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  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/stat.h>
29 #include <sys/types.h>
30 
31 #include <assert.h>
32 #include <ctype.h>
33 #include <errno.h>
34 #include <fcntl.h>
35 #include <libintl.h>
36 #include <libscf.h>
37 #include <libuutil.h>
38 #include <limits.h>
39 #include <md5.h>
40 #include <pthread.h>
41 #include <stdio.h>
42 #include <stdlib.h>
43 #include <string.h>
44 #include <strings.h>
45 #include <unistd.h>
46 
47 #include <manifest_hash.h>
48 
49 /*
50  * Translate a file name to property name.  Return an allocated string or NULL
51  * if realpath() fails.
52  */
53 char *
54 mhash_filename_to_propname(const char *in)
55 {
56 	char *out, *cp, *base;
57 	size_t len, piece_len;
58 
59 	out = uu_zalloc(PATH_MAX + 1);
60 	if (realpath(in, out) == NULL) {
61 		uu_free(out);
62 		return (NULL);
63 	}
64 
65 	base = getenv("PKG_INSTALL_ROOT");
66 
67 	/*
68 	 * We copy-shift over the basedir and the leading slash, since it's
69 	 * not relevant to when we boot with this repository.
70 	 */
71 
72 	cp = out + ((base != NULL)? strlen(base) : 0);
73 	if (*cp == '/')
74 		cp++;
75 	(void) memmove(out, cp, strlen(cp) + 1);
76 
77 	len = strlen(out);
78 	if (len > scf_limit(SCF_LIMIT_MAX_NAME_LENGTH)) {
79 		/* Use the first half and the second half. */
80 		piece_len = (scf_limit(SCF_LIMIT_MAX_NAME_LENGTH) - 3) / 2;
81 
82 		(void) strncpy(out + piece_len, "__", 2);
83 
84 		(void) memmove(out + piece_len + 2, out + (len - piece_len),
85 		    piece_len + 1);
86 	}
87 
88 	/*
89 	 * Translate non-property characters to '_', first making sure that
90 	 * we don't begin with '_'.
91 	 */
92 
93 	if (!isalpha(*out))
94 		*out = 'A';
95 
96 	for (cp = out + 1; *cp != '\0'; ++cp) {
97 		if (!(isalnum(*cp) || *cp == '_' || *cp == '-'))
98 			*cp = '_';
99 	}
100 
101 	return (out);
102 }
103 
104 int
105 mhash_retrieve_entry(scf_handle_t *hndl, const char *name, uchar_t *hash)
106 {
107 	scf_scope_t *scope;
108 	scf_service_t *svc;
109 	scf_propertygroup_t *pg;
110 	scf_property_t *prop;
111 	scf_value_t *val;
112 	ssize_t szret;
113 	int result = 0;
114 
115 	/*
116 	 * In this implementation the hash for name is the opaque value of
117 	 * svc:/MHASH_SVC/:properties/name/MHASH_PROP
118 	 */
119 
120 	if ((scope = scf_scope_create(hndl)) == NULL ||
121 	    (svc = scf_service_create(hndl)) == NULL ||
122 	    (pg = scf_pg_create(hndl)) == NULL ||
123 	    (prop = scf_property_create(hndl)) == NULL ||
124 	    (val = scf_value_create(hndl)) == NULL) {
125 		result = -1;
126 		goto out;
127 	}
128 
129 	if (scf_handle_get_local_scope(hndl, scope) < 0) {
130 		result = -1;
131 		goto out;
132 	}
133 
134 	if (scf_scope_get_service(scope, MHASH_SVC, svc) < 0) {
135 		result = -1;
136 		goto out;
137 	}
138 
139 	if (scf_service_get_pg(svc, name, pg) != SCF_SUCCESS) {
140 		result = -1;
141 		goto out;
142 	}
143 
144 	if (scf_pg_get_property(pg, MHASH_PROP, prop) != SCF_SUCCESS) {
145 		result = -1;
146 		goto out;
147 	}
148 
149 	if (scf_property_get_value(prop, val) != SCF_SUCCESS) {
150 		result = -1;
151 		goto out;
152 	}
153 
154 	szret = scf_value_get_opaque(val, hash, MHASH_SIZE);
155 	if (szret < 0) {
156 		result = -1;
157 		goto out;
158 	}
159 
160 	/*
161 	 * Make sure that the old hash is returned with
162 	 * remainder of the bytes zeroed.
