imgact_elf.c revision 204552
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: head/sys/kern/imgact_elf.c 204552 2010-03-02 06:58:58Z alfred $"); 33 34#include "opt_compat.h" 35#include "opt_core.h" 36 37#include <sys/param.h> 38#include <sys/exec.h> 39#include <sys/fcntl.h> 40#include <sys/imgact.h> 41#include <sys/imgact_elf.h> 42#include <sys/kernel.h> 43#include <sys/lock.h> 44#include <sys/malloc.h> 45#include <sys/mount.h> 46#include <sys/mutex.h> 47#include <sys/mman.h> 48#include <sys/namei.h> 49#include <sys/pioctl.h> 50#include <sys/proc.h> 51#include <sys/procfs.h> 52#include <sys/resourcevar.h> 53#include <sys/sf_buf.h> 54#include <sys/systm.h> 55#include <sys/signalvar.h> 56#include <sys/stat.h> 57#include <sys/sx.h> 58#include <sys/syscall.h> 59#include <sys/sysctl.h> 60#include <sys/sysent.h> 61#include <sys/vnode.h> 62#include <sys/syslog.h> 63#include <sys/eventhandler.h> 64 65#include <net/zlib.h> 66 67#include <vm/vm.h> 68#include <vm/vm_kern.h> 69#include <vm/vm_param.h> 70#include <vm/pmap.h> 71#include <vm/vm_map.h> 72#include <vm/vm_object.h> 73#include <vm/vm_extern.h> 74 75#include <machine/elf.h> 76#include <machine/md_var.h> 77 78#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 79#include <machine/fpu.h> 80#include <compat/ia32/ia32_reg.h> 81#endif 82 83#define OLD_EI_BRAND 8 84 85static int __elfN(check_header)(const Elf_Ehdr *hdr); 86static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 87 const char *interp, int32_t *osrel); 88static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 89 u_long *entry, size_t pagesize); 90static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object, 91 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 92 vm_prot_t prot, size_t pagesize); 93static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 94static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note, 95 int32_t *osrel); 96static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 97static boolean_t __elfN(check_note)(struct image_params *imgp, 98 Elf_Brandnote *checknote, int32_t *osrel); 99 100SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 101 ""); 102 103#ifdef COMPRESS_USER_CORES 104static int compress_core(gzFile, char *, char *, unsigned int, 105 struct thread * td); 106#define CORE_BUF_SIZE (16 * 1024) 107#endif 108 109int __elfN(fallback_brand) = -1; 110SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 111 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 112 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 113TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 114 &__elfN(fallback_brand)); 115 116static int elf_legacy_coredump = 0; 117SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 118 &elf_legacy_coredump, 0, ""); 119 120static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 121 122#define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 123#define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 124#define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a)) 125 126static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; 127 128Elf_Brandnote __elfN(freebsd_brandnote) = { 129 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 130 .hdr.n_descsz = sizeof(int32_t), 131 .hdr.n_type = 1, 132 .vendor = FREEBSD_ABI_VENDOR, 133 .flags = BN_TRANSLATE_OSREL, 134 .trans_osrel = __elfN(freebsd_trans_osrel) 135}; 136 137static boolean_t 138__elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 139{ 140 uintptr_t p; 141 142 p = (uintptr_t)(note + 1); 143 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 144 *osrel = *(const int32_t *)(p); 145 146 return (TRUE); 147} 148 149static const char GNU_ABI_VENDOR[] = "GNU"; 150static int GNU_KFREEBSD_ABI_DESC = 3; 151 152Elf_Brandnote __elfN(kfreebsd_brandnote) = { 153 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 154 .hdr.n_descsz = 16, /* XXX at least 16 */ 155 .hdr.n_type = 1, 156 .vendor = GNU_ABI_VENDOR, 157 .flags = BN_TRANSLATE_OSREL, 158 .trans_osrel = kfreebsd_trans_osrel 159}; 160 161static boolean_t 162kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 163{ 164 const Elf32_Word *desc; 165 uintptr_t p; 166 167 p = (uintptr_t)(note + 1); 168 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 169 170 desc = (const Elf32_Word *)p; 171 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 172 return (FALSE); 173 174 /* 175 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 176 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 177 */ 178 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 179 180 return (TRUE); 181} 182 183int 184__elfN(insert_brand_entry)(Elf_Brandinfo *entry) 185{ 186 int i; 187 188 for (i = 0; i < MAX_BRANDS; i++) { 189 if (elf_brand_list[i] == NULL) { 190 elf_brand_list[i] = entry; 191 break; 192 } 193 } 194 if (i == MAX_BRANDS) { 195 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 196 __func__, entry); 197 return (-1); 198 } 199 return (0); 200} 201 202int 203__elfN(remove_brand_entry)(Elf_Brandinfo *entry) 204{ 205 int i; 206 207 for (i = 0; i < MAX_BRANDS; i++) { 208 if (elf_brand_list[i] == entry) { 209 elf_brand_list[i] = NULL; 210 break; 211 } 212 } 213 if (i == MAX_BRANDS) 214 return (-1); 215 return (0); 216} 217 218int 219__elfN(brand_inuse)(Elf_Brandinfo *entry) 220{ 221 struct proc *p; 222 int rval = FALSE; 223 224 sx_slock(&allproc_lock); 225 FOREACH_PROC_IN_SYSTEM(p) { 226 if (p->p_sysent == entry->sysvec) { 227 rval = TRUE; 228 break; 229 } 230 } 231 sx_sunlock(&allproc_lock); 232 233 return (rval); 234} 235 236static Elf_Brandinfo * 237__elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 238 int32_t *osrel) 239{ 240 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 241 Elf_Brandinfo *bi; 242 boolean_t ret; 243 int i; 244 245 /* 246 * We support four types of branding -- (1) the ELF EI_OSABI field 247 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 248 * branding w/in the ELF header, (3) path of the `interp_path' 249 * field, and (4) the ".note.ABI-tag" ELF section. 