xref: /illumos-gate/usr/src/cmd/fs.d/ufs/newfs/newfs.c (revision 6451fdbc)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 
23 #pragma ident	"%Z%%M%	%I%	%E% SMI"
24 	/* from UCB 5.2 9/11/85 */
25 
26 /*
27  * newfs: friendly front end to mkfs
28  *
29  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
30  * Use is subject to license terms.
31  */
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <locale.h>
36 #include <sys/stat.h>
37 #include <sys/buf.h>
38 #include <sys/fs/ufs_fs.h>
39 #include <sys/vnode.h>
40 #include <sys/fs/ufs_inode.h>
41 #include <sys/sysmacros.h>
42 
43 #include <errno.h>
44 #include <stdio.h>
45 #include <string.h>
46 #include <stdlib.h>
47 #include <stdarg.h>
48 #include <stdio.h>
49 #include <fcntl.h>
50 #include <unistd.h>
51 #include <limits.h>
52 #include <libintl.h>
53 #include <sys/dkio.h>
54 #include <sys/vtoc.h>
55 #include <sys/mkdev.h>
56 #include <sys/efi_partition.h>
57 
58 #include <fslib.h>
59 
60 static unsigned int number(char *, char *, int, int);
61 static int64_t number64(char *, char *, int, int64_t);
62 static diskaddr_t getdiskbydev(char *);
63 static int  yes(void);
64 static int  notrand(char *);
65 static void usage();
66 static diskaddr_t get_device_size(int, char *);
67 static diskaddr_t brute_force_get_device_size(int);
68 static int validate_size(char *disk, diskaddr_t size);
69 static void exenv(void);
70 static struct fs *read_sb(char *);
71 /*PRINTFLIKE1*/
72 static void fatal(char *fmt, ...);
73 
74 #define	EPATH "PATH=/usr/sbin:/sbin:"
75 #define	CPATH "/sbin"					/* an EPATH element */
76 #define	MB (1024 * 1024)
77 #define	GBSEC ((1024 * 1024 * 1024) / DEV_BSIZE)	/* sectors in a GB */
78 #define	MINFREESEC ((64 * 1024 * 1024) / DEV_BSIZE)	/* sectors in 64 MB */
79 #define	MINCPG (16)	/* traditional */
80 #define	MAXDEFDENSITY (8 * 1024)	/* arbitrary */
81 #define	MINDENSITY (2 * 1024)	/* traditional */
82 #define	MIN_MTB_DENSITY (1024 * 1024)
83 #define	POWEROF2(num)	(((num) & ((num) - 1)) == 0)
84 #define	SECTORS_PER_TERABYTE	(1LL << 31)
85 /*
86  * The following constant specifies an upper limit for file system size
87  * that is actually a lot bigger than we expect to support with UFS. (Since
88  * it's specified in sectors, the file system size would be 2**44 * 512,
89  * which is 2**53, which is 8192 Terabytes.)  However, it's useful
90  * for checking the basic sanity of a size value that is input on the
91  * command line.
92  */
93 #define	FS_SIZE_UPPER_LIMIT	0x100000000000LL
94 
95 /* For use with number() */
96 #define	NR_NONE		0
97 #define	NR_PERCENT	0x01
98 
99 /*
100  * The following two constants set the default block and fragment sizes.
101  * Both constants must be a power of 2 and meet the following constraints:
102  *	MINBSIZE <= DESBLKSIZE <= MAXBSIZE
103  *	DEV_BSIZE <= DESFRAGSIZE <= DESBLKSIZE
104  *	DESBLKSIZE / DESFRAGSIZE <= 8
105  */
106 #define	DESBLKSIZE	8192
107 #define	DESFRAGSIZE	1024
108 
109 #define	dprintf(x)	if (debug) printf x
110 
111 int		debug = 0;	/* enable debug output */
112 static int	Nflag;		/* run mkfs without writing file system */
113 static int	Tflag;		/* set up file system for growth to over 1 TB */
114 static int	verbose;	/* show mkfs line before exec */
115 static int	fsize = 0;		/* fragment size */
116 static int	fsize_flag = 0;	/* fragment size was specified on cmd line */
117 static int	bsize;		/* block size */
118 static int	ntracks;	/* # tracks/cylinder */
119 static int	ntracks_set = 0; /* true if the user specified ntracks */
120 static int	optim = FS_OPTTIME;	/* optimization, t(ime) or s(pace) */
121 static int	nsectors;	/* # sectors/track */
122 static int	geom_nsectors;	/* # sectors/track as returned by GEOM ioctl */
123 static int	cpg;		/* cylinders/cylinder group */
124 static int	cpg_set = 0;	/* true if the user specified cpg */
125 static int	minfree = -1;	/* free space threshold */
126 static int	rpm;		/* revolutions/minute of drive */
127 static int	rpm_set = 0;	/* true if the user specified rpm */
128 static int	nrpos = 8;	/* # of distinguished rotational positions */
129 				/* 8 is the historical default */
130 static int	nrpos_set = 0;	/* true if the user specified nrpos */
131 static int	density = 0;	/* number of bytes per inode */
132 static int	apc;		/* alternates per cylinder */
133 static int	apc_set = 0;	/* true if the user specified apc */
134 static int 	rot = -1;	/* rotational delay (msecs) */
135 static int	rot_set = 0;	/* true if the user specified rot */
136 static int 	maxcontig = -1;	/* maximum number of contig blocks */
137 static int	text_sb = 0;	/* no disk changes; just final sb text dump */
138 static int	binary_sb = 0;	/* no disk changes; just final sb binary dump */
139 static int	label_type;	/* see types below */
140 
141 /*
142  * The variable use_efi_dflts is an indicator of whether to use EFI logic
143  * or the geometry logic in laying out the filesystem. This is decided
144  * based on the size of the disk and is used only for non-EFI labeled disks.
