size); 1846 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1847 PROC_LOCK(p); 1848 kern_proc_filedesc_out(p, sb, -1); 1849 sbuf_finish(sb); 1850 sbuf_delete(sb); 1851 *sizep = size; 1852 } else { 1853 structsize = sizeof(struct kinfo_file); 1854 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1855 PROC_LOCK(p); 1856 kern_proc_filedesc_out(p, sb, -1); 1857 } 1858} 1859 1860#ifdef KINFO_VMENTRY_SIZE 1861CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 1862#endif 1863 1864static void 1865note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 1866{ 1867 struct proc *p; 1868 size_t size; 1869 int structsize; 1870 1871 p = (struct proc *)arg; 1872 if (sb == NULL) { 1873 size = 0; 1874 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 1875 sbuf_set_drain(sb, sbuf_drain_count, &size); 1876 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1877 PROC_LOCK(p); 1878 kern_proc_vmmap_out(p, sb); 1879 sbuf_finish(sb); 1880 sbuf_delete(sb); 1881 *sizep = size; 1882 } else { 1883 structsize = sizeof(struct kinfo_vmentry); 1884 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1885 PROC_LOCK(p); 1886 kern_proc_vmmap_out(p, sb); 1887 } 1888} 1889 1890static void 1891note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 1892{ 1893 struct proc *p; 1894 size_t size; 1895 int structsize; 1896 1897 p = (struct proc *)arg; 1898 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 1899 if (sb != NULL) { 1900 KASSERT(*sizep == size, ("invalid size")); 1901 structsize = sizeof(gid_t); 1902 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1903 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 1904 sizeof(gid_t)); 1905 } 1906 *sizep = size; 1907} 1908 1909static void 1910note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 1911{ 1912 struct proc *p; 1913 size_t size; 1914 int structsize; 1915 1916 p = (struct proc *)arg; 1917 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask); 1918 if (sb != NULL) { 1919 KASSERT(*sizep == size, ("invalid size")); 1920 structsize = sizeof(p->p_fd->fd_cmask); 1921 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1922 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask)); 1923 } 1924 *sizep = size; 1925} 1926 1927static void 1928note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 1929{ 1930 struct proc *p; 1931 struct rlimit rlim[RLIM_NLIMITS]; 1932 size_t size; 1933 int structsize, i; 1934 1935 p = (struct proc *)arg; 1936 size = sizeof(structsize) + sizeof(rlim); 1937 if (sb != NULL) { 1938 KASSERT(*sizep == size, ("invalid size")); 1939 structsize = sizeof(rlim); 1940 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1941 PROC_LOCK(p); 1942 for (i = 0; i < RLIM_NLIMITS; i++) 1943 lim_rlimit(p, i, &rlim[i]); 1944 PROC_UNLOCK(p); 1945 sbuf_bcat(sb, rlim, sizeof(rlim)); 1946 } 1947 *sizep = size; 1948} 1949 1950static void 1951note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 1952{ 1953 struct proc *p; 1954 size_t size; 1955 int structsize; 1956 1957 p = (struct proc *)arg; 1958 size = sizeof(structsize) + sizeof(p->p_osrel); 1959 if (sb != NULL) { 1960 KASSERT(*sizep == size, ("invalid size")); 1961 structsize = sizeof(p->p_osrel); 1962 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1963 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 1964 } 1965 *sizep = size; 1966} 1967 1968static void 1969__elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 1970{ 1971 struct proc *p; 1972 elf_ps_strings_t ps_strings; 1973 size_t size; 1974 int structsize; 1975 1976 p = (struct proc *)arg; 1977 size = sizeof(structsize) + sizeof(ps_strings); 1978 if (sb != NULL) { 1979 KASSERT(*sizep == size, ("invalid size")); 1980 structsize = sizeof(ps_strings); 1981#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1982 ps_strings = PTROUT(p->p_sysent->sv_psstrings); 1983#else 1984 ps_strings = p->p_sysent->sv_psstrings; 1985#endif 1986 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1987 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 1988 } 1989 *sizep = size; 1990} 1991 1992static void 1993__elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 1994{ 1995 struct proc *p; 1996 size_t size; 1997 int structsize; 1998 1999 p = (struct proc *)arg; 2000 if (sb == NULL) { 2001 size = 0; 2002 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2003 sbuf_set_drain(sb, sbuf_drain_count, &size); 2004 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2005 PHOLD(p); 2006 proc_getauxv(curthread, p, sb); 2007 PRELE(p); 2008 sbuf_finish(sb); 2009 sbuf_delete(sb); 2010 *sizep = size; 2011 } else { 2012 structsize = sizeof(Elf_Auxinfo); 2013 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2014 PHOLD(p); 2015 proc_getauxv(curthread, p, sb); 2016 PRELE(p); 2017 } 2018} 2019 2020static boolean_t 2021__elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote, 2022 int32_t *osrel, const Elf_Phdr *pnote) 2023{ 2024 const Elf_Note *note, *note0, *note_end; 2025 const