xref: /illumos-gate/usr/src/uts/common/exec/elf/elf.c (revision 1b09309c5ebed5c0bf14e2b396bf626c5aa30034)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	   All Rights Reserved	*/
28 /*
29  * Copyright (c) 2019, Joyent, Inc.
30  * Copyright 2021 Oxide Computer Company
31  */
32 
33 #include <sys/types.h>
34 #include <sys/param.h>
35 #include <sys/thread.h>
36 #include <sys/sysmacros.h>
37 #include <sys/signal.h>
38 #include <sys/cred.h>
39 #include <sys/user.h>
40 #include <sys/errno.h>
41 #include <sys/vnode.h>
42 #include <sys/mman.h>
43 #include <sys/kmem.h>
44 #include <sys/proc.h>
45 #include <sys/pathname.h>
46 #include <sys/policy.h>
47 #include <sys/cmn_err.h>
48 #include <sys/systm.h>
49 #include <sys/elf.h>
50 #include <sys/vmsystm.h>
51 #include <sys/debug.h>
52 #include <sys/auxv.h>
53 #include <sys/exec.h>
54 #include <sys/prsystm.h>
55 #include <vm/as.h>
56 #include <vm/rm.h>
57 #include <vm/seg.h>
58 #include <vm/seg_vn.h>
59 #include <sys/modctl.h>
60 #include <sys/systeminfo.h>
61 #include <sys/vmparam.h>
62 #include <sys/machelf.h>
63 #include <sys/shm_impl.h>
64 #include <sys/archsystm.h>
65 #include <sys/fasttrap.h>
66 #include <sys/brand.h>
67 #include "elf_impl.h"
68 #include <sys/sdt.h>
69 #include <sys/siginfo.h>
70 #include <sys/random.h>
71 
72 #if defined(__x86)
73 #include <sys/comm_page_util.h>
74 #include <sys/fp.h>
75 #endif /* defined(__x86) */
76 
77 
78 extern int at_flags;
79 extern volatile size_t aslr_max_brk_skew;
80 
81 #define	ORIGIN_STR	"ORIGIN"
82 #define	ORIGIN_STR_SIZE	6
83 
84 static int getelfhead(vnode_t *, cred_t *, Ehdr *, int *, int *, int *);
85 static int getelfphdr(vnode_t *, cred_t *, const Ehdr *, int, caddr_t *,
86     ssize_t *);
87 static int getelfshdr(vnode_t *, cred_t *, const Ehdr *, int, int, caddr_t *,
88     ssize_t *, caddr_t *, ssize_t *);
89 static size_t elfsize(Ehdr *, int, caddr_t, uintptr_t *);
90 static int mapelfexec(vnode_t *, Ehdr *, int, caddr_t,
91     Phdr **, Phdr **, Phdr **, Phdr **, Phdr *,
92     caddr_t *, caddr_t *, intptr_t *, intptr_t *, size_t, long *, size_t *);
93 
94 typedef enum {
95 	STR_CTF,
96 	STR_SYMTAB,
97 	STR_DYNSYM,
98 	STR_STRTAB,
99 	STR_DYNSTR,
100 	STR_SHSTRTAB,
101 	STR_NUM
102 } shstrtype_t;
103 
104 static const char *shstrtab_data[] = {
105 	".SUNW_ctf",
106 	".symtab",
107 	".dynsym",
108 	".strtab",
109 	".dynstr",
110 	".shstrtab"
111 };
112 
113 typedef struct shstrtab {
114 	int	sst_ndx[STR_NUM];
115 	int	sst_cur;
116 } shstrtab_t;
117 
118 static void
119 shstrtab_init(shstrtab_t *s)
120 {
121 	bzero(&s->sst_ndx, sizeof (s->sst_ndx));
122 	s->sst_cur = 1;
123 }
124 
125 static int
126 shstrtab_ndx(shstrtab_t *s, shstrtype_t type)
127 {
128 	int ret;
129 
130 	if ((ret = s->sst_ndx[type]) != 0)
131 		return (ret);
132 
133 	ret = s->sst_ndx[type] = s->sst_cur;
134 	s->sst_cur += strlen(shstrtab_data[type]) + 1;
135 
136 	return (ret);
137 }
138 
139 static size_t
140 shstrtab_size(const shstrtab_t *s)
141 {
142 	return (s->sst_cur);
143 }
144 
145 static void
146 shstrtab_dump(const shstrtab_t *s, char *buf)
147 {
148 	int i, ndx;
149 
150 	*buf = '\0';
151 	for (i = 0; i < STR_NUM; i++) {
152 		if ((ndx = s->sst_ndx[i]) != 0)
153 			(void) strcpy(buf + ndx, shstrtab_data[i]);
154 	}
155 }
156 
157 static int
158 dtrace_safe_phdr(Phdr *phdrp, struct uarg *args, uintptr_t base)
159 {
160 	ASSERT(phdrp->p_type == PT_SUNWDTRACE);
161 
162 	/*
163 	 * See the comment in fasttrap.h for information on how to safely
164 	 * update this program header.
165 	 */
166 	if (phdrp->p_memsz < PT_SUNWDTRACE_SIZE ||
167 	    (phdrp->p_flags & (PF_R | PF_W | PF_X)) != (PF_R | PF_W | PF_X))
168 		return (-1);
169 
170 	args->thrptr = phdrp->p_vaddr + base;
171 
172 	return (0);
173 }
174 
175 static int
176 handle_secflag_dt(proc_t *p, uint_t dt, uint_t val)
177 {
178 	uint_t flag;
179 
180 	switch (dt) {
181 	case DT_SUNW_ASLR:
182 		flag = PROC_SEC_ASLR;
183 		break;
184 	default:
185 		return (EINVAL);
186 	}
187 
188 	if (val == 0) {
189 		if (secflag_isset(p->p_secflags.psf_lower, flag))
190 			return (EPERM);
191 		if ((secpolicy_psecflags(CRED(), p, p) != 0) &&
192 		    secflag_isset(p->p_secflags.psf_inherit, flag))
193 			return (EPERM);
194 
195 		secflag_clear(&p->p_secflags.psf_effective, flag);
196 	} else {
197 		if (!secflag_isset(p->p_secflags.psf_upper, flag))
198 			return (EPERM);
199 
200 		if ((secpolicy_psecflags(CRED(), p, p) != 0) &&
201 		    !secflag_isset(p->p_secflags.psf_inherit, flag))
202 			return (EPERM);
203 
204 		secflag_set(&p->p_secflags.psf_effective, flag);
205 	}
206 
207 	return (0);
208 }
209 
210 /*
211  * Map in the executable pointed to by vp. Returns 0 on success.
212  */
213 int
214 mapexec_brand(vnode_t *vp, uarg_t *args, Ehdr *ehdr, Addr *uphdr_vaddr,
215     intptr_t *voffset, caddr_t exec_file, int *interp, caddr_t *bssbase,
216     caddr_t *brkbase, size_t *brksize, uintptr_t *lddatap)
217 {
218 	size_t		len;
219 	struct vattr	vat;
220 	caddr_t		phdrbase = NULL;
221 	ssize_t		phdrsize;
222 	int		nshdrs, shstrndx, nphdrs;
223 	int		error = 0;
224 	Phdr		*uphdr = NULL;
225 	Phdr		*junk = NULL;
226 	Phdr		*dynphdr = NULL;
227 	Phdr		*dtrphdr = NULL;
228 	uintptr_t	lddata;
229 	long		execsz;
230 	intptr_t	minaddr;
231 
232 	if (lddatap != NULL)
233 		*lddatap = 0;
234 
235 	if (error = execpermissions(vp, &vat, args)) {
236 		uprintf("%s: Cannot execute %s\n", exec_file, args->pathname);
237 		return (error);
238 	}
239 
240 	if ((error = getelfhead(vp, CRED(), ehdr, &nshdrs, &shstrndx,
241 	    &nphdrs)) != 0 ||
242 	    (error = getelfphdr(vp, CRED(), ehdr, nphdrs, &phdrbase,
243 	    &phdrsize)) != 0) {
244 		uprintf("%s: Cannot read %s\n", exec_file, args->pathname);
245 		return (error);
246 	}
247 
248 	if ((len = elfsize(ehdr, nphdrs, phdrbase, &lddata)) == 0) {
249 		uprintf("%s: Nothing to load in %s", exec_file, args->pathname);
250 		kmem_free(phdrbase, phdrsize);
251 		return (ENOEXEC);
252 	}
253 	if (lddatap != NULL)
254 		*lddatap = lddata;
255 
256 	if (error = mapelfexec(vp, ehdr, nphdrs, phdrbase, &uphdr, &dynphdr,
257 	    &junk, &dtrphdr, NULL, bssbase, brkbase, voffset, &minaddr,
258 	    len, &execsz, brksize)) {
259 		uprintf("%s: Cannot map %s\n", exec_file, args->pathname);
260 		kmem_free(phdrbase, phdrsize);
261 		return (error);
262 	}
263 
264 	/*
265 	 * Inform our caller if the executable needs an interpreter.
266 	 */
267 	*interp = (dynphdr == NULL) ? 0 : 1;
268 
269 	/*
270 	 * If this is a statically linked executable, voffset should indicate
271 	 * the address of the executable itself (it normally holds the address
272 	 * of the interpreter).
273 	 */
274 	if (ehdr->e_type == ET_EXEC && *interp == 0)
275 		*voffset = minaddr;
276 
277 	if (uphdr != NULL) {
278 		*uphdr_vaddr = uphdr->p_vaddr;
279 	} else {
280 		*uphdr_vaddr = (Addr)-1;
281 	}
282 
283 	kmem_free(phdrbase, phdrsize);
284 	return (error);
285 }
286 
287 /*ARGSUSED*/
288 int
289 elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
290     int level, long *execsz, int setid, caddr_t exec_file, cred_t *cred,
291     int brand_action)
292 {
293 	caddr_t		phdrbase = NULL;
294 	caddr_t		bssbase = 0;
295 	caddr_t		brkbase = 0;
296 	size_t		brksize = 0;
297 	ssize_t		dlnsize;
298 	aux_entry_t	*aux;
299 	int		error;
300 	ssize_t		resid;
301 	int		fd = -1;
302 	intptr_t	voffset;
303 	Phdr		*intphdr = NULL;
304 	Phdr		*dynamicphdr = NULL;
305 	Phdr		*stphdr = NULL;
306 	Phdr		*uphdr = NULL;
307 	Phdr		*junk = NULL;
308 	size_t		len;
309 	size_t		i;
310 	ssize_t		phdrsize;
311 	int		postfixsize = 0;
312 	int		hsize;
313 	Phdr		*phdrp;
314 	Phdr		*dataphdrp = NULL;
315 	Phdr		*dtrphdr;
316 	Phdr		*capphdr = NULL;
317 	Cap		*cap = NULL;
318 	ssize_t		capsize;
319 	int		hasu = 0;
320 	int		hasauxv = 0;
321 	int		hasintp = 0;
322 	int		branded = 0;
323 
324 	struct proc *p = ttoproc(curthread);
325 	struct user *up = PTOU(p);
326 	struct bigwad {
327 		Ehdr	ehdr;
328 		aux_entry_t	elfargs[__KERN_NAUXV_IMPL];
329 		char		dl_name[MAXPATHLEN];
330 		char		pathbuf[MAXPATHLEN];
331 		struct vattr	vattr;
332 		struct execenv	exenv;
333 	} *bigwad;	/* kmem_alloc this behemoth so we don't blow stack */
334 	Ehdr		*ehdrp;
335 	int		nshdrs, shstrndx, nphdrs;
336 	char		*dlnp;
337 	char		*pathbufp;
338 	rlim64_t	limit;
339 	rlim64_t	roundlimit;
340 
341 	ASSERT(p->p_model == DATAMODEL_ILP32 || p->p_model == DATAMODEL_LP64);
342 
343 	bigwad = kmem_alloc(sizeof (struct bigwad), KM_SLEEP);
344 	ehdrp = &bigwad->ehdr;
345 	dlnp = bigwad->dl_name;
346 	pathbufp = bigwad->pathbuf;
347 
348 	/*
349 	 * Obtain ELF and program header information.
