xref: /illumos-gate/usr/src/uts/common/exec/elf/elf.c (revision ea4c6b78cebe2a3687fa43deeedf6212a124d817)
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) 2013, Joyent, Inc.  All rights reserved.
30  */
31 
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/thread.h>
35 #include <sys/sysmacros.h>
36 #include <sys/signal.h>
37 #include <sys/cred.h>
38 #include <sys/user.h>
39 #include <sys/errno.h>
40 #include <sys/vnode.h>
41 #include <sys/mman.h>
42 #include <sys/kmem.h>
43 #include <sys/proc.h>
44 #include <sys/pathname.h>
45 #include <sys/cmn_err.h>
46 #include <sys/systm.h>
47 #include <sys/elf.h>
48 #include <sys/vmsystm.h>
49 #include <sys/debug.h>
50 #include <sys/auxv.h>
51 #include <sys/exec.h>
52 #include <sys/prsystm.h>
53 #include <vm/as.h>
54 #include <vm/rm.h>
55 #include <vm/seg.h>
56 #include <vm/seg_vn.h>
57 #include <sys/modctl.h>
58 #include <sys/systeminfo.h>
59 #include <sys/vmparam.h>
60 #include <sys/machelf.h>
61 #include <sys/shm_impl.h>
62 #include <sys/archsystm.h>
63 #include <sys/fasttrap.h>
64 #include <sys/brand.h>
65 #include "elf_impl.h"
66 #include <sys/sdt.h>
67 #include <sys/siginfo.h>
68 
69 extern int at_flags;
70 
71 #define	ORIGIN_STR	"ORIGIN"
72 #define	ORIGIN_STR_SIZE	6
73 
74 static int getelfhead(vnode_t *, cred_t *, Ehdr *, int *, int *, int *);
75 static int getelfphdr(vnode_t *, cred_t *, const Ehdr *, int, caddr_t *,
76     ssize_t *);
77 static int getelfshdr(vnode_t *, cred_t *, const Ehdr *, int, int, caddr_t *,
78     ssize_t *, caddr_t *, ssize_t *);
79 static size_t elfsize(Ehdr *, int, caddr_t, uintptr_t *);
80 static int mapelfexec(vnode_t *, Ehdr *, int, caddr_t,
81     Phdr **, Phdr **, Phdr **, Phdr **, Phdr *,
82     caddr_t *, caddr_t *, intptr_t *, intptr_t *, size_t, long *, size_t *);
83 
84 typedef enum {
85 	STR_CTF,
86 	STR_SYMTAB,
87 	STR_DYNSYM,
88 	STR_STRTAB,
89 	STR_DYNSTR,
90 	STR_SHSTRTAB,
91 	STR_NUM
92 } shstrtype_t;
93 
94 static const char *shstrtab_data[] = {
95 	".SUNW_ctf",
96 	".symtab",
97 	".dynsym",
98 	".strtab",
99 	".dynstr",
100 	".shstrtab"
101 };
102 
103 typedef struct shstrtab {
104 	int	sst_ndx[STR_NUM];
105 	int	sst_cur;
106 } shstrtab_t;
107 
108 static void
109 shstrtab_init(shstrtab_t *s)
110 {
111 	bzero(&s->sst_ndx, sizeof (s->sst_ndx));
112 	s->sst_cur = 1;
113 }
114 
115 static int
116 shstrtab_ndx(shstrtab_t *s, shstrtype_t type)
117 {
118 	int ret;
119 
120 	if ((ret = s->sst_ndx[type]) != 0)
121 		return (ret);
122 
123 	ret = s->sst_ndx[type] = s->sst_cur;
124 	s->sst_cur += strlen(shstrtab_data[type]) + 1;
125 
126 	return (ret);
127 }
128 
129 static size_t
130 shstrtab_size(const shstrtab_t *s)
131 {
132 	return (s->sst_cur);
133 }
134 
135 static void
136 shstrtab_dump(const shstrtab_t *s, char *buf)
137 {
138 	int i, ndx;
139 
140 	*buf = '\0';
141 	for (i = 0; i < STR_NUM; i++) {
142 		if ((ndx = s->sst_ndx[i]) != 0)
143 			(void) strcpy(buf + ndx, shstrtab_data[i]);
144 	}
145 }
146 
147 static int
148 dtrace_safe_phdr(Phdr *phdrp, struct uarg *args, uintptr_t base)
149 {
150 	ASSERT(phdrp->p_type == PT_SUNWDTRACE);
151 
152 	/*
153 	 * See the comment in fasttrap.h for information on how to safely
154 	 * update this program header.
155 	 */
156 	if (phdrp->p_memsz < PT_SUNWDTRACE_SIZE ||
157 	    (phdrp->p_flags & (PF_R | PF_W | PF_X)) != (PF_R | PF_W | PF_X))
158 		return (-1);
159 
160 	args->thrptr = phdrp->p_vaddr + base;
161 
162 	return (0);
163 }
164 
165 /*
166  * Map in the executable pointed to by vp. Returns 0 on success.
167  */
168 int
169 mapexec_brand(vnode_t *vp, uarg_t *args, Ehdr *ehdr, Addr *uphdr_vaddr,
170     intptr_t *voffset, caddr_t exec_file, int *interp, caddr_t *bssbase,
171     caddr_t *brkbase, size_t *brksize, uintptr_t *lddatap)
172 {
173 	size_t		len;
174 	struct vattr	vat;
175 	caddr_t		phdrbase = NULL;
176 	ssize_t		phdrsize;
177 	int		nshdrs, shstrndx, nphdrs;
178 	int		error = 0;
179 	Phdr		*uphdr = NULL;
180 	Phdr		*junk = NULL;
181 	Phdr		*dynphdr = NULL;
182 	Phdr		*dtrphdr = NULL;
183 	uintptr_t	lddata;
184 	long		execsz;
185 	intptr_t	minaddr;
186 
187 	if (lddatap != NULL)
188 		*lddatap = NULL;
189 
190 	if (error = execpermissions(vp, &vat, args)) {
191 		uprintf("%s: Cannot execute %s\n", exec_file, args->pathname);
192 		return (error);
193 	}
194 
195 	if ((error = getelfhead(vp, CRED(), ehdr, &nshdrs, &shstrndx,
196 	    &nphdrs)) != 0 ||
197 	    (error = getelfphdr(vp, CRED(), ehdr, nphdrs, &phdrbase,
198 	    &phdrsize)) != 0) {
199 		uprintf("%s: Cannot read %s\n", exec_file, args->pathname);
200 		return (error);
201 	}
202 
203 	if ((len = elfsize(ehdr, nphdrs, phdrbase, &lddata)) == 0) {
204 		uprintf("%s: Nothing to load in %s", exec_file, args->pathname);
205 		kmem_free(phdrbase, phdrsize);
206 		return (ENOEXEC);
207 	}
208 	if (lddatap != NULL)
209 		*lddatap = lddata;
210 
211 	if (error = mapelfexec(vp, ehdr, nphdrs, phdrbase, &uphdr, &dynphdr,
212 	    &junk, &dtrphdr, NULL, bssbase, brkbase, voffset, &minaddr,
213 	    len, &execsz, brksize)) {
214 		uprintf("%s: Cannot map %s\n", exec_file, args->pathname);
215 		kmem_free(phdrbase, phdrsize);
216 		return (error);
217 	}
218 
219 	/*
220 	 * Inform our caller if the executable needs an interpreter.
221 	 */
222 	*interp = (dynphdr == NULL) ? 0 : 1;
223 
224 	/*
225 	 * If this is a statically linked executable, voffset should indicate
226 	 * the address of the executable itself (it normally holds the address
227 	 * of the interpreter).
228 	 */
229 	if (ehdr->e_type == ET_EXEC && *interp == 0)
230 		*voffset = minaddr;
231 
232 	if (uphdr != NULL) {
233 		*uphdr_vaddr = uphdr->p_vaddr;
234 	} else {
235 		*uphdr_vaddr = (Addr)-1;
236 	}
237 
238 	kmem_free(phdrbase, phdrsize);
239 	return (error);
240 }
241 
242 /*ARGSUSED*/
243 int
244 elfexec(vnode_t *vp, execa_t *uap, uarg_t *args, intpdata_t *idatap,
245     int level, long *execsz, int setid, caddr_t exec_file, cred_t *cred,
246     int brand_action)
247 {
248 	caddr_t		phdrbase = NULL;
249 	caddr_t 	bssbase = 0;
250 	caddr_t 	brkbase = 0;
251 	size_t		brksize = 0;
252 	ssize_t		dlnsize;
253 	aux_entry_t	*aux;
254 	int		error;
255 	ssize_t		resid;
256 	int		fd = -1;
257 	intptr_t	voffset;
258 	Phdr		*dyphdr = NULL;
259 	Phdr		*stphdr = NULL;
260 	Phdr		*uphdr = NULL;
261 	Phdr		*junk = NULL;
262 	size_t		len;
263 	ssize_t		phdrsize;
264 	int		postfixsize = 0;
265 	int		i, hsize;
266 	Phdr		*phdrp;
267 	Phdr		*dataphdrp = NULL;
268 	Phdr		*dtrphdr;
269 	Phdr		*capphdr = NULL;
270 	Cap		*cap = NULL;
271 	ssize_t		capsize;
272 	int		hasu = 0;
273 	int		hasauxv = 0;
274 	int		hasdy = 0;
275 	int		branded = 0;
276 
277 	struct proc *p = ttoproc(curthread);
278 	struct user *up = PTOU(p);
279 	struct bigwad {
280 		Ehdr	ehdr;
281 		aux_entry_t	elfargs[__KERN_NAUXV_IMPL];
282 		char		dl_name[MAXPATHLEN];
283 		char		pathbuf[MAXPATHLEN];
284 		struct vattr	vattr;
285 		struct execenv	exenv;
286 	} *bigwad;	/* kmem_alloc this behemoth so we don't blow stack */
287 	Ehdr		*ehdrp;
288 	int		nshdrs, shstrndx, nphdrs;
289 	char		*dlnp;
290 	char		*pathbufp;
291 	rlim64_t	limit;
292 	rlim64_t	roundlimit;
293 
294 	ASSERT(p->p_model == DATAMODEL_ILP32 || p->p_model == DATAMODEL_LP64);
295 
296 	bigwad = kmem_alloc(sizeof (struct bigwad), KM_SLEEP);
297 	ehdrp = &bigwad->ehdr;
298 	dlnp = bigwad->dl_name;
299 	pathbufp = bigwad->pathbuf;
300 
301 	/*
302 	 * Obtain ELF and program header information.
303 	 */
304 	if ((error = getelfhead(vp, CRED(), ehdrp, &nshdrs, &shstrndx,
305 	    &nphdrs)) != 0 ||
306 	    (error = getelfphdr(vp, CRED(), ehdrp, nphdrs, &phdrbase,
307 	    &phdrsize)) != 0)
308 		goto out;
309 
310 	/*
311 	 * Prevent executing an ELF file that has no entry point.
