xref: /illumos-gate/usr/src/uts/common/os/core.c (revision 657a8c206b913d1ee578fd725f0b25eca5b77253)
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 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
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 #ifdef SUN_SRC_COMPAT
440 		PTOU(curproc)->u_acflag |= ACORE;
441 #endif
442 		if ((eswp = PTOU(curproc)->u_execsw) == NULL ||
443 		    (eswp = findexec_by_magic(eswp->exec_magic)) == NULL) {
444 			error = ENOSYS;
445 		} else {
446 			error = eswp->exec_core(vp, p, credp, rlimit, sig,
447 			    content);
448 			rw_exit(eswp->exec_lock);
449 		}
450 
451 		closerr = VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, credp, NULL);
452 		VN_RELE(vp);
453 		if (error == 0)
454 			error = closerr;
455 	}
456 
457 	if (ocredp != NULL)
458 		crfree(credp);
459 
460 	return (error);
461 }
462 
463 /*
464  * Convert a core name pattern to a pathname.
465  */
466 static int
467 expand_string(const char *pat, char *fp, int size, cred_t *cr)
468 {
469 	proc_t *p = curproc;
470 	char buf[24];
471 	int len, i;
472 	char *s;
473 	char c;
474 
475 	while ((c = *pat++) != '\0') {
476 		if (size < 2)
477 			return (ENAMETOOLONG);
478 		if (c != '%') {
479 			size--;
480 			*fp++ = c;
481 			continue;
482 		}
483 		if ((c = *pat++) == '\0') {
484 			size--;
485 			*fp++ = '%';
486 			break;
487 		}
488 		switch (c) {
489 		case 'p':	/* pid */
490 			(void) sprintf((s = buf), "%d", p->p_pid);
491 			break;
492 		case 'u':	/* effective uid */
493 			(void) sprintf((s = buf), "%u", crgetuid(p->p_cred));
494 			break;
495 		case 'g':	/* effective gid */
496 			(void) sprintf((s = buf), "%u", crgetgid(p->p_cred));
497 			break;
498 		case 'f':	/* exec'd filename */
499 			s = PTOU(p)->u_comm;
500 			break;
501 		case 'd':	/* exec'd dirname */
502 			/*
503 			 * Even if pathname caching is disabled, we should
504 			 * be able to lookup the pathname for a directory.
505 			 */
506 			if (p->p_execdir != NULL && vnodetopath(NULL,
507 			    p->p_execdir, fp, size, cr) == 0) {
508 				len = (int)strlen(fp);
509 				ASSERT(len < size);
510 				ASSERT(len >= 1);
511 				ASSERT(fp[0] == '/');
512 
513 				/*
514 				 * Strip off the leading slash.
515 				 */
516 				for (i = 0; i < len; i++) {
517 					fp[i] = fp[i + 1];
518 				}
519 
520 				len--;
521 
522 				size -= len;
523 				fp += len;
524 			} else {
525 				*fp = '\0';
526 			}
527 
528 			continue;
529 		case 'n':	/* system nodename */
530 			s = uts_nodename();
531 			break;
532 		case 'm':	/* machine (sun4u, etc) */
533 			s = utsname.machine;
534 			break;
535 		case 't':	/* decimal value of time(2) */
536 			(void) sprintf((s = buf), "%ld", gethrestime_sec());
537 			break;
538 		case 'z':
539 			s = p->p_zone->zone_name;
540 			break;
541 		case '%':
542 			(void) strcpy((s = buf), "%");
543 			break;
544 		default:
545 			s = buf;
546 			buf[0] = '%';
547 			buf[1] = c;
548 			buf[2] = '\0';
549 			break;
550 		}
551 		len = (int)strlen(s);
552 		if ((size -= len) <= 0)
553 			return (ENAMETOOLONG);
554 		(void) strcpy(fp, s);
555 		fp += len;
556 	}
557 
558 	*fp = '\0';
