xref: /illumos-gate/usr/src/uts/common/os/core.c (revision 287247a826fa2ab8d01f6c8f276d405eb08420f8)
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  * Copyright (c) 2011, Joyent Inc. All rights reserved.
25  */
26 
27 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
28 /*	  All Rights Reserved  	*/
29 
30 #include <sys/param.h>
31 #include <sys/types.h>
32 #include <sys/time.h>
33 #include <sys/sysmacros.h>
34 #include <sys/proc.h>
35 #include <sys/systm.h>
36 #include <sys/cred.h>
37 #include <sys/user.h>
38 #include <sys/utsname.h>
39 #include <sys/errno.h>
40 #include <sys/signal.h>
41 #include <sys/siginfo.h>
42 #include <sys/fault.h>
43 #include <sys/syscall.h>
44 #include <sys/ucontext.h>
45 #include <sys/prsystm.h>
46 #include <sys/vnode.h>
47 #include <sys/var.h>
48 #include <sys/file.h>
49 #include <sys/pathname.h>
50 #include <sys/vfs.h>
51 #include <sys/exec.h>
52 #include <sys/debug.h>
53 #include <sys/stack.h>
54 #include <sys/kmem.h>
55 #include <sys/schedctl.h>
56 #include <sys/core.h>
57 #include <sys/corectl.h>
58 #include <sys/cmn_err.h>
59 #include <vm/as.h>
60 #include <sys/rctl.h>
61 #include <sys/nbmlock.h>
62 #include <sys/stat.h>
63 #include <sys/zone.h>
64 #include <sys/contract/process_impl.h>
65 #include <sys/ddi.h>
66 
67 /*
68  * Processes running within a zone potentially dump core in 3 locations,
69  * based on the per-process, per-zone, and the global zone's core settings.
70  *
71  * Per-zone and global zone settings are often referred to as "global"
72  * settings since they apply to the system (or zone) as a whole, as
73  * opposed to a particular process.
74  */
75 enum core_types {
76 	CORE_PROC,	/* Use per-process settings */
77 	CORE_ZONE,	/* Use per-zone settings */
78 	CORE_GLOBAL	/* Use global zone settings */
79 };
80 
81 /*
82  * Log information about "global" core dumps to syslog.
83  */
84 static void
85 core_log(struct core_globals *cg, int error, const char *why, const char *path,
86     zoneid_t zoneid)
87 {
88 	proc_t *p = curproc;
89 	pid_t pid = p->p_pid;
90 	char *fn = PTOU(p)->u_comm;
91 
92 	if (!(cg->core_options & CC_GLOBAL_LOG))
93 		return;
94 
95 	if (path == NULL)
96 		zcmn_err(zoneid, CE_NOTE, "core_log: %s[%d] %s", fn, pid, why);
97 	else if (error == 0)
98 		zcmn_err(zoneid, CE_NOTE, "core_log: %s[%d] %s: %s", fn, pid,
99 		    why, path);
100 	else
101 		zcmn_err(zoneid, CE_NOTE, "core_log: %s[%d] %s, errno=%d: %s",
102 		    fn, pid, why, error, path);
103 }
104 
105 /*
106  * Private version of vn_remove().
107  * Refuse to unlink a directory or an unwritable file.
108  * Also allow the process to access files normally inaccessible due to
109  * chroot(2) or Zone limitations.
110  */
111 static int
112 remove_core_file(char *fp, enum core_types core_type)
113 {
114 	vnode_t *vp = NULL;		/* entry vnode */
115 	vnode_t *dvp;			/* ptr to parent dir vnode */
116 	vfs_t *dvfsp;
117 	int error;
118 	int in_crit = 0;
119 	pathname_t pn;			/* name of entry */
120 	vnode_t *startvp, *rootvp;
121 
122 	if ((error = pn_get(fp, UIO_SYSSPACE, &pn)) != 0)
123 		return (error);
124 	/*
125 	 * Determine what rootvp to use.
126 	 */
127 	if (core_type == CORE_PROC) {
128 		rootvp = (PTOU(curproc)->u_rdir == NULL ?