350 	 */
351 	if ((error = getelfhead(vp, CRED(), ehdrp, &nshdrs, &shstrndx,
352 	    &nphdrs)) != 0 ||
353 	    (error = getelfphdr(vp, CRED(), ehdrp, nphdrs, &phdrbase,
354 	    &phdrsize)) != 0)
355 		goto out;
356 
357 	/*
358 	 * Prevent executing an ELF file that has no entry point.
359 	 */
360 	if (ehdrp->e_entry == 0) {
361 		uprintf("%s: Bad entry point\n", exec_file);
362 		goto bad;
363 	}
364 
365 	/*
366 	 * Put data model that we're exec-ing to into the args passed to
367 	 * exec_args(), so it will know what it is copying to on new stack.
368 	 * Now that we know whether we are exec-ing a 32-bit or 64-bit
369 	 * executable, we can set execsz with the appropriate NCARGS.
370 	 */
371 #ifdef	_LP64
372 	if (ehdrp->e_ident[EI_CLASS] == ELFCLASS32) {
373 		args->to_model = DATAMODEL_ILP32;
374 		*execsz = btopr(SINCR) + btopr(SSIZE) + btopr(NCARGS32-1);
375 	} else {
376 		args->to_model = DATAMODEL_LP64;
377 		args->stk_prot &= ~PROT_EXEC;
378 #if defined(__x86)
379 		args->dat_prot &= ~PROT_EXEC;
380 #endif
381 		*execsz = btopr(SINCR) + btopr(SSIZE) + btopr(NCARGS64-1);
382 	}
383 #else	/* _LP64 */
384 	args->to_model = DATAMODEL_ILP32;
385 	*execsz = btopr(SINCR) + btopr(SSIZE) + btopr(NCARGS-1);
386 #endif	/* _LP64 */
387 
388 	/*
389 	 * We delay invoking the brand callback until we've figured out
390 	 * what kind of elf binary we're trying to run, 32-bit or 64-bit.
391 	 * We do this because now the brand library can just check
392 	 * args->to_model to see if the target is 32-bit or 64-bit without
393 	 * having do duplicate all the code above.
394 	 *
395 	 * The level checks associated with brand handling below are used to
396 	 * prevent a loop since the brand elfexec function typically comes back
397 	 * through this function. We must check <= here since the nested
398 	 * handling in the #! interpreter code will increment the level before
399 	 * calling gexec to run the final elfexec interpreter.
400 	 */
401 	if ((level <= INTP_MAXDEPTH) &&
402 	    (brand_action != EBA_NATIVE) && (PROC_IS_BRANDED(p))) {
403 		error = BROP(p)->b_elfexec(vp, uap, args,
404 		    idatap, level + 1, execsz, setid, exec_file, cred,
405 		    brand_action);
406 		goto out;
407 	}
408 
409 	/*
410 	 * Determine aux size now so that stack can be built
411 	 * in one shot (except actual copyout of aux image),
412 	 * determine any non-default stack protections,
413 	 * and still have this code be machine independent.
414 	 */
415 	hsize = ehdrp->e_phentsize;
416 	phdrp = (Phdr *)phdrbase;
417 	for (i = nphdrs; i > 0; i--) {
418 		switch (phdrp->p_type) {
419 		case PT_INTERP:
420 			hasauxv = hasintp = 1;
421 			break;
422 		case PT_PHDR:
423 			hasu = 1;
424 			break;
425 		case PT_SUNWSTACK:
426 			args->stk_prot = PROT_USER;
427 			if (phdrp->p_flags & PF_R)
428 				args->stk_prot |= PROT_READ;
429 			if (phdrp->p_flags & PF_W)
430 				args->stk_prot |= PROT_WRITE;
431 			if (phdrp->p_flags & PF_X)
432 				args->stk_prot |= PROT_EXEC;
433 			break;
434 		case PT_LOAD:
435 			dataphdrp = phdrp;
436 			break;
437 		case PT_SUNWCAP:
438 			capphdr = phdrp;
439 			break;
440 		case PT_DYNAMIC:
441 			dynamicphdr = phdrp;
442 			break;
443 		}
444 		phdrp = (Phdr *)((caddr_t)phdrp + hsize);
445 	}
446 
447 	if (ehdrp->e_type != ET_EXEC) {
448 		dataphdrp = NULL;
449 		hasauxv = 1;
450 	}
451 
452 	/* Copy BSS permissions to args->dat_prot */
453 	if (dataphdrp != NULL) {
454 		args->dat_prot = PROT_USER;
455 		if (dataphdrp->p_flags & PF_R)
456 			args->dat_prot |= PROT_READ;
457 		if (dataphdrp->p_flags & PF_W)
458 			args->dat_prot |= PROT_WRITE;
459 		if (dataphdrp->p_flags & PF_X)
460 			args->dat_prot |= PROT_EXEC;
461 	}
462 
463 	/*
464 	 * If a auxvector will be required - reserve the space for
465 	 * it now.  This may be increased by exec_args if there are
466 	 * ISA-specific types (included in __KERN_NAUXV_IMPL).
467 	 */
468 	if (hasauxv) {
469 		/*
470 		 * If a AUX vector is being built - the base AUX
471 		 * entries are:
472 		 *
473 		 *	AT_BASE
474 		 *	AT_FLAGS
475 		 *	AT_PAGESZ
476 		 *	AT_SUN_AUXFLAGS
477 		 *	AT_SUN_HWCAP
478 		 *	AT_SUN_HWCAP2
479 		 *	AT_SUN_PLATFORM (added in stk_copyout)
480 		 *	AT_SUN_EXECNAME (added in stk_copyout)
481 		 *	AT_NULL
482 		 *
483 		 * total == 9
484 		 */
485 		if (hasintp && hasu) {
486 			/*
487 			 * Has PT_INTERP & PT_PHDR - the auxvectors that
488 			 * will be built are:
489 			 *
490 			 *	AT_PHDR
491 			 *	AT_PHENT
492 			 *	AT_PHNUM
493 			 *	AT_ENTRY
494 			 *	AT_LDDATA
495 			 *
496 			 * total = 5
497 			 */
498 			args->auxsize = (9 + 5) * sizeof (aux_entry_t);
499 		} else if (hasintp) {
500 			/*
501 			 * Has PT_INTERP but no PT_PHDR
502 			 *
503 			 *	AT_EXECFD
504 			 *	AT_LDDATA
505 			 *
506 			 * total = 2
507 			 */
508 			args->auxsize = (9 + 2) * sizeof (aux_entry_t);
509 		} else {
510 			args->auxsize = 9 * sizeof (aux_entry_t);
511 		}
512 	} else {
513 		args->auxsize = 0;
514 	}
515 
516 	/*
517 	 * If this binary is using an emulator, we need to add an
518 	 * AT_SUN_EMULATOR aux entry.
519 	 */
520 	if (args->emulator != NULL)
521 		args->auxsize += sizeof (aux_entry_t);
522 
523 	/*
524 	 * On supported kernels (x86_64) make room in the auxv for the
525 	 * AT_SUN_COMMPAGE entry.  This will go unpopulated on i86xpv systems
526 	 * which do not provide such functionality.
527 	 *
528 	 * Additionally cover the floating point information AT_SUN_FPSIZE and
529 	 * AT_SUN_FPTYPE.
530 	 */
531 #if defined(__amd64)
532 	args->auxsize += 3 * sizeof (aux_entry_t);
533 #endif /* defined(__amd64) */
534 
535 	if ((brand_action != EBA_NATIVE) && (PROC_IS_BRANDED(p))) {
536 		branded = 1;
537 		/*
538 		 * We will be adding 4 entries to the aux vectors.  One for
539 		 * the the brandname and 3 for the brand specific aux vectors.
540 		 */
541 		args->auxsize += 4 * sizeof (aux_entry_t);
542 	}
543 
544 	/* If the binary has an explicit ASLR flag, it must be honoured */
545 	if ((dynamicphdr != NULL) && (dynamicphdr->p_filesz > 0)) {
546 		const size_t dynfilesz = dynamicphdr->p_filesz;
547 		const size_t dynoffset = dynamicphdr->p_offset;
548 		Dyn *dyn, *dp;
549 
550 		if (dynoffset > MAXOFFSET_T ||
551 		    dynfilesz > MAXOFFSET_T ||
552 		    dynoffset + dynfilesz > MAXOFFSET_T) {
553 			uprintf("%s: cannot read full .dynamic section\n",
554 			    exec_file);
555 			error = EINVAL;
556 			goto out;
557 		}
558 
559 #define	DYN_STRIDE	100
560 		for (i = 0; i < dynfilesz; i += sizeof (*dyn) * DYN_STRIDE) {
561 			const size_t remdyns = (dynfilesz - i) / sizeof (*dyn);
562 			const size_t ndyns = MIN(DYN_STRIDE, remdyns);
563 			const size_t dynsize = ndyns * sizeof (*dyn);
564 
565 			dyn = kmem_alloc(dynsize, KM_SLEEP);
566 
567 			if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)dyn,
568 			    (ssize_t)dynsize, (offset_t)(dynoffset + i),
569 			    UIO_SYSSPACE, 0, (rlim64_t)0,
570 			    CRED(), &resid)) != 0) {
571 				uprintf("%s: cannot read .dynamic section\n",
572 				    exec_file);
573 				goto out;
574 			}
575 
576 			for (dp = dyn; dp < (dyn + ndyns); dp++) {
577 				if (dp->d_tag == DT_SUNW_ASLR) {
578 					if ((error = handle_secflag_dt(p,
579 					    DT_SUNW_ASLR,
580 					    dp->d_un.d_val)) != 0) {
581 						uprintf("%s: error setting "
582 						    "security-flag from "
583 						    "DT_SUNW_ASLR: %d\n",
584 						    exec_file, error);
585 						goto out;
586 					}
587 				}
588 			}
589 
590 			kmem_free(dyn, dynsize);
591 		}
592 	}
593 
594 	/* Hardware/Software capabilities */
595 	if (capphdr != NULL &&
596 	    (capsize = capphdr->p_filesz) > 0 &&
597 	    capsize <= 16 * sizeof (*cap)) {
598 		int ncaps = capsize / sizeof (*cap);
599 		Cap *cp;
600 
601 		cap = kmem_alloc(capsize, KM_SLEEP);
602 		if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)cap,
603 		    capsize, (offset_t)capphdr->p_offset,
604 		    UIO_SYSSPACE, 0, (rlim64_t)0, CRED(), &resid)) != 0) {
605 			uprintf("%s: Cannot read capabilities section\n",
606 			    exec_file);
607 			goto out;
608 		}
609 		for (cp = cap; cp < cap + ncaps; cp++) {
610 			if (cp->c_tag == CA_SUNW_SF_1 &&
611 			    (cp->c_un.c_val & SF1_SUNW_ADDR32)) {
612 				if (args->to_model == DATAMODEL_LP64)
613 					args->addr32 = 1;
614 				break;
615 			}
616 		}
617 	}
618 
619 	aux = bigwad->elfargs;
620 	/*
621 	 * Move args to the user's stack.
622 	 * This can fill in the AT_SUN_PLATFORM and AT_SUN_EXECNAME aux entries.
623 	 */
624 	if ((error = exec_args(uap, args, idatap, (void **)&aux)) != 0) {
625 		if (error == -1) {
626 			error = ENOEXEC;
627 			goto bad;
628 		}
629 		goto out;
630 	}
631 	/* we're single threaded after this point */
632 
633 	/*
634 	 * If this is an ET_DYN executable (shared object),
635 	 * determine its memory size so that mapelfexec() can load it.
