1 /* $NetBSD: exec_elf.c,v 1.102 2021/05/18 08:59:44 hannken Exp $ */ 2 3 /*- 4 * Copyright (c) 1994, 2000, 2005, 2015, 2020 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Christos Zoulas and Maxime Villard. 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Copyright (c) 1996 Christopher G. Demetriou 34 * All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. The name of the author may not be used to endorse or promote products 45 * derived from this software without specific prior written permission 46 * 47 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 48 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 49 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 50 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 51 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 52 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 53 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 54 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 55 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 56 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 57 */ 58 59 #include <sys/cdefs.h> 60 __KERNEL_RCSID(1, "$NetBSD: exec_elf.c,v 1.102 2021/05/18 08:59:44 hannken Exp $"); 61 62 #ifdef _KERNEL_OPT 63 #include "opt_pax.h" 64 #endif /* _KERNEL_OPT */ 65 66 #include <sys/param.h> 67 #include <sys/proc.h> 68 #include <sys/kmem.h> 69 #include <sys/namei.h> 70 #include <sys/vnode.h> 71 #include <sys/exec.h> 72 #include <sys/exec_elf.h> 73 #include <sys/syscall.h> 74 #include <sys/signalvar.h> 75 #include <sys/mount.h> 76 #include <sys/stat.h> 77 #include <sys/kauth.h> 78 #include <sys/bitops.h> 79 80 #include <sys/cpu.h> 81 #include <machine/reg.h> 82 83 #include <compat/common/compat_util.h> 84 85 #include <sys/pax.h> 86 #include <uvm/uvm_param.h> 87 88 extern struct emul emul_netbsd; 89 90 #define elf_check_header ELFNAME(check_header) 91 #define elf_copyargs ELFNAME(copyargs) 92 #define elf_populate_auxv ELFNAME(populate_auxv) 93 #define elf_load_interp ELFNAME(load_interp) 94 #define elf_load_psection ELFNAME(load_psection) 95 #define exec_elf_makecmds ELFNAME2(exec,makecmds) 96 #define netbsd_elf_signature ELFNAME2(netbsd,signature) 97 #define netbsd_elf_note ELFNAME2(netbsd,note) 98 #define netbsd_elf_probe ELFNAME2(netbsd,probe) 99 #define coredump ELFNAMEEND(coredump) 100 #define elf_free_emul_arg ELFNAME(free_emul_arg) 101 102 static int 103 elf_load_interp(struct lwp *, struct exec_package *, char *, 104 struct exec_vmcmd_set *, u_long *, Elf_Addr *); 105 static int 106 elf_load_psection(struct exec_vmcmd_set *, struct vnode *, const Elf_Phdr *, 107 Elf_Addr *, u_long *, int); 108 109 int netbsd_elf_signature(struct lwp *, struct exec_package *, Elf_Ehdr *); 110 int netbsd_elf_note(struct exec_package *, const Elf_Nhdr *, const char *, 111 const char *); 112 int netbsd_elf_probe(struct lwp *, struct exec_package *, void *, char *, 113 vaddr_t *); 114 115 static void elf_free_emul_arg(void *); 116 117 #ifdef DEBUG_ELF 118 #define DPRINTF(a, ...) printf("%s: " a "\n", __func__, ##__VA_ARGS__) 119 #else 120 #define DPRINTF(a, ...) 