145  */
146 static int	use_efi_dflts = 0;
147 
148 static char	device[MAXPATHLEN];
149 static char	cmd[BUFSIZ];
150 
151 extern	char	*getfullrawname(); /* from libadm */
152 
153 int
154 main(int argc, char *argv[])
155 {
156 	char *special, *name;
157 	struct stat64 st;
158 	int status;
159 	int option;
160 	int nsect;
161 	struct fs *sbp;	/* Pointer to superblock (if present) */
162 	diskaddr_t actual_fssize;
163 	diskaddr_t max_possible_fssize;
164 	diskaddr_t req_fssize = 0;
165 	diskaddr_t fssize = 0;
166 	char	*req_fssize_str = NULL; /* requested size argument */
167 
168 	(void) setlocale(LC_ALL, "");
169 
170 #if !defined(TEXT_DOMAIN)
171 #define	TEXT_DOMAIN	"SYS_TEST"
172 #endif
173 	(void) textdomain(TEXT_DOMAIN);
174 
175 	opterr = 0;	/* We print our own errors, disable getopt's message */
176 	while ((option = getopt(argc, argv,
177 	    "vNBSs:C:d:t:o:a:b:f:c:m:n:r:i:T")) != EOF) {
178 		switch (option) {
179 		case 'S':
180 			text_sb++;
181 			break;
182 		case 'B':
183 			binary_sb++;
184 			break;
185 		case 'v':
186 			verbose++;
187 			break;
188 
189 		case 'N':
190 			Nflag++;
191 			break;
192 
193 		case 's':
194 			/*
195 			 * The maximum file system size is a lot smaller
196 			 * than FS_SIZE_UPPER_LIMIT, but until we find out
197 			 * the device size and block size, we don't know
198 			 * what it is.  So save the requested size in a
199 			 * string so that we can print it out later if we
200 			 * determine it's too big.
201 			 */
202 			req_fssize = number64("fssize", optarg, NR_NONE,
203 			    FS_SIZE_UPPER_LIMIT);
204 			if (req_fssize < 1024)
205 				fatal(gettext(
206 				    "%s: fssize must be at least 1024"),
207 				    optarg);
208 			req_fssize_str = strdup(optarg);
209 			if (req_fssize_str == NULL)
210 				fatal(gettext(
211 				    "Insufficient memory for string copy."));
212 			break;
213 
214 		case 'C':
215 			maxcontig = number("maxcontig", optarg, NR_NONE, -1);
216 			if (maxcontig < 0)
217 				fatal(gettext("%s: bad maxcontig"), optarg);
218 			break;
219 
220 		case 'd':
221 			rot = number("rotdelay", optarg, NR_NONE, 0);
222 			rot_set = 1;
223 			if (rot < 0 || rot > 1000)
224 				fatal(gettext(
225 				    "%s: bad rotational delay"), optarg);
226 			break;
227 
228 		case 't':
229 			ntracks = number("ntrack", optarg, NR_NONE, 16);
230 			ntracks_set = 1;
231 			if ((ntracks < 0) ||
232 			    (ntracks > INT_MAX))
233 				fatal(gettext("%s: bad total tracks"), optarg);
234 			break;
235 
236 		case 'o':
237 			if (strcmp(optarg, "space") == 0)
238 			    optim = FS_OPTSPACE;
239 			else if (strcmp(optarg, "time") == 0)
240 			    optim = FS_OPTTIME;
241 			else
242 			    fatal(gettext(
243 "%s: bad optimization preference (options are `space' or `time')"),
244 				optarg);
245 			break;
246 
247 		case 'a':
248 			apc = number("apc", optarg, NR_NONE, 0);
249 			apc_set = 1;
250 			if (apc < 0 || apc > 32768) /* see mkfs.c */
251 				fatal(gettext(
252 				    "%s: bad alternates per cyl"), optarg);
253 			break;
254 
255 		case 'b':
256 			bsize = number("bsize", optarg, NR_NONE, DESBLKSIZE);
257 			if (bsize < MINBSIZE || bsize > MAXBSIZE)
258 				fatal(gettext(
259 				    "%s: bad block size"), optarg);
260 			break;
261 
262 		case 'f':
263 			fsize = number("fragsize", optarg, NR_NONE,
264 				DESFRAGSIZE);
265 			fsize_flag++;
266 			/* xxx ought to test against bsize for upper limit */
267 			if (fsize < DEV_BSIZE)
268 				fatal(gettext("%s: bad frag size"), optarg);
269 			break;
270 
271 		case 'c':
272 			cpg = number("cpg", optarg, NR_NONE, 16);
273 			cpg_set = 1;
274 			if (cpg < 1)
275 				fatal(gettext("%s: bad cylinders/group"),
276 				    optarg);
277 			break;
278 
279 		case 'm':
280 			minfree = number("minfree", optarg, NR_PERCENT, 10);
281 			if (minfree < 0 || minfree > 99)
282 				fatal(gettext("%s: bad free space %%"), optarg);
283 			break;
284 
285 		case 'n':
286 			nrpos = number("nrpos", optarg, NR_NONE, 8);
287 			nrpos_set = 1;
288 			if (nrpos <= 0)
289 				fatal(gettext(
290 				    "%s: bad number of rotational positions"),
291 				    optarg);
292 			break;
293 
294 		case 'r':
295 			rpm = number("rpm", optarg, NR_NONE, 3600);
296 			rpm_set = 1;
297 			if (rpm < 0)
298 				fatal(gettext("%s: bad revs/minute"), optarg);
299 			break;
300 
301 		case 'i':
302 			/* xxx ought to test against fsize */
303 			density = number("nbpi", optarg, NR_NONE, 2048);
304 			if (density < DEV_BSIZE)
305 				fatal(gettext("%s: bad bytes per inode"),
306 				    optarg);
307 			break;
308 
309 		case 'T':
310 			Tflag++;
311 			break;
312 
313 		default:
314 			usage();
315 			fatal(gettext("-%c: unknown flag"), optopt);
316 		}
317 	}
318 
319 	/* At this point, there should only be one argument left:	*/
320 	/* The raw-special-device itself. If not, print usage message.	*/
321 	if ((argc - optind) != 1) {
322 		usage();
323 		exit(1);
324 	}
325 
326 	name = argv[optind];
327 