250 */ 251 252 /* Look for an ".note.ABI-tag" ELF section */ 253 for (i = 0; i < MAX_BRANDS; i++) { 254 bi = elf_brand_list[i]; 255 if (bi == NULL) 256 continue; 257 if (hdr->e_machine == bi->machine && (bi->flags & 258 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 259 ret = __elfN(check_note)(imgp, bi->brand_note, osrel); 260 if (ret) 261 return (bi); 262 } 263 } 264 265 /* If the executable has a brand, search for it in the brand list. */ 266 for (i = 0; i < MAX_BRANDS; i++) { 267 bi = elf_brand_list[i]; 268 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 269 continue; 270 if (hdr->e_machine == bi->machine && 271 (hdr->e_ident[EI_OSABI] == bi->brand || 272 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 273 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 274 return (bi); 275 } 276 277 /* Lacking a known brand, search for a recognized interpreter. */ 278 if (interp != NULL) { 279 for (i = 0; i < MAX_BRANDS; i++) { 280 bi = elf_brand_list[i]; 281 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 282 continue; 283 if (hdr->e_machine == bi->machine && 284 strcmp(interp, bi->interp_path) == 0) 285 return (bi); 286 } 287 } 288 289 /* Lacking a recognized interpreter, try the default brand */ 290 for (i = 0; i < MAX_BRANDS; i++) { 291 bi = elf_brand_list[i]; 292 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 293 continue; 294 if (hdr->e_machine == bi->machine && 295 __elfN(fallback_brand) == bi->brand) 296 return (bi); 297 } 298 return (NULL); 299} 300 301static int 302__elfN(check_header)(const Elf_Ehdr *hdr) 303{ 304 Elf_Brandinfo *bi; 305 int i; 306 307 if (!IS_ELF(*hdr) || 308 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 309 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 310 hdr->e_ident[EI_VERSION] != EV_CURRENT || 311 hdr->e_phentsize != sizeof(Elf_Phdr) || 312 hdr->e_version != ELF_TARG_VER) 313 return (ENOEXEC); 314 315 /* 316 * Make sure we have at least one brand for this machine. 317 */ 318 319 for (i = 0; i < MAX_BRANDS; i++) { 320 bi = elf_brand_list[i]; 321 if (bi != NULL && bi->machine == hdr->e_machine) 322 break; 323 } 324 if (i == MAX_BRANDS) 325 return (ENOEXEC); 326 327 return (0); 328} 329 330static int 331__elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 332 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 333{ 334 struct sf_buf *sf; 335 int error; 336 vm_offset_t off; 337 338 /* 339 * Create the page if it doesn't exist yet. Ignore errors. 340 */ 341 vm_map_lock(map); 342 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), 343 VM_PROT_ALL, VM_PROT_ALL, 0); 344 vm_map_unlock(map); 345 346 /* 347 * Find the page from the underlying object. 348 */ 349 if (object) { 350 sf = vm_imgact_map_page(object, offset); 351 if (sf == NULL) 352 return (KERN_FAILURE); 353 off = offset - trunc_page(offset); 354 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 355 end - start); 356 vm_imgact_unmap_page(sf); 357 if (error) { 358 return (KERN_FAILURE); 359 } 360 } 361 362 return (KERN_SUCCESS); 363} 364 365static int 366__elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 367 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) 368{ 369 struct sf_buf *sf; 370 vm_offset_t off; 371 vm_size_t sz; 372 int error, rv; 373 374 if (start != trunc_page(start)) { 375 rv = __elfN(map_partial)(map, object, offset, start, 376 round_page(start), prot); 377 if (rv) 378 return (rv); 379 offset += round_page(start) - start; 380 start = round_page(start); 381 } 382 if (end != round_page(end)) { 383 rv = __elfN(map_partial)(map, object, offset + 384 trunc_page(end) - start, trunc_page(end), end, prot); 385 if (rv) 386 return (rv); 387 end = trunc_page(end); 388 } 389 if (end > start) { 390 if (offset & PAGE_MASK) { 391 /* 392 * The mapping is not page aligned. This means we have 393 * to copy the data. Sigh. 394 */ 395 rv = vm_map_find(map, NULL, 0, &start, end - start, 396 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0); 397 if (rv) 398 return (rv); 399 if (object == NULL) 400 return (KERN_SUCCESS); 401 for (; start < end; start += sz) { 402 sf = vm_imgact_map_page(object, offset); 403 if (sf == NULL) 404 return (KERN_FAILURE); 405 off = offset - trunc_page(offset); 406 sz = end - start; 407 if (sz > PAGE_SIZE - off) 408 sz = PAGE_SIZE - off; 409 error = copyout((caddr_t)sf_buf_kva(sf) + off, 410 (caddr_t)start, sz); 411 vm_imgact_unmap_page(sf); 412 if (error) { 413 return (KERN_FAILURE); 414 } 415 offset += sz; 416 } 417 rv = KERN_SUCCESS; 418 } else { 419 vm_object_reference(object); 420 vm_map_lock(map); 421 rv = vm_map_insert(map, object, offset, start, end, 422 prot, VM_PROT_ALL, cow); 423 vm_map_unlock(map); 424 if (rv != KERN_SUCCESS) 425 vm_object_deallocate(object); 426 } 427 return (rv); 428 } else { 429 return (KERN_SUCCESS); 430 } 431} 432 433static int 434__elfN(load_section)(struct vmspace *vmspace, 435 vm_object_t object, vm_offset_t offset, 436 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 437 size_t pagesize) 438{ 439 struct sf_buf *sf; 440 size_t map_len; 441 vm_offset_t map_addr; 442 int error, rv, cow; 443 size_t copy_len; 444 vm_offset_t file_addr; 445 446 /* 447 * It's necessary to fail if the filsz + offset taken from the 448 * header is greater than the actual file pager object's size. 449 * If we were to allow this, then the vm_map_find() below would 450 * walk right off the end of the file object and into the ether. 