xref: /titanic_41/usr/src/cmd/fs.d/ufs/mkfs/mkfs.c (revision 299d957d28501350bcae7fa4333ff183c062c427)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All Rights Reserved  	*/
28 
29 /*
30  * University Copyright- Copyright (c) 1982, 1986, 1988
31  * The Regents of the University of California
32  * All Rights Reserved
33  *
34  * University Acknowledgment- Portions of this document are derived from
35  * software developed by the University of California, Berkeley, and its
36  * contributors.
37  */
38 
39 /*
40  * The maximum supported file system size (in sectors) is the
41  * number of frags that can be represented in an int32_t field
42  * (INT_MAX) times the maximum number of sectors per frag.  Since
43  * the maximum frag size is MAXBSIZE, the maximum number of sectors
44  * per frag is MAXBSIZE/DEV_BSIZE.
45  */
46 #define	FS_MAX	(((diskaddr_t)INT_MAX) * (MAXBSIZE/DEV_BSIZE))
47 
48 /*
49  * make file system for cylinder-group style file systems
50  *
51  * usage:
52  *
53  *    mkfs [-F FSType] [-V] [-G [-P]] [-M dirname] [-m] [options]
54  *	[-o specific_options]  special size
55  *	[nsect ntrack bsize fsize cpg	minfree	rps nbpi opt apc rotdelay
56  *	  2     3      4     5     6	7	8   9	 10  11  12
57  *	nrpos maxcontig mtb]
58  *	13    14	15
59  *
60  *  where specific_options are:
61  *	N - no create
62  *	nsect - The number of sectors per track
63  *	ntrack - The number of tracks per cylinder
64  *	bsize - block size
65  *	fragsize - fragment size
66  *	cgsize - The number of disk cylinders per cylinder group.
67  * 	free - minimum free space
68  *	rps - rotational speed (rev/sec).
69  *	nbpi - number of data bytes per allocated inode
70  *	opt - optimization (space, time)
71  *	apc - number of alternates
72  *	gap - gap size
73  *	nrpos - number of rotational positions
74  *	maxcontig - maximum number of logical blocks that will be
75  *		allocated contiguously before inserting rotational delay
76  *	mtb - if "y", set up file system for eventual growth to over a
77  *		a terabyte
78  * -P Do not grow the file system, but print on stdout the maximal
79  *    size in sectors to which the file system can be increased. The calculated
80  *    size is limited by the value provided by the operand size.
81  *
82  * Note that -P is a project-private interface and together with -G intended
83  * to be used only by the growfs script. It is therefore purposely not
84  * documented in the man page.
85  * The -P option is covered by PSARC case 2003/422.
86  */
87 
88 /*
89  * The following constants set the defaults used for the number
90  * of sectors/track (fs_nsect), and number of tracks/cyl (fs_ntrak).
91  *
92  *			NSECT		NTRAK
93  *	72MB CDC	18		9
94  *	30MB CDC	18		5
95  *	720KB Diskette	9		2
96  *
97  * However the defaults will be different for disks larger than CHSLIMIT.
98  */
99 
100 #define	DFLNSECT	32
101 #define	DFLNTRAK	16
102 
103 /*
104  * The following default sectors and tracks values are used for
105  * non-efi disks that are larger than the CHS addressing limit. The
106  * existing default cpg of 16 (DESCPG) holds good for larger disks too.
107  */
108 #define	DEF_SECTORS_EFI	128
109 #define	DEF_TRACKS_EFI	48
110 
111 /*
112  * The maximum number of cylinders in a group depends upon how much
113  * information can be stored on a single cylinder. The default is to
114  * use 16 cylinders per group.  This is effectively tradition - it was
115  * the largest value acceptable under SunOs 4.1
116  */
117 #define	DESCPG		16	/* desired fs_cpg */
118 
119 /*
120  * The following two constants set the default block and fragment sizes.
121  * Both constants must be a power of 2 and meet the following constraints:
122  *	MINBSIZE <= DESBLKSIZE <= MAXBSIZE
123  *	DEV_BSIZE <= DESFRAGSIZE <= DESBLKSIZE
124  *	DESBLKSIZE / DESFRAGSIZE <= 8
125  */
126 #define	DESBLKSIZE	8192
127 #define	DESFRAGSIZE	1024
128 
129 /*
130  * MINFREE gives the minimum acceptable percentage of file system
131  * blocks which may be free. If the freelist drops below this level
132  * only the superuser may continue to allocate blocks. This may
133  * be set to 0 if no reserve of free blocks is deemed necessary,
134  * however throughput drops by fifty percent if the file system
135  * is run at between 90% and 100% full; thus the default value of
136  * fs_minfree is 10%. With 10% free space, fragmentation is not a
137  * problem, so we choose to optimize for time.
138  */
139 #define	MINFREE		10
140 #define	DEFAULTOPT	FS_OPTTIME
141 
142 /*
143  * ROTDELAY gives the minimum number of milliseconds to initiate
144  * another disk transfer on the same cylinder. It is no longer used
145  * and will always default to 0.
146  */
147 #define	ROTDELAY	0
148 
149 /*
150  * MAXBLKPG determines the maximum number of data blocks which are
151  * placed in a single cylinder group. The default is one indirect
152  * block worth of data blocks.
153  */
154 #define	MAXBLKPG(bsize)	((bsize) / sizeof (daddr32_t))
155 
156 /*
157  * Each file system has a number of inodes statically allocated.
158  * We allocate one inode slot per NBPI bytes, expecting this
159  * to be far more than we will ever need.
160  */
161 #define	NBPI		2048	/* Number Bytes Per Inode */
162 #define	MTB_NBPI	(MB)	/* Number Bytes Per Inode for multi-terabyte */
163 
164 /*
165  * Disks are assumed to rotate at 60HZ, unless otherwise specified.
166  */
167 #define	DEFHZ		60
168 
169 /*
170  * Cylinder group related limits.
171  *
172  * For each cylinder we keep track of the availability of blocks at different
173  * rotational positions, so that we can lay out the data to be picked
174  * up with minimum rotational latency.  NRPOS is the number of rotational
175  * positions which we distinguish.  With NRPOS 8 the resolution of our
176  * summary information is 2ms for a typical 3600 rpm drive.
177  */
178 #define	NRPOS		8	/* number distinct rotational positions */
179 
180 #ifdef DEBUG
181 #define	dprintf(x)	printf x
182 #else
183 #define	dprintf(x)
184 #endif
185 
186 /*
187  * For the -N option, when calculating the backup superblocks, do not print
188  * them if we are not really sure. We may have to try an alternate method of
189  * arriving at the superblocks. So defer printing till a handful of superblocks
190  * look good.
191  */
192 #define	tprintf(x)	if (Nflag && retry) \
193 				(void) strncat(tmpbuf, x, strlen(x)); \
194 			else \
195 				(void) fprintf(stderr, x);
196 
197 #define	ALTSB		32	/* Location of first backup superblock */
198 
199 /*
200  * range_check "user_supplied" flag values.
201  */
202 #define	RC_DEFAULT	0
203 #define	RC_KEYWORD	1
204 #define	RC_POSITIONAL	2
205 
206 /*
207  * ufs hole
208  */
209 #define	UFS_HOLE	-1
210 
211 #ifndef	STANDALONE
212 #include	<stdio.h>
213 #include	<sys/mnttab.h>
214 #endif
215 
216 #include	<stdlib.h>
217 #include	<unistd.h>
218 #include	<malloc.h>
219 #include	<string.h>
220 #include	<strings.h>
221 #include	<ctype.h>
222 #include	<errno.h>
223 #include	<sys/param.h>
224 #include	<time.h>
225 #include	<sys/types.h>
226 #include	<sys/sysmacros.h>
227 #include	<sys/vnode.h>
228 #include	<sys/fs/ufs_fsdir.h>
229 #include	<sys/fs/ufs_inode.h>
230 #include	<sys/fs/ufs_fs.h>
231 #include	<sys/fs/ufs_log.h>
232 #include	<sys/mntent.h>
233 #include	<sys/filio.h>
234 #include	<limits.h>
235 #include	<sys/int_const.h>
236 #include	<signal.h>
237 #include	<sys/efi_partition.h>
238 #include	"roll_log.h"
239 
240 #define	bcopy(f, t, n)    (void) memcpy(t, f, n)
241 #define	bzero(s, n)	(void) memset(s, 0, n)
242 #define	bcmp(s, d, n)	memcmp(s, d, n)
243 
244 #define	index(s, r)	strchr(s, r)
245 #define	rindex(s, r)	strrchr(s, r)
246 
247 #include	<sys/stat.h>
248 #include	<sys/statvfs.h>
249 #include	<locale.h>
250 #include	<fcntl.h>
251 #include 	<sys/isa_defs.h>	/* for ENDIAN defines */
252 #include	<sys/vtoc.h>
253 
254 #include	<sys/dkio.h>
255 #include	<sys/asynch.h>
256 
257 extern offset_t	llseek();
258 extern char	*getfullblkname();
259 extern long	lrand48();
260 
261 extern int	optind;
262 extern char	*optarg;
263 
264 
265 /*
266  * The size of a cylinder group is calculated by CGSIZE. The maximum size
267  * is limited by the fact that cylinder groups are at most one block.
268  * Its size is derived from the size of the maps maintained in the
269  * cylinder group and the (struct cg) size.
270  */
271 #define	CGSIZE(fs) \
272 	/* base cg		*/ (sizeof (struct cg) + \
273 	/* blktot size	*/ (fs)->fs_cpg * sizeof (long) + \
274 	/* blks size	*/ (fs)->fs_cpg * (fs)->fs_nrpos * sizeof (short) + \
275 	/* inode map	*/ howmany((fs)->fs_ipg, NBBY) + \
276 	/* block map */ howmany((fs)->fs_cpg * (fs)->fs_spc / NSPF(fs), NBBY))
277 
278 /*
279  * We limit the size of the inode map to be no more than a
280  * third of the cylinder group space, since we must leave at
281  * least an equal amount of space for the block map.
282  *
283  * N.B.: MAXIpG must be a multiple of INOPB(fs).
284  */
285 #define	MAXIpG(fs)	roundup((fs)->fs_bsize * NBBY / 3, INOPB(fs))
286 
287 /*
288  * Same as MAXIpG, but parameterized by the block size (b) and the
289  * cylinder group divisor (d), which is the reciprocal of the fraction of the
290  * cylinder group overhead block that is used for the inode map.  So for
291  * example, if d = 5, the macro's computation assumes that 1/5 of the
292  * cylinder group overhead block can be dedicated to the inode map.
293  */
294 #define	MAXIpG_B(b, d)	roundup((b) * NBBY / (d), (b) / sizeof (struct dinode))
295 
296 #define	UMASK		0755
297 #define	MAXINOPB	(MAXBSIZE / sizeof (struct dinode))
298 #define	POWEROF2(num)	(((num) & ((num) - 1)) == 0)
299 #define	MB		(1024*1024)
300 #define	BETWEEN(x, l, h)	((x) >= (l) && (x) <= (h))
301 
302 /*
303  * Used to set the inode generation number. Since both inodes and dinodes
304  * are dealt with, we really need a pointer to an icommon here.
305  */
306 #define	IRANDOMIZE(icp)	(icp)->ic_gen = lrand48();
307 
308 /*
309  * Flags for number()
310  */
311 #define	ALLOW_PERCENT	0x01	/* allow trailing `%' on number */
312 #define	ALLOW_MS1	0x02	/* allow trailing `ms', state 1 */
313 #define	ALLOW_MS2	0x04	/* allow trailing `ms', state 2 */
314 #define	ALLOW_END_ONLY	0x08	/* must be at end of number & suffixes */
315 
316 #define	MAXAIO	1000	/* maximum number of outstanding I/O's we'll manage */
317 #define	BLOCK	1	/* block in aiowait */
318 #define	NOBLOCK	0	/* don't block in aiowait */
319 
320 #define	RELEASE 1	/* free an aio buffer after use */
321 #define	SAVE	0	/* don't free the buffer */
322 
323 typedef struct aio_trans {
324 	aio_result_t resultbuf;
325 	diskaddr_t bno;
326 	char *buffer;
327 	int size;
328 	int release;
329 	struct aio_trans *next;
330 } aio_trans;
331 
332 typedef struct aio_results {
333 	int max;
334 	int outstanding;
335 	int maxpend;
336 	aio_trans *trans;
337 } aio_results;
338 
339 int aio_inited = 0;
340 aio_results results;
341 
342 /*
343  * Allow up to MAXBUF aio requests that each have a unique buffer.
344  * More aio's might be done, but not using memory through the getbuf()
345  * interface.  This can be raised, but you run into the potential of
346  * using more memory than is physically available on the machine,
347  * and if you start swapping, you can forget about performance.
348  * To prevent this, we also limit the total memory used for a given
349  * type of buffer to MAXBUFMEM.
350  *
351  * Tests indicate a cylinder group's worth of inodes takes:
352  *
353  *	NBPI	Size of Inode Buffer
354  *	 2k	1688k
355  *	 8k	 424k
356  *
357  * initcg() stores all the inodes for a cylinder group in one buffer,
358  * so allowing 20 buffers could take 32 MB if not limited by MAXBUFMEM.
359  */
360 #define	MAXBUF		20
361 #define	MAXBUFMEM	(8 * 1024 * 1024)
362 
363 /*
364  * header information for buffers managed by getbuf() and freebuf()
365  */
366 typedef struct bufhdr {
367 	struct bufhdr *head;
368 	struct bufhdr *next;
369 } bufhdr;
370 
371 int bufhdrsize;
372 
373 bufhdr inodebuf = { NULL, NULL };
374 bufhdr cgsumbuf = { NULL, NULL };
375 
376 #define	SECTORS_PER_TERABYTE	(1LL << 31)
377 /*
378  * The following constant specifies an upper limit for file system size
379  * that is actually a lot bigger than we expect to support with UFS. (Since
380  * it's specified in sectors, the file system size would be 2**44 * 512,
381  * which is 2**53, which is 8192 Terabytes.)  However, it's useful
382  * for checking the basic sanity of a size value that is input on the
383  * command line.
384  */
385 #define	FS_SIZE_UPPER_LIMIT	0x100000000000LL
386 
387 /*
388  * Forward declarations
389  */
390 static char *getbuf(bufhdr *bufhead, int size);
391 static void freebuf(char *buf);
392 static void freetrans(aio_trans *transp);
393 static aio_trans *get_aiop();
394 static aio_trans *wait_for_write(int block);
395 static void initcg(int cylno);
396 static void fsinit();
397 static int makedir(struct direct *protodir, int entries);
398 static void iput(struct inode *ip);
399 static void rdfs(diskaddr_t bno, int size, char *bf);
400 static void wtfs(diskaddr_t bno, int size, char *bf);
401 static void awtfs(diskaddr_t bno, int size, char *bf, int release);
402 static void wtfs_breakup(diskaddr_t bno, int size, char *bf);
403 static int isblock(struct fs *fs, unsigned char *cp, int h);
404 static void clrblock(struct fs *fs, unsigned char *cp, int h);
405 static void setblock(struct fs *fs, unsigned char *cp, int h);
406 static void usage();
407 static void dump_fscmd(char *fsys, int fsi);
408 static uint64_t number(uint64_t d_value, char *param, int flags);
409 static int match(char *s);
410 static char checkopt(char *optim);
411 static char checkmtb(char *mtbarg);
412 static void range_check(long *varp, char *name, long minimum,
413     long maximum, long def_val, int user_supplied);
414 static void range_check_64(uint64_t *varp, char *name, uint64_t minimum,
415     uint64_t maximum, uint64_t def_val, int user_supplied);
416 static daddr32_t alloc(int size, int mode);
417 static diskaddr_t get_max_size(int fd);
418 static long get_max_track_size(int fd);
419 static void block_sigint(sigset_t *old_mask);
420 static void unblock_sigint(sigset_t *old_mask);
421 static void recover_from_sigint(int signum);
422 static int confirm_abort(void);
423 static int getline(FILE *fp, char *loc, int maxlen);
424 static void flush_writes(void);
425 static long compute_maxcpg(long, long, long, long, long);
426 static int in_64bit_mode(void);
427 static int validate_size(int fd, diskaddr_t size);
428 static void dump_sblock(void);
429 
430 /*
431  * Workaround for mkfs to function properly on disks attached to XMIT 2.X
432  * controller. If the address is not aligned at 8 byte boundary, mkfs on
433  * disks attached to XMIT 2.X controller exhibts un-predictable behaviour.
434  */
435 #define	XMIT_2_X_ALIGN	8
436 #pragma	align XMIT_2_X_ALIGN(fsun, altfsun, cgun)
437 
438 union {
439 	struct fs fs;
440 	char pad[SBSIZE];
441 } fsun, altfsun;
442 #define	sblock	fsun.fs
443 #define	altsblock	altfsun.fs
444 
445 struct	csum *fscs;
446 
447 union cgun {
448 	struct cg cg;
449 	char pad[MAXBSIZE];
450 } cgun;
451 
452 #define	acg	cgun.cg
453 /*
454  * Size of screen in cols in which to fit output
455  */
456 #define	WIDTH	80
457 
458 struct dinode zino[MAXBSIZE / sizeof (struct dinode)];
459 
460 /*
461  * file descriptors used for rdfs(fsi) and wtfs(fso).
462  * Initialized to an illegal file descriptor number.
463  */
464 int	fsi = -1;
465 int	fso = -1;
466 
467 /*
468  * The BIG parameter is machine dependent.  It should be a longlong integer
469  * constant that can be used by the number parser to check the validity
470  * of numeric parameters.
471  */
472 
473 #define	BIG		0x7fffffffffffffffLL
474 
475 /* Used to indicate to number() that a bogus value should cause us to exit */
476 #define	NO_DEFAULT	LONG_MIN
477 
478 /*
479  * INVALIDSBLIMIT is the number of bad backup superblocks that will be
480  * tolerated before we decide to try arriving at a different set of them
481  * using a different logic. This is applicable for non-EFI disks only.
482  */
483 #define	INVALIDSBLIMIT	10
484 
485 /*
486  * The *_flag variables are used to indicate that the user specified
487  * the values, rather than that we made them up ourselves.  We can
488  * complain about the user giving us bogus values.
489  */
490 
491 /* semi-constants */
492 long	sectorsize = DEV_BSIZE;		/* bytes/sector from param.h */
493 long	bbsize = BBSIZE;		/* boot block size */
494 long	sbsize = SBSIZE;		/* superblock size */
495 
496 /* parameters */
497 diskaddr_t	fssize_db;		/* file system size in disk blocks */
498 diskaddr_t	fssize_frag;		/* file system size in frags */
499 long	cpg;				/* cylinders/cylinder group */
500 int	cpg_flag = RC_DEFAULT;
501 long	rotdelay = -1;			/* rotational delay between blocks */
502 int	rotdelay_flag = RC_DEFAULT;
503 long	maxcontig;			/* max contiguous blocks to allocate */
504 int	maxcontig_flag = RC_DEFAULT;
505 long	nsect = DFLNSECT;		/* sectors per track */
506 int	nsect_flag = RC_DEFAULT;
507 long	ntrack = DFLNTRAK;		/* tracks per cylinder group */
508 int	ntrack_flag = RC_DEFAULT;
509 long	bsize = DESBLKSIZE;		/* filesystem block size */
510 int	bsize_flag = RC_DEFAULT;
511 long	fragsize = DESFRAGSIZE; 	/* filesystem fragment size */
512 int	fragsize_flag = RC_DEFAULT;
513 long	minfree = MINFREE; 		/* fs_minfree */
514 int	minfree_flag = RC_DEFAULT;
515 long	rps = DEFHZ;			/* revolutions/second of drive */
516 int	rps_flag = RC_DEFAULT;
517 long	nbpi = NBPI;			/* number of bytes per inode */
518 int	nbpi_flag = RC_DEFAULT;
519 long	nrpos = NRPOS;			/* number of rotational positions */
520 int	nrpos_flag = RC_DEFAULT;
521 long	apc = 0;			/* alternate sectors per cylinder */
522 int	apc_flag = RC_DEFAULT;
523 char	opt = 't';			/* optimization style, `t' or `s' */
524 char	mtb = 'n';			/* multi-terabyte format, 'y' or 'n' */
525 #define	DEFAULT_SECT_TRAK_CPG	(nsect_flag == RC_DEFAULT && \
526 				ntrack_flag == RC_DEFAULT && \
527 				cpg_flag == RC_DEFAULT)
528 
529 long	debug = 0;			/* enable debugging output */
530 
531 int	spc_flag = 0;			/* alternate sectors specified or */
532 					/* found */
533 
534 /* global state */
535 int	Nflag;		/* do not write to disk */
536 int	mflag;		/* return the command line used to create this FS */
537 int	rflag;		/* report the superblock in an easily-parsed form */
538 int	Rflag;		/* dump the superblock in binary */
539 char	*fsys;
540 time_t	mkfstime;
541 char	*string;
542 int	label_type;
543 
544 /*
545  * logging support
546  */
547 int	ismdd;			/* true if device is a SVM device */
548 int	islog;			/* true if ufs or SVM logging is enabled */
549 int	islogok;		/* true if ufs/SVM log state is good */
550 
551 static int	isufslog;	/* true if ufs logging is enabled */
552 static int	waslog;		/* true when ufs logging disabled during grow */
553 
554 /*
555  * growfs defines, globals, and forward references
556  */
557 #define	NOTENOUGHSPACE 33
558 int		grow;
559 #define	GROW_WITH_DEFAULT_TRAK	(grow && ntrack_flag == RC_DEFAULT)
560 
561 static int	Pflag;		/* probe to which size the fs can be grown */
562 int		ismounted;
563 char		*directory;
564 diskaddr_t	grow_fssize;
565 long		grow_fs_size;
566 long		grow_fs_ncg;
567 diskaddr_t		grow_fs_csaddr;
568 long		grow_fs_cssize;
569 int		grow_fs_clean;
570 struct csum	*grow_fscs;
571 diskaddr_t		grow_sifrag;
572 int		test;
573 int		testforce;
574 diskaddr_t		testfrags;
575 int		inlockexit;
576 int		isbad;
577 
578 void		lockexit(int);
579 void		randomgeneration(void);
580 void		checksummarysize(void);
581 int		checksblock(struct fs, int);
582 void		growinit(char *);
583 void		checkdev(char *, char  *);
584 void		checkmount(struct mnttab *, char *);
585 struct dinode	*gdinode(ino_t);
586 int		csfraginrange(daddr32_t);
587 struct csfrag	*findcsfrag(daddr32_t, struct csfrag **);
588 void		checkindirect(ino_t, daddr32_t *, daddr32_t, int);
589 void		addcsfrag(ino_t, daddr32_t, struct csfrag **);
590 void		delcsfrag(daddr32_t, struct csfrag **);
591 void		checkdirect(ino_t, daddr32_t *, daddr32_t *, int);
592 void		findcsfragino(void);
593 void		fixindirect(daddr32_t, int);
594 void		fixdirect(caddr_t, daddr32_t, daddr32_t *, int);
595 void		fixcsfragino(void);
596 void		extendsummaryinfo(void);
597 int		notenoughspace(void);
598 void		unalloccsfragino(void);
599 void		unalloccsfragfree(void);
600 void		findcsfragfree(void);
601 void		copycsfragino(void);
602 void		rdcg(long);
603 void		wtcg(void);
604 void		flcg(void);
605 void		allocfrags(long, daddr32_t *, long *);
606 void		alloccsfragino(void);
607 void		alloccsfragfree(void);
608 void		freefrags(daddr32_t, long, long);
609 int		findfreerange(long *, long *);
610 void		resetallocinfo(void);
611 void		extendcg(long);
612 void		ulockfs(void);
613 void		wlockfs(void);
614 void		clockfs(void);
615 void		wtsb(void);
616 static int64_t	checkfragallocated(daddr32_t);
617 static struct csum 	*read_summaryinfo(struct fs *);
618 static diskaddr_t 	probe_summaryinfo();
619 
620 int
621 main(int argc, char *argv[])
622 {
623 	long i, mincpc, mincpg, ibpcl;
624 	long cylno, rpos, blk, j, warn = 0;
625 	long mincpgcnt, maxcpg;
626 	uint64_t used, bpcg, inospercg;
627 	long mapcramped, inodecramped;
628 	long postblsize, rotblsize, totalsbsize;
629 	FILE *mnttab;
630 	struct mnttab mntp;
631 	char *special;
632 	struct statvfs64 fs;
633 	struct dk_geom dkg;
634 	struct dk_cinfo dkcinfo;
635 	char pbuf[sizeof (uint64_t) * 3 + 1];
636 	char *tmpbuf;
637 	int width, plen;
638 	uint64_t num;
639 	int c, saverr;
640 	diskaddr_t max_fssize;
641 	long tmpmaxcontig = -1;
642 	struct sigaction sigact;
643 	uint64_t nbytes64;
644 	int remaining_cg;
645 	int do_dot = 0;
646 	int use_efi_dflts = 0, retry = 0, isremovable = 0, ishotpluggable = 0;
647 	int invalid_sb_cnt, ret, skip_this_sb, cg_too_small;
648 	int geom_nsect, geom_ntrack, geom_cpg;
649 
650 	(void) setlocale(LC_ALL, "");
651 
652 #if !defined(TEXT_DOMAIN)
653 #define	TEXT_DOMAIN "SYS_TEST"
654 #endif
655 	(void) textdomain(TEXT_DOMAIN);
656 
657 	while ((c = getopt(argc, argv, "F:bmo:VPGM:T:t:")) != EOF) {
658 		switch (c) {
659 
660 		case 'F':
661 			string = optarg;
662 			if (strcmp(string, "ufs") != 0)
663 				usage();
664 			break;
665 
666 		case 'm':	/* return command line used to create this FS */
667 			mflag++;
668 			break;
669 
670 		case 'o':
671 			/*
672 			 * ufs specific options.
673 			 */
674 			string = optarg;
675 			while (*string != '\0') {
676 				if (match("nsect=")) {
677 					nsect = number(DFLNSECT, "nsect", 0);
678 					nsect_flag = RC_KEYWORD;
679 				} else if (match("ntrack=")) {
680 					ntrack = number(DFLNTRAK, "ntrack", 0);
681 					ntrack_flag = RC_KEYWORD;
682 				} else if (match("bsize=")) {
683 					bsize = number(DESBLKSIZE, "bsize", 0);
684 					bsize_flag = RC_KEYWORD;
685 				} else if (match("fragsize=")) {
686 					fragsize = number(DESFRAGSIZE,
687 					    "fragsize", 0);
688 					fragsize_flag = RC_KEYWORD;
689 				} else if (match("cgsize=")) {
690 					cpg = number(DESCPG, "cgsize", 0);
691 					cpg_flag = RC_KEYWORD;
692 				} else if (match("free=")) {
693 					minfree = number(MINFREE, "free",
694 					    ALLOW_PERCENT);
695 					minfree_flag = RC_KEYWORD;
696 				} else if (match("maxcontig=")) {
697 					tmpmaxcontig =
698 					    number(-1, "maxcontig", 0);
699 					maxcontig_flag = RC_KEYWORD;
700 				} else if (match("nrpos=")) {
701 					nrpos = number(NRPOS, "nrpos", 0);
702 					nrpos_flag = RC_KEYWORD;
703 				} else if (match("rps=")) {
704 					rps = number(DEFHZ, "rps", 0);
705 					rps_flag = RC_KEYWORD;
706 				} else if (match("nbpi=")) {
707 					nbpi = number(NBPI, "nbpi", 0);
708 					nbpi_flag = RC_KEYWORD;
709 				} else if (match("opt=")) {
710 					opt = checkopt(string);
711 				} else if (match("mtb=")) {
712 					mtb = checkmtb(string);
713 				} else if (match("apc=")) {
714 					apc = number(0, "apc", 0);
715 					apc_flag = RC_KEYWORD;
716 				} else if (match("gap=")) {
717 					(void) number(0, "gap", ALLOW_MS1);
718 					rotdelay = ROTDELAY;
719 					rotdelay_flag = RC_DEFAULT;
720 				} else if (match("debug=")) {
721 					debug = number(0, "debug", 0);
722 				} else if (match("N")) {
723 					Nflag++;
724 				} else if (match("calcsb")) {
725 					rflag++;
726 					Nflag++;
727 				} else if (match("calcbinsb")) {
728 					rflag++;
729 					Rflag++;
730 					Nflag++;
731 				} else if (*string == '\0') {
732 					break;
733 				} else {
734 					(void) fprintf(stderr, gettext(
735 					    "illegal option: %s\n"), string);
736 					usage();
737 				}
738 
739 				if (*string == ',') string++;
740 				if (*string == ' ') string++;
741 			}
742 			break;
743 
744 		case 'V':
745 			{
746 				char	*opt_text;
747 				int	opt_count;
748 
749 				(void) fprintf(stdout, gettext("mkfs -F ufs "));
750 				for (opt_count = 1; opt_count < argc;
751 				    opt_count++) {
752 					opt_text = argv[opt_count];
753 					if (opt_text)
754 						(void) fprintf(stdout, " %s ",
755 						    opt_text);
756 				}
757 				(void) fprintf(stdout, "\n");
758 			}
759 			break;
760 
761 		case 'b':	/* do nothing for this */
762 			break;
763 
764 		case 'M':	/* grow the mounted file system */
765 			directory = optarg;
766 
767 			/* FALLTHROUGH */
768 		case 'G':	/* grow the file system */
769 			grow = 1;
770 			break;
771 		case 'P':	/* probe the file system growing size 	*/
772 			Pflag = 1;
773 			grow = 1; /* probe mode implies fs growing	*/
774 			break;
775 		case 'T':	/* For testing */
776 			testforce = 1;
777 
778 			/* FALLTHROUGH */
779 		case 't':
780 			test = 1;
781 			string = optarg;
782 			testfrags = number(NO_DEFAULT, "testfrags", 0);
783 			break;
784 
785 		case '?':
786 			usage();
787 			break;
788 		}
789 	}
790 #ifdef MKFS_DEBUG
791 	/*
792 	 * Turning on MKFS_DEBUG causes mkfs to produce a filesystem
793 	 * that can be reproduced by setting the time to 0 and seeding
794 	 * the random number generator to a constant.