328 	special = getfullrawname(name);
329 	if (special == NULL) {
330 		(void) fprintf(stderr, gettext("newfs: malloc failed\n"));
331 		exit(1);
332 	}
333 
334 	if (*special == '\0') {
335 		if (strchr(name, '/') != NULL) {
336 			if (stat64(name, &st) < 0) {
337 				(void) fprintf(stderr,
338 				    gettext("newfs: %s: %s\n"),
339 				    name, strerror(errno));
340 				exit(2);
341 			}
342 			fatal(gettext("%s: not a raw disk device"), name);
343 		}
344 		(void) sprintf(device, "/dev/rdsk/%s", name);
345 		if ((special = getfullrawname(device)) == NULL) {
346 			(void) fprintf(stderr,
347 			    gettext("newfs: malloc failed\n"));
348 			exit(1);
349 		}
350 
351 		if (*special == '\0') {
352 			(void) sprintf(device, "/dev/%s", name);
353 			if ((special = getfullrawname(device)) == NULL) {
354 				(void) fprintf(stderr,
355 				    gettext("newfs: malloc failed\n"));
356 				exit(1);
357 			}
358 			if (*special == '\0')
359 				fatal(gettext(
360 				    "%s: not a raw disk device"), name);
361 		}
362 	}
363 
364 	/*
365 	 * getdiskbydev() determines the characteristics of the special
366 	 * device on which the file system will be built.  In the case
367 	 * of devices with SMI labels (that is, non-EFI labels), the
368 	 * following characteristics are set (if they were not already
369 	 * set on the command line, since the command line settings
370 	 * take precedence):
371 	 *
372 	 *	nsectors - sectors per track
373 	 *	ntracks - tracks per cylinder
374 	 *	rpm - disk revolutions per minute
375 	 *
376 	 *	apc is NOT set
377 	 *
378 	 * getdiskbydev() also sets the following quantities for all
379 	 * devices, if not already set:
380 	 *
381 	 *	bsize - file system block size
382 	 *	maxcontig
383 	 *	label_type (efi, vtoc, or other)
384 	 *
385 	 * getdiskbydev() returns the actual size of the device, in
386 	 * sectors.
387 	 */
388 
389 	actual_fssize = getdiskbydev(special);
390 
391 	if (req_fssize == 0) {
392 		fssize = actual_fssize;
393 	} else {
394 		/*
395 		 * If the user specified a size larger than what we've
396 		 * determined as the actual size of the device, see if the
397 		 * size specified by the user can be read.  If so, use it,
398 		 * since some devices and volume managers may not support
399 		 * the vtoc and EFI interfaces we use to determine device
400 		 * size.
401 		 */
402 		if (req_fssize > actual_fssize &&
403 		    validate_size(special, req_fssize)) {
404 			(void) fprintf(stderr, gettext(
405 "Warning: the requested size of this file system\n"
406 "(%lld sectors) is greater than the size of the\n"
407 "device reported by the driver (%lld sectors).\n"
408 "However, a read of the device at the requested size\n"
409 "does succeed, so the requested size will be used.\n"),
410 			    req_fssize, actual_fssize);
411 			fssize = req_fssize;
412 		} else {
413 			fssize = MIN(req_fssize, actual_fssize);
414 		}
415 	}
416 
417 	if (label_type == LABEL_TYPE_VTOC) {
418 		if (!use_efi_dflts) {
419 			if (nsectors < 0)
420 				fatal(gettext("%s: no default #sectors/track"),
421 				    special);
422 			if (ntracks < 0)
423 				fatal(gettext("%s: no default #tracks"),
424 				    special);
425 		}
426 		if (rpm < 0)
427 			fatal(gettext(
428 			    "%s: no default revolutions/minute value"),
429 			    special);
430 		if (rpm < 60) {
431 			(void) fprintf(stderr,
432 			    gettext("Warning: setting rpm to 60\n"));
433 			rpm = 60;
434 		}
435 	}
436 	if (label_type == LABEL_TYPE_EFI || label_type == LABEL_TYPE_OTHER) {
437 		if (ntracks_set)
438 			(void) fprintf(stderr, gettext(
439 "Warning: ntracks is obsolete for this device and will be ignored.\n"));
440 		if (cpg_set)
441 			(void) fprintf(stderr, gettext(
442 "Warning: cylinders/group is obsolete for this device and will be ignored.\n"));
443 		if (rpm_set)
444 			(void) fprintf(stderr, gettext(
445 "Warning: rpm is obsolete for this device and will be ignored.\n"));
446 		if (rot_set)
447 			(void) fprintf(stderr, gettext(
448 "Warning: rotational delay is obsolete for this device and"
449 " will be ignored.\n"));
450 		if (nrpos_set)
451 			(void) fprintf(stderr, gettext(
452 "Warning: number of rotational positions is obsolete for this device and\n"
453 "will be ignored.\n"));
454 		if (apc_set)
455 			(void) fprintf(stderr, gettext(
456 "Warning: number of alternate sectors per cylinder is obsolete for this\n"
457 "device and will be ignored.\n"));
458 
459 		/*
460 		 * We need these for the call to mkfs, even though they are
461 		 * meaningless.
462 		 */
463 		rpm = 60;
464 		nrpos = 1;
465 		apc = 0;
466 		rot = -1;
467 
468 		/*
469 		 * These values are set to produce a file system with
470 		 * a cylinder group size of 48MB.   For disks with
471 		 * non-EFI labels, most geometries result in cylinder
472 		 * groups of around 40 - 50 MB, so we arbitrarily choose
473 		 * 48MB for disks with EFI labels.  mkfs will reduce
474 		 * cylinders per group even further if necessary.
