xref: /illumos-gate/usr/src/cmd/fs.d/ufs/fsck/pass3b.c (revision 2a8bcb4e)
1 /*
2  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
3  * Use is subject to license terms.
4  */
5 
6 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
7 /*	  All Rights Reserved  	*/
8 
9 /*
10  * Copyright (c) 1980, 1986, 1990 The Regents of the University of California.
11  * All rights reserved.
12  *
13  * Redistribution and use in source and binary forms are permitted
14  * provided that: (1) source distributions retain this entire copyright
15  * notice and comment, and (2) distributions including binaries display
16  * the following acknowledgement:  ``This product includes software
17  * developed by the University of California, Berkeley and its contributors''
18  * in the documentation or other materials provided with the distribution
19  * and in all advertising materials mentioning features or use of this
20  * software. Neither the name of the University nor the names of its
21  * contributors may be used to endorse or promote products derived
22  * from this software without specific prior written permission.
23  * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
25  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
26  */
27 
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <sys/param.h>
31 #include <sys/types.h>
32 #include <sys/mntent.h>
33 #include <sys/acl.h>
34 #include <sys/fs/ufs_acl.h>
35 #include <sys/fs/ufs_fs.h>
36 #include <sys/vnode.h>
37 #include <string.h>
38 #include <sys/fs/ufs_inode.h>
39 #include "fsck.h"
40 
41 /*
42  * We can be run on multiple filesystems (processed serially), so
43  * these need to be re-initialized each time we start the pass.
44  */
45 static caddr_t aclbuf;		/* hold acl's for parsing */
46 static int64_t aclbufoff;	/* offset into aclbuf */
47 static int64_t maxaclsize;	/* how big aclbuf is */
48 
49 static int aclblksort(const void *, const void *);
50 static int bufchk(char *, int64_t, fsck_ino_t);
51 static void clear_shadow_client(struct shadowclientinfo *,
52 	    struct shadowclients *, int);
53 
54 void
pass3b(void)55 pass3b(void)
56 {
57 	fsck_ino_t inumber;
58 	struct dinode *dp;
59 	struct inoinfo *aclp;
60 	struct inodesc curino;
61 	struct shadowclientinfo *sci;
62 	struct shadowclients *scc;
63 	int64_t acl_size_limit;
64 	int i;
65 
66 	/*
67 	 * Sort the acl list into disk block order.
68 	 */
69 	qsort((char *)aclpsort, (int)aclplast, sizeof (*aclpsort), aclblksort);
70 	/*
71 	 * Scan all the acl inodes, finding the largest acl file.
72 	 *
73 	 * The largest legal size is (4 * MAX_ACL_ENTRIES + 8) entries.
74 	 * The four are the categories of specific users, specific
75 	 * groups, default specific users, and default specific groups.
76 	 * The eight are the entries for the owning user/group/other/class
77 	 * plus the equivalent defaults.
78 	 *
79 	 * We double this to allow for a truly worst-case but legal
80 	 * situation of every single acl having its own fsd_t wrapper.
81 	 * Doubling is a bit pessimistic (sizeof (acl_t) > sizeof (fsd_t)).
82 	 */
83 	acl_size_limit = sizeof (ufs_acl_t) * (4 * MAX_ACL_ENTRIES + 8);
84 	acl_size_limit *= 2;
85 
86 	maxaclsize = 0;
87 	for (inumber = 0; inumber < aclplast; inumber++) {
88 		aclp = aclpsort[inumber];
89 		if ((int64_t)aclp->i_isize > acl_size_limit) {
90 			(void) printf(
91 			    "ACL I=%d is excessively large (%lld > %lld)",
92 			    inumber,
93 			    (longlong_t)aclp->i_isize,
94 			    (longlong_t)acl_size_limit);
95 			if (preen) {
96 				(void) printf(" (IGNORING)\n");
97 			} else if (reply("CLEAR") == 1) {
98 				freeino(inumber, TI_PARENT);
99 			} else {
100 				iscorrupt = 1;
101 				(void) printf("IGNORING SHADOW I=%d\n",
102 				    inumber);
103 			}
104 			continue;
105 		}
106 		if ((int64_t)aclp->i_isize > maxaclsize)
107 			maxaclsize = (int64_t)aclp->i_isize;
