1/* $OpenBSD: exec_elf.c,v 1.186 2024/04/02 08:39:16 deraadt Exp $ */ 2 3/* 4 * Copyright (c) 1996 Per Fogelstrom 5 * All rights reserved. 6 * 7 * Copyright (c) 1994 Christos Zoulas 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. The name of the author may not be used to endorse or promote products 19 * derived from this software without specific prior written permission 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 22 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 23 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 24 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 26 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 30 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 31 * 32 */ 33 34/* 35 * Copyright (c) 2001 Wasabi Systems, Inc. 36 * All rights reserved. 37 * 38 * Written by Jason R. Thorpe for Wasabi Systems, Inc. 39 * 40 * Redistribution and use in source and binary forms, with or without 41 * modification, are permitted provided that the following conditions 42 * are met: 43 * 1. Redistributions of source code must retain the above copyright 44 * notice, this list of conditions and the following disclaimer. 45 * 2. Redistributions in binary form must reproduce the above copyright 46 * notice, this list of conditions and the following disclaimer in the 47 * documentation and/or other materials provided with the distribution. 48 * 3. All advertising materials mentioning features or use of this software 49 * must display the following acknowledgement: 50 * This product includes software developed for the NetBSD Project by 51 * Wasabi Systems, Inc. 52 * 4. The name of Wasabi Systems, Inc. may not be used to endorse 53 * or promote products derived from this software without specific prior 54 * written permission. 55 * 56 * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND 57 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 58 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 59 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WASABI SYSTEMS, INC 60 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 61 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 62 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 63 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 64 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 65 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 66 * POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69#include <sys/param.h> 70#include <sys/systm.h> 71#include <sys/proc.h> 72#include <sys/malloc.h> 73#include <sys/pool.h> 74#include <sys/mount.h> 75#include <sys/namei.h> 76#include <sys/vnode.h> 77#include <sys/core.h> 78#include <sys/exec.h> 79#include <sys/exec_elf.h> 80#include <sys/fcntl.h> 81#include <sys/ptrace.h> 82#include <sys/signalvar.h> 83#include <sys/pledge.h> 84#include <sys/syscall.h> 85 86#include <sys/mman.h> 87 88#include <uvm/uvm_extern.h> 89 90#include <machine/reg.h> 91#include <machine/exec.h> 92 93int elf_load_file(struct proc *, char *, struct exec_package *, 94 struct elf_args *); 95int elf_check_header(Elf_Ehdr *); 96int elf_read_from(struct proc *, struct vnode *, u_long, void *, int); 97void elf_load_psection(struct exec_vmcmd_set *, struct vnode *, 98 Elf_Phdr *, Elf_Addr *, Elf_Addr *, int *, int); 99int elf_os_pt_note_name(Elf_Note *); 100int elf_os_pt_note(struct proc *, struct exec_package *, Elf_Ehdr *, int *); 101int elf_read_pintable(struct proc *p, struct vnode *vp, Elf_Phdr *pp, 102 u_int **pinp, int is_ldso, size_t len); 103 104/* round up and down to page boundaries. */ 105#define ELF_ROUND(a, b) (((a) + (b) - 1) & ~((b) - 1)) 106#define ELF_TRUNC(a, b) ((a) & ~((b) - 1)) 107 108/* 109 * We limit the number of program headers to 32, this should 110 * be a reasonable limit for ELF, the most we have seen so far is 12 111 */ 112#define ELF_MAX_VALID_PHDR 32 113 114#define ELF_NOTE_NAME_OPENBSD 0x01 115 116struct elf_note_name { 117 char *name; 118 int id; 119} elf_note_names[] = { 120 { "OpenBSD", ELF_NOTE_NAME_OPENBSD }, 121}; 122 123#define ELFROUNDSIZE sizeof(Elf_Word) 124#define elfround(x) roundup((x), ELFROUNDSIZE) 125 126 127/* 128 * Check header for validity; return 0 for ok, ENOEXEC if error 129 */ 130int 131elf_check_header(Elf_Ehdr *ehdr) 132{ 133 /* 134 * We need to check magic, class size, endianness, and version before 135 * we look at the rest of the Elf_Ehdr structure. These few elements 136 * are represented in a machine independent fashion. 137 */ 138 if (!IS_ELF(*ehdr) || 139 ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 140 ehdr->e_ident[EI_DATA] != ELF_TARG_DATA || 141 ehdr->e_ident[EI_VERSION] != ELF_TARG_VER) 142 return (ENOEXEC); 143 144 /* Now check the machine dependent header */ 145 if (ehdr->e_machine != ELF_TARG_MACH || 146 ehdr->e_version != ELF_TARG_VER) 147 return (ENOEXEC); 148 149 /* Don't allow an insane amount of sections. */ 150 if (ehdr->e_phnum > ELF_MAX_VALID_PHDR) 151 return (ENOEXEC); 152 153 return (0); 154} 155 156/* 157 * Load a psection at the appropriate address 158 */ 159void 160elf_load_psection(struct exec_vmcmd_set *vcset, struct vnode *vp, 161 Elf_Phdr *ph, Elf_Addr *addr, Elf_Addr *size, int *prot, int flags) 162{ 163 u_long msize, lsize, psize, rm, rf; 164 long diff, offset, bdiff; 165 Elf_Addr base; 166 167 /* 168 * If the user specified an address, then we load there. 