xref: /illumos-gate/usr/src/cmd/fs.d/ufs/mkfs/mkfs.c (revision 14b24e2b79293068c8e016a69ef1d872fb5e2fd5)
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 /*
23  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	All Rights Reserved	*/
28 
29 /*
30  * University Copyright- Copyright (c) 1982, 1986, 1988
31  * The Regents of the University of California
32  * All Rights Reserved
33  *
34  * University Acknowledgment- Portions of this document are derived from
35  * software developed by the University of California, Berkeley, and its
36  * contributors.
37  */
38 
39 /*
40  * The maximum supported file system size (in sectors) is the
41  * number of frags that can be represented in an int32_t field
42  * (INT_MAX) times the maximum number of sectors per frag.  Since
43  * the maximum frag size is MAXBSIZE, the maximum number of sectors
44  * per frag is MAXBSIZE/DEV_BSIZE.
45  */
46 #define	FS_MAX	(((diskaddr_t)INT_MAX) * (MAXBSIZE/DEV_BSIZE))
47 
48 /*
49  * make file system for cylinder-group style file systems
50  *
51  * usage:
52  *
53  *    mkfs [-F FSType] [-V] [-G [-P]] [-M dirname] [-m] [options]
54  *	[-o specific_options]  special size
55  *	[nsect ntrack bsize fsize cpg	minfree	rps nbpi opt apc rotdelay
56  *	  2     3      4     5     6	7	8   9	 10  11  12
57  *	nrpos maxcontig mtb]
58  *	13    14	15
59  *
60  *  where specific_options are:
61  *	N - no create
62  *	nsect - The number of sectors per track
63  *	ntrack - The number of tracks per cylinder
64  *	bsize - block size
65  *	fragsize - fragment size
66  *	cgsize - The number of disk cylinders per cylinder group.
67  *	free - minimum free space
68  *	rps - rotational speed (rev/sec).
69  *	nbpi - number of data bytes per allocated inode
70  *	opt - optimization (space, time)
71  *	apc - number of alternates
72  *	gap - gap size
73  *	nrpos - number of rotational positions
74  *	maxcontig - maximum number of logical blocks that will be
75  *		allocated contiguously before inserting rotational delay
76  *	mtb - if "y", set up file system for eventual growth to over a
77  *		a terabyte
78  * -P Do not grow the file system, but print on stdout the maximal
79  *    size in sectors to which the file system can be increased. The calculated
80  *    size is limited by the value provided by the operand size.
81  *
82  * Note that -P is a project-private interface and together with -G intended
83  * to be used only by the growfs script. It is therefore purposely not
84  * documented in the man page.
85  * The -P option is covered by PSARC case 2003/422.
86  */
87 
88 /*
89  * The following constants set the defaults used for the number
90  * of sectors/track (fs_nsect), and number of tracks/cyl (fs_ntrak).
91  *
92  *			NSECT		NTRAK
93  *	72MB CDC	18		9
94  *	30MB CDC	18		5
95  *	720KB Diskette	9		2
96  *
97  * However the defaults will be different for disks larger than CHSLIMIT.
98  */
99 
100 #define	DFLNSECT	32
101 #define	DFLNTRAK	16
102 
103 /*
104  * The following default sectors and tracks values are used for
105  * non-efi disks that are larger than the CHS addressing limit. The
106  * existing default cpg of 16 (DESCPG) holds good for larger disks too.
107  */
108 #define	DEF_SECTORS_EFI	128
109 #define	DEF_TRACKS_EFI	48
110 
111 /*
112  * The maximum number of cylinders in a group depends upon how much
113  * information can be stored on a single cylinder. The default is to
114  * use 16 cylinders per group.  This is effectively tradition - it was
115  * the largest value acceptable under SunOs 4.1
116  */
117 #define	DESCPG		16	/* desired fs_cpg */
118 
119 /*
120  * The following two constants set the default block and fragment sizes.
121  * Both constants must be a power of 2 and meet the following constraints:
122  *	MINBSIZE <= DESBLKSIZE <= MAXBSIZE
123  *	DEV_BSIZE <= DESFRAGSIZE <= DESBLKSIZE
124  *	DESBLKSIZE / DESFRAGSIZE <= 8
125  */
126 #define	DESBLKSIZE	8192
127 #define	DESFRAGSIZE	1024
128 
129 /*
130  * MINFREE gives the minimum acceptable percentage of file system
131  * blocks which may be free. If the freelist drops below this level
132  * only the superuser may continue to allocate blocks. This may
133  * be set to 0 if no reserve of free blocks is deemed necessary,
134  * however throughput drops by fifty percent if the file system
135  * is run at between 90% and 100% full; thus the default value of
136  * fs_minfree is 10%. With 10% free space, fragmentation is not a
137  * problem, so we choose to optimize for time.
138  */
139 #define	MINFREE		10
140 #define	DEFAULTOPT	FS_OPTTIME
141 
142 /*
143  * ROTDELAY gives the minimum number of milliseconds to initiate
144  * another disk transfer on the same cylinder. It is no longer used
145  * and will always default to 0.
146  */
147 #define	ROTDELAY	0
148 
149 /*
150  * MAXBLKPG determines the maximum number of data blocks which are
151  * placed in a single cylinder group. The default is one indirect
152  * block worth of data blocks.
153  */
154 #define	MAXBLKPG(bsize)	((bsize) / sizeof (daddr32_t))
155 
156 /*
157  * Each file system has a number of inodes statically allocated.
158  * We allocate one inode slot per NBPI bytes, expecting this
159  * to be far more than we will ever need.
160  */
161 #define	NBPI		2048	/* Number Bytes Per Inode */
162 #define	MTB_NBPI	(MB)	/* Number Bytes Per Inode for multi-terabyte */
163 
164 /*
165  * Disks are assumed to rotate at 60HZ, unless otherwise specified.
166  */
167 #define	DEFHZ		60
168 
169 /*
170  * Cylinder group related limits.
171  *
172  * For each cylinder we keep track of the availability of blocks at different
173  * rotational positions, so that we can lay out the data to be picked
174  * up with minimum rotational latency.  NRPOS is the number of rotational
175  * positions which we distinguish.  With NRPOS 8 the resolution of our
176  * summary information is 2ms for a typical 3600 rpm drive.
177  */
178 #define	NRPOS		8	/* number distinct rotational positions */
179 
180 #ifdef DEBUG
181 #define	dprintf(x)	printf x
182 #else
183 #define	dprintf(x)
184 #endif
185 
186 /*
187  * For the -N option, when calculating the backup superblocks, do not print
188  * them if we are not really sure. We may have to try an alternate method of
189  * arriving at the superblocks. So defer printing till a handful of superblocks
190  * look good.
191  */
192 #define	tprintf(x)	if (Nflag && retry) \
193 				(void) strncat(tmpbuf, x, strlen(x)); \
194 			else \
195 				(void) fprintf(stderr, x);
196 
197 #define	ALTSB		32	/* Location of first backup superblock */
198 
199 /*
200  * range_check "user_supplied" flag values.
201  */
202 #define	RC_DEFAULT	0
203 #define	RC_KEYWORD	1
204 #define	RC_POSITIONAL	2
205 
206 /*
207  * ufs hole
208  */
209 #define	UFS_HOLE	-1
210 
211 #ifndef	STANDALONE
212 #include	<stdio.h>
213 #include	<sys/mnttab.h>
214 #endif
215 
216 #include	<stdlib.h>
217 #include	<unistd.h>
218 #include	<malloc.h>
219 #include	<string.h>
220 #include	<strings.h>
221 #include	<ctype.h>
222 #include	<errno.h>
223 #include	<sys/param.h>
224 #include	<time.h>
225 #include	<sys/types.h>
226 #include	<sys/sysmacros.h>
227 #include	<sys/vnode.h>
228 #include	<sys/fs/ufs_fsdir.h>
229 #include	<sys/fs/ufs_inode.h>
230 #include	<sys/fs/ufs_fs.h>
231 #include	<sys/fs/ufs_log.h>
232 #include	<sys/mntent.h>
233 #include	<sys/filio.h>
234 #include	<limits.h>
235 #include	<sys/int_const.h>
236 #include	<signal.h>
237 #include	<sys/efi_partition.h>
238 #include	"roll_log.h"
239 
240 #define	bcopy(f, t, n)    (void) memcpy(t, f, n)
241 #define	bzero(s, n)	(void) memset(s, 0, n)
242 #define	bcmp(s, d, n)	memcmp(s, d, n)
243 
244 #define	index(s, r)	strchr(s, r)
245 #define	rindex(s, r)	strrchr(s, r)
246 
247 #include	<sys/stat.h>
248 #include	<sys/statvfs.h>
249 #include	<locale.h>
250 #include	<fcntl.h>
251 #include	<sys/isa_defs.h>	/* for ENDIAN defines */
252 #include	<sys/vtoc.h>
253 
254 #include	<sys/dkio.h>
255 #include	<sys/asynch.h>
256 
257 extern offset_t	llseek();
258 extern char	*getfullblkname();
259 extern long	lrand48();
260 
261 extern int	optind;
262 extern char	*optarg;
263 
264 
265 /*
266  * The size of a cylinder group is calculated by CGSIZE. The maximum size
267  * is limited by the fact that cylinder groups are at most one block.
268  * Its size is derived from the size of the maps maintained in the
269  * cylinder group and the (struct cg) size.
270  */
271 #define	CGSIZE(fs) \
272 	/* base cg		*/ (sizeof (struct cg) + \
273 	/* blktot size	*/ (fs)->fs_cpg * sizeof (long) + \
274 	/* blks size	*/ (fs)->fs_cpg * (fs)->fs_nrpos * sizeof (short) + \
275 	/* inode map	*/ howmany((fs)->fs_ipg, NBBY) + \
276 	/* block map */ howmany((fs)->fs_cpg * (fs)->fs_spc / NSPF(fs), NBBY))
277 
278 /*
279  * We limit the size of the inode map to be no more than a
280  * third of the cylinder group space, since we must leave at
281  * least an equal amount of space for the block map.
282  *
283  * N.B.: MAXIpG must be a multiple of INOPB(fs).
284  */
285 #define	MAXIpG(fs)	roundup((fs)->fs_bsize * NBBY / 3, INOPB(fs))
286 
287 /*
288  * Same as MAXIpG, but parameterized by the block size (b) and the
289  * cylinder group divisor (d), which is the reciprocal of the fraction of the
290  * cylinder group overhead block that is used for the inode map.  So for
291  * example, if d = 5, the macro's computation assumes that 1/5 of the
292  * cylinder group overhead block can be dedicated to the inode map.
293  */
294 #define	MAXIpG_B(b, d)	roundup((b) * NBBY / (d), (b) / sizeof (struct dinode))
295 
296 #define	UMASK		0755
297 #define	MAXINOPB	(MAXBSIZE / sizeof (struct dinode))
298 #define	POWEROF2(num)	(((num) & ((num) - 1)) == 0)
299 #define	MB		(1024*1024)
300 #define	BETWEEN(x, l, h)	((x) >= (l) && (x) <= (h))
301 
302 /*
303  * Used to set the inode generation number. Since both inodes and dinodes
304  * are dealt with, we really need a pointer to an icommon here.
305  */
306 #define	IRANDOMIZE(icp)	(icp)->ic_gen = lrand48();
307 
308 /*
309  * Flags for number()
310  */
311 #define	ALLOW_PERCENT	0x01	/* allow trailing `%' on number */
312 #define	ALLOW_MS1	0x02	/* allow trailing `ms', state 1 */
313 #define	ALLOW_MS2	0x04	/* allow trailing `ms', state 2 */
314 #define	ALLOW_END_ONLY	0x08	/* must be at end of number & suffixes */
315 
316 #define	MAXAIO	1000	/* maximum number of outstanding I/O's we'll manage */
317 #define	BLOCK	1	/* block in aiowait */
318 #define	NOBLOCK	0	/* don't block in aiowait */
319 
320 #define	RELEASE 1	/* free an aio buffer after use */
321 #define	SAVE	0	/* don't free the buffer */
322 
323 typedef struct aio_trans {
324 	aio_result_t resultbuf;
325 	diskaddr_t bno;
326 	char *buffer;
327 	int size;
328 	int release;
329 	struct aio_trans *next;
330 } aio_trans;
331 
332 typedef struct aio_results {
333 	int max;
334 	int outstanding;
335 	int maxpend;
336 	aio_trans *trans;
337 } aio_results;
338 
339 int aio_inited = 0;
340 aio_results results;
341 
342 /*
343  * Allow up to MAXBUF aio requests that each have a unique buffer.
344  * More aio's might be done, but not using memory through the getbuf()
345  * interface.  This can be raised, but you run into the potential of
346  * using more memory than is physically available on the machine,
347  * and if you start swapping, you can forget about performance.
348  * To prevent this, we also limit the total memory used for a given
349  * type of buffer to MAXBUFMEM.
350  *
351  * Tests indicate a cylinder group's worth of inodes takes:
352  *
353  *	NBPI	Size of Inode Buffer
354  *	 2k	1688k
355  *	 8k	 424k
356  *
357  * initcg() stores all the inodes for a cylinder group in one buffer,
358  * so allowing 20 buffers could take 32 MB if not limited by MAXBUFMEM.
359  */
360 #define	MAXBUF		20
361 #define	MAXBUFMEM	(8 * 1024 * 1024)
362 
363 /*
364  * header information for buffers managed by getbuf() and freebuf()
365  */
366 typedef struct bufhdr {
367 	struct bufhdr *head;
368 	struct bufhdr *next;
369 } bufhdr;
370 
371 int bufhdrsize;
372 
373 bufhdr inodebuf = { NULL, NULL };
374 bufhdr cgsumbuf = { NULL, NULL };
375 
376 #define	SECTORS_PER_TERABYTE	(1LL << 31)
377 /*
378  * The following constant specifies an upper limit for file system size
379  * that is actually a lot bigger than we expect to support with UFS. (Since
380  * it's specified in sectors, the file system size would be 2**44 * 512,
381  * which is 2**53, which is 8192 Terabytes.)  However, it's useful
382  * for checking the basic sanity of a size value that is input on the
383  * command line.
384  */
385 #define	FS_SIZE_UPPER_LIMIT	0x100000000000LL
386 
387 /*
388  * Forward declarations
389  */
390 static char *getbuf(bufhdr *bufhead, int size);
391 static void freebuf(char *buf);
392 static void freetrans(aio_trans *transp);
393 static aio_trans *get_aiop();
394 static aio_trans *wait_for_write(int block);
395 static void initcg(int cylno);
396 static void fsinit();
397 static int makedir(struct direct *protodir, int entries);
398 static void iput(struct inode *ip);
399 static void rdfs(diskaddr_t bno, int size, char *bf);
400 static void wtfs(diskaddr_t bno, int size, char *bf);
401 static void awtfs(diskaddr_t bno, int size, char *bf, int release);
402 static void wtfs_breakup(diskaddr_t bno, int size, char *bf);
403 static int isblock(struct fs *fs, unsigned char *cp, int h);
404 static void clrblock(struct fs *fs, unsigned char *cp, int h);
405 static void setblock(struct fs *fs, unsigned char *cp, int h);
406 static void usage(void) __NORETURN;
407 static void dump_fscmd(char *fsys, int fsi);
408 static uint64_t number(uint64_t d_value, char *param, int flags);
409 static int match(char *s);
410 static char checkopt(char *optim);
411 static char checkmtb(char *mtbarg);
412 static void range_check(long *varp, char *name, long minimum,
413     long maximum, long def_val, int user_supplied);
414 static void range_check_64(uint64_t *varp, char *name, uint64_t minimum,
415     uint64_t maximum, uint64_t def_val, int user_supplied);
416 static daddr32_t alloc(int size, int mode);
417 static diskaddr_t get_max_size(int fd);
418 static long get_max_track_size(int fd);
419 static void block_sigint(sigset_t *old_mask);
420 static void unblock_sigint(sigset_t *old_mask);
421 static void recover_from_sigint(int signum);
422 static int confirm_abort(void);
423 static int getaline(FILE *fp, char *loc, int maxlen);
424 static void flush_writes(void);
425 static long compute_maxcpg(long, long, long, long, long);
426 static int in_64bit_mode(void);
427 static int validate_size(int fd, diskaddr_t size);
428 static void dump_sblock(void);
429 
430 /*
431  * Workaround for mkfs to function properly on disks attached to XMIT 2.X
432  * controller. If the address is not aligned at 8 byte boundary, mkfs on
433  * disks attached to XMIT 2.X controller exhibts un-predictable behaviour.
434  */
435 #define	XMIT_2_X_ALIGN	8
436 #pragma	align XMIT_2_X_ALIGN(fsun, altfsun, cgun)
437 
438 union {
439 	struct fs fs;
440 	char pad[SBSIZE];
441 } fsun, altfsun;
442 #define	sblock	fsun.fs
443 #define	altsblock	altfsun.fs
444 
445 struct	csum *fscs;
446 
447 union cgun {
448 	struct cg cg;
449 	char pad[MAXBSIZE];
450 } cgun;
451 
452 #define	acg	cgun.cg
453 /*
454  * Size of screen in cols in which to fit output
455  */
456 #define	WIDTH	80
457 
458 struct dinode zino[MAXBSIZE / sizeof (struct dinode)];
459 
460 /*
461  * file descriptors used for rdfs(fsi) and wtfs(fso).
462  * Initialized to an illegal file descriptor number.
463  */
464 int	fsi = -1;
465 int	fso = -1;
466 
467 /*
468  * The BIG parameter is machine dependent.  It should be a longlong integer
469  * constant that can be used by the number parser to check the validity
470  * of numeric parameters.
471  */
472 
473 #define	BIG		0x7fffffffffffffffLL
474 
475 /* Used to indicate to number() that a bogus value should cause us to exit */
476 #define	NO_DEFAULT	LONG_MIN
477 
478 /*
479  * INVALIDSBLIMIT is the number of bad backup superblocks that will be
480  * tolerated before we decide to try arriving at a different set of them
481  * using a different logic. This is applicable for non-EFI disks only.
482  */
483 #define	INVALIDSBLIMIT	10
484 
485 /*
486  * The *_flag variables are used to indicate that the user specified
487  * the values, rather than that we made them up ourselves.  We can
488  * complain about the user giving us bogus values.
489  */
490 
491 /* semi-constants */
492 long	sectorsize = DEV_BSIZE;		/* bytes/sector from param.h */
493 long	bbsize = BBSIZE;		/* boot block size */
494 long	sbsize = SBSIZE;		/* superblock size */
495 
496 /* parameters */
497 diskaddr_t	fssize_db;		/* file system size in disk blocks */
498 diskaddr_t	fssize_frag;		/* file system size in frags */
499 long	cpg;				/* cylinders/cylinder group */
500 int	cpg_flag = RC_DEFAULT;
501 long	rotdelay = -1;			/* rotational delay between blocks */
502 int	rotdelay_flag = RC_DEFAULT;
503 long	maxcontig;			/* max contiguous blocks to allocate */
504 int	maxcontig_flag = RC_DEFAULT;
505 long	nsect = DFLNSECT;		/* sectors per track */
506 int	nsect_flag = RC_DEFAULT;
507 long	ntrack = DFLNTRAK;		/* tracks per cylinder group */
508 int	ntrack_flag = RC_DEFAULT;
509 long	bsize = DESBLKSIZE;		/* filesystem block size */
510 int	bsize_flag = RC_DEFAULT;
511 long	fragsize = DESFRAGSIZE;		/* filesystem fragment size */
512 int	fragsize_flag = RC_DEFAULT;
513 long	minfree = MINFREE;		/* fs_minfree */
514 int	minfree_flag = RC_DEFAULT;
515 long	rps = DEFHZ;			/* revolutions/second of drive */
516 int	rps_flag = RC_DEFAULT;
517 long	nbpi = NBPI;			/* number of bytes per inode */
518 int	nbpi_flag = RC_DEFAULT;
519 long	nrpos = NRPOS;			/* number of rotational positions */
520 int	nrpos_flag = RC_DEFAULT;
521 long	apc = 0;			/* alternate sectors per cylinder */
522 int	apc_flag = RC_DEFAULT;
523 char	opt = 't';			/* optimization style, `t' or `s' */
524 char	mtb = 'n';			/* multi-terabyte format, 'y' or 'n' */
525 #define	DEFAULT_SECT_TRAK_CPG	(nsect_flag == RC_DEFAULT && \
526 				ntrack_flag == RC_DEFAULT && \
527 				cpg_flag == RC_DEFAULT)
528 
529 long	debug = 0;			/* enable debugging output */
530 
531 int	spc_flag = 0;			/* alternate sectors specified or */
532 					/* found */
533 
534 /* global state */
535 int	Nflag;		/* do not write to disk */
536 int	mflag;		/* return the command line used to create this FS */
537 int	rflag;		/* report the superblock in an easily-parsed form */
538 int	Rflag;		/* dump the superblock in binary */
539 char	*fsys;
540 time_t	mkfstime;
541 char	*string;
542 int	label_type;
543 
544 /*
545  * logging support
546  */
547 int	islog;			/* true if ufs logging is enabled */
548 int	islogok;		/* true if ufs log state is good */
549 int	waslog;			/* true when ufs logging disabled during grow */
550 
551 /*
552  * growfs defines, globals, and forward references
553  */
554 #define	NOTENOUGHSPACE 33
555 int		grow;
556 #define	GROW_WITH_DEFAULT_TRAK	(grow && ntrack_flag == RC_DEFAULT)
557 
558 static int	Pflag;		/* probe to which size the fs can be grown */
559 int		ismounted;
560 char		*directory;
561 diskaddr_t	grow_fssize;
562 long		grow_fs_size;
563 long		grow_fs_ncg;
564 diskaddr_t		grow_fs_csaddr;
565 long		grow_fs_cssize;
566 int		grow_fs_clean;
567 struct csum	*grow_fscs;
568 diskaddr_t		grow_sifrag;
569 int		test;
570 int		testforce;
571 diskaddr_t		testfrags;
572 int		inlockexit;
573 int		isbad;
574 
575 void		lockexit(int) __NORETURN;
576 void		randomgeneration(void);
577 void		checksummarysize(void);
578 int		checksblock(struct fs, int);
579 void		growinit(char *);
580 void		checkdev(char *, char  *);
581 void		checkmount(struct mnttab *, char *);
582 struct dinode	*gdinode(ino_t);
583 int		csfraginrange(daddr32_t);
584 struct csfrag	*findcsfrag(daddr32_t, struct csfrag **);
585 void		checkindirect(ino_t, daddr32_t *, daddr32_t, int);
586 void		addcsfrag(ino_t, daddr32_t, struct csfrag **);
587 void		delcsfrag(daddr32_t, struct csfrag **);
588 void		checkdirect(ino_t, daddr32_t *, daddr32_t *, int);
589 void		findcsfragino(void);
590 void		fixindirect(daddr32_t, int);
591 void		fixdirect(caddr_t, daddr32_t, daddr32_t *, int);
592 void		fixcsfragino(void);
593 void		extendsummaryinfo(void);
594 int		notenoughspace(void);
595 void		unalloccsfragino(void);
596 void		unalloccsfragfree(void);
597 void		findcsfragfree(void);
598 void		copycsfragino(void);
599 void		rdcg(long);
600 void		wtcg(void);
601 void		flcg(void);
602 void		allocfrags(long, daddr32_t *, long *);
603 void		alloccsfragino(void);
604 void		alloccsfragfree(void);
605 void		freefrags(daddr32_t, long, long);
606 int		findfreerange(long *, long *);
607 void		resetallocinfo(void);
608 void		extendcg(long);
609 void		ulockfs(void);
610 void		wlockfs(void);
611 void		clockfs(void);
612 void		wtsb(void);
613 static int64_t	checkfragallocated(daddr32_t);
614 static struct csum	*read_summaryinfo(struct fs *);
615 static diskaddr_t	probe_summaryinfo();
616 
617 int
618 main(int argc, char *argv[])
619 {
620 	long i, mincpc, mincpg, ibpcl;
621 	long cylno, rpos, blk, j, warn = 0;
622 	long mincpgcnt, maxcpg;
623 	uint64_t used, bpcg, inospercg;
624 	long mapcramped, inodecramped;
625 	long postblsize, rotblsize, totalsbsize;
626 	FILE *mnttab;
627 	struct mnttab mntp;
628 	char *special;
629 	struct statvfs64 fs;
630 	struct dk_geom dkg;
631 	struct dk_minfo dkminfo;
632 	char pbuf[sizeof (uint64_t) * 3 + 1];
633 	char *tmpbuf;
634 	int width, plen;
635 	uint64_t num;
636 	int c, saverr;
637 	diskaddr_t max_fssize;
638 	long tmpmaxcontig = -1;
639 	struct sigaction sigact;
640 	uint64_t nbytes64;
641 	int remaining_cg;
642 	int do_dot = 0;
643 	int use_efi_dflts = 0, retry = 0, isremovable = 0, ishotpluggable = 0;
644 	int invalid_sb_cnt, ret, skip_this_sb, cg_too_small;
645 	int geom_nsect, geom_ntrack, geom_cpg;
646 
647 	(void) setlocale(LC_ALL, "");
648 
649 #if !defined(TEXT_DOMAIN)
650 #define	TEXT_DOMAIN "SYS_TEST"
651 #endif
652 	(void) textdomain(TEXT_DOMAIN);
653 
654 	while ((c = getopt(argc, argv, "F:bmo:VPGM:T:t:")) != EOF) {
655 		switch (c) {
656 
657 		case 'F':
658 			string = optarg;
659 			if (strcmp(string, "ufs") != 0)
660 				usage();
661 			break;
662 
663 		case 'm':	/* return command line used to create this FS */
664 			mflag++;
665 			break;
666 
667 		case 'o':
668 			/*
669 			 * ufs specific options.
670 			 */
671 			string = optarg;
672 			while (*string != '\0') {
673 				if (match("nsect=")) {
674 					nsect = number(DFLNSECT, "nsect", 0);
675 					nsect_flag = RC_KEYWORD;
676 				} else if (match("ntrack=")) {
677 					ntrack = number(DFLNTRAK, "ntrack", 0);
678 					ntrack_flag = RC_KEYWORD;
679 				} else if (match("bsize=")) {
680 					bsize = number(DESBLKSIZE, "bsize", 0);
681 					bsize_flag = RC_KEYWORD;
682 				} else if (match("fragsize=")) {
683 					fragsize = number(DESFRAGSIZE,
684 					    "fragsize", 0);
685 					fragsize_flag = RC_KEYWORD;
686 				} else if (match("cgsize=")) {
687 					cpg = number(DESCPG, "cgsize", 0);
688 					cpg_flag = RC_KEYWORD;
689 				} else if (match("free=")) {
690 					minfree = number(MINFREE, "free",
691 					    ALLOW_PERCENT);
692 					minfree_flag = RC_KEYWORD;
693 				} else if (match("maxcontig=")) {
694 					tmpmaxcontig =
695 					    number(-1, "maxcontig", 0);
696 					maxcontig_flag = RC_KEYWORD;
697 				} else if (match("nrpos=")) {
698 					nrpos = number(NRPOS, "nrpos", 0);
699 					nrpos_flag = RC_KEYWORD;
700 				} else if (match("rps=")) {
701 					rps = number(DEFHZ, "rps", 0);
702 					rps_flag = RC_KEYWORD;
703 				} else if (match("nbpi=")) {
704 					nbpi = number(NBPI, "nbpi", 0);
705 					nbpi_flag = RC_KEYWORD;
706 				} else if (match("opt=")) {
707 					opt = checkopt(string);
708 				} else if (match("mtb=")) {
709 					mtb = checkmtb(string);
710 				} else if (match("apc=")) {
711 					apc = number(0, "apc", 0);
712 					apc_flag = RC_KEYWORD;
713 				} else if (match("gap=")) {
714 					(void) number(0, "gap", ALLOW_MS1);
715 					rotdelay = ROTDELAY;
716 					rotdelay_flag = RC_DEFAULT;
717 				} else if (match("debug=")) {
718 					debug = number(0, "debug", 0);
719 				} else if (match("N")) {
720 					Nflag++;
721 				} else if (match("calcsb")) {
722 					rflag++;
723 					Nflag++;
724 				} else if (match("calcbinsb")) {
725 					rflag++;
726 					Rflag++;
727 					Nflag++;
728 				} else if (*string == '\0') {
729 					break;
730 				} else {
731 					(void) fprintf(stderr, gettext(
732 					    "illegal option: %s\n"), string);
733 					usage();
734 				}
735 
736 				if (*string == ',') string++;
737 				if (*string == ' ') string++;
738 			}
739 			break;
740 
741 		case 'V':
742 			{
743 				char	*opt_text;
744 				int	opt_count;
745 
746 				(void) fprintf(stdout, gettext("mkfs -F ufs "));
747 				for (opt_count = 1; opt_count < argc;
748 				    opt_count++) {
749 					opt_text = argv[opt_count];
750 					if (opt_text)
751 						(void) fprintf(stdout, " %s ",
752 						    opt_text);
753 				}
754 				(void) fprintf(stdout, "\n");
755 			}
756 			break;
757 
758 		case 'b':	/* do nothing for this */
759 			break;
760 
761 		case 'M':	/* grow the mounted file system */
762 			directory = optarg;
763 
764 			/* FALLTHROUGH */
765 		case 'G':	/* grow the file system */
766 			grow = 1;
767 			break;
768 		case 'P':	/* probe the file system growing size	*/
769 			Pflag = 1;
770 			grow = 1; /* probe mode implies fs growing	*/
771 			break;
772 		case 'T':	/* For testing */
773 			testforce = 1;
774 
775 			/* FALLTHROUGH */
776 		case 't':
777 			test = 1;
778 			string = optarg;
779 			testfrags = number(NO_DEFAULT, "testfrags", 0);
780 			break;
781 
782 		case '?':
783 			usage();
784 			break;
785 		}
786 	}
787 #ifdef MKFS_DEBUG
788 	/*
789 	 * Turning on MKFS_DEBUG causes mkfs to produce a filesystem
790 	 * that can be reproduced by setting the time to 0 and seeding
791 	 * the random number generator to a constant.