475 		 */
476 
477 		cpg = 16;
478 		nsectors = 128;
479 		ntracks = 48;
480 
481 		/*
482 		 * mkfs produces peculiar results for file systems
483 		 * that are smaller than one cylinder so don't allow
484 		 * them to be created (this check is only made for
485 		 * disks with EFI labels.  Eventually, it should probably
486 		 * be enforced for all disks.)
487 		 */
488 
489 		if (fssize < nsectors * ntracks) {
490 			fatal(gettext(
491 			    "file system size must be at least %d sectors"),
492 			    nsectors * ntracks);
493 		}
494 	}
495 
496 	if (fssize > INT_MAX)
497 		Tflag = 1;
498 
499 	/*
500 	 * If the user requested that the file system be set up for
501 	 * eventual growth to over a terabyte, or if it's already greater
502 	 * than a terabyte, set the inode density (nbpi) to MIN_MTB_DENSITY
503 	 * (unless the user has specified a larger nbpi), set the frag size
504 	 * equal to the block size, and set the cylinders-per-group value
505 	 * passed to mkfs to -1, which tells mkfs to make cylinder groups
506 	 * as large as possible.
507 	 */
508 	if (Tflag) {
509 		if (density < MIN_MTB_DENSITY)
510 			density = MIN_MTB_DENSITY;
511 		fsize = bsize;
512 		cpg = -1; 	/* says make cyl groups as big as possible */
513 	} else {
514 		if (fsize == 0)
515 			fsize = DESFRAGSIZE;
516 	}
517 
518 	if (!POWEROF2(fsize)) {
519 		(void) fprintf(stderr, gettext(
520 		    "newfs: fragment size must a power of 2, not %d\n"), fsize);
521 		fsize = bsize/8;
522 		(void) fprintf(stderr, gettext(
523 		    "newfs: fragsize reset to %ld\n"), fsize);
524 	}
525 
526 	/*
527 	 * The file system is limited in size by the fragment size.
528 	 * The number of fragments in the file system must fit into
529 	 * a signed 32-bit quantity, so the number of sectors in the
530 	 * file system is INT_MAX * the number of sectors in a frag.
531 	 */
532 
533 	max_possible_fssize = ((uint64_t)fsize)/DEV_BSIZE * INT_MAX;
534 	if (fssize > max_possible_fssize)
535 		fssize = max_possible_fssize;
536 
537 	/*
538 	 * Now fssize is the final size of the file system (in sectors).
539 	 * If it's less than what the user requested, print a message.
540 	 */
541 	if (fssize < req_fssize) {
542 		(void) fprintf(stderr, gettext(
543 		    "newfs: requested size of %s disk blocks is too large.\n"),
544 		    req_fssize_str);
545 		(void) fprintf(stderr, gettext(
546 		    "newfs: Resetting size to %lld\n"), fssize);
547 	}
548 
549 	/*
550 	 * fssize now equals the size (in sectors) of the file system
551 	 * that will be created.
552 	 */
553 
554 	/* XXX - following defaults are both here and in mkfs */
555 	if (density <= 0) {
556 		if (fssize < GBSEC)
557 			density = MINDENSITY;
558 		else
559 			density = (int)((((longlong_t)fssize + (GBSEC - 1)) /
560 						GBSEC) * MINDENSITY);
561 		if (density <= 0)
562 			density = MINDENSITY;
563 		if (density > MAXDEFDENSITY)
564 			density = MAXDEFDENSITY;
565 	}
566 	if (cpg == 0) {
567 		/*
568 		 * maxcpg calculation adapted from mkfs
569 		 * In the case of disks with EFI labels, cpg has
570 		 * already been set, so we won't enter this code.
571 		 */
572 		long maxcpg, maxipg;
573 
574 		maxipg = roundup(bsize * NBBY / 3,
575 		    bsize / sizeof (struct inode));
576 
577 		nsect = (nsectors == -1) ? geom_nsectors : nsectors;
578 
579 		maxcpg = (bsize - sizeof (struct cg) - howmany(maxipg, NBBY)) /
580 		    (sizeof (long) + nrpos * sizeof (short) +
581 			nsect / (MAXFRAG * NBBY));
582 		cpg = (fssize / GBSEC) * 32;
583 		if (cpg > maxcpg)
584 			cpg = maxcpg;
585 		if (cpg <= 0)
586 			cpg = MINCPG;
587 	}
588 	if (minfree < 0) {
589 		minfree = ((float)MINFREESEC / fssize) * 100;
590 		if (minfree > 10)
591 			minfree = 10;
592 		if (minfree <= 0)
593 			minfree = 1;
594 	}
595 #ifdef i386	/* Bug 1170182 */
596 	if (ntracks > 32 && (ntracks % 16) != 0) {
597 		ntracks -= (ntracks % 16);
598 	}
599 #endif
600 	/*
601 	 * Confirmation
602 	 */
603 	if (isatty(fileno(stdin)) && !Nflag) {
604 		/*
605 		 * If we can read a valid superblock, report the mount
606 		 * point on which this filesystem was last mounted.
607 		 */
608 		if (((sbp = read_sb(special)) != 0) &&
609 		    (*sbp->fs_fsmnt != '\0')) {
610 			(void) printf(gettext(
611 			    "newfs: %s last mounted as %s\n"),
612 			    special, sbp->fs_fsmnt);
613 		}
614 		(void) printf(gettext(
615 		    "newfs: construct a new file system %s: (y/n)? "),
616 		    special);
617 		(void) fflush(stdout);
618 		if (!yes())
619 			exit(0);
620 	}
621 	dprintf(("DeBuG newfs : nsect=%d ntrak=%d cpg=%d\n",
622 		nsectors, ntracks, cpg));
623 	/*
624 	 * If alternates-per-cylinder is ever implemented:
625 	 * need to get apc from dp->d_apc if no -a switch???