636 	 */
637 	if (ehdrp->e_type == ET_DYN)
638 		len = elfsize(ehdrp, nphdrs, phdrbase, NULL);
639 	else
640 		len = 0;
641 
642 	dtrphdr = NULL;
643 
644 	if ((error = mapelfexec(vp, ehdrp, nphdrs, phdrbase, &uphdr, &intphdr,
645 	    &stphdr, &dtrphdr, dataphdrp, &bssbase, &brkbase, &voffset, NULL,
646 	    len, execsz, &brksize)) != 0)
647 		goto bad;
648 
649 	if (uphdr != NULL && intphdr == NULL)
650 		goto bad;
651 
652 	if (dtrphdr != NULL && dtrace_safe_phdr(dtrphdr, args, voffset) != 0) {
653 		uprintf("%s: Bad DTrace phdr in %s\n", exec_file, exec_file);
654 		goto bad;
655 	}
656 
657 	if (intphdr != NULL) {
658 		size_t		len;
659 		uintptr_t	lddata;
660 		char		*p;
661 		struct vnode	*nvp;
662 
663 		dlnsize = intphdr->p_filesz;
664 
665 		if (dlnsize > MAXPATHLEN || dlnsize <= 0)
666 			goto bad;
667 
668 		/*
669 		 * Read in "interpreter" pathname.
670 		 */
671 		if ((error = vn_rdwr(UIO_READ, vp, dlnp, intphdr->p_filesz,
672 		    (offset_t)intphdr->p_offset, UIO_SYSSPACE, 0, (rlim64_t)0,
673 		    CRED(), &resid)) != 0) {
674 			uprintf("%s: Cannot obtain interpreter pathname\n",
675 			    exec_file);
676 			goto bad;
677 		}
678 
679 		if (resid != 0 || dlnp[dlnsize - 1] != '\0')
680 			goto bad;
681 
682 		/*
683 		 * Search for '$ORIGIN' token in interpreter path.
684 		 * If found, expand it.
685 		 */
686 		for (p = dlnp; p = strchr(p, '$'); ) {
687 			uint_t	len, curlen;
688 			char	*_ptr;
689 
690 			if (strncmp(++p, ORIGIN_STR, ORIGIN_STR_SIZE))
691 				continue;
692 
693 			/*
694 			 * We don't support $ORIGIN on setid programs to close
695 			 * a potential attack vector.
696 			 */
697 			if ((setid & EXECSETID_SETID) != 0) {
698 				error = ENOEXEC;
699 				goto bad;
700 			}
701 
702 			curlen = 0;
703 			len = p - dlnp - 1;
704 			if (len) {
705 				bcopy(dlnp, pathbufp, len);
706 				curlen += len;
707 			}
708 			if (_ptr = strrchr(args->pathname, '/')) {
709 				len = _ptr - args->pathname;
710 				if ((curlen + len) > MAXPATHLEN)
711 					break;
712 
713 				bcopy(args->pathname, &pathbufp[curlen], len);
714 				curlen += len;
715 			} else {
716 				/*
717 				 * executable is a basename found in the
718 				 * current directory.  So - just substitue
719 				 * '.' for ORIGIN.
720 				 */
721 				pathbufp[curlen] = '.';
722 				curlen++;
723 			}
724 			p += ORIGIN_STR_SIZE;
725 			len = strlen(p);
726 
727 			if ((curlen + len) > MAXPATHLEN)
728 				break;
729 			bcopy(p, &pathbufp[curlen], len);
730 			curlen += len;
731 			pathbufp[curlen++] = '\0';
732 			bcopy(pathbufp, dlnp, curlen);
733 		}
734 
735 		/*
736 		 * /usr/lib/ld.so.1 is known to be a symlink to /lib/ld.so.1
737 		 * (and /usr/lib/64/ld.so.1 is a symlink to /lib/64/ld.so.1).
738 		 * Just in case /usr is not mounted, change it now.
739 		 */
740 		if (strcmp(dlnp, USR_LIB_RTLD) == 0)
741 			dlnp += 4;
742 		error = lookupname(dlnp, UIO_SYSSPACE, FOLLOW, NULLVPP, &nvp);
743 		if (error && dlnp != bigwad->dl_name) {
744 			/* new kernel, old user-level */
745 			error = lookupname(dlnp -= 4, UIO_SYSSPACE, FOLLOW,
746 			    NULLVPP, &nvp);
747 		}
748 		if (error) {
749 			uprintf("%s: Cannot find %s\n", exec_file, dlnp);
750 			goto bad;
751 		}
752 
753 		/*
754 		 * Setup the "aux" vector.
755 		 */
756 		if (uphdr) {
757 			if (ehdrp->e_type == ET_DYN) {
758 				/* don't use the first page */
759 				bigwad->exenv.ex_brkbase = (caddr_t)PAGESIZE;
760 				bigwad->exenv.ex_bssbase = (caddr_t)PAGESIZE;
761 			} else {
762 				bigwad->exenv.ex_bssbase = bssbase;
763 				bigwad->exenv.ex_brkbase = brkbase;
764 			}
765 			bigwad->exenv.ex_brksize = brksize;
766 			bigwad->exenv.ex_magic = elfmagic;
767 			bigwad->exenv.ex_vp = vp;
768 			setexecenv(&bigwad->exenv);
769 
770 			ADDAUX(aux, AT_PHDR, uphdr->p_vaddr + voffset)
771 			ADDAUX(aux, AT_PHENT, ehdrp->e_phentsize)
772 			ADDAUX(aux, AT_PHNUM, nphdrs)
773 			ADDAUX(aux, AT_ENTRY, ehdrp->e_entry + voffset)
774 		} else {
775 			if ((error = execopen(&vp, &fd)) != 0) {
776 				VN_RELE(nvp);
777 				goto bad;
778 			}
779 
780 			ADDAUX(aux, AT_EXECFD, fd)
781 		}
782 
783 		if ((error = execpermissions(nvp, &bigwad->vattr, args)) != 0) {
784 			VN_RELE(nvp);
785 			uprintf("%s: Cannot execute %s\n", exec_file, dlnp);
786 			goto bad;
787 		}
788 
789 		/*
790 		 * Now obtain the ELF header along with the entire program
791 		 * header contained in "nvp".
792 		 */
793 		kmem_free(phdrbase, phdrsize);
794 		phdrbase = NULL;
795 		if ((error = getelfhead(nvp, CRED(), ehdrp, &nshdrs,
796 		    &shstrndx, &nphdrs)) != 0 ||
797 		    (error = getelfphdr(nvp, CRED(), ehdrp, nphdrs, &phdrbase,
798 		    &phdrsize)) != 0) {
799 			VN_RELE(nvp);
800 			uprintf("%s: Cannot read %s\n", exec_file, dlnp);
801 			goto bad;
802 		}
803 
804 		/*
805 		 * Determine memory size of the "interpreter's" loadable
806 		 * sections.  This size is then used to obtain the virtual
807 		 * address of a hole, in the user's address space, large
808 		 * enough to map the "interpreter".
809 		 */
810 		if ((len = elfsize(ehdrp, nphdrs, phdrbase, &lddata)) == 0) {
811 			VN_RELE(nvp);
812 			uprintf("%s: Nothing to load in %s\n", exec_file, dlnp);
813 			goto bad;
814 		}
815 
816 		dtrphdr = NULL;
817 
818 		error = mapelfexec(nvp, ehdrp, nphdrs, phdrbase, &junk, &junk,
819 		    &junk, &dtrphdr, NULL, NULL, NULL, &voffset, NULL, len,
820 		    execsz, NULL);
821 		if (error || junk != NULL) {
822 			VN_RELE(nvp);
823 			uprintf("%s: Cannot map %s\n", exec_file, dlnp);
824 			goto bad;
825 		}
826 
827 		/*
828 		 * We use the DTrace program header to initialize the
829 		 * architecture-specific user per-LWP location. The dtrace
830 		 * fasttrap provider requires ready access to per-LWP scratch
831 		 * space. We assume that there is only one such program header
832 		 * in the interpreter.
833 		 */
834 		if (dtrphdr != NULL &&
835 		    dtrace_safe_phdr(dtrphdr, args, voffset) != 0) {
836 			VN_RELE(nvp);
837 			uprintf("%s: Bad DTrace phdr in %s\n", exec_file, dlnp);
838 			goto bad;
839 		}
840 
841 		VN_RELE(nvp);
842 		ADDAUX(aux, AT_SUN_LDDATA, voffset + lddata)
843 	}
844 
845 	if (hasauxv) {
846 		int auxf = AF_SUN_HWCAPVERIFY;
847 #if defined(__amd64)
848 		size_t fpsize;
849 		int fptype;
850 #endif /* defined(__amd64) */
851 
852 		/*
853 		 * Note: AT_SUN_PLATFORM and AT_SUN_EXECNAME were filled in via
854 		 * exec_args()
855 		 */
856 		ADDAUX(aux, AT_BASE, voffset)
857 		ADDAUX(aux, AT_FLAGS, at_flags)
858 		ADDAUX(aux, AT_PAGESZ, PAGESIZE)
859 		/*
860 		 * Linker flags. (security)
861 		 * p_flag not yet set at this time.
862 		 * We rely on gexec() to provide us with the information.
863 		 * If the application is set-uid but this is not reflected
864 		 * in a mismatch between real/effective uids/gids, then
865 		 * don't treat this as a set-uid exec.  So we care about
866 		 * the EXECSETID_UGIDS flag but not the ...SETID flag.
867 		 */
868 		if ((setid &= ~EXECSETID_SETID) != 0)
869 			auxf |= AF_SUN_SETUGID;
870 
871 		/*
872 		 * If we're running a native process from within a branded
873 		 * zone under pfexec then we clear the AF_SUN_SETUGID flag so
874 		 * that the native ld.so.1 is able to link with the native
875 		 * libraries instead of using the brand libraries that are
876 		 * installed in the zone.  We only do this for processes
877 		 * which we trust because we see they are already running
878 		 * under pfexec (where uid != euid).  This prevents a
879 		 * malicious user within the zone from crafting a wrapper to
880 		 * run native suid commands with unsecure libraries interposed.
881 		 */
882 		if ((brand_action == EBA_NATIVE) && (PROC_IS_BRANDED(p) &&
883 		    (setid &= ~EXECSETID_SETID) != 0))
884 			auxf &= ~AF_SUN_SETUGID;
885 
886 		/*
887 		 * Record the user addr of the auxflags aux vector entry
888 		 * since brands may optionally want to manipulate this field.
889 		 */
890 		args->auxp_auxflags =
891 		    (char *)((char *)args->stackend +
892 		    ((char *)&aux->a_type -
893 		    (char *)bigwad->elfargs));
894 		ADDAUX(aux, AT_SUN_AUXFLAGS, auxf);
895 
896 		/*
897 		 * Hardware capability flag word (performance hints)
898 		 * Used for choosing faster library routines.
899 		 * (Potentially different between 32-bit and 64-bit ABIs)
900 		 */
901 #if defined(_LP64)
902 		if (args->to_model == DATAMODEL_NATIVE) {
903 			ADDAUX(aux, AT_SUN_HWCAP, auxv_hwcap)
904 			ADDAUX(aux, AT_SUN_HWCAP2, auxv_hwcap_2)
905 		} else {
906 			ADDAUX(aux, AT_SUN_HWCAP, auxv_hwcap32)
907 			ADDAUX(aux, AT_SUN_HWCAP2, auxv_hwcap32_2)
908 		}
909 #else
910 		ADDAUX(aux, AT_SUN_HWCAP, auxv_hwcap)
911 		ADDAUX(aux, AT_SUN_HWCAP2, auxv_hwcap_2)
912 #endif
913 		if (branded) {
914 			/*
915 			 * Reserve space for the brand-private aux vectors,
916 			 * and record the user addr of that space.
917 			 */
918 			args->auxp_brand =
919 			    (char *)((char *)args->stackend +
920 			    ((char *)&aux->a_type -
921 			    (char *)bigwad->elfargs));
922 			ADDAUX(aux, AT_SUN_BRAND_AUX1, 0)
923 			ADDAUX(aux, AT_SUN_BRAND_AUX2, 0)
924 			ADDAUX(aux, AT_SUN_BRAND_AUX3, 0)
925 		}
926 
927 		/*
928 		 * Add the comm page auxv entry, mapping it in if needed. Also
929 		 * take care of the FPU entries.
930 		 */
931 #if defined(__amd64)
932 		if (args->commpage != (uintptr_t)NULL ||
933 		    (args->commpage = (uintptr_t)comm_page_mapin()) !=
934 		    (uintptr_t)NULL) {
935 			ADDAUX(aux, AT_SUN_COMMPAGE, args->commpage)
936 		} else {
937 			/*
938 			 * If the comm page cannot be mapped, pad out the auxv
939 			 * to satisfy later size checks.
