xref: /illumos-gate/usr/src/uts/common/os/core.c (revision 4e18e297380f99d39327599157792494d044942a)
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) 2016 by Delphix. All rights reserved.
25  * Copyright 2019 Joyent Inc.
26  */
27 
28 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
29 /*	  All Rights Reserved	*/
30 
31 #include <sys/param.h>
32 #include <sys/types.h>
33 #include <sys/time.h>
34 #include <sys/sysmacros.h>
35 #include <sys/proc.h>
36 #include <sys/systm.h>
37 #include <sys/cred.h>
38 #include <sys/user.h>
39 #include <sys/utsname.h>
40 #include <sys/errno.h>
41 #include <sys/signal.h>
42 #include <sys/siginfo.h>
43 #include <sys/fault.h>
44 #include <sys/syscall.h>
45 #include <sys/ucontext.h>
46 #include <sys/prsystm.h>
47 #include <sys/vnode.h>
48 #include <sys/var.h>
49 #include <sys/file.h>
50 #include <sys/pathname.h>
51 #include <sys/vfs.h>
52 #include <sys/exec.h>
53 #include <sys/debug.h>
54 #include <sys/stack.h>
55 #include <sys/kmem.h>
56 #include <sys/schedctl.h>
57 #include <sys/core.h>
58 #include <sys/corectl.h>
59 #include <sys/cmn_err.h>
60 #include <vm/as.h>
61 #include <sys/rctl.h>
62 #include <sys/nbmlock.h>
63 #include <sys/stat.h>
64 #include <sys/zone.h>
65 #include <sys/contract/process_impl.h>
66 #include <sys/ddi.h>
67 
68 /*
69  * Processes running within a zone potentially dump core in 3 locations,
70  * based on the per-process, per-zone, and the global zone's core settings.
71  *
72  * Per-zone and global zone settings are often referred to as "global"
73  * settings since they apply to the system (or zone) as a whole, as
74  * opposed to a particular process.
75  */
76 enum core_types {
77 	CORE_PROC,	/* Use per-process settings */
78 	CORE_ZONE,	/* Use per-zone settings */
79 	CORE_GLOBAL	/* Use global zone settings */
80 };
81 
82 /*
83  * Log information about "global" core dumps to syslog.
84  */
85 static void
core_log(struct core_globals * cg,int error,const char * why,const char * path,zoneid_t zoneid)86 core_log(struct core_globals *cg, int error, const char *why, const char *path,
87     zoneid_t zoneid)
88 {
89 	proc_t *p = curproc;
90 	pid_t pid = p->p_pid;
91 	char *fn = PTOU(p)->u_comm;
92 
93 	if (!(cg->core_options & CC_GLOBAL_LOG))
94 		return;
95 
96 	if (path == NULL)
97 		zcmn_err(zoneid, CE_NOTE, "core_log: %s[%d] %s", fn, pid, why);
98 	else if (error == 0)
99 		zcmn_err(zoneid, CE_NOTE, "core_log: %s[%d] %s: %s", fn, pid,
100 		    why, path);
101 	else
102 		zcmn_err(zoneid, CE_NOTE, "core_log: %s[%d] %s, errno=%d: %s",
103 		    fn, pid, why, error, path);
104 }
105 
106 /*
107  * Private version of vn_remove().
108  * Refuse to unlink a directory or an unwritable file.
109  * Also allow the process to access files normally inaccessible due to
110  * chroot(2) or Zone limitations.
111  */
112 static int
remove_core_file(char * fp,enum core_types core_type)113 remove_core_file(char *fp, enum core_types core_type)
114 {
115 	vnode_t *vp = NULL;		/* entry vnode */
116 	vnode_t *dvp;			/* ptr to parent dir vnode */
117 	vfs_t *dvfsp;
118 	int error;
119 	int in_crit = 0;
120 	pathname_t pn;			/* name of entry */
121 	vnode_t *startvp, *rootvp;
122 
123 	if ((error = pn_get(fp, UIO_SYSSPACE, &pn)) != 0)
124 		return (error);
125 	/*
126 	 * Determine what rootvp to use.
127 	 */
128 	if (core_type == CORE_PROC) {
129 		rootvp = (PTOU(curproc)->u_rdir == NULL ?
