xref: /freebsd-14-stable/sys/ufs/ffs/ffs_subr.c (revision ae15f8ceaa85e6e02e056aa151d965d82e1f200e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)ffs_subr.c	8.5 (Berkeley) 3/21/95
32  */
33 
34 #include <sys/cdefs.h>
35 #include <sys/param.h>
36 #include <sys/endian.h>
37 #include <sys/limits.h>
38 
39 #ifndef _KERNEL
40 #include <stdbool.h>
41 #include <stdio.h>
42 #include <string.h>
43 #include <stdlib.h>
44 #include <time.h>
45 #include <sys/errno.h>
46 #include <ufs/ufs/dinode.h>
47 #include <ufs/ffs/fs.h>
48 
49 uint32_t calculate_crc32c(uint32_t, const void *, size_t);
50 uint32_t ffs_calc_sbhash(struct fs *);
51 struct malloc_type;
52 #define UFS_MALLOC(size, type, flags) malloc(size)
53 #define UFS_FREE(ptr, type) free(ptr)
54 #define maxphys MAXPHYS
55 
56 #else /* _KERNEL */
57 #include <sys/systm.h>
58 #include <sys/gsb_crc32.h>
59 #include <sys/lock.h>
60 #include <sys/malloc.h>
61 #include <sys/mount.h>
62 #include <sys/vnode.h>
63 #include <sys/bio.h>
64 #include <sys/buf.h>
65 #include <sys/ucred.h>
66 #include <sys/sysctl.h>
67 
68 #include <ufs/ufs/quota.h>
69 #include <ufs/ufs/inode.h>
70 #include <ufs/ufs/extattr.h>
71 #include <ufs/ufs/ufsmount.h>
72 #include <ufs/ufs/ufs_extern.h>
73 #include <ufs/ffs/ffs_extern.h>
74 #include <ufs/ffs/fs.h>
75 
76 #define UFS_MALLOC(size, type, flags) malloc(size, type, flags)
77 #define UFS_FREE(ptr, type) free(ptr, type)
78 
79 #endif /* _KERNEL */
80 
81 /*
82  * Verify an inode check-hash.
83  */
84 int
ffs_verify_dinode_ckhash(struct fs * fs,struct ufs2_dinode * dip)85 ffs_verify_dinode_ckhash(struct fs *fs, struct ufs2_dinode *dip)
86 {
87 	uint32_t ckhash, save_ckhash;
88 
89 	/*
90 	 * Return success if unallocated or we are not doing inode check-hash.
91 	 */
92 	if (dip->di_mode == 0 || (fs->fs_metackhash & CK_INODE) == 0)
93 		return (0);
94 	/*
95 	 * Exclude di_ckhash from the crc32 calculation, e.g., always use
96 	 * a check-hash value of zero when calculating the check-hash.
97 	 */
98 	save_ckhash = dip->di_ckhash;
99 	dip->di_ckhash = 0;
100 	ckhash = calculate_crc32c(~0L, (void *)dip, sizeof(*dip));
101 	dip->di_ckhash = save_ckhash;
102 	if (save_ckhash == ckhash)
103 		return (0);
104 	return (EINVAL);
105 }
106 
107 /*
108  * Update an inode check-hash.
109  */
110 void
ffs_update_dinode_ckhash(struct fs * fs,struct ufs2_dinode * dip)111 ffs_update_dinode_ckhash(struct fs *fs, struct ufs2_dinode *dip)
112 {
113 
114 	if (dip->di_mode == 0 || (fs->fs_metackhash & CK_INODE) == 0)
115 		return;
116 	/*
117 	 * Exclude old di_ckhash from the crc32 calculation, e.g., always use
118 	 * a check-hash value of zero when calculating the new check-hash.
119 	 */
120 	dip->di_ckhash = 0;
121 	dip->di_ckhash = calculate_crc32c(~0L, (void *)dip, sizeof(*dip));
122 }
123 
124 /*
125  * These are the low-level functions that actually read and write
126  * the superblock and its associated data.
127  */
128 static off_t sblock_try[] = SBLOCKSEARCH;
129 static int readsuper(void *, struct fs **, off_t, int,
130 	int (*)(void *, off_t, void **, int));
131 static void ffs_oldfscompat_read(struct fs *, ufs2_daddr_t);
132 static int validate_sblock(struct fs *, int);
133 
134 /*
135  * Read a superblock from the devfd device.
136  *
137  * If an alternate superblock is specified, it is read. Otherwise the
138  * set of locations given in the SBLOCKSEARCH list is searched for a
139  * superblock. Memory is allocated for the superblock by the readfunc and
140  * is returned. If filltype is non-NULL, additional memory is allocated
141  * of type filltype and filled in with the superblock summary information.
142  * All memory is freed when any error is returned.
143  *
144  * If a superblock is found, zero is returned. Otherwise one of the
145  * following error values is returned:
146  *     EIO: non-existent or truncated superblock.
147  *     EIO: error reading summary information.
148  *     ENOENT: no usable known superblock found.
149  *     EILSEQ: filesystem with wrong byte order found.
150  *     ENOMEM: failed to allocate space for the superblock.
151  *     EINVAL: The previous newfs operation on this volume did not complete.
152  *         The administrator must complete newfs before using this volume.
153  */
154 int
ffs_sbget(void * devfd,struct fs ** fsp,off_t sblock,int flags,struct malloc_type * filltype,int (* readfunc)(void * devfd,off_t loc,void ** bufp,int size))155 ffs_sbget(void *devfd, struct fs **fsp, off_t sblock, int flags,
156     struct malloc_type *filltype,
157     int (*readfunc)(void *devfd, off_t loc, void **bufp, int size))
158 {
159 	struct fs *fs;
160 	struct fs_summary_info *fs_si;
161 	int i, error;
162 	uint64_t size, blks;
163 	uint8_t *space;
164 	int32_t *lp;
165 	char *buf;
166 
167 	fs = NULL;
168 	*fsp = NULL;
169 	if (sblock != UFS_STDSB) {
170 		if ((error = readsuper(devfd, &fs, sblock,
171 		    flags | UFS_ALTSBLK, readfunc)) != 0) {
172 			if (fs != NULL)
173 				UFS_FREE(fs, filltype);
174 			return (error);
175 		}
176 	} else {
177 		for (i = 0; sblock_try[i] != -1; i++) {
178 			if ((error = readsuper(devfd, &fs, sblock_try[i],
179 			     flags, readfunc)) == 0) {
180 				if ((flags & UFS_NOCSUM) != 0) {
181 					*fsp = fs;
182 					return (0);
183 				}
184 				break;
185 			}
186 			if (fs != NULL) {
187 				UFS_FREE(fs, filltype);
188 				fs = NULL;
189 			}
190 			if (error == ENOENT)
191 				continue;
192 			return (error);
193 		}
194 		if (sblock_try[i] == -1)
195 			return (ENOENT);
196 	}
197 	/*
198 	 * Read in the superblock summary information.