121 #endif 122 123 /* round up and down to page boundaries. */ 124 #define ELF_ROUND(a, b) (((a) + (b) - 1) & ~((b) - 1)) 125 #define ELF_TRUNC(a, b) ((a) & ~((b) - 1)) 126 127 static int 128 elf_placedynexec(struct exec_package *epp, Elf_Ehdr *eh, Elf_Phdr *ph) 129 { 130 Elf_Addr align, offset; 131 int i; 132 133 for (align = 1, i = 0; i < eh->e_phnum; i++) 134 if (ph[i].p_type == PT_LOAD && ph[i].p_align > align) 135 align = ph[i].p_align; 136 137 offset = (Elf_Addr)pax_aslr_exec_offset(epp, align); 138 if (offset < epp->ep_vm_minaddr) 139 offset = roundup(epp->ep_vm_minaddr, align); 140 if ((offset & (align - 1)) != 0) { 141 DPRINTF("bad offset=%#jx align=%#jx", 142 (uintmax_t)offset, (uintmax_t)align); 143 return EINVAL; 144 } 145 146 for (i = 0; i < eh->e_phnum; i++) 147 ph[i].p_vaddr += offset; 148 epp->ep_entryoffset = offset; 149 eh->e_entry += offset; 150 return 0; 151 } 152 153 154 int 155 elf_populate_auxv(struct lwp *l, struct exec_package *pack, char **stackp) 156 { 157 size_t len, vlen; 158 AuxInfo ai[ELF_AUX_ENTRIES], *a, *execname; 159 struct elf_args *ap; 160 char *path = l->l_proc->p_path; 161 int error; 162 163 execname = NULL; 164 a = ai; 165 166 memset(ai, 0, sizeof(ai)); 167 168 /* 169 * Push extra arguments on the stack needed by dynamically 170 * linked binaries 171 */ 172 if ((ap = (struct elf_args *)pack->ep_emul_arg)) { 173 struct vattr *vap = pack->ep_vap; 174 175 a->a_type = AT_PHDR; 176 a->a_v = ap->arg_phaddr; 177 a++; 178 179 a->a_type = AT_PHENT; 180 a->a_v = ap->arg_phentsize; 181 a++; 182 183 a->a_type = AT_PHNUM; 184 a->a_v = ap->arg_phnum; 185 a++; 186 187 a->a_type = AT_PAGESZ; 188 a->a_v = PAGE_SIZE; 189 a++; 190 191 a->a_type = AT_BASE; 192 a->a_v = ap->arg_interp; 193 a++; 194 195 a->a_type = AT_FLAGS; 196 a->a_v = 0; 197 a++; 198 199 a->a_type = AT_ENTRY; 200 a->a_v = ap->arg_entry; 201 a++; 202 203 a->a_type = AT_EUID; 204 if (vap->va_mode & S_ISUID) 205 a->a_v = vap->va_uid; 206 else 207 a->a_v = kauth_cred_geteuid(l->l_cred); 208 a++; 209 210 a->a_type = AT_RUID; 211 a->a_v = kauth_cred_getuid(l->l_cred); 212 a++; 213 214 a->a_type = AT_EGID; 215 if (vap->va_mode & S_ISGID) 216 a->a_v = vap->va_gid; 217 else 218 a->a_v = kauth_cred_getegid(l->l_cred); 219 a++; 220 221 a->a_type = AT_RGID; 222 a->a_v = kauth_cred_getgid(l->l_cred); 223 a++; 224 225 a->a_type = AT_STACKBASE; 226 a->a_v = l->l_proc->p_stackbase; 227 a++; 228 229 /* "/" means fexecve(2) could not resolve the pathname */ 230 if (path[0] == '/' && path[1] != '\0') { 231 execname = a; 232 a->a_type = AT_SUN_EXECNAME; 233 a++; 234 } 235 236 exec_free_emul_arg(pack); 237 } 238 239 a->a_type = AT_NULL; 240 a->a_v = 0; 241 a++; 242 243 vlen = (a - ai) * sizeof(ai[0]); 244 245 KASSERT(vlen <= sizeof(ai)); 246 247 if (execname) { 248 execname->a_v = (uintptr_t)(*stackp + vlen); 249 len = strlen(path) + 1; 250 if ((error = copyout(path, (*stackp + vlen), len)) != 0) 251 return error; 252 len = ALIGN(len); 253 } else { 254 len = 0; 255 } 256 257 if ((error = copyout(ai, *stackp, vlen)) != 0) 258 return error; 259 *stackp += vlen + len; 260 261 return 0; 262 } 263 264 /* 265 * Copy arguments onto the stack in the normal way, but add some 266 * extra information in case of dynamic binding. 267 */ 268 int 269 elf_copyargs(struct lwp *l, struct exec_package *pack, 270 struct ps_strings *arginfo, char **stackp, void *argp) 271 { 272 int error; 273 274 if ((error = copyargs(l, pack, arginfo, stackp, argp)) != 0) 275 return error; 276 277 return elf_populate_auxv(l, pack, stackp); 278 } 279 280 /* 281 * elf_check_header(): 282 * 283 * Check header for validity; return 0 if ok, ENOEXEC if error 284 */ 285 int 286 elf_check_header(Elf_Ehdr *eh) 287 { 288 289 if (memcmp(eh->e_ident, ELFMAG, SELFMAG) != 0 || 290 eh->e_ident[EI_CLASS] != ELFCLASS) { 291 DPRINTF("bad magic e_ident[EI_MAG0,EI_MAG3] %#x%x%x%x, " 292 "e_ident[EI_CLASS] %#x", eh->e_ident[EI_MAG0], 293 eh->e_ident[EI_MAG1], eh->e_ident[EI_MAG2], 294 eh->e_ident[EI_MAG3], eh->e_ident[EI_CLASS]); 295 return ENOEXEC; 