163 	 */
164 	if (szret == MHASH_SIZE_OLD) {
165 		(void) memset(hash + MHASH_SIZE_OLD, 0,
166 		    MHASH_SIZE - MHASH_SIZE_OLD);
167 	} else if (szret != MHASH_SIZE) {
168 		scf_value_destroy(val);
169 		result = -1;
170 		goto out;
171 	}
172 
173 out:
174 	(void) scf_value_destroy(val);
175 	scf_property_destroy(prop);
176 	scf_pg_destroy(pg);
177 	scf_service_destroy(svc);
178 	scf_scope_destroy(scope);
179 
180 	return (result);
181 }
182 
183 int
184 mhash_store_entry(scf_handle_t *hndl, const char *name, uchar_t *hash,
185     char **errstr)
186 {
187 	scf_scope_t *scope = NULL;
188 	scf_service_t *svc = NULL;
189 	scf_propertygroup_t *pg = NULL;
190 	scf_property_t *prop = NULL;
191 	scf_value_t *val = NULL;
192 	scf_transaction_t *tx = NULL;
193 	scf_transaction_entry_t *e = NULL;
194 	int ret, result = 0;
195 
196 	int i;
197 
198 	if ((scope = scf_scope_create(hndl)) == NULL ||
199 	    (svc = scf_service_create(hndl)) == NULL ||
200 	    (pg = scf_pg_create(hndl)) == NULL ||
201 	    (prop = scf_property_create(hndl)) == NULL) {
202 		if (errstr != NULL)
203 			*errstr = gettext("Could not create scf objects");
204 		result = -1;
205 		goto out;
206 	}
207 
208 	if (scf_handle_get_local_scope(hndl, scope) != SCF_SUCCESS) {
209 		if (errstr != NULL)
210 			*errstr = gettext("Could not get local scope");
211 		result = -1;
212 		goto out;
213 	}
214 
215 	for (i = 0; i < 5; ++i) {
216 		scf_error_t err;
217 
218 		if (scf_scope_get_service(scope, MHASH_SVC, svc) ==
219 		    SCF_SUCCESS)
220 			break;
221 
222 		if (scf_error() != SCF_ERROR_NOT_FOUND) {
223 			if (errstr != NULL)
224 				*errstr = gettext("Could not get manifest hash "
225 				    "service");
226 			result = -1;
227 			goto out;
228 		}
229 
230 		if (scf_scope_add_service(scope, MHASH_SVC, svc) ==
231 		    SCF_SUCCESS)
232 			break;
233 
234 		err = scf_error();
235 
236 		if (err == SCF_ERROR_EXISTS)
237 			/* Try again. */
238 			continue;
239 		else if (err == SCF_ERROR_PERMISSION_DENIED) {
240 			if (errstr != NULL)
241 				*errstr = gettext("Could not store file hash: "
242 				    "permission denied.\n");
243 			result = -1;
244 			goto out;
245 		}
246 
247 		if (errstr != NULL)
248 			*errstr = gettext("Could not add manifest hash "
249 			    "service");
250 		result = -1;
251 		goto out;
252 	}
253 
254 	if (i == 5) {
255 		if (errstr != NULL)
256 			*errstr = gettext("Could not store file hash: "
257 			    "service addition contention.\n");
258 		result = -1;
259 		goto out;
260 	}
261 
262 	for (i = 0; i < 5; ++i) {
263 		scf_error_t err;
264 
265 		if (scf_service_get_pg(svc, name, pg) == SCF_SUCCESS)
266 			break;
267 
268 		if (scf_error() != SCF_ERROR_NOT_FOUND) {
269 			if (errstr != NULL)
270 				*errstr = gettext("Could not get service's "
271 				    "hash record)");
272 			result = -1;
273 			goto out;
274 		}
275 
276 		if (scf_service_add_pg(svc, name, MHASH_PG_TYPE,
277 		    MHASH_PG_FLAGS, pg) == SCF_SUCCESS)
278 			break;
279 
280 		err = scf_error();
281 
282 		if (err == SCF_ERROR_EXISTS)
283 			/* Try again. */
284 			continue;
