imgact_elf.c revision 186235
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 186235 2008-12-17 16:54:29Z peter $"); 33 34#include "opt_compat.h" 35 36#include <sys/param.h> 37#include <sys/exec.h> 38#include <sys/fcntl.h> 39#include <sys/imgact.h> 40#include <sys/imgact_elf.h> 41#include <sys/kernel.h> 42#include <sys/lock.h> 43#include <sys/malloc.h> 44#include <sys/mount.h> 45#include <sys/mutex.h> 46#include <sys/mman.h> 47#include <sys/namei.h> 48#include <sys/pioctl.h> 49#include <sys/proc.h> 50#include <sys/procfs.h> 51#include <sys/resourcevar.h> 52#include <sys/sf_buf.h> 53#include <sys/systm.h> 54#include <sys/signalvar.h> 55#include <sys/stat.h> 56#include <sys/sx.h> 57#include <sys/syscall.h> 58#include <sys/sysctl.h> 59#include <sys/sysent.h> 60#include <sys/vnode.h> 61 62#include <vm/vm.h> 63#include <vm/vm_kern.h> 64#include <vm/vm_param.h> 65#include <vm/pmap.h> 66#include <vm/vm_map.h> 67#include <vm/vm_object.h> 68#include <vm/vm_extern.h> 69 70#include <machine/elf.h> 71#include <machine/md_var.h> 72 73#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 74#include <machine/fpu.h> 75#include <compat/ia32/ia32_reg.h> 76#endif 77 78#define OLD_EI_BRAND 8 79 80static int __elfN(check_header)(const Elf_Ehdr *hdr); 81static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr, 82 const char *interp); 83static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 84 u_long *entry, size_t pagesize); 85static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object, 86 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 87 vm_prot_t prot, size_t pagesize); 88static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 89 90SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 91 ""); 92 93int __elfN(fallback_brand) = -1; 94SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 95 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 96 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 97TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 98 &__elfN(fallback_brand)); 99 100static int elf_legacy_coredump = 0; 101SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 102 &elf_legacy_coredump, 0, ""); 103 104static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 105 106#define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 107#define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 108#define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a)) 109 110int 111__elfN(insert_brand_entry)(Elf_Brandinfo *entry) 112{ 113 int i; 114 115 for (i = 0; i < MAX_BRANDS; i++) { 116 if (elf_brand_list[i] == NULL) { 117 elf_brand_list[i] = entry; 118 break; 119 } 120 } 121 if (i == MAX_BRANDS) 122 return (-1); 123 return (0); 124} 125 126int 127__elfN(remove_brand_entry)(Elf_Brandinfo *entry) 128{ 129 int i; 130 131 for (i = 0; i < MAX_BRANDS; i++) { 132 if (elf_brand_list[i] == entry) { 133 elf_brand_list[i] = NULL; 134 break; 135 } 136 } 137 if (i == MAX_BRANDS) 138 return (-1); 139 return (0); 140} 141 142int 143__elfN(brand_inuse)(Elf_Brandinfo *entry) 144{ 145 struct proc *p; 146 int rval = FALSE; 147 148 sx_slock(&allproc_lock); 149 FOREACH_PROC_IN_SYSTEM(p) { 150 if (p->p_sysent == entry->sysvec) { 151 rval = TRUE; 152 break; 153 } 154 } 155 sx_sunlock(&allproc_lock); 156 157 return (rval); 158} 159 160static Elf_Brandinfo * 161__elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp) 162{ 163 Elf_Brandinfo *bi; 164 int i; 165 166 /* 167 * We support three types of branding -- (1) the ELF EI_OSABI field 168 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 169 * branding w/in the ELF header, and (3) path of the `interp_path' 170 * field. We should also look for an ".note.ABI-tag" ELF section now 171 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones. 172 */ 173 174 /* If the executable has a brand, search for it in the brand list. */ 175 for (i = 0; i < MAX_BRANDS; i++) { 176 bi = elf_brand_list[i]; 177 if (bi != NULL && hdr->e_machine == bi->machine && 178 (hdr->e_ident[EI_OSABI] == bi->brand || 179 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 180 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 181 return (bi); 182 } 183 184 /* Lacking a known brand, search for a recognized interpreter. */ 185 if (interp != NULL) { 186 for (i = 0; i < MAX_BRANDS; i++) { 187 bi = elf_brand_list[i]; 188 if (bi != NULL && hdr->e_machine == bi->machine && 189 strcmp(interp, bi->interp_path) == 0) 190 return (bi); 191 } 192 } 193 194 /* Lacking a recognized interpreter, try the default brand */ 195 for (i = 0; i < MAX_BRANDS; i++) { 196 bi = elf_brand_list[i]; 197 if (bi != NULL && hdr->e_machine == bi->machine && 198 __elfN(fallback_brand) == bi->brand) 199 return (bi); 200 } 201 return (NULL); 202} 203 204static int 205__elfN(check_header)(const Elf_Ehdr *hdr) 206{ 207 Elf_Brandinfo *bi; 208 int i; 209 210 if (!IS_ELF(*hdr) || 211 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 212 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 213 hdr->e_ident[EI_VERSION] != EV_CURRENT || 214 hdr->e_phentsize != sizeof(Elf_Phdr) || 215 hdr->e_version != ELF_TARG_VER) 216 return (ENOEXEC); 217 218 /* 219 * Make sure we have at least one brand for this machine. 