FreeBSD/Linux Kernel Cross Reference
sys/kern/exec_elf.c
1 /* $NetBSD: exec_elf.c,v 1.104 2022/10/26 23:20:36 riastradh 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 1.104 2022/10/26 23:20:36 riastradh 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 #define elf_check_header ELFNAME(check_header)
89 #define elf_copyargs ELFNAME(copyargs)
90 #define elf_populate_auxv ELFNAME(populate_auxv)
91 #define elf_load_interp ELFNAME(load_interp)
92 #define elf_load_psection ELFNAME(load_psection)
93 #define exec_elf_makecmds ELFNAME2(exec,makecmds)
94 #define netbsd_elf_signature ELFNAME2(netbsd,signature)
95 #define netbsd_elf_note ELFNAME2(netbsd,note)
96 #define netbsd_elf_probe ELFNAME2(netbsd,probe)
97 #define coredump ELFNAMEEND(coredump)
98 #define elf_free_emul_arg ELFNAME(free_emul_arg)
99
100 static int
101 elf_load_interp(struct lwp *, struct exec_package *, char *,
102 struct exec_vmcmd_set *, u_long *, Elf_Addr *);
103 static int
104 elf_load_psection(struct exec_vmcmd_set *, struct vnode *, const Elf_Phdr *,
105 Elf_Addr *, u_long *, int);
106
107 int netbsd_elf_signature(struct lwp *, struct exec_package *, Elf_Ehdr *);
108 int netbsd_elf_note(struct exec_package *, const Elf_Nhdr *, const char *,
109 const char *);
110 int netbsd_elf_probe(struct lwp *, struct exec_package *, void *, char *,
111 vaddr_t *);
112
113 static void elf_free_emul_arg(void *);
114
115 #ifdef DEBUG_ELF
116 #define DPRINTF(a, ...) printf("%s: " a "\n", __func__, ##__VA_ARGS__)
117 #else
118 #define DPRINTF(a, ...)
119 #endif
120
121 /* round up and down to page boundaries. */
122 #define ELF_ROUND(a, b) (((a) + (b) - 1) & ~((b) - 1))
123 #define ELF_TRUNC(a, b) ((a) & ~((b) - 1))
124
125 static int
126 elf_placedynexec(struct exec_package *epp, Elf_Ehdr *eh, Elf_Phdr *ph)
127 {
128 Elf_Addr align, offset;
129 int i;
130
131 for (align = 1, i = 0; i < eh->e_phnum; i++)
132 if (ph[i].p_type == PT_LOAD && ph[i].p_align > align)
133 align = ph[i].p_align;
134
135 offset = (Elf_Addr)pax_aslr_exec_offset(epp, align);
136 if (offset < epp->ep_vm_minaddr)
137 offset = roundup(epp->ep_vm_minaddr, align);
138 if ((offset & (align - 1)) != 0) {
139 DPRINTF("bad offset=%#jx align=%#jx",
140 (uintmax_t)offset, (uintmax_t)align);
141 return EINVAL;
142 }
143
144 for (i = 0; i < eh->e_phnum; i++)
145 ph[i].p_vaddr += offset;
146 epp->ep_entryoffset = offset;
147 eh->e_entry += offset;
148 return 0;
149 }
150
151
152 int
153 elf_populate_auxv(struct lwp *l, struct exec_package *pack, char **stackp)
154 {
155 size_t len, vlen;
156 AuxInfo ai[ELF_AUX_ENTRIES], *a, *execname;
157 struct elf_args *ap;
158 char *path = l->l_proc->p_path;
159 int error;
160
161 execname = NULL;
162 a = ai;
163
164 memset(ai, 0, sizeof(ai));
165
166 /*
167 * Push extra arguments on the stack needed by dynamically
168 * linked binaries
169 */
170 if ((ap = (struct elf_args *)pack->ep_emul_arg)) {
171 struct vattr *vap = pack->ep_vap;
172
173 a->a_type = AT_PHDR;
174 a->a_v = ap->arg_phaddr;
175 a++;
176
177 a->a_type = AT_PHENT;
178 a->a_v = ap->arg_phentsize;
179 a++;
180
181 a->a_type = AT_PHNUM;
182 a->a_v = ap->arg_phnum;
183 a++;
184
185 a->a_type = AT_PAGESZ;
186 a->a_v = PAGE_SIZE;
187 a++;
188
189 a->a_type = AT_BASE;
190 a->a_v = ap->arg_interp;
191 a++;
192
193 a->a_type = AT_FLAGS;
194 a->a_v = 0;
195 a++;
