xref: /titanic_44/usr/src/cmd/fs.d/ufs/mkfs/mkfs.c (revision 663d253a1881379c900f0d0cff7a264dc34ca555)
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 2007 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 #pragma ident	"%Z%%M%	%I%	%E% SMI"
40 
41 
42 /*
43  * The maximum supported file system size (in sectors) is the
44  * number of frags that can be represented in an int32_t field
45  * (INT_MAX) times the maximum number of sectors per frag.  Since
46  * the maximum frag size is MAXBSIZE, the maximum number of sectors
47  * per frag is MAXBSIZE/DEV_BSIZE.
48  */
49 #define	FS_MAX	(((diskaddr_t)INT_MAX) * (MAXBSIZE/DEV_BSIZE))
50 
51 /*
52  * make file system for cylinder-group style file systems
53  *
54  * usage:
55  *
56  *    mkfs [-F FSType] [-V] [-G [-P]] [-M dirname] [-m] [options]
57  *	[-o specific_options]  special size
58  *	[nsect ntrack bsize fsize cpg	minfree	rps nbpi opt apc rotdelay
59  *	  2     3      4     5     6	7	8   9	 10  11  12
60  *	nrpos maxcontig mtb]
61  *	13    14	15
62  *
63  *  where specific_options are:
64  *	N - no create
65  *	nsect - The number of sectors per track
66  *	ntrack - The number of tracks per cylinder
67  *	bsize - block size
68  *	fragsize - fragment size
69  *	cgsize - The number of disk cylinders per cylinder group.
70  * 	free - minimum free space
71  *	rps - rotational speed (rev/sec).
72  *	nbpi - number of data bytes per allocated inode
73  *	opt - optimization (space, time)
74  *	apc - number of alternates
75  *	gap - gap size
76  *	nrpos - number of rotational positions
77  *	maxcontig - maximum number of logical blocks that will be
78  *		allocated contiguously before inserting rotational delay
79  *	mtb - if "y", set up file system for eventual growth to over a
80  *		a terabyte
81  * -P Do not grow the file system, but print on stdout the maximal
82  *    size in sectors to which the file system can be increased. The calculated
83  *    size is limited by the value provided by the operand size.
84  *
85  * Note that -P is a project-private interface and together with -G intended
86  * to be used only by the growfs script. It is therefore purposely not
87  * documented in the man page.
88  * The -P option is covered by PSARC case 2003/422.
89  */
90 
91 /*
92  * The following constants set the defaults used for the number
93  * of sectors/track (fs_nsect), and number of tracks/cyl (fs_ntrak).
94  *
95  *			NSECT		NTRAK
96  *	72MB CDC	18		9
97  *	30MB CDC	18		5
98  *	720KB Diskette	9		2
99  *
100  * However the defaults will be different for disks larger than CHSLIMIT.
101  */
102 
103 #define	DFLNSECT	32
104 #define	DFLNTRAK	16
105 
106 /*
107  * The following default sectors and tracks values are used for
108  * non-efi disks that are larger than the CHS addressing limit. The
109  * existing default cpg of 16 (DESCPG) holds good for larger disks too.
110  */
111 #define	DEF_SECTORS_EFI	128
112 #define	DEF_TRACKS_EFI	48
113 
114 /*
115  * The maximum number of cylinders in a group depends upon how much
116  * information can be stored on a single cylinder. The default is to
117  * use 16 cylinders per group.  This is effectively tradition - it was
118  * the largest value acceptable under SunOs 4.1
119  */
120 #define	DESCPG		16	/* desired fs_cpg */
121 
122 /*
123  * The following two constants set the default block and fragment sizes.
124  * Both constants must be a power of 2 and meet the following constraints:
125  *	MINBSIZE <= DESBLKSIZE <= MAXBSIZE
126  *	DEV_BSIZE <= DESFRAGSIZE <= DESBLKSIZE
127  *	DESBLKSIZE / DESFRAGSIZE <= 8
128  */
129 #define	DESBLKSIZE	8192
130 #define	DESFRAGSIZE	1024
131 
132 /*
133  * MINFREE gives the minimum acceptable percentage of file system
134  * blocks which may be free. If the freelist drops below this level
135  * only the superuser may continue to allocate blocks. This may
136  * be set to 0 if no reserve of free blocks is deemed necessary,
137  * however throughput drops by fifty percent if the file system
138  * is run at between 90% and 100% full; thus the default value of
139  * fs_minfree is 10%. With 10% free space, fragmentation is not a
140  * problem, so we choose to optimize for time.
141  */
142 #define	MINFREE		10
143 #define	DEFAULTOPT	FS_OPTTIME
144 
145 /*
146  * ROTDELAY gives the minimum number of milliseconds to initiate
147  * another disk transfer on the same cylinder. It is no longer used
148  * and will always default to 0.
149  */
150 #define	ROTDELAY	0
151 
152 /*
153  * MAXBLKPG determines the maximum number of data blocks which are
154  * placed in a single cylinder group. The default is one indirect
155  * block worth of data blocks.
156  */
157 #define	MAXBLKPG(bsize)	((bsize) / sizeof (daddr32_t))
158 
159 /*
160  * Each file system has a number of inodes statically allocated.
161  * We allocate one inode slot per NBPI bytes, expecting this
162  * to be far more than we will ever need.
163  */
164 #define	NBPI		2048	/* Number Bytes Per Inode */
165 #define	MTB_NBPI	(MB)	/* Number Bytes Per Inode for multi-terabyte */
166 
167 /*
168  * Disks are assumed to rotate at 60HZ, unless otherwise specified.
169  */
170 #define	DEFHZ		60
171 
172 /*
173  * Cylinder group related limits.
174  *
175  * For each cylinder we keep track of the availability of blocks at different
176  * rotational positions, so that we can lay out the data to be picked
177  * up with minimum rotational latency.  NRPOS is the number of rotational
178  * positions which we distinguish.  With NRPOS 8 the resolution of our
179  * summary information is 2ms for a typical 3600 rpm drive.
180  */
181 #define	NRPOS		8	/* number distinct rotational positions */
182 
183 #ifdef DEBUG
184 #define	dprintf(x)	printf x
185 #else
186 #define	dprintf(x)
187 #endif
188 
189 /*
190  * For the -N option, when calculating the backup superblocks, do not print
191  * them if we are not really sure. We may have to try an alternate method of
192  * arriving at the superblocks. So defer printing till a handful of superblocks
193  * look good.
194  */
195 #define	tprintf(x)	if (Nflag && retry) \
196 				(void) strncat(tmpbuf, x, strlen(x)); \
197 			else \
198 				(void) fprintf(stderr, x);
199 
200 #define	ALTSB		32	/* Location of first backup superblock */
201 
202 /*
203  * range_check "user_supplied" flag values.
204  */
205 #define	RC_DEFAULT	0
206 #define	RC_KEYWORD	1
207 #define	RC_POSITIONAL	2
208 
209 /*
210  * ufs hole
211  */
212 #define	UFS_HOLE	-1
213 
214 #ifndef	STANDALONE
215 #include	<stdio.h>
216 #include	<sys/mnttab.h>
217 #endif
218 
219 #include	<stdlib.h>
220 #include	<unistd.h>
221 #include	<malloc.h>
222 #include	<string.h>
223 #include	<strings.h>
224 #include	<ctype.h>
225 #include	<errno.h>
226 #include	<sys/param.h>
227 #include	<time.h>
228 #include	<sys/types.h>
229 #include	<sys/sysmacros.h>
230 #include	<sys/vnode.h>
231 #include	<sys/fs/ufs_fsdir.h>
232 #include	<sys/fs/ufs_inode.h>
233 #include	<sys/fs/ufs_fs.h>
234 #include	<sys/fs/ufs_log.h>
235 #include	<sys/mntent.h>
236 #include	<sys/filio.h>
237 #include	<limits.h>
238 #include	<sys/int_const.h>
239 #include	<signal.h>
240 #include	<sys/efi_partition.h>
241 #include	"roll_log.h"
242 
243 #define	bcopy(f, t, n)    (void) memcpy(t, f, n)
244 #define	bzero(s, n)	(void) memset(s, 0, n)
245 #define	bcmp(s, d, n)	memcmp(s, d, n)
246 
247 #define	index(s, r)	strchr(s, r)
248 #define	rindex(s, r)	strrchr(s, r)
249 
250 #include	<sys/stat.h>
251 #include	<sys/statvfs.h>
252 #include	<locale.h>
253 #include	<fcntl.h>
254 #include 	<sys/isa_defs.h>	/* for ENDIAN defines */
255 #include	<sys/vtoc.h>
256 
257 #include	<sys/dkio.h>
258 #include	<sys/asynch.h>
259 
260 extern offset_t	llseek();
261 extern char	*getfullblkname();
262 extern long	lrand48();
263 
264 extern int	optind;
265 extern char	*optarg;
266 
267 
268 /*
269  * The size of a cylinder group is calculated by CGSIZE. The maximum size
270  * is limited by the fact that cylinder groups are at most one block.
271  * Its size is derived from the size of the maps maintained in the
272  * cylinder group and the (struct cg) size.
273  */
274 #define	CGSIZE(fs) \
275 	/* base cg		*/ (sizeof (struct cg) + \
276 	/* blktot size	*/ (fs)->fs_cpg * sizeof (long) + \
277 	/* blks size	*/ (fs)->fs_cpg * (fs)->fs_nrpos * sizeof (short) + \
278 	/* inode map	*/ howmany((fs)->fs_ipg, NBBY) + \
279 	/* block map */ howmany((fs)->fs_cpg * (fs)->fs_spc / NSPF(fs), NBBY))
280 
281 /*
282  * We limit the size of the inode map to be no more than a
283  * third of the cylinder group space, since we must leave at
284  * least an equal amount of space for the block map.
285  *
286  * N.B.: MAXIpG must be a multiple of INOPB(fs).
287  */
288 #define	MAXIpG(fs)	roundup((fs)->fs_bsize * NBBY / 3, INOPB(fs))
289 
290 /*
291  * Same as MAXIpG, but parameterized by the block size (b) and the
292  * cylinder group divisor (d), which is the reciprocal of the fraction of the
293  * cylinder group overhead block that is used for the inode map.  So for
294  * example, if d = 5, the macro's computation assumes that 1/5 of the
295  * cylinder group overhead block can be dedicated to the inode map.
296  */
297 #define	MAXIpG_B(b, d)	roundup((b) * NBBY / (d), (b) / sizeof (struct dinode))
298 
299 #define	UMASK		0755
300 #define	MAXINOPB	(MAXBSIZE / sizeof (struct dinode))
301 #define	POWEROF2(num)	(((num) & ((num) - 1)) == 0)
302 #define	MB		(1024*1024)
303 #define	BETWEEN(x, l, h)	((x) >= (l) && (x) <= (h))
304 
305 /*
306  * Used to set the inode generation number. Since both inodes and dinodes
307  * are dealt with, we really need a pointer to an icommon here.
308  */
309 #define	IRANDOMIZE(icp)	(icp)->ic_gen = lrand48();
310 
311 /*
312  * Flags for number()
313  */
314 #define	ALLOW_PERCENT	0x01	/* allow trailing `%' on number */
315 #define	ALLOW_MS1	0x02	/* allow trailing `ms', state 1 */
316 #define	ALLOW_MS2	0x04	/* allow trailing `ms', state 2 */
317 #define	ALLOW_END_ONLY	0x08	/* must be at end of number & suffixes */
318 
319 #define	MAXAIO	1000	/* maximum number of outstanding I/O's we'll manage */
320 #define	BLOCK	1	/* block in aiowait */
321 #define	NOBLOCK	0	/* don't block in aiowait */
322 
323 #define	RELEASE 1	/* free an aio buffer after use */
324 #define	SAVE	0	/* don't free the buffer */
325 
326 typedef struct aio_trans {
327 	aio_result_t resultbuf;
328 	diskaddr_t bno;
329 	char *buffer;
330 	int size;
331 	int release;
332 	struct aio_trans *next;
333 } aio_trans;
334 
335 typedef struct aio_results {
336 	int max;
337 	int outstanding;
338 	int maxpend;
339 	aio_trans *trans;
340 } aio_results;
341 
342 int aio_inited = 0;
343 aio_results results;
344 
345 /*
346  * Allow up to MAXBUF aio requests that each have a unique buffer.
347  * More aio's might be done, but not using memory through the getbuf()
348  * interface.  This can be raised, but you run into the potential of
349  * using more memory than is physically available on the machine,
350  * and if you start swapping, you can forget about performance.
351  * To prevent this, we also limit the total memory used for a given
352  * type of buffer to MAXBUFMEM.
353  *
354  * Tests indicate a cylinder group's worth of inodes takes:
355  *
356  *	NBPI	Size of Inode Buffer
357  *	 2k	1688k
358  *	 8k	 424k
359  *
360  * initcg() stores all the inodes for a cylinder group in one buffer,
361  * so allowing 20 buffers could take 32 MB if not limited by MAXBUFMEM.
362  */
363 #define	MAXBUF		20
364 #define	MAXBUFMEM	(8 * 1024 * 1024)
365 
366 /*
367  * header information for buffers managed by getbuf() and freebuf()
368  */
369 typedef struct bufhdr {
370 	struct bufhdr *head;
371 	struct bufhdr *next;
372 } bufhdr;
373 
374 int bufhdrsize;
375 
376 bufhdr inodebuf = { NULL, NULL };
377 bufhdr cgsumbuf = { NULL, NULL };
378 
379 #define	SECTORS_PER_TERABYTE	(1LL << 31)
380 /*
381  * The following constant specifies an upper limit for file system size
382  * that is actually a lot bigger than we expect to support with UFS. (Since
383  * it's specified in sectors, the file system size would be 2**44 * 512,
384  * which is 2**53, which is 8192 Terabytes.)  However, it's useful
385  * for checking the basic sanity of a size value that is input on the
386  * command line.
387  */
388 #define	FS_SIZE_UPPER_LIMIT	0x100000000000LL
389 
390 /*
391  * Forward declarations
392  */
393 static char *getbuf(bufhdr *bufhead, int size);
394 static void freebuf(char *buf);
395 static void freetrans(aio_trans *transp);
396 static aio_trans *get_aiop();
397 static aio_trans *wait_for_write(int block);
398 static void initcg(int cylno);
399 static void fsinit();
400 static int makedir(struct direct *protodir, int entries);
401 static void iput(struct inode *ip);
402 static void rdfs(diskaddr_t bno, int size, char *bf);
403 static void wtfs(diskaddr_t bno, int size, char *bf);
404 static void awtfs(diskaddr_t bno, int size, char *bf, int release);
405 static void wtfs_breakup(diskaddr_t bno, int size, char *bf);
406 static int isblock(struct fs *fs, unsigned char *cp, int h);
407 static void clrblock(struct fs *fs, unsigned char *cp, int h);
408 static void setblock(struct fs *fs, unsigned char *cp, int h);
409 static void usage();
410 static void dump_fscmd(char *fsys, int fsi);
411 static uint64_t number(uint64_t d_value, char *param, int flags);
412 static int match(char *s);
413 static char checkopt(char *optim);
414 static char checkmtb(char *mtbarg);
415 static void range_check(long *varp, char *name, long minimum,
416     long maximum, long def_val, int user_supplied);
417 static void range_check_64(uint64_t *varp, char *name, uint64_t minimum,
418     uint64_t maximum, uint64_t def_val, int user_supplied);
419 static daddr32_t alloc(int size, int mode);
420 static diskaddr_t get_max_size(int fd);
421 static long get_max_track_size(int fd);
422 static void block_sigint(sigset_t *old_mask);
423 static void unblock_sigint(sigset_t *old_mask);
424 static void recover_from_sigint(int signum);
425 static int confirm_abort(void);
426 static int getline(FILE *fp, char *loc, int maxlen);
427 static void flush_writes(void);
428 static long compute_maxcpg(long, long, long, long, long);
429 static int in_64bit_mode(void);
430 static int validate_size(int fd, diskaddr_t size);
431 static void dump_sblock(void);
432 
433 /*
434  * Workaround for mkfs to function properly on disks attached to XMIT 2.X
435  * controller. If the address is not aligned at 8 byte boundary, mkfs on
436  * disks attached to XMIT 2.X controller exhibts un-predictable behaviour.
437  */
438 #define	XMIT_2_X_ALIGN	8
439 #pragma	align XMIT_2_X_ALIGN(fsun, altfsun, cgun)
440 
441 union {
442 	struct fs fs;
443 	char pad[SBSIZE];
444 } fsun, altfsun;
445 #define	sblock	fsun.fs
446 #define	altsblock	altfsun.fs
447 
448 struct	csum *fscs;
449 
450 union cgun {
451 	struct cg cg;
452 	char pad[MAXBSIZE];
453 } cgun;
454 
455 #define	acg	cgun.cg
456 /*
457  * Size of screen in cols in which to fit output
458  */
459 #define	WIDTH	80
460 
461 struct dinode zino[MAXBSIZE / sizeof (struct dinode)];
462 
463 /*
464  * file descriptors used for rdfs(fsi) and wtfs(fso).
465  * Initialized to an illegal file descriptor number.
466  */
467 int	fsi = -1;
468 int	fso = -1;
469 
470 /*
471  * The BIG parameter is machine dependent.  It should be a longlong integer
472  * constant that can be used by the number parser to check the validity
473  * of numeric parameters.
474  */
475 
476 #define	BIG		0x7fffffffffffffffLL
477 
478 /* Used to indicate to number() that a bogus value should cause us to exit */
479 #define	NO_DEFAULT	LONG_MIN
480 
481 /*
482  * INVALIDSBLIMIT is the number of bad backup superblocks that will be
483  * tolerated before we decide to try arriving at a different set of them
484  * using a different logic. This is applicable for non-EFI disks only.
485  */
486 #define	INVALIDSBLIMIT	10
487 
488 /*
489  * The *_flag variables are used to indicate that the user specified
490  * the values, rather than that we made them up ourselves.  We can
491  * complain about the user giving us bogus values.
492  */
493 
494 /* semi-constants */
495 long	sectorsize = DEV_BSIZE;		/* bytes/sector from param.h */
496 long	bbsize = BBSIZE;		/* boot block size */
497 long	sbsize = SBSIZE;		/* superblock size */
498 
499 /* parameters */
500 diskaddr_t	fssize_db;		/* file system size in disk blocks */
501 diskaddr_t	fssize_frag;		/* file system size in frags */
502 long	cpg;				/* cylinders/cylinder group */
503 int	cpg_flag = RC_DEFAULT;
504 long	rotdelay = -1;			/* rotational delay between blocks */
505 int	rotdelay_flag = RC_DEFAULT;
506 long	maxcontig;			/* max contiguous blocks to allocate */
507 int	maxcontig_flag = RC_DEFAULT;
508 long	nsect = DFLNSECT;		/* sectors per track */
509 int	nsect_flag = RC_DEFAULT;
510 long	ntrack = DFLNTRAK;		/* tracks per cylinder group */
511 int	ntrack_flag = RC_DEFAULT;
512 long	bsize = DESBLKSIZE;		/* filesystem block size */
513 int	bsize_flag = RC_DEFAULT;
514 long	fragsize = DESFRAGSIZE; 	/* filesystem fragment size */
515 int	fragsize_flag = RC_DEFAULT;
516 long	minfree = MINFREE; 		/* fs_minfree */
517 int	minfree_flag = RC_DEFAULT;
518 long	rps = DEFHZ;			/* revolutions/second of drive */
519 int	rps_flag = RC_DEFAULT;
520 long	nbpi = NBPI;			/* number of bytes per inode */
521 int	nbpi_flag = RC_DEFAULT;
522 long	nrpos = NRPOS;			/* number of rotational positions */
523 int	nrpos_flag = RC_DEFAULT;
524 long	apc = 0;			/* alternate sectors per cylinder */
525 int	apc_flag = RC_DEFAULT;
526 char	opt = 't';			/* optimization style, `t' or `s' */
527 char	mtb = 'n';			/* multi-terabyte format, 'y' or 'n' */
528 #define	DEFAULT_SECT_TRAK_CPG	(nsect_flag == RC_DEFAULT && \
529 				ntrack_flag == RC_DEFAULT && \
530 				cpg_flag == RC_DEFAULT)
531 
532 long	debug = 0;			/* enable debugging output */
533 
534 int	spc_flag = 0;			/* alternate sectors specified or */
535 					/* found */
536 
537 /* global state */
538 int	Nflag;		/* do not write to disk */
539 int	mflag;		/* return the command line used to create this FS */
540 int	rflag;		/* report the superblock in an easily-parsed form */
541 int	Rflag;		/* dump the superblock in binary */
542 char	*fsys;
543 time_t	mkfstime;
544 char	*string;
545 int	label_type;
546 
547 /*
548  * logging support
549  */
550 int	ismdd;			/* true if device is a SVM device */
551 int	islog;			/* true if ufs or SVM logging is enabled */
552 int	islogok;		/* true if ufs/SVM log state is good */
553 
554 static int	isufslog;	/* true if ufs logging is enabled */
555 static int	waslog;		/* true when ufs logging disabled during grow */
556 
557 /*
558  * growfs defines, globals, and forward references
559  */
560 #define	NOTENOUGHSPACE 33
561 int		grow;
562 #define	GROW_WITH_DEFAULT_TRAK	(grow && ntrack_flag == RC_DEFAULT)
563 
564 static int	Pflag;		/* probe to which size the fs can be grown */
565 int		ismounted;
566 char		*directory;
567 diskaddr_t	grow_fssize;
568 long		grow_fs_size;
569 long		grow_fs_ncg;
570 diskaddr_t		grow_fs_csaddr;
571 long		grow_fs_cssize;
572 int		grow_fs_clean;
573 struct csum	*grow_fscs;
574 diskaddr_t		grow_sifrag;
575 int		test;
576 int		testforce;
577 diskaddr_t		testfrags;
578 int		inlockexit;
579 int		isbad;
580 
581 void		lockexit(int);
582 void		randomgeneration(void);
583 void		checksummarysize(void);
584 int		checksblock(struct fs, int);
585 void		growinit(char *);
586 void		checkdev(char *, char  *);
587 void		checkmount(struct mnttab *, char *);
588 struct dinode	*gdinode(ino_t);
589 int		csfraginrange(daddr32_t);
590 struct csfrag	*findcsfrag(daddr32_t, struct csfrag **);
591 void		checkindirect(ino_t, daddr32_t *, daddr32_t, int);
592 void		addcsfrag(ino_t, daddr32_t, struct csfrag **);
593 void		delcsfrag(daddr32_t, struct csfrag **);
594 void		checkdirect(ino_t, daddr32_t *, daddr32_t *, int);
595 void		findcsfragino(void);
596 void		fixindirect(daddr32_t, int);
597 void		fixdirect(caddr_t, daddr32_t, daddr32_t *, int);
598 void		fixcsfragino(void);
599 void		extendsummaryinfo(void);
600 int		notenoughspace(void);
601 void		unalloccsfragino(void);
602 void		unalloccsfragfree(void);
603 void		findcsfragfree(void);
604 void		copycsfragino(void);
605 void		rdcg(long);
606 void		wtcg(void);
607 void		flcg(void);
608 void		allocfrags(long, daddr32_t *, long *);
609 void		alloccsfragino(void);
610 void		alloccsfragfree(void);
611 void		freefrags(daddr32_t, long, long);
612 int		findfreerange(long *, long *);
613 void		resetallocinfo(void);
614 void		extendcg(long);
615 void		ulockfs(void);
616 void		wlockfs(void);
617 void		clockfs(void);
618 void		wtsb(void);
619 static int64_t	checkfragallocated(daddr32_t);
620 static struct csum 	*read_summaryinfo(struct fs *);
621 static diskaddr_t 	probe_summaryinfo();
622 
623 int
624 main(int argc, char *argv[])
625 {
626 	long i, mincpc, mincpg, ibpcl;
627 	long cylno, rpos, blk, j, warn = 0;
628 	long mincpgcnt, maxcpg;
629 	uint64_t used, bpcg, inospercg;
630 	long mapcramped, inodecramped;
631 	long postblsize, rotblsize, totalsbsize;
632 	FILE *mnttab;
633 	struct mnttab mntp;
634 	char *special;
635 	struct statvfs64 fs;
636 	struct dk_geom dkg;
637 	struct dk_cinfo dkcinfo;
638 	char pbuf[sizeof (uint64_t) * 3 + 1];
639 	char *tmpbuf;
640 	int width, plen;
641 	uint64_t num;
642 	int c, saverr;
643 	diskaddr_t max_fssize;
644 	long tmpmaxcontig = -1;
645 	struct sigaction sigact;
646 	uint64_t nbytes64;
647 	int remaining_cg;
648 	int do_dot = 0;
649 	int use_efi_dflts = 0, retry = 0, isremovable = 0;
650 	int invalid_sb_cnt, ret, skip_this_sb, cg_too_small;
651 	int geom_nsect, geom_ntrack, geom_cpg;
652 
653 	(void) setlocale(LC_ALL, "");
654 
655 #if !defined(TEXT_DOMAIN)
656 #define	TEXT_DOMAIN "SYS_TEST"
657 #endif
658 	(void) textdomain(TEXT_DOMAIN);
659 
660 	while ((c = getopt(argc, argv, "F:bmo:VPGM:T:t:")) != EOF) {
661 		switch (c) {
662 
663 		case 'F':
664 			string = optarg;
665 			if (strcmp(string, "ufs") != 0)
666 				usage();
667 			break;
668 
669 		case 'm':	/* return command line used to create this FS */
670 			mflag++;
671 			break;
672 
673 		case 'o':
674 			/*
675 			 * ufs specific options.
676 			 */
677 			string = optarg;
678 			while (*string != '\0') {
679 				if (match("nsect=")) {
680 					nsect = number(DFLNSECT, "nsect", 0);
681 					nsect_flag = RC_KEYWORD;
682 				} else if (match("ntrack=")) {
683 					ntrack = number(DFLNTRAK, "ntrack", 0);
684 					ntrack_flag = RC_KEYWORD;
685 				} else if (match("bsize=")) {
686 					bsize = number(DESBLKSIZE, "bsize", 0);
687 					bsize_flag = RC_KEYWORD;
688 				} else if (match("fragsize=")) {
689 					fragsize = number(DESFRAGSIZE,
690 					    "fragsize", 0);
691 					fragsize_flag = RC_KEYWORD;
692 				} else if (match("cgsize=")) {
693 					cpg = number(DESCPG, "cgsize", 0);
694 					cpg_flag = RC_KEYWORD;
695 				} else if (match("free=")) {
696 					minfree = number(MINFREE, "free",
697 					    ALLOW_PERCENT);
698 					minfree_flag = RC_KEYWORD;
699 				} else if (match("maxcontig=")) {
700 					tmpmaxcontig =
701 					    number(-1, "maxcontig", 0);
702 					maxcontig_flag = RC_KEYWORD;
703 				} else if (match("nrpos=")) {
704 					nrpos = number(NRPOS, "nrpos", 0);
705 					nrpos_flag = RC_KEYWORD;
706 				} else if (match("rps=")) {
707 					rps = number(DEFHZ, "rps", 0);
708 					rps_flag = RC_KEYWORD;
709 				} else if (match("nbpi=")) {
710 					nbpi = number(NBPI, "nbpi", 0);
711 					nbpi_flag = RC_KEYWORD;
712 				} else if (match("opt=")) {
713 					opt = checkopt(string);
714 				} else if (match("mtb=")) {
715 					mtb = checkmtb(string);
716 				} else if (match("apc=")) {
717 					apc = number(0, "apc", 0);
718 					apc_flag = RC_KEYWORD;
719 				} else if (match("gap=")) {
720 					(void) number(0, "gap", ALLOW_MS1);
721 					rotdelay = ROTDELAY;
722 					rotdelay_flag = RC_DEFAULT;
723 				} else if (match("debug=")) {
724 					debug = number(0, "debug", 0);
725 				} else if (match("N")) {
726 					Nflag++;
727 				} else if (match("calcsb")) {
728 					rflag++;
729 					Nflag++;
730 				} else if (match("calcbinsb")) {
731 					rflag++;
732 					Rflag++;
733 					Nflag++;
734 				} else if (*string == '\0') {
735 					break;
736 				} else {
737 					(void) fprintf(stderr, gettext(
738 					    "illegal option: %s\n"), string);
739 					usage();
740 				}
741 
742 				if (*string == ',') string++;
743 				if (*string == ' ') string++;
744 			}
745 			break;
746 
747 		case 'V':
748 			{
749 				char	*opt_text;
750 				int	opt_count;
751 
752 				(void) fprintf(stdout, gettext("mkfs -F ufs "));
753 				for (opt_count = 1; opt_count < argc;
754 				    opt_count++) {
755 					opt_text = argv[opt_count];
756 					if (opt_text)
757 						(void) fprintf(stdout, " %s ",
758 						    opt_text);
759 				}
760 				(void) fprintf(stdout, "\n");
761 			}
762 			break;
763 
764 		case 'b':	/* do nothing for this */
765 			break;
766 
767 		case 'M':	/* grow the mounted file system */
768 			directory = optarg;
769 
770 			/* FALLTHROUGH */
771 		case 'G':	/* grow the file system */
772 			grow = 1;
773 			break;
774 		case 'P':	/* probe the file system growing size 	*/
775 			Pflag = 1;
776 			grow = 1; /* probe mode implies fs growing	*/
777 			break;
778 		case 'T':	/* For testing */
779 			testforce = 1;
780 
781 			/* FALLTHROUGH */
782 		case 't':
783 			test = 1;
784 			string = optarg;
785 			testfrags = number(NO_DEFAULT, "testfrags", 0);
786 			break;
787 
788 		case '?':
789 			usage();
790 			break;
791 		}
792 	}
793 #ifdef MKFS_DEBUG
794 	/*
795 	 * Turning on MKFS_DEBUG causes mkfs to produce a filesystem
796 	 * that can be reproduced by setting the time to 0 and seeding
797 	 * the random number generator to a constant.