220 */ 221 222 for (i = 0; i < MAX_BRANDS; i++) { 223 bi = elf_brand_list[i]; 224 if (bi != NULL && bi->machine == hdr->e_machine) 225 break; 226 } 227 if (i == MAX_BRANDS) 228 return (ENOEXEC); 229 230 return (0); 231} 232 233static int 234__elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 235 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 236{ 237 struct sf_buf *sf; 238 int error; 239 vm_offset_t off; 240 241 /* 242 * Create the page if it doesn't exist yet. Ignore errors. 243 */ 244 vm_map_lock(map); 245 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), 246 VM_PROT_ALL, VM_PROT_ALL, 0); 247 vm_map_unlock(map); 248 249 /* 250 * Find the page from the underlying object. 251 */ 252 if (object) { 253 sf = vm_imgact_map_page(object, offset); 254 if (sf == NULL) 255 return (KERN_FAILURE); 256 off = offset - trunc_page(offset); 257 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 258 end - start); 259 vm_imgact_unmap_page(sf); 260 if (error) { 261 return (KERN_FAILURE); 262 } 263 } 264 265 return (KERN_SUCCESS); 266} 267 268static int 269__elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 270 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) 271{ 272 struct sf_buf *sf; 273 vm_offset_t off; 274 vm_size_t sz; 275 int error, rv; 276 277 if (start != trunc_page(start)) { 278 rv = __elfN(map_partial)(map, object, offset, start, 279 round_page(start), prot); 280 if (rv) 281 return (rv); 282 offset += round_page(start) - start; 283 start = round_page(start); 284 } 285 if (end != round_page(end)) { 286 rv = __elfN(map_partial)(map, object, offset + 287 trunc_page(end) - start, trunc_page(end), end, prot); 288 if (rv) 289 return (rv); 290 end = trunc_page(end); 291 } 292 if (end > start) { 293 if (offset & PAGE_MASK) { 294 /* 295 * The mapping is not page aligned. This means we have 296 * to copy the data. Sigh. 297 */ 298 rv = vm_map_find(map, NULL, 0, &start, end - start, 299 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0); 300 if (rv) 301 return (rv); 302 if (object == NULL) 303 return (KERN_SUCCESS); 304 for (; start < end; start += sz) { 305 sf = vm_imgact_map_page(object, offset); 306 if (sf == NULL) 307 return (KERN_FAILURE); 308 off = offset - trunc_page(offset); 309 sz = end - start; 310 if (sz > PAGE_SIZE - off) 311 sz = PAGE_SIZE - off; 312 error = copyout((caddr_t)sf_buf_kva(sf) + off, 313 (caddr_t)start, sz); 314 vm_imgact_unmap_page(sf); 315 if (error) { 316 return (KERN_FAILURE); 317 } 318 offset += sz; 319 } 320 rv = KERN_SUCCESS; 321 } else { 322 vm_object_reference(object); 323 vm_map_lock(map); 324 rv = vm_map_insert(map, object, offset, start, end, 325 prot, VM_PROT_ALL, cow); 326 vm_map_unlock(map); 327 if (rv != KERN_SUCCESS) 328 vm_object_deallocate(object); 329 } 330 return (rv); 331 } else { 332 return (KERN_SUCCESS); 333 } 334} 335 336static int 337__elfN(load_section)(struct vmspace *vmspace, 338 vm_object_t object, vm_offset_t offset, 339 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 340 size_t pagesize) 341{ 342 struct sf_buf *sf; 343 size_t map_len; 344 vm_offset_t map_addr; 345 int error, rv, cow; 346 size_t copy_len; 347 vm_offset_t file_addr; 348 349 /* 350 * It's necessary to fail if the filsz + offset taken from the 351 * header is greater than the actual file pager object's size. 352 * If we were to allow this, then the vm_map_find() below would 353 * walk right off the end of the file object and into the ether. 354 * 355 * While I'm here, might as well check for something else that 356 * is invalid: filsz cannot be greater than memsz. 357 */ 358 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 359 filsz > memsz) { 360 uprintf("elf_load_section: truncated ELF file\n"); 361 return (ENOEXEC); 362 } 363 364 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 365 file_addr = trunc_page_ps(offset, pagesize); 366 367 /* 368 * We have two choices. We can either clear the data in the last page 369 * of an oversized mapping, or we can start the anon mapping a page 370 * early and copy the initialized data into that first page. We 371 * choose the second.. 