626 	 */
627 	(void) sprintf(cmd,
628 	"mkfs -F ufs %s%s%s%s %lld %d %d %d %d %d %d %d %d %s %d %d %d %d %s",
629 	    Nflag ? "-o N " : "", binary_sb ? "-o calcbinsb " : "",
630 	    text_sb ? "-o calcsb " : "", special,
631 	    fssize, nsectors, ntracks, bsize, fsize, cpg, minfree, rpm/60,
632 	    density, optim == FS_OPTSPACE ? "s" : "t", apc, rot, nrpos,
633 	    maxcontig, Tflag ? "y" : "n");
634 	if (verbose) {
635 		(void) printf("%s\n", cmd);
636 		(void) fflush(stdout);
637 	}
638 	exenv();
639 	if (status = system(cmd))
640 		exit(status >> 8);
641 	if (Nflag)
642 		exit(0);
643 	(void) sprintf(cmd, "/usr/sbin/fsirand %s", special);
644 	if (notrand(special) && (status = system(cmd)) != 0)
645 		(void) fprintf(stderr,
646 		    gettext("%s: failed, status = %d\n"),
647 		    cmd, status);
648 	return (0);
649 }
650 
651 static void
652 exenv(void)
653 {
654 	char *epath;				/* executable file path */
655 	char *cpath;				/* current path */
656 
657 	if ((cpath = getenv("PATH")) == NULL) {
658 		(void) fprintf(stderr, gettext("newfs: no PATH in env\n"));
659 		/*
660 		 * Background: the Bourne shell interpolates "." into
661 		 * the path where said path starts with a colon, ends
662 		 * with a colon, or has two adjacent colons.  Thus,
663 		 * the path ":/sbin::/usr/sbin:" is equivalent to
664 		 * ".:/sbin:.:/usr/sbin:.".  Now, we have no cpath,
665 		 * and epath ends in a colon (to make for easy
666 		 * catenation in the normal case).  By the above, if
667 		 * we use "", then "." becomes part of path.  That's
668 		 * bad, so use CPATH (which is just a duplicate of some
669 		 * element in EPATH).  No point in opening ourselves
670 		 * up to a Trojan horse attack when we don't have to....
671 		 */
672 		cpath = CPATH;
673 	}
674 	if ((epath = malloc(strlen(EPATH) + strlen(cpath) + 1)) == NULL) {
675 		(void) fprintf(stderr, gettext("newfs: malloc failed\n"));
676 		exit(1);
677 	}
678 	(void) strcpy(epath, EPATH);
679 	(void) strcat(epath, cpath);
680 	if (putenv(epath) < 0) {
681 		(void) fprintf(stderr, gettext("newfs: putenv failed\n"));
682 		exit(1);
683 	}
684 }
685 
686 static int
687 yes(void)
688 {
689 	int	i, b;
690 
691 	i = b = getchar();
692 	while (b != '\n' && b != '\0' && b != EOF)
693 		b = getchar();
694 	return (i == 'y');
695 }
696 
697 /*
698  * xxx Caller must run fmt through gettext(3) for us, if we ever
699  * xxx go the i18n route....
700  */
701 static void
702 fatal(char *fmt, ...)
703 {
704 	va_list pvar;
705 
706 	(void) fprintf(stderr, "newfs: ");
707 	va_start(pvar, fmt);
708 	(void) vfprintf(stderr, fmt, pvar);
709 	va_end(pvar);
710 	(void) putc('\n', stderr);
711 	exit(10);
712 }
713 
714 static diskaddr_t
715 getdiskbydev(char *disk)
716 {
717 	struct dk_geom g;
718 	struct dk_cinfo ci;
719 	diskaddr_t actual_size;
720 	int fd;
721 
722 	if ((fd = open64(disk, 0)) < 0) {
723 		perror(disk);
724 		exit(1);
725 	}
726 
727 	/*
728 	 * get_device_size() determines the actual size of the
729 	 * device, and also the disk's attributes, such as geometry.
730 	 */
731 	actual_size = get_device_size(fd, disk);
732 
733 	if (label_type == LABEL_TYPE_VTOC) {
734 		if (ioctl(fd, DKIOCGGEOM, &g))
735 			fatal(gettext(
736 			    "%s: Unable to read Disk geometry"), disk);
737 		dprintf(("DeBuG newfs : geom=%ld, CHSLIMIT=%d\n",
738 			g.dkg_ncyl * g.dkg_nhead * g.dkg_nsect, CHSLIMIT));
739 		if (((g.dkg_ncyl * g.dkg_nhead * g.dkg_nsect) > CHSLIMIT) &&
740 		    !Tflag) {
741 			dprintf(("DeBuG newfs : geom > CHSLIMIT\n"));
742 			use_efi_dflts = 1;
743 		}
744 		/*
745 		 * Though the nsector that is passed to mkfs is decided here
746 		 * based on use_efi_dflts, we still need the geometry
747 		 * based dkg_nsect for the cpg calculation later.
748 		 */
749 		geom_nsectors = g.dkg_nsect;
750 		if (nsectors == 0)
751 			nsectors = use_efi_dflts ? -1 : g.dkg_nsect;
752 		if (ntracks == 0)
753 			ntracks = use_efi_dflts ? -1 : g.dkg_nhead;
754 		if (rpm == 0)
755 			rpm = ((int)g.dkg_rpm <= 0) ? 3600: g.dkg_rpm;
756 	}
757 
758 	if (bsize == 0)
759 		bsize = DESBLKSIZE;
760 	/*
761 	 * Adjust maxcontig by the device's maxtransfer. If maxtransfer
762 	 * information is not available, default to the min of a MB and
763 	 * maxphys.
764 	 */
765 	if (maxcontig == -1 && ioctl(fd, DKIOCINFO, &ci) == 0) {
766 		maxcontig = ci.dki_maxtransfer * DEV_BSIZE;
767 		if (maxcontig < 0) {
768 			int	error, gotit, maxphys;
769 			gotit = fsgetmaxphys(&maxphys, &error);
770 
771 			/*
772 			 * If we cannot get the maxphys value, default
773 			 * to ufs_maxmaxphys (MB).