559 	return (0);
560 }
561 
562 static int
563 dump_one_core(int sig, rlim64_t rlimit, enum core_types core_type,
564     struct core_globals *cg, char **name)
565 {
566 	refstr_t *rp;
567 	proc_t *p = curproc;
568 	zoneid_t zoneid;
569 	int error;
570 	char *fp;
571 	cred_t *cr;
572 
573 	ASSERT(core_type == CORE_ZONE || core_type == CORE_GLOBAL);
574 	zoneid = (core_type == CORE_ZONE ? getzoneid() : GLOBAL_ZONEID);
575 
576 	mutex_enter(&cg->core_lock);
577 	if ((rp = cg->core_file) != NULL)
578 		refstr_hold(rp);
579 	mutex_exit(&cg->core_lock);
580 	if (rp == NULL) {
581 		core_log(cg, 0, "no global core file pattern exists", NULL,
582 		    zoneid);
583 		return (1);	/* core file not generated */
584 	}
585 	fp = kmem_alloc(MAXPATHLEN, KM_SLEEP);
586 	cr = zone_get_kcred(getzoneid());
587 	error = expand_string(refstr_value(rp), fp, MAXPATHLEN, cr);
588 	crfree(cr);
589 	if (error != 0) {
590 		core_log(cg, 0, "global core file pattern too long",
591 		    refstr_value(rp), zoneid);
592 	} else if ((error = do_core(fp, sig, core_type, cg)) == 0) {
593 		core_log(cg, 0, "core dumped", fp, zoneid);
594 	} else if (error == ENOTSUP) {
595 		core_log(cg, 0, "setid process, core not dumped", fp, zoneid);
596 	} else if (error == ENOSPC) {
597 		core_log(cg, 0, "no space left on device, core truncated",
598 		    fp, zoneid);
599 	} else if (error == EFBIG) {
600 		if (rlimit == 0)
601 			core_log(cg, 0, "core rlimit is zero, core not dumped",
602 			    fp, zoneid);
603 		else
604 			core_log(cg, 0, "core rlimit exceeded, core truncated",
605 			    fp, zoneid);
606 		/*
607 		 * In addition to the core result logging, we
608 		 * may also have explicit actions defined on
609 		 * core file size violations via the resource
610 		 * control framework.
611 		 */
612 		mutex_enter(&p->p_lock);
613 		(void) rctl_action(rctlproc_legacy[RLIMIT_CORE],
614 		    p->p_rctls, p, RCA_SAFE);
615 		mutex_exit(&p->p_lock);
616 	} else {
617 		core_log(cg, error, "core dump failed", fp, zoneid);
618 	}
619 	refstr_rele(rp);
620 	if (name != NULL)
621 		*name = fp;
622 	else
623 		kmem_free(fp, MAXPATHLEN);
624 	return (error);
625 }
626 
627 int
628 core(int sig, int ext)
629 {
630 	proc_t *p = curproc;
631 	klwp_t *lwp = ttolwp(curthread);
632 	refstr_t *rp;
633 	char *fp_process = NULL, *fp_global = NULL, *fp_zone = NULL;
634 	int error1 = 1;
635 	int error2 = 1;
636 	int error3 = 1;
637 	k_sigset_t sigmask;
638 	k_sigset_t sighold;
639 	rlim64_t rlimit;
640 	struct core_globals *my_cg, *global_cg;
641 
642 	global_cg = zone_getspecific(core_zone_key, global_zone);
643 	ASSERT(global_cg != NULL);
644 
645 	my_cg = zone_getspecific(core_zone_key, curproc->p_zone);
646 	ASSERT(my_cg != NULL);
647 
648 	/* core files suppressed? */
649 	if (!(my_cg->core_options & (CC_PROCESS_PATH|CC_GLOBAL_PATH)) &&
650 	    !(global_cg->core_options & CC_GLOBAL_PATH)) {
651 		if (!ext && p->p_ct_process != NULL)
652 			contract_process_core(p->p_ct_process, p, sig,
653 			    NULL, NULL, NULL);
654 		return (1);
655 	}
656 
657 	/*
658 	 * Block all signals except SIGHUP, SIGINT, SIGKILL, and SIGTERM.