792 	 */
793 	mkfstime = 0;	/* reproducible results */
794 #else
795 	(void) time(&mkfstime);
796 #endif
797 
798 	if (optind >= (argc - 1)) {
799 		if (optind > (argc - 1)) {
800 			(void) fprintf(stderr,
801 			    gettext("special not specified\n"));
802 			usage();
803 		} else if (mflag == 0) {
804 			(void) fprintf(stderr,
805 			    gettext("size not specified\n"));
806 			usage();
807 		}
808 	}
809 	argc -= optind;
810 	argv = &argv[optind];
811 
812 	fsys = argv[0];
813 	fsi = open64(fsys, O_RDONLY);
814 	if (fsi < 0) {
815 		(void) fprintf(stderr, gettext("%s: cannot open\n"), fsys);
816 		lockexit(32);
817 	}
818 
819 	if (mflag) {
820 		dump_fscmd(fsys, fsi);
821 		lockexit(0);
822 	}
823 
824 	/*
825 	 * The task of setting all of the configuration parameters for a
826 	 * UFS file system is basically a matter of solving n equations
827 	 * in m variables.  Typically, m is greater than n, so there is
828 	 * usually more than one valid solution.  Since this is usually
829 	 * an under-constrained problem, it's not always obvious what the
830 	 * "best" configuration is.
831 	 *
832 	 * In general, the approach is to
833 	 * 1. Determine the values for the file system parameters
834 	 *    that are externally contrained and therefore not adjustable
835 	 *    by mkfs (such as the device's size and maxtransfer size).
836 	 * 2. Acquire the user's requested setting for all configuration
837 	 *    values that can be set on the command line.
838 	 * 3. Determine the final value of all configuration values, by
839 	 *    the following approach:
840 	 *	- set the file system block size (fs_bsize).  Although
841 	 *	  this could be regarded as an adjustable parameter, in
842 	 *	  fact, it's pretty much a constant.  At this time, it's
843 	 *	  generally set to 8k (with older hardware, it can
844 	 *	  sometimes make sense to set it to 4k, but those
845 	 *	  situations are pretty rare now).
846 	 *	- re-adjust the maximum file system size based on the
847 	 *	  value of the file system block size.  Since the
848 	 *	  frag size can't be any larger than a file system
849 	 *	  block, and the number of frags in the file system
850 	 *	  has to fit into 31 bits, the file system block size
851 	 *	  affects the maximum file system size.
852 	 *	- now that the real maximum file system is known, set the
853 	 *	  actual size of the file system to be created to
854 	 *	  MIN(requested size, maximum file system size).
855 	 *	- now validate, and if necessary, adjust the following
856 	 *	  values:
857 	 *		rotdelay
858 	 *		nsect
859 	 *		maxcontig
860 	 *		apc
861 	 *		frag_size
862 	 *		rps
863 	 *		minfree
864 	 *		nrpos
865 	 *		nrack
866 	 *		nbpi
867 	 *	- calculate maxcpg (the maximum value of the cylinders-per-
868 	 *	  cylinder-group configuration parameters).  There are two
869 	 *	  algorithms for calculating maxcpg:  an old one, which is
870 	 *	  used for file systems of less than 1 terabyte, and a
871 	 *	  new one, implemented in the function compute_maxcpg(),
872 	 *	  which is used for file systems of greater than 1 TB.
873 	 *	  The difference between them is that compute_maxcpg()
874 	 *	  really tries to maximize the cpg value.  The old
875 	 *	  algorithm fails to take advantage of smaller frags and
876 	 *	  lower inode density when determining the maximum cpg,
877 	 *	  and thus comes up with much lower numbers in some
878 	 *	  configurations.  At some point, we might use the
879 	 *	  new algorithm for determining maxcpg for all file
880 	 *	  systems, but at this time, the changes implemented for
881 	 *	  multi-terabyte UFS are NOT being automatically applied
882 	 *	  to UFS file systems of less than a terabyte (in the
883 	 *	  interest of not changing existing UFS policy too much
884 	 *	  until the ramifications of the changes are well-understood
885 	 *	  and have been evaluated for their effects on performance.)
886 	 *	- check the current values of the configuration parameters
887 	 *	  against the various constraints imposed by UFS.  These
888 	 *	  include:
889 	 *		* There must be at least one inode in each
890 	 *		  cylinder group.
891 	 *		* The cylinder group overhead block, which
892 	 *		  contains the inode and frag bigmaps, must fit
893 	 *		  within one file system block.
894 	 *		* The space required for inode maps should
895 	 *		  occupy no more than a third of the cylinder
896 	 *		  group overhead block.
897 	 *		* The rotational position tables have to fit
898 	 *		  within the available space in the super block.
899 	 *	  Adjust the configuration values that can be adjusted
900 	 *	  so that these constraints are satisfied.  The
901 	 *	  configuration values that are adjustable are:
902 	 *		* frag size
903 	 *		* cylinders per group
904 	 *		* inode density (can be increased)
905 	 *		* number of rotational positions (the rotational
906 	 *		  position tables are eliminated altogether if
907 	 *		  there isn't enough room for them.)
908 	 * 4. Set the values for all the dependent configuration
909 	 *    values (those that aren't settable on the command
910 	 *    line and which are completely dependent on the
911 	 *    adjustable parameters).  This include cpc (cycles
912 	 *    per cylinder, spc (sectors-per-cylinder), and many others.
913 	 */
914 
915 	/*
916 	 * Figure out the partition size and initialize the label_type.
917 	 */
918 	max_fssize = get_max_size(fsi);
919 
920 	/*
921 	 * Get and check positional arguments, if any.
922 	 */
923 	switch (argc - 1) {
924 	default:
925 		usage();
926 		/*NOTREACHED*/
927 	case 15:
928 		mtb = checkmtb(argv[15]);
929 		/* FALLTHROUGH */
930 	case 14:
931 		string = argv[14];
932 		tmpmaxcontig = number(-1, "maxcontig", 0);
933 		maxcontig_flag = RC_POSITIONAL;
934 		/* FALLTHROUGH */
935 	case 13:
936 		string = argv[13];
937 		nrpos = number(NRPOS, "nrpos", 0);
938 		nrpos_flag = RC_POSITIONAL;
939 		/* FALLTHROUGH */
940 	case 12:
941 		string = argv[12];
942 		rotdelay = ROTDELAY;
943 		rotdelay_flag = RC_DEFAULT;
944 		/* FALLTHROUGH */
945 	case 11:
946 		string = argv[11];
947 		apc = number(0, "apc", 0);
948 		apc_flag = RC_POSITIONAL;
949 		/* FALLTHROUGH */
950 	case 10:
951 		opt = checkopt(argv[10]);
952 		/* FALLTHROUGH */
953 	case 9:
954 		string = argv[9];
955 		nbpi = number(NBPI, "nbpi", 0);
956 		nbpi_flag = RC_POSITIONAL;
957 		/* FALLTHROUGH */
958 	case 8:
959 		string = argv[8];
960 		rps = number(DEFHZ, "rps", 0);
961 		rps_flag = RC_POSITIONAL;
962 		/* FALLTHROUGH */
963 	case 7:
964 		string = argv[7];
965 		minfree = number(MINFREE, "free", ALLOW_PERCENT);
966 		minfree_flag = RC_POSITIONAL;
967 		/* FALLTHROUGH */
968 	case 6:
969 		string = argv[6];
970 		cpg = number(DESCPG, "cgsize", 0);
971 		cpg_flag = RC_POSITIONAL;
972 		/* FALLTHROUGH */
973 	case 5:
974 		string = argv[5];
975 		fragsize = number(DESFRAGSIZE, "fragsize", 0);
976 		fragsize_flag = RC_POSITIONAL;
977 		/* FALLTHROUGH */
978 	case 4:
979 		string = argv[4];
980 		bsize = number(DESBLKSIZE, "bsize", 0);
981 		bsize_flag = RC_POSITIONAL;
982 		/* FALLTHROUGH */
983 	case 3:
984 		string = argv[3];
985 		ntrack = number(DFLNTRAK, "ntrack", 0);
986 		ntrack_flag = RC_POSITIONAL;
987 		/* FALLTHROUGH */
988 	case 2:
989 		string = argv[2];
990 		nsect = number(DFLNSECT, "nsect", 0);
991 		nsect_flag = RC_POSITIONAL;
992 		/* FALLTHROUGH */
993 	case 1:
994 		string = argv[1];
995 		fssize_db = number(max_fssize, "size", 0);
996 	}
997 
998 	/*
999 	 * Initialize the parameters in the same way as newfs so that
1000 	 * newfs and mkfs would result in the same file system layout
1001 	 * for EFI labelled disks. Do this only in the absence of user
1002 	 * specified values for these parameters.
1003 	 */
1004 	if (label_type == LABEL_TYPE_EFI) {
1005 		if (apc_flag == RC_DEFAULT) apc = 0;
1006 		if (nrpos_flag == RC_DEFAULT) nrpos = 1;
1007 		if (ntrack_flag == RC_DEFAULT) ntrack = DEF_TRACKS_EFI;
1008 		if (rps_flag == RC_DEFAULT) rps = DEFHZ;
1009 		if (nsect_flag == RC_DEFAULT) nsect = DEF_SECTORS_EFI;
1010 	}
1011 
1012 	if ((maxcontig_flag == RC_DEFAULT) || (tmpmaxcontig == -1) ||
1013 	    (maxcontig == -1)) {
1014 		long maxtrax = get_max_track_size(fsi);
1015 		maxcontig = maxtrax / bsize;
1016 
1017 	} else {
1018 		maxcontig = tmpmaxcontig;
1019 	}
1020 	dprintf(("DeBuG maxcontig : %ld\n", maxcontig));
1021 
1022 	if (rotdelay == -1) {	/* default by newfs and mkfs */
1023 		rotdelay = ROTDELAY;
1024 	}
1025 
1026 	if (cpg_flag == RC_DEFAULT) { /* If not explicity set, use default */
1027 		cpg = DESCPG;
1028 	}
1029 	dprintf(("DeBuG cpg : %ld\n", cpg));
1030 
1031 	/*
1032 	 * Now that we have the semi-sane args, either positional, via -o,
1033 	 * or by defaulting, handle inter-dependencies and range checks.
1034 	 */
1035 
1036 	/*
1037 	 * Settle the file system block size first, since it's a fixed
1038 	 * parameter once set and so many other parameters, including
1039 	 * max_fssize, depend on it.
1040 	 */
1041 	range_check(&bsize, "bsize", MINBSIZE, MAXBSIZE, DESBLKSIZE,
1042 	    bsize_flag);
1043 
1044 	if (!POWEROF2(bsize)) {
1045 		(void) fprintf(stderr,
1046 		    gettext("block size must be a power of 2, not %ld\n"),
1047 		    bsize);
1048 		bsize = DESBLKSIZE;
1049 		(void) fprintf(stderr,
1050 		    gettext("mkfs: bsize reset to default %ld\n"),
1051 		    bsize);
1052 	}
1053 
1054 	if (fssize_db > max_fssize && validate_size(fsi, fssize_db)) {
1055 		(void) fprintf(stderr, gettext(
1056 		    "Warning: the requested size of this file system\n"
1057 		    "(%lld sectors) is greater than the size of the\n"
1058 		    "device reported by the driver (%lld sectors).\n"
1059 		    "However, a read of the device at the requested size\n"
1060 		    "does succeed, so the requested size will be used.\n"),
1061 		    fssize_db, max_fssize);
1062 		max_fssize = fssize_db;
1063 	}
1064 	/*
1065 	 * Since the maximum allocatable unit (the frag) must be less than
1066 	 * or equal to bsize, and the number of frags must be less than or
1067 	 * equal to INT_MAX, the total size of the file system (in
1068 	 * bytes) must be less than or equal to bsize * INT_MAX.
1069 	 */
1070 
1071 	if (max_fssize > ((diskaddr_t)bsize/DEV_BSIZE) * INT_MAX)
1072 		max_fssize = ((diskaddr_t)bsize/DEV_BSIZE) * INT_MAX;
1073 
1074 	range_check_64(&fssize_db, "size", 1024LL, max_fssize, max_fssize, 1);
1075 
1076 	if (fssize_db >= SECTORS_PER_TERABYTE) {
1077 		mtb = 'y';
1078 		if (!in_64bit_mode()) {
1079 			(void) fprintf(stderr, gettext(
1080 "mkfs:  Warning: Creating a file system greater than 1 terabyte on a\n"
1081 "       system running a 32-bit kernel.  This file system will not be\n"
1082 "       accessible until the system is rebooted with a 64-bit kernel.\n"));
1083 		}
1084 	}
1085 	dprintf(("DeBuG mtb : %c\n", mtb));
1086 
1087 	/*
1088 	 * With newer and much larger disks, the newfs(1M) and mkfs_ufs(1M)
1089 	 * commands had problems in correctly handling the "native" geometries
1090 	 * for various storage devices.
1091 	 *
1092 	 * To handle the new age disks, mkfs_ufs(1M) will use the EFI style
1093 	 * for non-EFI disks that are larger than the CHS addressing limit
1094 	 * ( > 8GB approx ) and ignore the disk geometry information for
1095 	 * these drives. This is what is currently done for multi-terrabyte
1096 	 * filesystems on EFI disks.
1097 	 *
1098 	 * However if the user asked for a specific layout by supplying values
1099 	 * for even one of the three parameters (nsect, ntrack, cpg), honour
1100 	 * the user supplied parameters.
1101 	 *
1102 	 * Choosing EFI style or native geometry style can make a lot of
1103 	 * difference, because the size of a cylinder group is dependent on
1104 	 * this choice. This in turn means that the position of alternate
1105 	 * superblocks varies depending on the style chosen. It is not
1106 	 * necessary that all disks of size > CHSLIMIT have EFI style layout.
1107 	 * There can be disks which are > CHSLIMIT size, but have native
1108 	 * geometry style layout, thereby warranting the need for alternate
1109 	 * logic in superblock detection.
1110 	 */
1111 	if (mtb != 'y' && (ntrack == -1 || GROW_WITH_DEFAULT_TRAK ||
1112 	    DEFAULT_SECT_TRAK_CPG)) {
1113 		/*
1114 		 * "-1" indicates that we were called from newfs and ntracks
1115 		 * was not specified in newfs command line. Calculate nsect
1116 		 * and ntrack in the same manner as newfs.
1117 		 *
1118 		 * This is required because, the defaults for nsect and ntrack
1119 		 * is hardcoded in mkfs, whereas to generate the alternate
1120 		 * superblock locations for the -N option, there is a need for
1121 		 * the geometry based values that newfs would have arrived at.
1122 		 * Newfs would have arrived at these values as below.
1123 		 */
1124 		if (label_type == LABEL_TYPE_EFI ||
1125 		    label_type == LABEL_TYPE_OTHER) {
1126 			use_efi_dflts = 1;
1127 			retry = 1;
1128 		} else if (ioctl(fsi, DKIOCGGEOM, &dkg)) {
1129 			dprintf(("%s: Unable to read Disk geometry", fsys));
1130 			perror(gettext("Unable to read Disk geometry"));
1131 			lockexit(32);
1132 		} else {
1133 			nsect = dkg.dkg_nsect;
1134 			ntrack = dkg.dkg_nhead;
1135 #ifdef i386	/* Bug 1170182 */
1136 			if (ntrack > 32 && (ntrack % 16) != 0) {
1137 				ntrack -= (ntrack % 16);
1138 			}
1139 #endif
1140 			if (ioctl(fsi, DKIOCREMOVABLE, &isremovable)) {
1141 				dprintf(("DeBuG Unable to determine if %s is"
1142 				    " Removable Media. Proceeding with system"
1143 				    " determined parameters.\n", fsys));
1144 				isremovable = 0;
1145 			}
1146 			if (ioctl(fsi, DKIOCHOTPLUGGABLE, &ishotpluggable)) {
1147 				dprintf(("DeBuG Unable to determine if %s is"
1148 				    " Hotpluggable Media. Proceeding with "
1149 				    "system determined parameters.\n", fsys));
1150 				ishotpluggable = 0;
1151 			}
1152 			if ((((diskaddr_t)dkg.dkg_ncyl * dkg.dkg_nhead *
1153 			    dkg.dkg_nsect) > CHSLIMIT) || isremovable ||
1154 			    ishotpluggable) {
1155 				use_efi_dflts = 1;
1156 				retry = 1;
1157 			}
1158 		}
1159 	}
1160 	dprintf(("DeBuG CHSLIMIT = %d geom = %llu\n", CHSLIMIT,
1161 	    (diskaddr_t)dkg.dkg_ncyl * dkg.dkg_nhead * dkg.dkg_nsect));
1162 	dprintf(("DeBuG label_type = %d isremovable = %d ishotpluggable = %d "
1163 	    "use_efi_dflts = %d\n", label_type, isremovable, ishotpluggable,
1164 	    use_efi_dflts));
1165 
1166 	/*
1167 	 * For the newfs -N case, even if the disksize is > CHSLIMIT, do not
1168 	 * blindly follow EFI style. If the fs_version indicates a geometry
1169 	 * based layout, try that one first. If it fails we can always try the
1170 	 * other logic.
1171 	 *
1172 	 * If we were called from growfs, we will have a problem if we mix
1173 	 * and match the filesystem creation and growth styles. For example,
1174 	 * if we create using EFI style, we have to also grow using EFI
1175 	 * style. So follow the style indicated by the fs_version.
1176 	 *
1177 	 * Read and verify the primary superblock. If it looks sane, use the
1178 	 * fs_version from the superblock. If the primary superblock does
1179 	 * not look good, read and verify the first alternate superblock at
1180 	 * ALTSB. Use the fs_version to decide whether to use the
1181 	 * EFI style logic or the old geometry based logic to calculate
1182 	 * the alternate superblock locations.
1183 	 */
1184 	if ((Nflag && use_efi_dflts) || (grow)) {
1185 		if (grow && ntrack_flag != RC_DEFAULT)
1186 			goto start_fs_creation;
1187 		rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize,
1188 		    (char *)&altsblock);
1189 		ret = checksblock(altsblock, 1);
1190 
1191 		if (!ret) {
1192 			if (altsblock.fs_magic == MTB_UFS_MAGIC) {
1193 				mtb = 'y';
1194 				goto start_fs_creation;
1195 			}
1196 			use_efi_dflts = (altsblock.fs_version ==
1197 			    UFS_EFISTYLE4NONEFI_VERSION_2) ? 1 : 0;
1198 		} else {
1199 			/*
1200 			 * The primary superblock didn't help in determining
1201 			 * the fs_version. Try the first alternate superblock.
1202 			 */
1203 			dprintf(("DeBuG checksblock() failed - error : %d"
1204 			    " for sb : %d\n", ret, SBOFF/sectorsize));
1205 			rdfs((diskaddr_t)ALTSB, (int)sbsize,
1206 			    (char *)&altsblock);
1207 			ret = checksblock(altsblock, 1);
1208 
1209 			if (!ret) {
1210 				if (altsblock.fs_magic == MTB_UFS_MAGIC) {
1211 					mtb = 'y';
1212 					goto start_fs_creation;
1213 				}
1214 				use_efi_dflts = (altsblock.fs_version ==
1215 				    UFS_EFISTYLE4NONEFI_VERSION_2) ? 1 : 0;
1216 			}
1217 			dprintf(("DeBuG checksblock() returned : %d"
1218 			    " for sb : %d\n", ret, ALTSB));
1219 		}
1220 	}
1221 
1222 	geom_nsect = nsect;
1223 	geom_ntrack = ntrack;
1224 	geom_cpg = cpg;
1225 	dprintf(("DeBuG geom_nsect=%d, geom_ntrack=%d, geom_cpg=%d\n",
1226 	    geom_nsect, geom_ntrack, geom_cpg));
1227 
1228 start_fs_creation:
1229 retry_alternate_logic:
1230 	invalid_sb_cnt = 0;
1231 	cg_too_small = 0;
1232 	if (use_efi_dflts) {
1233 		nsect = DEF_SECTORS_EFI;
1234 		ntrack = DEF_TRACKS_EFI;
1235 		cpg = DESCPG;
1236 		dprintf(("\nDeBuG Using EFI defaults\n"));
1237 	} else {
1238 		nsect = geom_nsect;
1239 		ntrack = geom_ntrack;
1240 		cpg = geom_cpg;
1241 		dprintf(("\nDeBuG Using Geometry\n"));
1242 		/*
1243 		 * 32K based on max block size of 64K, and rotational layout
1244 		 * test of nsect <= (256 * sectors/block).  Current block size
1245 		 * limit is not 64K, but it's growing soon.
1246 		 */
1247 		range_check(&nsect, "nsect", 1, 32768, DFLNSECT, nsect_flag);
1248 		/*
1249 		 * ntrack is the number of tracks per cylinder.
1250 		 * The ntrack value must be between 1 and the total number of
1251 		 * sectors in the file system.
1252 		 */
1253 		range_check(&ntrack, "ntrack", 1,
1254 		    fssize_db > INT_MAX ? INT_MAX : (uint32_t)fssize_db,
1255 		    DFLNTRAK, ntrack_flag);
1256 	}
1257 
1258 	range_check(&apc, "apc", 0, nsect - 1, 0, apc_flag);
1259 
1260 	if (mtb == 'y')
1261 		fragsize = bsize;
1262 
1263 	range_check(&fragsize, "fragsize", sectorsize, bsize,
1264 	    MAX(bsize / MAXFRAG, MIN(DESFRAGSIZE, bsize)), fragsize_flag);
1265 
1266 	if ((bsize / MAXFRAG) > fragsize) {
1267 		(void) fprintf(stderr, gettext(
1268 "fragment size %ld is too small, minimum with block size %ld is %ld\n"),
1269 		    fragsize, bsize, bsize / MAXFRAG);
1270 		(void) fprintf(stderr,
1271 		    gettext("mkfs: fragsize reset to minimum %ld\n"),
1272 		    bsize / MAXFRAG);
1273 		fragsize = bsize / MAXFRAG;
1274 	}
1275 
1276 	if (!POWEROF2(fragsize)) {
1277 		(void) fprintf(stderr,
1278 		    gettext("fragment size must be a power of 2, not %ld\n"),
1279 		    fragsize);
1280 		fragsize = MAX(bsize / MAXFRAG, MIN(DESFRAGSIZE, bsize));
1281 		(void) fprintf(stderr,
1282 		    gettext("mkfs: fragsize reset to %ld\n"),
1283 		    fragsize);
1284 	}
1285 
1286 	/* At this point, bsize must be >= fragsize, so no need to check it */
1287 
1288 	if (bsize < PAGESIZE) {
1289 		(void) fprintf(stderr, gettext(
1290 		    "WARNING: filesystem block size (%ld) is smaller than "
1291 		    "memory page size (%ld).\nResulting filesystem can not be "
1292 		    "mounted on this system.\n\n"),
1293 		    bsize, (long)PAGESIZE);
1294 	}
1295 
1296 	range_check(&rps, "rps", 1, 1000, DEFHZ, rps_flag);
1297 	range_check(&minfree, "free", 0, 99, MINFREE, minfree_flag);
1298 	range_check(&nrpos, "nrpos", 1, nsect, MIN(nsect, NRPOS), nrpos_flag);
1299 
1300 	/*
1301 	 * nbpi is variable, but 2MB seems a reasonable upper limit,
1302 	 * as 4MB tends to cause problems (using otherwise-default
1303 	 * parameters).  The true limit is where we end up with one
1304 	 * inode per cylinder group.  If this file system is being
1305 	 * configured for multi-terabyte access, nbpi must be at least 1MB.
1306 	 */
1307 	if (mtb == 'y' && nbpi < MTB_NBPI) {
1308 		if (nbpi_flag != RC_DEFAULT)
1309 			(void) fprintf(stderr, gettext("mkfs: bad value for "
1310 			    "nbpi: must be at least 1048576 for multi-terabyte,"
1311 			    " nbpi reset to default 1048576\n"));
1312 		nbpi = MTB_NBPI;
1313 	}
1314 
1315 	if (mtb == 'y')
1316 		range_check(&nbpi, "nbpi", MTB_NBPI, 2 * MB, MTB_NBPI,
1317 		    nbpi_flag);
1318 	else
1319 		range_check(&nbpi, "nbpi", DEV_BSIZE, 2 * MB, NBPI, nbpi_flag);
1320 
1321 	/*
1322 	 * maxcpg is another variably-limited parameter.  Calculate
1323 	 * the limit based on what we've got for its dependent
1324 	 * variables.  Effectively, it's how much space is left in the
1325 	 * superblock after all the other bits are accounted for.  We
1326 	 * only fill in sblock fields so we can use MAXIpG.
1327 	 *
1328 	 * If the calculation of maxcpg below (for the mtb == 'n'
1329 	 * case) is changed, update newfs as well.
1330 	 *
1331 	 * For old-style, non-MTB format file systems, use the old
1332 	 * algorithm for calculating the maximum cylinder group size,
1333 	 * even though it limits the cylinder group more than necessary.
1334 	 * Since layout can affect performance, we don't want to change
1335 	 * the default layout for non-MTB file systems at this time.
1336 	 * However, for MTB file systems, use the new maxcpg calculation,
1337 	 * which really maxes out the cylinder group size.
1338 	 */
1339 
1340 	sblock.fs_bsize = bsize;
1341 	sblock.fs_inopb = sblock.fs_bsize / sizeof (struct dinode);
1342 
1343 	if (mtb == 'n') {
1344 		maxcpg = (bsize - sizeof (struct cg) -
1345 		    howmany(MAXIpG(&sblock), NBBY)) /
1346 		    (sizeof (long) + nrpos * sizeof (short) +
1347 		    nsect / (MAXFRAG * NBBY));
1348 	} else {
1349 		maxcpg = compute_maxcpg(bsize, fragsize, nbpi, nrpos,
1350 		    nsect * ntrack);
1351 	}
1352 
1353 	dprintf(("DeBuG cpg : %ld\n", cpg));
1354 	/*
1355 	 * Increase the cpg to maxcpg if either newfs was invoked
1356 	 * with -T option or if mkfs wants to create a mtb file system
1357 	 * and if the user has not specified the cpg.
1358 	 */
1359 	if (cpg == -1 || (mtb == 'y' && cpg_flag == RC_DEFAULT))
1360 		cpg = maxcpg;
1361 	dprintf(("DeBuG cpg : %ld\n", cpg));
1362 
1363 	/*
1364 	 * mincpg is variable in complex ways, so we really can't
1365 	 * do a sane lower-end limit check at this point.
1366 	 */
1367 	range_check(&cpg, "cgsize", 1, maxcpg, MIN(maxcpg, DESCPG), cpg_flag);
1368 
1369 	/*
1370 	 * get the controller info
1371 	 */
1372 	islog = 0;
1373 	islogok = 0;
1374 	waslog = 0;
1375 
1376 	/*
1377 	 * Do not grow the file system, but print on stdout the maximum
1378 	 * size in sectors to which the file system can be increased.
1379 	 * The calculated size is limited by fssize_db.
1380 	 * Note that we don't lock the filesystem and therefore under rare
1381 	 * conditions (the filesystem is mounted, the free block count is
1382 	 * almost zero, and the superuser is still changing it) the calculated
1383 	 * size can be imprecise.
1384 	 */
1385 	if (Pflag) {
1386 		(void) printf("%llu\n", probe_summaryinfo());
1387 		exit(0);
1388 	}
1389 
1390 	/*
1391 	 * If we're growing an existing filesystem, then we're about
1392 	 * to start doing things that can require recovery efforts if
1393 	 * we get interrupted, so make sure we get a chance to do so.
1394 	 */
1395 	if (grow) {
1396 		sigact.sa_handler = recover_from_sigint;
1397 		sigemptyset(&sigact.sa_mask);
1398 		sigact.sa_flags = SA_RESTART;
1399 
1400 		if (sigaction(SIGINT, &sigact, (struct sigaction *)NULL) < 0) {
1401 			perror(gettext("Could not register SIGINT handler"));
1402 			lockexit(3);
1403 		}
1404 	}
1405 
1406 	if (!Nflag) {
1407 		/*
1408 		 * Check if MNTTAB is trustable
1409 		 */
1410 		if (statvfs64(MNTTAB, &fs) < 0) {
1411 			(void) fprintf(stderr, gettext("can't statvfs %s\n"),
1412 			    MNTTAB);
1413 			exit(32);
1414 		}
1415 
1416 		if (strcmp(MNTTYPE_MNTFS, fs.f_basetype) != 0) {
1417 			(void) fprintf(stderr, gettext(
1418 			    "%s file system type is not %s, can't mkfs\n"),
1419 			    MNTTAB, MNTTYPE_MNTFS);
1420 			exit(32);
1421 		}
1422 
1423 		special = getfullblkname(fsys);
1424 		checkdev(fsys, special);
1425 
1426 		/*
1427 		 * If we found the block device name,
1428 		 * then check the mount table.