169 */ 170 if (*addr != ELF_NO_ADDR) { 171 if (ph->p_align > 1) { 172 *addr = ELF_TRUNC(*addr, ph->p_align); 173 diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align); 174 /* page align vaddr */ 175 base = *addr + trunc_page(ph->p_vaddr) 176 - ELF_TRUNC(ph->p_vaddr, ph->p_align); 177 } else { 178 diff = 0; 179 base = *addr + trunc_page(ph->p_vaddr) - ph->p_vaddr; 180 } 181 } else { 182 *addr = ph->p_vaddr; 183 if (ph->p_align > 1) 184 *addr = ELF_TRUNC(*addr, ph->p_align); 185 base = trunc_page(ph->p_vaddr); 186 diff = ph->p_vaddr - *addr; 187 } 188 bdiff = ph->p_vaddr - trunc_page(ph->p_vaddr); 189 190 /* 191 * Enforce W^X and map W|X segments without X permission 192 * initially. The dynamic linker will make these read-only 193 * and add back X permission after relocation processing. 194 * Static executables with W|X segments will probably crash. 195 */ 196 *prot |= (ph->p_flags & PF_R) ? PROT_READ : 0; 197 *prot |= (ph->p_flags & PF_W) ? PROT_WRITE : 0; 198 if ((ph->p_flags & PF_W) == 0) 199 *prot |= (ph->p_flags & PF_X) ? PROT_EXEC : 0; 200 201 /* 202 * Apply immutability as much as possible, but not text/rodata 203 * segments of textrel binaries, or RELRO or PT_OPENBSD_MUTABLE 204 * sections, or LOADS marked PF_OPENBSD_MUTABLE, or LOADS which 205 * violate W^X. 206 * Userland (meaning crt0 or ld.so) will repair those regions. 207 */ 208 if ((ph->p_flags & (PF_X | PF_W)) != (PF_X | PF_W) && 209 ((ph->p_flags & PF_OPENBSD_MUTABLE) == 0)) 210 flags |= VMCMD_IMMUTABLE; 211 if ((flags & VMCMD_TEXTREL) && (ph->p_flags & PF_W) == 0) 212 flags &= ~VMCMD_IMMUTABLE; 213 214 msize = ph->p_memsz + diff; 215 offset = ph->p_offset - bdiff; 216 lsize = ph->p_filesz + bdiff; 217 psize = round_page(lsize); 218 219 /* 220 * Because the pagedvn pager can't handle zero fill of the last 221 * data page if it's not page aligned we map the last page readvn. 222 */ 223 if (ph->p_flags & PF_W) { 224 psize = trunc_page(lsize); 225 if (psize > 0) 226 NEW_VMCMD2(vcset, vmcmd_map_pagedvn, psize, base, vp, 227 offset, *prot, flags); 228 if (psize != lsize) { 229 NEW_VMCMD2(vcset, vmcmd_map_readvn, lsize - psize, 230 base + psize, vp, offset + psize, *prot, flags); 231 } 232 } else { 233 NEW_VMCMD2(vcset, vmcmd_map_pagedvn, psize, base, vp, offset, 234 *prot, flags); 235 } 236 237 /* 238 * Check if we need to extend the size of the segment 239 */ 240 rm = round_page(*addr + ph->p_memsz + diff); 241 rf = round_page(*addr + ph->p_filesz + diff); 242 243 if (rm != rf) { 244 NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP, 0, 245 *prot, flags); 246 } 247 *size = msize; 248} 249 250/* 251 * Read from vnode into buffer at offset. 252 */ 253int 254elf_read_from(struct proc *p, struct vnode *vp, u_long off, void *buf, 255 int size) 256{ 257 int error; 258 size_t resid; 259 260 if ((error = vn_rdwr(UIO_READ, vp, buf, size, off, UIO_SYSSPACE, 261 0, p->p_ucred, &resid, p)) != 0) 262 return error; 263 /* 264 * See if we got all of it 265 */ 266 if (resid != 0) 267 return (ENOEXEC); 268 return (0); 269} 270 271/* 272 * rebase the pin offsets inside a base,len window for the text segment only. 273 */ 274void 275elf_adjustpins(vaddr_t *basep, size_t *lenp, u_int *pins, int npins, u_int offset) 276{ 277 int i; 278 279 /* Adjust offsets, base, len */ 280 for (i = 0; i < npins; i++) { 281 if (pins[i] == -1 || pins[i] == 0) 282 continue; 283 pins[i] -= offset; 284 } 285 *basep += offset; 286 *lenp -= offset; 287} 288 289int 290elf_read_pintable(struct proc *p, struct vnode *vp, Elf_Phdr *pp, 291 u_int **pinp, int is_ldso, size_t len) 292{ 293 struct pinsyscalls { 294 u_int offset; 295 u_int sysno; 296 } *syscalls = NULL; 297 int i, nsyscalls = 0, npins = 0; 298 u_int *pins = NULL; 299 300 if (pp->p_filesz > SYS_MAXSYSCALL * 2 * sizeof(*syscalls) || 301 pp->p_filesz % sizeof(*syscalls) != 0) 302 goto bad; 303 nsyscalls = pp->p_filesz / sizeof(*syscalls); 304 syscalls = malloc(pp->p_filesz, M_PINSYSCALL, M_WAITOK); 305 if (elf_read_from(p, vp, pp->p_offset, syscalls, 306 pp->p_filesz) != 0) 307 goto bad; 308 309 /* Validate, and calculate pintable size */ 310 for (i = 0; i < nsyscalls; i++) { 311 if (syscalls[i].sysno <= 0 || 312 syscalls[i].sysno >= SYS_MAXSYSCALL || 313 syscalls[i].offset > len) 314 goto bad; 315 npins = MAX(npins, syscalls[i].sysno); 316 } 317 if (is_ldso) 318 npins = MAX(npins, SYS_kbind); /* XXX see ld.so/loader.c */ 319 npins++; 320 321 /* Fill pintable: 0 = invalid, -1 = allowed, else offset from base */ 322 pins = mallocarray(npins, sizeof(u_int), M_PINSYSCALL, M_WAITOK|M_ZERO); 323 for (i = 0; i < nsyscalls; i++) { 324 if (pins[syscalls[i].sysno]) 325 pins[syscalls[i].sysno] = -1; /* duplicated */ 326 else 327 pins[syscalls[i].sysno] = syscalls[i].offset; 328 } 329 if (is_ldso) 330 pins[SYS_kbind] = -1; /* XXX see ld.so/loader.c */ 331 *pinp = pins; 332 pins = NULL; 333bad: 334 free(syscalls, M_PINSYSCALL, nsyscalls * sizeof(*syscalls)); 335 free(pins, M_PINSYSCALL, npins * sizeof(u_int)); 336 return npins; 337} 338 339/* 340 * Load a file (interpreter/library) pointed to by path [stolen from 341 * coff_load_shlib()]. Made slightly generic so it might be used externally. 