451 * 452 * While I'm here, might as well check for something else that 453 * is invalid: filsz cannot be greater than memsz. 454 */ 455 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 456 filsz > memsz) { 457 uprintf("elf_load_section: truncated ELF file\n"); 458 return (ENOEXEC); 459 } 460 461 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 462 file_addr = trunc_page_ps(offset, pagesize); 463 464 /* 465 * We have two choices. We can either clear the data in the last page 466 * of an oversized mapping, or we can start the anon mapping a page 467 * early and copy the initialized data into that first page. We 468 * choose the second.. 469 */ 470 if (memsz > filsz) 471 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 472 else 473 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 474 475 if (map_len != 0) { 476 /* cow flags: don't dump readonly sections in core */ 477 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 478 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 479 480 rv = __elfN(map_insert)(&vmspace->vm_map, 481 object, 482 file_addr, /* file offset */ 483 map_addr, /* virtual start */ 484 map_addr + map_len,/* virtual end */ 485 prot, 486 cow); 487 if (rv != KERN_SUCCESS) 488 return (EINVAL); 489 490 /* we can stop now if we've covered it all */ 491 if (memsz == filsz) { 492 return (0); 493 } 494 } 495 496 497 /* 498 * We have to get the remaining bit of the file into the first part 499 * of the oversized map segment. This is normally because the .data 500 * segment in the file is extended to provide bss. It's a neat idea 501 * to try and save a page, but it's a pain in the behind to implement. 502 */ 503 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 504 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 505 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 506 map_addr; 507 508 /* This had damn well better be true! */ 509 if (map_len != 0) { 510 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr, 511 map_addr + map_len, VM_PROT_ALL, 0); 512 if (rv != KERN_SUCCESS) { 513 return (EINVAL); 514 } 515 } 516 517 if (copy_len != 0) { 518 vm_offset_t off; 519 520 sf = vm_imgact_map_page(object, offset + filsz); 521 if (sf == NULL) 522 return (EIO); 523 524 /* send the page fragment to user space */ 525 off = trunc_page_ps(offset + filsz, pagesize) - 526 trunc_page(offset + filsz); 527 error = copyout((caddr_t)sf_buf_kva(sf) + off, 528 (caddr_t)map_addr, copy_len); 529 vm_imgact_unmap_page(sf); 530 if (error) { 531 return (error); 532 } 533 } 534 535 /* 536 * set it to the specified protection. 537 * XXX had better undo the damage from pasting over the cracks here! 538 */ 539 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr), 540 round_page(map_addr + map_len), prot, FALSE); 541 542 return (0); 543} 544 545/* 546 * Load the file "file" into memory. It may be either a shared object 547 * or an executable. 548 * 549 * The "addr" reference parameter is in/out. On entry, it specifies 550 * the address where a shared object should be loaded. If the file is 551 * an executable, this value is ignored. On exit, "addr" specifies 552 * where the file was actually loaded. 553 * 554 * The "entry" reference parameter is out only. On exit, it specifies 555 * the entry point for the loaded file. 556 */ 557static int 558__elfN(load_file)(struct proc *p, const char *file, u_long *addr, 559 u_long *entry, size_t pagesize) 560{ 561 struct { 562 struct nameidata nd; 563 struct vattr attr; 564 struct image_params image_params; 565 } *tempdata; 566 const Elf_Ehdr *hdr = NULL; 567 const Elf_Phdr *phdr = NULL; 568 struct nameidata *nd; 569 struct vmspace *vmspace = p->p_vmspace; 570 struct vattr *attr; 571 struct image_params *imgp; 572 vm_prot_t prot; 573 u_long rbase; 574 u_long base_addr = 0; 575 int vfslocked, error, i, numsegs; 576 577 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 578 nd = &tempdata->nd; 579 attr = &tempdata->attr; 580 imgp = &tempdata->image_params; 581 582 /* 583 * Initialize part of the common data 584 */ 585 imgp->proc = p; 586 imgp->attr = attr; 587 imgp->firstpage = NULL; 588 imgp->image_header = NULL; 589 imgp->object = NULL; 590 imgp->execlabel = NULL; 591 592 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, 593 curthread); 594 vfslocked = 0; 595 if ((error = namei(nd)) != 0) { 596 nd->ni_vp = NULL; 597 goto fail; 598 } 599 vfslocked = NDHASGIANT(nd); 600 NDFREE(nd, NDF_ONLY_PNBUF); 601 imgp->vp = nd->ni_vp; 602 603 /* 604 * Check permissions, modes, uid, etc on the file, and "open" it. 605 */ 606 error = exec_check_permissions(imgp); 607 if (error) 608 goto fail; 609 610 error = exec_map_first_page(imgp); 611 if (error) 612 goto fail; 613 614 /* 615 * Also make certain that the interpreter stays the same, so set 616 * its VV_TEXT flag, too. 617 */ 618 nd->ni_vp->v_vflag |= VV_TEXT; 619 620 imgp->object = nd->ni_vp->v_object; 621 622 hdr = (const Elf_Ehdr *)imgp->image_header; 623 if ((error = __elfN(check_header)(hdr)) != 0) 624 goto fail; 625 if (hdr->e_type == ET_DYN) 626 rbase = *addr; 627 else if (hdr->e_type == ET_EXEC) 628 rbase = 0; 629 else { 630 error = ENOEXEC; 631 goto fail; 632 } 633 634 /* Only support headers that fit within first page for now */ 635 /* (multiplication of two Elf_Half fields will not overflow) */ 636 if ((hdr->e_phoff > PAGE_SIZE) || 637 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 638 error = ENOEXEC; 639 goto fail; 640 } 641 642 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 643 if (!aligned(phdr, Elf_Addr)) { 644 error = ENOEXEC; 645 goto fail; 646 } 647 648 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 649 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 650 /* Loadable segment */ 651 prot = 0; 652 if (phdr[i].p_flags & PF_X) 653 prot |= VM_PROT_EXECUTE; 654 if (phdr[i].p_flags & PF_W) 655 prot |= VM_PROT_WRITE; 656 if (phdr[i].p_flags & PF_R) 657 prot |= VM_PROT_READ; 658 659 if ((error = __elfN(load_section)(vmspace, 660 imgp->object, phdr[i].p_offset, 661 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 662 phdr[i].p_memsz, phdr[i].p_filesz, prot, 663 pagesize)) != 0) 664 goto fail; 665 /* 666 * Establish the base address if this is the 667 * first segment. 