659 	 * These signals are allowed to interrupt the core dump.
660 	 * SIGQUIT is not allowed because it is supposed to make a core.
661 	 * Additionally, get current limit on core file size for handling later
662 	 * error reporting.
663 	 */
664 	mutex_enter(&p->p_lock);
665 
666 	p->p_flag |= SDOCORE;
667 	schedctl_finish_sigblock(curthread);
668 	sigmask = curthread->t_hold;	/* remember for later */
669 	sigfillset(&sighold);
670 	if (!sigismember(&sigmask, SIGHUP))
671 		sigdelset(&sighold, SIGHUP);
672 	if (!sigismember(&sigmask, SIGINT))
673 		sigdelset(&sighold, SIGINT);
674 	if (!sigismember(&sigmask, SIGKILL))
675 		sigdelset(&sighold, SIGKILL);
676 	if (!sigismember(&sigmask, SIGTERM))
677 		sigdelset(&sighold, SIGTERM);
678 	curthread->t_hold = sighold;
679 
680 	rlimit = rctl_enforced_value(rctlproc_legacy[RLIMIT_CORE], p->p_rctls,
681 	    p);
682 
683 	mutex_exit(&p->p_lock);
684 
685 	/*
686 	 * Undo any watchpoints.
687 	 */
688 	pr_free_watched_pages(p);
689 
690 	/*
691 	 * The presence of a current signal prevents file i/o
692 	 * from succeeding over a network.  We copy the current
693 	 * signal information to the side and cancel the current
694 	 * signal so that the core dump will succeed.
695 	 */
696 	ASSERT(lwp->lwp_cursig == sig);
697 	lwp->lwp_cursig = 0;
698 	lwp->lwp_extsig = 0;
699 	if (lwp->lwp_curinfo == NULL) {
700 		bzero(&lwp->lwp_siginfo, sizeof (k_siginfo_t));
701 		lwp->lwp_siginfo.si_signo = sig;
702 		lwp->lwp_siginfo.si_code = SI_NOINFO;
703 	} else {
704 		bcopy(&lwp->lwp_curinfo->sq_info,
705 		    &lwp->lwp_siginfo, sizeof (k_siginfo_t));
706 		siginfofree(lwp->lwp_curinfo);
707 		lwp->lwp_curinfo = NULL;
708 	}
709 
710 	/*
711 	 * Convert the core file name patterns into path names
712 	 * and call do_core() to write the core files.
713 	 */
714 
715 	if (my_cg->core_options & CC_PROCESS_PATH) {
716 		mutex_enter(&p->p_lock);
717 		if (p->p_corefile != NULL)
718 			rp = corectl_path_value(p->p_corefile);
719 		else
720 			rp = NULL;
721 		mutex_exit(&p->p_lock);
722 		if (rp != NULL) {
723 			fp_process = kmem_alloc(MAXPATHLEN, KM_SLEEP);
724 			error1 = expand_string(refstr_value(rp),
725 			    fp_process, MAXPATHLEN, p->p_cred);
726 			if (error1 == 0)
727 				error1 = do_core(fp_process, sig, CORE_PROC,
728 				    my_cg);
729 			refstr_rele(rp);
730 		}
731 	}
732 
733 	if (my_cg->core_options & CC_GLOBAL_PATH)
734 		error2 = dump_one_core(sig, rlimit, CORE_ZONE, my_cg,
735 		    &fp_global);
736 	if (global_cg != my_cg && (global_cg->core_options & CC_GLOBAL_PATH))
737 		error3 = dump_one_core(sig, rlimit, CORE_GLOBAL, global_cg,
738 		    &fp_zone);
739 
740 	/*
741 	 * Restore the signal hold mask.