795 	 */
796 	mkfstime = 0;	/* reproducible results */
797 #else
798 	(void) time(&mkfstime);
799 #endif
800 
801 	if (optind >= (argc - 1)) {
802 		if (optind > (argc - 1)) {
803 			(void) fprintf(stderr,
804 			    gettext("special not specified\n"));
805 			usage();
806 		} else if (mflag == 0) {
807 			(void) fprintf(stderr,
808 			    gettext("size not specified\n"));
809 			usage();
810 		}
811 	}
812 	argc -= optind;
813 	argv = &argv[optind];
814 
815 	fsys = argv[0];
816 	fsi = open64(fsys, O_RDONLY);
817 	if (fsi < 0) {
818 		(void) fprintf(stderr, gettext("%s: cannot open\n"), fsys);
819 		lockexit(32);
820 	}
821 
822 	if (mflag) {
823 		dump_fscmd(fsys, fsi);
824 		lockexit(0);
825 	}
826 
827 	/*
828 	 * The task of setting all of the configuration parameters for a
829 	 * UFS file system is basically a matter of solving n equations
830 	 * in m variables.  Typically, m is greater than n, so there is
831 	 * usually more than one valid solution.  Since this is usually
832 	 * an under-constrained problem, it's not always obvious what the
833 	 * "best" configuration is.
834 	 *
835 	 * In general, the approach is to
836 	 * 1. Determine the values for the file system parameters
837 	 *    that are externally contrained and therefore not adjustable
838 	 *    by mkfs (such as the device's size and maxtransfer size).
839 	 * 2. Acquire the user's requested setting for all configuration
840 	 *    values that can be set on the command line.
841 	 * 3. Determine the final value of all configuration values, by
842 	 *    the following approach:
843 	 *	- set the file system block size (fs_bsize).  Although
844 	 *	  this could be regarded as an adjustable parameter, in
845 	 *	  fact, it's pretty much a constant.  At this time, it's
846 	 *	  generally set to 8k (with older hardware, it can
847 	 *	  sometimes make sense to set it to 4k, but those
848 	 *	  situations are pretty rare now).
849 	 *	- re-adjust the maximum file system size based on the
850 	 *	  value of the file system block size.  Since the
851 	 *	  frag size can't be any larger than a file system
852 	 *	  block, and the number of frags in the file system
853 	 *	  has to fit into 31 bits, the file system block size
854 	 *	  affects the maximum file system size.
855 	 *	- now that the real maximum file system is known, set the
856 	 *	  actual size of the file system to be created to
857 	 *	  MIN(requested size, maximum file system size).
858 	 *	- now validate, and if necessary, adjust the following
859 	 *	  values:
860 	 *		rotdelay
861 	 *		nsect
862 	 *		maxcontig
863 	 *		apc
864 	 *		frag_size
865 	 *		rps
866 	 *		minfree
867 	 *		nrpos
868 	 *		nrack
869 	 *		nbpi
870 	 *	- calculate maxcpg (the maximum value of the cylinders-per-
871 	 *	  cylinder-group configuration parameters).  There are two
872 	 *	  algorithms for calculating maxcpg:  an old one, which is
873 	 *	  used for file systems of less than 1 terabyte, and a
874 	 *	  new one, implemented in the function compute_maxcpg(),
875 	 *	  which is used for file systems of greater than 1 TB.
876 	 *	  The difference between them is that compute_maxcpg()
877 	 *	  really tries to maximize the cpg value.  The old
878 	 *	  algorithm fails to take advantage of smaller frags and
879 	 *	  lower inode density when determining the maximum cpg,
880 	 *	  and thus comes up with much lower numbers in some
881 	 *	  configurations.  At some point, we might use the
882 	 *	  new algorithm for determining maxcpg for all file
883 	 *	  systems, but at this time, the changes implemented for
884 	 *	  multi-terabyte UFS are NOT being automatically applied
885 	 *	  to UFS file systems of less than a terabyte (in the
886 	 *	  interest of not changing existing UFS policy too much
887 	 *	  until the ramifications of the changes are well-understood
888 	 *	  and have been evaluated for their effects on performance.)
889 	 *	- check the current values of the configuration parameters
890 	 *	  against the various constraints imposed by UFS.  These
891 	 *	  include:
892 	 *		* There must be at least one inode in each
893 	 *		  cylinder group.
894 	 *		* The cylinder group overhead block, which
895 	 *		  contains the inode and frag bigmaps, must fit
896 	 *		  within one file system block.
897 	 *		* The space required for inode maps should
898 	 *		  occupy no more than a third of the cylinder
899 	 *		  group overhead block.
900 	 *		* The rotational position tables have to fit
901 	 *		  within the available space in the super block.
902 	 *	  Adjust the configuration values that can be adjusted
903 	 *	  so that these constraints are satisfied.  The
904 	 *	  configuration values that are adjustable are:
905 	 *		* frag size
906 	 *		* cylinders per group
907 	 *		* inode density (can be increased)
908 	 *		* number of rotational positions (the rotational
909 	 *		  position tables are eliminated altogether if
910 	 *		  there isn't enough room for them.)
911 	 * 4. Set the values for all the dependent configuration
912 	 *    values (those that aren't settable on the command
913 	 *    line and which are completely dependent on the
914 	 *    adjustable parameters).  This include cpc (cycles
915 	 *    per cylinder, spc (sectors-per-cylinder), and many others.
916 	 */
917 
918 	/*
919 	 * Figure out the partition size and initialize the label_type.
920 	 */
921 	max_fssize = get_max_size(fsi);
922 
923 	/*
924 	 * Get and check positional arguments, if any.
925 	 */
926 	switch (argc - 1) {
927 	default:
928 		usage();
929 		/*NOTREACHED*/
930 	case 15:
931 		mtb = checkmtb(argv[15]);
932 		/* FALLTHROUGH */
933 	case 14:
934 		string = argv[14];
935 		tmpmaxcontig = number(-1, "maxcontig", 0);
936 		maxcontig_flag = RC_POSITIONAL;
937 		/* FALLTHROUGH */
938 	case 13:
939 		string = argv[13];
940 		nrpos = number(NRPOS, "nrpos", 0);
941 		nrpos_flag = RC_POSITIONAL;
942 		/* FALLTHROUGH */
943 	case 12:
944 		string = argv[12];
945 		rotdelay = ROTDELAY;
946 		rotdelay_flag = RC_DEFAULT;
947 		/* FALLTHROUGH */
948 	case 11:
949 		string = argv[11];
950 		apc = number(0, "apc", 0);
951 		apc_flag = RC_POSITIONAL;
952 		/* FALLTHROUGH */
953 	case 10:
954 		opt = checkopt(argv[10]);
955 		/* FALLTHROUGH */
956 	case 9:
957 		string = argv[9];
958 		nbpi = number(NBPI, "nbpi", 0);
959 		nbpi_flag = RC_POSITIONAL;
960 		/* FALLTHROUGH */
961 	case 8:
962 		string = argv[8];
963 		rps = number(DEFHZ, "rps", 0);
964 		rps_flag = RC_POSITIONAL;
965 		/* FALLTHROUGH */
966 	case 7:
967 		string = argv[7];
968 		minfree = number(MINFREE, "free", ALLOW_PERCENT);
969 		minfree_flag = RC_POSITIONAL;
970 		/* FALLTHROUGH */
971 	case 6:
972 		string = argv[6];
973 		cpg = number(DESCPG, "cgsize", 0);
974 		cpg_flag = RC_POSITIONAL;
975 		/* FALLTHROUGH */
976 	case 5:
977 		string = argv[5];
978 		fragsize = number(DESFRAGSIZE, "fragsize", 0);
979 		fragsize_flag = RC_POSITIONAL;
980 		/* FALLTHROUGH */
981 	case 4:
982 		string = argv[4];
983 		bsize = number(DESBLKSIZE, "bsize", 0);
984 		bsize_flag = RC_POSITIONAL;
985 		/* FALLTHROUGH */
986 	case 3:
987 		string = argv[3];
988 		ntrack = number(DFLNTRAK, "ntrack", 0);
989 		ntrack_flag = RC_POSITIONAL;
990 		/* FALLTHROUGH */
991 	case 2:
992 		string = argv[2];
993 		nsect = number(DFLNSECT, "nsect", 0);
994 		nsect_flag = RC_POSITIONAL;
995 		/* FALLTHROUGH */
996 	case 1:
997 		string = argv[1];
998 		fssize_db = number(max_fssize, "size", 0);
999 	}
1000 
1001 	/*
1002 	 * Initialize the parameters in the same way as newfs so that
1003 	 * newfs and mkfs would result in the same file system layout
1004 	 * for EFI labelled disks. Do this only in the absence of user
1005 	 * specified values for these parameters.
1006 	 */
1007 	if (label_type == LABEL_TYPE_EFI) {
1008 		if (apc_flag == RC_DEFAULT) apc = 0;
1009 		if (nrpos_flag == RC_DEFAULT) nrpos = 1;
1010 		if (ntrack_flag == RC_DEFAULT) ntrack = DEF_TRACKS_EFI;
1011 		if (rps_flag == RC_DEFAULT) rps = DEFHZ;
1012 		if (nsect_flag == RC_DEFAULT) nsect = DEF_SECTORS_EFI;
1013 	}
1014 
1015 	if ((maxcontig_flag == RC_DEFAULT) || (tmpmaxcontig == -1) ||
1016 	    (maxcontig == -1)) {
1017 		long maxtrax = get_max_track_size(fsi);
1018 		maxcontig = maxtrax / bsize;
1019 
1020 	} else {
1021 		maxcontig = tmpmaxcontig;
1022 	}
1023 	dprintf(("DeBuG maxcontig : %ld\n", maxcontig));
1024 
1025 	if (rotdelay == -1) {	/* default by newfs and mkfs */
1026 		rotdelay = ROTDELAY;
1027 	}
1028 
1029 	if (cpg_flag == RC_DEFAULT) { /* If not explicity set, use default */
1030 		cpg = DESCPG;
1031 	}
1032 	dprintf(("DeBuG cpg : %ld\n", cpg));
1033 
1034 	/*
1035 	 * Now that we have the semi-sane args, either positional, via -o,
1036 	 * or by defaulting, handle inter-dependencies and range checks.
1037 	 */
1038 
1039 	/*
1040 	 * Settle the file system block size first, since it's a fixed
1041 	 * parameter once set and so many other parameters, including
1042 	 * max_fssize, depend on it.
1043 	 */
1044 	range_check(&bsize, "bsize", MINBSIZE, MAXBSIZE, DESBLKSIZE,
1045 	    bsize_flag);
1046 
1047 	if (!POWEROF2(bsize)) {
1048 		(void) fprintf(stderr,
1049 		    gettext("block size must be a power of 2, not %ld\n"),
1050 		    bsize);
1051 		bsize = DESBLKSIZE;
1052 		(void) fprintf(stderr,
1053 		    gettext("mkfs: bsize reset to default %ld\n"),
1054 		    bsize);
1055 	}
1056 
1057 	if (fssize_db > max_fssize && validate_size(fsi, fssize_db)) {
1058 		(void) fprintf(stderr, gettext(
1059 		    "Warning: the requested size of this file system\n"
1060 		    "(%lld sectors) is greater than the size of the\n"
1061 		    "device reported by the driver (%lld sectors).\n"
1062 		    "However, a read of the device at the requested size\n"
1063 		    "does succeed, so the requested size will be used.\n"),
1064 		    fssize_db, max_fssize);
1065 		max_fssize = fssize_db;
1066 	}
1067 	/*
1068 	 * Since the maximum allocatable unit (the frag) must be less than
1069 	 * or equal to bsize, and the number of frags must be less than or
1070 	 * equal to INT_MAX, the total size of the file system (in
1071 	 * bytes) must be less than or equal to bsize * INT_MAX.
1072 	 */
1073 
1074 	if (max_fssize > ((diskaddr_t)bsize/DEV_BSIZE) * INT_MAX)
1075 		max_fssize = ((diskaddr_t)bsize/DEV_BSIZE) * INT_MAX;
1076 
1077 	range_check_64(&fssize_db, "size", 1024LL, max_fssize, max_fssize, 1);
1078 
1079 	if (fssize_db >= SECTORS_PER_TERABYTE) {
1080 		mtb = 'y';
1081 		if (!in_64bit_mode()) {
1082 			(void) fprintf(stderr, gettext(
1083 "mkfs:  Warning: Creating a file system greater than 1 terabyte on a\n"
1084 "       system running a 32-bit kernel.  This file system will not be\n"
1085 "       accessible until the system is rebooted with a 64-bit kernel.\n"));
1086 		}
1087 	}
1088 	dprintf(("DeBuG mtb : %c\n", mtb));
1089 
1090 	/*
1091 	 * With newer and much larger disks, the newfs(1M) and mkfs_ufs(1M)
1092 	 * commands had problems in correctly handling the "native" geometries
1093 	 * for various storage devices.
1094 	 *
1095 	 * To handle the new age disks, mkfs_ufs(1M) will use the EFI style
1096 	 * for non-EFI disks that are larger than the CHS addressing limit
1097 	 * ( > 8GB approx ) and ignore the disk geometry information for
1098 	 * these drives. This is what is currently done for multi-terrabyte
1099 	 * filesystems on EFI disks.
1100 	 *
1101 	 * However if the user asked for a specific layout by supplying values
1102 	 * for even one of the three parameters (nsect, ntrack, cpg), honour
1103 	 * the user supplied parameters.
1104 	 *
1105 	 * Choosing EFI style or native geometry style can make a lot of
1106 	 * difference, because the size of a cylinder group is dependent on
1107 	 * this choice. This in turn means that the position of alternate
1108 	 * superblocks varies depending on the style chosen. It is not
1109 	 * necessary that all disks of size > CHSLIMIT have EFI style layout.
1110 	 * There can be disks which are > CHSLIMIT size, but have native
1111 	 * geometry style layout, thereby warranting the need for alternate
1112 	 * logic in superblock detection.
1113 	 */
1114 	if (mtb != 'y' && (ntrack == -1 || GROW_WITH_DEFAULT_TRAK ||
1115 	    DEFAULT_SECT_TRAK_CPG)) {
1116 		/*
1117 		 * "-1" indicates that we were called from newfs and ntracks
1118 		 * was not specified in newfs command line. Calculate nsect
1119 		 * and ntrack in the same manner as newfs.
1120 		 *
1121 		 * This is required because, the defaults for nsect and ntrack
1122 		 * is hardcoded in mkfs, whereas to generate the alternate
1123 		 * superblock locations for the -N option, there is a need for
1124 		 * the geometry based values that newfs would have arrived at.
1125 		 * Newfs would have arrived at these values as below.
1126 		 */
1127 		if (label_type == LABEL_TYPE_EFI ||
1128 		    label_type == LABEL_TYPE_OTHER) {
1129 			use_efi_dflts = 1;
1130 			retry = 1;
1131 		} else if (ioctl(fsi, DKIOCGGEOM, &dkg)) {
1132 			dprintf(("%s: Unable to read Disk geometry", fsys));
1133 			perror(gettext("Unable to read Disk geometry"));
1134 			lockexit(32);
1135 		} else {
1136 			nsect = dkg.dkg_nsect;
1137 			ntrack = dkg.dkg_nhead;
1138 #ifdef i386	/* Bug 1170182 */
1139 			if (ntrack > 32 && (ntrack % 16) != 0) {
1140 				ntrack -= (ntrack % 16);
1141 			}
1142 #endif
1143 			if (ioctl(fsi, DKIOCREMOVABLE, &isremovable)) {
1144 				dprintf(("DeBuG Unable to determine if %s is"
1145 				    " Removable Media. Proceeding with system"
1146 				    " determined parameters.\n", fsys));
1147 				isremovable = 0;
1148 			}
1149 			if (ioctl(fsi, DKIOCHOTPLUGGABLE, &ishotpluggable)) {
1150 				dprintf(("DeBuG Unable to determine if %s is"
1151 				    " Hotpluggable Media. Proceeding with "
1152 				    "system determined parameters.\n", fsys));
1153 				ishotpluggable = 0;
1154 			}
1155 			if ((((diskaddr_t)dkg.dkg_ncyl * dkg.dkg_nhead *
1156 			    dkg.dkg_nsect) > CHSLIMIT) || isremovable ||
1157 			    ishotpluggable) {
1158 				use_efi_dflts = 1;
1159 				retry = 1;
1160 			}
1161 		}
1162 	}
1163 	dprintf(("DeBuG CHSLIMIT = %d geom = %llu\n", CHSLIMIT,
1164 	    (diskaddr_t)dkg.dkg_ncyl * dkg.dkg_nhead * dkg.dkg_nsect));
1165 	dprintf(("DeBuG label_type = %d isremovable = %d ishotpluggable = %d "
1166 	    "use_efi_dflts = %d\n", label_type, isremovable, ishotpluggable,
1167 	    use_efi_dflts));
1168 
1169 	/*
1170 	 * For the newfs -N case, even if the disksize is > CHSLIMIT, do not
1171 	 * blindly follow EFI style. If the fs_version indicates a geometry
1172 	 * based layout, try that one first. If it fails we can always try the
1173 	 * other logic.
1174 	 *
1175 	 * If we were called from growfs, we will have a problem if we mix
1176 	 * and match the filesystem creation and growth styles. For example,
1177 	 * if we create using EFI style, we have to also grow using EFI
1178 	 * style. So follow the style indicated by the fs_version.
1179 	 *
1180 	 * Read and verify the primary superblock. If it looks sane, use the
1181 	 * fs_version from the superblock. If the primary superblock does
1182 	 * not look good, read and verify the first alternate superblock at
1183 	 * ALTSB. Use the fs_version to decide whether to use the
1184 	 * EFI style logic or the old geometry based logic to calculate
1185 	 * the alternate superblock locations.
1186 	 */
1187 	if ((Nflag && use_efi_dflts) || (grow)) {
1188 		if (grow && ntrack_flag != RC_DEFAULT)
1189 			goto start_fs_creation;
1190 		rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize,
1191 		    (char *)&altsblock);
1192 		ret = checksblock(altsblock, 1);
1193 
1194 		if (!ret) {
1195 			if (altsblock.fs_magic == MTB_UFS_MAGIC) {
1196 				mtb = 'y';
1197 				goto start_fs_creation;
1198 			}
1199 			use_efi_dflts = (altsblock.fs_version ==
1200 			    UFS_EFISTYLE4NONEFI_VERSION_2) ? 1 : 0;
1201 		} else {
1202 			/*
1203 			 * The primary superblock didn't help in determining
1204 			 * the fs_version. Try the first alternate superblock.
1205 			 */
1206 			dprintf(("DeBuG checksblock() failed - error : %d"
1207 			    " for sb : %d\n", ret, SBOFF/sectorsize));
1208 			rdfs((diskaddr_t)ALTSB, (int)sbsize,
1209 			    (char *)&altsblock);
1210 			ret = checksblock(altsblock, 1);
1211 
1212 			if (!ret) {
1213 				if (altsblock.fs_magic == MTB_UFS_MAGIC) {
1214 					mtb = 'y';
1215 					goto start_fs_creation;
1216 				}
1217 				use_efi_dflts = (altsblock.fs_version ==
1218 				    UFS_EFISTYLE4NONEFI_VERSION_2) ? 1 : 0;
1219 			}
1220 			dprintf(("DeBuG checksblock() returned : %d"
1221 			    " for sb : %d\n", ret, ALTSB));
1222 		}
1223 	}
1224 
1225 	geom_nsect = nsect;
1226 	geom_ntrack = ntrack;
1227 	geom_cpg = cpg;
1228 	dprintf(("DeBuG geom_nsect=%d, geom_ntrack=%d, geom_cpg=%d\n",
1229 	    geom_nsect, geom_ntrack, geom_cpg));
1230 
1231 start_fs_creation:
1232 retry_alternate_logic:
1233 	invalid_sb_cnt = 0;
1234 	cg_too_small = 0;
1235 	if (use_efi_dflts) {
1236 		nsect = DEF_SECTORS_EFI;
1237 		ntrack = DEF_TRACKS_EFI;
1238 		cpg = DESCPG;
1239 		dprintf(("\nDeBuG Using EFI defaults\n"));
1240 	} else {
1241 		nsect = geom_nsect;
1242 		ntrack = geom_ntrack;
1243 		cpg = geom_cpg;
1244 		dprintf(("\nDeBuG Using Geometry\n"));
1245 		/*
1246 		 * 32K based on max block size of 64K, and rotational layout
1247 		 * test of nsect <= (256 * sectors/block).  Current block size
1248 		 * limit is not 64K, but it's growing soon.
1249 		 */
1250 		range_check(&nsect, "nsect", 1, 32768, DFLNSECT, nsect_flag);
1251 		/*
1252 		 * ntrack is the number of tracks per cylinder.
1253 		 * The ntrack value must be between 1 and the total number of
1254 		 * sectors in the file system.
1255 		 */
1256 		range_check(&ntrack, "ntrack", 1,
1257 		    fssize_db > INT_MAX ? INT_MAX : (uint32_t)fssize_db,
1258 		    DFLNTRAK, ntrack_flag);
1259 	}
1260 
1261 	range_check(&apc, "apc", 0, nsect - 1, 0, apc_flag);
1262 
1263 	if (mtb == 'y')
1264 		fragsize = bsize;
1265 
1266 	range_check(&fragsize, "fragsize", sectorsize, bsize,
1267 	    MAX(bsize / MAXFRAG, MIN(DESFRAGSIZE, bsize)), fragsize_flag);
1268 
1269 	if ((bsize / MAXFRAG) > fragsize) {
1270 		(void) fprintf(stderr, gettext(
1271 "fragment size %ld is too small, minimum with block size %ld is %ld\n"),
1272 		    fragsize, bsize, bsize / MAXFRAG);
1273 		(void) fprintf(stderr,
1274 		    gettext("mkfs: fragsize reset to minimum %ld\n"),
1275 		    bsize / MAXFRAG);
1276 		fragsize = bsize / MAXFRAG;
1277 	}
1278 
1279 	if (!POWEROF2(fragsize)) {
1280 		(void) fprintf(stderr,
1281 		    gettext("fragment size must be a power of 2, not %ld\n"),
1282 		    fragsize);
1283 		fragsize = MAX(bsize / MAXFRAG, MIN(DESFRAGSIZE, bsize));
1284 		(void) fprintf(stderr,
1285 		    gettext("mkfs: fragsize reset to %ld\n"),
1286 		    fragsize);
1287 	}
1288 
1289 	/* At this point, bsize must be >= fragsize, so no need to check it */
1290 
1291 	if (bsize < PAGESIZE) {
1292 		(void) fprintf(stderr, gettext(
1293 		    "WARNING: filesystem block size (%ld) is smaller than "
1294 		    "memory page size (%ld).\nResulting filesystem can not be "
1295 		    "mounted on this system.\n\n"),
1296 		    bsize, (long)PAGESIZE);
1297 	}
1298 
1299 	range_check(&rps, "rps", 1, 1000, DEFHZ, rps_flag);
1300 	range_check(&minfree, "free", 0, 99, MINFREE, minfree_flag);
1301 	range_check(&nrpos, "nrpos", 1, nsect, MIN(nsect, NRPOS), nrpos_flag);
1302 
1303 	/*
1304 	 * nbpi is variable, but 2MB seems a reasonable upper limit,
1305 	 * as 4MB tends to cause problems (using otherwise-default
1306 	 * parameters).  The true limit is where we end up with one
1307 	 * inode per cylinder group.  If this file system is being
1308 	 * configured for multi-terabyte access, nbpi must be at least 1MB.
1309 	 */
1310 	if (mtb == 'y' && nbpi < MTB_NBPI) {
1311 		if (nbpi_flag != RC_DEFAULT)
1312 			(void) fprintf(stderr, gettext("mkfs: bad value for "
1313 			    "nbpi: must be at least 1048576 for multi-terabyte,"
1314 			    " nbpi reset to default 1048576\n"));
1315 		nbpi = MTB_NBPI;
1316 	}
1317 
1318 	if (mtb == 'y')
1319 		range_check(&nbpi, "nbpi", MTB_NBPI, 2 * MB, MTB_NBPI,
1320 		    nbpi_flag);
1321 	else
1322 		range_check(&nbpi, "nbpi", DEV_BSIZE, 2 * MB, NBPI, nbpi_flag);
1323 
1324 	/*
1325 	 * maxcpg is another variably-limited parameter.  Calculate
1326 	 * the limit based on what we've got for its dependent
1327 	 * variables.  Effectively, it's how much space is left in the
1328 	 * superblock after all the other bits are accounted for.  We
1329 	 * only fill in sblock fields so we can use MAXIpG.
1330 	 *
1331 	 * If the calculation of maxcpg below (for the mtb == 'n'
1332 	 * case) is changed, update newfs as well.
1333 	 *
1334 	 * For old-style, non-MTB format file systems, use the old
1335 	 * algorithm for calculating the maximum cylinder group size,
1336 	 * even though it limits the cylinder group more than necessary.
1337 	 * Since layout can affect performance, we don't want to change
1338 	 * the default layout for non-MTB file systems at this time.
1339 	 * However, for MTB file systems, use the new maxcpg calculation,
1340 	 * which really maxes out the cylinder group size.
1341 	 */
1342 
1343 	sblock.fs_bsize = bsize;
1344 	sblock.fs_inopb = sblock.fs_bsize / sizeof (struct dinode);
1345 
1346 	if (mtb == 'n') {
1347 		maxcpg = (bsize - sizeof (struct cg) -
1348 		    howmany(MAXIpG(&sblock), NBBY)) /
1349 		    (sizeof (long) + nrpos * sizeof (short) +
1350 		    nsect / (MAXFRAG * NBBY));
1351 	} else {
1352 		maxcpg = compute_maxcpg(bsize, fragsize, nbpi, nrpos,
1353 		    nsect * ntrack);
1354 	}
1355 
1356 	dprintf(("DeBuG cpg : %ld\n", cpg));
1357 	/*
1358 	 * Increase the cpg to maxcpg if either newfs was invoked
1359 	 * with -T option or if mkfs wants to create a mtb file system
1360 	 * and if the user has not specified the cpg.
1361 	 */
1362 	if (cpg == -1 || (mtb == 'y' && cpg_flag == RC_DEFAULT))
1363 		cpg = maxcpg;
1364 	dprintf(("DeBuG cpg : %ld\n", cpg));
1365 
1366 	/*
1367 	 * mincpg is variable in complex ways, so we really can't
1368 	 * do a sane lower-end limit check at this point.
1369 	 */
1370 	range_check(&cpg, "cgsize", 1, maxcpg, MIN(maxcpg, DESCPG), cpg_flag);
1371 
1372 	/*
1373 	 * get the controller info
1374 	 */
1375 	ismdd = 0;
1376 	islog = 0;
1377 	islogok = 0;
1378 	waslog = 0;
1379 
1380 	if (ioctl(fsi, DKIOCINFO, &dkcinfo) == 0)
1381 		/*
1382 		 * if it is an MDD (disksuite) device
1383 		 */
1384 		if (dkcinfo.dki_ctype == DKC_MD) {
1385 			ismdd++;
1386 			/*
1387 			 * check the logging device
1388 			 */
1389 			if (ioctl(fsi, _FIOISLOG, NULL) == 0) {
1390 				islog++;
1391 				if (ioctl(fsi, _FIOISLOGOK, NULL) == 0)
1392 					islogok++;
1393 			}
1394 		}
1395 
1396 	/*
1397 	 * Do not grow the file system, but print on stdout the maximum
1398 	 * size in sectors to which the file system can be increased.
1399 	 * The calculated size is limited by fssize_db.
1400 	 * Note that we don't lock the filesystem and therefore under rare
1401 	 * conditions (the filesystem is mounted, the free block count is
1402 	 * almost zero, and the superuser is still changing it) the calculated
1403 	 * size can be imprecise.
1404 	 */
1405 	if (Pflag) {
1406 		(void) printf("%llu\n", probe_summaryinfo());
1407 		exit(0);
1408 	}
1409 
1410 	/*
1411 	 * If we're growing an existing filesystem, then we're about
1412 	 * to start doing things that can require recovery efforts if
1413 	 * we get interrupted, so make sure we get a chance to do so.
1414 	 */
1415 	if (grow) {
1416 		sigact.sa_handler = recover_from_sigint;
1417 		sigemptyset(&sigact.sa_mask);
1418 		sigact.sa_flags = SA_RESTART;
1419 
1420 		if (sigaction(SIGINT, &sigact, (struct sigaction *)NULL) < 0) {
1421 			perror(gettext("Could not register SIGINT handler"));
1422 			lockexit(3);
1423 		}
1424 	}
1425 
1426 	if (!Nflag) {
1427 		/*
1428 		 * Check if MNTTAB is trustable
1429 		 */
1430 		if (statvfs64(MNTTAB, &fs) < 0) {
1431 			(void) fprintf(stderr, gettext("can't statvfs %s\n"),
1432 			    MNTTAB);
1433 			exit(32);
1434 		}
1435 
1436 		if (strcmp(MNTTYPE_MNTFS, fs.f_basetype) != 0) {
1437 			(void) fprintf(stderr, gettext(
1438 			    "%s file system type is not %s, can't mkfs\n"),
1439 			    MNTTAB, MNTTYPE_MNTFS);
1440 			exit(32);
1441 		}
1442 
1443 		special = getfullblkname(fsys);
1444 		checkdev(fsys, special);
1445 
1446 		/*
1447 		 * If we found the block device name,
1448 		 * then check the mount table.