196
197 a->a_type = AT_ENTRY;
198 a->a_v = ap->arg_entry;
199 a++;
200
201 a->a_type = AT_EUID;
202 if (vap->va_mode & S_ISUID)
203 a->a_v = vap->va_uid;
204 else
205 a->a_v = kauth_cred_geteuid(l->l_cred);
206 a++;
207
208 a->a_type = AT_RUID;
209 a->a_v = kauth_cred_getuid(l->l_cred);
210 a++;
211
212 a->a_type = AT_EGID;
213 if (vap->va_mode & S_ISGID)
214 a->a_v = vap->va_gid;
215 else
216 a->a_v = kauth_cred_getegid(l->l_cred);
217 a++;
218
219 a->a_type = AT_RGID;
220 a->a_v = kauth_cred_getgid(l->l_cred);
221 a++;
222
223 a->a_type = AT_STACKBASE;
224 a->a_v = l->l_proc->p_stackbase;
225 a++;
226
227 /* "/" means fexecve(2) could not resolve the pathname */
228 if (path[0] == '/' && path[1] != '\0') {
229 execname = a;
230 a->a_type = AT_SUN_EXECNAME;
231 a++;
232 }
233
234 exec_free_emul_arg(pack);
235 }
236
237 a->a_type = AT_NULL;
238 a->a_v = 0;
239 a++;
240
241 vlen = (a - ai) * sizeof(ai[0]);
242
243 KASSERT(vlen <= sizeof(ai));
244
245 if (execname) {
246 execname->a_v = (uintptr_t)(*stackp + vlen);
247 len = strlen(path) + 1;
248 if ((error = copyout(path, (*stackp + vlen), len)) != 0)
249 return error;
250 len = ALIGN(len);
251 } else {
252 len = 0;
253 }
254
255 if ((error = copyout(ai, *stackp, vlen)) != 0)
256 return error;
257 *stackp += vlen + len;
258
259 return 0;
260 }
261
262 /*
263 * Copy arguments onto the stack in the normal way, but add some
264 * extra information in case of dynamic binding.
265 */
266 int
267 elf_copyargs(struct lwp *l, struct exec_package *pack,
268 struct ps_strings *arginfo, char **stackp, void *argp)
269 {
270 int error;
271
272 if ((error = copyargs(l, pack, arginfo, stackp, argp)) != 0)
273 return error;
274
275 return elf_populate_auxv(l, pack, stackp);
276 }
277
278 /*
279 * elf_check_header():
280 *
281 * Check header for validity; return 0 if ok, ENOEXEC if error
282 */
283 int
284 elf_check_header(Elf_Ehdr *eh)
285 {
286
287 if (memcmp(eh->e_ident, ELFMAG, SELFMAG) != 0 ||
288 eh->e_ident[EI_CLASS] != ELFCLASS) {
289 DPRINTF("bad magic e_ident[EI_MAG0,EI_MAG3] %#x%x%x%x, "
290 "e_ident[EI_CLASS] %#x", eh->e_ident[EI_MAG0],
291 eh->e_ident[EI_MAG1], eh->e_ident[EI_MAG2],
292 eh->e_ident[EI_MAG3], eh->e_ident[EI_CLASS]);
293 return ENOEXEC;
294 }
295
296 switch (eh->e_machine) {
297
298 ELFDEFNNAME(MACHDEP_ID_CASES)
299
300 default:
301 DPRINTF("bad machine %#x", eh->e_machine);
302 return ENOEXEC;
303 }
304
305 if (ELF_EHDR_FLAGS_OK(eh) == 0) {
306 DPRINTF("bad flags %#x", eh->e_flags);
307 return ENOEXEC;
308 }
309
310 if (eh->e_shnum > ELF_MAXSHNUM || eh->e_phnum > ELF_MAXPHNUM) {
311 DPRINTF("bad shnum/phnum %#x/%#x", eh->e_shnum, eh->e_phnum);
312 return ENOEXEC;
313 }
314
315 return 0;
316 }
317
318 /*
319 * elf_load_psection():
320 *
321 * Load a psection at the appropriate address
322 */
323 static int
324 elf_load_psection(struct exec_vmcmd_set *vcset, struct vnode *vp,
325 const Elf_Phdr *ph, Elf_Addr *addr, u_long *size, int flags)
326 {
327 u_long msize, psize, rm, rf;
328 long diff, offset;
329 int vmprot = 0;
330
331 KASSERT(VOP_ISLOCKED(vp) != LK_NONE);
332
333 /*
334 * If the user specified an address, then we load there.
335 */
336 if (*addr == ELFDEFNNAME(NO_ADDR))
337 *addr = ph->p_vaddr;
338
339 if (ph->p_align > 1) {
340 /*
341 * Make sure we are virtually aligned as we are supposed to be.