130 		    curproc->p_zone->zone_rootvp : PTOU(curproc)->u_rdir);
131 		startvp = (fp[0] == '/' ? rootvp : PTOU(curproc)->u_cdir);
132 	} else if (core_type == CORE_ZONE) {
133 		startvp = curproc->p_zone->zone_rootvp;
134 		rootvp = curproc->p_zone->zone_rootvp;
135 	} else {
136 		ASSERT(core_type == CORE_GLOBAL);
137 		startvp = rootdir;
138 		rootvp = rootdir;
139 	}
140 	VN_HOLD(startvp);
141 	if (rootvp != rootdir)
142 		VN_HOLD(rootvp);
143 	if ((error = lookuppnvp(&pn, NULL, NO_FOLLOW, &dvp, &vp, rootvp,
144 	    startvp, CRED())) != 0) {
145 		pn_free(&pn);
146 		return (error);
147 	}
148 	/*
149 	 * Succeed if there is no file.
150 	 * Fail if the file is not a regular file.
151 	 * Fail if the filesystem is mounted read-only.
152 	 * Fail if the file is not writeable.
153 	 * Fail if the file has NBMAND share reservations.
154 	 */
155 	if (vp == NULL)
156 		error = 0;
157 	else if (vp->v_type != VREG)
158 		error = EACCES;
159 	else if ((dvfsp = dvp->v_vfsp) != NULL &&
160 	    (dvfsp->vfs_flag & VFS_RDONLY))
161 		error = EROFS;
162 	else if ((error = VOP_ACCESS(vp, VWRITE, 0, CRED(), NULL)) == 0) {
163 		if (nbl_need_check(vp)) {
164 			nbl_start_crit(vp, RW_READER);
165 			in_crit = 1;
166 			if (nbl_share_conflict(vp, NBL_REMOVE, NULL)) {
167 				error = EACCES;
168 			}
169 		}
170 		if (!error) {
171 			error = VOP_REMOVE(dvp, pn.pn_path, CRED(), NULL, 0);
172 		}
173 	}
174 
175 	pn_free(&pn);
176 	if (vp != NULL) {
177 		if (in_crit)
178 			nbl_end_crit(vp);
179 		VN_RELE(vp);
180 	}
181 	VN_RELE(dvp);
182 	return (error);
183 }
184 
185 /*
186  * Create the core file in a location that may be normally inaccessible due
187  * to chroot(2) or Zone limitations.
188  */
189 static int
create_core_file(char * fp,enum core_types core_type,vnode_t ** vpp)190 create_core_file(char *fp, enum core_types core_type, vnode_t **vpp)
191 {
192 	int error;
193 	mode_t perms = (S_IRUSR | S_IWUSR);
194 	pathname_t pn;
195 	char *file;
196 	vnode_t *vp;
197 	vnode_t *dvp;
198 	vattr_t vattr;
199 	cred_t *credp = CRED();
200 
201 	if (core_type == CORE_PROC) {
202 		file = fp;
203 		dvp = NULL;	/* regular lookup */
204 	} else {
205 		vnode_t *startvp, *rootvp;
206 
207 		ASSERT(core_type == CORE_ZONE || core_type == CORE_GLOBAL);
208 		/*
209 		 * This is tricky because we want to dump the core in
210 		 * a location which may normally be inaccessible
211 		 * to us (due to chroot(2) limitations, or zone
212 		 * membership), and hence need to overcome u_rdir
213 		 * restrictions.  The basic idea is to separate
214 		 * the path from the filename, lookup the
215 		 * pathname separately (starting from the global
216 		 * zone's root directory), and then open the
217 		 * file starting at the directory vnode.
218 		 */
219 		if (error = pn_get(fp, UIO_SYSSPACE, &pn))
220 			return (error);
221 
222 		if (core_type == CORE_ZONE) {
223 			startvp = rootvp = curproc->p_zone->zone_rootvp;
224 		} else {
225 			startvp = rootvp = rootdir;
226 		}
227 		/*
228 		 * rootvp and startvp will be VN_RELE()'d by lookuppnvp() if
229 		 * necessary.