1449 		 * if mounted, and growing write lock the file system
1450 		 *
1451 		 */
1452 		if ((special != NULL) && (*special != '\0')) {
1453 			if ((mnttab = fopen(MNTTAB, "r")) == NULL) {
1454 				(void) fprintf(stderr, gettext(
1455 				    "can't open %s\n"), MNTTAB);
1456 				exit(32);
1457 			}
1458 			while ((getmntent(mnttab, &mntp)) == NULL) {
1459 				if (grow) {
1460 					checkmount(&mntp, special);
1461 					continue;
1462 				}
1463 				if (strcmp(special, mntp.mnt_special) == 0) {
1464 					(void) fprintf(stderr, gettext(
1465 					    "%s is mounted, can't mkfs\n"),
1466 					    special);
1467 					exit(32);
1468 				}
1469 			}
1470 			(void) fclose(mnttab);
1471 		}
1472 
1473 		if (directory && (ismounted == 0)) {
1474 			(void) fprintf(stderr, gettext("%s is not mounted\n"),
1475 			    special);
1476 			lockexit(32);
1477 		}
1478 
1479 		fso = (grow) ? open64(fsys, O_WRONLY) : creat64(fsys, 0666);
1480 		if (fso < 0) {
1481 			saverr = errno;
1482 			(void) fprintf(stderr,
1483 			    gettext("%s: cannot create: %s\n"),
1484 			    fsys, strerror(saverr));
1485 			lockexit(32);
1486 		}
1487 
1488 	} else {
1489 
1490 		/*
1491 		 * For the -N case, a file descriptor is needed for the llseek()
1492 		 * in wtfs(). See the comment in wtfs() for more information.
1493 		 *
1494 		 * Get a file descriptor that's read-only so that this code
1495 		 * doesn't accidentally write to the file.
1496 		 */
1497 		fso = open64(fsys, O_RDONLY);
1498 		if (fso < 0) {
1499 			saverr = errno;
1500 			(void) fprintf(stderr, gettext("%s: cannot open: %s\n"),
1501 			    fsys, strerror(saverr));
1502 			lockexit(32);
1503 		}
1504 	}
1505 
1506 	/*
1507 	 * seed random # generator (for ic_generation)
1508 	 */
1509 #ifdef MKFS_DEBUG
1510 	srand48(12962);	/* reproducible results */
1511 #else
1512 	srand48((long)(time((time_t *)NULL) + getpid()));
1513 #endif
1514 
1515 	if (grow) {
1516 		growinit(fsys);
1517 		goto grow00;
1518 	}
1519 
1520 	/*
1521 	 * Validate the given file system size.
1522 	 * Verify that its last block can actually be accessed.
1523 	 *
1524 	 * Note: it's ok to use sblock as a buffer because it is immediately
1525 	 * overwritten by the rdfs() of the superblock in the next line.
1526 	 *
1527 	 * ToDo: Because the size checking is done in rdfs()/wtfs(), the
1528 	 * error message for specifying an illegal size is very unfriendly.
1529 	 * In the future, one could replace the rdfs()/wtfs() calls
1530 	 * below with in-line calls to read() or write(). This allows better
1531 	 * error messages to be put in place.
1532 	 */
1533 	rdfs(fssize_db - 1, (int)sectorsize, (char *)&sblock);
1534 
1535 	/*
1536 	 * make the fs unmountable
1537 	 */
1538 	rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
1539 	sblock.fs_magic = -1;
1540 	sblock.fs_clean = FSBAD;
1541 	sblock.fs_state = FSOKAY - sblock.fs_time;
1542 	wtfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
1543 	bzero(&sblock, (size_t)sbsize);
1544 
1545 	sblock.fs_nsect = nsect;
1546 	sblock.fs_ntrak = ntrack;
1547 
1548 	/*
1549 	 * Validate specified/determined spc
1550 	 * and calculate minimum cylinders per group.
1551 	 */
1552 
1553 	/*
1554 	 * sectors/cyl = tracks/cyl * sectors/track
1555 	 */
1556 	sblock.fs_spc = sblock.fs_ntrak * sblock.fs_nsect;
1557 
1558 grow00:
1559 	if (apc_flag) {
1560 		sblock.fs_spc -= apc;
1561 	}
1562 	/*
1563 	 * Have to test for this separately from apc_flag, due to
1564 	 * the growfs case....
1565 	 */
1566 	if (sblock.fs_spc != sblock.fs_ntrak * sblock.fs_nsect) {
1567 		spc_flag = 1;
1568 	}
1569 	if (grow)
1570 		goto grow10;
1571 
1572 	sblock.fs_nrpos = nrpos;
1573 	sblock.fs_bsize = bsize;
1574 	sblock.fs_fsize = fragsize;
1575 	sblock.fs_minfree = minfree;
1576 
1577 grow10:
1578 	if (nbpi < sblock.fs_fsize) {
1579 		(void) fprintf(stderr, gettext(
1580 		"warning: wasteful data byte allocation / inode (nbpi):\n"));
1581 		(void) fprintf(stderr, gettext(
1582 		    "%ld smaller than allocatable fragment size of %d\n"),
1583 		    nbpi, sblock.fs_fsize);
1584 	}
1585 	if (grow)
1586 		goto grow20;
1587 
1588 	if (opt == 's')
1589 		sblock.fs_optim = FS_OPTSPACE;
1590 	else
1591 		sblock.fs_optim = FS_OPTTIME;
1592 
1593 	sblock.fs_bmask = ~(sblock.fs_bsize - 1);
1594 	sblock.fs_fmask = ~(sblock.fs_fsize - 1);
1595 	/*
1596 	 * Planning now for future expansion.
1597 	 */
1598 #if defined(_BIG_ENDIAN)
1599 		sblock.fs_qbmask.val[0] = 0;
1600 		sblock.fs_qbmask.val[1] = ~sblock.fs_bmask;
1601 		sblock.fs_qfmask.val[0] = 0;
1602 		sblock.fs_qfmask.val[1] = ~sblock.fs_fmask;
1603 #endif
1604 #if defined(_LITTLE_ENDIAN)
1605 		sblock.fs_qbmask.val[0] = ~sblock.fs_bmask;
1606 		sblock.fs_qbmask.val[1] = 0;
1607 		sblock.fs_qfmask.val[0] = ~sblock.fs_fmask;
1608 		sblock.fs_qfmask.val[1] = 0;
1609 #endif
1610 	for (sblock.fs_bshift = 0, i = sblock.fs_bsize; i > 1; i >>= 1)
1611 		sblock.fs_bshift++;
1612 	for (sblock.fs_fshift = 0, i = sblock.fs_fsize; i > 1; i >>= 1)
1613 		sblock.fs_fshift++;
1614 	sblock.fs_frag = numfrags(&sblock, sblock.fs_bsize);
1615 	for (sblock.fs_fragshift = 0, i = sblock.fs_frag; i > 1; i >>= 1)
1616 		sblock.fs_fragshift++;
1617 	if (sblock.fs_frag > MAXFRAG) {
1618 		(void) fprintf(stderr, gettext(
1619 	"fragment size %d is too small, minimum with block size %d is %d\n"),
1620 		    sblock.fs_fsize, sblock.fs_bsize,
1621 		    sblock.fs_bsize / MAXFRAG);
1622 		lockexit(32);
1623 	}
1624 	sblock.fs_nindir = sblock.fs_bsize / sizeof (daddr32_t);
1625 	sblock.fs_inopb = sblock.fs_bsize / sizeof (struct dinode);
1626 	sblock.fs_nspf = sblock.fs_fsize / sectorsize;
1627 	for (sblock.fs_fsbtodb = 0, i = NSPF(&sblock); i > 1; i >>= 1)
1628 		sblock.fs_fsbtodb++;
1629 
1630 	/*
1631 	 * Compute the super-block, cylinder group, and inode blocks.
1632 	 * Note that these "blkno" are really fragment addresses.
1633 	 * For example, on an 8K/1K (block/fragment) system, fs_sblkno is 16,
1634 	 * fs_cblkno is 24, and fs_iblkno is 32. This is why CGSIZE is so
1635 	 * important: only 1 FS block is allocated for the cg struct (fragment
1636 	 * numbers 24 through 31).
1637 	 */
1638 	sblock.fs_sblkno =
1639 	    roundup(howmany(bbsize + sbsize, sblock.fs_fsize), sblock.fs_frag);
1640 	sblock.fs_cblkno = (daddr32_t)(sblock.fs_sblkno +
1641 	    roundup(howmany(sbsize, sblock.fs_fsize), sblock.fs_frag));
1642 	sblock.fs_iblkno = sblock.fs_cblkno + sblock.fs_frag;
1643 
1644 	sblock.fs_cgoffset = roundup(
1645 	    howmany(sblock.fs_nsect, NSPF(&sblock)), sblock.fs_frag);
1646 	for (sblock.fs_cgmask = -1, i = sblock.fs_ntrak; i > 1; i >>= 1)
1647 		sblock.fs_cgmask <<= 1;
1648 	if (!POWEROF2(sblock.fs_ntrak))
1649 		sblock.fs_cgmask <<= 1;
1650 	/*
1651 	 * Validate specified/determined spc
1652 	 * and calculate minimum cylinders per group.
1653 	 */
1654 
1655 	for (sblock.fs_cpc = NSPB(&sblock), i = sblock.fs_spc;
1656 	    sblock.fs_cpc > 1 && (i & 1) == 0;
1657 	    sblock.fs_cpc >>= 1, i >>= 1)
1658 		/* void */;
1659 	mincpc = sblock.fs_cpc;
1660 
1661 	/* if these calculations are changed, check dump_fscmd also */
1662 	bpcg = (uint64_t)sblock.fs_spc * sectorsize;
1663 	inospercg = (uint64_t)roundup(bpcg / sizeof (struct dinode),
1664 	    INOPB(&sblock));
1665 	if (inospercg > MAXIpG(&sblock))
1666 		inospercg = MAXIpG(&sblock);
1667 	used = (uint64_t)(sblock.fs_iblkno + inospercg /
1668 	    INOPF(&sblock)) * NSPF(&sblock);
1669 	mincpgcnt = (long)howmany((uint64_t)sblock.fs_cgoffset *
1670 	    (~sblock.fs_cgmask) + used, sblock.fs_spc);
1671 	mincpg = roundup(mincpgcnt, mincpc);
1672 	/*
1673 	 * Insure that cylinder group with mincpg has enough space
1674 	 * for block maps
1675 	 */
1676 	sblock.fs_cpg = mincpg;
1677 	sblock.fs_ipg = (int32_t)inospercg;
1678 	mapcramped = 0;
1679 
1680 	/*
1681 	 * Make sure the cg struct fits within the file system block.
1682 	 * Use larger block sizes until it fits
1683 	 */
1684 	while (CGSIZE(&sblock) > sblock.fs_bsize) {
1685 		mapcramped = 1;
1686 		if (sblock.fs_bsize < MAXBSIZE) {
1687 			sblock.fs_bsize <<= 1;
1688 			if ((i & 1) == 0) {
1689 				i >>= 1;
1690 			} else {
1691 				sblock.fs_cpc <<= 1;
1692 				mincpc <<= 1;
1693 				mincpg = roundup(mincpgcnt, mincpc);
1694 				sblock.fs_cpg = mincpg;
1695 			}
1696 			sblock.fs_frag <<= 1;
1697 			sblock.fs_fragshift += 1;
1698 			if (sblock.fs_frag <= MAXFRAG)
1699 				continue;
1700 		}
1701 
1702 		/*
1703 		 * Looped far enough. The fragment is now as large as the
1704 		 * filesystem block!
1705 		 */
1706 		if (sblock.fs_fsize == sblock.fs_bsize) {
1707 			(void) fprintf(stderr, gettext(
1708 		    "There is no block size that can support this disk\n"));
1709 			lockexit(32);
1710 		}
1711 
1712 		/*
1713 		 * Try a larger fragment. Double the fragment size.
1714 		 */
1715 		sblock.fs_frag >>= 1;
1716 		sblock.fs_fragshift -= 1;
1717 		sblock.fs_fsize <<= 1;
1718 		sblock.fs_nspf <<= 1;
1719 	}
1720 	/*
1721 	 * Insure that cylinder group with mincpg has enough space for inodes
1722 	 */
1723 	inodecramped = 0;
1724 	used *= sectorsize;
1725 	nbytes64 = (uint64_t)mincpg * bpcg - used;
1726 	inospercg = (uint64_t)roundup((nbytes64 / nbpi), INOPB(&sblock));
1727 	sblock.fs_ipg = (int32_t)inospercg;
1728 	while (inospercg > MAXIpG(&sblock)) {
1729 		inodecramped = 1;
1730 		if (mincpc == 1 || sblock.fs_frag == 1 ||
1731 		    sblock.fs_bsize == MINBSIZE)
1732 			break;
1733 		nbytes64 = (uint64_t)mincpg * bpcg - used;
1734 		(void) fprintf(stderr,
1735 		    gettext("With a block size of %d %s %lu\n"),
1736 		    sblock.fs_bsize, gettext("minimum bytes per inode is"),
1737 		    (uint32_t)(nbytes64 / MAXIpG(&sblock) + 1));
1738 		sblock.fs_bsize >>= 1;
1739 		sblock.fs_frag >>= 1;
1740 		sblock.fs_fragshift -= 1;
1741 		mincpc >>= 1;
1742 		sblock.fs_cpg = roundup(mincpgcnt, mincpc);
1743 		if (CGSIZE(&sblock) > sblock.fs_bsize) {
1744 			sblock.fs_bsize <<= 1;
1745 			break;
1746 		}
1747 		mincpg = sblock.fs_cpg;
1748 		nbytes64 = (uint64_t)mincpg * bpcg - used;
1749 		inospercg = (uint64_t)roundup((nbytes64 / nbpi),
1750 		    INOPB(&sblock));
1751 		sblock.fs_ipg = (int32_t)inospercg;
1752 	}
1753 	if (inodecramped) {
1754 		if (inospercg > MAXIpG(&sblock)) {
1755 			nbytes64 = (uint64_t)mincpg * bpcg - used;
1756 			(void) fprintf(stderr, gettext(
1757 			    "Minimum bytes per inode is %d\n"),
1758 			    (uint32_t)(nbytes64 / MAXIpG(&sblock) + 1));
1759 		} else if (!mapcramped) {
1760 			(void) fprintf(stderr, gettext(
1761 	    "With %ld bytes per inode, minimum cylinders per group is %ld\n"),
1762 			    nbpi, mincpg);
1763 		}
1764 	}
1765 	if (mapcramped) {
1766 		(void) fprintf(stderr, gettext(
1767 		    "With %d sectors per cylinder, minimum cylinders "
1768 		    "per group is %ld\n"),
1769 		    sblock.fs_spc, mincpg);
1770 	}
1771 	if (inodecramped || mapcramped) {
1772 		/*
1773 		 * To make this at least somewhat comprehensible in
1774 		 * the world of i18n, figure out what we're going to
1775 		 * say and then say it all at one time.  The days of
1776 		 * needing to scrimp on string space are behind us....
1777 		 */
1778 		if ((sblock.fs_bsize != bsize) &&
1779 		    (sblock.fs_fsize != fragsize)) {
1780 			(void) fprintf(stderr, gettext(
1781 	    "This requires the block size to be changed from %ld to %d\n"
1782 	    "and the fragment size to be changed from %ld to %d\n"),
1783 			    bsize, sblock.fs_bsize,
1784 			    fragsize, sblock.fs_fsize);
1785 		} else if (sblock.fs_bsize != bsize) {
1786 			(void) fprintf(stderr, gettext(
1787 	    "This requires the block size to be changed from %ld to %d\n"),
1788 			    bsize, sblock.fs_bsize);
1789 		} else if (sblock.fs_fsize != fragsize) {
1790 			(void) fprintf(stderr, gettext(
1791 	    "This requires the fragment size to be changed from %ld to %d\n"),
1792 			    fragsize, sblock.fs_fsize);
1793 		} else {
1794 			(void) fprintf(stderr, gettext(
1795 	    "Unable to make filesystem fit with the given constraints\n"));
1796 		}
1797 		(void) fprintf(stderr, gettext(
1798 		    "Please re-run mkfs with corrected parameters\n"));
1799 		lockexit(32);
1800 	}
1801 	/*
1802 	 * Calculate the number of cylinders per group
1803 	 */
1804 	sblock.fs_cpg = cpg;
1805 	if (sblock.fs_cpg % mincpc != 0) {
1806 		(void) fprintf(stderr, gettext(
1807 		    "Warning: cylinder groups must have a multiple "
1808 		    "of %ld cylinders with the given\n         parameters\n"),
1809 		    mincpc);
1810 		sblock.fs_cpg = roundup(sblock.fs_cpg, mincpc);
1811 		(void) fprintf(stderr, gettext("Rounded cgsize up to %d\n"),
1812 		    sblock.fs_cpg);
1813 	}
1814 	/*
1815 	 * Must insure there is enough space for inodes
1816 	 */
1817 	/* if these calculations are changed, check dump_fscmd also */
1818 	nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
1819 	sblock.fs_ipg = roundup((uint32_t)(nbytes64 / nbpi), INOPB(&sblock));
1820 
1821 	/*
1822 	 * Slim down cylinders per group, until the inodes can fit.
1823 	 */
1824 	while (sblock.fs_ipg > MAXIpG(&sblock)) {
1825 		inodecramped = 1;
1826 		sblock.fs_cpg -= mincpc;
1827 		nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
1828 		sblock.fs_ipg = roundup((uint32_t)(nbytes64 / nbpi),
1829 		    INOPB(&sblock));
1830 	}
1831 	/*
1832 	 * Must insure there is enough space to hold block map.
1833 	 * Cut down on cylinders per group, until the cg struct fits in a
1834 	 * filesystem block.
1835 	 */
1836 	while (CGSIZE(&sblock) > sblock.fs_bsize) {
1837 		mapcramped = 1;
1838 		sblock.fs_cpg -= mincpc;
1839 		nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
1840 		sblock.fs_ipg = roundup((uint32_t)(nbytes64 / nbpi),
1841 		    INOPB(&sblock));
1842 	}
1843 	sblock.fs_fpg = (sblock.fs_cpg * sblock.fs_spc) / NSPF(&sblock);
1844 	if ((sblock.fs_cpg * sblock.fs_spc) % NSPB(&sblock) != 0) {
1845 		(void) fprintf(stderr,
1846 		gettext("newfs: panic (fs_cpg * fs_spc) %% NSPF != 0\n"));
1847 		lockexit(32);
1848 	}
1849 	if (sblock.fs_cpg < mincpg) {
1850 		(void) fprintf(stderr, gettext(
1851 "With the given parameters, cgsize must be at least %ld; please re-run mkfs\n"),
1852 		    mincpg);
1853 		lockexit(32);
1854 	}
1855 	sblock.fs_cgsize = fragroundup(&sblock, CGSIZE(&sblock));
1856 grow20:
1857 	/*
1858 	 * Now have size for file system and nsect and ntrak.
1859 	 * Determine number of cylinders and blocks in the file system.
1860 	 */
1861 	fssize_frag = (int64_t)dbtofsb(&sblock, fssize_db);
1862 	if (fssize_frag > INT_MAX) {
1863 		(void) fprintf(stderr, gettext(
1864 "There are too many fragments in the system, increase fragment size\n"),
1865 		    mincpg);
1866 		lockexit(32);
1867 	}
1868 	sblock.fs_size = (int32_t)fssize_frag;
1869 	sblock.fs_ncyl = (int32_t)(fssize_frag * NSPF(&sblock) / sblock.fs_spc);
1870 	if (fssize_frag * NSPF(&sblock) >
1871 	    (uint64_t)sblock.fs_ncyl * sblock.fs_spc) {
1872 		sblock.fs_ncyl++;
1873 		warn = 1;
1874 	}
1875 	if (sblock.fs_ncyl < 1) {
1876 		(void) fprintf(stderr, gettext(
1877 		    "file systems must have at least one cylinder\n"));
1878 		lockexit(32);
1879 	}
1880 	if (grow)
1881 		goto grow30;
1882 	/*
1883 	 * Determine feasability/values of rotational layout tables.
1884 	 *
1885 	 * The size of the rotational layout tables is limited by the size
1886 	 * of the file system block, fs_bsize.  The amount of space
1887 	 * available for tables is calculated as (fs_bsize - sizeof (struct
1888 	 * fs)).  The size of these tables is inversely proportional to the
1889 	 * block size of the file system. The size increases if sectors per
1890 	 * track are not powers of two, because more cylinders must be
1891 	 * described by the tables before the rotational pattern repeats
1892 	 * (fs_cpc).
1893 	 */
1894 	sblock.fs_postblformat = FS_DYNAMICPOSTBLFMT;
1895 	sblock.fs_sbsize = fragroundup(&sblock, sizeof (struct fs));
1896 	sblock.fs_npsect = sblock.fs_nsect;
1897 	if (sblock.fs_ntrak == 1) {
1898 		sblock.fs_cpc = 0;
1899 		goto next;
1900 	}
1901 	postblsize = sblock.fs_nrpos * sblock.fs_cpc * sizeof (short);
1902 	rotblsize = sblock.fs_cpc * sblock.fs_spc / NSPB(&sblock);
1903 	totalsbsize = sizeof (struct fs) + rotblsize;
1904 
1905 	/* do static allocation if nrpos == 8 and fs_cpc == 16  */
1906 	if (sblock.fs_nrpos == 8 && sblock.fs_cpc <= 16) {
1907 		/* use old static table space */
1908 		sblock.fs_postbloff = (char *)(&sblock.fs_opostbl[0][0]) -
1909 		    (char *)(&sblock.fs_link);
1910 		sblock.fs_rotbloff = &sblock.fs_space[0] -
1911 		    (uchar_t *)(&sblock.fs_link);
1912 	} else {
1913 		/* use 4.3 dynamic table space */
1914 		sblock.fs_postbloff = &sblock.fs_space[0] -
1915 		    (uchar_t *)(&sblock.fs_link);
1916 		sblock.fs_rotbloff = sblock.fs_postbloff + postblsize;
1917 		totalsbsize += postblsize;
1918 	}
1919 	if (totalsbsize > sblock.fs_bsize ||
1920 	    sblock.fs_nsect > (1 << NBBY) * NSPB(&sblock)) {
1921 		(void) fprintf(stderr, gettext(
1922 		    "Warning: insufficient space in super block for\n"
1923 		    "rotational layout tables with nsect %d, ntrack %d, "
1924 		    "and nrpos %d.\nOmitting tables - file system "
1925 		    "performance may be impaired.\n"),
1926 		    sblock.fs_nsect, sblock.fs_ntrak, sblock.fs_nrpos);
1927 
1928 		/*
1929 		 * Setting fs_cpc to 0 tells alloccgblk() in ufs_alloc.c to
1930 		 * ignore the positional layout table and rotational
1931 		 * position table.
1932 		 */
1933 		sblock.fs_cpc = 0;
1934 		goto next;
1935 	}
1936 	sblock.fs_sbsize = fragroundup(&sblock, totalsbsize);
1937 
1938 
1939 	/*
1940 	 * calculate the available blocks for each rotational position
1941 	 */
1942 	for (cylno = 0; cylno < sblock.fs_cpc; cylno++)
1943 		for (rpos = 0; rpos < sblock.fs_nrpos; rpos++)
1944 			fs_postbl(&sblock, cylno)[rpos] = -1;
1945 	for (i = (rotblsize - 1) * sblock.fs_frag;
1946 	    i >= 0; i -= sblock.fs_frag) {
1947 		cylno = cbtocylno(&sblock, i);
1948 		rpos = cbtorpos(&sblock, i);
1949 		blk = fragstoblks(&sblock, i);
1950 		if (fs_postbl(&sblock, cylno)[rpos] == -1)
1951 			fs_rotbl(&sblock)[blk] = 0;
1952 		else
1953 			fs_rotbl(&sblock)[blk] =
1954 			    fs_postbl(&sblock, cylno)[rpos] - blk;
1955 		fs_postbl(&sblock, cylno)[rpos] = blk;
1956 	}
1957 next:
1958 grow30:
1959 	/*
1960 	 * Compute/validate number of cylinder groups.
1961 	 * Note that if an excessively large filesystem is specified
1962 	 * (e.g., more than 16384 cylinders for an 8K filesystem block), it
1963 	 * does not get detected until checksummarysize()
1964 	 */
1965 	sblock.fs_ncg = sblock.fs_ncyl / sblock.fs_cpg;
1966 	if (sblock.fs_ncyl % sblock.fs_cpg)
1967 		sblock.fs_ncg++;
1968 	sblock.fs_dblkno = sblock.fs_iblkno + sblock.fs_ipg / INOPF(&sblock);
1969 	i = MIN(~sblock.fs_cgmask, sblock.fs_ncg - 1);
1970 	ibpcl = cgdmin(&sblock, i) - cgbase(&sblock, i);
1971 	if (ibpcl >= sblock.fs_fpg) {
1972 		(void) fprintf(stderr, gettext(
1973 		    "inode blocks/cyl group (%d) >= data blocks (%d)\n"),
1974 		    cgdmin(&sblock, i) - cgbase(&sblock, i) / sblock.fs_frag,
1975 		    sblock.fs_fpg / sblock.fs_frag);
1976 		if ((ibpcl < 0) || (sblock.fs_fpg < 0)) {
1977 			(void) fprintf(stderr, gettext(
1978 	    "number of cylinders per cylinder group (%d) must be decreased.\n"),
1979 			    sblock.fs_cpg);
1980 		} else {
1981 			(void) fprintf(stderr, gettext(
1982 	    "number of cylinders per cylinder group (%d) must be increased.\n"),
1983 			    sblock.fs_cpg);
1984 		}
1985 		(void) fprintf(stderr, gettext(
1986 "Note that cgsize may have been adjusted to allow struct cg to fit.\n"));
1987 		lockexit(32);
1988 	}
1989 	j = sblock.fs_ncg - 1;
1990 	if ((i = fssize_frag - j * sblock.fs_fpg) < sblock.fs_fpg &&
1991 	    cgdmin(&sblock, j) - cgbase(&sblock, j) > i) {
1992 		(void) fprintf(stderr, gettext(
1993 		    "Warning: inode blocks/cyl group (%d) >= data "
1994 		    "blocks (%ld) in last\n    cylinder group. This "
1995 		    "implies %ld sector(s) cannot be allocated.\n"),
1996 		    (cgdmin(&sblock, j) - cgbase(&sblock, j)) / sblock.fs_frag,
1997 		    i / sblock.fs_frag, i * NSPF(&sblock));
1998 		/*
1999 		 * If there is only one cylinder group and that is not even
2000 		 * big enough to hold the inodes, exit.
2001 		 */
2002 		if (sblock.fs_ncg == 1)
2003 			cg_too_small = 1;
2004 		sblock.fs_ncg--;
2005 		sblock.fs_ncyl -= sblock.fs_ncyl % sblock.fs_cpg;
2006 		sblock.fs_size = fssize_frag =
2007 		    (int64_t)sblock.fs_ncyl * (int64_t)sblock.fs_spc /
2008 		    (int64_t)NSPF(&sblock);
2009 		warn = 0;
2010 	}
2011 	if (warn && !spc_flag) {
2012 		(void) fprintf(stderr, gettext(
2013 		    "Warning: %d sector(s) in last cylinder unallocated\n"),
2014 		    sblock.fs_spc - (uint32_t)(fssize_frag * NSPF(&sblock) -
2015 		    (uint64_t)(sblock.fs_ncyl - 1) * sblock.fs_spc));
2016 	}
2017 	/*
2018 	 * fill in remaining fields of the super block
2019 	 */
2020 
2021 	/*
2022 	 * The csum records are stored in cylinder group 0, starting at
2023 	 * cgdmin, the first data block.
2024 	 */
2025 	sblock.fs_csaddr = cgdmin(&sblock, 0);
2026 	sblock.fs_cssize =
2027 	    fragroundup(&sblock, sblock.fs_ncg * sizeof (struct csum));
2028 	i = sblock.fs_bsize / sizeof (struct csum);
2029 	sblock.fs_csmask = ~(i - 1);
2030 	for (sblock.fs_csshift = 0; i > 1; i >>= 1)
2031 		sblock.fs_csshift++;
2032 	fscs = (struct csum *)calloc(1, sblock.fs_cssize);
2033 
2034 	checksummarysize();
2035 	if (mtb == 'y') {
2036 		sblock.fs_magic = MTB_UFS_MAGIC;
2037 		sblock.fs_version = MTB_UFS_VERSION_1;
2038 	} else {
2039 		sblock.fs_magic = FS_MAGIC;
2040 		if (use_efi_dflts)
2041 			sblock.fs_version = UFS_EFISTYLE4NONEFI_VERSION_2;
2042 		else
2043 			sblock.fs_version = UFS_VERSION_MIN;
2044 	}
2045 
2046 	if (grow) {
2047 		bcopy((caddr_t)grow_fscs, (caddr_t)fscs, (int)grow_fs_cssize);
2048 		extendsummaryinfo();
2049 		goto grow40;
2050 	}
2051 	sblock.fs_rotdelay = rotdelay;
2052 	sblock.fs_maxcontig = maxcontig;
2053 	sblock.fs_maxbpg = MAXBLKPG(sblock.fs_bsize);
2054 
2055 	sblock.fs_rps = rps;
2056 	sblock.fs_cgrotor = 0;
2057 	sblock.fs_cstotal.cs_ndir = 0;
2058 	sblock.fs_cstotal.cs_nbfree = 0;
2059 	sblock.fs_cstotal.cs_nifree = 0;
2060 	sblock.fs_cstotal.cs_nffree = 0;
2061 	sblock.fs_fmod = 0;
2062 	sblock.fs_ronly = 0;
2063 	sblock.fs_time = mkfstime;
2064 	sblock.fs_state = FSOKAY - sblock.fs_time;
2065 	sblock.fs_clean = FSCLEAN;
2066 grow40:
2067 
2068 	/*
2069 	 * If all that's needed is a dump of the superblock we
2070 	 * would use by default, we've got it now.  So, splat it
2071 	 * out and leave.
