xref: /illumos-gate/usr/src/cmd/fs.d/ufs/newfs/newfs.c (revision db60a39d)
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 #pragma ident	"%Z%%M%	%I%	%E% SMI"
23 	/* from UCB 5.2 9/11/85 */
24 
25 /*
26  * newfs: friendly front end to mkfs
27  *
28  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
29  * Use is subject to license terms.
30  */
31 
32 #include <sys/param.h>
33 #include <sys/types.h>
34 #include <locale.h>
35 #include <sys/stat.h>
36 #include <sys/buf.h>
37 #include <sys/fs/ufs_fs.h>
38 #include <sys/vnode.h>
39 #include <sys/fs/ufs_inode.h>
40 #include <sys/sysmacros.h>
41 
42 #include <errno.h>
43 #include <stdio.h>
44 #include <string.h>
45 #include <stdlib.h>
46 #include <stdarg.h>
47 #include <stdio.h>
48 #include <fcntl.h>
49 #include <unistd.h>
50 #include <limits.h>
51 #include <libintl.h>
52 #include <sys/dkio.h>
53 #include <sys/vtoc.h>
54 #include <sys/mkdev.h>
55 #include <sys/efi_partition.h>
56 
57 #include <fslib.h>
58 
59 static unsigned int number(char *, char *, int, int);
60 static int64_t number64(char *, char *, int, int64_t);
61 static diskaddr_t getdiskbydev(char *);
62 static int  yes(void);
63 static int  notrand(char *);
64 static void usage();
65 static diskaddr_t get_device_size(int, char *);
66 static diskaddr_t brute_force_get_device_size(int);
67 static int validate_size(char *disk, diskaddr_t size);
68 static void exenv(void);
69 static struct fs *read_sb(char *);
70 /*PRINTFLIKE1*/
71 static void fatal(char *fmt, ...);
72 
73 #define	EPATH "PATH=/usr/sbin:/sbin:"
74 #define	CPATH "/sbin"					/* an EPATH element */
75 #define	MB (1024 * 1024)
76 #define	GBSEC ((1024 * 1024 * 1024) / DEV_BSIZE)	/* sectors in a GB */
77 #define	MINFREESEC ((64 * 1024 * 1024) / DEV_BSIZE)	/* sectors in 64 MB */
78 #define	MINCPG (16)	/* traditional */
79 #define	MAXDEFDENSITY (8 * 1024)	/* arbitrary */
80 #define	MINDENSITY (2 * 1024)	/* traditional */
81 #define	MIN_MTB_DENSITY (1024 * 1024)
82 #define	POWEROF2(num)	(((num) & ((num) - 1)) == 0)
83 #define	SECTORS_PER_TERABYTE	(1LL << 31)
84 /*
85  * The following constant specifies an upper limit for file system size
86  * that is actually a lot bigger than we expect to support with UFS. (Since
87  * it's specified in sectors, the file system size would be 2**44 * 512,
88  * which is 2**53, which is 8192 Terabytes.)  However, it's useful
89  * for checking the basic sanity of a size value that is input on the
90  * command line.
91  */
92 #define	FS_SIZE_UPPER_LIMIT	0x100000000000LL
93 
94 /* For use with number() */
95 #define	NR_NONE		0
96 #define	NR_PERCENT	0x01
97 
98 /*
99  * The following two constants set the default block and fragment sizes.
100  * Both constants must be a power of 2 and meet the following constraints:
101  *	MINBSIZE <= DESBLKSIZE <= MAXBSIZE
102  *	DEV_BSIZE <= DESFRAGSIZE <= DESBLKSIZE
103  *	DESBLKSIZE / DESFRAGSIZE <= 8
104  */
105 #define	DESBLKSIZE	8192
106 #define	DESFRAGSIZE	1024
107 
108 #ifdef DEBUG
109 #define	dprintf(x)	printf x
110 #else
111 #define	dprintf(x)
112 #endif
113 
114 static int	Nflag;		/* run mkfs without writing file system */
115 static int	Tflag;		/* set up file system for growth to over 1 TB */
116 static int	verbose;	/* show mkfs line before exec */
117 static int	fsize = 0;		/* fragment size */
118 static int	fsize_flag = 0;	/* fragment size was specified on cmd line */
119 static int	bsize;		/* block size */
120 static int	ntracks;	/* # tracks/cylinder */
121 static int	ntracks_set = 0; /* true if the user specified ntracks */
122 static int	optim = FS_OPTTIME;	/* optimization, t(ime) or s(pace) */
123 static int	nsectors;	/* # sectors/track */
124 static int	cpg;		/* cylinders/cylinder group */
125 static int	cpg_set = 0;	/* true if the user specified cpg */
126 static int	minfree = -1;	/* free space threshold */
127 static int	rpm;		/* revolutions/minute of drive */
128 static int	rpm_set = 0;	/* true if the user specified rpm */
129 static int	nrpos = 8;	/* # of distinguished rotational positions */
130 				/* 8 is the historical default */
131 static int	nrpos_set = 0;	/* true if the user specified nrpos */
132 static int	density = 0;	/* number of bytes per inode */
133 static int	apc;		/* alternates per cylinder */
134 static int	apc_set = 0;	/* true if the user specified apc */
135 static int 	rot = -1;	/* rotational delay (msecs) */
136 static int	rot_set = 0;	/* true if the user specified rot */
137 static int 	maxcontig = -1;	/* maximum number of contig blocks */
138 static int	text_sb = 0;	/* no disk changes; just final sb text dump */
139 static int	binary_sb = 0;	/* no disk changes; just final sb binary dump */
140 static int	label_type;	/* see types below */
141 
142 /*
143  * The variable use_efi_dflts is an indicator of whether to use EFI logic
144  * or the geometry logic in laying out the filesystem. This is decided
145  * based on the size/type of the disk and is used only for non-EFI labeled
146  * disks and removable media.
