imgact_elf.c revision 280966
1/*- 2 * Copyright (c) 2000 David O'Brien 3 * Copyright (c) 1995-1996 S��ren Schmidt 4 * Copyright (c) 1996 Peter Wemm 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer 12 * in this position and unchanged. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. The name of the author may not be used to endorse or promote products 17 * derived from this software without specific prior written permission 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 29 */ 30 31#include <sys/cdefs.h> 32__FBSDID("$FreeBSD: stable/10/sys/kern/imgact_elf.c 280966 2015-04-01 19:48:19Z jhb $"); 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 break; 780 } 781 } 782 783 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len, 784 &osrel); 785 if (brand_info == NULL) { 786 uprintf("ELF binary type \"%u\" not known.\n", 787 hdr->e_ident[EI_OSABI]); 788 return (ENOEXEC); 789 } 790 if (hdr->e_type == ET_DYN) { 791 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) 792 return (ENOEXEC); 793 /* 794 * Honour the base load address from the dso if it is 795 * non-zero for some reason. 796 */ 797 if (baddr == 0) 798 et_dyn_addr = ET_DYN_LOAD_ADDR; 799 else 800 et_dyn_addr = 0; 801 } else 802 et_dyn_addr = 0; 803 sv = brand_info->sysvec; 804 if (interp != NULL && brand_info->interp_newpath != NULL) 805 newinterp = brand_info->interp_newpath; 806 807 /* 808 * Avoid a possible deadlock if the current address space is destroyed 809 * and that address space maps the locked vnode. In the common case, 810 * the locked vnode's v_usecount is decremented but remains greater 811 * than zero. Consequently, the vnode lock is not needed by vrele(). 812 * However, in cases where the vnode lock is external, such as nullfs, 813 * v_usecount may become zero. 814 * 815 * The VV_TEXT flag prevents modifications to the executable while 816 * the vnode is unlocked. 817 */ 818 VOP_UNLOCK(imgp->vp, 0); 819 820 error = exec_new_vmspace(imgp, sv); 821 imgp->proc->p_sysent = sv; 822 823 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 824 if (error) 825 return (error); 826 827 for (i = 0; i < hdr->e_phnum; i++) { 828 switch (phdr[i].p_type) { 829 case PT_LOAD: /* Loadable segment */ 830 if (phdr[i].p_memsz == 0) 831 break; 832 prot = __elfN(trans_prot)(phdr[i].p_flags); 833 error = __elfN(load_section)(imgp, phdr[i].p_offset, 834 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 835 phdr[i].p_memsz, phdr[i].p_filesz, prot, 836 sv->sv_pagesize); 837 if (error != 0) 838 return (error); 839 840 /* 841 * If this segment contains the program headers, 842 * remember their virtual address for the AT_PHDR 843 * aux entry. Static binaries don't usually include 844 * a PT_PHDR entry. 845 */ 846 if (phdr[i].p_offset == 0 && 847 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 848 <= phdr[i].p_filesz) 849 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 850 et_dyn_addr; 851 852 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 853 seg_size = round_page(phdr[i].p_memsz + 854 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 855 856 /* 857 * Make the largest executable segment the official 858 * text segment and all others data. 859 * 860 * Note that obreak() assumes that data_addr + 861 * data_size == end of data load area, and the ELF 862 * file format expects segments to be sorted by 863 * address. If multiple data segments exist, the 864 * last one will be used. 865 */ 866 867 if (phdr[i].p_flags & PF_X && text_size < seg_size) { 868 text_size = seg_size; 869 text_addr = seg_addr; 870 } else { 871 data_size = seg_size; 872 data_addr = seg_addr; 873 } 874 total_size += seg_size; 875 break; 876 case PT_PHDR: /* Program header table info */ 877 proghdr = phdr[i].p_vaddr + et_dyn_addr; 878 break; 879 default: 880 break; 881 } 882 } 883 884 if (data_addr == 0 && data_size == 0) { 885 data_addr = text_addr; 886 data_size = text_size; 887 } 888 889 entry = (u_long)hdr->e_entry + et_dyn_addr; 890 891 /* 892 * Check limits. It should be safe to check the 893 * limits after loading the segments since we do 894 * not actually fault in all the segments pages. 895 */ 896 PROC_LOCK(imgp->proc); 897 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 898 text_size > maxtsiz || 899 total_size > lim_cur(imgp->proc, RLIMIT_VMEM) || 900 racct_set(imgp->proc, RACCT_DATA, data_size) != 0 || 901 racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) { 902 PROC_UNLOCK(imgp->proc); 903 return (ENOMEM); 904 } 905 906 vmspace = imgp->proc->p_vmspace; 907 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 908 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 909 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 910 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 911 912 /* 913 * We load the dynamic linker where a userland call 914 * to mmap(0, ...) would put it. The rationale behind this 915 * calculation is that it leaves room for the heap to grow to 916 * its maximum allowed size. 917 */ 918 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc, 919 RLIMIT_DATA)); 920 PROC_UNLOCK(imgp->proc); 921 922 imgp->entry_addr = entry; 923 924 if (interp != NULL) { 925 int have_interp = FALSE; 926 VOP_UNLOCK(imgp->vp, 0); 927 if (brand_info->emul_path != NULL && 928 brand_info->emul_path[0] != '\0') { 929 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 