2072 	 */
2073 	if (rflag) {
2074 		dump_sblock();
2075 		lockexit(0);
2076 	}
2077 	/*
2078 	 * Dump out summary information about file system.
2079 	 */
2080 	(void) fprintf(stderr, gettext(
2081 	    "%s:\t%lld sectors in %d cylinders of %d tracks, %d sectors\n"),
2082 	    fsys, (uint64_t)sblock.fs_size * NSPF(&sblock), sblock.fs_ncyl,
2083 	    sblock.fs_ntrak, sblock.fs_nsect);
2084 	(void) fprintf(stderr, gettext(
2085 	    "\t%.1fMB in %d cyl groups (%d c/g, %.2fMB/g, %d i/g)\n"),
2086 	    (float)sblock.fs_size * sblock.fs_fsize / MB, sblock.fs_ncg,
2087 	    sblock.fs_cpg, (float)sblock.fs_fpg * sblock.fs_fsize / MB,
2088 	    sblock.fs_ipg);
2089 
2090 	tmpbuf = calloc(sblock.fs_ncg / 50 + 500, 1);
2091 	if (tmpbuf == NULL) {
2092 		perror("calloc");
2093 		lockexit(32);
2094 	}
2095 	if (cg_too_small) {
2096 		(void) fprintf(stderr, gettext("File system creation failed. "
2097 		    "There is only one cylinder group and\nthat is "
2098 		    "not even big enough to hold the inodes.\n"));
2099 		lockexit(32);
2100 	}
2101 	/*
2102 	 * Now build the cylinders group blocks and
2103 	 * then print out indices of cylinder groups.
2104 	 */
2105 	tprintf(gettext(
2106 	    "super-block backups (for fsck -F ufs -o b=#) at:\n"));
2107 	for (width = cylno = 0; cylno < sblock.fs_ncg && cylno < 10; cylno++) {
2108 		if ((grow == 0) || (cylno >= grow_fs_ncg))
2109 			initcg(cylno);
2110 		num = fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno));
2111 		/*
2112 		 * If Nflag and if the disk is larger than the CHSLIMIT,
2113 		 * then sanity test the superblocks before reporting. If there
2114 		 * are too many superblocks which look insane, we have
2115 		 * to retry with alternate logic. If both methods have
2116 		 * failed, then our efforts to arrive at alternate
2117 		 * superblocks failed, so complain and exit.
2118 		 */
2119 		if (Nflag && retry) {
2120 			skip_this_sb = 0;
2121 			rdfs((diskaddr_t)num, sbsize, (char *)&altsblock);
2122 			ret = checksblock(altsblock, 1);
2123 			if (ret) {
2124 				skip_this_sb = 1;
2125 				invalid_sb_cnt++;
2126 				dprintf(("DeBuG checksblock() failed - error :"
2127 				    " %d for sb : %llu invalid_sb_cnt : %d\n",
2128 				    ret, num, invalid_sb_cnt));
2129 			} else {
2130 				/*
2131 				 * Though the superblock looks sane, verify if
2132 				 * the fs_version in the superblock and the
2133 				 * logic that we are using to arrive at the
2134 				 * superblocks match.
2135 				 */
2136 				if (use_efi_dflts && altsblock.fs_version
2137 				    != UFS_EFISTYLE4NONEFI_VERSION_2) {
2138 					skip_this_sb = 1;
2139 					invalid_sb_cnt++;
2140 				}
2141 			}
2142 			if (invalid_sb_cnt >= INVALIDSBLIMIT) {
2143 				if (retry > 1) {
2144 					(void) fprintf(stderr, gettext(
2145 					    "Error determining alternate "
2146 					    "superblock locations\n"));
2147 					free(tmpbuf);
2148 					lockexit(32);
2149 				}
2150 				retry++;
2151 				use_efi_dflts = !use_efi_dflts;
2152 				free(tmpbuf);
2153 				goto retry_alternate_logic;
2154 			}
2155 			if (skip_this_sb)
2156 				continue;
2157 		}
2158 		(void) sprintf(pbuf, " %llu,", num);
2159 		plen = strlen(pbuf);
2160 		if ((width + plen) > (WIDTH - 1)) {
2161 			width = plen;
2162 			tprintf("\n");
2163 		} else {
2164 			width += plen;
2165 		}
2166 		if (Nflag && retry)
2167 			(void) strncat(tmpbuf, pbuf, strlen(pbuf));
2168 		else
2169 			(void) fprintf(stderr, "%s", pbuf);
2170 	}
2171 	tprintf("\n");
2172 
2173 	remaining_cg = sblock.fs_ncg - cylno;
2174 
2175 	/*
2176 	 * If there are more than 300 cylinder groups still to be
2177 	 * initialized, print a "." for every 50 cylinder groups.
2178 	 */
2179 	if (remaining_cg > 300) {
2180 		tprintf(gettext("Initializing cylinder groups:\n"));
2181 		do_dot = 1;
2182 	}
2183 
2184 	/*
2185 	 * Now initialize all cylinder groups between the first ten
2186 	 * and the last ten.
2187 	 *
2188 	 * If the number of cylinder groups was less than 10, all of the
2189 	 * cylinder group offsets would have printed in the last loop
2190 	 * and cylno will already be equal to sblock.fs_ncg and so this
2191 	 * loop will not be entered.  If there are less than 20 cylinder
2192 	 * groups, cylno is already less than fs_ncg - 10, so this loop
2193 	 * won't be entered in that case either.
2194 	 */
2195 
2196 	i = 0;
2197 	for (; cylno < sblock.fs_ncg - 10; cylno++) {
2198 		if ((grow == 0) || (cylno >= grow_fs_ncg))
2199 			initcg(cylno);
2200 		if (do_dot && cylno % 50 == 0) {
2201 			tprintf(".");
2202 			i++;
2203 			if (i == WIDTH - 1) {
2204 				tprintf("\n");
2205 				i = 0;
2206 			}
2207 		}
2208 	}
2209 
2210 	/*
2211 	 * Now print the cylinder group offsets for the last 10
2212 	 * cylinder groups, if any are left.
2213 	 */
2214 
2215 	if (do_dot) {
2216 		tprintf(gettext(
2217 	    "\nsuper-block backups for last 10 cylinder groups at:\n"));
2218 	}
2219 	for (width = 0; cylno < sblock.fs_ncg; cylno++) {
2220 		if ((grow == 0) || (cylno >= grow_fs_ncg))
2221 			initcg(cylno);
2222 		num = fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno));
2223 		if (Nflag && retry) {
2224 			skip_this_sb = 0;
2225 			rdfs((diskaddr_t)num, sbsize, (char *)&altsblock);
2226 			ret = checksblock(altsblock, 1);
2227 			if (ret) {
2228 				skip_this_sb = 1;
2229 				invalid_sb_cnt++;
2230 				dprintf(("DeBuG checksblock() failed - error :"
2231 				    " %d for sb : %llu invalid_sb_cnt : %d\n",
2232 				    ret, num, invalid_sb_cnt));
2233 			} else {
2234 				/*
2235 				 * Though the superblock looks sane, verify if
2236 				 * the fs_version in the superblock and the
2237 				 * logic that we are using to arrive at the
2238 				 * superblocks match.
2239 				 */
2240 				if (use_efi_dflts && altsblock.fs_version
2241 				    != UFS_EFISTYLE4NONEFI_VERSION_2) {
2242 					skip_this_sb = 1;
2243 					invalid_sb_cnt++;
2244 				}
2245 			}
2246 			if (invalid_sb_cnt >= INVALIDSBLIMIT) {
2247 				if (retry > 1) {
2248 					(void) fprintf(stderr, gettext(
2249 					    "Error determining alternate "
2250 					    "superblock locations\n"));
2251 					free(tmpbuf);
2252 					lockexit(32);
2253 				}
2254 				retry++;
2255 				use_efi_dflts = !use_efi_dflts;
2256 				free(tmpbuf);
2257 				goto retry_alternate_logic;
2258 			}
2259 			if (skip_this_sb)
2260 				continue;
2261 		}
2262 		/* Don't print ',' for the last superblock */
2263 		if (cylno == sblock.fs_ncg-1)
2264 			(void) sprintf(pbuf, " %llu", num);
2265 		else
2266 			(void) sprintf(pbuf, " %llu,", num);
2267 		plen = strlen(pbuf);
2268 		if ((width + plen) > (WIDTH - 1)) {
2269 			width = plen;
2270 			tprintf("\n");
2271 		} else {
2272 			width += plen;
2273 		}
2274 		if (Nflag && retry)
2275 			(void) strncat(tmpbuf, pbuf, strlen(pbuf));
2276 		else
2277 			(void) fprintf(stderr, "%s", pbuf);
2278 	}
2279 	tprintf("\n");
2280 	if (Nflag) {
2281 		if (retry)
2282 			(void) fprintf(stderr, "%s", tmpbuf);
2283 		free(tmpbuf);
2284 		lockexit(0);
2285 	}
2286 
2287 	free(tmpbuf);
2288 	if (grow)
2289 		goto grow50;
2290 
2291 	/*
2292 	 * Now construct the initial file system,
2293 	 * then write out the super-block.
2294 	 */
2295 	fsinit();
2296 grow50:
2297 	/*
2298 	 * write the superblock and csum information
2299 	 */
2300 	wtsb();
2301 
2302 	/*
2303 	 * extend the last cylinder group in the original file system
2304 	 */
2305 	if (grow) {
2306 		extendcg(grow_fs_ncg-1);
2307 		wtsb();
2308 	}
2309 
2310 	/*
2311 	 * Write out the duplicate super blocks to the first 10
2312 	 * cylinder groups (or fewer, if there are fewer than 10
2313 	 * cylinder groups).
2314 	 */
2315 	for (cylno = 0; cylno < sblock.fs_ncg && cylno < 10; cylno++)
2316 		awtfs(fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno)),
2317 		    (int)sbsize, (char *)&sblock, SAVE);
2318 
2319 	/*
2320 	 * Now write out duplicate super blocks to the remaining
2321 	 * cylinder groups.  In the case of multi-terabyte file
2322 	 * systems, just write out the super block to the last ten
2323 	 * cylinder groups (or however many are left).
2324 	 */
2325 	if (mtb == 'y') {
2326 		if (sblock.fs_ncg <= 10)
2327 			cylno = sblock.fs_ncg;
2328 		else if (sblock.fs_ncg <= 20)
2329 			cylno = 10;
2330 		else
2331 			cylno = sblock.fs_ncg - 10;
2332 	}
2333 
2334 	for (; cylno < sblock.fs_ncg; cylno++)
2335 		awtfs(fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno)),
2336 		    (int)sbsize, (char *)&sblock, SAVE);
2337 
2338 	/*
2339 	 * Flush out all the AIO writes we've done.  It's not
2340 	 * necessary to do this explicitly, but it's the only
2341 	 * way to report any errors from those writes.
2342 	 */
2343 	flush_writes();
2344 
2345 	/*
2346 	 * set clean flag
2347 	 */
2348 	if (grow)
2349 		sblock.fs_clean = grow_fs_clean;
2350 	else
2351 		sblock.fs_clean = FSCLEAN;
2352 	sblock.fs_time = mkfstime;
2353 	sblock.fs_state = FSOKAY - sblock.fs_time;
2354 	wtfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
2355 	isbad = 0;
2356 
2357 	if (fsync(fso) == -1) {
2358 		saverr = errno;
2359 		(void) fprintf(stderr,
2360 		    gettext("mkfs: fsync failed on write disk: %s\n"),
2361 		    strerror(saverr));
2362 		/* we're just cleaning up, so keep going */
2363 	}
2364 	if (close(fsi) == -1) {
2365 		saverr = errno;
2366 		(void) fprintf(stderr,
2367 		    gettext("mkfs: close failed on read disk: %s\n"),
2368 		    strerror(saverr));
2369 		/* we're just cleaning up, so keep going */
2370 	}
2371 	if (close(fso) == -1) {
2372 		saverr = errno;
2373 		(void) fprintf(stderr,
2374 		    gettext("mkfs: close failed on write disk: %s\n"),
2375 		    strerror(saverr));
2376 		/* we're just cleaning up, so keep going */
2377 	}
2378 	fsi = fso = -1;
2379 
2380 #ifndef STANDALONE
2381 	lockexit(0);
2382 #endif
2383 
2384 	return (0);
2385 }
2386 
2387 /*
2388  * Figure out how big the partition we're dealing with is.
2389  * The value returned is in disk blocks (sectors);
2390  */
2391 static diskaddr_t
2392 get_max_size(int fd)
2393 {
2394 	struct extvtoc vtoc;
2395 	dk_gpt_t *efi_vtoc;
2396 	diskaddr_t	slicesize;
2397 
2398 	int index = read_extvtoc(fd, &vtoc);
2399 
2400 	if (index >= 0) {
2401 		label_type = LABEL_TYPE_VTOC;
2402 	} else {
2403 		if (index == VT_ENOTSUP || index == VT_ERROR) {
2404 			/* it might be an EFI label */
2405 			index = efi_alloc_and_read(fd, &efi_vtoc);
2406 			label_type = LABEL_TYPE_EFI;
2407 		}
2408 	}
2409 
2410 	if (index < 0) {
2411 		switch (index) {
2412 		case VT_ERROR:
2413 			break;
2414 		case VT_EIO:
2415 			errno = EIO;
2416 			break;
2417 		case VT_EINVAL:
2418 			errno = EINVAL;
2419 		}
2420 		perror(gettext("Can not determine partition size"));
2421 		lockexit(32);
2422 	}
2423 
2424 	if (label_type == LABEL_TYPE_EFI) {
2425 		slicesize = efi_vtoc->efi_parts[index].p_size;
2426 		efi_free(efi_vtoc);
2427 	} else {
2428 		/*
2429 		 * In the vtoc struct, p_size is a 32-bit signed quantity.
2430 		 * In the dk_gpt struct (efi's version of the vtoc), p_size
2431 		 * is an unsigned 64-bit quantity.  By casting the vtoc's
2432 		 * psize to an unsigned 32-bit quantity, it will be copied
2433 		 * to 'slicesize' (an unsigned 64-bit diskaddr_t) without
2434 		 * sign extension.
2435 		 */
2436 
2437 		slicesize = (uint32_t)vtoc.v_part[index].p_size;
2438 	}
2439 
2440 	dprintf(("DeBuG get_max_size index = %d, p_size = %lld, dolimit = %d\n",
2441 	    index, slicesize, (slicesize > FS_MAX)));
2442 
2443 	/*
2444 	 * The next line limits a UFS file system to the maximum
2445 	 * supported size.
2446 	 */
2447 
2448 	if (slicesize > FS_MAX)
2449 		return (FS_MAX);
2450 	return (slicesize);
2451 }
2452 
2453 static long
2454 get_max_track_size(int fd)
2455 {
2456 	struct dk_cinfo ci;
2457 	long track_size = -1;
2458 
2459 	if (ioctl(fd, DKIOCINFO, &ci) == 0) {
2460 		track_size = ci.dki_maxtransfer * DEV_BSIZE;
2461 	}
2462 
2463 	if ((track_size < 0)) {
2464 		int	error = 0;
2465 		int	maxphys;
2466 		int	gotit = 0;
2467 
2468 		gotit = fsgetmaxphys(&maxphys, &error);
2469 		if (gotit) {
2470 			track_size = MIN(MB, maxphys);
2471 		} else {
2472 			(void) fprintf(stderr, gettext(
2473 "Warning: Could not get system value for maxphys. The value for\n"
2474 "maxcontig will default to 1MB.\n"));
2475 			track_size = MB;
2476 		}
2477 	}
2478 	return (track_size);
2479 }
2480 
2481 /*
2482  * Initialize a cylinder group.
2483  */
2484 static void
2485 initcg(int cylno)
2486 {
2487 	diskaddr_t cbase, d;
2488 	diskaddr_t dlower;	/* last data block before cg metadata */
2489 	diskaddr_t dupper;	/* first data block after cg metadata */
2490 	diskaddr_t dmax;
2491 	int64_t i;
2492 	struct csum *cs;
2493 	struct dinode *inode_buffer;
2494 	int size;
2495 
2496 	/*
2497 	 * Variables used to store intermediate results as a part of
2498 	 * the internal implementation of the cbtocylno() macros.
2499 	 */
2500 	diskaddr_t bno;		/* UFS block number (not sector number) */
2501 	int	cbcylno;	/* current cylinder number */
2502 	int	cbcylno_sect;	/* sector offset within cylinder */
2503 	int	cbsect_incr;	/* amount to increment sector offset */
2504 
2505 	/*
2506 	 * Variables used to store intermediate results as a part of
2507 	 * the internal implementation of the cbtorpos() macros.
2508 	 */
2509 	short	*cgblks;	/* pointer to array of free blocks in cg */
2510 	int	trackrpos;	/* tmp variable for rotation position */
2511 	int	trackoff;	/* offset within a track */
2512 	int	trackoff_incr;	/* amount to increment trackoff */
2513 	int	rpos;		/* rotation position of current block */
2514 	int	rpos_incr;	/* amount to increment rpos per block */
2515 
2516 	union cgun *icgun;	/* local pointer to a cg summary block */
2517 #define	icg	(icgun->cg)
2518 
2519 	icgun = (union cgun *)getbuf(&cgsumbuf, sizeof (union cgun));
2520 
2521 	/*
2522 	 * Determine block bounds for cylinder group.
2523 	 * Allow space for super block summary information in first
2524 	 * cylinder group.
2525 	 */
2526 	cbase = cgbase(&sblock, cylno);
2527 	dmax = cbase + sblock.fs_fpg;
2528 	if (dmax > sblock.fs_size)	/* last cg may be smaller than normal */
2529 		dmax = sblock.fs_size;
2530 	dlower = cgsblock(&sblock, cylno) - cbase;
2531 	dupper = cgdmin(&sblock, cylno) - cbase;
2532 	if (cylno == 0)
2533 		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
2534 	cs = fscs + cylno;
2535 	icg.cg_time = mkfstime;
2536 	icg.cg_magic = CG_MAGIC;
2537 	icg.cg_cgx = cylno;
2538 	/* last one gets whatever's left */
2539 	if (cylno == sblock.fs_ncg - 1)
2540 		icg.cg_ncyl = sblock.fs_ncyl - (sblock.fs_cpg * cylno);
2541 	else
2542 		icg.cg_ncyl = sblock.fs_cpg;
2543 	icg.cg_niblk = sblock.fs_ipg;
2544 	icg.cg_ndblk = dmax - cbase;
2545 	icg.cg_cs.cs_ndir = 0;
2546 	icg.cg_cs.cs_nffree = 0;
2547 	icg.cg_cs.cs_nbfree = 0;
2548 	icg.cg_cs.cs_nifree = 0;
2549 	icg.cg_rotor = 0;
2550 	icg.cg_frotor = 0;
2551 	icg.cg_irotor = 0;
2552 	icg.cg_btotoff = &icg.cg_space[0] - (uchar_t *)(&icg.cg_link);
2553 	icg.cg_boff = icg.cg_btotoff + sblock.fs_cpg * sizeof (long);
2554 	icg.cg_iusedoff = icg.cg_boff +
2555 	    sblock.fs_cpg * sblock.fs_nrpos * sizeof (short);
2556 	icg.cg_freeoff = icg.cg_iusedoff + howmany(sblock.fs_ipg, NBBY);
2557 	icg.cg_nextfreeoff = icg.cg_freeoff +
2558 	    howmany(sblock.fs_cpg * sblock.fs_spc / NSPF(&sblock), NBBY);
2559 	for (i = 0; i < sblock.fs_frag; i++) {
2560 		icg.cg_frsum[i] = 0;
2561 	}
2562 	bzero((caddr_t)cg_inosused(&icg), icg.cg_freeoff - icg.cg_iusedoff);
2563 	icg.cg_cs.cs_nifree += sblock.fs_ipg;
2564 	if (cylno == 0)
2565 		for (i = 0; i < UFSROOTINO; i++) {
2566 			setbit(cg_inosused(&icg), i);
2567 			icg.cg_cs.cs_nifree--;
2568 		}
2569 
2570 	/*
2571 	 * Initialize all the inodes in the cylinder group using
2572 	 * random numbers.
2573 	 */
2574 	size = sblock.fs_ipg * sizeof (struct dinode);
2575 	inode_buffer = (struct dinode *)getbuf(&inodebuf, size);
2576 
2577 	for (i = 0; i < sblock.fs_ipg; i++) {
2578 		IRANDOMIZE(&(inode_buffer[i].di_ic));
2579 	}
2580 
2581 	/*
2582 	 * Write all inodes in a single write for performance.
2583 	 */
2584 	awtfs(fsbtodb(&sblock, (uint64_t)cgimin(&sblock, cylno)), (int)size,
2585 	    (char *)inode_buffer, RELEASE);
2586 
2587 	bzero((caddr_t)cg_blktot(&icg), icg.cg_boff - icg.cg_btotoff);
2588 	bzero((caddr_t)cg_blks(&sblock, &icg, 0),
2589 	    icg.cg_iusedoff - icg.cg_boff);
2590 	bzero((caddr_t)cg_blksfree(&icg), icg.cg_nextfreeoff - icg.cg_freeoff);
2591 
2592 	if (cylno > 0) {
2593 		for (d = 0; d < dlower; d += sblock.fs_frag) {
2594 			setblock(&sblock, cg_blksfree(&icg), d/sblock.fs_frag);
2595 			icg.cg_cs.cs_nbfree++;
2596 			cg_blktot(&icg)[cbtocylno(&sblock, d)]++;
2597 			cg_blks(&sblock, &icg, cbtocylno(&sblock, d))
2598 			    [cbtorpos(&sblock, d)]++;
2599 		}
2600 		sblock.fs_dsize += dlower;
2601 	}
2602 	sblock.fs_dsize += icg.cg_ndblk - dupper;
2603 	if ((i = dupper % sblock.fs_frag) != 0) {
2604 		icg.cg_frsum[sblock.fs_frag - i]++;
2605 		for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
2606 			setbit(cg_blksfree(&icg), dupper);
2607 			icg.cg_cs.cs_nffree++;
2608 		}
2609 	}
2610 
2611 	/*
2612 	 * WARNING: The following code is somewhat confusing, but
2613 	 * results in a substantial performance improvement in mkfs.
2614 	 *
2615 	 * Instead of using cbtocylno() and cbtorpos() macros, we
2616 	 * keep track of all the intermediate state of those macros
2617 	 * in some variables.  This allows simple addition to be
2618 	 * done to calculate the results as we step through the
2619 	 * blocks in an orderly fashion instead of the slower
2620 	 * multiplication and division the macros are forced to
2621 	 * used so they can support random input.  (Multiplication,
2622 	 * division, and remainder operations typically take about
2623 	 * 10x as many processor cycles as other operations.)
2624 	 *
2625 	 * The basic idea is to take code:
2626 	 *
2627 	 *	for (x = starting_x; x < max; x++)
2628 	 *		y = (x * c) / z
2629 	 *
2630 	 * and rewrite it to take advantage of the fact that
2631 	 * the variable x is incrementing in an orderly way:
2632 	 *
2633 	 *	intermediate = starting_x * c
2634 	 *	yval = intermediate / z
2635 	 *	for (x = starting_x; x < max; x++) {
2636 	 *		y = yval;
2637 	 *		intermediate += c
2638 	 *		if (intermediate > z) {
2639 	 *			yval++;
2640 	 *			intermediate -= z
2641 	 *		}
2642 	 *	}
2643 	 *
2644 	 * Performance has improved as much as 4X using this code.
2645 	 */
2646 
2647 	/*
2648 	 * Initialize the starting points for all the cbtocylno()
2649 	 * macro variables and figure out the increments needed each
2650 	 * time through the loop.
2651 	 */
2652 	cbcylno_sect = dupper * NSPF(&sblock);
2653 	cbsect_incr = sblock.fs_frag * NSPF(&sblock);
2654 	cbcylno = cbcylno_sect / sblock.fs_spc;
2655 	cbcylno_sect %= sblock.fs_spc;
2656 	cgblks = cg_blks(&sblock, &icg, cbcylno);
2657 	bno = dupper / sblock.fs_frag;
2658 
2659 	/*
2660 	 * Initialize the starting points for all the cbtorpos()
2661 	 * macro variables and figure out the increments needed each
2662 	 * time through the loop.
2663 	 *
2664 	 * It's harder to simplify the cbtorpos() macro if there were
2665 	 * alternate sectors specified (or if they previously existed
2666 	 * in the growfs case).  Since this is rare, we just revert to
2667 	 * using the macros in this case and skip the variable setup.
2668 	 */
2669 	if (!spc_flag) {
2670 		trackrpos = (cbcylno_sect % sblock.fs_nsect) * sblock.fs_nrpos;
2671 		rpos = trackrpos / sblock.fs_nsect;
2672 		trackoff = trackrpos % sblock.fs_nsect;
2673 		trackoff_incr = cbsect_incr * sblock.fs_nrpos;
2674 		rpos_incr = (trackoff_incr / sblock.fs_nsect) % sblock.fs_nrpos;
2675 		trackoff_incr = trackoff_incr % sblock.fs_nsect;
2676 	}
2677 
2678 	/*
2679 	 * Loop through all the blocks, marking them free and
2680 	 * updating totals kept in the superblock and cg summary.
2681 	 */
2682 	for (d = dupper; d + sblock.fs_frag <= dmax - cbase; ) {
2683 		setblock(&sblock, cg_blksfree(&icg),  bno);
2684 		icg.cg_cs.cs_nbfree++;
2685 
2686 		cg_blktot(&icg)[cbcylno]++;
2687 
2688 		if (!spc_flag)
2689 			cgblks[rpos]++;
2690 		else
2691 			cg_blks(&sblock, &icg, cbtocylno(&sblock, d))
2692 			    [cbtorpos(&sblock, d)]++;
2693 
2694 		d += sblock.fs_frag;
2695 		bno++;
2696 
2697 		/*
2698 		 * Increment the sector offset within the cylinder
2699 		 * for the cbtocylno() macro reimplementation.  If
2700 		 * we're beyond the end of the cylinder, update the
2701 		 * cylinder number, calculate the offset in the
2702 		 * new cylinder, and update the cgblks pointer
2703 		 * to the next rotational position.
2704 		 */
2705 		cbcylno_sect += cbsect_incr;
2706 		if (cbcylno_sect >= sblock.fs_spc) {
2707 			cbcylno++;
2708 			cbcylno_sect -= sblock.fs_spc;
2709 			cgblks += sblock.fs_nrpos;
2710 		}
2711 
2712 		/*
2713 		 * If there aren't alternate sectors, increment the
2714 		 * rotational position variables for the cbtorpos()
2715 		 * reimplementation.  Note that we potentially
2716 		 * increment rpos twice.  Once by rpos_incr, and one
2717 		 * more time when we wrap to a new track because
2718 		 * trackoff >= fs_nsect.