342 */
343 diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align);
344 if (*addr - diff != ELF_TRUNC(*addr, ph->p_align)) {
345 DPRINTF("bad alignment %#jx != %#jx\n",
346 (uintptr_t)(*addr - diff),
347 (uintptr_t)ELF_TRUNC(*addr, ph->p_align));
348 return EINVAL;
349 }
350 /*
351 * But make sure to not map any pages before the start of the
352 * psection by limiting the difference to within a page.
353 */
354 diff &= PAGE_MASK;
355 } else
356 diff = 0;
357
358 vmprot |= (ph->p_flags & PF_R) ? VM_PROT_READ : 0;
359 vmprot |= (ph->p_flags & PF_W) ? VM_PROT_WRITE : 0;
360 vmprot |= (ph->p_flags & PF_X) ? VM_PROT_EXECUTE : 0;
361
362 /*
363 * Adjust everything so it all starts on a page boundary.
364 */
365 *addr -= diff;
366 offset = ph->p_offset - diff;
367 *size = ph->p_filesz + diff;
368 msize = ph->p_memsz + diff;
369
370 if (ph->p_align >= PAGE_SIZE) {
371 if ((ph->p_flags & PF_W) != 0) {
372 /*
373 * Because the pagedvn pager can't handle zero fill
374 * of the last data page if it's not page aligned we
375 * map the last page readvn.
376 */
377 psize = trunc_page(*size);
378 } else {
379 psize = round_page(*size);
380 }
381 } else {
382 psize = *size;
383 }
384
385 if (psize > 0) {
386 NEW_VMCMD2(vcset, ph->p_align < PAGE_SIZE ?
387 vmcmd_map_readvn : vmcmd_map_pagedvn, psize, *addr, vp,
388 offset, vmprot, flags);
389 flags &= VMCMD_RELATIVE;
390 }
391 if (psize < *size) {
392 NEW_VMCMD2(vcset, vmcmd_map_readvn, *size - psize,
393 *addr + psize, vp, offset + psize, vmprot, flags);
394 }
395
396 /*
397 * Check if we need to extend the size of the segment (does
398 * bss extend page the next page boundary)?
399 */
400 rm = round_page(*addr + msize);
401 rf = round_page(*addr + *size);
402
403 if (rm != rf) {
404 NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP,
405 0, vmprot, flags & VMCMD_RELATIVE);
406 *size = msize;
407 }
408 return 0;
409 }
410
411 /*
412 * elf_load_interp():
413 *
414 * Load an interpreter pointed to by path.
415 */
416 static int
417 elf_load_interp(struct lwp *l, struct exec_package *epp, char *path,
418 struct exec_vmcmd_set *vcset, u_long *entryoff, Elf_Addr *last)
419 {
420 int error, i;
421 struct vnode *vp;
422 Elf_Ehdr eh;
423 Elf_Phdr *ph = NULL;
424 const Elf_Phdr *base_ph;
425 const Elf_Phdr *last_ph;
426 u_long phsize;
427 Elf_Addr addr = *last;
428 struct proc *p;
429 bool use_topdown;
430
431 p = l->l_proc;
432
433 KASSERT(p->p_vmspace);
434 KASSERT(p->p_vmspace != proc0.p_vmspace);
435
436 #ifdef __USE_TOPDOWN_VM
437 use_topdown = epp->ep_flags & EXEC_TOPDOWN_VM;
438 #else
439 use_topdown = false;
440 #endif
441
442 /*
443 * 1. open file
444 * 2. read filehdr
445 * 3. map text, data, and bss out of it using VM_*
446 */
447 vp = epp->ep_interp;
448 if (vp == NULL) {
449 error = emul_find_interp(l, epp, path);
450 if (error != 0)
451 return error;
452 vp = epp->ep_interp;
453 }
454 /* We'll tidy this ourselves - otherwise we have locking issues */
455 epp->ep_interp = NULL;
456 vn_lock(vp, LK_SHARED | LK_RETRY);
457
458 /*
459 * Similarly, if it's not marked as executable, or it's not a regular
460 * file, we don't allow it to be used.
461 */
462 if (vp->v_type != VREG) {
463 error = EACCES;
464 goto bad;
465 }
466 if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
467 goto bad;
468
469 /*
470 * Check mount point. Though we're not trying to exec this binary,
471 * we will be executing code from it, so if the mount point
472 * disallows execution or set-id-ness, we punt or kill the set-id.