129 		    curproc->p_zone->zone_rootvp : PTOU(curproc)->u_rdir);
130 		startvp = (fp[0] == '/' ? rootvp : PTOU(curproc)->u_cdir);
131 	} else if (core_type == CORE_ZONE) {
132 		startvp = curproc->p_zone->zone_rootvp;
133 		rootvp = curproc->p_zone->zone_rootvp;
134 	} else {
135 		ASSERT(core_type == CORE_GLOBAL);
136 		startvp = rootdir;
137 		rootvp = rootdir;
138 	}
139 	VN_HOLD(startvp);
140 	if (rootvp != rootdir)
141 		VN_HOLD(rootvp);
142 	if ((error = lookuppnvp(&pn, NULL, NO_FOLLOW, &dvp, &vp, rootvp,
143 	    startvp, CRED())) != 0) {
144 		pn_free(&pn);
145 		return (error);
146 	}
147 	/*
148 	 * Succeed if there is no file.
149 	 * Fail if the file is not a regular file.
150 	 * Fail if the filesystem is mounted read-only.
151 	 * Fail if the file is not writeable.
152 	 * Fail if the file has NBMAND share reservations.
153 	 */
154 	if (vp == NULL)
155 		error = 0;
156 	else if (vp->v_type != VREG)
157 		error = EACCES;
158 	else if ((dvfsp = dvp->v_vfsp) != NULL &&
159 	    (dvfsp->vfs_flag & VFS_RDONLY))
160 		error = EROFS;
161 	else if ((error = VOP_ACCESS(vp, VWRITE, 0, CRED(), NULL)) == 0) {
162 		if (nbl_need_check(vp)) {
163 			nbl_start_crit(vp, RW_READER);
164 			in_crit = 1;
165 			if (nbl_share_conflict(vp, NBL_REMOVE, NULL)) {
166 				error = EACCES;
167 			}
168 		}
169 		if (!error) {
170 			error = VOP_REMOVE(dvp, pn.pn_path, CRED(), NULL, 0);
171 		}
172 	}
173 
174 	pn_free(&pn);
175 	if (vp != NULL) {
176 		if (in_crit)
177 			nbl_end_crit(vp);
178 		VN_RELE(vp);
179 	}
180 	VN_RELE(dvp);
181 	return (error);
182 }
183 
184 /*
185  * Create the core file in a location that may be normally inaccessible due
186  * to chroot(2) or Zone limitations.
187  */
188 static int
189 create_core_file(char *fp, enum core_types core_type, vnode_t **vpp)
190 {
191 	int error;
192 	mode_t perms = (S_IRUSR | S_IWUSR);
193 	pathname_t pn;
194 	char *file;
195 	vnode_t *vp;
196 	vnode_t *dvp;
197 	vattr_t vattr;
198 	cred_t *credp = CRED();
199 
200 	if (core_type == CORE_PROC) {
201 		file = fp;
202 		dvp = NULL;	/* regular lookup */
203 	} else {
204 		vnode_t *startvp, *rootvp;
205 
206 		ASSERT(core_type == CORE_ZONE || core_type == CORE_GLOBAL);
207 		/*
208 		 * This is tricky because we want to dump the core in
209 		 * a location which may normally be inaccessible
210 		 * to us (due to chroot(2) limitations, or zone
211 		 * membership), and hence need to overcome u_rdir
212 		 * restrictions.  The basic idea is to separate
213 		 * the path from the filename, lookup the
214 		 * pathname separately (starting from the global
215 		 * zone's root directory), and then open the
216 		 * file starting at the directory vnode.
217 		 */
218 		if (error = pn_get(fp, UIO_SYSSPACE, &pn))
219 			return (error);
220 
221 		if (core_type == CORE_ZONE) {
222 			startvp = rootvp = curproc->p_zone->zone_rootvp;
223 		} else {
224 			startvp = rootvp = rootdir;
225 		}
226 		/*
227 		 * rootvp and startvp will be VN_RELE()'d by lookuppnvp() if
228 		 * necessary.