930 snprintf(path, MAXPATHLEN, "%s%s", 931 brand_info->emul_path, interp); 932 error = __elfN(load_file)(imgp->proc, path, &addr, 933 &imgp->entry_addr, sv->sv_pagesize); 934 free(path, M_TEMP); 935 if (error == 0) 936 have_interp = TRUE; 937 } 938 if (!have_interp && newinterp != NULL) { 939 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 940 &imgp->entry_addr, sv->sv_pagesize); 941 if (error == 0) 942 have_interp = TRUE; 943 } 944 if (!have_interp) { 945 error = __elfN(load_file)(imgp->proc, interp, &addr, 946 &imgp->entry_addr, sv->sv_pagesize); 947 } 948 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 949 if (error != 0) { 950 uprintf("ELF interpreter %s not found\n", interp); 951 return (error); 952 } 953 } else 954 addr = et_dyn_addr; 955 956 /* 957 * Construct auxargs table (used by the fixup routine) 958 */ 959 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 960 elf_auxargs->execfd = -1; 961 elf_auxargs->phdr = proghdr; 962 elf_auxargs->phent = hdr->e_phentsize; 963 elf_auxargs->phnum = hdr->e_phnum; 964 elf_auxargs->pagesz = PAGE_SIZE; 965 elf_auxargs->base = addr; 966 elf_auxargs->flags = 0; 967 elf_auxargs->entry = entry; 968 969 imgp->auxargs = elf_auxargs; 970 imgp->interpreted = 0; 971 imgp->reloc_base = addr; 972 imgp->proc->p_osrel = osrel; 973 974 return (error); 975} 976 977#define suword __CONCAT(suword, __ELF_WORD_SIZE) 978 979int 980__elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 981{ 982 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 983 Elf_Addr *base; 984 Elf_Addr *pos; 985 986 base = (Elf_Addr *)*stack_base; 987 pos = base + (imgp->args->argc + imgp->args->envc + 2); 988 989 if (args->execfd != -1) 990 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 991 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 992 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 993 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 994 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 995 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 996 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 997 AUXARGS_ENTRY(pos, AT_BASE, args->base); 998 if (imgp->execpathp != 0) 999 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 1000 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1001 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1002 if (imgp->canary != 0) { 1003 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary); 1004 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1005 } 1006 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1007 if (imgp->pagesizes != 0) { 1008 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes); 1009 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1010 } 1011 if (imgp->sysent->sv_timekeep_base != 0) { 1012 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1013 imgp->sysent->sv_timekeep_base); 1014 } 1015 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1016 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1017 imgp->sysent->sv_stackprot); 1018 AUXARGS_ENTRY(pos, AT_NULL, 0); 1019 1020 free(imgp->auxargs, M_TEMP); 1021 imgp->auxargs = NULL; 1022 1023 base--; 1024 suword(base, (long)imgp->args->argc); 1025 *stack_base = (register_t *)base; 1026 return (0); 1027} 1028 1029/* 1030 * Code for generating ELF core dumps. 1031 */ 1032 1033typedef void (*segment_callback)(vm_map_entry_t, void *); 1034 1035/* Closure for cb_put_phdr(). */ 1036struct phdr_closure { 1037 Elf_Phdr *phdr; /* Program header to fill in */ 1038 Elf_Off offset; /* Offset of segment in core file */ 1039}; 1040 1041/* Closure for cb_size_segment(). */ 1042struct sseg_closure { 1043 int count; /* Count of writable segments. */ 1044 size_t size; /* Total size of all writable segments. */ 1045}; 1046 1047typedef void (*outfunc_t)(void *, struct sbuf *, size_t *); 1048 1049struct note_info { 1050 int type; /* Note type. */ 1051 outfunc_t outfunc; /* Output function. */ 1052 void *outarg; /* Argument for the output function. */ 1053 size_t outsize; /* Output size. */ 1054 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1055}; 1056 1057TAILQ_HEAD(note_info_list, note_info); 1058 1059static void cb_put_phdr(vm_map_entry_t, void *); 1060static void cb_size_segment(vm_map_entry_t, void *); 1061static void each_writable_segment(struct thread *, segment_callback, void *); 1062static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 1063 int, void *, size_t, struct note_info_list *, size_t, gzFile); 1064static void __elfN(prepare_notes)(struct thread *, struct note_info_list *, 1065 size_t *); 1066static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t); 1067static void __elfN(putnote)(struct note_info *, struct sbuf *); 1068static size_t register_note(struct note_info_list *, int, outfunc_t, void *); 1069static int sbuf_drain_core_output(void *, const char *, int); 1070static int sbuf_drain_count(void *arg, const char *data, int len); 1071 1072static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *); 1073static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1074static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *); 1075static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1076static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *); 