147  */
148 static int	use_efi_dflts = 0;
149 static int	isremovable = 0;
150 
151 static char	device[MAXPATHLEN];
152 static char	cmd[BUFSIZ];
153 
154 extern	char	*getfullrawname(); /* from libadm */
155 
156 int
157 main(int argc, char *argv[])
158 {
159 	char *special, *name;
160 	struct stat64 st;
161 	int status;
162 	int option;
163 	struct fs *sbp;	/* Pointer to superblock (if present) */
164 	diskaddr_t actual_fssize;
165 	diskaddr_t max_possible_fssize;
166 	diskaddr_t req_fssize = 0;
167 	diskaddr_t fssize = 0;
168 	char	*req_fssize_str = NULL; /* requested size argument */
169 
170 	(void) setlocale(LC_ALL, "");
171 
172 #if !defined(TEXT_DOMAIN)
173 #define	TEXT_DOMAIN	"SYS_TEST"
174 #endif
175 	(void) textdomain(TEXT_DOMAIN);
176 
177 	opterr = 0;	/* We print our own errors, disable getopt's message */
178 	while ((option = getopt(argc, argv,
179 	    "vNBSs:C:d:t:o:a:b:f:c:m:n:r:i:T")) != EOF) {
180 		switch (option) {
181 		case 'S':
182 			text_sb++;
183 			break;
184 		case 'B':
185 			binary_sb++;
186 			break;
187 		case 'v':
188 			verbose++;
189 			break;
190 
191 		case 'N':
192 			Nflag++;
193 			break;
194 
195 		case 's':
196 			/*
197 			 * The maximum file system size is a lot smaller
198 			 * than FS_SIZE_UPPER_LIMIT, but until we find out
199 			 * the device size and block size, we don't know
200 			 * what it is.  So save the requested size in a
201 			 * string so that we can print it out later if we
202 			 * determine it's too big.
203 			 */
204 			req_fssize = number64("fssize", optarg, NR_NONE,
205 			    FS_SIZE_UPPER_LIMIT);
206 			if (req_fssize < 1024)
207 				fatal(gettext(
208 				    "%s: fssize must be at least 1024"),
209 				    optarg);
210 			req_fssize_str = strdup(optarg);
211 			if (req_fssize_str == NULL)
212 				fatal(gettext(
213 				    "Insufficient memory for string copy."));
214 			break;
215 
216 		case 'C':
217 			maxcontig = number("maxcontig", optarg, NR_NONE, -1);
218 			if (maxcontig < 0)
219 				fatal(gettext("%s: bad maxcontig"), optarg);
220 			break;
221 
222 		case 'd':
223 			rot = number("rotdelay", optarg, NR_NONE, 0);
224 			rot_set = 1;
225 			if (rot < 0 || rot > 1000)
226 				fatal(gettext(
227 				    "%s: bad rotational delay"), optarg);
228 			break;
229 
230 		case 't':
231 			ntracks = number("ntrack", optarg, NR_NONE, 16);
232 			ntracks_set = 1;
233 			if ((ntracks < 0) ||
234 			    (ntracks > INT_MAX))
235 				fatal(gettext("%s: bad total tracks"), optarg);
236 			break;
237 
238 		case 'o':
239 			if (strcmp(optarg, "space") == 0)
240 				optim = FS_OPTSPACE;
241 			else if (strcmp(optarg, "time") == 0)
242 				optim = FS_OPTTIME;
243 			else
244 				fatal(gettext(
245 "%s: bad optimization preference (options are `space' or `time')"), optarg);
246 			break;
247 
248 		case 'a':
249 			apc = number("apc", optarg, NR_NONE, 0);
250 			apc_set = 1;
251 			if (apc < 0 || apc > 32768) /* see mkfs.c */
252 				fatal(gettext(
253 				    "%s: bad alternates per cyl"), optarg);
254 			break;
255 
256 		case 'b':
257 			bsize = number("bsize", optarg, NR_NONE, DESBLKSIZE);
258 			if (bsize < MINBSIZE || bsize > MAXBSIZE)
259 				fatal(gettext(
260 				    "%s: bad block size"), optarg);
261 			break;
262 
263 		case 'f':
264 			fsize = number("fragsize", optarg, NR_NONE,
265 			    DESFRAGSIZE);
266 			fsize_flag++;
267 			/* xxx ought to test against bsize for upper limit */
268 			if (fsize < DEV_BSIZE)
269 				fatal(gettext("%s: bad frag size"), optarg);
270 			break;
271 
272 		case 'c':
273 			cpg = number("cpg", optarg, NR_NONE, 16);
274 			cpg_set = 1;
275 			if (cpg < 1)
276 				fatal(gettext("%s: bad cylinders/group"),