229 		 */
230 		VN_HOLD(startvp);
231 		if (rootvp != rootdir)
232 			VN_HOLD(rootvp);
233 		/*
234 		 * Do a lookup on the full path, ignoring the actual file, but
235 		 * finding the vnode for the directory.  It's OK if the file
236 		 * doesn't exist -- it most likely won't since we just removed
237 		 * it.
238 		 */
239 		error = lookuppnvp(&pn, NULL, FOLLOW, &dvp, NULLVPP,
240 		    rootvp, startvp, credp);
241 		pn_free(&pn);
242 		if (error != 0)
243 			return (error);
244 		ASSERT(dvp != NULL);
245 		/*
246 		 * Now find the final component in the path (ie, the name of
247 		 * the core file).
248 		 */
249 		if (error = pn_get(fp, UIO_SYSSPACE, &pn)) {
250 			VN_RELE(dvp);
251 			return (error);
252 		}
253 		pn_setlast(&pn);
254 		file = pn.pn_path;
255 	}
256 	error =  vn_openat(file, UIO_SYSSPACE,
257 	    FWRITE | FTRUNC | FEXCL | FCREAT | FOFFMAX,
258 	    perms, &vp, CRCREAT, PTOU(curproc)->u_cmask, dvp, -1);
259 	if (core_type != CORE_PROC) {
260 		VN_RELE(dvp);
261 		pn_free(&pn);
262 	}
263 	/*
264 	 * Don't dump a core file owned by "nobody".
265 	 */
266 	vattr.va_mask = AT_UID;
267 	if (error == 0 &&
268 	    (VOP_GETATTR(vp, &vattr, 0, credp, NULL) != 0 ||
269 	    vattr.va_uid != crgetuid(credp))) {
270 		(void) VOP_CLOSE(vp, FWRITE, 1, (offset_t)0,
271 		    credp, NULL);
272 		VN_RELE(vp);
273 		(void) remove_core_file(fp, core_type);
274 		error = EACCES;
275 	}
276 	*vpp = vp;
277 	return (error);
278 }
279 
280 /*
281  * Install the specified held cred into the process, and return a pointer to
282  * the held cred which was previously the value of p->p_cred.
283  */
284 static cred_t *
285 set_cred(proc_t *p, cred_t *newcr)
286 {
287 	cred_t *oldcr;
288 	uid_t olduid, newuid;
289 
290 	/*
291 	 * Place a hold on the existing cred, and then install the new
292 	 * cred into the proc structure.
293 	 */
294 	mutex_enter(&p->p_crlock);
295 	oldcr = p->p_cred;
296 	crhold(oldcr);
297 	p->p_cred = newcr;
298 	mutex_exit(&p->p_crlock);
299 
300 	ASSERT(crgetzoneid(oldcr) == crgetzoneid(newcr));
301 
302 	/*
303 	 * If the real uid is changing, keep the per-user process
304 	 * counts accurate.
305 	 */
306 	olduid = crgetruid(oldcr);
307 	newuid = crgetruid(newcr);
308 	if (olduid != newuid) {
309 		zoneid_t zoneid = crgetzoneid(newcr);
310 
311 		mutex_enter(&pidlock);
312 		upcount_dec(olduid, zoneid);
313 		upcount_inc(newuid, zoneid);
314 		mutex_exit(&pidlock);
315 	}
316 
317 	/*
318 	 * Broadcast the new cred to all the other threads.  The old
319 	 * cred can be safely returned because we have a hold on it.