798 	 */
799 	mkfstime = 0;	/* reproducible results */
800 #else
801 	(void) time(&mkfstime);
802 #endif
803 
804 	if (optind >= (argc - 1)) {
805 		if (optind > (argc - 1)) {
806 			(void) fprintf(stderr,
807 			    gettext("special not specified\n"));
808 			usage();
809 		} else if (mflag == 0) {
810 			(void) fprintf(stderr,
811 			    gettext("size not specified\n"));
812 			usage();
813 		}
814 	}
815 	argc -= optind;
816 	argv = &argv[optind];
817 
818 	fsys = argv[0];
819 	fsi = open64(fsys, O_RDONLY);
820 	if (fsi < 0) {
821 		(void) fprintf(stderr, gettext("%s: cannot open\n"), fsys);
822 		lockexit(32);
823 	}
824 
825 	if (mflag) {
826 		dump_fscmd(fsys, fsi);
827 		lockexit(0);
828 	}
829 
830 	/*
831 	 * The task of setting all of the configuration parameters for a
832 	 * UFS file system is basically a matter of solving n equations
833 	 * in m variables.  Typically, m is greater than n, so there is
834 	 * usually more than one valid solution.  Since this is usually
835 	 * an under-constrained problem, it's not always obvious what the
836 	 * "best" configuration is.
837 	 *
838 	 * In general, the approach is to
839 	 * 1. Determine the values for the file system parameters
840 	 *    that are externally contrained and therefore not adjustable
841 	 *    by mkfs (such as the device's size and maxtransfer size).
842 	 * 2. Acquire the user's requested setting for all configuration
843 	 *    values that can be set on the command line.
844 	 * 3. Determine the final value of all configuration values, by
845 	 *    the following approach:
846 	 *	- set the file system block size (fs_bsize).  Although
847 	 *	  this could be regarded as an adjustable parameter, in
848 	 *	  fact, it's pretty much a constant.  At this time, it's
849 	 *	  generally set to 8k (with older hardware, it can
850 	 *	  sometimes make sense to set it to 4k, but those
851 	 *	  situations are pretty rare now).
852 	 *	- re-adjust the maximum file system size based on the
853 	 *	  value of the file system block size.  Since the
854 	 *	  frag size can't be any larger than a file system
855 	 *	  block, and the number of frags in the file system
856 	 *	  has to fit into 31 bits, the file system block size
857 	 *	  affects the maximum file system size.
858 	 *	- now that the real maximum file system is known, set the
859 	 *	  actual size of the file system to be created to
860 	 *	  MIN(requested size, maximum file system size).
861 	 *	- now validate, and if necessary, adjust the following
862 	 *	  values:
863 	 *		rotdelay
864 	 *		nsect
865 	 *		maxcontig
866 	 *		apc
867 	 *		frag_size
868 	 *		rps
869 	 *		minfree
870 	 *		nrpos
871 	 *		nrack
872 	 *		nbpi
873 	 *	- calculate maxcpg (the maximum value of the cylinders-per-
874 	 *	  cylinder-group configuration parameters).  There are two
875 	 *	  algorithms for calculating maxcpg:  an old one, which is
876 	 *	  used for file systems of less than 1 terabyte, and a
877 	 *	  new one, implemented in the function compute_maxcpg(),
878 	 *	  which is used for file systems of greater than 1 TB.
879 	 *	  The difference between them is that compute_maxcpg()
880 	 *	  really tries to maximize the cpg value.  The old
881 	 *	  algorithm fails to take advantage of smaller frags and
882 	 *	  lower inode density when determining the maximum cpg,
883 	 *	  and thus comes up with much lower numbers in some
884 	 *	  configurations.  At some point, we might use the
885 	 *	  new algorithm for determining maxcpg for all file
886 	 *	  systems, but at this time, the changes implemented for
887 	 *	  multi-terabyte UFS are NOT being automatically applied
888 	 *	  to UFS file systems of less than a terabyte (in the
889 	 *	  interest of not changing existing UFS policy too much
890 	 *	  until the ramifications of the changes are well-understood
891 	 *	  and have been evaluated for their effects on performance.)
892 	 *	- check the current values of the configuration parameters
893 	 *	  against the various constraints imposed by UFS.  These
894 	 *	  include:
895 	 *		* There must be at least one inode in each
896 	 *		  cylinder group.
897 	 *		* The cylinder group overhead block, which
898 	 *		  contains the inode and frag bigmaps, must fit
899 	 *		  within one file system block.
900 	 *		* The space required for inode maps should
901 	 *		  occupy no more than a third of the cylinder
902 	 *		  group overhead block.
903 	 *		* The rotational position tables have to fit
904 	 *		  within the available space in the super block.
905 	 *	  Adjust the configuration values that can be adjusted
906 	 *	  so that these constraints are satisfied.  The
907 	 *	  configuration values that are adjustable are:
908 	 *		* frag size
909 	 *		* cylinders per group
910 	 *		* inode density (can be increased)
911 	 *		* number of rotational positions (the rotational
912 	 *		  position tables are eliminated altogether if
913 	 *		  there isn't enough room for them.)
914 	 * 4. Set the values for all the dependent configuration
915 	 *    values (those that aren't settable on the command
916 	 *    line and which are completely dependent on the
917 	 *    adjustable parameters).  This include cpc (cycles
918 	 *    per cylinder, spc (sectors-per-cylinder), and many others.
919 	 */
920 
921 	/*
922 	 * Figure out the partition size and initialize the label_type.
923 	 */
924 	max_fssize = get_max_size(fsi);
925 
926 	/*
927 	 * Get and check positional arguments, if any.
928 	 */
929 	switch (argc - 1) {
930 	default:
931 		usage();
932 		/*NOTREACHED*/
933 	case 15:
934 		mtb = checkmtb(argv[15]);
935 		/* FALLTHROUGH */
936 	case 14:
937 		string = argv[14];
938 		tmpmaxcontig = number(-1, "maxcontig", 0);
939 		maxcontig_flag = RC_POSITIONAL;
940 		/* FALLTHROUGH */
941 	case 13:
942 		string = argv[13];
943 		nrpos = number(NRPOS, "nrpos", 0);
944 		nrpos_flag = RC_POSITIONAL;
945 		/* FALLTHROUGH */
946 	case 12:
947 		string = argv[12];
948 		rotdelay = ROTDELAY;
949 		rotdelay_flag = RC_DEFAULT;
950 		/* FALLTHROUGH */
951 	case 11:
952 		string = argv[11];
953 		apc = number(0, "apc", 0);
954 		apc_flag = RC_POSITIONAL;
955 		/* FALLTHROUGH */
956 	case 10:
957 		opt = checkopt(argv[10]);
958 		/* FALLTHROUGH */
959 	case 9:
960 		string = argv[9];
961 		nbpi = number(NBPI, "nbpi", 0);
962 		nbpi_flag = RC_POSITIONAL;
963 		/* FALLTHROUGH */
964 	case 8:
965 		string = argv[8];
966 		rps = number(DEFHZ, "rps", 0);
967 		rps_flag = RC_POSITIONAL;
968 		/* FALLTHROUGH */
969 	case 7:
970 		string = argv[7];
971 		minfree = number(MINFREE, "free", ALLOW_PERCENT);
972 		minfree_flag = RC_POSITIONAL;
973 		/* FALLTHROUGH */
974 	case 6:
975 		string = argv[6];
976 		cpg = number(DESCPG, "cgsize", 0);
977 		cpg_flag = RC_POSITIONAL;
978 		/* FALLTHROUGH */
979 	case 5:
980 		string = argv[5];
981 		fragsize = number(DESFRAGSIZE, "fragsize", 0);
982 		fragsize_flag = RC_POSITIONAL;
983 		/* FALLTHROUGH */
984 	case 4:
985 		string = argv[4];
986 		bsize = number(DESBLKSIZE, "bsize", 0);
987 		bsize_flag = RC_POSITIONAL;
988 		/* FALLTHROUGH */
989 	case 3:
990 		string = argv[3];
991 		ntrack = number(DFLNTRAK, "ntrack", 0);
992 		ntrack_flag = RC_POSITIONAL;
993 		/* FALLTHROUGH */
994 	case 2:
995 		string = argv[2];
996 		nsect = number(DFLNSECT, "nsect", 0);
997 		nsect_flag = RC_POSITIONAL;
998 		/* FALLTHROUGH */
999 	case 1:
1000 		string = argv[1];
1001 		fssize_db = number(max_fssize, "size", 0);
1002 	}
1003 
1004 	/*
1005 	 * Initialize the parameters in the same way as newfs so that
1006 	 * newfs and mkfs would result in the same file system layout
1007 	 * for EFI labelled disks. Do this only in the absence of user
1008 	 * specified values for these parameters.
1009 	 */
1010 	if (label_type == LABEL_TYPE_EFI) {
1011 		if (apc_flag == RC_DEFAULT) apc = 0;
1012 		if (nrpos_flag == RC_DEFAULT) nrpos = 1;
1013 		if (ntrack_flag == RC_DEFAULT) ntrack = DEF_TRACKS_EFI;
1014 		if (rps_flag == RC_DEFAULT) rps = DEFHZ;
1015 		if (nsect_flag == RC_DEFAULT) nsect = DEF_SECTORS_EFI;
1016 	}
1017 
1018 	if ((maxcontig_flag == RC_DEFAULT) || (tmpmaxcontig == -1) ||
1019 	    (maxcontig == -1)) {
1020 		long maxtrax = get_max_track_size(fsi);
1021 		maxcontig = maxtrax / bsize;
1022 
1023 	} else {
1024 		maxcontig = tmpmaxcontig;
1025 	}
1026 	dprintf(("DeBuG maxcontig : %ld\n", maxcontig));
1027 
1028 	if (rotdelay == -1) {	/* default by newfs and mkfs */
1029 		rotdelay = ROTDELAY;
1030 	}
1031 
1032 	if (cpg_flag == RC_DEFAULT) { /* If not explicity set, use default */
1033 		cpg = DESCPG;
1034 	}
1035 	dprintf(("DeBuG cpg : %ld\n", cpg));
1036 
1037 	/*
1038 	 * Now that we have the semi-sane args, either positional, via -o,
1039 	 * or by defaulting, handle inter-dependencies and range checks.
1040 	 */
1041 
1042 	/*
1043 	 * Settle the file system block size first, since it's a fixed
1044 	 * parameter once set and so many other parameters, including
1045 	 * max_fssize, depend on it.
1046 	 */
1047 	range_check(&bsize, "bsize", MINBSIZE, MAXBSIZE, DESBLKSIZE,
1048 	    bsize_flag);
1049 
1050 	if (!POWEROF2(bsize)) {
1051 		(void) fprintf(stderr,
1052 		    gettext("block size must be a power of 2, not %ld\n"),
1053 		    bsize);
1054 		bsize = DESBLKSIZE;
1055 		(void) fprintf(stderr,
1056 		    gettext("mkfs: bsize reset to default %ld\n"),
1057 		    bsize);
1058 	}
1059 
1060 	if (fssize_db > max_fssize && validate_size(fsi, fssize_db)) {
1061 		(void) fprintf(stderr, gettext(
1062 		    "Warning: the requested size of this file system\n"
1063 		    "(%lld sectors) is greater than the size of the\n"
1064 		    "device reported by the driver (%lld sectors).\n"
1065 		    "However, a read of the device at the requested size\n"
1066 		    "does succeed, so the requested size will be used.\n"),
1067 		    fssize_db, max_fssize);
1068 		max_fssize = fssize_db;
1069 	}
1070 	/*
1071 	 * Since the maximum allocatable unit (the frag) must be less than
1072 	 * or equal to bsize, and the number of frags must be less than or
1073 	 * equal to INT_MAX, the total size of the file system (in
1074 	 * bytes) must be less than or equal to bsize * INT_MAX.
1075 	 */
1076 
1077 	if (max_fssize > ((diskaddr_t)bsize/DEV_BSIZE) * INT_MAX)
1078 		max_fssize = ((diskaddr_t)bsize/DEV_BSIZE) * INT_MAX;
1079 
1080 	range_check_64(&fssize_db, "size", 1024LL, max_fssize, max_fssize, 1);
1081 
1082 	if (fssize_db >= SECTORS_PER_TERABYTE) {
1083 		mtb = 'y';
1084 		if (!in_64bit_mode()) {
1085 			(void) fprintf(stderr, gettext(
1086 "mkfs:  Warning: Creating a file system greater than 1 terabyte on a\n"
1087 "       system running a 32-bit kernel.  This file system will not be\n"
1088 "       accessible until the system is rebooted with a 64-bit kernel.\n"));
1089 		}
1090 	}
1091 	dprintf(("DeBuG mtb : %c\n", mtb));
1092 
1093 	/*
1094 	 * With newer and much larger disks, the newfs(1M) and mkfs_ufs(1M)
1095 	 * commands had problems in correctly handling the "native" geometries
1096 	 * for various storage devices.
1097 	 *
1098 	 * To handle the new age disks, mkfs_ufs(1M) will use the EFI style
1099 	 * for non-EFI disks that are larger than the CHS addressing limit
1100 	 * ( > 8GB approx ) and ignore the disk geometry information for
1101 	 * these drives. This is what is currently done for multi-terrabyte
1102 	 * filesystems on EFI disks.
1103 	 *
1104 	 * However if the user asked for a specific layout by supplying values
1105 	 * for even one of the three parameters (nsect, ntrack, cpg), honour
1106 	 * the user supplied parameters.
1107 	 *
1108 	 * Choosing EFI style or native geometry style can make a lot of
1109 	 * difference, because the size of a cylinder group is dependent on
1110 	 * this choice. This in turn means that the position of alternate
1111 	 * superblocks varies depending on the style chosen. It is not
1112 	 * necessary that all disks of size > CHSLIMIT have EFI style layout.
1113 	 * There can be disks which are > CHSLIMIT size, but have native
1114 	 * geometry style layout, thereby warranting the need for alternate
1115 	 * logic in superblock detection.
1116 	 */
1117 	if (mtb != 'y' && (ntrack == -1 || GROW_WITH_DEFAULT_TRAK ||
1118 	    DEFAULT_SECT_TRAK_CPG)) {
1119 		/*
1120 		 * "-1" indicates that we were called from newfs and ntracks
1121 		 * was not specified in newfs command line. Calculate nsect
1122 		 * and ntrack in the same manner as newfs.
1123 		 *
1124 		 * This is required because, the defaults for nsect and ntrack
1125 		 * is hardcoded in mkfs, whereas to generate the alternate
1126 		 * superblock locations for the -N option, there is a need for
1127 		 * the geometry based values that newfs would have arrived at.
1128 		 * Newfs would have arrived at these values as below.
1129 		 */
1130 		if (label_type == LABEL_TYPE_EFI ||
1131 		    label_type == LABEL_TYPE_OTHER) {
1132 			use_efi_dflts = 1;
1133 			retry = 1;
1134 		} else if (ioctl(fsi, DKIOCGGEOM, &dkg)) {
1135 			dprintf(("%s: Unable to read Disk geometry", fsys));
1136 			perror(gettext("Unable to read Disk geometry"));
1137 			lockexit(32);
1138 		} else {
1139 			nsect = dkg.dkg_nsect;
1140 			ntrack = dkg.dkg_nhead;
1141 #ifdef i386	/* Bug 1170182 */
1142 			if (ntrack > 32 && (ntrack % 16) != 0) {
1143 				ntrack -= (ntrack % 16);
1144 			}
1145 #endif
1146 			if (ioctl(fsi, DKIOCREMOVABLE, &isremovable)) {
1147 				(void) fprintf(stderr, gettext(
1148 				    "%s: Unable to find Media type. "
1149 				    "Proceeding with system determined "
1150 				    "parameters.\n"), fsys);
1151 				isremovable = 0;
1152 			}
1153 			if (((dkg.dkg_ncyl * dkg.dkg_nhead * dkg.dkg_nsect)
1154 			    > CHSLIMIT) || isremovable) {
1155 				use_efi_dflts = 1;
1156 				retry = 1;
1157 			}
1158 		}
1159 	}
1160 	dprintf(("DeBuG CHSLIMIT = %d geom = %ld\n", CHSLIMIT,
1161 	    dkg.dkg_ncyl * dkg.dkg_nhead * dkg.dkg_nsect));
1162 	dprintf(("DeBuG label_type = %d isremovable = %d use_efi_dflts = %d\n",
1163 	    label_type, isremovable, use_efi_dflts));
1164 
1165 	/*
1166 	 * For the newfs -N case, even if the disksize is > CHSLIMIT, do not
1167 	 * blindly follow EFI style. If the fs_version indicates a geometry
1168 	 * based layout, try that one first. If it fails we can always try the
1169 	 * other logic.
1170 	 *
1171 	 * If we were called from growfs, we will have a problem if we mix
1172 	 * and match the filesystem creation and growth styles. For example,
1173 	 * if we create using EFI style, we have to also grow using EFI
1174 	 * style. So follow the style indicated by the fs_version.
1175 	 *
1176 	 * Read and verify the primary superblock. If it looks sane, use the
1177 	 * fs_version from the superblock. If the primary superblock does
1178 	 * not look good, read and verify the first alternate superblock at
1179 	 * ALTSB. Use the fs_version to decide whether to use the
1180 	 * EFI style logic or the old geometry based logic to calculate
1181 	 * the alternate superblock locations.
1182 	 */
1183 	if ((Nflag && use_efi_dflts) || (grow)) {
1184 		if (grow && ntrack_flag != RC_DEFAULT)
1185 			goto start_fs_creation;
1186 		rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize,
1187 		    (char *)&altsblock);
1188 		ret = checksblock(altsblock, 1);
1189 
1190 		if (!ret) {
1191 			if (altsblock.fs_magic == MTB_UFS_MAGIC) {
1192 				mtb = 'y';
1193 				goto start_fs_creation;
1194 			}
1195 			use_efi_dflts = (altsblock.fs_version ==
1196 			    UFS_EFISTYLE4NONEFI_VERSION_2) ? 1 : 0;
1197 		} else {
1198 			/*
1199 			 * The primary superblock didn't help in determining
1200 			 * the fs_version. Try the first alternate superblock.
1201 			 */
1202 			dprintf(("DeBuG checksblock() failed - error : %d"
1203 			    " for sb : %d\n", ret, SBOFF/sectorsize));
1204 			rdfs((diskaddr_t)ALTSB, (int)sbsize,
1205 			    (char *)&altsblock);
1206 			ret = checksblock(altsblock, 1);
1207 
1208 			if (!ret) {
1209 				if (altsblock.fs_magic == MTB_UFS_MAGIC) {
1210 					mtb = 'y';
1211 					goto start_fs_creation;
1212 				}
1213 				use_efi_dflts = (altsblock.fs_version ==
1214 				    UFS_EFISTYLE4NONEFI_VERSION_2) ? 1 : 0;
1215 			}
1216 			dprintf(("DeBuG checksblock() returned : %d"
1217 			    " for sb : %d\n", ret, ALTSB));
1218 		}
1219 	}
1220 
1221 	geom_nsect = nsect;
1222 	geom_ntrack = ntrack;
1223 	geom_cpg = cpg;
1224 	dprintf(("DeBuG geom_nsect=%d, geom_ntrack=%d, geom_cpg=%d\n",
1225 	    geom_nsect, geom_ntrack, geom_cpg));
1226 
1227 start_fs_creation:
1228 retry_alternate_logic:
1229 	invalid_sb_cnt = 0;
1230 	cg_too_small = 0;
1231 	if (use_efi_dflts) {
1232 		nsect = DEF_SECTORS_EFI;
1233 		ntrack = DEF_TRACKS_EFI;
1234 		cpg = DESCPG;
1235 		dprintf(("\nDeBuG Using EFI defaults\n"));
1236 	} else {
1237 		nsect = geom_nsect;
1238 		ntrack = geom_ntrack;
1239 		cpg = geom_cpg;
1240 		dprintf(("\nDeBuG Using Geometry\n"));
1241 		/*
1242 		 * 32K based on max block size of 64K, and rotational layout
1243 		 * test of nsect <= (256 * sectors/block).  Current block size
1244 		 * limit is not 64K, but it's growing soon.
1245 		 */
1246 		range_check(&nsect, "nsect", 1, 32768, DFLNSECT, nsect_flag);
1247 		/*
1248 		 * ntrack is the number of tracks per cylinder.
1249 		 * The ntrack value must be between 1 and the total number of
1250 		 * sectors in the file system.
1251 		 */
1252 		range_check(&ntrack, "ntrack", 1,
1253 		    fssize_db > INT_MAX ? INT_MAX : (uint32_t)fssize_db,
1254 		    DFLNTRAK, ntrack_flag);
1255 	}
1256 
1257 	range_check(&apc, "apc", 0, nsect - 1, 0, apc_flag);
1258 
1259 	if (mtb == 'y')
1260 		fragsize = bsize;
1261 
1262 	range_check(&fragsize, "fragsize", sectorsize, bsize,
1263 	    MAX(bsize / MAXFRAG, MIN(DESFRAGSIZE, bsize)), fragsize_flag);
1264 
1265 	if ((bsize / MAXFRAG) > fragsize) {
1266 		(void) fprintf(stderr, gettext(
1267 "fragment size %ld is too small, minimum with block size %ld is %ld\n"),
1268 		    fragsize, bsize, bsize / MAXFRAG);
1269 		(void) fprintf(stderr,
1270 		    gettext("mkfs: fragsize reset to minimum %ld\n"),
1271 		    bsize / MAXFRAG);
1272 		fragsize = bsize / MAXFRAG;
1273 	}
1274 
1275 	if (!POWEROF2(fragsize)) {
1276 		(void) fprintf(stderr,
1277 		    gettext("fragment size must be a power of 2, not %ld\n"),
1278 		    fragsize);
1279 		fragsize = MAX(bsize / MAXFRAG, MIN(DESFRAGSIZE, bsize));
1280 		(void) fprintf(stderr,
1281 		    gettext("mkfs: fragsize reset to %ld\n"),
1282 		    fragsize);
1283 	}
1284 
1285 	/* At this point, bsize must be >= fragsize, so no need to check it */
1286 
1287 	if (bsize < PAGESIZE) {
1288 		(void) fprintf(stderr, gettext(
1289 		    "WARNING: filesystem block size (%ld) is smaller than "
1290 		    "memory page size (%ld).\nResulting filesystem can not be "
1291 		    "mounted on this system.\n\n"),
1292 		    bsize, (long)PAGESIZE);
1293 	}
1294 
1295 	range_check(&rps, "rps", 1, 1000, DEFHZ, rps_flag);
1296 	range_check(&minfree, "free", 0, 99, MINFREE, minfree_flag);
1297 	range_check(&nrpos, "nrpos", 1, nsect, MIN(nsect, NRPOS), nrpos_flag);
1298 
1299 	/*
1300 	 * nbpi is variable, but 2MB seems a reasonable upper limit,
1301 	 * as 4MB tends to cause problems (using otherwise-default
1302 	 * parameters).  The true limit is where we end up with one
1303 	 * inode per cylinder group.  If this file system is being
1304 	 * configured for multi-terabyte access, nbpi must be at least 1MB.
1305 	 */
1306 	if (mtb == 'y' && nbpi < MTB_NBPI) {
1307 		if (nbpi_flag != RC_DEFAULT)
1308 			(void) fprintf(stderr, gettext("mkfs: bad value for "
1309 			    "nbpi: must be at least 1048576 for multi-terabyte,"
1310 			    " nbpi reset to default 1048576\n"));
1311 		nbpi = MTB_NBPI;
1312 	}
1313 
1314 	if (mtb == 'y')
1315 		range_check(&nbpi, "nbpi", MTB_NBPI, 2 * MB, MTB_NBPI,
1316 		    nbpi_flag);
1317 	else
1318 		range_check(&nbpi, "nbpi", DEV_BSIZE, 2 * MB, NBPI, nbpi_flag);
1319 
1320 	/*
1321 	 * maxcpg is another variably-limited parameter.  Calculate
1322 	 * the limit based on what we've got for its dependent
1323 	 * variables.  Effectively, it's how much space is left in the
1324 	 * superblock after all the other bits are accounted for.  We
1325 	 * only fill in sblock fields so we can use MAXIpG.
1326 	 *
1327 	 * If the calculation of maxcpg below (for the mtb == 'n'
1328 	 * case) is changed, update newfs as well.
1329 	 *
1330 	 * For old-style, non-MTB format file systems, use the old
1331 	 * algorithm for calculating the maximum cylinder group size,
1332 	 * even though it limits the cylinder group more than necessary.
1333 	 * Since layout can affect performance, we don't want to change
1334 	 * the default layout for non-MTB file systems at this time.
1335 	 * However, for MTB file systems, use the new maxcpg calculation,
1336 	 * which really maxes out the cylinder group size.
1337 	 */
1338 
1339 	sblock.fs_bsize = bsize;
1340 	sblock.fs_inopb = sblock.fs_bsize / sizeof (struct dinode);
1341 
1342 	if (mtb == 'n') {
1343 		maxcpg = (bsize - sizeof (struct cg) -
1344 		    howmany(MAXIpG(&sblock), NBBY)) /
1345 		    (sizeof (long) + nrpos * sizeof (short) +
1346 		    nsect / (MAXFRAG * NBBY));
1347 	} else {
1348 		maxcpg = compute_maxcpg(bsize, fragsize, nbpi, nrpos,
1349 		    nsect * ntrack);
1350 	}
1351 
1352 	dprintf(("DeBuG cpg : %ld\n", cpg));
1353 	/*
1354 	 * Increase the cpg to maxcpg if either newfs was invoked
1355 	 * with -T option or if mkfs wants to create a mtb file system
1356 	 * and if the user has not specified the cpg.
1357 	 */
1358 	if (cpg == -1 || (mtb == 'y' && cpg_flag == RC_DEFAULT))
1359 		cpg = maxcpg;
1360 	dprintf(("DeBuG cpg : %ld\n", cpg));
1361 
1362 	/*
1363 	 * mincpg is variable in complex ways, so we really can't
1364 	 * do a sane lower-end limit check at this point.
1365 	 */
1366 	range_check(&cpg, "cgsize", 1, maxcpg, MIN(maxcpg, DESCPG), cpg_flag);
1367 
1368 	/*
1369 	 * get the controller info
1370 	 */
1371 	ismdd = 0;
1372 	islog = 0;
1373 	islogok = 0;
1374 	waslog = 0;
1375 
1376 	if (ioctl(fsi, DKIOCINFO, &dkcinfo) == 0)
1377 		/*
1378 		 * if it is an MDD (disksuite) device
1379 		 */
1380 		if (dkcinfo.dki_ctype == DKC_MD) {
1381 			ismdd++;
1382 			/*
1383 			 * check the logging device
1384 			 */
1385 			if (ioctl(fsi, _FIOISLOG, NULL) == 0) {
1386 				islog++;
1387 				if (ioctl(fsi, _FIOISLOGOK, NULL) == 0)
1388 					islogok++;
1389 			}
1390 		}
1391 
1392 	/*
1393 	 * Do not grow the file system, but print on stdout the maximum
1394 	 * size in sectors to which the file system can be increased.
1395 	 * The calculated size is limited by fssize_db.
1396 	 * Note that we don't lock the filesystem and therefore under rare
1397 	 * conditions (the filesystem is mounted, the free block count is
1398 	 * almost zero, and the superuser is still changing it) the calculated
1399 	 * size can be imprecise.
1400 	 */
1401 	if (Pflag) {
1402 		(void) printf("%llu\n", probe_summaryinfo());
1403 		exit(0);
1404 	}
1405 
1406 	/*
1407 	 * If we're growing an existing filesystem, then we're about
1408 	 * to start doing things that can require recovery efforts if
1409 	 * we get interrupted, so make sure we get a chance to do so.
1410 	 */
1411 	if (grow) {
1412 		sigact.sa_handler = recover_from_sigint;
1413 		sigemptyset(&sigact.sa_mask);
1414 		sigact.sa_flags = SA_RESTART;
1415 
1416 		if (sigaction(SIGINT, &sigact, (struct sigaction *)NULL) < 0) {
1417 			perror(gettext("Could not register SIGINT handler"));
1418 			lockexit(3);
1419 		}
1420 	}
1421 
1422 	if (!Nflag) {
1423 		/*
1424 		 * Check if MNTTAB is trustable
1425 		 */
1426 		if (statvfs64(MNTTAB, &fs) < 0) {
1427 			(void) fprintf(stderr, gettext("can't statvfs %s\n"),
1428 			    MNTTAB);
1429 			exit(32);
1430 		}
1431 
1432 		if (strcmp(MNTTYPE_MNTFS, fs.f_basetype) != 0) {
1433 			(void) fprintf(stderr, gettext(
1434 			    "%s file system type is not %s, can't mkfs\n"),
1435 			    MNTTAB, MNTTYPE_MNTFS);
1436 			exit(32);
1437 		}
1438 
1439 		special = getfullblkname(fsys);
1440 		checkdev(fsys, special);
1441 
1442 		/*
1443 		 * If we found the block device name,
1444 		 * then check the mount table.