108 	}
109 
110 	maxaclsize = ((maxaclsize / sblock.fs_bsize) + 1) * sblock.fs_bsize;
111 	if (maxaclsize == 0)
112 		goto noacls;
113 
114 	if (aclbuf != NULL) {
115 		free((void *)aclbuf);
116 	}
117 	if ((aclbuf = malloc(maxaclsize)) == NULL) {
118 		errexit("cannot alloc %lld bytes for aclbuf\n",
119 			(longlong_t)maxaclsize);
120 	}
121 	/*
122 	 * Scan all the acl inodes, checking contents
123 	 */
124 	for (inumber = 0; inumber < aclplast; inumber++) {
125 		aclp = aclpsort[inumber];
126 		if ((int64_t)aclp->i_isize > acl_size_limit) {
127 			continue;
128 		}
129 		if ((statemap[aclp->i_number] & STMASK) != SSTATE) {
130 			continue;
131 		}
132 		dp = ginode(aclp->i_number);
133 		init_inodesc(&curino);
134 		curino.id_fix = FIX;
135 		curino.id_type = ACL;
136 		curino.id_func = pass3bcheck;
137 		curino.id_number = aclp->i_number;
138 		curino.id_filesize = aclp->i_isize;
139 		aclbufoff = 0;
140 		(void) memset(aclbuf, 0, (size_t)maxaclsize);
141 		if ((ckinode(dp, &curino, CKI_TRAVERSE) & KEEPON) == 0 ||
142 		    bufchk(aclbuf, (int64_t)aclp->i_isize, aclp->i_number)) {
143 			dp = ginode(aclp->i_number); /* defensive no-op */
144 			if (dp->di_nlink <= 0) {
145 				statemap[aclp->i_number] = FSTATE;
146 				continue;
147 			}
148 			(void) printf("ACL I=%d BAD/CORRUPT", aclp->i_number);
149 			if (preen || reply("CLEAR") == 1) {
150 				if (preen)
151 					(void) printf("\n");
152 				freeino(aclp->i_number, TI_PARENT);
153 			} else {
154 				iscorrupt = 1;
155 			}
156 		}
157 	}
158 	/*
159 	 * Now scan all shadow inodes, checking that any inodes that previously
160 	 * had an acl still have an acl.
161 	 */
162 noacls:
163 	for (sci = shadowclientinfo; sci; sci = sci->next) {
164 		if ((statemap[sci->shadow] & STMASK) != SSTATE) {
165 			for (scc = sci->clients; scc; scc = scc->next) {
166 				for (i = 0; i < scc->nclients; i++) {
167 					clear_shadow_client(sci, scc, i);
168 				}
169 			}
170 		}
171 	}
172 	free((void *)aclbuf);
173 	aclbuf = NULL;
174 }
175 
176 static void
clear_shadow_client(struct shadowclientinfo * sci,struct shadowclients * scc,int client)177 clear_shadow_client(struct shadowclientinfo *sci, struct shadowclients *scc,
178 	int client)
179 {
180 	int suppress_update = 0;
181 	caddr_t flow;
182 	struct inodesc ldesc;
183 	struct dinode *dp;
184 
185 	(void) printf("I=%d HAS BAD/CLEARED ACL I=%d",
186 	    scc->client[client], sci->shadow);
187 	if (preen || reply("FIX") == 1) {
188 		if (preen)
189 			(void) printf("\n");
190 
191 		/*
192 		 * If we clear the ACL, then the permissions should
193 		 * be as restrictive as possible until the user can
194 		 * set it to something reasonable.  If we keep the
195 		 * ACL, then the permissions are pretty much
196 		 * irrelevant.  So, just always clear the permission
197 		 * bits.
198 		 */
199 		dp = ginode(scc->client[client]);
200 		dp->di_mode &= IFMT;
201 		dp->di_shadow = 0;
202 		inodirty();
203 
204 		/*
205 		 * Decrement in-memory link count - pass1 made sure
206 		 * the shadow inode # is a valid inode number.  But
207 		 * first, see if we're going to overflow our sixteen
208 		 * bits.
209 		 */
210 		LINK_RANGE(flow, lncntp[dp->di_shadow], 1);
211 		if (flow != NULL) {
212 			LINK_CLEAR(flow, scc->client[client], dp->di_mode,
213 			    &ldesc);
214 			if (statemap[scc->client[client]] == USTATE)
215 				suppress_update = 1;
216 		}
217 
218 		/*
219 		 * We don't touch the shadow's on-disk link count,
220 		 * because we've already cleared its state in pass3b().
221 		 * Here we're just trying to keep lncntp[] in sync, so
222 		 * we can detect spurious links.
223 		 */
224 		if (!suppress_update)
225 			TRACK_LNCNTP(sci->shadow, lncntp[sci->shadow]++);
226 	} else {
227 		iscorrupt = 1;
228 	}
229 }
230 
231 /*
232  * Collect all the (data) blocks of an acl file into a buffer.