372 */ 373 if (memsz > filsz) 374 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 375 else 376 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 377 378 if (map_len != 0) { 379 /* cow flags: don't dump readonly sections in core */ 380 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 381 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 382 383 rv = __elfN(map_insert)(&vmspace->vm_map, 384 object, 385 file_addr, /* file offset */ 386 map_addr, /* virtual start */ 387 map_addr + map_len,/* virtual end */ 388 prot, 389 cow); 390 if (rv != KERN_SUCCESS) 391 return (EINVAL); 392 393 /* we can stop now if we've covered it all */ 394 if (memsz == filsz) { 395 return (0); 396 } 397 } 398 399 400 /* 401 * We have to get the remaining bit of the file into the first part 402 * of the oversized map segment. This is normally because the .data 403 * segment in the file is extended to provide bss. It's a neat idea 404 * to try and save a page, but it's a pain in the behind to implement. 405 */ 406 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 407 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 408 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 409 map_addr; 410 411 /* This had damn well better be true! */ 412 if (map_len != 0) { 413 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr, 414 map_addr + map_len, VM_PROT_ALL, 0); 415 if (rv != KERN_SUCCESS) { 416 return (EINVAL); 417 } 418 } 419 420 if (copy_len != 0) { 421 vm_offset_t off; 422 423 sf = vm_imgact_map_page(object, offset + filsz); 424 if (sf == NULL) 425 return (EIO); 426 427 /* send the page fragment to user space */ 428 off = trunc_page_ps(offset + filsz, pagesize) - 429 trunc_page(offset + filsz); 430 error = copyout((caddr_t)sf_buf_kva(sf) + off, 431 (caddr_t)map_addr, copy_len); 432 vm_imgact_unmap_page(sf); 433 if (error) { 434 return (error); 435 } 436 } 437 438 /* 439 * set it to the specified protection. 440 * XXX had better undo the damage from pasting over the cracks here! 441 */ 442 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr), 443 round_page(map_addr + map_len), prot, FALSE); 444 445 return (0); 446} 447 448/* 449 * Load the file "file" into memory. It may be either a shared object 450 * or an executable. 451 * 452 * The "addr" reference parameter is in/out. On entry, it specifies 453 * the address where a shared object should be loaded. If the file is 454 * an executable, this value is ignored. On exit, "addr" specifies 455 * where the file was actually loaded. 456 * 457 * The "entry" reference parameter is out only. On exit, it specifies 458 * the entry point for the loaded file. 459 */ 460static int 461__elfN(load_file)(struct proc *p, const char *file, u_long *addr, 462 u_long *entry, size_t pagesize) 463{ 464 struct { 465 struct nameidata nd; 466 struct vattr attr; 467 struct image_params image_params; 468 } *tempdata; 469 const Elf_Ehdr *hdr = NULL; 470 const Elf_Phdr *phdr = NULL; 471 struct nameidata *nd; 472 struct vmspace *vmspace = p->p_vmspace; 473 struct vattr *attr; 474 struct image_params *imgp; 475 vm_prot_t prot; 476 u_long rbase; 477 u_long base_addr = 0; 478 int vfslocked, error, i, numsegs; 479 480 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 481 nd = &tempdata->nd; 482 attr = &tempdata->attr; 483 imgp = &tempdata->image_params; 484 485 /* 486 * Initialize part of the common data 487 */ 488 imgp->proc = p; 489 imgp->attr = attr; 490 imgp->firstpage = NULL; 491 imgp->image_header = NULL; 492 imgp->object = NULL; 493 imgp->execlabel = NULL; 494 495 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, 496 curthread); 497 vfslocked = 0; 498 if ((error = namei(nd)) != 0) { 499 nd->ni_vp = NULL; 500 goto fail; 501 } 502 vfslocked = NDHASGIANT(nd); 503 NDFREE(nd, NDF_ONLY_PNBUF); 504 imgp->vp = nd->ni_vp; 505 506 /* 507 * Check permissions, modes, uid, etc on the file, and "open" it. 508 */ 509 error = exec_check_permissions(imgp); 510 if (error) 511 goto fail; 512 513 error = exec_map_first_page(imgp); 514 if (error) 515 goto fail; 516 517 /* 518 * Also make certain that the interpreter stays the same, so set 519 * its VV_TEXT flag, too. 520 */ 521 nd->ni_vp->v_vflag |= VV_TEXT; 522 523 imgp->object = nd->ni_vp->v_object; 524 525 hdr = (const Elf_Ehdr *)imgp->image_header; 526 if ((error = __elfN(check_header)(hdr)) != 0) 527 goto fail; 528 if (hdr->e_type == ET_DYN) 529 rbase = *addr; 530 else if (hdr->e_type == ET_EXEC) 531 rbase = 0; 532 else { 533 error = ENOEXEC; 534 goto fail; 535 } 536 537 /* Only support headers that fit within first page for now */ 538 /* (multiplication of two Elf_Half fields will not overflow) */ 539 if ((hdr->e_phoff > PAGE_SIZE) || 540 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 541 error = ENOEXEC; 542 goto fail; 543 } 544 545 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 546 if (!aligned(phdr, Elf_Addr)) { 547 error = ENOEXEC; 548 goto fail; 549 } 550 551 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 552 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 553 prot = 0; 554 if (phdr[i].p_flags & PF_X) 555 prot |= VM_PROT_EXECUTE; 556 if (phdr[i].p_flags & PF_W) 557 prot |= VM_PROT_WRITE; 558 if (phdr[i].p_flags & PF_R) 559 prot |= VM_PROT_READ; 560 561 if ((error = __elfN(load_section)(vmspace, 562 imgp->object, phdr[i].p_offset, 563 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 564 phdr[i].p_memsz, phdr[i].p_filesz, prot, 565 pagesize)) != 0) 566 goto fail; 567 /* 568 * Establish the base address if this is the 569 * first segment. 