1077static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1078static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1079static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1080static void note_procstat_files(void *, struct sbuf *, size_t *); 1081static void note_procstat_groups(void *, struct sbuf *, size_t *); 1082static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1083static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1084static void note_procstat_umask(void *, struct sbuf *, size_t *); 1085static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1086 1087#ifdef COMPRESS_USER_CORES 1088extern int compress_user_cores; 1089extern int compress_user_cores_gzlevel; 1090#endif 1091 1092static int 1093core_output(struct vnode *vp, void *base, size_t len, off_t offset, 1094 struct ucred *active_cred, struct ucred *file_cred, 1095 struct thread *td, char *core_buf, gzFile gzfile) { 1096 1097 int error; 1098 if (gzfile) { 1099#ifdef COMPRESS_USER_CORES 1100 error = compress_core(gzfile, base, core_buf, len, td); 1101#else 1102 panic("shouldn't be here"); 1103#endif 1104 } else { 1105 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset, 1106 UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred, 1107 NULL, td); 1108 } 1109 return (error); 1110} 1111 1112/* Coredump output parameters for sbuf drain routine. */ 1113struct sbuf_drain_core_params { 1114 off_t offset; 1115 struct ucred *active_cred; 1116 struct ucred *file_cred; 1117 struct thread *td; 1118 struct vnode *vp; 1119#ifdef COMPRESS_USER_CORES 1120 gzFile gzfile; 1121#endif 1122}; 1123 1124/* 1125 * Drain into a core file. 1126 */ 1127static int 1128sbuf_drain_core_output(void *arg, const char *data, int len) 1129{ 1130 struct sbuf_drain_core_params *p; 1131 int error, locked; 1132 1133 p = (struct sbuf_drain_core_params *)arg; 1134 1135 /* 1136 * Some kern_proc out routines that print to this sbuf may 1137 * call us with the process lock held. Draining with the 1138 * non-sleepable lock held is unsafe. The lock is needed for 1139 * those routines when dumping a live process. In our case we 1140 * can safely release the lock before draining and acquire 1141 * again after. 1142 */ 1143 locked = PROC_LOCKED(p->td->td_proc); 1144 if (locked) 1145 PROC_UNLOCK(p->td->td_proc); 1146#ifdef COMPRESS_USER_CORES 1147 if (p->gzfile != Z_NULL) 1148 error = compress_core(p->gzfile, NULL, __DECONST(char *, data), 1149 len, p->td); 1150 else 1151#endif 1152 error = vn_rdwr_inchunks(UIO_WRITE, p->vp, 1153 __DECONST(void *, data), len, p->offset, UIO_SYSSPACE, 1154 IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL, 1155 p->td); 1156 if (locked) 1157 PROC_LOCK(p->td->td_proc); 1158 if (error != 0) 1159 return (-error); 1160 p->offset += len; 1161 return (len); 1162} 1163 1164/* 1165 * Drain into a counter. 1166 */ 1167static int 1168sbuf_drain_count(void *arg, const char *data __unused, int len) 1169{ 1170 size_t *sizep; 1171 1172 sizep = (size_t *)arg; 1173 *sizep += len; 1174 return (len); 1175} 1176 1177int 1178__elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1179{ 1180 struct ucred *cred = td->td_ucred; 1181 int error = 0; 1182 struct sseg_closure seginfo; 1183 struct note_info_list notelst; 1184 struct note_info *ninfo; 1185 void *hdr; 1186 size_t hdrsize, notesz, coresize; 1187 1188 gzFile gzfile = Z_NULL; 1189 char *core_buf = NULL; 1190#ifdef COMPRESS_USER_CORES 1191 char gzopen_flags[8]; 1192 char *p; 1193 int doing_compress = flags & IMGACT_CORE_COMPRESS; 1194#endif 1195 1196 hdr = NULL; 1197 TAILQ_INIT(¬elst); 1198 1199#ifdef COMPRESS_USER_CORES 1200 if (doing_compress) { 1201 p = gzopen_flags; 1202 *p++ = 'w'; 1203 if (compress_user_cores_gzlevel >= 0 && 1204 compress_user_cores_gzlevel <= 9) 1205 *p++ = '0' + compress_user_cores_gzlevel; 1206 *p = 0; 1207 gzfile = gz_open("", gzopen_flags, vp); 1208 if (gzfile == Z_NULL) { 1209 error = EFAULT; 1210 goto done; 1211 } 1212 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1213 if (!core_buf) { 1214 error = ENOMEM; 1215 goto done; 1216 } 1217 } 1218#endif 1219 1220 /* Size the program segments. */ 1221 seginfo.count = 0; 1222 seginfo.size = 0; 1223 each_writable_segment(td, cb_size_segment, &seginfo); 1224 1225 /* 1226 * Collect info about the core file header area. 1227 */ 1228 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1229 __elfN(prepare_notes)(td, ¬elst, ¬esz); 1230 coresize = round_page(hdrsize + notesz) + seginfo.size; 1231 1232#ifdef RACCT 1233 PROC_LOCK(td->td_proc); 1234 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1235 PROC_UNLOCK(td->td_proc); 1236 if (error != 0) { 1237 error = EFAULT; 1238 goto done; 1239 } 1240#endif 1241 if (coresize >= limit) { 1242 error = EFAULT; 1243 goto done; 1244 } 1245 1246 /* 1247 * Allocate memory for building the header, fill it up, 1248 * and write it out following the notes. 