233  * Later we'll scan the buffer and validate the acl data.
234  */
235 int
pass3bcheck(struct inodesc * idesc)236 pass3bcheck(struct inodesc *idesc)
237 {
238 	struct bufarea *bp;
239 	size_t size, bsize;
240 
241 	if (aclbufoff == idesc->id_filesize) {
242 		return (STOP);
243 	}
244 	bsize = size = sblock.fs_fsize * idesc->id_numfrags;
245 	if ((size + aclbufoff) > idesc->id_filesize)
246 		size = idesc->id_filesize - aclbufoff;
247 	if (aclbufoff + size > maxaclsize)
248 		errexit("acl size %lld exceeds maximum calculated "
249 			"size of %lld bytes",
250 			(longlong_t)aclbufoff + size, (longlong_t)maxaclsize);
251 	bp = getdatablk(idesc->id_blkno, bsize);
252 	if (bp->b_errs != 0) {
253 		brelse(bp);
254 		return (STOP);
255 	}
256 	(void) memmove((void *)(aclbuf + aclbufoff), (void *)bp->b_un.b_buf,
257 		(size_t)size);
258 	aclbufoff += size;
259 	brelse(bp);
260 	return (KEEPON);
261 }
262 
263 /*
264  * Routine to sort disk blocks.
265  */
266 static int
aclblksort(const void * pp1,const void * pp2)267 aclblksort(const void *pp1, const void *pp2)
268 {
269 	const struct inoinfo **aclpp1 = (const struct inoinfo **)pp1;
270 	const struct inoinfo **aclpp2 = (const struct inoinfo **)pp2;
271 
272 	return ((*aclpp1)->i_blks[0] - (*aclpp2)->i_blks[0]);
273 }
274 
275 /*
276  * Scan a chunk of a shadow file.  Return zero if no ACLs were found,
277  * or when all that were found were valid.
278  */
279 static int
bufchk(char * buf,int64_t len,fsck_ino_t inum)280 bufchk(char *buf, int64_t len, fsck_ino_t inum)
281 {
282 	ufs_fsd_t *fsdp;
283 	ufs_acl_t *ufsaclp = NULL;
284 	int numacls;
285 	int curacl;
286 	struct type_counts_s {
287 		int nuser_objs;
288 		int ngroup_objs;
289 		int nother_objs;
290 		int nclass_objs;
291 		int ndef_user_objs;
292 		int ndef_group_objs;
293 		int ndef_other_objs;
294 		int ndef_class_objs;
295 		int nusers;
296 		int ngroups;
297 		int ndef_users;
298 		int ndef_groups;
299 	} type_counts[3];	/* indexed by FSD_ACL and FSD_DFACL */
300 	struct type_counts_s *tcp, *tcp_all, *tcp_def, *tcp_norm;
301 	int numdefs;
302 	caddr_t bad;
303 	caddr_t end = buf + len;
304 	int64_t recsz = 0;
305 	int64_t min_recsz = FSD_RECSZ(fsdp, sizeof (*fsdp));
306 	struct shadowclientinfo *sci;
307 	struct shadowclients *scc;
308 	fsck_ino_t target;
309 	int numtargets = 0;
310 
311 	/*
312 	 * check we have a non-zero length for this shadow inode
313 	 */
314 	if (len == 0) {
315 		pwarn("ACL I=%d HAS ZERO LENGTH\n", inum);
316 		return (1);
317 	}
318 
319 	(void) memset(type_counts, 0, sizeof (type_counts));
320 
321 	/* LINTED pointer cast alignment (aligned buffer always passed in) */
322 	for (fsdp = (ufs_fsd_t *)buf;
323 	    (caddr_t)fsdp < end;
324 	    /* LINTED as per the above */
325 	    fsdp = (ufs_fsd_t *)((caddr_t)fsdp + recsz)) {
326 
327 		recsz = FSD_RECSZ(fsdp, fsdp->fsd_size);
328 		if ((recsz < min_recsz) ||
329 		    (((caddr_t)fsdp + recsz) > (buf + len))) {
330 			pwarn("Bad FSD entry size %lld in shadow inode %d",
331 			    recsz, inum);
332 			if (reply("CLEAR SHADOW INODE") == 1) {
333 				freeino(inum, TI_PARENT);
334 			} else {
335 				/*
336 				 * Bad size can cause the kernel to
337 				 * go traipsing off into never-never land.