312 	 */
313 	if (ehdrp->e_entry == 0) {
314 		uprintf("%s: Bad entry point\n", exec_file);
315 		goto bad;
316 	}
317 
318 	/*
319 	 * Put data model that we're exec-ing to into the args passed to
320 	 * exec_args(), so it will know what it is copying to on new stack.
321 	 * Now that we know whether we are exec-ing a 32-bit or 64-bit
322 	 * executable, we can set execsz with the appropriate NCARGS.
323 	 */
324 #ifdef	_LP64
325 	if (ehdrp->e_ident[EI_CLASS] == ELFCLASS32) {
326 		args->to_model = DATAMODEL_ILP32;
327 		*execsz = btopr(SINCR) + btopr(SSIZE) + btopr(NCARGS32-1);
328 	} else {
329 		args->to_model = DATAMODEL_LP64;
330 		args->stk_prot &= ~PROT_EXEC;
331 #if defined(__i386) || defined(__amd64)
332 		args->dat_prot &= ~PROT_EXEC;
333 #endif
334 		*execsz = btopr(SINCR) + btopr(SSIZE) + btopr(NCARGS64-1);
335 	}
336 #else	/* _LP64 */
337 	args->to_model = DATAMODEL_ILP32;
338 	*execsz = btopr(SINCR) + btopr(SSIZE) + btopr(NCARGS-1);
339 #endif	/* _LP64 */
340 
341 	/*
342 	 * We delay invoking the brand callback until we've figured out
343 	 * what kind of elf binary we're trying to run, 32-bit or 64-bit.
344 	 * We do this because now the brand library can just check
345 	 * args->to_model to see if the target is 32-bit or 64-bit without
346 	 * having do duplicate all the code above.
347 	 *
348 	 * The level checks associated with brand handling below are used to
349 	 * prevent a loop since the brand elfexec function typically comes back
350 	 * through this function. We must check <= here since the nested
351 	 * handling in the #! interpreter code will increment the level before
352 	 * calling gexec to run the final elfexec interpreter.
353 	 */
354 	if ((level <= INTP_MAXDEPTH) &&
355 	    (brand_action != EBA_NATIVE) && (PROC_IS_BRANDED(p))) {
356 		error = BROP(p)->b_elfexec(vp, uap, args,
357 		    idatap, level + 1, execsz, setid, exec_file, cred,
358 		    brand_action);
359 		goto out;
360 	}
361 
362 	/*
363 	 * Determine aux size now so that stack can be built
364 	 * in one shot (except actual copyout of aux image),
365 	 * determine any non-default stack protections,
366 	 * and still have this code be machine independent.
367 	 */
368 	hsize = ehdrp->e_phentsize;
369 	phdrp = (Phdr *)phdrbase;
370 	for (i = nphdrs; i > 0; i--) {
371 		switch (phdrp->p_type) {
372 		case PT_INTERP:
373 			hasauxv = hasdy = 1;
374 			break;
375 		case PT_PHDR:
376 			hasu = 1;
377 			break;
378 		case PT_SUNWSTACK:
379 			args->stk_prot = PROT_USER;
380 			if (phdrp->p_flags & PF_R)
381 				args->stk_prot |= PROT_READ;
382 			if (phdrp->p_flags & PF_W)
383 				args->stk_prot |= PROT_WRITE;
384 			if (phdrp->p_flags & PF_X)
385 				args->stk_prot |= PROT_EXEC;
386 			break;
387 		case PT_LOAD:
388 			dataphdrp = phdrp;
389 			break;
390 		case PT_SUNWCAP:
391 			capphdr = phdrp;
392 			break;
393 		}
394 		phdrp = (Phdr *)((caddr_t)phdrp + hsize);
395 	}
396 
397 	if (ehdrp->e_type != ET_EXEC) {
398 		dataphdrp = NULL;
399 		hasauxv = 1;
400 	}
401 
402 	/* Copy BSS permissions to args->dat_prot */
403 	if (dataphdrp != NULL) {
404 		args->dat_prot = PROT_USER;
405 		if (dataphdrp->p_flags & PF_R)
406 			args->dat_prot |= PROT_READ;
407 		if (dataphdrp->p_flags & PF_W)
408 			args->dat_prot |= PROT_WRITE;
409 		if (dataphdrp->p_flags & PF_X)
410 			args->dat_prot |= PROT_EXEC;
411 	}
412 
413 	/*
414 	 * If a auxvector will be required - reserve the space for
415 	 * it now.  This may be increased by exec_args if there are
416 	 * ISA-specific types (included in __KERN_NAUXV_IMPL).
417 	 */
418 	if (hasauxv) {
419 		/*
420 		 * If a AUX vector is being built - the base AUX
421 		 * entries are:
422 		 *
423 		 *	AT_BASE
424 		 *	AT_FLAGS
425 		 *	AT_PAGESZ
426 		 *	AT_SUN_AUXFLAGS
427 		 *	AT_SUN_HWCAP
428 		 *	AT_SUN_HWCAP2
429 		 *	AT_SUN_PLATFORM (added in stk_copyout)
430 		 *	AT_SUN_EXECNAME (added in stk_copyout)
431 		 *	AT_NULL
432 		 *
433 		 * total == 9
434 		 */
435 		if (hasdy && hasu) {
436 			/*
437 			 * Has PT_INTERP & PT_PHDR - the auxvectors that
438 			 * will be built are:
439 			 *
440 			 *	AT_PHDR
441 			 *	AT_PHENT
442 			 *	AT_PHNUM
443 			 *	AT_ENTRY
444 			 *	AT_LDDATA
445 			 *
446 			 * total = 5
447 			 */
448 			args->auxsize = (9 + 5) * sizeof (aux_entry_t);
449 		} else if (hasdy) {
450 			/*
451 			 * Has PT_INTERP but no PT_PHDR
452 			 *
453 			 *	AT_EXECFD
454 			 *	AT_LDDATA
455 			 *
456 			 * total = 2
457 			 */
458 			args->auxsize = (9 + 2) * sizeof (aux_entry_t);
459 		} else {
460 			args->auxsize = 9 * sizeof (aux_entry_t);
461 		}
462 	} else {
463 		args->auxsize = 0;
464 	}
465 
466 	/*
467 	 * If this binary is using an emulator, we need to add an
468 	 * AT_SUN_EMULATOR aux entry.
469 	 */
470 	if (args->emulator != NULL)
471 		args->auxsize += sizeof (aux_entry_t);
472 
473 	if ((brand_action != EBA_NATIVE) && (PROC_IS_BRANDED(p))) {
474 		branded = 1;
475 		/*
476 		 * We will be adding 4 entries to the aux vectors.  One for
477 		 * the the brandname and 3 for the brand specific aux vectors.
478 		 */
479 		args->auxsize += 4 * sizeof (aux_entry_t);
480 	}
481 
482 	/* Hardware/Software capabilities */
483 	if (capphdr != NULL &&
484 	    (capsize = capphdr->p_filesz) > 0 &&
485 	    capsize <= 16 * sizeof (*cap)) {
486 		int ncaps = capsize / sizeof (*cap);
487 		Cap *cp;
488 
489 		cap = kmem_alloc(capsize, KM_SLEEP);
490 		if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)cap,
491 		    capsize, (offset_t)capphdr->p_offset,
492 		    UIO_SYSSPACE, 0, (rlim64_t)0, CRED(), &resid)) != 0) {
493 			uprintf("%s: Cannot read capabilities section\n",
494 			    exec_file);
495 			goto out;
496 		}
497 		for (cp = cap; cp < cap + ncaps; cp++) {
498 			if (cp->c_tag == CA_SUNW_SF_1 &&
499 			    (cp->c_un.c_val & SF1_SUNW_ADDR32)) {
500 				if (args->to_model == DATAMODEL_LP64)
501 					args->addr32 = 1;
502 				break;
503 			}
504 		}
505 	}
506 
507 	aux = bigwad->elfargs;
508 	/*
509 	 * Move args to the user's stack.
510 	 * This can fill in the AT_SUN_PLATFORM and AT_SUN_EXECNAME aux entries.
511 	 */
512 	if ((error = exec_args(uap, args, idatap, (void **)&aux)) != 0) {
513 		if (error == -1) {
514 			error = ENOEXEC;
515 			goto bad;
516 		}
517 		goto out;
518 	}
519 	/* we're single threaded after this point */
520 
521 	/*
522 	 * If this is an ET_DYN executable (shared object),
523 	 * determine its memory size so that mapelfexec() can load it.
524 	 */
525 	if (ehdrp->e_type == ET_DYN)
526 		len = elfsize(ehdrp, nphdrs, phdrbase, NULL);
527 	else
528 		len = 0;
529 
530 	dtrphdr = NULL;
531 
532 	if ((error = mapelfexec(vp, ehdrp, nphdrs, phdrbase, &uphdr, &dyphdr,
533 	    &stphdr, &dtrphdr, dataphdrp, &bssbase, &brkbase, &voffset, NULL,
534 	    len, execsz, &brksize)) != 0)
535 		goto bad;
536 
537 	if (uphdr != NULL && dyphdr == NULL)
538 		goto bad;
539 
540 	if (dtrphdr != NULL && dtrace_safe_phdr(dtrphdr, args, voffset) != 0) {
541 		uprintf("%s: Bad DTrace phdr in %s\n", exec_file, exec_file);
542 		goto bad;
543 	}
544 
545 	if (dyphdr != NULL) {
546 		size_t		len;
547 		uintptr_t	lddata;
548 		char		*p;
549 		struct vnode	*nvp;
550 
551 		dlnsize = dyphdr->p_filesz;
552 
553 		if (dlnsize > MAXPATHLEN || dlnsize <= 0)
554 			goto bad;
555 
556 		/*
557 		 * Read in "interpreter" pathname.
558 		 */
559 		if ((error = vn_rdwr(UIO_READ, vp, dlnp, dyphdr->p_filesz,
560 		    (offset_t)dyphdr->p_offset, UIO_SYSSPACE, 0, (rlim64_t)0,
561 		    CRED(), &resid)) != 0) {
562 			uprintf("%s: Cannot obtain interpreter pathname\n",
563 			    exec_file);
564 			goto bad;
565 		}
566 
567 		if (resid != 0 || dlnp[dlnsize - 1] != '\0')
568 			goto bad;
569 
570 		/*
571 		 * Search for '$ORIGIN' token in interpreter path.
572 		 * If found, expand it.
573 		 */
574 		for (p = dlnp; p = strchr(p, '$'); ) {
575 			uint_t	len, curlen;
576 			char	*_ptr;
577 
578 			if (strncmp(++p, ORIGIN_STR, ORIGIN_STR_SIZE))
579 				continue;
580 
581 			/*
582 			 * We don't support $ORIGIN on setid programs to close
583 			 * a potential attack vector.