1445 		 * if mounted, and growing write lock the file system
1446 		 *
1447 		 */
1448 		if ((special != NULL) && (*special != '\0')) {
1449 			if ((mnttab = fopen(MNTTAB, "r")) == NULL) {
1450 				(void) fprintf(stderr, gettext(
1451 				    "can't open %s\n"), MNTTAB);
1452 				exit(32);
1453 			}
1454 			while ((getmntent(mnttab, &mntp)) == NULL) {
1455 				if (grow) {
1456 					checkmount(&mntp, special);
1457 					continue;
1458 				}
1459 				if (strcmp(special, mntp.mnt_special) == 0) {
1460 					(void) fprintf(stderr, gettext(
1461 					    "%s is mounted, can't mkfs\n"),
1462 					    special);
1463 					exit(32);
1464 				}
1465 			}
1466 			(void) fclose(mnttab);
1467 		}
1468 
1469 		if (directory && (ismounted == 0)) {
1470 			(void) fprintf(stderr, gettext("%s is not mounted\n"),
1471 			    special);
1472 			lockexit(32);
1473 		}
1474 
1475 		fso = (grow) ? open64(fsys, O_WRONLY) : creat64(fsys, 0666);
1476 		if (fso < 0) {
1477 			saverr = errno;
1478 			(void) fprintf(stderr,
1479 			    gettext("%s: cannot create: %s\n"),
1480 			    fsys, strerror(saverr));
1481 			lockexit(32);
1482 		}
1483 
1484 	} else {
1485 
1486 		/*
1487 		 * For the -N case, a file descriptor is needed for the llseek()
1488 		 * in wtfs(). See the comment in wtfs() for more information.
1489 		 *
1490 		 * Get a file descriptor that's read-only so that this code
1491 		 * doesn't accidentally write to the file.
1492 		 */
1493 		fso = open64(fsys, O_RDONLY);
1494 		if (fso < 0) {
1495 			saverr = errno;
1496 			(void) fprintf(stderr, gettext("%s: cannot open: %s\n"),
1497 			    fsys, strerror(saverr));
1498 			lockexit(32);
1499 		}
1500 	}
1501 
1502 	/*
1503 	 * seed random # generator (for ic_generation)
1504 	 */
1505 #ifdef MKFS_DEBUG
1506 	srand48(12962);	/* reproducible results */
1507 #else
1508 	srand48((long)(time((time_t *)NULL) + getpid()));
1509 #endif
1510 
1511 	if (grow) {
1512 		growinit(fsys);
1513 		goto grow00;
1514 	}
1515 
1516 	/*
1517 	 * Validate the given file system size.
1518 	 * Verify that its last block can actually be accessed.
1519 	 *
1520 	 * Note: it's ok to use sblock as a buffer because it is immediately
1521 	 * overwritten by the rdfs() of the superblock in the next line.
1522 	 *
1523 	 * ToDo: Because the size checking is done in rdfs()/wtfs(), the
1524 	 * error message for specifying an illegal size is very unfriendly.
1525 	 * In the future, one could replace the rdfs()/wtfs() calls
1526 	 * below with in-line calls to read() or write(). This allows better
1527 	 * error messages to be put in place.
1528 	 */
1529 	rdfs(fssize_db - 1, (int)sectorsize, (char *)&sblock);
1530 
1531 	/*
1532 	 * make the fs unmountable
1533 	 */
1534 	rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
1535 	sblock.fs_magic = -1;
1536 	sblock.fs_clean = FSBAD;
1537 	sblock.fs_state = FSOKAY - sblock.fs_time;
1538 	wtfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
1539 	bzero(&sblock, (size_t)sbsize);
1540 
1541 	sblock.fs_nsect = nsect;
1542 	sblock.fs_ntrak = ntrack;
1543 
1544 	/*
1545 	 * Validate specified/determined spc
1546 	 * and calculate minimum cylinders per group.
1547 	 */
1548 
1549 	/*
1550 	 * sectors/cyl = tracks/cyl * sectors/track
1551 	 */
1552 	sblock.fs_spc = sblock.fs_ntrak * sblock.fs_nsect;
1553 
1554 grow00:
1555 	if (apc_flag) {
1556 		sblock.fs_spc -= apc;
1557 	}
1558 	/*
1559 	 * Have to test for this separately from apc_flag, due to
1560 	 * the growfs case....
1561 	 */
1562 	if (sblock.fs_spc != sblock.fs_ntrak * sblock.fs_nsect) {
1563 		spc_flag = 1;
1564 	}
1565 	if (grow)
1566 		goto grow10;
1567 
1568 	sblock.fs_nrpos = nrpos;
1569 	sblock.fs_bsize = bsize;
1570 	sblock.fs_fsize = fragsize;
1571 	sblock.fs_minfree = minfree;
1572 
1573 grow10:
1574 	if (nbpi < sblock.fs_fsize) {
1575 		(void) fprintf(stderr, gettext(
1576 		"warning: wasteful data byte allocation / inode (nbpi):\n"));
1577 		(void) fprintf(stderr, gettext(
1578 		    "%ld smaller than allocatable fragment size of %d\n"),
1579 		    nbpi, sblock.fs_fsize);
1580 	}
1581 	if (grow)
1582 		goto grow20;
1583 
1584 	if (opt == 's')
1585 		sblock.fs_optim = FS_OPTSPACE;
1586 	else
1587 		sblock.fs_optim = FS_OPTTIME;
1588 
1589 	sblock.fs_bmask = ~(sblock.fs_bsize - 1);
1590 	sblock.fs_fmask = ~(sblock.fs_fsize - 1);
1591 	/*
1592 	 * Planning now for future expansion.
1593 	 */
1594 #if defined(_BIG_ENDIAN)
1595 		sblock.fs_qbmask.val[0] = 0;
1596 		sblock.fs_qbmask.val[1] = ~sblock.fs_bmask;
1597 		sblock.fs_qfmask.val[0] = 0;
1598 		sblock.fs_qfmask.val[1] = ~sblock.fs_fmask;
1599 #endif
1600 #if defined(_LITTLE_ENDIAN)
1601 		sblock.fs_qbmask.val[0] = ~sblock.fs_bmask;
1602 		sblock.fs_qbmask.val[1] = 0;
1603 		sblock.fs_qfmask.val[0] = ~sblock.fs_fmask;
1604 		sblock.fs_qfmask.val[1] = 0;
1605 #endif
1606 	for (sblock.fs_bshift = 0, i = sblock.fs_bsize; i > 1; i >>= 1)
1607 		sblock.fs_bshift++;
1608 	for (sblock.fs_fshift = 0, i = sblock.fs_fsize; i > 1; i >>= 1)
1609 		sblock.fs_fshift++;
1610 	sblock.fs_frag = numfrags(&sblock, sblock.fs_bsize);
1611 	for (sblock.fs_fragshift = 0, i = sblock.fs_frag; i > 1; i >>= 1)
1612 		sblock.fs_fragshift++;
1613 	if (sblock.fs_frag > MAXFRAG) {
1614 		(void) fprintf(stderr, gettext(
1615 	"fragment size %d is too small, minimum with block size %d is %d\n"),
1616 		    sblock.fs_fsize, sblock.fs_bsize,
1617 		    sblock.fs_bsize / MAXFRAG);
1618 		lockexit(32);
1619 	}
1620 	sblock.fs_nindir = sblock.fs_bsize / sizeof (daddr32_t);
1621 	sblock.fs_inopb = sblock.fs_bsize / sizeof (struct dinode);
1622 	sblock.fs_nspf = sblock.fs_fsize / sectorsize;
1623 	for (sblock.fs_fsbtodb = 0, i = NSPF(&sblock); i > 1; i >>= 1)
1624 		sblock.fs_fsbtodb++;
1625 
1626 	/*
1627 	 * Compute the super-block, cylinder group, and inode blocks.
1628 	 * Note that these "blkno" are really fragment addresses.
1629 	 * For example, on an 8K/1K (block/fragment) system, fs_sblkno is 16,
1630 	 * fs_cblkno is 24, and fs_iblkno is 32. This is why CGSIZE is so
1631 	 * important: only 1 FS block is allocated for the cg struct (fragment
1632 	 * numbers 24 through 31).
1633 	 */
1634 	sblock.fs_sblkno =
1635 	    roundup(howmany(bbsize + sbsize, sblock.fs_fsize), sblock.fs_frag);
1636 	sblock.fs_cblkno = (daddr32_t)(sblock.fs_sblkno +
1637 	    roundup(howmany(sbsize, sblock.fs_fsize), sblock.fs_frag));
1638 	sblock.fs_iblkno = sblock.fs_cblkno + sblock.fs_frag;
1639 
1640 	sblock.fs_cgoffset = roundup(
1641 	    howmany(sblock.fs_nsect, NSPF(&sblock)), sblock.fs_frag);
1642 	for (sblock.fs_cgmask = -1, i = sblock.fs_ntrak; i > 1; i >>= 1)
1643 		sblock.fs_cgmask <<= 1;
1644 	if (!POWEROF2(sblock.fs_ntrak))
1645 		sblock.fs_cgmask <<= 1;
1646 	/*
1647 	 * Validate specified/determined spc
1648 	 * and calculate minimum cylinders per group.
1649 	 */
1650 
1651 	for (sblock.fs_cpc = NSPB(&sblock), i = sblock.fs_spc;
1652 	    sblock.fs_cpc > 1 && (i & 1) == 0;
1653 	    sblock.fs_cpc >>= 1, i >>= 1)
1654 		/* void */;
1655 	mincpc = sblock.fs_cpc;
1656 
1657 	/* if these calculations are changed, check dump_fscmd also */
1658 	bpcg = (uint64_t)sblock.fs_spc * sectorsize;
1659 	inospercg = (uint64_t)roundup(bpcg / sizeof (struct dinode),
1660 	    INOPB(&sblock));
1661 	if (inospercg > MAXIpG(&sblock))
1662 		inospercg = MAXIpG(&sblock);
1663 	used = (uint64_t)(sblock.fs_iblkno + inospercg /
1664 	    INOPF(&sblock)) * NSPF(&sblock);
1665 	mincpgcnt = (long)howmany((uint64_t)sblock.fs_cgoffset *
1666 	    (~sblock.fs_cgmask) + used, sblock.fs_spc);
1667 	mincpg = roundup(mincpgcnt, mincpc);
1668 	/*
1669 	 * Insure that cylinder group with mincpg has enough space
1670 	 * for block maps
1671 	 */
1672 	sblock.fs_cpg = mincpg;
1673 	sblock.fs_ipg = (int32_t)inospercg;
1674 	mapcramped = 0;
1675 
1676 	/*
1677 	 * Make sure the cg struct fits within the file system block.
1678 	 * Use larger block sizes until it fits
1679 	 */
1680 	while (CGSIZE(&sblock) > sblock.fs_bsize) {
1681 		mapcramped = 1;
1682 		if (sblock.fs_bsize < MAXBSIZE) {
1683 			sblock.fs_bsize <<= 1;
1684 			if ((i & 1) == 0) {
1685 				i >>= 1;
1686 			} else {
1687 				sblock.fs_cpc <<= 1;
1688 				mincpc <<= 1;
1689 				mincpg = roundup(mincpgcnt, mincpc);
1690 				sblock.fs_cpg = mincpg;
1691 			}
1692 			sblock.fs_frag <<= 1;
1693 			sblock.fs_fragshift += 1;
1694 			if (sblock.fs_frag <= MAXFRAG)
1695 				continue;
1696 		}
1697 
1698 		/*
1699 		 * Looped far enough. The fragment is now as large as the
1700 		 * filesystem block!
1701 		 */
1702 		if (sblock.fs_fsize == sblock.fs_bsize) {
1703 			(void) fprintf(stderr, gettext(
1704 		    "There is no block size that can support this disk\n"));
1705 			lockexit(32);
1706 		}
1707 
1708 		/*
1709 		 * Try a larger fragment. Double the fragment size.
1710 		 */
1711 		sblock.fs_frag >>= 1;
1712 		sblock.fs_fragshift -= 1;
1713 		sblock.fs_fsize <<= 1;
1714 		sblock.fs_nspf <<= 1;
1715 	}
1716 	/*
1717 	 * Insure that cylinder group with mincpg has enough space for inodes
1718 	 */
1719 	inodecramped = 0;
1720 	used *= sectorsize;
1721 	nbytes64 = (uint64_t)mincpg * bpcg - used;
1722 	inospercg = (uint64_t)roundup((nbytes64 / nbpi), INOPB(&sblock));
1723 	sblock.fs_ipg = (int32_t)inospercg;
1724 	while (inospercg > MAXIpG(&sblock)) {
1725 		inodecramped = 1;
1726 		if (mincpc == 1 || sblock.fs_frag == 1 ||
1727 		    sblock.fs_bsize == MINBSIZE)
1728 			break;
1729 		nbytes64 = (uint64_t)mincpg * bpcg - used;
1730 		(void) fprintf(stderr,
1731 		    gettext("With a block size of %d %s %lu\n"),
1732 		    sblock.fs_bsize, gettext("minimum bytes per inode is"),
1733 		    (uint32_t)(nbytes64 / MAXIpG(&sblock) + 1));
1734 		sblock.fs_bsize >>= 1;
1735 		sblock.fs_frag >>= 1;
1736 		sblock.fs_fragshift -= 1;
1737 		mincpc >>= 1;
1738 		sblock.fs_cpg = roundup(mincpgcnt, mincpc);
1739 		if (CGSIZE(&sblock) > sblock.fs_bsize) {
1740 			sblock.fs_bsize <<= 1;
1741 			break;
1742 		}
1743 		mincpg = sblock.fs_cpg;
1744 		nbytes64 = (uint64_t)mincpg * bpcg - used;
1745 		inospercg = (uint64_t)roundup((nbytes64 / nbpi),
1746 		    INOPB(&sblock));
1747 		sblock.fs_ipg = (int32_t)inospercg;
1748 	}
1749 	if (inodecramped) {
1750 		if (inospercg > MAXIpG(&sblock)) {
1751 			nbytes64 = (uint64_t)mincpg * bpcg - used;
1752 			(void) fprintf(stderr, gettext(
1753 			    "Minimum bytes per inode is %d\n"),
1754 			    (uint32_t)(nbytes64 / MAXIpG(&sblock) + 1));
1755 		} else if (!mapcramped) {
1756 			(void) fprintf(stderr, gettext(
1757 	    "With %ld bytes per inode, minimum cylinders per group is %ld\n"),
1758 			    nbpi, mincpg);
1759 		}
1760 	}
1761 	if (mapcramped) {
1762 		(void) fprintf(stderr, gettext(
1763 		    "With %d sectors per cylinder, minimum cylinders "
1764 		    "per group is %ld\n"),
1765 		    sblock.fs_spc, mincpg);
1766 	}
1767 	if (inodecramped || mapcramped) {
1768 		/*
1769 		 * To make this at least somewhat comprehensible in
1770 		 * the world of i18n, figure out what we're going to
1771 		 * say and then say it all at one time.  The days of
1772 		 * needing to scrimp on string space are behind us....
1773 		 */
1774 		if ((sblock.fs_bsize != bsize) &&
1775 		    (sblock.fs_fsize != fragsize)) {
1776 			(void) fprintf(stderr, gettext(
1777 	    "This requires the block size to be changed from %ld to %d\n"
1778 	    "and the fragment size to be changed from %ld to %d\n"),
1779 			    bsize, sblock.fs_bsize,
1780 			    fragsize, sblock.fs_fsize);
1781 		} else if (sblock.fs_bsize != bsize) {
1782 			(void) fprintf(stderr, gettext(
1783 	    "This requires the block size to be changed from %ld to %d\n"),
1784 			    bsize, sblock.fs_bsize);
1785 		} else if (sblock.fs_fsize != fragsize) {
1786 			(void) fprintf(stderr, gettext(
1787 	    "This requires the fragment size to be changed from %ld to %d\n"),
1788 			    fragsize, sblock.fs_fsize);
1789 		} else {
1790 			(void) fprintf(stderr, gettext(
1791 	    "Unable to make filesystem fit with the given constraints\n"));
1792 		}
1793 		(void) fprintf(stderr, gettext(
1794 		    "Please re-run mkfs with corrected parameters\n"));
1795 		lockexit(32);
1796 	}
1797 	/*
1798 	 * Calculate the number of cylinders per group
1799 	 */
1800 	sblock.fs_cpg = cpg;
1801 	if (sblock.fs_cpg % mincpc != 0) {
1802 		(void) fprintf(stderr, gettext(
1803 		    "Warning: cylinder groups must have a multiple "
1804 		    "of %ld cylinders with the given\n         parameters\n"),
1805 		    mincpc);
1806 		sblock.fs_cpg = roundup(sblock.fs_cpg, mincpc);
1807 		(void) fprintf(stderr, gettext("Rounded cgsize up to %d\n"),
1808 		    sblock.fs_cpg);
1809 	}
1810 	/*
1811 	 * Must insure there is enough space for inodes
1812 	 */
1813 	/* if these calculations are changed, check dump_fscmd also */
1814 	nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
1815 	sblock.fs_ipg = roundup((uint32_t)(nbytes64 / nbpi), INOPB(&sblock));
1816 
1817 	/*
1818 	 * Slim down cylinders per group, until the inodes can fit.
1819 	 */
1820 	while (sblock.fs_ipg > MAXIpG(&sblock)) {
1821 		inodecramped = 1;
1822 		sblock.fs_cpg -= mincpc;
1823 		nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
1824 		sblock.fs_ipg = roundup((uint32_t)(nbytes64 / nbpi),
1825 		    INOPB(&sblock));
1826 	}
1827 	/*
1828 	 * Must insure there is enough space to hold block map.
1829 	 * Cut down on cylinders per group, until the cg struct fits in a
1830 	 * filesystem block.
1831 	 */
1832 	while (CGSIZE(&sblock) > sblock.fs_bsize) {
1833 		mapcramped = 1;
1834 		sblock.fs_cpg -= mincpc;
1835 		nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
1836 		sblock.fs_ipg = roundup((uint32_t)(nbytes64 / nbpi),
1837 		    INOPB(&sblock));
1838 	}
1839 	sblock.fs_fpg = (sblock.fs_cpg * sblock.fs_spc) / NSPF(&sblock);
1840 	if ((sblock.fs_cpg * sblock.fs_spc) % NSPB(&sblock) != 0) {
1841 		(void) fprintf(stderr,
1842 		gettext("newfs: panic (fs_cpg * fs_spc) %% NSPF != 0\n"));
1843 		lockexit(32);
1844 	}
1845 	if (sblock.fs_cpg < mincpg) {
1846 		(void) fprintf(stderr, gettext(
1847 "With the given parameters, cgsize must be at least %ld; please re-run mkfs\n"),
1848 		    mincpg);
1849 		lockexit(32);
1850 	}
1851 	sblock.fs_cgsize = fragroundup(&sblock, CGSIZE(&sblock));
1852 grow20:
1853 	/*
1854 	 * Now have size for file system and nsect and ntrak.
1855 	 * Determine number of cylinders and blocks in the file system.
1856 	 */
1857 	fssize_frag = (int64_t)dbtofsb(&sblock, fssize_db);
1858 	if (fssize_frag > INT_MAX) {
1859 		(void) fprintf(stderr, gettext(
1860 "There are too many fragments in the system, increase fragment size\n"),
1861 		    mincpg);
1862 		lockexit(32);
1863 	}
1864 	sblock.fs_size = (int32_t)fssize_frag;
1865 	sblock.fs_ncyl = (int32_t)(fssize_frag * NSPF(&sblock) / sblock.fs_spc);
1866 	if (fssize_frag * NSPF(&sblock) >
1867 	    (uint64_t)sblock.fs_ncyl * sblock.fs_spc) {
1868 		sblock.fs_ncyl++;
1869 		warn = 1;
1870 	}
1871 	if (sblock.fs_ncyl < 1) {
1872 		(void) fprintf(stderr, gettext(
1873 		    "file systems must have at least one cylinder\n"));
1874 		lockexit(32);
1875 	}
1876 	if (grow)
1877 		goto grow30;
1878 	/*
1879 	 * Determine feasability/values of rotational layout tables.
1880 	 *
1881 	 * The size of the rotational layout tables is limited by the size
1882 	 * of the file system block, fs_bsize.  The amount of space
1883 	 * available for tables is calculated as (fs_bsize - sizeof (struct
1884 	 * fs)).  The size of these tables is inversely proportional to the
1885 	 * block size of the file system. The size increases if sectors per
1886 	 * track are not powers of two, because more cylinders must be
1887 	 * described by the tables before the rotational pattern repeats
1888 	 * (fs_cpc).
1889 	 */
1890 	sblock.fs_postblformat = FS_DYNAMICPOSTBLFMT;
1891 	sblock.fs_sbsize = fragroundup(&sblock, sizeof (struct fs));
1892 	sblock.fs_npsect = sblock.fs_nsect;
1893 	if (sblock.fs_ntrak == 1) {
1894 		sblock.fs_cpc = 0;
1895 		goto next;
1896 	}
1897 	postblsize = sblock.fs_nrpos * sblock.fs_cpc * sizeof (short);
1898 	rotblsize = sblock.fs_cpc * sblock.fs_spc / NSPB(&sblock);
1899 	totalsbsize = sizeof (struct fs) + rotblsize;
1900 
1901 	/* do static allocation if nrpos == 8 and fs_cpc == 16  */
1902 	if (sblock.fs_nrpos == 8 && sblock.fs_cpc <= 16) {
1903 		/* use old static table space */
1904 		sblock.fs_postbloff = (char *)(&sblock.fs_opostbl[0][0]) -
1905 		    (char *)(&sblock.fs_link);
1906 		sblock.fs_rotbloff = &sblock.fs_space[0] -
1907 		    (uchar_t *)(&sblock.fs_link);
1908 	} else {
1909 		/* use 4.3 dynamic table space */
1910 		sblock.fs_postbloff = &sblock.fs_space[0] -
1911 		    (uchar_t *)(&sblock.fs_link);
1912 		sblock.fs_rotbloff = sblock.fs_postbloff + postblsize;
1913 		totalsbsize += postblsize;
1914 	}
1915 	if (totalsbsize > sblock.fs_bsize ||
1916 	    sblock.fs_nsect > (1 << NBBY) * NSPB(&sblock)) {
1917 		(void) fprintf(stderr, gettext(
1918 		    "Warning: insufficient space in super block for\n"
1919 		    "rotational layout tables with nsect %d, ntrack %d, "
1920 		    "and nrpos %d.\nOmitting tables - file system "
1921 		    "performance may be impaired.\n"),
1922 		    sblock.fs_nsect, sblock.fs_ntrak, sblock.fs_nrpos);
1923 
1924 		/*
1925 		 * Setting fs_cpc to 0 tells alloccgblk() in ufs_alloc.c to
1926 		 * ignore the positional layout table and rotational
1927 		 * position table.
1928 		 */
1929 		sblock.fs_cpc = 0;
1930 		goto next;
1931 	}
1932 	sblock.fs_sbsize = fragroundup(&sblock, totalsbsize);
1933 
1934 
1935 	/*
1936 	 * calculate the available blocks for each rotational position
1937 	 */
1938 	for (cylno = 0; cylno < sblock.fs_cpc; cylno++)
1939 		for (rpos = 0; rpos < sblock.fs_nrpos; rpos++)
1940 			fs_postbl(&sblock, cylno)[rpos] = -1;
1941 	for (i = (rotblsize - 1) * sblock.fs_frag;
1942 	    i >= 0; i -= sblock.fs_frag) {
1943 		cylno = cbtocylno(&sblock, i);
1944 		rpos = cbtorpos(&sblock, i);
1945 		blk = fragstoblks(&sblock, i);
1946 		if (fs_postbl(&sblock, cylno)[rpos] == -1)
1947 			fs_rotbl(&sblock)[blk] = 0;
1948 		else
1949 			fs_rotbl(&sblock)[blk] =
1950 			    fs_postbl(&sblock, cylno)[rpos] - blk;
1951 		fs_postbl(&sblock, cylno)[rpos] = blk;
1952 	}
1953 next:
1954 grow30:
1955 	/*
1956 	 * Compute/validate number of cylinder groups.
1957 	 * Note that if an excessively large filesystem is specified
1958 	 * (e.g., more than 16384 cylinders for an 8K filesystem block), it
1959 	 * does not get detected until checksummarysize()
1960 	 */
1961 	sblock.fs_ncg = sblock.fs_ncyl / sblock.fs_cpg;
1962 	if (sblock.fs_ncyl % sblock.fs_cpg)
1963 		sblock.fs_ncg++;
1964 	sblock.fs_dblkno = sblock.fs_iblkno + sblock.fs_ipg / INOPF(&sblock);
1965 	i = MIN(~sblock.fs_cgmask, sblock.fs_ncg - 1);
1966 	ibpcl = cgdmin(&sblock, i) - cgbase(&sblock, i);
1967 	if (ibpcl >= sblock.fs_fpg) {
1968 		(void) fprintf(stderr, gettext(
1969 		    "inode blocks/cyl group (%d) >= data blocks (%d)\n"),
1970 		    cgdmin(&sblock, i) - cgbase(&sblock, i) / sblock.fs_frag,
1971 		    sblock.fs_fpg / sblock.fs_frag);
1972 		if ((ibpcl < 0) || (sblock.fs_fpg < 0)) {
1973 			(void) fprintf(stderr, gettext(
1974 	    "number of cylinders per cylinder group (%d) must be decreased.\n"),
1975 			    sblock.fs_cpg);
1976 		} else {
1977 			(void) fprintf(stderr, gettext(
1978 	    "number of cylinders per cylinder group (%d) must be increased.\n"),
1979 			    sblock.fs_cpg);
1980 		}
1981 		(void) fprintf(stderr, gettext(
1982 "Note that cgsize may have been adjusted to allow struct cg to fit.\n"));
1983 		lockexit(32);
1984 	}
1985 	j = sblock.fs_ncg - 1;
1986 	if ((i = fssize_frag - j * sblock.fs_fpg) < sblock.fs_fpg &&
1987 	    cgdmin(&sblock, j) - cgbase(&sblock, j) > i) {
1988 		(void) fprintf(stderr, gettext(
1989 		    "Warning: inode blocks/cyl group (%d) >= data "
1990 		    "blocks (%ld) in last\n    cylinder group. This "
1991 		    "implies %ld sector(s) cannot be allocated.\n"),
1992 		    (cgdmin(&sblock, j) - cgbase(&sblock, j)) / sblock.fs_frag,
1993 		    i / sblock.fs_frag, i * NSPF(&sblock));
1994 		/*
1995 		 * If there is only one cylinder group and that is not even
1996 		 * big enough to hold the inodes, exit.
1997 		 */
1998 		if (sblock.fs_ncg == 1)
1999 			cg_too_small = 1;
2000 		sblock.fs_ncg--;
2001 		sblock.fs_ncyl -= sblock.fs_ncyl % sblock.fs_cpg;
2002 		sblock.fs_size = fssize_frag =
2003 		    (int64_t)sblock.fs_ncyl * (int64_t)sblock.fs_spc /
2004 		    (int64_t)NSPF(&sblock);
2005 		warn = 0;
2006 	}
2007 	if (warn && !spc_flag) {
2008 		(void) fprintf(stderr, gettext(
2009 		    "Warning: %d sector(s) in last cylinder unallocated\n"),
2010 		    sblock.fs_spc - (uint32_t)(fssize_frag * NSPF(&sblock) -
2011 		    (uint64_t)(sblock.fs_ncyl - 1) * sblock.fs_spc));
2012 	}
2013 	/*
2014 	 * fill in remaining fields of the super block
2015 	 */
2016 
2017 	/*
2018 	 * The csum records are stored in cylinder group 0, starting at
2019 	 * cgdmin, the first data block.
2020 	 */
2021 	sblock.fs_csaddr = cgdmin(&sblock, 0);
2022 	sblock.fs_cssize =
2023 	    fragroundup(&sblock, sblock.fs_ncg * sizeof (struct csum));
2024 	i = sblock.fs_bsize / sizeof (struct csum);
2025 	sblock.fs_csmask = ~(i - 1);
2026 	for (sblock.fs_csshift = 0; i > 1; i >>= 1)
2027 		sblock.fs_csshift++;
2028 	fscs = (struct csum *)calloc(1, sblock.fs_cssize);
2029 
2030 	checksummarysize();
2031 	if (mtb == 'y') {
2032 		sblock.fs_magic = MTB_UFS_MAGIC;
2033 		sblock.fs_version = MTB_UFS_VERSION_1;
2034 	} else {
2035 		sblock.fs_magic = FS_MAGIC;
2036 		if (use_efi_dflts)
2037 			sblock.fs_version = UFS_EFISTYLE4NONEFI_VERSION_2;
2038 		else
2039 			sblock.fs_version = UFS_VERSION_MIN;
2040 	}
2041 
2042 	if (grow) {
2043 		bcopy((caddr_t)grow_fscs, (caddr_t)fscs, (int)grow_fs_cssize);
2044 		extendsummaryinfo();
2045 		goto grow40;
2046 	}
2047 	sblock.fs_rotdelay = rotdelay;
2048 	sblock.fs_maxcontig = maxcontig;
2049 	sblock.fs_maxbpg = MAXBLKPG(sblock.fs_bsize);
2050 
2051 	sblock.fs_rps = rps;
2052 	sblock.fs_cgrotor = 0;
2053 	sblock.fs_cstotal.cs_ndir = 0;
2054 	sblock.fs_cstotal.cs_nbfree = 0;
2055 	sblock.fs_cstotal.cs_nifree = 0;
2056 	sblock.fs_cstotal.cs_nffree = 0;
2057 	sblock.fs_fmod = 0;
2058 	sblock.fs_ronly = 0;
2059 	sblock.fs_time = mkfstime;
2060 	sblock.fs_state = FSOKAY - sblock.fs_time;
2061 	sblock.fs_clean = FSCLEAN;
2062 grow40:
2063 
2064 	/*
2065 	 * If all that's needed is a dump of the superblock we
2066 	 * would use by default, we've got it now.  So, splat it
2067 	 * out and leave.