2719 		 */
2720 		if (!spc_flag) {
2721 			trackoff += trackoff_incr;
2722 			rpos += rpos_incr;
2723 			if (trackoff >= sblock.fs_nsect) {
2724 				trackoff -= sblock.fs_nsect;
2725 				rpos++;
2726 			}
2727 			if (rpos >= sblock.fs_nrpos)
2728 				rpos -= sblock.fs_nrpos;
2729 		}
2730 	}
2731 
2732 	if (d < dmax - cbase) {
2733 		icg.cg_frsum[dmax - cbase - d]++;
2734 		for (; d < dmax - cbase; d++) {
2735 			setbit(cg_blksfree(&icg), d);
2736 			icg.cg_cs.cs_nffree++;
2737 		}
2738 	}
2739 	sblock.fs_cstotal.cs_ndir += icg.cg_cs.cs_ndir;
2740 	sblock.fs_cstotal.cs_nffree += icg.cg_cs.cs_nffree;
2741 	sblock.fs_cstotal.cs_nbfree += icg.cg_cs.cs_nbfree;
2742 	sblock.fs_cstotal.cs_nifree += icg.cg_cs.cs_nifree;
2743 	*cs = icg.cg_cs;
2744 	awtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, cylno)),
2745 	    sblock.fs_bsize, (char *)&icg, RELEASE);
2746 }
2747 
2748 /*
2749  * initialize the file system
2750  */
2751 struct inode node;
2752 
2753 #define	LOSTDIR
2754 #ifdef LOSTDIR
2755 #define	PREDEFDIR 3
2756 #else
2757 #define	PREDEFDIR 2
2758 #endif
2759 
2760 struct direct root_dir[] = {
2761 	{ UFSROOTINO, sizeof (struct direct), 1, "." },
2762 	{ UFSROOTINO, sizeof (struct direct), 2, ".." },
2763 #ifdef LOSTDIR
2764 	{ LOSTFOUNDINO, sizeof (struct direct), 10, "lost+found" },
2765 #endif
2766 };
2767 #ifdef LOSTDIR
2768 struct direct lost_found_dir[] = {
2769 	{ LOSTFOUNDINO, sizeof (struct direct), 1, "." },
2770 	{ UFSROOTINO, sizeof (struct direct), 2, ".." },
2771 	{ 0, DIRBLKSIZ, 0, 0 },
2772 };
2773 #endif
2774 char buf[MAXBSIZE];
2775 
2776 static void
2777 fsinit()
2778 {
2779 	int i;
2780 
2781 
2782 	/*
2783 	 * initialize the node
2784 	 */
2785 	node.i_atime = mkfstime;
2786 	node.i_mtime = mkfstime;
2787 	node.i_ctime = mkfstime;
2788 #ifdef LOSTDIR
2789 	/*
2790 	 * create the lost+found directory
2791 	 */
2792 	(void) makedir(lost_found_dir, 2);
2793 	for (i = DIRBLKSIZ; i < sblock.fs_bsize; i += DIRBLKSIZ) {
2794 		bcopy(&lost_found_dir[2], &buf[i], DIRSIZ(&lost_found_dir[2]));
2795 	}
2796 	node.i_number = LOSTFOUNDINO;
2797 	node.i_smode = node.i_mode = IFDIR | 0700;
2798 	node.i_nlink = 2;
2799 	node.i_size = sblock.fs_bsize;
2800 	node.i_db[0] = alloc((int)node.i_size, node.i_mode);
2801 	node.i_blocks = btodb(fragroundup(&sblock, (int)node.i_size));
2802 	IRANDOMIZE(&node.i_ic);
2803 	wtfs(fsbtodb(&sblock, (uint64_t)node.i_db[0]), (int)node.i_size, buf);
2804 	iput(&node);
2805 #endif
2806 	/*
2807 	 * create the root directory
2808 	 */
2809 	node.i_number = UFSROOTINO;
2810 	node.i_mode = node.i_smode = IFDIR | UMASK;
2811 	node.i_nlink = PREDEFDIR;
2812 	node.i_size = makedir(root_dir, PREDEFDIR);
2813 	node.i_db[0] = alloc(sblock.fs_fsize, node.i_mode);
2814 	/* i_size < 2GB because we are initializing the file system */
2815 	node.i_blocks = btodb(fragroundup(&sblock, (int)node.i_size));
2816 	IRANDOMIZE(&node.i_ic);
2817 	wtfs(fsbtodb(&sblock, (uint64_t)node.i_db[0]), sblock.fs_fsize, buf);
2818 	iput(&node);
2819 }
2820 
2821 /*
2822  * construct a set of directory entries in "buf".
2823  * return size of directory.
2824  */
2825 static int
2826 makedir(struct direct *protodir, int entries)
2827 {
2828 	char *cp;
2829 	int i;
2830 	ushort_t spcleft;
2831 
2832 	spcleft = DIRBLKSIZ;
2833 	for (cp = buf, i = 0; i < entries - 1; i++) {
2834 		protodir[i].d_reclen = DIRSIZ(&protodir[i]);
2835 		bcopy(&protodir[i], cp, protodir[i].d_reclen);
2836 		cp += protodir[i].d_reclen;
2837 		spcleft -= protodir[i].d_reclen;
2838 	}
2839 	protodir[i].d_reclen = spcleft;
2840 	bcopy(&protodir[i], cp, DIRSIZ(&protodir[i]));
2841 	return (DIRBLKSIZ);
2842 }
2843 
2844 /*
2845  * allocate a block or frag
2846  */
2847 static daddr32_t
2848 alloc(int size, int mode)
2849 {
2850 	int i, frag;
2851 	daddr32_t d;
2852 
2853 	rdfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2854 	    (char *)&acg);
2855 	if (acg.cg_magic != CG_MAGIC) {
2856 		(void) fprintf(stderr, gettext("cg 0: bad magic number\n"));
2857 		lockexit(32);
2858 	}
2859 	if (acg.cg_cs.cs_nbfree == 0) {
2860 		(void) fprintf(stderr,
2861 		    gettext("first cylinder group ran out of space\n"));
2862 		lockexit(32);
2863 	}
2864 	for (d = 0; d < acg.cg_ndblk; d += sblock.fs_frag)
2865 		if (isblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag))
2866 			goto goth;
2867 	(void) fprintf(stderr,
2868 	    gettext("internal error: can't find block in cyl 0\n"));
2869 	lockexit(32);
2870 goth:
2871 	clrblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag);
2872 	acg.cg_cs.cs_nbfree--;
2873 	sblock.fs_cstotal.cs_nbfree--;
2874 	fscs[0].cs_nbfree--;
2875 	if (mode & IFDIR) {
2876 		acg.cg_cs.cs_ndir++;
2877 		sblock.fs_cstotal.cs_ndir++;
2878 		fscs[0].cs_ndir++;
2879 	}
2880 	cg_blktot(&acg)[cbtocylno(&sblock, d)]--;
2881 	cg_blks(&sblock, &acg, cbtocylno(&sblock, d))[cbtorpos(&sblock, d)]--;
2882 	if (size != sblock.fs_bsize) {
2883 		frag = howmany(size, sblock.fs_fsize);
2884 		fscs[0].cs_nffree += sblock.fs_frag - frag;
2885 		sblock.fs_cstotal.cs_nffree += sblock.fs_frag - frag;
2886 		acg.cg_cs.cs_nffree += sblock.fs_frag - frag;
2887 		acg.cg_frsum[sblock.fs_frag - frag]++;
2888 		for (i = frag; i < sblock.fs_frag; i++)
2889 			setbit(cg_blksfree(&acg), d + i);
2890 	}
2891 	wtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2892 	    (char *)&acg);
2893 	return (d);
2894 }
2895 
2896 /*
2897  * Allocate an inode on the disk
2898  */
2899 static void
2900 iput(struct inode *ip)
2901 {
2902 	struct dinode buf[MAXINOPB];
2903 	diskaddr_t d;
2904 
2905 	rdfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2906 	    (char *)&acg);
2907 	if (acg.cg_magic != CG_MAGIC) {
2908 		(void) fprintf(stderr, gettext("cg 0: bad magic number\n"));
2909 		lockexit(32);
2910 	}
2911 	acg.cg_cs.cs_nifree--;
2912 	setbit(cg_inosused(&acg), ip->i_number);
2913 	wtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2914 	    (char *)&acg);
2915 	sblock.fs_cstotal.cs_nifree--;
2916 	fscs[0].cs_nifree--;
2917 	if ((int)ip->i_number >= sblock.fs_ipg * sblock.fs_ncg) {
2918 		(void) fprintf(stderr,
2919 		    gettext("fsinit: inode value out of range (%d).\n"),
2920 		    ip->i_number);
2921 		lockexit(32);
2922 	}
2923 	d = fsbtodb(&sblock, (uint64_t)itod(&sblock, (int)ip->i_number));
2924 	rdfs(d, sblock.fs_bsize, (char *)buf);
2925 	buf[itoo(&sblock, (int)ip->i_number)].di_ic = ip->i_ic;
2926 	wtfs(d, sblock.fs_bsize, (char *)buf);
2927 }
2928 
2929 /*
2930  * getbuf()	-- Get a buffer for use in an AIO operation.  Buffer
2931  *		is zero'd the first time returned, left with whatever
2932  *		was in memory after that.  This function actually gets
2933  *		enough memory the first time it's called to support
2934  *		MAXBUF buffers like a slab allocator.  When all the
2935  *		buffers are in use, it waits for an aio to complete
2936  *		and make a buffer available.
2937  *
2938  *		Never returns an error.  Either succeeds or exits.
2939  */
2940 static char *
2941 getbuf(bufhdr *bufhead, int size)
2942 {
2943 	bufhdr *pbuf;
2944 	bufhdr *prev;
2945 	int i;
2946 	int buf_size, max_bufs;
2947 
2948 	/*
2949 	 * Initialize all the buffers
2950 	 */
2951 	if (bufhead->head == NULL) {
2952 		/*
2953 		 * round up the size of our buffer header to a
2954 		 * 16 byte boundary so the address we return to
2955 		 * the caller is "suitably aligned".
2956 		 */
2957 		bufhdrsize = (sizeof (bufhdr) + 15) & ~15;
2958 
2959 		/*
2960 		 * Add in our header to the buffer and round it all up to
2961 		 * a 16 byte boundry so each member of the slab is aligned.
2962 		 */
2963 		buf_size = (size + bufhdrsize + 15) & ~15;
2964 
2965 		/*
2966 		 * Limit number of buffers to lesser of MAXBUFMEM's worth
2967 		 * or MAXBUF, whichever is less.
2968 		 */
2969 		max_bufs = MAXBUFMEM / buf_size;
2970 		if (max_bufs > MAXBUF)
2971 			max_bufs = MAXBUF;
2972 
2973 		pbuf = (bufhdr *)calloc(max_bufs, buf_size);
2974 		if (pbuf == NULL) {
2975 			perror("calloc");
2976 			lockexit(32);
2977 		}
2978 
2979 		bufhead->head = bufhead;
2980 		prev = bufhead;
2981 		for (i = 0; i < max_bufs; i++) {
2982 			pbuf->head = bufhead;
2983 			prev->next = pbuf;
2984 			prev = pbuf;
2985 			pbuf = (bufhdr *)((char *)pbuf + buf_size);
2986 		}
2987 	}
2988 
2989 	/*
2990 	 * Get an available buffer, waiting for I/O if necessary
2991 	 */
2992 	wait_for_write(NOBLOCK);
2993 	while (bufhead->next == NULL)
2994 		wait_for_write(BLOCK);
2995 
2996 	/*
2997 	 * Take the buffer off the list
2998 	 */
2999 	pbuf = bufhead->next;
3000 	bufhead->next = pbuf->next;
3001 	pbuf->next = NULL;
3002 
3003 	/*
3004 	 * return the empty buffer space just past the header
3005 	 */
3006 	return ((char *)pbuf + bufhdrsize);
3007 }
3008 
3009 /*
3010  * freebuf()	-- Free a buffer gotten previously through getbuf.
3011  *		Puts the buffer back on the appropriate list for
3012  *		later use.  Never calls free().
3013  *
3014  * Assumes that SIGINT is blocked.
3015  */
3016 static void
3017 freebuf(char *buf)
3018 {
3019 	bufhdr *pbuf;
3020 	bufhdr *bufhead;
3021 
3022 	/*
3023 	 * get the header for this buffer
3024 	 */
3025 	pbuf = (bufhdr *)(buf - bufhdrsize);
3026 
3027 	/*
3028 	 * Put it back on the list of available buffers
3029 	 */
3030 	bufhead = pbuf->head;
3031 	pbuf->next = bufhead->next;
3032 	bufhead->next = pbuf;
3033 }
3034 
3035 /*
3036  * freetrans()	-- Free a transaction gotten previously through getaiop.
3037  *		Puts the transaction struct back on the appropriate list for
3038  *		later use.  Never calls free().
3039  *
3040  * Assumes that SIGINT is blocked.
3041  */
3042 static void
3043 freetrans(aio_trans *transp)
3044 {
3045 	/*
3046 	 * free the buffer associated with this AIO if needed
3047 	 */
3048 	if (transp->release == RELEASE)
3049 		freebuf(transp->buffer);
3050 
3051 	/*
3052 	 * Put transaction on the free list
3053 	 */
3054 	transp->next = results.trans;
3055 	results.trans = transp;
3056 }
3057 
3058 /*
3059  * wait_for_write()	-- Wait for an aio write to complete.  Return
3060  *			the transaction structure for that write.
3061  *
3062  * Blocks SIGINT if necessary.
3063  */
3064 aio_trans *
3065 wait_for_write(int block)
3066 {
3067 	aio_trans	*transp;
3068 	aio_result_t	*resultp;
3069 	static struct timeval  zero_wait = { 0, 0 };
3070 	sigset_t	old_mask;
3071 
3072 	/*
3073 	 * If we know there aren't any outstanding transactions, just return
3074 	 */
3075 	if (results.outstanding == 0)
3076 		return ((aio_trans *) 0);
3077 
3078 	block_sigint(&old_mask);
3079 
3080 	resultp = aiowait(block ? NULL : &zero_wait);
3081 	if (resultp == NULL ||
3082 	    (resultp == (aio_result_t *)-1 && errno == EINVAL)) {
3083 		unblock_sigint(&old_mask);
3084 		return ((aio_trans *) 0);
3085 	}
3086 
3087 	results.outstanding--;
3088 	transp = (aio_trans *)resultp;
3089 
3090 	if (resultp->aio_return != transp->size) {
3091 		if (resultp->aio_return == -1) {
3092 			/*
3093 			 * The aiowrite() may have failed because the
3094 			 * kernel didn't have enough memory to do the job.
3095 			 * Flush all pending writes and try a normal
3096 			 * write().  wtfs_breakup() will call exit if it
3097 			 * fails, so we don't worry about errors here.
3098 			 */
3099 			flush_writes();
3100 			wtfs_breakup(transp->bno, transp->size, transp->buffer);
3101 		} else {
3102 			(void) fprintf(stderr, gettext(
3103 			    "short write (%d of %d bytes) on sector %lld\n"),
3104 			    resultp->aio_return, transp->size,
3105 			    transp->bno);
3106 			/*
3107 			 * Don't unblock SIGINT, to avoid potential
3108 			 * looping due to queued interrupts and
3109 			 * error handling.
3110 			 */
3111 			lockexit(32);
3112 		}
3113 	}
3114 
3115 	resultp->aio_return = 0;
3116 	freetrans(transp);
3117 	unblock_sigint(&old_mask);
3118 	return (transp);
3119 }
3120 
3121 /*
3122  * flush_writes()	-- flush all the outstanding aio writes.
3123  */
3124 static void
3125 flush_writes(void)
3126 {
3127 	while (wait_for_write(BLOCK))
3128 		;
3129 }
3130 
3131 /*
3132  * get_aiop()	-- find and return an aio_trans structure on which a new
3133  *		aio can be done.  Blocks on aiowait() if needed.  Reaps
3134  *		all outstanding completed aio's.
3135  *
3136  * Assumes that SIGINT is blocked.
3137  */
3138 aio_trans *
3139 get_aiop()
3140 {
3141 	int i;
3142 	aio_trans *transp;
3143 	aio_trans *prev;
3144 
3145 	/*
3146 	 * initialize aio stuff
3147 	 */
3148 	if (!aio_inited) {
3149 		aio_inited = 1;
3150 
3151 		results.maxpend = 0;
3152 		results.outstanding = 0;
3153 		results.max = MAXAIO;
3154 
3155 		results.trans = (aio_trans *)calloc(results.max,
3156 		    sizeof (aio_trans));
3157 		if (results.trans == NULL) {
3158 			perror("calloc");
3159 			lockexit(32);
3160 		}
3161 
3162 		/*
3163 		 * Initialize the linked list of aio transaction
3164 		 * structures.  Note that the final "next" pointer
3165 		 * will be NULL since we got the buffer from calloc().
3166 		 */
3167 		prev = results.trans;
3168 		for (i = 1; i < results.max; i++) {
3169 			prev->next = &(results.trans[i]);
3170 			prev = prev->next;
3171 		}
3172 	}
3173 
3174 	wait_for_write(NOBLOCK);
3175 	while (results.trans == NULL)
3176 		wait_for_write(BLOCK);
3177 	transp = results.trans;
3178 	results.trans = results.trans->next;
3179 
3180 	transp->next = 0;
3181 	transp->resultbuf.aio_return = AIO_INPROGRESS;
3182 	return (transp);
3183 }
3184 
3185 /*
3186  * read a block from the file system
3187  */
3188 static void
3189 rdfs(diskaddr_t bno, int size, char *bf)
3190 {
3191 	int n, saverr;
3192 
3193 	/*
3194 	 * In case we need any data that's pending in an aiowrite(),
3195 	 * we wait for them all to complete before doing a read.
3196 	 */
3197 	flush_writes();
3198 
3199 	/*
3200 	 * Note: the llseek() can succeed, even if the offset is out of range.
3201 	 * It's not until the file i/o operation (the read()) that one knows
3202 	 * for sure if the raw device can handle the offset.
3203 	 */
3204 	if (llseek(fsi, (offset_t)bno * sectorsize, 0) < 0) {
3205 		saverr = errno;
3206 		(void) fprintf(stderr,
3207 		    gettext("seek error on sector %lld: %s\n"),
3208 		    bno, strerror(saverr));
3209 		lockexit(32);
3210 	}
3211 	n = read(fsi, bf, size);
3212 	if (n != size) {
3213 		saverr = errno;
3214 		if (n == -1)
3215 			(void) fprintf(stderr,
3216 			    gettext("read error on sector %lld: %s\n"),
3217 			    bno, strerror(saverr));
3218 		else
3219 			(void) fprintf(stderr, gettext(
3220 			    "short read (%d of %d bytes) on sector %lld\n"),
3221 			    n, size, bno);
3222 		lockexit(32);
3223 	}
3224 }
3225 
3226 /*
3227  * write a block to the file system
3228  */
3229 static void
3230 wtfs(diskaddr_t bno, int size, char *bf)
3231 {
3232 	int n, saverr;
3233 
3234 	if (fso == -1)
3235 		return;
3236 
3237 	/*
3238 	 * Note: the llseek() can succeed, even if the offset is out of range.
3239 	 * It's not until the file i/o operation (the write()) that one knows
3240 	 * for sure if the raw device can handle the offset.
3241 	 */
3242 	if (llseek(fso, (offset_t)bno * sectorsize, 0) < 0) {
3243 		saverr = errno;
3244 		(void) fprintf(stderr,
3245 		    gettext("seek error on sector %lld: %s\n"),
3246 		    bno, strerror(saverr));
3247 		lockexit(32);
3248 	}
3249 	if (Nflag)
3250 		return;
3251 	n = write(fso, bf, size);
3252 	if (n != size) {
3253 		saverr = errno;
3254 		if (n == -1)
3255 			(void) fprintf(stderr,
3256 			    gettext("write error on sector %lld: %s\n"),
3257 			    bno, strerror(saverr));
3258 		else
3259 			(void) fprintf(stderr, gettext(
3260 			    "short write (%d of %d bytes) on sector %lld\n"),
3261 			    n, size, bno);
3262 		lockexit(32);
3263 	}
3264 }
3265 
3266 /*
3267  * write a block to the file system -- buffered with aio
3268  */
3269 static void
3270 awtfs(diskaddr_t bno, int size, char *bf, int release)
3271 {
3272 	int n;
3273 	aio_trans 	*transp;
3274 	sigset_t 	old_mask;
3275 
3276 	if (fso == -1)
3277 		return;
3278 
3279 	/*
3280 	 * We need to keep things consistent if we get interrupted,
3281 	 * so defer any expected interrupts for the time being.
3282 	 */
3283 	block_sigint(&old_mask);
3284 
3285 	if (Nflag) {
3286 		if (release == RELEASE)
3287 			freebuf(bf);
3288 	} else {
3289 		transp = get_aiop();
3290 		transp->bno = bno;
3291 		transp->buffer = bf;
3292 		transp->size = size;
3293 		transp->release = release;
3294 
3295 		n = aiowrite(fso, bf, size, (off_t)bno * sectorsize,
3296 		    SEEK_SET, &transp->resultbuf);
3297 
3298 		if (n < 0) {
3299 			/*
3300 			 * The aiowrite() may have failed because the
3301 			 * kernel didn't have enough memory to do the job.
3302 			 * Flush all pending writes and try a normal
3303 			 * write().  wtfs_breakup() will call exit if it
3304 			 * fails, so we don't worry about errors here.
3305 			 */
3306 			flush_writes();
3307 			wtfs_breakup(transp->bno, transp->size, transp->buffer);
3308 			freetrans(transp);
3309 		} else {
3310 			/*
3311 			 * Keep track of our pending writes.
3312 			 */
3313 			results.outstanding++;
3314 			if (results.outstanding > results.maxpend)
3315 				results.maxpend = results.outstanding;
3316 		}
3317 	}
3318 
3319 	unblock_sigint(&old_mask);
3320 }
3321 
3322 
3323 /*
3324  * write a block to the file system, but break it up into sbsize
3325  * chunks to avoid forcing a large amount of memory to be locked down.
3326  * Only used as a fallback when an aio write has failed.
3327  */
3328 static void
3329 wtfs_breakup(diskaddr_t bno, int size, char *bf)
3330 {
3331 	int n, saverr;
3332 	int wsize;
3333 	int block_incr = sbsize / sectorsize;
3334 
3335 	if (size < sbsize)
3336 		wsize = size;
3337 	else
3338 		wsize = sbsize;
3339 
3340 	n = 0;
3341 	while (size) {
3342 		/*
3343 		 * Note: the llseek() can succeed, even if the offset is
3344 		 * out of range.  It's not until the file i/o operation
3345 		 * (the write()) that one knows for sure if the raw device
3346 		 * can handle the offset.
3347 		 */
3348 		if (llseek(fso, (offset_t)bno * sectorsize, 0) < 0) {
3349 			saverr = errno;
3350 			(void) fprintf(stderr,
3351 			    gettext("seek error on sector %lld: %s\n"),
3352 			    bno, strerror(saverr));
3353 			lockexit(32);
3354 		}
3355 
3356 		n = write(fso, bf, wsize);
3357 		if (n == -1) {
3358 			saverr = errno;
3359 			(void) fprintf(stderr,
3360 			    gettext("write error on sector %lld: %s\n"),
3361 			    bno, strerror(saverr));
3362 			lockexit(32);
3363 		}
3364 		if (n != wsize) {
3365 			saverr = errno;
3366 			(void) fprintf(stderr, gettext(
3367 			    "short write (%d of %d bytes) on sector %lld\n"),
3368 			    n, size, bno);
3369 			lockexit(32);
3370 		}
3371 
3372 		bno += block_incr;
3373 		bf += wsize;
3374 		size -= wsize;
3375 		if (size < wsize)
3376 			wsize = size;
3377 	}
3378 }
3379 
3380 
3381 /*
3382  * check if a block is available
3383  */
3384 static int
3385 isblock(struct fs *fs, unsigned char *cp, int h)
3386 {
3387 	unsigned char mask;
3388 
3389 	switch (fs->fs_frag) {
3390 	case 8:
3391 		return (cp[h] == 0xff);
3392 	case 4:
3393 		mask = 0x0f << ((h & 0x1) << 2);
3394 		return ((cp[h >> 1] & mask) == mask);
3395 	case 2:
3396 		mask = 0x03 << ((h & 0x3) << 1);
3397 		return ((cp[h >> 2] & mask) == mask);
3398 	case 1:
3399 		mask = 0x01 << (h & 0x7);
3400 		return ((cp[h >> 3] & mask) == mask);
3401 	default:
3402 		(void) fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
3403 		return (0);
3404 	}
3405 }
3406 
3407 /*
3408  * take a block out of the map
3409  */
3410 static void
3411 clrblock(struct fs *fs, unsigned char *cp, int h)
3412 {
3413 	switch ((fs)->fs_frag) {
3414 	case 8:
3415 		cp[h] = 0;
3416 		return;
3417 	case 4:
3418 		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
3419 		return;
3420 	case 2:
3421 		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
3422 		return;
3423 	case 1:
3424 		cp[h >> 3] &= ~(0x01 << (h & 0x7));
3425 		return;
3426 	default:
3427 		(void) fprintf(stderr,
3428 		    gettext("clrblock: bad fs_frag value %d\n"), fs->fs_frag);
3429 		return;
3430 	}
3431 }
3432 
3433 /*
3434  * put a block into the map
3435  */
3436 static void
3437 setblock(struct fs *fs, unsigned char *cp, int h)
3438 {
3439 	switch (fs->fs_frag) {
3440 	case 8:
3441 		cp[h] = 0xff;
3442 		return;
3443 	case 4:
3444 		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
3445 		return;
3446 	case 2:
3447 		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
3448 		return;
3449 	case 1:
3450 		cp[h >> 3] |= (0x01 << (h & 0x7));
3451 		return;
3452 	default:
3453 		(void) fprintf(stderr,
3454 		    gettext("setblock: bad fs_frag value %d\n"), fs->fs_frag);
3455 		return;
3456 	}
3457 }
3458 
3459 static void
3460 usage()
3461 {
3462 	(void) fprintf(stderr,
3463 	    gettext("ufs usage: mkfs [-F FSType] [-V] [-m] [-o options] "
3464 	    "special "				/* param 0 */
3465 	    "size(sectors) \\ \n"));		/* param 1 */
3466 	(void) fprintf(stderr,
3467 	    "[nsect "				/* param 2 */
3468 	    "ntrack "				/* param 3 */
3469 	    "bsize "				/* param 4 */
3470 	    "fragsize "				/* param 5 */
3471 	    "cpg "				/* param 6 */
3472 	    "free "				/* param 7 */
3473 	    "rps "				/* param 8 */
3474 	    "nbpi "				/* param 9 */
3475 	    "opt "				/* param 10 */
3476 	    "apc "				/* param 11 */
3477 	    "gap "				/* param 12 */
3478 	    "nrpos "				/* param 13 */
3479 	    "maxcontig "			/* param 14 */
3480 	    "mtb]\n");				/* param 15 */
3481 	(void) fprintf(stderr,
3482 	    gettext(" -m : dump fs cmd line used to make this partition\n"
3483 	    " -V :print this command line and return\n"
3484 	    " -o :ufs options: :nsect=%d,ntrack=%d,bsize=%d,fragsize=%d\n"
3485 	    " -o :ufs options: :cgsize=%d,free=%d,rps=%d,nbpi=%d,opt=%c\n"
3486 	    " -o :ufs options: :apc=%d,gap=%d,nrpos=%d,maxcontig=%d\n"
3487 	    " -o :ufs options: :mtb=%c,calcsb,calcbinsb\n"
3488 "NOTE that all -o suboptions: must be separated only by commas so as to\n"
3489 "be parsed as a single argument\n"),
3490 	    nsect, ntrack, bsize, fragsize, cpg, sblock.fs_minfree, rps,
3491 	    nbpi, opt, apc, (rotdelay == -1) ? 0 : rotdelay,
3492 	    sblock.fs_nrpos, maxcontig, mtb);
3493 	lockexit(32);
3494 }
3495 
3496 /*ARGSUSED*/
3497 static void
3498 dump_fscmd(char *fsys, int fsi)
3499 {
3500 	int64_t used, bpcg, inospercg;
3501 	int64_t nbpi;
3502 	uint64_t nbytes64;
3503 
3504 	bzero((char *)&sblock, sizeof (sblock));
3505 	rdfs((diskaddr_t)SBLOCK, SBSIZE, (char *)&sblock);
3506 
3507 	/*
3508 	 * ensure a valid file system and if not, exit with error or else
3509 	 * we will end up computing block numbers etc and dividing by zero
3510 	 * which will cause floating point errors in this routine.
3511 	 */
3512 
3513 	if ((sblock.fs_magic != FS_MAGIC) &&
3514 	    (sblock.fs_magic != MTB_UFS_MAGIC)) {
3515 		(void) fprintf(stderr, gettext(
3516 		    "[not currently a valid file system - bad superblock]\n"));
3517 		lockexit(32);
3518 	}
3519 
3520 	if (sblock.fs_magic == FS_MAGIC &&
3521 	    (sblock.fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
3522 	    sblock.fs_version != UFS_VERSION_MIN)) {
3523 		(void) fprintf(stderr, gettext(
3524 		    "Unknown version of UFS format: %d\n"), sblock.fs_version);
3525 		lockexit(32);
3526 	}
3527 
3528 	if (sblock.fs_magic == MTB_UFS_MAGIC &&
3529 	    (sblock.fs_version > MTB_UFS_VERSION_1 ||
3530 	    sblock.fs_version < MTB_UFS_VERSION_MIN)) {
3531 		(void) fprintf(stderr, gettext(
3532 		    "Unknown version of UFS format: %d\n"), sblock.fs_version);
3533 		lockexit(32);
3534 	}
3535 
3536 	/*
3537 	 * Compute a reasonable nbpi value.
3538 	 * The algorithm for "used" is copied from code
3539 	 * in main() verbatim.
3540 	 * The nbpi equation is taken from main where the
3541 	 * fs_ipg value is set for the last time.  The INOPB(...) - 1
3542 	 * is used to account for the roundup.
3543 	 * The problem is that a range of nbpi values map to
3544 	 * the same file system layout.  So it is not possible
3545 	 * to calculate the exact value specified when the file
3546 	 * system was created.  So instead we determine the top
3547 	 * end of the range of values.
3548 	 */
3549 	bpcg = sblock.fs_spc * sectorsize;
3550 	inospercg = (int64_t)roundup(bpcg / sizeof (struct dinode),
3551 	    INOPB(&sblock));
3552 	if (inospercg > MAXIpG(&sblock))
3553 		inospercg = MAXIpG(&sblock);
3554 	used = (int64_t)
3555 	    (sblock.fs_iblkno + inospercg / INOPF(&sblock)) * NSPF(&sblock);
3556 	used *= sectorsize;
3557 	nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
3558 
3559 	/*
3560 	 * The top end of the range of values for nbpi may not be
3561 	 * a valid command line value for mkfs. Report the bottom
3562 	 * end instead.