1249 */ 1250 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1251 if (hdr == NULL) { 1252 error = EINVAL; 1253 goto done; 1254 } 1255 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize, 1256 ¬elst, notesz, gzfile); 1257 1258 /* Write the contents of all of the writable segments. */ 1259 if (error == 0) { 1260 Elf_Phdr *php; 1261 off_t offset; 1262 int i; 1263 1264 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1265 offset = round_page(hdrsize + notesz); 1266 for (i = 0; i < seginfo.count; i++) { 1267 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr, 1268 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile); 1269 if (error != 0) 1270 break; 1271 offset += php->p_filesz; 1272 php++; 1273 } 1274 } 1275 if (error) { 1276 log(LOG_WARNING, 1277 "Failed to write core file for process %s (error %d)\n", 1278 curproc->p_comm, error); 1279 } 1280 1281done: 1282#ifdef COMPRESS_USER_CORES 1283 if (core_buf) 1284 free(core_buf, M_TEMP); 1285 if (gzfile) 1286 gzclose(gzfile); 1287#endif 1288 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1289 TAILQ_REMOVE(¬elst, ninfo, link); 1290 free(ninfo, M_TEMP); 1291 } 1292 if (hdr != NULL) 1293 free(hdr, M_TEMP); 1294 1295 return (error); 1296} 1297 1298/* 1299 * A callback for each_writable_segment() to write out the segment's 1300 * program header entry. 1301 */ 1302static void 1303cb_put_phdr(entry, closure) 1304 vm_map_entry_t entry; 1305 void *closure; 1306{ 1307 struct phdr_closure *phc = (struct phdr_closure *)closure; 1308 Elf_Phdr *phdr = phc->phdr; 1309 1310 phc->offset = round_page(phc->offset); 1311 1312 phdr->p_type = PT_LOAD; 1313 phdr->p_offset = phc->offset; 1314 phdr->p_vaddr = entry->start; 1315 phdr->p_paddr = 0; 1316 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1317 phdr->p_align = PAGE_SIZE; 1318 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1319 1320 phc->offset += phdr->p_filesz; 1321 phc->phdr++; 1322} 1323 1324/* 1325 * A callback for each_writable_segment() to gather information about 1326 * the number of segments and their total size. 1327 */ 1328static void 1329cb_size_segment(entry, closure) 1330 vm_map_entry_t entry; 1331 void *closure; 1332{ 1333 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1334 1335 ssc->count++; 1336 ssc->size += entry->end - entry->start; 1337} 1338 1339/* 1340 * For each writable segment in the process's memory map, call the given 1341 * function with a pointer to the map entry and some arbitrary 1342 * caller-supplied data. 1343 */ 1344static void 1345each_writable_segment(td, func, closure) 1346 struct thread *td; 1347 segment_callback func; 1348 void *closure; 1349{ 1350 struct proc *p = td->td_proc; 1351 vm_map_t map = &p->p_vmspace->vm_map; 1352 vm_map_entry_t entry; 1353 vm_object_t backing_object, object; 1354 boolean_t ignore_entry; 1355 1356 vm_map_lock_read(map); 1357 for (entry = map->header.next; entry != &map->header; 1358 entry = entry->next) { 1359 /* 1360 * Don't dump inaccessible mappings, deal with legacy 1361 * coredump mode. 1362 * 1363 * Note that read-only segments related to the elf binary 1364 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1365 * need to arbitrarily ignore such segments. 1366 */ 1367 if (elf_legacy_coredump) { 1368 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1369 continue; 1370 } else { 1371 if ((entry->protection & VM_PROT_ALL) == 0) 1372 continue; 1373 } 1374 1375 /* 1376 * Dont include memory segment in the coredump if 1377 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1378 * madvise(2). Do not dump submaps (i.e. parts of the 1379 * kernel map). 1380 */ 1381 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1382 continue; 1383 1384 if ((object = entry->object.vm_object) == NULL) 1385 continue; 1386 1387 /* Ignore memory-mapped devices and such things. */ 1388 VM_OBJECT_RLOCK(object); 1389 while ((backing_object = object->backing_object) != NULL) { 1390 VM_OBJECT_RLOCK(backing_object); 1391 VM_OBJECT_RUNLOCK(object); 1392 object = backing_object; 1393 } 1394 ignore_entry = object->type != OBJT_DEFAULT && 1395 object->type != OBJT_SWAP && object->type != OBJT_VNODE && 1396 object->type != OBJT_PHYS; 1397 VM_OBJECT_RUNLOCK(object); 1398 if (ignore_entry) 1399 continue; 1400 1401 (*func)(entry, closure); 1402 } 1403 vm_map_unlock_read(map); 1404} 1405 1406/* 1407 * Write the core file header to the file, including padding up to 1408 * the page boundary. 1409 */ 1410static int 1411__elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred, 1412 int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst, 1413 size_t notesz, gzFile gzfile) 1414{ 1415 struct sbuf_drain_core_params params; 1416 struct note_info *ninfo; 1417 struct sbuf *sb; 1418 int error; 1419 1420 /* Fill in the header. */ 1421 bzero(hdr, hdrsize); 1422 __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz); 1423 1424 params.offset = 0; 1425 params.active_cred = cred; 1426 params.file_cred = NOCRED; 1427 params.td = td; 1428 params.vp = vp; 1429#ifdef COMPRESS_USER_CORES 1430 params.gzfile = gzfile; 1431#endif 1432 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1433 sbuf_set_drain(sb, sbuf_drain_core_output, ¶ms); 1434 sbuf_start_section(sb, NULL); 1435 sbuf_bcat(sb, hdr, hdrsize); 1436 TAILQ_FOREACH(ninfo, notelst, link) 1437 __elfN(putnote)(ninfo, sb); 1438 /* Align up to a page boundary for the program segments. */ 1439 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1440 error = sbuf_finish(sb); 1441 sbuf_delete(sb); 1442 1443 return (error); 1444} 1445 1446static void 1447__elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1448 size_t *sizep) 1449{ 1450 struct proc *p; 1451 struct thread *thr; 1452 size_t size; 1453 1454 p = td->td_proc; 1455 size = 0; 1456 1457 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p); 1458 1459 /* 1460 * To have the debugger select the right thread (LWP) as the initial 1461 * thread, we dump the state of the thread passed to us in td first. 