668 */ 669 if (numsegs == 0) 670 base_addr = trunc_page(phdr[i].p_vaddr + 671 rbase); 672 numsegs++; 673 } 674 } 675 *addr = base_addr; 676 *entry = (unsigned long)hdr->e_entry + rbase; 677 678fail: 679 if (imgp->firstpage) 680 exec_unmap_first_page(imgp); 681 682 if (nd->ni_vp) 683 vput(nd->ni_vp); 684 685 VFS_UNLOCK_GIANT(vfslocked); 686 free(tempdata, M_TEMP); 687 688 return (error); 689} 690 691static int 692__CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 693{ 694 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 695 const Elf_Phdr *phdr; 696 Elf_Auxargs *elf_auxargs; 697 struct vmspace *vmspace; 698 vm_prot_t prot; 699 u_long text_size = 0, data_size = 0, total_size = 0; 700 u_long text_addr = 0, data_addr = 0; 701 u_long seg_size, seg_addr; 702 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0; 703 int32_t osrel = 0; 704 int error = 0, i, n; 705 const char *interp = NULL, *newinterp = NULL; 706 Elf_Brandinfo *brand_info; 707 char *path; 708 struct sysentvec *sv; 709 710 /* 711 * Do we have a valid ELF header ? 712 * 713 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 714 * if particular brand doesn't support it. 715 */ 716 if (__elfN(check_header)(hdr) != 0 || 717 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 718 return (-1); 719 720 /* 721 * From here on down, we return an errno, not -1, as we've 722 * detected an ELF file. 723 */ 724 725 if ((hdr->e_phoff > PAGE_SIZE) || 726 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 727 /* Only support headers in first page for now */ 728 return (ENOEXEC); 729 } 730 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 731 if (!aligned(phdr, Elf_Addr)) 732 return (ENOEXEC); 733 n = 0; 734 baddr = 0; 735 for (i = 0; i < hdr->e_phnum; i++) { 736 if (phdr[i].p_type == PT_LOAD) { 737 if (n == 0) 738 baddr = phdr[i].p_vaddr; 739 n++; 740 continue; 741 } 742 if (phdr[i].p_type == PT_INTERP) { 743 /* Path to interpreter */ 744 if (phdr[i].p_filesz > MAXPATHLEN || 745 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) 746 return (ENOEXEC); 747 interp = imgp->image_header + phdr[i].p_offset; 748 continue; 749 } 750 } 751 752 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel); 753 if (brand_info == NULL) { 754 uprintf("ELF binary type \"%u\" not known.\n", 755 hdr->e_ident[EI_OSABI]); 756 return (ENOEXEC); 757 } 758 if (hdr->e_type == ET_DYN) { 759 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) 760 return (ENOEXEC); 761 /* 762 * Honour the base load address from the dso if it is 763 * non-zero for some reason. 764 */ 765 if (baddr == 0) 766 et_dyn_addr = ET_DYN_LOAD_ADDR; 767 else 768 et_dyn_addr = 0; 769 } else 770 et_dyn_addr = 0; 771 sv = brand_info->sysvec; 772 if (interp != NULL && brand_info->interp_newpath != NULL) 773 newinterp = brand_info->interp_newpath; 774 775 /* 776 * Avoid a possible deadlock if the current address space is destroyed 777 * and that address space maps the locked vnode. In the common case, 778 * the locked vnode's v_usecount is decremented but remains greater 779 * than zero. Consequently, the vnode lock is not needed by vrele(). 780 * However, in cases where the vnode lock is external, such as nullfs, 781 * v_usecount may become zero. 782 */ 783 VOP_UNLOCK(imgp->vp, 0); 784 785 error = exec_new_vmspace(imgp, sv); 786 imgp->proc->p_sysent = sv; 787 788 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 789 if (error) 790 return (error); 791 792 vmspace = imgp->proc->p_vmspace; 793 794 for (i = 0; i < hdr->e_phnum; i++) { 795 switch (phdr[i].p_type) { 796 case PT_LOAD: /* Loadable segment */ 797 if (phdr[i].p_memsz == 0) 798 break; 799 prot = 0; 800 if (phdr[i].p_flags & PF_X) 801 prot |= VM_PROT_EXECUTE; 802 if (phdr[i].p_flags & PF_W) 803 prot |= VM_PROT_WRITE; 804 if (phdr[i].p_flags & PF_R) 805 prot |= VM_PROT_READ; 806 807#if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER) 808 /* 809 * Some x86 binaries assume read == executable, 810 * notably the M3 runtime and therefore cvsup 811 */ 812 if (prot & VM_PROT_READ) 813 prot |= VM_PROT_EXECUTE; 814#endif 815 816 if ((error = __elfN(load_section)(vmspace, 817 imgp->object, phdr[i].p_offset, 818 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 819 phdr[i].p_memsz, phdr[i].p_filesz, prot, 820 sv->sv_pagesize)) != 0) 821 return (error); 822 823 /* 824 * If this segment contains the program headers, 825 * remember their virtual address for the AT_PHDR 826 * aux entry. Static binaries don't usually include 827 * a PT_PHDR entry. 828 */ 829 if (phdr[i].p_offset == 0 && 830 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 831 <= phdr[i].p_filesz) 832 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 833 et_dyn_addr; 834 835 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 836 seg_size = round_page(phdr[i].p_memsz + 837 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 838 839 /* 840 * Is this .text or .data? We can't use 841 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 842 * alpha terribly and possibly does other bad 843 * things so we stick to the old way of figuring 844 * it out: If the segment contains the program 845 * entry point, it's a text segment, otherwise it 846 * is a data segment. 