940 			 */
941 			ADDAUX(aux, AT_NULL, 0)
942 		}
943 
944 		fptype = AT_386_FPINFO_NONE;
945 		fpu_auxv_info(&fptype, &fpsize);
946 		if (fptype != AT_386_FPINFO_NONE) {
947 			ADDAUX(aux, AT_SUN_FPTYPE, fptype)
948 			ADDAUX(aux, AT_SUN_FPSIZE, fpsize)
949 		} else {
950 			ADDAUX(aux, AT_NULL, 0)
951 			ADDAUX(aux, AT_NULL, 0)
952 		}
953 #endif /* defined(__amd64) */
954 
955 		ADDAUX(aux, AT_NULL, 0)
956 		postfixsize = (char *)aux - (char *)bigwad->elfargs;
957 
958 		/*
959 		 * We make assumptions above when we determine how many aux
960 		 * vector entries we will be adding. However, if we have an
961 		 * invalid elf file, it is possible that mapelfexec might
962 		 * behave differently (but not return an error), in which case
963 		 * the number of aux entries we actually add will be different.
964 		 * We detect that now and error out.
965 		 */
966 		if (postfixsize != args->auxsize) {
967 			DTRACE_PROBE2(elfexec_badaux, int, postfixsize,
968 			    int, args->auxsize);
969 			goto bad;
970 		}
971 		ASSERT(postfixsize <= __KERN_NAUXV_IMPL * sizeof (aux_entry_t));
972 	}
973 
974 	/*
975 	 * For the 64-bit kernel, the limit is big enough that rounding it up
976 	 * to a page can overflow the 64-bit limit, so we check for btopr()
977 	 * overflowing here by comparing it with the unrounded limit in pages.
978 	 * If it hasn't overflowed, compare the exec size with the rounded up
979 	 * limit in pages.  Otherwise, just compare with the unrounded limit.
980 	 */
981 	limit = btop(p->p_vmem_ctl);
982 	roundlimit = btopr(p->p_vmem_ctl);
983 	if ((roundlimit > limit && *execsz > roundlimit) ||
984 	    (roundlimit < limit && *execsz > limit)) {
985 		mutex_enter(&p->p_lock);
986 		(void) rctl_action(rctlproc_legacy[RLIMIT_VMEM], p->p_rctls, p,
987 		    RCA_SAFE);
988 		mutex_exit(&p->p_lock);
989 		error = ENOMEM;
990 		goto bad;
991 	}
992 
993 	bzero(up->u_auxv, sizeof (up->u_auxv));
994 	up->u_commpagep = args->commpage;
995 	if (postfixsize) {
996 		int num_auxv;
997 
998 		/*
999 		 * Copy the aux vector to the user stack.
1000 		 */
1001 		error = execpoststack(args, bigwad->elfargs, postfixsize);
1002 		if (error)
1003 			goto bad;
1004 
1005 		/*
1006 		 * Copy auxv to the process's user structure for use by /proc.
1007 		 * If this is a branded process, the brand's exec routine will
1008 		 * copy it's private entries to the user structure later. It
1009 		 * relies on the fact that the blank entries are at the end.
1010 		 */
1011 		num_auxv = postfixsize / sizeof (aux_entry_t);
1012 		ASSERT(num_auxv <= sizeof (up->u_auxv) / sizeof (auxv_t));
1013 		aux = bigwad->elfargs;
1014 		for (i = 0; i < num_auxv; i++) {
1015 			up->u_auxv[i].a_type = aux[i].a_type;
1016 			up->u_auxv[i].a_un.a_val = (aux_val_t)aux[i].a_un.a_val;
1017 		}
1018 	}
1019 
1020 	/*
1021 	 * Pass back the starting address so we can set the program counter.
1022 	 */
1023 	args->entry = (uintptr_t)(ehdrp->e_entry + voffset);
1024 
1025 	if (!uphdr) {
1026 		if (ehdrp->e_type == ET_DYN) {
1027 			/*
1028 			 * If we are executing a shared library which doesn't
1029 			 * have a interpreter (probably ld.so.1) then
1030 			 * we don't set the brkbase now.  Instead we
1031 			 * delay it's setting until the first call
1032 			 * via grow.c::brk().  This permits ld.so.1 to
1033 			 * initialize brkbase to the tail of the executable it
1034 			 * loads (which is where it needs to be).
1035 			 */
1036 			bigwad->exenv.ex_brkbase = (caddr_t)0;
1037 			bigwad->exenv.ex_bssbase = (caddr_t)0;
1038 			bigwad->exenv.ex_brksize = 0;
1039 		} else {
1040 			bigwad->exenv.ex_brkbase = brkbase;
1041 			bigwad->exenv.ex_bssbase = bssbase;
1042 			bigwad->exenv.ex_brksize = brksize;
1043 		}
1044 		bigwad->exenv.ex_magic = elfmagic;
1045 		bigwad->exenv.ex_vp = vp;
1046 		setexecenv(&bigwad->exenv);
1047 	}
1048 
1049 	ASSERT(error == 0);
1050 	goto out;
1051 
1052 bad:
1053 	if (fd != -1)		/* did we open the a.out yet */
1054 		(void) execclose(fd);
1055 
1056 	psignal(p, SIGKILL);
1057 
1058 	if (error == 0)
1059 		error = ENOEXEC;
1060 out:
1061 	if (phdrbase != NULL)
1062 		kmem_free(phdrbase, phdrsize);
1063 	if (cap != NULL)
1064 		kmem_free(cap, capsize);
1065 	kmem_free(bigwad, sizeof (struct bigwad));
1066 	return (error);
1067 }
1068 
1069 /*
1070  * Compute the memory size requirement for the ELF file.
1071  */
1072 static size_t
1073 elfsize(Ehdr *ehdrp, int nphdrs, caddr_t phdrbase, uintptr_t *lddata)
1074 {
1075 	size_t	len;
1076 	Phdr	*phdrp = (Phdr *)phdrbase;
1077 	int	hsize = ehdrp->e_phentsize;
1078 	int	first = 1;
1079 	int	dfirst = 1;	/* first data segment */
1080 	uintptr_t loaddr = 0;
1081 	uintptr_t hiaddr = 0;
1082 	uintptr_t lo, hi;
1083 	int	i;
1084 
1085 	for (i = nphdrs; i > 0; i--) {
1086 		if (phdrp->p_type == PT_LOAD) {
1087 			lo = phdrp->p_vaddr;
1088 			hi = lo + phdrp->p_memsz;
1089 			if (first) {
1090 				loaddr = lo;
1091 				hiaddr = hi;
1092 				first = 0;
1093 			} else {
1094 				if (loaddr > lo)
1095 					loaddr = lo;
1096 				if (hiaddr < hi)
1097 					hiaddr = hi;
1098 			}
1099 
1100 			/*
1101 			 * save the address of the first data segment
1102 			 * of a object - used for the AT_SUNW_LDDATA
1103 			 * aux entry.
1104 			 */
1105 			if ((lddata != NULL) && dfirst &&
1106 			    (phdrp->p_flags & PF_W)) {
1107 				*lddata = lo;
1108 				dfirst = 0;
1109 			}
1110 		}
1111 		phdrp = (Phdr *)((caddr_t)phdrp + hsize);
1112 	}
1113 
1114 	len = hiaddr - (loaddr & PAGEMASK);
1115 	len = roundup(len, PAGESIZE);
1116 
1117 	return (len);
1118 }
1119 
1120 /*
1121  * Read in the ELF header and program header table.
1122  * SUSV3 requires:
1123  *	ENOEXEC	File format is not recognized
1124  *	EINVAL	Format recognized but execution not supported
1125  */
1126 static int
1127 getelfhead(vnode_t *vp, cred_t *credp, Ehdr *ehdr, int *nshdrs, int *shstrndx,
1128     int *nphdrs)
1129 {
1130 	int error;
1131 	ssize_t resid;
1132 
1133 	/*
1134 	 * We got here by the first two bytes in ident,
1135 	 * now read the entire ELF header.
1136 	 */
1137 	if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)ehdr,
1138 	    sizeof (Ehdr), (offset_t)0, UIO_SYSSPACE, 0,
1139 	    (rlim64_t)0, credp, &resid)) != 0)
1140 		return (error);
1141 
1142 	/*
1143 	 * Since a separate version is compiled for handling 32-bit and
1144 	 * 64-bit ELF executables on a 64-bit kernel, the 64-bit version
1145 	 * doesn't need to be able to deal with 32-bit ELF files.
1146 	 */
1147 	if (resid != 0 ||
1148 	    ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
1149 	    ehdr->e_ident[EI_MAG3] != ELFMAG3)
1150 		return (ENOEXEC);
1151 
1152 	if ((ehdr->e_type != ET_EXEC && ehdr->e_type != ET_DYN) ||
1153 #if defined(_ILP32) || defined(_ELF32_COMPAT)
1154 	    ehdr->e_ident[EI_CLASS] != ELFCLASS32 ||
1155 #else
1156 	    ehdr->e_ident[EI_CLASS] != ELFCLASS64 ||
1157 #endif
1158 	    !elfheadcheck(ehdr->e_ident[EI_DATA], ehdr->e_machine,
1159 	    ehdr->e_flags))
1160 		return (EINVAL);
1161 
1162 	*nshdrs = ehdr->e_shnum;
1163 	*shstrndx = ehdr->e_shstrndx;
1164 	*nphdrs = ehdr->e_phnum;
1165 
1166 	/*
1167 	 * If e_shnum, e_shstrndx, or e_phnum is its sentinel value, we need
1168 	 * to read in the section header at index zero to acces the true
1169 	 * values for those fields.
1170 	 */
1171 	if ((*nshdrs == 0 && ehdr->e_shoff != 0) ||
1172 	    *shstrndx == SHN_XINDEX || *nphdrs == PN_XNUM) {
1173 		Shdr shdr;
1174 
1175 		if (ehdr->e_shoff == 0)
1176 			return (EINVAL);
1177 
1178 		if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)&shdr,
1179 		    sizeof (shdr), (offset_t)ehdr->e_shoff, UIO_SYSSPACE, 0,
1180 		    (rlim64_t)0, credp, &resid)) != 0)
1181 			return (error);
1182 
1183 		if (*nshdrs == 0)
1184 			*nshdrs = shdr.sh_size;
1185 		if (*shstrndx == SHN_XINDEX)
1186 			*shstrndx = shdr.sh_link;
1187 		if (*nphdrs == PN_XNUM && shdr.sh_info != 0)
1188 			*nphdrs = shdr.sh_info;
1189 	}
1190 
1191 	return (0);
1192 }
1193 
1194 #ifdef _ELF32_COMPAT
1195 extern size_t elf_nphdr_max;
1196 #else
1197 size_t elf_nphdr_max = 1000;
1198 #endif
1199 
1200 static int
1201 getelfphdr(vnode_t *vp, cred_t *credp, const Ehdr *ehdr, int nphdrs,
1202     caddr_t *phbasep, ssize_t *phsizep)
1203 {
1204 	ssize_t resid, minsize;
1205 	int err;
1206 
1207 	/*
1208 	 * Since we're going to be using e_phentsize to iterate down the
1209 	 * array of program headers, it must be 8-byte aligned or else
1210 	 * a we might cause a misaligned access. We use all members through
1211 	 * p_flags on 32-bit ELF files and p_memsz on 64-bit ELF files so
1212 	 * e_phentsize must be at least large enough to include those
1213 	 * members.