584 			 */
585 			if ((setid & EXECSETID_SETID) != 0) {
586 				error = ENOEXEC;
587 				goto bad;
588 			}
589 
590 			curlen = 0;
591 			len = p - dlnp - 1;
592 			if (len) {
593 				bcopy(dlnp, pathbufp, len);
594 				curlen += len;
595 			}
596 			if (_ptr = strrchr(args->pathname, '/')) {
597 				len = _ptr - args->pathname;
598 				if ((curlen + len) > MAXPATHLEN)
599 					break;
600 
601 				bcopy(args->pathname, &pathbufp[curlen], len);
602 				curlen += len;
603 			} else {
604 				/*
605 				 * executable is a basename found in the
606 				 * current directory.  So - just substitue
607 				 * '.' for ORIGIN.
608 				 */
609 				pathbufp[curlen] = '.';
610 				curlen++;
611 			}
612 			p += ORIGIN_STR_SIZE;
613 			len = strlen(p);
614 
615 			if ((curlen + len) > MAXPATHLEN)
616 				break;
617 			bcopy(p, &pathbufp[curlen], len);
618 			curlen += len;
619 			pathbufp[curlen++] = '\0';
620 			bcopy(pathbufp, dlnp, curlen);
621 		}
622 
623 		/*
624 		 * /usr/lib/ld.so.1 is known to be a symlink to /lib/ld.so.1
625 		 * (and /usr/lib/64/ld.so.1 is a symlink to /lib/64/ld.so.1).
626 		 * Just in case /usr is not mounted, change it now.
627 		 */
628 		if (strcmp(dlnp, USR_LIB_RTLD) == 0)
629 			dlnp += 4;
630 		error = lookupname(dlnp, UIO_SYSSPACE, FOLLOW, NULLVPP, &nvp);
631 		if (error && dlnp != bigwad->dl_name) {
632 			/* new kernel, old user-level */
633 			error = lookupname(dlnp -= 4, UIO_SYSSPACE, FOLLOW,
634 			    NULLVPP, &nvp);
635 		}
636 		if (error) {
637 			uprintf("%s: Cannot find %s\n", exec_file, dlnp);
638 			goto bad;
639 		}
640 
641 		/*
642 		 * Setup the "aux" vector.
643 		 */
644 		if (uphdr) {
645 			if (ehdrp->e_type == ET_DYN) {
646 				/* don't use the first page */
647 				bigwad->exenv.ex_brkbase = (caddr_t)PAGESIZE;
648 				bigwad->exenv.ex_bssbase = (caddr_t)PAGESIZE;
649 			} else {
650 				bigwad->exenv.ex_bssbase = bssbase;
651 				bigwad->exenv.ex_brkbase = brkbase;
652 			}
653 			bigwad->exenv.ex_brksize = brksize;
654 			bigwad->exenv.ex_magic = elfmagic;
655 			bigwad->exenv.ex_vp = vp;
656 			setexecenv(&bigwad->exenv);
657 
658 			ADDAUX(aux, AT_PHDR, uphdr->p_vaddr + voffset)
659 			ADDAUX(aux, AT_PHENT, ehdrp->e_phentsize)
660 			ADDAUX(aux, AT_PHNUM, nphdrs)
661 			ADDAUX(aux, AT_ENTRY, ehdrp->e_entry + voffset)
662 		} else {
663 			if ((error = execopen(&vp, &fd)) != 0) {
664 				VN_RELE(nvp);
665 				goto bad;
666 			}
667 
668 			ADDAUX(aux, AT_EXECFD, fd)
669 		}
670 
671 		if ((error = execpermissions(nvp, &bigwad->vattr, args)) != 0) {
672 			VN_RELE(nvp);
673 			uprintf("%s: Cannot execute %s\n", exec_file, dlnp);
674 			goto bad;
675 		}
676 
677 		/*
678 		 * Now obtain the ELF header along with the entire program
679 		 * header contained in "nvp".
680 		 */
681 		kmem_free(phdrbase, phdrsize);
682 		phdrbase = NULL;
683 		if ((error = getelfhead(nvp, CRED(), ehdrp, &nshdrs,
684 		    &shstrndx, &nphdrs)) != 0 ||
685 		    (error = getelfphdr(nvp, CRED(), ehdrp, nphdrs, &phdrbase,
686 		    &phdrsize)) != 0) {
687 			VN_RELE(nvp);
688 			uprintf("%s: Cannot read %s\n", exec_file, dlnp);
689 			goto bad;
690 		}
691 
692 		/*
693 		 * Determine memory size of the "interpreter's" loadable
694 		 * sections.  This size is then used to obtain the virtual
695 		 * address of a hole, in the user's address space, large
696 		 * enough to map the "interpreter".
697 		 */
698 		if ((len = elfsize(ehdrp, nphdrs, phdrbase, &lddata)) == 0) {
699 			VN_RELE(nvp);
700 			uprintf("%s: Nothing to load in %s\n", exec_file, dlnp);
701 			goto bad;
702 		}
703 
704 		dtrphdr = NULL;
705 
706 		error = mapelfexec(nvp, ehdrp, nphdrs, phdrbase, &junk, &junk,
707 		    &junk, &dtrphdr, NULL, NULL, NULL, &voffset, NULL, len,
708 		    execsz, NULL);
709 		if (error || junk != NULL) {
710 			VN_RELE(nvp);
711 			uprintf("%s: Cannot map %s\n", exec_file, dlnp);
712 			goto bad;
713 		}
714 
715 		/*
716 		 * We use the DTrace program header to initialize the
717 		 * architecture-specific user per-LWP location. The dtrace
718 		 * fasttrap provider requires ready access to per-LWP scratch
719 		 * space. We assume that there is only one such program header
720 		 * in the interpreter.
721 		 */
722 		if (dtrphdr != NULL &&
723 		    dtrace_safe_phdr(dtrphdr, args, voffset) != 0) {
724 			VN_RELE(nvp);
725 			uprintf("%s: Bad DTrace phdr in %s\n", exec_file, dlnp);
726 			goto bad;
727 		}
728 
729 		VN_RELE(nvp);
730 		ADDAUX(aux, AT_SUN_LDDATA, voffset + lddata)
731 	}
732 
733 	if (hasauxv) {
734 		int auxf = AF_SUN_HWCAPVERIFY;
735 		/*
736 		 * Note: AT_SUN_PLATFORM and AT_SUN_EXECNAME were filled in via
737 		 * exec_args()
738 		 */
739 		ADDAUX(aux, AT_BASE, voffset)
740 		ADDAUX(aux, AT_FLAGS, at_flags)
741 		ADDAUX(aux, AT_PAGESZ, PAGESIZE)
742 		/*
743 		 * Linker flags. (security)
744 		 * p_flag not yet set at this time.
745 		 * We rely on gexec() to provide us with the information.
746 		 * If the application is set-uid but this is not reflected
747 		 * in a mismatch between real/effective uids/gids, then
748 		 * don't treat this as a set-uid exec.  So we care about
749 		 * the EXECSETID_UGIDS flag but not the ...SETID flag.
750 		 */
751 		if ((setid &= ~EXECSETID_SETID) != 0)
752 			auxf |= AF_SUN_SETUGID;
753 
754 		/*
755 		 * If we're running a native process from within a branded
756 		 * zone under pfexec then we clear the AF_SUN_SETUGID flag so
757 		 * that the native ld.so.1 is able to link with the native
758 		 * libraries instead of using the brand libraries that are
759 		 * installed in the zone.  We only do this for processes
760 		 * which we trust because we see they are already running
761 		 * under pfexec (where uid != euid).  This prevents a
762 		 * malicious user within the zone from crafting a wrapper to
763 		 * run native suid commands with unsecure libraries interposed.
764 		 */
765 		if ((brand_action == EBA_NATIVE) && (PROC_IS_BRANDED(p) &&
766 		    (setid &= ~EXECSETID_SETID) != 0))
767 			auxf &= ~AF_SUN_SETUGID;
768 
769 		/*
770 		 * Record the user addr of the auxflags aux vector entry
771 		 * since brands may optionally want to manipulate this field.
772 		 */
773 		args->auxp_auxflags =
774 		    (char *)((char *)args->stackend +
775 		    ((char *)&aux->a_type -
776 		    (char *)bigwad->elfargs));
777 		ADDAUX(aux, AT_SUN_AUXFLAGS, auxf);
778 		/*
779 		 * Hardware capability flag word (performance hints)
780 		 * Used for choosing faster library routines.
781 		 * (Potentially different between 32-bit and 64-bit ABIs)
782 		 */
783 #if defined(_LP64)
784 		if (args->to_model == DATAMODEL_NATIVE) {
785 			ADDAUX(aux, AT_SUN_HWCAP, auxv_hwcap)
786 			ADDAUX(aux, AT_SUN_HWCAP2, auxv_hwcap_2)
787 		} else {
788 			ADDAUX(aux, AT_SUN_HWCAP, auxv_hwcap32)
789 			ADDAUX(aux, AT_SUN_HWCAP2, auxv_hwcap32_2)
790 		}
791 #else
792 		ADDAUX(aux, AT_SUN_HWCAP, auxv_hwcap)
793 		ADDAUX(aux, AT_SUN_HWCAP2, auxv_hwcap_2)
794 #endif
795 		if (branded) {
796 			/*
797 			 * Reserve space for the brand-private aux vectors,
798 			 * and record the user addr of that space.
799 			 */
800 			args->auxp_brand =
801 			    (char *)((char *)args->stackend +
802 			    ((char *)&aux->a_type -
803 			    (char *)bigwad->elfargs));
804 			ADDAUX(aux, AT_SUN_BRAND_AUX1, 0)
805 			ADDAUX(aux, AT_SUN_BRAND_AUX2, 0)
806 			ADDAUX(aux, AT_SUN_BRAND_AUX3, 0)
807 		}
808 
809 		ADDAUX(aux, AT_NULL, 0)
810 		postfixsize = (char *)aux - (char *)bigwad->elfargs;
811 
812 		/*
813 		 * We make assumptions above when we determine how many aux
814 		 * vector entries we will be adding. However, if we have an
815 		 * invalid elf file, it is possible that mapelfexec might
816 		 * behave differently (but not return an error), in which case
817 		 * the number of aux entries we actually add will be different.
818 		 * We detect that now and error out.
819 		 */
820 		if (postfixsize != args->auxsize) {
821 			DTRACE_PROBE2(elfexec_badaux, int, postfixsize,
822 			    int, args->auxsize);
823 			goto bad;
824 		}
825 		ASSERT(postfixsize <= __KERN_NAUXV_IMPL * sizeof (aux_entry_t));
826 	}
827 
828 	/*
829 	 * For the 64-bit kernel, the limit is big enough that rounding it up
830 	 * to a page can overflow the 64-bit limit, so we check for btopr()
831 	 * overflowing here by comparing it with the unrounded limit in pages.
832 	 * If it hasn't overflowed, compare the exec size with the rounded up
833 	 * limit in pages.  Otherwise, just compare with the unrounded limit.