2068 	 */
2069 	if (rflag) {
2070 		dump_sblock();
2071 		lockexit(0);
2072 	}
2073 	/*
2074 	 * Dump out summary information about file system.
2075 	 */
2076 	(void) fprintf(stderr, gettext(
2077 	    "%s:\t%lld sectors in %d cylinders of %d tracks, %d sectors\n"),
2078 	    fsys, (uint64_t)sblock.fs_size * NSPF(&sblock), sblock.fs_ncyl,
2079 	    sblock.fs_ntrak, sblock.fs_nsect);
2080 	(void) fprintf(stderr, gettext(
2081 	    "\t%.1fMB in %d cyl groups (%d c/g, %.2fMB/g, %d i/g)\n"),
2082 	    (float)sblock.fs_size * sblock.fs_fsize / MB, sblock.fs_ncg,
2083 	    sblock.fs_cpg, (float)sblock.fs_fpg * sblock.fs_fsize / MB,
2084 	    sblock.fs_ipg);
2085 
2086 	tmpbuf = calloc(sblock.fs_ncg / 50 + 500, 1);
2087 	if (tmpbuf == NULL) {
2088 		perror("calloc");
2089 		lockexit(32);
2090 	}
2091 	if (cg_too_small) {
2092 		(void) fprintf(stderr, gettext("File system creation failed. "
2093 		    "There is only one cylinder group and\nthat is "
2094 		    "not even big enough to hold the inodes.\n"));
2095 		lockexit(32);
2096 	}
2097 	/*
2098 	 * Now build the cylinders group blocks and
2099 	 * then print out indices of cylinder groups.
2100 	 */
2101 	tprintf(gettext(
2102 	    "super-block backups (for fsck -F ufs -o b=#) at:\n"));
2103 	for (width = cylno = 0; cylno < sblock.fs_ncg && cylno < 10; cylno++) {
2104 		if ((grow == 0) || (cylno >= grow_fs_ncg))
2105 			initcg(cylno);
2106 		num = fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno));
2107 		/*
2108 		 * If Nflag and if the disk is larger than the CHSLIMIT,
2109 		 * then sanity test the superblocks before reporting. If there
2110 		 * are too many superblocks which look insane, we have
2111 		 * to retry with alternate logic. If both methods have
2112 		 * failed, then our efforts to arrive at alternate
2113 		 * superblocks failed, so complain and exit.
2114 		 */
2115 		if (Nflag && retry) {
2116 			skip_this_sb = 0;
2117 			rdfs((diskaddr_t)num, sbsize, (char *)&altsblock);
2118 			ret = checksblock(altsblock, 1);
2119 			if (ret) {
2120 				skip_this_sb = 1;
2121 				invalid_sb_cnt++;
2122 				dprintf(("DeBuG checksblock() failed - error :"
2123 				    " %d for sb : %llu invalid_sb_cnt : %d\n",
2124 				    ret, num, invalid_sb_cnt));
2125 			} else {
2126 				/*
2127 				 * Though the superblock looks sane, verify if
2128 				 * the fs_version in the superblock and the
2129 				 * logic that we are using to arrive at the
2130 				 * superblocks match.
2131 				 */
2132 				if (use_efi_dflts && altsblock.fs_version
2133 				    != UFS_EFISTYLE4NONEFI_VERSION_2) {
2134 					skip_this_sb = 1;
2135 					invalid_sb_cnt++;
2136 				}
2137 			}
2138 			if (invalid_sb_cnt >= INVALIDSBLIMIT) {
2139 				if (retry > 1) {
2140 					(void) fprintf(stderr, gettext(
2141 					    "Error determining alternate "
2142 					    "superblock locations\n"));
2143 					free(tmpbuf);
2144 					lockexit(32);
2145 				}
2146 				retry++;
2147 				use_efi_dflts = !use_efi_dflts;
2148 				free(tmpbuf);
2149 				goto retry_alternate_logic;
2150 			}
2151 			if (skip_this_sb)
2152 				continue;
2153 		}
2154 		(void) sprintf(pbuf, " %llu,", num);
2155 		plen = strlen(pbuf);
2156 		if ((width + plen) > (WIDTH - 1)) {
2157 			width = plen;
2158 			tprintf("\n");
2159 		} else {
2160 			width += plen;
2161 		}
2162 		if (Nflag && retry)
2163 			(void) strncat(tmpbuf, pbuf, strlen(pbuf));
2164 		else
2165 			(void) fprintf(stderr, "%s", pbuf);
2166 	}
2167 	tprintf("\n");
2168 
2169 	remaining_cg = sblock.fs_ncg - cylno;
2170 
2171 	/*
2172 	 * If there are more than 300 cylinder groups still to be
2173 	 * initialized, print a "." for every 50 cylinder groups.
2174 	 */
2175 	if (remaining_cg > 300) {
2176 		tprintf(gettext("Initializing cylinder groups:\n"));
2177 		do_dot = 1;
2178 	}
2179 
2180 	/*
2181 	 * Now initialize all cylinder groups between the first ten
2182 	 * and the last ten.
2183 	 *
2184 	 * If the number of cylinder groups was less than 10, all of the
2185 	 * cylinder group offsets would have printed in the last loop
2186 	 * and cylno will already be equal to sblock.fs_ncg and so this
2187 	 * loop will not be entered.  If there are less than 20 cylinder
2188 	 * groups, cylno is already less than fs_ncg - 10, so this loop
2189 	 * won't be entered in that case either.
2190 	 */
2191 
2192 	i = 0;
2193 	for (; cylno < sblock.fs_ncg - 10; cylno++) {
2194 		if ((grow == 0) || (cylno >= grow_fs_ncg))
2195 			initcg(cylno);
2196 		if (do_dot && cylno % 50 == 0) {
2197 			tprintf(".");
2198 			i++;
2199 			if (i == WIDTH - 1) {
2200 				tprintf("\n");
2201 				i = 0;
2202 			}
2203 		}
2204 	}
2205 
2206 	/*
2207 	 * Now print the cylinder group offsets for the last 10
2208 	 * cylinder groups, if any are left.
2209 	 */
2210 
2211 	if (do_dot) {
2212 		tprintf(gettext(
2213 	    "\nsuper-block backups for last 10 cylinder groups at:\n"));
2214 	}
2215 	for (width = 0; cylno < sblock.fs_ncg; cylno++) {
2216 		if ((grow == 0) || (cylno >= grow_fs_ncg))
2217 			initcg(cylno);
2218 		num = fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno));
2219 		if (Nflag && retry) {
2220 			skip_this_sb = 0;
2221 			rdfs((diskaddr_t)num, sbsize, (char *)&altsblock);
2222 			ret = checksblock(altsblock, 1);
2223 			if (ret) {
2224 				skip_this_sb = 1;
2225 				invalid_sb_cnt++;
2226 				dprintf(("DeBuG checksblock() failed - error :"
2227 				    " %d for sb : %llu invalid_sb_cnt : %d\n",
2228 				    ret, num, invalid_sb_cnt));
2229 			} else {
2230 				/*
2231 				 * Though the superblock looks sane, verify if
2232 				 * the fs_version in the superblock and the
2233 				 * logic that we are using to arrive at the
2234 				 * superblocks match.
2235 				 */
2236 				if (use_efi_dflts && altsblock.fs_version
2237 				    != UFS_EFISTYLE4NONEFI_VERSION_2) {
2238 					skip_this_sb = 1;
2239 					invalid_sb_cnt++;
2240 				}
2241 			}
2242 			if (invalid_sb_cnt >= INVALIDSBLIMIT) {
2243 				if (retry > 1) {
2244 					(void) fprintf(stderr, gettext(
2245 					    "Error determining alternate "
2246 					    "superblock locations\n"));
2247 					free(tmpbuf);
2248 					lockexit(32);
2249 				}
2250 				retry++;
2251 				use_efi_dflts = !use_efi_dflts;
2252 				free(tmpbuf);
2253 				goto retry_alternate_logic;
2254 			}
2255 			if (skip_this_sb)
2256 				continue;
2257 		}
2258 		/* Don't print ',' for the last superblock */
2259 		if (cylno == sblock.fs_ncg-1)
2260 			(void) sprintf(pbuf, " %llu", num);
2261 		else
2262 			(void) sprintf(pbuf, " %llu,", num);
2263 		plen = strlen(pbuf);
2264 		if ((width + plen) > (WIDTH - 1)) {
2265 			width = plen;
2266 			tprintf("\n");
2267 		} else {
2268 			width += plen;
2269 		}
2270 		if (Nflag && retry)
2271 			(void) strncat(tmpbuf, pbuf, strlen(pbuf));
2272 		else
2273 			(void) fprintf(stderr, "%s", pbuf);
2274 	}
2275 	tprintf("\n");
2276 	if (Nflag) {
2277 		if (retry)
2278 			(void) fprintf(stderr, "%s", tmpbuf);
2279 		free(tmpbuf);
2280 		lockexit(0);
2281 	}
2282 
2283 	free(tmpbuf);
2284 	if (grow)
2285 		goto grow50;
2286 
2287 	/*
2288 	 * Now construct the initial file system,
2289 	 * then write out the super-block.
2290 	 */
2291 	fsinit();
2292 grow50:
2293 	/*
2294 	 * write the superblock and csum information
2295 	 */
2296 	wtsb();
2297 
2298 	/*
2299 	 * extend the last cylinder group in the original file system
2300 	 */
2301 	if (grow) {
2302 		extendcg(grow_fs_ncg-1);
2303 		wtsb();
2304 	}
2305 
2306 	/*
2307 	 * Write out the duplicate super blocks to the first 10
2308 	 * cylinder groups (or fewer, if there are fewer than 10
2309 	 * cylinder groups).
2310 	 */
2311 	for (cylno = 0; cylno < sblock.fs_ncg && cylno < 10; cylno++)
2312 		awtfs(fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno)),
2313 		    (int)sbsize, (char *)&sblock, SAVE);
2314 
2315 	/*
2316 	 * Now write out duplicate super blocks to the remaining
2317 	 * cylinder groups.  In the case of multi-terabyte file
2318 	 * systems, just write out the super block to the last ten
2319 	 * cylinder groups (or however many are left).
2320 	 */
2321 	if (mtb == 'y') {
2322 		if (sblock.fs_ncg <= 10)
2323 			cylno = sblock.fs_ncg;
2324 		else if (sblock.fs_ncg <= 20)
2325 			cylno = 10;
2326 		else
2327 			cylno = sblock.fs_ncg - 10;
2328 	}
2329 
2330 	for (; cylno < sblock.fs_ncg; cylno++)
2331 		awtfs(fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno)),
2332 		    (int)sbsize, (char *)&sblock, SAVE);
2333 
2334 	/*
2335 	 * Flush out all the AIO writes we've done.  It's not
2336 	 * necessary to do this explicitly, but it's the only
2337 	 * way to report any errors from those writes.
2338 	 */
2339 	flush_writes();
2340 
2341 	/*
2342 	 * set clean flag
2343 	 */
2344 	if (grow)
2345 		sblock.fs_clean = grow_fs_clean;
2346 	else
2347 		sblock.fs_clean = FSCLEAN;
2348 	sblock.fs_time = mkfstime;
2349 	sblock.fs_state = FSOKAY - sblock.fs_time;
2350 	wtfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
2351 	isbad = 0;
2352 
2353 	if (fsync(fso) == -1) {
2354 		saverr = errno;
2355 		(void) fprintf(stderr,
2356 		    gettext("mkfs: fsync failed on write disk: %s\n"),
2357 		    strerror(saverr));
2358 		/* we're just cleaning up, so keep going */
2359 	}
2360 	if (close(fsi) == -1) {
2361 		saverr = errno;
2362 		(void) fprintf(stderr,
2363 		    gettext("mkfs: close failed on read disk: %s\n"),
2364 		    strerror(saverr));
2365 		/* we're just cleaning up, so keep going */
2366 	}
2367 	if (close(fso) == -1) {
2368 		saverr = errno;
2369 		(void) fprintf(stderr,
2370 		    gettext("mkfs: close failed on write disk: %s\n"),
2371 		    strerror(saverr));
2372 		/* we're just cleaning up, so keep going */
2373 	}
2374 	fsi = fso = -1;
2375 
2376 #ifndef STANDALONE
2377 	lockexit(0);
2378 #endif
2379 
2380 	return (0);
2381 }
2382 
2383 /*
2384  * Figure out how big the partition we're dealing with is.
2385  * The value returned is in disk blocks (sectors);
2386  */
2387 static diskaddr_t
2388 get_max_size(int fd)
2389 {
2390 	struct vtoc vtoc;
2391 	dk_gpt_t *efi_vtoc;
2392 	diskaddr_t	slicesize;
2393 
2394 	int index = read_vtoc(fd, &vtoc);
2395 
2396 	if (index >= 0) {
2397 		label_type = LABEL_TYPE_VTOC;
2398 	} else {
2399 		if (index == VT_ENOTSUP || index == VT_ERROR) {
2400 			/* it might be an EFI label */
2401 			index = efi_alloc_and_read(fd, &efi_vtoc);
2402 			label_type = LABEL_TYPE_EFI;
2403 		}
2404 	}
2405 
2406 	if (index < 0) {
2407 		switch (index) {
2408 		case VT_ERROR:
2409 			break;
2410 		case VT_EIO:
2411 			errno = EIO;
2412 			break;
2413 		case VT_EINVAL:
2414 			errno = EINVAL;
2415 		}
2416 		perror(gettext("Can not determine partition size"));
2417 		lockexit(32);
2418 	}
2419 
2420 	if (label_type == LABEL_TYPE_EFI) {
2421 		slicesize = efi_vtoc->efi_parts[index].p_size;
2422 		efi_free(efi_vtoc);
2423 	} else {
2424 		/*
2425 		 * In the vtoc struct, p_size is a 32-bit signed quantity.
2426 		 * In the dk_gpt struct (efi's version of the vtoc), p_size
2427 		 * is an unsigned 64-bit quantity.  By casting the vtoc's
2428 		 * psize to an unsigned 32-bit quantity, it will be copied
2429 		 * to 'slicesize' (an unsigned 64-bit diskaddr_t) without
2430 		 * sign extension.
2431 		 */
2432 
2433 		slicesize = (uint32_t)vtoc.v_part[index].p_size;
2434 	}
2435 
2436 	dprintf(("DeBuG get_max_size index = %d, p_size = %lld, dolimit = %d\n",
2437 	    index, slicesize, (slicesize > FS_MAX)));
2438 
2439 	/*
2440 	 * The next line limits a UFS file system to the maximum
2441 	 * supported size.
2442 	 */
2443 
2444 	if (slicesize > FS_MAX)
2445 		return (FS_MAX);
2446 	return (slicesize);
2447 }
2448 
2449 static long
2450 get_max_track_size(int fd)
2451 {
2452 	struct dk_cinfo ci;
2453 	long track_size = -1;
2454 
2455 	if (ioctl(fd, DKIOCINFO, &ci) == 0) {
2456 		track_size = ci.dki_maxtransfer * DEV_BSIZE;
2457 	}
2458 
2459 	if ((track_size < 0)) {
2460 		int	error = 0;
2461 		int	maxphys;
2462 		int	gotit = 0;
2463 
2464 		gotit = fsgetmaxphys(&maxphys, &error);
2465 		if (gotit) {
2466 			track_size = MIN(MB, maxphys);
2467 		} else {
2468 			(void) fprintf(stderr, gettext(
2469 "Warning: Could not get system value for maxphys. The value for\n"
2470 "maxcontig will default to 1MB.\n"));
2471 			track_size = MB;
2472 		}
2473 	}
2474 	return (track_size);
2475 }
2476 
2477 /*
2478  * Initialize a cylinder group.
2479  */
2480 static void
2481 initcg(int cylno)
2482 {
2483 	diskaddr_t cbase, d;
2484 	diskaddr_t dlower;	/* last data block before cg metadata */
2485 	diskaddr_t dupper;	/* first data block after cg metadata */
2486 	diskaddr_t dmax;
2487 	int64_t i;
2488 	struct csum *cs;
2489 	struct dinode *inode_buffer;
2490 	int size;
2491 
2492 	/*
2493 	 * Variables used to store intermediate results as a part of
2494 	 * the internal implementation of the cbtocylno() macros.
2495 	 */
2496 	diskaddr_t bno;		/* UFS block number (not sector number) */
2497 	int	cbcylno;	/* current cylinder number */
2498 	int	cbcylno_sect;	/* sector offset within cylinder */
2499 	int	cbsect_incr;	/* amount to increment sector offset */
2500 
2501 	/*
2502 	 * Variables used to store intermediate results as a part of
2503 	 * the internal implementation of the cbtorpos() macros.
2504 	 */
2505 	short	*cgblks;	/* pointer to array of free blocks in cg */
2506 	int	trackrpos;	/* tmp variable for rotation position */
2507 	int	trackoff;	/* offset within a track */
2508 	int	trackoff_incr;	/* amount to increment trackoff */
2509 	int	rpos;		/* rotation position of current block */
2510 	int	rpos_incr;	/* amount to increment rpos per block */
2511 
2512 	union cgun *icgun;	/* local pointer to a cg summary block */
2513 #define	icg	(icgun->cg)
2514 
2515 	icgun = (union cgun *)getbuf(&cgsumbuf, sizeof (union cgun));
2516 
2517 	/*
2518 	 * Determine block bounds for cylinder group.
2519 	 * Allow space for super block summary information in first
2520 	 * cylinder group.
2521 	 */
2522 	cbase = cgbase(&sblock, cylno);
2523 	dmax = cbase + sblock.fs_fpg;
2524 	if (dmax > sblock.fs_size)	/* last cg may be smaller than normal */
2525 		dmax = sblock.fs_size;
2526 	dlower = cgsblock(&sblock, cylno) - cbase;
2527 	dupper = cgdmin(&sblock, cylno) - cbase;
2528 	if (cylno == 0)
2529 		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
2530 	cs = fscs + cylno;
2531 	icg.cg_time = mkfstime;
2532 	icg.cg_magic = CG_MAGIC;
2533 	icg.cg_cgx = cylno;
2534 	/* last one gets whatever's left */
2535 	if (cylno == sblock.fs_ncg - 1)
2536 		icg.cg_ncyl = sblock.fs_ncyl - (sblock.fs_cpg * cylno);
2537 	else
2538 		icg.cg_ncyl = sblock.fs_cpg;
2539 	icg.cg_niblk = sblock.fs_ipg;
2540 	icg.cg_ndblk = dmax - cbase;
2541 	icg.cg_cs.cs_ndir = 0;
2542 	icg.cg_cs.cs_nffree = 0;
2543 	icg.cg_cs.cs_nbfree = 0;
2544 	icg.cg_cs.cs_nifree = 0;
2545 	icg.cg_rotor = 0;
2546 	icg.cg_frotor = 0;
2547 	icg.cg_irotor = 0;
2548 	icg.cg_btotoff = &icg.cg_space[0] - (uchar_t *)(&icg.cg_link);
2549 	icg.cg_boff = icg.cg_btotoff + sblock.fs_cpg * sizeof (long);
2550 	icg.cg_iusedoff = icg.cg_boff +
2551 	    sblock.fs_cpg * sblock.fs_nrpos * sizeof (short);
2552 	icg.cg_freeoff = icg.cg_iusedoff + howmany(sblock.fs_ipg, NBBY);
2553 	icg.cg_nextfreeoff = icg.cg_freeoff +
2554 	    howmany(sblock.fs_cpg * sblock.fs_spc / NSPF(&sblock), NBBY);
2555 	for (i = 0; i < sblock.fs_frag; i++) {
2556 		icg.cg_frsum[i] = 0;
2557 	}
2558 	bzero((caddr_t)cg_inosused(&icg), icg.cg_freeoff - icg.cg_iusedoff);
2559 	icg.cg_cs.cs_nifree += sblock.fs_ipg;
2560 	if (cylno == 0)
2561 		for (i = 0; i < UFSROOTINO; i++) {
2562 			setbit(cg_inosused(&icg), i);
2563 			icg.cg_cs.cs_nifree--;
2564 		}
2565 
2566 	/*
2567 	 * Initialize all the inodes in the cylinder group using
2568 	 * random numbers.
2569 	 */
2570 	size = sblock.fs_ipg * sizeof (struct dinode);
2571 	inode_buffer = (struct dinode *)getbuf(&inodebuf, size);
2572 
2573 	for (i = 0; i < sblock.fs_ipg; i++) {
2574 		IRANDOMIZE(&(inode_buffer[i].di_ic));
2575 	}
2576 
2577 	/*
2578 	 * Write all inodes in a single write for performance.
2579 	 */
2580 	awtfs(fsbtodb(&sblock, (uint64_t)cgimin(&sblock, cylno)), (int)size,
2581 	    (char *)inode_buffer, RELEASE);
2582 
2583 	bzero((caddr_t)cg_blktot(&icg), icg.cg_boff - icg.cg_btotoff);
2584 	bzero((caddr_t)cg_blks(&sblock, &icg, 0),
2585 	    icg.cg_iusedoff - icg.cg_boff);
2586 	bzero((caddr_t)cg_blksfree(&icg), icg.cg_nextfreeoff - icg.cg_freeoff);
2587 
2588 	if (cylno > 0) {
2589 		for (d = 0; d < dlower; d += sblock.fs_frag) {
2590 			setblock(&sblock, cg_blksfree(&icg), d/sblock.fs_frag);
2591 			icg.cg_cs.cs_nbfree++;
2592 			cg_blktot(&icg)[cbtocylno(&sblock, d)]++;
2593 			cg_blks(&sblock, &icg, cbtocylno(&sblock, d))
2594 			    [cbtorpos(&sblock, d)]++;
2595 		}
2596 		sblock.fs_dsize += dlower;
2597 	}
2598 	sblock.fs_dsize += icg.cg_ndblk - dupper;
2599 	if ((i = dupper % sblock.fs_frag) != 0) {
2600 		icg.cg_frsum[sblock.fs_frag - i]++;
2601 		for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
2602 			setbit(cg_blksfree(&icg), dupper);
2603 			icg.cg_cs.cs_nffree++;
2604 		}
2605 	}
2606 
2607 	/*
2608 	 * WARNING: The following code is somewhat confusing, but
2609 	 * results in a substantial performance improvement in mkfs.
2610 	 *
2611 	 * Instead of using cbtocylno() and cbtorpos() macros, we
2612 	 * keep track of all the intermediate state of those macros
2613 	 * in some variables.  This allows simple addition to be
2614 	 * done to calculate the results as we step through the
2615 	 * blocks in an orderly fashion instead of the slower
2616 	 * multiplication and division the macros are forced to
2617 	 * used so they can support random input.  (Multiplication,
2618 	 * division, and remainder operations typically take about
2619 	 * 10x as many processor cycles as other operations.)
2620 	 *
2621 	 * The basic idea is to take code:
2622 	 *
2623 	 *	for (x = starting_x; x < max; x++)
2624 	 *		y = (x * c) / z
2625 	 *
2626 	 * and rewrite it to take advantage of the fact that
2627 	 * the variable x is incrementing in an orderly way:
2628 	 *
2629 	 *	intermediate = starting_x * c
2630 	 *	yval = intermediate / z
2631 	 *	for (x = starting_x; x < max; x++) {
2632 	 *		y = yval;
2633 	 *		intermediate += c
2634 	 *		if (intermediate > z) {
2635 	 *			yval++;
2636 	 *			intermediate -= z
2637 	 *		}
2638 	 *	}
2639 	 *
2640 	 * Performance has improved as much as 4X using this code.
2641 	 */
2642 
2643 	/*
2644 	 * Initialize the starting points for all the cbtocylno()
2645 	 * macro variables and figure out the increments needed each
2646 	 * time through the loop.
2647 	 */
2648 	cbcylno_sect = dupper * NSPF(&sblock);
2649 	cbsect_incr = sblock.fs_frag * NSPF(&sblock);
2650 	cbcylno = cbcylno_sect / sblock.fs_spc;
2651 	cbcylno_sect %= sblock.fs_spc;
2652 	cgblks = cg_blks(&sblock, &icg, cbcylno);
2653 	bno = dupper / sblock.fs_frag;
2654 
2655 	/*
2656 	 * Initialize the starting points for all the cbtorpos()
2657 	 * macro variables and figure out the increments needed each
2658 	 * time through the loop.
2659 	 *
2660 	 * It's harder to simplify the cbtorpos() macro if there were
2661 	 * alternate sectors specified (or if they previously existed
2662 	 * in the growfs case).  Since this is rare, we just revert to
2663 	 * using the macros in this case and skip the variable setup.
2664 	 */
2665 	if (!spc_flag) {
2666 		trackrpos = (cbcylno_sect % sblock.fs_nsect) * sblock.fs_nrpos;
2667 		rpos = trackrpos / sblock.fs_nsect;
2668 		trackoff = trackrpos % sblock.fs_nsect;
2669 		trackoff_incr = cbsect_incr * sblock.fs_nrpos;
2670 		rpos_incr = (trackoff_incr / sblock.fs_nsect) % sblock.fs_nrpos;
2671 		trackoff_incr = trackoff_incr % sblock.fs_nsect;
2672 	}
2673 
2674 	/*
2675 	 * Loop through all the blocks, marking them free and
2676 	 * updating totals kept in the superblock and cg summary.
2677 	 */
2678 	for (d = dupper; d + sblock.fs_frag <= dmax - cbase; ) {
2679 		setblock(&sblock, cg_blksfree(&icg),  bno);
2680 		icg.cg_cs.cs_nbfree++;
2681 
2682 		cg_blktot(&icg)[cbcylno]++;
2683 
2684 		if (!spc_flag)
2685 			cgblks[rpos]++;
2686 		else
2687 			cg_blks(&sblock, &icg, cbtocylno(&sblock, d))
2688 			    [cbtorpos(&sblock, d)]++;
2689 
2690 		d += sblock.fs_frag;
2691 		bno++;
2692 
2693 		/*
2694 		 * Increment the sector offset within the cylinder
2695 		 * for the cbtocylno() macro reimplementation.  If
2696 		 * we're beyond the end of the cylinder, update the
2697 		 * cylinder number, calculate the offset in the
2698 		 * new cylinder, and update the cgblks pointer
2699 		 * to the next rotational position.
2700 		 */
2701 		cbcylno_sect += cbsect_incr;
2702 		if (cbcylno_sect >= sblock.fs_spc) {
2703 			cbcylno++;
2704 			cbcylno_sect -= sblock.fs_spc;
2705 			cgblks += sblock.fs_nrpos;
2706 		}
2707 
2708 		/*
2709 		 * If there aren't alternate sectors, increment the
2710 		 * rotational position variables for the cbtorpos()
2711 		 * reimplementation.  Note that we potentially
2712 		 * increment rpos twice.  Once by rpos_incr, and one
2713 		 * more time when we wrap to a new track because
2714 		 * trackoff >= fs_nsect.