847 * 848 * Note that obreak() assumes that data_addr + 849 * data_size == end of data load area, and the ELF 850 * file format expects segments to be sorted by 851 * address. If multiple data segments exist, the 852 * last one will be used. 853 */ 854 if (hdr->e_entry >= phdr[i].p_vaddr && 855 hdr->e_entry < (phdr[i].p_vaddr + 856 phdr[i].p_memsz)) { 857 text_size = seg_size; 858 text_addr = seg_addr; 859 entry = (u_long)hdr->e_entry + et_dyn_addr; 860 } else { 861 data_size = seg_size; 862 data_addr = seg_addr; 863 } 864 total_size += seg_size; 865 break; 866 case PT_PHDR: /* Program header table info */ 867 proghdr = phdr[i].p_vaddr + et_dyn_addr; 868 break; 869 default: 870 break; 871 } 872 } 873 874 if (data_addr == 0 && data_size == 0) { 875 data_addr = text_addr; 876 data_size = text_size; 877 } 878 879 /* 880 * Check limits. It should be safe to check the 881 * limits after loading the segments since we do 882 * not actually fault in all the segments pages. 883 */ 884 PROC_LOCK(imgp->proc); 885 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 886 text_size > maxtsiz || 887 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) { 888 PROC_UNLOCK(imgp->proc); 889 return (ENOMEM); 890 } 891 892 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 893 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 894 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 895 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 896 897 /* 898 * We load the dynamic linker where a userland call 899 * to mmap(0, ...) would put it. The rationale behind this 900 * calculation is that it leaves room for the heap to grow to 901 * its maximum allowed size. 902 */ 903 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr + 904 lim_max(imgp->proc, RLIMIT_DATA)); 905 PROC_UNLOCK(imgp->proc); 906 907 imgp->entry_addr = entry; 908 909 if (interp != NULL) { 910 int have_interp = FALSE; 911 VOP_UNLOCK(imgp->vp, 0); 912 if (brand_info->emul_path != NULL && 913 brand_info->emul_path[0] != '\0') { 914 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 915 snprintf(path, MAXPATHLEN, "%s%s", 916 brand_info->emul_path, interp); 917 error = __elfN(load_file)(imgp->proc, path, &addr, 918 &imgp->entry_addr, sv->sv_pagesize); 919 free(path, M_TEMP); 920 if (error == 0) 921 have_interp = TRUE; 922 } 923 if (!have_interp && newinterp != NULL) { 924 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 925 &imgp->entry_addr, sv->sv_pagesize); 926 if (error == 0) 927 have_interp = TRUE; 928 } 929 if (!have_interp) { 930 error = __elfN(load_file)(imgp->proc, interp, &addr, 931 &imgp->entry_addr, sv->sv_pagesize); 932 } 933 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 934 if (error != 0) { 935 uprintf("ELF interpreter %s not found\n", interp); 936 return (error); 937 } 938 } else 939 addr = et_dyn_addr; 940 941 /* 942 * Construct auxargs table (used by the fixup routine) 943 */ 944 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 945 elf_auxargs->execfd = -1; 946 elf_auxargs->phdr = proghdr; 947 elf_auxargs->phent = hdr->e_phentsize; 948 elf_auxargs->phnum = hdr->e_phnum; 949 elf_auxargs->pagesz = PAGE_SIZE; 950 elf_auxargs->base = addr; 951 elf_auxargs->flags = 0; 952 elf_auxargs->entry = entry; 953 954 imgp->auxargs = elf_auxargs; 955 imgp->interpreted = 0; 956 imgp->proc->p_osrel = osrel; 957 958 return (error); 959} 960 961#define suword __CONCAT(suword, __ELF_WORD_SIZE) 962 963int 964__elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 965{ 966 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 967 Elf_Addr *base; 968 Elf_Addr *pos; 969 970 base = (Elf_Addr *)*stack_base; 971 pos = base + (imgp->args->argc + imgp->args->envc + 2); 972 973 if (args->execfd != -1) 974 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 975 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 976 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 977 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 978 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 979 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 980 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 981 AUXARGS_ENTRY(pos, AT_BASE, args->base); 982 if (imgp->execpathp != 0) 983 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 984 AUXARGS_ENTRY(pos, AT_NULL, 0); 985 986 free(imgp->auxargs, M_TEMP); 987 imgp->auxargs = NULL; 988 989 base--; 990 suword(base, (long)imgp->args->argc); 991 *stack_base = (register_t *)base; 992 return (0); 993} 994 995/* 996 * Code for generating ELF core dumps. 997 */ 998 999typedef void (*segment_callback)(vm_map_entry_t, void *); 1000 1001/* Closure for cb_put_phdr(). */ 1002struct phdr_closure { 1003 Elf_Phdr *phdr; /* Program header to fill in */ 1004 Elf_Off offset; /* Offset of segment in core file */ 1005}; 1006 1007/* Closure for cb_size_segment(). */ 1008struct sseg_closure { 1009 int count; /* Count of writable segments. */ 1010 size_t size; /* Total size of all writable segments. */ 1011}; 1012 1013static void cb_put_phdr(vm_map_entry_t, void *); 1014static void cb_size_segment(vm_map_entry_t, void *); 1015static void each_writable_segment(struct thread *, segment_callback, void *); 1016static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 1017 int, void *, size_t, gzFile); 1018static void __elfN(puthdr)(struct thread *, void *, size_t *, int); 1019static void __elfN(putnote)(void *, size_t *, const char *, int, 1020 const void *, size_t); 1021 1022#ifdef COMPRESS_USER_CORES 1023extern int compress_user_cores; 1024extern int compress_user_cores_gzlevel; 1025#endif 1026 1027static int 1028core_output(struct vnode *vp, void *base, size_t len, off_t offset, 1029 struct ucred *active_cred, struct ucred *file_cred, 1030 