473 */
474 if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
475 error = EACCES;
476 goto bad;
477 }
478 if (vp->v_mount->mnt_flag & MNT_NOSUID)
479 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
480
481 error = vn_marktext(vp);
482 if (error)
483 goto bad;
484
485 error = exec_read(l, vp, 0, &eh, sizeof(eh), IO_NODELOCKED);
486 if (error != 0)
487 goto bad;
488
489 if ((error = elf_check_header(&eh)) != 0)
490 goto bad;
491 if (eh.e_type != ET_DYN || eh.e_phnum == 0) {
492 DPRINTF("bad interpreter type %#x", eh.e_type);
493 error = ENOEXEC;
494 goto bad;
495 }
496
497 phsize = eh.e_phnum * sizeof(Elf_Phdr);
498 ph = kmem_alloc(phsize, KM_SLEEP);
499
500 error = exec_read(l, vp, eh.e_phoff, ph, phsize, IO_NODELOCKED);
501 if (error != 0)
502 goto bad;
503
504 #ifdef ELF_INTERP_NON_RELOCATABLE
505 /*
506 * Evil hack: Only MIPS should be non-relocatable, and the
507 * psections should have a high address (typically 0x5ffe0000).
508 * If it's now relocatable, it should be linked at 0 and the
509 * psections should have zeros in the upper part of the address.
510 * Otherwise, force the load at the linked address.
511 */
512 if (*last == ELF_LINK_ADDR && (ph->p_vaddr & 0xffff0000) == 0)
513 *last = ELFDEFNNAME(NO_ADDR);
514 #endif
515
516 /*
517 * If no position to load the interpreter was set by a probe
518 * function, pick the same address that a non-fixed mmap(0, ..)
519 * would (i.e. something safely out of the way).
520 */
521 if (*last == ELFDEFNNAME(NO_ADDR)) {
522 u_long limit = 0;
523 /*
524 * Find the start and ending addresses of the psections to
525 * be loaded. This will give us the size.
526 */
527 for (i = 0, base_ph = NULL; i < eh.e_phnum; i++) {
528 if (ph[i].p_type == PT_LOAD) {
529 u_long psize = ph[i].p_vaddr + ph[i].p_memsz;
530 if (base_ph == NULL)
531 base_ph = &ph[i];
532 if (psize > limit)
533 limit = psize;
534 }
535 }
536
537 if (base_ph == NULL) {
538 DPRINTF("no interpreter loadable sections");
539 error = ENOEXEC;
540 goto bad;
541 }
542
543 /*
544 * Now compute the size and load address.
545 */
546 addr = (*epp->ep_esch->es_emul->e_vm_default_addr)(p,
547 epp->ep_daddr,
548 round_page(limit) - trunc_page(base_ph->p_vaddr),
549 use_topdown);
550 addr += (Elf_Addr)pax_aslr_rtld_offset(epp, base_ph->p_align,
551 use_topdown);
552 } else {
553 addr = *last; /* may be ELF_LINK_ADDR */
554 }
555
556 /*
557 * Load all the necessary sections
558 */
559 for (i = 0, base_ph = NULL, last_ph = NULL; i < eh.e_phnum; i++) {
560 switch (ph[i].p_type) {
561 case PT_LOAD: {
562 u_long size;
563 int flags;
564
565 if (base_ph == NULL) {
566 /*
567 * First encountered psection is always the
568 * base psection. Make sure it's aligned
569 * properly (align down for topdown and align
570 * upwards for not topdown).
571 */
572 base_ph = &ph[i];
573 flags = VMCMD_BASE;
574 if (addr == ELF_LINK_ADDR)
575 addr = ph[i].p_vaddr;
576 if (use_topdown)
577 addr = ELF_TRUNC(addr, ph[i].p_align);
578 else
579 addr = ELF_ROUND(addr, ph[i].p_align);
580 } else {
581 u_long limit = round_page(last_ph->p_vaddr
582 + last_ph->p_memsz);
583 u_long base = trunc_page(ph[i].p_vaddr);
584
585 /*
586 * If there is a gap in between the psections,
587 * map it as inaccessible so nothing else
588 * mmap'ed will be placed there.
589 */
590 if (limit != base) {
591 NEW_VMCMD2(vcset, vmcmd_map_zero,
592 base - limit,
593 limit - base_ph->p_vaddr, NULLVP,
594 0, VM_PROT_NONE, VMCMD_RELATIVE);
595 }
596
597 addr = ph[i].p_vaddr - base_ph->p_vaddr;
598 flags = VMCMD_RELATIVE;
599 }
600 last_ph = &ph[i];
601 if ((error = elf_load_psection(vcset, vp, &ph[i], &addr,
602 &size, flags)) != 0)
603 goto bad;
604 /*
605 * If entry is within this psection then this
606 * must contain the .text section. *entryoff is
607 * relative to the base psection.
608 */
609 if (eh.e_entry >= ph[i].p_vaddr &&
610 eh.e_entry < (ph[i].p_vaddr + size)) {
611 *entryoff = eh.e_entry - base_ph->p_vaddr;
612 }
613 addr += size;
614 break;
615 }
616
617 default:
618 break;
619 }
620 }
621
622 kmem_free(ph, phsize);
623 /*
624 * This value is ignored if TOPDOWN.