199 	 */
200 	size = fs->fs_cssize;
201 	blks = howmany(size, fs->fs_fsize);
202 	if (fs->fs_contigsumsize > 0)
203 		size += fs->fs_ncg * sizeof(int32_t);
204 	size += fs->fs_ncg * sizeof(uint8_t);
205 	if ((fs_si = UFS_MALLOC(sizeof(*fs_si), filltype, M_NOWAIT)) == NULL) {
206 		UFS_FREE(fs, filltype);
207 		return (ENOMEM);
208 	}
209 	bzero(fs_si, sizeof(*fs_si));
210 	fs->fs_si = fs_si;
211 	if ((space = UFS_MALLOC(size, filltype, M_NOWAIT)) == NULL) {
212 		UFS_FREE(fs->fs_si, filltype);
213 		UFS_FREE(fs, filltype);
214 		return (ENOMEM);
215 	}
216 	fs->fs_csp = (struct csum *)space;
217 	for (i = 0; i < blks; i += fs->fs_frag) {
218 		size = fs->fs_bsize;
219 		if (i + fs->fs_frag > blks)
220 			size = (blks - i) * fs->fs_fsize;
221 		buf = NULL;
222 		error = (*readfunc)(devfd,
223 		    dbtob(fsbtodb(fs, fs->fs_csaddr + i)), (void **)&buf, size);
224 		if (error) {
225 			if (buf != NULL)
226 				UFS_FREE(buf, filltype);
227 			UFS_FREE(fs->fs_csp, filltype);
228 			UFS_FREE(fs->fs_si, filltype);
229 			UFS_FREE(fs, filltype);
230 			return (error);
231 		}
232 		memcpy(space, buf, size);
233 		UFS_FREE(buf, filltype);
234 		space += size;
235 	}
236 	if (fs->fs_contigsumsize > 0) {
237 		fs->fs_maxcluster = lp = (int32_t *)space;
238 		for (i = 0; i < fs->fs_ncg; i++)
239 			*lp++ = fs->fs_contigsumsize;
240 		space = (uint8_t *)lp;
241 	}
242 	size = fs->fs_ncg * sizeof(uint8_t);
243 	fs->fs_contigdirs = (uint8_t *)space;
244 	bzero(fs->fs_contigdirs, size);
245 	*fsp = fs;
246 	return (0);
247 }
248 
249 /*
250  * Try to read a superblock from the location specified by sblockloc.
251  * Return zero on success or an errno on failure.
252  */
253 static int
readsuper(void * devfd,struct fs ** fsp,off_t sblockloc,int flags,int (* readfunc)(void * devfd,off_t loc,void ** bufp,int size))254 readsuper(void *devfd, struct fs **fsp, off_t sblockloc, int flags,
255     int (*readfunc)(void *devfd, off_t loc, void **bufp, int size))
256 {
257 	struct fs *fs;
258 	int error, res;
259 	uint32_t ckhash;
260 
261 	error = (*readfunc)(devfd, sblockloc, (void **)fsp, SBLOCKSIZE);
262 	if (error != 0)
263 		return (error);
264 	fs = *fsp;
265 	if (fs->fs_magic == FS_BAD_MAGIC)
266 		return (EINVAL);
267 	/*
268 	 * For UFS1 with a 65536 block size, the first backup superblock
269 	 * is at the same location as the UFS2 superblock. Since SBLOCK_UFS2
270 	 * is the first location checked, the first backup is the superblock
271 	 * that will be accessed. Here we fail the lookup so that we can
272 	 * retry with the correct location for the UFS1 superblock.
273 	 */
274 	if (fs->fs_magic == FS_UFS1_MAGIC && (flags & UFS_ALTSBLK) == 0 &&
275 	    fs->fs_bsize == SBLOCK_UFS2 && sblockloc == SBLOCK_UFS2)
276 		return (ENOENT);
277 	ffs_oldfscompat_read(fs, sblockloc);
278 	if ((error = validate_sblock(fs, flags)) > 0)
279 		return (error);
280 	/*
281 	 * If the filesystem has been run on a kernel without
282 	 * metadata check hashes, disable them.
283 	 */
284 	if ((fs->fs_flags & FS_METACKHASH) == 0)
285 		fs->fs_metackhash = 0;
286 	/*
287 	 * Clear any check-hashes that are not maintained
288 	 * by this kernel. Also clear any unsupported flags.
289 	 */
290 	fs->fs_metackhash &= CK_SUPPORTED;
291 	fs->fs_flags &= FS_SUPPORTED;
292 	if (fs->fs_ckhash != (ckhash = ffs_calc_sbhash(fs))) {
293 		if ((flags & (UFS_NOMSG | UFS_NOHASHFAIL)) ==
294 		    (UFS_NOMSG | UFS_NOHASHFAIL))
295 			return (0);
296 		if ((flags & UFS_NOMSG) != 0)
297 			return (EINTEGRITY);
298 #ifdef _KERNEL
299 		res = uprintf("Superblock check-hash failed: recorded "
300 		    "check-hash 0x%x != computed check-hash 0x%x%s\n",
301 		    fs->fs_ckhash, ckhash,
302 		    (flags & UFS_NOHASHFAIL) != 0 ? " (Ignored)" : "");
303 #else
304 		res = 0;
305 #endif
306 		/*
307 		 * Print check-hash failure if no controlling terminal
308 		 * in kernel or always if in user-mode (libufs).
309 		 */
310 		if (res == 0)
311 			printf("Superblock check-hash failed: recorded "
312 			    "check-hash 0x%x != computed check-hash "
313 			    "0x%x%s\n", fs->fs_ckhash, ckhash,
314 			    (flags & UFS_NOHASHFAIL) ? " (Ignored)" : "");
315 		if ((flags & UFS_NOHASHFAIL) != 0)
316 			return (0);
317 		return (EINTEGRITY);
318 	}
319 	/* Have to set for old filesystems that predate this field */
320 	fs->fs_sblockactualloc = sblockloc;
321 	/* Not yet any summary information */
322 	fs->fs_si = NULL;
323 	return (0);
324 }
325 
326 /*
327  * Sanity checks for loading old filesystem superblocks.
328  * See ffs_oldfscompat_write below for unwound actions.
329  *
330  * XXX - Parts get retired eventually.
331  * Unfortunately new bits get added.
332  */
333 static void
ffs_oldfscompat_read(struct fs * fs,ufs2_daddr_t sblockloc)334 ffs_oldfscompat_read(struct fs *fs, ufs2_daddr_t sblockloc)
335 {
336 	uint64_t maxfilesize;
337 
338 	/*
339 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
340 	 */
341 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
342 		fs->fs_flags = fs->fs_old_flags;
343 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
344 		fs->fs_sblockloc = sblockloc;
345 	}
346 	switch (fs->fs_magic) {
347 	case FS_UFS2_MAGIC:
348 		/* No changes for now */
349 		break;
350 
351 	case FS_UFS1_MAGIC:
352 		/*
353 		 * If not yet done, update UFS1 superblock with new wider fields
354 		 */
355 		if (fs->fs_maxbsize != fs->fs_bsize) {
356 			fs->fs_maxbsize = fs->fs_bsize;
357 			fs->fs_time = fs->fs_old_time;
358 			fs->fs_size = fs->fs_old_size;
359 			fs->fs_dsize = fs->fs_old_dsize;
360 			fs->fs_csaddr = fs->fs_old_csaddr;
361 			fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
362 			fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
363 			fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
364 			fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
365 		}
366 		if (fs->fs_old_inodefmt < FS_44INODEFMT) {
367 			fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
368 			fs->fs_qbmask = ~fs->fs_bmask;
369 			fs->fs_qfmask = ~fs->fs_fmask;
370 		}
371 		fs->fs_save_maxfilesize = fs->fs_maxfilesize;
372 		maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
373 		if (fs->fs_maxfilesize > maxfilesize)
374 			fs->fs_maxfilesize = maxfilesize;
375 		break;
376 	}
377 	/* Compatibility for old filesystems */
378 	if (fs->fs_avgfilesize <= 0)
379 		fs->fs_avgfilesize = AVFILESIZ;
380 	if (fs->fs_avgfpdir <= 0)
381 		fs->fs_avgfpdir = AFPDIR;
382 }
383 
384 /*
385  * Unwinding superblock updates for old filesystems.