570 */ 571 if (numsegs == 0) 572 base_addr = trunc_page(phdr[i].p_vaddr + 573 rbase); 574 numsegs++; 575 } 576 } 577 *addr = base_addr; 578 *entry = (unsigned long)hdr->e_entry + rbase; 579 580fail: 581 if (imgp->firstpage) 582 exec_unmap_first_page(imgp); 583 584 if (nd->ni_vp) 585 vput(nd->ni_vp); 586 587 VFS_UNLOCK_GIANT(vfslocked); 588 free(tempdata, M_TEMP); 589 590 return (error); 591} 592 593static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; 594 595static int 596__CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 597{ 598 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 599 const Elf_Phdr *phdr, *pnote = NULL; 600 Elf_Auxargs *elf_auxargs; 601 struct vmspace *vmspace; 602 vm_prot_t prot; 603 u_long text_size = 0, data_size = 0, total_size = 0; 604 u_long text_addr = 0, data_addr = 0; 605 u_long seg_size, seg_addr; 606 u_long addr, entry = 0, proghdr = 0; 607 int error = 0, i; 608 const char *interp = NULL, *newinterp = NULL; 609 Elf_Brandinfo *brand_info; 610 const Elf_Note *note, *note_end; 611 char *path; 612 const char *note_name; 613 struct sysentvec *sv; 614 615 /* 616 * Do we have a valid ELF header ? 617 * 618 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 619 * if particular brand doesn't support it. 620 */ 621 if (__elfN(check_header)(hdr) != 0 || 622 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 623 return (-1); 624 625 /* 626 * From here on down, we return an errno, not -1, as we've 627 * detected an ELF file. 628 */ 629 630 if ((hdr->e_phoff > PAGE_SIZE) || 631 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 632 /* Only support headers in first page for now */ 633 return (ENOEXEC); 634 } 635 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 636 if (!aligned(phdr, Elf_Addr)) 637 return (ENOEXEC); 638 for (i = 0; i < hdr->e_phnum; i++) { 639 if (phdr[i].p_type == PT_INTERP) { 640 /* Path to interpreter */ 641 if (phdr[i].p_filesz > MAXPATHLEN || 642 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) 643 return (ENOEXEC); 644 interp = imgp->image_header + phdr[i].p_offset; 645 break; 646 } 647 } 648 649 brand_info = __elfN(get_brandinfo)(hdr, interp); 650 if (brand_info == NULL) { 651 uprintf("ELF binary type \"%u\" not known.\n", 652 hdr->e_ident[EI_OSABI]); 653 return (ENOEXEC); 654 } 655 if (hdr->e_type == ET_DYN && 656 (brand_info->flags & BI_CAN_EXEC_DYN) == 0) 657 return (ENOEXEC); 658 sv = brand_info->sysvec; 659 if (interp != NULL && brand_info->interp_newpath != NULL) 660 newinterp = brand_info->interp_newpath; 661 662 /* 663 * Avoid a possible deadlock if the current address space is destroyed 664 * and that address space maps the locked vnode. In the common case, 665 * the locked vnode's v_usecount is decremented but remains greater 666 * than zero. Consequently, the vnode lock is not needed by vrele(). 667 * However, in cases where the vnode lock is external, such as nullfs, 668 * v_usecount may become zero. 669 */ 670 VOP_UNLOCK(imgp->vp, 0); 671 672 error = exec_new_vmspace(imgp, sv); 673 imgp->proc->p_sysent = sv; 674 675 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 676 if (error) 677 return (error); 678 679 vmspace = imgp->proc->p_vmspace; 680 681 for (i = 0; i < hdr->e_phnum; i++) { 682 switch (phdr[i].p_type) { 683 case PT_LOAD: /* Loadable segment */ 684 prot = 0; 685 if (phdr[i].p_flags & PF_X) 686 prot |= VM_PROT_EXECUTE; 687 if (phdr[i].p_flags & PF_W) 688 prot |= VM_PROT_WRITE; 689 if (phdr[i].p_flags & PF_R) 690 prot |= VM_PROT_READ; 691 692#if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER) 693 /* 694 * Some x86 binaries assume read == executable, 695 * notably the M3 runtime and therefore cvsup 696 */ 697 if (prot & VM_PROT_READ) 698 prot |= VM_PROT_EXECUTE; 699#endif 700 701 if ((error = __elfN(load_section)(vmspace, 702 imgp->object, phdr[i].p_offset, 703 (caddr_t)(uintptr_t)phdr[i].p_vaddr, 704 phdr[i].p_memsz, phdr[i].p_filesz, prot, 705 sv->sv_pagesize)) != 0) 706 return (error); 707 708 /* 709 * If this segment contains the program headers, 710 * remember their virtual address for the AT_PHDR 711 * aux entry. Static binaries don't usually include 712 * a PT_PHDR entry. 