296 } 297 298 switch (eh->e_machine) { 299 300 ELFDEFNNAME(MACHDEP_ID_CASES) 301 302 default: 303 DPRINTF("bad machine %#x", eh->e_machine); 304 return ENOEXEC; 305 } 306 307 if (ELF_EHDR_FLAGS_OK(eh) == 0) { 308 DPRINTF("bad flags %#x", eh->e_flags); 309 return ENOEXEC; 310 } 311 312 if (eh->e_shnum > ELF_MAXSHNUM || eh->e_phnum > ELF_MAXPHNUM) { 313 DPRINTF("bad shnum/phnum %#x/%#x", eh->e_shnum, eh->e_phnum); 314 return ENOEXEC; 315 } 316 317 return 0; 318 } 319 320 /* 321 * elf_load_psection(): 322 * 323 * Load a psection at the appropriate address 324 */ 325 static int 326 elf_load_psection(struct exec_vmcmd_set *vcset, struct vnode *vp, 327 const Elf_Phdr *ph, Elf_Addr *addr, u_long *size, int flags) 328 { 329 u_long msize, psize, rm, rf; 330 long diff, offset; 331 int vmprot = 0; 332 333 KASSERT(VOP_ISLOCKED(vp) != LK_NONE); 334 335 /* 336 * If the user specified an address, then we load there. 337 */ 338 if (*addr == ELFDEFNNAME(NO_ADDR)) 339 *addr = ph->p_vaddr; 340 341 if (ph->p_align > 1) { 342 /* 343 * Make sure we are virtually aligned as we are supposed to be. 344 */ 345 diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align); 346 if (*addr - diff != ELF_TRUNC(*addr, ph->p_align)) { 347 DPRINTF("bad alignment %#jx != %#jx\n", 348 (uintptr_t)(*addr - diff), 349 (uintptr_t)ELF_TRUNC(*addr, ph->p_align)); 350 return EINVAL; 351 } 352 /* 353 * But make sure to not map any pages before the start of the 354 * psection by limiting the difference to within a page. 355 */ 356 diff &= PAGE_MASK; 357 } else 358 diff = 0; 359 360 vmprot |= (ph->p_flags & PF_R) ? VM_PROT_READ : 0; 361 vmprot |= (ph->p_flags & PF_W) ? VM_PROT_WRITE : 0; 362 vmprot |= (ph->p_flags & PF_X) ? VM_PROT_EXECUTE : 0; 363 364 /* 365 * Adjust everything so it all starts on a page boundary. 366 */ 367 *addr -= diff; 368 offset = ph->p_offset - diff; 369 *size = ph->p_filesz + diff; 370 msize = ph->p_memsz + diff; 371 372 if (ph->p_align >= PAGE_SIZE) { 373 if ((ph->p_flags & PF_W) != 0) { 374 /* 375 * Because the pagedvn pager can't handle zero fill 376 * of the last data page if it's not page aligned we 377 * map the last page readvn. 378 */ 379 psize = trunc_page(*size); 380 } else { 381 psize = round_page(*size); 382 } 383 } else { 384 psize = *size; 385 } 386 387 if (psize > 0) { 388 NEW_VMCMD2(vcset, ph->p_align < PAGE_SIZE ? 389 vmcmd_map_readvn : vmcmd_map_pagedvn, psize, *addr, vp, 390 offset, vmprot, flags); 391 flags &= VMCMD_RELATIVE; 392 } 393 if (psize < *size) { 394 NEW_VMCMD2(vcset, vmcmd_map_readvn, *size - psize, 395 *addr + psize, vp, offset + psize, vmprot, flags); 396 } 397 398 /* 399 * Check if we need to extend the size of the segment (does 400 * bss extend page the next page boundary)? 401 */ 402 rm = round_page(*addr + msize); 403 rf = round_page(*addr + *size); 404 405 if (rm != rf) { 406 NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP, 407 0, vmprot, flags & VMCMD_RELATIVE); 408 *size = msize; 409 } 410 return 0; 411 } 412 413 /* 414 * elf_load_interp(): 415 * 416 * Load an interpreter pointed to by path. 417 */ 418 static int 419 elf_load_interp(struct lwp *l, struct exec_package *epp, char *path, 420 struct exec_vmcmd_set *vcset, u_long *entryoff, Elf_Addr *last) 421 { 422 int error, i; 423 struct vnode *vp; 424 Elf_Ehdr eh; 425 Elf_Phdr *ph = NULL; 426 const Elf_Phdr *base_ph; 427 const Elf_Phdr *last_ph; 428 u_long phsize; 429 Elf_Addr addr = *last; 430 struct proc *p; 431 bool use_topdown; 432 433 p = l->l_proc; 434 435 KASSERT(p->p_vmspace); 436 KASSERT(p->p_vmspace != proc0.p_vmspace); 