834 	 */
835 	limit = btop(p->p_vmem_ctl);
836 	roundlimit = btopr(p->p_vmem_ctl);
837 	if ((roundlimit > limit && *execsz > roundlimit) ||
838 	    (roundlimit < limit && *execsz > limit)) {
839 		mutex_enter(&p->p_lock);
840 		(void) rctl_action(rctlproc_legacy[RLIMIT_VMEM], p->p_rctls, p,
841 		    RCA_SAFE);
842 		mutex_exit(&p->p_lock);
843 		error = ENOMEM;
844 		goto bad;
845 	}
846 
847 	bzero(up->u_auxv, sizeof (up->u_auxv));
848 	if (postfixsize) {
849 		int num_auxv;
850 
851 		/*
852 		 * Copy the aux vector to the user stack.
853 		 */
854 		error = execpoststack(args, bigwad->elfargs, postfixsize);
855 		if (error)
856 			goto bad;
857 
858 		/*
859 		 * Copy auxv to the process's user structure for use by /proc.
860 		 * If this is a branded process, the brand's exec routine will
861 		 * copy it's private entries to the user structure later. It
862 		 * relies on the fact that the blank entries are at the end.
863 		 */
864 		num_auxv = postfixsize / sizeof (aux_entry_t);
865 		ASSERT(num_auxv <= sizeof (up->u_auxv) / sizeof (auxv_t));
866 		aux = bigwad->elfargs;
867 		for (i = 0; i < num_auxv; i++) {
868 			up->u_auxv[i].a_type = aux[i].a_type;
869 			up->u_auxv[i].a_un.a_val = (aux_val_t)aux[i].a_un.a_val;
870 		}
871 	}
872 
873 	/*
874 	 * Pass back the starting address so we can set the program counter.
875 	 */
876 	args->entry = (uintptr_t)(ehdrp->e_entry + voffset);
877 
878 	if (!uphdr) {
879 		if (ehdrp->e_type == ET_DYN) {
880 			/*
881 			 * If we are executing a shared library which doesn't
882 			 * have a interpreter (probably ld.so.1) then
883 			 * we don't set the brkbase now.  Instead we
884 			 * delay it's setting until the first call
885 			 * via grow.c::brk().  This permits ld.so.1 to
886 			 * initialize brkbase to the tail of the executable it
887 			 * loads (which is where it needs to be).
888 			 */
889 			bigwad->exenv.ex_brkbase = (caddr_t)0;
890 			bigwad->exenv.ex_bssbase = (caddr_t)0;
891 			bigwad->exenv.ex_brksize = 0;
892 		} else {
893 			bigwad->exenv.ex_brkbase = brkbase;
894 			bigwad->exenv.ex_bssbase = bssbase;
895 			bigwad->exenv.ex_brksize = brksize;
896 		}
897 		bigwad->exenv.ex_magic = elfmagic;
898 		bigwad->exenv.ex_vp = vp;
899 		setexecenv(&bigwad->exenv);
900 	}
901 
902 	ASSERT(error == 0);
903 	goto out;
904 
905 bad:
906 	if (fd != -1)		/* did we open the a.out yet */
907 		(void) execclose(fd);
908 
909 	psignal(p, SIGKILL);
910 
911 	if (error == 0)
912 		error = ENOEXEC;
913 out:
914 	if (phdrbase != NULL)
915 		kmem_free(phdrbase, phdrsize);
916 	if (cap != NULL)
917 		kmem_free(cap, capsize);
918 	kmem_free(bigwad, sizeof (struct bigwad));
919 	return (error);
920 }
921 
922 /*
923  * Compute the memory size requirement for the ELF file.
924  */
925 static size_t
926 elfsize(Ehdr *ehdrp, int nphdrs, caddr_t phdrbase, uintptr_t *lddata)
927 {
928 	size_t	len;
929 	Phdr	*phdrp = (Phdr *)phdrbase;
930 	int	hsize = ehdrp->e_phentsize;
931 	int	first = 1;
932 	int	dfirst = 1;	/* first data segment */
933 	uintptr_t loaddr = 0;
934 	uintptr_t hiaddr = 0;
935 	uintptr_t lo, hi;
936 	int	i;
937 
938 	for (i = nphdrs; i > 0; i--) {
939 		if (phdrp->p_type == PT_LOAD) {
940 			lo = phdrp->p_vaddr;
941 			hi = lo + phdrp->p_memsz;
942 			if (first) {
943 				loaddr = lo;
944 				hiaddr = hi;
945 				first = 0;
946 			} else {
947 				if (loaddr > lo)
948 					loaddr = lo;
949 				if (hiaddr < hi)
950 					hiaddr = hi;
951 			}
952 
953 			/*
954 			 * save the address of the first data segment
955 			 * of a object - used for the AT_SUNW_LDDATA
956 			 * aux entry.
957 			 */
958 			if ((lddata != NULL) && dfirst &&
959 			    (phdrp->p_flags & PF_W)) {
960 				*lddata = lo;
961 				dfirst = 0;
962 			}
963 		}
964 		phdrp = (Phdr *)((caddr_t)phdrp + hsize);
965 	}
966 
967 	len = hiaddr - (loaddr & PAGEMASK);
968 	len = roundup(len, PAGESIZE);
969 
970 	return (len);
971 }
972 
973 /*
974  * Read in the ELF header and program header table.
975  * SUSV3 requires:
976  *	ENOEXEC	File format is not recognized
977  *	EINVAL	Format recognized but execution not supported
978  */
979 static int
980 getelfhead(vnode_t *vp, cred_t *credp, Ehdr *ehdr, int *nshdrs, int *shstrndx,
981     int *nphdrs)
982 {
983 	int error;
984 	ssize_t resid;
985 
986 	/*
987 	 * We got here by the first two bytes in ident,
988 	 * now read the entire ELF header.
989 	 */
990 	if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)ehdr,
991 	    sizeof (Ehdr), (offset_t)0, UIO_SYSSPACE, 0,
992 	    (rlim64_t)0, credp, &resid)) != 0)
993 		return (error);
994 
995 	/*
996 	 * Since a separate version is compiled for handling 32-bit and
997 	 * 64-bit ELF executables on a 64-bit kernel, the 64-bit version
998 	 * doesn't need to be able to deal with 32-bit ELF files.
999 	 */
1000 	if (resid != 0 ||
1001 	    ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
1002 	    ehdr->e_ident[EI_MAG3] != ELFMAG3)
1003 		return (ENOEXEC);
1004 
1005 	if ((ehdr->e_type != ET_EXEC && ehdr->e_type != ET_DYN) ||
1006 #if defined(_ILP32) || defined(_ELF32_COMPAT)
1007 	    ehdr->e_ident[EI_CLASS] != ELFCLASS32 ||
1008 #else
1009 	    ehdr->e_ident[EI_CLASS] != ELFCLASS64 ||
1010 #endif
1011 	    !elfheadcheck(ehdr->e_ident[EI_DATA], ehdr->e_machine,
1012 	    ehdr->e_flags))
1013 		return (EINVAL);
1014 
1015 	*nshdrs = ehdr->e_shnum;
1016 	*shstrndx = ehdr->e_shstrndx;
1017 	*nphdrs = ehdr->e_phnum;
1018 
1019 	/*
1020 	 * If e_shnum, e_shstrndx, or e_phnum is its sentinel value, we need
1021 	 * to read in the section header at index zero to acces the true
1022 	 * values for those fields.
1023 	 */
1024 	if ((*nshdrs == 0 && ehdr->e_shoff != 0) ||
1025 	    *shstrndx == SHN_XINDEX || *nphdrs == PN_XNUM) {
1026 		Shdr shdr;
1027 
1028 		if (ehdr->e_shoff == 0)
1029 			return (EINVAL);
1030 
1031 		if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)&shdr,
1032 		    sizeof (shdr), (offset_t)ehdr->e_shoff, UIO_SYSSPACE, 0,
1033 		    (rlim64_t)0, credp, &resid)) != 0)
1034 			return (error);
1035 
1036 		if (*nshdrs == 0)
1037 			*nshdrs = shdr.sh_size;
1038 		if (*shstrndx == SHN_XINDEX)
1039 			*shstrndx = shdr.sh_link;
1040 		if (*nphdrs == PN_XNUM && shdr.sh_info != 0)
1041 			*nphdrs = shdr.sh_info;
1042 	}
1043 
1044 	return (0);
1045 }
1046 
1047 #ifdef _ELF32_COMPAT
1048 extern size_t elf_nphdr_max;
1049 #else
1050 size_t elf_nphdr_max = 1000;
1051 #endif
1052 
1053 static int
1054 getelfphdr(vnode_t *vp, cred_t *credp, const Ehdr *ehdr, int nphdrs,
1055     caddr_t *phbasep, ssize_t *phsizep)
1056 {
1057 	ssize_t resid, minsize;
1058 	int err;
1059 
1060 	/*
1061 	 * Since we're going to be using e_phentsize to iterate down the
1062 	 * array of program headers, it must be 8-byte aligned or else
1063 	 * a we might cause a misaligned access. We use all members through
1064 	 * p_flags on 32-bit ELF files and p_memsz on 64-bit ELF files so
1065 	 * e_phentsize must be at least large enough to include those
1066 	 * members.
1067 	 */
1068 #if !defined(_LP64) || defined(_ELF32_COMPAT)
1069 	minsize = offsetof(Phdr, p_flags) + sizeof (((Phdr *)NULL)->p_flags);
1070 #else
1071 	minsize = offsetof(Phdr, p_memsz) + sizeof (((Phdr *)NULL)->p_memsz);
1072 #endif
1073 	if (ehdr->e_phentsize < minsize || (ehdr->e_phentsize & 3))
1074 		return (EINVAL);
1075 
1076 	*phsizep = nphdrs * ehdr->e_phentsize;
1077 
1078 	if (*phsizep > sizeof (Phdr) * elf_nphdr_max) {
1079 		if ((*phbasep = kmem_alloc(*phsizep, KM_NOSLEEP)) == NULL)
1080 			return (ENOMEM);
1081 	} else {
1082 		*phbasep = kmem_alloc(*phsizep, KM_SLEEP);
1083 	}
1084 
1085 	if ((err = vn_rdwr(UIO_READ, vp, *phbasep, *phsizep,
1086 	    (offset_t)ehdr->e_phoff, UIO_SYSSPACE, 0, (rlim64_t)0,
1087 	    credp, &resid)) != 0) {
1088 		kmem_free(*phbasep, *phsizep);
1089 		*phbasep = NULL;
1090 		return (err);
1091 	}
1092 
1093 	return (0);
1094 }
1095 
1096 #ifdef _ELF32_COMPAT
1097 extern size_t elf_nshdr_max;
1098 extern size_t elf_shstrtab_max;
1099 #else
1100 size_t elf_nshdr_max = 10000;
1101 size_t elf_shstrtab_max = 100 * 1024;
1102 #endif
1103 
1104 
1105 static int
1106 getelfshdr(vnode_t *vp, cred_t *credp, const Ehdr *ehdr,
1107     int nshdrs, int shstrndx, caddr_t *shbasep, ssize_t *shsizep,
1108     char **shstrbasep, ssize_t *shstrsizep)
1109 {
1110 	ssize_t resid, minsize;
1111 	int err;
1112 	Shdr *shdr;
1113 
1114 	/*
1115 	 * Since we're going to be using e_shentsize to iterate down the
1116 	 * array of section headers, it must be 8-byte aligned or else
1117 	 * a we might cause a misaligned access. We use all members through
1118 	 * sh_entsize (on both 32- and 64-bit ELF files) so e_shentsize
1119 	 * must be at least large enough to include that member. The index
1120 	 * of the string table section must also be valid.