320 	 */
321 	crset(p, newcr);
322 	return (oldcr);
323 }
324 
325 static int
326 do_core(char *fp, int sig, enum core_types core_type, struct core_globals *cg)
327 {
328 	proc_t *p = curproc;
329 	cred_t *credp = CRED();
330 	rlim64_t rlimit;
331 	vnode_t *vp;
332 	int error = 0;
333 	struct execsw *eswp;
334 	cred_t *ocredp = NULL;
335 	int is_setid = 0;
336 	core_content_t content;
337 	uid_t uid;
338 	gid_t gid;
339 
340 	if (core_type == CORE_GLOBAL || core_type == CORE_ZONE) {
341 		mutex_enter(&cg->core_lock);
342 		content = cg->core_content;
343 		mutex_exit(&cg->core_lock);
344 		rlimit = cg->core_rlimit;
345 	} else {
346 		mutex_enter(&p->p_lock);
347 		rlimit = rctl_enforced_value(rctlproc_legacy[RLIMIT_CORE],
348 		    p->p_rctls, p);
349 		content = corectl_content_value(p->p_content);
350 		mutex_exit(&p->p_lock);
351 	}
352 
353 	if (rlimit == 0)
354 		return (EFBIG);
355 
356 	/*
357 	 * If SNOCD is set, or if the effective, real, and saved ids do
358 	 * not match up, no one but a privileged user is allowed to view
359 	 * this core file.  Set the credentials and the owner to root.
360 	 */
361 	if ((p->p_flag & SNOCD) ||
362 	    (uid = crgetuid(credp)) != crgetruid(credp) ||
363 	    uid != crgetsuid(credp) ||
364 	    (gid = crgetgid(credp)) != crgetrgid(credp) ||
365 	    gid != crgetsgid(credp)) {
366 		/*
367 		 * Because this is insecure against certain forms of file
368 		 * system attack, do it only if set-id core files have been
369 		 * enabled via corectl(CC_GLOBAL_SETID | CC_PROCESS_SETID).
370 		 */
371 		if (((core_type == CORE_GLOBAL || core_type == CORE_ZONE) &&
372 		    !(cg->core_options & CC_GLOBAL_SETID)) ||
373 		    (core_type == CORE_PROC &&
374 		    !(cg->core_options & CC_PROCESS_SETID)))
375 			return (ENOTSUP);
376 
377 		is_setid = 1;
378 	}
379 
380 	/*
381 	 * If we are doing a "global" core dump or a set-id core dump,
382 	 * use kcred to do the dumping.
383 	 */
384 	if (core_type == CORE_GLOBAL || core_type == CORE_ZONE || is_setid) {
385 		/*
386 		 * Use the zone's "kcred" to prevent privilege
387 		 * escalation.
388 		 */
389 		credp = zone_get_kcred(getzoneid());
390 		ASSERT(credp != NULL);
391 		ocredp = set_cred(p, credp);
392 	}
393 
394 	/*
395 	 * First remove any existing core file, then
396 	 * open the new core file with (O_EXCL|O_CREAT).
397 	 *
398 	 * The reasons for doing this are manifold:
399 	 *
400 	 * For security reasons, we don't want root processes
401 	 * to dump core through a symlink because that would
402 	 * allow a malicious user to clobber any file on
403 	 * the system if s/he could convince a root process,
404 	 * perhaps a set-uid root process that s/he started,
405 	 * to dump core in a directory writable by that user.
406 	 * Similar security reasons apply to hard links.
407 	 * For symmetry we do this unconditionally, not
408 	 * just for root processes.
409 	 *
410 	 * If the process has the core file mmap()d into the
411 	 * address space, we would be modifying the address
412 	 * space that we are trying to dump if we did not first
413 	 * remove the core file.  (The command "file core"
414 	 * is the canonical example of this possibility.)
415 	 *
416 	 * Opening the core file with O_EXCL|O_CREAT ensures than
417 	 * two concurrent core dumps don't clobber each other.
418 	 * One is bound to lose; we don't want to make both lose.
419 	 */
420 	if ((error = remove_core_file(fp, core_type)) == 0) {
421 		error = create_core_file(fp, core_type, &vp);
422 	}
423 
424 	/*
425 	 * Now that vn_open is complete, reset the process's credentials if
426 	 * we changed them, and make 'credp' point to kcred used
427 	 * above.  We use 'credp' to do i/o on the core file below, but leave
428 	 * p->p_cred set to the original credential to allow the core file
429 	 * to record this information.