437 438 #ifdef __USE_TOPDOWN_VM 439 use_topdown = epp->ep_flags & EXEC_TOPDOWN_VM; 440 #else 441 use_topdown = false; 442 #endif 443 444 /* 445 * 1. open file 446 * 2. read filehdr 447 * 3. map text, data, and bss out of it using VM_* 448 */ 449 vp = epp->ep_interp; 450 if (vp == NULL) { 451 error = emul_find_interp(l, epp, path); 452 if (error != 0) 453 return error; 454 vp = epp->ep_interp; 455 } 456 /* We'll tidy this ourselves - otherwise we have locking issues */ 457 epp->ep_interp = NULL; 458 vn_lock(vp, LK_SHARED | LK_RETRY); 459 460 /* 461 * Similarly, if it's not marked as executable, or it's not a regular 462 * file, we don't allow it to be used. 463 */ 464 if (vp->v_type != VREG) { 465 error = EACCES; 466 goto bad; 467 } 468 if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0) 469 goto bad; 470 471 /* 472 * Check mount point. Though we're not trying to exec this binary, 473 * we will be executing code from it, so if the mount point 474 * disallows execution or set-id-ness, we punt or kill the set-id. 475 */ 476 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 477 error = EACCES; 478 goto bad; 479 } 480 if (vp->v_mount->mnt_flag & MNT_NOSUID) 481 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID); 482 483 error = vn_marktext(vp); 484 if (error) 485 goto bad; 486 487 error = exec_read(l, vp, 0, &eh, sizeof(eh), IO_NODELOCKED); 488 if (error != 0) 489 goto bad; 490 491 if ((error = elf_check_header(&eh)) != 0) 492 goto bad; 493 if (eh.e_type != ET_DYN || eh.e_phnum == 0) { 494 DPRINTF("bad interpreter type %#x", eh.e_type); 495 error = ENOEXEC; 496 goto bad; 497 } 498 499 phsize = eh.e_phnum * sizeof(Elf_Phdr); 500 ph = kmem_alloc(phsize, KM_SLEEP); 501 502 error = exec_read(l, vp, eh.e_phoff, ph, phsize, IO_NODELOCKED); 503 if (error != 0) 504 goto bad; 505 506 #ifdef ELF_INTERP_NON_RELOCATABLE 507 /* 508 * Evil hack: Only MIPS should be non-relocatable, and the 509 * psections should have a high address (typically 0x5ffe0000). 510 * If it's now relocatable, it should be linked at 0 and the 511 * psections should have zeros in the upper part of the address. 512 * Otherwise, force the load at the linked address. 513 */ 514 if (*last == ELF_LINK_ADDR && (ph->p_vaddr & 0xffff0000) == 0) 515 *last = ELFDEFNNAME(NO_ADDR); 516 #endif 517 518 /* 519 * If no position to load the interpreter was set by a probe 520 * function, pick the same address that a non-fixed mmap(0, ..) 521 * would (i.e. something safely out of the way). 522 */ 523 if (*last == ELFDEFNNAME(NO_ADDR)) { 524 u_long limit = 0; 525 /* 526 * Find the start and ending addresses of the psections to 527 * be loaded. This will give us the size. 528 */ 529 for (i = 0, base_ph = NULL; i < eh.e_phnum; i++) { 530 if (ph[i].p_type == PT_LOAD) { 531 u_long psize = ph[i].p_vaddr + ph[i].p_memsz; 532 if (base_ph == NULL) 533 base_ph = &ph[i]; 534 if (psize > limit) 535 limit = psize; 536 } 537 } 538 539 if (base_ph == NULL) { 540 DPRINTF("no interpreter loadable sections"); 541 error = ENOEXEC; 542 goto bad; 543 } 544 545 /* 546 * Now compute the size and load address. 547 */ 548 addr = (*epp->ep_esch->es_emul->e_vm_default_addr)(p, 549 epp->ep_daddr, 550 round_page(limit) - trunc_page(base_ph->p_vaddr), 551 use_topdown); 552 addr += (Elf_Addr)pax_aslr_rtld_offset(epp, base_ph->p_align, 553 use_topdown); 554 } else { 555 addr = *last; /* may be ELF_LINK_ADDR */ 556 } 557 558 /* 559 * Load all the necessary sections 560 */ 561 for (i = 0, base_ph = NULL, last_ph = NULL; i < eh.e_phnum; i++) { 562 switch (ph[i].p_type) { 563 case PT_LOAD: { 564 u_long size; 565 int flags; 566 567 if (base_ph == NULL) { 568 /* 569 * First encountered psection is always the 570 * base psection. Make sure it's aligned 571 * properly (align down for topdown and align 572 * upwards for not topdown). 