230 		 */
231 		VN_HOLD(startvp);
232 		if (rootvp != rootdir)
233 			VN_HOLD(rootvp);
234 		/*
235 		 * Do a lookup on the full path, ignoring the actual file, but
236 		 * finding the vnode for the directory.  It's OK if the file
237 		 * doesn't exist -- it most likely won't since we just removed
238 		 * it.
239 		 */
240 		error = lookuppnvp(&pn, NULL, FOLLOW, &dvp, NULLVPP,
241 		    rootvp, startvp, credp);
242 		pn_free(&pn);
243 		if (error != 0)
244 			return (error);
245 		ASSERT(dvp != NULL);
246 		/*
247 		 * Now find the final component in the path (ie, the name of
248 		 * the core file).
249 		 */
250 		if (error = pn_get(fp, UIO_SYSSPACE, &pn)) {
251 			VN_RELE(dvp);
252 			return (error);
253 		}
254 		pn_setlast(&pn);
255 		file = pn.pn_path;
256 	}
257 	error =  vn_openat(file, UIO_SYSSPACE,
258 	    FWRITE | FTRUNC | FEXCL | FCREAT | FOFFMAX,
259 	    perms, &vp, CRCREAT, PTOU(curproc)->u_cmask, dvp, -1);
260 	if (core_type != CORE_PROC) {
261 		VN_RELE(dvp);
262 		pn_free(&pn);
263 	}
264 	/*
265 	 * Don't dump a core file owned by "nobody".
266 	 */
267 	vattr.va_mask = AT_UID;
268 	if (error == 0 &&
269 	    (VOP_GETATTR(vp, &vattr, 0, credp, NULL) != 0 ||
270 	    vattr.va_uid != crgetuid(credp))) {
271 		(void) VOP_CLOSE(vp, FWRITE, 1, (offset_t)0,
272 		    credp, NULL);
273 		VN_RELE(vp);
274 		(void) remove_core_file(fp, core_type);
275 		error = EACCES;
276 	}
277 	*vpp = vp;
278 	return (error);
279 }
280 
281 /*
282  * Install the specified held cred into the process, and return a pointer to
283  * the held cred which was previously the value of p->p_cred.
284  */
285 static cred_t *
set_cred(proc_t * p,cred_t * newcr)286 set_cred(proc_t *p, cred_t *newcr)
287 {
288 	cred_t *oldcr;
289 	uid_t olduid, newuid;
290 
291 	/*
292 	 * Place a hold on the existing cred, and then install the new
293 	 * cred into the proc structure.
294 	 */
295 	mutex_enter(&p->p_crlock);
296 	oldcr = p->p_cred;
297 	crhold(oldcr);
298 	p->p_cred = newcr;
299 	mutex_exit(&p->p_crlock);
300 
301 	ASSERT(crgetzoneid(oldcr) == crgetzoneid(newcr));
302 
303 	/*
304 	 * If the real uid is changing, keep the per-user process
305 	 * counts accurate.
306 	 */
307 	olduid = crgetruid(oldcr);
308 	newuid = crgetruid(newcr);
309 	if (olduid != newuid) {
310 		zoneid_t zoneid = crgetzoneid(newcr);
311 
312 		mutex_enter(&pidlock);
313 		upcount_dec(olduid, zoneid);
314 		upcount_inc(newuid, zoneid);
315 		mutex_exit(&pidlock);
316 	}
317 
318 	/*
319 	 * Broadcast the new cred to all the other threads.  The old
320 	 * cred can be safely returned because we have a hold on it.
321 	 */
322 	crset(p, newcr);
323 	return (oldcr);
324 }
325 
326 static int
do_core(char * fp,int sig,enum core_types core_type,struct core_globals * cg)327 do_core(char *fp, int sig, enum core_types core_type, struct core_globals *cg)
328 {
329 	proc_t *p = curproc;
330 	cred_t *credp = CRED();
331 	rlim64_t rlimit;
332 	vnode_t *vp;
333 	int error = 0;
334 	struct execsw *eswp;
335 	cred_t *ocredp = NULL;
336 	int is_setid = 0;
337 	core_content_t content;
338 	uid_t uid;
339 	gid_t gid;
340 
341 	if (core_type == CORE_GLOBAL || core_type == CORE_ZONE) {
342 		mutex_enter(&cg->core_lock);
343 		content = cg->core_content;
344 		mutex_exit(&cg->core_lock);
345 		rlimit = cg->core_rlimit;
346 	} else {
347 		mutex_enter(&p->p_lock);
348 		rlimit = rctl_enforced_value(rctlproc_legacy[RLIMIT_CORE],
349 		    p->p_rctls, p);
350 		content = corectl_content_value(p->p_content);
351 		mutex_exit(&p->p_lock);
352 	}
353 
354 	if (rlimit == 0)
355 		return (EFBIG);
356 
357 	/*
358 	 * If SNOCD is set, or if the effective, real, and saved ids do
359 	 * not match up, no one but a privileged user is allowed to view
360 	 * this core file.  Set the credentials and the owner to root.