277 				    optarg);
278 			break;
279 
280 		case 'm':
281 			minfree = number("minfree", optarg, NR_PERCENT, 10);
282 			if (minfree < 0 || minfree > 99)
283 				fatal(gettext("%s: bad free space %%"), optarg);
284 			break;
285 
286 		case 'n':
287 			nrpos = number("nrpos", optarg, NR_NONE, 8);
288 			nrpos_set = 1;
289 			if (nrpos <= 0)
290 				fatal(gettext(
291 				    "%s: bad number of rotational positions"),
292 				    optarg);
293 			break;
294 
295 		case 'r':
296 			rpm = number("rpm", optarg, NR_NONE, 3600);
297 			rpm_set = 1;
298 			if (rpm < 0)
299 				fatal(gettext("%s: bad revs/minute"), optarg);
300 			break;
301 
302 		case 'i':
303 			/* xxx ought to test against fsize */
304 			density = number("nbpi", optarg, NR_NONE, 2048);
305 			if (density < DEV_BSIZE)
306 				fatal(gettext("%s: bad bytes per inode"),
307 				    optarg);
308 			break;
309 
310 		case 'T':
311 			Tflag++;
312 			break;
313 
314 		default:
315 			usage();
316 			fatal(gettext("-%c: unknown flag"), optopt);
317 		}
318 	}
319 
320 	/* At this point, there should only be one argument left:	*/
321 	/* The raw-special-device itself. If not, print usage message.	*/
322 	if ((argc - optind) != 1) {
323 		usage();
324 		exit(1);
325 	}
326 
327 	name = argv[optind];
328 
329 	special = getfullrawname(name);
330 	if (special == NULL) {
331 		(void) fprintf(stderr, gettext("newfs: malloc failed\n"));
332 		exit(1);
333 	}
334 
335 	if (*special == '\0') {
336 		if (strchr(name, '/') != NULL) {
337 			if (stat64(name, &st) < 0) {
338 				(void) fprintf(stderr,
339 				    gettext("newfs: %s: %s\n"),
340 				    name, strerror(errno));
341 				exit(2);
342 			}
343 			fatal(gettext("%s: not a raw disk device"), name);
344 		}
345 		(void) snprintf(device, sizeof (device), "/dev/rdsk/%s", name);
346 		if ((special = getfullrawname(device)) == NULL) {
347 			(void) fprintf(stderr,
348 			    gettext("newfs: malloc failed\n"));
349 			exit(1);
350 		}
351 
352 		if (*special == '\0') {
353 			(void) snprintf(device, sizeof (device), "/dev/%s",
354 			    name);
355 			if ((special = getfullrawname(device)) == NULL) {
356 				(void) fprintf(stderr,
357 				    gettext("newfs: malloc failed\n"));
358 				exit(1);
359 			}
360 			if (*special == '\0')
361 				fatal(gettext(
362 				    "%s: not a raw disk device"), name);
363 		}
364 	}
365 
366 	/*
367 	 * getdiskbydev() determines the characteristics of the special
368 	 * device on which the file system will be built.  In the case
369 	 * of devices with SMI labels (that is, non-EFI labels), the
370 	 * following characteristics are set (if they were not already
371 	 * set on the command line, since the command line settings
372 	 * take precedence):
373 	 *
374 	 *	nsectors - sectors per track
375 	 *	ntracks - tracks per cylinder
376 	 *	rpm - disk revolutions per minute
377 	 *
378 	 *	apc is NOT set
379 	 *
380 	 * getdiskbydev() also sets the following quantities for all
381 	 * devices, if not already set:
382 	 *
383 	 *	bsize - file system block size
384 	 *	maxcontig
385 	 *	label_type (efi, vtoc, or other)
386 	 *
387 	 * getdiskbydev() returns the actual size of the device, in
388 	 * sectors.
389 	 */
390 
391 	actual_fssize = getdiskbydev(special);
392 
393 	if (req_fssize == 0) {
394 		fssize = actual_fssize;
395 	} else {
396 		/*
397 		 * If the user specified a size larger than what we've
398 		 * determined as the actual size of the device, see if the
399 		 * size specified by the user can be read.  If so, use it,
400 		 * since some devices and volume managers may not support
401 		 * the vtoc and EFI interfaces we use to determine device
402 		 * size.