713 */ 714 if (phdr[i].p_offset == 0 && 715 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 716 <= phdr[i].p_filesz) 717 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 718 719 seg_addr = trunc_page(phdr[i].p_vaddr); 720 seg_size = round_page(phdr[i].p_memsz + 721 phdr[i].p_vaddr - seg_addr); 722 723 /* 724 * Is this .text or .data? We can't use 725 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 726 * alpha terribly and possibly does other bad 727 * things so we stick to the old way of figuring 728 * it out: If the segment contains the program 729 * entry point, it's a text segment, otherwise it 730 * is a data segment. 731 * 732 * Note that obreak() assumes that data_addr + 733 * data_size == end of data load area, and the ELF 734 * file format expects segments to be sorted by 735 * address. If multiple data segments exist, the 736 * last one will be used. 737 */ 738 if (hdr->e_entry >= phdr[i].p_vaddr && 739 hdr->e_entry < (phdr[i].p_vaddr + 740 phdr[i].p_memsz)) { 741 text_size = seg_size; 742 text_addr = seg_addr; 743 entry = (u_long)hdr->e_entry; 744 } else { 745 data_size = seg_size; 746 data_addr = seg_addr; 747 } 748 total_size += seg_size; 749 break; 750 case PT_PHDR: /* Program header table info */ 751 proghdr = phdr[i].p_vaddr; 752 break; 753 case PT_NOTE: 754 pnote = &phdr[i]; 755 break; 756 default: 757 break; 758 } 759 } 760 761 if (data_addr == 0 && data_size == 0) { 762 data_addr = text_addr; 763 data_size = text_size; 764 } 765 766 /* 767 * Check limits. It should be safe to check the 768 * limits after loading the segments since we do 769 * not actually fault in all the segments pages. 770 */ 771 PROC_LOCK(imgp->proc); 772 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 773 text_size > maxtsiz || 774 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) { 775 PROC_UNLOCK(imgp->proc); 776 return (ENOMEM); 777 } 778 779 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 780 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 781 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 782 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 783 784 /* 785 * We load the dynamic linker where a userland call 786 * to mmap(0, ...) would put it. The rationale behind this 787 * calculation is that it leaves room for the heap to grow to 788 * its maximum allowed size. 789 */ 790 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr + 791 lim_max(imgp->proc, RLIMIT_DATA)); 792 PROC_UNLOCK(imgp->proc); 793 794 imgp->entry_addr = entry; 795 796 if (interp != NULL) { 797 int have_interp = FALSE; 798 VOP_UNLOCK(imgp->vp, 0); 799 if (brand_info->emul_path != NULL && 800 brand_info->emul_path[0] != '\0') { 801 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 802 snprintf(path, MAXPATHLEN, "%s%s", 803 brand_info->emul_path, interp); 804 error = __elfN(load_file)(imgp->proc, path, &addr, 805 &imgp->entry_addr, sv->sv_pagesize); 806 free(path, M_TEMP); 807 if (error == 0) 808 have_interp = TRUE; 809 } 810 if (!have_interp && newinterp != NULL) { 811 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 812 &imgp->entry_addr, sv->sv_pagesize); 813 if (error == 0) 814 have_interp = TRUE; 815 } 816 if (!have_interp) { 817 error = __elfN(load_file)(imgp->proc, interp, &addr, 818 &imgp->entry_addr, sv->sv_pagesize); 819 } 820 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 821 if (error != 0) { 822 uprintf("ELF interpreter %s not found\n", interp); 823 return (error); 824 } 825 } 826 827 /* 828 * Construct auxargs table (used by the fixup routine) 829 */ 830 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 831 elf_auxargs->execfd = -1; 832 elf_auxargs->phdr = proghdr; 833 elf_auxargs->phent = hdr->e_phentsize; 834 elf_auxargs->phnum = hdr->e_phnum; 835 elf_auxargs->pagesz = PAGE_SIZE; 836 elf_auxargs->base = addr; 837 elf_auxargs->flags = 0; 838 elf_auxargs->entry = entry; 839 840 imgp->auxargs = elf_auxargs; 841 imgp->interpreted = 0; 842 843 /* 844 * Try to fetch the osreldate for FreeBSD binary from the ELF 845 * OSABI-note. Only the first page of the image is searched, 846 * the same as for headers. 