573 */ 574 base_ph = &ph[i]; 575 flags = VMCMD_BASE; 576 if (addr == ELF_LINK_ADDR) 577 addr = ph[i].p_vaddr; 578 if (use_topdown) 579 addr = ELF_TRUNC(addr, ph[i].p_align); 580 else 581 addr = ELF_ROUND(addr, ph[i].p_align); 582 } else { 583 u_long limit = round_page(last_ph->p_vaddr 584 + last_ph->p_memsz); 585 u_long base = trunc_page(ph[i].p_vaddr); 586 587 /* 588 * If there is a gap in between the psections, 589 * map it as inaccessible so nothing else 590 * mmap'ed will be placed there. 591 */ 592 if (limit != base) { 593 NEW_VMCMD2(vcset, vmcmd_map_zero, 594 base - limit, 595 limit - base_ph->p_vaddr, NULLVP, 596 0, VM_PROT_NONE, VMCMD_RELATIVE); 597 } 598 599 addr = ph[i].p_vaddr - base_ph->p_vaddr; 600 flags = VMCMD_RELATIVE; 601 } 602 last_ph = &ph[i]; 603 if ((error = elf_load_psection(vcset, vp, &ph[i], &addr, 604 &size, flags)) != 0) 605 goto bad; 606 /* 607 * If entry is within this psection then this 608 * must contain the .text section. *entryoff is 609 * relative to the base psection. 610 */ 611 if (eh.e_entry >= ph[i].p_vaddr && 612 eh.e_entry < (ph[i].p_vaddr + size)) { 613 *entryoff = eh.e_entry - base_ph->p_vaddr; 614 } 615 addr += size; 616 break; 617 } 618 619 default: 620 break; 621 } 622 } 623 624 kmem_free(ph, phsize); 625 /* 626 * This value is ignored if TOPDOWN. 627 */ 628 *last = addr; 629 vput(vp); 630 return 0; 631 632 bad: 633 if (ph != NULL) 634 kmem_free(ph, phsize); 635 vput(vp); 636 return error; 637 } 638 639 /* 640 * exec_elf_makecmds(): Prepare an Elf binary's exec package 641 * 642 * First, set of the various offsets/lengths in the exec package. 643 * 644 * Then, mark the text image busy (so it can be demand paged) or error 645 * out if this is not possible. Finally, set up vmcmds for the 646 * text, data, bss, and stack segments. 647 */ 648 int 649 exec_elf_makecmds(struct lwp *l, struct exec_package *epp) 650 { 651 Elf_Ehdr *eh = epp->ep_hdr; 652 Elf_Phdr *ph, *pp; 653 Elf_Addr phdr = 0, computed_phdr = 0, pos = 0, end_text = 0; 654 int error, i; 655 char *interp = NULL; 656 u_long phsize; 657 struct elf_args *ap; 658 bool is_dyn = false; 659 660 if (epp->ep_hdrvalid < sizeof(Elf_Ehdr)) { 661 DPRINTF("small header %#x", epp->ep_hdrvalid); 662 return ENOEXEC; 663 } 664 if ((error = elf_check_header(eh)) != 0) 665 return error; 666 667 if (eh->e_type == ET_DYN) 668 /* PIE, and some libs have an entry point */ 669 is_dyn = true; 670 else if (eh->e_type != ET_EXEC) { 671 DPRINTF("bad type %#x", eh->e_type); 672 return ENOEXEC; 673 } 674 675 if (eh->e_phnum == 0) { 676 DPRINTF("no program headers"); 677 return ENOEXEC; 678 } 679 680 /* XXX only LK_EXCLUSIVE to match all others - allow spinning */ 681 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY); 682 error = vn_marktext(epp->ep_vp); 683 if (error) { 684 VOP_UNLOCK(epp->ep_vp); 685 return error; 686 } 687 688 /* 689 * Allocate space to hold all the program headers, and read them 690 * from the file 691 */ 692 phsize = eh->e_phnum * sizeof(Elf_Phdr); 693 ph = kmem_alloc(phsize, KM_SLEEP); 694 695 error = exec_read(l, epp->ep_vp, eh->e_phoff, ph, phsize, 696 IO_NODELOCKED); 697 if (error != 0) { 698 VOP_UNLOCK(epp->ep_vp); 699 goto bad; 700 } 701 702 epp->ep_taddr = epp->ep_tsize = ELFDEFNNAME(NO_ADDR); 703 epp->ep_daddr = epp->ep_dsize = ELFDEFNNAME(NO_ADDR); 704 705 