430 	 */
431 	if (ocredp != NULL)
432 		credp = set_cred(p, ocredp);
433 
434 	if (error == 0) {
435 		int closerr;
436 #if defined(__sparc)
437 		(void) flush_user_windows_to_stack(NULL);
438 #endif
439 		if ((eswp = PTOU(curproc)->u_execsw) == NULL ||
440 		    (eswp = findexec_by_magic(eswp->exec_magic)) == NULL) {
441 			error = ENOSYS;
442 		} else {
443 			error = eswp->exec_core(vp, p, credp, rlimit, sig,
444 			    content);
445 			rw_exit(eswp->exec_lock);
446 		}
447 
448 		closerr = VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, credp, NULL);
449 		VN_RELE(vp);
450 		if (error == 0)
451 			error = closerr;
452 	}
453 
454 	if (ocredp != NULL)
455 		crfree(credp);
456 
457 	return (error);
458 }
459 
460 /*
461  * Convert a core name pattern to a pathname.
462  */
463 static int
464 expand_string(const char *pat, char *fp, int size, cred_t *cr)
465 {
466 	proc_t *p = curproc;
467 	char buf[24];
468 	int len, i;
469 	char *s;
470 	char c;
471 
472 	while ((c = *pat++) != '\0') {
473 		if (size < 2)
474 			return (ENAMETOOLONG);
475 		if (c != '%') {
476 			size--;
477 			*fp++ = c;
478 			continue;
479 		}
480 		if ((c = *pat++) == '\0') {
481 			size--;
482 			*fp++ = '%';
483 			break;
484 		}
485 		switch (c) {
486 		case 'p':	/* pid */
487 			(void) sprintf((s = buf), "%d", p->p_pid);
488 			break;
489 		case 'u':	/* effective uid */
490 			(void) sprintf((s = buf), "%u", crgetuid(p->p_cred));
491 			break;
492 		case 'g':	/* effective gid */
493 			(void) sprintf((s = buf), "%u", crgetgid(p->p_cred));
494 			break;
495 		case 'f':	/* exec'd filename */
496 			s = PTOU(p)->u_comm;
497 			break;
498 		case 'd':	/* exec'd dirname */
499 			/*
500 			 * Even if pathname caching is disabled, we should
501 			 * be able to lookup the pathname for a directory.
502 			 */
503 			if (p->p_execdir != NULL && vnodetopath(NULL,
504 			    p->p_execdir, fp, size, cr) == 0) {
505 				len = (int)strlen(fp);
506 				ASSERT(len < size);
507 				ASSERT(len >= 1);
508 				ASSERT(fp[0] == '/');
509 
510 				/*
511 				 * Strip off the leading slash.
512 				 */
513 				for (i = 0; i < len; i++) {
514 					fp[i] = fp[i + 1];
515 				}
516 
517 				len--;
518 
519 				size -= len;
520 				fp += len;
521 			} else {
522 				*fp = '\0';
523 			}
524 
525 			continue;
526 		case 'n':	/* system nodename */
527 			s = uts_nodename();
528 			break;
529 		case 'm':	/* machine (sun4u, etc) */
530 			s = utsname.machine;
531 			break;
532 		case 't':	/* decimal value of time(2) */
533 			(void) sprintf((s = buf), "%ld", gethrestime_sec());
534 			break;
535 		case 'z':
536 			s = p->p_zone->zone_name;
537 			break;
538 		case 'Z':
539 			/* This is zonepath + "/root/", except for GZ */
540 			s = p->p_zone->zone_rootpath;
541 			break;
542 		case '%':
543 			(void) strcpy((s = buf), "%");
544 			break;
545 		default:
546 			s = buf;
547 			buf[0] = '%';
548 			buf[1] = c;
549 			buf[2] = '\0';
550 			break;
551 		}
552 		len = (int)strlen(s);
553 		if ((size -= len) <= 0)
554 			return (ENAMETOOLONG);
555 		(void) strcpy(fp, s);
556 		/* strip trailing "/root/" from non-GZ zonepath string */
557 		if (c == 'Z' && len > 6) {
558 			len -= 6;
559 			ASSERT(strncmp(fp + len, "/root/", 6) == 0);
560 		}
561 		fp += len;
562 	}
563 
564 	*fp = '\0';
565 	return (0);
566 }
567 
568 static int