1429 		 * if mounted, and growing write lock the file system
1430 		 *
1431 		 */
1432 		if ((special != NULL) && (*special != '\0')) {
1433 			if ((mnttab = fopen(MNTTAB, "r")) == NULL) {
1434 				(void) fprintf(stderr, gettext(
1435 				    "can't open %s\n"), MNTTAB);
1436 				exit(32);
1437 			}
1438 			while ((getmntent(mnttab, &mntp)) == NULL) {
1439 				if (grow) {
1440 					checkmount(&mntp, special);
1441 					continue;
1442 				}
1443 				if (strcmp(special, mntp.mnt_special) == 0) {
1444 					(void) fprintf(stderr, gettext(
1445 					    "%s is mounted, can't mkfs\n"),
1446 					    special);
1447 					exit(32);
1448 				}
1449 			}
1450 			(void) fclose(mnttab);
1451 		}
1452 
1453 		if (directory && (ismounted == 0)) {
1454 			(void) fprintf(stderr, gettext("%s is not mounted\n"),
1455 			    special);
1456 			lockexit(32);
1457 		}
1458 
1459 		fso = (grow) ? open64(fsys, O_WRONLY) : creat64(fsys, 0666);
1460 		if (fso < 0) {
1461 			saverr = errno;
1462 			(void) fprintf(stderr,
1463 			    gettext("%s: cannot create: %s\n"),
1464 			    fsys, strerror(saverr));
1465 			lockexit(32);
1466 		}
1467 
1468 	} else {
1469 
1470 		/*
1471 		 * For the -N case, a file descriptor is needed for the llseek()
1472 		 * in wtfs(). See the comment in wtfs() for more information.
1473 		 *
1474 		 * Get a file descriptor that's read-only so that this code
1475 		 * doesn't accidentally write to the file.
1476 		 */
1477 		fso = open64(fsys, O_RDONLY);
1478 		if (fso < 0) {
1479 			saverr = errno;
1480 			(void) fprintf(stderr, gettext("%s: cannot open: %s\n"),
1481 			    fsys, strerror(saverr));
1482 			lockexit(32);
1483 		}
1484 	}
1485 
1486 	/*
1487 	 * Check the media sector size
1488 	 */
1489 	if (ioctl(fso, DKIOCGMEDIAINFO, &dkminfo) != -1) {
1490 		if (dkminfo.dki_lbsize != 0 &&
1491 		    POWEROF2(dkminfo.dki_lbsize / DEV_BSIZE) &&
1492 		    dkminfo.dki_lbsize != DEV_BSIZE) {
1493 			fprintf(stderr,
1494 			    gettext("The device sector size %u is not "
1495 			    "supported by ufs!\n"), dkminfo.dki_lbsize);
1496 			(void) close(fso);
1497 			exit(1);
1498 		}
1499 	}
1500 
1501 	/*
1502 	 * seed random # generator (for ic_generation)
1503 	 */
1504 #ifdef MKFS_DEBUG
1505 	srand48(12962);	/* reproducible results */
1506 #else
1507 	srand48((long)(time((time_t *)NULL) + getpid()));
1508 #endif
1509 
1510 	if (grow) {
1511 		growinit(fsys);
1512 		goto grow00;
1513 	}
1514 
1515 	/*
1516 	 * Validate the given file system size.
1517 	 * Verify that its last block can actually be accessed.
1518 	 *
1519 	 * Note: it's ok to use sblock as a buffer because it is immediately
1520 	 * overwritten by the rdfs() of the superblock in the next line.
1521 	 *
1522 	 * ToDo: Because the size checking is done in rdfs()/wtfs(), the
1523 	 * error message for specifying an illegal size is very unfriendly.
1524 	 * In the future, one could replace the rdfs()/wtfs() calls
1525 	 * below with in-line calls to read() or write(). This allows better
1526 	 * error messages to be put in place.
1527 	 */
1528 	rdfs(fssize_db - 1, (int)sectorsize, (char *)&sblock);
1529 
1530 	/*
1531 	 * make the fs unmountable
1532 	 */
1533 	rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
1534 	sblock.fs_magic = -1;
1535 	sblock.fs_clean = FSBAD;
1536 	sblock.fs_state = FSOKAY - sblock.fs_time;
1537 	wtfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
1538 	bzero(&sblock, (size_t)sbsize);
1539 
1540 	sblock.fs_nsect = nsect;
1541 	sblock.fs_ntrak = ntrack;
1542 
1543 	/*
1544 	 * Validate specified/determined spc
1545 	 * and calculate minimum cylinders per group.
1546 	 */
1547 
1548 	/*
1549 	 * sectors/cyl = tracks/cyl * sectors/track
1550 	 */
1551 	sblock.fs_spc = sblock.fs_ntrak * sblock.fs_nsect;
1552 
1553 grow00:
1554 	if (apc_flag) {
1555 		sblock.fs_spc -= apc;
1556 	}
1557 	/*
1558 	 * Have to test for this separately from apc_flag, due to
1559 	 * the growfs case....
1560 	 */
1561 	if (sblock.fs_spc != sblock.fs_ntrak * sblock.fs_nsect) {
1562 		spc_flag = 1;
1563 	}
1564 	if (grow)
1565 		goto grow10;
1566 
1567 	sblock.fs_nrpos = nrpos;
1568 	sblock.fs_bsize = bsize;
1569 	sblock.fs_fsize = fragsize;
1570 	sblock.fs_minfree = minfree;
1571 
1572 grow10:
1573 	if (nbpi < sblock.fs_fsize) {
1574 		(void) fprintf(stderr, gettext(
1575 		"warning: wasteful data byte allocation / inode (nbpi):\n"));
1576 		(void) fprintf(stderr, gettext(
1577 		    "%ld smaller than allocatable fragment size of %d\n"),
1578 		    nbpi, sblock.fs_fsize);
1579 	}
1580 	if (grow)
1581 		goto grow20;
1582 
1583 	if (opt == 's')
1584 		sblock.fs_optim = FS_OPTSPACE;
1585 	else
1586 		sblock.fs_optim = FS_OPTTIME;
1587 
1588 	sblock.fs_bmask = ~(sblock.fs_bsize - 1);
1589 	sblock.fs_fmask = ~(sblock.fs_fsize - 1);
1590 	/*
1591 	 * Planning now for future expansion.
1592 	 */
1593 #if defined(_BIG_ENDIAN)
1594 		sblock.fs_qbmask.val[0] = 0;
1595 		sblock.fs_qbmask.val[1] = ~sblock.fs_bmask;
1596 		sblock.fs_qfmask.val[0] = 0;
1597 		sblock.fs_qfmask.val[1] = ~sblock.fs_fmask;
1598 #endif
1599 #if defined(_LITTLE_ENDIAN)
1600 		sblock.fs_qbmask.val[0] = ~sblock.fs_bmask;
1601 		sblock.fs_qbmask.val[1] = 0;
1602 		sblock.fs_qfmask.val[0] = ~sblock.fs_fmask;
1603 		sblock.fs_qfmask.val[1] = 0;
1604 #endif
1605 	for (sblock.fs_bshift = 0, i = sblock.fs_bsize; i > 1; i >>= 1)
1606 		sblock.fs_bshift++;
1607 	for (sblock.fs_fshift = 0, i = sblock.fs_fsize; i > 1; i >>= 1)
1608 		sblock.fs_fshift++;
1609 	sblock.fs_frag = numfrags(&sblock, sblock.fs_bsize);
1610 	for (sblock.fs_fragshift = 0, i = sblock.fs_frag; i > 1; i >>= 1)
1611 		sblock.fs_fragshift++;
1612 	if (sblock.fs_frag > MAXFRAG) {
1613 		(void) fprintf(stderr, gettext(
1614 	"fragment size %d is too small, minimum with block size %d is %d\n"),
1615 		    sblock.fs_fsize, sblock.fs_bsize,
1616 		    sblock.fs_bsize / MAXFRAG);
1617 		lockexit(32);
1618 	}
1619 	sblock.fs_nindir = sblock.fs_bsize / sizeof (daddr32_t);
1620 	sblock.fs_inopb = sblock.fs_bsize / sizeof (struct dinode);
1621 	sblock.fs_nspf = sblock.fs_fsize / sectorsize;
1622 	for (sblock.fs_fsbtodb = 0, i = NSPF(&sblock); i > 1; i >>= 1)
1623 		sblock.fs_fsbtodb++;
1624 
1625 	/*
1626 	 * Compute the super-block, cylinder group, and inode blocks.
1627 	 * Note that these "blkno" are really fragment addresses.
1628 	 * For example, on an 8K/1K (block/fragment) system, fs_sblkno is 16,
1629 	 * fs_cblkno is 24, and fs_iblkno is 32. This is why CGSIZE is so
1630 	 * important: only 1 FS block is allocated for the cg struct (fragment
1631 	 * numbers 24 through 31).
1632 	 */
1633 	sblock.fs_sblkno =
1634 	    roundup(howmany(bbsize + sbsize, sblock.fs_fsize), sblock.fs_frag);
1635 	sblock.fs_cblkno = (daddr32_t)(sblock.fs_sblkno +
1636 	    roundup(howmany(sbsize, sblock.fs_fsize), sblock.fs_frag));
1637 	sblock.fs_iblkno = sblock.fs_cblkno + sblock.fs_frag;
1638 
1639 	sblock.fs_cgoffset = roundup(
1640 	    howmany(sblock.fs_nsect, NSPF(&sblock)), sblock.fs_frag);
1641 	for (sblock.fs_cgmask = -1, i = sblock.fs_ntrak; i > 1; i >>= 1)
1642 		sblock.fs_cgmask <<= 1;
1643 	if (!POWEROF2(sblock.fs_ntrak))
1644 		sblock.fs_cgmask <<= 1;
1645 	/*
1646 	 * Validate specified/determined spc
1647 	 * and calculate minimum cylinders per group.
1648 	 */
1649 
1650 	for (sblock.fs_cpc = NSPB(&sblock), i = sblock.fs_spc;
1651 	    sblock.fs_cpc > 1 && (i & 1) == 0;
1652 	    sblock.fs_cpc >>= 1, i >>= 1)
1653 		/* void */;
1654 	mincpc = sblock.fs_cpc;
1655 
1656 	/* if these calculations are changed, check dump_fscmd also */
1657 	bpcg = (uint64_t)sblock.fs_spc * sectorsize;
1658 	inospercg = (uint64_t)roundup(bpcg / sizeof (struct dinode),
1659 	    INOPB(&sblock));
1660 	if (inospercg > MAXIpG(&sblock))
1661 		inospercg = MAXIpG(&sblock);
1662 	used = (uint64_t)(sblock.fs_iblkno + inospercg /
1663 	    INOPF(&sblock)) * NSPF(&sblock);
1664 	mincpgcnt = (long)howmany((uint64_t)sblock.fs_cgoffset *
1665 	    (~sblock.fs_cgmask) + used, sblock.fs_spc);
1666 	mincpg = roundup(mincpgcnt, mincpc);
1667 	/*
1668 	 * Insure that cylinder group with mincpg has enough space
1669 	 * for block maps
1670 	 */
1671 	sblock.fs_cpg = mincpg;
1672 	sblock.fs_ipg = (int32_t)inospercg;
1673 	mapcramped = 0;
1674 
1675 	/*
1676 	 * Make sure the cg struct fits within the file system block.
1677 	 * Use larger block sizes until it fits
1678 	 */
1679 	while (CGSIZE(&sblock) > sblock.fs_bsize) {
1680 		mapcramped = 1;
1681 		if (sblock.fs_bsize < MAXBSIZE) {
1682 			sblock.fs_bsize <<= 1;
1683 			if ((i & 1) == 0) {
1684 				i >>= 1;
1685 			} else {
1686 				sblock.fs_cpc <<= 1;
1687 				mincpc <<= 1;
1688 				mincpg = roundup(mincpgcnt, mincpc);
1689 				sblock.fs_cpg = mincpg;
1690 			}
1691 			sblock.fs_frag <<= 1;
1692 			sblock.fs_fragshift += 1;
1693 			if (sblock.fs_frag <= MAXFRAG)
1694 				continue;
1695 		}
1696 
1697 		/*
1698 		 * Looped far enough. The fragment is now as large as the
1699 		 * filesystem block!
1700 		 */
1701 		if (sblock.fs_fsize == sblock.fs_bsize) {
1702 			(void) fprintf(stderr, gettext(
1703 		    "There is no block size that can support this disk\n"));
1704 			lockexit(32);
1705 		}
1706 
1707 		/*
1708 		 * Try a larger fragment. Double the fragment size.
1709 		 */
1710 		sblock.fs_frag >>= 1;
1711 		sblock.fs_fragshift -= 1;
1712 		sblock.fs_fsize <<= 1;
1713 		sblock.fs_nspf <<= 1;
1714 	}
1715 	/*
1716 	 * Insure that cylinder group with mincpg has enough space for inodes
1717 	 */
1718 	inodecramped = 0;
1719 	used *= sectorsize;
1720 	nbytes64 = (uint64_t)mincpg * bpcg - used;
1721 	inospercg = (uint64_t)roundup((nbytes64 / nbpi), INOPB(&sblock));
1722 	sblock.fs_ipg = (int32_t)inospercg;
1723 	while (inospercg > MAXIpG(&sblock)) {
1724 		inodecramped = 1;
1725 		if (mincpc == 1 || sblock.fs_frag == 1 ||
1726 		    sblock.fs_bsize == MINBSIZE)
1727 			break;
1728 		nbytes64 = (uint64_t)mincpg * bpcg - used;
1729 		(void) fprintf(stderr,
1730 		    gettext("With a block size of %d %s %lu\n"),
1731 		    sblock.fs_bsize, gettext("minimum bytes per inode is"),
1732 		    (uint32_t)(nbytes64 / MAXIpG(&sblock) + 1));
1733 		sblock.fs_bsize >>= 1;
1734 		sblock.fs_frag >>= 1;
1735 		sblock.fs_fragshift -= 1;
1736 		mincpc >>= 1;
1737 		sblock.fs_cpg = roundup(mincpgcnt, mincpc);
1738 		if (CGSIZE(&sblock) > sblock.fs_bsize) {
1739 			sblock.fs_bsize <<= 1;
1740 			break;
1741 		}
1742 		mincpg = sblock.fs_cpg;
1743 		nbytes64 = (uint64_t)mincpg * bpcg - used;
1744 		inospercg = (uint64_t)roundup((nbytes64 / nbpi),
1745 		    INOPB(&sblock));
1746 		sblock.fs_ipg = (int32_t)inospercg;
1747 	}
1748 	if (inodecramped) {
1749 		if (inospercg > MAXIpG(&sblock)) {
1750 			nbytes64 = (uint64_t)mincpg * bpcg - used;
1751 			(void) fprintf(stderr, gettext(
1752 			    "Minimum bytes per inode is %d\n"),
1753 			    (uint32_t)(nbytes64 / MAXIpG(&sblock) + 1));
1754 		} else if (!mapcramped) {
1755 			(void) fprintf(stderr, gettext(
1756 	    "With %ld bytes per inode, minimum cylinders per group is %ld\n"),
1757 			    nbpi, mincpg);
1758 		}
1759 	}
1760 	if (mapcramped) {
1761 		(void) fprintf(stderr, gettext(
1762 		    "With %d sectors per cylinder, minimum cylinders "
1763 		    "per group is %ld\n"),
1764 		    sblock.fs_spc, mincpg);
1765 	}
1766 	if (inodecramped || mapcramped) {
1767 		/*
1768 		 * To make this at least somewhat comprehensible in
1769 		 * the world of i18n, figure out what we're going to
1770 		 * say and then say it all at one time.  The days of
1771 		 * needing to scrimp on string space are behind us....
1772 		 */
1773 		if ((sblock.fs_bsize != bsize) &&
1774 		    (sblock.fs_fsize != fragsize)) {
1775 			(void) fprintf(stderr, gettext(
1776 	    "This requires the block size to be changed from %ld to %d\n"
1777 	    "and the fragment size to be changed from %ld to %d\n"),
1778 			    bsize, sblock.fs_bsize,
1779 			    fragsize, sblock.fs_fsize);
1780 		} else if (sblock.fs_bsize != bsize) {
1781 			(void) fprintf(stderr, gettext(
1782 	    "This requires the block size to be changed from %ld to %d\n"),
1783 			    bsize, sblock.fs_bsize);
1784 		} else if (sblock.fs_fsize != fragsize) {
1785 			(void) fprintf(stderr, gettext(
1786 	    "This requires the fragment size to be changed from %ld to %d\n"),
1787 			    fragsize, sblock.fs_fsize);
1788 		} else {
1789 			(void) fprintf(stderr, gettext(
1790 	    "Unable to make filesystem fit with the given constraints\n"));
1791 		}
1792 		(void) fprintf(stderr, gettext(
1793 		    "Please re-run mkfs with corrected parameters\n"));
1794 		lockexit(32);
1795 	}
1796 	/*
1797 	 * Calculate the number of cylinders per group
1798 	 */
1799 	sblock.fs_cpg = cpg;
1800 	if (sblock.fs_cpg % mincpc != 0) {
1801 		(void) fprintf(stderr, gettext(
1802 		    "Warning: cylinder groups must have a multiple "
1803 		    "of %ld cylinders with the given\n         parameters\n"),
1804 		    mincpc);
1805 		sblock.fs_cpg = roundup(sblock.fs_cpg, mincpc);
1806 		(void) fprintf(stderr, gettext("Rounded cgsize up to %d\n"),
1807 		    sblock.fs_cpg);
1808 	}
1809 	/*
1810 	 * Must insure there is enough space for inodes
1811 	 */
1812 	/* if these calculations are changed, check dump_fscmd also */
1813 	nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
1814 	sblock.fs_ipg = roundup((uint32_t)(nbytes64 / nbpi), INOPB(&sblock));
1815 
1816 	/*
1817 	 * Slim down cylinders per group, until the inodes can fit.
1818 	 */
1819 	while (sblock.fs_ipg > MAXIpG(&sblock)) {
1820 		inodecramped = 1;
1821 		sblock.fs_cpg -= mincpc;
1822 		nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
1823 		sblock.fs_ipg = roundup((uint32_t)(nbytes64 / nbpi),
1824 		    INOPB(&sblock));
1825 	}
1826 	/*
1827 	 * Must insure there is enough space to hold block map.
1828 	 * Cut down on cylinders per group, until the cg struct fits in a
1829 	 * filesystem block.
1830 	 */
1831 	while (CGSIZE(&sblock) > sblock.fs_bsize) {
1832 		mapcramped = 1;
1833 		sblock.fs_cpg -= mincpc;
1834 		nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
1835 		sblock.fs_ipg = roundup((uint32_t)(nbytes64 / nbpi),
1836 		    INOPB(&sblock));
1837 	}
1838 	sblock.fs_fpg = (sblock.fs_cpg * sblock.fs_spc) / NSPF(&sblock);
1839 	if ((sblock.fs_cpg * sblock.fs_spc) % NSPB(&sblock) != 0) {
1840 		(void) fprintf(stderr,
1841 		gettext("newfs: panic (fs_cpg * fs_spc) %% NSPF != 0\n"));
1842 		lockexit(32);
1843 	}
1844 	if (sblock.fs_cpg < mincpg) {
1845 		(void) fprintf(stderr, gettext(
1846 "With the given parameters, cgsize must be at least %ld; please re-run mkfs\n"),
1847 		    mincpg);
1848 		lockexit(32);
1849 	}
1850 	sblock.fs_cgsize = fragroundup(&sblock, CGSIZE(&sblock));
1851 grow20:
1852 	/*
1853 	 * Now have size for file system and nsect and ntrak.
1854 	 * Determine number of cylinders and blocks in the file system.
1855 	 */
1856 	fssize_frag = (int64_t)dbtofsb(&sblock, fssize_db);
1857 	if (fssize_frag > INT_MAX) {
1858 		(void) fprintf(stderr, gettext(
1859 "There are too many fragments in the system, increase fragment size\n"),
1860 		    mincpg);
1861 		lockexit(32);
1862 	}
1863 	sblock.fs_size = (int32_t)fssize_frag;
1864 	sblock.fs_ncyl = (int32_t)(fssize_frag * NSPF(&sblock) / sblock.fs_spc);
1865 	if (fssize_frag * NSPF(&sblock) >
1866 	    (uint64_t)sblock.fs_ncyl * sblock.fs_spc) {
1867 		sblock.fs_ncyl++;
1868 		warn = 1;
1869 	}
1870 	if (sblock.fs_ncyl < 1) {
1871 		(void) fprintf(stderr, gettext(
1872 		    "file systems must have at least one cylinder\n"));
1873 		lockexit(32);
1874 	}
1875 	if (grow)
1876 		goto grow30;
1877 	/*
1878 	 * Determine feasability/values of rotational layout tables.
1879 	 *
1880 	 * The size of the rotational layout tables is limited by the size
1881 	 * of the file system block, fs_bsize.  The amount of space
1882 	 * available for tables is calculated as (fs_bsize - sizeof (struct
1883 	 * fs)).  The size of these tables is inversely proportional to the
1884 	 * block size of the file system. The size increases if sectors per
1885 	 * track are not powers of two, because more cylinders must be
1886 	 * described by the tables before the rotational pattern repeats
1887 	 * (fs_cpc).
1888 	 */
1889 	sblock.fs_postblformat = FS_DYNAMICPOSTBLFMT;
1890 	sblock.fs_sbsize = fragroundup(&sblock, sizeof (struct fs));
1891 	sblock.fs_npsect = sblock.fs_nsect;
1892 	if (sblock.fs_ntrak == 1) {
1893 		sblock.fs_cpc = 0;
1894 		goto next;
1895 	}
1896 	postblsize = sblock.fs_nrpos * sblock.fs_cpc * sizeof (short);
1897 	rotblsize = sblock.fs_cpc * sblock.fs_spc / NSPB(&sblock);
1898 	totalsbsize = sizeof (struct fs) + rotblsize;
1899 
1900 	/* do static allocation if nrpos == 8 and fs_cpc == 16  */
1901 	if (sblock.fs_nrpos == 8 && sblock.fs_cpc <= 16) {
1902 		/* use old static table space */
1903 		sblock.fs_postbloff = (char *)(&sblock.fs_opostbl[0][0]) -
1904 		    (char *)(&sblock.fs_link);
1905 		sblock.fs_rotbloff = &sblock.fs_space[0] -
1906 		    (uchar_t *)(&sblock.fs_link);
1907 	} else {
1908 		/* use 4.3 dynamic table space */
1909 		sblock.fs_postbloff = &sblock.fs_space[0] -
1910 		    (uchar_t *)(&sblock.fs_link);
1911 		sblock.fs_rotbloff = sblock.fs_postbloff + postblsize;
1912 		totalsbsize += postblsize;
1913 	}
1914 	if (totalsbsize > sblock.fs_bsize ||
1915 	    sblock.fs_nsect > (1 << NBBY) * NSPB(&sblock)) {
1916 		(void) fprintf(stderr, gettext(
1917 		    "Warning: insufficient space in super block for\n"
1918 		    "rotational layout tables with nsect %d, ntrack %d, "
1919 		    "and nrpos %d.\nOmitting tables - file system "
1920 		    "performance may be impaired.\n"),
1921 		    sblock.fs_nsect, sblock.fs_ntrak, sblock.fs_nrpos);
1922 
1923 		/*
1924 		 * Setting fs_cpc to 0 tells alloccgblk() in ufs_alloc.c to
1925 		 * ignore the positional layout table and rotational
1926 		 * position table.
1927 		 */
1928 		sblock.fs_cpc = 0;
1929 		goto next;
1930 	}
1931 	sblock.fs_sbsize = fragroundup(&sblock, totalsbsize);
1932 
1933 
1934 	/*
1935 	 * calculate the available blocks for each rotational position
1936 	 */
1937 	for (cylno = 0; cylno < sblock.fs_cpc; cylno++)
1938 		for (rpos = 0; rpos < sblock.fs_nrpos; rpos++)
1939 			fs_postbl(&sblock, cylno)[rpos] = -1;
1940 	for (i = (rotblsize - 1) * sblock.fs_frag;
1941 	    i >= 0; i -= sblock.fs_frag) {
1942 		cylno = cbtocylno(&sblock, i);
1943 		rpos = cbtorpos(&sblock, i);
1944 		blk = fragstoblks(&sblock, i);
1945 		if (fs_postbl(&sblock, cylno)[rpos] == -1)
1946 			fs_rotbl(&sblock)[blk] = 0;
1947 		else
1948 			fs_rotbl(&sblock)[blk] =
1949 			    fs_postbl(&sblock, cylno)[rpos] - blk;
1950 		fs_postbl(&sblock, cylno)[rpos] = blk;
1951 	}
1952 next:
1953 grow30:
1954 	/*
1955 	 * Compute/validate number of cylinder groups.
1956 	 * Note that if an excessively large filesystem is specified
1957 	 * (e.g., more than 16384 cylinders for an 8K filesystem block), it
1958 	 * does not get detected until checksummarysize()
1959 	 */
1960 	sblock.fs_ncg = sblock.fs_ncyl / sblock.fs_cpg;
1961 	if (sblock.fs_ncyl % sblock.fs_cpg)
1962 		sblock.fs_ncg++;
1963 	sblock.fs_dblkno = sblock.fs_iblkno + sblock.fs_ipg / INOPF(&sblock);
1964 	i = MIN(~sblock.fs_cgmask, sblock.fs_ncg - 1);
1965 	ibpcl = cgdmin(&sblock, i) - cgbase(&sblock, i);
1966 	if (ibpcl >= sblock.fs_fpg) {
1967 		(void) fprintf(stderr, gettext(
1968 		    "inode blocks/cyl group (%d) >= data blocks (%d)\n"),
1969 		    cgdmin(&sblock, i) - cgbase(&sblock, i) / sblock.fs_frag,
1970 		    sblock.fs_fpg / sblock.fs_frag);
1971 		if ((ibpcl < 0) || (sblock.fs_fpg < 0)) {
1972 			(void) fprintf(stderr, gettext(
1973 	    "number of cylinders per cylinder group (%d) must be decreased.\n"),
1974 			    sblock.fs_cpg);
1975 		} else {
1976 			(void) fprintf(stderr, gettext(
1977 	    "number of cylinders per cylinder group (%d) must be increased.\n"),
1978 			    sblock.fs_cpg);
1979 		}
1980 		(void) fprintf(stderr, gettext(
1981 "Note that cgsize may have been adjusted to allow struct cg to fit.\n"));
1982 		lockexit(32);
1983 	}
1984 	j = sblock.fs_ncg - 1;
1985 	if ((i = fssize_frag - j * sblock.fs_fpg) < sblock.fs_fpg &&
1986 	    cgdmin(&sblock, j) - cgbase(&sblock, j) > i) {
1987 		(void) fprintf(stderr, gettext(
1988 		    "Warning: inode blocks/cyl group (%d) >= data "
1989 		    "blocks (%ld) in last\n    cylinder group. This "
1990 		    "implies %ld sector(s) cannot be allocated.\n"),
1991 		    (cgdmin(&sblock, j) - cgbase(&sblock, j)) / sblock.fs_frag,
1992 		    i / sblock.fs_frag, i * NSPF(&sblock));
1993 		/*
1994 		 * If there is only one cylinder group and that is not even
1995 		 * big enough to hold the inodes, exit.
1996 		 */
1997 		if (sblock.fs_ncg == 1)
1998 			cg_too_small = 1;
1999 		sblock.fs_ncg--;
2000 		sblock.fs_ncyl = sblock.fs_ncg * sblock.fs_cpg;
2001 		sblock.fs_size = fssize_frag =
2002 		    (int64_t)sblock.fs_ncyl * (int64_t)sblock.fs_spc /
2003 		    (int64_t)NSPF(&sblock);
2004 		warn = 0;
2005 	}
2006 	if (warn && !spc_flag) {
2007 		(void) fprintf(stderr, gettext(
2008 		    "Warning: %d sector(s) in last cylinder unallocated\n"),
2009 		    sblock.fs_spc - (uint32_t)(fssize_frag * NSPF(&sblock) -
2010 		    (uint64_t)(sblock.fs_ncyl - 1) * sblock.fs_spc));
2011 	}
2012 	/*
2013 	 * fill in remaining fields of the super block
2014 	 */
2015 
2016 	/*
2017 	 * The csum records are stored in cylinder group 0, starting at
2018 	 * cgdmin, the first data block.
2019 	 */
2020 	sblock.fs_csaddr = cgdmin(&sblock, 0);
2021 	sblock.fs_cssize =
2022 	    fragroundup(&sblock, sblock.fs_ncg * sizeof (struct csum));
2023 	i = sblock.fs_bsize / sizeof (struct csum);
2024 	sblock.fs_csmask = ~(i - 1);
2025 	for (sblock.fs_csshift = 0; i > 1; i >>= 1)
2026 		sblock.fs_csshift++;
2027 	fscs = (struct csum *)calloc(1, sblock.fs_cssize);
2028 
2029 	checksummarysize();
2030 	if (mtb == 'y') {
2031 		sblock.fs_magic = MTB_UFS_MAGIC;
2032 		sblock.fs_version = MTB_UFS_VERSION_1;
2033 	} else {
2034 		sblock.fs_magic = FS_MAGIC;
2035 		if (use_efi_dflts)
2036 			sblock.fs_version = UFS_EFISTYLE4NONEFI_VERSION_2;
2037 		else
2038 			sblock.fs_version = UFS_VERSION_MIN;
2039 	}
2040 
2041 	if (grow) {
2042 		bcopy((caddr_t)grow_fscs, (caddr_t)fscs, (int)grow_fs_cssize);
2043 		extendsummaryinfo();
2044 		goto grow40;
2045 	}
2046 	sblock.fs_rotdelay = rotdelay;
2047 	sblock.fs_maxcontig = maxcontig;
2048 	sblock.fs_maxbpg = MAXBLKPG(sblock.fs_bsize);
2049 
2050 	sblock.fs_rps = rps;
2051 	sblock.fs_cgrotor = 0;
2052 	sblock.fs_cstotal.cs_ndir = 0;
2053 	sblock.fs_cstotal.cs_nbfree = 0;
2054 	sblock.fs_cstotal.cs_nifree = 0;
2055 	sblock.fs_cstotal.cs_nffree = 0;
2056 	sblock.fs_fmod = 0;
2057 	sblock.fs_ronly = 0;
2058 	sblock.fs_time = mkfstime;
2059 	sblock.fs_state = FSOKAY - sblock.fs_time;
2060 	sblock.fs_clean = FSCLEAN;
2061 grow40:
2062 
2063 	/*
2064 	 * If all that's needed is a dump of the superblock we
2065 	 * would use by default, we've got it now.  So, splat it
2066 	 * out and leave.