1214 	 */
1215 #if !defined(_LP64) || defined(_ELF32_COMPAT)
1216 	minsize = offsetof(Phdr, p_flags) + sizeof (((Phdr *)NULL)->p_flags);
1217 #else
1218 	minsize = offsetof(Phdr, p_memsz) + sizeof (((Phdr *)NULL)->p_memsz);
1219 #endif
1220 	if (ehdr->e_phentsize < minsize || (ehdr->e_phentsize & 3))
1221 		return (EINVAL);
1222 
1223 	*phsizep = nphdrs * ehdr->e_phentsize;
1224 
1225 	if (*phsizep > sizeof (Phdr) * elf_nphdr_max) {
1226 		if ((*phbasep = kmem_alloc(*phsizep, KM_NOSLEEP)) == NULL)
1227 			return (ENOMEM);
1228 	} else {
1229 		*phbasep = kmem_alloc(*phsizep, KM_SLEEP);
1230 	}
1231 
1232 	if ((err = vn_rdwr(UIO_READ, vp, *phbasep, *phsizep,
1233 	    (offset_t)ehdr->e_phoff, UIO_SYSSPACE, 0, (rlim64_t)0,
1234 	    credp, &resid)) != 0) {
1235 		kmem_free(*phbasep, *phsizep);
1236 		*phbasep = NULL;
1237 		return (err);
1238 	}
1239 
1240 	return (0);
1241 }
1242 
1243 #ifdef _ELF32_COMPAT
1244 extern size_t elf_nshdr_max;
1245 extern size_t elf_shstrtab_max;
1246 #else
1247 size_t elf_nshdr_max = 10000;
1248 size_t elf_shstrtab_max = 100 * 1024;
1249 #endif
1250 
1251 
1252 static int
1253 getelfshdr(vnode_t *vp, cred_t *credp, const Ehdr *ehdr,
1254     int nshdrs, int shstrndx, caddr_t *shbasep, ssize_t *shsizep,
1255     char **shstrbasep, ssize_t *shstrsizep)
1256 {
1257 	ssize_t resid, minsize;
1258 	int err;
1259 	Shdr *shdr;
1260 
1261 	/*
1262 	 * Since we're going to be using e_shentsize to iterate down the
1263 	 * array of section headers, it must be 8-byte aligned or else
1264 	 * a we might cause a misaligned access. We use all members through
1265 	 * sh_entsize (on both 32- and 64-bit ELF files) so e_shentsize
1266 	 * must be at least large enough to include that member. The index
1267 	 * of the string table section must also be valid.
1268 	 */
1269 	minsize = offsetof(Shdr, sh_entsize) + sizeof (shdr->sh_entsize);
1270 	if (ehdr->e_shentsize < minsize || (ehdr->e_shentsize & 3) ||
1271 	    shstrndx >= nshdrs)
1272 		return (EINVAL);
1273 
1274 	*shsizep = nshdrs * ehdr->e_shentsize;
1275 
1276 	if (*shsizep > sizeof (Shdr) * elf_nshdr_max) {
1277 		if ((*shbasep = kmem_alloc(*shsizep, KM_NOSLEEP)) == NULL)
1278 			return (ENOMEM);
1279 	} else {
1280 		*shbasep = kmem_alloc(*shsizep, KM_SLEEP);
1281 	}
1282 
1283 	if ((err = vn_rdwr(UIO_READ, vp, *shbasep, *shsizep,
1284 	    (offset_t)ehdr->e_shoff, UIO_SYSSPACE, 0, (rlim64_t)0,
1285 	    credp, &resid)) != 0) {
1286 		kmem_free(*shbasep, *shsizep);
1287 		return (err);
1288 	}
1289 
1290 	/*
1291 	 * Pull the section string table out of the vnode; fail if the size
1292 	 * is zero.
1293 	 */
1294 	shdr = (Shdr *)(*shbasep + shstrndx * ehdr->e_shentsize);
1295 	if ((*shstrsizep = shdr->sh_size) == 0) {
1296 		kmem_free(*shbasep, *shsizep);
1297 		return (EINVAL);
1298 	}
1299 
1300 	if (*shstrsizep > elf_shstrtab_max) {
1301 		if ((*shstrbasep = kmem_alloc(*shstrsizep,
1302 		    KM_NOSLEEP)) == NULL) {
1303 			kmem_free(*shbasep, *shsizep);
1304 			return (ENOMEM);
1305 		}
1306 	} else {
1307 		*shstrbasep = kmem_alloc(*shstrsizep, KM_SLEEP);
1308 	}
1309 
1310 	if ((err = vn_rdwr(UIO_READ, vp, *shstrbasep, *shstrsizep,
1311 	    (offset_t)shdr->sh_offset, UIO_SYSSPACE, 0, (rlim64_t)0,
1312 	    credp, &resid)) != 0) {
1313 		kmem_free(*shbasep, *shsizep);
1314 		kmem_free(*shstrbasep, *shstrsizep);
1315 		return (err);
1316 	}
1317 
1318 	/*
1319 	 * Make sure the strtab is null-terminated to make sure we
1320 	 * don't run off the end of the table.
1321 	 */
1322 	(*shstrbasep)[*shstrsizep - 1] = '\0';
1323 
1324 	return (0);
1325 }
1326 
1327 static int
1328 mapelfexec(
1329 	vnode_t *vp,
1330 	Ehdr *ehdr,
1331 	int nphdrs,
1332 	caddr_t phdrbase,
1333 	Phdr **uphdr,
1334 	Phdr **intphdr,
1335 	Phdr **stphdr,
1336 	Phdr **dtphdr,
1337 	Phdr *dataphdrp,
1338 	caddr_t *bssbase,
1339 	caddr_t *brkbase,
1340 	intptr_t *voffset,
1341 	intptr_t *minaddr,
1342 	size_t len,
1343 	long *execsz,
1344 	size_t *brksize)
1345 {
1346 	Phdr *phdr;
1347 	int i, prot, error;
1348 	caddr_t addr = NULL;
1349 	size_t zfodsz;
1350 	int ptload = 0;
1351 	int page;
1352 	off_t offset;
1353 	int hsize = ehdr->e_phentsize;
1354 	caddr_t mintmp = (caddr_t)-1;
1355 	extern int use_brk_lpg;
1356 
1357 	if (ehdr->e_type == ET_DYN) {
1358 		secflagset_t flags = 0;
1359 		/*
1360 		 * Obtain the virtual address of a hole in the
1361 		 * address space to map the "interpreter".
1362 		 */
1363 		if (secflag_enabled(curproc, PROC_SEC_ASLR))
1364 			flags |= _MAP_RANDOMIZE;
1365 
1366 		map_addr(&addr, len, (offset_t)0, 1, flags);
1367 		if (addr == NULL)
1368 			return (ENOMEM);
1369 		*voffset = (intptr_t)addr;
1370 
1371 		/*
1372 		 * Calculate the minimum vaddr so it can be subtracted out.
1373 		 * According to the ELF specification, since PT_LOAD sections
1374 		 * must be sorted by increasing p_vaddr values, this is
1375 		 * guaranteed to be the first PT_LOAD section.
1376 		 */
1377 		phdr = (Phdr *)phdrbase;
1378 		for (i = nphdrs; i > 0; i--) {
1379 			if (phdr->p_type == PT_LOAD) {
1380 				*voffset -= (uintptr_t)phdr->p_vaddr;
1381 				break;
1382 			}
1383 			phdr = (Phdr *)((caddr_t)phdr + hsize);
1384 		}
1385 
1386 	} else {
1387 		*voffset = 0;
1388 	}
1389 	phdr = (Phdr *)phdrbase;
1390 	for (i = nphdrs; i > 0; i--) {
1391 		switch (phdr->p_type) {
1392 		case PT_LOAD:
1393 			if ((*intphdr != NULL) && (*uphdr == NULL))
1394 				return (0);
1395 
1396 			ptload = 1;
1397 			prot = PROT_USER;
1398 			if (phdr->p_flags & PF_R)
1399 				prot |= PROT_READ;
1400 			if (phdr->p_flags & PF_W)
1401 				prot |= PROT_WRITE;
1402 			if (phdr->p_flags & PF_X)
1403 				prot |= PROT_EXEC;
1404 
1405 			addr = (caddr_t)((uintptr_t)phdr->p_vaddr + *voffset);
1406 
1407 			/*
1408 			 * Keep track of the segment with the lowest starting
1409 			 * address.
1410 			 */
1411 			if (addr < mintmp)
1412 				mintmp = addr;
1413 
1414 			zfodsz = (size_t)phdr->p_memsz - phdr->p_filesz;
1415 
1416 			offset = phdr->p_offset;
1417 			if (((uintptr_t)offset & PAGEOFFSET) ==
1418 			    ((uintptr_t)addr & PAGEOFFSET) &&
1419 			    (!(vp->v_flag & VNOMAP))) {
1420 				page = 1;
1421 			} else {
1422 				page = 0;
1423 			}
1424 
1425 			/*
1426 			 * Set the heap pagesize for OOB when the bss size
1427 			 * is known and use_brk_lpg is not 0.
1428 			 */
1429 			if (brksize != NULL && use_brk_lpg &&
1430 			    zfodsz != 0 && phdr == dataphdrp &&
1431 			    (prot & PROT_WRITE)) {
1432 				size_t tlen = P2NPHASE((uintptr_t)addr +
1433 				    phdr->p_filesz, PAGESIZE);
1434 
1435 				if (zfodsz > tlen) {
1436 					curproc->p_brkpageszc =
1437 					    page_szc(map_pgsz(MAPPGSZ_HEAP,
1438 					    curproc, addr + phdr->p_filesz +
1439 					    tlen, zfodsz - tlen, 0));
1440 				}
1441 			}
1442 
1443 			if (curproc->p_brkpageszc != 0 && phdr == dataphdrp &&
1444 			    (prot & PROT_WRITE)) {
1445 				uint_t	szc = curproc->p_brkpageszc;
1446 				size_t pgsz = page_get_pagesize(szc);
1447 				caddr_t ebss = addr + phdr->p_memsz;
1448 				/*
1449 				 * If we need extra space to keep the BSS an
1450 				 * integral number of pages in size, some of
1451 				 * that space may fall beyond p_brkbase, so we
1452 				 * need to set p_brksize to account for it
1453 				 * being (logically) part of the brk.
1454 				 */
1455 				size_t extra_zfodsz;
1456 
1457 				ASSERT(pgsz > PAGESIZE);
1458 
1459 				extra_zfodsz = P2NPHASE((uintptr_t)ebss, pgsz);
1460 
1461 				if (error = execmap(vp, addr, phdr->p_filesz,
1462 				    zfodsz + extra_zfodsz, phdr->p_offset,
1463 				    prot, page, szc))
1464 					goto bad;
1465 				if (brksize != NULL)
1466 					*brksize = extra_zfodsz;
1467 			} else {
1468 				if (error = execmap(vp, addr, phdr->p_filesz,
1469 				    zfodsz, phdr->p_offset, prot, page, 0))
1470 					goto bad;
1471 			}
1472 
1473 			if (bssbase != NULL && addr >= *bssbase &&
1474 			    phdr == dataphdrp) {
1475 				*bssbase = addr + phdr->p_filesz;
1476 			}
1477 			if (brkbase != NULL && addr >= *brkbase) {
1478 				*brkbase = addr + phdr->p_memsz;
1479 			}
1480 
1481 			*execsz += btopr(phdr->p_memsz);
1482 			break;
1483 
1484 		case PT_INTERP:
1485 			if (ptload)
1486 				goto bad;
1487 			*intphdr = phdr;
1488 			break;
1489 
1490 		case PT_SHLIB:
1491 			*stphdr = phdr;
1492 			break;
1493 
1494 		case PT_PHDR:
1495 			if (ptload)
1496 				goto bad;
1497 			*uphdr = phdr;
1498 			break;
1499 
1500 		case PT_NULL:
1501 		case PT_DYNAMIC:
1502 		case PT_NOTE:
1503 			break;
1504 
1505 		case PT_SUNWDTRACE:
1506 			if (dtphdr != NULL)
1507 				*dtphdr = phdr;
1508 			break;
1509 
1510 		default:
1511 			break;
1512 		}
1513 		phdr = (Phdr *)((caddr_t)phdr + hsize);
1514 	}
1515 
1516 	if (minaddr != NULL) {
1517 		ASSERT(mintmp != (caddr_t)-1);
1518 		*minaddr = (intptr_t)mintmp;
1519 	}
1520 
1521 	if (brkbase != NULL && secflag_enabled(curproc, PROC_SEC_ASLR)) {
1522 		size_t off;
1523 		uintptr_t base = (uintptr_t)*brkbase;
1524 		uintptr_t oend = base + *brksize;
1525 
1526 		ASSERT(ISP2(aslr_max_brk_skew));
1527 
1528 		(void) random_get_pseudo_bytes((uint8_t *)&off, sizeof (off));
1529 		base += P2PHASE(off, aslr_max_brk_skew);
1530 		base = P2ROUNDUP(base, PAGESIZE);
1531 		*brkbase = (caddr_t)base;
1532 		/*
1533 		 * Above, we set *brksize to account for the possibility we
1534 		 * had to grow the 'brk' in padding out the BSS to a page
1535 		 * boundary.