2715 		 */
2716 		if (!spc_flag) {
2717 			trackoff += trackoff_incr;
2718 			rpos += rpos_incr;
2719 			if (trackoff >= sblock.fs_nsect) {
2720 				trackoff -= sblock.fs_nsect;
2721 				rpos++;
2722 			}
2723 			if (rpos >= sblock.fs_nrpos)
2724 				rpos -= sblock.fs_nrpos;
2725 		}
2726 	}
2727 
2728 	if (d < dmax - cbase) {
2729 		icg.cg_frsum[dmax - cbase - d]++;
2730 		for (; d < dmax - cbase; d++) {
2731 			setbit(cg_blksfree(&icg), d);
2732 			icg.cg_cs.cs_nffree++;
2733 		}
2734 	}
2735 	sblock.fs_cstotal.cs_ndir += icg.cg_cs.cs_ndir;
2736 	sblock.fs_cstotal.cs_nffree += icg.cg_cs.cs_nffree;
2737 	sblock.fs_cstotal.cs_nbfree += icg.cg_cs.cs_nbfree;
2738 	sblock.fs_cstotal.cs_nifree += icg.cg_cs.cs_nifree;
2739 	*cs = icg.cg_cs;
2740 	awtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, cylno)),
2741 	    sblock.fs_bsize, (char *)&icg, RELEASE);
2742 }
2743 
2744 /*
2745  * initialize the file system
2746  */
2747 struct inode node;
2748 
2749 #define	LOSTDIR
2750 #ifdef LOSTDIR
2751 #define	PREDEFDIR 3
2752 #else
2753 #define	PREDEFDIR 2
2754 #endif
2755 
2756 struct direct root_dir[] = {
2757 	{ UFSROOTINO, sizeof (struct direct), 1, "." },
2758 	{ UFSROOTINO, sizeof (struct direct), 2, ".." },
2759 #ifdef LOSTDIR
2760 	{ LOSTFOUNDINO, sizeof (struct direct), 10, "lost+found" },
2761 #endif
2762 };
2763 #ifdef LOSTDIR
2764 struct direct lost_found_dir[] = {
2765 	{ LOSTFOUNDINO, sizeof (struct direct), 1, "." },
2766 	{ UFSROOTINO, sizeof (struct direct), 2, ".." },
2767 	{ 0, DIRBLKSIZ, 0, 0 },
2768 };
2769 #endif
2770 char buf[MAXBSIZE];
2771 
2772 static void
2773 fsinit()
2774 {
2775 	int i;
2776 
2777 
2778 	/*
2779 	 * initialize the node
2780 	 */
2781 	node.i_atime = mkfstime;
2782 	node.i_mtime = mkfstime;
2783 	node.i_ctime = mkfstime;
2784 #ifdef LOSTDIR
2785 	/*
2786 	 * create the lost+found directory
2787 	 */
2788 	(void) makedir(lost_found_dir, 2);
2789 	for (i = DIRBLKSIZ; i < sblock.fs_bsize; i += DIRBLKSIZ) {
2790 		bcopy(&lost_found_dir[2], &buf[i], DIRSIZ(&lost_found_dir[2]));
2791 	}
2792 	node.i_number = LOSTFOUNDINO;
2793 	node.i_smode = node.i_mode = IFDIR | 0700;
2794 	node.i_nlink = 2;
2795 	node.i_size = sblock.fs_bsize;
2796 	node.i_db[0] = alloc((int)node.i_size, node.i_mode);
2797 	node.i_blocks = btodb(fragroundup(&sblock, (int)node.i_size));
2798 	IRANDOMIZE(&node.i_ic);
2799 	wtfs(fsbtodb(&sblock, (uint64_t)node.i_db[0]), (int)node.i_size, buf);
2800 	iput(&node);
2801 #endif
2802 	/*
2803 	 * create the root directory
2804 	 */
2805 	node.i_number = UFSROOTINO;
2806 	node.i_mode = node.i_smode = IFDIR | UMASK;
2807 	node.i_nlink = PREDEFDIR;
2808 	node.i_size = makedir(root_dir, PREDEFDIR);
2809 	node.i_db[0] = alloc(sblock.fs_fsize, node.i_mode);
2810 	/* i_size < 2GB because we are initializing the file system */
2811 	node.i_blocks = btodb(fragroundup(&sblock, (int)node.i_size));
2812 	IRANDOMIZE(&node.i_ic);
2813 	wtfs(fsbtodb(&sblock, (uint64_t)node.i_db[0]), sblock.fs_fsize, buf);
2814 	iput(&node);
2815 }
2816 
2817 /*
2818  * construct a set of directory entries in "buf".
2819  * return size of directory.
2820  */
2821 static int
2822 makedir(struct direct *protodir, int entries)
2823 {
2824 	char *cp;
2825 	int i;
2826 	ushort_t spcleft;
2827 
2828 	spcleft = DIRBLKSIZ;
2829 	for (cp = buf, i = 0; i < entries - 1; i++) {
2830 		protodir[i].d_reclen = DIRSIZ(&protodir[i]);
2831 		bcopy(&protodir[i], cp, protodir[i].d_reclen);
2832 		cp += protodir[i].d_reclen;
2833 		spcleft -= protodir[i].d_reclen;
2834 	}
2835 	protodir[i].d_reclen = spcleft;
2836 	bcopy(&protodir[i], cp, DIRSIZ(&protodir[i]));
2837 	return (DIRBLKSIZ);
2838 }
2839 
2840 /*
2841  * allocate a block or frag
2842  */
2843 static daddr32_t
2844 alloc(int size, int mode)
2845 {
2846 	int i, frag;
2847 	daddr32_t d;
2848 
2849 	rdfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2850 	    (char *)&acg);
2851 	if (acg.cg_magic != CG_MAGIC) {
2852 		(void) fprintf(stderr, gettext("cg 0: bad magic number\n"));
2853 		lockexit(32);
2854 	}
2855 	if (acg.cg_cs.cs_nbfree == 0) {
2856 		(void) fprintf(stderr,
2857 		    gettext("first cylinder group ran out of space\n"));
2858 		lockexit(32);
2859 	}
2860 	for (d = 0; d < acg.cg_ndblk; d += sblock.fs_frag)
2861 		if (isblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag))
2862 			goto goth;
2863 	(void) fprintf(stderr,
2864 	    gettext("internal error: can't find block in cyl 0\n"));
2865 	lockexit(32);
2866 goth:
2867 	clrblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag);
2868 	acg.cg_cs.cs_nbfree--;
2869 	sblock.fs_cstotal.cs_nbfree--;
2870 	fscs[0].cs_nbfree--;
2871 	if (mode & IFDIR) {
2872 		acg.cg_cs.cs_ndir++;
2873 		sblock.fs_cstotal.cs_ndir++;
2874 		fscs[0].cs_ndir++;
2875 	}
2876 	cg_blktot(&acg)[cbtocylno(&sblock, d)]--;
2877 	cg_blks(&sblock, &acg, cbtocylno(&sblock, d))[cbtorpos(&sblock, d)]--;
2878 	if (size != sblock.fs_bsize) {
2879 		frag = howmany(size, sblock.fs_fsize);
2880 		fscs[0].cs_nffree += sblock.fs_frag - frag;
2881 		sblock.fs_cstotal.cs_nffree += sblock.fs_frag - frag;
2882 		acg.cg_cs.cs_nffree += sblock.fs_frag - frag;
2883 		acg.cg_frsum[sblock.fs_frag - frag]++;
2884 		for (i = frag; i < sblock.fs_frag; i++)
2885 			setbit(cg_blksfree(&acg), d + i);
2886 	}
2887 	wtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2888 	    (char *)&acg);
2889 	return (d);
2890 }
2891 
2892 /*
2893  * Allocate an inode on the disk
2894  */
2895 static void
2896 iput(struct inode *ip)
2897 {
2898 	struct dinode buf[MAXINOPB];
2899 	diskaddr_t d;
2900 
2901 	rdfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2902 	    (char *)&acg);
2903 	if (acg.cg_magic != CG_MAGIC) {
2904 		(void) fprintf(stderr, gettext("cg 0: bad magic number\n"));
2905 		lockexit(32);
2906 	}
2907 	acg.cg_cs.cs_nifree--;
2908 	setbit(cg_inosused(&acg), ip->i_number);
2909 	wtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2910 	    (char *)&acg);
2911 	sblock.fs_cstotal.cs_nifree--;
2912 	fscs[0].cs_nifree--;
2913 	if ((int)ip->i_number >= sblock.fs_ipg * sblock.fs_ncg) {
2914 		(void) fprintf(stderr,
2915 		    gettext("fsinit: inode value out of range (%d).\n"),
2916 		    ip->i_number);
2917 		lockexit(32);
2918 	}
2919 	d = fsbtodb(&sblock, (uint64_t)itod(&sblock, (int)ip->i_number));
2920 	rdfs(d, sblock.fs_bsize, (char *)buf);
2921 	buf[itoo(&sblock, (int)ip->i_number)].di_ic = ip->i_ic;
2922 	wtfs(d, sblock.fs_bsize, (char *)buf);
2923 }
2924 
2925 /*
2926  * getbuf()	-- Get a buffer for use in an AIO operation.  Buffer
2927  *		is zero'd the first time returned, left with whatever
2928  *		was in memory after that.  This function actually gets
2929  *		enough memory the first time it's called to support
2930  *		MAXBUF buffers like a slab allocator.  When all the
2931  *		buffers are in use, it waits for an aio to complete
2932  *		and make a buffer available.
2933  *
2934  *		Never returns an error.  Either succeeds or exits.
2935  */
2936 static char *
2937 getbuf(bufhdr *bufhead, int size)
2938 {
2939 	bufhdr *pbuf;
2940 	bufhdr *prev;
2941 	int i;
2942 	int buf_size, max_bufs;
2943 
2944 	/*
2945 	 * Initialize all the buffers
2946 	 */
2947 	if (bufhead->head == NULL) {
2948 		/*
2949 		 * round up the size of our buffer header to a
2950 		 * 16 byte boundary so the address we return to
2951 		 * the caller is "suitably aligned".
2952 		 */
2953 		bufhdrsize = (sizeof (bufhdr) + 15) & ~15;
2954 
2955 		/*
2956 		 * Add in our header to the buffer and round it all up to
2957 		 * a 16 byte boundry so each member of the slab is aligned.
2958 		 */
2959 		buf_size = (size + bufhdrsize + 15) & ~15;
2960 
2961 		/*
2962 		 * Limit number of buffers to lesser of MAXBUFMEM's worth
2963 		 * or MAXBUF, whichever is less.
2964 		 */
2965 		max_bufs = MAXBUFMEM / buf_size;
2966 		if (max_bufs > MAXBUF)
2967 			max_bufs = MAXBUF;
2968 
2969 		pbuf = (bufhdr *)calloc(max_bufs, buf_size);
2970 		if (pbuf == NULL) {
2971 			perror("calloc");
2972 			lockexit(32);
2973 		}
2974 
2975 		bufhead->head = bufhead;
2976 		prev = bufhead;
2977 		for (i = 0; i < max_bufs; i++) {
2978 			pbuf->head = bufhead;
2979 			prev->next = pbuf;
2980 			prev = pbuf;
2981 			pbuf = (bufhdr *)((char *)pbuf + buf_size);
2982 		}
2983 	}
2984 
2985 	/*
2986 	 * Get an available buffer, waiting for I/O if necessary
2987 	 */
2988 	wait_for_write(NOBLOCK);
2989 	while (bufhead->next == NULL)
2990 		wait_for_write(BLOCK);
2991 
2992 	/*
2993 	 * Take the buffer off the list
2994 	 */
2995 	pbuf = bufhead->next;
2996 	bufhead->next = pbuf->next;
2997 	pbuf->next = NULL;
2998 
2999 	/*
3000 	 * return the empty buffer space just past the header
3001 	 */
3002 	return ((char *)pbuf + bufhdrsize);
3003 }
3004 
3005 /*
3006  * freebuf()	-- Free a buffer gotten previously through getbuf.
3007  *		Puts the buffer back on the appropriate list for
3008  *		later use.  Never calls free().
3009  *
3010  * Assumes that SIGINT is blocked.
3011  */
3012 static void
3013 freebuf(char *buf)
3014 {
3015 	bufhdr *pbuf;
3016 	bufhdr *bufhead;
3017 
3018 	/*
3019 	 * get the header for this buffer
3020 	 */
3021 	pbuf = (bufhdr *)(buf - bufhdrsize);
3022 
3023 	/*
3024 	 * Put it back on the list of available buffers
3025 	 */
3026 	bufhead = pbuf->head;
3027 	pbuf->next = bufhead->next;
3028 	bufhead->next = pbuf;
3029 }
3030 
3031 /*
3032  * freetrans()	-- Free a transaction gotten previously through getaiop.
3033  *		Puts the transaction struct back on the appropriate list for
3034  *		later use.  Never calls free().
3035  *
3036  * Assumes that SIGINT is blocked.
3037  */
3038 static void
3039 freetrans(aio_trans *transp)
3040 {
3041 	/*
3042 	 * free the buffer associated with this AIO if needed
3043 	 */
3044 	if (transp->release == RELEASE)
3045 		freebuf(transp->buffer);
3046 
3047 	/*
3048 	 * Put transaction on the free list
3049 	 */
3050 	transp->next = results.trans;
3051 	results.trans = transp;
3052 }
3053 
3054 /*
3055  * wait_for_write()	-- Wait for an aio write to complete.  Return
3056  *			the transaction structure for that write.
3057  *
3058  * Blocks SIGINT if necessary.
3059  */
3060 aio_trans *
3061 wait_for_write(int block)
3062 {
3063 	aio_trans	*transp;
3064 	aio_result_t	*resultp;
3065 	static struct timeval  zero_wait = { 0, 0 };
3066 	sigset_t	old_mask;
3067 
3068 	/*
3069 	 * If we know there aren't any outstanding transactions, just return
3070 	 */
3071 	if (results.outstanding == 0)
3072 		return ((aio_trans *) 0);
3073 
3074 	block_sigint(&old_mask);
3075 
3076 	resultp = aiowait(block ? NULL : &zero_wait);
3077 	if (resultp == NULL ||
3078 	    (resultp == (aio_result_t *)-1 && errno == EINVAL)) {
3079 		unblock_sigint(&old_mask);
3080 		return ((aio_trans *) 0);
3081 	}
3082 
3083 	results.outstanding--;
3084 	transp = (aio_trans *)resultp;
3085 
3086 	if (resultp->aio_return != transp->size) {
3087 		if (resultp->aio_return == -1) {
3088 			/*
3089 			 * The aiowrite() may have failed because the
3090 			 * kernel didn't have enough memory to do the job.
3091 			 * Flush all pending writes and try a normal
3092 			 * write().  wtfs_breakup() will call exit if it
3093 			 * fails, so we don't worry about errors here.
3094 			 */
3095 			flush_writes();
3096 			wtfs_breakup(transp->bno, transp->size, transp->buffer);
3097 		} else {
3098 			(void) fprintf(stderr, gettext(
3099 			    "short write (%d of %d bytes) on sector %lld\n"),
3100 			    resultp->aio_return, transp->size,
3101 			    transp->bno);
3102 			/*
3103 			 * Don't unblock SIGINT, to avoid potential
3104 			 * looping due to queued interrupts and
3105 			 * error handling.
3106 			 */
3107 			lockexit(32);
3108 		}
3109 	}
3110 
3111 	resultp->aio_return = 0;
3112 	freetrans(transp);
3113 	unblock_sigint(&old_mask);
3114 	return (transp);
3115 }
3116 
3117 /*
3118  * flush_writes()	-- flush all the outstanding aio writes.
3119  */
3120 static void
3121 flush_writes(void)
3122 {
3123 	while (wait_for_write(BLOCK))
3124 		;
3125 }
3126 
3127 /*
3128  * get_aiop()	-- find and return an aio_trans structure on which a new
3129  *		aio can be done.  Blocks on aiowait() if needed.  Reaps
3130  *		all outstanding completed aio's.
3131  *
3132  * Assumes that SIGINT is blocked.
3133  */
3134 aio_trans *
3135 get_aiop()
3136 {
3137 	int i;
3138 	aio_trans *transp;
3139 	aio_trans *prev;
3140 
3141 	/*
3142 	 * initialize aio stuff
3143 	 */
3144 	if (!aio_inited) {
3145 		aio_inited = 1;
3146 
3147 		results.maxpend = 0;
3148 		results.outstanding = 0;
3149 		results.max = MAXAIO;
3150 
3151 		results.trans = (aio_trans *)calloc(results.max,
3152 		    sizeof (aio_trans));
3153 		if (results.trans == NULL) {
3154 			perror("calloc");
3155 			lockexit(32);
3156 		}
3157 
3158 		/*
3159 		 * Initialize the linked list of aio transaction
3160 		 * structures.  Note that the final "next" pointer
3161 		 * will be NULL since we got the buffer from calloc().
3162 		 */
3163 		prev = results.trans;
3164 		for (i = 1; i < results.max; i++) {
3165 			prev->next = &(results.trans[i]);
3166 			prev = prev->next;
3167 		}
3168 	}
3169 
3170 	wait_for_write(NOBLOCK);
3171 	while (results.trans == NULL)
3172 		wait_for_write(BLOCK);
3173 	transp = results.trans;
3174 	results.trans = results.trans->next;
3175 
3176 	transp->next = 0;
3177 	transp->resultbuf.aio_return = AIO_INPROGRESS;
3178 	return (transp);
3179 }
3180 
3181 /*
3182  * read a block from the file system
3183  */
3184 static void
3185 rdfs(diskaddr_t bno, int size, char *bf)
3186 {
3187 	int n, saverr;
3188 
3189 	/*
3190 	 * In case we need any data that's pending in an aiowrite(),
3191 	 * we wait for them all to complete before doing a read.
3192 	 */
3193 	flush_writes();
3194 
3195 	/*
3196 	 * Note: the llseek() can succeed, even if the offset is out of range.
3197 	 * It's not until the file i/o operation (the read()) that one knows
3198 	 * for sure if the raw device can handle the offset.
3199 	 */
3200 	if (llseek(fsi, (offset_t)bno * sectorsize, 0) < 0) {
3201 		saverr = errno;
3202 		(void) fprintf(stderr,
3203 		    gettext("seek error on sector %lld: %s\n"),
3204 		    bno, strerror(saverr));
3205 		lockexit(32);
3206 	}
3207 	n = read(fsi, bf, size);
3208 	if (n != size) {
3209 		saverr = errno;
3210 		if (n == -1)
3211 			(void) fprintf(stderr,
3212 			    gettext("read error on sector %lld: %s\n"),
3213 			    bno, strerror(saverr));
3214 		else
3215 			(void) fprintf(stderr, gettext(
3216 			    "short read (%d of %d bytes) on sector %lld\n"),
3217 			    n, size, bno);
3218 		lockexit(32);
3219 	}
3220 }
3221 
3222 /*
3223  * write a block to the file system
3224  */
3225 static void
3226 wtfs(diskaddr_t bno, int size, char *bf)
3227 {
3228 	int n, saverr;
3229 
3230 	if (fso == -1)
3231 		return;
3232 
3233 	/*
3234 	 * Note: the llseek() can succeed, even if the offset is out of range.
3235 	 * It's not until the file i/o operation (the write()) that one knows
3236 	 * for sure if the raw device can handle the offset.
3237 	 */
3238 	if (llseek(fso, (offset_t)bno * sectorsize, 0) < 0) {
3239 		saverr = errno;
3240 		(void) fprintf(stderr,
3241 		    gettext("seek error on sector %lld: %s\n"),
3242 		    bno, strerror(saverr));
3243 		lockexit(32);
3244 	}
3245 	if (Nflag)
3246 		return;
3247 	n = write(fso, bf, size);
3248 	if (n != size) {
3249 		saverr = errno;
3250 		if (n == -1)
3251 			(void) fprintf(stderr,
3252 			    gettext("write error on sector %lld: %s\n"),
3253 			    bno, strerror(saverr));
3254 		else
3255 			(void) fprintf(stderr, gettext(
3256 			    "short write (%d of %d bytes) on sector %lld\n"),
3257 			    n, size, bno);
3258 		lockexit(32);
3259 	}
3260 }
3261 
3262 /*
3263  * write a block to the file system -- buffered with aio
3264  */
3265 static void
3266 awtfs(diskaddr_t bno, int size, char *bf, int release)
3267 {
3268 	int n;
3269 	aio_trans 	*transp;
3270 	sigset_t 	old_mask;
3271 
3272 	if (fso == -1)
3273 		return;
3274 
3275 	/*
3276 	 * We need to keep things consistent if we get interrupted,
3277 	 * so defer any expected interrupts for the time being.
3278 	 */
3279 	block_sigint(&old_mask);
3280 
3281 	if (Nflag) {
3282 		if (release == RELEASE)
3283 			freebuf(bf);
3284 	} else {
3285 		transp = get_aiop();
3286 		transp->bno = bno;
3287 		transp->buffer = bf;
3288 		transp->size = size;
3289 		transp->release = release;
3290 
3291 		n = aiowrite(fso, bf, size, (off_t)bno * sectorsize,
3292 		    SEEK_SET, &transp->resultbuf);
3293 
3294 		if (n < 0) {
3295 			/*
3296 			 * The aiowrite() may have failed because the
3297 			 * kernel didn't have enough memory to do the job.
3298 			 * Flush all pending writes and try a normal
3299 			 * write().  wtfs_breakup() will call exit if it
3300 			 * fails, so we don't worry about errors here.
3301 			 */
3302 			flush_writes();
3303 			wtfs_breakup(transp->bno, transp->size, transp->buffer);
3304 			freetrans(transp);
3305 		} else {
3306 			/*
3307 			 * Keep track of our pending writes.
3308 			 */
3309 			results.outstanding++;
3310 			if (results.outstanding > results.maxpend)
3311 				results.maxpend = results.outstanding;
3312 		}
3313 	}
3314 
3315 	unblock_sigint(&old_mask);
3316 }
3317 
3318 
3319 /*
3320  * write a block to the file system, but break it up into sbsize
3321  * chunks to avoid forcing a large amount of memory to be locked down.
3322  * Only used as a fallback when an aio write has failed.
3323  */
3324 static void
3325 wtfs_breakup(diskaddr_t bno, int size, char *bf)
3326 {
3327 	int n, saverr;
3328 	int wsize;
3329 	int block_incr = sbsize / sectorsize;
3330 
3331 	if (size < sbsize)
3332 		wsize = size;
3333 	else
3334 		wsize = sbsize;
3335 
3336 	n = 0;
3337 	while (size) {
3338 		/*
3339 		 * Note: the llseek() can succeed, even if the offset is
3340 		 * out of range.  It's not until the file i/o operation
3341 		 * (the write()) that one knows for sure if the raw device
3342 		 * can handle the offset.
3343 		 */
3344 		if (llseek(fso, (offset_t)bno * sectorsize, 0) < 0) {
3345 			saverr = errno;
3346 			(void) fprintf(stderr,
3347 			    gettext("seek error on sector %lld: %s\n"),
3348 			    bno, strerror(saverr));
3349 			lockexit(32);
3350 		}
3351 
3352 		n = write(fso, bf, wsize);
3353 		if (n == -1) {
3354 			saverr = errno;
3355 			(void) fprintf(stderr,
3356 			    gettext("write error on sector %lld: %s\n"),
3357 			    bno, strerror(saverr));
3358 			lockexit(32);
3359 		}
3360 		if (n != wsize) {
3361 			saverr = errno;
3362 			(void) fprintf(stderr, gettext(
3363 			    "short write (%d of %d bytes) on sector %lld\n"),
3364 			    n, size, bno);
3365 			lockexit(32);
3366 		}
3367 
3368 		bno += block_incr;
3369 		bf += wsize;
3370 		size -= wsize;
3371 		if (size < wsize)
3372 			wsize = size;
3373 	}
3374 }
3375 
3376 
3377 /*
3378  * check if a block is available
3379  */
3380 static int
3381 isblock(struct fs *fs, unsigned char *cp, int h)
3382 {
3383 	unsigned char mask;
3384 
3385 	switch (fs->fs_frag) {
3386 	case 8:
3387 		return (cp[h] == 0xff);
3388 	case 4:
3389 		mask = 0x0f << ((h & 0x1) << 2);
3390 		return ((cp[h >> 1] & mask) == mask);
3391 	case 2:
3392 		mask = 0x03 << ((h & 0x3) << 1);
3393 		return ((cp[h >> 2] & mask) == mask);
3394 	case 1:
3395 		mask = 0x01 << (h & 0x7);
3396 		return ((cp[h >> 3] & mask) == mask);
3397 	default:
3398 		(void) fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
3399 		return (0);
3400 	}
3401 }
3402 
3403 /*
3404  * take a block out of the map
3405  */
3406 static void
3407 clrblock(struct fs *fs, unsigned char *cp, int h)
3408 {
3409 	switch ((fs)->fs_frag) {
3410 	case 8:
3411 		cp[h] = 0;
3412 		return;
3413 	case 4:
3414 		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
3415 		return;
3416 	case 2:
3417 		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
3418 		return;
3419 	case 1:
3420 		cp[h >> 3] &= ~(0x01 << (h & 0x7));
3421 		return;
3422 	default:
3423 		(void) fprintf(stderr,
3424 		    gettext("clrblock: bad fs_frag value %d\n"), fs->fs_frag);
3425 		return;
3426 	}
3427 }
3428 
3429 /*
3430  * put a block into the map
3431  */
3432 static void
3433 setblock(struct fs *fs, unsigned char *cp, int h)
3434 {
3435 	switch (fs->fs_frag) {
3436 	case 8:
3437 		cp[h] = 0xff;
3438 		return;
3439 	case 4:
3440 		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
3441 		return;
3442 	case 2:
3443 		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
3444 		return;
3445 	case 1:
3446 		cp[h >> 3] |= (0x01 << (h & 0x7));
3447 		return;
3448 	default:
3449 		(void) fprintf(stderr,
3450 		    gettext("setblock: bad fs_frag value %d\n"), fs->fs_frag);
3451 		return;
3452 	}
3453 }
3454 
3455 static void
3456 usage()
3457 {
3458 	(void) fprintf(stderr,
3459 	    gettext("ufs usage: mkfs [-F FSType] [-V] [-m] [-o options] "
3460 	    "special "				/* param 0 */
3461 	    "size(sectors) \\ \n"));		/* param 1 */
3462 	(void) fprintf(stderr,
3463 	    "[nsect "				/* param 2 */
3464 	    "ntrack "				/* param 3 */
3465 	    "bsize "				/* param 4 */
3466 	    "fragsize "				/* param 5 */
3467 	    "cpg "				/* param 6 */
3468 	    "free "				/* param 7 */
3469 	    "rps "				/* param 8 */
3470 	    "nbpi "				/* param 9 */
3471 	    "opt "				/* param 10 */
3472 	    "apc "				/* param 11 */
3473 	    "gap "				/* param 12 */
3474 	    "nrpos "				/* param 13 */
3475 	    "maxcontig "			/* param 14 */
3476 	    "mtb]\n");				/* param 15 */
3477 	(void) fprintf(stderr,
3478 	    gettext(" -m : dump fs cmd line used to make this partition\n"
3479 	    " -V :print this command line and return\n"
3480 	    " -o :ufs options: :nsect=%d,ntrack=%d,bsize=%d,fragsize=%d\n"
3481 	    " -o :ufs options: :cgsize=%d,free=%d,rps=%d,nbpi=%d,opt=%c\n"
3482 	    " -o :ufs options: :apc=%d,gap=%d,nrpos=%d,maxcontig=%d\n"
3483 	    " -o :ufs options: :mtb=%c,calcsb,calcbinsb\n"
3484 "NOTE that all -o suboptions: must be separated only by commas so as to\n"
3485 "be parsed as a single argument\n"),
3486 	    nsect, ntrack, bsize, fragsize, cpg, sblock.fs_minfree, rps,
3487 	    nbpi, opt, apc, (rotdelay == -1) ? 0 : rotdelay,
3488 	    sblock.fs_nrpos, maxcontig, mtb);
3489 	lockexit(32);
3490 }
3491 
3492 /*ARGSUSED*/
3493 static void
3494 dump_fscmd(char *fsys, int fsi)
3495 {
3496 	int64_t used, bpcg, inospercg;
3497 	int64_t nbpi;
3498 	uint64_t nbytes64;
3499 
3500 	bzero((char *)&sblock, sizeof (sblock));
3501 	rdfs((diskaddr_t)SBLOCK, SBSIZE, (char *)&sblock);
3502 
3503 	/*
3504 	 * ensure a valid file system and if not, exit with error or else
3505 	 * we will end up computing block numbers etc and dividing by zero
3506 	 * which will cause floating point errors in this routine.
3507 	 */
3508 
3509 	if ((sblock.fs_magic != FS_MAGIC) &&
3510 	    (sblock.fs_magic != MTB_UFS_MAGIC)) {
3511 		(void) fprintf(stderr, gettext(
3512 		    "[not currently a valid file system - bad superblock]\n"));
3513 		lockexit(32);
3514 	}
3515 
3516 	if (sblock.fs_magic == FS_MAGIC &&
3517 	    (sblock.fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
3518 	    sblock.fs_version != UFS_VERSION_MIN)) {
3519 		(void) fprintf(stderr, gettext(
3520 		    "Unknown version of UFS format: %d\n"), sblock.fs_version);
3521 		lockexit(32);
3522 	}
3523 
3524 	if (sblock.fs_magic == MTB_UFS_MAGIC &&
3525 	    (sblock.fs_version > MTB_UFS_VERSION_1 ||
3526 	    sblock.fs_version < MTB_UFS_VERSION_MIN)) {
3527 		(void) fprintf(stderr, gettext(
3528 		    "Unknown version of UFS format: %d\n"), sblock.fs_version);
3529 		lockexit(32);
3530 	}
3531 
3532 	/*
3533 	 * Compute a reasonable nbpi value.
3534 	 * The algorithm for "used" is copied from code
3535 	 * in main() verbatim.
3536 	 * The nbpi equation is taken from main where the
3537 	 * fs_ipg value is set for the last time.  The INOPB(...) - 1
3538 	 * is used to account for the roundup.
3539 	 * The problem is that a range of nbpi values map to
3540 	 * the same file system layout.  So it is not possible
3541 	 * to calculate the exact value specified when the file
3542 	 * system was created.  So instead we determine the top
3543 	 * end of the range of values.
3544 	 */
3545 	bpcg = sblock.fs_spc * sectorsize;
3546 	inospercg = (int64_t)roundup(bpcg / sizeof (struct dinode),
3547 	    INOPB(&sblock));
3548 	if (inospercg > MAXIpG(&sblock))
3549 		inospercg = MAXIpG(&sblock);
3550 	used = (int64_t)
3551 	    (sblock.fs_iblkno + inospercg / INOPF(&sblock)) * NSPF(&sblock);
3552 	used *= sectorsize;
3553 	nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
3554 
3555 	/*
3556 	 * The top end of the range of values for nbpi may not be
3557 	 * a valid command line value for mkfs. Report the bottom
3558 	 * end instead.