1121 	 */
1122 	minsize = offsetof(Shdr, sh_entsize) + sizeof (shdr->sh_entsize);
1123 	if (ehdr->e_shentsize < minsize || (ehdr->e_shentsize & 3) ||
1124 	    shstrndx >= nshdrs)
1125 		return (EINVAL);
1126 
1127 	*shsizep = nshdrs * ehdr->e_shentsize;
1128 
1129 	if (*shsizep > sizeof (Shdr) * elf_nshdr_max) {
1130 		if ((*shbasep = kmem_alloc(*shsizep, KM_NOSLEEP)) == NULL)
1131 			return (ENOMEM);
1132 	} else {
1133 		*shbasep = kmem_alloc(*shsizep, KM_SLEEP);
1134 	}
1135 
1136 	if ((err = vn_rdwr(UIO_READ, vp, *shbasep, *shsizep,
1137 	    (offset_t)ehdr->e_shoff, UIO_SYSSPACE, 0, (rlim64_t)0,
1138 	    credp, &resid)) != 0) {
1139 		kmem_free(*shbasep, *shsizep);
1140 		return (err);
1141 	}
1142 
1143 	/*
1144 	 * Pull the section string table out of the vnode; fail if the size
1145 	 * is zero.
1146 	 */
1147 	shdr = (Shdr *)(*shbasep + shstrndx * ehdr->e_shentsize);
1148 	if ((*shstrsizep = shdr->sh_size) == 0) {
1149 		kmem_free(*shbasep, *shsizep);
1150 		return (EINVAL);
1151 	}
1152 
1153 	if (*shstrsizep > elf_shstrtab_max) {
1154 		if ((*shstrbasep = kmem_alloc(*shstrsizep,
1155 		    KM_NOSLEEP)) == NULL) {
1156 			kmem_free(*shbasep, *shsizep);
1157 			return (ENOMEM);
1158 		}
1159 	} else {
1160 		*shstrbasep = kmem_alloc(*shstrsizep, KM_SLEEP);
1161 	}
1162 
1163 	if ((err = vn_rdwr(UIO_READ, vp, *shstrbasep, *shstrsizep,
1164 	    (offset_t)shdr->sh_offset, UIO_SYSSPACE, 0, (rlim64_t)0,
1165 	    credp, &resid)) != 0) {
1166 		kmem_free(*shbasep, *shsizep);
1167 		kmem_free(*shstrbasep, *shstrsizep);
1168 		return (err);
1169 	}
1170 
1171 	/*
1172 	 * Make sure the strtab is null-terminated to make sure we
1173 	 * don't run off the end of the table.
1174 	 */
1175 	(*shstrbasep)[*shstrsizep - 1] = '\0';
1176 
1177 	return (0);
1178 }
1179 
1180 static int
1181 mapelfexec(
1182 	vnode_t *vp,
1183 	Ehdr *ehdr,
1184 	int nphdrs,
1185 	caddr_t phdrbase,
1186 	Phdr **uphdr,
1187 	Phdr **dyphdr,
1188 	Phdr **stphdr,
1189 	Phdr **dtphdr,
1190 	Phdr *dataphdrp,
1191 	caddr_t *bssbase,
1192 	caddr_t *brkbase,
1193 	intptr_t *voffset,
1194 	intptr_t *minaddr,
1195 	size_t len,
1196 	long *execsz,
1197 	size_t *brksize)
1198 {
1199 	Phdr *phdr;
1200 	int i, prot, error;
1201 	caddr_t addr = NULL;
1202 	size_t zfodsz;
1203 	int ptload = 0;
1204 	int page;
1205 	off_t offset;
1206 	int hsize = ehdr->e_phentsize;
1207 	caddr_t mintmp = (caddr_t)-1;
1208 	extern int use_brk_lpg;
1209 
1210 	if (ehdr->e_type == ET_DYN) {
1211 		/*
1212 		 * Obtain the virtual address of a hole in the
1213 		 * address space to map the "interpreter".
1214 		 */
1215 		map_addr(&addr, len, (offset_t)0, 1, 0);
1216 		if (addr == NULL)
1217 			return (ENOMEM);
1218 		*voffset = (intptr_t)addr;
1219 
1220 		/*
1221 		 * Calculate the minimum vaddr so it can be subtracted out.
1222 		 * According to the ELF specification, since PT_LOAD sections
1223 		 * must be sorted by increasing p_vaddr values, this is
1224 		 * guaranteed to be the first PT_LOAD section.
1225 		 */
1226 		phdr = (Phdr *)phdrbase;
1227 		for (i = nphdrs; i > 0; i--) {
1228 			if (phdr->p_type == PT_LOAD) {
1229 				*voffset -= (uintptr_t)phdr->p_vaddr;
1230 				break;
1231 			}
1232 			phdr = (Phdr *)((caddr_t)phdr + hsize);
1233 		}
1234 
1235 	} else {
1236 		*voffset = 0;
1237 	}
1238 	phdr = (Phdr *)phdrbase;
1239 	for (i = nphdrs; i > 0; i--) {
1240 		switch (phdr->p_type) {
1241 		case PT_LOAD:
1242 			if ((*dyphdr != NULL) && (*uphdr == NULL))
1243 				return (0);
1244 
1245 			ptload = 1;
1246 			prot = PROT_USER;
1247 			if (phdr->p_flags & PF_R)
1248 				prot |= PROT_READ;
1249 			if (phdr->p_flags & PF_W)
1250 				prot |= PROT_WRITE;
1251 			if (phdr->p_flags & PF_X)
1252 				prot |= PROT_EXEC;
1253 
1254 			addr = (caddr_t)((uintptr_t)phdr->p_vaddr + *voffset);
1255 
1256 			/*
1257 			 * Keep track of the segment with the lowest starting
1258 			 * address.
1259 			 */
1260 			if (addr < mintmp)
1261 				mintmp = addr;
1262 
1263 			zfodsz = (size_t)phdr->p_memsz - phdr->p_filesz;
1264 
1265 			offset = phdr->p_offset;
1266 			if (((uintptr_t)offset & PAGEOFFSET) ==
1267 			    ((uintptr_t)addr & PAGEOFFSET) &&
1268 			    (!(vp->v_flag & VNOMAP))) {
1269 				page = 1;
1270 			} else {
1271 				page = 0;
1272 			}
1273 
1274 			/*
1275 			 * Set the heap pagesize for OOB when the bss size
1276 			 * is known and use_brk_lpg is not 0.
1277 			 */
1278 			if (brksize != NULL && use_brk_lpg &&
1279 			    zfodsz != 0 && phdr == dataphdrp &&
1280 			    (prot & PROT_WRITE)) {
1281 				size_t tlen = P2NPHASE((uintptr_t)addr +
1282 				    phdr->p_filesz, PAGESIZE);
1283 
1284 				if (zfodsz > tlen) {
1285 					curproc->p_brkpageszc =
1286 					    page_szc(map_pgsz(MAPPGSZ_HEAP,
1287 					    curproc, addr + phdr->p_filesz +
1288 					    tlen, zfodsz - tlen, 0));
1289 				}
1290 			}
1291 
1292 			if (curproc->p_brkpageszc != 0 && phdr == dataphdrp &&
1293 			    (prot & PROT_WRITE)) {
1294 				uint_t	szc = curproc->p_brkpageszc;
1295 				size_t pgsz = page_get_pagesize(szc);
1296 				caddr_t ebss = addr + phdr->p_memsz;
1297 				size_t extra_zfodsz;
1298 
1299 				ASSERT(pgsz > PAGESIZE);
1300 
1301 				extra_zfodsz = P2NPHASE((uintptr_t)ebss, pgsz);
1302 
1303 				if (error = execmap(vp, addr, phdr->p_filesz,
1304 				    zfodsz + extra_zfodsz, phdr->p_offset,
1305 				    prot, page, szc))
1306 					goto bad;
1307 				if (brksize != NULL)
1308 					*brksize = extra_zfodsz;
1309 			} else {
1310 				if (error = execmap(vp, addr, phdr->p_filesz,
1311 				    zfodsz, phdr->p_offset, prot, page, 0))
1312 					goto bad;
1313 			}
1314 
1315 			if (bssbase != NULL && addr >= *bssbase &&
1316 			    phdr == dataphdrp) {
1317 				*bssbase = addr + phdr->p_filesz;
1318 			}
1319 			if (brkbase != NULL && addr >= *brkbase) {
1320 				*brkbase = addr + phdr->p_memsz;
1321 			}
1322 
1323 			*execsz += btopr(phdr->p_memsz);
1324 			break;
1325 
1326 		case PT_INTERP:
1327 			if (ptload)
1328 				goto bad;
1329 			*dyphdr = phdr;
1330 			break;
1331 
1332 		case PT_SHLIB:
1333 			*stphdr = phdr;
1334 			break;
1335 
1336 		case PT_PHDR:
1337 			if (ptload)
1338 				goto bad;
1339 			*uphdr = phdr;
1340 			break;
1341 
1342 		case PT_NULL:
1343 		case PT_DYNAMIC:
1344 		case PT_NOTE:
1345 			break;
1346 
1347 		case PT_SUNWDTRACE:
1348 			if (dtphdr != NULL)
1349 				*dtphdr = phdr;
1350 			break;
1351 
1352 		default:
1353 			break;
1354 		}
1355 		phdr = (Phdr *)((caddr_t)phdr + hsize);
1356 	}
1357 
1358 	if (minaddr != NULL) {
1359 		ASSERT(mintmp != (caddr_t)-1);
1360 		*minaddr = (intptr_t)mintmp;
1361 	}
1362 
1363 	return (0);
1364 bad:
1365 	if (error == 0)
1366 		error = EINVAL;
1367 	return (error);
1368 }
1369 
1370 int
1371 elfnote(vnode_t *vp, offset_t *offsetp, int type, int descsz, void *desc,
1372     rlim64_t rlimit, cred_t *credp)
1373 {
1374 	Note note;
1375 	int error;
1376 
1377 	bzero(&note, sizeof (note));
1378 	bcopy("CORE", note.name, 4);
1379 	note.nhdr.n_type = type;
1380 	/*
1381 	 * The System V ABI states that n_namesz must be the length of the
1382 	 * string that follows the Nhdr structure including the terminating
1383 	 * null. The ABI also specifies that sufficient padding should be
1384 	 * included so that the description that follows the name string
1385 	 * begins on a 4- or 8-byte boundary for 32- and 64-bit binaries
1386 	 * respectively. However, since this change was not made correctly
1387 	 * at the time of the 64-bit port, both 32- and 64-bit binaries
1388 	 * descriptions are only guaranteed to begin on a 4-byte boundary.