1536 		 *
1537 		 * We now need to adjust that based on where we now are
1538 		 * actually putting the brk.
1539 		 */
1540 		if (oend > base)
1541 			*brksize = oend - base;
1542 		else
1543 			*brksize = 0;
1544 	}
1545 
1546 	return (0);
1547 bad:
1548 	if (error == 0)
1549 		error = EINVAL;
1550 	return (error);
1551 }
1552 
1553 int
1554 elfnote(vnode_t *vp, offset_t *offsetp, int type, int descsz, void *desc,
1555     rlim64_t rlimit, cred_t *credp)
1556 {
1557 	Note note;
1558 	int error;
1559 
1560 	bzero(&note, sizeof (note));
1561 	bcopy("CORE", note.name, 4);
1562 	note.nhdr.n_type = type;
1563 	/*
1564 	 * The System V ABI states that n_namesz must be the length of the
1565 	 * string that follows the Nhdr structure including the terminating
1566 	 * null. The ABI also specifies that sufficient padding should be
1567 	 * included so that the description that follows the name string
1568 	 * begins on a 4- or 8-byte boundary for 32- and 64-bit binaries
1569 	 * respectively. However, since this change was not made correctly
1570 	 * at the time of the 64-bit port, both 32- and 64-bit binaries
1571 	 * descriptions are only guaranteed to begin on a 4-byte boundary.
1572 	 */
1573 	note.nhdr.n_namesz = 5;
1574 	note.nhdr.n_descsz = roundup(descsz, sizeof (Word));
1575 
1576 	if (error = core_write(vp, UIO_SYSSPACE, *offsetp, &note,
1577 	    sizeof (note), rlimit, credp))
1578 		return (error);
1579 
1580 	*offsetp += sizeof (note);
1581 
1582 	if (error = core_write(vp, UIO_SYSSPACE, *offsetp, desc,
1583 	    note.nhdr.n_descsz, rlimit, credp))
1584 		return (error);
1585 
1586 	*offsetp += note.nhdr.n_descsz;
1587 	return (0);
1588 }
1589 
1590 /*
1591  * Copy the section data from one vnode to the section of another vnode.
1592  */
1593 static void
1594 copy_scn(Shdr *src, vnode_t *src_vp, Shdr *dst, vnode_t *dst_vp, Off *doffset,
1595     void *buf, size_t size, cred_t *credp, rlim64_t rlimit)
1596 {
1597 	ssize_t resid;
1598 	size_t len, n = src->sh_size;
1599 	offset_t off = 0;
1600 
1601 	while (n != 0) {
1602 		len = MIN(size, n);
1603 		if (vn_rdwr(UIO_READ, src_vp, buf, len, src->sh_offset + off,
1604 		    UIO_SYSSPACE, 0, (rlim64_t)0, credp, &resid) != 0 ||
1605 		    resid >= len ||
1606 		    core_write(dst_vp, UIO_SYSSPACE, *doffset + off,
1607 		    buf, len - resid, rlimit, credp) != 0) {
1608 			dst->sh_size = 0;
1609 			dst->sh_offset = 0;
1610 			return;
1611 		}
1612 
1613 		ASSERT(n >= len - resid);
1614 
1615 		n -= len - resid;
1616 		off += len - resid;
1617 	}
1618 
1619 	*doffset += src->sh_size;
1620 }
1621 
1622 #ifdef _ELF32_COMPAT
1623 extern size_t elf_datasz_max;
1624 extern size_t elf_zeropg_sz;
1625 #else
1626 size_t elf_datasz_max = 1 * 1024 * 1024;
1627 size_t elf_zeropg_sz = 4 * 1024;
1628 #endif
1629 
1630 /*
1631  * This function processes mappings that correspond to load objects to
1632  * examine their respective sections for elfcore(). It's called once with
1633  * v set to NULL to count the number of sections that we're going to need
1634  * and then again with v set to some allocated buffer that we fill in with
1635  * all the section data.
1636  */
1637 static int
1638 process_scns(core_content_t content, proc_t *p, cred_t *credp, vnode_t *vp,
1639     Shdr *v, int nv, rlim64_t rlimit, Off *doffsetp, int *nshdrsp)
1640 {
1641 	vnode_t *lastvp = NULL;
1642 	struct seg *seg;
1643 	int i, j;
1644 	void *data = NULL;
1645 	size_t datasz = 0;
1646 	shstrtab_t shstrtab;
1647 	struct as *as = p->p_as;
1648 	int error = 0;
1649 
1650 	if (v != NULL)
1651 		shstrtab_init(&shstrtab);
1652 
1653 	i = 1;
1654 	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
1655 		uint_t prot;
1656 		vnode_t *mvp;
1657 		void *tmp = NULL;
1658 		caddr_t saddr = seg->s_base;
1659 		caddr_t naddr;
1660 		caddr_t eaddr;
1661 		size_t segsize;
1662 
1663 		Ehdr ehdr;
1664 		int nshdrs, shstrndx, nphdrs;
1665 		caddr_t shbase;
1666 		ssize_t shsize;
1667 		char *shstrbase;
1668 		ssize_t shstrsize;
1669 
1670 		Shdr *shdr;
1671 		const char *name;
1672 		size_t sz;
1673 		uintptr_t off;
1674 
1675 		int ctf_ndx = 0;
1676 		int symtab_ndx = 0;
1677 
1678 		/*
1679 		 * Since we're just looking for text segments of load
1680 		 * objects, we only care about the protection bits; we don't
1681 		 * care about the actual size of the segment so we use the
1682 		 * reserved size. If the segment's size is zero, there's
1683 		 * something fishy going on so we ignore this segment.
1684 		 */
1685 		if (seg->s_ops != &segvn_ops ||
1686 		    SEGOP_GETVP(seg, seg->s_base, &mvp) != 0 ||
1687 		    mvp == lastvp || mvp == NULL || mvp->v_type != VREG ||
1688 		    (segsize = pr_getsegsize(seg, 1)) == 0)
1689 			continue;
1690 
1691 		eaddr = saddr + segsize;
1692 		prot = pr_getprot(seg, 1, &tmp, &saddr, &naddr, eaddr);
1693 		pr_getprot_done(&tmp);
1694 
1695 		/*
1696 		 * Skip this segment unless the protection bits look like
1697 		 * what we'd expect for a text segment.
1698 		 */
1699 		if ((prot & (PROT_WRITE | PROT_EXEC)) != PROT_EXEC)
1700 			continue;
1701 
1702 		if (getelfhead(mvp, credp, &ehdr, &nshdrs, &shstrndx,
1703 		    &nphdrs) != 0 ||
1704 		    getelfshdr(mvp, credp, &ehdr, nshdrs, shstrndx,
1705 		    &shbase, &shsize, &shstrbase, &shstrsize) != 0)
1706 			continue;
1707 
1708 		off = ehdr.e_shentsize;
1709 		for (j = 1; j < nshdrs; j++, off += ehdr.e_shentsize) {
1710 			Shdr *symtab = NULL, *strtab;
1711 			size_t allocsz;
1712 
1713 			shdr = (Shdr *)(shbase + off);
1714 			allocsz = MIN(shdr->sh_size, elf_datasz_max);
1715 
1716 			if (shdr->sh_name >= shstrsize)
1717 				continue;
1718 
1719 			name = shstrbase + shdr->sh_name;
1720 
1721 			if (strcmp(name, shstrtab_data[STR_CTF]) == 0) {
1722 				if ((content & CC_CONTENT_CTF) == 0 ||
1723 				    ctf_ndx != 0)
1724 					continue;
1725 
1726 				if (shdr->sh_link > 0 &&
1727 				    shdr->sh_link < nshdrs) {
1728 					symtab = (Shdr *)(shbase +
1729 					    shdr->sh_link * ehdr.e_shentsize);
1730 				}
1731 
1732 				if (v != NULL && i < nv - 1) {
1733 					if (allocsz > datasz) {
1734 						if (data != NULL)
1735 							kmem_free(data, datasz);
1736 
1737 						datasz = allocsz;
1738 						data = kmem_alloc(datasz,
1739 						    KM_SLEEP);
1740 					}
1741 
1742 					v[i].sh_name = shstrtab_ndx(&shstrtab,
1743 					    STR_CTF);
1744 					v[i].sh_addr = (Addr)(uintptr_t)saddr;
1745 					v[i].sh_type = SHT_PROGBITS;
1746 					v[i].sh_addralign = 4;
1747 					*doffsetp = roundup(*doffsetp,
1748 					    v[i].sh_addralign);
1749 					v[i].sh_offset = *doffsetp;
1750 					v[i].sh_size = shdr->sh_size;
1751 					if (symtab == NULL)  {
1752 						v[i].sh_link = 0;
1753 					} else if (symtab->sh_type ==
1754 					    SHT_SYMTAB &&
1755 					    symtab_ndx != 0) {
1756 						v[i].sh_link =
1757 						    symtab_ndx;
1758 					} else {
1759 						v[i].sh_link = i + 1;
1760 					}
1761 
1762 					copy_scn(shdr, mvp, &v[i], vp,
1763 					    doffsetp, data, datasz, credp,
1764 					    rlimit);
1765 				}
1766 
1767 				ctf_ndx = i++;
1768 
1769 				/*
1770 				 * We've already dumped the symtab.