char *note_name; 2026 int i; 2027 2028 if (pnote == NULL || pnote->p_offset > PAGE_SIZE || 2029 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) 2030 return (FALSE); 2031 2032 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset); 2033 note_end = (const Elf_Note *)(imgp->image_header + 2034 pnote->p_offset + pnote->p_filesz); 2035 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 2036 if (!aligned(note, Elf32_Addr) || (const char *)note_end - 2037 (const char *)note < sizeof(Elf_Note)) 2038 return (FALSE); 2039 if (note->n_namesz != checknote->hdr.n_namesz || 2040 note->n_descsz != checknote->hdr.n_descsz || 2041 note->n_type != checknote->hdr.n_type) 2042 goto nextnote; 2043 note_name = (const char *)(note + 1); 2044 if (note_name + checknote->hdr.n_namesz >= 2045 (const char *)note_end || strncmp(checknote->vendor, 2046 note_name, checknote->hdr.n_namesz) != 0) 2047 goto nextnote; 2048 2049 /* 2050 * Fetch the osreldate for binary 2051 * from the ELF OSABI-note if necessary. 2052 */ 2053 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 2054 checknote->trans_osrel != NULL) 2055 return (checknote->trans_osrel(note, osrel)); 2056 return (TRUE); 2057 2058nextnote: 2059 note = (const Elf_Note *)((const char *)(note + 1) + 2060 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2061 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2062 } 2063 2064 return (FALSE); 2065} 2066 2067/* 2068 * Try to find the appropriate ABI-note section for checknote, 2069 * fetch the osreldate for binary from the ELF OSABI-note. Only the 2070 * first page of the image is searched, the same as for headers. 2071 */ 2072static boolean_t 2073__elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 2074 int32_t *osrel) 2075{ 2076 const Elf_Phdr *phdr; 2077 const Elf_Ehdr *hdr; 2078 int i; 2079 2080 hdr = (const Elf_Ehdr *)imgp->image_header; 2081 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2082 2083 for (i = 0; i < hdr->e_phnum; i++) { 2084 if (phdr[i].p_type == PT_NOTE && 2085 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i])) 2086 return (TRUE); 2087 } 2088 return (FALSE); 2089 2090} 2091 2092/* 2093 * Tell kern_execve.c about it, with a little help from the linker. 2094 */ 2095static struct execsw __elfN(execsw) = { 2096 __CONCAT(exec_, __elfN(imgact)), 2097 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2098}; 2099EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2100 2101#ifdef COMPRESS_USER_CORES 2102/* 2103 * Compress and write out a core segment for a user process. 2104 * 2105 * 'inbuf' is the starting address of a VM segment in the process' address 2106 * space that is to be compressed and written out to the core file. 'dest_buf' 2107 * is a buffer in the kernel's address space. The segment is copied from 2108 * 'inbuf' to 'dest_buf' first before being processed by the compression 2109 * routine gzwrite(). This copying is necessary because the content of the VM 2110 * segment may change between the compression pass and the crc-computation pass 2111 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel. 2112 * 2113 * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'. 2114 */ 2115static int 2116compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len, 2117 struct thread *td) 2118{ 2119 int len_compressed; 2120 int error = 0; 2121 unsigned int chunk_len; 2122 2123 while (len) { 2124 if (inbuf != NULL) { 2125 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len; 2126 copyin(inbuf, dest_buf, chunk_len); 2127 inbuf += chunk_len; 2128 } else { 2129 chunk_len = len; 2130 } 2131 len_compressed = gzwrite(file, dest_buf, chunk_len); 2132 2133 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed); 2134 2135 if ((unsigned int)len_compressed != chunk_len) { 2136 log(LOG_WARNING, 2137 "compress_core: length mismatch (0x%x returned, " 2138 "0x%x expected)\n", len_compressed, chunk_len); 2139 EVENTHANDLER_INVOKE(app_coredump_error, td, 2140 "compress_core: length mismatch %x -> %x", 2141 chunk_len, len_compressed); 2142 error = EFAULT; 2143 break; 2144 } 2145 len -= chunk_len; 2146 maybe_yield(); 2147 } 2148 2149 return (error); 2150} 2151#endif /* COMPRESS_USER_CORES */ 2152 2153static vm_prot_t 2154__elfN(trans_prot)(Elf_Word flags) 2155{ 2156 vm_prot_t prot; 2157 2158 prot = 0; 2159 if (flags & PF_X) 2160 prot |= VM_PROT_EXECUTE; 2161 if (flags & PF_W) 2162 prot |= VM_PROT_WRITE; 2163 if (flags & PF_R) 2164 prot |= VM_PROT_READ; 2165#if __ELF_WORD_SIZE == 32 2166#if defined(__amd64__) || defined(__ia64__) 2167 if (i386_read_exec && (flags & PF_R)) 2168 prot |= VM_PROT_EXECUTE; 2169#endif 2170#endif 2171 return (prot); 2172} 2173 2174static Elf_Word 2175__elfN(untrans_prot)(vm_prot_t prot) 2176{ 2177 Elf_Word flags; 2178 2179 flags = 0; 2180 if (prot & VM_PROT_EXECUTE) 2181 flags |= PF_X; 2182 if (prot & VM_PROT_READ) 2183 flags |= PF_R; 2184 if (prot & VM_PROT_WRITE) 2185 flags |= PF_W; 2186 return (flags); 2187}
|