2067 	 */
2068 	if (rflag) {
2069 		dump_sblock();
2070 		lockexit(0);
2071 	}
2072 	/*
2073 	 * Dump out summary information about file system.
2074 	 */
2075 	(void) fprintf(stderr, gettext(
2076 	    "%s:\t%lld sectors in %d cylinders of %d tracks, %d sectors\n"),
2077 	    fsys, (uint64_t)sblock.fs_size * NSPF(&sblock), sblock.fs_ncyl,
2078 	    sblock.fs_ntrak, sblock.fs_nsect);
2079 	(void) fprintf(stderr, gettext(
2080 	    "\t%.1fMB in %d cyl groups (%d c/g, %.2fMB/g, %d i/g)\n"),
2081 	    (float)sblock.fs_size * sblock.fs_fsize / MB, sblock.fs_ncg,
2082 	    sblock.fs_cpg, (float)sblock.fs_fpg * sblock.fs_fsize / MB,
2083 	    sblock.fs_ipg);
2084 
2085 	tmpbuf = calloc(sblock.fs_ncg / 50 + 500, 1);
2086 	if (tmpbuf == NULL) {
2087 		perror("calloc");
2088 		lockexit(32);
2089 	}
2090 	if (cg_too_small) {
2091 		(void) fprintf(stderr, gettext("File system creation failed. "
2092 		    "There is only one cylinder group and\nthat is "
2093 		    "not even big enough to hold the inodes.\n"));
2094 		lockexit(32);
2095 	}
2096 	/*
2097 	 * Now build the cylinders group blocks and
2098 	 * then print out indices of cylinder groups.
2099 	 */
2100 	tprintf(gettext(
2101 	    "super-block backups (for fsck -F ufs -o b=#) at:\n"));
2102 	for (width = cylno = 0; cylno < sblock.fs_ncg && cylno < 10; cylno++) {
2103 		if ((grow == 0) || (cylno >= grow_fs_ncg))
2104 			initcg(cylno);
2105 		num = fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno));
2106 		/*
2107 		 * If Nflag and if the disk is larger than the CHSLIMIT,
2108 		 * then sanity test the superblocks before reporting. If there
2109 		 * are too many superblocks which look insane, we have
2110 		 * to retry with alternate logic. If both methods have
2111 		 * failed, then our efforts to arrive at alternate
2112 		 * superblocks failed, so complain and exit.
2113 		 */
2114 		if (Nflag && retry) {
2115 			skip_this_sb = 0;
2116 			rdfs((diskaddr_t)num, sbsize, (char *)&altsblock);
2117 			ret = checksblock(altsblock, 1);
2118 			if (ret) {
2119 				skip_this_sb = 1;
2120 				invalid_sb_cnt++;
2121 				dprintf(("DeBuG checksblock() failed - error :"
2122 				    " %d for sb : %llu invalid_sb_cnt : %d\n",
2123 				    ret, num, invalid_sb_cnt));
2124 			} else {
2125 				/*
2126 				 * Though the superblock looks sane, verify if
2127 				 * the fs_version in the superblock and the
2128 				 * logic that we are using to arrive at the
2129 				 * superblocks match.
2130 				 */
2131 				if (use_efi_dflts && altsblock.fs_version
2132 				    != UFS_EFISTYLE4NONEFI_VERSION_2) {
2133 					skip_this_sb = 1;
2134 					invalid_sb_cnt++;
2135 				}
2136 			}
2137 			if (invalid_sb_cnt >= INVALIDSBLIMIT) {
2138 				if (retry > 1) {
2139 					(void) fprintf(stderr, gettext(
2140 					    "Error determining alternate "
2141 					    "superblock locations\n"));
2142 					free(tmpbuf);
2143 					lockexit(32);
2144 				}
2145 				retry++;
2146 				use_efi_dflts = !use_efi_dflts;
2147 				free(tmpbuf);
2148 				goto retry_alternate_logic;
2149 			}
2150 			if (skip_this_sb)
2151 				continue;
2152 		}
2153 		(void) sprintf(pbuf, " %llu,", num);
2154 		plen = strlen(pbuf);
2155 		if ((width + plen) > (WIDTH - 1)) {
2156 			width = plen;
2157 			tprintf("\n");
2158 		} else {
2159 			width += plen;
2160 		}
2161 		if (Nflag && retry)
2162 			(void) strncat(tmpbuf, pbuf, strlen(pbuf));
2163 		else
2164 			(void) fprintf(stderr, "%s", pbuf);
2165 	}
2166 	tprintf("\n");
2167 
2168 	remaining_cg = sblock.fs_ncg - cylno;
2169 
2170 	/*
2171 	 * If there are more than 300 cylinder groups still to be
2172 	 * initialized, print a "." for every 50 cylinder groups.
2173 	 */
2174 	if (remaining_cg > 300) {
2175 		tprintf(gettext("Initializing cylinder groups:\n"));
2176 		do_dot = 1;
2177 	}
2178 
2179 	/*
2180 	 * Now initialize all cylinder groups between the first ten
2181 	 * and the last ten.
2182 	 *
2183 	 * If the number of cylinder groups was less than 10, all of the
2184 	 * cylinder group offsets would have printed in the last loop
2185 	 * and cylno will already be equal to sblock.fs_ncg and so this
2186 	 * loop will not be entered.  If there are less than 20 cylinder
2187 	 * groups, cylno is already less than fs_ncg - 10, so this loop
2188 	 * won't be entered in that case either.
2189 	 */
2190 
2191 	i = 0;
2192 	for (; cylno < sblock.fs_ncg - 10; cylno++) {
2193 		if ((grow == 0) || (cylno >= grow_fs_ncg))
2194 			initcg(cylno);
2195 		if (do_dot && cylno % 50 == 0) {
2196 			tprintf(".");
2197 			i++;
2198 			if (i == WIDTH - 1) {
2199 				tprintf("\n");
2200 				i = 0;
2201 			}
2202 		}
2203 	}
2204 
2205 	/*
2206 	 * Now print the cylinder group offsets for the last 10
2207 	 * cylinder groups, if any are left.
2208 	 */
2209 
2210 	if (do_dot) {
2211 		tprintf(gettext(
2212 	    "\nsuper-block backups for last 10 cylinder groups at:\n"));
2213 	}
2214 	for (width = 0; cylno < sblock.fs_ncg; cylno++) {
2215 		if ((grow == 0) || (cylno >= grow_fs_ncg))
2216 			initcg(cylno);
2217 		num = fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno));
2218 		if (Nflag && retry) {
2219 			skip_this_sb = 0;
2220 			rdfs((diskaddr_t)num, sbsize, (char *)&altsblock);
2221 			ret = checksblock(altsblock, 1);
2222 			if (ret) {
2223 				skip_this_sb = 1;
2224 				invalid_sb_cnt++;
2225 				dprintf(("DeBuG checksblock() failed - error :"
2226 				    " %d for sb : %llu invalid_sb_cnt : %d\n",
2227 				    ret, num, invalid_sb_cnt));
2228 			} else {
2229 				/*
2230 				 * Though the superblock looks sane, verify if
2231 				 * the fs_version in the superblock and the
2232 				 * logic that we are using to arrive at the
2233 				 * superblocks match.
2234 				 */
2235 				if (use_efi_dflts && altsblock.fs_version
2236 				    != UFS_EFISTYLE4NONEFI_VERSION_2) {
2237 					skip_this_sb = 1;
2238 					invalid_sb_cnt++;
2239 				}
2240 			}
2241 			if (invalid_sb_cnt >= INVALIDSBLIMIT) {
2242 				if (retry > 1) {
2243 					(void) fprintf(stderr, gettext(
2244 					    "Error determining alternate "
2245 					    "superblock locations\n"));
2246 					free(tmpbuf);
2247 					lockexit(32);
2248 				}
2249 				retry++;
2250 				use_efi_dflts = !use_efi_dflts;
2251 				free(tmpbuf);
2252 				goto retry_alternate_logic;
2253 			}
2254 			if (skip_this_sb)
2255 				continue;
2256 		}
2257 		/* Don't print ',' for the last superblock */
2258 		if (cylno == sblock.fs_ncg-1)
2259 			(void) sprintf(pbuf, " %llu", num);
2260 		else
2261 			(void) sprintf(pbuf, " %llu,", num);
2262 		plen = strlen(pbuf);
2263 		if ((width + plen) > (WIDTH - 1)) {
2264 			width = plen;
2265 			tprintf("\n");
2266 		} else {
2267 			width += plen;
2268 		}
2269 		if (Nflag && retry)
2270 			(void) strncat(tmpbuf, pbuf, strlen(pbuf));
2271 		else
2272 			(void) fprintf(stderr, "%s", pbuf);
2273 	}
2274 	tprintf("\n");
2275 	if (Nflag) {
2276 		if (retry)
2277 			(void) fprintf(stderr, "%s", tmpbuf);
2278 		free(tmpbuf);
2279 		lockexit(0);
2280 	}
2281 
2282 	free(tmpbuf);
2283 	if (grow)
2284 		goto grow50;
2285 
2286 	/*
2287 	 * Now construct the initial file system,
2288 	 * then write out the super-block.
2289 	 */
2290 	fsinit();
2291 grow50:
2292 	/*
2293 	 * write the superblock and csum information
2294 	 */
2295 	wtsb();
2296 
2297 	/*
2298 	 * extend the last cylinder group in the original file system
2299 	 */
2300 	if (grow) {
2301 		extendcg(grow_fs_ncg-1);
2302 		wtsb();
2303 	}
2304 
2305 	/*
2306 	 * Write out the duplicate super blocks to the first 10
2307 	 * cylinder groups (or fewer, if there are fewer than 10
2308 	 * cylinder groups).
2309 	 */
2310 	for (cylno = 0; cylno < sblock.fs_ncg && cylno < 10; cylno++)
2311 		awtfs(fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno)),
2312 		    (int)sbsize, (char *)&sblock, SAVE);
2313 
2314 	/*
2315 	 * Now write out duplicate super blocks to the remaining
2316 	 * cylinder groups.  In the case of multi-terabyte file
2317 	 * systems, just write out the super block to the last ten
2318 	 * cylinder groups (or however many are left).
2319 	 */
2320 	if (mtb == 'y') {
2321 		if (sblock.fs_ncg <= 10)
2322 			cylno = sblock.fs_ncg;
2323 		else if (sblock.fs_ncg <= 20)
2324 			cylno = 10;
2325 		else
2326 			cylno = sblock.fs_ncg - 10;
2327 	}
2328 
2329 	for (; cylno < sblock.fs_ncg; cylno++)
2330 		awtfs(fsbtodb(&sblock, (uint64_t)cgsblock(&sblock, cylno)),
2331 		    (int)sbsize, (char *)&sblock, SAVE);
2332 
2333 	/*
2334 	 * Flush out all the AIO writes we've done.  It's not
2335 	 * necessary to do this explicitly, but it's the only
2336 	 * way to report any errors from those writes.
2337 	 */
2338 	flush_writes();
2339 
2340 	/*
2341 	 * set clean flag
2342 	 */
2343 	if (grow)
2344 		sblock.fs_clean = grow_fs_clean;
2345 	else
2346 		sblock.fs_clean = FSCLEAN;
2347 	sblock.fs_time = mkfstime;
2348 	sblock.fs_state = FSOKAY - sblock.fs_time;
2349 	wtfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
2350 	isbad = 0;
2351 
2352 	if (fsync(fso) == -1) {
2353 		saverr = errno;
2354 		(void) fprintf(stderr,
2355 		    gettext("mkfs: fsync failed on write disk: %s\n"),
2356 		    strerror(saverr));
2357 		/* we're just cleaning up, so keep going */
2358 	}
2359 	if (close(fsi) == -1) {
2360 		saverr = errno;
2361 		(void) fprintf(stderr,
2362 		    gettext("mkfs: close failed on read disk: %s\n"),
2363 		    strerror(saverr));
2364 		/* we're just cleaning up, so keep going */
2365 	}
2366 	if (close(fso) == -1) {
2367 		saverr = errno;
2368 		(void) fprintf(stderr,
2369 		    gettext("mkfs: close failed on write disk: %s\n"),
2370 		    strerror(saverr));
2371 		/* we're just cleaning up, so keep going */
2372 	}
2373 	fsi = fso = -1;
2374 
2375 #ifndef STANDALONE
2376 	lockexit(0);
2377 #endif
2378 
2379 	return (0);
2380 }
2381 
2382 static diskaddr_t
2383 get_device_size(int fd)
2384 {
2385 	struct dk_minfo	disk_info;
2386 
2387 	if ((ioctl(fd, DKIOCGMEDIAINFO, (caddr_t)&disk_info)) == -1)
2388 		return (0);
2389 
2390 	return (disk_info.dki_capacity);
2391 }
2392 
2393 /*
2394  * Figure out how big the partition we're dealing with is.
2395  * The value returned is in disk blocks (sectors);
2396  */
2397 static diskaddr_t
2398 get_max_size(int fd)
2399 {
2400 	struct extvtoc vtoc;
2401 	dk_gpt_t *efi_vtoc;
2402 	diskaddr_t	slicesize;
2403 
2404 	int index = read_extvtoc(fd, &vtoc);
2405 
2406 	if (index >= 0) {
2407 		label_type = LABEL_TYPE_VTOC;
2408 	} else {
2409 		if (index == VT_ENOTSUP || index == VT_ERROR) {
2410 			/* it might be an EFI label */
2411 			index = efi_alloc_and_read(fd, &efi_vtoc);
2412 			label_type = LABEL_TYPE_EFI;
2413 		}
2414 	}
2415 
2416 	if (index < 0) {
2417 		/*
2418 		 * Since both attempts to read the label failed, we're
2419 		 * going to use DKIOCGMEDIAINFO to get device size.
2420 		 */
2421 
2422 		label_type = LABEL_TYPE_OTHER;
2423 		slicesize = get_device_size(fd);
2424 		if (slicesize == 0) {
2425 			switch (index) {
2426 			case VT_ERROR:
2427 				break;
2428 			case VT_EIO:
2429 				errno = EIO;
2430 				break;
2431 			case VT_EINVAL:
2432 				errno = EINVAL;
2433 			}
2434 			perror(gettext("Can not determine partition size"));
2435 			lockexit(32);
2436 		}
2437 	}
2438 
2439 	if (label_type == LABEL_TYPE_EFI) {
2440 		slicesize = efi_vtoc->efi_parts[index].p_size;
2441 		efi_free(efi_vtoc);
2442 	} else if (label_type == LABEL_TYPE_VTOC) {
2443 		/*
2444 		 * In the vtoc struct, p_size is a 32-bit signed quantity.
2445 		 * In the dk_gpt struct (efi's version of the vtoc), p_size
2446 		 * is an unsigned 64-bit quantity.  By casting the vtoc's
2447 		 * psize to an unsigned 32-bit quantity, it will be copied
2448 		 * to 'slicesize' (an unsigned 64-bit diskaddr_t) without
2449 		 * sign extension.
2450 		 */
2451 
2452 		slicesize = (uint32_t)vtoc.v_part[index].p_size;
2453 	}
2454 
2455 	dprintf(("DeBuG get_max_size index = %d, p_size = %lld, dolimit = %d\n",
2456 	    index, slicesize, (slicesize > FS_MAX)));
2457 
2458 	/*
2459 	 * The next line limits a UFS file system to the maximum
2460 	 * supported size.
2461 	 */
2462 
2463 	if (slicesize > FS_MAX)
2464 		return (FS_MAX);
2465 	return (slicesize);
2466 }
2467 
2468 static long
2469 get_max_track_size(int fd)
2470 {
2471 	struct dk_cinfo ci;
2472 	long track_size = -1;
2473 
2474 	if (ioctl(fd, DKIOCINFO, &ci) == 0) {
2475 		track_size = ci.dki_maxtransfer * DEV_BSIZE;
2476 	}
2477 
2478 	if ((track_size < 0)) {
2479 		int	error = 0;
2480 		int	maxphys;
2481 		int	gotit = 0;
2482 
2483 		gotit = fsgetmaxphys(&maxphys, &error);
2484 		if (gotit) {
2485 			track_size = MIN(MB, maxphys);
2486 		} else {
2487 			(void) fprintf(stderr, gettext(
2488 "Warning: Could not get system value for maxphys. The value for\n"
2489 "maxcontig will default to 1MB.\n"));
2490 			track_size = MB;
2491 		}
2492 	}
2493 	return (track_size);
2494 }
2495 
2496 /*
2497  * Initialize a cylinder group.
2498  */
2499 static void
2500 initcg(int cylno)
2501 {
2502 	diskaddr_t cbase, d;
2503 	diskaddr_t dlower;	/* last data block before cg metadata */
2504 	diskaddr_t dupper;	/* first data block after cg metadata */
2505 	diskaddr_t dmax;
2506 	int64_t i;
2507 	struct csum *cs;
2508 	struct dinode *inode_buffer;
2509 	int size;
2510 
2511 	/*
2512 	 * Variables used to store intermediate results as a part of
2513 	 * the internal implementation of the cbtocylno() macros.
2514 	 */
2515 	diskaddr_t bno;		/* UFS block number (not sector number) */
2516 	int	cbcylno;	/* current cylinder number */
2517 	int	cbcylno_sect;	/* sector offset within cylinder */
2518 	int	cbsect_incr;	/* amount to increment sector offset */
2519 
2520 	/*
2521 	 * Variables used to store intermediate results as a part of
2522 	 * the internal implementation of the cbtorpos() macros.
2523 	 */
2524 	short	*cgblks;	/* pointer to array of free blocks in cg */
2525 	int	trackrpos;	/* tmp variable for rotation position */
2526 	int	trackoff;	/* offset within a track */
2527 	int	trackoff_incr;	/* amount to increment trackoff */
2528 	int	rpos;		/* rotation position of current block */
2529 	int	rpos_incr;	/* amount to increment rpos per block */
2530 
2531 	union cgun *icgun;	/* local pointer to a cg summary block */
2532 #define	icg	(icgun->cg)
2533 
2534 	icgun = (union cgun *)getbuf(&cgsumbuf, sizeof (union cgun));
2535 
2536 	/*
2537 	 * Determine block bounds for cylinder group.
2538 	 * Allow space for super block summary information in first
2539 	 * cylinder group.
2540 	 */
2541 	cbase = cgbase(&sblock, cylno);
2542 	dmax = cbase + sblock.fs_fpg;
2543 	if (dmax > sblock.fs_size)	/* last cg may be smaller than normal */
2544 		dmax = sblock.fs_size;
2545 	dlower = cgsblock(&sblock, cylno) - cbase;
2546 	dupper = cgdmin(&sblock, cylno) - cbase;
2547 	if (cylno == 0)
2548 		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
2549 	cs = fscs + cylno;
2550 	icg.cg_time = mkfstime;
2551 	icg.cg_magic = CG_MAGIC;
2552 	icg.cg_cgx = cylno;
2553 	/* last one gets whatever's left */
2554 	if (cylno == sblock.fs_ncg - 1)
2555 		icg.cg_ncyl = sblock.fs_ncyl - (sblock.fs_cpg * cylno);
2556 	else
2557 		icg.cg_ncyl = sblock.fs_cpg;
2558 	icg.cg_niblk = sblock.fs_ipg;
2559 	icg.cg_ndblk = dmax - cbase;
2560 	icg.cg_cs.cs_ndir = 0;
2561 	icg.cg_cs.cs_nffree = 0;
2562 	icg.cg_cs.cs_nbfree = 0;
2563 	icg.cg_cs.cs_nifree = 0;
2564 	icg.cg_rotor = 0;
2565 	icg.cg_frotor = 0;
2566 	icg.cg_irotor = 0;
2567 	icg.cg_btotoff = &icg.cg_space[0] - (uchar_t *)(&icg.cg_link);
2568 	icg.cg_boff = icg.cg_btotoff + sblock.fs_cpg * sizeof (long);
2569 	icg.cg_iusedoff = icg.cg_boff +
2570 	    sblock.fs_cpg * sblock.fs_nrpos * sizeof (short);
2571 	icg.cg_freeoff = icg.cg_iusedoff + howmany(sblock.fs_ipg, NBBY);
2572 	icg.cg_nextfreeoff = icg.cg_freeoff +
2573 	    howmany(sblock.fs_cpg * sblock.fs_spc / NSPF(&sblock), NBBY);
2574 	for (i = 0; i < sblock.fs_frag; i++) {
2575 		icg.cg_frsum[i] = 0;
2576 	}
2577 	bzero((caddr_t)cg_inosused(&icg), icg.cg_freeoff - icg.cg_iusedoff);
2578 	icg.cg_cs.cs_nifree += sblock.fs_ipg;
2579 	if (cylno == 0)
2580 		for (i = 0; i < UFSROOTINO; i++) {
2581 			setbit(cg_inosused(&icg), i);
2582 			icg.cg_cs.cs_nifree--;
2583 		}
2584 
2585 	/*
2586 	 * Initialize all the inodes in the cylinder group using
2587 	 * random numbers.
2588 	 */
2589 	size = sblock.fs_ipg * sizeof (struct dinode);
2590 	inode_buffer = (struct dinode *)getbuf(&inodebuf, size);
2591 
2592 	for (i = 0; i < sblock.fs_ipg; i++) {
2593 		IRANDOMIZE(&(inode_buffer[i].di_ic));
2594 	}
2595 
2596 	/*
2597 	 * Write all inodes in a single write for performance.
2598 	 */
2599 	awtfs(fsbtodb(&sblock, (uint64_t)cgimin(&sblock, cylno)), (int)size,
2600 	    (char *)inode_buffer, RELEASE);
2601 
2602 	bzero((caddr_t)cg_blktot(&icg), icg.cg_boff - icg.cg_btotoff);
2603 	bzero((caddr_t)cg_blks(&sblock, &icg, 0),
2604 	    icg.cg_iusedoff - icg.cg_boff);
2605 	bzero((caddr_t)cg_blksfree(&icg), icg.cg_nextfreeoff - icg.cg_freeoff);
2606 
2607 	if (cylno > 0) {
2608 		for (d = 0; d < dlower; d += sblock.fs_frag) {
2609 			setblock(&sblock, cg_blksfree(&icg), d/sblock.fs_frag);
2610 			icg.cg_cs.cs_nbfree++;
2611 			cg_blktot(&icg)[cbtocylno(&sblock, d)]++;
2612 			cg_blks(&sblock, &icg, cbtocylno(&sblock, d))
2613 			    [cbtorpos(&sblock, d)]++;
2614 		}
2615 		sblock.fs_dsize += dlower;
2616 	}
2617 	sblock.fs_dsize += icg.cg_ndblk - dupper;
2618 	if ((i = dupper % sblock.fs_frag) != 0) {
2619 		icg.cg_frsum[sblock.fs_frag - i]++;
2620 		for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
2621 			setbit(cg_blksfree(&icg), dupper);
2622 			icg.cg_cs.cs_nffree++;
2623 		}
2624 	}
2625 
2626 	/*
2627 	 * WARNING: The following code is somewhat confusing, but
2628 	 * results in a substantial performance improvement in mkfs.
2629 	 *
2630 	 * Instead of using cbtocylno() and cbtorpos() macros, we
2631 	 * keep track of all the intermediate state of those macros
2632 	 * in some variables.  This allows simple addition to be
2633 	 * done to calculate the results as we step through the
2634 	 * blocks in an orderly fashion instead of the slower
2635 	 * multiplication and division the macros are forced to
2636 	 * used so they can support random input.  (Multiplication,
2637 	 * division, and remainder operations typically take about
2638 	 * 10x as many processor cycles as other operations.)
2639 	 *
2640 	 * The basic idea is to take code:
2641 	 *
2642 	 *	for (x = starting_x; x < max; x++)
2643 	 *		y = (x * c) / z
2644 	 *
2645 	 * and rewrite it to take advantage of the fact that
2646 	 * the variable x is incrementing in an orderly way:
2647 	 *
2648 	 *	intermediate = starting_x * c
2649 	 *	yval = intermediate / z
2650 	 *	for (x = starting_x; x < max; x++) {
2651 	 *		y = yval;
2652 	 *		intermediate += c
2653 	 *		if (intermediate > z) {
2654 	 *			yval++;
2655 	 *			intermediate -= z
2656 	 *		}
2657 	 *	}
2658 	 *
2659 	 * Performance has improved as much as 4X using this code.
2660 	 */
2661 
2662 	/*
2663 	 * Initialize the starting points for all the cbtocylno()
2664 	 * macro variables and figure out the increments needed each
2665 	 * time through the loop.
2666 	 */
2667 	cbcylno_sect = dupper * NSPF(&sblock);
2668 	cbsect_incr = sblock.fs_frag * NSPF(&sblock);
2669 	cbcylno = cbcylno_sect / sblock.fs_spc;
2670 	cbcylno_sect %= sblock.fs_spc;
2671 	cgblks = cg_blks(&sblock, &icg, cbcylno);
2672 	bno = dupper / sblock.fs_frag;
2673 
2674 	/*
2675 	 * Initialize the starting points for all the cbtorpos()
2676 	 * macro variables and figure out the increments needed each
2677 	 * time through the loop.
2678 	 *
2679 	 * It's harder to simplify the cbtorpos() macro if there were
2680 	 * alternate sectors specified (or if they previously existed
2681 	 * in the growfs case).  Since this is rare, we just revert to
2682 	 * using the macros in this case and skip the variable setup.
2683 	 */
2684 	if (!spc_flag) {
2685 		trackrpos = (cbcylno_sect % sblock.fs_nsect) * sblock.fs_nrpos;
2686 		rpos = trackrpos / sblock.fs_nsect;
2687 		trackoff = trackrpos % sblock.fs_nsect;
2688 		trackoff_incr = cbsect_incr * sblock.fs_nrpos;
2689 		rpos_incr = (trackoff_incr / sblock.fs_nsect) % sblock.fs_nrpos;
2690 		trackoff_incr = trackoff_incr % sblock.fs_nsect;
2691 	}
2692 
2693 	/*
2694 	 * Loop through all the blocks, marking them free and
2695 	 * updating totals kept in the superblock and cg summary.
2696 	 */
2697 	for (d = dupper; d + sblock.fs_frag <= dmax - cbase; ) {
2698 		setblock(&sblock, cg_blksfree(&icg),  bno);
2699 		icg.cg_cs.cs_nbfree++;
2700 
2701 		cg_blktot(&icg)[cbcylno]++;
2702 
2703 		if (!spc_flag)
2704 			cgblks[rpos]++;
2705 		else
2706 			cg_blks(&sblock, &icg, cbtocylno(&sblock, d))
2707 			    [cbtorpos(&sblock, d)]++;
2708 
2709 		d += sblock.fs_frag;
2710 		bno++;
2711 
2712 		/*
2713 		 * Increment the sector offset within the cylinder
2714 		 * for the cbtocylno() macro reimplementation.  If
2715 		 * we're beyond the end of the cylinder, update the
2716 		 * cylinder number, calculate the offset in the
2717 		 * new cylinder, and update the cgblks pointer
2718 		 * to the next rotational position.
2719 		 */
2720 		cbcylno_sect += cbsect_incr;
2721 		if (cbcylno_sect >= sblock.fs_spc) {
2722 			cbcylno++;
2723 			cbcylno_sect -= sblock.fs_spc;
2724 			cgblks += sblock.fs_nrpos;
2725 		}
2726 
2727 		/*
2728 		 * If there aren't alternate sectors, increment the
2729 		 * rotational position variables for the cbtorpos()
2730 		 * reimplementation.  Note that we potentially
2731 		 * increment rpos twice.  Once by rpos_incr, and one
2732 		 * more time when we wrap to a new track because
2733 		 * trackoff >= fs_nsect.