774 			 */
775 			if (gotit) {
776 				maxcontig = MIN(maxphys, MB);
777 			} else {
778 				(void) fprintf(stderr, gettext(
779 "Warning: Could not get system value for maxphys. The value for maxcontig\n"
780 "will default to 1MB.\n"));
781 			maxcontig = MB;
782 			}
783 		}
784 		maxcontig /= bsize;
785 	}
786 	(void) close(fd);
787 	return (actual_size);
788 }
789 
790 /*
791  * Figure out how big the partition we're dealing with is.
792  */
793 static diskaddr_t
794 get_device_size(int fd, char *name)
795 {
796 	struct vtoc vtoc;
797 	dk_gpt_t *efi_vtoc;
798 	diskaddr_t	slicesize;
799 
800 	int index = read_vtoc(fd, &vtoc);
801 
802 	if (index >= 0) {
803 		label_type = LABEL_TYPE_VTOC;
804 	} else {
805 		if (index == VT_ENOTSUP || index == VT_ERROR) {
806 			/* it might be an EFI label */
807 			index = efi_alloc_and_read(fd, &efi_vtoc);
808 			if (index >= 0)
809 				label_type = LABEL_TYPE_EFI;
810 		}
811 	}
812 
813 	if (index < 0) {
814 		/*
815 		 * Since both attempts to read the label failed, we're
816 		 * going to fall back to a brute force approach to
817 		 * determining the device's size:  see how far out we can
818 		 * perform reads on the device.
819 		 */
820 
821 		slicesize = brute_force_get_device_size(fd);
822 		if (slicesize == 0) {
823 			switch (index) {
824 			case VT_ERROR:
825 				(void) fprintf(stderr, gettext(
826 				    "newfs: %s: %s\n"), name, strerror(errno));
827 				exit(10);
828 				/*NOTREACHED*/
829 			case VT_EIO:
830 				fatal(gettext(
831 				    "%s: I/O error accessing VTOC"), name);
832 				/*NOTREACHED*/
833 			case VT_EINVAL:
834 				fatal(gettext(
835 				    "%s: Invalid field in VTOC"), name);
836 				/*NOTREACHED*/
837 			default:
838 				fatal(gettext(
839 				    "%s: unknown error accessing VTOC"),
840 				    name);
841 				/*NOTREACHED*/
842 			}
843 		} else {
844 			label_type = LABEL_TYPE_OTHER;
845 		}
846 	}
847 
848 	if (label_type == LABEL_TYPE_EFI) {
849 		slicesize = efi_vtoc->efi_parts[index].p_size;
850 		efi_free(efi_vtoc);
851 	} else if (label_type == LABEL_TYPE_VTOC) {
852 		/*
853 		 * In the vtoc struct, p_size is a 32-bit signed quantity.
854 		 * In the dk_gpt struct (efi's version of the vtoc), p_size
855 		 * is an unsigned 64-bit quantity.  By casting the vtoc's
856 		 * psize to an unsigned 32-bit quantity, it will be copied
857 		 * to 'slicesize' (an unsigned 64-bit diskaddr_t) without
858 		 * sign extension.
859 		 */
860 
861 		slicesize = (uint32_t)vtoc.v_part[index].p_size;
862 	}
863 
864 	return (slicesize);
865 }
866 
867 /*
868  * brute_force_get_device_size
869  *
870  * Determine the size of the device by seeing how far we can
871  * read.  Doing an llseek( , , SEEK_END) would probably work
872  * in most cases, but we've seen at least one third-party driver
873  * which doesn't correctly support the SEEK_END option when the
874  * the device is greater than a terabyte.
875  */
876 
877 static diskaddr_t
878 brute_force_get_device_size(int fd)
879 {
880 	diskaddr_t	min_fail = 0;
881 	diskaddr_t	max_succeed = 0;
882 	diskaddr_t	cur_db_off;
883 	char 		buf[DEV_BSIZE];
884 
885 	/*
886 	 * First, see if we can read the device at all, just to
887 	 * eliminate errors that have nothing to do with the
888 	 * device's size.
889 	 */
890 
891 	if (((llseek(fd, (offset_t)0, SEEK_SET)) == -1) ||
892 	    ((read(fd, buf, DEV_BSIZE)) == -1))
893 		return (0);  /* can't determine size */
894 
895 	/*
896 	 * Now, go sequentially through the multiples of 4TB
897 	 * to find the first read that fails (this isn't strictly
898 	 * the most efficient way to find the actual size if the
899 	 * size really could be anything between 0 and 2**64 bytes.
900 	 * We expect the sizes to be less than 16 TB for some time,
901 	 * so why do a bunch of reads that are larger than that?
902 	 * However, this algorithm *will* work for sizes of greater
903 	 * than 16 TB.  We're just not optimizing for those sizes.)
904 	 */
905 
906 	for (cur_db_off = SECTORS_PER_TERABYTE * 4;
907 	    min_fail == 0 && cur_db_off < FS_SIZE_UPPER_LIMIT;
908 	    cur_db_off += 4 * SECTORS_PER_TERABYTE) {
909 		if (((llseek(fd, (offset_t)(cur_db_off * DEV_BSIZE),
910 		    SEEK_SET)) == -1) ||
911 		    ((read(fd, buf, DEV_BSIZE)) != DEV_BSIZE))
912 			min_fail = cur_db_off;
913 		else
914 			max_succeed = cur_db_off;
915 	}
916 
917 	if (min_fail == 0)
918 		return (0);
919 
920 	/*
921 	 * We now know that the size of the device is less than
922 	 * min_fail and greater than or equal to max_succeed.  Now
923 	 * keep splitting the difference until the actual size in
924 	 * sectors in known.  We also know that the difference
925 	 * between max_succeed and min_fail at this time is
926 	 * 4 * SECTORS_PER_TERABYTE, which is a power of two, which
927 	 * simplifies the math below.