342 */ 343int 344elf_load_file(struct proc *p, char *path, struct exec_package *epp, 345 struct elf_args *ap) 346{ 347 int error, i; 348 struct nameidata nd; 349 Elf_Ehdr eh; 350 Elf_Phdr *ph = NULL, *syscall_ph = NULL; 351 u_long phsize = 0; 352 Elf_Addr addr; 353 struct vnode *vp; 354 Elf_Phdr *base_ph = NULL; 355 struct interp_ld_sec { 356 Elf_Addr vaddr; 357 u_long memsz; 358 } loadmap[ELF_MAX_VALID_PHDR]; 359 int nload, idx = 0; 360 Elf_Addr pos; 361 int file_align; 362 int loop; 363 size_t randomizequota = ELF_RANDOMIZE_LIMIT; 364 vaddr_t text_start = -1, text_end = 0; 365 366 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path, p); 367 nd.ni_pledge = PLEDGE_RPATH; 368 nd.ni_unveil = UNVEIL_READ; 369 if ((error = namei(&nd)) != 0) { 370 return (error); 371 } 372 vp = nd.ni_vp; 373 if (vp->v_type != VREG) { 374 error = EACCES; 375 goto bad; 376 } 377 if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0) 378 goto bad; 379 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 380 error = EACCES; 381 goto bad; 382 } 383 if ((error = VOP_ACCESS(vp, VREAD, p->p_ucred, p)) != 0) 384 goto bad1; 385 if ((error = elf_read_from(p, nd.ni_vp, 0, &eh, sizeof(eh))) != 0) 386 goto bad1; 387 388 if (elf_check_header(&eh) || eh.e_type != ET_DYN) { 389 error = ENOEXEC; 390 goto bad1; 391 } 392 393 ph = mallocarray(eh.e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK); 394 phsize = eh.e_phnum * sizeof(Elf_Phdr); 395 396 if ((error = elf_read_from(p, nd.ni_vp, eh.e_phoff, ph, phsize)) != 0) 397 goto bad1; 398 399 for (i = 0; i < eh.e_phnum; i++) { 400 if ((ph[i].p_align > 1) && !powerof2(ph[i].p_align)) { 401 error = EINVAL; 402 goto bad1; 403 } 404 405 if (ph[i].p_type == PT_LOAD) { 406 if (ph[i].p_filesz > ph[i].p_memsz || 407 ph[i].p_memsz == 0) { 408 error = EINVAL; 409 goto bad1; 410 } 411 loadmap[idx].vaddr = trunc_page(ph[i].p_vaddr); 412 loadmap[idx].memsz = round_page (ph[i].p_vaddr + 413 ph[i].p_memsz - loadmap[idx].vaddr); 414 file_align = ph[i].p_align; 415 idx++; 416 } 417 } 418 nload = idx; 419 420 /* 421 * Load the interpreter where a non-fixed mmap(NULL, ...) 422 * would (i.e. something safely out of the way). 423 */ 424 pos = uvm_map_hint(p->p_vmspace, PROT_EXEC, VM_MIN_ADDRESS, 425 VM_MAXUSER_ADDRESS); 426 pos = ELF_ROUND(pos, file_align); 427 428 loop = 0; 429 for (i = 0; i < nload;/**/) { 430 vaddr_t addr; 431 struct uvm_object *uobj; 432 off_t uoff; 433 size_t size; 434 435#ifdef this_needs_fixing 436 if (i == 0) { 437 uobj = &vp->v_uvm.u_obj; 438 /* need to fix uoff */ 439 } else { 440#endif 441 uobj = NULL; 442 uoff = 0; 443#ifdef this_needs_fixing 444 } 445#endif 446 447 addr = trunc_page(pos + loadmap[i].vaddr); 448 size = round_page(addr + loadmap[i].memsz) - addr; 449 450 /* CRAP - map_findspace does not avoid daddr+BRKSIZ */ 451 if ((addr + size > (vaddr_t)p->p_vmspace->vm_daddr) && 452 (addr < (vaddr_t)p->p_vmspace->vm_daddr + BRKSIZ)) 453 addr = round_page((vaddr_t)p->p_vmspace->vm_daddr + 454 BRKSIZ); 455 456 if (uvm_map_mquery(&p->p_vmspace->vm_map, &addr, size, 457 (i == 0 ? uoff : UVM_UNKNOWN_OFFSET), 0) != 0) { 458 if (loop == 0) { 459 loop = 1; 460 i = 0; 461 pos = 0; 462 continue; 463 } 464 error = ENOMEM; 465 goto bad1; 466 } 467 if (addr != pos + loadmap[i].vaddr) { 468 /* base changed. */ 469 pos = addr - trunc_page(loadmap[i].vaddr); 470 pos = ELF_ROUND(pos,file_align); 471 i = 0; 472 continue; 473 } 474 475 i++; 476 } 477 478 /* 479 * Load all the necessary sections 480 */ 481 for (i = 0; i < eh.e_phnum; i++) { 482 Elf_Addr size = 0; 483 int prot = 0; 484 int flags; 485 486 switch (ph[i].p_type) { 487 case PT_LOAD: 488 if (base_ph == NULL) { 489 flags = VMCMD_BASE; 490 addr = pos; 491 base_ph = &ph[i]; 492 } else { 493 flags = VMCMD_RELATIVE; 494 addr = ph[i].p_vaddr - base_ph->p_vaddr; 495 } 496 elf_load_psection(&epp->ep_vmcmds, nd.ni_vp, 497 &ph[i], &addr, &size, &prot, flags); 498 /* If entry is within this section it must be text */ 499 if (eh.e_entry >= ph[i].p_vaddr && 500 eh.e_entry < (ph[i].p_vaddr + size)) { 501 /* LOAD containing e_entry may not be writable */ 502 if (prot & PROT_WRITE) { 503 error = ENOEXEC; 504 goto bad1; 505 } 506 epp->ep_entry = addr + eh.e_entry - 507 ELF_TRUNC(ph[i].p_vaddr,ph[i].p_align); 508 if (flags == VMCMD_RELATIVE) 509 epp->ep_entry += pos; 510 ap->arg_interp = pos; 511 } 512 if (prot & PROT_EXEC) { 513 if (addr < text_start) 514 text_start = addr; 515 if (addr+size >= text_end) 516 text_end = addr + size; 517 } 518 addr += size; 519 break; 520 521 case PT_PHDR: 522 case PT_NOTE: 523 break; 524 525 case PT_OPENBSD_RANDOMIZE: 526 if (ph[i].p_memsz > randomizequota) { 527 error = ENOMEM; 528 goto bad1; 529 } 530 randomizequota -= ph[i].p_memsz; 531 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_randomize, 532 ph[i].p_memsz, ph[i].p_vaddr + pos, NULLVP, 0, 0); 533 break; 534 535 case PT_DYNAMIC: 536#if defined (__mips__) 537 /* DT_DEBUG is not ready on mips */ 538 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_mutable, 539 ph[i].p_memsz, ph[i].p_vaddr + pos, NULLVP, 0, 0); 540#endif 541 break; 542 case PT_GNU_RELRO: 543 case PT_OPENBSD_MUTABLE: 544 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_mutable, 545 ph[i].p_memsz, ph[i].p_vaddr + pos, NULLVP, 0, 0); 546 break; 547 case PT_OPENBSD_SYSCALLS: 548 syscall_ph = &ph[i]; 549 break; 550 default: 551 break; 552 } 553 } 554 555 if (syscall_ph) { 556 struct process *pr = p->p_p; 557 vaddr_t base = pos; 558 size_t len = text_end; 559 u_int *pins; 560 int npins; 561 562 npins = elf_read_pintable(p, nd.ni_vp, syscall_ph, 563 &pins, 1, len); 564 if (npins) { 565 elf_adjustpins(&base, &len, pins, npins, 566 text_start); 567 pr->ps_pin.pn_start = base; 568 pr->ps_pin.pn_end = base + len; 569 pr->ps_pin.pn_pins = pins; 570 pr->ps_pin.pn_npins = npins; 571 pr->ps_flags |= PS_PIN; 572 } 573 } 574 575 vn_marktext(nd.ni_vp); 576 577bad1: 578 VOP_CLOSE(nd.ni_vp, FREAD, p->p_ucred, p); 579bad: 580 free(ph, M_TEMP, phsize); 581 582 vput(nd.ni_vp); 583 return (error); 584} 585 586/* 587 * Prepare an Elf binary's exec package 588 * 589 * First, set of the various offsets/lengths in the exec package. 