361 	 */
362 	if ((p->p_flag & SNOCD) ||
363 	    (uid = crgetuid(credp)) != crgetruid(credp) ||
364 	    uid != crgetsuid(credp) ||
365 	    (gid = crgetgid(credp)) != crgetrgid(credp) ||
366 	    gid != crgetsgid(credp)) {
367 		/*
368 		 * Because this is insecure against certain forms of file
369 		 * system attack, do it only if set-id core files have been
370 		 * enabled via corectl(CC_GLOBAL_SETID | CC_PROCESS_SETID).
371 		 */
372 		if (((core_type == CORE_GLOBAL || core_type == CORE_ZONE) &&
373 		    !(cg->core_options & CC_GLOBAL_SETID)) ||
374 		    (core_type == CORE_PROC &&
375 		    !(cg->core_options & CC_PROCESS_SETID)))
376 			return (ENOTSUP);
377 
378 		is_setid = 1;
379 	}
380 
381 	/*
382 	 * If we are doing a "global" core dump or a set-id core dump,
383 	 * use kcred to do the dumping.
384 	 */
385 	if (core_type == CORE_GLOBAL || core_type == CORE_ZONE || is_setid) {
386 		/*
387 		 * Use the zone's "kcred" to prevent privilege
388 		 * escalation.
389 		 */
390 		credp = zone_get_kcred(getzoneid());
391 		ASSERT(credp != NULL);
392 		ocredp = set_cred(p, credp);
393 	}
394 
395 	/*
396 	 * First remove any existing core file, then
397 	 * open the new core file with (O_EXCL|O_CREAT).
398 	 *
399 	 * The reasons for doing this are manifold:
400 	 *
401 	 * For security reasons, we don't want root processes
402 	 * to dump core through a symlink because that would
403 	 * allow a malicious user to clobber any file on
404 	 * the system if they could convince a root process,
405 	 * perhaps a set-uid root process that they started,
406 	 * to dump core in a directory writable by that user.
407 	 * Similar security reasons apply to hard links.
408 	 * For symmetry we do this unconditionally, not
409 	 * just for root processes.
410 	 *
411 	 * If the process has the core file mmap()d into the
412 	 * address space, we would be modifying the address
413 	 * space that we are trying to dump if we did not first
414 	 * remove the core file.  (The command "file core"
415 	 * is the canonical example of this possibility.)
416 	 *
417 	 * Opening the core file with O_EXCL|O_CREAT ensures than
418 	 * two concurrent core dumps don't clobber each other.
419 	 * One is bound to lose; we don't want to make both lose.
420 	 */
421 	if ((error = remove_core_file(fp, core_type)) == 0) {
422 		error = create_core_file(fp, core_type, &vp);
423 	}
424 
425 	/*
426 	 * Now that vn_open is complete, reset the process's credentials if
427 	 * we changed them, and make 'credp' point to kcred used
428 	 * above.  We use 'credp' to do i/o on the core file below, but leave
429 	 * p->p_cred set to the original credential to allow the core file
430 	 * to record this information.