847 */ 848 if (pnote != NULL && pnote->p_offset < PAGE_SIZE && 849 pnote->p_offset + pnote->p_filesz < PAGE_SIZE ) { 850 note = (const Elf_Note *)(imgp->image_header + pnote->p_offset); 851 if (!aligned(note, Elf32_Addr)) { 852 free(imgp->auxargs, M_TEMP); 853 imgp->auxargs = NULL; 854 return (ENOEXEC); 855 } 856 note_end = (const Elf_Note *)(imgp->image_header + pnote->p_offset + 857 pnote->p_filesz); 858 while (note < note_end) { 859 if (note->n_namesz == sizeof(FREEBSD_ABI_VENDOR) && 860 note->n_descsz == sizeof(int32_t) && 861 note->n_type == 1 /* ABI_NOTETYPE */) { 862 note_name = (const char *)(note + 1); 863 if (strncmp(FREEBSD_ABI_VENDOR, note_name, 864 sizeof(FREEBSD_ABI_VENDOR)) == 0) { 865 imgp->proc->p_osrel = *(const int32_t *) 866 (note_name + 867 round_page_ps(sizeof(FREEBSD_ABI_VENDOR), 868 sizeof(Elf32_Addr))); 869 break; 870 } 871 } 872 note = (const Elf_Note *)((const char *)(note + 1) + 873 round_page_ps(note->n_namesz, sizeof(Elf32_Addr)) + 874 round_page_ps(note->n_descsz, sizeof(Elf32_Addr))); 875 } 876 } 877 878 return (error); 879} 880 881#define suword __CONCAT(suword, __ELF_WORD_SIZE) 882 883int 884__elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 885{ 886 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 887 Elf_Addr *base; 888 Elf_Addr *pos; 889 890 base = (Elf_Addr *)*stack_base; 891 pos = base + (imgp->args->argc + imgp->args->envc + 2); 892 893 if (args->execfd != -1) 894 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 895 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 896 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 897 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 898 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 899 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 900 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 901 AUXARGS_ENTRY(pos, AT_BASE, args->base); 902 AUXARGS_ENTRY(pos, AT_NULL, 0); 903 904 free(imgp->auxargs, M_TEMP); 905 imgp->auxargs = NULL; 906 907 base--; 908 suword(base, (long)imgp->args->argc); 909 *stack_base = (register_t *)base; 910 return (0); 911} 912 913/* 914 * Code for generating ELF core dumps. 915 */ 916 917typedef void (*segment_callback)(vm_map_entry_t, void *); 918 919/* Closure for cb_put_phdr(). */ 920struct phdr_closure { 921 Elf_Phdr *phdr; /* Program header to fill in */ 922 Elf_Off offset; /* Offset of segment in core file */ 923}; 924 925/* Closure for cb_size_segment(). */ 926struct sseg_closure { 927 int count; /* Count of writable segments. */ 928 size_t size; /* Total size of all writable segments. */ 929}; 930 931static void cb_put_phdr(vm_map_entry_t, void *); 932static void cb_size_segment(vm_map_entry_t, void *); 933static void each_writable_segment(struct thread *, segment_callback, void *); 934static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 935 int, void *, size_t); 936static void __elfN(puthdr)(struct thread *, void *, size_t *, int); 937static void __elfN(putnote)(void *, size_t *, const char *, int, 938 const void *, size_t); 939 940int 941__elfN(coredump)(td, vp, limit) 942 struct thread *td; 943 struct vnode *vp; 944 off_t limit; 945{ 946 struct ucred *cred = td->td_ucred; 947 int error = 0; 948 struct sseg_closure seginfo; 949 void *hdr; 950 size_t hdrsize; 951 952 /* Size the program segments. */ 953 seginfo.count = 0; 954 seginfo.size = 0; 955 each_writable_segment(td, cb_size_segment, &seginfo); 956 957 /* 958 * Calculate the size of the core file header area by making 959 * a dry run of generating it. Nothing is written, but the 960 * size is calculated. 961 */ 962 hdrsize = 0; 963 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count); 964 965 if (hdrsize + seginfo.size >= limit) 966 return (EFAULT); 967 968 /* 969 * Allocate memory for building the header, fill it up, 970 * and write it out. 971 */ 972 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 973 if (hdr == NULL) { 974 return (EINVAL); 975 } 976 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize); 977 978 /* Write the contents of all of the writable segments. */ 979 if (error == 0) { 980 Elf_Phdr *php; 981 off_t offset; 982 int i; 983 984 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 985 offset = hdrsize; 986 for (i = 0; i < seginfo.count; i++) { 987 error = vn_rdwr_inchunks(UIO_WRITE, vp, 988 (caddr_t)(uintptr_t)php->p_vaddr, 989 php->p_filesz, offset, UIO_USERSPACE, 990 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 991 curthread); 992 if (error != 0) 993 break; 994 offset += php->p_filesz; 995 php++; 996 } 997 } 998 free(hdr, M_TEMP); 999 1000 return (error); 1001} 1002 1003/* 1004 * A callback for each_writable_segment() to write out the segment's 1005 * program header entry. 