3563 	 */
3564 	nbpi = (int64_t)(nbytes64 / (sblock.fs_ipg));
3565 
3566 	(void) fprintf(stdout, gettext("mkfs -F ufs -o "), fsys);
3567 	(void) fprintf(stdout, "nsect=%d,ntrack=%d,",
3568 	    sblock.fs_nsect, sblock.fs_ntrak);
3569 	(void) fprintf(stdout, "bsize=%d,fragsize=%d,cgsize=%d,free=%d,",
3570 	    sblock.fs_bsize, sblock.fs_fsize, sblock.fs_cpg, sblock.fs_minfree);
3571 	(void) fprintf(stdout, "rps=%d,nbpi=%lld,opt=%c,apc=%d,gap=%d,",
3572 	    sblock.fs_rps, nbpi, (sblock.fs_optim == FS_OPTSPACE) ? 's' : 't',
3573 	    (sblock.fs_ntrak * sblock.fs_nsect) - sblock.fs_spc,
3574 	    sblock.fs_rotdelay);
3575 	(void) fprintf(stdout, "nrpos=%d,maxcontig=%d,mtb=%c ",
3576 	    sblock.fs_nrpos, sblock.fs_maxcontig,
3577 	    ((sblock.fs_magic == MTB_UFS_MAGIC) ? 'y' : 'n'));
3578 	(void) fprintf(stdout, "%s %lld\n", fsys,
3579 	    fsbtodb(&sblock, sblock.fs_size));
3580 
3581 	bzero((char *)&sblock, sizeof (sblock));
3582 }
3583 
3584 /* number ************************************************************* */
3585 /*									*/
3586 /* Convert a numeric string arg to binary				*/
3587 /*									*/
3588 /* Args:	d_value - default value, if have parse error		*/
3589 /*		param - the name of the argument, for error messages	*/
3590 /*		flags - parser state and what's allowed in the arg	*/
3591 /* Global arg:  string - pointer to command arg				*/
3592 /*									*/
3593 /* Valid forms: 123 | 123k | 123*123 | 123x123				*/
3594 /*									*/
3595 /* Return:	converted number					*/
3596 /*									*/
3597 /* ******************************************************************** */
3598 
3599 static uint64_t
3600 number(uint64_t d_value, char *param, int flags)
3601 {
3602 	char *cs;
3603 	uint64_t n, t;
3604 	uint64_t cut = BIG / 10;    /* limit to avoid overflow */
3605 	int minus = 0;
3606 
3607 	cs = string;
3608 	if (*cs == '-') {
3609 		minus = 1;
3610 		cs += 1;
3611 	}
3612 	if ((*cs < '0') || (*cs > '9')) {
3613 		goto bail_out;
3614 	}
3615 	n = 0;
3616 	while ((*cs >= '0') && (*cs <= '9') && (n <= cut)) {
3617 		n = n*10 + *cs++ - '0';
3618 	}
3619 	if (minus)
3620 		n = -n;
3621 	for (;;) {
3622 		switch (*cs++) {
3623 		case 'k':
3624 			if (flags & ALLOW_END_ONLY)
3625 				goto bail_out;
3626 			if (n > (BIG / 1024))
3627 				goto overflow;
3628 			n *= 1024;
3629 			continue;
3630 
3631 		case '*':
3632 		case 'x':
3633 			if (flags & ALLOW_END_ONLY)
3634 				goto bail_out;
3635 			string = cs;
3636 			t = number(d_value, param, flags);
3637 			if (n > (BIG / t))
3638 				goto overflow;
3639 			n *= t;
3640 			cs = string + 1; /* adjust for -- below */
3641 
3642 			/* recursion has read rest of expression */
3643 			/* FALLTHROUGH */
3644 
3645 		case ',':
3646 		case '\0':
3647 			cs--;
3648 			string = cs;
3649 			return (n);
3650 
3651 		case '%':
3652 			if (flags & ALLOW_END_ONLY)
3653 				goto bail_out;
3654 			if (flags & ALLOW_PERCENT) {
3655 				flags &= ~ALLOW_PERCENT;
3656 				flags |= ALLOW_END_ONLY;
3657 				continue;
3658 			}
3659 			goto bail_out;
3660 
3661 		case 'm':
3662 			if (flags & ALLOW_END_ONLY)
3663 				goto bail_out;
3664 			if (flags & ALLOW_MS1) {
3665 				flags &= ~ALLOW_MS1;
3666 				flags |= ALLOW_MS2;
3667 				continue;
3668 			}
3669 			goto bail_out;
3670 
3671 		case 's':
3672 			if (flags & ALLOW_END_ONLY)
3673 				goto bail_out;
3674 			if (flags & ALLOW_MS2) {
3675 				flags &= ~ALLOW_MS2;
3676 				flags |= ALLOW_END_ONLY;
3677 				continue;
3678 			}
3679 			goto bail_out;
3680 
3681 		case '0': case '1': case '2': case '3': case '4':
3682 		case '5': case '6': case '7': case '8': case '9':
3683 overflow:
3684 			(void) fprintf(stderr,
3685 			    gettext("mkfs: value for %s overflowed\n"),
3686 			    param);
3687 			while ((*cs != '\0') && (*cs != ','))
3688 				cs++;
3689 			string = cs;
3690 			return (BIG);
3691 
3692 		default:
3693 bail_out:
3694 			(void) fprintf(stderr, gettext(
3695 			    "mkfs: bad numeric arg for %s: \"%s\"\n"),
3696 			    param, string);
3697 			while ((*cs != '\0') && (*cs != ','))
3698 				cs++;
3699 			string = cs;
3700 			if (d_value != NO_DEFAULT) {
3701 				(void) fprintf(stderr,
3702 				    gettext("mkfs: %s reset to default %lld\n"),
3703 				    param, d_value);
3704 				return (d_value);
3705 			}
3706 			lockexit(2);
3707 
3708 		}
3709 	} /* never gets here */
3710 }
3711 
3712 /* match ************************************************************** */
3713 /*									*/
3714 /* Compare two text strings for equality				*/
3715 /*									*/
3716 /* Arg:	 s - pointer to string to match with a command arg		*/
3717 /* Global arg:  string - pointer to command arg				*/
3718 /*									*/
3719 /* Return:	1 if match, 0 if no match				*/
3720 /*		If match, also reset `string' to point to the text	*/
3721 /*		that follows the matching text.				*/
3722 /*									*/
3723 /* ******************************************************************** */
3724 
3725 static int
3726 match(char *s)
3727 {
3728 	char *cs;
3729 
3730 	cs = string;
3731 	while (*cs++ == *s) {
3732 		if (*s++ == '\0') {
3733 			goto true;
3734 		}
3735 	}
3736 	if (*s != '\0') {
3737 		return (0);
3738 	}
3739 
3740 true:
3741 	cs--;
3742 	string = cs;
3743 	return (1);
3744 }
3745 
3746 /*
3747  * GROWFS ROUTINES
3748  */
3749 
3750 /* ARGSUSED */
3751 void
3752 lockexit(int exitstatus)
3753 {
3754 	if (Pflag) {
3755 		/* the probe mode neither changes nor locks the filesystem */
3756 		exit(exitstatus);
3757 	}
3758 
3759 	/*
3760 	 * flush the dirty cylinder group
3761 	 */
3762 	if (inlockexit == 0) {
3763 		inlockexit = 1;
3764 		flcg();
3765 	}
3766 
3767 	if (aio_inited) {
3768 		flush_writes();
3769 	}
3770 
3771 	/*
3772 	 * make sure the file system is unlocked before exiting
3773 	 */
3774 	if ((inlockexit == 1) && (!isbad)) {
3775 		inlockexit = 2;
3776 		ulockfs();
3777 		/*
3778 		 * if logging was enabled, then re-enable it
3779 		 */
3780 		if (waslog) {
3781 			if (rl_log_control(fsys, _FIOLOGENABLE) != RL_SUCCESS) {
3782 				(void) fprintf(stderr, gettext(
3783 				    "failed to re-enable logging\n"));
3784 			}
3785 		}
3786 	} else if (grow) {
3787 		if (isbad) {
3788 			(void) fprintf(stderr, gettext(
3789 			    "Filesystem is currently inconsistent.  It "
3790 			    "must be repaired with fsck(1M)\nbefore being "
3791 			    "used.  Use the following command to "
3792 			    "do this:\n\n\tfsck %s\n\n"), fsys);
3793 
3794 			if (ismounted) {
3795 				(void) fprintf(stderr, gettext(
3796 				    "You will be told that the filesystem "
3797 				    "is already mounted, and asked if you\n"
3798 				    "wish to continue.  Answer `yes' to "
3799 				    "this question.\n\n"));
3800 			}
3801 
3802 			(void) fprintf(stderr, gettext(
3803 			    "One problem should be reported, that the summary "
3804 			    "information is bad.\nYou will then be asked if it "
3805 			    "should be salvaged.  Answer `yes' to\nthis "
3806 			    "question.\n\n"));
3807 		}
3808 
3809 		if (ismounted) {
3810 			/*
3811 			 * In theory, there's no way to get here without
3812 			 * isbad also being set, but be robust in the
3813 			 * face of future code changes.
3814 			 */
3815 			(void) fprintf(stderr, gettext(
3816 			    "The filesystem is currently mounted "
3817 			    "read-only and write-locked.  "));
3818 			if (isbad) {
3819 				(void) fprintf(stderr, gettext(
3820 				    "After\nrunning fsck, unlock the "
3821 				    "filesystem and "));
3822 			} else {
3823 				(void) fprintf(stderr, gettext(
3824 				    "Unlock the filesystem\nand "));
3825 			}
3826 
3827 			(void) fprintf(stderr, gettext(
3828 			    "re-enable writing with\nthe following "
3829 			    "command:\n\n\tlockfs -u %s\n\n"), directory);
3830 		}
3831 	}
3832 
3833 	exit(exitstatus);
3834 }
3835 
3836 void
3837 randomgeneration()
3838 {
3839 	int		 i;
3840 	struct dinode	*dp;
3841 
3842 	/*
3843 	 * always perform fsirand(1) function... newfs will notice that
3844 	 * the inodes have been randomized and will not call fsirand itself
3845 	 */
3846 	for (i = 0, dp = zino; i < sblock.fs_inopb; ++i, ++dp)
3847 		IRANDOMIZE(&dp->di_ic);
3848 }
3849 
3850 /*
3851  * Check the size of the summary information.
3852  * Fields in sblock are not changed in this function.
3853  *
3854  * For an 8K filesystem block, the maximum number of cylinder groups is 16384.
3855  *     MAXCSBUFS {32}  *   8K  {FS block size}
3856  *                         divided by (sizeof csum) {16}
3857  *
3858  * Note that MAXCSBUFS is not used in the kernel; as of Solaris 2.6 build 32,
3859  * this is the only place where it's referenced.
3860  */
3861 void
3862 checksummarysize()
3863 {
3864 	diskaddr_t	dmax;
3865 	diskaddr_t	dmin;
3866 	int64_t	cg0frags;
3867 	int64_t	cg0blocks;
3868 	int64_t	maxncg;
3869 	int64_t	maxfrags;
3870 	uint64_t	fs_size;
3871 	uint64_t maxfs_blocks; /* filesystem blocks for max filesystem size */
3872 
3873 	/*
3874 	 * compute the maximum summary info size
3875 	 */
3876 	dmin = cgdmin(&sblock, 0);
3877 	dmax = cgbase(&sblock, 0) + sblock.fs_fpg;
3878 	fs_size = (grow) ? grow_fs_size : sblock.fs_size;
3879 	if (dmax > fs_size)
3880 		dmax = fs_size;
3881 	cg0frags  = dmax - dmin;
3882 	cg0blocks = cg0frags / sblock.fs_frag;
3883 	cg0frags = cg0blocks * sblock.fs_frag;
3884 	maxncg   = (longlong_t)cg0blocks *
3885 	    (longlong_t)(sblock.fs_bsize / sizeof (struct csum));
3886 
3887 	maxfs_blocks = FS_MAX;
3888 
3889 	if (maxncg > ((longlong_t)maxfs_blocks / (longlong_t)sblock.fs_fpg) + 1)
3890 		maxncg = ((longlong_t)maxfs_blocks /
3891 		    (longlong_t)sblock.fs_fpg) + 1;
3892 
3893 	maxfrags = maxncg * (longlong_t)sblock.fs_fpg;
3894 
3895 	if (maxfrags > maxfs_blocks)
3896 		maxfrags = maxfs_blocks;
3897 
3898 
3899 	/*
3900 	 * remember for later processing in extendsummaryinfo()
3901 	 */
3902 	if (test)
3903 		grow_sifrag = dmin + (cg0blocks * sblock.fs_frag);
3904 	if (testfrags == 0)
3905 		testfrags = cg0frags;
3906 	if (testforce)
3907 		if (testfrags > cg0frags) {
3908 			(void) fprintf(stderr,
3909 			    gettext("Too many test frags (%lld); "
3910 			    "try %lld\n"), testfrags, cg0frags);
3911 			lockexit(32);
3912 		}
3913 
3914 	/*
3915 	 * if summary info is too large (too many cg's) tell the user and exit
3916 	 */
3917 	if ((longlong_t)sblock.fs_size > maxfrags) {
3918 		(void) fprintf(stderr, gettext(
3919 		    "Too many cylinder groups with %llu sectors;\n    try "
3920 		    "increasing cgsize, or decreasing fssize to %llu\n"),
3921 		    fsbtodb(&sblock, (uint64_t)sblock.fs_size),
3922 		    fsbtodb(&sblock, (uint64_t)maxfrags));
3923 		lockexit(32);
3924 	}
3925 }
3926 
3927 /*
3928  * checksblock() has two uses:
3929  *	- One is to sanity test the superblock and is used when newfs(1M)
3930  *	  is invoked with the "-N" option. If any discrepancy was found,
3931  *	  just return whatever error was found and do not exit.
3932  *	- the other use of it is in places where you expect the superblock
3933  *	  to be sane, and if it isn't, then we exit.
3934  * Which of the above two actions to take is indicated with the second argument.
3935  */
3936 
3937 int
3938 checksblock(struct fs sb, int proceed)
3939 {
3940 	int err = 0;
3941 	char *errmsg;
3942 
3943 	if ((sb.fs_magic != FS_MAGIC) && (sb.fs_magic != MTB_UFS_MAGIC)) {
3944 		err = 1;
3945 		errmsg = gettext("Bad superblock; magic number wrong\n");
3946 	} else if ((sb.fs_magic == FS_MAGIC &&
3947 	    (sb.fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
3948 	    sb.fs_version != UFS_VERSION_MIN)) ||
3949 	    (sb.fs_magic == MTB_UFS_MAGIC &&
3950 	    (sb.fs_version > MTB_UFS_VERSION_1 ||
3951 	    sb.fs_version < MTB_UFS_VERSION_MIN))) {
3952 		err = 2;
3953 		errmsg = gettext("Unrecognized version of UFS\n");
3954 	} else if (sb.fs_ncg < 1) {
3955 		err = 3;
3956 		errmsg = gettext("Bad superblock; ncg out of range\n");
3957 	} else if (sb.fs_cpg < 1) {
3958 		err = 4;
3959 		errmsg = gettext("Bad superblock; cpg out of range\n");
3960 	} else if (sb.fs_ncg * sb.fs_cpg < sb.fs_ncyl ||
3961 	    (sb.fs_ncg - 1) * sb.fs_cpg >= sb.fs_ncyl) {
3962 		err = 5;
3963 		errmsg = gettext("Bad superblock; ncyl out of range\n");
3964 	} else if (sb.fs_sbsize <= 0 || sb.fs_sbsize > sb.fs_bsize) {
3965 		err = 6;
3966 		errmsg = gettext("Bad superblock; superblock size out of "
3967 		    "range\n");
3968 	}
3969 
3970 	if (proceed) {
3971 		if (err) dprintf(("%s", errmsg));
3972 		return (err);
3973 	}
3974 
3975 	if (err) {
3976 		fprintf(stderr, "%s", errmsg);
3977 		lockexit(32);
3978 	}
3979 	return (32);
3980 }
3981 
3982 /*
3983  * Roll the embedded log, if any, and set up the global variables
3984  * islog, islogok and isufslog.
3985  */
3986 static void
3987 logsetup(char *devstr)
3988 {
3989 	void		*buf, *ud_buf;
3990 	extent_block_t	*ebp;
3991 	ml_unit_t	*ul;
3992 	ml_odunit_t	*ud;
3993 
3994 	/*
3995 	 * Does the superblock indicate that we are supposed to have a log ?
3996 	 */
3997 	if (sblock.fs_logbno == 0) {
3998 		/*
3999 		 * No log present, nothing to do.
4000 		 */
4001 		islogok = 0;
4002 		islog = 0;
4003 		isufslog = 0;
4004 		return;
4005 	} else {
4006 		/*
4007 		 * There's a log in a yet unknown state, attempt to roll it.
4008 		 */
4009 		islog = 1;
4010 		islogok = 0;
4011 		isufslog = 0;
4012 
4013 		/*
4014 		 * We failed to roll the log, bail out.
4015 		 */
4016 		if (rl_roll_log(devstr) != RL_SUCCESS)
4017 			return;
4018 
4019 		isufslog = 1;
4020 
4021 		/* log is not okay; check the fs */
4022 		if ((FSOKAY != (sblock.fs_state + sblock.fs_time)) ||
4023 		    (sblock.fs_clean != FSLOG))
4024 			return;
4025 
4026 		/* get the log allocation block */
4027 		buf = (void *)malloc(DEV_BSIZE);
4028 		if (buf == (void *) NULL)
4029 			return;
4030 
4031 		ud_buf = (void *)malloc(DEV_BSIZE);
4032 		if (ud_buf == (void *) NULL) {
4033 			free(buf);
4034 			return;
4035 		}
4036 
4037 		rdfs((diskaddr_t)logbtodb(&sblock, sblock.fs_logbno),
4038 		    DEV_BSIZE, buf);
4039 		ebp = (extent_block_t *)buf;
4040 
4041 		/* log allocation block is not okay; check the fs */
4042 		if (ebp->type != LUFS_EXTENTS) {
4043 			free(buf);
4044 			free(ud_buf);
4045 			return;
4046 		}
4047 
4048 		/* get the log state block(s) */
4049 		rdfs((diskaddr_t)logbtodb(&sblock, ebp->extents[0].pbno),
4050 		    DEV_BSIZE, ud_buf);
4051 		ud = (ml_odunit_t *)ud_buf;
4052 		ul = (ml_unit_t *)malloc(sizeof (*ul));
4053 		ul->un_ondisk = *ud;
4054 
4055 		/* log state is okay */
4056 		if ((ul->un_chksum == ul->un_head_ident + ul->un_tail_ident) &&
4057 		    (ul->un_version == LUFS_VERSION_LATEST) &&
4058 		    (ul->un_badlog == 0))
4059 			islogok = 1;
4060 		free(ud_buf);
4061 		free(buf);
4062 		free(ul);
4063 	}
4064 }
4065 
4066 void
4067 growinit(char *devstr)
4068 {
4069 	int	i;
4070 	char	buf[DEV_BSIZE];
4071 
4072 	/*
4073 	 * Read and verify the superblock
4074 	 */
4075 	rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
4076 	(void) checksblock(sblock, 0);
4077 	if (sblock.fs_postblformat != FS_DYNAMICPOSTBLFMT) {
4078 		(void) fprintf(stderr,
4079 		    gettext("old file system format; can't growfs\n"));
4080 		lockexit(32);
4081 	}
4082 
4083 	/*
4084 	 * can't shrink a file system
4085 	 */
4086 	grow_fssize = fsbtodb(&sblock, (uint64_t)sblock.fs_size);
4087 	if (fssize_db < grow_fssize) {
4088 		(void) fprintf(stderr,
4089 		    gettext("%lld sectors < current size of %lld sectors\n"),
4090 		    fssize_db, grow_fssize);
4091 		lockexit(32);
4092 	}
4093 
4094 	/*
4095 	 * can't grow a system to over a terabyte unless it was set up
4096 	 * as an MTB UFS file system.
4097 	 */
4098 	if (mtb == 'y' && sblock.fs_magic != MTB_UFS_MAGIC) {
4099 		if (fssize_db >= SECTORS_PER_TERABYTE) {
4100 			(void) fprintf(stderr, gettext(
4101 "File system was not set up with the multi-terabyte format.\n"));
4102 			(void) fprintf(stderr, gettext(
4103 "Its size cannot be increased to a terabyte or more.\n"));
4104 		} else {
4105 			(void) fprintf(stderr, gettext(
4106 "Cannot convert file system to multi-terabyte format.\n"));
4107 		}
4108 		lockexit(32);
4109 	}
4110 
4111 	logsetup(devstr);
4112 
4113 	/*
4114 	 * can't growfs when logging device has errors
4115 	 */
4116 	if ((islog && !islogok) ||
4117 	    ((FSOKAY == (sblock.fs_state + sblock.fs_time)) &&
4118 	    (sblock.fs_clean == FSLOG && !islog))) {
4119 		(void) fprintf(stderr,
4120 		    gettext("logging device has errors; can't growfs\n"));
4121 		lockexit(32);
4122 	}
4123 
4124 	/*
4125 	 * disable ufs logging for growing
4126 	 */
4127 	if (isufslog) {
4128 		if (rl_log_control(devstr, _FIOLOGDISABLE) != RL_SUCCESS) {
4129 			(void) fprintf(stderr, gettext(
4130 			    "failed to disable logging\n"));
4131 			lockexit(32);
4132 		}
4133 		islog = 0;
4134 		waslog = 1;
4135 	}
4136 
4137 	/*
4138 	 * if mounted write lock the file system to be grown
4139 	 */
4140 	if (ismounted)
4141 		wlockfs();
4142 
4143 	/*
4144 	 * refresh dynamic superblock state - disabling logging will have
4145 	 * changed the amount of free space available in the file system
4146 	 */
4147 	rdfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
4148 
4149 	/*
4150 	 * make sure device is big enough
4151 	 */
4152 	rdfs((diskaddr_t)fssize_db - 1, DEV_BSIZE, buf);
4153 	wtfs((diskaddr_t)fssize_db - 1, DEV_BSIZE, buf);
4154 
4155 	/*
4156 	 * read current summary information
4157 	 */
4158 	grow_fscs = read_summaryinfo(&sblock);
4159 
4160 	/*
4161 	 * save some current size related fields from the superblock
4162 	 * These are used in extendsummaryinfo()
4163 	 */
4164 	grow_fs_size	= sblock.fs_size;
4165 	grow_fs_ncg	= sblock.fs_ncg;
4166 	grow_fs_csaddr	= (diskaddr_t)sblock.fs_csaddr;
4167 	grow_fs_cssize	= sblock.fs_cssize;
4168 
4169 	/*
4170 	 * save and reset the clean flag
4171 	 */
4172 	if (FSOKAY == (sblock.fs_state + sblock.fs_time))
4173 		grow_fs_clean = sblock.fs_clean;
4174 	else
4175 		grow_fs_clean = FSBAD;
4176 	sblock.fs_clean = FSBAD;
4177 	sblock.fs_state = FSOKAY - sblock.fs_time;
4178 	isbad = 1;
4179 	wtfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
4180 }
4181 
4182 void
4183 checkdev(char *rdev, char *bdev)
4184 {
4185 	struct stat64	statarea;
4186 
4187 	if (stat64(bdev, &statarea) < 0) {
4188 		(void) fprintf(stderr, gettext("can't check mount point; "));
4189 		(void) fprintf(stderr, gettext("can't stat %s\n"), bdev);
4190 		lockexit(32);
4191 	}
4192 	if ((statarea.st_mode & S_IFMT) != S_IFBLK) {
4193 		(void) fprintf(stderr, gettext(
4194 		    "can't check mount point; %s is not a block device\n"),
4195 		    bdev);
4196 		lockexit(32);
4197 	}
4198 	if (stat64(rdev, &statarea) < 0) {
4199 		(void) fprintf(stderr, gettext("can't stat %s\n"), rdev);
4200 		lockexit(32);
4201 	}
4202 	if ((statarea.st_mode & S_IFMT) != S_IFCHR) {
4203 		(void) fprintf(stderr,
4204 		    gettext("%s is not a character device\n"), rdev);
4205 		lockexit(32);
4206 	}
4207 }
4208 
4209 void
4210 checkmount(struct mnttab *mntp, char *bdevname)
4211 {
4212 	struct stat64	statdir;
4213 	struct stat64	statdev;
4214 
4215 	if (strcmp(bdevname, mntp->mnt_special) == 0) {
4216 		if (stat64(mntp->mnt_mountp, &statdir) == -1) {
4217 			(void) fprintf(stderr, gettext("can't stat %s\n"),
4218 			    mntp->mnt_mountp);
4219 			lockexit(32);
4220 		}
4221 		if (stat64(mntp->mnt_special, &statdev) == -1) {
4222 			(void) fprintf(stderr, gettext("can't stat %s\n"),
4223 			    mntp->mnt_special);
4224 			lockexit(32);
4225 		}
4226 		if (statdir.st_dev != statdev.st_rdev) {
4227 			(void) fprintf(stderr, gettext(
4228 			    "%s is not mounted on %s; mnttab(4) wrong\n"),
4229 			    mntp->mnt_special, mntp->mnt_mountp);
4230 			lockexit(32);
4231 		}
4232 		ismounted = 1;
4233 		if (directory) {
4234 			if (strcmp(mntp->mnt_mountp, directory) != 0) {
4235 				(void) fprintf(stderr,
4236 				    gettext("%s is mounted on %s, not %s\n"),
4237 				    bdevname, mntp->mnt_mountp, directory);
4238 				lockexit(32);
4239 			}
4240 		} else {
4241 			if (grow)
4242 				(void) fprintf(stderr, gettext(
4243 				    "%s is mounted on %s; can't growfs\n"),
4244 				    bdevname, mntp->mnt_mountp);
4245 			else
4246 				(void) fprintf(stderr,
4247 				    gettext("%s is mounted, can't mkfs\n"),
4248 				    bdevname);
4249 			lockexit(32);
4250 		}
4251 	}
4252 }
4253 
4254 struct dinode	*dibuf	= 0;
4255 diskaddr_t	difrag	= 0;
4256 
4257 struct dinode *
4258 gdinode(ino_t ino)
4259 {
4260 	/*
4261 	 * read the block of inodes containing inode number ino
4262 	 */
4263 	if (dibuf == 0)
4264 		dibuf = (struct dinode *)malloc((unsigned)sblock.fs_bsize);
4265 	if (itod(&sblock, ino) != difrag) {
4266 		difrag = itod(&sblock, ino);
4267 		rdfs(fsbtodb(&sblock, (uint64_t)difrag), (int)sblock.fs_bsize,
4268 		    (char *)dibuf);
4269 	}
4270 	return (dibuf + (ino % INOPB(&sblock)));
4271 }
4272 
4273 /*
4274  * structure that manages the frags we need for extended summary info
4275  *	These frags can be:
4276  *		free
4277  *		data  block
4278  *		alloc block
4279  */
4280 struct csfrag {
4281 	struct csfrag	*next;		/* next entry */
4282 	daddr32_t	 ofrag;		/* old frag */
4283 	daddr32_t	 nfrag;		/* new frag */
4284 	long		 cylno;		/* cylno of nfrag */
4285 	long		 frags;		/* number of frags */
4286 	long		 size;		/* size in bytes */
4287 	ino_t		 ino;		/* inode number */
4288 	long		 fixed;		/* Boolean - Already fixed? */
4289 };
4290 struct csfrag	*csfrag;		/* state unknown */
4291 struct csfrag	*csfragino;		/* frags belonging to an inode */
4292 struct csfrag	*csfragfree;		/* frags that are free */
4293 
4294 daddr32_t maxcsfrag	= 0;		/* maximum in range */
4295 daddr32_t mincsfrag	= 0x7fffffff;	/* minimum in range */
4296 
4297 int
4298 csfraginrange(daddr32_t frag)
4299 {
4300 	return ((frag >= mincsfrag) && (frag <= maxcsfrag));
4301 }
4302 
4303 struct csfrag *
4304 findcsfrag(daddr32_t frag, struct csfrag **cfap)
4305 {
4306 	struct csfrag	*cfp;
4307 
4308 	if (!csfraginrange(frag))
4309 		return (NULL);
4310 
4311 	for (cfp = *cfap; cfp; cfp = cfp->next)
4312 		if (cfp->ofrag == frag)
4313 			return (cfp);
4314 	return (NULL);
4315 }
4316 
4317 void
4318 checkindirect(ino_t ino, daddr32_t *fragsp, daddr32_t frag, int level)
4319 {
4320 	int			i;
4321 	int			ne	= sblock.fs_bsize / sizeof (daddr32_t);
4322 	daddr32_t			fsb[MAXBSIZE / sizeof (daddr32_t)];
4323 
4324 	if (frag == 0)
4325 		return;
4326 
4327 	rdfs(fsbtodb(&sblock, frag), (int)sblock.fs_bsize,
4328 	    (char *)fsb);
4329 
4330 	checkdirect(ino, fragsp, fsb, sblock.fs_bsize / sizeof (daddr32_t));
4331 
4332 	if (level)
4333 		for (i = 0; i < ne && *fragsp; ++i)
4334 			checkindirect(ino, fragsp, fsb[i], level-1);
4335 }
4336 
4337 void
4338 addcsfrag(ino_t ino, daddr32_t frag, struct csfrag **cfap)
4339 {
4340 	struct csfrag	*cfp, *curr, *prev;
4341 
4342 	/*
4343 	 * establish a range for faster checking in csfraginrange()
4344 	 */
4345 	if (frag > maxcsfrag)
4346 		maxcsfrag = frag;
4347 	if (frag < mincsfrag)
4348 		mincsfrag = frag;
4349 
4350 	/*
4351 	 * if this frag belongs to an inode and is not the start of a block
4352 	 *	then see if it is part of a frag range for this inode
4353 	 */
4354 	if (ino && (frag % sblock.fs_frag))
4355 		for (cfp = *cfap; cfp; cfp = cfp->next) {
4356 			if (ino != cfp->ino)
4357 				continue;
4358 			if (frag != cfp->ofrag + cfp->frags)
4359 				continue;
4360 			cfp->frags++;
4361 			cfp->size += sblock.fs_fsize;
4362 			return;
4363 		}
4364 	/*
4365 	 * allocate a csfrag entry and insert it in an increasing order into the
4366 	 * specified list
4367 	 */
4368 	cfp = (struct csfrag *)calloc(1, sizeof (struct csfrag));
4369 	cfp->ino	= ino;
4370 	cfp->ofrag	= frag;
4371 	cfp->frags	= 1;
4372 	cfp->size	= sblock.fs_fsize;
4373 	for (prev = NULL, curr = *cfap; curr != NULL;
4374 	    prev = curr, curr = curr->next) {
4375 		if (frag < curr->ofrag) {
4376 			cfp->next = curr;
4377 			if (prev)
4378 				prev->next = cfp;	/* middle element */
4379 			else
4380 				*cfap = cfp;		/* first element */
4381 			break;
4382 		}
4383 		if (curr->next == NULL) {
4384 			curr->next = cfp;		/* last element	*/
4385 			break;
4386 		}
4387 	}
4388 	if (*cfap == NULL)	/* will happen only once */
4389 		*cfap = cfp;
4390 }
4391 
4392 void
4393 delcsfrag(daddr32_t frag, struct csfrag **cfap)
4394 {
4395 	struct csfrag	*cfp;
4396 	struct csfrag	**cfpp;
4397 
4398 	/*
4399 	 * free up entry whose beginning frag matches
4400 	 */
4401 	for (cfpp = cfap; *cfpp; cfpp = &(*cfpp)->next) {
4402 		if (frag == (*cfpp)->ofrag) {
4403 			cfp = *cfpp;
4404 			*cfpp = (*cfpp)->next;
4405 			free((char *)cfp);
4406 			return;
4407 		}
4408 	}
4409 }
4410 
4411 /*
4412  * See whether any of the direct blocks in the array pointed by "db" and of
4413  * length "ne" are within the range of frags needed to extend the cylinder
4414  * summary. If so, remove those frags from the "as-yet-unclassified" list
4415  * (csfrag) and add them to the "owned-by-inode" list (csfragino).