590 * 591 * Then, mark the text image busy (so it can be demand paged) or error out if 592 * this is not possible. Finally, set up vmcmds for the text, data, bss, and 593 * stack segments. 594 */ 595int 596exec_elf_makecmds(struct proc *p, struct exec_package *epp) 597{ 598 Elf_Ehdr *eh = epp->ep_hdr; 599 Elf_Phdr *ph, *pp, *base_ph = NULL, *syscall_ph = NULL; 600 Elf_Addr phdr = 0, exe_base = 0, exe_end = 0; 601 int error, i, has_phdr = 0, names = 0, textrel = 0; 602 char *interp = NULL; 603 u_long phsize; 604 size_t randomizequota = ELF_RANDOMIZE_LIMIT; 605 606 if (epp->ep_hdrvalid < sizeof(Elf_Ehdr)) 607 return (ENOEXEC); 608 609 if (elf_check_header(eh) || 610 (eh->e_type != ET_EXEC && eh->e_type != ET_DYN)) 611 return (ENOEXEC); 612 613 /* 614 * check if vnode is in open for writing, because we want to demand- 615 * page out of it. if it is, don't do it, for various reasons. 616 */ 617 if (epp->ep_vp->v_writecount != 0) { 618#ifdef DIAGNOSTIC 619 if (epp->ep_vp->v_flag & VTEXT) 620 panic("exec: a VTEXT vnode has writecount != 0"); 621#endif 622 return (ETXTBSY); 623 } 624 /* 625 * Allocate space to hold all the program headers, and read them 626 * from the file 627 */ 628 ph = mallocarray(eh->e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK); 629 phsize = eh->e_phnum * sizeof(Elf_Phdr); 630 631 if ((error = elf_read_from(p, epp->ep_vp, eh->e_phoff, ph, 632 phsize)) != 0) 633 goto bad; 634 635 epp->ep_tsize = ELF_NO_ADDR; 636 epp->ep_dsize = ELF_NO_ADDR; 637 638 for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) { 639 if ((pp->p_align > 1) && !powerof2(pp->p_align)) { 640 error = EINVAL; 641 goto bad; 642 } 643 644 if (pp->p_type == PT_INTERP && !interp) { 645 if (pp->p_filesz < 2 || pp->p_filesz > MAXPATHLEN) 646 goto bad; 647 interp = pool_get(&namei_pool, PR_WAITOK); 648 if ((error = elf_read_from(p, epp->ep_vp, 649 pp->p_offset, interp, pp->p_filesz)) != 0) { 650 goto bad; 651 } 652 if (interp[pp->p_filesz - 1] != '\0') 653 goto bad; 654 } else if (pp->p_type == PT_LOAD) { 655 if (pp->p_filesz > pp->p_memsz || 656 pp->p_memsz == 0) { 657 error = EINVAL; 658 goto bad; 659 } 660 if (base_ph == NULL) 661 base_ph = pp; 662 } else if (pp->p_type == PT_PHDR) { 663 has_phdr = 1; 664 } 665 } 666 667 /* 668 * Verify this is an OpenBSD executable. If it's marked that way 669 * via a PT_NOTE then also check for a PT_OPENBSD_WXNEEDED segment. 670 */ 671 if ((error = elf_os_pt_note(p, epp, epp->ep_hdr, &names)) != 0) 672 goto bad; 673 if (eh->e_ident[EI_OSABI] == ELFOSABI_OPENBSD) 674 names |= ELF_NOTE_NAME_OPENBSD; 675 676 if (eh->e_type == ET_DYN) { 677 /* need phdr and load sections for PIE */ 678 if (!has_phdr || base_ph == NULL || base_ph->p_vaddr != 0) { 679 error = EINVAL; 680 goto bad; 681 } 682 /* randomize exe_base for PIE */ 683 exe_base = uvm_map_pie(base_ph->p_align); 684 685 /* 686 * Check if DYNAMIC contains DT_TEXTREL 687 */ 688 for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) { 689 Elf_Dyn *dt; 690 int j; 691 692 switch (pp->p_type) { 693 case PT_DYNAMIC: 694 if (pp->p_filesz > 64*1024) 695 break; 696 dt = malloc(pp->p_filesz, M_TEMP, M_WAITOK); 697 error = vn_rdwr(UIO_READ, epp->ep_vp, 698 (caddr_t)dt, pp->p_filesz, pp->p_offset, 699 UIO_SYSSPACE, IO_UNIT, p->p_ucred, NULL, p); 700 if (error) { 701 free(dt, M_TEMP, pp->p_filesz); 702 break; 703 } 704 for (j = 0; j < pp->p_filesz / sizeof(*dt); j++) { 705 if (dt[j].d_tag == DT_TEXTREL) { 706 textrel = VMCMD_TEXTREL; 707 break; 708 } 709 } 710 free(dt, M_TEMP, pp->p_filesz); 711 break; 712 default: 713 break; 714 } 715 } 716 } 717 718 /* 719 * Load all the necessary sections 720 */ 721 for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) { 722 Elf_Addr addr, size = 0; 723 int prot = 0; 724 int flags = 0; 725 726 switch (pp->p_type) { 727 case PT_LOAD: 728 if (exe_base != 0) { 729 if (pp == base_ph) { 730 flags = VMCMD_BASE; 731 addr = exe_base; 732 } else { 733 flags = VMCMD_RELATIVE; 734 addr = pp->p_vaddr - base_ph->p_vaddr; 735 } 736 } else 737 addr = ELF_NO_ADDR; 738 739 /* Static binaries may not call pinsyscalls() */ 740 if (interp == NULL) 741 p->p_vmspace->vm_map.flags |= VM_MAP_PINSYSCALL_ONCE; 742 743 /* 744 * Calculates size of text and data segments 745 * by starting at first and going to end of last. 746 * 'rwx' sections are treated as data. 747 * this is correct for BSS_PLT, but may not be 748 * for DATA_PLT, is fine for TEXT_PLT. 749 */ 750 elf_load_psection(&epp->ep_vmcmds, epp->ep_vp, 751 pp, &addr, &size, &prot, flags | textrel); 752 753 /* 754 * Update exe_base in case alignment was off. 755 * For PIE, addr is relative to exe_base so 756 * adjust it (non PIE exe_base is 0 so no change). 