625 */
626 *last = addr;
627 vput(vp);
628 return 0;
629
630 bad:
631 if (ph != NULL)
632 kmem_free(ph, phsize);
633 vput(vp);
634 return error;
635 }
636
637 /*
638 * exec_elf_makecmds(): Prepare an Elf binary's exec package
639 *
640 * First, set of the various offsets/lengths in the exec package.
641 *
642 * Then, mark the text image busy (so it can be demand paged) or error
643 * out if this is not possible. Finally, set up vmcmds for the
644 * text, data, bss, and stack segments.
645 */
646 int
647 exec_elf_makecmds(struct lwp *l, struct exec_package *epp)
648 {
649 Elf_Ehdr *eh = epp->ep_hdr;
650 Elf_Phdr *ph, *pp;
651 Elf_Addr phdr = 0, computed_phdr = 0, pos = 0, end_text = 0;
652 int error, i;
653 char *interp = NULL;
654 u_long phsize;
655 struct elf_args *ap;
656 bool is_dyn = false;
657
658 if (epp->ep_hdrvalid < sizeof(Elf_Ehdr)) {
659 DPRINTF("small header %#x", epp->ep_hdrvalid);
660 return ENOEXEC;
661 }
662 if ((error = elf_check_header(eh)) != 0)
663 return error;
664
665 if (eh->e_type == ET_DYN)
666 /* PIE, and some libs have an entry point */
667 is_dyn = true;
668 else if (eh->e_type != ET_EXEC) {
669 DPRINTF("bad type %#x", eh->e_type);
670 return ENOEXEC;
671 }
672
673 if (eh->e_phnum == 0) {
674 DPRINTF("no program headers");
675 return ENOEXEC;
676 }
677
678 /* XXX only LK_EXCLUSIVE to match all others - allow spinning */
679 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
680 error = vn_marktext(epp->ep_vp);
681 if (error) {
682 VOP_UNLOCK(epp->ep_vp);
683 return error;
684 }
685
686 /*
687 * Allocate space to hold all the program headers, and read them
688 * from the file
689 */
690 phsize = eh->e_phnum * sizeof(Elf_Phdr);
691 ph = kmem_alloc(phsize, KM_SLEEP);
692
693 error = exec_read(l, epp->ep_vp, eh->e_phoff, ph, phsize,
694 IO_NODELOCKED);
695 if (error != 0) {
696 VOP_UNLOCK(epp->ep_vp);
697 goto bad;
698 }
699
700 epp->ep_taddr = epp->ep_tsize = ELFDEFNNAME(NO_ADDR);
701 epp->ep_daddr = epp->ep_dsize = ELFDEFNNAME(NO_ADDR);
702
703 for (i = 0; i < eh->e_phnum; i++) {
704 pp = &ph[i];
705 if (pp->p_type == PT_INTERP) {
706 if (pp->p_filesz < 2 || pp->p_filesz > MAXPATHLEN) {
707 DPRINTF("bad interpreter namelen %#jx",
708 (uintmax_t)pp->p_filesz);
709 error = ENOEXEC;
710 VOP_UNLOCK(epp->ep_vp);
711 goto bad;
712 }
713 interp = PNBUF_GET();
714 error = exec_read(l, epp->ep_vp, pp->p_offset, interp,
715 pp->p_filesz, IO_NODELOCKED);
716 if (error != 0) {
717 VOP_UNLOCK(epp->ep_vp);
718 goto bad;
719 }
720 /* Ensure interp is NUL-terminated and of the expected length */
721 if (strnlen(interp, pp->p_filesz) != pp->p_filesz - 1) {
722 DPRINTF("bad interpreter name");
723 error = ENOEXEC;
724 VOP_UNLOCK(epp->ep_vp);
725 goto bad;
726 }
727 break;
728 }
729 }
730
731 /*
732 * On the same architecture, we may be emulating different systems.
733 * See which one will accept this executable.
734 *
735 * Probe functions would normally see if the interpreter (if any)
736 * exists. Emulation packages may possibly replace the interpreter in
737 * interp with a changed path (/emul/xxx/<path>).