386  * See ffs_oldfscompat_read above for details.
387  *
388  * XXX - Parts get retired eventually.
389  * Unfortunately new bits get added.
390  */
391 void
ffs_oldfscompat_write(struct fs * fs)392 ffs_oldfscompat_write(struct fs *fs)
393 {
394 
395 	switch (fs->fs_magic) {
396 	case FS_UFS1_MAGIC:
397 		if (fs->fs_sblockloc != SBLOCK_UFS1 &&
398 		    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
399 			printf(
400 			"WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
401 			    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
402 			fs->fs_sblockloc = SBLOCK_UFS1;
403 		}
404 		/*
405 		 * Copy back UFS2 updated fields that UFS1 inspects.
406 		 */
407 		fs->fs_old_time = fs->fs_time;
408 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
409 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
410 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
411 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
412 		if (fs->fs_save_maxfilesize != 0)
413 			fs->fs_maxfilesize = fs->fs_save_maxfilesize;
414 		break;
415 	case FS_UFS2_MAGIC:
416 		if (fs->fs_sblockloc != SBLOCK_UFS2 &&
417 		    (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
418 			printf(
419 			"WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
420 			    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
421 			fs->fs_sblockloc = SBLOCK_UFS2;
422 		}
423 		break;
424 	}
425 }
426 
427 /*
428  * Sanity checks for loading old filesystem inodes.
429  *
430  * XXX - Parts get retired eventually.
431  * Unfortunately new bits get added.
432  */
433 static int prttimechgs = 0;
434 #ifdef _KERNEL
435 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
436     "FFS filesystem");
437 SYSCTL_INT(_vfs_ffs, OID_AUTO, prttimechgs, CTLFLAG_RWTUN, &prttimechgs, 0,
438     "print UFS1 time changes made to inodes");
439 #endif /* _KERNEL */
440 bool
ffs_oldfscompat_inode_read(struct fs * fs,union dinodep dp,time_t now)441 ffs_oldfscompat_inode_read(struct fs *fs, union dinodep dp, time_t now)
442 {
443 	bool change;
444 
445 	change = false;
446 	switch (fs->fs_magic) {
447 	case FS_UFS2_MAGIC:
448 		/* No changes for now */
449 		break;
450 
451 	case FS_UFS1_MAGIC:
452 		/*
453 		 * With the change to unsigned time values in UFS1, times set
454 		 * before Jan 1, 1970 will appear to be in the future. Check
455 		 * for future times and set them to be the current time.
456 		 */
457 		if (dp.dp1->di_ctime > now) {
458 			if (prttimechgs)
459 				printf("ctime %ud changed to %ld\n",
460 				    dp.dp1->di_ctime, (long)now);
461 			dp.dp1->di_ctime = now;
462 			change = true;
463 		}
464 		if (dp.dp1->di_mtime > now) {
465 			if (prttimechgs)
466 				printf("mtime %ud changed to %ld\n",
467 				    dp.dp1->di_mtime, (long)now);
468 			dp.dp1->di_mtime = now;
469 			dp.dp1->di_ctime = now;
470 			change = true;
471 		}
472 		if (dp.dp1->di_atime > now) {
473 			if (prttimechgs)
474 				printf("atime %ud changed to %ld\n",
475 				    dp.dp1->di_atime, (long)now);
476 			dp.dp1->di_atime = now;
477 			dp.dp1->di_ctime = now;
478 			change = true;
479 		}
480 		break;
481 	}
482 	return (change);
483 }
484 
485 /*
486  * Verify the filesystem values.
487  */
488 #define ILOG2(num)	(fls(num) - 1)
489 #ifdef STANDALONE_SMALL
490 #define MPRINT(...)	do { } while (0)
491 #else
492 #define MPRINT(...)	if (prtmsg) printf(__VA_ARGS__)
493 #endif
494 #define FCHK(lhs, op, rhs, fmt)						\
495 	if (lhs op rhs) {						\
496 		MPRINT("UFS%d superblock failed: %s (" #fmt ") %s %s ("	\
497 		    #fmt ")\n", fs->fs_magic == FS_UFS1_MAGIC ? 1 : 2,	\
498 		    #lhs, (intmax_t)lhs, #op, #rhs, (intmax_t)rhs);	\
499 		if (error < 0)						\
500 			return (ENOENT);				\
501 		if (error == 0)						\
502 			error = ENOENT;					\
503 	}
504 #define WCHK(lhs, op, rhs, fmt)						\
505 	if (lhs op rhs) {						\
506 		MPRINT("UFS%d superblock failed: %s (" #fmt ") %s %s ("	\
507 		    #fmt ")%s\n", fs->fs_magic == FS_UFS1_MAGIC ? 1 : 2,\
508 		    #lhs, (intmax_t)lhs, #op, #rhs, (intmax_t)rhs, wmsg);\
509 		if (error == 0)						\
510 			error = warnerr;				\
511 		if (warnerr == 0)					\
512 			lhs = rhs;					\
513 	}
514 #define FCHK2(lhs1, op1, rhs1, lhs2, op2, rhs2, fmt)			\
515 	if (lhs1 op1 rhs1 && lhs2 op2 rhs2) {				\
516 		MPRINT("UFS%d superblock failed: %s (" #fmt ") %s %s ("	\
517 		    #fmt ") && %s (" #fmt ") %s %s (" #fmt ")\n",	\
518 		    fs->fs_magic == FS_UFS1_MAGIC ? 1 : 2, #lhs1,	\
519 		    (intmax_t)lhs1, #op1, #rhs1, (intmax_t)rhs1, #lhs2,	\
520 		    (intmax_t)lhs2, #op2, #rhs2, (intmax_t)rhs2);	\
521 		if (error < 0)						\
522 			return (ENOENT);				\
523 		if (error == 0)						\
524 			error = ENOENT;					\
525 	}
526 
527 static int
validate_sblock(struct fs * fs,int flags)528 validate_sblock(struct fs *fs, int flags)
529 {
530 	uint64_t i, sectorsize;
531 	uint64_t maxfilesize, sizepb;
532 	int error, prtmsg, warnerr;
533 	char *wmsg;
534 
535 	error = 0;
536 	sectorsize = dbtob(1);
537 	prtmsg = ((flags & UFS_NOMSG) == 0);
538 	warnerr = (flags & UFS_NOWARNFAIL) == UFS_NOWARNFAIL ? 0 : ENOENT;
539 	wmsg = warnerr ? "" : " (Ignored)";
540 	/*
541 	 * Check for endian mismatch between machine and filesystem.
542 	 */
543 	if (((fs->fs_magic != FS_UFS2_MAGIC) &&
544 	    (bswap32(fs->fs_magic) == FS_UFS2_MAGIC)) ||
545 	    ((fs->fs_magic != FS_UFS1_MAGIC) &&
546 	    (bswap32(fs->fs_magic) == FS_UFS1_MAGIC))) {
547 		MPRINT("UFS superblock failed due to endian mismatch "
548 		    "between machine and filesystem\n");
549 		return(EILSEQ);
550 	}
551 	/*
552 	 * If just validating for recovery, then do just the minimal
553 	 * checks needed for the superblock fields needed to find
554 	 * alternate superblocks.