569 dump_one_core(int sig, rlim64_t rlimit, enum core_types core_type,
570     struct core_globals *cg, char **name)
571 {
572 	refstr_t *rp;
573 	proc_t *p = curproc;
574 	zoneid_t zoneid;
575 	int error;
576 	char *fp;
577 	cred_t *cr;
578 
579 	ASSERT(core_type == CORE_ZONE || core_type == CORE_GLOBAL);
580 	zoneid = (core_type == CORE_ZONE ? getzoneid() : GLOBAL_ZONEID);
581 
582 	mutex_enter(&cg->core_lock);
583 	if ((rp = cg->core_file) != NULL)
584 		refstr_hold(rp);
585 	mutex_exit(&cg->core_lock);
586 	if (rp == NULL) {
587 		core_log(cg, 0, "no global core file pattern exists", NULL,
588 		    zoneid);
589 		return (1);	/* core file not generated */
590 	}
591 	fp = kmem_alloc(MAXPATHLEN, KM_SLEEP);
592 	cr = zone_get_kcred(getzoneid());
593 	error = expand_string(refstr_value(rp), fp, MAXPATHLEN, cr);
594 	crfree(cr);
595 	if (error != 0) {
596 		core_log(cg, 0, "global core file pattern too long",
597 		    refstr_value(rp), zoneid);
598 	} else if ((error = do_core(fp, sig, core_type, cg)) == 0) {
599 		core_log(cg, 0, "core dumped", fp, zoneid);
600 	} else if (error == ENOTSUP) {
601 		core_log(cg, 0, "setid process, core not dumped", fp, zoneid);
602 	} else if (error == ENOSPC) {
603 		core_log(cg, 0, "no space left on device, core truncated",
604 		    fp, zoneid);
605 	} else if (error == EFBIG) {
606 		if (rlimit == 0)
607 			core_log(cg, 0, "core rlimit is zero, core not dumped",
608 			    fp, zoneid);
609 		else
610 			core_log(cg, 0, "core rlimit exceeded, core truncated",
611 			    fp, zoneid);
612 		/*
613 		 * In addition to the core result logging, we
614 		 * may also have explicit actions defined on
615 		 * core file size violations via the resource
616 		 * control framework.
617 		 */
618 		mutex_enter(&p->p_lock);
619 		(void) rctl_action(rctlproc_legacy[RLIMIT_CORE],
620 		    p->p_rctls, p, RCA_SAFE);
621 		mutex_exit(&p->p_lock);
622 	} else {
623 		core_log(cg, error, "core dump failed", fp, zoneid);
624 	}
625 	refstr_rele(rp);
626 	if (name != NULL)
627 		*name = fp;
628 	else
629 		kmem_free(fp, MAXPATHLEN);
630 	return (error);
631 }
632 
633 int
634 core(int sig, int ext)
635 {
636 	proc_t *p = curproc;
637 	klwp_t *lwp = ttolwp(curthread);
638 	refstr_t *rp;
639 	char *fp_process = NULL, *fp_global = NULL, *fp_zone = NULL;
640 	int error1 = 1;
641 	int error2 = 1;
642 	int error3 = 1;
643 	k_sigset_t sigmask;
644 	k_sigset_t sighold;
645 	rlim64_t rlimit;
646 	struct core_globals *my_cg, *global_cg;
647 
648 	global_cg = zone_getspecific(core_zone_key, global_zone);
649 	ASSERT(global_cg != NULL);
650 
651 	my_cg = zone_getspecific(core_zone_key, curproc->p_zone);
652 	ASSERT(my_cg != NULL);
653 
654 	/* core files suppressed? */
655 	if (!(my_cg->core_options & (CC_PROCESS_PATH|CC_GLOBAL_PATH)) &&
656 	    !(global_cg->core_options & CC_GLOBAL_PATH)) {
657 		if (!ext && p->p_ct_process != NULL)
658 			contract_process_core(p->p_ct_process, p, sig,
659 			    NULL, NULL, NULL);
660 		return (1);
661 	}
662 
663 	/*
664 	 * Block all signals except SIGHUP, SIGINT, SIGKILL, and SIGTERM; no
665 	 * other signal may interrupt a core dump.  For each signal, we
666 	 * explicitly unblock it and set it in p_siginfo to allow for some
667 	 * minimal error reporting.  Additionally, we get the current limit on
668 	 * core file size for handling later error reporting.