1462 * This is the thread that causes the core dump and thus likely to 1463 * be the right thread one wants to have selected in the debugger. 1464 */ 1465 thr = td; 1466 while (thr != NULL) { 1467 size += register_note(list, NT_PRSTATUS, 1468 __elfN(note_prstatus), thr); 1469 size += register_note(list, NT_FPREGSET, 1470 __elfN(note_fpregset), thr); 1471 size += register_note(list, NT_THRMISC, 1472 __elfN(note_thrmisc), thr); 1473 size += register_note(list, -1, 1474 __elfN(note_threadmd), thr); 1475 1476 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1477 TAILQ_NEXT(thr, td_plist); 1478 if (thr == td) 1479 thr = TAILQ_NEXT(thr, td_plist); 1480 } 1481 1482 size += register_note(list, NT_PROCSTAT_PROC, 1483 __elfN(note_procstat_proc), p); 1484 size += register_note(list, NT_PROCSTAT_FILES, 1485 note_procstat_files, p); 1486 size += register_note(list, NT_PROCSTAT_VMMAP, 1487 note_procstat_vmmap, p); 1488 size += register_note(list, NT_PROCSTAT_GROUPS, 1489 note_procstat_groups, p); 1490 size += register_note(list, NT_PROCSTAT_UMASK, 1491 note_procstat_umask, p); 1492 size += register_note(list, NT_PROCSTAT_RLIMIT, 1493 note_procstat_rlimit, p); 1494 size += register_note(list, NT_PROCSTAT_OSREL, 1495 note_procstat_osrel, p); 1496 size += register_note(list, NT_PROCSTAT_PSSTRINGS, 1497 __elfN(note_procstat_psstrings), p); 1498 size += register_note(list, NT_PROCSTAT_AUXV, 1499 __elfN(note_procstat_auxv), p); 1500 1501 *sizep = size; 1502} 1503 1504static void 1505__elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1506 size_t notesz) 1507{ 1508 Elf_Ehdr *ehdr; 1509 Elf_Phdr *phdr; 1510 struct phdr_closure phc; 1511 1512 ehdr = (Elf_Ehdr *)hdr; 1513 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)); 1514 1515 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1516 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1517 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1518 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1519 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1520 ehdr->e_ident[EI_DATA] = ELF_DATA; 1521 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1522 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1523 ehdr->e_ident[EI_ABIVERSION] = 0; 1524 ehdr->e_ident[EI_PAD] = 0; 1525 ehdr->e_type = ET_CORE; 1526#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1527 ehdr->e_machine = ELF_ARCH32; 1528#else 1529 ehdr->e_machine = ELF_ARCH; 1530#endif 1531 ehdr->e_version = EV_CURRENT; 1532 ehdr->e_entry = 0; 1533 ehdr->e_phoff = sizeof(Elf_Ehdr); 1534 ehdr->e_flags = 0; 1535 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1536 ehdr->e_phentsize = sizeof(Elf_Phdr); 1537 ehdr->e_phnum = numsegs + 1; 1538 ehdr->e_shentsize = sizeof(Elf_Shdr); 1539 ehdr->e_shnum = 0; 1540 ehdr->e_shstrndx = SHN_UNDEF; 1541 1542 /* 1543 * Fill in the program header entries. 1544 */ 1545 1546 /* The note segement. */ 1547 phdr->p_type = PT_NOTE; 1548 phdr->p_offset = hdrsize; 1549 phdr->p_vaddr = 0; 1550 phdr->p_paddr = 0; 1551 phdr->p_filesz = notesz; 1552 phdr->p_memsz = 0; 1553 phdr->p_flags = PF_R; 1554 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1555 phdr++; 1556 1557 /* All the writable segments from the program. */ 1558 phc.phdr = phdr; 1559 phc.offset = round_page(hdrsize + notesz); 1560 each_writable_segment(td, cb_put_phdr, &phc); 1561} 1562 1563static size_t 1564register_note(struct note_info_list *list, int type, outfunc_t out, void *arg) 1565{ 1566 struct note_info *ninfo; 1567 size_t size, notesize; 1568 1569 size = 0; 1570 out(arg, NULL, &size); 1571 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 1572 ninfo->type = type; 1573 ninfo->outfunc = out; 1574 ninfo->outarg = arg; 1575 ninfo->outsize = size; 1576 TAILQ_INSERT_TAIL(list, ninfo, link); 1577 1578 if (type == -1) 1579 return (size); 1580 1581 notesize = sizeof(Elf_Note) + /* note header */ 1582 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1583 /* note name */ 1584 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1585 1586 return (notesize); 1587} 1588 1589static size_t 1590append_note_data(const void *src, void *dst, size_t len) 1591{ 1592 size_t padded_len; 1593 1594 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 1595 if (dst != NULL) { 1596 bcopy(src, dst, len); 1597 bzero((char *)dst + len, padded_len - len); 1598 } 1599 return (padded_len); 1600} 1601 1602size_t 1603__elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 1604{ 1605 Elf_Note *note; 1606 char *buf; 1607 size_t notesize; 1608 1609 buf = dst; 1610 if (buf != NULL) { 1611 note = (Elf_Note *)buf; 1612 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1613 note->n_descsz = size; 1614 note->n_type = type; 1615 buf += sizeof(*note); 1616 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 1617 sizeof(FREEBSD_ABI_VENDOR)); 1618 append_note_data(src, buf, size); 1619 if (descp != NULL) 1620 *descp = buf; 1621 } 1622 1623 notesize = sizeof(Elf_Note) + /* note header */ 1624 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1625 /* note name */ 1626 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1627 1628 return (notesize); 1629} 1630 1631static void 1632__elfN(putnote)(struct note_info *ninfo, struct sbuf *sb) 1633{ 1634 Elf_Note note; 1635 ssize_t old_len; 1636 1637 if (ninfo->type == -1) { 1638 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1639 return; 1640 } 1641 1642 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1643 note.n_descsz = ninfo->outsize; 1644 note.n_type = ninfo->type; 1645 1646 sbuf_bcat(sb, ¬e, sizeof(note)); 1647 sbuf_start_section(sb, &old_len); 1648 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR)); 1649 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1650 if (note.n_descsz == 0) 1651 return; 1652 sbuf_start_section(sb, &old_len); 1653 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1654 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1655} 1656 1657/* 1658 * Miscellaneous note out functions. 