431 	 */
432 	if (ocredp != NULL)
433 		credp = set_cred(p, ocredp);
434 
435 	if (error == 0) {
436 		int closerr;
437 #if defined(__sparc)
438 		(void) flush_user_windows_to_stack(NULL);
439 #endif
440 		if ((eswp = PTOU(curproc)->u_execsw) == NULL ||
441 		    (eswp = findexec_by_magic(eswp->exec_magic)) == NULL) {
442 			error = ENOSYS;
443 		} else {
444 			error = eswp->exec_core(vp, p, credp, rlimit, sig,
445 			    content);
446 			rw_exit(eswp->exec_lock);
447 		}
448 
449 		closerr = VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, credp, NULL);
450 		VN_RELE(vp);
451 		if (error == 0)
452 			error = closerr;
453 	}
454 
455 	if (ocredp != NULL)
456 		crfree(credp);
457 
458 	return (error);
459 }
460 
461 /*
462  * Convert a core name pattern to a pathname.
463  */
464 static int
expand_string(const char * pat,char * fp,int size,cred_t * cr)465 expand_string(const char *pat, char *fp, int size, cred_t *cr)
466 {
467 	proc_t *p = curproc;
468 	char buf[24];
469 	int len, i;
470 	char *s;
471 	char c;
472 
473 	while ((c = *pat++) != '\0') {
474 		if (size < 2)
475 			return (ENAMETOOLONG);
476 		if (c != '%') {
477 			size--;
478 			*fp++ = c;
479 			continue;
480 		}
481 		if ((c = *pat++) == '\0') {
482 			size--;
483 			*fp++ = '%';
484 			break;
485 		}
486 		switch (c) {
487 		case 'p':	/* pid */
488 			(void) sprintf((s = buf), "%d", p->p_pid);
489 			break;
490 		case 'u':	/* effective uid */
491 			(void) sprintf((s = buf), "%u", crgetuid(p->p_cred));
492 			break;
493 		case 'g':	/* effective gid */
494 			(void) sprintf((s = buf), "%u", crgetgid(p->p_cred));
495 			break;
496 		case 'f':	/* exec'd filename */
497 			s = PTOU(p)->u_comm;
498 			break;
499 		case 'd':	/* exec'd dirname */
500 			/*
501 			 * Even if pathname caching is disabled, we should
502 			 * be able to lookup the pathname for a directory.
503 			 */
504 			if (p->p_execdir != NULL && vnodetopath(NULL,
505 			    p->p_execdir, fp, size, cr) == 0) {
506 				len = (int)strlen(fp);
507 				ASSERT(len < size);
508 				ASSERT(len >= 1);
509 				ASSERT(fp[0] == '/');
510 
511 				/*
512 				 * Strip off the leading slash.
513 				 */
514 				for (i = 0; i < len; i++) {
515 					fp[i] = fp[i + 1];
516 				}
517 
518 				len--;
519 
520 				size -= len;
521 				fp += len;
522 			} else {
523 				*fp = '\0';
524 			}
525 
526 			continue;
527 		case 'n':	/* system nodename */
528 			s = uts_nodename();
529 			break;
530 		case 'm':	/* machine (sun4u, etc) */
531 			s = utsname.machine;
532 			break;
533 		case 't':	/* decimal value of time(2) */
534 			(void) sprintf((s = buf), "%ld", gethrestime_sec());
535 			break;
536 		case 'z':
537 			s = p->p_zone->zone_name;
538 			break;
539 		case 'Z':
540 			/* This is zonepath + "/root/", except for GZ */
541 			s = p->p_zone->zone_rootpath;
542 			break;
543 		case '%':
544 			(void) strcpy((s = buf), "%");
545 			break;
546 		default:
547 			s = buf;
548 			buf[0] = '%';
549 			buf[1] = c;
550 			buf[2] = '\0';
551 			break;
552 		}
553 		len = (int)strlen(s);
554 		if ((size -= len) <= 0)
555 			return (ENAMETOOLONG);
556 		(void) strcpy(fp, s);
557 		/* strip trailing "/root/" from non-GZ zonepath string */
558 		if (c == 'Z' && len > 6) {
559 			len -= 6;
560 			ASSERT(strncmp(fp + len, "/root/", 6) == 0);
561 		}
562 		fp += len;
563 	}
564 
565 	*fp = '\0';
566 	return (0);
567 }
568 
569 static int