1389 	 */
1390 	note.nhdr.n_namesz = 5;
1391 	note.nhdr.n_descsz = roundup(descsz, sizeof (Word));
1392 
1393 	if (error = core_write(vp, UIO_SYSSPACE, *offsetp, &note,
1394 	    sizeof (note), rlimit, credp))
1395 		return (error);
1396 
1397 	*offsetp += sizeof (note);
1398 
1399 	if (error = core_write(vp, UIO_SYSSPACE, *offsetp, desc,
1400 	    note.nhdr.n_descsz, rlimit, credp))
1401 		return (error);
1402 
1403 	*offsetp += note.nhdr.n_descsz;
1404 	return (0);
1405 }
1406 
1407 /*
1408  * Copy the section data from one vnode to the section of another vnode.
1409  */
1410 static void
1411 copy_scn(Shdr *src, vnode_t *src_vp, Shdr *dst, vnode_t *dst_vp, Off *doffset,
1412     void *buf, size_t size, cred_t *credp, rlim64_t rlimit)
1413 {
1414 	ssize_t resid;
1415 	size_t len, n = src->sh_size;
1416 	offset_t off = 0;
1417 
1418 	while (n != 0) {
1419 		len = MIN(size, n);
1420 		if (vn_rdwr(UIO_READ, src_vp, buf, len, src->sh_offset + off,
1421 		    UIO_SYSSPACE, 0, (rlim64_t)0, credp, &resid) != 0 ||
1422 		    resid >= len ||
1423 		    core_write(dst_vp, UIO_SYSSPACE, *doffset + off,
1424 		    buf, len - resid, rlimit, credp) != 0) {
1425 			dst->sh_size = 0;
1426 			dst->sh_offset = 0;
1427 			return;
1428 		}
1429 
1430 		ASSERT(n >= len - resid);
1431 
1432 		n -= len - resid;
1433 		off += len - resid;
1434 	}
1435 
1436 	*doffset += src->sh_size;
1437 }
1438 
1439 #ifdef _ELF32_COMPAT
1440 extern size_t elf_datasz_max;
1441 #else
1442 size_t elf_datasz_max = 1 * 1024 * 1024;
1443 #endif
1444 
1445 /*
1446  * This function processes mappings that correspond to load objects to
1447  * examine their respective sections for elfcore(). It's called once with
1448  * v set to NULL to count the number of sections that we're going to need
1449  * and then again with v set to some allocated buffer that we fill in with
1450  * all the section data.
1451  */
1452 static int
1453 process_scns(core_content_t content, proc_t *p, cred_t *credp, vnode_t *vp,
1454     Shdr *v, int nv, rlim64_t rlimit, Off *doffsetp, int *nshdrsp)
1455 {
1456 	vnode_t *lastvp = NULL;
1457 	struct seg *seg;
1458 	int i, j;
1459 	void *data = NULL;
1460 	size_t datasz = 0;
1461 	shstrtab_t shstrtab;
1462 	struct as *as = p->p_as;
1463 	int error = 0;
1464 
1465 	if (v != NULL)
1466 		shstrtab_init(&shstrtab);
1467 
1468 	i = 1;
1469 	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
1470 		uint_t prot;
1471 		vnode_t *mvp;
1472 		void *tmp = NULL;
1473 		caddr_t saddr = seg->s_base;
1474 		caddr_t naddr;
1475 		caddr_t eaddr;
1476 		size_t segsize;
1477 
1478 		Ehdr ehdr;
1479 		int nshdrs, shstrndx, nphdrs;
1480 		caddr_t shbase;
1481 		ssize_t shsize;
1482 		char *shstrbase;
1483 		ssize_t shstrsize;
1484 
1485 		Shdr *shdr;
1486 		const char *name;
1487 		size_t sz;
1488 		uintptr_t off;
1489 
1490 		int ctf_ndx = 0;
1491 		int symtab_ndx = 0;
1492 
1493 		/*
1494 		 * Since we're just looking for text segments of load
1495 		 * objects, we only care about the protection bits; we don't
1496 		 * care about the actual size of the segment so we use the
1497 		 * reserved size. If the segment's size is zero, there's
1498 		 * something fishy going on so we ignore this segment.
1499 		 */
1500 		if (seg->s_ops != &segvn_ops ||
1501 		    SEGOP_GETVP(seg, seg->s_base, &mvp) != 0 ||
1502 		    mvp == lastvp || mvp == NULL || mvp->v_type != VREG ||
1503 		    (segsize = pr_getsegsize(seg, 1)) == 0)
1504 			continue;
1505 
1506 		eaddr = saddr + segsize;
1507 		prot = pr_getprot(seg, 1, &tmp, &saddr, &naddr, eaddr);
1508 		pr_getprot_done(&tmp);
1509 
1510 		/*
1511 		 * Skip this segment unless the protection bits look like
1512 		 * what we'd expect for a text segment.
1513 		 */
1514 		if ((prot & (PROT_WRITE | PROT_EXEC)) != PROT_EXEC)
1515 			continue;
1516 
1517 		if (getelfhead(mvp, credp, &ehdr, &nshdrs, &shstrndx,
1518 		    &nphdrs) != 0 ||
1519 		    getelfshdr(mvp, credp, &ehdr, nshdrs, shstrndx,
1520 		    &shbase, &shsize, &shstrbase, &shstrsize) != 0)
1521 			continue;
1522 
1523 		off = ehdr.e_shentsize;
1524 		for (j = 1; j < nshdrs; j++, off += ehdr.e_shentsize) {
1525 			Shdr *symtab = NULL, *strtab;
1526 
1527 			shdr = (Shdr *)(shbase + off);
1528 
1529 			if (shdr->sh_name >= shstrsize)
1530 				continue;
1531 
1532 			name = shstrbase + shdr->sh_name;
1533 
1534 			if (strcmp(name, shstrtab_data[STR_CTF]) == 0) {
1535 				if ((content & CC_CONTENT_CTF) == 0 ||
1536 				    ctf_ndx != 0)
1537 					continue;
1538 
1539 				if (shdr->sh_link > 0 &&
1540 				    shdr->sh_link < nshdrs) {
1541 					symtab = (Shdr *)(shbase +
1542 					    shdr->sh_link * ehdr.e_shentsize);
1543 				}
1544 
1545 				if (v != NULL && i < nv - 1) {
1546 					if (shdr->sh_size > datasz &&
1547 					    shdr->sh_size <= elf_datasz_max) {
1548 						if (data != NULL)
1549 							kmem_free(data, datasz);
1550 
1551 						datasz = shdr->sh_size;
1552 						data = kmem_alloc(datasz,
1553 						    KM_SLEEP);
1554 					}
1555 
1556 					v[i].sh_name = shstrtab_ndx(&shstrtab,
1557 					    STR_CTF);
1558 					v[i].sh_addr = (Addr)(uintptr_t)saddr;
1559 					v[i].sh_type = SHT_PROGBITS;
1560 					v[i].sh_addralign = 4;
1561 					*doffsetp = roundup(*doffsetp,
1562 					    v[i].sh_addralign);
1563 					v[i].sh_offset = *doffsetp;
1564 					v[i].sh_size = shdr->sh_size;
1565 					if (symtab == NULL)  {
1566 						v[i].sh_link = 0;
1567 					} else if (symtab->sh_type ==
1568 					    SHT_SYMTAB &&
1569 					    symtab_ndx != 0) {
1570 						v[i].sh_link =
1571 						    symtab_ndx;
1572 					} else {
1573 						v[i].sh_link = i + 1;
1574 					}
1575 
1576 					copy_scn(shdr, mvp, &v[i], vp,
1577 					    doffsetp, data, datasz, credp,
1578 					    rlimit);
1579 				}
1580 
1581 				ctf_ndx = i++;
1582 
1583 				/*
1584 				 * We've already dumped the symtab.
1585 				 */
1586 				if (symtab != NULL &&
1587 				    symtab->sh_type == SHT_SYMTAB &&
1588 				    symtab_ndx != 0)
1589 					continue;
1590 
1591 			} else if (strcmp(name,
1592 			    shstrtab_data[STR_SYMTAB]) == 0) {
1593 				if ((content & CC_CONTENT_SYMTAB) == 0 ||
1594 				    symtab != 0)
1595 					continue;
1596 
1597 				symtab = shdr;
1598 			}
1599 
1600 			if (symtab != NULL) {
1601 				if ((symtab->sh_type != SHT_DYNSYM &&
1602 				    symtab->sh_type != SHT_SYMTAB) ||
1603 				    symtab->sh_link == 0 ||
1604 				    symtab->sh_link >= nshdrs)
1605 					continue;
1606 
1607 				strtab = (Shdr *)(shbase +
1608 				    symtab->sh_link * ehdr.e_shentsize);
1609 
1610 				if (strtab->sh_type != SHT_STRTAB)
1611 					continue;
1612 
1613 				if (v != NULL && i < nv - 2) {
1614 					sz = MAX(symtab->sh_size,
1615 					    strtab->sh_size);
1616 					if (sz > datasz &&
1617 					    sz <= elf_datasz_max) {
1618 						if (data != NULL)
1619 							kmem_free(data, datasz);
1620 
1621 						datasz = sz;
1622 						data = kmem_alloc(datasz,
1623 						    KM_SLEEP);
1624 					}
1625 
1626 					if (symtab->sh_type == SHT_DYNSYM) {
1627 						v[i].sh_name = shstrtab_ndx(
1628 						    &shstrtab, STR_DYNSYM);
1629 						v[i + 1].sh_name = shstrtab_ndx(
1630 						    &shstrtab, STR_DYNSTR);
1631 					} else {
1632 						v[i].sh_name = shstrtab_ndx(
1633 						    &shstrtab, STR_SYMTAB);
1634 						v[i + 1].sh_name = shstrtab_ndx(
1635 						    &shstrtab, STR_STRTAB);
1636 					}
1637 
1638 					v[i].sh_type = symtab->sh_type;
1639 					v[i].sh_addr = symtab->sh_addr;
1640 					if (ehdr.e_type == ET_DYN ||
1641 					    v[i].sh_addr == 0)
1642 						v[i].sh_addr +=
1643 						    (Addr)(uintptr_t)saddr;
1644 					v[i].sh_addralign =
1645 					    symtab->sh_addralign;
1646 					*doffsetp = roundup(*doffsetp,
1647 					    v[i].sh_addralign);
1648 					v[i].sh_offset = *doffsetp;
1649 					v[i].sh_size = symtab->sh_size;
1650 					v[i].sh_link = i + 1;
1651 					v[i].sh_entsize = symtab->sh_entsize;
1652 					v[i].sh_info = symtab->sh_info;
1653 
1654 					copy_scn(symtab, mvp, &v[i], vp,
1655 					    doffsetp, data, datasz, credp,
1656 					    rlimit);