2734 		 */
2735 		if (!spc_flag) {
2736 			trackoff += trackoff_incr;
2737 			rpos += rpos_incr;
2738 			if (trackoff >= sblock.fs_nsect) {
2739 				trackoff -= sblock.fs_nsect;
2740 				rpos++;
2741 			}
2742 			if (rpos >= sblock.fs_nrpos)
2743 				rpos -= sblock.fs_nrpos;
2744 		}
2745 	}
2746 
2747 	if (d < dmax - cbase) {
2748 		icg.cg_frsum[dmax - cbase - d]++;
2749 		for (; d < dmax - cbase; d++) {
2750 			setbit(cg_blksfree(&icg), d);
2751 			icg.cg_cs.cs_nffree++;
2752 		}
2753 	}
2754 	sblock.fs_cstotal.cs_ndir += icg.cg_cs.cs_ndir;
2755 	sblock.fs_cstotal.cs_nffree += icg.cg_cs.cs_nffree;
2756 	sblock.fs_cstotal.cs_nbfree += icg.cg_cs.cs_nbfree;
2757 	sblock.fs_cstotal.cs_nifree += icg.cg_cs.cs_nifree;
2758 	*cs = icg.cg_cs;
2759 	awtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, cylno)),
2760 	    sblock.fs_bsize, (char *)&icg, RELEASE);
2761 }
2762 
2763 /*
2764  * initialize the file system
2765  */
2766 struct inode node;
2767 
2768 #define	LOSTDIR
2769 #ifdef LOSTDIR
2770 #define	PREDEFDIR 3
2771 #else
2772 #define	PREDEFDIR 2
2773 #endif
2774 
2775 struct direct root_dir[] = {
2776 	{ UFSROOTINO, sizeof (struct direct), 1, "." },
2777 	{ UFSROOTINO, sizeof (struct direct), 2, ".." },
2778 #ifdef LOSTDIR
2779 	{ LOSTFOUNDINO, sizeof (struct direct), 10, "lost+found" },
2780 #endif
2781 };
2782 #ifdef LOSTDIR
2783 struct direct lost_found_dir[] = {
2784 	{ LOSTFOUNDINO, sizeof (struct direct), 1, "." },
2785 	{ UFSROOTINO, sizeof (struct direct), 2, ".." },
2786 	{ 0, DIRBLKSIZ, 0, 0 },
2787 };
2788 #endif
2789 char buf[MAXBSIZE];
2790 
2791 static void
2792 fsinit()
2793 {
2794 	int i;
2795 
2796 
2797 	/*
2798 	 * initialize the node
2799 	 */
2800 	node.i_atime = mkfstime;
2801 	node.i_mtime = mkfstime;
2802 	node.i_ctime = mkfstime;
2803 #ifdef LOSTDIR
2804 	/*
2805 	 * create the lost+found directory
2806 	 */
2807 	(void) makedir(lost_found_dir, 2);
2808 	for (i = DIRBLKSIZ; i < sblock.fs_bsize; i += DIRBLKSIZ) {
2809 		bcopy(&lost_found_dir[2], &buf[i], DIRSIZ(&lost_found_dir[2]));
2810 	}
2811 	node.i_number = LOSTFOUNDINO;
2812 	node.i_smode = IFDIR | 0700;
2813 	node.i_nlink = 2;
2814 	node.i_size = sblock.fs_bsize;
2815 	node.i_db[0] = alloc((int)node.i_size, node.i_mode);
2816 	node.i_blocks = btodb(fragroundup(&sblock, (int)node.i_size));
2817 	IRANDOMIZE(&node.i_ic);
2818 	wtfs(fsbtodb(&sblock, (uint64_t)node.i_db[0]), (int)node.i_size, buf);
2819 	iput(&node);
2820 #endif
2821 	/*
2822 	 * create the root directory
2823 	 */
2824 	node.i_number = UFSROOTINO;
2825 	node.i_mode = IFDIR | UMASK;
2826 	node.i_nlink = PREDEFDIR;
2827 	node.i_size = makedir(root_dir, PREDEFDIR);
2828 	node.i_db[0] = alloc(sblock.fs_fsize, node.i_mode);
2829 	/* i_size < 2GB because we are initializing the file system */
2830 	node.i_blocks = btodb(fragroundup(&sblock, (int)node.i_size));
2831 	IRANDOMIZE(&node.i_ic);
2832 	wtfs(fsbtodb(&sblock, (uint64_t)node.i_db[0]), sblock.fs_fsize, buf);
2833 	iput(&node);
2834 }
2835 
2836 /*
2837  * construct a set of directory entries in "buf".
2838  * return size of directory.
2839  */
2840 static int
2841 makedir(struct direct *protodir, int entries)
2842 {
2843 	char *cp;
2844 	int i;
2845 	ushort_t spcleft;
2846 
2847 	spcleft = DIRBLKSIZ;
2848 	for (cp = buf, i = 0; i < entries - 1; i++) {
2849 		protodir[i].d_reclen = DIRSIZ(&protodir[i]);
2850 		bcopy(&protodir[i], cp, protodir[i].d_reclen);
2851 		cp += protodir[i].d_reclen;
2852 		spcleft -= protodir[i].d_reclen;
2853 	}
2854 	protodir[i].d_reclen = spcleft;
2855 	bcopy(&protodir[i], cp, DIRSIZ(&protodir[i]));
2856 	return (DIRBLKSIZ);
2857 }
2858 
2859 /*
2860  * allocate a block or frag
2861  */
2862 static daddr32_t
2863 alloc(int size, int mode)
2864 {
2865 	int i, frag;
2866 	daddr32_t d;
2867 
2868 	rdfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2869 	    (char *)&acg);
2870 	if (acg.cg_magic != CG_MAGIC) {
2871 		(void) fprintf(stderr, gettext("cg 0: bad magic number\n"));
2872 		lockexit(32);
2873 	}
2874 	if (acg.cg_cs.cs_nbfree == 0) {
2875 		(void) fprintf(stderr,
2876 		    gettext("first cylinder group ran out of space\n"));
2877 		lockexit(32);
2878 	}
2879 	for (d = 0; d < acg.cg_ndblk; d += sblock.fs_frag)
2880 		if (isblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag))
2881 			goto goth;
2882 	(void) fprintf(stderr,
2883 	    gettext("internal error: can't find block in cyl 0\n"));
2884 	lockexit(32);
2885 goth:
2886 	clrblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag);
2887 	acg.cg_cs.cs_nbfree--;
2888 	sblock.fs_cstotal.cs_nbfree--;
2889 	fscs[0].cs_nbfree--;
2890 	if (mode & IFDIR) {
2891 		acg.cg_cs.cs_ndir++;
2892 		sblock.fs_cstotal.cs_ndir++;
2893 		fscs[0].cs_ndir++;
2894 	}
2895 	cg_blktot(&acg)[cbtocylno(&sblock, d)]--;
2896 	cg_blks(&sblock, &acg, cbtocylno(&sblock, d))[cbtorpos(&sblock, d)]--;
2897 	if (size != sblock.fs_bsize) {
2898 		frag = howmany(size, sblock.fs_fsize);
2899 		fscs[0].cs_nffree += sblock.fs_frag - frag;
2900 		sblock.fs_cstotal.cs_nffree += sblock.fs_frag - frag;
2901 		acg.cg_cs.cs_nffree += sblock.fs_frag - frag;
2902 		acg.cg_frsum[sblock.fs_frag - frag]++;
2903 		for (i = frag; i < sblock.fs_frag; i++)
2904 			setbit(cg_blksfree(&acg), d + i);
2905 	}
2906 	wtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2907 	    (char *)&acg);
2908 	return (d);
2909 }
2910 
2911 /*
2912  * Allocate an inode on the disk
2913  */
2914 static void
2915 iput(struct inode *ip)
2916 {
2917 	struct dinode buf[MAXINOPB];
2918 	diskaddr_t d;
2919 
2920 	rdfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2921 	    (char *)&acg);
2922 	if (acg.cg_magic != CG_MAGIC) {
2923 		(void) fprintf(stderr, gettext("cg 0: bad magic number\n"));
2924 		lockexit(32);
2925 	}
2926 	acg.cg_cs.cs_nifree--;
2927 	setbit(cg_inosused(&acg), ip->i_number);
2928 	wtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, 0)), sblock.fs_cgsize,
2929 	    (char *)&acg);
2930 	sblock.fs_cstotal.cs_nifree--;
2931 	fscs[0].cs_nifree--;
2932 	if ((int)ip->i_number >= sblock.fs_ipg * sblock.fs_ncg) {
2933 		(void) fprintf(stderr,
2934 		    gettext("fsinit: inode value out of range (%d).\n"),
2935 		    ip->i_number);
2936 		lockexit(32);
2937 	}
2938 	d = fsbtodb(&sblock, (uint64_t)itod(&sblock, (int)ip->i_number));
2939 	rdfs(d, sblock.fs_bsize, (char *)buf);
2940 	buf[itoo(&sblock, (int)ip->i_number)].di_ic = ip->i_ic;
2941 	wtfs(d, sblock.fs_bsize, (char *)buf);
2942 }
2943 
2944 /*
2945  * getbuf()	-- Get a buffer for use in an AIO operation.  Buffer
2946  *		is zero'd the first time returned, left with whatever
2947  *		was in memory after that.  This function actually gets
2948  *		enough memory the first time it's called to support
2949  *		MAXBUF buffers like a slab allocator.  When all the
2950  *		buffers are in use, it waits for an aio to complete
2951  *		and make a buffer available.
2952  *
2953  *		Never returns an error.  Either succeeds or exits.
2954  */
2955 static char *
2956 getbuf(bufhdr *bufhead, int size)
2957 {
2958 	bufhdr *pbuf;
2959 	bufhdr *prev;
2960 	int i;
2961 	int buf_size, max_bufs;
2962 
2963 	/*
2964 	 * Initialize all the buffers
2965 	 */
2966 	if (bufhead->head == NULL) {
2967 		/*
2968 		 * round up the size of our buffer header to a
2969 		 * 16 byte boundary so the address we return to
2970 		 * the caller is "suitably aligned".
2971 		 */
2972 		bufhdrsize = (sizeof (bufhdr) + 15) & ~15;
2973 
2974 		/*
2975 		 * Add in our header to the buffer and round it all up to
2976 		 * a 16 byte boundry so each member of the slab is aligned.
2977 		 */
2978 		buf_size = (size + bufhdrsize + 15) & ~15;
2979 
2980 		/*
2981 		 * Limit number of buffers to lesser of MAXBUFMEM's worth
2982 		 * or MAXBUF, whichever is less.
2983 		 */
2984 		max_bufs = MAXBUFMEM / buf_size;
2985 		if (max_bufs > MAXBUF)
2986 			max_bufs = MAXBUF;
2987 
2988 		pbuf = (bufhdr *)calloc(max_bufs, buf_size);
2989 		if (pbuf == NULL) {
2990 			perror("calloc");
2991 			lockexit(32);
2992 		}
2993 
2994 		bufhead->head = bufhead;
2995 		prev = bufhead;
2996 		for (i = 0; i < max_bufs; i++) {
2997 			pbuf->head = bufhead;
2998 			prev->next = pbuf;
2999 			prev = pbuf;
3000 			pbuf = (bufhdr *)((char *)pbuf + buf_size);
3001 		}
3002 	}
3003 
3004 	/*
3005 	 * Get an available buffer, waiting for I/O if necessary
3006 	 */
3007 	wait_for_write(NOBLOCK);
3008 	while (bufhead->next == NULL)
3009 		wait_for_write(BLOCK);
3010 
3011 	/*
3012 	 * Take the buffer off the list
3013 	 */
3014 	pbuf = bufhead->next;
3015 	bufhead->next = pbuf->next;
3016 	pbuf->next = NULL;
3017 
3018 	/*
3019 	 * return the empty buffer space just past the header
3020 	 */
3021 	return ((char *)pbuf + bufhdrsize);
3022 }
3023 
3024 /*
3025  * freebuf()	-- Free a buffer gotten previously through getbuf.
3026  *		Puts the buffer back on the appropriate list for
3027  *		later use.  Never calls free().
3028  *
3029  * Assumes that SIGINT is blocked.
3030  */
3031 static void
3032 freebuf(char *buf)
3033 {
3034 	bufhdr *pbuf;
3035 	bufhdr *bufhead;
3036 
3037 	/*
3038 	 * get the header for this buffer
3039 	 */
3040 	pbuf = (bufhdr *)(buf - bufhdrsize);
3041 
3042 	/*
3043 	 * Put it back on the list of available buffers
3044 	 */
3045 	bufhead = pbuf->head;
3046 	pbuf->next = bufhead->next;
3047 	bufhead->next = pbuf;
3048 }
3049 
3050 /*
3051  * freetrans()	-- Free a transaction gotten previously through getaiop.
3052  *		Puts the transaction struct back on the appropriate list for
3053  *		later use.  Never calls free().
3054  *
3055  * Assumes that SIGINT is blocked.
3056  */
3057 static void
3058 freetrans(aio_trans *transp)
3059 {
3060 	/*
3061 	 * free the buffer associated with this AIO if needed
3062 	 */
3063 	if (transp->release == RELEASE)
3064 		freebuf(transp->buffer);
3065 
3066 	/*
3067 	 * Put transaction on the free list
3068 	 */
3069 	transp->next = results.trans;
3070 	results.trans = transp;
3071 }
3072 
3073 /*
3074  * wait_for_write()	-- Wait for an aio write to complete.  Return
3075  *			the transaction structure for that write.
3076  *
3077  * Blocks SIGINT if necessary.
3078  */
3079 aio_trans *
3080 wait_for_write(int block)
3081 {
3082 	aio_trans	*transp;
3083 	aio_result_t	*resultp;
3084 	static struct timeval  zero_wait = { 0, 0 };
3085 	sigset_t	old_mask;
3086 
3087 	/*
3088 	 * If we know there aren't any outstanding transactions, just return
3089 	 */
3090 	if (results.outstanding == 0)
3091 		return ((aio_trans *) 0);
3092 
3093 	block_sigint(&old_mask);
3094 
3095 	resultp = aiowait(block ? NULL : &zero_wait);
3096 	if (resultp == NULL ||
3097 	    (resultp == (aio_result_t *)-1 && errno == EINVAL)) {
3098 		unblock_sigint(&old_mask);
3099 		return ((aio_trans *) 0);
3100 	}
3101 
3102 	results.outstanding--;
3103 	transp = (aio_trans *)resultp;
3104 
3105 	if (resultp->aio_return != transp->size) {
3106 		if (resultp->aio_return == -1) {
3107 			/*
3108 			 * The aiowrite() may have failed because the
3109 			 * kernel didn't have enough memory to do the job.
3110 			 * Flush all pending writes and try a normal
3111 			 * write().  wtfs_breakup() will call exit if it
3112 			 * fails, so we don't worry about errors here.
3113 			 */
3114 			flush_writes();
3115 			wtfs_breakup(transp->bno, transp->size, transp->buffer);
3116 		} else {
3117 			(void) fprintf(stderr, gettext(
3118 			    "short write (%d of %d bytes) on sector %lld\n"),
3119 			    resultp->aio_return, transp->size,
3120 			    transp->bno);
3121 			/*
3122 			 * Don't unblock SIGINT, to avoid potential
3123 			 * looping due to queued interrupts and
3124 			 * error handling.
3125 			 */
3126 			lockexit(32);
3127 		}
3128 	}
3129 
3130 	resultp->aio_return = 0;
3131 	freetrans(transp);
3132 	unblock_sigint(&old_mask);
3133 	return (transp);
3134 }
3135 
3136 /*
3137  * flush_writes()	-- flush all the outstanding aio writes.
3138  */
3139 static void
3140 flush_writes(void)
3141 {
3142 	while (wait_for_write(BLOCK))
3143 		;
3144 }
3145 
3146 /*
3147  * get_aiop()	-- find and return an aio_trans structure on which a new
3148  *		aio can be done.  Blocks on aiowait() if needed.  Reaps
3149  *		all outstanding completed aio's.
3150  *
3151  * Assumes that SIGINT is blocked.
3152  */
3153 aio_trans *
3154 get_aiop()
3155 {
3156 	int i;
3157 	aio_trans *transp;
3158 	aio_trans *prev;
3159 
3160 	/*
3161 	 * initialize aio stuff
3162 	 */
3163 	if (!aio_inited) {
3164 		aio_inited = 1;
3165 
3166 		results.maxpend = 0;
3167 		results.outstanding = 0;
3168 		results.max = MAXAIO;
3169 
3170 		results.trans = (aio_trans *)calloc(results.max,
3171 		    sizeof (aio_trans));
3172 		if (results.trans == NULL) {
3173 			perror("calloc");
3174 			lockexit(32);
3175 		}
3176 
3177 		/*
3178 		 * Initialize the linked list of aio transaction
3179 		 * structures.  Note that the final "next" pointer
3180 		 * will be NULL since we got the buffer from calloc().
3181 		 */
3182 		prev = results.trans;
3183 		for (i = 1; i < results.max; i++) {
3184 			prev->next = &(results.trans[i]);
3185 			prev = prev->next;
3186 		}
3187 	}
3188 
3189 	wait_for_write(NOBLOCK);
3190 	while (results.trans == NULL)
3191 		wait_for_write(BLOCK);
3192 	transp = results.trans;
3193 	results.trans = results.trans->next;
3194 
3195 	transp->next = 0;
3196 	transp->resultbuf.aio_return = AIO_INPROGRESS;
3197 	return (transp);
3198 }
3199 
3200 /*
3201  * read a block from the file system
3202  */
3203 static void
3204 rdfs(diskaddr_t bno, int size, char *bf)
3205 {
3206 	int n, saverr;
3207 
3208 	/*
3209 	 * In case we need any data that's pending in an aiowrite(),
3210 	 * we wait for them all to complete before doing a read.
3211 	 */
3212 	flush_writes();
3213 
3214 	/*
3215 	 * Note: the llseek() can succeed, even if the offset is out of range.
3216 	 * It's not until the file i/o operation (the read()) that one knows
3217 	 * for sure if the raw device can handle the offset.
3218 	 */
3219 	if (llseek(fsi, (offset_t)bno * sectorsize, 0) < 0) {
3220 		saverr = errno;
3221 		(void) fprintf(stderr,
3222 		    gettext("seek error on sector %lld: %s\n"),
3223 		    bno, strerror(saverr));
3224 		lockexit(32);
3225 	}
3226 	n = read(fsi, bf, size);
3227 	if (n != size) {
3228 		saverr = errno;
3229 		if (n == -1)
3230 			(void) fprintf(stderr,
3231 			    gettext("read error on sector %lld: %s\n"),
3232 			    bno, strerror(saverr));
3233 		else
3234 			(void) fprintf(stderr, gettext(
3235 			    "short read (%d of %d bytes) on sector %lld\n"),
3236 			    n, size, bno);
3237 		lockexit(32);
3238 	}
3239 }
3240 
3241 /*
3242  * write a block to the file system
3243  */
3244 static void
3245 wtfs(diskaddr_t bno, int size, char *bf)
3246 {
3247 	int n, saverr;
3248 
3249 	if (fso == -1)
3250 		return;
3251 
3252 	/*
3253 	 * Note: the llseek() can succeed, even if the offset is out of range.
3254 	 * It's not until the file i/o operation (the write()) that one knows
3255 	 * for sure if the raw device can handle the offset.
3256 	 */
3257 	if (llseek(fso, (offset_t)bno * sectorsize, 0) < 0) {
3258 		saverr = errno;
3259 		(void) fprintf(stderr,
3260 		    gettext("seek error on sector %lld: %s\n"),
3261 		    bno, strerror(saverr));
3262 		lockexit(32);
3263 	}
3264 	if (Nflag)
3265 		return;
3266 	n = write(fso, bf, size);
3267 	if (n != size) {
3268 		saverr = errno;
3269 		if (n == -1)
3270 			(void) fprintf(stderr,
3271 			    gettext("write error on sector %lld: %s\n"),
3272 			    bno, strerror(saverr));
3273 		else
3274 			(void) fprintf(stderr, gettext(
3275 			    "short write (%d of %d bytes) on sector %lld\n"),
3276 			    n, size, bno);
3277 		lockexit(32);
3278 	}
3279 }
3280 
3281 /*
3282  * write a block to the file system -- buffered with aio
3283  */
3284 static void
3285 awtfs(diskaddr_t bno, int size, char *bf, int release)
3286 {
3287 	int n;
3288 	aio_trans	*transp;
3289 	sigset_t	old_mask;
3290 
3291 	if (fso == -1)
3292 		return;
3293 
3294 	/*
3295 	 * We need to keep things consistent if we get interrupted,
3296 	 * so defer any expected interrupts for the time being.
3297 	 */
3298 	block_sigint(&old_mask);
3299 
3300 	if (Nflag) {
3301 		if (release == RELEASE)
3302 			freebuf(bf);
3303 	} else {
3304 		transp = get_aiop();
3305 		transp->bno = bno;
3306 		transp->buffer = bf;
3307 		transp->size = size;
3308 		transp->release = release;
3309 
3310 		n = aiowrite(fso, bf, size, (off_t)bno * sectorsize,
3311 		    SEEK_SET, &transp->resultbuf);
3312 
3313 		if (n < 0) {
3314 			/*
3315 			 * The aiowrite() may have failed because the
3316 			 * kernel didn't have enough memory to do the job.
3317 			 * Flush all pending writes and try a normal
3318 			 * write().  wtfs_breakup() will call exit if it
3319 			 * fails, so we don't worry about errors here.
3320 			 */
3321 			flush_writes();
3322 			wtfs_breakup(transp->bno, transp->size, transp->buffer);
3323 			freetrans(transp);
3324 		} else {
3325 			/*
3326 			 * Keep track of our pending writes.
3327 			 */
3328 			results.outstanding++;
3329 			if (results.outstanding > results.maxpend)
3330 				results.maxpend = results.outstanding;
3331 		}
3332 	}
3333 
3334 	unblock_sigint(&old_mask);
3335 }
3336 
3337 
3338 /*
3339  * write a block to the file system, but break it up into sbsize
3340  * chunks to avoid forcing a large amount of memory to be locked down.
3341  * Only used as a fallback when an aio write has failed.
3342  */
3343 static void
3344 wtfs_breakup(diskaddr_t bno, int size, char *bf)
3345 {
3346 	int n, saverr;
3347 	int wsize;
3348 	int block_incr = sbsize / sectorsize;
3349 
3350 	if (size < sbsize)
3351 		wsize = size;
3352 	else
3353 		wsize = sbsize;
3354 
3355 	n = 0;
3356 	while (size) {
3357 		/*
3358 		 * Note: the llseek() can succeed, even if the offset is
3359 		 * out of range.  It's not until the file i/o operation
3360 		 * (the write()) that one knows for sure if the raw device
3361 		 * can handle the offset.
3362 		 */
3363 		if (llseek(fso, (offset_t)bno * sectorsize, 0) < 0) {
3364 			saverr = errno;
3365 			(void) fprintf(stderr,
3366 			    gettext("seek error on sector %lld: %s\n"),
3367 			    bno, strerror(saverr));
3368 			lockexit(32);
3369 		}
3370 
3371 		n = write(fso, bf, wsize);
3372 		if (n == -1) {
3373 			saverr = errno;
3374 			(void) fprintf(stderr,
3375 			    gettext("write error on sector %lld: %s\n"),
3376 			    bno, strerror(saverr));
3377 			lockexit(32);
3378 		}
3379 		if (n != wsize) {
3380 			saverr = errno;
3381 			(void) fprintf(stderr, gettext(
3382 			    "short write (%d of %d bytes) on sector %lld\n"),
3383 			    n, size, bno);
3384 			lockexit(32);
3385 		}
3386 
3387 		bno += block_incr;
3388 		bf += wsize;
3389 		size -= wsize;
3390 		if (size < wsize)
3391 			wsize = size;
3392 	}
3393 }
3394 
3395 
3396 /*
3397  * check if a block is available
3398  */
3399 static int
3400 isblock(struct fs *fs, unsigned char *cp, int h)
3401 {
3402 	unsigned char mask;
3403 
3404 	switch (fs->fs_frag) {
3405 	case 8:
3406 		return (cp[h] == 0xff);
3407 	case 4:
3408 		mask = 0x0f << ((h & 0x1) << 2);
3409 		return ((cp[h >> 1] & mask) == mask);
3410 	case 2:
3411 		mask = 0x03 << ((h & 0x3) << 1);
3412 		return ((cp[h >> 2] & mask) == mask);
3413 	case 1:
3414 		mask = 0x01 << (h & 0x7);
3415 		return ((cp[h >> 3] & mask) == mask);
3416 	default:
3417 		(void) fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
3418 		return (0);
3419 	}
3420 }
3421 
3422 /*
3423  * take a block out of the map
3424  */
3425 static void
3426 clrblock(struct fs *fs, unsigned char *cp, int h)
3427 {
3428 	switch ((fs)->fs_frag) {
3429 	case 8:
3430 		cp[h] = 0;
3431 		return;
3432 	case 4:
3433 		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
3434 		return;
3435 	case 2:
3436 		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
3437 		return;
3438 	case 1:
3439 		cp[h >> 3] &= ~(0x01 << (h & 0x7));
3440 		return;
3441 	default:
3442 		(void) fprintf(stderr,
3443 		    gettext("clrblock: bad fs_frag value %d\n"), fs->fs_frag);
3444 		return;
3445 	}
3446 }
3447 
3448 /*
3449  * put a block into the map
3450  */
3451 static void
3452 setblock(struct fs *fs, unsigned char *cp, int h)
3453 {
3454 	switch (fs->fs_frag) {
3455 	case 8:
3456 		cp[h] = 0xff;
3457 		return;
3458 	case 4:
3459 		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
3460 		return;
3461 	case 2:
3462 		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
3463 		return;
3464 	case 1:
3465 		cp[h >> 3] |= (0x01 << (h & 0x7));
3466 		return;
3467 	default:
3468 		(void) fprintf(stderr,
3469 		    gettext("setblock: bad fs_frag value %d\n"), fs->fs_frag);
3470 		return;
3471 	}
3472 }
3473 
3474 static void
3475 usage(void)
3476 {
3477 	(void) fprintf(stderr,
3478 	    gettext("ufs usage: mkfs [-F FSType] [-V] [-m] [-o options] "
3479 	    "special "				/* param 0 */
3480 	    "size(sectors) \\ \n"));		/* param 1 */
3481 	(void) fprintf(stderr,
3482 	    "[nsect "				/* param 2 */
3483 	    "ntrack "				/* param 3 */
3484 	    "bsize "				/* param 4 */
3485 	    "fragsize "				/* param 5 */
3486 	    "cpg "				/* param 6 */
3487 	    "free "				/* param 7 */
3488 	    "rps "				/* param 8 */
3489 	    "nbpi "				/* param 9 */
3490 	    "opt "				/* param 10 */
3491 	    "apc "				/* param 11 */
3492 	    "gap "				/* param 12 */
3493 	    "nrpos "				/* param 13 */
3494 	    "maxcontig "			/* param 14 */
3495 	    "mtb]\n");				/* param 15 */
3496 	(void) fprintf(stderr,
3497 	    gettext(" -m : dump fs cmd line used to make this partition\n"
3498 	    " -V :print this command line and return\n"
3499 	    " -o :ufs options: :nsect=%d,ntrack=%d,bsize=%d,fragsize=%d\n"
3500 	    " -o :ufs options: :cgsize=%d,free=%d,rps=%d,nbpi=%d,opt=%c\n"
3501 	    " -o :ufs options: :apc=%d,gap=%d,nrpos=%d,maxcontig=%d\n"
3502 	    " -o :ufs options: :mtb=%c,calcsb,calcbinsb\n"
3503 "NOTE that all -o suboptions: must be separated only by commas so as to\n"
3504 "be parsed as a single argument\n"),
3505 	    nsect, ntrack, bsize, fragsize, cpg, sblock.fs_minfree, rps,
3506 	    nbpi, opt, apc, (rotdelay == -1) ? 0 : rotdelay,
3507 	    sblock.fs_nrpos, maxcontig, mtb);
3508 	lockexit(32);
3509 }
3510 
3511 /*ARGSUSED*/
3512 static void
3513 dump_fscmd(char *fsys, int fsi)
3514 {
3515 	int64_t used, bpcg, inospercg;
3516 	int64_t nbpi;
3517 	uint64_t nbytes64;
3518 
3519 	bzero((char *)&sblock, sizeof (sblock));
3520 	rdfs((diskaddr_t)SBLOCK, SBSIZE, (char *)&sblock);
3521 
3522 	/*
3523 	 * ensure a valid file system and if not, exit with error or else
3524 	 * we will end up computing block numbers etc and dividing by zero
3525 	 * which will cause floating point errors in this routine.
3526 	 */
3527 
3528 	if ((sblock.fs_magic != FS_MAGIC) &&
3529 	    (sblock.fs_magic != MTB_UFS_MAGIC)) {
3530 		(void) fprintf(stderr, gettext(
3531 		    "[not currently a valid file system - bad superblock]\n"));
3532 		lockexit(32);
3533 	}
3534 
3535 	if (sblock.fs_magic == FS_MAGIC &&
3536 	    (sblock.fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
3537 	    sblock.fs_version != UFS_VERSION_MIN)) {
3538 		(void) fprintf(stderr, gettext(
3539 		    "Unknown version of UFS format: %d\n"), sblock.fs_version);
3540 		lockexit(32);
3541 	}
3542 
3543 	if (sblock.fs_magic == MTB_UFS_MAGIC &&
3544 	    (sblock.fs_version > MTB_UFS_VERSION_1 ||
3545 	    sblock.fs_version < MTB_UFS_VERSION_MIN)) {
3546 		(void) fprintf(stderr, gettext(
3547 		    "Unknown version of UFS format: %d\n"), sblock.fs_version);
3548 		lockexit(32);
3549 	}
3550 
3551 	/*
3552 	 * Compute a reasonable nbpi value.
3553 	 * The algorithm for "used" is copied from code
3554 	 * in main() verbatim.
3555 	 * The nbpi equation is taken from main where the
3556 	 * fs_ipg value is set for the last time.  The INOPB(...) - 1
3557 	 * is used to account for the roundup.
3558 	 * The problem is that a range of nbpi values map to
3559 	 * the same file system layout.  So it is not possible
3560 	 * to calculate the exact value specified when the file
3561 	 * system was created.  So instead we determine the top
3562 	 * end of the range of values.
3563 	 */
3564 	bpcg = sblock.fs_spc * sectorsize;
3565 	inospercg = (int64_t)roundup(bpcg / sizeof (struct dinode),
3566 	    INOPB(&sblock));
3567 	if (inospercg > MAXIpG(&sblock))
3568 		inospercg = MAXIpG(&sblock);
3569 	used = (int64_t)
3570 	    (sblock.fs_iblkno + inospercg / INOPF(&sblock)) * NSPF(&sblock);
3571 	used *= sectorsize;
3572 	nbytes64 = (uint64_t)sblock.fs_cpg * bpcg - used;
3573 
3574 	/*
3575 	 * The top end of the range of values for nbpi may not be
3576 	 * a valid command line value for mkfs. Report the bottom
3577 	 * end instead.