738 */
739 pos = ELFDEFNNAME(NO_ADDR);
740 if (epp->ep_esch->u.elf_probe_func) {
741 vaddr_t startp = (vaddr_t)pos;
742
743 error = (*epp->ep_esch->u.elf_probe_func)(l, epp, eh, interp,
744 &startp);
745 if (error) {
746 VOP_UNLOCK(epp->ep_vp);
747 goto bad;
748 }
749 pos = (Elf_Addr)startp;
750 }
751
752 if (is_dyn && (error = elf_placedynexec(epp, eh, ph)) != 0) {
753 VOP_UNLOCK(epp->ep_vp);
754 goto bad;
755 }
756
757 /*
758 * Load all the necessary sections
759 */
760 for (i = 0; i < eh->e_phnum; i++) {
761 Elf_Addr addr = ELFDEFNNAME(NO_ADDR);
762 u_long size = 0;
763
764 switch (ph[i].p_type) {
765 case PT_LOAD:
766 if ((error = elf_load_psection(&epp->ep_vmcmds,
767 epp->ep_vp, &ph[i], &addr, &size, VMCMD_FIXED))
768 != 0) {
769 VOP_UNLOCK(epp->ep_vp);
770 goto bad;
771 }
772
773 /*
774 * Consider this as text segment, if it is executable.
775 * If there is more than one text segment, pick the
776 * largest.
777 */
778 if (ph[i].p_flags & PF_X) {
779 if (epp->ep_taddr == ELFDEFNNAME(NO_ADDR) ||
780 size > epp->ep_tsize) {
781 epp->ep_taddr = addr;
782 epp->ep_tsize = size;
783 }
784 end_text = addr + size;
785 } else {
786 epp->ep_daddr = addr;
787 epp->ep_dsize = size;
788 }
789 if (ph[i].p_offset == 0) {
790 computed_phdr = ph[i].p_vaddr + eh->e_phoff;
791 }
792 break;
793
794 case PT_SHLIB:
795 /* SCO has these sections. */
796 case PT_INTERP:
797 /* Already did this one. */
798 case PT_DYNAMIC:
799 case PT_NOTE:
800 break;
801 case PT_PHDR:
802 /* Note address of program headers (in text segment) */
803 phdr = ph[i].p_vaddr;
804 break;
805
806 default:
807 /*
808 * Not fatal; we don't need to understand everything.
809 */
810 break;
811 }
812 }
813
814 /* Now done with the vnode. */
815 VOP_UNLOCK(epp->ep_vp);
816
817 if (epp->ep_vmcmds.evs_used == 0) {
818 /* No VMCMD; there was no PT_LOAD section, or those
819 * sections were empty */
820 DPRINTF("no vmcommands");
821 error = ENOEXEC;
822 goto bad;
823 }
824
825 if (epp->ep_daddr == ELFDEFNNAME(NO_ADDR)) {
826 epp->ep_daddr = round_page(end_text);
827 epp->ep_dsize = 0;
828 }
829
830 /*
831 * Check if we found a dynamically linked binary and arrange to load
832 * its interpreter
833 */
834 if (interp) {
835 u_int nused = epp->ep_vmcmds.evs_used;
836 u_long interp_offset = 0;
837
838 if ((error = elf_load_interp(l, epp, interp,
839 &epp->ep_vmcmds, &interp_offset, &pos)) != 0) {
840 goto bad;
841 }
842 if (epp->ep_vmcmds.evs_used == nused) {
843 /* elf_load_interp() has not set up any new VMCMD */
844 DPRINTF("no vmcommands for interpreter");
845 error = ENOEXEC;
846 goto bad;
847 }
848
849 ap = kmem_alloc(sizeof(*ap), KM_SLEEP);
850 ap->arg_interp = epp->ep_vmcmds.evs_cmds[nused].ev_addr;
851 epp->ep_entryoffset = interp_offset;
852 epp->ep_entry = ap->arg_interp + interp_offset;
853 PNBUF_PUT(interp);
854 interp = NULL;
855 } else {
856 epp->ep_entry = eh->e_entry;
857 if (epp->ep_flags & EXEC_FORCEAUX) {
858 ap = kmem_zalloc(sizeof(*ap), KM_SLEEP);
859 ap->arg_interp = (vaddr_t)NULL;
860 } else {
861 ap = NULL;
862 }
863 }
864
865 if (ap) {
866 ap->arg_phaddr = phdr ? phdr : computed_phdr;
867 ap->arg_phentsize = eh->e_phentsize;
868 ap->arg_phnum = eh->e_phnum;