1659 */ 1660 1661#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1662#include <compat/freebsd32/freebsd32.h> 1663 1664typedef struct prstatus32 elf_prstatus_t; 1665typedef struct prpsinfo32 elf_prpsinfo_t; 1666typedef struct fpreg32 elf_prfpregset_t; 1667typedef struct fpreg32 elf_fpregset_t; 1668typedef struct reg32 elf_gregset_t; 1669typedef struct thrmisc32 elf_thrmisc_t; 1670#define ELF_KERN_PROC_MASK KERN_PROC_MASK32 1671typedef struct kinfo_proc32 elf_kinfo_proc_t; 1672typedef uint32_t elf_ps_strings_t; 1673#else 1674typedef prstatus_t elf_prstatus_t; 1675typedef prpsinfo_t elf_prpsinfo_t; 1676typedef prfpregset_t elf_prfpregset_t; 1677typedef prfpregset_t elf_fpregset_t; 1678typedef gregset_t elf_gregset_t; 1679typedef thrmisc_t elf_thrmisc_t; 1680#define ELF_KERN_PROC_MASK 0 1681typedef struct kinfo_proc elf_kinfo_proc_t; 1682typedef vm_offset_t elf_ps_strings_t; 1683#endif 1684 1685static void 1686__elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 1687{ 1688 struct proc *p; 1689 elf_prpsinfo_t *psinfo; 1690 1691 p = (struct proc *)arg; 1692 if (sb != NULL) { 1693 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 1694 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 1695 psinfo->pr_version = PRPSINFO_VERSION; 1696 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1697 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1698 /* 1699 * XXX - We don't fill in the command line arguments properly 1700 * yet. 1701 */ 1702 strlcpy(psinfo->pr_psargs, p->p_comm, 1703 sizeof(psinfo->pr_psargs)); 1704 1705 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 1706 free(psinfo, M_TEMP); 1707 } 1708 *sizep = sizeof(*psinfo); 1709} 1710 1711static void 1712__elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep) 1713{ 1714 struct thread *td; 1715 elf_prstatus_t *status; 1716 1717 td = (struct thread *)arg; 1718 if (sb != NULL) { 1719 KASSERT(*sizep == sizeof(*status), ("invalid size")); 1720 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK); 1721 status->pr_version = PRSTATUS_VERSION; 1722 status->pr_statussz = sizeof(elf_prstatus_t); 1723 status->pr_gregsetsz = sizeof(elf_gregset_t); 1724 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1725 status->pr_osreldate = osreldate; 1726 status->pr_cursig = td->td_proc->p_sig; 1727 status->pr_pid = td->td_tid; 1728#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1729 fill_regs32(td, &status->pr_reg); 1730#else 1731 fill_regs(td, &status->pr_reg); 1732#endif 1733 sbuf_bcat(sb, status, sizeof(*status)); 1734 free(status, M_TEMP); 1735 } 1736 *sizep = sizeof(*status); 1737} 1738 1739static void 1740__elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep) 1741{ 1742 struct thread *td; 1743 elf_prfpregset_t *fpregset; 1744 1745 td = (struct thread *)arg; 1746 if (sb != NULL) { 1747 KASSERT(*sizep == sizeof(*fpregset), ("invalid size")); 1748 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK); 1749#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1750 fill_fpregs32(td, fpregset); 1751#else 1752 fill_fpregs(td, fpregset); 1753#endif 1754 sbuf_bcat(sb, fpregset, sizeof(*fpregset)); 1755 free(fpregset, M_TEMP); 1756 } 1757 *sizep = sizeof(*fpregset); 1758} 1759 1760static void 1761__elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep) 1762{ 1763 struct thread *td; 1764 elf_thrmisc_t thrmisc; 1765 1766 td = (struct thread *)arg; 1767 if (sb != NULL) { 1768 KASSERT(*sizep == sizeof(thrmisc), ("invalid size")); 1769 bzero(&thrmisc._pad, sizeof(thrmisc._pad)); 1770 strcpy(thrmisc.pr_tname, td->td_name); 1771 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc)); 1772 } 1773 *sizep = sizeof(thrmisc); 1774} 1775 1776/* 1777 * Allow for MD specific notes, as well as any MD 1778 * specific preparations for writing MI notes. 1779 */ 1780static void 1781__elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 1782{ 1783 struct thread *td; 1784 void *buf; 1785 size_t size; 1786 1787 td = (struct thread *)arg; 1788 size = *sizep; 1789 if (size != 0 && sb != NULL) 1790 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 1791 else 1792 buf = NULL; 1793 size = 0; 1794 __elfN(dump_thread)(td, buf, &size); 1795 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 1796 if (size != 0 && sb != NULL) 1797 sbuf_bcat(sb, buf, size); 1798 free(buf, M_TEMP); 1799 *sizep = size; 1800} 1801 1802#ifdef KINFO_PROC_SIZE 1803CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 1804#endif 1805 1806static void 1807__elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 1808{ 1809 struct proc *p; 1810 size_t size; 1811 int structsize; 1812 1813 p = (struct proc *)arg; 1814 size = sizeof(structsize) + p->p_numthreads * 1815 sizeof(elf_kinfo_proc_t); 1816 1817 if (sb != NULL) { 1818 KASSERT(*sizep == size, ("invalid size")); 1819 structsize = sizeof(elf_kinfo_proc_t); 1820 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1821 PROC_LOCK(p); 1822 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 1823 } 1824 *sizep = size; 1825} 1826 1827#ifdef KINFO_FILE_SIZE 1828CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 1829#endif 1830 1831static void 1832note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 1833{ 1834 struct proc *p; 1835 size_t size; 1836 int structsize; 1837 1838 p = (struct proc *)arg; 1839 if (sb == NULL) { 1840 size = 0; 1841 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 1842 sbuf_set_drain(sb, sbuf_drain_count, &size); 1843 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1844 PROC_LOCK(p); 1845 kern_proc_filedesc_out(p, sb, -1); 1846 sbuf_finish(sb); 1847 sbuf_delete(sb); 1848 *sizep = size; 1849 } else { 1850 structsize = sizeof(struct kinfo_file); 1851 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1852 PROC_LOCK(p); 1853 kern_proc_filedesc_out(p, sb, -1); 1854 } 1855} 1856 1857#ifdef KINFO_VMENTRY_SIZE 1858CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 1859#endif 1860 1861static void 1862note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 1863{ 1864 struct proc *p; 1865 size_t size; 1866 int structsize; 1867 1868 p = (struct proc *)arg; 1869 if (sb == NULL) { 1870 size = 0; 1871 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 1872 sbuf_set_drain(sb, sbuf_drain_count, &size); 1873 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1874 PROC_LOCK(p); 1875 kern_proc_vmmap_out(p, sb); 1876 sbuf_finish(sb); 1877 sbuf_delete(sb); 1878 *sizep = size; 1879 } else { 1880 structsize = sizeof(struct kinfo_vmentry); 1881 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1882 PROC_LOCK(p); 1883 kern_proc_vmmap_out(p, sb); 1884 } 1885} 1886 1887static void 1888note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 1889{ 1890 struct proc *p; 1891 size_t size; 1892 int structsize; 1893 1894 p = (struct proc *)arg; 1895 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 1896 if (sb != NULL) { 1897 KASSERT(*sizep == size, ("invalid size")); 1898 structsize = sizeof(gid_t); 1899 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1900 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 1901 sizeof(gid_t)); 1902 } 1903 *sizep = size; 1904} 1905 1906static void 1907note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 1908{ 1909 struct proc *p; 1910 size_t size; 1911 int structsize; 1912 1913 p = (struct proc *)arg; 1914 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask); 1915 if (sb != NULL) { 1916 KASSERT(*sizep == size, ("invalid size")); 1917 structsize = sizeof(p->p_fd->fd_cmask); 1918 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1919 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask)); 1920 } 1921 *sizep = size; 1922} 1923 1924static void 1925note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 1926{ 1927 struct proc *p; 1928 struct rlimit rlim[RLIM_NLIMITS]; 1929 size_t size; 1930 int structsize, i; 1931 1932 p = (struct proc *)arg; 1933 size = sizeof(structsize) + sizeof(rlim); 1934 if (sb != NULL) { 1935 KASSERT(*sizep == size, ("invalid size")); 1936 structsize = sizeof(rlim); 1937 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1938 PROC_LOCK(p); 1939 for (i = 0; i < RLIM_NLIMITS; i++) 1940 lim_rlimit(p, i, &rlim[i]); 1941 PROC_UNLOCK(p); 1942 sbuf_bcat(sb, rlim, sizeof(rlim)); 1943 } 1944 *sizep = size; 1945} 1946 1947static void 1948note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 1949{ 1950 struct proc *p; 1951 size_t size; 1952 int structsize; 1953 1954 p = (struct proc *)arg; 1955 size = sizeof(structsize) + sizeof(p->p_osrel); 1956 if (sb != NULL) { 1957 KASSERT(*sizep == size, ("invalid size")); 1958 structsize = sizeof(p->p_osrel); 1959 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1960 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 1961 } 1962 *sizep = size; 1963} 1964 1965static void 1966__elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 1967{ 1968 struct proc *p; 1969 elf_ps_strings_t ps_strings; 1970 size_t size; 1971 int structsize; 1972 1973 p = (struct proc *)arg; 1974 size = sizeof(structsize) + sizeof(ps_strings); 1975 if (sb != NULL) { 1976 KASSERT(*sizep == size, ("invalid size")); 1977 structsize = sizeof(ps_strings); 1978#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1979 ps_strings = PTROUT(p->p_sysent->sv_psstrings); 1980#else 1981 ps_strings = p->p_sysent->sv_psstrings; 1982#endif 1983 sbuf_bcat(sb, &structsize, sizeof(structsize)); 1984 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 1985 } 1986 *sizep = size; 1987} 1988 1989static void 1990__elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 1991{ 1992 struct proc *p; 1993 size_t size; 1994 int structsize; 1995 1996 p = (struct proc *)arg; 1997 if (sb == NULL) { 1998 size = 0; 1999 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2000 sbuf_set_drain(sb, sbuf_drain_count, &size); 2001 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2002 PHOLD(p); 2003 proc_getauxv(curthread, p, sb); 2004 PRELE(p); 2005 sbuf_finish(sb); 2006 sbuf_delete(sb); 2007 *sizep = size; 2008 } else { 2009 structsize = sizeof(Elf_Auxinfo); 2010 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2011 PHOLD(p); 2012 proc_getauxv(curthread, p, sb); 2013 PRELE(p); 2014 } 2015} 2016 2017static boolean_t 2018__elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote, 2019 int32_t *osrel, const Elf_Phdr *pnote) 2020{ 2021 const Elf_Note *note, *note0, *note_end; 2022 const char *note_name; 2023 int i; 2024 2025 if (pnote == NULL || pnote->p_offset > PAGE_SIZE || 2026 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) 2027 return (FALSE); 2028 2029 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset); 2030 note_end = (const Elf_Note *)(imgp->image_header + 2031 pnote->p_offset + pnote->p_filesz); 2032 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 2033 if (!aligned(note, Elf32_Addr) || (const char *)note_end - 2034 (const char *)note < sizeof(Elf_Note)) 2035 return (FALSE); 2036 if (note->n_namesz != checknote->hdr.n_namesz || 2037 note->n_descsz != checknote->hdr.n_descsz || 2038 note->n_type != checknote->hdr.n_type) 2039 goto nextnote; 2040 note_name = (const char *)(note + 1); 2041 if (note_name + checknote->hdr.n_namesz >= 2042 (const char *)note_end || strncmp(checknote->vendor, 2043 note_name, checknote->hdr.n_namesz) != 0) 2044 goto nextnote; 2045 2046 /* 2047 * Fetch the osreldate for binary 2048 * from the ELF OSABI-note if necessary. 2049 */ 2050 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 2051 checknote->trans_osrel != NULL) 2052 return (checknote->trans_osrel(note, osrel)); 2053 return (TRUE); 2054 2055nextnote: 2056 note = (const Elf_Note *)((const char *)(note + 1) + 2057 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2058 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2059 } 2060 2061 return (FALSE); 2062} 2063 2064/* 2065 * Try to find the appropriate ABI-note section for checknote, 2066 * fetch the osreldate for binary from the ELF OSABI-note. Only the 2067 * first page of the image is searched, the same as for headers. 2068 */ 2069static boolean_t 2070__elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 2071 int32_t *osrel) 2072{ 2073 const Elf_Phdr *phdr; 2074 const Elf_Ehdr *hdr; 2075 int i; 2076 2077 hdr = (const Elf_Ehdr *)imgp->image_header; 2078 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2079 2080 for (i = 0; i < hdr->e_phnum; i++) { 2081 if (phdr[i].p_type == PT_NOTE && 2082 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i])) 2083 return (TRUE); 2084 } 2085 return (FALSE); 2086 2087} 2088 2089/* 2090 * Tell kern_execve.c about it, with a little help from the linker. 2091 */ 2092static struct execsw __elfN(execsw) = { 2093 __CONCAT(exec_, __elfN(imgact)), 2094 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2095}; 2096EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2097 2098#ifdef COMPRESS_USER_CORES 2099/* 2100 * Compress and write out a core segment for a user process. 2101 * 2102 * 'inbuf' is the starting address of a VM segment in the process' address 2103 * space that is to be compressed and written out to the core file. 'dest_buf' 2104 * is a buffer in the kernel's address space. The segment is copied from 2105 * 'inbuf' to 'dest_buf' first before being processed by the compression 2106 * routine gzwrite(). This copying is necessary because the content of the VM 2107 * segment may change between the compression pass and the crc-computation pass 2108 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel. 2109 * 2110 * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'. 2111 */ 2112static int 2113compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len, 2114 struct thread *td) 2115{ 2116 int len_compressed; 2117 int error = 0; 2118 unsigned int chunk_len; 2119 2120 while (len) { 2121 if (inbuf != NULL) { 2122 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len; 2123 copyin(inbuf, dest_buf, chunk_len); 2124 inbuf += chunk_len; 2125 } else { 2126 chunk_len = len; 2127 } 2128 len_compressed = gzwrite(file, dest_buf, chunk_len); 2129 2130 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed); 2131 2132 if ((unsigned int)len_compressed != chunk_len) { 2133 log(LOG_WARNING, 2134 "compress_core: length mismatch (0x%x returned, " 2135 "0x%x expected)\n", len_compressed, chunk_len); 2136 EVENTHANDLER_INVOKE(app_coredump_error, td, 2137 "compress_core: length mismatch %x -> %x", 2138 chunk_len, len_compressed); 2139 error = EFAULT; 2140 break; 2141 } 2142 len -= chunk_len; 2143 maybe_yield(); 2144 } 2145 2146 return (error); 2147} 2148#endif /* COMPRESS_USER_CORES */ 2149 2150static vm_prot_t 2151__elfN(trans_prot)(Elf_Word flags) 2152{ 2153 vm_prot_t prot; 2154 2155 prot = 0; 2156 if (flags & PF_X) 2157 prot |= VM_PROT_EXECUTE; 2158 if (flags & PF_W) 2159 prot |= VM_PROT_WRITE; 2160 if (flags & PF_R) 2161 prot |= VM_PROT_READ; 2162#if __ELF_WORD_SIZE == 32 2163#if defined(__amd64__) || defined(__ia64__) 2164 if (i386_read_exec && (flags & PF_R)) 2165 prot |= VM_PROT_EXECUTE; 2166#endif 2167#endif 2168 return (prot); 2169} 2170 2171static Elf_Word 2172__elfN(untrans_prot)(vm_prot_t prot) 2173{ 2174 Elf_Word flags; 2175 2176 flags = 0; 2177 if (prot & VM_PROT_EXECUTE) 2178 flags |= PF_X; 2179 if (prot & VM_PROT_READ) 2180 flags |= PF_R; 2181 if (prot & VM_PROT_WRITE) 2182 flags |= PF_W; 2183 return (flags); 2184} 2185