struct thread *td, char *core_buf, gzFile gzfile) { 1031 1032 int error; 1033 if (gzfile) { 1034#ifdef COMPRESS_USER_CORES 1035 error = compress_core(gzfile, base, core_buf, len, td); 1036#else 1037 panic("shouldn't be here"); 1038#endif 1039 } else { 1040 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset, 1041 UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred, 1042 NULL, td); 1043 } 1044 return (error); 1045} 1046 1047int 1048__elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1049{ 1050 struct ucred *cred = td->td_ucred; 1051 int error = 0; 1052 struct sseg_closure seginfo; 1053 void *hdr; 1054 size_t hdrsize; 1055 1056 gzFile gzfile = Z_NULL; 1057 char *core_buf = NULL; 1058#ifdef COMPRESS_USER_CORES 1059 char gzopen_flags[8]; 1060 char *p; 1061 int doing_compress = flags & IMGACT_CORE_COMPRESS; 1062#endif 1063 1064 hdr = NULL; 1065 1066#ifdef COMPRESS_USER_CORES 1067 if (doing_compress) { 1068 p = gzopen_flags; 1069 *p++ = 'w'; 1070 if (compress_user_cores_gzlevel >= 0 && 1071 compress_user_cores_gzlevel <= 9) 1072 *p++ = '0' + compress_user_cores_gzlevel; 1073 *p = 0; 1074 gzfile = gz_open("", gzopen_flags, vp); 1075 if (gzfile == Z_NULL) { 1076 error = EFAULT; 1077 goto done; 1078 } 1079 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1080 if (!core_buf) { 1081 error = ENOMEM; 1082 goto done; 1083 } 1084 } 1085#endif 1086 1087 /* Size the program segments. */ 1088 seginfo.count = 0; 1089 seginfo.size = 0; 1090 each_writable_segment(td, cb_size_segment, &seginfo); 1091 1092 /* 1093 * Calculate the size of the core file header area by making 1094 * a dry run of generating it. Nothing is written, but the 1095 * size is calculated. 1096 */ 1097 hdrsize = 0; 1098 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count); 1099 1100 if (hdrsize + seginfo.size >= limit) 1101 return (EFAULT); 1102 1103 /* 1104 * Allocate memory for building the header, fill it up, 1105 * and write it out. 1106 */ 1107 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1108 if (hdr == NULL) { 1109 return (EINVAL); 1110 } 1111 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize, 1112 gzfile); 1113 1114 /* Write the contents of all of the writable segments. */ 1115 if (error == 0) { 1116 Elf_Phdr *php; 1117 off_t offset; 1118 int i; 1119 1120 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1121 offset = hdrsize; 1122 for (i = 0; i < seginfo.count; i++) { 1123 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr, 1124 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile); 1125 if (error != 0) 1126 break; 1127 offset += php->p_filesz; 1128 php++; 1129 } 1130 } 1131 if (error) { 1132 log(LOG_WARNING, 1133 "Failed to write core file for process %s (error %d)\n", 1134 curproc->p_comm, error); 1135 } 1136 1137#ifdef COMPRESS_USER_CORES 1138done: 1139#endif 1140 if (core_buf) 1141 free(core_buf, M_TEMP); 1142 if (gzfile) 1143 gzclose(gzfile); 1144 1145 free(hdr, M_TEMP); 1146 1147 return (error); 1148} 1149 1150/* 1151 * A callback for each_writable_segment() to write out the segment's 1152 * program header entry. 1153 */ 1154static void 1155cb_put_phdr(entry, closure) 1156 vm_map_entry_t entry; 1157 void *closure; 1158{ 1159 struct phdr_closure *phc = (struct phdr_closure *)closure; 1160 Elf_Phdr *phdr = phc->phdr; 1161 1162 phc->offset = round_page(phc->offset); 1163 1164 phdr->p_type = PT_LOAD; 1165 phdr->p_offset = phc->offset; 1166 phdr->p_vaddr = entry->start; 1167 phdr->p_paddr = 0; 1168 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1169 phdr->p_align = PAGE_SIZE; 1170 phdr->p_flags = 0; 1171 if (entry->protection & VM_PROT_READ) 1172 phdr->p_flags |= PF_R; 1173 if (entry->protection & VM_PROT_WRITE) 1174 phdr->p_flags |= PF_W; 1175 if (entry->protection & VM_PROT_EXECUTE) 1176 phdr->p_flags |= PF_X; 1177 1178 phc->offset += phdr->p_filesz; 1179 phc->phdr++; 1180} 1181 1182/* 1183 * A callback for each_writable_segment() to gather information about 1184 * the number of segments and their total size. 1185 */ 1186static void 1187cb_size_segment(entry, closure) 1188 vm_map_entry_t entry; 1189 void *closure; 1190{ 1191 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1192 1193 ssc->count++; 1194 ssc->size += entry->end - entry->start; 1195} 1196 1197/* 1198 * For each writable segment in the process's memory map, call the given 1199 * function with a pointer to the map entry and some arbitrary 1200 * caller-supplied data. 1201 */ 1202static void 1203each_writable_segment(td, func, closure) 1204 struct thread *td; 1205 segment_callback func; 1206 void *closure; 1207{ 1208 struct proc *p = td->td_proc; 1209 vm_map_t map = &p->p_vmspace->vm_map; 1210 vm_map_entry_t entry; 1211 vm_object_t backing_object, object; 1212 boolean_t ignore_entry; 1213 1214 vm_map_lock_read(map); 1215 for (entry = map->header.next; entry != &map->header; 1216 entry = entry->next) { 1217 /* 1218 * Don't dump inaccessible mappings, deal with legacy 1219 * coredump mode. 1220 * 1221 * Note that read-only segments related to the elf binary 1222 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1223 * need to arbitrarily ignore such segments. 1224 */ 1225 if (elf_legacy_coredump) { 1226 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1227 continue; 1228 } else { 1229 if ((entry->protection & VM_PROT_ALL) == 0) 1230 continue; 1231 } 1232 1233 /* 1234 * Dont include memory segment in the coredump if 1235 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1236 * madvise(2). Do not dump submaps (i.e. parts of the 1237 * kernel map). 