1771 				 */
1772 				if (symtab != NULL &&
1773 				    symtab->sh_type == SHT_SYMTAB &&
1774 				    symtab_ndx != 0)
1775 					continue;
1776 
1777 			} else if (strcmp(name,
1778 			    shstrtab_data[STR_SYMTAB]) == 0) {
1779 				if ((content & CC_CONTENT_SYMTAB) == 0 ||
1780 				    symtab != 0)
1781 					continue;
1782 
1783 				symtab = shdr;
1784 			}
1785 
1786 			if (symtab != NULL) {
1787 				if ((symtab->sh_type != SHT_DYNSYM &&
1788 				    symtab->sh_type != SHT_SYMTAB) ||
1789 				    symtab->sh_link == 0 ||
1790 				    symtab->sh_link >= nshdrs)
1791 					continue;
1792 
1793 				strtab = (Shdr *)(shbase +
1794 				    symtab->sh_link * ehdr.e_shentsize);
1795 
1796 				if (strtab->sh_type != SHT_STRTAB)
1797 					continue;
1798 
1799 				if (v != NULL && i < nv - 2) {
1800 					sz = MAX(symtab->sh_size,
1801 					    strtab->sh_size);
1802 					allocsz = MIN(sz, elf_datasz_max);
1803 					if (allocsz > datasz) {
1804 						if (data != NULL)
1805 							kmem_free(data, datasz);
1806 
1807 						datasz = allocsz;
1808 						data = kmem_alloc(datasz,
1809 						    KM_SLEEP);
1810 					}
1811 
1812 					if (symtab->sh_type == SHT_DYNSYM) {
1813 						v[i].sh_name = shstrtab_ndx(
1814 						    &shstrtab, STR_DYNSYM);
1815 						v[i + 1].sh_name = shstrtab_ndx(
1816 						    &shstrtab, STR_DYNSTR);
1817 					} else {
1818 						v[i].sh_name = shstrtab_ndx(
1819 						    &shstrtab, STR_SYMTAB);
1820 						v[i + 1].sh_name = shstrtab_ndx(
1821 						    &shstrtab, STR_STRTAB);
1822 					}
1823 
1824 					v[i].sh_type = symtab->sh_type;
1825 					v[i].sh_addr = symtab->sh_addr;
1826 					if (ehdr.e_type == ET_DYN ||
1827 					    v[i].sh_addr == 0)
1828 						v[i].sh_addr +=
1829 						    (Addr)(uintptr_t)saddr;
1830 					v[i].sh_addralign =
1831 					    symtab->sh_addralign;
1832 					*doffsetp = roundup(*doffsetp,
1833 					    v[i].sh_addralign);
1834 					v[i].sh_offset = *doffsetp;
1835 					v[i].sh_size = symtab->sh_size;
1836 					v[i].sh_link = i + 1;
1837 					v[i].sh_entsize = symtab->sh_entsize;
1838 					v[i].sh_info = symtab->sh_info;
1839 
1840 					copy_scn(symtab, mvp, &v[i], vp,
1841 					    doffsetp, data, datasz, credp,
1842 					    rlimit);
1843 
1844 					v[i + 1].sh_type = SHT_STRTAB;
1845 					v[i + 1].sh_flags = SHF_STRINGS;
1846 					v[i + 1].sh_addr = symtab->sh_addr;
1847 					if (ehdr.e_type == ET_DYN ||
1848 					    v[i + 1].sh_addr == 0)
1849 						v[i + 1].sh_addr +=
1850 						    (Addr)(uintptr_t)saddr;
1851 					v[i + 1].sh_addralign =
1852 					    strtab->sh_addralign;
1853 					*doffsetp = roundup(*doffsetp,
1854 					    v[i + 1].sh_addralign);
1855 					v[i + 1].sh_offset = *doffsetp;
1856 					v[i + 1].sh_size = strtab->sh_size;
1857 
1858 					copy_scn(strtab, mvp, &v[i + 1], vp,
1859 					    doffsetp, data, datasz, credp,
1860 					    rlimit);
1861 				}
1862 
1863 				if (symtab->sh_type == SHT_SYMTAB)
1864 					symtab_ndx = i;
1865 				i += 2;
1866 			}
1867 		}
1868 
1869 		kmem_free(shstrbase, shstrsize);
1870 		kmem_free(shbase, shsize);
1871 
1872 		lastvp = mvp;
1873 	}
1874 
1875 	if (v == NULL) {
1876 		if (i == 1)
1877 			*nshdrsp = 0;
1878 		else
1879 			*nshdrsp = i + 1;
1880 		goto done;
1881 	}
1882 
1883 	if (i != nv - 1) {
1884 		cmn_err(CE_WARN, "elfcore: core dump failed for "
1885 		    "process %d; address space is changing", p->p_pid);
1886 		error = EIO;
1887 		goto done;
1888 	}
1889 
1890 	v[i].sh_name = shstrtab_ndx(&shstrtab, STR_SHSTRTAB);
1891 	v[i].sh_size = shstrtab_size(&shstrtab);
1892 	v[i].sh_addralign = 1;
1893 	*doffsetp = roundup(*doffsetp, v[i].sh_addralign);
1894 	v[i].sh_offset = *doffsetp;
1895 	v[i].sh_flags = SHF_STRINGS;
1896 	v[i].sh_type = SHT_STRTAB;
1897 
1898 	if (v[i].sh_size > datasz) {
1899 		if (data != NULL)
1900 			kmem_free(data, datasz);
1901 
1902 		datasz = v[i].sh_size;
1903 		data = kmem_alloc(datasz,
1904 		    KM_SLEEP);
1905 	}
1906 
1907 	shstrtab_dump(&shstrtab, data);
1908 
1909 	if ((error = core_write(vp, UIO_SYSSPACE, *doffsetp,
1910 	    data, v[i].sh_size, rlimit, credp)) != 0)
1911 		goto done;
1912 
1913 	*doffsetp += v[i].sh_size;
1914 
1915 done:
1916 	if (data != NULL)
1917 		kmem_free(data, datasz);
1918 
1919 	return (error);
1920 }
1921 
1922 int
1923 elfcore(vnode_t *vp, proc_t *p, cred_t *credp, rlim64_t rlimit, int sig,
1924     core_content_t content)
1925 {
1926 	offset_t poffset, soffset;
1927 	Off doffset;
1928 	int error, i, nphdrs, nshdrs;
1929 	int overflow = 0;
1930 	struct seg *seg;
1931 	struct as *as = p->p_as;
1932 	union {
1933 		Ehdr ehdr;
1934 		Phdr phdr[1];
1935 		Shdr shdr[1];
1936 	} *bigwad;
1937 	size_t bigsize;
1938 	size_t phdrsz, shdrsz;
1939 	Ehdr *ehdr;
1940 	Phdr *v;
1941 	void *zeropg = NULL;
1942 	caddr_t brkbase;
1943 	size_t brksize;
1944 	caddr_t stkbase;
1945 	size_t stksize;
1946 	int ntries = 0;
1947 	klwp_t *lwp = ttolwp(curthread);
1948 
1949 top:
1950 	/*
1951 	 * Make sure we have everything we need (registers, etc.).
1952 	 * All other lwps have already stopped and are in an orderly state.
1953 	 */
1954 	ASSERT(p == ttoproc(curthread));
1955 	prstop(0, 0);
1956 
1957 	AS_LOCK_ENTER(as, RW_WRITER);
1958 	nphdrs = prnsegs(as, 0) + 2;		/* two CORE note sections */
1959 
1960 	/*
1961 	 * Count the number of section headers we're going to need.
1962 	 */
1963 	nshdrs = 0;
1964 	if (content & (CC_CONTENT_CTF | CC_CONTENT_SYMTAB)) {
1965 		(void) process_scns(content, p, credp, NULL, NULL, 0, 0,
1966 		    NULL, &nshdrs);
1967 	}
1968 	AS_LOCK_EXIT(as);
1969 
1970 	ASSERT(nshdrs == 0 || nshdrs > 1);
1971 
1972 	/*
1973 	 * The core file contents may required zero section headers, but if
1974 	 * we overflow the 16 bits allotted to the program header count in
1975 	 * the ELF header, we'll need that program header at index zero.
1976 	 */
1977 	if (nshdrs == 0 && nphdrs >= PN_XNUM)
1978 		nshdrs = 1;
1979 
1980 	phdrsz = nphdrs * sizeof (Phdr);
1981 	shdrsz = nshdrs * sizeof (Shdr);
1982 
1983 	bigsize = MAX(sizeof (*bigwad), MAX(phdrsz, shdrsz));
1984 	bigwad = kmem_alloc(bigsize, KM_SLEEP);
1985 
1986 	ehdr = &bigwad->ehdr;
1987 	bzero(ehdr, sizeof (*ehdr));
1988 
1989 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1990 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1991 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1992 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1993 	ehdr->e_ident[EI_CLASS] = ELFCLASS;
1994 	ehdr->e_type = ET_CORE;
1995 
1996 #if !defined(_LP64) || defined(_ELF32_COMPAT)
1997 
1998 #if defined(__sparc)
1999 	ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
2000 	ehdr->e_machine = EM_SPARC;
2001 #elif defined(__i386_COMPAT)
2002 	ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
2003 	ehdr->e_machine = EM_386;
2004 #else
2005 #error "no recognized machine type is defined"
2006 #endif
2007 
2008 #else	/* !defined(_LP64) || defined(_ELF32_COMPAT) */
2009 
2010 #if defined(__sparc)
2011 	ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
2012 	ehdr->e_machine = EM_SPARCV9;
2013 #elif defined(__amd64)
2014 	ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
2015 	ehdr->e_machine = EM_AMD64;
2016 #else
2017 #error "no recognized 64-bit machine type is defined"
2018 #endif
2019 
2020 #endif	/* !defined(_LP64) || defined(_ELF32_COMPAT) */
2021 
2022 	/*
2023 	 * If the count of program headers or section headers or the index
2024 	 * of the section string table can't fit in the mere 16 bits
2025 	 * shortsightedly allotted to them in the ELF header, we use the
2026 	 * extended formats and put the real values in the section header
2027 	 * as index 0.
2028 	 */
2029 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
2030 	ehdr->e_version = EV_CURRENT;
2031 	ehdr->e_ehsize = sizeof (Ehdr);
2032 
2033 	if (nphdrs >= PN_XNUM)
2034 		ehdr->e_phnum = PN_XNUM;
2035 	else
2036 		ehdr->e_phnum = (unsigned short)nphdrs;
2037 
2038 	ehdr->e_phoff = sizeof (Ehdr);
2039 	ehdr->e_phentsize = sizeof (Phdr);
2040 
2041 	if (nshdrs > 0) {
2042 		if (nshdrs >= SHN_LORESERVE)
2043 			ehdr->e_shnum = 0;
2044 		else
2045 			ehdr->e_shnum = (unsigned short)nshdrs;
2046 
2047 		if (nshdrs - 1 >= SHN_LORESERVE)
2048 			ehdr->e_shstrndx = SHN_XINDEX;
2049 		else
2050 			ehdr->e_shstrndx = (unsigned short)(nshdrs - 1);
2051 
2052 		ehdr->e_shoff = ehdr->e_phoff + ehdr->e_phentsize * nphdrs;
2053 		ehdr->e_shentsize = sizeof (Shdr);
2054 	}
2055 
2056 	if (error = core_write(vp, UIO_SYSSPACE, (offset_t)0, ehdr,
2057 	    sizeof (Ehdr), rlimit, credp))
2058 		goto done;
2059 
2060 	poffset = sizeof (Ehdr);
2061 	soffset = sizeof (Ehdr) + phdrsz;
2062 	doffset = sizeof (Ehdr) + phdrsz + shdrsz;
2063 
2064 	v = &bigwad->phdr[0];
2065 	bzero(v, phdrsz);
2066 
2067 	setup_old_note_header(&v[0], p);
2068 	v[0].p_offset = doffset = roundup(doffset, sizeof (Word));
2069 	doffset += v[0].p_filesz;
2070 
2071 	setup_note_header(&v[1], p);
2072 	v[1].p_offset = doffset = roundup(doffset, sizeof (Word));
2073 	doffset += v[1].p_filesz;
2074 
2075 	mutex_enter(&p->p_lock);
2076 
2077 	brkbase = p->p_brkbase;
2078 	brksize = p->p_brksize;
2079 
2080 	stkbase = p->p_usrstack - p->p_stksize;
2081 	stksize = p->p_stksize;
2082 
2083 	mutex_exit(&p->p_lock);
2084 
2085 	AS_LOCK_ENTER(as, RW_WRITER);
2086 	i = 2;
2087 	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
2088 		caddr_t eaddr = seg->s_base + pr_getsegsize(seg, 0);
2089 		caddr_t saddr, naddr;
2090 		void *tmp = NULL;
2091 		extern struct seg_ops segspt_shmops;
2092 
2093 		if ((seg->s_flags & S_HOLE) != 0) {
2094 			continue;
2095 		}
2096 
2097 		for (saddr = seg->s_base; saddr < eaddr; saddr = naddr) {
2098 			uint_t prot;
2099 			size_t size;
2100 			int type;
2101 			vnode_t *mvp;
2102 
2103 			prot = pr_getprot(seg, 0, &tmp, &saddr, &naddr, eaddr);
2104 			prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
2105 			if ((size = (size_t)(naddr - saddr)) == 0)
2106 				continue;
2107 			if (i == nphdrs) {
2108 				overflow++;
2109 				continue;
2110 			}
2111 			v[i].p_type = PT_LOAD;
2112 			v[i].p_vaddr = (Addr)(uintptr_t)saddr;
2113 			v[i].p_memsz = size;
2114 			if (prot & PROT_READ)
2115 				v[i].p_flags |= PF_R;
2116 			if (prot & PROT_WRITE)
2117 				v[i].p_flags |= PF_W;
2118 			if (prot & PROT_EXEC)
2119 				v[i].p_flags |= PF_X;
2120 
2121 			/*
2122 			 * Figure out which mappings to include in the core.