3578 	 */
3579 	nbpi = (int64_t)(nbytes64 / (sblock.fs_ipg));
3580 
3581 	(void) fprintf(stdout, gettext("mkfs -F ufs -o "), fsys);
3582 	(void) fprintf(stdout, "nsect=%d,ntrack=%d,",
3583 	    sblock.fs_nsect, sblock.fs_ntrak);
3584 	(void) fprintf(stdout, "bsize=%d,fragsize=%d,cgsize=%d,free=%d,",
3585 	    sblock.fs_bsize, sblock.fs_fsize, sblock.fs_cpg, sblock.fs_minfree);
3586 	(void) fprintf(stdout, "rps=%d,nbpi=%lld,opt=%c,apc=%d,gap=%d,",
3587 	    sblock.fs_rps, nbpi, (sblock.fs_optim == FS_OPTSPACE) ? 's' : 't',
3588 	    (sblock.fs_ntrak * sblock.fs_nsect) - sblock.fs_spc,
3589 	    sblock.fs_rotdelay);
3590 	(void) fprintf(stdout, "nrpos=%d,maxcontig=%d,mtb=%c ",
3591 	    sblock.fs_nrpos, sblock.fs_maxcontig,
3592 	    ((sblock.fs_magic == MTB_UFS_MAGIC) ? 'y' : 'n'));
3593 	(void) fprintf(stdout, "%s %lld\n", fsys,
3594 	    fsbtodb(&sblock, sblock.fs_size));
3595 
3596 	bzero((char *)&sblock, sizeof (sblock));
3597 }
3598 
3599 /* number ************************************************************* */
3600 /*									*/
3601 /* Convert a numeric string arg to binary				*/
3602 /*									*/
3603 /* Args:	d_value - default value, if have parse error		*/
3604 /*		param - the name of the argument, for error messages	*/
3605 /*		flags - parser state and what's allowed in the arg	*/
3606 /* Global arg:  string - pointer to command arg				*/
3607 /*									*/
3608 /* Valid forms: 123 | 123k | 123*123 | 123x123				*/
3609 /*									*/
3610 /* Return:	converted number					*/
3611 /*									*/
3612 /* ******************************************************************** */
3613 
3614 static uint64_t
3615 number(uint64_t d_value, char *param, int flags)
3616 {
3617 	char *cs;
3618 	uint64_t n, t;
3619 	uint64_t cut = BIG / 10;    /* limit to avoid overflow */
3620 	int minus = 0;
3621 
3622 	cs = string;
3623 	if (*cs == '-') {
3624 		minus = 1;
3625 		cs += 1;
3626 	}
3627 	if ((*cs < '0') || (*cs > '9')) {
3628 		goto bail_out;
3629 	}
3630 	n = 0;
3631 	while ((*cs >= '0') && (*cs <= '9') && (n <= cut)) {
3632 		n = n*10 + *cs++ - '0';
3633 	}
3634 	if (minus)
3635 		n = -n;
3636 	for (;;) {
3637 		switch (*cs++) {
3638 		case 'k':
3639 			if (flags & ALLOW_END_ONLY)
3640 				goto bail_out;
3641 			if (n > (BIG / 1024))
3642 				goto overflow;
3643 			n *= 1024;
3644 			continue;
3645 
3646 		case '*':
3647 		case 'x':
3648 			if (flags & ALLOW_END_ONLY)
3649 				goto bail_out;
3650 			string = cs;
3651 			t = number(d_value, param, flags);
3652 			if (n > (BIG / t))
3653 				goto overflow;
3654 			n *= t;
3655 			cs = string + 1; /* adjust for -- below */
3656 
3657 			/* recursion has read rest of expression */
3658 			/* FALLTHROUGH */
3659 
3660 		case ',':
3661 		case '\0':
3662 			cs--;
3663 			string = cs;
3664 			return (n);
3665 
3666 		case '%':
3667 			if (flags & ALLOW_END_ONLY)
3668 				goto bail_out;
3669 			if (flags & ALLOW_PERCENT) {
3670 				flags &= ~ALLOW_PERCENT;
3671 				flags |= ALLOW_END_ONLY;
3672 				continue;
3673 			}
3674 			goto bail_out;
3675 
3676 		case 'm':
3677 			if (flags & ALLOW_END_ONLY)
3678 				goto bail_out;
3679 			if (flags & ALLOW_MS1) {
3680 				flags &= ~ALLOW_MS1;
3681 				flags |= ALLOW_MS2;
3682 				continue;
3683 			}
3684 			goto bail_out;
3685 
3686 		case 's':
3687 			if (flags & ALLOW_END_ONLY)
3688 				goto bail_out;
3689 			if (flags & ALLOW_MS2) {
3690 				flags &= ~ALLOW_MS2;
3691 				flags |= ALLOW_END_ONLY;
3692 				continue;
3693 			}
3694 			goto bail_out;
3695 
3696 		case '0': case '1': case '2': case '3': case '4':
3697 		case '5': case '6': case '7': case '8': case '9':
3698 overflow:
3699 			(void) fprintf(stderr,
3700 			    gettext("mkfs: value for %s overflowed\n"),
3701 			    param);
3702 			while ((*cs != '\0') && (*cs != ','))
3703 				cs++;
3704 			string = cs;
3705 			return (BIG);
3706 
3707 		default:
3708 bail_out:
3709 			(void) fprintf(stderr, gettext(
3710 			    "mkfs: bad numeric arg for %s: \"%s\"\n"),
3711 			    param, string);
3712 			while ((*cs != '\0') && (*cs != ','))
3713 				cs++;
3714 			string = cs;
3715 			if (d_value != NO_DEFAULT) {
3716 				(void) fprintf(stderr,
3717 				    gettext("mkfs: %s reset to default %lld\n"),
3718 				    param, d_value);
3719 				return (d_value);
3720 			}
3721 			lockexit(2);
3722 
3723 		}
3724 	} /* never gets here */
3725 }
3726 
3727 /* match ************************************************************** */
3728 /*									*/
3729 /* Compare two text strings for equality				*/
3730 /*									*/
3731 /* Arg:	 s - pointer to string to match with a command arg		*/
3732 /* Global arg:  string - pointer to command arg				*/
3733 /*									*/
3734 /* Return:	1 if match, 0 if no match				*/
3735 /*		If match, also reset `string' to point to the text	*/
3736 /*		that follows the matching text.				*/
3737 /*									*/
3738 /* ******************************************************************** */
3739 
3740 static int
3741 match(char *s)
3742 {
3743 	char *cs;
3744 
3745 	cs = string;
3746 	while (*cs++ == *s) {
3747 		if (*s++ == '\0') {
3748 			goto true;
3749 		}
3750 	}
3751 	if (*s != '\0') {
3752 		return (0);
3753 	}
3754 
3755 true:
3756 	cs--;
3757 	string = cs;
3758 	return (1);
3759 }
3760 
3761 /*
3762  * GROWFS ROUTINES
3763  */
3764 
3765 /* ARGSUSED */
3766 void
3767 lockexit(int exitstatus)
3768 {
3769 	if (Pflag) {
3770 		/* the probe mode neither changes nor locks the filesystem */
3771 		exit(exitstatus);
3772 	}
3773 
3774 	/*
3775 	 * flush the dirty cylinder group
3776 	 */
3777 	if (inlockexit == 0) {
3778 		inlockexit = 1;
3779 		flcg();
3780 	}
3781 
3782 	if (aio_inited) {
3783 		flush_writes();
3784 	}
3785 
3786 	/*
3787 	 * make sure the file system is unlocked before exiting
3788 	 */
3789 	if ((inlockexit == 1) && (!isbad)) {
3790 		inlockexit = 2;
3791 		ulockfs();
3792 		/*
3793 		 * if logging was enabled, then re-enable it
3794 		 */
3795 		if (waslog) {
3796 			if (rl_log_control(fsys, _FIOLOGENABLE) != RL_SUCCESS) {
3797 				(void) fprintf(stderr, gettext(
3798 				    "failed to re-enable logging\n"));
3799 			}
3800 		}
3801 	} else if (grow) {
3802 		if (isbad) {
3803 			(void) fprintf(stderr, gettext(
3804 			    "Filesystem is currently inconsistent.  It "
3805 			    "must be repaired with fsck(1M)\nbefore being "
3806 			    "used.  Use the following command to "
3807 			    "do this:\n\n\tfsck %s\n\n"), fsys);
3808 
3809 			if (ismounted) {
3810 				(void) fprintf(stderr, gettext(
3811 				    "You will be told that the filesystem "
3812 				    "is already mounted, and asked if you\n"
3813 				    "wish to continue.  Answer `yes' to "
3814 				    "this question.\n\n"));
3815 			}
3816 
3817 			(void) fprintf(stderr, gettext(
3818 			    "One problem should be reported, that the summary "
3819 			    "information is bad.\nYou will then be asked if it "
3820 			    "should be salvaged.  Answer `yes' to\nthis "
3821 			    "question.\n\n"));
3822 		}
3823 
3824 		if (ismounted) {
3825 			/*
3826 			 * In theory, there's no way to get here without
3827 			 * isbad also being set, but be robust in the
3828 			 * face of future code changes.
3829 			 */
3830 			(void) fprintf(stderr, gettext(
3831 			    "The filesystem is currently mounted "
3832 			    "read-only and write-locked.  "));
3833 			if (isbad) {
3834 				(void) fprintf(stderr, gettext(
3835 				    "After\nrunning fsck, unlock the "
3836 				    "filesystem and "));
3837 			} else {
3838 				(void) fprintf(stderr, gettext(
3839 				    "Unlock the filesystem\nand "));
3840 			}
3841 
3842 			(void) fprintf(stderr, gettext(
3843 			    "re-enable writing with\nthe following "
3844 			    "command:\n\n\tlockfs -u %s\n\n"), directory);
3845 		}
3846 	}
3847 
3848 	exit(exitstatus);
3849 }
3850 
3851 void
3852 randomgeneration()
3853 {
3854 	int		 i;
3855 	struct dinode	*dp;
3856 
3857 	/*
3858 	 * always perform fsirand(1) function... newfs will notice that
3859 	 * the inodes have been randomized and will not call fsirand itself
3860 	 */
3861 	for (i = 0, dp = zino; i < sblock.fs_inopb; ++i, ++dp)
3862 		IRANDOMIZE(&dp->di_ic);
3863 }
3864 
3865 /*
3866  * Check the size of the summary information.
3867  * Fields in sblock are not changed in this function.
3868  *
3869  * For an 8K filesystem block, the maximum number of cylinder groups is 16384.
3870  *     MAXCSBUFS {32}  *   8K  {FS block size}
3871  *                         divided by (sizeof csum) {16}
3872  *
3873  * Note that MAXCSBUFS is not used in the kernel; as of Solaris 2.6 build 32,
3874  * this is the only place where it's referenced.
3875  */
3876 void
3877 checksummarysize()
3878 {
3879 	diskaddr_t	dmax;
3880 	diskaddr_t	dmin;
3881 	int64_t	cg0frags;
3882 	int64_t	cg0blocks;
3883 	int64_t	maxncg;
3884 	int64_t	maxfrags;
3885 	uint64_t	fs_size;
3886 	uint64_t maxfs_blocks; /* filesystem blocks for max filesystem size */
3887 
3888 	/*
3889 	 * compute the maximum summary info size
3890 	 */
3891 	dmin = cgdmin(&sblock, 0);
3892 	dmax = cgbase(&sblock, 0) + sblock.fs_fpg;
3893 	fs_size = (grow) ? grow_fs_size : sblock.fs_size;
3894 	if (dmax > fs_size)
3895 		dmax = fs_size;
3896 	cg0frags  = dmax - dmin;
3897 	cg0blocks = cg0frags / sblock.fs_frag;
3898 	cg0frags = cg0blocks * sblock.fs_frag;
3899 	maxncg   = (longlong_t)cg0blocks *
3900 	    (longlong_t)(sblock.fs_bsize / sizeof (struct csum));
3901 
3902 	maxfs_blocks = FS_MAX;
3903 
3904 	if (maxncg > ((longlong_t)maxfs_blocks / (longlong_t)sblock.fs_fpg) + 1)
3905 		maxncg = ((longlong_t)maxfs_blocks /
3906 		    (longlong_t)sblock.fs_fpg) + 1;
3907 
3908 	maxfrags = maxncg * (longlong_t)sblock.fs_fpg;
3909 
3910 	if (maxfrags > maxfs_blocks)
3911 		maxfrags = maxfs_blocks;
3912 
3913 
3914 	/*
3915 	 * remember for later processing in extendsummaryinfo()
3916 	 */
3917 	if (test)
3918 		grow_sifrag = dmin + (cg0blocks * sblock.fs_frag);
3919 	if (testfrags == 0)
3920 		testfrags = cg0frags;
3921 	if (testforce)
3922 		if (testfrags > cg0frags) {
3923 			(void) fprintf(stderr,
3924 			    gettext("Too many test frags (%lld); "
3925 			    "try %lld\n"), testfrags, cg0frags);
3926 			lockexit(32);
3927 		}
3928 
3929 	/*
3930 	 * if summary info is too large (too many cg's) tell the user and exit
3931 	 */
3932 	if ((longlong_t)sblock.fs_size > maxfrags) {
3933 		(void) fprintf(stderr, gettext(
3934 		    "Too many cylinder groups with %llu sectors;\n    try "
3935 		    "increasing cgsize, or decreasing fssize to %llu\n"),
3936 		    fsbtodb(&sblock, (uint64_t)sblock.fs_size),
3937 		    fsbtodb(&sblock, (uint64_t)maxfrags));
3938 		lockexit(32);
3939 	}
3940 }
3941 
3942 /*
3943  * checksblock() has two uses:
3944  *	- One is to sanity test the superblock and is used when newfs(1M)
3945  *	  is invoked with the "-N" option. If any discrepancy was found,
3946  *	  just return whatever error was found and do not exit.
3947  *	- the other use of it is in places where you expect the superblock
3948  *	  to be sane, and if it isn't, then we exit.
3949  * Which of the above two actions to take is indicated with the second argument.
3950  */
3951 
3952 int
3953 checksblock(struct fs sb, int proceed)
3954 {
3955 	int err = 0;
3956 	char *errmsg;
3957 
3958 	if ((sb.fs_magic != FS_MAGIC) && (sb.fs_magic != MTB_UFS_MAGIC)) {
3959 		err = 1;
3960 		errmsg = gettext("Bad superblock; magic number wrong\n");
3961 	} else if ((sb.fs_magic == FS_MAGIC &&
3962 	    (sb.fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
3963 	    sb.fs_version != UFS_VERSION_MIN)) ||
3964 	    (sb.fs_magic == MTB_UFS_MAGIC &&
3965 	    (sb.fs_version > MTB_UFS_VERSION_1 ||
3966 	    sb.fs_version < MTB_UFS_VERSION_MIN))) {
3967 		err = 2;
3968 		errmsg = gettext("Unrecognized version of UFS\n");
3969 	} else if (sb.fs_ncg < 1) {
3970 		err = 3;
3971 		errmsg = gettext("Bad superblock; ncg out of range\n");
3972 	} else if (sb.fs_cpg < 1) {
3973 		err = 4;
3974 		errmsg = gettext("Bad superblock; cpg out of range\n");
3975 	} else if (sb.fs_ncg * sb.fs_cpg < sb.fs_ncyl ||
3976 	    (sb.fs_ncg - 1) * sb.fs_cpg >= sb.fs_ncyl) {
3977 		err = 5;
3978 		errmsg = gettext("Bad superblock; ncyl out of range\n");
3979 	} else if (sb.fs_sbsize <= 0 || sb.fs_sbsize > sb.fs_bsize) {
3980 		err = 6;
3981 		errmsg = gettext("Bad superblock; superblock size out of "
3982 		    "range\n");
3983 	}
3984 
3985 	if (proceed) {
3986 		if (err) dprintf(("%s", errmsg));
3987 		return (err);
3988 	}
3989 
3990 	if (err) {
3991 		fprintf(stderr, "%s", errmsg);
3992 		lockexit(32);
3993 	}
3994 	return (32);
3995 }
3996 
3997 /*
3998  * Roll the embedded log, if any, and set up the global variables
3999  * islog and islogok.
4000  */
4001 static void
4002 logsetup(char *devstr)
4003 {
4004 	void		*buf, *ud_buf;
4005 	extent_block_t	*ebp;
4006 	ml_unit_t	*ul;
4007 	ml_odunit_t	*ud;
4008 
4009 	/*
4010 	 * Does the superblock indicate that we are supposed to have a log ?
4011 	 */
4012 	if (sblock.fs_logbno == 0) {
4013 		/*
4014 		 * No log present, nothing to do.
4015 		 */
4016 		islog = 0;
4017 		islogok = 0;
4018 		return;
4019 	} else {
4020 		/*
4021 		 * There's a log in a yet unknown state, attempt to roll it.
4022 		 */
4023 		islogok = 0;
4024 
4025 		/*
4026 		 * We failed to roll the log, bail out.
4027 		 */
4028 		if (rl_roll_log(devstr) != RL_SUCCESS)
4029 			return;
4030 
4031 		islog = 1;
4032 
4033 		/* log is not okay; check the fs */
4034 		if ((FSOKAY != (sblock.fs_state + sblock.fs_time)) ||
4035 		    (sblock.fs_clean != FSLOG))
4036 			return;
4037 
4038 		/* get the log allocation block */
4039 		buf = (void *)malloc(DEV_BSIZE);
4040 		if (buf == (void *) NULL)
4041 			return;
4042 
4043 		ud_buf = (void *)malloc(DEV_BSIZE);
4044 		if (ud_buf == (void *) NULL) {
4045 			free(buf);
4046 			return;
4047 		}
4048 
4049 		rdfs((diskaddr_t)logbtodb(&sblock, sblock.fs_logbno),
4050 		    DEV_BSIZE, buf);
4051 		ebp = (extent_block_t *)buf;
4052 
4053 		/* log allocation block is not okay; check the fs */
4054 		if (ebp->type != LUFS_EXTENTS) {
4055 			free(buf);
4056 			free(ud_buf);
4057 			return;
4058 		}
4059 
4060 		/* get the log state block(s) */
4061 		rdfs((diskaddr_t)logbtodb(&sblock, ebp->extents[0].pbno),
4062 		    DEV_BSIZE, ud_buf);
4063 		ud = (ml_odunit_t *)ud_buf;
4064 		ul = (ml_unit_t *)malloc(sizeof (*ul));
4065 		ul->un_ondisk = *ud;
4066 
4067 		/* log state is okay */
4068 		if ((ul->un_chksum == ul->un_head_ident + ul->un_tail_ident) &&
4069 		    (ul->un_version == LUFS_VERSION_LATEST) &&
4070 		    (ul->un_badlog == 0))
4071 			islogok = 1;
4072 		free(ud_buf);
4073 		free(buf);
4074 		free(ul);
4075 	}
4076 }
4077 
4078 void
4079 growinit(char *devstr)
4080 {
4081 	int	i;
4082 	char	buf[DEV_BSIZE];
4083 
4084 	/*
4085 	 * Read and verify the superblock
4086 	 */
4087 	rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
4088 	(void) checksblock(sblock, 0);
4089 	if (sblock.fs_postblformat != FS_DYNAMICPOSTBLFMT) {
4090 		(void) fprintf(stderr,
4091 		    gettext("old file system format; can't growfs\n"));
4092 		lockexit(32);
4093 	}
4094 
4095 	/*
4096 	 * can't shrink a file system
4097 	 */
4098 	grow_fssize = fsbtodb(&sblock, (uint64_t)sblock.fs_size);
4099 	if (fssize_db < grow_fssize) {
4100 		(void) fprintf(stderr,
4101 		    gettext("%lld sectors < current size of %lld sectors\n"),
4102 		    fssize_db, grow_fssize);
4103 		lockexit(32);
4104 	}
4105 
4106 	/*
4107 	 * can't grow a system to over a terabyte unless it was set up
4108 	 * as an MTB UFS file system.
4109 	 */
4110 	if (mtb == 'y' && sblock.fs_magic != MTB_UFS_MAGIC) {
4111 		if (fssize_db >= SECTORS_PER_TERABYTE) {
4112 			(void) fprintf(stderr, gettext(
4113 "File system was not set up with the multi-terabyte format.\n"));
4114 			(void) fprintf(stderr, gettext(
4115 "Its size cannot be increased to a terabyte or more.\n"));
4116 		} else {
4117 			(void) fprintf(stderr, gettext(
4118 "Cannot convert file system to multi-terabyte format.\n"));
4119 		}
4120 		lockexit(32);
4121 	}
4122 
4123 	logsetup(devstr);
4124 
4125 	/*
4126 	 * can't growfs when logging device has errors
4127 	 */
4128 	if ((islog && !islogok) ||
4129 	    ((FSOKAY == (sblock.fs_state + sblock.fs_time)) &&
4130 	    (sblock.fs_clean == FSLOG && !islog))) {
4131 		(void) fprintf(stderr,
4132 		    gettext("logging device has errors; can't growfs\n"));
4133 		lockexit(32);
4134 	}
4135 
4136 	/*
4137 	 * disable ufs logging for growing
4138 	 */
4139 	if (islog) {
4140 		if (rl_log_control(devstr, _FIOLOGDISABLE) != RL_SUCCESS) {
4141 			(void) fprintf(stderr, gettext(
4142 			    "failed to disable logging\n"));
4143 			lockexit(32);
4144 		}
4145 		islog = 0;
4146 		waslog = 1;
4147 	}
4148 
4149 	/*
4150 	 * if mounted write lock the file system to be grown
4151 	 */
4152 	if (ismounted)
4153 		wlockfs();
4154 
4155 	/*
4156 	 * refresh dynamic superblock state - disabling logging will have
4157 	 * changed the amount of free space available in the file system
4158 	 */
4159 	rdfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
4160 
4161 	/*
4162 	 * make sure device is big enough
4163 	 */
4164 	rdfs((diskaddr_t)fssize_db - 1, DEV_BSIZE, buf);
4165 	wtfs((diskaddr_t)fssize_db - 1, DEV_BSIZE, buf);
4166 
4167 	/*
4168 	 * read current summary information
4169 	 */
4170 	grow_fscs = read_summaryinfo(&sblock);
4171 
4172 	/*
4173 	 * save some current size related fields from the superblock
4174 	 * These are used in extendsummaryinfo()
4175 	 */
4176 	grow_fs_size	= sblock.fs_size;
4177 	grow_fs_ncg	= sblock.fs_ncg;
4178 	grow_fs_csaddr	= (diskaddr_t)sblock.fs_csaddr;
4179 	grow_fs_cssize	= sblock.fs_cssize;
4180 
4181 	/*
4182 	 * save and reset the clean flag
4183 	 */
4184 	if (FSOKAY == (sblock.fs_state + sblock.fs_time))
4185 		grow_fs_clean = sblock.fs_clean;
4186 	else
4187 		grow_fs_clean = FSBAD;
4188 	sblock.fs_clean = FSBAD;
4189 	sblock.fs_state = FSOKAY - sblock.fs_time;
4190 	isbad = 1;
4191 	wtfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
4192 }
4193 
4194 void
4195 checkdev(char *rdev, char *bdev)
4196 {
4197 	struct stat64	statarea;
4198 
4199 	if (stat64(bdev, &statarea) < 0) {
4200 		(void) fprintf(stderr, gettext("can't check mount point; "));
4201 		(void) fprintf(stderr, gettext("can't stat %s\n"), bdev);
4202 		lockexit(32);
4203 	}
4204 	if ((statarea.st_mode & S_IFMT) != S_IFBLK) {
4205 		(void) fprintf(stderr, gettext(
4206 		    "can't check mount point; %s is not a block device\n"),
4207 		    bdev);
4208 		lockexit(32);
4209 	}
4210 	if (stat64(rdev, &statarea) < 0) {
4211 		(void) fprintf(stderr, gettext("can't stat %s\n"), rdev);
4212 		lockexit(32);
4213 	}
4214 	if ((statarea.st_mode & S_IFMT) != S_IFCHR) {
4215 		(void) fprintf(stderr,
4216 		    gettext("%s is not a character device\n"), rdev);
4217 		lockexit(32);
4218 	}
4219 }
4220 
4221 void
4222 checkmount(struct mnttab *mntp, char *bdevname)
4223 {
4224 	struct stat64	statdir;
4225 	struct stat64	statdev;
4226 
4227 	if (strcmp(bdevname, mntp->mnt_special) == 0) {
4228 		if (stat64(mntp->mnt_mountp, &statdir) == -1) {
4229 			(void) fprintf(stderr, gettext("can't stat %s\n"),
4230 			    mntp->mnt_mountp);
4231 			lockexit(32);
4232 		}
4233 		if (stat64(mntp->mnt_special, &statdev) == -1) {
4234 			(void) fprintf(stderr, gettext("can't stat %s\n"),
4235 			    mntp->mnt_special);
4236 			lockexit(32);
4237 		}
4238 		if (statdir.st_dev != statdev.st_rdev) {
4239 			(void) fprintf(stderr, gettext(
4240 			    "%s is not mounted on %s; mnttab(4) wrong\n"),
4241 			    mntp->mnt_special, mntp->mnt_mountp);
4242 			lockexit(32);
4243 		}
4244 		ismounted = 1;
4245 		if (directory) {
4246 			if (strcmp(mntp->mnt_mountp, directory) != 0) {
4247 				(void) fprintf(stderr,
4248 				    gettext("%s is mounted on %s, not %s\n"),
4249 				    bdevname, mntp->mnt_mountp, directory);
4250 				lockexit(32);
4251 			}
4252 		} else {
4253 			if (grow)
4254 				(void) fprintf(stderr, gettext(
4255 				    "%s is mounted on %s; can't growfs\n"),
4256 				    bdevname, mntp->mnt_mountp);
4257 			else
4258 				(void) fprintf(stderr,
4259 				    gettext("%s is mounted, can't mkfs\n"),
4260 				    bdevname);
4261 			lockexit(32);
4262 		}
4263 	}
4264 }
4265 
4266 struct dinode	*dibuf	= 0;
4267 diskaddr_t	difrag	= 0;
4268 
4269 struct dinode *
4270 gdinode(ino_t ino)
4271 {
4272 	/*
4273 	 * read the block of inodes containing inode number ino
4274 	 */
4275 	if (dibuf == 0)
4276 		dibuf = (struct dinode *)malloc((unsigned)sblock.fs_bsize);
4277 	if (itod(&sblock, ino) != difrag) {
4278 		difrag = itod(&sblock, ino);
4279 		rdfs(fsbtodb(&sblock, (uint64_t)difrag), (int)sblock.fs_bsize,
4280 		    (char *)dibuf);
4281 	}
4282 	return (dibuf + (ino % INOPB(&sblock)));
4283 }
4284 
4285 /*
4286  * structure that manages the frags we need for extended summary info
4287  *	These frags can be:
4288  *		free
4289  *		data  block
4290  *		alloc block
4291  */
4292 struct csfrag {
4293 	struct csfrag	*next;		/* next entry */
4294 	daddr32_t	 ofrag;		/* old frag */
4295 	daddr32_t	 nfrag;		/* new frag */
4296 	long		 cylno;		/* cylno of nfrag */
4297 	long		 frags;		/* number of frags */
4298 	long		 size;		/* size in bytes */
4299 	ino_t		 ino;		/* inode number */
4300 	long		 fixed;		/* Boolean - Already fixed? */
4301 };
4302 struct csfrag	*csfrag;		/* state unknown */
4303 struct csfrag	*csfragino;		/* frags belonging to an inode */
4304 struct csfrag	*csfragfree;		/* frags that are free */
4305 
4306 daddr32_t maxcsfrag	= 0;		/* maximum in range */
4307 daddr32_t mincsfrag	= 0x7fffffff;	/* minimum in range */
4308 
4309 int
4310 csfraginrange(daddr32_t frag)
4311 {
4312 	return ((frag >= mincsfrag) && (frag <= maxcsfrag));
4313 }
4314 
4315 struct csfrag *
4316 findcsfrag(daddr32_t frag, struct csfrag **cfap)
4317 {
4318 	struct csfrag	*cfp;
4319 
4320 	if (!csfraginrange(frag))
4321 		return (NULL);
4322 
4323 	for (cfp = *cfap; cfp; cfp = cfp->next)
4324 		if (cfp->ofrag == frag)
4325 			return (cfp);
4326 	return (NULL);
4327 }
4328 
4329 void
4330 checkindirect(ino_t ino, daddr32_t *fragsp, daddr32_t frag, int level)
4331 {
4332 	int			i;
4333 	int			ne	= sblock.fs_bsize / sizeof (daddr32_t);
4334 	daddr32_t			fsb[MAXBSIZE / sizeof (daddr32_t)];
4335 
4336 	if (frag == 0)
4337 		return;
4338 
4339 	rdfs(fsbtodb(&sblock, frag), (int)sblock.fs_bsize,
4340 	    (char *)fsb);
4341 
4342 	checkdirect(ino, fragsp, fsb, sblock.fs_bsize / sizeof (daddr32_t));
4343 
4344 	if (level)
4345 		for (i = 0; i < ne && *fragsp; ++i)
4346 			checkindirect(ino, fragsp, fsb[i], level-1);
4347 }
4348 
4349 void
4350 addcsfrag(ino_t ino, daddr32_t frag, struct csfrag **cfap)
4351 {
4352 	struct csfrag	*cfp, *curr, *prev;
4353 
4354 	/*
4355 	 * establish a range for faster checking in csfraginrange()
4356 	 */
4357 	if (frag > maxcsfrag)
4358 		maxcsfrag = frag;
4359 	if (frag < mincsfrag)
4360 		mincsfrag = frag;
4361 
4362 	/*
4363 	 * if this frag belongs to an inode and is not the start of a block
4364 	 *	then see if it is part of a frag range for this inode
4365 	 */
4366 	if (ino && (frag % sblock.fs_frag))
4367 		for (cfp = *cfap; cfp; cfp = cfp->next) {
4368 			if (ino != cfp->ino)
4369 				continue;
4370 			if (frag != cfp->ofrag + cfp->frags)
4371 				continue;
4372 			cfp->frags++;
4373 			cfp->size += sblock.fs_fsize;
4374 			return;
4375 		}
4376 	/*
4377 	 * allocate a csfrag entry and insert it in an increasing order into the
4378 	 * specified list
4379 	 */
4380 	cfp = (struct csfrag *)calloc(1, sizeof (struct csfrag));
4381 	cfp->ino	= ino;
4382 	cfp->ofrag	= frag;
4383 	cfp->frags	= 1;
4384 	cfp->size	= sblock.fs_fsize;
4385 	for (prev = NULL, curr = *cfap; curr != NULL;
4386 	    prev = curr, curr = curr->next) {
4387 		if (frag < curr->ofrag) {
4388 			cfp->next = curr;
4389 			if (prev)
4390 				prev->next = cfp;	/* middle element */
4391 			else
4392 				*cfap = cfp;		/* first element */
4393 			break;
4394 		}
4395 		if (curr->next == NULL) {
4396 			curr->next = cfp;		/* last element	*/
4397 			break;
4398 		}
4399 	}
4400 	if (*cfap == NULL)	/* will happen only once */
4401 		*cfap = cfp;
4402 }
4403 
4404 void
4405 delcsfrag(daddr32_t frag, struct csfrag **cfap)
4406 {
4407 	struct csfrag	*cfp;
4408 	struct csfrag	**cfpp;
4409 
4410 	/*
4411 	 * free up entry whose beginning frag matches
4412 	 */
4413 	for (cfpp = cfap; *cfpp; cfpp = &(*cfpp)->next) {
4414 		if (frag == (*cfpp)->ofrag) {
4415 			cfp = *cfpp;
4416 			*cfpp = (*cfpp)->next;
4417 			free((char *)cfp);
4418 			return;
4419 		}
4420 	}
4421 }
4422 
4423 /*
4424  * See whether any of the direct blocks in the array pointed by "db" and of
4425  * length "ne" are within the range of frags needed to extend the cylinder
4426  * summary. If so, remove those frags from the "as-yet-unclassified" list
4427  * (csfrag) and add them to the "owned-by-inode" list (csfragino).