555 	 */
556 	if ((flags & UFS_FSRONLY) == UFS_FSRONLY &&
557 	    (fs->fs_magic == FS_UFS1_MAGIC || fs->fs_magic == FS_UFS2_MAGIC)) {
558 		error = -1; /* fail on first error */
559 		if (fs->fs_magic == FS_UFS2_MAGIC) {
560 			FCHK(fs->fs_sblockloc, !=, SBLOCK_UFS2, %#jx);
561 		} else if (fs->fs_magic == FS_UFS1_MAGIC) {
562 			FCHK(fs->fs_sblockloc, <, 0, %jd);
563 			FCHK(fs->fs_sblockloc, >, SBLOCK_UFS1, %jd);
564 		}
565 		FCHK(fs->fs_frag, <, 1, %jd);
566 		FCHK(fs->fs_frag, >, MAXFRAG, %jd);
567 		FCHK(fs->fs_bsize, <, MINBSIZE, %jd);
568 		FCHK(fs->fs_bsize, >, MAXBSIZE, %jd);
569 		FCHK(fs->fs_bsize, <, roundup(sizeof(struct fs), DEV_BSIZE),
570 		    %jd);
571 		FCHK(fs->fs_fsize, <, sectorsize, %jd);
572 		FCHK(fs->fs_fsize * fs->fs_frag, !=, fs->fs_bsize, %jd);
573 		FCHK(powerof2(fs->fs_fsize), ==, 0, %jd);
574 		FCHK(fs->fs_sbsize, >, SBLOCKSIZE, %jd);
575 		FCHK(fs->fs_sbsize, <, (signed)sizeof(struct fs), %jd);
576 		FCHK(fs->fs_sbsize % sectorsize, !=, 0, %jd);
577 		FCHK(fs->fs_fpg, <, 3 * fs->fs_frag, %jd);
578 		FCHK(fs->fs_ncg, <, 1, %jd);
579 		FCHK(fs->fs_fsbtodb, !=, ILOG2(fs->fs_fsize / sectorsize), %jd);
580 		FCHK(fs->fs_old_cgoffset, <, 0, %jd);
581 		FCHK2(fs->fs_old_cgoffset, >, 0, ~fs->fs_old_cgmask, <, 0, %jd);
582 		FCHK(fs->fs_old_cgoffset * (~fs->fs_old_cgmask), >, fs->fs_fpg,
583 		    %jd);
584 		FCHK(fs->fs_sblkno, !=, roundup(
585 		    howmany(fs->fs_sblockloc + SBLOCKSIZE, fs->fs_fsize),
586 		    fs->fs_frag), %jd);
587 		FCHK(CGSIZE(fs), >, fs->fs_bsize, %jd);
588 		/* Only need to validate these if reading in csum data */
589 		if ((flags & UFS_NOCSUM) != 0)
590 			return (error);
591 		FCHK((uint64_t)fs->fs_ipg * fs->fs_ncg, >,
592 		    (((int64_t)(1)) << 32) - INOPB(fs), %jd);
593 		FCHK(fs->fs_cstotal.cs_nifree, <, 0, %jd);
594 		FCHK(fs->fs_cstotal.cs_nifree, >,
595 		    (uint64_t)fs->fs_ipg * fs->fs_ncg, %jd);
596 		FCHK(fs->fs_cstotal.cs_ndir, >,
597 		    ((uint64_t)fs->fs_ipg * fs->fs_ncg) -
598 		    fs->fs_cstotal.cs_nifree, %jd);
599 		FCHK(fs->fs_size, <, 8 * fs->fs_frag, %jd);
600 		FCHK(fs->fs_size, <=, ((int64_t)fs->fs_ncg - 1) * fs->fs_fpg,
601 		    %jd);
602 		FCHK(fs->fs_size, >, (int64_t)fs->fs_ncg * fs->fs_fpg, %jd);
603 		FCHK(fs->fs_csaddr, <, 0, %jd);
604 		FCHK(fs->fs_cssize, !=,
605 		    fragroundup(fs, fs->fs_ncg * sizeof(struct csum)), %jd);
606 		FCHK(fs->fs_csaddr + howmany(fs->fs_cssize, fs->fs_fsize), >,
607 		    fs->fs_size, %jd);
608 		FCHK(fs->fs_csaddr, <, cgdmin(fs, dtog(fs, fs->fs_csaddr)),
609 		    %jd);
610 		FCHK(dtog(fs, fs->fs_csaddr + howmany(fs->fs_cssize,
611 		    fs->fs_fsize)), >, dtog(fs, fs->fs_csaddr), %jd);
612 		return (error);
613 	}
614 	if (fs->fs_magic == FS_UFS2_MAGIC) {
615 		if ((flags & UFS_ALTSBLK) == 0)
616 			FCHK2(fs->fs_sblockactualloc, !=, SBLOCK_UFS2,
617 			    fs->fs_sblockactualloc, !=, 0, %jd);
618 		FCHK(fs->fs_sblockloc, !=, SBLOCK_UFS2, %#jx);
619 		FCHK(fs->fs_maxsymlinklen, !=, ((UFS_NDADDR + UFS_NIADDR) *
620 			sizeof(ufs2_daddr_t)), %jd);
621 		FCHK(fs->fs_nindir, !=, fs->fs_bsize / sizeof(ufs2_daddr_t),
622 		    %jd);
623 		FCHK(fs->fs_inopb, !=,
624 		    fs->fs_bsize / sizeof(struct ufs2_dinode), %jd);
625 	} else if (fs->fs_magic == FS_UFS1_MAGIC) {
626 		if ((flags & UFS_ALTSBLK) == 0)
627 			FCHK(fs->fs_sblockactualloc, >, SBLOCK_UFS1, %jd);
628 		FCHK(fs->fs_sblockloc, <, 0, %jd);
629 		FCHK(fs->fs_sblockloc, >, SBLOCK_UFS1, %jd);
630 		FCHK(fs->fs_nindir, !=, fs->fs_bsize / sizeof(ufs1_daddr_t),
631 		    %jd);
632 		FCHK(fs->fs_inopb, !=,
633 		    fs->fs_bsize / sizeof(struct ufs1_dinode), %jd);
634 		FCHK(fs->fs_maxsymlinklen, !=, ((UFS_NDADDR + UFS_NIADDR) *
635 			sizeof(ufs1_daddr_t)), %jd);
636 		WCHK(fs->fs_old_inodefmt, !=, FS_44INODEFMT, %jd);
637 		WCHK(fs->fs_old_rotdelay, !=, 0, %jd);
638 		WCHK(fs->fs_old_rps, !=, 60, %jd);
639 		WCHK(fs->fs_old_nspf, !=, fs->fs_fsize / sectorsize, %jd);
640 		WCHK(fs->fs_old_interleave, !=, 1, %jd);
641 		WCHK(fs->fs_old_trackskew, !=, 0, %jd);
642 		WCHK(fs->fs_old_cpc, !=, 0, %jd);
643 		WCHK(fs->fs_old_postblformat, !=, 1, %jd);
644 		FCHK(fs->fs_old_nrpos, !=, 1, %jd);
645 		WCHK(fs->fs_old_nsect, !=, fs->fs_old_spc, %jd);
646 		WCHK(fs->fs_old_npsect, !=, fs->fs_old_spc, %jd);
647 	} else {
648 		/* Bad magic number, so assume not a superblock */
649 		return (ENOENT);
650 	}
651 	FCHK(fs->fs_bsize, <, MINBSIZE, %jd);
652 	FCHK(fs->fs_bsize, >, MAXBSIZE, %jd);
653 	FCHK(fs->fs_bsize, <, roundup(sizeof(struct fs), DEV_BSIZE), %jd);
654 	FCHK(powerof2(fs->fs_bsize), ==, 0, %jd);