403 		 */
404 		if (req_fssize > actual_fssize &&
405 		    validate_size(special, req_fssize)) {
406 			(void) fprintf(stderr, gettext(
407 "Warning: the requested size of this file system\n"
408 "(%lld sectors) is greater than the size of the\n"
409 "device reported by the driver (%lld sectors).\n"
410 "However, a read of the device at the requested size\n"
411 "does succeed, so the requested size will be used.\n"),
412 			    req_fssize, actual_fssize);
413 			fssize = req_fssize;
414 		} else {
415 			fssize = MIN(req_fssize, actual_fssize);
416 		}
417 	}
418 
419 	if (label_type == LABEL_TYPE_VTOC) {
420 		if (nsectors < 0)
421 			fatal(gettext("%s: no default #sectors/track"),
422 			    special);
423 		if (!use_efi_dflts) {
424 			if (ntracks < 0)
425 				fatal(gettext("%s: no default #tracks"),
426 				    special);
427 		}
428 		if (rpm < 0)
429 			fatal(gettext(
430 			    "%s: no default revolutions/minute value"),
431 			    special);
432 		if (rpm < 60) {
433 			(void) fprintf(stderr,
434 			    gettext("Warning: setting rpm to 60\n"));
435 			rpm = 60;
436 		}
437 	}
438 	if (label_type == LABEL_TYPE_EFI || label_type == LABEL_TYPE_OTHER) {
439 		if (ntracks_set)
440 			(void) fprintf(stderr, gettext(
441 "Warning: ntracks is obsolete for this device and will be ignored.\n"));
442 		if (cpg_set)
443 			(void) fprintf(stderr, gettext(
444 "Warning: cylinders/group is obsolete for this device and will be ignored.\n"));
445 		if (rpm_set)
446 			(void) fprintf(stderr, gettext(
447 "Warning: rpm is obsolete for this device and will be ignored.\n"));
448 		if (rot_set)
449 			(void) fprintf(stderr, gettext(
450 "Warning: rotational delay is obsolete for this device and"
451 " will be ignored.\n"));
452 		if (nrpos_set)
453 			(void) fprintf(stderr, gettext(
454 "Warning: number of rotational positions is obsolete for this device and\n"
455 "will be ignored.\n"));
456 		if (apc_set)
457 			(void) fprintf(stderr, gettext(
458 "Warning: number of alternate sectors per cylinder is obsolete for this\n"
459 "device and will be ignored.\n"));
460 
461 		/*
462 		 * We need these for the call to mkfs, even though they are
463 		 * meaningless.
464 		 */
465 		rpm = 60;
466 		nrpos = 1;
467 		apc = 0;
468 		rot = -1;
469 
470 		/*
471 		 * These values are set to produce a file system with
472 		 * a cylinder group size of 48MB.   For disks with
473 		 * non-EFI labels, most geometries result in cylinder
474 		 * groups of around 40 - 50 MB, so we arbitrarily choose
475 		 * 48MB for disks with EFI labels.  mkfs will reduce
476 		 * cylinders per group even further if necessary.
477 		 */
478 
479 		cpg = 16;
480 		nsectors = 128;
481 		ntracks = 48;
482 
483 		/*
484 		 * mkfs produces peculiar results for file systems
485 		 * that are smaller than one cylinder so don't allow
486 		 * them to be created (this check is only made for
487 		 * disks with EFI labels.  Eventually, it should probably
488 		 * be enforced for all disks.)
489 		 */
490 
491 		if (fssize < nsectors * ntracks) {
492 			fatal(gettext(
493 			    "file system size must be at least %d sectors"),
494 			    nsectors * ntracks);
495 		}
496 	}
497 
498 	if (fssize > INT_MAX)
499 		Tflag = 1;
500 
501 	/*
502 	 * If the user requested that the file system be set up for
503 	 * eventual growth to over a terabyte, or if it's already greater
504 	 * than a terabyte, set the inode density (nbpi) to MIN_MTB_DENSITY
505 	 * (unless the user has specified a larger nbpi), set the frag size
506 	 * equal to the block size, and set the cylinders-per-group value
507 	 * passed to mkfs to -1, which tells mkfs to make cylinder groups
508 	 * as large as possible.
509 	 */
510 	if (Tflag) {
511 		if (density < MIN_MTB_DENSITY)
512 			density = MIN_MTB_DENSITY;
513 		fsize = bsize;
514 		cpg = -1; 	/* says make cyl groups as big as possible */
515 	} else {
516 		if (fsize == 0)
517 			fsize = DESFRAGSIZE;
518 	}
519 
520 	if (!POWEROF2(fsize)) {
521 		(void) fprintf(stderr, gettext(
522 		    "newfs: fragment size must a power of 2, not %d\n"), fsize);
523 		fsize = bsize/8;
524 		(void) fprintf(stderr, gettext(
525 		    "newfs: fragsize reset to %ld\n"), fsize);
526 	}
527 
528 	/*
529 	 * The file system is limited in size by the fragment size.
530 	 * The number of fragments in the file system must fit into
531 	 * a signed 32-bit quantity, so the number of sectors in the
532 	 * file system is INT_MAX * the number of sectors in a frag.
533 	 */
534 
535 	max_possible_fssize = ((uint64_t)fsize)/DEV_BSIZE * INT_MAX;
536 	if (fssize > max_possible_fssize)
537 		fssize = max_possible_fssize;
538 
539 	/*
540 	 * Now fssize is the final size of the file system (in sectors).
541 	 * If it's less than what the user requested, print a message.