dump_one_core(int sig,rlim64_t rlimit,enum core_types core_type,struct core_globals * cg,char ** name)570 dump_one_core(int sig, rlim64_t rlimit, enum core_types core_type,
571     struct core_globals *cg, char **name)
572 {
573 	refstr_t *rp;
574 	proc_t *p = curproc;
575 	zoneid_t zoneid;
576 	int error;
577 	char *fp;
578 	cred_t *cr;
579 
580 	ASSERT(core_type == CORE_ZONE || core_type == CORE_GLOBAL);
581 	zoneid = (core_type == CORE_ZONE ? getzoneid() : GLOBAL_ZONEID);
582 
583 	mutex_enter(&cg->core_lock);
584 	if ((rp = cg->core_file) != NULL)
585 		refstr_hold(rp);
586 	mutex_exit(&cg->core_lock);
587 	if (rp == NULL) {
588 		core_log(cg, 0, "no global core file pattern exists", NULL,
589 		    zoneid);
590 		return (1);	/* core file not generated */
591 	}
592 	fp = kmem_alloc(MAXPATHLEN, KM_SLEEP);
593 	cr = zone_get_kcred(getzoneid());
594 	error = expand_string(refstr_value(rp), fp, MAXPATHLEN, cr);
595 	crfree(cr);
596 	if (error != 0) {
597 		core_log(cg, 0, "global core file pattern too long",
598 		    refstr_value(rp), zoneid);
599 	} else if ((error = do_core(fp, sig, core_type, cg)) == 0) {
600 		core_log(cg, 0, "core dumped", fp, zoneid);
601 	} else if (error == ENOTSUP) {
602 		core_log(cg, 0, "setid process, core not dumped", fp, zoneid);
603 	} else if (error == ENOSPC) {
604 		core_log(cg, 0, "no space left on device, core truncated",
605 		    fp, zoneid);
606 	} else if (error == EFBIG) {
607 		if (rlimit == 0)
608 			core_log(cg, 0, "core rlimit is zero, core not dumped",
609 			    fp, zoneid);
610 		else
611 			core_log(cg, 0, "core rlimit exceeded, core truncated",
612 			    fp, zoneid);
613 		/*
614 		 * In addition to the core result logging, we
615 		 * may also have explicit actions defined on
616 		 * core file size violations via the resource
617 		 * control framework.
618 		 */
619 		mutex_enter(&p->p_lock);
620 		(void) rctl_action(rctlproc_legacy[RLIMIT_CORE],
621 		    p->p_rctls, p, RCA_SAFE);
622 		mutex_exit(&p->p_lock);
623 	} else {
624 		core_log(cg, error, "core dump failed", fp, zoneid);
625 	}
626 	refstr_rele(rp);
627 	if (name != NULL)
628 		*name = fp;
629 	else
630 		kmem_free(fp, MAXPATHLEN);
631 	return (error);
632 }
633 
634 int
core(int sig,int ext)635 core(int sig, int ext)
636 {
637 	proc_t *p = curproc;
638 	klwp_t *lwp = ttolwp(curthread);
639 	refstr_t *rp;
640 	char *fp_process = NULL, *fp_global = NULL, *fp_zone = NULL;
641 	int error1 = 1;
642 	int error2 = 1;
643 	int error3 = 1;
644 	k_sigset_t sigmask;
645 	k_sigset_t sighold;
646 	rlim64_t rlimit;
647 	struct core_globals *my_cg, *global_cg;
648 
649 	global_cg = zone_getspecific(core_zone_key, global_zone);
650 	ASSERT(global_cg != NULL);
651 
652 	my_cg = zone_getspecific(core_zone_key, curproc->p_zone);
653 	ASSERT(my_cg != NULL);
654 
655 	/* core files suppressed? */
656 	if (!(my_cg->core_options & (CC_PROCESS_PATH|CC_GLOBAL_PATH)) &&
657 	    !(global_cg->core_options & CC_GLOBAL_PATH)) {
658 		if (!ext && p->p_ct_process != NULL)
659 			contract_process_core(p->p_ct_process, p, sig,
660 			    NULL, NULL, NULL);
661 		return (1);
662 	}
663 
664 	/*
665 	 * Block all signals except SIGHUP, SIGINT, SIGKILL, and SIGTERM; no
666 	 * other signal may interrupt a core dump.  For each signal, we
667 	 * explicitly unblock it and set it in p_siginfo to allow for some
668 	 * minimal error reporting.  Additionally, we get the current limit on
669 	 * core file size for handling later error reporting.
670 	 */
671 	mutex_enter(&p->p_lock);
672 
673 	p->p_flag |= SDOCORE;
674 	schedctl_finish_sigblock(curthread);
675 	sigmask = curthread->t_hold;	/* remember for later */
676 	sigfillset(&sighold);
677 	if (!sigismember(&sigmask, SIGHUP))
678 		sigdelset(&sighold, SIGHUP);
679 	if (!sigismember(&sigmask, SIGINT))
680 		sigdelset(&sighold, SIGINT);
681 	if (!sigismember(&sigmask, SIGKILL))
682 		sigdelset(&sighold, SIGKILL);
683 	if (!sigismember(&sigmask, SIGTERM))
684 		sigdelset(&sighold, SIGTERM);
685 
686 	sigaddset(&p->p_siginfo, SIGHUP);
687 	sigaddset(&p->p_siginfo, SIGINT);
688 	sigaddset(&p->p_siginfo, SIGKILL);
689 	sigaddset(&p->p_siginfo, SIGTERM);
690 
691 	curthread->t_hold = sighold;
692 
693 	rlimit = rctl_enforced_value(rctlproc_legacy[RLIMIT_CORE], p->p_rctls,
694 	    p);
695 
696 	mutex_exit(&p->p_lock);
697 
698 	/*
699 	 * Undo any watchpoints.
700 	 */
701 	pr_free_watched_pages(p);
702 
703 	/*
704 	 * The presence of a current signal prevents file i/o
705 	 * from succeeding over a network.  We copy the current
706 	 * signal information to the side and cancel the current
707 	 * signal so that the core dump will succeed.
708 	 */
709 	ASSERT(lwp->lwp_cursig == sig);
710 	lwp->lwp_cursig = 0;
711 	lwp->lwp_extsig = 0;
712 	if (lwp->lwp_curinfo == NULL) {
713 		bzero(&lwp->lwp_siginfo, sizeof (k_siginfo_t));
714 		lwp->lwp_siginfo.si_signo = sig;
715 		lwp->lwp_siginfo.si_code = SI_NOINFO;
716 	} else {
717 		bcopy(&lwp->lwp_curinfo->sq_info,
718 		    &lwp->lwp_siginfo, sizeof (k_siginfo_t));
719 		siginfofree(lwp->lwp_curinfo);
720 		lwp->lwp_curinfo = NULL;
721 	}
722 
723 	/*
724 	 * Convert the core file name patterns into path names
725 	 * and call do_core() to write the core files.
726 	 */
727 
728 	if (my_cg->core_options & CC_PROCESS_PATH) {
729 		mutex_enter(&p->p_lock);
730 		if (p->p_corefile != NULL)
731 			rp = corectl_path_value(p->p_corefile);
732 		else
733 			rp = NULL;
734 		mutex_exit(&p->p_lock);
735 		if (rp != NULL) {
736 			fp_process = kmem_alloc(MAXPATHLEN, KM_SLEEP);
737 			error1 = expand_string(refstr_value(rp),
738 			    fp_process, MAXPATHLEN, p->p_cred);
739 			if (error1 == 0)
740 				error1 = do_core(fp_process, sig, CORE_PROC,
741 				    my_cg);
742 			refstr_rele(rp);
743 		}
744 	}
745 
746 	if (my_cg->core_options & CC_GLOBAL_PATH)
747 		error2 = dump_one_core(sig, rlimit, CORE_ZONE, my_cg,
748 		    &fp_global);
749 	if (global_cg != my_cg && (global_cg->core_options & CC_GLOBAL_PATH))
750 		error3 = dump_one_core(sig, rlimit, CORE_GLOBAL, global_cg,
751 		    &fp_zone);
752 
753 	/*
754 	 * Restore the signal hold mask.
755 	 */
756 	mutex_enter(&p->p_lock);
757 	curthread->t_hold = sigmask;
758 	mutex_exit(&p->p_lock);
759 
760 	if (!ext && p->p_ct_process != NULL)
761 		contract_process_core(p->p_ct_process, p, sig,
762 		    error1 == 0 ? fp_process : NULL,
763 		    error2 == 0 ? fp_global : NULL,
764 		    error3 == 0 ? fp_zone : NULL);
765 
766 	if (fp_process != NULL)
767 		kmem_free(fp_process, MAXPATHLEN);
768 	if (fp_global != NULL)
769 		kmem_free(fp_global, MAXPATHLEN);
770 	if (fp_zone != NULL)
771 		kmem_free(fp_zone, MAXPATHLEN);
772 
773 	/*
774 	 * Return non-zero if no core file was created.