285 		else if (err == SCF_ERROR_PERMISSION_DENIED) {
286 			if (errstr != NULL)
287 				*errstr = gettext("Could not store file hash: "
288 				    "permission denied.\n");
289 			result = -1;
290 			goto out;
291 		}
292 
293 		if (errstr != NULL)
294 			*errstr = gettext("Could not store file hash");
295 		result = -1;
296 		goto out;
297 	}
298 	if (i == 5) {
299 		if (errstr != NULL)
300 			*errstr = gettext("Could not store file hash: "
301 			    "property group addition contention.\n");
302 		result = -1;
303 		goto out;
304 	}
305 
306 	if ((e = scf_entry_create(hndl)) == NULL ||
307 	    (val = scf_value_create(hndl)) == NULL) {
308 		if (errstr != NULL)
309 			*errstr = gettext("Could not store file hash: "
310 			    "permission denied.\n");
311 		result = -1;
312 		goto out;
313 	}
314 
315 	ret = scf_value_set_opaque(val, hash, MHASH_SIZE);
316 	assert(ret == SCF_SUCCESS);
317 
318 	tx = scf_transaction_create(hndl);
319 	if (tx == NULL) {
320 		if (errstr != NULL)
321 			*errstr = gettext("Could not create transaction");
322 		result = -1;
323 		goto out;
324 	}
325 
326 	do {
327 		if (scf_pg_update(pg) == -1) {
328 			if (errstr != NULL)
329 				*errstr = gettext("Could not update hash "
330 				    "entry");
331 			result = -1;
332 			goto out;
333 		}
334 		if (scf_transaction_start(tx, pg) != SCF_SUCCESS) {
335 			if (scf_error() != SCF_ERROR_PERMISSION_DENIED) {
336 				if (errstr != NULL)
337 					*errstr = gettext("Could not start "
338 					    "hash transaction.\n");
339 				result = -1;
340 				goto out;
341 			}
342 
343 			if (errstr != NULL)
344 				*errstr = gettext("Could not store file hash: "
345 				    "permission denied.\n");
346 			result = -1;
347 
348 			scf_transaction_destroy(tx);
349 			(void) scf_entry_destroy(e);
350 			goto out;
351 		}
352 
353 		if (scf_transaction_property_new(tx, e, MHASH_PROP,
354 		    SCF_TYPE_OPAQUE) != SCF_SUCCESS &&
355 		    scf_transaction_property_change_type(tx, e, MHASH_PROP,
356 		    SCF_TYPE_OPAQUE) != SCF_SUCCESS) {
357 			if (errstr != NULL)
358 				*errstr = gettext("Could not modify hash "
359 				    "entry");
360 			result = -1;
361 			goto out;
362 		}
363 
364 		ret = scf_entry_add_value(e, val);
365 		assert(ret == SCF_SUCCESS);
366 
367 		ret = scf_transaction_commit(tx);
368 
369 		if (ret == 0)
370 			scf_transaction_reset(tx);
371 	} while (ret == 0);
372 
373 	if (ret < 0) {
374 		if (scf_error() != SCF_ERROR_PERMISSION_DENIED) {
375 			if (errstr != NULL)
376 				*errstr = gettext("Could not store file hash: "
377 				    "permission denied.\n");
378 			result = -1;
379 			goto out;
380 		}
381 
382 		if (errstr != NULL)
383 			*errstr = gettext("Could not commit transaction");
384 		result = -1;
385 	}
386 
387 	scf_transaction_destroy(tx);
388 	(void) scf_entry_destroy(e);
389 
390 out:
391 	(void) scf_value_destroy(val);
392 	scf_property_destroy(prop);
393 	scf_pg_destroy(pg);
394 	scf_service_destroy(svc);
395 	scf_scope_destroy(scope);
396 
397 	return (result);
398 }
399 
400 /*
401  * Generate the md5 hash of a file; manifest files are smallish
402  * so we can read them in one gulp.