542 	 */
543 	if (fssize < req_fssize) {
544 		(void) fprintf(stderr, gettext(
545 		    "newfs: requested size of %s disk blocks is too large.\n"),
546 		    req_fssize_str);
547 		(void) fprintf(stderr, gettext(
548 		    "newfs: Resetting size to %lld\n"), fssize);
549 	}
550 
551 	/*
552 	 * fssize now equals the size (in sectors) of the file system
553 	 * that will be created.
554 	 */
555 
556 	/* XXX - following defaults are both here and in mkfs */
557 	if (density <= 0) {
558 		if (fssize < GBSEC)
559 			density = MINDENSITY;
560 		else
561 			density = (int)((((longlong_t)fssize + (GBSEC - 1)) /
562 			    GBSEC) * MINDENSITY);
563 		if (density <= 0)
564 			density = MINDENSITY;
565 		if (density > MAXDEFDENSITY)
566 			density = MAXDEFDENSITY;
567 	}
568 	if (cpg == 0) {
569 		/*
570 		 * maxcpg calculation adapted from mkfs
571 		 * In the case of disks with EFI labels, cpg has
572 		 * already been set, so we won't enter this code.
573 		 */
574 		long maxcpg, maxipg;
575 
576 		maxipg = roundup(bsize * NBBY / 3,
577 		    bsize / sizeof (struct inode));
578 		maxcpg = (bsize - sizeof (struct cg) - howmany(maxipg, NBBY)) /
579 		    (sizeof (long) + nrpos * sizeof (short) +
580 		    nsectors / (MAXFRAG * NBBY));
581 		cpg = (fssize / GBSEC) * 32;
582 		if (cpg > maxcpg)
583 			cpg = maxcpg;
584 		if (cpg <= 0)
585 			cpg = MINCPG;
586 	}
587 	if (minfree < 0) {
588 		minfree = (int)(((float)MINFREESEC / fssize) * 100);
589 		if (minfree > 10)
590 			minfree = 10;
591 		if (minfree <= 0)
592 			minfree = 1;
593 	}
594 #ifdef i386	/* Bug 1170182 */
595 	if (ntracks > 32 && (ntracks % 16) != 0) {
596 		ntracks -= (ntracks % 16);
597 	}
598 #endif
599 	/*
600 	 * Confirmation
601 	 */
602 	if (isatty(fileno(stdin)) && !Nflag) {
603 		/*
604 		 * If we can read a valid superblock, report the mount
605 		 * point on which this filesystem was last mounted.
606 		 */
607 		if (((sbp = read_sb(special)) != 0) &&
608 		    (*sbp->fs_fsmnt != '\0')) {
609 			(void) printf(gettext(
610 			    "newfs: %s last mounted as %s\n"),
611 			    special, sbp->fs_fsmnt);
612 		}
613 		(void) printf(gettext(
614 		    "newfs: construct a new file system %s: (y/n)? "),
615 		    special);
616 		(void) fflush(stdout);
617 		if (!yes())
618 			exit(0);
619 	}
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) snprintf(cmd, sizeof (cmd), "pfexec mkfs -F ufs "
628 	    "%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) snprintf(cmd, sizeof (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 		/*
735 		 * Geometry information does not make sense for removable media
736 		 * anyway, so indicate mkfs to use EFI default parameters.
737 		 */
738 		if (ioctl(fd, DKIOCREMOVABLE, &isremovable)) {
739 			dprintf(("DeBuG newfs : Unable to determine if %s is"
740 			    " Removable Media. Proceeding with system"
741 			    " determined parameters.\n", disk));
742 			isremovable = 0;
743 		} else if (isremovable && !Tflag)
744 			use_efi_dflts = 1;
745 
746 		if (ioctl(fd, DKIOCGGEOM, &g))
747 			fatal(gettext(
748 			    "%s: Unable to read Disk geometry"), disk);
749 		if (((g.dkg_ncyl * g.dkg_nhead * g.dkg_nsect) > CHSLIMIT) &&
750 		    !Tflag) {
751 			use_efi_dflts = 1;
752 		}
753 		dprintf(("DeBuG newfs : geom=%ld, CHSLIMIT=%d "
754 		    "isremovable = %d use_efi_dflts = %d\n",
755 		    g.dkg_ncyl * g.dkg_nhead * g.dkg_nsect, CHSLIMIT,
756 		    isremovable, use_efi_dflts));
757 		/*
758 		 * The ntracks that is passed to mkfs is decided here based
759 		 * on 'use_efi_dflts' and whether ntracks was specified as a
760 		 * command line parameter to newfs.
761 		 * If ntracks of -1 is passed to mkfs, mkfs uses DEF_TRACKS_EFI
762 		 * and DEF_SECTORS_EFI for ntracks and nsectors respectively.
763 		 */
764 		if (nsectors == 0)
765 			nsectors = g.dkg_nsect;
766 		if (ntracks == 0)
767 			ntracks = use_efi_dflts ? -1 : g.dkg_nhead;
768 		if (rpm == 0)
769 			rpm = ((int)g.dkg_rpm <= 0) ? 3600: g.dkg_rpm;
770 	}
771 
772 	if (bsize == 0)
773 		bsize = DESBLKSIZE;
774 	/*
775 	 * Adjust maxcontig by the device's maxtransfer. If maxtransfer
776 	 * information is not available, default to the min of a MB and
777 	 * maxphys.