655 	FCHK(fs->fs_frag, <, 1, %jd);
656 	FCHK(fs->fs_frag, >, MAXFRAG, %jd);
657 	FCHK(fs->fs_frag, !=, numfrags(fs, fs->fs_bsize), %jd);
658 	FCHK(fs->fs_fsize, <, sectorsize, %jd);
659 	FCHK(fs->fs_fsize * fs->fs_frag, !=, fs->fs_bsize, %jd);
660 	FCHK(powerof2(fs->fs_fsize), ==, 0, %jd);
661 	FCHK(fs->fs_fpg, <, 3 * fs->fs_frag, %jd);
662 	FCHK(fs->fs_ncg, <, 1, %jd);
663 	FCHK(fs->fs_ipg, <, fs->fs_inopb, %jd);
664 	FCHK((uint64_t)fs->fs_ipg * fs->fs_ncg, >,
665 	    (((int64_t)(1)) << 32) - INOPB(fs), %jd);
666 	FCHK(fs->fs_cstotal.cs_nifree, <, 0, %jd);
667 	FCHK(fs->fs_cstotal.cs_nifree, >, (uint64_t)fs->fs_ipg * fs->fs_ncg,
668 	    %jd);
669 	FCHK(fs->fs_cstotal.cs_ndir, <, 0, %jd);
670 	FCHK(fs->fs_cstotal.cs_ndir, >,
671 	    ((uint64_t)fs->fs_ipg * fs->fs_ncg) - fs->fs_cstotal.cs_nifree,
672 	    %jd);
673 	FCHK(fs->fs_sbsize, >, SBLOCKSIZE, %jd);
674 	FCHK(fs->fs_sbsize, <, (signed)sizeof(struct fs), %jd);
675 	/* fix for misconfigured filesystems */
676 	if (fs->fs_maxbsize == 0)
677 		fs->fs_maxbsize = fs->fs_bsize;
678 	FCHK(fs->fs_maxbsize, <, fs->fs_bsize, %jd);
679 	FCHK(powerof2(fs->fs_maxbsize), ==, 0, %jd);
680 	FCHK(fs->fs_maxbsize, >, FS_MAXCONTIG * fs->fs_bsize, %jd);
681 	FCHK(fs->fs_bmask, !=, ~(fs->fs_bsize - 1), %#jx);
682 	FCHK(fs->fs_fmask, !=, ~(fs->fs_fsize - 1), %#jx);
683 	FCHK(fs->fs_qbmask, !=, ~fs->fs_bmask, %#jx);
684 	FCHK(fs->fs_qfmask, !=, ~fs->fs_fmask, %#jx);
685 	FCHK(fs->fs_bshift, !=, ILOG2(fs->fs_bsize), %jd);
686 	FCHK(fs->fs_fshift, !=, ILOG2(fs->fs_fsize), %jd);
687 	FCHK(fs->fs_fragshift, !=, ILOG2(fs->fs_frag), %jd);
688 	FCHK(fs->fs_fsbtodb, !=, ILOG2(fs->fs_fsize / sectorsize), %jd);
689 	FCHK(fs->fs_old_cgoffset, <, 0, %jd);
690 	FCHK2(fs->fs_old_cgoffset, >, 0, ~fs->fs_old_cgmask, <, 0, %jd);
691 	FCHK(fs->fs_old_cgoffset * (~fs->fs_old_cgmask), >, fs->fs_fpg, %jd);
692 	FCHK(CGSIZE(fs), >, fs->fs_bsize, %jd);
693 	/*
694 	 * If anything has failed up to this point, it is usafe to proceed
695 	 * as checks below may divide by zero or make other fatal calculations.
696 	 * So if we have any errors at this point, give up.
697 	 */
698 	if (error)
699 		return (error);
700 	FCHK(fs->fs_sbsize % sectorsize, !=, 0, %jd);
701 	FCHK(fs->fs_ipg % fs->fs_inopb, !=, 0, %jd);
702 	FCHK(fs->fs_sblkno, !=, roundup(
703 	    howmany(fs->fs_sblockloc + SBLOCKSIZE, fs->fs_fsize),
704 	    fs->fs_frag), %jd);
705 	FCHK(fs->fs_cblkno, !=, fs->fs_sblkno +
706 	    roundup(howmany(SBLOCKSIZE, fs->fs_fsize), fs->fs_frag), %jd);
707 	FCHK(fs->fs_iblkno, !=, fs->fs_cblkno + fs->fs_frag, %jd);
708 	FCHK(fs->fs_dblkno, !=, fs->fs_iblkno + fs->fs_ipg / INOPF(fs), %jd);
709 	FCHK(fs->fs_cgsize, >, fs->fs_bsize, %jd);
710 	FCHK(fs->fs_cgsize, <, fs->fs_fsize, %jd);
711 	FCHK(fs->fs_cgsize % fs->fs_fsize, !=, 0, %jd);
712 	/*
713 	 * This test is valid, however older versions of growfs failed
714 	 * to correctly update fs_dsize so will fail this test. Thus we
715 	 * exclude it from the requirements.
716 	 */
717 #ifdef notdef
718 	WCHK(fs->fs_dsize, !=, fs->fs_size - fs->fs_sblkno -
719 		fs->fs_ncg * (fs->fs_dblkno - fs->fs_sblkno) -
720 		howmany(fs->fs_cssize, fs->fs_fsize), %jd);
721 #endif
722 	WCHK(fs->fs_metaspace, <, 0, %jd);
723 	WCHK(fs->fs_metaspace, >, fs->fs_fpg / 2, %jd);
724 	WCHK(fs->fs_minfree, >, 99, %jd%%);
725 	maxfilesize = fs->fs_bsize * UFS_NDADDR - 1;
726 	for (sizepb = fs->fs_bsize, i = 0; i < UFS_NIADDR; i++) {
727 		sizepb *= NINDIR(fs);
728 		maxfilesize += sizepb;
729 	}
730 	WCHK(fs->fs_maxfilesize, >, maxfilesize, %jd);
731 	/*
732 	 * These values have a tight interaction with each other that
733 	 * makes it hard to tightly bound them. So we can only check
734 	 * that they are within a broader possible range.
735 	 *
736 	 * The size cannot always be accurately determined, but ensure
737 	 * that it is consistent with the number of cylinder groups (fs_ncg)
738 	 * and the number of fragments per cylinder group (fs_fpg). Ensure
739 	 * that the summary information size is correct and that it starts
740 	 * and ends in the data area of the same cylinder group.
741 	 */
742 	FCHK(fs->fs_size, <, 8 * fs->fs_frag, %jd);
743 	FCHK(fs->fs_size, <=, ((int64_t)fs->fs_ncg - 1) * fs->fs_fpg, %jd);
744 	FCHK(fs->fs_size, >, (int64_t)fs->fs_ncg * fs->fs_fpg, %jd);
745 	/*
746 	 * If we are not requested to read in the csum data stop here
747 	 * as the correctness of the remaining values is only important
748 	 * to bound the space needed to be allocated to hold the csum data.