4416  * For each such frag found, decrement the frag count pointed to by fragsp.
4417  * "ino" is the inode that contains (either directly or indirectly) the frags
4418  * being checked.
4419  */
4420 void
4421 checkdirect(ino_t ino, daddr32_t *fragsp, daddr32_t *db, int ne)
4422 {
4423 	int	 i;
4424 	int	 j;
4425 	int	 found;
4426 	diskaddr_t	 frag;
4427 
4428 	/*
4429 	 * scan for allocation within the new summary info range
4430 	 */
4431 	for (i = 0; i < ne && *fragsp; ++i) {
4432 		if ((frag = *db++) != 0) {
4433 			found = 0;
4434 			for (j = 0; j < sblock.fs_frag && *fragsp; ++j) {
4435 				if (found || (found = csfraginrange(frag))) {
4436 					addcsfrag(ino, frag, &csfragino);
4437 					delcsfrag(frag, &csfrag);
4438 				}
4439 				++frag;
4440 				--(*fragsp);
4441 			}
4442 		}
4443 	}
4444 }
4445 
4446 void
4447 findcsfragino()
4448 {
4449 	int		 i;
4450 	int		 j;
4451 	daddr32_t		 frags;
4452 	struct dinode	*dp;
4453 
4454 	/*
4455 	 * scan all old inodes looking for allocations in the new
4456 	 * summary info range.  Move the affected frag from the
4457 	 * generic csfrag list onto the `owned-by-inode' list csfragino.
4458 	 */
4459 	for (i = UFSROOTINO; i < grow_fs_ncg*sblock.fs_ipg && csfrag; ++i) {
4460 		dp = gdinode((ino_t)i);
4461 		switch (dp->di_mode & IFMT) {
4462 			case IFSHAD	:
4463 			case IFLNK 	:
4464 			case IFDIR 	:
4465 			case IFREG 	: break;
4466 			default		: continue;
4467 		}
4468 
4469 		frags   = dbtofsb(&sblock, dp->di_blocks);
4470 
4471 		checkdirect((ino_t)i, &frags, &dp->di_db[0], NDADDR+NIADDR);
4472 		for (j = 0; j < NIADDR && frags; ++j) {
4473 			/* Negate the block if its an fallocate'd block */
4474 			if (dp->di_ib[j] < 0 && dp->di_ib[j] != UFS_HOLE)
4475 				checkindirect((ino_t)i, &frags,
4476 				    -(dp->di_ib[j]), j);
4477 			else
4478 				checkindirect((ino_t)i, &frags,
4479 				    dp->di_ib[j], j);
4480 		}
4481 	}
4482 }
4483 
4484 void
4485 fixindirect(daddr32_t frag, int level)
4486 {
4487 	int			 i;
4488 	int			 ne	= sblock.fs_bsize / sizeof (daddr32_t);
4489 	daddr32_t			fsb[MAXBSIZE / sizeof (daddr32_t)];
4490 
4491 	if (frag == 0)
4492 		return;
4493 
4494 	rdfs(fsbtodb(&sblock, (uint64_t)frag), (int)sblock.fs_bsize,
4495 	    (char *)fsb);
4496 
4497 	fixdirect((caddr_t)fsb, frag, fsb, ne);
4498 
4499 	if (level)
4500 		for (i = 0; i < ne; ++i)
4501 			fixindirect(fsb[i], level-1);
4502 }
4503 
4504 void
4505 fixdirect(caddr_t bp, daddr32_t frag, daddr32_t *db, int ne)
4506 {
4507 	int	 i;
4508 	struct csfrag	*cfp;
4509 
4510 	for (i = 0; i < ne; ++i, ++db) {
4511 		if (*db == 0)
4512 			continue;
4513 		if ((cfp = findcsfrag(*db, &csfragino)) == NULL)
4514 			continue;
4515 		*db = cfp->nfrag;
4516 		cfp->fixed = 1;
4517 		wtfs(fsbtodb(&sblock, (uint64_t)frag), (int)sblock.fs_bsize,
4518 		    bp);
4519 	}
4520 }
4521 
4522 void
4523 fixcsfragino()
4524 {
4525 	int		 i;
4526 	struct dinode	*dp;
4527 	struct csfrag	*cfp;
4528 
4529 	for (cfp = csfragino; cfp; cfp = cfp->next) {
4530 		if (cfp->fixed)
4531 			continue;
4532 		dp = gdinode((ino_t)cfp->ino);
4533 		fixdirect((caddr_t)dibuf, difrag, dp->di_db, NDADDR+NIADDR);
4534 		for (i = 0; i < NIADDR; ++i)
4535 			fixindirect(dp->di_ib[i], i);
4536 	}
4537 }
4538 
4539 /*
4540  * Read the cylinders summary information specified by settings in the
4541  * passed 'fs' structure into a new allocated array of csum structures.
4542  * The caller is responsible for freeing the returned array.
4543  * Return a pointer to an array of csum structures.
4544  */
4545 static struct csum *
4546 read_summaryinfo(struct	fs *fsp)
4547 {
4548 	struct csum 	*csp;
4549 	int		i;
4550 
4551 	if ((csp = malloc((size_t)fsp->fs_cssize)) == NULL) {
4552 		(void) fprintf(stderr, gettext("cannot create csum list,"
4553 		    " not enough memory\n"));
4554 		exit(32);
4555 	}
4556 
4557 	for (i = 0; i < fsp->fs_cssize; i += fsp->fs_bsize) {
4558 		rdfs(fsbtodb(fsp,
4559 		    (uint64_t)(fsp->fs_csaddr + numfrags(fsp, i))),
4560 		    (int)(fsp->fs_cssize - i < fsp->fs_bsize ?
4561 		    fsp->fs_cssize - i : fsp->fs_bsize), ((caddr_t)csp) + i);
4562 	}
4563 
4564 	return (csp);
4565 }
4566 
4567 /*
4568  * Check the allocation of fragments that are to be made part of a csum block.
4569  * A fragment is allocated if it is either in the csfragfree list or, it is
4570  * in the csfragino list and has new frags associated with it.
4571  * Return the number of allocated fragments.
4572  */
4573 int64_t
4574 checkfragallocated(daddr32_t frag)
4575 {
4576 	struct 	csfrag	*cfp;
4577 	/*
4578 	 * Since the lists are sorted we can break the search if the asked
4579 	 * frag is smaller then the one in the list.
4580 	 */
4581 	for (cfp = csfragfree; cfp != NULL && frag >= cfp->ofrag;
4582 	    cfp = cfp->next) {
4583 		if (frag == cfp->ofrag)
4584 			return (1);
4585 	}
4586 	for (cfp = csfragino; cfp != NULL && frag >= cfp->ofrag;
4587 	    cfp = cfp->next) {
4588 		if (frag == cfp->ofrag && cfp->nfrag != 0)
4589 			return (cfp->frags);
4590 	}
4591 
4592 	return (0);
4593 }
4594 
4595 /*
4596  * Figure out how much the filesystem can be grown. The limiting factor is
4597  * the available free space needed to extend the cg summary info block.
4598  * The free space is determined in three steps:
4599  * - Try to extend the cg summary block to the required size.
4600  * - Find free blocks in last cg.
4601  * - Find free space in the last already allocated fragment of the summary info
4602  *   block, and use it for additional csum structures.
4603  * Return the maximum size of the new filesystem or 0 if it can't be grown.
4604  * Please note that this function leaves the global list pointers csfrag,
4605  * csfragfree, and csfragino initialized, and the caller is responsible for
4606  * freeing the lists.
4607  */
4608 diskaddr_t
4609 probe_summaryinfo()
4610 {
4611 	/* fragments by which the csum block can be extended. */
4612 	int64_t 	growth_csum_frags = 0;
4613 	/* fragments by which the filesystem can be extended. */
4614 	int64_t		growth_fs_frags = 0;
4615 	int64_t		new_fs_cssize;	/* size of csum blk in the new FS */
4616 	int64_t		new_fs_ncg;	/* number of cg in the new FS */
4617 	int64_t 	spare_csum;
4618 	daddr32_t	oldfrag_daddr;
4619 	daddr32_t	newfrag_daddr;
4620 	daddr32_t	daddr;
4621 	int		i;
4622 
4623 	/*
4624 	 * read and verify the superblock
4625 	 */
4626 	rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
4627 	(void) checksblock(sblock, 0);
4628 
4629 	/*
4630 	 * check how much we can extend the cg summary info block
4631 	 */
4632 
4633 	/*
4634 	 * read current summary information
4635 	 */
4636 	fscs = read_summaryinfo(&sblock);
4637 
4638 	/*
4639 	 * build list of frags needed for cg summary info block extension
4640 	 */
4641 	oldfrag_daddr = howmany(sblock.fs_cssize, sblock.fs_fsize) +
4642 	    sblock.fs_csaddr;
4643 	new_fs_ncg = howmany(dbtofsb(&sblock, fssize_db), sblock.fs_fpg);
4644 	new_fs_cssize = fragroundup(&sblock, new_fs_ncg * sizeof (struct csum));
4645 	newfrag_daddr = howmany(new_fs_cssize, sblock.fs_fsize) +
4646 	    sblock.fs_csaddr;
4647 	/*
4648 	 * add all of the frags that are required to grow the cyl summary to the
4649 	 * csfrag list, which is the generic/unknown list, since at this point
4650 	 * we don't yet know the state of those frags.
4651 	 */
4652 	for (daddr = oldfrag_daddr; daddr < newfrag_daddr; daddr++)
4653 		addcsfrag((ino_t)0, daddr, &csfrag);
4654 
4655 	/*
4656 	 * filter free fragments and allocate them. Note that the free frags
4657 	 * must be allocated first otherwise they could be grabbed by
4658 	 * alloccsfragino() for data frags.
4659 	 */
4660 	findcsfragfree();
4661 	alloccsfragfree();
4662 
4663 	/*
4664 	 * filter fragments owned by inodes and allocate them
4665 	 */
4666 	grow_fs_ncg = sblock.fs_ncg; /* findcsfragino() needs this glob. var. */
4667 	findcsfragino();
4668 	alloccsfragino();
4669 
4670 	if (notenoughspace()) {
4671 		/*
4672 		 * check how many consecutive fragments could be allocated
4673 		 * in both lists.
4674 		 */
4675 		int64_t tmp_frags;
4676 		for (daddr = oldfrag_daddr; daddr < newfrag_daddr;
4677 		    daddr += tmp_frags) {
4678 			if ((tmp_frags = checkfragallocated(daddr)) > 0)
4679 				growth_csum_frags += tmp_frags;
4680 			else
4681 				break;
4682 		}
4683 	} else {
4684 		/*
4685 		 * We have all we need for the new desired size,
4686 		 * so clean up and report back.
4687 		 */
4688 		return (fssize_db);
4689 	}
4690 
4691 	/*
4692 	 * given the number of fragments by which the csum block can be grown
4693 	 * compute by how many new fragments the FS can be increased.
4694 	 * It is the number of csum instances per fragment multiplied by
4695 	 * `growth_csum_frags' and the number of fragments per cylinder group.
4696 	 */
4697 	growth_fs_frags = howmany(sblock.fs_fsize, sizeof (struct csum)) *
4698 	    growth_csum_frags * sblock.fs_fpg;
4699 
4700 	/*
4701 	 * compute free fragments in the last cylinder group
4702 	 */
4703 	rdcg(sblock.fs_ncg - 1);
4704 	growth_fs_frags += sblock.fs_fpg - acg.cg_ndblk;
4705 
4706 	/*
4707 	 * compute how many csum instances are unused in the old csum block.
4708 	 * For each unused csum instance the FS can be grown by one cylinder
4709 	 * group without extending the csum block.
4710 	 */
4711 	spare_csum = howmany(sblock.fs_cssize, sizeof (struct csum)) -
4712 	    sblock.fs_ncg;
4713 	if (spare_csum > 0)
4714 		growth_fs_frags += spare_csum * sblock.fs_fpg;
4715 
4716 	/*
4717 	 * recalculate the new filesystem size in sectors, shorten it by
4718 	 * the requested size `fssize_db' if necessary.
4719 	 */
4720 	if (growth_fs_frags > 0) {
4721 		diskaddr_t sect;
4722 		sect = (sblock.fs_size + growth_fs_frags) * sblock.fs_nspf;
4723 		return ((sect > fssize_db) ? fssize_db : sect);
4724 	}
4725 
4726 	return (0);
4727 }
4728 
4729 void
4730 extendsummaryinfo()
4731 {
4732 	int64_t		i;
4733 	int		localtest	= test;
4734 	int64_t		frags;
4735 	daddr32_t		oldfrag;
4736 	daddr32_t		newfrag;
4737 
4738 	/*
4739 	 * if no-write (-N), don't bother
4740 	 */
4741 	if (Nflag)
4742 		return;
4743 
4744 again:
4745 	flcg();
4746 	/*
4747 	 * summary info did not change size -- do nothing unless in test mode
4748 	 */
4749 	if (grow_fs_cssize == sblock.fs_cssize)
4750 		if (!localtest)
4751 			return;
4752 
4753 	/*
4754 	 * build list of frags needed for additional summary information
4755 	 */
4756 	oldfrag = howmany(grow_fs_cssize, sblock.fs_fsize) + grow_fs_csaddr;
4757 	newfrag = howmany(sblock.fs_cssize, sblock.fs_fsize) + grow_fs_csaddr;
4758 	/*
4759 	 * add all of the frags that are required to grow the cyl summary to the
4760 	 * csfrag list, which is the generic/unknown list, since at this point
4761 	 * we don't yet know the state of those frags.
4762 	 */
4763 	for (i = oldfrag, frags = 0; i < newfrag; ++i, ++frags)
4764 		addcsfrag((ino_t)0, (diskaddr_t)i, &csfrag);
4765 	/*
4766 	 * reduce the number of data blocks in the file system (fs_dsize) by
4767 	 * the number of frags that need to be added to the cyl summary
4768 	 */
4769 	sblock.fs_dsize -= (newfrag - oldfrag);
4770 
4771 	/*
4772 	 * In test mode, we move more data than necessary from
4773 	 * cylinder group 0.  The lookup/allocate/move code can be
4774 	 * better stressed without having to create HUGE file systems.
4775 	 */
4776 	if (localtest)
4777 		for (i = newfrag; i < grow_sifrag; ++i) {
4778 			if (frags >= testfrags)
4779 				break;
4780 			frags++;
4781 			addcsfrag((ino_t)0, (diskaddr_t)i, &csfrag);
4782 		}
4783 
4784 	/*
4785 	 * move frags to free or inode lists, depending on owner
4786 	 */
4787 	findcsfragfree();
4788 	findcsfragino();
4789 
4790 	/*
4791 	 * if not all frags can be located, file system must be inconsistent
4792 	 */
4793 	if (csfrag) {
4794 		isbad = 1;	/* should already be set, but make sure */
4795 		lockexit(32);
4796 	}
4797 
4798 	/*
4799 	 * allocate the free frags. Note that the free frags must be allocated
4800 	 * first otherwise they could be grabbed by alloccsfragino() for data
4801 	 * frags.
4802 	 */
4803 	alloccsfragfree();
4804 	/*
4805 	 * allocate extra space for inode frags
4806 	 */
4807 	alloccsfragino();
4808 
4809 	/*
4810 	 * not enough space
4811 	 */
4812 	if (notenoughspace()) {
4813 		unalloccsfragfree();
4814 		unalloccsfragino();
4815 		if (localtest && !testforce) {
4816 			localtest = 0;
4817 			goto again;
4818 		}
4819 		(void) fprintf(stderr, gettext("Not enough free space\n"));
4820 		lockexit(NOTENOUGHSPACE);
4821 	}
4822 
4823 	/*
4824 	 * copy the data from old frags to new frags
4825 	 */
4826 	copycsfragino();
4827 
4828 	/*
4829 	 * fix the inodes to point to the new frags
4830 	 */
4831 	fixcsfragino();
4832 
4833 	/*
4834 	 * We may have moved more frags than we needed.  Free them.
4835 	 */
4836 	rdcg((long)0);
4837 	for (i = newfrag; i <= maxcsfrag; ++i)
4838 		setbit(cg_blksfree(&acg), i-cgbase(&sblock, 0));
4839 	wtcg();
4840 
4841 	flcg();
4842 }
4843 
4844 /*
4845  * Check if all fragments in the `csfragino' list were reallocated.
4846  */
4847 int
4848 notenoughspace()
4849 {
4850 	struct csfrag	*cfp;
4851 
4852 	/*
4853 	 * If any element in the csfragino array has a "new frag location"
4854 	 * of 0, the allocfrags() function was unsuccessful in allocating
4855 	 * space for moving the frag represented by this array element.
4856 	 */
4857 	for (cfp = csfragino; cfp; cfp = cfp->next)
4858 		if (cfp->nfrag == 0)
4859 			return (1);
4860 	return (0);
4861 }
4862 
4863 void
4864 unalloccsfragino()
4865 {
4866 	struct csfrag	*cfp;
4867 
4868 	while ((cfp = csfragino) != NULL) {
4869 		if (cfp->nfrag)
4870 			freefrags(cfp->nfrag, cfp->frags, cfp->cylno);
4871 		delcsfrag(cfp->ofrag, &csfragino);
4872 	}
4873 }
4874 
4875 void
4876 unalloccsfragfree()
4877 {
4878 	struct csfrag	*cfp;
4879 
4880 	while ((cfp = csfragfree) != NULL) {
4881 		freefrags(cfp->ofrag, cfp->frags, cfp->cylno);
4882 		delcsfrag(cfp->ofrag, &csfragfree);
4883 	}
4884 }
4885 
4886 /*
4887  * For each frag in the "as-yet-unclassified" list (csfrag), see if
4888  * it's free (i.e., its bit is set in the free frag bit map).  If so,
4889  * move it from the "as-yet-unclassified" list to the csfragfree list.
4890  */
4891 void
4892 findcsfragfree()
4893 {
4894 	struct csfrag	*cfp;
4895 	struct csfrag	*cfpnext;
4896 
4897 	/*
4898 	 * move free frags onto the free-frag list
4899 	 */
4900 	rdcg((long)0);
4901 	for (cfp = csfrag; cfp; cfp = cfpnext) {
4902 		cfpnext = cfp->next;
4903 		if (isset(cg_blksfree(&acg), cfp->ofrag - cgbase(&sblock, 0))) {
4904 			addcsfrag(cfp->ino, cfp->ofrag, &csfragfree);
4905 			delcsfrag(cfp->ofrag, &csfrag);
4906 		}
4907 	}
4908 }
4909 
4910 void
4911 copycsfragino()
4912 {
4913 	struct csfrag	*cfp;
4914 	char		buf[MAXBSIZE];
4915 
4916 	/*
4917 	 * copy data from old frags to newly allocated frags
4918 	 */
4919 	for (cfp = csfragino; cfp; cfp = cfp->next) {
4920 		rdfs(fsbtodb(&sblock, (uint64_t)cfp->ofrag), (int)cfp->size,
4921 		    buf);
4922 		wtfs(fsbtodb(&sblock, (uint64_t)cfp->nfrag), (int)cfp->size,
4923 		    buf);
4924 	}
4925 }
4926 
4927 long	curcylno	= -1;
4928 int	cylnodirty	= 0;
4929 
4930 void
4931 rdcg(long cylno)
4932 {
4933 	if (cylno != curcylno) {
4934 		flcg();
4935 		curcylno = cylno;
4936 		rdfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, curcylno)),
4937 		    (int)sblock.fs_cgsize, (char *)&acg);
4938 	}
4939 }
4940 
4941 void
4942 flcg()
4943 {
4944 	if (cylnodirty) {
4945 		if (debug && Pflag) {
4946 			(void) fprintf(stderr,
4947 			    "Assert: cylnodirty set in probe mode\n");
4948 			return;
4949 		}
4950 		resetallocinfo();
4951 		wtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, curcylno)),
4952 		    (int)sblock.fs_cgsize, (char *)&acg);
4953 		cylnodirty = 0;
4954 	}
4955 	curcylno = -1;
4956 }
4957 
4958 void
4959 wtcg()
4960 {
4961 	if (!Pflag) {
4962 		/* probe mode should never write to disk */
4963 		cylnodirty = 1;
4964 	}
4965 }
4966 
4967 void
4968 allocfrags(long frags, daddr32_t *fragp, long *cylnop)
4969 {
4970 	int	 i;
4971 	int	 j;
4972 	long	 bits;
4973 	long	 bit;
4974 
4975 	/*
4976 	 * Allocate a free-frag range in an old cylinder group
4977 	 */
4978 	for (i = 0, *fragp = 0; i < grow_fs_ncg; ++i) {
4979 		if (((fscs+i)->cs_nffree < frags) && ((fscs+i)->cs_nbfree == 0))
4980 			continue;
4981 		rdcg((long)i);
4982 		bit = bits = 0;
4983 		while (findfreerange(&bit, &bits)) {
4984 			if (frags <= bits)  {
4985 				for (j = 0; j < frags; ++j)
4986 					clrbit(cg_blksfree(&acg), bit+j);
4987 				wtcg();
4988 				*cylnop = i;
4989 				*fragp  = bit + cgbase(&sblock, i);
4990 				return;
4991 			}
4992 			bit += bits;
4993 		}
4994 	}
4995 }
4996 
4997 /*
4998  * Allocate space for frags that need to be moved in order to free up space for
4999  * expanding the cylinder summary info.
5000  * For each frag that needs to be moved (each frag or range of frags in
5001  * the csfragino list), allocate a new location and store the frag number
5002  * of that new location in the nfrag field of the csfrag struct.
5003  * If a new frag can't be allocated for any element in the csfragino list,
5004  * set the new frag number for that element to 0 and return immediately.
5005  * The notenoughspace() function will detect this condition.
5006  */
5007 void
5008 alloccsfragino()
5009 {
5010 	struct csfrag	*cfp;
5011 
5012 	/*
5013 	 * allocate space for inode frag ranges
5014 	 */
5015 	for (cfp = csfragino; cfp; cfp = cfp->next) {
5016 		allocfrags(cfp->frags, &cfp->nfrag, &cfp->cylno);
5017 		if (cfp->nfrag == 0)
5018 			break;
5019 	}
5020 }
5021 
5022 void
5023 alloccsfragfree()
5024 {
5025 	struct csfrag	*cfp;
5026 
5027 	/*
5028 	 * allocate the free frags needed for extended summary info
5029 	 */
5030 	rdcg((long)0);
5031 
5032 	for (cfp = csfragfree; cfp; cfp = cfp->next)
5033 		clrbit(cg_blksfree(&acg), cfp->ofrag - cgbase(&sblock, 0));
5034 
5035 	wtcg();
5036 }
5037 
5038 void
5039 freefrags(daddr32_t frag, long frags, long cylno)
5040 {
5041 	int	i;
5042 
5043 	/*
5044 	 * free frags
5045 	 */
5046 	rdcg(cylno);
5047 	for (i = 0; i < frags; ++i) {
5048 		setbit(cg_blksfree(&acg), (frag+i) - cgbase(&sblock, cylno));
5049 	}
5050 	wtcg();
5051 }
5052 
5053 int
5054 findfreerange(long *bitp, long *bitsp)
5055 {
5056 	long	 bit;
5057 
5058 	/*
5059 	 * find a range of free bits in a cylinder group bit map
5060 	 */
5061 	for (bit = *bitp, *bitsp = 0; bit < acg.cg_ndblk; ++bit)
5062 		if (isset(cg_blksfree(&acg), bit))
5063 			break;
5064 
5065 	if (bit >= acg.cg_ndblk)
5066 		return (0);
5067 
5068 	*bitp  = bit;
5069 	*bitsp = 1;
5070 	for (++bit; bit < acg.cg_ndblk; ++bit, ++(*bitsp)) {
5071 		if ((bit % sblock.fs_frag) == 0)
5072 			break;
5073 		if (isclr(cg_blksfree(&acg), bit))
5074 			break;
5075 	}
5076 	return (1);
5077 }
5078 
5079 void
5080 resetallocinfo()
5081 {
5082 	long	cno;
5083 	long	bit;
5084 	long	bits;
5085 
5086 	/*
5087 	 * Compute the free blocks/frags info and update the appropriate
5088 	 * inmemory superblock, summary info, and cylinder group fields
5089 	 */
5090 	sblock.fs_cstotal.cs_nffree -= acg.cg_cs.cs_nffree;
5091 	sblock.fs_cstotal.cs_nbfree -= acg.cg_cs.cs_nbfree;
5092 
5093 	acg.cg_cs.cs_nffree = 0;
5094 	acg.cg_cs.cs_nbfree = 0;
5095 
5096 	bzero((caddr_t)acg.cg_frsum, sizeof (acg.cg_frsum));
5097 	bzero((caddr_t)cg_blktot(&acg), (int)(acg.cg_iusedoff-acg.cg_btotoff));
5098 
5099 	bit = bits = 0;
5100 	while (findfreerange(&bit, &bits)) {
5101 		if (bits == sblock.fs_frag) {
5102 			acg.cg_cs.cs_nbfree++;
5103 			cno = cbtocylno(&sblock, bit);
5104 			cg_blktot(&acg)[cno]++;
5105 			cg_blks(&sblock, &acg, cno)[cbtorpos(&sblock, bit)]++;
5106 		} else {
5107 			acg.cg_cs.cs_nffree += bits;
5108 			acg.cg_frsum[bits]++;
5109 		}
5110 		bit += bits;
5111 	}
5112 
5113 	*(fscs + acg.cg_cgx) = acg.cg_cs;
5114 
5115 	sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
5116 	sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
5117 }
5118 
5119 void
5120 extendcg(long cylno)
5121 {
5122 	int	i;
5123 	diskaddr_t	dupper;
5124 	diskaddr_t	cbase;
5125 	diskaddr_t	dmax;
5126 
5127 	/*
5128 	 * extend the cylinder group at the end of the old file system
5129 	 * if it was partially allocated becase of lack of space
5130 	 */
5131 	flcg();
5132 	rdcg(cylno);
5133 
5134 	dupper = acg.cg_ndblk;
5135 	if (cylno == sblock.fs_ncg - 1)
5136 		acg.cg_ncyl = sblock.fs_ncyl - (sblock.fs_cpg * cylno);
5137 	else
5138 		acg.cg_ncyl = sblock.fs_cpg;
5139 	cbase = cgbase(&sblock, cylno);
5140 	dmax = cbase + sblock.fs_fpg;
5141 	if (dmax > sblock.fs_size)
5142 		dmax = sblock.fs_size;
5143 	acg.cg_ndblk = dmax - cbase;
5144 
5145 	for (i = dupper; i < acg.cg_ndblk; ++i)
5146 		setbit(cg_blksfree(&acg), i);
5147 
5148 	sblock.fs_dsize += (acg.cg_ndblk - dupper);
5149 
5150 	wtcg();
5151 	flcg();
5152 }
5153 
5154 struct lockfs	lockfs;
5155 int		lockfd;
5156 int		islocked;
5157 int		lockfskey;
5158 char		lockfscomment[128];
5159 
5160 void
5161 ulockfs()
5162 {
5163 	/*
5164 	 * if the file system was locked, unlock it before exiting
5165 	 */
5166 	if (islocked == 0)
5167 		return;
5168 
5169 	/*
5170 	 * first, check if the lock held
5171 	 */
5172 	lockfs.lf_flags = LOCKFS_MOD;
5173 	if (ioctl(lockfd, _FIOLFSS, &lockfs) == -1) {
5174 		perror(directory);
5175 		lockexit(32);
5176 	}
5177 
5178 	if (LOCKFS_IS_MOD(&lockfs)) {
5179 		(void) fprintf(stderr,
5180 		    gettext("FILE SYSTEM CHANGED DURING GROWFS!\n"));
5181 		(void) fprintf(stderr,
5182 		    gettext("   See lockfs(1), umount(1), and fsck(1)\n"));
5183 		lockexit(32);
5184 	}
5185 	/*
5186 	 * unlock the file system
5187 	 */
5188 	lockfs.lf_lock  = LOCKFS_ULOCK;
5189 	lockfs.lf_flags = 0;
5190 	lockfs.lf_key   = lockfskey;
5191 	clockfs();
5192 	if (ioctl(lockfd, _FIOLFS, &lockfs) == -1) {
5193 		perror(directory);
5194 		lockexit(32);
5195 	}
5196 }
5197 
5198 void
5199 wlockfs()
5200 {
5201 
5202 	/*
5203 	 * if no-write (-N), don't bother
5204 	 */
5205 	if (Nflag)
5206 		return;
5207 	/*
5208 	 * open the mountpoint, and write lock the file system
5209 	 */
5210 	if ((lockfd = open64(directory, O_RDONLY)) == -1) {
5211 		perror(directory);
5212 		lockexit(32);
5213 	}
5214 
5215 	/*
5216 	 * check if it is already locked
5217 	 */
5218 	if (ioctl(lockfd, _FIOLFSS, &lockfs) == -1) {
5219 		perror(directory);
5220 		lockexit(32);
5221 	}
5222 
5223 	if (lockfs.lf_lock != LOCKFS_WLOCK) {
5224 		lockfs.lf_lock  = LOCKFS_WLOCK;
5225 		lockfs.lf_flags = 0;
5226 		lockfs.lf_key   = 0;
5227 		clockfs();
5228 		if (ioctl(lockfd, _FIOLFS, &lockfs) == -1) {
5229 			perror(directory);
5230 			lockexit(32);
5231 		}
5232 	}
5233 	islocked = 1;
5234 	lockfskey = lockfs.lf_key;
5235 }
5236 
5237 void
5238 clockfs()
5239 {
5240 	time_t	t;
5241 	char	*ct;
5242 
5243 	(void) time(&t);
5244 	ct = ctime(&t);
5245 	ct[strlen(ct)-1] = '\0';
5246 
5247 	(void) sprintf(lockfscomment, "%s -- mkfs pid %d", ct, getpid());
5248 	lockfs.lf_comlen  = strlen(lockfscomment)+1;
5249 	lockfs.lf_comment = lockfscomment;
5250 }
5251 
5252 /*
5253  * Write the csum records and the superblock
5254  */
5255 void
5256 wtsb()
5257 {
5258 	long	i;
5259 
5260 	/*
5261 	 * write summary information
5262 	 */
5263 	for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize)
5264 		wtfs(fsbtodb(&sblock, (uint64_t)(sblock.fs_csaddr +
5265 		    numfrags(&sblock, i))),
5266 		    (int)(sblock.fs_cssize - i < sblock.fs_bsize ?