1006 */ 1007static void 1008cb_put_phdr(entry, closure) 1009 vm_map_entry_t entry; 1010 void *closure; 1011{ 1012 struct phdr_closure *phc = (struct phdr_closure *)closure; 1013 Elf_Phdr *phdr = phc->phdr; 1014 1015 phc->offset = round_page(phc->offset); 1016 1017 phdr->p_type = PT_LOAD; 1018 phdr->p_offset = phc->offset; 1019 phdr->p_vaddr = entry->start; 1020 phdr->p_paddr = 0; 1021 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1022 phdr->p_align = PAGE_SIZE; 1023 phdr->p_flags = 0; 1024 if (entry->protection & VM_PROT_READ) 1025 phdr->p_flags |= PF_R; 1026 if (entry->protection & VM_PROT_WRITE) 1027 phdr->p_flags |= PF_W; 1028 if (entry->protection & VM_PROT_EXECUTE) 1029 phdr->p_flags |= PF_X; 1030 1031 phc->offset += phdr->p_filesz; 1032 phc->phdr++; 1033} 1034 1035/* 1036 * A callback for each_writable_segment() to gather information about 1037 * the number of segments and their total size. 1038 */ 1039static void 1040cb_size_segment(entry, closure) 1041 vm_map_entry_t entry; 1042 void *closure; 1043{ 1044 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1045 1046 ssc->count++; 1047 ssc->size += entry->end - entry->start; 1048} 1049 1050/* 1051 * For each writable segment in the process's memory map, call the given 1052 * function with a pointer to the map entry and some arbitrary 1053 * caller-supplied data. 1054 */ 1055static void 1056each_writable_segment(td, func, closure) 1057 struct thread *td; 1058 segment_callback func; 1059 void *closure; 1060{ 1061 struct proc *p = td->td_proc; 1062 vm_map_t map = &p->p_vmspace->vm_map; 1063 vm_map_entry_t entry; 1064 vm_object_t backing_object, object; 1065 boolean_t ignore_entry; 1066 1067 vm_map_lock_read(map); 1068 for (entry = map->header.next; entry != &map->header; 1069 entry = entry->next) { 1070 /* 1071 * Don't dump inaccessible mappings, deal with legacy 1072 * coredump mode. 1073 * 1074 * Note that read-only segments related to the elf binary 1075 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1076 * need to arbitrarily ignore such segments. 1077 */ 1078 if (elf_legacy_coredump) { 1079 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1080 continue; 1081 } else { 1082 if ((entry->protection & VM_PROT_ALL) == 0) 1083 continue; 1084 } 1085 1086 /* 1087 * Dont include memory segment in the coredump if 1088 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1089 * madvise(2). Do not dump submaps (i.e. parts of the 1090 * kernel map). 1091 */ 1092 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1093 continue; 1094 1095 if ((object = entry->object.vm_object) == NULL) 1096 continue; 1097 1098 /* Ignore memory-mapped devices and such things. */ 1099 VM_OBJECT_LOCK(object); 1100 while ((backing_object = object->backing_object) != NULL) { 1101 VM_OBJECT_LOCK(backing_object); 1102 VM_OBJECT_UNLOCK(object); 1103 object = backing_object; 1104 } 1105 ignore_entry = object->type != OBJT_DEFAULT && 1106 object->type != OBJT_SWAP && object->type != OBJT_VNODE; 1107 VM_OBJECT_UNLOCK(object); 1108 if (ignore_entry) 1109 continue; 1110 1111 (*func)(entry, closure); 1112 } 1113 vm_map_unlock_read(map); 1114} 1115 1116/* 1117 * Write the core file header to the file, including padding up to 1118 * the page boundary. 1119 */ 1120static int 1121__elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize) 1122 struct thread *td; 1123 struct vnode *vp; 1124 struct ucred *cred; 1125 int numsegs; 1126 size_t hdrsize; 1127 void *hdr; 1128{ 1129 size_t off; 1130 1131 /* Fill in the header. */ 1132 bzero(hdr, hdrsize); 1133 off = 0; 1134 __elfN(puthdr)(td, hdr, &off, numsegs); 1135 1136 /* Write it to the core file. */ 1137 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1138 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1139 td)); 1140} 1141 1142#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1143typedef struct prstatus32 elf_prstatus_t; 1144typedef struct prpsinfo32 elf_prpsinfo_t; 1145typedef struct fpreg32 elf_prfpregset_t; 1146typedef struct fpreg32 elf_fpregset_t; 1147typedef struct reg32 elf_gregset_t; 1148#else 1149typedef prstatus_t elf_prstatus_t; 1150typedef prpsinfo_t elf_prpsinfo_t; 1151typedef prfpregset_t elf_prfpregset_t; 1152typedef prfpregset_t elf_fpregset_t; 1153typedef gregset_t elf_gregset_t; 1154#endif 1155 1156static void 1157__elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs) 1158{ 1159 struct { 1160 elf_prstatus_t status; 1161 elf_prfpregset_t fpregset; 1162 elf_prpsinfo_t psinfo; 1163 } *tempdata; 1164 elf_prstatus_t *status; 1165 elf_prfpregset_t *fpregset; 1166 elf_prpsinfo_t *psinfo; 1167 struct proc *p; 1168 struct thread *thr; 1169 size_t ehoff, noteoff, notesz, phoff; 1170 1171 p = td->td_proc; 1172 1173 ehoff = *off; 1174 *off += sizeof(Elf_Ehdr); 1175 1176 phoff = *off; 1177 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1178 1179 noteoff = *off; 1180 /* 1181 * Don't allocate space for the notes if we're just calculating 1182 * the size of the header. We also don't collect the data. 1183 */ 1184 if (dst != NULL) { 1185 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK); 1186 status = &tempdata->status; 1187 fpregset = &tempdata->fpregset; 1188 psinfo = &tempdata->psinfo; 1189 } else { 1190 tempdata = NULL; 1191 status = NULL; 1192 fpregset = NULL; 1193 psinfo = NULL; 1194 } 1195 1196 if (dst != NULL) { 1197 psinfo->pr_version = PRPSINFO_VERSION; 1198 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1199 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1200 /* 1201 * XXX - We don't fill in the command line arguments properly 1202 * yet. 1203 */ 1204 strlcpy(psinfo->pr_psargs, p->p_comm, 1205 sizeof(psinfo->pr_psargs)); 1206 } 1207 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1208 sizeof *psinfo); 1209 1210 /* 1211 * To have the debugger select the right thread (LWP) as the initial 1212 * thread, we dump the state of the thread passed to us in td first. 