928 	 */
929 
930 	while (min_fail - max_succeed > 1) {
931 		cur_db_off = max_succeed + (min_fail - max_succeed)/2;
932 		if (((llseek(fd, (offset_t)(cur_db_off * DEV_BSIZE),
933 		    SEEK_SET)) == -1) ||
934 		    ((read(fd, buf, DEV_BSIZE)) != DEV_BSIZE))
935 			min_fail = cur_db_off;
936 		else
937 			max_succeed = cur_db_off;
938 	}
939 
940 	/* the size is the last successfully read sector offset plus one */
941 	return (max_succeed + 1);
942 }
943 
944 /*
945  * validate_size
946  *
947  * Return 1 if the device appears to be at least "size" sectors long.
948  * Return 0 if it's shorter or we can't read it.
949  */
950 
951 static int
952 validate_size(char *disk, diskaddr_t size)
953 {
954 	char 		buf[DEV_BSIZE];
955 	int fd, rc;
956 
957 	if ((fd = open64(disk, O_RDONLY)) < 0) {
958 		perror(disk);
959 		exit(1);
960 	}
961 
962 	if ((llseek(fd, (offset_t)((size - 1) * DEV_BSIZE), SEEK_SET) == -1) ||
963 	    (read(fd, buf, DEV_BSIZE)) != DEV_BSIZE)
964 		rc = 0;
965 	else
966 		rc = 1;
967 	(void) close(fd);
968 	return (rc);
969 }
970 
971 /*
972  * read_sb(char * rawdev) - Attempt to read the superblock from a raw device
973  *
974  * Returns:
975  *	0 :
976  *		Could not read a valid superblock for a variety of reasons.
977  *		Since 'newfs' handles any fatal conditions, we're not going
978  *		to make any guesses as to why this is failing or what should
979  *		be done about it.
980  *
981  *	struct fs *:
982  *		A pointer to (what we think is) a valid superblock. The
983  *		space for the superblock is static (inside the function)
984  *		since we will only be reading the values from it.
985  */
986 
987 struct fs *
988 read_sb(char *fsdev)
989 {
990 	static struct fs	sblock;
991 	struct stat64		statb;
992 	int			dskfd;
993 	char			*bufp = NULL;
994 	int			bufsz = 0;
995 
996 	if (stat64(fsdev, &statb) < 0)
997 		return (0);
998 
999 	if ((dskfd = open64(fsdev, O_RDONLY)) < 0)
1000 		return (0);
1001 
1002 	/*
1003 	 * We need a buffer whose size is a multiple of DEV_BSIZE in order
1004 	 * to read from a raw device (which we were probably passed).
1005 	 */
1006 	bufsz = ((sizeof (sblock) / DEV_BSIZE) + 1) * DEV_BSIZE;
1007 	if ((bufp = malloc(bufsz)) == NULL) {
1008 		(void) close(dskfd);
1009 		return (0);
1010 	}
1011 
1012 	if (llseek(dskfd, (offset_t)SBOFF, SEEK_SET) < 0 ||
1013 	    read(dskfd, bufp, bufsz) < 0) {
1014 		(void) close(dskfd);
1015 		free(bufp);
1016 		return (0);
1017 	}
1018 	(void) close(dskfd);	/* Done with the file */
1019 
1020 	(void) memcpy(&sblock, bufp, sizeof (sblock));
1021 	free(bufp);	/* Don't need this anymore */
1022 
1023 	if (((sblock.fs_magic != FS_MAGIC) &&
1024 	    (sblock.fs_magic != MTB_UFS_MAGIC)) ||
1025 	    sblock.fs_ncg < 1 || sblock.fs_cpg < 1)
1026 		return (0);
1027 
1028 	if (sblock.fs_ncg * sblock.fs_cpg < sblock.fs_ncyl ||
1029 	    (sblock.fs_ncg - 1) * sblock.fs_cpg >= sblock.fs_ncyl)
1030 		return (0);
1031 
1032 	if (sblock.fs_sbsize < 0 || sblock.fs_sbsize > SBSIZE)
1033 		return (0);
1034 
1035 	return (&sblock);
1036 }
1037 
1038 /*
1039  * Read the UFS file system on the raw device SPECIAL.  If it does not
1040  * appear to be a UFS file system, return non-zero, indicating that
1041  * fsirand should be called (and it will spit out an error message).
1042  * If it is a UFS file system, take a look at the inodes in the first
1043  * cylinder group.  If they appear to be randomized (non-zero), return
1044  * zero, which will cause fsirand to not be called.  If the inode generation
1045  * counts are all zero, then we must call fsirand, so return non-zero.