1238 */ 1239 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1240 continue; 1241 1242 if ((object = entry->object.vm_object) == NULL) 1243 continue; 1244 1245 /* Ignore memory-mapped devices and such things. */ 1246 VM_OBJECT_LOCK(object); 1247 while ((backing_object = object->backing_object) != NULL) { 1248 VM_OBJECT_LOCK(backing_object); 1249 VM_OBJECT_UNLOCK(object); 1250 object = backing_object; 1251 } 1252 ignore_entry = object->type != OBJT_DEFAULT && 1253 object->type != OBJT_SWAP && object->type != OBJT_VNODE; 1254 VM_OBJECT_UNLOCK(object); 1255 if (ignore_entry) 1256 continue; 1257 1258 (*func)(entry, closure); 1259 } 1260 vm_map_unlock_read(map); 1261} 1262 1263/* 1264 * Write the core file header to the file, including padding up to 1265 * the page boundary. 1266 */ 1267static int 1268__elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize, gzfile) 1269 struct thread *td; 1270 struct vnode *vp; 1271 struct ucred *cred; 1272 int numsegs; 1273 size_t hdrsize; 1274 void *hdr; 1275 gzFile gzfile; 1276{ 1277 size_t off; 1278 1279 /* Fill in the header. */ 1280 bzero(hdr, hdrsize); 1281 off = 0; 1282 __elfN(puthdr)(td, hdr, &off, numsegs); 1283 1284 if (!gzfile) { 1285 /* Write it to the core file. */ 1286 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1287 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1288 td)); 1289 } else { 1290#ifdef COMPRESS_USER_CORES 1291 if (gzwrite(gzfile, hdr, hdrsize) != hdrsize) { 1292 log(LOG_WARNING, 1293 "Failed to compress core file header for process" 1294 " %s.\n", curproc->p_comm); 1295 return (EFAULT); 1296 } 1297 else { 1298 return (0); 1299 } 1300#else 1301 panic("shouldn't be here"); 1302#endif 1303 } 1304} 1305 1306#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1307typedef struct prstatus32 elf_prstatus_t; 1308typedef struct prpsinfo32 elf_prpsinfo_t; 1309typedef struct fpreg32 elf_prfpregset_t; 1310typedef struct fpreg32 elf_fpregset_t; 1311typedef struct reg32 elf_gregset_t; 1312#else 1313typedef prstatus_t elf_prstatus_t; 1314typedef prpsinfo_t elf_prpsinfo_t; 1315typedef prfpregset_t elf_prfpregset_t; 1316typedef prfpregset_t elf_fpregset_t; 1317typedef gregset_t elf_gregset_t; 1318#endif 1319 1320static void 1321__elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs) 1322{ 1323 struct { 1324 elf_prstatus_t status; 1325 elf_prfpregset_t fpregset; 1326 elf_prpsinfo_t psinfo; 1327 } *tempdata; 1328 elf_prstatus_t *status; 1329 elf_prfpregset_t *fpregset; 1330 elf_prpsinfo_t *psinfo; 1331 struct proc *p; 1332 struct thread *thr; 1333 size_t ehoff, noteoff, notesz, phoff; 1334 1335 p = td->td_proc; 1336 1337 ehoff = *off; 1338 *off += sizeof(Elf_Ehdr); 1339 1340 phoff = *off; 1341 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1342 1343 noteoff = *off; 1344 /* 1345 * Don't allocate space for the notes if we're just calculating 1346 * the size of the header. We also don't collect the data. 1347 */ 1348 if (dst != NULL) { 1349 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK); 1350 status = &tempdata->status; 1351 fpregset = &tempdata->fpregset; 1352 psinfo = &tempdata->psinfo; 1353 } else { 1354 tempdata = NULL; 1355 status = NULL; 1356 fpregset = NULL; 1357 psinfo = NULL; 1358 } 1359 1360 if (dst != NULL) { 1361 psinfo->pr_version = PRPSINFO_VERSION; 1362 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1363 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1364 /* 1365 * XXX - We don't fill in the command line arguments properly 1366 * yet. 1367 */ 1368 strlcpy(psinfo->pr_psargs, p->p_comm, 1369 sizeof(psinfo->pr_psargs)); 1370 } 1371 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1372 sizeof *psinfo); 1373 1374 /* 1375 * To have the debugger select the right thread (LWP) as the initial 1376 * thread, we dump the state of the thread passed to us in td first. 1377 * This is the thread that causes the core dump and thus likely to 1378 * be the right thread one wants to have selected in the debugger. 1379 */ 1380 thr = td; 1381 while (thr != NULL) { 1382 if (dst != NULL) { 1383 status->pr_version = PRSTATUS_VERSION; 1384 status->pr_statussz = sizeof(elf_prstatus_t); 1385 status->pr_gregsetsz = sizeof(elf_gregset_t); 1386 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1387 status->pr_osreldate = osreldate; 1388 status->pr_cursig = p->p_sig; 1389 status->pr_pid = thr->td_tid; 1390#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1391 fill_regs32(thr, &status->pr_reg); 1392 fill_fpregs32(thr, fpregset); 1393#else 1394 fill_regs(thr, &status->pr_reg); 1395 fill_fpregs(thr, fpregset); 1396#endif 1397 } 1398 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1399 sizeof *status); 1400 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1401 sizeof *fpregset); 1402 /* 1403 * Allow for MD specific notes, as well as any MD 1404 * specific preparations for writing MI notes. 1405 */ 1406 __elfN(dump_thread)(thr, dst, off); 1407 1408 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1409 TAILQ_NEXT(thr, td_plist); 1410 if (thr == td) 1411 thr = TAILQ_NEXT(thr, td_plist); 1412 } 1413 1414 notesz = *off - noteoff; 1415 1416 if (dst != NULL) 1417 free(tempdata, M_TEMP); 1418 1419 /* Align up to a page boundary for the program segments. */ 1420 *off = round_page(*off); 1421 1422 if (dst != NULL) { 1423 Elf_Ehdr *ehdr; 1424 Elf_Phdr *phdr; 1425 struct phdr_closure phc; 1426 1427 /* 1428 * Fill in the ELF header. 