757 */ 758 if (flags == VMCMD_BASE) 759 exe_base = addr; 760 else 761 addr += exe_base; 762 763 /* 764 * Decide whether it's text or data by looking 765 * at the protection of the section 766 */ 767 if (prot & PROT_WRITE) { 768 /* data section */ 769 if (epp->ep_dsize == ELF_NO_ADDR) { 770 epp->ep_daddr = addr; 771 epp->ep_dsize = size; 772 } else { 773 if (addr < epp->ep_daddr) { 774 epp->ep_dsize = 775 epp->ep_dsize + 776 epp->ep_daddr - 777 addr; 778 epp->ep_daddr = addr; 779 } else 780 epp->ep_dsize = addr+size - 781 epp->ep_daddr; 782 } 783 } else if (prot & PROT_EXEC) { 784 /* text section */ 785 if (epp->ep_tsize == ELF_NO_ADDR) { 786 epp->ep_taddr = addr; 787 epp->ep_tsize = size; 788 } else { 789 if (addr < epp->ep_taddr) { 790 epp->ep_tsize = 791 epp->ep_tsize + 792 epp->ep_taddr - 793 addr; 794 epp->ep_taddr = addr; 795 } else 796 epp->ep_tsize = addr+size - 797 epp->ep_taddr; 798 } 799 if (interp == NULL) 800 exe_end = epp->ep_taddr + 801 epp->ep_tsize; /* end of TEXT */ 802 } 803 break; 804 805 case PT_SHLIB: 806 error = ENOEXEC; 807 goto bad; 808 809 case PT_INTERP: 810 /* Already did this one */ 811 case PT_NOTE: 812 break; 813 814 case PT_PHDR: 815 /* Note address of program headers (in text segment) */ 816 phdr = pp->p_vaddr; 817 break; 818 819 case PT_OPENBSD_RANDOMIZE: 820 if (ph[i].p_memsz > randomizequota) { 821 error = ENOMEM; 822 goto bad; 823 } 824 randomizequota -= ph[i].p_memsz; 825 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_randomize, 826 ph[i].p_memsz, ph[i].p_vaddr + exe_base, NULLVP, 0, 0); 827 break; 828 829 case PT_DYNAMIC: 830#if defined (__mips__) 831 /* DT_DEBUG is not ready on mips */ 832 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_mutable, 833 ph[i].p_memsz, ph[i].p_vaddr + exe_base, NULLVP, 0, 0); 834#endif 835 break; 836 case PT_GNU_RELRO: 837 case PT_OPENBSD_MUTABLE: 838 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_mutable, 839 ph[i].p_memsz, ph[i].p_vaddr + exe_base, NULLVP, 0, 0); 840 break; 841 case PT_OPENBSD_SYSCALLS: 842 if (interp == NULL) 843 syscall_ph = &ph[i]; 844 break; 845 default: 846 /* 847 * Not fatal, we don't need to understand everything 848 * :-) 849 */ 850 break; 851 } 852 } 853 854 if (syscall_ph) { 855 vaddr_t base = exe_base; 856 size_t len = exe_end - exe_base; 857 u_int *pins; 858 int npins; 859 860 npins = elf_read_pintable(p, epp->ep_vp, syscall_ph, 861 &pins, 0, len); 862 if (npins) { 863 elf_adjustpins(&base, &len, pins, npins, 864 epp->ep_taddr - exe_base); 865 epp->ep_pinstart = base; 866 epp->ep_pinend = base + len; 867 epp->ep_pins = pins; 868 epp->ep_npins = npins; 869 p->p_p->ps_flags |= PS_PIN; 870 } 871 } 872 873 phdr += exe_base; 874 875 /* 876 * Strangely some linux programs may have all load sections marked 877 * writeable, in this case, textsize is not -1, but rather 0; 878 */ 879 if (epp->ep_tsize == ELF_NO_ADDR) 880 epp->ep_tsize = 0; 881 /* 882 * Another possibility is that it has all load sections marked 883 * read-only. Fake a zero-sized data segment right after the 884 * text segment. 885 */ 886 if (epp->ep_dsize == ELF_NO_ADDR) { 887 epp->ep_daddr = round_page(epp->ep_taddr + epp->ep_tsize); 888 epp->ep_dsize = 0; 889 } 890 891 epp->ep_interp = interp; 892 epp->ep_entry = eh->e_entry + exe_base; 893 894 /* 895 * Check if we found a dynamically linked binary and arrange to load 896 * its interpreter when the exec file is released. 897 */ 898 if (interp || eh->e_type == ET_DYN) { 899 struct elf_args *ap; 900 901 ap = malloc(sizeof(*ap), M_TEMP, M_WAITOK); 902 903 ap->arg_phaddr = phdr; 904 ap->arg_phentsize = eh->e_phentsize; 905 ap->arg_phnum = eh->e_phnum; 906 ap->arg_entry = eh->e_entry + exe_base; 907 ap->arg_interp = exe_base; 908 909 epp->ep_args = ap; 910 } 911 912 free(ph, M_TEMP, phsize); 913 vn_marktext(epp->ep_vp); 914 return (exec_setup_stack(p, epp)); 915 916bad: 917 if (interp) 918 pool_put(&namei_pool, interp); 919 free(ph, M_TEMP, phsize); 920 kill_vmcmds(&epp->ep_vmcmds); 921 if (error == 0) 922 return (ENOEXEC); 923 return (error); 924} 925 926/* 927 * Phase II of load. It is now safe to load the interpreter. Info collected 928 * when loading the program is available for setup of the interpreter. 929 */ 930int 931exec_elf_fixup(struct proc *p, struct exec_package *epp) 932{ 933 char *interp; 934 int error = 0; 935 struct elf_args *ap; 936 AuxInfo ai[ELF_AUX_ENTRIES], *a; 937 938 ap = epp->ep_args; 939 if (ap == NULL) { 940 return (0); 941 } 942 943 interp = epp->ep_interp; 944 945 /* disable kbind in programs that don't use ld.so */ 946 if (interp == NULL) 947 p->p_p->ps_kbind_addr = BOGO_PC; 948 949 if (interp && 950 (error = elf_load_file(p, interp, epp, ap)) != 0) { 951 uprintf("execve: cannot load %s\n", interp); 952 free(ap, M_TEMP, sizeof *ap); 953 pool_put(&namei_pool, interp); 954 kill_vmcmds(&epp->ep_vmcmds); 955 return (error); 956 } 957 /* 958 * We have to do this ourselves... 959 */ 960 error = exec_process_vmcmds(p, epp); 961 962 /* 963 * Push extra arguments on the stack needed by dynamically 964 * linked binaries 965 */ 966 if (error == 0) { 967 memset(&ai, 0, sizeof ai); 968 a = ai; 969 970 a->au_id = AUX_phdr; 971 a->au_v = ap->arg_phaddr; 972 a++; 973 974 a->au_id = AUX_phent; 975 a->au_v = ap->arg_phentsize; 976 a++; 977 978 a->au_id = AUX_phnum; 979 a->au_v = ap->arg_phnum; 980 a++; 981 982 a->au_id = AUX_pagesz; 983 a->au_v = PAGE_SIZE; 984 a++; 985 986 a->au_id = AUX_base; 987 a->au_v = ap->arg_interp; 988 a++; 989 990 a->au_id = AUX_flags; 991 a->au_v = 0; 992 a++; 993 994 a->au_id = AUX_entry; 995 a->au_v = ap->arg_entry; 996 a++; 997 998 a->au_id = AUX_openbsd_timekeep; 999 a->au_v = p->p_p->ps_timekeep; 1000 a++; 1001 1002 a->au_id = AUX_null; 1003 a->au_v = 0; 1004 a++; 1005 1006 error = copyout(ai, epp->ep_auxinfo, sizeof ai); 1007 } 1008 free(ap, M_TEMP, sizeof *ap); 1009 if (interp) 1010 pool_put(&namei_pool, interp); 1011 return (error); 1012} 1013 1014int 1015elf_os_pt_note_name(Elf_Note *np) 1016{ 1017 int i, j; 1018 1019 for (i = 0; i < nitems(elf_note_names); i++) { 1020 size_t namlen = strlen(elf_note_names[i].name); 1021 if (np->namesz < namlen) 1022 continue; 1023 /* verify name padding (after the NUL) is NUL */ 1024 for (j = namlen + 1; j < elfround(np->namesz); j++) 1025 if (((char *)(np + 1))[j] != '\0') 1026 continue; 1027 /* verify desc padding is NUL */ 1028 for (j = np->descsz; j < elfround(np->descsz); j++) 1029 if (((char *)(np + 1))[j] != '\0') 1030 continue; 1031 if (strcmp((char *)(np + 1), elf_note_names[i].name) == 0) 1032 return elf_note_names[i].id; 1033 } 1034 return (0); 1035} 1036 1037int 1038elf_os_pt_note(struct proc *p, struct exec_package *epp, Elf_Ehdr *eh, int *namesp) 1039{ 1040 Elf_Phdr *hph, *ph; 1041 Elf_Note *np = NULL; 1042 size_t phsize, offset, pfilesz = 0, total; 1043 int error, names = 0; 1044 1045 hph = mallocarray(eh->e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK); 1046 phsize = eh->e_phnum * sizeof(Elf_Phdr); 1047 if ((error = elf_read_from(p, epp->ep_vp, eh->e_phoff, 1048 hph, phsize)) != 0) 1049 goto out1; 1050 1051 for (ph = hph; ph < &hph[eh->e_phnum]; ph++) { 1052 if (ph->p_type == PT_OPENBSD_WXNEEDED) { 1053 epp->ep_flags |= EXEC_WXNEEDED; 1054 continue; 1055 } 1056 if (ph->p_type == PT_OPENBSD_NOBTCFI) { 1057 epp->ep_flags |= EXEC_NOBTCFI; 1058 continue; 1059 } 1060 1061 if (ph->p_type != PT_NOTE || ph->p_filesz > 1024) 1062 continue; 1063 1064 if (np && ph->p_filesz != pfilesz) { 1065 free(np, M_TEMP, pfilesz); 1066 np = NULL; 1067 } 1068 if (!np) 1069 np = malloc(ph->p_filesz, M_TEMP, M_WAITOK); 1070 pfilesz = ph->p_filesz; 1071 if ((error = elf_read_from(p, epp->ep_vp, ph->p_offset, 1072 np, ph->p_filesz)) != 0) 1073 goto out2; 1074 1075 for (offset = 0; offset < ph->p_filesz; offset += total) { 1076 Elf_Note *np2 = (Elf_Note *)((char *)np + offset); 1077 1078 if (offset + sizeof(Elf_Note) > ph->p_filesz) 1079 break; 1080 total = sizeof(Elf_Note) + elfround(np2->namesz) + 1081 elfround(np2->descsz); 1082 if (offset + total > ph->p_filesz) 1083 break; 1084 names |= elf_os_pt_note_name(np2); 1085 } 1086 } 1087 1088out2: 1089 free(np, M_TEMP, pfilesz); 1090out1: 1091 free(hph, M_TEMP, phsize); 1092 *namesp = names; 1093 return ((names & ELF_NOTE_NAME_OPENBSD) ? 0 : ENOEXEC); 1094} 1095 1096/* 1097 * Start of routines related to dumping core 1098 */ 1099 1100#ifdef SMALL_KERNEL 1101int 1102coredump_elf(struct proc *p, void *cookie) 1103{ 1104 return EPERM; 1105} 1106#else /* !SMALL_KERNEL */ 1107 1108struct writesegs_state { 1109 off_t notestart; 1110 off_t secstart; 1111 off_t secoff; 1112 struct proc *p; 1113 void *iocookie; 1114 Elf_Phdr *psections; 1115 size_t psectionslen; 1116 size_t notesize; 1117 int npsections; 1118}; 1119 1120uvm_coredump_setup_cb coredump_setup_elf; 1121uvm_coredump_walk_cb coredump_walk_elf; 1122 1123int coredump_notes_elf(struct proc *, void *, size_t *); 1124int coredump_note_elf(struct proc *, void *, size_t *); 1125int coredump_writenote_elf(struct proc *, void *, Elf_Note *, 1126 const char *, void *); 1127 1128extern vaddr_t sigcode_va; 1129extern vsize_t sigcode_sz; 1130 1131int 1132coredump_elf(struct proc *p, void *cookie) 1133{ 1134#ifdef DIAGNOSTIC 1135 off_t offset; 1136#endif 1137 struct writesegs_state ws; 1138 size_t notesize; 1139 int error, i; 1140 1141 ws.p = p; 1142 ws.iocookie = cookie; 1143 ws.psections = NULL; 1144 1145 /* 1146 * Walk the map to get all the segment offsets and lengths, 1147 * write out the ELF header. 1148 */ 1149 error = uvm_coredump_walkmap(p, coredump_setup_elf, 1150 coredump_walk_elf, &ws); 1151 if (error) 1152 goto out; 1153 1154 error = coredump_write(cookie, UIO_SYSSPACE, ws.psections, 1155 ws.psectionslen, 0); 1156 if (error) 1157 goto out; 1158 1159 /* Write out the notes. */ 1160 error = coredump_notes_elf(p, cookie, ¬esize); 1161 if (error) 1162 goto out; 1163 1164#ifdef DIAGNOSTIC 1165 if (notesize != ws.notesize) 1166 panic("coredump: notesize changed: %zu != %zu", 1167 ws.notesize, notesize); 1168 offset = ws.notestart + notesize; 1169 if (offset != ws.secstart) 1170 panic("coredump: offset %lld != secstart %lld", 1171 (long long) offset, (long long) ws.secstart); 1172#endif 1173 1174 /* Pass 3: finally, write the sections themselves. */ 1175 for (i = 0; i < ws.npsections - 1; i++) { 1176 Elf_Phdr *pent = &ws.psections[i]; 1177 if (pent->p_filesz == 0) 1178 continue; 1179 1180#ifdef DIAGNOSTIC 1181 if (offset != pent->p_offset) 1182 panic("coredump: offset %lld != p_offset[%d] %lld", 1183 (long long) offset, i, 1184 (long long) pent->p_filesz); 1185#endif 1186 1187 /* 1188 * Since the sigcode is mapped execute-only, we can't 1189 * read it. So use the kernel mapping for it instead. 1190 */ 1191 if (pent->p_vaddr == p->p_p->ps_sigcode && 1192 pent->p_filesz == sigcode_sz) { 1193 error = coredump_write(cookie, UIO_SYSSPACE, 1194 (void *)sigcode_va, sigcode_sz, 0); 1195 } else { 1196 error = coredump_write(cookie, UIO_USERSPACE, 1197 (void *)(vaddr_t)pent->p_vaddr, pent->p_filesz, 1198 (pent->p_flags & PF_ISVNODE)); 1199 } 1200 if (error) 1201 goto out; 1202 1203 coredump_unmap(cookie, (vaddr_t)pent->p_vaddr, 1204 (vaddr_t)pent->p_vaddr + pent->p_filesz); 1205 1206#ifdef DIAGNOSTIC 1207 offset += ws.psections[i].p_filesz; 1208#endif 1209 } 1210 1211out: 1212 free(ws.psections, M_TEMP, ws.psectionslen); 1213 return (error); 1214} 1215 1216 1217/* 1218 * Normally we lay out core files like this: 1219 * [ELF Header] [Program headers] [Notes] [data for PT_LOAD segments] 1220 * 1221 * However, if there's >= 65535 segments then it overflows the field 1222 * in the ELF header, so the standard specifies putting a magic 1223 * number there and saving the real count in the .sh_info field of 1224 * the first *section* header...which requires generating a section 1225 * header. To avoid confusing tools, we include an .shstrtab section 1226 * as well so all the indexes look valid. So in this case we lay 1227 * out the core file like this: 1228 * [ELF Header] [Section Headers] [.shstrtab] [Program headers] \ 1229 * [Notes] [data for PT_LOAD segments] 1230 * 1231 * The 'shstrtab' structure below is data for the second of the two 1232 * section headers, plus the .shstrtab itself, in one const buffer. 