749 	 */
750 	if ((flags & UFS_NOCSUM) != 0)
751 		return (error);
752 	FCHK(fs->fs_csaddr, <, 0, %jd);
753 	FCHK(fs->fs_cssize, !=,
754 	    fragroundup(fs, fs->fs_ncg * sizeof(struct csum)), %jd);
755 	FCHK(fs->fs_csaddr + howmany(fs->fs_cssize, fs->fs_fsize), >,
756 	    fs->fs_size, %jd);
757 	FCHK(fs->fs_csaddr, <, cgdmin(fs, dtog(fs, fs->fs_csaddr)), %jd);
758 	FCHK(dtog(fs, fs->fs_csaddr + howmany(fs->fs_cssize, fs->fs_fsize)), >,
759 	    dtog(fs, fs->fs_csaddr), %jd);
760 	/*
761 	 * With file system clustering it is possible to allocate
762 	 * many contiguous blocks. The kernel variable maxphys defines
763 	 * the maximum transfer size permitted by the controller and/or
764 	 * buffering. The fs_maxcontig parameter controls the maximum
765 	 * number of blocks that the filesystem will read or write
766 	 * in a single transfer. It is calculated when the filesystem
767 	 * is created as maxphys / fs_bsize. The loader uses a maxphys
768 	 * of 128K even when running on a system that supports larger
769 	 * values. If the filesystem was built on a system that supports
770 	 * a larger maxphys (1M is typical) it will have configured
771 	 * fs_maxcontig for that larger system. So we bound the upper
772 	 * allowable limit for fs_maxconfig to be able to at least
773 	 * work with a 1M maxphys on the smallest block size filesystem:
774 	 * 1M / 4096 == 256. There is no harm in allowing the mounting of
775 	 * filesystems that make larger than maxphys I/O requests because
776 	 * those (mostly 32-bit machines) can (very slowly) handle I/O
777 	 * requests that exceed maxphys.
778 	 */
779 	WCHK(fs->fs_maxcontig, <, 0, %jd);
780 	WCHK(fs->fs_maxcontig, >, MAX(256, maxphys / fs->fs_bsize), %jd);
781 	FCHK2(fs->fs_maxcontig, ==, 0, fs->fs_contigsumsize, !=, 0, %jd);
782 	FCHK2(fs->fs_maxcontig, >, 1, fs->fs_contigsumsize, !=,
783 	    MIN(fs->fs_maxcontig, FS_MAXCONTIG), %jd);
784 	return (error);
785 }
786 
787 /*
788  * Make an extensive search to find a superblock. If the superblock
789  * in the standard place cannot be used, try looking for one of the
790  * backup superblocks.
791  *
792  * Flags are made up of the following or'ed together options:
793  *
794  * UFS_NOMSG indicates that superblock inconsistency error messages
795  *    should not be printed.
796  *
797  * UFS_NOCSUM causes only the superblock itself to be returned, but does
798  *    not read in any auxillary data structures like the cylinder group
799  *    summary information.
800  */
801 int
ffs_sbsearch(void * devfd,struct fs ** fsp,int reqflags,struct malloc_type * filltype,int (* readfunc)(void * devfd,off_t loc,void ** bufp,int size))802 ffs_sbsearch(void *devfd, struct fs **fsp, int reqflags,
803     struct malloc_type *filltype,
804     int (*readfunc)(void *devfd, off_t loc, void **bufp, int size))
805 {
806 	struct fsrecovery *fsr;
807 	struct fs *protofs;
808 	void *fsrbuf;
809 	char *cp;
810 	long nocsum, flags, msg, cg;
811 	off_t sblk, secsize;
812 	int error;
813 
814 	msg = (reqflags & UFS_NOMSG) == 0;
815 	nocsum = reqflags & UFS_NOCSUM;
816 	/*
817 	 * Try normal superblock read and return it if it works.
818 	 *
819 	 * Suppress messages if it fails until we find out if
820 	 * failure can be avoided.
821 	 */
822 	flags = UFS_NOMSG | nocsum;
823 	error = ffs_sbget(devfd, fsp, UFS_STDSB, flags, filltype, readfunc);
824 	/*
825 	 * If successful or endian error, no need to try further.
826 	 */
827 	if (error == 0 || error == EILSEQ) {
828 		if (msg && error == EILSEQ)
829 			printf("UFS superblock failed due to endian mismatch "
830 			    "between machine and filesystem\n");
831 		return (error);
832 	}
833 	/*
834 	 * First try: ignoring hash failures.
835 	 */
836 	flags |= UFS_NOHASHFAIL;
837 	if (msg)
838 		flags &= ~UFS_NOMSG;
839 	if (ffs_sbget(devfd, fsp, UFS_STDSB, flags, filltype, readfunc) == 0)
840 		return (0);
841 	/*
842 	 * Next up is to check if fields of the superblock that are
843 	 * needed to find backup superblocks are usable.
844 	 */
845 	if (msg)
846 		printf("Attempted recovery for standard superblock: failed\n");
847 	flags = UFS_FSRONLY | UFS_NOHASHFAIL | UFS_NOCSUM | UFS_NOMSG;
848 	if (ffs_sbget(devfd, &protofs, UFS_STDSB, flags, filltype,
849 	    readfunc) == 0) {
850 		if (msg)
851 			printf("Attempt extraction of recovery data from "
852 			    "standard superblock.\n");
853 	} else {
854 		/*
855 		 * Final desperation is to see if alternate superblock
856 		 * parameters have been saved in the boot area.
857 		 */
858 		if (msg)
859 			printf("Attempted extraction of recovery data from "
860 			    "standard superblock: failed\nAttempt to find "
861 			    "boot zone recovery data.\n");
862 		/*
863 		 * Look to see if recovery information has been saved.
864 		 * If so we can generate a prototype superblock based
865 		 * on that information.
866 		 *
867 		 * We need fragments-per-group, number of cylinder groups,
868 		 * location of the superblock within the cylinder group, and
869 		 * the conversion from filesystem fragments to disk blocks.
870 		 *
871 		 * When building a UFS2 filesystem, newfs(8) stores these
872 		 * details at the end of the boot block area at the start
873 		 * of the filesystem partition. If they have been overwritten
874 		 * by a boot block, we fail.  But usually they are there
875 		 * and we can use them.
876 		 *
877 		 * We could ask the underlying device for its sector size,
878 		 * but some devices lie. So we just try a plausible range.
879 		 */
880 		error = ENOENT;
881 		fsrbuf = NULL;
882 		for (secsize = dbtob(1); secsize <= SBLOCKSIZE; secsize *= 2)
883 			if ((error = (*readfunc)(devfd, (SBLOCK_UFS2 - secsize),
884 			    &fsrbuf, secsize)) == 0)
885 				break;
886 		if (error != 0)
887 			goto trynowarn;
888 		cp = fsrbuf; /* type change to keep compiler happy */
889 		fsr = (struct fsrecovery *)&cp[secsize - sizeof *fsr];
890 		if (fsr->fsr_magic != FS_UFS2_MAGIC ||
891 		    (protofs = UFS_MALLOC(SBLOCKSIZE, filltype, M_NOWAIT))
892 		    == NULL) {
893 			UFS_FREE(fsrbuf, filltype);
894 			goto trynowarn;
895 		}
896 		memset(protofs, 0, sizeof(struct fs));
897 		protofs->fs_fpg = fsr->fsr_fpg;
898 		protofs->fs_fsbtodb = fsr->fsr_fsbtodb;
899 		protofs->fs_sblkno = fsr->fsr_sblkno;
900 		protofs->fs_magic = fsr->fsr_magic;
901 		protofs->fs_ncg = fsr->fsr_ncg;
902 		UFS_FREE(fsrbuf, filltype);
903 	}
904 	/*
905 	 * Scan looking for alternative superblocks.