1429 */ 1430 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1431 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1432 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1433 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1434 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1435 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1436 ehdr->e_ident[EI_DATA] = ELF_DATA; 1437 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1438 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1439 ehdr->e_ident[EI_ABIVERSION] = 0; 1440 ehdr->e_ident[EI_PAD] = 0; 1441 ehdr->e_type = ET_CORE; 1442#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1443 ehdr->e_machine = EM_386; 1444#else 1445 ehdr->e_machine = ELF_ARCH; 1446#endif 1447 ehdr->e_version = EV_CURRENT; 1448 ehdr->e_entry = 0; 1449 ehdr->e_phoff = phoff; 1450 ehdr->e_flags = 0; 1451 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1452 ehdr->e_phentsize = sizeof(Elf_Phdr); 1453 ehdr->e_phnum = numsegs + 1; 1454 ehdr->e_shentsize = sizeof(Elf_Shdr); 1455 ehdr->e_shnum = 0; 1456 ehdr->e_shstrndx = SHN_UNDEF; 1457 1458 /* 1459 * Fill in the program header entries. 1460 */ 1461 phdr = (Elf_Phdr *)((char *)dst + phoff); 1462 1463 /* The note segement. */ 1464 phdr->p_type = PT_NOTE; 1465 phdr->p_offset = noteoff; 1466 phdr->p_vaddr = 0; 1467 phdr->p_paddr = 0; 1468 phdr->p_filesz = notesz; 1469 phdr->p_memsz = 0; 1470 phdr->p_flags = 0; 1471 phdr->p_align = 0; 1472 phdr++; 1473 1474 /* All the writable segments from the program. */ 1475 phc.phdr = phdr; 1476 phc.offset = *off; 1477 each_writable_segment(td, cb_put_phdr, &phc); 1478 } 1479} 1480 1481static void 1482__elfN(putnote)(void *dst, size_t *off, const char *name, int type, 1483 const void *desc, size_t descsz) 1484{ 1485 Elf_Note note; 1486 1487 note.n_namesz = strlen(name) + 1; 1488 note.n_descsz = descsz; 1489 note.n_type = type; 1490 if (dst != NULL) 1491 bcopy(¬e, (char *)dst + *off, sizeof note); 1492 *off += sizeof note; 1493 if (dst != NULL) 1494 bcopy(name, (char *)dst + *off, note.n_namesz); 1495 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1496 if (dst != NULL) 1497 bcopy(desc, (char *)dst + *off, note.n_descsz); 1498 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1499} 1500 1501/* 1502 * Try to find the appropriate ABI-note section for checknote, 1503 * fetch the osreldate for binary from the ELF OSABI-note. Only the 1504 * first page of the image is searched, the same as for headers. 1505 */ 1506static boolean_t 1507__elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 1508 int32_t *osrel) 1509{ 1510 const Elf_Note *note, *note0, *note_end; 1511 const Elf_Phdr *phdr, *pnote; 1512 const Elf_Ehdr *hdr; 1513 const char *note_name; 1514 int i; 1515 1516 pnote = NULL; 1517 hdr = (const Elf_Ehdr *)imgp->image_header; 1518 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1519 1520 for (i = 0; i < hdr->e_phnum; i++) { 1521 if (phdr[i].p_type == PT_NOTE) { 1522 pnote = &phdr[i]; 1523 break; 1524 } 1525 } 1526 1527 if (pnote == NULL || pnote->p_offset >= PAGE_SIZE || 1528 pnote->p_offset + pnote->p_filesz >= PAGE_SIZE) 1529 return (FALSE); 1530 1531 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset); 1532 note_end = (const Elf_Note *)(imgp->image_header + 1533 pnote->p_offset + pnote->p_filesz); 1534 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 1535 if (!aligned(note, Elf32_Addr)) 1536 return (FALSE); 1537 if (note->n_namesz != checknote->hdr.n_namesz || 1538 note->n_descsz != checknote->hdr.n_descsz || 1539 note->n_type != checknote->hdr.n_type) 1540 goto nextnote; 1541 note_name = (const char *)(note + 1); 1542 if (strncmp(checknote->vendor, note_name, 1543 checknote->hdr.n_namesz) != 0) 1544 goto nextnote; 1545 1546 /* 1547 * Fetch the osreldate for binary 1548 * from the ELF OSABI-note if necessary. 1549 */ 1550 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 1551 checknote->trans_osrel != NULL) 1552 return (checknote->trans_osrel(note, osrel)); 1553 return (TRUE); 1554 1555nextnote: 1556 note = (const Elf_Note *)((const char *)(note + 1) + 1557 roundup2(note->n_namesz, sizeof(Elf32_Addr)) + 1558 roundup2(note->n_descsz, sizeof(Elf32_Addr))); 1559 } 1560 1561 return (FALSE); 1562} 1563 1564/* 1565 * Tell kern_execve.c about it, with a little help from the linker. 1566 */ 1567static struct execsw __elfN(execsw) = { 1568 __CONCAT(exec_, __elfN(imgact)), 1569 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 1570}; 1571EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 1572 1573#ifdef COMPRESS_USER_CORES 1574/* 1575 * Compress and write out a core segment for a user process. 1576 * 1577 * 'inbuf' is the starting address of a VM segment in the process' address 1578 * space that is to be compressed and written out to the core file. 'dest_buf' 1579 * is a buffer in the kernel's address space. The segment is copied from 1580 * 'inbuf' to 'dest_buf' first before being processed by the compression 1581 * routine gzwrite(). This copying is necessary because the content of the VM 1582 * segment may change between the compression pass and the crc-computation pass 1583 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel. 1584 */ 1585static int 1586compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len, 1587 struct thread *td) 1588{ 1589 int len_compressed; 1590 int error = 0; 1591 unsigned int chunk_len; 1592 1593 while (len) { 1594 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len; 1595 copyin(inbuf, dest_buf, chunk_len); 1596 len_compressed = gzwrite(file, dest_buf, chunk_len); 1597 1598 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed); 1599 1600 if ((unsigned int)len_compressed != chunk_len) { 1601 log(LOG_WARNING, 1602 "compress_core: length mismatch (0x%x returned, " 1603 "0x%x expected)\n", len_compressed, chunk_len); 1604 EVENTHANDLER_INVOKE(app_coredump_error, td, 1605 "compress_core: length mismatch %x -> %x", 1606 chunk_len, len_compressed); 1607 error = EFAULT; 1608 break; 1609 } 1610 inbuf += chunk_len; 1611 len -= chunk_len; 1612 if (ticks - PCPU_GET(switchticks) >= hogticks) 1613 uio_yield(); 1614 } 1615 1616 return (error); 1617} 1618#endif /* COMPRESS_USER_CORES */ 1619