403  */
404 static int
405 md5_hash_file(const char *file, off64_t sz, uchar_t *hash)
406 {
407 	char *buf;
408 	int fd;
409 	ssize_t res;
410 	int ret;
411 
412 	fd = open(file, O_RDONLY);
413 	if (fd < 0)
414 		return (-1);
415 
416 	buf = malloc(sz);
417 	if (buf == NULL) {
418 		(void) close(fd);
419 		return (-1);
420 	}
421 
422 	res = read(fd, buf, (size_t)sz);
423 
424 	(void) close(fd);
425 
426 	if (res == sz) {
427 		ret = 0;
428 		md5_calc(hash, (uchar_t *)buf, (unsigned int) sz);
429 	} else {
430 		ret = -1;
431 	}
432 
433 	free(buf);
434 	return (ret);
435 }
436 
437 /*
438  * int mhash_test_file(scf_handle_t *, const char *, uint_t, char **, uchar_t *)
439  *   Test the given filename against the hashed metadata in the repository.
440  *   The behaviours for import and apply are slightly different.  For imports,
441  *   if the hash value is absent or different, then the import operation
442  *   continues.  For profile application, the operation continues only if the
443  *   hash value for the file is absent.
444  *
445  *   We keep two hashes: one which can be quickly test: the metadata hash,
446  *   and one which is more expensive to test: the file contents hash.
447  *
448  *   If either hash matches, the file does not need to be re-read.
449  *   If only one of the hashes matches, a side effect of this function
450  *   is to store the newly computed hash.
451  *   If neither hash matches, the hash computed for the new file is returned
452  *   and not stored.
453  *
454  *   Return values:
455  *	MHASH_NEWFILE	- the file no longer matches the hash or no hash existed
456  *			  ONLY in this case we return the new file's hash.
457  *	MHASH_FAILURE	- an internal error occurred, or the file was not found.
458  *	MHASH_RECONCILED- based on the metadata/file hash, the file does
459  *			  not need to be re-read; if necessary,
460  *			  the hash was upgraded or reconciled.
461  *
462  * NOTE: no hash is returned UNLESS MHASH_NEWFILE is returned.
463  */
464 int
465 mhash_test_file(scf_handle_t *hndl, const char *file, uint_t is_profile,
466     char **pnamep, uchar_t *hashbuf)
467 {
468 	boolean_t do_hash;
469 	struct stat64 st;
470 	char *cp;
471 	char *data;
472 	uchar_t stored_hash[MHASH_SIZE];
473 	uchar_t hash[MHASH_SIZE];
474 	char *pname;
475 	int ret;
476 	int hashash;
477 	int metahashok = 0;
478 
479 	/*
480 	 * In the case where we are doing automated imports, we reduce the UID,
481 	 * the GID, the size, and the mtime into a string (to eliminate
482 	 * endianness) which we then make opaque as a single MD5 digest.
483 	 *
484 	 * The previous hash was composed of the inode number, the UID, the file
485 	 * size, and the mtime.  This formulation was found to be insufficiently
486 	 * portable for use in highly replicated deployments.  The current
487 	 * algorithm will allow matches of this "v1" hash, but always returns
488 	 * the effective "v2" hash, such that updates result in the more
489 	 * portable hash being used.
490 	 *
491 	 * An unwanted side effect of a hash based solely on the file
492 	 * meta data is the fact that we pay no attention to the contents
493 	 * which may remain the same despite meta data changes.  This happens
494 	 * with (live) upgrades.  We extend the V2 hash with an additional
495 	 * digest of the file contents and the code retrieving the hash
496 	 * from the repository zero fills the remainder so we can detect
497 	 * it is missing.
498 	 *
499 	 * If the the V2 digest matches, we check for the presence of
500 	 * the contents digest and compute and store it if missing.
501 	 *
502 	 * If the V2 digest doesn't match but we also have a non-zero
503 	 * file hash, we match the file content digest.  If it matches,
504 	 * we compute and store the new complete hash so that later
505 	 * checks will find the meta data digest correct.
506 	 *
507 	 * If the above matches fail and the V1 hash doesn't match either,
508 	 * we consider the test to have failed, implying that some aspect
509 	 * of the manifest has changed.
510 	 */
511 
512 	cp = getenv("SVCCFG_CHECKHASH");
513 	do_hash = (cp != NULL && *cp != '\0');
514 	if (!do_hash) {
515 		if (pnamep != NULL)
516 			*pnamep = NULL;
517 		return (MHASH_NEWFILE);
518 	}
519 
520 	pname = mhash_filename_to_propname(file);
521 	if (pname == NULL)
522 		return (MHASH_FAILURE);
523 
524 	hashash = mhash_retrieve_entry(hndl, pname, stored_hash) == 0;
525 
526 	/* Never reread a profile. */
527 	if (hashash && is_profile) {
528 		uu_free(pname);
529 		return (MHASH_RECONCILED);
530 	}
531 
532 	/*
533 	 * No hash and not interested in one, then don't bother computing it.