906 	 */
907 	flags = nocsum;
908 	if (!msg)
909 		flags |= UFS_NOMSG;
910 	for (cg = 0; cg < protofs->fs_ncg; cg++) {
911 		sblk = fsbtodb(protofs, cgsblock(protofs, cg));
912 		if (msg)
913 			printf("Try cg %ld at sblock loc %jd\n", cg,
914 			    (intmax_t)sblk);
915 		if (ffs_sbget(devfd, fsp, dbtob(sblk), flags, filltype,
916 		    readfunc) == 0) {
917 			if (msg)
918 				printf("Succeeded with alternate superblock "
919 				    "at %jd\n", (intmax_t)sblk);
920 			UFS_FREE(protofs, filltype);
921 			return (0);
922 		}
923 	}
924 	UFS_FREE(protofs, filltype);
925 	/*
926 	 * Our alternate superblock strategies failed. Our last ditch effort
927 	 * is to see if the standard superblock has only non-critical errors.
928 	 */
929 trynowarn:
930 	flags = UFS_NOWARNFAIL | UFS_NOMSG | nocsum;
931 	if (msg) {
932 		printf("Finding an alternate superblock failed.\nCheck for "
933 		    "only non-critical errors in standard superblock\n");
934 		flags &= ~UFS_NOMSG;
935 	}
936 	if (ffs_sbget(devfd, fsp, UFS_STDSB, flags, filltype, readfunc) != 0) {
937 		if (msg)
938 			printf("Failed, superblock has critical errors\n");
939 		return (ENOENT);
940 	}
941 	if (msg)
942 		printf("Success, using standard superblock with "
943 		    "non-critical errors.\n");
944 	return (0);
945 }
946 
947 /*
948  * Write a superblock to the devfd device from the memory pointed to by fs.
949  * Write out the superblock summary information if it is present.
950  *
951  * If the write is successful, zero is returned. Otherwise one of the
952  * following error values is returned:
953  *     EIO: failed to write superblock.
954  *     EIO: failed to write superblock summary information.
955  */
956 int
ffs_sbput(void * devfd,struct fs * fs,off_t loc,int (* writefunc)(void * devfd,off_t loc,void * buf,int size))957 ffs_sbput(void *devfd, struct fs *fs, off_t loc,
958     int (*writefunc)(void *devfd, off_t loc, void *buf, int size))
959 {
960 	struct fs_summary_info *fs_si;
961 	int i, error, blks, size;
962 	uint8_t *space;
963 
964 	/*
965 	 * If there is summary information, write it first, so if there
966 	 * is an error, the superblock will not be marked as clean.
967 	 */
968 	if (fs->fs_si != NULL && fs->fs_csp != NULL) {
969 		blks = howmany(fs->fs_cssize, fs->fs_fsize);
970 		space = (uint8_t *)fs->fs_csp;
971 		for (i = 0; i < blks; i += fs->fs_frag) {
972 			size = fs->fs_bsize;
973 			if (i + fs->fs_frag > blks)
974 				size = (blks - i) * fs->fs_fsize;
975 			if ((error = (*writefunc)(devfd,
976 			     dbtob(fsbtodb(fs, fs->fs_csaddr + i)),
977 			     space, size)) != 0)
978 				return (error);
979 			space += size;
980 		}
981 	}
982 	fs->fs_fmod = 0;
983 	ffs_oldfscompat_write(fs);
984 #ifdef _KERNEL
985 	fs->fs_time = time_second;
986 #else /* User Code */
987 	fs->fs_time = time(NULL);
988 #endif
989 	/* Clear the pointers for the duration of writing. */
990 	fs_si = fs->fs_si;
991 	fs->fs_si = NULL;
992 	fs->fs_ckhash = ffs_calc_sbhash(fs);
993 	error = (*writefunc)(devfd, loc, fs, fs->fs_sbsize);
994 	/*
995 	 * A negative error code is returned when a copy of the
996 	 * superblock has been made which is discarded when the I/O
997 	 * is done. So the fs_si field does not and indeed cannot be
998 	 * restored after the write is done. Convert the error code
999 	 * back to its usual positive value when returning it.
1000 	 */
1001 	if (error < 0)
1002 		return (-error - 1);
1003 	fs->fs_si = fs_si;
1004 	return (error);
1005 }
1006 
1007 /*
1008  * Calculate the check-hash for a superblock.
1009  */
1010 uint32_t
ffs_calc_sbhash(struct fs * fs)1011 ffs_calc_sbhash(struct fs *fs)
1012 {
1013 	uint32_t ckhash, save_ckhash;
1014 
1015 	/*
1016 	 * A filesystem that was using a superblock ckhash may be moved
1017 	 * to an older kernel that does not support ckhashes. The
1018 	 * older kernel will clear the FS_METACKHASH flag indicating
1019 	 * that it does not update hashes. When the disk is moved back
1020 	 * to a kernel capable of ckhashes it disables them on mount:
1021 	 *
1022 	 *	if ((fs->fs_flags & FS_METACKHASH) == 0)
1023 	 *		fs->fs_metackhash = 0;
1024 	 *
1025 	 * This leaves (fs->fs_metackhash & CK_SUPERBLOCK) == 0) with an
1026 	 * old stale value in the fs->fs_ckhash field. Thus the need to
1027 	 * just accept what is there.
1028 	 */
1029 	if ((fs->fs_metackhash & CK_SUPERBLOCK) == 0)
1030 		return (fs->fs_ckhash);
1031 
1032 	save_ckhash = fs->fs_ckhash;
1033 	fs->fs_ckhash = 0;
1034 	/*
1035 	 * If newly read from disk, the caller is responsible for
1036 	 * verifying that fs->fs_sbsize <= SBLOCKSIZE.
1037 	 */
1038 	ckhash = calculate_crc32c(~0L, (void *)fs, fs->fs_sbsize);
1039 	fs->fs_ckhash = save_ckhash;
1040 	return (ckhash);
1041 }
1042 
1043 /*
1044  * Update the frsum fields to reflect addition or deletion
1045  * of some frags.