3559 	 */
3560 	nbpi = (int64_t)(nbytes64 / (sblock.fs_ipg));
3561 
3562 	(void) fprintf(stdout, gettext("mkfs -F ufs -o "), fsys);
3563 	(void) fprintf(stdout, "nsect=%d,ntrack=%d,",
3564 	    sblock.fs_nsect, sblock.fs_ntrak);
3565 	(void) fprintf(stdout, "bsize=%d,fragsize=%d,cgsize=%d,free=%d,",
3566 	    sblock.fs_bsize, sblock.fs_fsize, sblock.fs_cpg, sblock.fs_minfree);
3567 	(void) fprintf(stdout, "rps=%d,nbpi=%lld,opt=%c,apc=%d,gap=%d,",
3568 	    sblock.fs_rps, nbpi, (sblock.fs_optim == FS_OPTSPACE) ? 's' : 't',
3569 	    (sblock.fs_ntrak * sblock.fs_nsect) - sblock.fs_spc,
3570 	    sblock.fs_rotdelay);
3571 	(void) fprintf(stdout, "nrpos=%d,maxcontig=%d,mtb=%c ",
3572 	    sblock.fs_nrpos, sblock.fs_maxcontig,
3573 	    ((sblock.fs_magic == MTB_UFS_MAGIC) ? 'y' : 'n'));
3574 	(void) fprintf(stdout, "%s %lld\n", fsys,
3575 	    fsbtodb(&sblock, sblock.fs_size));
3576 
3577 	bzero((char *)&sblock, sizeof (sblock));
3578 }
3579 
3580 /* number ************************************************************* */
3581 /*									*/
3582 /* Convert a numeric string arg to binary				*/
3583 /*									*/
3584 /* Args:	d_value - default value, if have parse error		*/
3585 /*		param - the name of the argument, for error messages	*/
3586 /*		flags - parser state and what's allowed in the arg	*/
3587 /* Global arg:  string - pointer to command arg				*/
3588 /*									*/
3589 /* Valid forms: 123 | 123k | 123*123 | 123x123				*/
3590 /*									*/
3591 /* Return:	converted number					*/
3592 /*									*/
3593 /* ******************************************************************** */
3594 
3595 static uint64_t
3596 number(uint64_t d_value, char *param, int flags)
3597 {
3598 	char *cs;
3599 	uint64_t n, t;
3600 	uint64_t cut = BIG / 10;    /* limit to avoid overflow */
3601 	int minus = 0;
3602 
3603 	cs = string;
3604 	if (*cs == '-') {
3605 		minus = 1;
3606 		cs += 1;
3607 	}
3608 	if ((*cs < '0') || (*cs > '9')) {
3609 		goto bail_out;
3610 	}
3611 	n = 0;
3612 	while ((*cs >= '0') && (*cs <= '9') && (n <= cut)) {
3613 		n = n*10 + *cs++ - '0';
3614 	}
3615 	if (minus)
3616 		n = -n;
3617 	for (;;) {
3618 		switch (*cs++) {
3619 		case 'k':
3620 			if (flags & ALLOW_END_ONLY)
3621 				goto bail_out;
3622 			if (n > (BIG / 1024))
3623 				goto overflow;
3624 			n *= 1024;
3625 			continue;
3626 
3627 		case '*':
3628 		case 'x':
3629 			if (flags & ALLOW_END_ONLY)
3630 				goto bail_out;
3631 			string = cs;
3632 			t = number(d_value, param, flags);
3633 			if (n > (BIG / t))
3634 				goto overflow;
3635 			n *= t;
3636 			cs = string + 1; /* adjust for -- below */
3637 
3638 			/* recursion has read rest of expression */
3639 			/* FALLTHROUGH */
3640 
3641 		case ',':
3642 		case '\0':
3643 			cs--;
3644 			string = cs;
3645 			return (n);
3646 
3647 		case '%':
3648 			if (flags & ALLOW_END_ONLY)
3649 				goto bail_out;
3650 			if (flags & ALLOW_PERCENT) {
3651 				flags &= ~ALLOW_PERCENT;
3652 				flags |= ALLOW_END_ONLY;
3653 				continue;
3654 			}
3655 			goto bail_out;
3656 
3657 		case 'm':
3658 			if (flags & ALLOW_END_ONLY)
3659 				goto bail_out;
3660 			if (flags & ALLOW_MS1) {
3661 				flags &= ~ALLOW_MS1;
3662 				flags |= ALLOW_MS2;
3663 				continue;
3664 			}
3665 			goto bail_out;
3666 
3667 		case 's':
3668 			if (flags & ALLOW_END_ONLY)
3669 				goto bail_out;
3670 			if (flags & ALLOW_MS2) {
3671 				flags &= ~ALLOW_MS2;
3672 				flags |= ALLOW_END_ONLY;
3673 				continue;
3674 			}
3675 			goto bail_out;
3676 
3677 		case '0': case '1': case '2': case '3': case '4':
3678 		case '5': case '6': case '7': case '8': case '9':
3679 overflow:
3680 			(void) fprintf(stderr,
3681 			    gettext("mkfs: value for %s overflowed\n"),
3682 			    param);
3683 			while ((*cs != '\0') && (*cs != ','))
3684 				cs++;
3685 			string = cs;
3686 			return (BIG);
3687 
3688 		default:
3689 bail_out:
3690 			(void) fprintf(stderr, gettext(
3691 			    "mkfs: bad numeric arg for %s: \"%s\"\n"),
3692 			    param, string);
3693 			while ((*cs != '\0') && (*cs != ','))
3694 				cs++;
3695 			string = cs;
3696 			if (d_value != NO_DEFAULT) {
3697 				(void) fprintf(stderr,
3698 				    gettext("mkfs: %s reset to default %lld\n"),
3699 				    param, d_value);
3700 				return (d_value);
3701 			}
3702 			lockexit(2);
3703 
3704 		}
3705 	} /* never gets here */
3706 }
3707 
3708 /* match ************************************************************** */
3709 /*									*/
3710 /* Compare two text strings for equality				*/
3711 /*									*/
3712 /* Arg:	 s - pointer to string to match with a command arg		*/
3713 /* Global arg:  string - pointer to command arg				*/
3714 /*									*/
3715 /* Return:	1 if match, 0 if no match				*/
3716 /*		If match, also reset `string' to point to the text	*/
3717 /*		that follows the matching text.				*/
3718 /*									*/
3719 /* ******************************************************************** */
3720 
3721 static int
3722 match(char *s)
3723 {
3724 	char *cs;
3725 
3726 	cs = string;
3727 	while (*cs++ == *s) {
3728 		if (*s++ == '\0') {
3729 			goto true;
3730 		}
3731 	}
3732 	if (*s != '\0') {
3733 		return (0);
3734 	}
3735 
3736 true:
3737 	cs--;
3738 	string = cs;
3739 	return (1);
3740 }
3741 
3742 /*
3743  * GROWFS ROUTINES
3744  */
3745 
3746 /* ARGSUSED */
3747 void
3748 lockexit(int exitstatus)
3749 {
3750 	if (Pflag) {
3751 		/* the probe mode neither changes nor locks the filesystem */
3752 		exit(exitstatus);
3753 	}
3754 
3755 	/*
3756 	 * flush the dirty cylinder group
3757 	 */
3758 	if (inlockexit == 0) {
3759 		inlockexit = 1;
3760 		flcg();
3761 	}
3762 
3763 	if (aio_inited) {
3764 		flush_writes();
3765 	}
3766 
3767 	/*
3768 	 * make sure the file system is unlocked before exiting
3769 	 */
3770 	if ((inlockexit == 1) && (!isbad)) {
3771 		inlockexit = 2;
3772 		ulockfs();
3773 		/*
3774 		 * if logging was enabled, then re-enable it
3775 		 */
3776 		if (waslog) {
3777 			if (rl_log_control(fsys, _FIOLOGENABLE) != RL_SUCCESS) {
3778 				(void) fprintf(stderr, gettext(
3779 				    "failed to re-enable logging\n"));
3780 			}
3781 		}
3782 	} else if (grow) {
3783 		if (isbad) {
3784 			(void) fprintf(stderr, gettext(
3785 			    "Filesystem is currently inconsistent.  It "
3786 			    "must be repaired with fsck(1M)\nbefore being "
3787 			    "used.  Use the following command to "
3788 			    "do this:\n\n\tfsck %s\n\n"), fsys);
3789 
3790 			if (ismounted) {
3791 				(void) fprintf(stderr, gettext(
3792 				    "You will be told that the filesystem "
3793 				    "is already mounted, and asked if you\n"
3794 				    "wish to continue.  Answer `yes' to "
3795 				    "this question.\n\n"));
3796 			}
3797 
3798 			(void) fprintf(stderr, gettext(
3799 			    "One problem should be reported, that the summary "
3800 			    "information is bad.\nYou will then be asked if it "
3801 			    "should be salvaged.  Answer `yes' to\nthis "
3802 			    "question.\n\n"));
3803 		}
3804 
3805 		if (ismounted) {
3806 			/*
3807 			 * In theory, there's no way to get here without
3808 			 * isbad also being set, but be robust in the
3809 			 * face of future code changes.
3810 			 */
3811 			(void) fprintf(stderr, gettext(
3812 			    "The filesystem is currently mounted "
3813 			    "read-only and write-locked.  "));
3814 			if (isbad) {
3815 				(void) fprintf(stderr, gettext(
3816 				    "After\nrunning fsck, unlock the "
3817 				    "filesystem and "));
3818 			} else {
3819 				(void) fprintf(stderr, gettext(
3820 				    "Unlock the filesystem\nand "));
3821 			}
3822 
3823 			(void) fprintf(stderr, gettext(
3824 			    "re-enable writing with\nthe following "
3825 			    "command:\n\n\tlockfs -u %s\n\n"), directory);
3826 		}
3827 	}
3828 
3829 	exit(exitstatus);
3830 }
3831 
3832 void
3833 randomgeneration()
3834 {
3835 	int		 i;
3836 	struct dinode	*dp;
3837 
3838 	/*
3839 	 * always perform fsirand(1) function... newfs will notice that
3840 	 * the inodes have been randomized and will not call fsirand itself
3841 	 */
3842 	for (i = 0, dp = zino; i < sblock.fs_inopb; ++i, ++dp)
3843 		IRANDOMIZE(&dp->di_ic);
3844 }
3845 
3846 /*
3847  * Check the size of the summary information.
3848  * Fields in sblock are not changed in this function.
3849  *
3850  * For an 8K filesystem block, the maximum number of cylinder groups is 16384.
3851  *     MAXCSBUFS {32}  *   8K  {FS block size}
3852  *                         divided by (sizeof csum) {16}
3853  *
3854  * Note that MAXCSBUFS is not used in the kernel; as of Solaris 2.6 build 32,
3855  * this is the only place where it's referenced.
3856  */
3857 void
3858 checksummarysize()
3859 {
3860 	diskaddr_t	dmax;
3861 	diskaddr_t	dmin;
3862 	int64_t	cg0frags;
3863 	int64_t	cg0blocks;
3864 	int64_t	maxncg;
3865 	int64_t	maxfrags;
3866 	uint64_t	fs_size;
3867 	uint64_t maxfs_blocks; /* filesystem blocks for max filesystem size */
3868 
3869 	/*
3870 	 * compute the maximum summary info size
3871 	 */
3872 	dmin = cgdmin(&sblock, 0);
3873 	dmax = cgbase(&sblock, 0) + sblock.fs_fpg;
3874 	fs_size = (grow) ? grow_fs_size : sblock.fs_size;
3875 	if (dmax > fs_size)
3876 		dmax = fs_size;
3877 	cg0frags  = dmax - dmin;
3878 	cg0blocks = cg0frags / sblock.fs_frag;
3879 	cg0frags = cg0blocks * sblock.fs_frag;
3880 	maxncg   = (longlong_t)cg0blocks *
3881 	    (longlong_t)(sblock.fs_bsize / sizeof (struct csum));
3882 
3883 	maxfs_blocks = FS_MAX;
3884 
3885 	if (maxncg > ((longlong_t)maxfs_blocks / (longlong_t)sblock.fs_fpg) + 1)
3886 		maxncg = ((longlong_t)maxfs_blocks /
3887 		    (longlong_t)sblock.fs_fpg) + 1;
3888 
3889 	maxfrags = maxncg * (longlong_t)sblock.fs_fpg;
3890 
3891 	if (maxfrags > maxfs_blocks)
3892 		maxfrags = maxfs_blocks;
3893 
3894 
3895 	/*
3896 	 * remember for later processing in extendsummaryinfo()
3897 	 */
3898 	if (test)
3899 		grow_sifrag = dmin + (cg0blocks * sblock.fs_frag);
3900 	if (testfrags == 0)
3901 		testfrags = cg0frags;
3902 	if (testforce)
3903 		if (testfrags > cg0frags) {
3904 			(void) fprintf(stderr,
3905 			    gettext("Too many test frags (%lld); "
3906 			    "try %lld\n"), testfrags, cg0frags);
3907 			lockexit(32);
3908 		}
3909 
3910 	/*
3911 	 * if summary info is too large (too many cg's) tell the user and exit
3912 	 */
3913 	if ((longlong_t)sblock.fs_size > maxfrags) {
3914 		(void) fprintf(stderr, gettext(
3915 		    "Too many cylinder groups with %llu sectors;\n    try "
3916 		    "increasing cgsize, or decreasing fssize to %llu\n"),
3917 		    fsbtodb(&sblock, (uint64_t)sblock.fs_size),
3918 		    fsbtodb(&sblock, (uint64_t)maxfrags));
3919 		lockexit(32);
3920 	}
3921 }
3922 
3923 /*
3924  * checksblock() has two uses:
3925  *	- One is to sanity test the superblock and is used when newfs(1M)
3926  *	  is invoked with the "-N" option. If any discrepancy was found,
3927  *	  just return whatever error was found and do not exit.
3928  *	- the other use of it is in places where you expect the superblock
3929  *	  to be sane, and if it isn't, then we exit.
3930  * Which of the above two actions to take is indicated with the second argument.
3931  */
3932 
3933 int
3934 checksblock(struct fs sb, int proceed)
3935 {
3936 	int err = 0;
3937 	char *errmsg;
3938 
3939 	if ((sb.fs_magic != FS_MAGIC) && (sb.fs_magic != MTB_UFS_MAGIC)) {
3940 		err = 1;
3941 		errmsg = gettext("Bad superblock; magic number wrong\n");
3942 	} else if ((sb.fs_magic == FS_MAGIC &&
3943 	    (sb.fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
3944 	    sb.fs_version != UFS_VERSION_MIN)) ||
3945 	    (sb.fs_magic == MTB_UFS_MAGIC &&
3946 	    (sb.fs_version > MTB_UFS_VERSION_1 ||
3947 	    sb.fs_version < MTB_UFS_VERSION_MIN))) {
3948 		err = 2;
3949 		errmsg = gettext("Unrecognized version of UFS\n");
3950 	} else if (sb.fs_ncg < 1) {
3951 		err = 3;
3952 		errmsg = gettext("Bad superblock; ncg out of range\n");
3953 	} else if (sb.fs_cpg < 1) {
3954 		err = 4;
3955 		errmsg = gettext("Bad superblock; cpg out of range\n");
3956 	} else if (sb.fs_ncg * sb.fs_cpg < sb.fs_ncyl ||
3957 	    (sb.fs_ncg - 1) * sb.fs_cpg >= sb.fs_ncyl) {
3958 		err = 5;
3959 		errmsg = gettext("Bad superblock; ncyl out of range\n");
3960 	} else if (sb.fs_sbsize <= 0 || sb.fs_sbsize > sb.fs_bsize) {
3961 		err = 6;
3962 		errmsg = gettext("Bad superblock; superblock size out of "
3963 		    "range\n");
3964 	}
3965 
3966 	if (proceed) {
3967 		if (err) dprintf(("%s", errmsg));
3968 		return (err);
3969 	}
3970 
3971 	if (err) {
3972 		fprintf(stderr, "%s", errmsg);
3973 		lockexit(32);
3974 	}
3975 	return (32);
3976 }
3977 
3978 /*
3979  * Roll the embedded log, if any, and set up the global variables
3980  * islog, islogok and isufslog.
3981  */
3982 static void
3983 logsetup(char *devstr)
3984 {
3985 	void		*buf, *ud_buf;
3986 	extent_block_t	*ebp;
3987 	ml_unit_t	*ul;
3988 	ml_odunit_t	*ud;
3989 
3990 	/*
3991 	 * Does the superblock indicate that we are supposed to have a log ?
3992 	 */
3993 	if (sblock.fs_logbno == 0) {
3994 		/*
3995 		 * No log present, nothing to do.
3996 		 */
3997 		islogok = 0;
3998 		islog = 0;
3999 		isufslog = 0;
4000 		return;
4001 	} else {
4002 		/*
4003 		 * There's a log in a yet unknown state, attempt to roll it.
4004 		 */
4005 		islog = 1;
4006 		islogok = 0;
4007 		isufslog = 0;
4008 
4009 		/*
4010 		 * We failed to roll the log, bail out.
4011 		 */
4012 		if (rl_roll_log(devstr) != RL_SUCCESS)
4013 			return;
4014 
4015 		isufslog = 1;
4016 
4017 		/* log is not okay; check the fs */
4018 		if ((FSOKAY != (sblock.fs_state + sblock.fs_time)) ||
4019 		    (sblock.fs_clean != FSLOG))
4020 			return;
4021 
4022 		/* get the log allocation block */
4023 		buf = (void *)malloc(DEV_BSIZE);
4024 		if (buf == (void *) NULL)
4025 			return;
4026 
4027 		ud_buf = (void *)malloc(DEV_BSIZE);
4028 		if (ud_buf == (void *) NULL) {
4029 			free(buf);
4030 			return;
4031 		}
4032 
4033 		rdfs((diskaddr_t)logbtodb(&sblock, sblock.fs_logbno),
4034 		    DEV_BSIZE, buf);
4035 		ebp = (extent_block_t *)buf;
4036 
4037 		/* log allocation block is not okay; check the fs */
4038 		if (ebp->type != LUFS_EXTENTS) {
4039 			free(buf);
4040 			free(ud_buf);
4041 			return;
4042 		}
4043 
4044 		/* get the log state block(s) */
4045 		rdfs((diskaddr_t)logbtodb(&sblock, ebp->extents[0].pbno),
4046 		    DEV_BSIZE, ud_buf);
4047 		ud = (ml_odunit_t *)ud_buf;
4048 		ul = (ml_unit_t *)malloc(sizeof (*ul));
4049 		ul->un_ondisk = *ud;
4050 
4051 		/* log state is okay */
4052 		if ((ul->un_chksum == ul->un_head_ident + ul->un_tail_ident) &&
4053 		    (ul->un_version == LUFS_VERSION_LATEST) &&
4054 		    (ul->un_badlog == 0))
4055 			islogok = 1;
4056 		free(ud_buf);
4057 		free(buf);
4058 		free(ul);
4059 	}
4060 }
4061 
4062 void
4063 growinit(char *devstr)
4064 {
4065 	int	i;
4066 	char	buf[DEV_BSIZE];
4067 
4068 	/*
4069 	 * Read and verify the superblock
4070 	 */
4071 	rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
4072 	(void) checksblock(sblock, 0);
4073 	if (sblock.fs_postblformat != FS_DYNAMICPOSTBLFMT) {
4074 		(void) fprintf(stderr,
4075 		    gettext("old file system format; can't growfs\n"));
4076 		lockexit(32);
4077 	}
4078 
4079 	/*
4080 	 * can't shrink a file system
4081 	 */
4082 	grow_fssize = fsbtodb(&sblock, (uint64_t)sblock.fs_size);
4083 	if (fssize_db < grow_fssize) {
4084 		(void) fprintf(stderr,
4085 		    gettext("%lld sectors < current size of %lld sectors\n"),
4086 		    fssize_db, grow_fssize);
4087 		lockexit(32);
4088 	}
4089 
4090 	/*
4091 	 * can't grow a system to over a terabyte unless it was set up
4092 	 * as an MTB UFS file system.
4093 	 */
4094 	if (mtb == 'y' && sblock.fs_magic != MTB_UFS_MAGIC) {
4095 		if (fssize_db >= SECTORS_PER_TERABYTE) {
4096 			(void) fprintf(stderr, gettext(
4097 "File system was not set up with the multi-terabyte format.\n"));
4098 			(void) fprintf(stderr, gettext(
4099 "Its size cannot be increased to a terabyte or more.\n"));
4100 		} else {
4101 			(void) fprintf(stderr, gettext(
4102 "Cannot convert file system to multi-terabyte format.\n"));
4103 		}
4104 		lockexit(32);
4105 	}
4106 
4107 	logsetup(devstr);
4108 
4109 	/*
4110 	 * can't growfs when logging device has errors
4111 	 */
4112 	if ((islog && !islogok) ||
4113 	    ((FSOKAY == (sblock.fs_state + sblock.fs_time)) &&
4114 	    (sblock.fs_clean == FSLOG && !islog))) {
4115 		(void) fprintf(stderr,
4116 		    gettext("logging device has errors; can't growfs\n"));
4117 		lockexit(32);
4118 	}
4119 
4120 	/*
4121 	 * disable ufs logging for growing
4122 	 */
4123 	if (isufslog) {
4124 		if (rl_log_control(devstr, _FIOLOGDISABLE) != RL_SUCCESS) {
4125 			(void) fprintf(stderr, gettext(
4126 			    "failed to disable logging\n"));
4127 			lockexit(32);
4128 		}
4129 		islog = 0;
4130 		waslog = 1;
4131 	}
4132 
4133 	/*
4134 	 * if mounted write lock the file system to be grown
4135 	 */
4136 	if (ismounted)
4137 		wlockfs();
4138 
4139 	/*
4140 	 * refresh dynamic superblock state - disabling logging will have
4141 	 * changed the amount of free space available in the file system
4142 	 */
4143 	rdfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
4144 
4145 	/*
4146 	 * make sure device is big enough
4147 	 */
4148 	rdfs((diskaddr_t)fssize_db - 1, DEV_BSIZE, buf);
4149 	wtfs((diskaddr_t)fssize_db - 1, DEV_BSIZE, buf);
4150 
4151 	/*
4152 	 * read current summary information
4153 	 */
4154 	grow_fscs = read_summaryinfo(&sblock);
4155 
4156 	/*
4157 	 * save some current size related fields from the superblock
4158 	 * These are used in extendsummaryinfo()
4159 	 */
4160 	grow_fs_size	= sblock.fs_size;
4161 	grow_fs_ncg	= sblock.fs_ncg;
4162 	grow_fs_csaddr	= (diskaddr_t)sblock.fs_csaddr;
4163 	grow_fs_cssize	= sblock.fs_cssize;
4164 
4165 	/*
4166 	 * save and reset the clean flag
4167 	 */
4168 	if (FSOKAY == (sblock.fs_state + sblock.fs_time))
4169 		grow_fs_clean = sblock.fs_clean;
4170 	else
4171 		grow_fs_clean = FSBAD;
4172 	sblock.fs_clean = FSBAD;
4173 	sblock.fs_state = FSOKAY - sblock.fs_time;
4174 	isbad = 1;
4175 	wtfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
4176 }
4177 
4178 void
4179 checkdev(char *rdev, char *bdev)
4180 {
4181 	struct stat64	statarea;
4182 
4183 	if (stat64(bdev, &statarea) < 0) {
4184 		(void) fprintf(stderr, gettext("can't check mount point; "));
4185 		(void) fprintf(stderr, gettext("can't stat %s\n"), bdev);
4186 		lockexit(32);
4187 	}
4188 	if ((statarea.st_mode & S_IFMT) != S_IFBLK) {
4189 		(void) fprintf(stderr, gettext(
4190 		    "can't check mount point; %s is not a block device\n"),
4191 		    bdev);
4192 		lockexit(32);
4193 	}
4194 	if (stat64(rdev, &statarea) < 0) {
4195 		(void) fprintf(stderr, gettext("can't stat %s\n"), rdev);
4196 		lockexit(32);
4197 	}
4198 	if ((statarea.st_mode & S_IFMT) != S_IFCHR) {
4199 		(void) fprintf(stderr,
4200 		    gettext("%s is not a character device\n"), rdev);
4201 		lockexit(32);
4202 	}
4203 }
4204 
4205 void
4206 checkmount(struct mnttab *mntp, char *bdevname)
4207 {
4208 	struct stat64	statdir;
4209 	struct stat64	statdev;
4210 
4211 	if (strcmp(bdevname, mntp->mnt_special) == 0) {
4212 		if (stat64(mntp->mnt_mountp, &statdir) == -1) {
4213 			(void) fprintf(stderr, gettext("can't stat %s\n"),
4214 			    mntp->mnt_mountp);
4215 			lockexit(32);
4216 		}
4217 		if (stat64(mntp->mnt_special, &statdev) == -1) {
4218 			(void) fprintf(stderr, gettext("can't stat %s\n"),
4219 			    mntp->mnt_special);
4220 			lockexit(32);
4221 		}
4222 		if (statdir.st_dev != statdev.st_rdev) {
4223 			(void) fprintf(stderr, gettext(
4224 			    "%s is not mounted on %s; mnttab(4) wrong\n"),
4225 			    mntp->mnt_special, mntp->mnt_mountp);
4226 			lockexit(32);
4227 		}
4228 		ismounted = 1;
4229 		if (directory) {
4230 			if (strcmp(mntp->mnt_mountp, directory) != 0) {
4231 				(void) fprintf(stderr,
4232 				    gettext("%s is mounted on %s, not %s\n"),
4233 				    bdevname, mntp->mnt_mountp, directory);
4234 				lockexit(32);
4235 			}
4236 		} else {
4237 			if (grow)
4238 				(void) fprintf(stderr, gettext(
4239 				    "%s is mounted on %s; can't growfs\n"),
4240 				    bdevname, mntp->mnt_mountp);
4241 			else
4242 				(void) fprintf(stderr,
4243 				    gettext("%s is mounted, can't mkfs\n"),
4244 				    bdevname);
4245 			lockexit(32);
4246 		}
4247 	}
4248 }
4249 
4250 struct dinode	*dibuf	= 0;
4251 diskaddr_t	difrag	= 0;
4252 
4253 struct dinode *
4254 gdinode(ino_t ino)
4255 {
4256 	/*
4257 	 * read the block of inodes containing inode number ino
4258 	 */
4259 	if (dibuf == 0)
4260 		dibuf = (struct dinode *)malloc((unsigned)sblock.fs_bsize);
4261 	if (itod(&sblock, ino) != difrag) {
4262 		difrag = itod(&sblock, ino);
4263 		rdfs(fsbtodb(&sblock, (uint64_t)difrag), (int)sblock.fs_bsize,
4264 		    (char *)dibuf);
4265 	}
4266 	return (dibuf + (ino % INOPB(&sblock)));
4267 }
4268 
4269 /*
4270  * structure that manages the frags we need for extended summary info
4271  *	These frags can be:
4272  *		free
4273  *		data  block
4274  *		alloc block
4275  */
4276 struct csfrag {
4277 	struct csfrag	*next;		/* next entry */
4278 	daddr32_t	 ofrag;		/* old frag */
4279 	daddr32_t	 nfrag;		/* new frag */
4280 	long		 cylno;		/* cylno of nfrag */
4281 	long		 frags;		/* number of frags */
4282 	long		 size;		/* size in bytes */
4283 	ino_t		 ino;		/* inode number */
4284 	long		 fixed;		/* Boolean - Already fixed? */
4285 };
4286 struct csfrag	*csfrag;		/* state unknown */
4287 struct csfrag	*csfragino;		/* frags belonging to an inode */
4288 struct csfrag	*csfragfree;		/* frags that are free */
4289 
4290 daddr32_t maxcsfrag	= 0;		/* maximum in range */
4291 daddr32_t mincsfrag	= 0x7fffffff;	/* minimum in range */
4292 
4293 int
4294 csfraginrange(daddr32_t frag)
4295 {
4296 	return ((frag >= mincsfrag) && (frag <= maxcsfrag));
4297 }
4298 
4299 struct csfrag *
4300 findcsfrag(daddr32_t frag, struct csfrag **cfap)
4301 {
4302 	struct csfrag	*cfp;
4303 
4304 	if (!csfraginrange(frag))
4305 		return (NULL);
4306 
4307 	for (cfp = *cfap; cfp; cfp = cfp->next)
4308 		if (cfp->ofrag == frag)
4309 			return (cfp);
4310 	return (NULL);
4311 }
4312 
4313 void
4314 checkindirect(ino_t ino, daddr32_t *fragsp, daddr32_t frag, int level)
4315 {
4316 	int			i;
4317 	int			ne	= sblock.fs_bsize / sizeof (daddr32_t);
4318 	daddr32_t			fsb[MAXBSIZE / sizeof (daddr32_t)];
4319 
4320 	if (frag == 0)
4321 		return;
4322 
4323 	rdfs(fsbtodb(&sblock, frag), (int)sblock.fs_bsize,
4324 	    (char *)fsb);
4325 
4326 	checkdirect(ino, fragsp, fsb, sblock.fs_bsize / sizeof (daddr32_t));
4327 
4328 	if (level)
4329 		for (i = 0; i < ne && *fragsp; ++i)
4330 			checkindirect(ino, fragsp, fsb[i], level-1);
4331 }
4332 
4333 void
4334 addcsfrag(ino_t ino, daddr32_t frag, struct csfrag **cfap)
4335 {
4336 	struct csfrag	*cfp, *curr, *prev;
4337 
4338 	/*
4339 	 * establish a range for faster checking in csfraginrange()
4340 	 */
4341 	if (frag > maxcsfrag)
4342 		maxcsfrag = frag;
4343 	if (frag < mincsfrag)
4344 		mincsfrag = frag;
4345 
4346 	/*
4347 	 * if this frag belongs to an inode and is not the start of a block
4348 	 *	then see if it is part of a frag range for this inode
4349 	 */
4350 	if (ino && (frag % sblock.fs_frag))
4351 		for (cfp = *cfap; cfp; cfp = cfp->next) {
4352 			if (ino != cfp->ino)
4353 				continue;
4354 			if (frag != cfp->ofrag + cfp->frags)
4355 				continue;
4356 			cfp->frags++;
4357 			cfp->size += sblock.fs_fsize;
4358 			return;
4359 		}
4360 	/*
4361 	 * allocate a csfrag entry and insert it in an increasing order into the
4362 	 * specified list
4363 	 */
4364 	cfp = (struct csfrag *)calloc(1, sizeof (struct csfrag));
4365 	cfp->ino	= ino;
4366 	cfp->ofrag	= frag;
4367 	cfp->frags	= 1;
4368 	cfp->size	= sblock.fs_fsize;
4369 	for (prev = NULL, curr = *cfap; curr != NULL;
4370 	    prev = curr, curr = curr->next) {
4371 		if (frag < curr->ofrag) {
4372 			cfp->next = curr;
4373 			if (prev)
4374 				prev->next = cfp;	/* middle element */
4375 			else
4376 				*cfap = cfp;		/* first element */
4377 			break;
4378 		}
4379 		if (curr->next == NULL) {
4380 			curr->next = cfp;		/* last element	*/
4381 			break;
4382 		}
4383 	}
4384 	if (*cfap == NULL)	/* will happen only once */
4385 		*cfap = cfp;
4386 }
4387 
4388 void
4389 delcsfrag(daddr32_t frag, struct csfrag **cfap)
4390 {
4391 	struct csfrag	*cfp;
4392 	struct csfrag	**cfpp;
4393 
4394 	/*
4395 	 * free up entry whose beginning frag matches
4396 	 */
4397 	for (cfpp = cfap; *cfpp; cfpp = &(*cfpp)->next) {
4398 		if (frag == (*cfpp)->ofrag) {
4399 			cfp = *cfpp;
4400 			*cfpp = (*cfpp)->next;
4401 			free((char *)cfp);
4402 			return;
4403 		}
4404 	}
4405 }
4406 
4407 /*
4408  * See whether any of the direct blocks in the array pointed by "db" and of
4409  * length "ne" are within the range of frags needed to extend the cylinder
4410  * summary. If so, remove those frags from the "as-yet-unclassified" list
4411  * (csfrag) and add them to the "owned-by-inode" list (csfragino).