742 	 */
743 	mutex_enter(&p->p_lock);
744 	curthread->t_hold = sigmask;
745 	mutex_exit(&p->p_lock);
746 
747 	if (!ext && p->p_ct_process != NULL)
748 		contract_process_core(p->p_ct_process, p, sig,
749 		    error1 == 0 ? fp_process : NULL,
750 		    error2 == 0 ? fp_global : NULL,
751 		    error3 == 0 ? fp_zone : NULL);
752 
753 	if (fp_process != NULL)
754 		kmem_free(fp_process, MAXPATHLEN);
755 	if (fp_global != NULL)
756 		kmem_free(fp_global, MAXPATHLEN);
757 	if (fp_zone != NULL)
758 		kmem_free(fp_zone, MAXPATHLEN);
759 
760 	/*
761 	 * Return non-zero if no core file was created.
762 	 */
763 	return (error1 != 0 && error2 != 0 && error3 != 0);
764 }
765 
766 /*
767  * Maximum chunk size for dumping core files,
768  * size in pages, patchable in /etc/system
769  */
770 uint_t	core_chunk = 32;
771 
772 /*
773  * The delay between core_write() calls, in microseconds.  The default
774  * matches one "normal" clock tick, or 10 milliseconds.
775  */
776 clock_t	core_delay_usec = 10000;
777 
778 /*
779  * Common code to core dump process memory.  The core_seg routine does i/o
780  * using core_write() below, and so it has the same failure semantics.
781  */
782 int
783 core_seg(proc_t *p, vnode_t *vp, offset_t offset, caddr_t addr, size_t size,
784     rlim64_t rlimit, cred_t *credp)
785 {
786 	caddr_t eaddr;
787 	caddr_t base;
788 	size_t len;
789 	int err = 0;
790 
791 	eaddr = addr + size;
792 	for (base = addr; base < eaddr; base += len) {
793 		len = eaddr - base;
794 		if (as_memory(p->p_as, &base, &len) != 0)
795 			return (0);
796 		/*
797 		 * Reduce len to a reasonable value so that we don't
798 		 * overwhelm the VM system with a monstrously large
799 		 * single write and cause pageout to stop running.
800 		 */
801 		if (len > (size_t)core_chunk * PAGESIZE)
802 			len = (size_t)core_chunk * PAGESIZE;
803 
804 		err = core_write(vp, UIO_USERSPACE,
805 		    offset + (size_t)(base - addr), base, len, rlimit, credp);
806 
807 		if (err == 0) {
808 			/*
809 			 * Give pageout a chance to run.
810 			 * Also allow core dumping to be interruptible.
811 			 */
812 			err = delay_sig(drv_usectohz(core_delay_usec));
813 		}
814 		if (err)
815 			return (err);
816 	}
817 	return (0);
818 }
819 
820 /*
821  * Wrapper around vn_rdwr to perform writes to a core file.  For core files,
822  * we always want to write as much as we possibly can, and then make sure to
823  * return either 0 to the caller (for success), or the actual errno value.
824  * By using this function, the caller can omit additional code for handling
825  * retries and errors for partial writes returned by vn_rdwr.  If vn_rdwr
826  * unexpectedly returns zero but no progress has been made, we return ENOSPC.
827  */
828 int
829 core_write(vnode_t *vp, enum uio_seg segflg, offset_t offset,
830     const void *buf, size_t len, rlim64_t rlimit, cred_t *credp)
831 {
832 	ssize_t resid = len;
833 	int error = 0;
834 
835 	while (len != 0) {
836 		error = vn_rdwr(UIO_WRITE, vp, (caddr_t)buf, len, offset,
837 		    segflg, 0, rlimit, credp, &resid);
838 
839 		if (error != 0)
840 			break;
841 
842 		if (resid >= len)
843 			return (ENOSPC);
844 
845 		buf = (const char *)buf + len - resid;
846 		offset += len - resid;
847 		len = resid;
848 	}
849 
850 	return (error);
851 }
852