1046  */
1047 void
ffs_fragacct(struct fs * fs,int fragmap,int32_t fraglist[],int cnt)1048 ffs_fragacct(struct fs *fs, int fragmap, int32_t fraglist[], int cnt)
1049 {
1050 	int inblk;
1051 	int field, subfield;
1052 	int siz, pos;
1053 
1054 	inblk = (int)(fragtbl[fs->fs_frag][fragmap]) << 1;
1055 	fragmap <<= 1;
1056 	for (siz = 1; siz < fs->fs_frag; siz++) {
1057 		if ((inblk & (1 << (siz + (fs->fs_frag % NBBY)))) == 0)
1058 			continue;
1059 		field = around[siz];
1060 		subfield = inside[siz];
1061 		for (pos = siz; pos <= fs->fs_frag; pos++) {
1062 			if ((fragmap & field) == subfield) {
1063 				fraglist[siz] += cnt;
1064 				pos += siz;
1065 				field <<= siz;
1066 				subfield <<= siz;
1067 			}
1068 			field <<= 1;
1069 			subfield <<= 1;
1070 		}
1071 	}
1072 }
1073 
1074 /*
1075  * block operations
1076  *
1077  * check if a block is available
1078  */
1079 int
ffs_isblock(struct fs * fs,unsigned char * cp,ufs1_daddr_t h)1080 ffs_isblock(struct fs *fs, unsigned char *cp, ufs1_daddr_t h)
1081 {
1082 	unsigned char mask;
1083 
1084 	switch ((int)fs->fs_frag) {
1085 	case 8:
1086 		return (cp[h] == 0xff);
1087 	case 4:
1088 		mask = 0x0f << ((h & 0x1) << 2);
1089 		return ((cp[h >> 1] & mask) == mask);
1090 	case 2:
1091 		mask = 0x03 << ((h & 0x3) << 1);
1092 		return ((cp[h >> 2] & mask) == mask);
1093 	case 1:
1094 		mask = 0x01 << (h & 0x7);
1095 		return ((cp[h >> 3] & mask) == mask);
1096 	default:
1097 #ifdef _KERNEL
1098 		panic("ffs_isblock");
1099 #endif
1100 		break;
1101 	}
1102 	return (0);
1103 }
1104 
1105 /*
1106  * check if a block is free
1107  */
1108 int
ffs_isfreeblock(struct fs * fs,uint8_t * cp,ufs1_daddr_t h)1109 ffs_isfreeblock(struct fs *fs, uint8_t *cp, ufs1_daddr_t h)
1110 {
1111 
1112 	switch ((int)fs->fs_frag) {
1113 	case 8:
1114 		return (cp[h] == 0);
1115 	case 4:
1116 		return ((cp[h >> 1] & (0x0f << ((h & 0x1) << 2))) == 0);
1117 	case 2:
1118 		return ((cp[h >> 2] & (0x03 << ((h & 0x3) << 1))) == 0);
1119 	case 1:
1120 		return ((cp[h >> 3] & (0x01 << (h & 0x7))) == 0);
1121 	default:
1122 #ifdef _KERNEL
1123 		panic("ffs_isfreeblock");
1124 #endif
1125 		break;
1126 	}
1127 	return (0);
1128 }
1129 
1130 /*
1131  * take a block out of the map
1132  */
1133 void
ffs_clrblock(struct fs * fs,uint8_t * cp,ufs1_daddr_t h)1134 ffs_clrblock(struct fs *fs, uint8_t *cp, ufs1_daddr_t h)
1135 {
1136 
1137 	switch ((int)fs->fs_frag) {
1138 	case 8:
1139 		cp[h] = 0;
1140 		return;
1141 	case 4:
1142 		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1143 		return;
1144 	case 2:
1145 		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1146 		return;
1147 	case 1:
1148 		cp[h >> 3] &= ~(0x01 << (h & 0x7));
1149 		return;
1150 	default:
1151 #ifdef _KERNEL
1152 		panic("ffs_clrblock");
1153 #endif
1154 		break;
1155 	}
1156 }
1157 
1158 /*
1159  * put a block into the map
1160  */
1161 void
ffs_setblock(struct fs * fs,unsigned char * cp,ufs1_daddr_t h)1162 ffs_setblock(struct fs *fs, unsigned char *cp, ufs1_daddr_t h)
1163 {
1164 
1165 	switch ((int)fs->fs_frag) {
1166 	case 8:
1167 		cp[h] = 0xff;
1168 		return;
1169 	case 4:
1170 		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1171 		return;
1172 	case 2:
1173 		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1174 		return;
1175 	case 1:
1176 		cp[h >> 3] |= (0x01 << (h & 0x7));
1177 		return;
1178 	default:
1179 #ifdef _KERNEL
1180 		panic("ffs_setblock");
1181 #endif
1182 		break;
1183 	}
1184 }
1185 
1186 /*
1187  * Update the cluster map because of an allocation or free.
1188  *
1189  * Cnt == 1 means free; cnt == -1 means allocating.
1190  */
1191 void
ffs_clusteracct(struct fs * fs,struct cg * cgp,ufs1_daddr_t blkno,int cnt)1192 ffs_clusteracct(struct fs *fs, struct cg *cgp, ufs1_daddr_t blkno, int cnt)
1193 {
1194 	int32_t *sump;
1195 	int32_t *lp;
1196 	uint8_t *freemapp, *mapp;
1197 	int i, start, end, forw, back, map;
1198 	uint64_t bit;
1199 
1200 	if (fs->fs_contigsumsize <= 0)
1201 		return;
1202 	freemapp = cg_clustersfree(cgp);
1203 	sump = cg_clustersum(cgp);
1204 	/*
1205 	 * Allocate or clear the actual block.
1206 	 */
1207 	if (cnt > 0)
1208 		setbit(freemapp, blkno);
1209 	else
1210 		clrbit(freemapp, blkno);
1211 	/*
1212 	 * Find the size of the cluster going forward.
1213 	 */
1214 	start = blkno + 1;
1215 	end = start + fs->fs_contigsumsize;
1216 	if (end >= cgp->cg_nclusterblks)
1217 		end = cgp->cg_nclusterblks;
1218 	mapp = &freemapp[start / NBBY];
1219 	map = *mapp++;
1220 	bit = 1U << (start % NBBY);
1221 	for (i = start; i < end; i++) {
1222 		if ((map & bit) == 0)
1223 			break;
1224 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
1225 			bit <<= 1;
1226 		} else {
1227 			map = *mapp++;
1228 			bit = 1;
1229 		}
1230 	}
1231 	forw = i - start;
1232 	/*
1233 	 * Find the size of the cluster going backward.
1234 	 */
1235 	start = blkno - 1;
1236 	end = start - fs->fs_contigsumsize;
1237 	if (end < 0)
1238 		end = -1;
1239 	mapp = &freemapp[start / NBBY];
1240 	map = *mapp--;
1241 	bit = 1U << (start % NBBY);
1242 	for (i = start; i > end; i--) {
1243 		if ((map & bit) == 0)
1244 			break;
1245 		if ((i & (NBBY - 1)) != 0) {
1246 			bit >>= 1;
1247 		} else {
1248 			map = *mapp--;
1249 			bit = 1U << (NBBY - 1);
1250 		}
1251 	}
1252 	back = start - i;
1253 	/*
1254 	 * Account for old cluster and the possibly new forward and
1255 	 * back clusters.
1256 	 */
1257 	i = back + forw + 1;
1258 	if (i > fs->fs_contigsumsize)
1259 		i = fs->fs_contigsumsize;
1260 	sump[i] += cnt;
1261 	if (back > 0)
1262 		sump[back] -= cnt;
1263 	if (forw > 0)
1264 		sump[forw] -= cnt;
1265 	/*
1266 	 * Update cluster summary information.
1267 	 */
1268 	lp = &sump[fs->fs_contigsumsize];
1269 	for (i = fs->fs_contigsumsize; i > 0; i--)
1270 		if (*lp-- > 0)
1271 			break;
1272 	fs->fs_maxcluster[cgp->cg_cgx] = i;
1273 }
1274