4428  * For each such frag found, decrement the frag count pointed to by fragsp.
4429  * "ino" is the inode that contains (either directly or indirectly) the frags
4430  * being checked.
4431  */
4432 void
4433 checkdirect(ino_t ino, daddr32_t *fragsp, daddr32_t *db, int ne)
4434 {
4435 	int	 i;
4436 	int	 j;
4437 	int	 found;
4438 	diskaddr_t	 frag;
4439 
4440 	/*
4441 	 * scan for allocation within the new summary info range
4442 	 */
4443 	for (i = 0; i < ne && *fragsp; ++i) {
4444 		if ((frag = *db++) != 0) {
4445 			found = 0;
4446 			for (j = 0; j < sblock.fs_frag && *fragsp; ++j) {
4447 				if (found || (found = csfraginrange(frag))) {
4448 					addcsfrag(ino, frag, &csfragino);
4449 					delcsfrag(frag, &csfrag);
4450 				}
4451 				++frag;
4452 				--(*fragsp);
4453 			}
4454 		}
4455 	}
4456 }
4457 
4458 void
4459 findcsfragino()
4460 {
4461 	int		 i;
4462 	int		 j;
4463 	daddr32_t		 frags;
4464 	struct dinode	*dp;
4465 
4466 	/*
4467 	 * scan all old inodes looking for allocations in the new
4468 	 * summary info range.  Move the affected frag from the
4469 	 * generic csfrag list onto the `owned-by-inode' list csfragino.
4470 	 */
4471 	for (i = UFSROOTINO; i < grow_fs_ncg*sblock.fs_ipg && csfrag; ++i) {
4472 		dp = gdinode((ino_t)i);
4473 		switch (dp->di_mode & IFMT) {
4474 			case IFSHAD	:
4475 			case IFLNK	:
4476 			case IFDIR	:
4477 			case IFREG	: break;
4478 			default		: continue;
4479 		}
4480 
4481 		frags   = dbtofsb(&sblock, dp->di_blocks);
4482 
4483 		checkdirect((ino_t)i, &frags, &dp->di_db[0], NDADDR+NIADDR);
4484 		for (j = 0; j < NIADDR && frags; ++j) {
4485 			/* Negate the block if its an fallocate'd block */
4486 			if (dp->di_ib[j] < 0 && dp->di_ib[j] != UFS_HOLE)
4487 				checkindirect((ino_t)i, &frags,
4488 				    -(dp->di_ib[j]), j);
4489 			else
4490 				checkindirect((ino_t)i, &frags,
4491 				    dp->di_ib[j], j);
4492 		}
4493 	}
4494 }
4495 
4496 void
4497 fixindirect(daddr32_t frag, int level)
4498 {
4499 	int			 i;
4500 	int			 ne	= sblock.fs_bsize / sizeof (daddr32_t);
4501 	daddr32_t			fsb[MAXBSIZE / sizeof (daddr32_t)];
4502 
4503 	if (frag == 0)
4504 		return;
4505 
4506 	rdfs(fsbtodb(&sblock, (uint64_t)frag), (int)sblock.fs_bsize,
4507 	    (char *)fsb);
4508 
4509 	fixdirect((caddr_t)fsb, frag, fsb, ne);
4510 
4511 	if (level)
4512 		for (i = 0; i < ne; ++i)
4513 			fixindirect(fsb[i], level-1);
4514 }
4515 
4516 void
4517 fixdirect(caddr_t bp, daddr32_t frag, daddr32_t *db, int ne)
4518 {
4519 	int	 i;
4520 	struct csfrag	*cfp;
4521 
4522 	for (i = 0; i < ne; ++i, ++db) {
4523 		if (*db == 0)
4524 			continue;
4525 		if ((cfp = findcsfrag(*db, &csfragino)) == NULL)
4526 			continue;
4527 		*db = cfp->nfrag;
4528 		cfp->fixed = 1;
4529 		wtfs(fsbtodb(&sblock, (uint64_t)frag), (int)sblock.fs_bsize,
4530 		    bp);
4531 	}
4532 }
4533 
4534 void
4535 fixcsfragino()
4536 {
4537 	int		 i;
4538 	struct dinode	*dp;
4539 	struct csfrag	*cfp;
4540 
4541 	for (cfp = csfragino; cfp; cfp = cfp->next) {
4542 		if (cfp->fixed)
4543 			continue;
4544 		dp = gdinode((ino_t)cfp->ino);
4545 		fixdirect((caddr_t)dibuf, difrag, dp->di_db, NDADDR+NIADDR);
4546 		for (i = 0; i < NIADDR; ++i)
4547 			fixindirect(dp->di_ib[i], i);
4548 	}
4549 }
4550 
4551 /*
4552  * Read the cylinders summary information specified by settings in the
4553  * passed 'fs' structure into a new allocated array of csum structures.
4554  * The caller is responsible for freeing the returned array.
4555  * Return a pointer to an array of csum structures.
4556  */
4557 static struct csum *
4558 read_summaryinfo(struct	fs *fsp)
4559 {
4560 	struct csum	*csp;
4561 	int		i;
4562 
4563 	if ((csp = malloc((size_t)fsp->fs_cssize)) == NULL) {
4564 		(void) fprintf(stderr, gettext("cannot create csum list,"
4565 		    " not enough memory\n"));
4566 		exit(32);
4567 	}
4568 
4569 	for (i = 0; i < fsp->fs_cssize; i += fsp->fs_bsize) {
4570 		rdfs(fsbtodb(fsp,
4571 		    (uint64_t)(fsp->fs_csaddr + numfrags(fsp, i))),
4572 		    (int)(fsp->fs_cssize - i < fsp->fs_bsize ?
4573 		    fsp->fs_cssize - i : fsp->fs_bsize), ((caddr_t)csp) + i);
4574 	}
4575 
4576 	return (csp);
4577 }
4578 
4579 /*
4580  * Check the allocation of fragments that are to be made part of a csum block.
4581  * A fragment is allocated if it is either in the csfragfree list or, it is
4582  * in the csfragino list and has new frags associated with it.
4583  * Return the number of allocated fragments.
4584  */
4585 int64_t
4586 checkfragallocated(daddr32_t frag)
4587 {
4588 	struct	csfrag	*cfp;
4589 	/*
4590 	 * Since the lists are sorted we can break the search if the asked
4591 	 * frag is smaller then the one in the list.
4592 	 */
4593 	for (cfp = csfragfree; cfp != NULL && frag >= cfp->ofrag;
4594 	    cfp = cfp->next) {
4595 		if (frag == cfp->ofrag)
4596 			return (1);
4597 	}
4598 	for (cfp = csfragino; cfp != NULL && frag >= cfp->ofrag;
4599 	    cfp = cfp->next) {
4600 		if (frag == cfp->ofrag && cfp->nfrag != 0)
4601 			return (cfp->frags);
4602 	}
4603 
4604 	return (0);
4605 }
4606 
4607 /*
4608  * Figure out how much the filesystem can be grown. The limiting factor is
4609  * the available free space needed to extend the cg summary info block.
4610  * The free space is determined in three steps:
4611  * - Try to extend the cg summary block to the required size.
4612  * - Find free blocks in last cg.
4613  * - Find free space in the last already allocated fragment of the summary info
4614  *   block, and use it for additional csum structures.
4615  * Return the maximum size of the new filesystem or 0 if it can't be grown.
4616  * Please note that this function leaves the global list pointers csfrag,
4617  * csfragfree, and csfragino initialized, and the caller is responsible for
4618  * freeing the lists.
4619  */
4620 diskaddr_t
4621 probe_summaryinfo()
4622 {
4623 	/* fragments by which the csum block can be extended. */
4624 	int64_t		growth_csum_frags = 0;
4625 	/* fragments by which the filesystem can be extended. */
4626 	int64_t		growth_fs_frags = 0;
4627 	int64_t		new_fs_cssize;	/* size of csum blk in the new FS */
4628 	int64_t		new_fs_ncg;	/* number of cg in the new FS */
4629 	int64_t		spare_csum;
4630 	daddr32_t	oldfrag_daddr;
4631 	daddr32_t	newfrag_daddr;
4632 	daddr32_t	daddr;
4633 	int		i;
4634 
4635 	/*
4636 	 * read and verify the superblock
4637 	 */
4638 	rdfs((diskaddr_t)(SBOFF / sectorsize), (int)sbsize, (char *)&sblock);
4639 	(void) checksblock(sblock, 0);
4640 
4641 	/*
4642 	 * check how much we can extend the cg summary info block
4643 	 */
4644 
4645 	/*
4646 	 * read current summary information
4647 	 */
4648 	fscs = read_summaryinfo(&sblock);
4649 
4650 	/*
4651 	 * build list of frags needed for cg summary info block extension
4652 	 */
4653 	oldfrag_daddr = howmany(sblock.fs_cssize, sblock.fs_fsize) +
4654 	    sblock.fs_csaddr;
4655 	new_fs_ncg = howmany(dbtofsb(&sblock, fssize_db), sblock.fs_fpg);
4656 	new_fs_cssize = fragroundup(&sblock, new_fs_ncg * sizeof (struct csum));
4657 	newfrag_daddr = howmany(new_fs_cssize, sblock.fs_fsize) +
4658 	    sblock.fs_csaddr;
4659 	/*
4660 	 * add all of the frags that are required to grow the cyl summary to the
4661 	 * csfrag list, which is the generic/unknown list, since at this point
4662 	 * we don't yet know the state of those frags.
4663 	 */
4664 	for (daddr = oldfrag_daddr; daddr < newfrag_daddr; daddr++)
4665 		addcsfrag((ino_t)0, daddr, &csfrag);
4666 
4667 	/*
4668 	 * filter free fragments and allocate them. Note that the free frags
4669 	 * must be allocated first otherwise they could be grabbed by
4670 	 * alloccsfragino() for data frags.
4671 	 */
4672 	findcsfragfree();
4673 	alloccsfragfree();
4674 
4675 	/*
4676 	 * filter fragments owned by inodes and allocate them
4677 	 */
4678 	grow_fs_ncg = sblock.fs_ncg; /* findcsfragino() needs this glob. var. */
4679 	findcsfragino();
4680 	alloccsfragino();
4681 
4682 	if (notenoughspace()) {
4683 		/*
4684 		 * check how many consecutive fragments could be allocated
4685 		 * in both lists.
4686 		 */
4687 		int64_t tmp_frags;
4688 		for (daddr = oldfrag_daddr; daddr < newfrag_daddr;
4689 		    daddr += tmp_frags) {
4690 			if ((tmp_frags = checkfragallocated(daddr)) > 0)
4691 				growth_csum_frags += tmp_frags;
4692 			else
4693 				break;
4694 		}
4695 	} else {
4696 		/*
4697 		 * We have all we need for the new desired size,
4698 		 * so clean up and report back.
4699 		 */
4700 		return (fssize_db);
4701 	}
4702 
4703 	/*
4704 	 * given the number of fragments by which the csum block can be grown
4705 	 * compute by how many new fragments the FS can be increased.
4706 	 * It is the number of csum instances per fragment multiplied by
4707 	 * `growth_csum_frags' and the number of fragments per cylinder group.
4708 	 */
4709 	growth_fs_frags = howmany(sblock.fs_fsize, sizeof (struct csum)) *
4710 	    growth_csum_frags * sblock.fs_fpg;
4711 
4712 	/*
4713 	 * compute free fragments in the last cylinder group
4714 	 */
4715 	rdcg(sblock.fs_ncg - 1);
4716 	growth_fs_frags += sblock.fs_fpg - acg.cg_ndblk;
4717 
4718 	/*
4719 	 * compute how many csum instances are unused in the old csum block.
4720 	 * For each unused csum instance the FS can be grown by one cylinder
4721 	 * group without extending the csum block.
4722 	 */
4723 	spare_csum = howmany(sblock.fs_cssize, sizeof (struct csum)) -
4724 	    sblock.fs_ncg;
4725 	if (spare_csum > 0)
4726 		growth_fs_frags += spare_csum * sblock.fs_fpg;
4727 
4728 	/*
4729 	 * recalculate the new filesystem size in sectors, shorten it by
4730 	 * the requested size `fssize_db' if necessary.
4731 	 */
4732 	if (growth_fs_frags > 0) {
4733 		diskaddr_t sect;
4734 		sect = (sblock.fs_size + growth_fs_frags) * sblock.fs_nspf;
4735 		return ((sect > fssize_db) ? fssize_db : sect);
4736 	}
4737 
4738 	return (0);
4739 }
4740 
4741 void
4742 extendsummaryinfo()
4743 {
4744 	int64_t		i;
4745 	int		localtest	= test;
4746 	int64_t		frags;
4747 	daddr32_t		oldfrag;
4748 	daddr32_t		newfrag;
4749 
4750 	/*
4751 	 * if no-write (-N), don't bother
4752 	 */
4753 	if (Nflag)
4754 		return;
4755 
4756 again:
4757 	flcg();
4758 	/*
4759 	 * summary info did not change size -- do nothing unless in test mode
4760 	 */
4761 	if (grow_fs_cssize == sblock.fs_cssize)
4762 		if (!localtest)
4763 			return;
4764 
4765 	/*
4766 	 * build list of frags needed for additional summary information
4767 	 */
4768 	oldfrag = howmany(grow_fs_cssize, sblock.fs_fsize) + grow_fs_csaddr;
4769 	newfrag = howmany(sblock.fs_cssize, sblock.fs_fsize) + grow_fs_csaddr;
4770 	/*
4771 	 * add all of the frags that are required to grow the cyl summary to the
4772 	 * csfrag list, which is the generic/unknown list, since at this point
4773 	 * we don't yet know the state of those frags.
4774 	 */
4775 	for (i = oldfrag, frags = 0; i < newfrag; ++i, ++frags)
4776 		addcsfrag((ino_t)0, (diskaddr_t)i, &csfrag);
4777 	/*
4778 	 * reduce the number of data blocks in the file system (fs_dsize) by
4779 	 * the number of frags that need to be added to the cyl summary
4780 	 */
4781 	sblock.fs_dsize -= (newfrag - oldfrag);
4782 
4783 	/*
4784 	 * In test mode, we move more data than necessary from
4785 	 * cylinder group 0.  The lookup/allocate/move code can be
4786 	 * better stressed without having to create HUGE file systems.
4787 	 */
4788 	if (localtest)
4789 		for (i = newfrag; i < grow_sifrag; ++i) {
4790 			if (frags >= testfrags)
4791 				break;
4792 			frags++;
4793 			addcsfrag((ino_t)0, (diskaddr_t)i, &csfrag);
4794 		}
4795 
4796 	/*
4797 	 * move frags to free or inode lists, depending on owner
4798 	 */
4799 	findcsfragfree();
4800 	findcsfragino();
4801 
4802 	/*
4803 	 * if not all frags can be located, file system must be inconsistent
4804 	 */
4805 	if (csfrag) {
4806 		isbad = 1;	/* should already be set, but make sure */
4807 		lockexit(32);
4808 	}
4809 
4810 	/*
4811 	 * allocate the free frags. Note that the free frags must be allocated
4812 	 * first otherwise they could be grabbed by alloccsfragino() for data
4813 	 * frags.
4814 	 */
4815 	alloccsfragfree();
4816 	/*
4817 	 * allocate extra space for inode frags
4818 	 */
4819 	alloccsfragino();
4820 
4821 	/*
4822 	 * not enough space
4823 	 */
4824 	if (notenoughspace()) {
4825 		unalloccsfragfree();
4826 		unalloccsfragino();
4827 		if (localtest && !testforce) {
4828 			localtest = 0;
4829 			goto again;
4830 		}
4831 		(void) fprintf(stderr, gettext("Not enough free space\n"));
4832 		lockexit(NOTENOUGHSPACE);
4833 	}
4834 
4835 	/*
4836 	 * copy the data from old frags to new frags
4837 	 */
4838 	copycsfragino();
4839 
4840 	/*
4841 	 * fix the inodes to point to the new frags
4842 	 */
4843 	fixcsfragino();
4844 
4845 	/*
4846 	 * We may have moved more frags than we needed.  Free them.
4847 	 */
4848 	rdcg((long)0);
4849 	for (i = newfrag; i <= maxcsfrag; ++i)
4850 		setbit(cg_blksfree(&acg), i-cgbase(&sblock, 0));
4851 	wtcg();
4852 
4853 	flcg();
4854 }
4855 
4856 /*
4857  * Check if all fragments in the `csfragino' list were reallocated.
4858  */
4859 int
4860 notenoughspace()
4861 {
4862 	struct csfrag	*cfp;
4863 
4864 	/*
4865 	 * If any element in the csfragino array has a "new frag location"
4866 	 * of 0, the allocfrags() function was unsuccessful in allocating
4867 	 * space for moving the frag represented by this array element.
4868 	 */
4869 	for (cfp = csfragino; cfp; cfp = cfp->next)
4870 		if (cfp->nfrag == 0)
4871 			return (1);
4872 	return (0);
4873 }
4874 
4875 void
4876 unalloccsfragino()
4877 {
4878 	struct csfrag	*cfp;
4879 
4880 	while ((cfp = csfragino) != NULL) {
4881 		if (cfp->nfrag)
4882 			freefrags(cfp->nfrag, cfp->frags, cfp->cylno);
4883 		delcsfrag(cfp->ofrag, &csfragino);
4884 	}
4885 }
4886 
4887 void
4888 unalloccsfragfree()
4889 {
4890 	struct csfrag	*cfp;
4891 
4892 	while ((cfp = csfragfree) != NULL) {
4893 		freefrags(cfp->ofrag, cfp->frags, cfp->cylno);
4894 		delcsfrag(cfp->ofrag, &csfragfree);
4895 	}
4896 }
4897 
4898 /*
4899  * For each frag in the "as-yet-unclassified" list (csfrag), see if
4900  * it's free (i.e., its bit is set in the free frag bit map).  If so,
4901  * move it from the "as-yet-unclassified" list to the csfragfree list.
4902  */
4903 void
4904 findcsfragfree()
4905 {
4906 	struct csfrag	*cfp;
4907 	struct csfrag	*cfpnext;
4908 
4909 	/*
4910 	 * move free frags onto the free-frag list
4911 	 */
4912 	rdcg((long)0);
4913 	for (cfp = csfrag; cfp; cfp = cfpnext) {
4914 		cfpnext = cfp->next;
4915 		if (isset(cg_blksfree(&acg), cfp->ofrag - cgbase(&sblock, 0))) {
4916 			addcsfrag(cfp->ino, cfp->ofrag, &csfragfree);
4917 			delcsfrag(cfp->ofrag, &csfrag);
4918 		}
4919 	}
4920 }
4921 
4922 void
4923 copycsfragino()
4924 {
4925 	struct csfrag	*cfp;
4926 	char		buf[MAXBSIZE];
4927 
4928 	/*
4929 	 * copy data from old frags to newly allocated frags
4930 	 */
4931 	for (cfp = csfragino; cfp; cfp = cfp->next) {
4932 		rdfs(fsbtodb(&sblock, (uint64_t)cfp->ofrag), (int)cfp->size,
4933 		    buf);
4934 		wtfs(fsbtodb(&sblock, (uint64_t)cfp->nfrag), (int)cfp->size,
4935 		    buf);
4936 	}
4937 }
4938 
4939 long	curcylno	= -1;
4940 int	cylnodirty	= 0;
4941 
4942 void
4943 rdcg(long cylno)
4944 {
4945 	if (cylno != curcylno) {
4946 		flcg();
4947 		curcylno = cylno;
4948 		rdfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, curcylno)),
4949 		    (int)sblock.fs_cgsize, (char *)&acg);
4950 	}
4951 }
4952 
4953 void
4954 flcg()
4955 {
4956 	if (cylnodirty) {
4957 		if (debug && Pflag) {
4958 			(void) fprintf(stderr,
4959 			    "Assert: cylnodirty set in probe mode\n");
4960 			return;
4961 		}
4962 		resetallocinfo();
4963 		wtfs(fsbtodb(&sblock, (uint64_t)cgtod(&sblock, curcylno)),
4964 		    (int)sblock.fs_cgsize, (char *)&acg);
4965 		cylnodirty = 0;
4966 	}
4967 	curcylno = -1;
4968 }
4969 
4970 void
4971 wtcg()
4972 {
4973 	if (!Pflag) {
4974 		/* probe mode should never write to disk */
4975 		cylnodirty = 1;
4976 	}
4977 }
4978 
4979 void
4980 allocfrags(long frags, daddr32_t *fragp, long *cylnop)
4981 {
4982 	int	 i;
4983 	int	 j;
4984 	long	 bits;
4985 	long	 bit;
4986 
4987 	/*
4988 	 * Allocate a free-frag range in an old cylinder group
4989 	 */
4990 	for (i = 0, *fragp = 0; i < grow_fs_ncg; ++i) {
4991 		if (((fscs+i)->cs_nffree < frags) && ((fscs+i)->cs_nbfree == 0))
4992 			continue;
4993 		rdcg((long)i);
4994 		bit = bits = 0;
4995 		while (findfreerange(&bit, &bits)) {
4996 			if (frags <= bits)  {
4997 				for (j = 0; j < frags; ++j)
4998 					clrbit(cg_blksfree(&acg), bit+j);
4999 				wtcg();
5000 				*cylnop = i;
5001 				*fragp  = bit + cgbase(&sblock, i);
5002 				return;
5003 			}
5004 			bit += bits;
5005 		}
5006 	}
5007 }
5008 
5009 /*
5010  * Allocate space for frags that need to be moved in order to free up space for
5011  * expanding the cylinder summary info.
5012  * For each frag that needs to be moved (each frag or range of frags in
5013  * the csfragino list), allocate a new location and store the frag number
5014  * of that new location in the nfrag field of the csfrag struct.
5015  * If a new frag can't be allocated for any element in the csfragino list,
5016  * set the new frag number for that element to 0 and return immediately.
5017  * The notenoughspace() function will detect this condition.
5018  */
5019 void
5020 alloccsfragino()
5021 {
5022 	struct csfrag	*cfp;
5023 
5024 	/*
5025 	 * allocate space for inode frag ranges
5026 	 */
5027 	for (cfp = csfragino; cfp; cfp = cfp->next) {
5028 		allocfrags(cfp->frags, &cfp->nfrag, &cfp->cylno);
5029 		if (cfp->nfrag == 0)
5030 			break;
5031 	}
5032 }
5033 
5034 void
5035 alloccsfragfree()
5036 {
5037 	struct csfrag	*cfp;
5038 
5039 	/*
5040 	 * allocate the free frags needed for extended summary info
5041 	 */
5042 	rdcg((long)0);
5043 
5044 	for (cfp = csfragfree; cfp; cfp = cfp->next)
5045 		clrbit(cg_blksfree(&acg), cfp->ofrag - cgbase(&sblock, 0));
5046 
5047 	wtcg();
5048 }
5049 
5050 void
5051 freefrags(daddr32_t frag, long frags, long cylno)
5052 {
5053 	int	i;
5054 
5055 	/*
5056 	 * free frags
5057 	 */
5058 	rdcg(cylno);
5059 	for (i = 0; i < frags; ++i) {
5060 		setbit(cg_blksfree(&acg), (frag+i) - cgbase(&sblock, cylno));
5061 	}
5062 	wtcg();
5063 }
5064 
5065 int
5066 findfreerange(long *bitp, long *bitsp)
5067 {
5068 	long	 bit;
5069 
5070 	/*
5071 	 * find a range of free bits in a cylinder group bit map
5072 	 */
5073 	for (bit = *bitp, *bitsp = 0; bit < acg.cg_ndblk; ++bit)
5074 		if (isset(cg_blksfree(&acg), bit))
5075 			break;
5076 
5077 	if (bit >= acg.cg_ndblk)
5078 		return (0);
5079 
5080 	*bitp  = bit;
5081 	*bitsp = 1;
5082 	for (++bit; bit < acg.cg_ndblk; ++bit, ++(*bitsp)) {
5083 		if ((bit % sblock.fs_frag) == 0)
5084 			break;
5085 		if (isclr(cg_blksfree(&acg), bit))
5086 			break;
5087 	}
5088 	return (1);
5089 }
5090 
5091 void
5092 resetallocinfo()
5093 {
5094 	long	cno;
5095 	long	bit;
5096 	long	bits;
5097 
5098 	/*
5099 	 * Compute the free blocks/frags info and update the appropriate
5100 	 * inmemory superblock, summary info, and cylinder group fields
5101 	 */
5102 	sblock.fs_cstotal.cs_nffree -= acg.cg_cs.cs_nffree;
5103 	sblock.fs_cstotal.cs_nbfree -= acg.cg_cs.cs_nbfree;
5104 
5105 	acg.cg_cs.cs_nffree = 0;
5106 	acg.cg_cs.cs_nbfree = 0;
5107 
5108 	bzero((caddr_t)acg.cg_frsum, sizeof (acg.cg_frsum));
5109 	bzero((caddr_t)cg_blktot(&acg), (int)(acg.cg_iusedoff-acg.cg_btotoff));
5110 
5111 	bit = bits = 0;
5112 	while (findfreerange(&bit, &bits)) {
5113 		if (bits == sblock.fs_frag) {
5114 			acg.cg_cs.cs_nbfree++;
5115 			cno = cbtocylno(&sblock, bit);
5116 			cg_blktot(&acg)[cno]++;
5117 			cg_blks(&sblock, &acg, cno)[cbtorpos(&sblock, bit)]++;
5118 		} else {
5119 			acg.cg_cs.cs_nffree += bits;
5120 			acg.cg_frsum[bits]++;
5121 		}
5122 		bit += bits;
5123 	}
5124 
5125 	*(fscs + acg.cg_cgx) = acg.cg_cs;
5126 
5127 	sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
5128 	sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
5129 }
5130 
5131 void
5132 extendcg(long cylno)
5133 {
5134 	int	i;
5135 	diskaddr_t	dupper;
5136 	diskaddr_t	cbase;
5137 	diskaddr_t	dmax;
5138 
5139 	/*
5140 	 * extend the cylinder group at the end of the old file system
5141 	 * if it was partially allocated becase of lack of space
5142 	 */
5143 	flcg();
5144 	rdcg(cylno);
5145 
5146 	dupper = acg.cg_ndblk;
5147 	if (cylno == sblock.fs_ncg - 1)
5148 		acg.cg_ncyl = sblock.fs_ncyl - (sblock.fs_cpg * cylno);
5149 	else
5150 		acg.cg_ncyl = sblock.fs_cpg;
5151 	cbase = cgbase(&sblock, cylno);
5152 	dmax = cbase + sblock.fs_fpg;
5153 	if (dmax > sblock.fs_size)
5154 		dmax = sblock.fs_size;
5155 	acg.cg_ndblk = dmax - cbase;
5156 
5157 	for (i = dupper; i < acg.cg_ndblk; ++i)
5158 		setbit(cg_blksfree(&acg), i);
5159 
5160 	sblock.fs_dsize += (acg.cg_ndblk - dupper);
5161 
5162 	wtcg();
5163 	flcg();
5164 }
5165 
5166 struct lockfs	lockfs;
5167 int		lockfd;
5168 int		islocked;
5169 int		lockfskey;
5170 char		lockfscomment[128];
5171 
5172 void
5173 ulockfs()
5174 {
5175 	/*
5176 	 * if the file system was locked, unlock it before exiting
5177 	 */
5178 	if (islocked == 0)
5179 		return;
5180 
5181 	/*
5182 	 * first, check if the lock held
5183 	 */
5184 	lockfs.lf_flags = LOCKFS_MOD;
5185 	if (ioctl(lockfd, _FIOLFSS, &lockfs) == -1) {
5186 		perror(directory);
5187 		lockexit(32);
5188 	}
5189 
5190 	if (LOCKFS_IS_MOD(&lockfs)) {
5191 		(void) fprintf(stderr,
5192 		    gettext("FILE SYSTEM CHANGED DURING GROWFS!\n"));
5193 		(void) fprintf(stderr,
5194 		    gettext("   See lockfs(1), umount(1), and fsck(1)\n"));
5195 		lockexit(32);
5196 	}
5197 	/*
5198 	 * unlock the file system
5199 	 */
5200 	lockfs.lf_lock  = LOCKFS_ULOCK;
5201 	lockfs.lf_flags = 0;
5202 	lockfs.lf_key   = lockfskey;
5203 	clockfs();
5204 	if (ioctl(lockfd, _FIOLFS, &lockfs) == -1) {
5205 		perror(directory);
5206 		lockexit(32);
5207 	}
5208 }
5209 
5210 void
5211 wlockfs()
5212 {
5213 
5214 	/*
5215 	 * if no-write (-N), don't bother
5216 	 */
5217 	if (Nflag)
5218 		return;
5219 	/*
5220 	 * open the mountpoint, and write lock the file system
5221 	 */
5222 	if ((lockfd = open64(directory, O_RDONLY)) == -1) {
5223 		perror(directory);
5224 		lockexit(32);
5225 	}
5226 
5227 	/*
5228 	 * check if it is already locked
5229 	 */
5230 	if (ioctl(lockfd, _FIOLFSS, &lockfs) == -1) {
5231 		perror(directory);
5232 		lockexit(32);
5233 	}
5234 
5235 	if (lockfs.lf_lock != LOCKFS_WLOCK) {
5236 		lockfs.lf_lock  = LOCKFS_WLOCK;
5237 		lockfs.lf_flags = 0;
5238 		lockfs.lf_key   = 0;
5239 		clockfs();
5240 		if (ioctl(lockfd, _FIOLFS, &lockfs) == -1) {
5241 			perror(directory);
5242 			lockexit(32);
5243 		}
5244 	}
5245 	islocked = 1;
5246 	lockfskey = lockfs.lf_key;
5247 }
5248 
5249 void
5250 clockfs()
5251 {
5252 	time_t	t;
5253 	char	*ct;
5254 
5255 	(void) time(&t);
5256 	ct = ctime(&t);
5257 	ct[strlen(ct)-1] = '\0';
5258 
5259 	(void) sprintf(lockfscomment, "%s -- mkfs pid %d", ct, getpid());
5260 	lockfs.lf_comlen  = strlen(lockfscomment)+1;
5261 	lockfs.lf_comment = lockfscomment;
5262 }
5263 
5264 /*
5265  * Write the csum records and the superblock
5266  */
5267 void
5268 wtsb()
5269 {
5270 	long	i;
5271 
5272 	/*
5273 	 * write summary information
5274 	 */
5275 	for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize)
5276 		wtfs(fsbtodb(&sblock, (uint64_t)(sblock.fs_csaddr +
5277 		    numfrags(&sblock, i))),
5278 		    (int)(sblock.fs_cssize - i < sblock.fs_bsize ?