4412  * For each such frag found, decrement the frag count pointed to by fragsp.
4413  * "ino" is the inode that contains (either directly or indirectly) the frags
4414  * being checked.
4415  */
4416 void
4417 checkdirect(ino_t ino, daddr32_t *fragsp, daddr32_t *db, int ne)
4418 {
4419 	int	 i;
4420 	int	 j;
4421 	int	 found;
4422 	diskaddr_t	 frag;
4423 
4424 	/*
4425 	 * scan for allocation within the new summary info range
4426 	 */
4427 	for (i = 0; i < ne && *fragsp; ++i) {
4428 		if ((frag = *db++) != 0) {
4429 			found = 0;
4430 			for (j = 0; j < sblock.fs_frag && *fragsp; ++j) {
4431 				if (found || (found = csfraginrange(frag))) {
4432 					addcsfrag(ino, frag, &csfragino);
4433 					delcsfrag(frag, &csfrag);
4434 				}
4435 				++frag;
4436 				--(*fragsp);
4437 			}
4438 		}
4439 	}
4440 }
4441 
4442 void
4443 findcsfragino()
4444 {
4445 	int		 i;
4446 	int		 j;
4447 	daddr32_t		 frags;
4448 	struct dinode	*dp;
4449 
4450 	/*
4451 	 * scan all old inodes looking for allocations in the new
4452 	 * summary info range.  Move the affected frag from the
4453 	 * generic csfrag list onto the `owned-by-inode' list csfragino.
4454 	 */
4455 	for (i = UFSROOTINO; i < grow_fs_ncg*sblock.fs_ipg && csfrag; ++i) {
4456 		dp = gdinode((ino_t)i);
4457 		switch (dp->di_mode & IFMT) {
4458 			case IFSHAD	:
4459 			case IFLNK 	:
4460 			case IFDIR 	:
4461 			case IFREG 	: break;
4462 			default		: continue;
4463 		}
4464 
4465 		frags   = dbtofsb(&sblock, dp->di_blocks);
4466 
4467 		checkdirect((ino_t)i, &frags, &dp->di_db[0], NDADDR+NIADDR);
4468 		for (j = 0; j < NIADDR && frags; ++j) {
4469 			/* Negate the block if its an fallocate'd block */
4470 			if (dp->di_ib[j] < 0 && dp->di_ib[j] != UFS_HOLE)
4471 				checkindirect((ino_t)i, &frags,
4472 				    -(dp->di_ib[j]), j);
4473 			else
4474 				checkindirect((ino_t)i, &frags,
4475 				    dp->di_ib[j], j);
4476 		}
4477 	}
4478 }
4479 
4480 void
4481 fixindirect(daddr32_t frag, int level)
4482 {
4483 	int			 i;
4484 	int			 ne	= sblock.fs_bsize / sizeof (daddr32_t);
4485 	daddr32_t			fsb[MAXBSIZE / sizeof (daddr32_t)];
4486 
4487 	if (frag == 0)
4488 		return;
4489 
4490 	rdfs(fsbtodb(&sblock, (uint64_t)frag), (int)sblock.fs_bsize,
4491 	    (char *)fsb);
4492 
4493 	fixdirect((caddr_t)fsb, frag, fsb, ne);
4494 
4495 	if (level)
4496 		for (i = 0; i < ne; ++i)
4497 			fixindirect(fsb[i], level-1);
4498 }
4499 
4500 void
4501 fixdirect(caddr_t bp, daddr32_t frag, daddr32_t *db, int ne)
4502 {
4503 	int	 i;
4504 	struct csfrag	*cfp;
4505 
4506 	for (i = 0; i < ne; ++i, ++db) {
4507 		if (*db == 0)
4508 			continue;
4509 		if ((cfp = findcsfrag(*db, &csfragino)) == NULL)
4510 			continue;
4511 		*db = cfp->nfrag;
4512 		cfp->fixed = 1;
4513 		wtfs(fsbtodb(&sblock, (uint64_t)frag), (int)sblock.fs_bsize,
4514 		    bp);
4515 	}
4516 }
4517 
4518 void
4519 fixcsfragino()
4520 {
4521 	int		 i;
4522 	struct dinode	*dp;
4523 	struct csfrag	*cfp;
4524 
4525 	for (cfp = csfragino; cfp; cfp = cfp->next) {
4526 		if (cfp->fixed)
4527 			continue;
4528 		dp = gdinode((ino_t)cfp->ino);
4529 		fixdirect((caddr_t)dibuf, difrag, dp->di_db, NDADDR+NIADDR);
4530 		for (i = 0; i < NIADDR; ++i)
4531 			fixindirect(dp->di_ib[i], i);
4532 	}
4533 }
4534 
4535 /*
4536  * Read the cylinders summary information specified by settings in the
4537  * passed 'fs' structure into a new allocated array of csum structures.
4538  * The caller is responsible for freeing the returned array.
4539  * Return a pointer to an array of csum structures.
4540  */
4541 static struct csum *
4542 read_summaryinfo(struct	fs *fsp)
4543 {
4544 	struct csum 	*csp;
4545 	int		i;
4546 
4547 	if ((csp = malloc((size_t)fsp->fs_cssize)) == NULL) {
4548 		(void) fprintf(stderr, gettext("cannot create csum list,"
4549 		    " not enough memory\n"));
4550 		exit(32);
4551 	}
4552 
4553 	for (i = 0; i < fsp->fs_cssize; i += fsp->fs_bsize) {
4554 		rdfs(fsbtodb(fsp,
4555 		    (uint64_t)(fsp->fs_csaddr + numfrags(fsp, i))),
4556 		    (int)(fsp->fs_cssize - i < fsp->fs_bsize ?
4557 		    fsp->fs_cssize - i : fsp->fs_bsize), ((caddr_t)csp) + i);
4558 	}
4559 
4560 	return (csp);
4561 }
4562 
4563 /*
4564  * Check the allocation of fragments that are to be made part of a csum block.
4565  * A fragment is allocated if it is either in the csfragfree list or, it is
4566  * in the csfragino list and has new frags associated with it.
4567  * Return the number of allocated fragments.
4568  */
4569 int64_t
4570 checkfragallocated(daddr32_t frag)
4571 {
4572 	struct 	csfrag	*cfp;
4573 	/*
4574 	 * Since the lists are sorted we can break the search if the asked
4575 	 * frag is smaller then the one in the list.
4576 	 */
4577 	for (cfp = csfragfree; cfp != NULL && frag >= cfp->ofrag;
4578 	    cfp = cfp->next) {
4579 		if (frag == cfp->ofrag)
4580 			return (1);
4581 	}
4582 	for (cfp = csfragino; cfp != NULL && frag >= cfp->ofrag;
4583 	    cfp = cfp->next) {
4584 		if (frag == cfp->ofrag && cfp->nfrag != 0)
4585 			return (cfp->frags);
4586 	}
4587 
4588 	return (0);
4589 }
4590 
4591 /*
4592  * Figure out how much the filesystem can be grown. The limiting factor is
4593  * the available free space needed to extend the cg summary info block.
4594  * The free space is determined in three steps:
4595  * - Try to extend the cg summary block to the required size.
4596  * - Find free blocks in last cg.
4597  * - Find free space in the last already allocated fragment of the summary info
4598  *   block, and use it for additional csum structures.
4599  * Return the maximum size of the new filesystem or 0 if it can't be grown.
4600  * Please note that this function leaves the global list pointers csfrag,
4601  * csfragfree, and csfragino initialized, and the caller is responsible for
4602  * freeing the lists.
4603  */
4604 diskaddr_t
4605 probe_summaryinfo()
4606 {
4607 	/* fragments by which the csum block can be extended. */
4608 	int64_t 	growth_csum_frags = 0;
4609 	/* fragments by which the filesystem can be extended. */
4610 	int64_t		growth_fs_frags = 0;
4611 	int64_t		new_fs_cssize;	/* size of csum blk in the new FS */
4612 	int64_t		new_fs_ncg;	/* number of cg in the new FS */
4613 	int64_t 	spare_csum;
4614 	daddr32_t	oldfrag_daddr;
4615 	daddr32_t	newfrag_daddr;
4616 	daddr32_t	daddr;
4617 	int		i;
4618 
4619 	/*
4620 	 * read and verify the superblock
4621 	 */
4622 	rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
4623 	(void) checksblock(sblock, 0);
4624 
4625 	/*
4626 	 * check how much we can extend the cg summary info block
4627 	 */
4628 
4629 	/*
4630 	 * read current summary information
4631 	 */
4632 	fscs = read_summaryinfo(&sblock);
4633 
4634 	/*
4635 	 * build list of frags needed for cg summary info block extension
4636 	 */
4637 	oldfrag_daddr = howmany(sblock.fs_cssize, sblock.fs_fsize) +
4638 	    sblock.fs_csaddr;
4639 	new_fs_ncg = howmany(dbtofsb(&sblock, fssize_db), sblock.fs_fpg);
4640 	new_fs_cssize = fragroundup(&sblock, new_fs_ncg * sizeof (struct csum));
4641 	newfrag_daddr = howmany(new_fs_cssize, sblock.fs_fsize) +
4642 	    sblock.fs_csaddr;
4643 	/*
4644 	 * add all of the frags that are required to grow the cyl summary to the
4645 	 * csfrag list, which is the generic/unknown list, since at this point
4646 	 * we don't yet know the state of those frags.
4647 	 */
4648 	for (daddr = oldfrag_daddr; daddr < newfrag_daddr; daddr++)
4649 		addcsfrag((ino_t)0, daddr, &csfrag);
4650 
4651 	/*
4652 	 * filter free fragments and allocate them. Note that the free frags
4653 	 * must be allocated first otherwise they could be grabbed by
4654 	 * alloccsfragino() for data frags.
4655 	 */
4656 	findcsfragfree();
4657 	alloccsfragfree();
4658 
4659 	/*
4660 	 * filter fragments owned by inodes and allocate them
4661 	 */
4662 	grow_fs_ncg = sblock.fs_ncg; /* findcsfragino() needs this glob. var. */
4663 	findcsfragino();
4664 	alloccsfragino();
4665 
4666 	if (notenoughspace()) {
4667 		/*
4668 		 * check how many consecutive fragments could be allocated
4669 		 * in both lists.
4670 		 */
4671 		int64_t tmp_frags;
4672 		for (daddr = oldfrag_daddr; daddr < newfrag_daddr;
4673 		    daddr += tmp_frags) {
4674 			if ((tmp_frags = checkfragallocated(daddr)) > 0)
4675 				growth_csum_frags += tmp_frags;
4676 			else
4677 				break;
4678 		}
4679 	} else {
4680 		/*
4681 		 * We have all we need for the new desired size,
4682 		 * so clean up and report back.
4683 		 */
4684 		return (fssize_db);
4685 	}
4686 
4687 	/*
4688 	 * given the number of fragments by which the csum block can be grown
4689 	 * compute by how many new fragments the FS can be increased.
4690 	 * It is the number of csum instances per fragment multiplied by
4691 	 * `growth_csum_frags' and the number of fragments per cylinder group.
4692 	 */
4693 	growth_fs_frags = howmany(sblock.fs_fsize, sizeof (struct csum)) *
4694 	    growth_csum_frags * sblock.fs_fpg;
4695 
4696 	/*
4697 	 * compute free fragments in the last cylinder group
4698 	 */
4699 	rdcg(sblock.fs_ncg - 1);
4700 	growth_fs_frags += sblock.fs_fpg - acg.cg_ndblk;
4701 
4702 	/*
4703 	 * compute how many csum instances are unused in the old csum block.
4704 	 * For each unused csum instance the FS can be grown by one cylinder
4705 	 * group without extending the csum block.
4706 	 */
4707 	spare_csum = howmany(sblock.fs_cssize, sizeof (struct csum)) -
4708 	    sblock.fs_ncg;
4709 	if (spare_csum > 0)
4710 		growth_fs_frags += spare_csum * sblock.fs_fpg;
4711 
4712 	/*
4713 	 * recalculate the new filesystem size in sectors, shorten it by
4714 	 * the requested size `fssize_db' if necessary.
4715 	 */
4716 	if (growth_fs_frags > 0) {
4717 		diskaddr_t sect;
4718 		sect = (sblock.fs_size + growth_fs_frags) * sblock.fs_nspf;
4719 		return ((sect > fssize_db) ? fssize_db : sect);
4720 	}
4721 
4722 	return (0);
4723 }
4724 
4725 void
4726 extendsummaryinfo()
4727 {
4728 	int64_t		i;
4729 	int		localtest	= test;
4730 	int64_t		frags;
4731 	daddr32_t		oldfrag;
4732 	daddr32_t		newfrag;
4733 
4734 	/*
4735 	 * if no-write (-N), don't bother
4736 	 */
4737 	if (Nflag)
4738 		return;
4739 
4740 again:
4741 	flcg();
4742 	/*
4743 	 * summary info did not change size -- do nothing unless in test mode
4744 	 */
4745 	if (grow_fs_cssize == sblock.fs_cssize)
4746 		if (!localtest)
4747 			return;
4748 
4749 	/*
4750 	 * build list of frags needed for additional summary information
4751 	 */
4752 	oldfrag = howmany(grow_fs_cssize, sblock.fs_fsize) + grow_fs_csaddr;
4753 	newfrag = howmany(sblock.fs_cssize, sblock.fs_fsize) + grow_fs_csaddr;
4754 	/*
4755 	 * add all of the frags that are required to grow the cyl summary to the
4756 	 * csfrag list, which is the generic/unknown list, since at this point
4757 	 * we don't yet know the state of those frags.
4758 	 */
4759 	for (i = oldfrag, frags = 0; i < newfrag; ++i, ++frags)
4760 		addcsfrag((ino_t)0, (diskaddr_t)i, &csfrag);
4761 	/*
4762 	 * reduce the number of data blocks in the file system (fs_dsize) by
4763 	 * the number of frags that need to be added to the cyl summary
4764 	 */
4765 	sblock.fs_dsize -= (newfrag - oldfrag);
4766 
4767 	/*
4768 	 * In test mode, we move more data than necessary from
4769 	 * cylinder group 0.  The lookup/allocate/move code can be
4770 	 * better stressed without having to create HUGE file systems.
4771 	 */
4772 	if (localtest)
4773 		for (i = newfrag; i < grow_sifrag; ++i) {
4774 			if (frags >= testfrags)
4775 				break;
4776 			frags++;
4777 			addcsfrag((ino_t)0, (diskaddr_t)i, &csfrag);
4778 		}
4779 
4780 	/*
4781 	 * move frags to free or inode lists, depending on owner
4782 	 */
4783 	findcsfragfree();
4784 	findcsfragino();
4785 
4786 	/*
4787 	 * if not all frags can be located, file system must be inconsistent
4788 	 */
4789 	if (csfrag) {
4790 		isbad = 1;	/* should already be set, but make sure */
4791 		lockexit(32);
4792 	}
4793 
4794 	/*
4795 	 * allocate the free frags. Note that the free frags must be allocated
4796 	 * first otherwise they could be grabbed by alloccsfragino() for data
4797 	 * frags.
4798 	 */
4799 	alloccsfragfree();
4800 	/*
4801 	 * allocate extra space for inode frags
4802 	 */
4803 	alloccsfragino();
4804 
4805 	/*
4806 	 * not enough space
4807 	 */
4808 	if (notenoughspace()) {
4809 		unalloccsfragfree();
4810 		unalloccsfragino();
4811 		if (localtest && !testforce) {
4812 			localtest = 0;
4813 			goto again;
4814 		}
4815 		(void) fprintf(stderr, gettext("Not enough free space\n"));
4816 		lockexit(NOTENOUGHSPACE);
4817 	}
4818 
4819 	/*
4820 	 * copy the data from old frags to new frags
4821 	 */
4822 	copycsfragino();
4823 
4824 	/*
4825 	 * fix the inodes to point to the new frags
4826 	 */
4827 	fixcsfragino();
4828 
4829 	/*
4830 	 * We may have moved more frags than we needed.  Free them.
4831 	 */
4832 	rdcg((long)0);
4833 	for (i = newfrag; i <= maxcsfrag; ++i)
4834 		setbit(cg_blksfree(&acg), i-cgbase(&sblock, 0));
4835 	wtcg();
4836 
4837 	flcg();
4838 }
4839 
4840 /*
4841  * Check if all fragments in the `csfragino' list were reallocated.
4842  */
4843 int
4844 notenoughspace()
4845 {
4846 	struct csfrag	*cfp;
4847 
4848 	/*
4849 	 * If any element in the csfragino array has a "new frag location"
4850 	 * of 0, the allocfrags() function was unsuccessful in allocating
4851 	 * space for moving the frag represented by this array element.
4852 	 */
4853 	for (cfp = csfragino; cfp; cfp = cfp->next)
4854 		if (cfp->nfrag == 0)
4855 			return (1);
4856 	return (0);
4857 }
4858 
4859 void
4860 unalloccsfragino()
4861 {
4862 	struct csfrag	*cfp;
4863 
4864 	while ((cfp = csfragino) != NULL) {
4865 		if (cfp->nfrag)
4866 			freefrags(cfp->nfrag, cfp->frags, cfp->cylno);
4867 		delcsfrag(cfp->ofrag, &csfragino);
4868 	}
4869 }
4870 
4871 void
4872 unalloccsfragfree()
4873 {
4874 	struct csfrag	*cfp;
4875 
4876 	while ((cfp = csfragfree) != NULL) {
4877 		freefrags(cfp->ofrag, cfp->frags, cfp->cylno);
4878 		delcsfrag(cfp->ofrag, &csfragfree);
4879 	}
4880 }
4881 
4882 /*
4883  * For each frag in the "as-yet-unclassified" list (csfrag), see if
4884  * it's free (i.e., its bit is set in the free frag bit map).  If so,
4885  * move it from the "as-yet-unclassified" list to the csfragfree list.
4886  */
4887 void
4888 findcsfragfree()
4889 {
4890 	struct csfrag	*cfp;
4891 	struct csfrag	*cfpnext;
4892 
4893 	/*
4894 	 * move free frags onto the free-frag list
4895 	 */
4896 	rdcg((long)0);
4897 	for (cfp = csfrag; cfp; cfp = cfpnext) {
4898 		cfpnext = cfp->next;
4899 		if (isset(cg_blksfree(&acg), cfp->ofrag - cgbase(&sblock, 0))) {
4900 			addcsfrag(cfp->ino, cfp->ofrag, &csfragfree);
4901 			delcsfrag(cfp->ofrag, &csfrag);
4902 		}
4903 	}
4904 }
4905 
4906 void
4907 copycsfragino()
4908 {
4909 	struct csfrag	*cfp;
4910 	char		buf[MAXBSIZE];
4911 
4912 	/*
4913 	 * copy data from old frags to newly allocated frags
4914 	 */
4915 	for (cfp = csfragino; cfp; cfp = cfp->next) {
4916 		rdfs(fsbtodb(&sblock, (uint64_t)cfp->ofrag), (int)cfp->size,
4917 		    buf);
4918 		wtfs(fsbtodb(&sblock, (uint64_t)cfp->nfrag), (int)cfp->size,
4919 		    buf);
4920 	}
4921 }
4922 
4923 long	curcylno	= -1;
4924 int	cylnodirty	= 0;
4925 
4926 void
4927 rdcg(long cylno)
4928 {
4929 	if (cylno != curcylno) {
4930 		flcg();
4931 		curcylno = cylno;
4932 		rdfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, curcylno)),
4933 		    (int)sblock.fs_cgsize, (char *)&acg);
4934 	}
4935 }
4936 
4937 void
4938 flcg()
4939 {
4940 	if (cylnodirty) {
4941 		if (debug && Pflag) {
4942 			(void) fprintf(stderr,
4943 			    "Assert: cylnodirty set in probe mode\n");
4944 			return;
4945 		}
4946 		resetallocinfo();
4947 		wtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, curcylno)),
4948 		    (int)sblock.fs_cgsize, (char *)&acg);
4949 		cylnodirty = 0;
4950 	}
4951 	curcylno = -1;
4952 }
4953 
4954 void
4955 wtcg()
4956 {
4957 	if (!Pflag) {
4958 		/* probe mode should never write to disk */
4959 		cylnodirty = 1;
4960 	}
4961 }
4962 
4963 void
4964 allocfrags(long frags, daddr32_t *fragp, long *cylnop)
4965 {
4966 	int	 i;
4967 	int	 j;
4968 	long	 bits;
4969 	long	 bit;
4970 
4971 	/*
4972 	 * Allocate a free-frag range in an old cylinder group
4973 	 */
4974 	for (i = 0, *fragp = 0; i < grow_fs_ncg; ++i) {
4975 		if (((fscs+i)->cs_nffree < frags) && ((fscs+i)->cs_nbfree == 0))
4976 			continue;
4977 		rdcg((long)i);
4978 		bit = bits = 0;
4979 		while (findfreerange(&bit, &bits)) {
4980 			if (frags <= bits)  {
4981 				for (j = 0; j < frags; ++j)
4982 					clrbit(cg_blksfree(&acg), bit+j);
4983 				wtcg();
4984 				*cylnop = i;
4985 				*fragp  = bit + cgbase(&sblock, i);
4986 				return;
4987 			}
4988 			bit += bits;
4989 		}
4990 	}
4991 }
4992 
4993 /*
4994  * Allocate space for frags that need to be moved in order to free up space for
4995  * expanding the cylinder summary info.
4996  * For each frag that needs to be moved (each frag or range of frags in
4997  * the csfragino list), allocate a new location and store the frag number
4998  * of that new location in the nfrag field of the csfrag struct.
4999  * If a new frag can't be allocated for any element in the csfragino list,
5000  * set the new frag number for that element to 0 and return immediately.
5001  * The notenoughspace() function will detect this condition.
5002  */
5003 void
5004 alloccsfragino()
5005 {
5006 	struct csfrag	*cfp;
5007 
5008 	/*
5009 	 * allocate space for inode frag ranges
5010 	 */
5011 	for (cfp = csfragino; cfp; cfp = cfp->next) {
5012 		allocfrags(cfp->frags, &cfp->nfrag, &cfp->cylno);
5013 		if (cfp->nfrag == 0)
5014 			break;
5015 	}
5016 }
5017 
5018 void
5019 alloccsfragfree()
5020 {
5021 	struct csfrag	*cfp;
5022 
5023 	/*
5024 	 * allocate the free frags needed for extended summary info
5025 	 */
5026 	rdcg((long)0);
5027 
5028 	for (cfp = csfragfree; cfp; cfp = cfp->next)
5029 		clrbit(cg_blksfree(&acg), cfp->ofrag - cgbase(&sblock, 0));
5030 
5031 	wtcg();
5032 }
5033 
5034 void
5035 freefrags(daddr32_t frag, long frags, long cylno)
5036 {
5037 	int	i;
5038 
5039 	/*
5040 	 * free frags
5041 	 */
5042 	rdcg(cylno);
5043 	for (i = 0; i < frags; ++i) {
5044 		setbit(cg_blksfree(&acg), (frag+i) - cgbase(&sblock, cylno));
5045 	}
5046 	wtcg();
5047 }
5048 
5049 int
5050 findfreerange(long *bitp, long *bitsp)
5051 {
5052 	long	 bit;
5053 
5054 	/*
5055 	 * find a range of free bits in a cylinder group bit map
5056 	 */
5057 	for (bit = *bitp, *bitsp = 0; bit < acg.cg_ndblk; ++bit)
5058 		if (isset(cg_blksfree(&acg), bit))
5059 			break;
5060 
5061 	if (bit >= acg.cg_ndblk)
5062 		return (0);
5063 
5064 	*bitp  = bit;
5065 	*bitsp = 1;
5066 	for (++bit; bit < acg.cg_ndblk; ++bit, ++(*bitsp)) {
5067 		if ((bit % sblock.fs_frag) == 0)
5068 			break;
5069 		if (isclr(cg_blksfree(&acg), bit))
5070 			break;
5071 	}
5072 	return (1);
5073 }
5074 
5075 void
5076 resetallocinfo()
5077 {
5078 	long	cno;
5079 	long	bit;
5080 	long	bits;
5081 
5082 	/*
5083 	 * Compute the free blocks/frags info and update the appropriate
5084 	 * inmemory superblock, summary info, and cylinder group fields
5085 	 */
5086 	sblock.fs_cstotal.cs_nffree -= acg.cg_cs.cs_nffree;
5087 	sblock.fs_cstotal.cs_nbfree -= acg.cg_cs.cs_nbfree;
5088 
5089 	acg.cg_cs.cs_nffree = 0;
5090 	acg.cg_cs.cs_nbfree = 0;
5091 
5092 	bzero((caddr_t)acg.cg_frsum, sizeof (acg.cg_frsum));
5093 	bzero((caddr_t)cg_blktot(&acg), (int)(acg.cg_iusedoff-acg.cg_btotoff));
5094 
5095 	bit = bits = 0;
5096 	while (findfreerange(&bit, &bits)) {
5097 		if (bits == sblock.fs_frag) {
5098 			acg.cg_cs.cs_nbfree++;
5099 			cno = cbtocylno(&sblock, bit);
5100 			cg_blktot(&acg)[cno]++;
5101 			cg_blks(&sblock, &acg, cno)[cbtorpos(&sblock, bit)]++;
5102 		} else {
5103 			acg.cg_cs.cs_nffree += bits;
5104 			acg.cg_frsum[bits]++;
5105 		}
5106 		bit += bits;
5107 	}
5108 
5109 	*(fscs + acg.cg_cgx) = acg.cg_cs;
5110 
5111 	sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
5112 	sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
5113 }
5114 
5115 void
5116 extendcg(long cylno)
5117 {
5118 	int	i;
5119 	diskaddr_t	dupper;
5120 	diskaddr_t	cbase;
5121 	diskaddr_t	dmax;
5122 
5123 	/*
5124 	 * extend the cylinder group at the end of the old file system
5125 	 * if it was partially allocated becase of lack of space
5126 	 */
5127 	flcg();
5128 	rdcg(cylno);
5129 
5130 	dupper = acg.cg_ndblk;
5131 	if (cylno == sblock.fs_ncg - 1)
5132 		acg.cg_ncyl = sblock.fs_ncyl - (sblock.fs_cpg * cylno);
5133 	else
5134 		acg.cg_ncyl = sblock.fs_cpg;
5135 	cbase = cgbase(&sblock, cylno);
5136 	dmax = cbase + sblock.fs_fpg;
5137 	if (dmax > sblock.fs_size)
5138 		dmax = sblock.fs_size;
5139 	acg.cg_ndblk = dmax - cbase;
5140 
5141 	for (i = dupper; i < acg.cg_ndblk; ++i)
5142 		setbit(cg_blksfree(&acg), i);
5143 
5144 	sblock.fs_dsize += (acg.cg_ndblk - dupper);
5145 
5146 	wtcg();
5147 	flcg();
5148 }
5149 
5150 struct lockfs	lockfs;
5151 int		lockfd;
5152 int		islocked;
5153 int		lockfskey;
5154 char		lockfscomment[128];
5155 
5156 void
5157 ulockfs()
5158 {
5159 	/*
5160 	 * if the file system was locked, unlock it before exiting
5161 	 */
5162 	if (islocked == 0)
5163 		return;
5164 
5165 	/*
5166 	 * first, check if the lock held
5167 	 */
5168 	lockfs.lf_flags = LOCKFS_MOD;
5169 	if (ioctl(lockfd, _FIOLFSS, &lockfs) == -1) {
5170 		perror(directory);
5171 		lockexit(32);
5172 	}
5173 
5174 	if (LOCKFS_IS_MOD(&lockfs)) {
5175 		(void) fprintf(stderr,
5176 		    gettext("FILE SYSTEM CHANGED DURING GROWFS!\n"));
5177 		(void) fprintf(stderr,
5178 		    gettext("   See lockfs(1), umount(1), and fsck(1)\n"));
5179 		lockexit(32);
5180 	}
5181 	/*
5182 	 * unlock the file system
5183 	 */
5184 	lockfs.lf_lock  = LOCKFS_ULOCK;
5185 	lockfs.lf_flags = 0;
5186 	lockfs.lf_key   = lockfskey;
5187 	clockfs();
5188 	if (ioctl(lockfd, _FIOLFS, &lockfs) == -1) {
5189 		perror(directory);
5190 		lockexit(32);
5191 	}
5192 }
5193 
5194 void
5195 wlockfs()
5196 {
5197 
5198 	/*
5199 	 * if no-write (-N), don't bother
5200 	 */
5201 	if (Nflag)
5202 		return;
5203 	/*
5204 	 * open the mountpoint, and write lock the file system
5205 	 */
5206 	if ((lockfd = open64(directory, O_RDONLY)) == -1) {
5207 		perror(directory);
5208 		lockexit(32);
5209 	}
5210 
5211 	/*
5212 	 * check if it is already locked
5213 	 */
5214 	if (ioctl(lockfd, _FIOLFSS, &lockfs) == -1) {
5215 		perror(directory);
5216 		lockexit(32);
5217 	}
5218 
5219 	if (lockfs.lf_lock != LOCKFS_WLOCK) {
5220 		lockfs.lf_lock  = LOCKFS_WLOCK;
5221 		lockfs.lf_flags = 0;
5222 		lockfs.lf_key   = 0;
5223 		clockfs();
5224 		if (ioctl(lockfd, _FIOLFS, &lockfs) == -1) {
5225 			perror(directory);
5226 			lockexit(32);
5227 		}
5228 	}
5229 	islocked = 1;
5230 	lockfskey = lockfs.lf_key;
5231 }
5232 
5233 void
5234 clockfs()
5235 {
5236 	time_t	t;
5237 	char	*ct;
5238 
5239 	(void) time(&t);
5240 	ct = ctime(&t);
5241 	ct[strlen(ct)-1] = '\0';
5242 
5243 	(void) sprintf(lockfscomment, "%s -- mkfs pid %d", ct, getpid());
5244 	lockfs.lf_comlen  = strlen(lockfscomment)+1;
5245 	lockfs.lf_comment = lockfscomment;
5246 }
5247 
5248 /*
5249  * Write the csum records and the superblock
5250  */
5251 void
5252 wtsb()
5253 {
5254 	long	i;
5255 
5256 	/*
5257 	 * write summary information
5258 	 */
5259 	for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize)
5260 		wtfs(fsbtodb(&sblock, (uint64_t)(sblock.fs_csaddr +
5261 		    numfrags(&sblock, i))),
5262 		    (int)(sblock.fs_cssize - i < sblock.fs_bsize ?