for (i = 0; i < eh->e_phnum; i++) { 706 pp = &ph[i]; 707 if (pp->p_type == PT_INTERP) { 708 if (pp->p_filesz < 2 || pp->p_filesz > MAXPATHLEN) { 709 DPRINTF("bad interpreter namelen %#jx", 710 (uintmax_t)pp->p_filesz); 711 error = ENOEXEC; 712 VOP_UNLOCK(epp->ep_vp); 713 goto bad; 714 } 715 interp = PNBUF_GET(); 716 error = exec_read(l, epp->ep_vp, pp->p_offset, interp, 717 pp->p_filesz, IO_NODELOCKED); 718 if (error != 0) { 719 VOP_UNLOCK(epp->ep_vp); 720 goto bad; 721 } 722 /* Ensure interp is NUL-terminated and of the expected length */ 723 if (strnlen(interp, pp->p_filesz) != pp->p_filesz - 1) { 724 DPRINTF("bad interpreter name"); 725 error = ENOEXEC; 726 VOP_UNLOCK(epp->ep_vp); 727 goto bad; 728 } 729 break; 730 } 731 } 732 733 /* 734 * On the same architecture, we may be emulating different systems. 735 * See which one will accept this executable. 736 * 737 * Probe functions would normally see if the interpreter (if any) 738 * exists. Emulation packages may possibly replace the interpreter in 739 * interp with a changed path (/emul/xxx/<path>). 740 */ 741 pos = ELFDEFNNAME(NO_ADDR); 742 if (epp->ep_esch->u.elf_probe_func) { 743 vaddr_t startp = (vaddr_t)pos; 744 745 error = (*epp->ep_esch->u.elf_probe_func)(l, epp, eh, interp, 746 &startp); 747 if (error) { 748 VOP_UNLOCK(epp->ep_vp); 749 goto bad; 750 } 751 pos = (Elf_Addr)startp; 752 } 753 754 if (is_dyn && (error = elf_placedynexec(epp, eh, ph)) != 0) { 755 VOP_UNLOCK(epp->ep_vp); 756 goto bad; 757 } 758 759 /* 760 * Load all the necessary sections 761 */ 762 for (i = 0; i < eh->e_phnum; i++) { 763 Elf_Addr addr = ELFDEFNNAME(NO_ADDR); 764 u_long size = 0; 765 766 switch (ph[i].p_type) { 767 case PT_LOAD: 768 if ((error = elf_load_psection(&epp->ep_vmcmds, 769 epp->ep_vp, &ph[i], &addr, &size, VMCMD_FIXED)) 770 != 0) { 771 VOP_UNLOCK(epp->ep_vp); 772 goto bad; 773 } 774 775 /* 776 * Consider this as text segment, if it is executable. 777 * If there is more than one text segment, pick the 778 * largest. 779 */ 780 if (ph[i].p_flags & PF_X) { 781 if (epp->ep_taddr == ELFDEFNNAME(NO_ADDR) || 782 size > epp->ep_tsize) { 783 epp->ep_taddr = addr; 784 epp->ep_tsize = size; 785 } 786 end_text = addr + size; 787 } else { 788 epp->ep_daddr = addr; 789 epp->ep_dsize = size; 790 } 791 if (ph[i].p_offset == 0) { 792 computed_phdr = ph[i].p_vaddr + eh->e_phoff; 793 } 794 break; 795 796 case PT_SHLIB: 797 /* SCO has these sections. */ 798 case PT_INTERP: 799 /* Already did this one. */ 800 case PT_DYNAMIC: 801 case PT_NOTE: 802 break; 803 case PT_PHDR: 804 /* Note address of program headers (in text segment) */ 805 phdr = ph[i].p_vaddr; 806 break; 807 808 default: 809 /* 810 * Not fatal; we don't need to understand everything. 811 */ 812 break; 813 } 814 } 815 816 /* Now done with the vnode. */ 817 VOP_UNLOCK(epp->ep_vp); 818 819 if (epp->ep_vmcmds.evs_used == 0) { 820 /* No VMCMD; there was no PT_LOAD section, or those 821 * sections were empty */ 822 DPRINTF("no vmcommands"); 823 error = ENOEXEC; 824 goto bad; 825 } 826 827 if (epp->ep_daddr == ELFDEFNNAME(NO_ADDR)) { 828 epp->ep_daddr = round_page(end_text); 829 epp->ep_dsize = 0; 830 } 831 832 /* 833 * Check if we found a dynamically linked binary and arrange to load 834 * its interpreter 835 */ 836 if (interp) { 837 u_int nused = epp->ep_vmcmds.evs_used; 838 u_long interp_offset = 0; 839 840 if ((error = elf_load_interp(l, epp, interp, 841 &epp->ep_vmcmds, &interp_offset, &pos)) != 0) { 842 goto bad; 843 } 844 if (epp->ep_vmcmds.evs_used == nused) { 845 /* elf_load_interp() has not set up any new VMCMD */ 