778 	 */
779 	if (maxcontig == -1 && ioctl(fd, DKIOCINFO, &ci) == 0) {
780 		maxcontig = ci.dki_maxtransfer * DEV_BSIZE;
781 		if (maxcontig < 0) {
782 			int	error, gotit, maxphys;
783 			gotit = fsgetmaxphys(&maxphys, &error);
784 
785 			/*
786 			 * If we cannot get the maxphys value, default
787 			 * to ufs_maxmaxphys (MB).
788 			 */
789 			if (gotit) {
790 				maxcontig = MIN(maxphys, MB);
791 			} else {
792 				(void) fprintf(stderr, gettext(
793 "Warning: Could not get system value for maxphys. The value for maxcontig\n"
794 "will default to 1MB.\n"));
795 			maxcontig = MB;
796 			}
797 		}
798 		maxcontig /= bsize;
799 	}
800 	(void) close(fd);
801 	return (actual_size);
802 }
803 
804 /*
805  * Figure out how big the partition we're dealing with is.
806  */
807 static diskaddr_t
808 get_device_size(int fd, char *name)
809 {
810 	struct vtoc vtoc;
811 	dk_gpt_t *efi_vtoc;
812 	diskaddr_t	slicesize;
813 
814 	int index = read_vtoc(fd, &vtoc);
815 
816 	if (index >= 0) {
817 		label_type = LABEL_TYPE_VTOC;
818 	} else {
819 		if (index == VT_ENOTSUP || index == VT_ERROR) {
820 			/* it might be an EFI label */
821 			index = efi_alloc_and_read(fd, &efi_vtoc);
822 			if (index >= 0)
823 				label_type = LABEL_TYPE_EFI;
824 		}
825 	}
826 
827 	if (index < 0) {
828 		/*
829 		 * Since both attempts to read the label failed, we're
830 		 * going to fall back to a brute force approach to
831 		 * determining the device's size:  see how far out we can
832 		 * perform reads on the device.
833 		 */
834 
835 		slicesize = brute_force_get_device_size(fd);
836 		if (slicesize == 0) {
837 			switch (index) {
838 			case VT_ERROR:
839 				(void) fprintf(stderr, gettext(
840 				    "newfs: %s: %s\n"), name, strerror(errno));
841 				exit(10);
842 				/*NOTREACHED*/
843 			case VT_EIO:
844 				fatal(gettext(
845 				    "%s: I/O error accessing VTOC"), name);
846 				/*NOTREACHED*/
847 			case VT_EINVAL:
848 				fatal(gettext(
849 				    "%s: Invalid field in VTOC"), name);
850 				/*NOTREACHED*/
851 			default:
852 				fatal(gettext(
853 				    "%s: unknown error accessing VTOC"),
854 				    name);
855 				/*NOTREACHED*/
856 			}
857 		} else {
858 			label_type = LABEL_TYPE_OTHER;
859 		}
860 	}
861 
862 	if (label_type == LABEL_TYPE_EFI) {
863 		slicesize = efi_vtoc->efi_parts[index].p_size;
864 		efi_free(efi_vtoc);
865 	} else if (label_type == LABEL_TYPE_VTOC) {
866 		/*
867 		 * In the vtoc struct, p_size is a 32-bit signed quantity.
868 		 * In the dk_gpt struct (efi's version of the vtoc), p_size
869 		 * is an unsigned 64-bit quantity.  By casting the vtoc's
870 		 * psize to an unsigned 32-bit quantity, it will be copied
871 		 * to 'slicesize' (an unsigned 64-bit diskaddr_t) without
872 		 * sign extension.
873 		 */
874 
875 		slicesize = (uint32_t)vtoc.v_part[index].p_size;
876 	}
877 
878 	return (slicesize);
879 }
880 
881 /*
882  * brute_force_get_device_size
883  *
884  * Determine the size of the device by seeing how far we can
885  * read.  Doing an llseek( , , SEEK_END) would probably work
886  * in most cases, but we've seen at least one third-party driver
887  * which doesn't correctly support the SEEK_END option when the
888  * the device is greater than a terabyte.
889  */
890 
891 static diskaddr_t
892 brute_force_get_device_size(int fd)
893 {
894 	diskaddr_t	min_fail = 0;
895 	diskaddr_t	max_succeed = 0;
896 	diskaddr_t	cur_db_off;
897 	char 		buf[DEV_BSIZE];
898 
899 	/*
900 	 * First, see if we can read the device at all, just to
901 	 * eliminate errors that have nothing to do with the
902 	 * device's size.
903 	 */
904 
905 	if (((llseek(fd, (offset_t)0, SEEK_SET)) == -1) ||
906 	    ((read(fd, buf, DEV_BSIZE)) == -1))
907 		return (0);  /* can't determine size */
908 
909 	/*
910 	 * Now, go sequentially through the multiples of 4TB
911 	 * to find the first read that fails (this isn't strictly
912 	 * the most efficient way to find the actual size if the
913 	 * size really could be anything between 0 and 2**64 bytes.