2123 			 */
2124 			type = SEGOP_GETTYPE(seg, saddr);
2125 
2126 			if (saddr == stkbase && size == stksize) {
2127 				if (!(content & CC_CONTENT_STACK))
2128 					goto exclude;
2129 
2130 			} else if (saddr == brkbase && size == brksize) {
2131 				if (!(content & CC_CONTENT_HEAP))
2132 					goto exclude;
2133 
2134 			} else if (seg->s_ops == &segspt_shmops) {
2135 				if (type & MAP_NORESERVE) {
2136 					if (!(content & CC_CONTENT_DISM))
2137 						goto exclude;
2138 				} else {
2139 					if (!(content & CC_CONTENT_ISM))
2140 						goto exclude;
2141 				}
2142 
2143 			} else if (seg->s_ops != &segvn_ops) {
2144 				goto exclude;
2145 
2146 			} else if (type & MAP_SHARED) {
2147 				if (shmgetid(p, saddr) != SHMID_NONE) {
2148 					if (!(content & CC_CONTENT_SHM))
2149 						goto exclude;
2150 
2151 				} else if (SEGOP_GETVP(seg, seg->s_base,
2152 				    &mvp) != 0 || mvp == NULL ||
2153 				    mvp->v_type != VREG) {
2154 					if (!(content & CC_CONTENT_SHANON))
2155 						goto exclude;
2156 
2157 				} else {
2158 					if (!(content & CC_CONTENT_SHFILE))
2159 						goto exclude;
2160 				}
2161 
2162 			} else if (SEGOP_GETVP(seg, seg->s_base, &mvp) != 0 ||
2163 			    mvp == NULL || mvp->v_type != VREG) {
2164 				if (!(content & CC_CONTENT_ANON))
2165 					goto exclude;
2166 
2167 			} else if (prot == (PROT_READ | PROT_EXEC)) {
2168 				if (!(content & CC_CONTENT_TEXT))
2169 					goto exclude;
2170 
2171 			} else if (prot == PROT_READ) {
2172 				if (!(content & CC_CONTENT_RODATA))
2173 					goto exclude;
2174 
2175 			} else {
2176 				if (!(content & CC_CONTENT_DATA))
2177 					goto exclude;
2178 			}
2179 
2180 			doffset = roundup(doffset, sizeof (Word));
2181 			v[i].p_offset = doffset;
2182 			v[i].p_filesz = size;
2183 			doffset += size;
2184 exclude:
2185 			i++;
2186 		}
2187 		ASSERT(tmp == NULL);
2188 	}
2189 	AS_LOCK_EXIT(as);
2190 
2191 	if (overflow || i != nphdrs) {
2192 		if (ntries++ == 0) {
2193 			kmem_free(bigwad, bigsize);
2194 			overflow = 0;
2195 			goto top;
2196 		}
2197 		cmn_err(CE_WARN, "elfcore: core dump failed for "
2198 		    "process %d; address space is changing", p->p_pid);
2199 		error = EIO;
2200 		goto done;
2201 	}
2202 
2203 	if ((error = core_write(vp, UIO_SYSSPACE, poffset,
2204 	    v, phdrsz, rlimit, credp)) != 0)
2205 		goto done;
2206 
2207 	if ((error = write_old_elfnotes(p, sig, vp, v[0].p_offset, rlimit,
2208 	    credp)) != 0)
2209 		goto done;
2210 
2211 	if ((error = write_elfnotes(p, sig, vp, v[1].p_offset, rlimit,
2212 	    credp, content)) != 0)
2213 		goto done;
2214 
2215 	for (i = 2; i < nphdrs; i++) {
2216 		prkillinfo_t killinfo;
2217 		sigqueue_t *sq;
2218 		int sig, j;
2219 
2220 		if (v[i].p_filesz == 0)
2221 			continue;
2222 
2223 		/*
2224 		 * If we hit a region that was mapped PROT_NONE then we cannot
2225 		 * continue dumping this normally as the kernel would be unable
2226 		 * to read from the page and that would result in us failing to
2227 		 * dump the page. As such, any region mapped PROT_NONE, we dump
2228 		 * as a zero-filled page such that this is still represented in
2229 		 * the map.
2230 		 *
2231 		 * If dumping out this segment fails, rather than failing
2232 		 * the core dump entirely, we reset the size of the mapping
2233 		 * to zero to indicate that the data is absent from the core
2234 		 * file and or in the PF_SUNW_FAILURE flag to differentiate
2235 		 * this from mappings that were excluded due to the core file
2236 		 * content settings.
2237 		 */
2238 		if ((v[i].p_flags & (PF_R | PF_W | PF_X)) == 0) {
2239 			size_t towrite = v[i].p_filesz;
2240 			size_t curoff = 0;
2241 
2242 			if (zeropg == NULL) {
2243 				zeropg = kmem_zalloc(elf_zeropg_sz, KM_SLEEP);
2244 			}
2245 
2246 			error = 0;
2247 			while (towrite != 0) {
2248 				size_t len = MIN(towrite, elf_zeropg_sz);
2249 
2250 				error = core_write(vp, UIO_SYSSPACE,
2251 				    v[i].p_offset + curoff, zeropg, len, rlimit,
2252 				    credp);
2253 				if (error != 0)
2254 					break;
2255 
2256 				towrite -= len;
2257 				curoff += len;
2258 			}
2259 
2260 			if (error == 0)
2261 				continue;
2262 		} else {
2263 			error = core_seg(p, vp, v[i].p_offset,
2264 			    (caddr_t)(uintptr_t)v[i].p_vaddr, v[i].p_filesz,
2265 			    rlimit, credp);
2266 			if (error == 0)
2267 				continue;
2268 		}
2269 
2270 		if ((sig = lwp->lwp_cursig) == 0) {
2271 			/*
2272 			 * We failed due to something other than a signal.
2273 			 * Since the space reserved for the segment is now
2274 			 * unused, we stash the errno in the first four
2275 			 * bytes. This undocumented interface will let us
2276 			 * understand the nature of the failure.
2277 			 */
2278 			(void) core_write(vp, UIO_SYSSPACE, v[i].p_offset,
2279 			    &error, sizeof (error), rlimit, credp);
2280 
2281 			v[i].p_filesz = 0;
2282 			v[i].p_flags |= PF_SUNW_FAILURE;
2283 			if ((error = core_write(vp, UIO_SYSSPACE,
2284 			    poffset + sizeof (v[i]) * i, &v[i], sizeof (v[i]),
2285 			    rlimit, credp)) != 0)
2286 				goto done;
2287 
2288 			continue;
2289 		}
2290 
2291 		/*
2292 		 * We took a signal.  We want to abort the dump entirely, but
2293 		 * we also want to indicate what failed and why.  We therefore
2294 		 * use the space reserved for the first failing segment to
2295 		 * write our error (which, for purposes of compatability with
2296 		 * older core dump readers, we set to EINTR) followed by any
2297 		 * siginfo associated with the signal.
2298 		 */
2299 		bzero(&killinfo, sizeof (killinfo));
2300 		killinfo.prk_error = EINTR;
2301 
2302 		sq = sig == SIGKILL ? curproc->p_killsqp : lwp->lwp_curinfo;
2303 
2304 		if (sq != NULL) {
2305 			bcopy(&sq->sq_info, &killinfo.prk_info,
2306 			    sizeof (sq->sq_info));
2307 		} else {
2308 			killinfo.prk_info.si_signo = lwp->lwp_cursig;
2309 			killinfo.prk_info.si_code = SI_NOINFO;
2310 		}
2311 
2312 #if (defined(_SYSCALL32_IMPL) || defined(_LP64))
2313 		/*
2314 		 * If this is a 32-bit process, we need to translate from the
2315 		 * native siginfo to the 32-bit variant.  (Core readers must
2316 		 * always have the same data model as their target or must
2317 		 * be aware of -- and compensate for -- data model differences.)
2318 		 */
2319 		if (curproc->p_model == DATAMODEL_ILP32) {
2320 			siginfo32_t si32;
2321 
2322 			siginfo_kto32((k_siginfo_t *)&killinfo.prk_info, &si32);
2323 			bcopy(&si32, &killinfo.prk_info, sizeof (si32));
2324 		}
2325 #endif
2326 
2327 		(void) core_write(vp, UIO_SYSSPACE, v[i].p_offset,
2328 		    &killinfo, sizeof (killinfo), rlimit, credp);
2329 
2330 		/*
2331 		 * For the segment on which we took the signal, indicate that
2332 		 * its data now refers to a siginfo.
2333 		 */
2334 		v[i].p_filesz = 0;
2335 		v[i].p_flags |= PF_SUNW_FAILURE | PF_SUNW_KILLED |
2336 		    PF_SUNW_SIGINFO;
2337 
2338 		/*
2339 		 * And for every other segment, indicate that its absence
2340 		 * is due to a signal.
2341 		 */
2342 		for (j = i + 1; j < nphdrs; j++) {
2343 			v[j].p_filesz = 0;
2344 			v[j].p_flags |= PF_SUNW_FAILURE | PF_SUNW_KILLED;
2345 		}
2346 
2347 		/*
2348 		 * Finally, write out our modified program headers.
2349 		 */
2350 		if ((error = core_write(vp, UIO_SYSSPACE,
2351 		    poffset + sizeof (v[i]) * i, &v[i],
2352 		    sizeof (v[i]) * (nphdrs - i), rlimit, credp)) != 0)
2353 			goto done;
2354 
2355 		break;
2356 	}
2357 
2358 	if (nshdrs > 0) {
2359 		bzero(&bigwad->shdr[0], shdrsz);
2360 
2361 		if (nshdrs >= SHN_LORESERVE)
2362 			bigwad->shdr[0].sh_size = nshdrs;
2363 
2364 		if (nshdrs - 1 >= SHN_LORESERVE)
2365 			bigwad->shdr[0].sh_link = nshdrs - 1;
2366 
2367 		if (nphdrs >= PN_XNUM)
2368 			bigwad->shdr[0].sh_info = nphdrs;
2369 
2370 		if (nshdrs > 1) {
2371 			AS_LOCK_ENTER(as, RW_WRITER);
2372 			if ((error = process_scns(content, p, credp, vp,
2373 			    &bigwad->shdr[0], nshdrs, rlimit, &doffset,
2374 			    NULL)) != 0) {
2375 				AS_LOCK_EXIT(as);
2376 				goto done;
2377 			}
2378 			AS_LOCK_EXIT(as);
2379 		}
2380 
2381 		if ((error = core_write(vp, UIO_SYSSPACE, soffset,
2382 		    &bigwad->shdr[0], shdrsz, rlimit, credp)) != 0)
2383 			goto done;
2384 	}
2385 
2386 done:
2387 	if (zeropg != NULL) {
2388 		kmem_free(zeropg, elf_zeropg_sz);
2389 	}
2390 	kmem_free(bigwad, bigsize);
2391 	return (error);
2392 }
2393 
2394 #ifndef	_ELF32_COMPAT
2395 
2396 static struct execsw esw = {
2397 #ifdef	_LP64
2398 	elf64magicstr,
2399 #else	/* _LP64 */
2400 	elf32magicstr,
2401 #endif	/* _LP64 */
2402 	0,
2403 	5,
2404 	elfexec,
2405 	elfcore
2406 };
2407 
2408 static struct modlexec modlexec = {
2409 	&mod_execops, "exec module for elf", &esw
2410 };
2411 
2412 #ifdef	_LP64
2413 extern int elf32exec(vnode_t *vp, execa_t *uap, uarg_t *args,
2414 			intpdata_t *idatap, int level, long *execsz,
2415 			int setid, caddr_t exec_file, cred_t *cred,
2416 			int brand_action);
2417 extern int elf32core(vnode_t *vp, proc_t *p, cred_t *credp,
2418 			rlim64_t rlimit, int sig, core_content_t content);
2419 
2420 static struct execsw esw32 = {
2421 	elf32magicstr,
2422 	0,
2423 	5,
2424 	elf32exec,
2425 	elf32core
2426 };
2427 
2428 static struct modlexec modlexec32 = {
2429 	&mod_execops, "32-bit exec module for elf", &esw32
2430 };
2431 #endif	/* _LP64 */
2432 
2433 static struct modlinkage modlinkage = {
2434 	MODREV_1,
2435 	(void *)&modlexec,
2436 #ifdef	_LP64
2437 	(void *)&modlexec32,
2438 #endif	/* _LP64 */
2439 	NULL
2440 };
2441 
2442 int
2443 _init(void)
2444 {
2445 	return (mod_install(&modlinkage));
2446 }
2447 
2448 int
2449 _fini(void)
2450 {
2451 	return (mod_remove(&modlinkage));
2452 }
2453 
2454 int
2455 _info(struct modinfo *modinfop)
2456 {
2457 	return (mod_info(&modlinkage, modinfop));
2458 }
2459 
2460 #endif	/* !_ELF32_COMPAT */
2461