338 				 */
339 				iscorrupt = 1;
340 			}
341 			return (0);
342 		}
343 
344 		switch (fsdp->fsd_type) {
345 		case FSD_FREE:	/* ignore empty slots */
346 			break;
347 		case FSD_ACL:
348 		case FSD_DFACL:
349 			/*
350 			 * Subtract out the two ints in the fsd_type,
351 			 * leaving us just the size of fsd_data[].
352 			 */
353 			numacls = (fsdp->fsd_size - 2 * sizeof (int)) /
354 							sizeof (ufs_acl_t);
355 			tcp = &type_counts[fsdp->fsd_type];
356 			curacl = 0;
357 			/* LINTED pointer cast alignment */
358 			for (ufsaclp = (ufs_acl_t *)fsdp->fsd_data;
359 						numacls; ufsaclp++, curacl++) {
360 				switch (ufsaclp->acl_tag) {
361 				case USER_OBJ:		/* Owner */
362 					tcp->nuser_objs++;
363 					break;
364 				case GROUP_OBJ:		/* Group */
365 					tcp->ngroup_objs++;
366 					break;
367 				case OTHER_OBJ:		/* Other */
368 					tcp->nother_objs++;
369 					break;
370 				case CLASS_OBJ:		/* Mask */
371 					tcp->nclass_objs++;
372 					break;
373 				case DEF_USER_OBJ:	/* Default Owner */
374 					tcp->ndef_user_objs++;
375 					break;
376 				case DEF_GROUP_OBJ:	/* Default Group */
377 					tcp->ndef_group_objs++;
378 					break;
379 				case DEF_OTHER_OBJ:	/* Default Other */
380 					tcp->ndef_other_objs++;
381 					break;
382 				case DEF_CLASS_OBJ:	/* Default Mask */
383 					tcp->ndef_class_objs++;
384 					break;
385 				case USER:		/* Users */
386 					tcp->nusers++;
387 					break;
388 				case GROUP:		/* Groups */
389 					tcp->ngroups++;
390 					break;
391 				case DEF_USER:		/* Default Users */
392 					tcp->ndef_users++;
393 					break;
394 				case DEF_GROUP:		/* Default Groups */
395 					tcp->ndef_groups++;
396 					break;
397 				default:
398 					return (1);
399 				}
400 
401 				if ((ufsaclp->acl_perm & ~07) != 0) {
402 					/*
403 					 * Caller will report inode, etc
404 					 */
405 					pwarn("Bad permission 0%o in ACL\n",
406 					    ufsaclp->acl_perm);
407 					return (1);
408 				}
409 
410 				numacls--;
411 			}
412 			break;
413 		default:
414 			if (fsdp->fsd_type >= FSD_RESERVED3 &&
415 			    fsdp->fsd_type <= FSD_RESERVED7)
416 				bad = "Unexpected";
417 			else
418 				bad = "Unknown";
419 			pwarn("%s FSD type %d in shadow inode %d",
420 			    bad, fsdp->fsd_type, inum);
421 			/*
422 			 * This is relatively harmless, since the
423 			 * kernel will ignore any entries it doesn't
424 			 * recognize.  Don't bother with iscorrupt.
425 			 */
426 			if (preen) {
427 				(void) printf(" (IGNORED)\n");
428 			} else if (reply("IGNORE") == 0) {
429 				if (reply("CLEAR SHADOW INODE") == 1) {
430 					freeino(inum, TI_PARENT);
431 				}
432 				return (0);
433 			}
434 			break;
435 		}
436 	}
437 	if ((caddr_t)fsdp != (buf + len)) {
438 		return (1);
439 	}
440 
441 	/* If we didn't find any acls, ignore the unknown attribute */
442 	if (ufsaclp == NULL)
443 		return (0);
444 
445 	/*
446 	 * Should only have default ACLs in FSD_DFACL records.
447 	 * However, the kernel can handle it, so just report that
448 	 * something odd might be going on.
449 	 */
450 	tcp = &type_counts[FSD_DFACL];
451 	if (verbose &&
452 	    (tcp->nuser_objs != 0 ||
453 	    tcp->ngroup_objs != 0 ||
454 	    tcp->nother_objs != 0 ||
455 	    tcp->nclass_objs != 0 ||
456 	    tcp->nusers != 0 ||
457 	    tcp->ngroups != 0)) {
458 		(void) printf("NOTE: ACL I=%d has miscategorized ACLs.  ",
459 		    inum);
460 		(void) printf("This is harmless, but not normal.\n");
461 	}
462 
463 	/*
464 	 * Similarly for default ACLs in FSD_ACL records.