534 	 * We also skip returning the property name in that case.
535 	 */
536 	if (!hashash && hashbuf == NULL) {
537 		uu_free(pname);
538 		return (MHASH_NEWFILE);
539 	}
540 
541 	do
542 		ret = stat64(file, &st);
543 	while (ret < 0 && errno == EINTR);
544 	if (ret < 0) {
545 		uu_free(pname);
546 		return (MHASH_FAILURE);
547 	}
548 
549 	data = uu_msprintf(MHASH_FORMAT_V2, st.st_uid, st.st_gid,
550 	    st.st_size, st.st_mtime);
551 	if (data == NULL) {
552 		uu_free(pname);
553 		return (MHASH_FAILURE);
554 	}
555 
556 	(void) memset(hash, 0, MHASH_SIZE);
557 	md5_calc(hash, (uchar_t *)data, strlen(data));
558 
559 	uu_free(data);
560 
561 	/*
562 	 * Verify the meta data hash.
563 	 */
564 	if (hashash && memcmp(hash, stored_hash, MD5_DIGEST_LENGTH) == 0) {
565 		int i;
566 
567 		metahashok = 1;
568 		/*
569 		 * The metadata hash matches; now we see if there was a
570 		 * content hash; if not, we will continue on and compute and
571 		 * store the updated hash.
572 		 * If there was no content hash, mhash_retrieve_entry()
573 		 * will have zero filled it.
574 		 */
575 		for (i = 0; i < MD5_DIGEST_LENGTH; i++) {
576 			if (stored_hash[MD5_DIGEST_LENGTH+i] != 0) {
577 				uu_free(pname);
578 				return (MHASH_RECONCILED);
579 			}
580 		}
581 	}
582 
583 	/*
584 	 * Compute the file hash as we can no longer avoid having to know it.
585 	 * Note: from this point on "hash" contains the full, current, hash.
586 	 */
587 	if (md5_hash_file(file, st.st_size, &hash[MHASH_SIZE_OLD]) != 0) {
588 		uu_free(pname);
589 		return (MHASH_FAILURE);
590 	}
591 	if (hashash) {
592 		uchar_t hash_v1[MHASH_SIZE_OLD];
593 
594 		if (metahashok ||
595 		    memcmp(&hash[MHASH_SIZE_OLD], &stored_hash[MHASH_SIZE_OLD],
596 		    MD5_DIGEST_LENGTH) == 0) {
597 
598 			/*
599 			 * Reconcile entry: we get here when either the
600 			 * meta data hash matches or the content hash matches;
601 			 * we then update the database with the complete
602 			 * new hash so we can be a bit quicker next time.
603 			 */
604 			(void) mhash_store_entry(hndl, pname, hash, NULL);
605 			uu_free(pname);
606 			return (MHASH_RECONCILED);
607 		}
608 
609 		/*
610 		 * No match on V2 hash or file content; compare V1 hash.
611 		 */
612 		data = uu_msprintf(MHASH_FORMAT_V1, st.st_ino, st.st_uid,
613 		    st.st_size, st.st_mtime);
614 		if (data == NULL) {
615 			uu_free(pname);
616 			return (MHASH_FAILURE);
617 		}
618 
619 		md5_calc(hash_v1, (uchar_t *)data, strlen(data));
620 
621 		uu_free(data);
622 
623 		if (memcmp(hash_v1, stored_hash, MD5_DIGEST_LENGTH) == 0) {
624 			/*
625 			 * Update the new entry so we don't have to go through
626 			 * all this trouble next time.
627 			 */
628 			(void) mhash_store_entry(hndl, pname, hash, NULL);
629 			uu_free(pname);
630 			return (MHASH_RECONCILED);
631 		}
632 	}
633 
634 	if (pnamep != NULL)
635 		*pnamep = pname;
636 	else
637 		uu_free(pname);
638 
639 	if (hashbuf != NULL)
640 		(void) memcpy(hashbuf, hash, MHASH_SIZE);
641 
642 	return (MHASH_NEWFILE);
643 }
644