869 ap->arg_entry = eh->e_entry;
870 epp->ep_emul_arg = ap;
871 epp->ep_emul_arg_free = elf_free_emul_arg;
872 }
873
874 #ifdef ELF_MAP_PAGE_ZERO
875 /* Dell SVR4 maps page zero, yeuch! */
876 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, PAGE_SIZE, 0,
877 epp->ep_vp, 0, VM_PROT_READ);
878 #endif
879
880 error = (*epp->ep_esch->es_setup_stack)(l, epp);
881 if (error)
882 goto bad;
883
884 kmem_free(ph, phsize);
885 return 0;
886
887 bad:
888 if (interp)
889 PNBUF_PUT(interp);
890 exec_free_emul_arg(epp);
891 kmem_free(ph, phsize);
892 kill_vmcmds(&epp->ep_vmcmds);
893 return error;
894 }
895
896 int
897 netbsd_elf_signature(struct lwp *l, struct exec_package *epp,
898 Elf_Ehdr *eh)
899 {
900 size_t i;
901 Elf_Phdr *ph;
902 size_t phsize;
903 char *nbuf;
904 int error;
905 int isnetbsd = 0;
906
907 epp->ep_pax_flags = 0;
908
909 if (eh->e_phnum > ELF_MAXPHNUM || eh->e_phnum == 0) {
910 DPRINTF("no signature %#x", eh->e_phnum);
911 return ENOEXEC;
912 }
913
914 phsize = eh->e_phnum * sizeof(Elf_Phdr);
915 ph = kmem_alloc(phsize, KM_SLEEP);
916 error = exec_read(l, epp->ep_vp, eh->e_phoff, ph, phsize,
917 IO_NODELOCKED);
918 if (error)
919 goto out;
920
921 nbuf = kmem_alloc(ELF_MAXNOTESIZE, KM_SLEEP);
922 for (i = 0; i < eh->e_phnum; i++) {
923 const char *nptr;
924 size_t nlen;
925
926 if (ph[i].p_type != PT_NOTE ||
927 ph[i].p_filesz > ELF_MAXNOTESIZE)
928 continue;
929
930 nlen = ph[i].p_filesz;
931 error = exec_read(l, epp->ep_vp, ph[i].p_offset, nbuf, nlen,
932 IO_NODELOCKED);
933 if (error)
934 continue;
935
936 nptr = nbuf;
937 while (nlen > 0) {
938 const Elf_Nhdr *np;
939 const char *ndata, *ndesc;
940
941 /* note header */
942 np = (const Elf_Nhdr *)nptr;
943 if (nlen < sizeof(*np)) {
944 break;
945 }
946 nptr += sizeof(*np);
947 nlen -= sizeof(*np);
948
949 /* note name */
950 ndata = nptr;
951 if (nlen < roundup(np->n_namesz, 4)) {
952 break;
953 }
954 nptr += roundup(np->n_namesz, 4);
955 nlen -= roundup(np->n_namesz, 4);
956
957 /* note description */
958 ndesc = nptr;
959 if (nlen < roundup(np->n_descsz, 4)) {
960 break;
961 }
962 nptr += roundup(np->n_descsz, 4);
963 nlen -= roundup(np->n_descsz, 4);
964
965 isnetbsd |= netbsd_elf_note(epp, np, ndata, ndesc);
966 }
967 }
968 kmem_free(nbuf, ELF_MAXNOTESIZE);
969
970 error = isnetbsd ? 0 : ENOEXEC;
971 #ifdef DEBUG_ELF
972 if (error)
973 DPRINTF("not netbsd");
974 #endif
975 out:
976 kmem_free(ph, phsize);
977 return error;
978 }
979
980 int
981 netbsd_elf_note(struct exec_package *epp,
982 const Elf_Nhdr *np, const char *ndata, const char *ndesc)
983 {
984 int isnetbsd = 0;
985
986 #ifdef DIAGNOSTIC
987 const char *badnote;
988 #define BADNOTE(n) badnote = (n)
989 #else
990 #define BADNOTE(n)
991 #endif
992
993 switch (np->n_type) {
994 case ELF_NOTE_TYPE_NETBSD_TAG:
995 /* It is us */
996 if (np->n_namesz == ELF_NOTE_NETBSD_NAMESZ &&
997 np->n_descsz == ELF_NOTE_NETBSD_DESCSZ &&
998 memcmp(ndata, ELF_NOTE_NETBSD_NAME,
999 ELF_NOTE_NETBSD_NAMESZ) == 0) {
1000 memcpy(&epp->ep_osversion, ndesc,
1001 ELF_NOTE_NETBSD_DESCSZ);
1002 isnetbsd = 1;
1003 break;
1004 }
1005
1006 /*
1007 * Ignore SuSE tags; SuSE's n_type is the same the
1008 * NetBSD one.