914 	 * We expect the sizes to be less than 16 TB for some time,
915 	 * so why do a bunch of reads that are larger than that?
916 	 * However, this algorithm *will* work for sizes of greater
917 	 * than 16 TB.  We're just not optimizing for those sizes.)
918 	 */
919 
920 	for (cur_db_off = SECTORS_PER_TERABYTE * 4;
921 	    min_fail == 0 && cur_db_off < FS_SIZE_UPPER_LIMIT;
922 	    cur_db_off += 4 * SECTORS_PER_TERABYTE) {
923 		if (((llseek(fd, (offset_t)(cur_db_off * DEV_BSIZE),
924 		    SEEK_SET)) == -1) ||
925 		    ((read(fd, buf, DEV_BSIZE)) != DEV_BSIZE))
926 			min_fail = cur_db_off;
927 		else
928 			max_succeed = cur_db_off;
929 	}
930 
931 	if (min_fail == 0)
932 		return (0);
933 
934 	/*
935 	 * We now know that the size of the device is less than
936 	 * min_fail and greater than or equal to max_succeed.  Now
937 	 * keep splitting the difference until the actual size in
938 	 * sectors in known.  We also know that the difference
939 	 * between max_succeed and min_fail at this time is
940 	 * 4 * SECTORS_PER_TERABYTE, which is a power of two, which
941 	 * simplifies the math below.
942 	 */
943 
944 	while (min_fail - max_succeed > 1) {
945 		cur_db_off = max_succeed + (min_fail - max_succeed)/2;
946 		if (((llseek(fd, (offset_t)(cur_db_off * DEV_BSIZE),
947 		    SEEK_SET)) == -1) ||
948 		    ((read(fd, buf, DEV_BSIZE)) != DEV_BSIZE))
949 			min_fail = cur_db_off;
950 		else
951 			max_succeed = cur_db_off;
952 	}
953 
954 	/* the size is the last successfully read sector offset plus one */
955 	return (max_succeed + 1);
956 }
957 
958 /*
959  * validate_size
960  *
961  * Return 1 if the device appears to be at least "size" sectors long.
962  * Return 0 if it's shorter or we can't read it.
963  */
964 
965 static int
966 validate_size(char *disk, diskaddr_t size)
967 {
968 	char 		buf[DEV_BSIZE];
969 	int fd, rc;
970 
971 	if ((fd = open64(disk, O_RDONLY)) < 0) {
972 		perror(disk);
973 		exit(1);
974 	}
975 
976 	if ((llseek(fd, (offset_t)((size - 1) * DEV_BSIZE), SEEK_SET) == -1) ||
977 	    (read(fd, buf, DEV_BSIZE)) != DEV_BSIZE)
978 		rc = 0;
979 	else
980 		rc = 1;
981 	(void) close(fd);
982 	return (rc);
983 }
984 
985 /*
986  * read_sb(char * rawdev) - Attempt to read the superblock from a raw device
987  *
988  * Returns:
989  *	0 :
990  *		Could not read a valid superblock for a variety of reasons.
991  *		Since 'newfs' handles any fatal conditions, we're not going
992  *		to make any guesses as to why this is failing or what should
993  *		be done about it.
994  *
995  *	struct fs *:
996  *		A pointer to (what we think is) a valid superblock. The
997  *		space for the superblock is static (inside the function)
998  *		since we will only be reading the values from it.
999  */
1000 
1001 struct fs *
1002 read_sb(char *fsdev)
1003 {
1004 	static struct fs	sblock;
1005 	struct stat64		statb;
1006 	int			dskfd;
1007 	char			*bufp = NULL;
1008 	int			bufsz = 0;
1009 
1010 	if (stat64(fsdev, &statb) < 0)
1011 		return (0);
1012 
1013 	if ((dskfd = open64(fsdev, O_RDONLY)) < 0)
1014 		return (0);
1015 
1016 	/*
1017 	 * We need a buffer whose size is a multiple of DEV_BSIZE in order
1018 	 * to read from a raw device (which we were probably passed).
1019 	 */
1020 	bufsz = ((sizeof (sblock) / DEV_BSIZE) + 1) * DEV_BSIZE;
1021 	if ((bufp = malloc(bufsz)) == NULL) {
1022 		(void) close(dskfd);
1023 		return (0);
1024 	}
1025 
1026 	if (llseek(dskfd, (offset_t)SBOFF, SEEK_SET) < 0 ||
1027 	    read(dskfd, bufp, bufsz) < 0) {
1028 		(void) close(dskfd);
1029 		free(bufp);
1030 		return (0);
1031 	}
1032 	(void) close(dskfd);	/* Done with the file */
1033 
1034 	(void) memcpy(&sblock, bufp, sizeof (sblock));
1035 	free(bufp);	/* Don't need this anymore */
1036 
1037 	if (((sblock.fs_magic != FS_MAGIC) &&
1038 	    (sblock.fs_magic != MTB_UFS_MAGIC)) ||
1039 	    sblock.fs_ncg < 1 || sblock.fs_cpg < 1)
1040 		return (0);
1041 
1042 	if (sblock.fs_ncg * sblock.fs_cpg < sblock.fs_ncyl ||
1043 	    (sblock.fs_ncg - 1) * sblock.fs_cpg >= sblock.fs_ncyl)
1044 		return (0);
1045 
1046 	if (sblock.fs_sbsize < 0 || sblock.fs_sbsize > SBSIZE)
1047 		return (0);
1048 
1049 	return (&sblock);
1050 }
1051 
1052 /*
1053  * Read the UFS file system on the raw device SPECIAL.  If it does not
1054  * appear to be a UFS file system, return non-zero, indicating that
1055  * fsirand should be called (and it will spit out an error message).