465 	 */
466 	tcp = &type_counts[FSD_ACL];
467 	if (verbose &&
468 	    (tcp->ndef_user_objs != 0 ||
469 	    tcp->ndef_group_objs != 0 ||
470 	    tcp->ndef_other_objs != 0 ||
471 	    tcp->ndef_class_objs != 0 ||
472 	    tcp->ndef_users != 0 ||
473 	    tcp->ndef_groups != 0)) {
474 		(void) printf("NOTE: ACL I=%d has miscategorized ACLs.",
475 		    inum);
476 		(void) printf("  This is harmless, but not normal.\n");
477 	}
478 
479 	/*
480 	 * Get consolidated totals, now that we're done with checking
481 	 * the segregation above.  Assumes that neither FSD_ACL nor
482 	 * FSD_DFACL are zero.
483 	 */
484 	tcp_all = &type_counts[0];
485 	tcp_norm = &type_counts[FSD_ACL];
486 	tcp_def = &type_counts[FSD_DFACL];
487 
488 	tcp_all->nuser_objs = tcp_def->nuser_objs + tcp_norm->nuser_objs;
489 	tcp_all->ngroup_objs = tcp_def->ngroup_objs + tcp_norm->ngroup_objs;
490 	tcp_all->nother_objs = tcp_def->nother_objs + tcp_norm->nother_objs;
491 	tcp_all->nclass_objs = tcp_def->nclass_objs + tcp_norm->nclass_objs;
492 	tcp_all->ndef_user_objs =
493 		tcp_def->ndef_user_objs + tcp_norm->ndef_user_objs;
494 	tcp_all->ndef_group_objs =
495 		tcp_def->ndef_group_objs + tcp_norm->ndef_group_objs;
496 	tcp_all->ndef_other_objs =
497 		tcp_def->ndef_other_objs + tcp_norm->ndef_other_objs;
498 	tcp_all->ndef_class_objs =
499 		tcp_def->ndef_class_objs + tcp_norm->ndef_class_objs;
500 	tcp_all->nusers = tcp_def->nusers + tcp_norm->nusers;
501 	tcp_all->ngroups = tcp_def->ngroups + tcp_norm->ngroups;
502 	tcp_all->ndef_users = tcp_def->ndef_users + tcp_norm->ndef_users;
503 	tcp_all->ndef_groups = tcp_def->ndef_groups + tcp_norm->ndef_groups;
504 
505 	/*
506 	 * Check relationships among acls
507 	 */
508 	if (tcp_all->nuser_objs != 1 ||
509 	    tcp_all->ngroup_objs != 1 ||
510 	    tcp_all->nother_objs != 1 ||
511 	    tcp_all->nclass_objs > 1) {
512 		return (1);
513 	}
514 
515 	if (tcp_all->ngroups && !tcp_all->nclass_objs) {
516 		return (1);
517 	}
518 
519 	if (tcp_all->ndef_user_objs > 1 ||
520 	    tcp_all->ndef_group_objs > 1 ||
521 	    tcp_all->ndef_other_objs > 1 ||
522 	    tcp_all->ndef_class_objs > 1) {
523 		return (1);
524 	}
525 
526 	/*
527 	 * Check relationships among default acls
528 	 */
529 	numdefs = tcp_all->ndef_other_objs + tcp_all->ndef_user_objs +
530 		tcp_all->ndef_group_objs;
531 
532 	if (numdefs != 0 && numdefs != 3) {
533 		return (1);
534 	}
535 
536 	/*
537 	 * If there are default acls, then the shadow inode's clients
538 	 * must be a directory or an xattr directory.
539 	 */
540 	if (numdefs != 0) {
541 		/* This is an ACL so find it's clients */
542 		for (sci = shadowclientinfo; sci != NULL; sci = sci->next)
543 			if (sci->shadow == inum)
544 			    break;
545 		if ((sci ==  NULL) || (sci->clients == NULL))
546 			return (1);
547 
548 		/* Got shadow info, now look at clients */
549 		for (scc = sci->clients; scc != NULL; scc = scc->next) {
550 			for (numtargets = 0; numtargets < scc->nclients;
551 			    numtargets++) {
552 				target = scc->client[numtargets];
553 				if (!INO_IS_DVALID(target))
554 					return (1);
555 			}
556 		}
557 	}
558 
559 	if (tcp_all->ndef_groups && !tcp_all->ndef_class_objs) {
560 		return (1);
561 	}
562 
563 	if ((tcp_all->ndef_users || tcp_all->ndef_groups) &&
564 	    ((numdefs != 3) && !tcp_all->ndef_class_objs)) {
565 		return (1);
566 	}
567 
568 	return (0);
569 }
570