1009 */
1010 if (np->n_namesz == ELF_NOTE_SUSE_NAMESZ &&
1011 memcmp(ndata, ELF_NOTE_SUSE_NAME,
1012 ELF_NOTE_SUSE_NAMESZ) == 0)
1013 break;
1014 /*
1015 * Ignore old GCC
1016 */
1017 if (np->n_namesz == ELF_NOTE_OGCC_NAMESZ &&
1018 memcmp(ndata, ELF_NOTE_OGCC_NAME,
1019 ELF_NOTE_OGCC_NAMESZ) == 0)
1020 break;
1021 BADNOTE("NetBSD tag");
1022 goto bad;
1023
1024 case ELF_NOTE_TYPE_PAX_TAG:
1025 if (np->n_namesz == ELF_NOTE_PAX_NAMESZ &&
1026 np->n_descsz == ELF_NOTE_PAX_DESCSZ &&
1027 memcmp(ndata, ELF_NOTE_PAX_NAME,
1028 ELF_NOTE_PAX_NAMESZ) == 0) {
1029 uint32_t flags;
1030 memcpy(&flags, ndesc, sizeof(flags));
1031 /* Convert the flags and insert them into
1032 * the exec package. */
1033 pax_setup_elf_flags(epp, flags);
1034 break;
1035 }
1036 BADNOTE("PaX tag");
1037 goto bad;
1038
1039 case ELF_NOTE_TYPE_MARCH_TAG:
1040 /* Copy the machine arch into the package. */
1041 if (np->n_namesz == ELF_NOTE_MARCH_NAMESZ
1042 && memcmp(ndata, ELF_NOTE_MARCH_NAME,
1043 ELF_NOTE_MARCH_NAMESZ) == 0) {
1044 /* Do not truncate the buffer */
1045 if (np->n_descsz > sizeof(epp->ep_machine_arch)) {
1046 BADNOTE("description size limit");
1047 goto bad;
1048 }
1049 /*
1050 * Ensure ndesc is NUL-terminated and of the
1051 * expected length.
1052 */
1053 if (strnlen(ndesc, np->n_descsz) + 1 !=
1054 np->n_descsz) {
1055 BADNOTE("description size");
1056 goto bad;
1057 }
1058 strlcpy(epp->ep_machine_arch, ndesc,
1059 sizeof(epp->ep_machine_arch));
1060 break;
1061 }
1062 BADNOTE("march tag");
1063 goto bad;
1064
1065 case ELF_NOTE_TYPE_MCMODEL_TAG:
1066 /* arch specific check for code model */
1067 #ifdef ELF_MD_MCMODEL_CHECK
1068 if (np->n_namesz == ELF_NOTE_MCMODEL_NAMESZ
1069 && memcmp(ndata, ELF_NOTE_MCMODEL_NAME,
1070 ELF_NOTE_MCMODEL_NAMESZ) == 0) {
1071 ELF_MD_MCMODEL_CHECK(epp, ndesc, np->n_descsz);
1072 break;
1073 }
1074 BADNOTE("mcmodel tag");
1075 goto bad;
1076 #endif
1077 break;
1078
1079 case ELF_NOTE_TYPE_SUSE_VERSION_TAG:
1080 break;
1081
1082 case ELF_NOTE_TYPE_GO_BUILDID_TAG:
1083 break;
1084
1085 case ELF_NOTE_TYPE_FDO_PACKAGING_METADATA:
1086 break;
1087
1088 case ELF_NOTE_TYPE_NETBSD_EMUL_TAG:
1089 /* Ancient NetBSD version tag */
1090 break;
1091
1092 default:
1093 BADNOTE("unknown tag");
1094 bad:
1095 #ifdef DIAGNOSTIC
1096 /* Ignore GNU tags */
1097 if (np->n_namesz == ELF_NOTE_GNU_NAMESZ &&
1098 memcmp(ndata, ELF_NOTE_GNU_NAME,
1099 ELF_NOTE_GNU_NAMESZ) == 0)
1100 break;
1101
1102 int ns = (int)np->n_namesz;
1103 printf("%s: Unknown elf note type %d (%s): "
1104 "[namesz=%d, descsz=%d name=%-*.*s]\n",
1105 epp->ep_kname, np->n_type, badnote, np->n_namesz,
1106 np->n_descsz, ns, ns, ndata);
1107 #endif
1108 break;
1109 }
1110
1111 return isnetbsd;
1112 }
1113
1114 int
1115 netbsd_elf_probe(struct lwp *l, struct exec_package *epp, void *eh, char *itp,
1116 vaddr_t *pos)
1117 {
1118 int error;
1119
1120 if ((error = netbsd_elf_signature(l, epp, eh)) != 0)
1121 return error;
1122 #ifdef ELF_MD_PROBE_FUNC
1123 if ((error = ELF_MD_PROBE_FUNC(l, epp, eh, itp, pos)) != 0)
1124 return error;
1125 #elif defined(ELF_INTERP_NON_RELOCATABLE)
1126 *pos = ELF_LINK_ADDR;
1127 #endif
1128 epp->ep_flags |= EXEC_FORCEAUX;
1129 return 0;
1130 }
1131
1132 void
1133 elf_free_emul_arg(void *arg)
1134 {
1135 struct elf_args *ap = arg;
1136 KASSERT(ap != NULL);
1137 kmem_free(ap, sizeof(*ap));
1138 }
Cache object: 00cc41b24f5f698d978f8050ed16810d
|