5267 		    sblock.fs_cssize - i : sblock.fs_bsize),
5268 		    ((char *)fscs) + i);
5269 
5270 	/*
5271 	 * write superblock
5272 	 */
5273 	sblock.fs_time = mkfstime;
5274 	wtfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
5275 }
5276 
5277 /*
5278  * Verify that the optimization selection is reasonable, and advance
5279  * the global "string" appropriately.
5280  */
5281 static char
5282 checkopt(char *optim)
5283 {
5284 	char	opt;
5285 	int	limit = strcspn(optim, ",");
5286 
5287 	switch (limit) {
5288 	case 0:	/* missing indicator (have comma or nul) */
5289 		(void) fprintf(stderr, gettext(
5290 		    "mkfs: missing optimization flag reset to `t' (time)\n"));
5291 		opt = 't';
5292 		break;
5293 
5294 	case 1: /* single-character indicator */
5295 		opt = *optim;
5296 		if ((opt != 's') && (opt != 't')) {
5297 			(void) fprintf(stderr, gettext(
5298 		    "mkfs: bad optimization value `%c' reset to `t' (time)\n"),
5299 			    opt);
5300 			opt = 't';
5301 		}
5302 		break;
5303 
5304 	default: /* multi-character indicator */
5305 		(void) fprintf(stderr, gettext(
5306 	    "mkfs: bad optimization value `%*.*s' reset to `t' (time)\n"),
5307 		    limit, limit, optim);
5308 		opt = 't';
5309 		break;
5310 	}
5311 
5312 	string += limit;
5313 
5314 	return (opt);
5315 }
5316 
5317 /*
5318  * Verify that the mtb selection is reasonable, and advance
5319  * the global "string" appropriately.
5320  */
5321 static char
5322 checkmtb(char *mtbarg)
5323 {
5324 	char	mtbc;
5325 	int	limit = strcspn(mtbarg, ",");
5326 
5327 	switch (limit) {
5328 	case 0:	/* missing indicator (have comma or nul) */
5329 		(void) fprintf(stderr, gettext(
5330 		    "mkfs: missing mtb flag reset to `n' (no mtb support)\n"));
5331 		mtbc = 'n';
5332 		break;
5333 
5334 	case 1: /* single-character indicator */
5335 		mtbc = tolower(*mtbarg);
5336 		if ((mtbc != 'y') && (mtbc != 'n')) {
5337 			(void) fprintf(stderr, gettext(
5338 		    "mkfs: bad mtb value `%c' reset to `n' (no mtb support)\n"),
5339 			    mtbc);
5340 			mtbc = 'n';
5341 		}
5342 		break;
5343 
5344 	default: /* multi-character indicator */
5345 		(void) fprintf(stderr, gettext(
5346 	    "mkfs: bad mtb value `%*.*s' reset to `n' (no mtb support)\n"),
5347 		    limit, limit, mtbarg);
5348 		opt = 'n';
5349 		break;
5350 	}
5351 
5352 	string += limit;
5353 
5354 	return (mtbc);
5355 }
5356 
5357 /*
5358  * Verify that a value is in a range.  If it is not, resets it to
5359  * its default value if one is supplied, exits otherwise.
5360  *
5361  * When testing, can compare user_supplied to RC_KEYWORD or RC_POSITIONAL.
5362  */
5363 static void
5364 range_check(long *varp, char *name, long minimum, long maximum,
5365     long def_val, int user_supplied)
5366 {
5367 	dprintf(("DeBuG %s : %ld (%ld %ld %ld)\n",
5368 	    name, *varp, minimum, maximum, def_val));
5369 
5370 	if ((*varp < minimum) || (*varp > maximum)) {
5371 		if (user_supplied != RC_DEFAULT) {
5372 			(void) fprintf(stderr, gettext(
5373 	    "mkfs: bad value for %s: %ld must be between %ld and %ld\n"),
5374 			    name, *varp, minimum, maximum);
5375 		}
5376 		if (def_val != NO_DEFAULT) {
5377 			if (user_supplied) {
5378 				(void) fprintf(stderr,
5379 				    gettext("mkfs: %s reset to default %ld\n"),
5380 				    name, def_val);
5381 			}
5382 			*varp = def_val;
5383 			dprintf(("DeBuG %s : %ld\n", name, *varp));
5384 			return;
5385 		}
5386 		lockexit(2);
5387 		/*NOTREACHED*/
5388 	}
5389 }
5390 
5391 /*
5392  * Verify that a value is in a range.  If it is not, resets it to
5393  * its default value if one is supplied, exits otherwise.
5394  *
5395  * When testing, can compare user_supplied to RC_KEYWORD or RC_POSITIONAL.
5396  */
5397 static void
5398 range_check_64(uint64_t *varp, char *name, uint64_t minimum, uint64_t maximum,
5399     uint64_t def_val, int user_supplied)
5400 {
5401 	if ((*varp < minimum) || (*varp > maximum)) {
5402 		if (user_supplied != RC_DEFAULT) {
5403 			(void) fprintf(stderr, gettext(
5404 	    "mkfs: bad value for %s: %lld must be between %lld and %lld\n"),
5405 			    name, *varp, minimum, maximum);
5406 		}
5407 		if (def_val != NO_DEFAULT) {
5408 			if (user_supplied) {
5409 				(void) fprintf(stderr,
5410 				    gettext("mkfs: %s reset to default %lld\n"),
5411 				    name, def_val);
5412 			}
5413 			*varp = def_val;
5414 			return;
5415 		}
5416 		lockexit(2);
5417 		/*NOTREACHED*/
5418 	}
5419 }
5420 
5421 /*
5422  * Blocks SIGINT from delivery.  Returns the previous mask in the
5423  * buffer provided, so that mask may be later restored.
5424  */
5425 static void
5426 block_sigint(sigset_t *old_mask)
5427 {
5428 	sigset_t block_mask;
5429 
5430 	if (sigemptyset(&block_mask) < 0) {
5431 		fprintf(stderr, gettext("Could not clear signal mask\n"));
5432 		lockexit(3);
5433 	}
5434 	if (sigaddset(&block_mask, SIGINT) < 0) {
5435 		fprintf(stderr, gettext("Could not set signal mask\n"));
5436 		lockexit(3);
5437 	}
5438 	if (sigprocmask(SIG_BLOCK, &block_mask, old_mask) < 0) {
5439 		fprintf(stderr, gettext("Could not block SIGINT\n"));
5440 		lockexit(3);
5441 	}
5442 }
5443 
5444 /*
5445  * Restores the signal mask that was in force before a call
5446  * to block_sigint().  This may actually still have SIGINT blocked,
5447  * if we've been recursively invoked.
5448  */
5449 static void
5450 unblock_sigint(sigset_t *old_mask)
5451 {
5452 	if (sigprocmask(SIG_UNBLOCK, old_mask, (sigset_t *)NULL) < 0) {
5453 		fprintf(stderr, gettext("Could not restore signal mask\n"));
5454 		lockexit(3);
5455 	}
5456 }
5457 
5458 /*
5459  * Attempt to be somewhat graceful about being interrupted, rather than
5460  * just silently leaving the filesystem in an unusable state.
5461  *
5462  * The kernel has blocked SIGINT upon entry, so we don't have to worry
5463  * about recursion if the user starts pounding on the keyboard.
5464  */
5465 static void
5466 recover_from_sigint(int signum)
5467 {
5468 	if (fso > -1) {
5469 		if ((Nflag != 0) || confirm_abort()) {
5470 			lockexit(4);
5471 		}
5472 	}
5473 }
5474 
5475 static int
5476 confirm_abort(void)
5477 {
5478 	char line[80];
5479 
5480 	printf(gettext("\n\nAborting at this point will leave the filesystem "
5481 	    "in an inconsistent\nstate.  If you do choose to stop, "
5482 	    "you will be given instructions on how to\nrecover "
5483 	    "the filesystem.  Do you wish to cancel the filesystem "
5484 	    "grow\noperation (y/n)?"));
5485 	if (getline(stdin, line, sizeof (line)) == EOF)
5486 		line[0] = 'y';
5487 
5488 	printf("\n");
5489 	if (line[0] == 'y' || line[0] == 'Y')
5490 		return (1);
5491 	else {
5492 		return (0);
5493 	}
5494 }
5495 
5496 static int
5497 getline(FILE *fp, char *loc, int maxlen)
5498 {
5499 	int n;
5500 	char *p, *lastloc;
5501 
5502 	p = loc;
5503 	lastloc = &p[maxlen-1];
5504 	while ((n = getc(fp)) != '\n') {
5505 		if (n == EOF)
5506 			return (EOF);
5507 		if (!isspace(n) && p < lastloc)
5508 			*p++ = n;
5509 	}
5510 	*p = 0;
5511 	return (p - loc);
5512 }
5513 
5514 /*
5515  * Calculate the maximum value of cylinders-per-group for a file
5516  * system with the characteristics:
5517  *
5518  *	bsize - file system block size
5519  *	fragsize - frag size
5520  *	nbpi - number of bytes of disk space per inode
5521  *	nrpos - number of rotational positions
5522  *	spc - sectors per cylinder
5523  *
5524  * These five characteristic are not adjustable (by this function).
5525  * The only attribute of the file system which IS adjusted by this
5526  * function in order to maximize cylinders-per-group is the proportion
5527  * of the cylinder group overhead block used for the inode map.  The
5528  * inode map cannot occupy more than one-third of the cylinder group
5529  * overhead block, but it's OK for it to occupy less than one-third
5530  * of the overhead block.
5531  *
5532  * The setting of nbpi determines one possible value for the maximum
5533  * size of a cylinder group.  It does so because it determines the total
5534  * number of inodes in the file system (file system size is fixed, and
5535  * nbpi is fixed, so the total number of inodes is fixed too).  The
5536  * cylinder group has to be small enough so that the number of inodes
5537  * in the cylinder group is less than or equal to the number of bits
5538  * in one-third (or whatever proportion is assumed) of a file system
5539  * block.  The details of the calculation are:
5540  *
5541  *     The macro MAXIpG_B(bsize, inode_divisor) determines the maximum
5542  *     number of inodes that can be in a cylinder group, given the
5543  *     proportion of the cylinder group overhead block used for the
5544  *     inode bitmaps (an inode_divisor of 3 means that 1/3 of the
5545  *     block is used for inode bitmaps; an inode_divisor of 12 means
5546  *     that 1/12 of the block is used for inode bitmaps.)
5547  *
5548  *     Once the number of inodes per cylinder group is known, the
5549  *     maximum value of cylinders-per-group (determined by nbpi)
5550  *     is calculated by the formula
5551  *
5552  *     maxcpg_given_nbpi = (size of a cylinder group)/(size of a cylinder)
5553  *
5554  *			 = (inodes-per-cg * nbpi)/(spc * DEV_BSIZE)
5555  *
5556  *     (Interestingly, the size of the file system never enters
5557  *     into this calculation.)
5558  *
5559  * Another possible value for the maximum cylinder group size is determined
5560  * by frag_size and nrpos.  The frags in the cylinder group must be
5561  * representable in the frag bitmaps in the cylinder overhead block and the
5562  * rotational positions for each cylinder must be represented in the
5563  * rotational position tables.  The calculation of the maximum cpg
5564  * value, given the frag and nrpos vales, is:
5565  *
5566  *     maxcpg_given_fragsize =
5567  *	  (available space in the overhead block) / (size of per-cylinder data)
5568  *
5569  *     The available space in the overhead block =
5570  *	  bsize - sizeof (struct cg) - space_used_for_inode_bitmaps
5571  *
5572  *     The size of the per-cylinder data is:
5573  *	    sizeof(long)            # for the "blocks avail per cylinder" field
5574  *	    + nrpos * sizeof(short)   # for the rotational position table entry
5575  *	    + frags-per-cylinder/NBBY # number of bytes to represent this
5576  *				      # cylinder in the frag bitmap
5577  *
5578  * The two calculated maximum values of cylinder-per-group will typically
5579  * turn out to be different, since they are derived from two different
5580  * constraints.  Usually, maxcpg_given_nbpi is much bigger than
5581  * maxcpg_given_fragsize.  But they can be brought together by
5582  * adjusting the proportion of the overhead block dedicated to
5583  * the inode bitmaps.  Decreasing the proportion of the cylinder
5584  * group overhead block used for inode maps will decrease
5585  * maxcpg_given_nbpi and increase maxcpg_given_fragsize.
5586  *
5587  * This function calculates the initial values of maxcpg_given_nbpi
5588  * and maxcpg_given_fragsize assuming that 1/3 of the cg overhead
5589  * block is used for inode bitmaps.  Then it decreases the proportion
5590  * of the cg overhead block used for inode bitmaps (by increasing
5591  * the value of inode_divisor) until maxcpg_given_nbpi and
5592  * maxcpg_given_fragsize are the same, or stop changing, or
5593  * maxcpg_given_nbpi is less than maxcpg_given_fragsize.
5594  *
5595  * The loop terminates when any of the following occur:
5596  *	* maxcpg_given_fragsize is greater than or equal to
5597  *	  maxcpg_given_nbpi
5598  *	* neither maxcpg_given_fragsize nor maxcpg_given_nbpi
5599  *	  change in the expected direction
5600  *
5601  * The loop is guaranteed to terminate because it only continues
5602  * while maxcpg_given_fragsize and maxcpg_given_nbpi are approaching
5603  * each other.  As soon they cross each other, or neither one changes
5604  * in the direction of the other, or one of them moves in the wrong
5605  * direction, the loop completes.
5606  */
5607 
5608 static long
5609 compute_maxcpg(long bsize, long fragsize, long nbpi, long nrpos, long spc)
5610 {
5611 	int	maxcpg_given_nbpi;	/* in cylinders */
5612 	int	maxcpg_given_fragsize;	/* in cylinders */
5613 	int	spf;			/* sectors per frag */
5614 	int	inode_divisor;
5615 	int	old_max_given_frag = 0;
5616 	int	old_max_given_nbpi = INT_MAX;
5617 
5618 	spf = fragsize / DEV_BSIZE;
5619 	inode_divisor = 3;
5620 
5621 	while (1) {
5622 		maxcpg_given_nbpi =
5623 		    (((int64_t)(MAXIpG_B(bsize, inode_divisor))) * nbpi) /
5624 		    (DEV_BSIZE * ((int64_t)spc));
5625 		maxcpg_given_fragsize =
5626 		    (bsize - (sizeof (struct cg)) - (bsize / inode_divisor)) /
5627 		    (sizeof (long) + nrpos * sizeof (short) +
5628 		    (spc / spf) / NBBY);
5629 
5630 		if (maxcpg_given_fragsize >= maxcpg_given_nbpi)
5631 			return (maxcpg_given_nbpi);
5632 
5633 		/*
5634 		 * If neither value moves toward the other, return the
5635 		 * least of the old values (we use the old instead of the
5636 		 * new because: if the old is the same as the new, it
5637 		 * doesn't matter which ones we use.  If one of the
5638 		 * values changed, but in the wrong direction, the
5639 		 * new values are suspect.  Better use the old.  This
5640 		 * shouldn't happen, but it's best to check.
5641 		 */
5642 
5643 		if (!(maxcpg_given_nbpi < old_max_given_nbpi) &&
5644 		    !(maxcpg_given_fragsize > old_max_given_frag))
5645 			return (MIN(old_max_given_nbpi, old_max_given_frag));
5646 
5647 		/*
5648 		 * This is probably impossible, but if one of the maxcpg
5649 		 * values moved in the "right" direction and one moved
5650 		 * in the "wrong" direction (that is, the two values moved
5651 		 * in the same direction), the previous conditional won't
5652 		 * recognize that the values aren't converging (since at
5653 		 * least one value moved in the "right" direction, the
5654 		 * last conditional says "keep going").
5655 		 *
5656 		 * Just to make absolutely certain that the loop terminates,
5657 		 * check for one of the values moving in the "wrong" direction
5658 		 * and terminate the loop if it happens.
5659 		 */
5660 
5661 		if (maxcpg_given_nbpi > old_max_given_nbpi ||
5662 		    maxcpg_given_fragsize < old_max_given_frag)
5663 			return (MIN(old_max_given_nbpi, old_max_given_frag));
5664 
5665 		old_max_given_nbpi = maxcpg_given_nbpi;
5666 		old_max_given_frag = maxcpg_given_fragsize;
5667 
5668 		inode_divisor++;
5669 	}
5670 }
5671 
5672 static int
5673 in_64bit_mode(void)
5674 {
5675 	/*  cmd must be an absolute path, for security */
5676 	char *cmd = "/usr/bin/isainfo -b";
5677 	char buf[BUFSIZ];
5678 	FILE *ptr;
5679 	int retval = 0;
5680 
5681 	putenv("IFS= \t");
5682 	if ((ptr = popen(cmd, "r")) != NULL) {
5683 		if (fgets(buf, BUFSIZ, ptr) != NULL &&
5684 		    strncmp(buf, "64", 2) == 0)
5685 			retval = 1;
5686 		(void) pclose(ptr);
5687 	}
5688 	return (retval);
5689 }
5690 
5691 /*
5692  * validate_size
5693  *
5694  * Return 1 if the device appears to be at least "size" sectors long.
5695  * Return 0 if it's shorter or we can't read it.
5696  */
5697 
5698 static int
5699 validate_size(int fd, diskaddr_t size)
5700 {
5701 	char 		buf[DEV_BSIZE];
5702 	int rc;
5703 
5704 	if ((llseek(fd, (offset_t)((size - 1) * DEV_BSIZE), SEEK_SET) == -1) ||
5705 	    (read(fd, buf, DEV_BSIZE)) != DEV_BSIZE)
5706 		rc = 0;
5707 	else
5708 		rc = 1;
5709 	return (rc);
5710 }
5711 
5712 /*
5713  * Print every field of the calculated superblock, along with
5714  * its value.  To make parsing easier on the caller, the value
5715  * is printed first, then the name.  Additionally, there's only
5716  * one name/value pair per line.  All values are reported in
5717  * hexadecimal (with the traditional 0x prefix), as that's slightly
5718  * easier for humans to read.  Not that they're expected to, but
5719  * debugging happens.
5720  */
5721 static void
5722 dump_sblock(void)
5723 {
5724 	int row, column, pending, written;
5725 	caddr_t source;
5726 
5727 	if (Rflag) {
5728 		pending = sizeof (sblock);
5729 		source = (caddr_t)&sblock;
5730 		do {
5731 			written = write(fileno(stdout), source, pending);
5732 			pending -= written;
5733 			source += written;
5734 		} while ((pending > 0) && (written > 0));
5735 
5736 		if (written < 0) {
5737 			perror(gettext("Binary dump of superblock failed"));
5738 			lockexit(1);
5739 		}
5740 		return;
5741 	} else {
5742 		printf("0x%x sblock.fs_link\n", sblock.fs_link);
5743 		printf("0x%x sblock.fs_rolled\n", sblock.fs_rolled);
5744 		printf("0x%x sblock.fs_sblkno\n", sblock.fs_sblkno);
5745 		printf("0x%x sblock.fs_cblkno\n", sblock.fs_cblkno);
5746 		printf("0x%x sblock.fs_iblkno\n", sblock.fs_iblkno);
5747 		printf("0x%x sblock.fs_dblkno\n", sblock.fs_dblkno);
5748 		printf("0x%x sblock.fs_cgoffset\n", sblock.fs_cgoffset);
5749 		printf("0x%x sblock.fs_cgmask\n", sblock.fs_cgmask);
5750 		printf("0x%x sblock.fs_time\n", sblock.fs_time);
5751 		printf("0x%x sblock.fs_size\n", sblock.fs_size);
5752 		printf("0x%x sblock.fs_dsize\n", sblock.fs_dsize);
5753 		printf("0x%x sblock.fs_ncg\n", sblock.fs_ncg);
5754 		printf("0x%x sblock.fs_bsize\n", sblock.fs_bsize);
5755 		printf("0x%x sblock.fs_fsize\n", sblock.fs_fsize);
5756 		printf("0x%x sblock.fs_frag\n", sblock.fs_frag);
5757 		printf("0x%x sblock.fs_minfree\n", sblock.fs_minfree);
5758 		printf("0x%x sblock.fs_rotdelay\n", sblock.fs_rotdelay);
5759 		printf("0x%x sblock.fs_rps\n", sblock.fs_rps);
5760 		printf("0x%x sblock.fs_bmask\n", sblock.fs_bmask);
5761 		printf("0x%x sblock.fs_fmask\n", sblock.fs_fmask);
5762 		printf("0x%x sblock.fs_bshift\n", sblock.fs_bshift);
5763 		printf("0x%x sblock.fs_fshift\n", sblock.fs_fshift);
5764 		printf("0x%x sblock.fs_maxcontig\n", sblock.fs_maxcontig);
5765 		printf("0x%x sblock.fs_maxbpg\n", sblock.fs_maxbpg);
5766 		printf("0x%x sblock.fs_fragshift\n", sblock.fs_fragshift);
5767 		printf("0x%x sblock.fs_fsbtodb\n", sblock.fs_fsbtodb);
5768 		printf("0x%x sblock.fs_sbsize\n", sblock.fs_sbsize);
5769 		printf("0x%x sblock.fs_csmask\n", sblock.fs_csmask);
5770 		printf("0x%x sblock.fs_csshift\n", sblock.fs_csshift);
5771 		printf("0x%x sblock.fs_nindir\n", sblock.fs_nindir);
5772 		printf("0x%x sblock.fs_inopb\n", sblock.fs_inopb);
5773 		printf("0x%x sblock.fs_nspf\n", sblock.fs_nspf);
5774 		printf("0x%x sblock.fs_optim\n", sblock.fs_optim);
5775 #ifdef _LITTLE_ENDIAN
5776 		printf("0x%x sblock.fs_state\n", sblock.fs_state);
5777 #else
5778 		printf("0x%x sblock.fs_npsect\n", sblock.fs_npsect);
5779 #endif
5780 		printf("0x%x sblock.fs_si\n", sblock.fs_si);
5781 		printf("0x%x sblock.fs_trackskew\n", sblock.fs_trackskew);
5782 		printf("0x%x sblock.fs_id[0]\n", sblock.fs_id[0]);
5783 		printf("0x%x sblock.fs_id[1]\n", sblock.fs_id[1]);
5784 		printf("0x%x sblock.fs_csaddr\n", sblock.fs_csaddr);
5785 		printf("0x%x sblock.fs_cssize\n", sblock.fs_cssize);
5786 		printf("0x%x sblock.fs_cgsize\n", sblock.fs_cgsize);
5787 		printf("0x%x sblock.fs_ntrak\n", sblock.fs_ntrak);
5788 		printf("0x%x sblock.fs_nsect\n", sblock.fs_nsect);
5789 		printf("0x%x sblock.fs_spc\n", sblock.fs_spc);
5790 		printf("0x%x sblock.fs_ncyl\n", sblock.fs_ncyl);
5791 		printf("0x%x sblock.fs_cpg\n", sblock.fs_cpg);
5792 		printf("0x%x sblock.fs_ipg\n", sblock.fs_ipg);
5793 		printf("0x%x sblock.fs_fpg\n", sblock.fs_fpg);
5794 		printf("0x%x sblock.fs_cstotal\n", sblock.fs_cstotal);
5795 		printf("0x%x sblock.fs_fmod\n", sblock.fs_fmod);
5796 		printf("0x%x sblock.fs_clean\n", sblock.fs_clean);
5797 		printf("0x%x sblock.fs_ronly\n", sblock.fs_ronly);
5798 		printf("0x%x sblock.fs_flags\n", sblock.fs_flags);
5799 		printf("0x%x sblock.fs_fsmnt\n", sblock.fs_fsmnt);
5800 		printf("0x%x sblock.fs_cgrotor\n", sblock.fs_cgrotor);
5801 		printf("0x%x sblock.fs_u.fs_csp\n", sblock.fs_u.fs_csp);
5802 		printf("0x%x sblock.fs_cpc\n", sblock.fs_cpc);
5803 
5804 		/*
5805 		 * No macros are defined for the dimensions of the
5806 		 * opostbl array.
5807 		 */
5808 		for (row = 0; row < 16; row++) {
5809 			for (column = 0; column < 8; column++) {
5810 				printf("0x%x sblock.fs_opostbl[%d][%d]\n",
5811 				    sblock.fs_opostbl[row][column],
5812 				    row, column);
5813 			}
5814 		}
5815 
5816 		/*
5817 		 * Ditto the size of sparecon.
5818 		 */
5819 		for (row = 0; row < 51; row++) {
5820 			printf("0x%x sblock.fs_sparecon[%d]\n",
5821 			    sblock.fs_sparecon[row], row);
5822 		}
5823 
5824 		printf("0x%x sblock.fs_version\n", sblock.fs_version);
5825 		printf("0x%x sblock.fs_logbno\n", sblock.fs_logbno);
5826 		printf("0x%x sblock.fs_reclaim\n", sblock.fs_reclaim);
5827 		printf("0x%x sblock.fs_sparecon2\n", sblock.fs_sparecon2);
5828 #ifdef _LITTLE_ENDIAN
5829 		printf("0x%x sblock.fs_npsect\n", sblock.fs_npsect);
5830 #else
5831 		printf("0x%x sblock.fs_state\n", sblock.fs_state);
5832 #endif
5833 		printf("0x%llx sblock.fs_qbmask\n", sblock.fs_qbmask);
5834 		printf("0x%llx sblock.fs_qfmask\n", sblock.fs_qfmask);
5835 		printf("0x%x sblock.fs_postblformat\n", sblock.fs_postblformat);
5836 		printf("0x%x sblock.fs_nrpos\n", sblock.fs_nrpos);
5837 		printf("0x%x sblock.fs_postbloff\n", sblock.fs_postbloff);
5838 		printf("0x%x sblock.fs_rotbloff\n", sblock.fs_rotbloff);
5839 		printf("0x%x sblock.fs_magic\n", sblock.fs_magic);
5840 
5841 		/*
5842 		 * fs_space isn't of much use in this context, so we'll
5843 		 * just ignore it for now.
5844 		 */
5845 	}
5846 }
5847