1657 
1658 					v[i + 1].sh_type = SHT_STRTAB;
1659 					v[i + 1].sh_flags = SHF_STRINGS;
1660 					v[i + 1].sh_addr = symtab->sh_addr;
1661 					if (ehdr.e_type == ET_DYN ||
1662 					    v[i + 1].sh_addr == 0)
1663 						v[i + 1].sh_addr +=
1664 						    (Addr)(uintptr_t)saddr;
1665 					v[i + 1].sh_addralign =
1666 					    strtab->sh_addralign;
1667 					*doffsetp = roundup(*doffsetp,
1668 					    v[i + 1].sh_addralign);
1669 					v[i + 1].sh_offset = *doffsetp;
1670 					v[i + 1].sh_size = strtab->sh_size;
1671 
1672 					copy_scn(strtab, mvp, &v[i + 1], vp,
1673 					    doffsetp, data, datasz, credp,
1674 					    rlimit);
1675 				}
1676 
1677 				if (symtab->sh_type == SHT_SYMTAB)
1678 					symtab_ndx = i;
1679 				i += 2;
1680 			}
1681 		}
1682 
1683 		kmem_free(shstrbase, shstrsize);
1684 		kmem_free(shbase, shsize);
1685 
1686 		lastvp = mvp;
1687 	}
1688 
1689 	if (v == NULL) {
1690 		if (i == 1)
1691 			*nshdrsp = 0;
1692 		else
1693 			*nshdrsp = i + 1;
1694 		goto done;
1695 	}
1696 
1697 	if (i != nv - 1) {
1698 		cmn_err(CE_WARN, "elfcore: core dump failed for "
1699 		    "process %d; address space is changing", p->p_pid);
1700 		error = EIO;
1701 		goto done;
1702 	}
1703 
1704 	v[i].sh_name = shstrtab_ndx(&shstrtab, STR_SHSTRTAB);
1705 	v[i].sh_size = shstrtab_size(&shstrtab);
1706 	v[i].sh_addralign = 1;
1707 	*doffsetp = roundup(*doffsetp, v[i].sh_addralign);
1708 	v[i].sh_offset = *doffsetp;
1709 	v[i].sh_flags = SHF_STRINGS;
1710 	v[i].sh_type = SHT_STRTAB;
1711 
1712 	if (v[i].sh_size > datasz) {
1713 		if (data != NULL)
1714 			kmem_free(data, datasz);
1715 
1716 		datasz = v[i].sh_size;
1717 		data = kmem_alloc(datasz,
1718 		    KM_SLEEP);
1719 	}
1720 
1721 	shstrtab_dump(&shstrtab, data);
1722 
1723 	if ((error = core_write(vp, UIO_SYSSPACE, *doffsetp,
1724 	    data, v[i].sh_size, rlimit, credp)) != 0)
1725 		goto done;
1726 
1727 	*doffsetp += v[i].sh_size;
1728 
1729 done:
1730 	if (data != NULL)
1731 		kmem_free(data, datasz);
1732 
1733 	return (error);
1734 }
1735 
1736 int
1737 elfcore(vnode_t *vp, proc_t *p, cred_t *credp, rlim64_t rlimit, int sig,
1738     core_content_t content)
1739 {
1740 	offset_t poffset, soffset;
1741 	Off doffset;
1742 	int error, i, nphdrs, nshdrs;
1743 	int overflow = 0;
1744 	struct seg *seg;
1745 	struct as *as = p->p_as;
1746 	union {
1747 		Ehdr ehdr;
1748 		Phdr phdr[1];
1749 		Shdr shdr[1];
1750 	} *bigwad;
1751 	size_t bigsize;
1752 	size_t phdrsz, shdrsz;
1753 	Ehdr *ehdr;
1754 	Phdr *v;
1755 	caddr_t brkbase;
1756 	size_t brksize;
1757 	caddr_t stkbase;
1758 	size_t stksize;
1759 	int ntries = 0;
1760 	klwp_t *lwp = ttolwp(curthread);
1761 
1762 top:
1763 	/*
1764 	 * Make sure we have everything we need (registers, etc.).
1765 	 * All other lwps have already stopped and are in an orderly state.
1766 	 */
1767 	ASSERT(p == ttoproc(curthread));
1768 	prstop(0, 0);
1769 
1770 	AS_LOCK_ENTER(as, RW_WRITER);
1771 	nphdrs = prnsegs(as, 0) + 2;		/* two CORE note sections */
1772 
1773 	/*
1774 	 * Count the number of section headers we're going to need.
1775 	 */
1776 	nshdrs = 0;
1777 	if (content & (CC_CONTENT_CTF | CC_CONTENT_SYMTAB)) {
1778 		(void) process_scns(content, p, credp, NULL, NULL, NULL, 0,
1779 		    NULL, &nshdrs);
1780 	}
1781 	AS_LOCK_EXIT(as);
1782 
1783 	ASSERT(nshdrs == 0 || nshdrs > 1);
1784 
1785 	/*
1786 	 * The core file contents may required zero section headers, but if
1787 	 * we overflow the 16 bits allotted to the program header count in
1788 	 * the ELF header, we'll need that program header at index zero.
1789 	 */
1790 	if (nshdrs == 0 && nphdrs >= PN_XNUM)
1791 		nshdrs = 1;
1792 
1793 	phdrsz = nphdrs * sizeof (Phdr);
1794 	shdrsz = nshdrs * sizeof (Shdr);
1795 
1796 	bigsize = MAX(sizeof (*bigwad), MAX(phdrsz, shdrsz));
1797 	bigwad = kmem_alloc(bigsize, KM_SLEEP);
1798 
1799 	ehdr = &bigwad->ehdr;
1800 	bzero(ehdr, sizeof (*ehdr));
1801 
1802 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1803 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1804 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1805 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1806 	ehdr->e_ident[EI_CLASS] = ELFCLASS;
1807 	ehdr->e_type = ET_CORE;
1808 
1809 #if !defined(_LP64) || defined(_ELF32_COMPAT)
1810 
1811 #if defined(__sparc)
1812 	ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
1813 	ehdr->e_machine = EM_SPARC;
1814 #elif defined(__i386) || defined(__i386_COMPAT)
1815 	ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1816 	ehdr->e_machine = EM_386;
1817 #else
1818 #error "no recognized machine type is defined"
1819 #endif
1820 
1821 #else	/* !defined(_LP64) || defined(_ELF32_COMPAT) */
1822 
1823 #if defined(__sparc)
1824 	ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
1825 	ehdr->e_machine = EM_SPARCV9;
1826 #elif defined(__amd64)
1827 	ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1828 	ehdr->e_machine = EM_AMD64;
1829 #else
1830 #error "no recognized 64-bit machine type is defined"
1831 #endif
1832 
1833 #endif	/* !defined(_LP64) || defined(_ELF32_COMPAT) */
1834 
1835 	/*
1836 	 * If the count of program headers or section headers or the index
1837 	 * of the section string table can't fit in the mere 16 bits
1838 	 * shortsightedly allotted to them in the ELF header, we use the
1839 	 * extended formats and put the real values in the section header
1840 	 * as index 0.
1841 	 */
1842 	ehdr->e_version = EV_CURRENT;
1843 	ehdr->e_ehsize = sizeof (Ehdr);
1844 
1845 	if (nphdrs >= PN_XNUM)
1846 		ehdr->e_phnum = PN_XNUM;
1847 	else
1848 		ehdr->e_phnum = (unsigned short)nphdrs;
1849 
1850 	ehdr->e_phoff = sizeof (Ehdr);
1851 	ehdr->e_phentsize = sizeof (Phdr);
1852 
1853 	if (nshdrs > 0) {
1854 		if (nshdrs >= SHN_LORESERVE)
1855 			ehdr->e_shnum = 0;
1856 		else
1857 			ehdr->e_shnum = (unsigned short)nshdrs;
1858 
1859 		if (nshdrs - 1 >= SHN_LORESERVE)
1860 			ehdr->e_shstrndx = SHN_XINDEX;
1861 		else
1862 			ehdr->e_shstrndx = (unsigned short)(nshdrs - 1);
1863 
1864 		ehdr->e_shoff = ehdr->e_phoff + ehdr->e_phentsize * nphdrs;
1865 		ehdr->e_shentsize = sizeof (Shdr);
1866 	}
1867 
1868 	if (error = core_write(vp, UIO_SYSSPACE, (offset_t)0, ehdr,
1869 	    sizeof (Ehdr), rlimit, credp))
1870 		goto done;
1871 
1872 	poffset = sizeof (Ehdr);
1873 	soffset = sizeof (Ehdr) + phdrsz;
1874 	doffset = sizeof (Ehdr) + phdrsz + shdrsz;
1875 
1876 	v = &bigwad->phdr[0];
1877 	bzero(v, phdrsz);
1878 
1879 	setup_old_note_header(&v[0], p);
1880 	v[0].p_offset = doffset = roundup(doffset, sizeof (Word));
1881 	doffset += v[0].p_filesz;
1882 
1883 	setup_note_header(&v[1], p);
1884 	v[1].p_offset = doffset = roundup(doffset, sizeof (Word));
1885 	doffset += v[1].p_filesz;
1886 
1887 	mutex_enter(&p->p_lock);
1888 
1889 	brkbase = p->p_brkbase;
1890 	brksize = p->p_brksize;
1891 
1892 	stkbase = p->p_usrstack - p->p_stksize;
1893 	stksize = p->p_stksize;
1894 
1895 	mutex_exit(&p->p_lock);
1896 
1897 	AS_LOCK_ENTER(as, RW_WRITER);
1898 	i = 2;
1899 	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
1900 		caddr_t eaddr = seg->s_base + pr_getsegsize(seg, 0);
1901 		caddr_t saddr, naddr;
1902 		void *tmp = NULL;
1903 		extern struct seg_ops segspt_shmops;
1904 
1905 		for (saddr = seg->s_base; saddr < eaddr; saddr = naddr) {
1906 			uint_t prot;
1907 			size_t size;
1908 			int type;
1909 			vnode_t *mvp;
1910 
1911 			prot = pr_getprot(seg, 0, &tmp, &saddr, &naddr, eaddr);
1912 			prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
1913 			if ((size = (size_t)(naddr - saddr)) == 0)
1914 				continue;
1915 			if (i == nphdrs) {
1916 				overflow++;
1917 				continue;
1918 			}
1919 			v[i].p_type = PT_LOAD;
1920 			v[i].p_vaddr = (Addr)(uintptr_t)saddr;
1921 			v[i].p_memsz = size;
1922 			if (prot & PROT_READ)
1923 				v[i].p_flags |= PF_R;
1924 			if (prot & PROT_WRITE)
1925 				v[i].p_flags |= PF_W;
1926 			if (prot & PROT_EXEC)
1927 				v[i].p_flags |= PF_X;
1928 
1929 			/*
1930 			 * Figure out which mappings to include in the core.