1233 */ 1234static const struct { 1235 Elf_Shdr shdr; 1236 char shstrtab[sizeof(ELF_SHSTRTAB) + 1]; 1237} shstrtab = { 1238 .shdr = { 1239 .sh_name = 1, /* offset in .shstrtab below */ 1240 .sh_type = SHT_STRTAB, 1241 .sh_offset = sizeof(Elf_Ehdr) + 2*sizeof(Elf_Shdr), 1242 .sh_size = sizeof(ELF_SHSTRTAB) + 1, 1243 .sh_addralign = 1, 1244 }, 1245 .shstrtab = "\0" ELF_SHSTRTAB, 1246}; 1247 1248int 1249coredump_setup_elf(int segment_count, void *cookie) 1250{ 1251 Elf_Ehdr ehdr; 1252 struct writesegs_state *ws = cookie; 1253 Elf_Phdr *note; 1254 int error; 1255 1256 /* Get the count of segments, plus one for the PT_NOTE */ 1257 ws->npsections = segment_count + 1; 1258 1259 /* Get the size of the notes. */ 1260 error = coredump_notes_elf(ws->p, NULL, &ws->notesize); 1261 if (error) 1262 return error; 1263 1264 /* Setup the ELF header */ 1265 memset(&ehdr, 0, sizeof(ehdr)); 1266 memcpy(ehdr.e_ident, ELFMAG, SELFMAG); 1267 ehdr.e_ident[EI_CLASS] = ELF_TARG_CLASS; 1268 ehdr.e_ident[EI_DATA] = ELF_TARG_DATA; 1269 ehdr.e_ident[EI_VERSION] = EV_CURRENT; 1270 /* XXX Should be the OSABI/ABI version of the executable. */ 1271 ehdr.e_ident[EI_OSABI] = ELFOSABI_SYSV; 1272 ehdr.e_ident[EI_ABIVERSION] = 0; 1273 ehdr.e_type = ET_CORE; 1274 /* XXX This should be the e_machine of the executable. */ 1275 ehdr.e_machine = ELF_TARG_MACH; 1276 ehdr.e_version = EV_CURRENT; 1277 ehdr.e_entry = 0; 1278 ehdr.e_flags = 0; 1279 ehdr.e_ehsize = sizeof(ehdr); 1280 ehdr.e_phentsize = sizeof(Elf_Phdr); 1281 1282 if (ws->npsections < PN_XNUM) { 1283 ehdr.e_phoff = sizeof(ehdr); 1284 ehdr.e_shoff = 0; 1285 ehdr.e_phnum = ws->npsections; 1286 ehdr.e_shentsize = 0; 1287 ehdr.e_shnum = 0; 1288 ehdr.e_shstrndx = 0; 1289 } else { 1290 /* too many segments, use extension setup */ 1291 ehdr.e_shoff = sizeof(ehdr); 1292 ehdr.e_phnum = PN_XNUM; 1293 ehdr.e_shentsize = sizeof(Elf_Shdr); 1294 ehdr.e_shnum = 2; 1295 ehdr.e_shstrndx = 1; 1296 ehdr.e_phoff = shstrtab.shdr.sh_offset + shstrtab.shdr.sh_size; 1297 } 1298 1299 /* Write out the ELF header. */ 1300 error = coredump_write(ws->iocookie, UIO_SYSSPACE, &ehdr, sizeof(ehdr), 0); 1301 if (error) 1302 return error; 1303 1304 /* 1305 * If an section header is needed to store extension info, write 1306 * it out after the ELF header and before the program header. 1307 */ 1308 if (ehdr.e_shnum != 0) { 1309 Elf_Shdr shdr = { .sh_info = ws->npsections }; 1310 error = coredump_write(ws->iocookie, UIO_SYSSPACE, &shdr, 1311 sizeof shdr, 0); 1312 if (error) 1313 return error; 1314 error = coredump_write(ws->iocookie, UIO_SYSSPACE, &shstrtab, 1315 sizeof(shstrtab.shdr) + sizeof(shstrtab.shstrtab), 0); 1316 if (error) 1317 return error; 1318 } 1319 1320 /* 1321 * Allocate the segment header array and setup to collect 1322 * the section sizes and offsets 1323 */ 1324 ws->psections = mallocarray(ws->npsections, sizeof(Elf_Phdr), 1325 M_TEMP, M_WAITOK|M_CANFAIL|M_ZERO); 1326 if (ws->psections == NULL) 1327 return ENOMEM; 1328 ws->psectionslen = ws->npsections * sizeof(Elf_Phdr); 1329 1330 ws->notestart = ehdr.e_phoff + ws->psectionslen; 1331 ws->secstart = ws->notestart + ws->notesize; 1332 ws->secoff = ws->secstart; 1333 1334 /* Fill in the PT_NOTE segment header in the last slot */ 1335 note = &ws->psections[ws->npsections - 1]; 1336 note->p_type = PT_NOTE; 1337 note->p_offset = ws->notestart; 1338 note->p_vaddr = 0; 1339 note->p_paddr = 0; 1340 note->p_filesz = ws->notesize; 1341 note->p_memsz = 0; 1342 note->p_flags = PF_R; 1343 note->p_align = ELFROUNDSIZE; 1344 1345 return (0); 1346} 1347 1348int 1349coredump_walk_elf(vaddr_t start, vaddr_t realend, vaddr_t end, vm_prot_t prot, 1350 int isvnode, int nsegment, void *cookie) 1351{ 1352 struct writesegs_state *ws = cookie; 1353 Elf_Phdr phdr; 1354 vsize_t size, realsize; 1355 1356 size = end - start; 1357 realsize = realend - start; 1358 1359 phdr.p_type = PT_LOAD; 1360 phdr.p_offset = ws->secoff; 1361 phdr.p_vaddr = start; 1362 phdr.p_paddr = 0; 1363 phdr.p_filesz = realsize; 1364 phdr.p_memsz = size; 1365 phdr.p_flags = 0; 1366 if (prot & PROT_READ) 1367 phdr.p_flags |= PF_R; 1368 if (prot & PROT_WRITE) 1369 phdr.p_flags |= PF_W; 1370 if (prot & PROT_EXEC) 1371 phdr.p_flags |= PF_X; 1372 if (isvnode) 1373 phdr.p_flags |= PF_ISVNODE; 1374 phdr.p_align = PAGE_SIZE; 1375 1376 ws->secoff += phdr.p_filesz; 1377 ws->psections[nsegment] = phdr; 1378 1379 return (0); 1380} 1381 1382int 1383coredump_notes_elf(struct proc *p, void *iocookie, size_t *sizep) 1384{ 1385 struct elfcore_procinfo cpi; 1386 Elf_Note nhdr; 1387 struct process *pr = p->p_p; 1388 struct proc *q; 1389 size_t size, notesize; 1390 int error; 1391 1392 KASSERT(!P_HASSIBLING(p) || pr->ps_single != NULL); 1393 size = 0; 1394 1395 /* First, write an elfcore_procinfo. */ 1396 notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) + 1397 elfround(sizeof(cpi)); 1398 if (iocookie) { 1399 memset(&cpi, 0, sizeof(cpi)); 1400 1401 cpi.cpi_version = ELFCORE_PROCINFO_VERSION; 1402 cpi.cpi_cpisize = sizeof(cpi); 1403 cpi.cpi_signo = p->p_sisig; 1404 cpi.cpi_sigcode = p->p_sicode; 1405 1406 cpi.cpi_sigpend = p->p_siglist | pr->ps_siglist; 1407 cpi.cpi_sigmask = p->p_sigmask; 1408 cpi.cpi_sigignore = pr->ps_sigacts->ps_sigignore; 1409 cpi.cpi_sigcatch = pr->ps_sigacts->ps_sigcatch; 1410 1411 cpi.cpi_pid = pr->ps_pid; 1412 cpi.cpi_ppid = pr->ps_ppid; 1413 cpi.cpi_pgrp = pr->ps_pgid; 1414 if (pr->ps_session->s_leader) 1415 cpi.cpi_sid = pr->ps_session->s_leader->ps_pid; 1416 else 1417 cpi.cpi_sid = 0; 1418 1419 cpi.cpi_ruid = p->p_ucred->cr_ruid; 1420 cpi.cpi_euid = p->p_ucred->cr_uid; 1421 cpi.cpi_svuid = p->p_ucred->cr_svuid; 1422 1423 cpi.cpi_rgid = p->p_ucred->cr_rgid; 1424 cpi.cpi_egid = p->p_ucred->cr_gid; 1425 cpi.cpi_svgid = p->p_ucred->cr_svgid; 1426 1427 (void)strlcpy(cpi.cpi_name, pr->ps_comm, sizeof(cpi.cpi_name)); 1428 1429 nhdr.namesz = sizeof("OpenBSD"); 1430 nhdr.descsz = sizeof(cpi); 1431 nhdr.type = NT_OPENBSD_PROCINFO; 1432 1433 error = coredump_writenote_elf(p, iocookie, &nhdr, 1434 "OpenBSD", &cpi); 1435 if (error) 1436 return (error); 1437 } 1438 size += notesize; 1439 1440 /* Second, write an NT_OPENBSD_AUXV note. */ 1441 notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) + 1442 elfround(ELF_AUX_WORDS * sizeof(char *)); 1443 if (iocookie && pr->ps_auxinfo) { 1444 1445 nhdr.namesz = sizeof("OpenBSD"); 1446 nhdr.descsz = ELF_AUX_WORDS * sizeof(char *); 1447 nhdr.type = NT_OPENBSD_AUXV; 1448 1449 error = coredump_write(iocookie, UIO_SYSSPACE, 1450 &nhdr, sizeof(nhdr), 0); 1451 if (error) 1452 return (error); 1453 1454 error = coredump_write(iocookie, UIO_SYSSPACE, 1455 "OpenBSD", elfround(nhdr.namesz), 0); 1456 if (error) 1457 return (error); 1458 1459 error = coredump_write(iocookie, UIO_USERSPACE, 1460 (caddr_t)pr->ps_auxinfo, nhdr.descsz, 0); 1461 if (error) 1462 return (error); 1463 } 1464 size += notesize; 1465 1466#ifdef PT_WCOOKIE 1467 notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) + 1468 elfround(sizeof(register_t)); 1469 if (iocookie) { 1470 register_t wcookie; 1471 1472 nhdr.namesz = sizeof("OpenBSD"); 1473 nhdr.descsz = sizeof(register_t); 1474 nhdr.type = NT_OPENBSD_WCOOKIE; 1475 1476 wcookie = process_get_wcookie(p); 1477 error = coredump_writenote_elf(p, iocookie, &nhdr, 1478 "OpenBSD", &wcookie); 1479 if (error) 1480 return (error); 1481 } 1482 size += notesize; 1483#endif 1484 1485 /* 1486 * Now write the register info for the thread that caused the 1487 * coredump. 1488 */ 1489 error = coredump_note_elf(p, iocookie, ¬esize); 1490 if (error) 1491 return (error); 1492 size += notesize; 1493 1494 /* 1495 * Now, for each thread, write the register info and any other 1496 * per-thread notes. Since we're dumping core, all the other 1497 * threads in the process have been stopped and the list can't 1498 * change. 1499 */ 1500 TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) { 1501 if (q == p) /* we've taken care of this thread */ 1502 continue; 1503 error = coredump_note_elf(q, iocookie, ¬esize); 1504 if (error) 1505 return (error); 1506 size += notesize; 1507 } 1508 1509 *sizep = size; 1510 return (0); 1511} 1512 1513int 1514coredump_note_elf(struct proc *p, void *iocookie, size_t *sizep) 1515{ 1516 Elf_Note nhdr; 1517 int size, notesize, error; 1518 int namesize; 1519 char name[64+ELFROUNDSIZE]; 1520 struct reg intreg; 1521#ifdef PT_GETFPREGS 1522 struct fpreg freg; 1523#endif 1524#ifdef PT_PACMASK 1525 register_t pacmask[2]; 1526#endif 1527 1528 size = 0; 1529 1530 snprintf(name, sizeof(name)-ELFROUNDSIZE, "%s@%d", 1531 "OpenBSD", p->p_tid + THREAD_PID_OFFSET); 1532 namesize = strlen(name) + 1; 1533 memset(name + namesize, 0, elfround(namesize) - namesize); 1534 1535 notesize = sizeof(nhdr) + elfround(namesize) + elfround(sizeof(intreg)); 1536 if (iocookie) { 1537 error = process_read_regs(p, &intreg); 1538 if (error) 1539 return (error); 1540 1541 nhdr.namesz = namesize; 1542 nhdr.descsz = sizeof(intreg); 1543 nhdr.type = NT_OPENBSD_REGS; 1544 1545 error = coredump_writenote_elf(p, iocookie, &nhdr, 1546 name, &intreg); 1547 if (error) 1548 return (error); 1549 1550 } 1551 size += notesize; 1552 1553#ifdef PT_GETFPREGS 1554 notesize = sizeof(nhdr) + elfround(namesize) + elfround(sizeof(freg)); 1555 if (iocookie) { 1556 error = process_read_fpregs(p, &freg); 1557 if (error) 1558 return (error); 1559 1560 nhdr.namesz = namesize; 1561 nhdr.descsz = sizeof(freg); 1562 nhdr.type = NT_OPENBSD_FPREGS; 1563 1564 error = coredump_writenote_elf(p, iocookie, &nhdr, name, &freg); 1565 if (error) 1566 return (error); 1567 } 1568 size += notesize; 1569#endif 1570 1571#ifdef PT_PACMASK 1572 notesize = sizeof(nhdr) + elfround(namesize) + 1573 elfround(sizeof(pacmask)); 1574 if (iocookie) { 1575 pacmask[0] = pacmask[1] = process_get_pacmask(p); 1576 1577 nhdr.namesz = namesize; 1578 nhdr.descsz = sizeof(pacmask); 1579 nhdr.type = NT_OPENBSD_PACMASK; 1580 1581 error = coredump_writenote_elf(p, iocookie, &nhdr, 1582 name, &pacmask); 1583 if (error) 1584 return (error); 1585 } 1586 size += notesize; 1587#endif 1588 1589 *sizep = size; 1590 /* XXX Add hook for machdep per-LWP notes. */ 1591 return (0); 1592} 1593 1594int 1595coredump_writenote_elf(struct proc *p, void *cookie, Elf_Note *nhdr, 1596 const char *name, void *data) 1597{ 1598 int error; 1599 1600 error = coredump_write(cookie, UIO_SYSSPACE, nhdr, sizeof(*nhdr), 0); 1601 if (error) 1602 return error; 1603 1604 error = coredump_write(cookie, UIO_SYSSPACE, name, 1605 elfround(nhdr->namesz), 0); 1606 if (error) 1607 return error; 1608 1609 return coredump_write(cookie, UIO_SYSSPACE, data, nhdr->descsz, 0); 1610} 1611#endif /* !SMALL_KERNEL */ 1612