775 	 */
776 	return (error1 != 0 && error2 != 0 && error3 != 0);
777 }
778 
779 /*
780  * Maximum chunk size for dumping core files,
781  * size in pages, patchable in /etc/system
782  */
783 uint_t	core_chunk = 32;
784 
785 /*
786  * The delay between core_write() calls, in microseconds.  The default
787  * matches one "normal" clock tick, or 10 milliseconds.
788  */
789 clock_t	core_delay_usec = 10000;
790 
791 /*
792  * Common code to core dump process memory.  The core_seg routine does i/o
793  * using core_write() below, and so it has the same failure semantics.
794  */
795 int
core_seg(proc_t * p,vnode_t * vp,u_offset_t offset,caddr_t addr,size_t size,rlim64_t rlimit,cred_t * credp)796 core_seg(proc_t *p, vnode_t *vp, u_offset_t offset, caddr_t addr, size_t size,
797     rlim64_t rlimit, cred_t *credp)
798 {
799 	caddr_t eaddr;
800 	caddr_t base;
801 	size_t len;
802 	int err = 0;
803 
804 	if (offset > OFF_MAX || offset + size > OFF_MAX ||
805 	    offset + size < offset) {
806 		return (EOVERFLOW);
807 	}
808 
809 	eaddr = addr + size;
810 	for (base = addr; base < eaddr; base += len) {
811 		len = eaddr - base;
812 		if (as_memory(p->p_as, &base, &len) != 0)
813 			return (0);
814 
815 		/*
816 		 * Reduce len to a reasonable value so that we don't
817 		 * overwhelm the VM system with a monstrously large
818 		 * single write and cause pageout to stop running.
819 		 */
820 		if (len > (size_t)core_chunk * PAGESIZE)
821 			len = (size_t)core_chunk * PAGESIZE;
822 
823 		err = core_write(vp, UIO_USERSPACE,
824 		    offset + (size_t)(base - addr), base, len, rlimit, credp);
825 
826 		if (err)
827 			return (err);
828 
829 		/*
830 		 * If we have taken a signal, return EINTR to allow the dump
831 		 * to be aborted.
832 		 */
833 		if (issig(JUSTLOOKING) && issig(FORREAL))
834 			return (EINTR);
835 	}
836 
837 	return (0);
838 }
839 
840 /*
841  * Wrapper around vn_rdwr to perform writes to a core file.  For core files,
842  * we always want to write as much as we possibly can, and then make sure to
843  * return either 0 to the caller (for success), or the actual errno value.
844  * By using this function, the caller can omit additional code for handling
845  * retries and errors for partial writes returned by vn_rdwr.  If vn_rdwr
846  * unexpectedly returns zero but no progress has been made, we return ENOSPC.
847  */
848 int
core_write(vnode_t * vp,enum uio_seg segflg,u_offset_t offset,const void * buf,size_t len,rlim64_t rlimit,cred_t * credp)849 core_write(vnode_t *vp, enum uio_seg segflg, u_offset_t offset,
850     const void *buf, size_t len, rlim64_t rlimit, cred_t *credp)
851 {
852 	ssize_t resid = len;
853 	int error = 0;
854 
855 	if (offset > OFF_MAX || offset + len > OFF_MAX ||
856 	    offset + len < offset) {
857 		return (EOVERFLOW);
858 	}
859 
860 	while (len != 0) {
861 		error = vn_rdwr(UIO_WRITE, vp, (caddr_t)buf, len,
862 		    (offset_t)offset, segflg, 0, rlimit, credp, &resid);
863 
864 		if (error != 0)
865 			break;
866 
867 		if (resid >= len)
868 			return (ENOSPC);
869 
870 		buf = (const char *)buf + len - resid;
871 		offset += len - resid;
872 		len = resid;
873 	}
874 
875 	return (error);
876 }
877