669 	 */
670 	mutex_enter(&p->p_lock);
671 
672 	p->p_flag |= SDOCORE;
673 	schedctl_finish_sigblock(curthread);
674 	sigmask = curthread->t_hold;	/* remember for later */
675 	sigfillset(&sighold);
676 	if (!sigismember(&sigmask, SIGHUP))
677 		sigdelset(&sighold, SIGHUP);
678 	if (!sigismember(&sigmask, SIGINT))
679 		sigdelset(&sighold, SIGINT);
680 	if (!sigismember(&sigmask, SIGKILL))
681 		sigdelset(&sighold, SIGKILL);
682 	if (!sigismember(&sigmask, SIGTERM))
683 		sigdelset(&sighold, SIGTERM);
684 
685 	sigaddset(&p->p_siginfo, SIGHUP);
686 	sigaddset(&p->p_siginfo, SIGINT);
687 	sigaddset(&p->p_siginfo, SIGKILL);
688 	sigaddset(&p->p_siginfo, SIGTERM);
689 
690 	curthread->t_hold = sighold;
691 
692 	rlimit = rctl_enforced_value(rctlproc_legacy[RLIMIT_CORE], p->p_rctls,
693 	    p);
694 
695 	mutex_exit(&p->p_lock);
696 
697 	/*
698 	 * Undo any watchpoints.
699 	 */
700 	pr_free_watched_pages(p);
701 
702 	/*
703 	 * The presence of a current signal prevents file i/o
704 	 * from succeeding over a network.  We copy the current
705 	 * signal information to the side and cancel the current
706 	 * signal so that the core dump will succeed.
707 	 */
708 	ASSERT(lwp->lwp_cursig == sig);
709 	lwp->lwp_cursig = 0;
710 	lwp->lwp_extsig = 0;
711 	if (lwp->lwp_curinfo == NULL) {
712 		bzero(&lwp->lwp_siginfo, sizeof (k_siginfo_t));
713 		lwp->lwp_siginfo.si_signo = sig;
714 		lwp->lwp_siginfo.si_code = SI_NOINFO;
715 	} else {
716 		bcopy(&lwp->lwp_curinfo->sq_info,
717 		    &lwp->lwp_siginfo, sizeof (k_siginfo_t));
718 		siginfofree(lwp->lwp_curinfo);
719 		lwp->lwp_curinfo = NULL;
720 	}
721 
722 	/*
723 	 * Convert the core file name patterns into path names
724 	 * and call do_core() to write the core files.
725 	 */
726 
727 	if (my_cg->core_options & CC_PROCESS_PATH) {
728 		mutex_enter(&p->p_lock);
729 		if (p->p_corefile != NULL)
730 			rp = corectl_path_value(p->p_corefile);
731 		else
732 			rp = NULL;
733 		mutex_exit(&p->p_lock);
734 		if (rp != NULL) {
735 			fp_process = kmem_alloc(MAXPATHLEN, KM_SLEEP);
736 			error1 = expand_string(refstr_value(rp),
737 			    fp_process, MAXPATHLEN, p->p_cred);
738 			if (error1 == 0)
739 				error1 = do_core(fp_process, sig, CORE_PROC,
740 				    my_cg);
741 			refstr_rele(rp);
742 		}
743 	}
744 
745 	if (my_cg->core_options & CC_GLOBAL_PATH)
746 		error2 = dump_one_core(sig, rlimit, CORE_ZONE, my_cg,
747 		    &fp_global);
748 	if (global_cg != my_cg && (global_cg->core_options & CC_GLOBAL_PATH))
749 		error3 = dump_one_core(sig, rlimit, CORE_GLOBAL, global_cg,
750 		    &fp_zone);
751 
752 	/*
753 	 * Restore the signal hold mask.