1931 			 */
1932 			type = SEGOP_GETTYPE(seg, saddr);
1933 
1934 			if (saddr == stkbase && size == stksize) {
1935 				if (!(content & CC_CONTENT_STACK))
1936 					goto exclude;
1937 
1938 			} else if (saddr == brkbase && size == brksize) {
1939 				if (!(content & CC_CONTENT_HEAP))
1940 					goto exclude;
1941 
1942 			} else if (seg->s_ops == &segspt_shmops) {
1943 				if (type & MAP_NORESERVE) {
1944 					if (!(content & CC_CONTENT_DISM))
1945 						goto exclude;
1946 				} else {
1947 					if (!(content & CC_CONTENT_ISM))
1948 						goto exclude;
1949 				}
1950 
1951 			} else if (seg->s_ops != &segvn_ops) {
1952 				goto exclude;
1953 
1954 			} else if (type & MAP_SHARED) {
1955 				if (shmgetid(p, saddr) != SHMID_NONE) {
1956 					if (!(content & CC_CONTENT_SHM))
1957 						goto exclude;
1958 
1959 				} else if (SEGOP_GETVP(seg, seg->s_base,
1960 				    &mvp) != 0 || mvp == NULL ||
1961 				    mvp->v_type != VREG) {
1962 					if (!(content & CC_CONTENT_SHANON))
1963 						goto exclude;
1964 
1965 				} else {
1966 					if (!(content & CC_CONTENT_SHFILE))
1967 						goto exclude;
1968 				}
1969 
1970 			} else if (SEGOP_GETVP(seg, seg->s_base, &mvp) != 0 ||
1971 			    mvp == NULL || mvp->v_type != VREG) {
1972 				if (!(content & CC_CONTENT_ANON))
1973 					goto exclude;
1974 
1975 			} else if (prot == (PROT_READ | PROT_EXEC)) {
1976 				if (!(content & CC_CONTENT_TEXT))
1977 					goto exclude;
1978 
1979 			} else if (prot == PROT_READ) {
1980 				if (!(content & CC_CONTENT_RODATA))
1981 					goto exclude;
1982 
1983 			} else {
1984 				if (!(content & CC_CONTENT_DATA))
1985 					goto exclude;
1986 			}
1987 
1988 			doffset = roundup(doffset, sizeof (Word));
1989 			v[i].p_offset = doffset;
1990 			v[i].p_filesz = size;
1991 			doffset += size;
1992 exclude:
1993 			i++;
1994 		}
1995 		ASSERT(tmp == NULL);
1996 	}
1997 	AS_LOCK_EXIT(as);
1998 
1999 	if (overflow || i != nphdrs) {
2000 		if (ntries++ == 0) {
2001 			kmem_free(bigwad, bigsize);
2002 			overflow = 0;
2003 			goto top;
2004 		}
2005 		cmn_err(CE_WARN, "elfcore: core dump failed for "
2006 		    "process %d; address space is changing", p->p_pid);
2007 		error = EIO;
2008 		goto done;
2009 	}
2010 
2011 	if ((error = core_write(vp, UIO_SYSSPACE, poffset,
2012 	    v, phdrsz, rlimit, credp)) != 0)
2013 		goto done;
2014 
2015 	if ((error = write_old_elfnotes(p, sig, vp, v[0].p_offset, rlimit,
2016 	    credp)) != 0)
2017 		goto done;
2018 
2019 	if ((error = write_elfnotes(p, sig, vp, v[1].p_offset, rlimit,
2020 	    credp, content)) != 0)
2021 		goto done;
2022 
2023 	for (i = 2; i < nphdrs; i++) {
2024 		prkillinfo_t killinfo;
2025 		sigqueue_t *sq;
2026 		int sig, j;
2027 
2028 		if (v[i].p_filesz == 0)
2029 			continue;
2030 
2031 		/*
2032 		 * If dumping out this segment fails, rather than failing
2033 		 * the core dump entirely, we reset the size of the mapping
2034 		 * to zero to indicate that the data is absent from the core
2035 		 * file and or in the PF_SUNW_FAILURE flag to differentiate
2036 		 * this from mappings that were excluded due to the core file
2037 		 * content settings.
2038 		 */
2039 		if ((error = core_seg(p, vp, v[i].p_offset,
2040 		    (caddr_t)(uintptr_t)v[i].p_vaddr, v[i].p_filesz,
2041 		    rlimit, credp)) == 0) {
2042 			continue;
2043 		}
2044 
2045 		if ((sig = lwp->lwp_cursig) == 0) {
2046 			/*
2047 			 * We failed due to something other than a signal.
2048 			 * Since the space reserved for the segment is now
2049 			 * unused, we stash the errno in the first four
2050 			 * bytes. This undocumented interface will let us
2051 			 * understand the nature of the failure.
2052 			 */
2053 			(void) core_write(vp, UIO_SYSSPACE, v[i].p_offset,
2054 			    &error, sizeof (error), rlimit, credp);
2055 
2056 			v[i].p_filesz = 0;
2057 			v[i].p_flags |= PF_SUNW_FAILURE;
2058 			if ((error = core_write(vp, UIO_SYSSPACE,
2059 			    poffset + sizeof (v[i]) * i, &v[i], sizeof (v[i]),
2060 			    rlimit, credp)) != 0)
2061 				goto done;
2062 
2063 			continue;
2064 		}
2065 
2066 		/*
2067 		 * We took a signal.  We want to abort the dump entirely, but
2068 		 * we also want to indicate what failed and why.  We therefore
2069 		 * use the space reserved for the first failing segment to
2070 		 * write our error (which, for purposes of compatability with
2071 		 * older core dump readers, we set to EINTR) followed by any
2072 		 * siginfo associated with the signal.
2073 		 */
2074 		bzero(&killinfo, sizeof (killinfo));
2075 		killinfo.prk_error = EINTR;
2076 
2077 		sq = sig == SIGKILL ? curproc->p_killsqp : lwp->lwp_curinfo;
2078 
2079 		if (sq != NULL) {
2080 			bcopy(&sq->sq_info, &killinfo.prk_info,
2081 			    sizeof (sq->sq_info));
2082 		} else {
2083 			killinfo.prk_info.si_signo = lwp->lwp_cursig;
2084 			killinfo.prk_info.si_code = SI_NOINFO;
2085 		}
2086 
2087 #if (defined(_SYSCALL32_IMPL) || defined(_LP64))
2088 		/*
2089 		 * If this is a 32-bit process, we need to translate from the
2090 		 * native siginfo to the 32-bit variant.  (Core readers must
2091 		 * always have the same data model as their target or must
2092 		 * be aware of -- and compensate for -- data model differences.)
2093 		 */
2094 		if (curproc->p_model == DATAMODEL_ILP32) {
2095 			siginfo32_t si32;
2096 
2097 			siginfo_kto32((k_siginfo_t *)&killinfo.prk_info, &si32);
2098 			bcopy(&si32, &killinfo.prk_info, sizeof (si32));
2099 		}
2100 #endif
2101 
2102 		(void) core_write(vp, UIO_SYSSPACE, v[i].p_offset,
2103 		    &killinfo, sizeof (killinfo), rlimit, credp);
2104 
2105 		/*
2106 		 * For the segment on which we took the signal, indicate that
2107 		 * its data now refers to a siginfo.
2108 		 */
2109 		v[i].p_filesz = 0;
2110 		v[i].p_flags |= PF_SUNW_FAILURE | PF_SUNW_KILLED |
2111 		    PF_SUNW_SIGINFO;
2112 
2113 		/*
2114 		 * And for every other segment, indicate that its absence
2115 		 * is due to a signal.
2116 		 */
2117 		for (j = i + 1; j < nphdrs; j++) {
2118 			v[j].p_filesz = 0;
2119 			v[j].p_flags |= PF_SUNW_FAILURE | PF_SUNW_KILLED;
2120 		}
2121 
2122 		/*
2123 		 * Finally, write out our modified program headers.
2124 		 */
2125 		if ((error = core_write(vp, UIO_SYSSPACE,
2126 		    poffset + sizeof (v[i]) * i, &v[i],
2127 		    sizeof (v[i]) * (nphdrs - i), rlimit, credp)) != 0)
2128 			goto done;
2129 
2130 		break;
2131 	}
2132 
2133 	if (nshdrs > 0) {
2134 		bzero(&bigwad->shdr[0], shdrsz);
2135 
2136 		if (nshdrs >= SHN_LORESERVE)
2137 			bigwad->shdr[0].sh_size = nshdrs;
2138 
2139 		if (nshdrs - 1 >= SHN_LORESERVE)
2140 			bigwad->shdr[0].sh_link = nshdrs - 1;
2141 
2142 		if (nphdrs >= PN_XNUM)
2143 			bigwad->shdr[0].sh_info = nphdrs;
2144 
2145 		if (nshdrs > 1) {
2146 			AS_LOCK_ENTER(as, RW_WRITER);
2147 			if ((error = process_scns(content, p, credp, vp,
2148 			    &bigwad->shdr[0], nshdrs, rlimit, &doffset,
2149 			    NULL)) != 0) {
2150 				AS_LOCK_EXIT(as);
2151 				goto done;
2152 			}
2153 			AS_LOCK_EXIT(as);
2154 		}
2155 
2156 		if ((error = core_write(vp, UIO_SYSSPACE, soffset,
2157 		    &bigwad->shdr[0], shdrsz, rlimit, credp)) != 0)
2158 			goto done;
2159 	}
2160 
2161 done:
2162 	kmem_free(bigwad, bigsize);
2163 	return (error);
2164 }
2165 
2166 #ifndef	_ELF32_COMPAT
2167 
2168 static struct execsw esw = {
2169 #ifdef	_LP64
2170 	elf64magicstr,
2171 #else	/* _LP64 */
2172 	elf32magicstr,
2173 #endif	/* _LP64 */
2174 	0,
2175 	5,
2176 	elfexec,
2177 	elfcore
2178 };
2179 
2180 static struct modlexec modlexec = {
2181 	&mod_execops, "exec module for elf", &esw
2182 };
2183 
2184 #ifdef	_LP64
2185 extern int elf32exec(vnode_t *vp, execa_t *uap, uarg_t *args,
2186 			intpdata_t *idatap, int level, long *execsz,
2187 			int setid, caddr_t exec_file, cred_t *cred,
2188 			int brand_action);
2189 extern int elf32core(vnode_t *vp, proc_t *p, cred_t *credp,
2190 			rlim64_t rlimit, int sig, core_content_t content);
2191 
2192 static struct execsw esw32 = {
2193 	elf32magicstr,
2194 	0,
2195 	5,
2196 	elf32exec,
2197 	elf32core
2198 };
2199 
2200 static struct modlexec modlexec32 = {
2201 	&mod_execops, "32-bit exec module for elf", &esw32
2202 };
2203 #endif	/* _LP64 */
2204 
2205 static struct modlinkage modlinkage = {
2206 	MODREV_1,
2207 	(void *)&modlexec,
2208 #ifdef	_LP64
2209 	(void *)&modlexec32,
2210 #endif	/* _LP64 */
2211 	NULL
2212 };
2213 
2214 int
2215 _init(void)
2216 {
2217 	return (mod_install(&modlinkage));
2218 }
2219 
2220 int
2221 _fini(void)
2222 {
2223 	return (mod_remove(&modlinkage));
2224 }
2225 
2226 int
2227 _info(struct modinfo *modinfop)
2228 {
2229 	return (mod_info(&modlinkage, modinfop));
2230 }
2231 
2232 #endif	/* !_ELF32_COMPAT */
2233