846 DPRINTF("no vmcommands for interpreter"); 847 error = ENOEXEC; 848 goto bad; 849 } 850 851 ap = kmem_alloc(sizeof(*ap), KM_SLEEP); 852 ap->arg_interp = epp->ep_vmcmds.evs_cmds[nused].ev_addr; 853 epp->ep_entryoffset = interp_offset; 854 epp->ep_entry = ap->arg_interp + interp_offset; 855 PNBUF_PUT(interp); 856 interp = NULL; 857 } else { 858 epp->ep_entry = eh->e_entry; 859 if (epp->ep_flags & EXEC_FORCEAUX) { 860 ap = kmem_zalloc(sizeof(*ap), KM_SLEEP); 861 ap->arg_interp = (vaddr_t)NULL; 862 } else { 863 ap = NULL; 864 } 865 } 866 867 if (ap) { 868 ap->arg_phaddr = phdr ? phdr : computed_phdr; 869 ap->arg_phentsize = eh->e_phentsize; 870 ap->arg_phnum = eh->e_phnum; 871 ap->arg_entry = eh->e_entry; 872 epp->ep_emul_arg = ap; 873 epp->ep_emul_arg_free = elf_free_emul_arg; 874 } 875 876 #ifdef ELF_MAP_PAGE_ZERO 877 /* Dell SVR4 maps page zero, yeuch! */ 878 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, PAGE_SIZE, 0, 879 epp->ep_vp, 0, VM_PROT_READ); 880 #endif 881 882 error = (*epp->ep_esch->es_setup_stack)(l, epp); 883 if (error) 884 goto bad; 885 886 kmem_free(ph, phsize); 887 return 0; 888 889 bad: 890 if (interp) 891 PNBUF_PUT(interp); 892 exec_free_emul_arg(epp); 893 kmem_free(ph, phsize); 894 kill_vmcmds(&epp->ep_vmcmds); 895 return error; 896 } 897 898 int 899 netbsd_elf_signature(struct lwp *l, struct exec_package *epp, 900 Elf_Ehdr *eh) 901 { 902 size_t i; 903 Elf_Phdr *ph; 904 size_t phsize; 905 char *nbuf; 906 int error; 907 int isnetbsd = 0; 908 909 epp->ep_pax_flags = 0; 910 911 if (eh->e_phnum > ELF_MAXPHNUM || eh->e_phnum == 0) { 912 DPRINTF("no signature %#x", eh->e_phnum); 913 return ENOEXEC; 914 } 915 916 phsize = eh->e_phnum * sizeof(Elf_Phdr); 917 ph = kmem_alloc(phsize, KM_SLEEP); 918 error = exec_read(l, epp->ep_vp, eh->e_phoff, ph, phsize, 919 IO_NODELOCKED); 920 if (error) 921 goto out; 922 923 nbuf = kmem_alloc(ELF_MAXNOTESIZE, KM_SLEEP); 924 for (i = 0; i < eh->e_phnum; i++) { 925 const char *nptr; 926 size_t nlen; 927 928 if (ph[i].p_type != PT_NOTE || 929 ph[i].p_filesz > ELF_MAXNOTESIZE) 930 continue; 931 932 nlen = ph[i].p_filesz; 933 error = exec_read(l, epp->ep_vp, ph[i].p_offset, nbuf, nlen, 934 IO_NODELOCKED); 935 if (error) 936 continue; 937 938 nptr = nbuf; 939 while (nlen > 0) { 940 const Elf_Nhdr *np; 941 const char *ndata, *ndesc; 942 943 /* note header */ 944 np = (const Elf_Nhdr *)nptr; 945 if (nlen < sizeof(*np)) { 946 break; 947 } 948 nptr += sizeof(*np); 949 nlen -= sizeof(*np); 950 951 /* note name */ 952 ndata = nptr; 953 if (nlen < roundup(np->n_namesz, 4)) { 954 break; 955 } 956 nptr += roundup(np->n_namesz, 4); 957 nlen -= roundup(np->n_namesz, 4); 958 959 /* note description */ 960 ndesc = nptr; 961 if (nlen < roundup(np->n_descsz, 4)) { 962 break; 963 } 964 nptr += roundup(np->n_descsz, 4); 965 nlen -= roundup(np->n_descsz, 4); 966 967 isnetbsd |= netbsd_elf_note(epp, np, ndata, ndesc); 968 } 969 } 970 kmem_free(nbuf, ELF_MAXNOTESIZE); 971 972 error = isnetbsd ? 0 : ENOEXEC; 973 #ifdef DEBUG_ELF 974 if (error) 975 DPRINTF("not netbsd"); 976 #endif 977 out: 978 kmem_free(ph, phsize); 979 return error; 980 } 981 982 int 983 netbsd_elf_note(struct exec_package *epp, 984 const Elf_Nhdr *np, const char *ndata, const char *ndesc) 985 { 986 int isnetbsd = 0; 987 988 #ifdef DIAGNOSTIC 989 const char *badnote; 990 #define BADNOTE(n) badnote = (n) 991 #else 992 #define BADNOTE(n) 993 #endif 994 995 switch (np->n_type) { 996 case ELF_NOTE_TYPE_NETBSD_TAG: 997 /* It is us */ 998 if (np->n_namesz == ELF_NOTE_NETBSD_NAMESZ && 999 np->n_descsz == ELF_NOTE_NETBSD_DESCSZ && 1000 memcmp(ndata, ELF_NOTE_NETBSD_NAME, 1001 ELF_NOTE_NETBSD_NAMESZ) == 0) { 1002 memcpy(&epp->ep_osversion, ndesc, 1003 ELF_NOTE_NETBSD_DESCSZ); 1004 isnetbsd = 1; 1005 break; 1006 } 1007 1008 /* 1009 * Ignore SuSE tags; SuSE's n_type is the same the 1010 * NetBSD one. 1011 */ 1012 if (np->n_namesz == ELF_NOTE_SUSE_NAMESZ && 1013 memcmp(ndata, ELF_NOTE_SUSE_NAME, 1014 ELF_NOTE_SUSE_NAMESZ) == 0) 1015 break; 1016 /* 1017 * Ignore old GCC 1018 */ 1019 if (np->n_namesz == ELF_NOTE_OGCC_NAMESZ && 1020 memcmp(ndata, ELF_NOTE_OGCC_NAME, 1021 ELF_NOTE_OGCC_NAMESZ) == 0) 1022 break; 1023 BADNOTE("NetBSD tag"); 1024 goto bad; 1025 1026 case ELF_NOTE_TYPE_PAX_TAG: 1027 if (np->n_namesz == ELF_NOTE_PAX_NAMESZ && 1028 np->n_descsz == ELF_NOTE_PAX_DESCSZ && 1029 memcmp(ndata, ELF_NOTE_PAX_NAME, 1030 ELF_NOTE_PAX_NAMESZ) == 0) { 1031 uint32_t flags; 1032 memcpy(&flags, ndesc, sizeof(flags)); 1033 /* Convert the flags and insert them into 1034 * the exec package. */ 1035 pax_setup_elf_flags(epp, flags); 1036 break; 1037 } 1038 BADNOTE("PaX tag"); 1039 goto bad; 1040 1041 case ELF_NOTE_TYPE_MARCH_TAG: 1042 /* Copy the machine arch into the package. */ 1043 if (np->n_namesz == ELF_NOTE_MARCH_NAMESZ 1044 && memcmp(ndata, ELF_NOTE_MARCH_NAME, 1045 ELF_NOTE_MARCH_NAMESZ) == 0) { 1046 /* Do not truncate the buffer */ 1047 if (np->n_descsz > sizeof(epp->ep_machine_arch)) { 1048 BADNOTE("description size limit"); 1049 goto bad; 1050 } 1051 /* 1052 * Ensure ndesc is NUL-terminated and of the 1053 * expected length. 1054 */ 1055 if (strnlen(ndesc, np->n_descsz) + 1 != 1056 np->n_descsz) { 1057 BADNOTE("description size"); 1058 goto bad; 1059 } 1060 strlcpy(epp->ep_machine_arch, ndesc, 1061 sizeof(epp->ep_machine_arch)); 1062 break; 1063 } 1064 BADNOTE("march tag"); 1065 goto bad; 1066 1067 case ELF_NOTE_TYPE_MCMODEL_TAG: 1068 /* arch specific check for code model */ 1069 #ifdef ELF_MD_MCMODEL_CHECK 1070 if (np->n_namesz == ELF_NOTE_MCMODEL_NAMESZ 1071 && memcmp(ndata, ELF_NOTE_MCMODEL_NAME, 1072 ELF_NOTE_MCMODEL_NAMESZ) == 0) { 1073 ELF_MD_MCMODEL_CHECK(epp, ndesc, np->n_descsz); 1074 break; 1075 } 1076 BADNOTE("mcmodel tag"); 1077 goto bad; 1078 #endif 1079 break; 1080 1081 case ELF_NOTE_TYPE_SUSE_VERSION_TAG: 1082 break; 1083 1084 case ELF_NOTE_TYPE_GO_BUILDID_TAG: 1085 break; 1086 1087 case ELF_NOTE_TYPE_NETBSD_EMUL_TAG: 1088 /* Ancient NetBSD version tag */ 1089 break; 1090 1091 default: 1092 BADNOTE("unknown tag"); 1093 bad: 1094 #ifdef DIAGNOSTIC 1095 /* Ignore GNU tags */ 1096 if (np->n_namesz == ELF_NOTE_GNU_NAMESZ && 1097 memcmp(ndata, ELF_NOTE_GNU_NAME, 1098 ELF_NOTE_GNU_NAMESZ) == 0) 1099 break; 1100 1101 int ns = (int)np->n_namesz; 1102 printf("%s: Unknown elf note type %d (%s): " 1103 "[namesz=%d, descsz=%d name=%-*.*s]\n", 1104 epp->ep_kname, np->n_type, badnote, np->n_namesz, 1105 np->n_descsz, ns, ns, ndata); 1106 #endif 1107 break; 1108 } 1109 1110 return isnetbsd; 1111 } 1112 1113 int 1114 netbsd_elf_probe(struct lwp *l, struct exec_package *epp, void *eh, char *itp, 1115 vaddr_t *pos) 1116 { 1117 int error; 1118 1119 if ((error = netbsd_elf_signature(l, epp, eh)) != 0) 1120 return error; 1121 #ifdef ELF_MD_PROBE_FUNC 1122 if ((error = ELF_MD_PROBE_FUNC(l, epp, eh, itp, pos)) != 0) 1123 return error; 1124 #elif defined(ELF_INTERP_NON_RELOCATABLE) 1125 *pos = ELF_LINK_ADDR; 1126 #endif 1127 epp->ep_flags |= EXEC_FORCEAUX; 1128 return 0; 1129 } 1130 1131 void 1132 elf_free_emul_arg(void *arg) 1133 { 1134 struct elf_args *ap = arg; 1135 KASSERT(ap != NULL); 1136 kmem_free(ap, sizeof(*ap)); 1137 } 1138