754 	 */
755 	mutex_enter(&p->p_lock);
756 	curthread->t_hold = sigmask;
757 	mutex_exit(&p->p_lock);
758 
759 	if (!ext && p->p_ct_process != NULL)
760 		contract_process_core(p->p_ct_process, p, sig,
761 		    error1 == 0 ? fp_process : NULL,
762 		    error2 == 0 ? fp_global : NULL,
763 		    error3 == 0 ? fp_zone : NULL);
764 
765 	if (fp_process != NULL)
766 		kmem_free(fp_process, MAXPATHLEN);
767 	if (fp_global != NULL)
768 		kmem_free(fp_global, MAXPATHLEN);
769 	if (fp_zone != NULL)
770 		kmem_free(fp_zone, MAXPATHLEN);
771 
772 	/*
773 	 * Return non-zero if no core file was created.
774 	 */
775 	return (error1 != 0 && error2 != 0 && error3 != 0);
776 }
777 
778 /*
779  * Maximum chunk size for dumping core files,
780  * size in pages, patchable in /etc/system
781  */
782 uint_t	core_chunk = 32;
783 
784 /*
785  * The delay between core_write() calls, in microseconds.  The default
786  * matches one "normal" clock tick, or 10 milliseconds.
787  */
788 clock_t	core_delay_usec = 10000;
789 
790 /*
791  * Common code to core dump process memory.  The core_seg routine does i/o
792  * using core_write() below, and so it has the same failure semantics.
793  */
794 int
795 core_seg(proc_t *p, vnode_t *vp, offset_t offset, caddr_t addr, size_t size,
796     rlim64_t rlimit, cred_t *credp)
797 {
798 	caddr_t eaddr;
799 	caddr_t base;
800 	size_t len;
801 	int err = 0;
802 
803 	eaddr = addr + size;
804 	for (base = addr; base < eaddr; base += len) {
805 		len = eaddr - base;
806 		if (as_memory(p->p_as, &base, &len) != 0)
807 			return (0);
808 
809 		/*
810 		 * Reduce len to a reasonable value so that we don't
811 		 * overwhelm the VM system with a monstrously large
812 		 * single write and cause pageout to stop running.
813 		 */
814 		if (len > (size_t)core_chunk * PAGESIZE)
815 			len = (size_t)core_chunk * PAGESIZE;
816 
817 		err = core_write(vp, UIO_USERSPACE,
818 		    offset + (size_t)(base - addr), base, len, rlimit, credp);
819 
820 		if (err)
821 			return (err);
822 
823 		/*
824 		 * If we have taken a signal, return EINTR to allow the dump
825 		 * to be aborted.
826 		 */
827 		if (issig(JUSTLOOKING) && issig(FORREAL))
828 			return (EINTR);
829 	}
830 
831 	return (0);
832 }
833 
834 /*
835  * Wrapper around vn_rdwr to perform writes to a core file.  For core files,
836  * we always want to write as much as we possibly can, and then make sure to
837  * return either 0 to the caller (for success), or the actual errno value.
838  * By using this function, the caller can omit additional code for handling
839  * retries and errors for partial writes returned by vn_rdwr.  If vn_rdwr
840  * unexpectedly returns zero but no progress has been made, we return ENOSPC.
841  */
842 int
843 core_write(vnode_t *vp, enum uio_seg segflg, offset_t offset,
844     const void *buf, size_t len, rlim64_t rlimit, cred_t *credp)
845 {
846 	ssize_t resid = len;
847 	int error = 0;
848 
849 	while (len != 0) {
850 		error = vn_rdwr(UIO_WRITE, vp, (caddr_t)buf, len, offset,
851 		    segflg, 0, rlimit, credp, &resid);
852 
853 		if (error != 0)
854 			break;
855 
856 		if (resid >= len)
857 			return (ENOSPC);
858 
859 		buf = (const char *)buf + len - resid;
860 		offset += len - resid;
861 		len = resid;
862 	}
863 
864 	return (error);
865 }
866