1213 * This is the thread that causes the core dump and thus likely to 1214 * be the right thread one wants to have selected in the debugger. 1215 */ 1216 thr = td; 1217 while (thr != NULL) { 1218 if (dst != NULL) { 1219 status->pr_version = PRSTATUS_VERSION; 1220 status->pr_statussz = sizeof(elf_prstatus_t); 1221 status->pr_gregsetsz = sizeof(elf_gregset_t); 1222 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1223 status->pr_osreldate = osreldate; 1224 status->pr_cursig = p->p_sig; 1225 status->pr_pid = thr->td_tid; 1226#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1227 fill_regs32(thr, &status->pr_reg); 1228 fill_fpregs32(thr, fpregset); 1229#else 1230 fill_regs(thr, &status->pr_reg); 1231 fill_fpregs(thr, fpregset); 1232#endif 1233 } 1234 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1235 sizeof *status); 1236 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1237 sizeof *fpregset); 1238 /* 1239 * Allow for MD specific notes, as well as any MD 1240 * specific preparations for writing MI notes. 1241 */ 1242 __elfN(dump_thread)(thr, dst, off); 1243 1244 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1245 TAILQ_NEXT(thr, td_plist); 1246 if (thr == td) 1247 thr = TAILQ_NEXT(thr, td_plist); 1248 } 1249 1250 notesz = *off - noteoff; 1251 1252 if (dst != NULL) 1253 free(tempdata, M_TEMP); 1254 1255 /* Align up to a page boundary for the program segments. */ 1256 *off = round_page(*off); 1257 1258 if (dst != NULL) { 1259 Elf_Ehdr *ehdr; 1260 Elf_Phdr *phdr; 1261 struct phdr_closure phc; 1262 1263 /* 1264 * Fill in the ELF header. 1265 */ 1266 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1267 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1268 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1269 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1270 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1271 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1272 ehdr->e_ident[EI_DATA] = ELF_DATA; 1273 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1274 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1275 ehdr->e_ident[EI_ABIVERSION] = 0; 1276 ehdr->e_ident[EI_PAD] = 0; 1277 ehdr->e_type = ET_CORE; 1278#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1279 ehdr->e_machine = EM_386; 1280#else 1281 ehdr->e_machine = ELF_ARCH; 1282#endif 1283 ehdr->e_version = EV_CURRENT; 1284 ehdr->e_entry = 0; 1285 ehdr->e_phoff = phoff; 1286 ehdr->e_flags = 0; 1287 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1288 ehdr->e_phentsize = sizeof(Elf_Phdr); 1289 ehdr->e_phnum = numsegs + 1; 1290 ehdr->e_shentsize = sizeof(Elf_Shdr); 1291 ehdr->e_shnum = 0; 1292 ehdr->e_shstrndx = SHN_UNDEF; 1293 1294 /* 1295 * Fill in the program header entries. 1296 */ 1297 phdr = (Elf_Phdr *)((char *)dst + phoff); 1298 1299 /* The note segement. */ 1300 phdr->p_type = PT_NOTE; 1301 phdr->p_offset = noteoff; 1302 phdr->p_vaddr = 0; 1303 phdr->p_paddr = 0; 1304 phdr->p_filesz = notesz; 1305 phdr->p_memsz = 0; 1306 phdr->p_flags = 0; 1307 phdr->p_align = 0; 1308 phdr++; 1309 1310 /* All the writable segments from the program. */ 1311 phc.phdr = phdr; 1312 phc.offset = *off; 1313 each_writable_segment(td, cb_put_phdr, &phc); 1314 } 1315} 1316 1317static void 1318__elfN(putnote)(void *dst, size_t *off, const char *name, int type, 1319 const void *desc, size_t descsz) 1320{ 1321 Elf_Note note; 1322 1323 note.n_namesz = strlen(name) + 1; 1324 note.n_descsz = descsz; 1325 note.n_type = type; 1326 if (dst != NULL) 1327 bcopy(¬e, (char *)dst + *off, sizeof note); 1328 *off += sizeof note; 1329 if (dst != NULL) 1330 bcopy(name, (char *)dst + *off, note.n_namesz); 1331 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1332 if (dst != NULL) 1333 bcopy(desc, (char *)dst + *off, note.n_descsz); 1334 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1335} 1336 1337/* 1338 * Tell kern_execve.c about it, with a little help from the linker. 1339 */ 1340static struct execsw __elfN(execsw) = { 1341 __CONCAT(exec_, __elfN(imgact)), 1342 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 1343}; 1344EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 1345