1046  */
1047 
1048 #define	RANDOMIZED	0
1049 #define	NOT_RANDOMIZED	1
1050 
1051 static int
1052 notrand(char *special)
1053 {
1054 	long fsbuf[SBSIZE / sizeof (long)];
1055 	struct dinode dibuf[MAXBSIZE/sizeof (struct dinode)];
1056 	struct fs *fs;
1057 	struct dinode *dip;
1058 	offset_t seekaddr;
1059 	int bno, inum;
1060 	int fd;
1061 
1062 	fs = (struct fs *)fsbuf;
1063 	if ((fd = open64(special, 0)) == -1)
1064 		return (NOT_RANDOMIZED);
1065 	if (llseek(fd, (offset_t)SBLOCK * DEV_BSIZE, 0) == -1 ||
1066 	    read(fd, (char *)fs, SBSIZE) != SBSIZE ||
1067 	    ((fs->fs_magic != FS_MAGIC) && (fs->fs_magic != MTB_UFS_MAGIC))) {
1068 		(void) close(fd);
1069 		return (NOT_RANDOMIZED);
1070 	}
1071 
1072 	/* looks like a UFS file system; read the first cylinder group */
1073 	bsize = INOPB(fs) * sizeof (struct dinode);
1074 	inum = 0;
1075 	while (inum < fs->fs_ipg) {
1076 		bno = itod(fs, inum);
1077 		seekaddr = (offset_t)fsbtodb(fs, bno) * DEV_BSIZE;
1078 		if (llseek(fd, seekaddr, 0) == -1 ||
1079 		    read(fd, (char *)dibuf, bsize) != bsize) {
1080 			(void) close(fd);
1081 			return (NOT_RANDOMIZED);
1082 		}
1083 		for (dip = dibuf; dip < &dibuf[INOPB(fs)]; dip++) {
1084 			if (dip->di_gen != 0) {
1085 				(void) close(fd);
1086 				return (RANDOMIZED);
1087 			}
1088 			inum++;
1089 		}
1090 	}
1091 	(void) close(fd);
1092 	return (NOT_RANDOMIZED);
1093 }
1094 
1095 static void
1096 usage(void)
1097 {
1098 	(void) fprintf(stderr, gettext(
1099 	    "usage: newfs [ -v ] [ mkfs-options ] raw-special-device\n"));
1100 	(void) fprintf(stderr, gettext("where mkfs-options are:\n"));
1101 	(void) fprintf(stderr, gettext(
1102 	    "\t-N do not create file system, just print out parameters\n"));
1103 	(void) fprintf(stderr, gettext(
1104 "\t-T configure file system for eventual growth to over a terabyte\n"));
1105 	(void) fprintf(stderr, gettext("\t-s file system size (sectors)\n"));
1106 	(void) fprintf(stderr, gettext("\t-b block size\n"));
1107 	(void) fprintf(stderr, gettext("\t-f frag size\n"));
1108 	(void) fprintf(stderr, gettext("\t-t tracks/cylinder\n"));
1109 	(void) fprintf(stderr, gettext("\t-c cylinders/group\n"));
1110 	(void) fprintf(stderr, gettext("\t-m minimum free space %%\n"));
1111 	(void) fprintf(stderr, gettext(
1112 	    "\t-o optimization preference (`space' or `time')\n"));
1113 	(void) fprintf(stderr, gettext("\t-r revolutions/minute\n"));
1114 	(void) fprintf(stderr, gettext("\t-i number of bytes per inode\n"));
1115 	(void) fprintf(stderr, gettext(
1116 	    "\t-a number of alternates per cylinder\n"));
1117 	(void) fprintf(stderr, gettext("\t-C maxcontig\n"));
1118 	(void) fprintf(stderr, gettext("\t-d rotational delay\n"));
1119 	(void) fprintf(stderr, gettext(
1120 	    "\t-n number of rotational positions\n"));
1121 	(void) fprintf(stderr, gettext(
1122 "\t-S print a textual version of the calculated superblock to stdout\n"));
1123 	(void) fprintf(stderr, gettext(
1124 "\t-B dump a binary version of the calculated superblock to stdout\n"));
1125 }
1126 
1127 /*
1128  * Error-detecting version of atoi(3).  Adapted from mkfs' number().
1129  */
1130 static unsigned int
1131 number(char *param, char *value, int flags, int def_value)
1132 {
1133 	char *cs;
1134 	int n;
1135 	int cut = INT_MAX / 10;    /* limit to avoid overflow */
1136 	int minus = 0;
1137 
1138 	cs = value;
1139 	if (*cs == '-') {
1140 		minus = 1;
1141 		cs += 1;
1142 	}
1143 	if ((*cs < '0') || (*cs > '9')) {
1144 		goto bail_out;
1145 	}
1146 	n = 0;
1147 	while ((*cs >= '0') && (*cs <= '9') && (n <= cut)) {
1148 		n = n*10 + *cs++ - '0';
1149 	}
1150 	if (minus)
1151 	    n = -n;
1152 	for (;;) {
1153 		switch (*cs++) {
1154 		case '\0':
1155 			return (n);
1156 
1157 		case '0': case '1': case '2': case '3': case '4':
1158 		case '5': case '6': case '7': case '8': case '9':
1159 			(void) fprintf(stderr, gettext(
1160 			    "newfs: value for %s overflowed, using %d\n"),
1161 			    param, def_value);
1162 			return (def_value);
1163 
1164 		case '%':
1165 			if (flags & NR_PERCENT)
1166 				break;
1167 			/* FALLTHROUGH */
1168 
1169 		default:
1170 bail_out:
1171 			fatal(gettext("bad numeric arg for %s: \"%s\""),
1172 			    param, value);
1173 
1174 		}
1175 	}
1176 	/* NOTREACHED */
1177 }
1178 
1179 /*
1180  * Error-detecting version of atoi(3).  Adapted from mkfs' number().
1181  */
1182 static int64_t
1183 number64(char *param, char *value, int flags, int64_t def_value)
1184 {
1185 	char *cs;
1186 	int64_t n;
1187 	int64_t cut = FS_SIZE_UPPER_LIMIT/ 10;    /* limit to avoid overflow */
1188 	int minus = 0;
1189 
1190 	cs = value;
1191 	if (*cs == '-') {
1192 		minus = 1;
1193 		cs += 1;
1194 	}
1195 	if ((*cs < '0') || (*cs > '9')) {
1196 		goto bail_out;
1197 	}
1198 	n = 0;
1199 	while ((*cs >= '0') && (*cs <= '9') && (n <= cut)) {
1200 		n = n*10 + *cs++ - '0';
1201 	}
1202 	if (minus)
1203 	    n = -n;
1204 	for (;;) {
1205 		switch (*cs++) {
1206 		case '\0':
1207 			return (n);
1208 
1209 		case '0': case '1': case '2': case '3': case '4':
1210 		case '5': case '6': case '7': case '8': case '9':
1211 			(void) fprintf(stderr, gettext(
1212 			    "newfs: value for %s overflowed, using %d\n"),
1213 			    param, def_value);
1214 			return (def_value);
1215 
1216 		case '%':
1217 			if (flags & NR_PERCENT)
1218 				break;
1219 			/* FALLTHROUGH */
1220 
1221 		default:
1222 bail_out:
1223 			fatal(gettext("bad numeric arg for %s: \"%s\""),
1224 			    param, value);
1225 
1226 		}
1227 	}
1228 	/* NOTREACHED */
1229 }
1230