5279 		    sblock.fs_cssize - i : sblock.fs_bsize),
5280 		    ((char *)fscs) + i);
5281 
5282 	/*
5283 	 * write superblock
5284 	 */
5285 	sblock.fs_time = mkfstime;
5286 	wtfs((diskaddr_t)(SBOFF / sectorsize), sbsize, (char *)&sblock);
5287 }
5288 
5289 /*
5290  * Verify that the optimization selection is reasonable, and advance
5291  * the global "string" appropriately.
5292  */
5293 static char
5294 checkopt(char *optim)
5295 {
5296 	char	opt;
5297 	int	limit = strcspn(optim, ",");
5298 
5299 	switch (limit) {
5300 	case 0:	/* missing indicator (have comma or nul) */
5301 		(void) fprintf(stderr, gettext(
5302 		    "mkfs: missing optimization flag reset to `t' (time)\n"));
5303 		opt = 't';
5304 		break;
5305 
5306 	case 1: /* single-character indicator */
5307 		opt = *optim;
5308 		if ((opt != 's') && (opt != 't')) {
5309 			(void) fprintf(stderr, gettext(
5310 		    "mkfs: bad optimization value `%c' reset to `t' (time)\n"),
5311 			    opt);
5312 			opt = 't';
5313 		}
5314 		break;
5315 
5316 	default: /* multi-character indicator */
5317 		(void) fprintf(stderr, gettext(
5318 	    "mkfs: bad optimization value `%*.*s' reset to `t' (time)\n"),
5319 		    limit, limit, optim);
5320 		opt = 't';
5321 		break;
5322 	}
5323 
5324 	string += limit;
5325 
5326 	return (opt);
5327 }
5328 
5329 /*
5330  * Verify that the mtb selection is reasonable, and advance
5331  * the global "string" appropriately.
5332  */
5333 static char
5334 checkmtb(char *mtbarg)
5335 {
5336 	char	mtbc;
5337 	int	limit = strcspn(mtbarg, ",");
5338 
5339 	switch (limit) {
5340 	case 0:	/* missing indicator (have comma or nul) */
5341 		(void) fprintf(stderr, gettext(
5342 		    "mkfs: missing mtb flag reset to `n' (no mtb support)\n"));
5343 		mtbc = 'n';
5344 		break;
5345 
5346 	case 1: /* single-character indicator */
5347 		mtbc = tolower(*mtbarg);
5348 		if ((mtbc != 'y') && (mtbc != 'n')) {
5349 			(void) fprintf(stderr, gettext(
5350 		    "mkfs: bad mtb value `%c' reset to `n' (no mtb support)\n"),
5351 			    mtbc);
5352 			mtbc = 'n';
5353 		}
5354 		break;
5355 
5356 	default: /* multi-character indicator */
5357 		(void) fprintf(stderr, gettext(
5358 	    "mkfs: bad mtb value `%*.*s' reset to `n' (no mtb support)\n"),
5359 		    limit, limit, mtbarg);
5360 		opt = 'n';
5361 		break;
5362 	}
5363 
5364 	string += limit;
5365 
5366 	return (mtbc);
5367 }
5368 
5369 /*
5370  * Verify that a value is in a range.  If it is not, resets it to
5371  * its default value if one is supplied, exits otherwise.
5372  *
5373  * When testing, can compare user_supplied to RC_KEYWORD or RC_POSITIONAL.
5374  */
5375 static void
5376 range_check(long *varp, char *name, long minimum, long maximum,
5377     long def_val, int user_supplied)
5378 {
5379 	dprintf(("DeBuG %s : %ld (%ld %ld %ld)\n",
5380 	    name, *varp, minimum, maximum, def_val));
5381 
5382 	if ((*varp < minimum) || (*varp > maximum)) {
5383 		if (user_supplied != RC_DEFAULT) {
5384 			(void) fprintf(stderr, gettext(
5385 	    "mkfs: bad value for %s: %ld must be between %ld and %ld\n"),
5386 			    name, *varp, minimum, maximum);
5387 		}
5388 		if (def_val != NO_DEFAULT) {
5389 			if (user_supplied) {
5390 				(void) fprintf(stderr,
5391 				    gettext("mkfs: %s reset to default %ld\n"),
5392 				    name, def_val);
5393 			}
5394 			*varp = def_val;
5395 			dprintf(("DeBuG %s : %ld\n", name, *varp));
5396 			return;
5397 		}
5398 		lockexit(2);
5399 		/*NOTREACHED*/
5400 	}
5401 }
5402 
5403 /*
5404  * Verify that a value is in a range.  If it is not, resets it to
5405  * its default value if one is supplied, exits otherwise.
5406  *
5407  * When testing, can compare user_supplied to RC_KEYWORD or RC_POSITIONAL.
5408  */
5409 static void
5410 range_check_64(uint64_t *varp, char *name, uint64_t minimum, uint64_t maximum,
5411     uint64_t def_val, int user_supplied)
5412 {
5413 	if ((*varp < minimum) || (*varp > maximum)) {
5414 		if (user_supplied != RC_DEFAULT) {
5415 			(void) fprintf(stderr, gettext(
5416 	    "mkfs: bad value for %s: %lld must be between %lld and %lld\n"),
5417 			    name, *varp, minimum, maximum);
5418 		}
5419 		if (def_val != NO_DEFAULT) {
5420 			if (user_supplied) {
5421 				(void) fprintf(stderr,
5422 				    gettext("mkfs: %s reset to default %lld\n"),
5423 				    name, def_val);
5424 			}
5425 			*varp = def_val;
5426 			return;
5427 		}
5428 		lockexit(2);
5429 		/*NOTREACHED*/
5430 	}
5431 }
5432 
5433 /*
5434  * Blocks SIGINT from delivery.  Returns the previous mask in the
5435  * buffer provided, so that mask may be later restored.
5436  */
5437 static void
5438 block_sigint(sigset_t *old_mask)
5439 {
5440 	sigset_t block_mask;
5441 
5442 	if (sigemptyset(&block_mask) < 0) {
5443 		fprintf(stderr, gettext("Could not clear signal mask\n"));
5444 		lockexit(3);
5445 	}
5446 	if (sigaddset(&block_mask, SIGINT) < 0) {
5447 		fprintf(stderr, gettext("Could not set signal mask\n"));
5448 		lockexit(3);
5449 	}
5450 	if (sigprocmask(SIG_BLOCK, &block_mask, old_mask) < 0) {
5451 		fprintf(stderr, gettext("Could not block SIGINT\n"));
5452 		lockexit(3);
5453 	}
5454 }
5455 
5456 /*
5457  * Restores the signal mask that was in force before a call
5458  * to block_sigint().  This may actually still have SIGINT blocked,
5459  * if we've been recursively invoked.
5460  */
5461 static void
5462 unblock_sigint(sigset_t *old_mask)
5463 {
5464 	if (sigprocmask(SIG_UNBLOCK, old_mask, (sigset_t *)NULL) < 0) {
5465 		fprintf(stderr, gettext("Could not restore signal mask\n"));
5466 		lockexit(3);
5467 	}
5468 }
5469 
5470 /*
5471  * Attempt to be somewhat graceful about being interrupted, rather than
5472  * just silently leaving the filesystem in an unusable state.
5473  *
5474  * The kernel has blocked SIGINT upon entry, so we don't have to worry
5475  * about recursion if the user starts pounding on the keyboard.
5476  */
5477 static void
5478 recover_from_sigint(int signum)
5479 {
5480 	if (fso > -1) {
5481 		if ((Nflag != 0) || confirm_abort()) {
5482 			lockexit(4);
5483 		}
5484 	}
5485 }
5486 
5487 static int
5488 confirm_abort(void)
5489 {
5490 	char line[80];
5491 
5492 	printf(gettext("\n\nAborting at this point will leave the filesystem "
5493 	    "in an inconsistent\nstate.  If you do choose to stop, "
5494 	    "you will be given instructions on how to\nrecover "
5495 	    "the filesystem.  Do you wish to cancel the filesystem "
5496 	    "grow\noperation (y/n)?"));
5497 	if (getaline(stdin, line, sizeof (line)) == EOF)
5498 		line[0] = 'y';
5499 
5500 	printf("\n");
5501 	if (line[0] == 'y' || line[0] == 'Y')
5502 		return (1);
5503 	else {
5504 		return (0);
5505 	}
5506 }
5507 
5508 static int
5509 getaline(FILE *fp, char *loc, int maxlen)
5510 {
5511 	int n;
5512 	char *p, *lastloc;
5513 
5514 	p = loc;
5515 	lastloc = &p[maxlen-1];
5516 	while ((n = getc(fp)) != '\n') {
5517 		if (n == EOF)
5518 			return (EOF);
5519 		if (!isspace(n) && p < lastloc)
5520 			*p++ = n;
5521 	}
5522 	*p = 0;
5523 	return (p - loc);
5524 }
5525 
5526 /*
5527  * Calculate the maximum value of cylinders-per-group for a file
5528  * system with the characteristics:
5529  *
5530  *	bsize - file system block size
5531  *	fragsize - frag size
5532  *	nbpi - number of bytes of disk space per inode
5533  *	nrpos - number of rotational positions
5534  *	spc - sectors per cylinder
5535  *
5536  * These five characteristic are not adjustable (by this function).
5537  * The only attribute of the file system which IS adjusted by this
5538  * function in order to maximize cylinders-per-group is the proportion
5539  * of the cylinder group overhead block used for the inode map.  The
5540  * inode map cannot occupy more than one-third of the cylinder group
5541  * overhead block, but it's OK for it to occupy less than one-third
5542  * of the overhead block.
5543  *
5544  * The setting of nbpi determines one possible value for the maximum
5545  * size of a cylinder group.  It does so because it determines the total
5546  * number of inodes in the file system (file system size is fixed, and
5547  * nbpi is fixed, so the total number of inodes is fixed too).  The
5548  * cylinder group has to be small enough so that the number of inodes
5549  * in the cylinder group is less than or equal to the number of bits
5550  * in one-third (or whatever proportion is assumed) of a file system
5551  * block.  The details of the calculation are:
5552  *
5553  *     The macro MAXIpG_B(bsize, inode_divisor) determines the maximum
5554  *     number of inodes that can be in a cylinder group, given the
5555  *     proportion of the cylinder group overhead block used for the
5556  *     inode bitmaps (an inode_divisor of 3 means that 1/3 of the
5557  *     block is used for inode bitmaps; an inode_divisor of 12 means
5558  *     that 1/12 of the block is used for inode bitmaps.)
5559  *
5560  *     Once the number of inodes per cylinder group is known, the
5561  *     maximum value of cylinders-per-group (determined by nbpi)
5562  *     is calculated by the formula
5563  *
5564  *     maxcpg_given_nbpi = (size of a cylinder group)/(size of a cylinder)
5565  *
5566  *			 = (inodes-per-cg * nbpi)/(spc * DEV_BSIZE)
5567  *
5568  *     (Interestingly, the size of the file system never enters
5569  *     into this calculation.)
5570  *
5571  * Another possible value for the maximum cylinder group size is determined
5572  * by frag_size and nrpos.  The frags in the cylinder group must be
5573  * representable in the frag bitmaps in the cylinder overhead block and the
5574  * rotational positions for each cylinder must be represented in the
5575  * rotational position tables.  The calculation of the maximum cpg
5576  * value, given the frag and nrpos vales, is:
5577  *
5578  *     maxcpg_given_fragsize =
5579  *	  (available space in the overhead block) / (size of per-cylinder data)
5580  *
5581  *     The available space in the overhead block =
5582  *	  bsize - sizeof (struct cg) - space_used_for_inode_bitmaps
5583  *
5584  *     The size of the per-cylinder data is:
5585  *	    sizeof(long)            # for the "blocks avail per cylinder" field
5586  *	    + nrpos * sizeof(short)   # for the rotational position table entry
5587  *	    + frags-per-cylinder/NBBY # number of bytes to represent this
5588  *				      # cylinder in the frag bitmap
5589  *
5590  * The two calculated maximum values of cylinder-per-group will typically
5591  * turn out to be different, since they are derived from two different
5592  * constraints.  Usually, maxcpg_given_nbpi is much bigger than
5593  * maxcpg_given_fragsize.  But they can be brought together by
5594  * adjusting the proportion of the overhead block dedicated to
5595  * the inode bitmaps.  Decreasing the proportion of the cylinder
5596  * group overhead block used for inode maps will decrease
5597  * maxcpg_given_nbpi and increase maxcpg_given_fragsize.
5598  *
5599  * This function calculates the initial values of maxcpg_given_nbpi
5600  * and maxcpg_given_fragsize assuming that 1/3 of the cg overhead
5601  * block is used for inode bitmaps.  Then it decreases the proportion
5602  * of the cg overhead block used for inode bitmaps (by increasing
5603  * the value of inode_divisor) until maxcpg_given_nbpi and
5604  * maxcpg_given_fragsize are the same, or stop changing, or
5605  * maxcpg_given_nbpi is less than maxcpg_given_fragsize.
5606  *
5607  * The loop terminates when any of the following occur:
5608  *	* maxcpg_given_fragsize is greater than or equal to
5609  *	  maxcpg_given_nbpi
5610  *	* neither maxcpg_given_fragsize nor maxcpg_given_nbpi
5611  *	  change in the expected direction
5612  *
5613  * The loop is guaranteed to terminate because it only continues
5614  * while maxcpg_given_fragsize and maxcpg_given_nbpi are approaching
5615  * each other.  As soon they cross each other, or neither one changes
5616  * in the direction of the other, or one of them moves in the wrong
5617  * direction, the loop completes.
5618  */
5619 
5620 static long
5621 compute_maxcpg(long bsize, long fragsize, long nbpi, long nrpos, long spc)
5622 {
5623 	int	maxcpg_given_nbpi;	/* in cylinders */
5624 	int	maxcpg_given_fragsize;	/* in cylinders */
5625 	int	spf;			/* sectors per frag */
5626 	int	inode_divisor;
5627 	int	old_max_given_frag = 0;
5628 	int	old_max_given_nbpi = INT_MAX;
5629 
5630 	spf = fragsize / DEV_BSIZE;
5631 	inode_divisor = 3;
5632 
5633 	while (1) {
5634 		maxcpg_given_nbpi =
5635 		    (((int64_t)(MAXIpG_B(bsize, inode_divisor))) * nbpi) /
5636 		    (DEV_BSIZE * ((int64_t)spc));
5637 		maxcpg_given_fragsize =
5638 		    (bsize - (sizeof (struct cg)) - (bsize / inode_divisor)) /
5639 		    (sizeof (long) + nrpos * sizeof (short) +
5640 		    (spc / spf) / NBBY);
5641 
5642 		if (maxcpg_given_fragsize >= maxcpg_given_nbpi)
5643 			return (maxcpg_given_nbpi);
5644 
5645 		/*
5646 		 * If neither value moves toward the other, return the
5647 		 * least of the old values (we use the old instead of the
5648 		 * new because: if the old is the same as the new, it
5649 		 * doesn't matter which ones we use.  If one of the
5650 		 * values changed, but in the wrong direction, the
5651 		 * new values are suspect.  Better use the old.  This
5652 		 * shouldn't happen, but it's best to check.
5653 		 */
5654 
5655 		if (!(maxcpg_given_nbpi < old_max_given_nbpi) &&
5656 		    !(maxcpg_given_fragsize > old_max_given_frag))
5657 			return (MIN(old_max_given_nbpi, old_max_given_frag));
5658 
5659 		/*
5660 		 * This is probably impossible, but if one of the maxcpg
5661 		 * values moved in the "right" direction and one moved
5662 		 * in the "wrong" direction (that is, the two values moved
5663 		 * in the same direction), the previous conditional won't
5664 		 * recognize that the values aren't converging (since at
5665 		 * least one value moved in the "right" direction, the
5666 		 * last conditional says "keep going").
5667 		 *
5668 		 * Just to make absolutely certain that the loop terminates,
5669 		 * check for one of the values moving in the "wrong" direction
5670 		 * and terminate the loop if it happens.
5671 		 */
5672 
5673 		if (maxcpg_given_nbpi > old_max_given_nbpi ||
5674 		    maxcpg_given_fragsize < old_max_given_frag)
5675 			return (MIN(old_max_given_nbpi, old_max_given_frag));
5676 
5677 		old_max_given_nbpi = maxcpg_given_nbpi;
5678 		old_max_given_frag = maxcpg_given_fragsize;
5679 
5680 		inode_divisor++;
5681 	}
5682 }
5683 
5684 static int
5685 in_64bit_mode(void)
5686 {
5687 	/*  cmd must be an absolute path, for security */
5688 	char *cmd = "/usr/bin/isainfo -b";
5689 	char buf[BUFSIZ];
5690 	FILE *ptr;
5691 	int retval = 0;
5692 
5693 	putenv("IFS= \t");
5694 	if ((ptr = popen(cmd, "r")) != NULL) {
5695 		if (fgets(buf, BUFSIZ, ptr) != NULL &&
5696 		    strncmp(buf, "64", 2) == 0)
5697 			retval = 1;
5698 		(void) pclose(ptr);
5699 	}
5700 	return (retval);
5701 }
5702 
5703 /*
5704  * validate_size
5705  *
5706  * Return 1 if the device appears to be at least "size" sectors long.
5707  * Return 0 if it's shorter or we can't read it.
5708  */
5709 
5710 static int
5711 validate_size(int fd, diskaddr_t size)
5712 {
5713 	char	buf[DEV_BSIZE];
5714 	int rc;
5715 
5716 	if ((llseek(fd, (offset_t)((size - 1) * DEV_BSIZE), SEEK_SET) == -1) ||
5717 	    (read(fd, buf, DEV_BSIZE)) != DEV_BSIZE)
5718 		rc = 0;
5719 	else
5720 		rc = 1;
5721 	return (rc);
5722 }
5723 
5724 /*
5725  * Print every field of the calculated superblock, along with
5726  * its value.  To make parsing easier on the caller, the value
5727  * is printed first, then the name.  Additionally, there's only
5728  * one name/value pair per line.  All values are reported in
5729  * hexadecimal (with the traditional 0x prefix), as that's slightly
5730  * easier for humans to read.  Not that they're expected to, but
5731  * debugging happens.
5732  */
5733 static void
5734 dump_sblock(void)
5735 {
5736 	int row, column, pending, written;
5737 	caddr_t source;
5738 
5739 	if (Rflag) {
5740 		pending = sizeof (sblock);
5741 		source = (caddr_t)&sblock;
5742 		do {
5743 			written = write(fileno(stdout), source, pending);
5744 			pending -= written;
5745 			source += written;
5746 		} while ((pending > 0) && (written > 0));
5747 
5748 		if (written < 0) {
5749 			perror(gettext("Binary dump of superblock failed"));
5750 			lockexit(1);
5751 		}
5752 		return;
5753 	} else {
5754 		printf("0x%x sblock.fs_link\n", sblock.fs_link);
5755 		printf("0x%x sblock.fs_rolled\n", sblock.fs_rolled);
5756 		printf("0x%x sblock.fs_sblkno\n", sblock.fs_sblkno);
5757 		printf("0x%x sblock.fs_cblkno\n", sblock.fs_cblkno);
5758 		printf("0x%x sblock.fs_iblkno\n", sblock.fs_iblkno);
5759 		printf("0x%x sblock.fs_dblkno\n", sblock.fs_dblkno);
5760 		printf("0x%x sblock.fs_cgoffset\n", sblock.fs_cgoffset);
5761 		printf("0x%x sblock.fs_cgmask\n", sblock.fs_cgmask);
5762 		printf("0x%x sblock.fs_time\n", sblock.fs_time);
5763 		printf("0x%x sblock.fs_size\n", sblock.fs_size);
5764 		printf("0x%x sblock.fs_dsize\n", sblock.fs_dsize);
5765 		printf("0x%x sblock.fs_ncg\n", sblock.fs_ncg);
5766 		printf("0x%x sblock.fs_bsize\n", sblock.fs_bsize);
5767 		printf("0x%x sblock.fs_fsize\n", sblock.fs_fsize);
5768 		printf("0x%x sblock.fs_frag\n", sblock.fs_frag);
5769 		printf("0x%x sblock.fs_minfree\n", sblock.fs_minfree);
5770 		printf("0x%x sblock.fs_rotdelay\n", sblock.fs_rotdelay);
5771 		printf("0x%x sblock.fs_rps\n", sblock.fs_rps);
5772 		printf("0x%x sblock.fs_bmask\n", sblock.fs_bmask);
5773 		printf("0x%x sblock.fs_fmask\n", sblock.fs_fmask);
5774 		printf("0x%x sblock.fs_bshift\n", sblock.fs_bshift);
5775 		printf("0x%x sblock.fs_fshift\n", sblock.fs_fshift);
5776 		printf("0x%x sblock.fs_maxcontig\n", sblock.fs_maxcontig);
5777 		printf("0x%x sblock.fs_maxbpg\n", sblock.fs_maxbpg);
5778 		printf("0x%x sblock.fs_fragshift\n", sblock.fs_fragshift);
5779 		printf("0x%x sblock.fs_fsbtodb\n", sblock.fs_fsbtodb);
5780 		printf("0x%x sblock.fs_sbsize\n", sblock.fs_sbsize);
5781 		printf("0x%x sblock.fs_csmask\n", sblock.fs_csmask);
5782 		printf("0x%x sblock.fs_csshift\n", sblock.fs_csshift);
5783 		printf("0x%x sblock.fs_nindir\n", sblock.fs_nindir);
5784 		printf("0x%x sblock.fs_inopb\n", sblock.fs_inopb);
5785 		printf("0x%x sblock.fs_nspf\n", sblock.fs_nspf);
5786 		printf("0x%x sblock.fs_optim\n", sblock.fs_optim);
5787 #ifdef _LITTLE_ENDIAN
5788 		printf("0x%x sblock.fs_state\n", sblock.fs_state);
5789 #else
5790 		printf("0x%x sblock.fs_npsect\n", sblock.fs_npsect);
5791 #endif
5792 		printf("0x%x sblock.fs_si\n", sblock.fs_si);
5793 		printf("0x%x sblock.fs_trackskew\n", sblock.fs_trackskew);
5794 		printf("0x%x sblock.fs_id[0]\n", sblock.fs_id[0]);
5795 		printf("0x%x sblock.fs_id[1]\n", sblock.fs_id[1]);
5796 		printf("0x%x sblock.fs_csaddr\n", sblock.fs_csaddr);
5797 		printf("0x%x sblock.fs_cssize\n", sblock.fs_cssize);
5798 		printf("0x%x sblock.fs_cgsize\n", sblock.fs_cgsize);
5799 		printf("0x%x sblock.fs_ntrak\n", sblock.fs_ntrak);
5800 		printf("0x%x sblock.fs_nsect\n", sblock.fs_nsect);
5801 		printf("0x%x sblock.fs_spc\n", sblock.fs_spc);
5802 		printf("0x%x sblock.fs_ncyl\n", sblock.fs_ncyl);
5803 		printf("0x%x sblock.fs_cpg\n", sblock.fs_cpg);
5804 		printf("0x%x sblock.fs_ipg\n", sblock.fs_ipg);
5805 		printf("0x%x sblock.fs_fpg\n", sblock.fs_fpg);
5806 		printf("0x%x sblock.fs_cstotal\n", sblock.fs_cstotal);
5807 		printf("0x%x sblock.fs_fmod\n", sblock.fs_fmod);
5808 		printf("0x%x sblock.fs_clean\n", sblock.fs_clean);
5809 		printf("0x%x sblock.fs_ronly\n", sblock.fs_ronly);
5810 		printf("0x%x sblock.fs_flags\n", sblock.fs_flags);
5811 		printf("0x%x sblock.fs_fsmnt\n", sblock.fs_fsmnt);
5812 		printf("0x%x sblock.fs_cgrotor\n", sblock.fs_cgrotor);
5813 		printf("0x%x sblock.fs_u.fs_csp\n", sblock.fs_u.fs_csp);
5814 		printf("0x%x sblock.fs_cpc\n", sblock.fs_cpc);
5815 
5816 		/*
5817 		 * No macros are defined for the dimensions of the
5818 		 * opostbl array.
5819 		 */
5820 		for (row = 0; row < 16; row++) {
5821 			for (column = 0; column < 8; column++) {
5822 				printf("0x%x sblock.fs_opostbl[%d][%d]\n",
5823 				    sblock.fs_opostbl[row][column],
5824 				    row, column);
5825 			}
5826 		}
5827 
5828 		/*
5829 		 * Ditto the size of sparecon.
5830 		 */
5831 		for (row = 0; row < 51; row++) {
5832 			printf("0x%x sblock.fs_sparecon[%d]\n",
5833 			    sblock.fs_sparecon[row], row);
5834 		}
5835 
5836 		printf("0x%x sblock.fs_version\n", sblock.fs_version);
5837 		printf("0x%x sblock.fs_logbno\n", sblock.fs_logbno);
5838 		printf("0x%x sblock.fs_reclaim\n", sblock.fs_reclaim);
5839 		printf("0x%x sblock.fs_sparecon2\n", sblock.fs_sparecon2);
5840 #ifdef _LITTLE_ENDIAN
5841 		printf("0x%x sblock.fs_npsect\n", sblock.fs_npsect);
5842 #else
5843 		printf("0x%x sblock.fs_state\n", sblock.fs_state);
5844 #endif
5845 		printf("0x%llx sblock.fs_qbmask\n", sblock.fs_qbmask);
5846 		printf("0x%llx sblock.fs_qfmask\n", sblock.fs_qfmask);
5847 		printf("0x%x sblock.fs_postblformat\n", sblock.fs_postblformat);
5848 		printf("0x%x sblock.fs_nrpos\n", sblock.fs_nrpos);
5849 		printf("0x%x sblock.fs_postbloff\n", sblock.fs_postbloff);
5850 		printf("0x%x sblock.fs_rotbloff\n", sblock.fs_rotbloff);
5851 		printf("0x%x sblock.fs_magic\n", sblock.fs_magic);
5852 
5853 		/*
5854 		 * fs_space isn't of much use in this context, so we'll
5855 		 * just ignore it for now.
5856 		 */
5857 	}
5858 }
5859