5263 		    sblock.fs_cssize - i : sblock.fs_bsize),
5264 		    ((char *)fscs) + i);
5265 
5266 	/*
5267 	 * write superblock
5268 	 */
5269 	sblock.fs_time = mkfstime;
5270 	wtfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
5271 }
5272 
5273 /*
5274  * Verify that the optimization selection is reasonable, and advance
5275  * the global "string" appropriately.
5276  */
5277 static char
5278 checkopt(char *optim)
5279 {
5280 	char	opt;
5281 	int	limit = strcspn(optim, ",");
5282 
5283 	switch (limit) {
5284 	case 0:	/* missing indicator (have comma or nul) */
5285 		(void) fprintf(stderr, gettext(
5286 		    "mkfs: missing optimization flag reset to `t' (time)\n"));
5287 		opt = 't';
5288 		break;
5289 
5290 	case 1: /* single-character indicator */
5291 		opt = *optim;
5292 		if ((opt != 's') && (opt != 't')) {
5293 			(void) fprintf(stderr, gettext(
5294 		    "mkfs: bad optimization value `%c' reset to `t' (time)\n"),
5295 			    opt);
5296 			opt = 't';
5297 		}
5298 		break;
5299 
5300 	default: /* multi-character indicator */
5301 		(void) fprintf(stderr, gettext(
5302 	    "mkfs: bad optimization value `%*.*s' reset to `t' (time)\n"),
5303 		    limit, limit, optim);
5304 		opt = 't';
5305 		break;
5306 	}
5307 
5308 	string += limit;
5309 
5310 	return (opt);
5311 }
5312 
5313 /*
5314  * Verify that the mtb selection is reasonable, and advance
5315  * the global "string" appropriately.
5316  */
5317 static char
5318 checkmtb(char *mtbarg)
5319 {
5320 	char	mtbc;
5321 	int	limit = strcspn(mtbarg, ",");
5322 
5323 	switch (limit) {
5324 	case 0:	/* missing indicator (have comma or nul) */
5325 		(void) fprintf(stderr, gettext(
5326 		    "mkfs: missing mtb flag reset to `n' (no mtb support)\n"));
5327 		mtbc = 'n';
5328 		break;
5329 
5330 	case 1: /* single-character indicator */
5331 		mtbc = tolower(*mtbarg);
5332 		if ((mtbc != 'y') && (mtbc != 'n')) {
5333 			(void) fprintf(stderr, gettext(
5334 		    "mkfs: bad mtb value `%c' reset to `n' (no mtb support)\n"),
5335 			    mtbc);
5336 			mtbc = 'n';
5337 		}
5338 		break;
5339 
5340 	default: /* multi-character indicator */
5341 		(void) fprintf(stderr, gettext(
5342 	    "mkfs: bad mtb value `%*.*s' reset to `n' (no mtb support)\n"),
5343 		    limit, limit, mtbarg);
5344 		opt = 'n';
5345 		break;
5346 	}
5347 
5348 	string += limit;
5349 
5350 	return (mtbc);
5351 }
5352 
5353 /*
5354  * Verify that a value is in a range.  If it is not, resets it to
5355  * its default value if one is supplied, exits otherwise.
5356  *
5357  * When testing, can compare user_supplied to RC_KEYWORD or RC_POSITIONAL.
5358  */
5359 static void
5360 range_check(long *varp, char *name, long minimum, long maximum,
5361     long def_val, int user_supplied)
5362 {
5363 	dprintf(("DeBuG %s : %ld (%ld %ld %ld)\n",
5364 	    name, *varp, minimum, maximum, def_val));
5365 
5366 	if ((*varp < minimum) || (*varp > maximum)) {
5367 		if (user_supplied != RC_DEFAULT) {
5368 			(void) fprintf(stderr, gettext(
5369 	    "mkfs: bad value for %s: %ld must be between %ld and %ld\n"),
5370 			    name, *varp, minimum, maximum);
5371 		}
5372 		if (def_val != NO_DEFAULT) {
5373 			if (user_supplied) {
5374 				(void) fprintf(stderr,
5375 				    gettext("mkfs: %s reset to default %ld\n"),
5376 				    name, def_val);
5377 			}
5378 			*varp = def_val;
5379 			dprintf(("DeBuG %s : %ld\n", name, *varp));
5380 			return;
5381 		}
5382 		lockexit(2);
5383 		/*NOTREACHED*/
5384 	}
5385 }
5386 
5387 /*
5388  * Verify that a value is in a range.  If it is not, resets it to
5389  * its default value if one is supplied, exits otherwise.
5390  *
5391  * When testing, can compare user_supplied to RC_KEYWORD or RC_POSITIONAL.
5392  */
5393 static void
5394 range_check_64(uint64_t *varp, char *name, uint64_t minimum, uint64_t maximum,
5395     uint64_t def_val, int user_supplied)
5396 {
5397 	if ((*varp < minimum) || (*varp > maximum)) {
5398 		if (user_supplied != RC_DEFAULT) {
5399 			(void) fprintf(stderr, gettext(
5400 	    "mkfs: bad value for %s: %lld must be between %lld and %lld\n"),
5401 			    name, *varp, minimum, maximum);
5402 		}
5403 		if (def_val != NO_DEFAULT) {
5404 			if (user_supplied) {
5405 				(void) fprintf(stderr,
5406 				    gettext("mkfs: %s reset to default %lld\n"),
5407 				    name, def_val);
5408 			}
5409 			*varp = def_val;
5410 			return;
5411 		}
5412 		lockexit(2);
5413 		/*NOTREACHED*/
5414 	}
5415 }
5416 
5417 /*
5418  * Blocks SIGINT from delivery.  Returns the previous mask in the
5419  * buffer provided, so that mask may be later restored.
5420  */
5421 static void
5422 block_sigint(sigset_t *old_mask)
5423 {
5424 	sigset_t block_mask;
5425 
5426 	if (sigemptyset(&block_mask) < 0) {
5427 		fprintf(stderr, gettext("Could not clear signal mask\n"));
5428 		lockexit(3);
5429 	}
5430 	if (sigaddset(&block_mask, SIGINT) < 0) {
5431 		fprintf(stderr, gettext("Could not set signal mask\n"));
5432 		lockexit(3);
5433 	}
5434 	if (sigprocmask(SIG_BLOCK, &block_mask, old_mask) < 0) {
5435 		fprintf(stderr, gettext("Could not block SIGINT\n"));
5436 		lockexit(3);
5437 	}
5438 }
5439 
5440 /*
5441  * Restores the signal mask that was in force before a call
5442  * to block_sigint().  This may actually still have SIGINT blocked,
5443  * if we've been recursively invoked.
5444  */
5445 static void
5446 unblock_sigint(sigset_t *old_mask)
5447 {
5448 	if (sigprocmask(SIG_UNBLOCK, old_mask, (sigset_t *)NULL) < 0) {
5449 		fprintf(stderr, gettext("Could not restore signal mask\n"));
5450 		lockexit(3);
5451 	}
5452 }
5453 
5454 /*
5455  * Attempt to be somewhat graceful about being interrupted, rather than
5456  * just silently leaving the filesystem in an unusable state.
5457  *
5458  * The kernel has blocked SIGINT upon entry, so we don't have to worry
5459  * about recursion if the user starts pounding on the keyboard.
5460  */
5461 static void
5462 recover_from_sigint(int signum)
5463 {
5464 	if (fso > -1) {
5465 		if ((Nflag != 0) || confirm_abort()) {
5466 			lockexit(4);
5467 		}
5468 	}
5469 }
5470 
5471 static int
5472 confirm_abort(void)
5473 {
5474 	char line[80];
5475 
5476 	printf(gettext("\n\nAborting at this point will leave the filesystem "
5477 	    "in an inconsistent\nstate.  If you do choose to stop, "
5478 	    "you will be given instructions on how to\nrecover "
5479 	    "the filesystem.  Do you wish to cancel the filesystem "
5480 	    "grow\noperation (y/n)?"));
5481 	if (getline(stdin, line, sizeof (line)) == EOF)
5482 		line[0] = 'y';
5483 
5484 	printf("\n");
5485 	if (line[0] == 'y' || line[0] == 'Y')
5486 		return (1);
5487 	else {
5488 		return (0);
5489 	}
5490 }
5491 
5492 static int
5493 getline(FILE *fp, char *loc, int maxlen)
5494 {
5495 	int n;
5496 	char *p, *lastloc;
5497 
5498 	p = loc;
5499 	lastloc = &p[maxlen-1];
5500 	while ((n = getc(fp)) != '\n') {
5501 		if (n == EOF)
5502 			return (EOF);
5503 		if (!isspace(n) && p < lastloc)
5504 			*p++ = n;
5505 	}
5506 	*p = 0;
5507 	return (p - loc);
5508 }
5509 
5510 /*
5511  * Calculate the maximum value of cylinders-per-group for a file
5512  * system with the characteristics:
5513  *
5514  *	bsize - file system block size
5515  *	fragsize - frag size
5516  *	nbpi - number of bytes of disk space per inode
5517  *	nrpos - number of rotational positions
5518  *	spc - sectors per cylinder
5519  *
5520  * These five characteristic are not adjustable (by this function).
5521  * The only attribute of the file system which IS adjusted by this
5522  * function in order to maximize cylinders-per-group is the proportion
5523  * of the cylinder group overhead block used for the inode map.  The
5524  * inode map cannot occupy more than one-third of the cylinder group
5525  * overhead block, but it's OK for it to occupy less than one-third
5526  * of the overhead block.
5527  *
5528  * The setting of nbpi determines one possible value for the maximum
5529  * size of a cylinder group.  It does so because it determines the total
5530  * number of inodes in the file system (file system size is fixed, and
5531  * nbpi is fixed, so the total number of inodes is fixed too).  The
5532  * cylinder group has to be small enough so that the number of inodes
5533  * in the cylinder group is less than or equal to the number of bits
5534  * in one-third (or whatever proportion is assumed) of a file system
5535  * block.  The details of the calculation are:
5536  *
5537  *     The macro MAXIpG_B(bsize, inode_divisor) determines the maximum
5538  *     number of inodes that can be in a cylinder group, given the
5539  *     proportion of the cylinder group overhead block used for the
5540  *     inode bitmaps (an inode_divisor of 3 means that 1/3 of the
5541  *     block is used for inode bitmaps; an inode_divisor of 12 means
5542  *     that 1/12 of the block is used for inode bitmaps.)
5543  *
5544  *     Once the number of inodes per cylinder group is known, the
5545  *     maximum value of cylinders-per-group (determined by nbpi)
5546  *     is calculated by the formula
5547  *
5548  *     maxcpg_given_nbpi = (size of a cylinder group)/(size of a cylinder)
5549  *
5550  *			 = (inodes-per-cg * nbpi)/(spc * DEV_BSIZE)
5551  *
5552  *     (Interestingly, the size of the file system never enters
5553  *     into this calculation.)
5554  *
5555  * Another possible value for the maximum cylinder group size is determined
5556  * by frag_size and nrpos.  The frags in the cylinder group must be
5557  * representable in the frag bitmaps in the cylinder overhead block and the
5558  * rotational positions for each cylinder must be represented in the
5559  * rotational position tables.  The calculation of the maximum cpg
5560  * value, given the frag and nrpos vales, is:
5561  *
5562  *     maxcpg_given_fragsize =
5563  *	  (available space in the overhead block) / (size of per-cylinder data)
5564  *
5565  *     The available space in the overhead block =
5566  *	  bsize - sizeof (struct cg) - space_used_for_inode_bitmaps
5567  *
5568  *     The size of the per-cylinder data is:
5569  *	    sizeof(long)            # for the "blocks avail per cylinder" field
5570  *	    + nrpos * sizeof(short)   # for the rotational position table entry
5571  *	    + frags-per-cylinder/NBBY # number of bytes to represent this
5572  *				      # cylinder in the frag bitmap
5573  *
5574  * The two calculated maximum values of cylinder-per-group will typically
5575  * turn out to be different, since they are derived from two different
5576  * constraints.  Usually, maxcpg_given_nbpi is much bigger than
5577  * maxcpg_given_fragsize.  But they can be brought together by
5578  * adjusting the proportion of the overhead block dedicated to
5579  * the inode bitmaps.  Decreasing the proportion of the cylinder
5580  * group overhead block used for inode maps will decrease
5581  * maxcpg_given_nbpi and increase maxcpg_given_fragsize.
5582  *
5583  * This function calculates the initial values of maxcpg_given_nbpi
5584  * and maxcpg_given_fragsize assuming that 1/3 of the cg overhead
5585  * block is used for inode bitmaps.  Then it decreases the proportion
5586  * of the cg overhead block used for inode bitmaps (by increasing
5587  * the value of inode_divisor) until maxcpg_given_nbpi and
5588  * maxcpg_given_fragsize are the same, or stop changing, or
5589  * maxcpg_given_nbpi is less than maxcpg_given_fragsize.
5590  *
5591  * The loop terminates when any of the following occur:
5592  *	* maxcpg_given_fragsize is greater than or equal to
5593  *	  maxcpg_given_nbpi
5594  *	* neither maxcpg_given_fragsize nor maxcpg_given_nbpi
5595  *	  change in the expected direction
5596  *
5597  * The loop is guaranteed to terminate because it only continues
5598  * while maxcpg_given_fragsize and maxcpg_given_nbpi are approaching
5599  * each other.  As soon they cross each other, or neither one changes
5600  * in the direction of the other, or one of them moves in the wrong
5601  * direction, the loop completes.
5602  */
5603 
5604 static long
5605 compute_maxcpg(long bsize, long fragsize, long nbpi, long nrpos, long spc)
5606 {
5607 	int	maxcpg_given_nbpi;	/* in cylinders */
5608 	int	maxcpg_given_fragsize;	/* in cylinders */
5609 	int	spf;			/* sectors per frag */
5610 	int	inode_divisor;
5611 	int	old_max_given_frag = 0;
5612 	int	old_max_given_nbpi = INT_MAX;
5613 
5614 	spf = fragsize / DEV_BSIZE;
5615 	inode_divisor = 3;
5616 
5617 	while (1) {
5618 		maxcpg_given_nbpi =
5619 		    (((int64_t)(MAXIpG_B(bsize, inode_divisor))) * nbpi) /
5620 		    (DEV_BSIZE * ((int64_t)spc));
5621 		maxcpg_given_fragsize =
5622 		    (bsize - (sizeof (struct cg)) - (bsize / inode_divisor)) /
5623 		    (sizeof (long) + nrpos * sizeof (short) +
5624 		    (spc / spf) / NBBY);
5625 
5626 		if (maxcpg_given_fragsize >= maxcpg_given_nbpi)
5627 			return (maxcpg_given_nbpi);
5628 
5629 		/*
5630 		 * If neither value moves toward the other, return the
5631 		 * least of the old values (we use the old instead of the
5632 		 * new because: if the old is the same as the new, it
5633 		 * doesn't matter which ones we use.  If one of the
5634 		 * values changed, but in the wrong direction, the
5635 		 * new values are suspect.  Better use the old.  This
5636 		 * shouldn't happen, but it's best to check.
5637 		 */
5638 
5639 		if (!(maxcpg_given_nbpi < old_max_given_nbpi) &&
5640 		    !(maxcpg_given_fragsize > old_max_given_frag))
5641 			return (MIN(old_max_given_nbpi, old_max_given_frag));
5642 
5643 		/*
5644 		 * This is probably impossible, but if one of the maxcpg
5645 		 * values moved in the "right" direction and one moved
5646 		 * in the "wrong" direction (that is, the two values moved
5647 		 * in the same direction), the previous conditional won't
5648 		 * recognize that the values aren't converging (since at
5649 		 * least one value moved in the "right" direction, the
5650 		 * last conditional says "keep going").
5651 		 *
5652 		 * Just to make absolutely certain that the loop terminates,
5653 		 * check for one of the values moving in the "wrong" direction
5654 		 * and terminate the loop if it happens.
5655 		 */
5656 
5657 		if (maxcpg_given_nbpi > old_max_given_nbpi ||
5658 		    maxcpg_given_fragsize < old_max_given_frag)
5659 			return (MIN(old_max_given_nbpi, old_max_given_frag));
5660 
5661 		old_max_given_nbpi = maxcpg_given_nbpi;
5662 		old_max_given_frag = maxcpg_given_fragsize;
5663 
5664 		inode_divisor++;
5665 	}
5666 }
5667 
5668 static int
5669 in_64bit_mode(void)
5670 {
5671 	/*  cmd must be an absolute path, for security */
5672 	char *cmd = "/usr/bin/isainfo -b";
5673 	char buf[BUFSIZ];
5674 	FILE *ptr;
5675 	int retval = 0;
5676 
5677 	putenv("IFS= \t");
5678 	if ((ptr = popen(cmd, "r")) != NULL) {
5679 		if (fgets(buf, BUFSIZ, ptr) != NULL &&
5680 		    strncmp(buf, "64", 2) == 0)
5681 			retval = 1;
5682 		(void) pclose(ptr);
5683 	}
5684 	return (retval);
5685 }
5686 
5687 /*
5688  * validate_size
5689  *
5690  * Return 1 if the device appears to be at least "size" sectors long.
5691  * Return 0 if it's shorter or we can't read it.
5692  */
5693 
5694 static int
5695 validate_size(int fd, diskaddr_t size)
5696 {
5697 	char 		buf[DEV_BSIZE];
5698 	int rc;
5699 
5700 	if ((llseek(fd, (offset_t)((size - 1) * DEV_BSIZE), SEEK_SET) == -1) ||
5701 	    (read(fd, buf, DEV_BSIZE)) != DEV_BSIZE)
5702 		rc = 0;
5703 	else
5704 		rc = 1;
5705 	return (rc);
5706 }
5707 
5708 /*
5709  * Print every field of the calculated superblock, along with
5710  * its value.  To make parsing easier on the caller, the value
5711  * is printed first, then the name.  Additionally, there's only
5712  * one name/value pair per line.  All values are reported in
5713  * hexadecimal (with the traditional 0x prefix), as that's slightly
5714  * easier for humans to read.  Not that they're expected to, but
5715  * debugging happens.
5716  */
5717 static void
5718 dump_sblock(void)
5719 {
5720 	int row, column, pending, written;
5721 	caddr_t source;
5722 
5723 	if (Rflag) {
5724 		pending = sizeof (sblock);
5725 		source = (caddr_t)&sblock;
5726 		do {
5727 			written = write(fileno(stdout), source, pending);
5728 			pending -= written;
5729 			source += written;
5730 		} while ((pending > 0) && (written > 0));
5731 
5732 		if (written < 0) {
5733 			perror(gettext("Binary dump of superblock failed"));
5734 			lockexit(1);
5735 		}
5736 		return;
5737 	} else {
5738 		printf("0x%x sblock.fs_link\n", sblock.fs_link);
5739 		printf("0x%x sblock.fs_rolled\n", sblock.fs_rolled);
5740 		printf("0x%x sblock.fs_sblkno\n", sblock.fs_sblkno);
5741 		printf("0x%x sblock.fs_cblkno\n", sblock.fs_cblkno);
5742 		printf("0x%x sblock.fs_iblkno\n", sblock.fs_iblkno);
5743 		printf("0x%x sblock.fs_dblkno\n", sblock.fs_dblkno);
5744 		printf("0x%x sblock.fs_cgoffset\n", sblock.fs_cgoffset);
5745 		printf("0x%x sblock.fs_cgmask\n", sblock.fs_cgmask);
5746 		printf("0x%x sblock.fs_time\n", sblock.fs_time);
5747 		printf("0x%x sblock.fs_size\n", sblock.fs_size);
5748 		printf("0x%x sblock.fs_dsize\n", sblock.fs_dsize);
5749 		printf("0x%x sblock.fs_ncg\n", sblock.fs_ncg);
5750 		printf("0x%x sblock.fs_bsize\n", sblock.fs_bsize);
5751 		printf("0x%x sblock.fs_fsize\n", sblock.fs_fsize);
5752 		printf("0x%x sblock.fs_frag\n", sblock.fs_frag);
5753 		printf("0x%x sblock.fs_minfree\n", sblock.fs_minfree);
5754 		printf("0x%x sblock.fs_rotdelay\n", sblock.fs_rotdelay);
5755 		printf("0x%x sblock.fs_rps\n", sblock.fs_rps);
5756 		printf("0x%x sblock.fs_bmask\n", sblock.fs_bmask);
5757 		printf("0x%x sblock.fs_fmask\n", sblock.fs_fmask);
5758 		printf("0x%x sblock.fs_bshift\n", sblock.fs_bshift);
5759 		printf("0x%x sblock.fs_fshift\n", sblock.fs_fshift);
5760 		printf("0x%x sblock.fs_maxcontig\n", sblock.fs_maxcontig);
5761 		printf("0x%x sblock.fs_maxbpg\n", sblock.fs_maxbpg);
5762 		printf("0x%x sblock.fs_fragshift\n", sblock.fs_fragshift);
5763 		printf("0x%x sblock.fs_fsbtodb\n", sblock.fs_fsbtodb);
5764 		printf("0x%x sblock.fs_sbsize\n", sblock.fs_sbsize);
5765 		printf("0x%x sblock.fs_csmask\n", sblock.fs_csmask);
5766 		printf("0x%x sblock.fs_csshift\n", sblock.fs_csshift);
5767 		printf("0x%x sblock.fs_nindir\n", sblock.fs_nindir);
5768 		printf("0x%x sblock.fs_inopb\n", sblock.fs_inopb);
5769 		printf("0x%x sblock.fs_nspf\n", sblock.fs_nspf);
5770 		printf("0x%x sblock.fs_optim\n", sblock.fs_optim);
5771 #ifdef _LITTLE_ENDIAN
5772 		printf("0x%x sblock.fs_state\n", sblock.fs_state);
5773 #else
5774 		printf("0x%x sblock.fs_npsect\n", sblock.fs_npsect);
5775 #endif
5776 		printf("0x%x sblock.fs_si\n", sblock.fs_si);
5777 		printf("0x%x sblock.fs_trackskew\n", sblock.fs_trackskew);
5778 		printf("0x%x sblock.fs_id[0]\n", sblock.fs_id[0]);
5779 		printf("0x%x sblock.fs_id[1]\n", sblock.fs_id[1]);
5780 		printf("0x%x sblock.fs_csaddr\n", sblock.fs_csaddr);
5781 		printf("0x%x sblock.fs_cssize\n", sblock.fs_cssize);
5782 		printf("0x%x sblock.fs_cgsize\n", sblock.fs_cgsize);
5783 		printf("0x%x sblock.fs_ntrak\n", sblock.fs_ntrak);
5784 		printf("0x%x sblock.fs_nsect\n", sblock.fs_nsect);
5785 		printf("0x%x sblock.fs_spc\n", sblock.fs_spc);
5786 		printf("0x%x sblock.fs_ncyl\n", sblock.fs_ncyl);
5787 		printf("0x%x sblock.fs_cpg\n", sblock.fs_cpg);
5788 		printf("0x%x sblock.fs_ipg\n", sblock.fs_ipg);
5789 		printf("0x%x sblock.fs_fpg\n", sblock.fs_fpg);
5790 		printf("0x%x sblock.fs_cstotal\n", sblock.fs_cstotal);
5791 		printf("0x%x sblock.fs_fmod\n", sblock.fs_fmod);
5792 		printf("0x%x sblock.fs_clean\n", sblock.fs_clean);
5793 		printf("0x%x sblock.fs_ronly\n", sblock.fs_ronly);
5794 		printf("0x%x sblock.fs_flags\n", sblock.fs_flags);
5795 		printf("0x%x sblock.fs_fsmnt\n", sblock.fs_fsmnt);
5796 		printf("0x%x sblock.fs_cgrotor\n", sblock.fs_cgrotor);
5797 		printf("0x%x sblock.fs_u.fs_csp\n", sblock.fs_u.fs_csp);
5798 		printf("0x%x sblock.fs_cpc\n", sblock.fs_cpc);
5799 
5800 		/*
5801 		 * No macros are defined for the dimensions of the
5802 		 * opostbl array.
5803 		 */
5804 		for (row = 0; row < 16; row++) {
5805 			for (column = 0; column < 8; column++) {
5806 				printf("0x%x sblock.fs_opostbl[%d][%d]\n",
5807 				    sblock.fs_opostbl[row][column],
5808 				    row, column);
5809 			}
5810 		}
5811 
5812 		/*
5813 		 * Ditto the size of sparecon.
5814 		 */
5815 		for (row = 0; row < 51; row++) {
5816 			printf("0x%x sblock.fs_sparecon[%d]\n",
5817 			    sblock.fs_sparecon[row], row);
5818 		}
5819 
5820 		printf("0x%x sblock.fs_version\n", sblock.fs_version);
5821 		printf("0x%x sblock.fs_logbno\n", sblock.fs_logbno);
5822 		printf("0x%x sblock.fs_reclaim\n", sblock.fs_reclaim);
5823 		printf("0x%x sblock.fs_sparecon2\n", sblock.fs_sparecon2);
5824 #ifdef _LITTLE_ENDIAN
5825 		printf("0x%x sblock.fs_npsect\n", sblock.fs_npsect);
5826 #else
5827 		printf("0x%x sblock.fs_state\n", sblock.fs_state);
5828 #endif
5829 		printf("0x%llx sblock.fs_qbmask\n", sblock.fs_qbmask);
5830 		printf("0x%llx sblock.fs_qfmask\n", sblock.fs_qfmask);
5831 		printf("0x%x sblock.fs_postblformat\n", sblock.fs_postblformat);
5832 		printf("0x%x sblock.fs_nrpos\n", sblock.fs_nrpos);
5833 		printf("0x%x sblock.fs_postbloff\n", sblock.fs_postbloff);
5834 		printf("0x%x sblock.fs_rotbloff\n", sblock.fs_rotbloff);
5835 		printf("0x%x sblock.fs_magic\n", sblock.fs_magic);
5836 
5837 		/*
5838 		 * fs_space isn't of much use in this context, so we'll
5839 		 * just ignore it for now.
5840 		 */
5841 	}
5842 }
5843