1056  * If it is a UFS file system, take a look at the inodes in the first
1057  * cylinder group.  If they appear to be randomized (non-zero), return
1058  * zero, which will cause fsirand to not be called.  If the inode generation
1059  * counts are all zero, then we must call fsirand, so return non-zero.
1060  */
1061 
1062 #define	RANDOMIZED	0
1063 #define	NOT_RANDOMIZED	1
1064 
1065 static int
1066 notrand(char *special)
1067 {
1068 	long fsbuf[SBSIZE / sizeof (long)];
1069 	struct dinode dibuf[MAXBSIZE/sizeof (struct dinode)];
1070 	struct fs *fs;
1071 	struct dinode *dip;
1072 	offset_t seekaddr;
1073 	int bno, inum;
1074 	int fd;
1075 
1076 	fs = (struct fs *)fsbuf;
1077 	if ((fd = open64(special, 0)) == -1)
1078 		return (NOT_RANDOMIZED);
1079 	if (llseek(fd, (offset_t)SBLOCK * DEV_BSIZE, 0) == -1 ||
1080 	    read(fd, (char *)fs, SBSIZE) != SBSIZE ||
1081 	    ((fs->fs_magic != FS_MAGIC) && (fs->fs_magic != MTB_UFS_MAGIC))) {
1082 		(void) close(fd);
1083 		return (NOT_RANDOMIZED);
1084 	}
1085 
1086 	/* looks like a UFS file system; read the first cylinder group */
1087 	bsize = INOPB(fs) * sizeof (struct dinode);
1088 	inum = 0;
1089 	while (inum < fs->fs_ipg) {
1090 		bno = itod(fs, inum);
1091 		seekaddr = (offset_t)fsbtodb(fs, bno) * DEV_BSIZE;
1092 		if (llseek(fd, seekaddr, 0) == -1 ||
1093 		    read(fd, (char *)dibuf, bsize) != bsize) {
1094 			(void) close(fd);
1095 			return (NOT_RANDOMIZED);
1096 		}
1097 		for (dip = dibuf; dip < &dibuf[INOPB(fs)]; dip++) {
1098 			if (dip->di_gen != 0) {
1099 				(void) close(fd);
1100 				return (RANDOMIZED);
1101 			}
1102 			inum++;
1103 		}
1104 	}
1105 	(void) close(fd);
1106 	return (NOT_RANDOMIZED);
1107 }
1108 
1109 static void
1110 usage(void)
1111 {
1112 	(void) fprintf(stderr, gettext(
1113 	    "usage: newfs [ -v ] [ mkfs-options ] raw-special-device\n"));
1114 	(void) fprintf(stderr, gettext("where mkfs-options are:\n"));
1115 	(void) fprintf(stderr, gettext(
1116 	    "\t-N do not create file system, just print out parameters\n"));
1117 	(void) fprintf(stderr, gettext(
1118 "\t-T configure file system for eventual growth to over a terabyte\n"));
1119 	(void) fprintf(stderr, gettext("\t-s file system size (sectors)\n"));
1120 	(void) fprintf(stderr, gettext("\t-b block size\n"));
1121 	(void) fprintf(stderr, gettext("\t-f frag size\n"));
1122 	(void) fprintf(stderr, gettext("\t-t tracks/cylinder\n"));
1123 	(void) fprintf(stderr, gettext("\t-c cylinders/group\n"));
1124 	(void) fprintf(stderr, gettext("\t-m minimum free space %%\n"));
1125 	(void) fprintf(stderr, gettext(
1126 	    "\t-o optimization preference (`space' or `time')\n"));
1127 	(void) fprintf(stderr, gettext("\t-r revolutions/minute\n"));
1128 	(void) fprintf(stderr, gettext("\t-i number of bytes per inode\n"));
1129 	(void) fprintf(stderr, gettext(
1130 	    "\t-a number of alternates per cylinder\n"));
1131 	(void) fprintf(stderr, gettext("\t-C maxcontig\n"));
1132 	(void) fprintf(stderr, gettext("\t-d rotational delay\n"));
1133 	(void) fprintf(stderr, gettext(
1134 	    "\t-n number of rotational positions\n"));
1135 	(void) fprintf(stderr, gettext(
1136 "\t-S print a textual version of the calculated superblock to stdout\n"));
1137 	(void) fprintf(stderr, gettext(
1138 "\t-B dump a binary version of the calculated superblock to stdout\n"));
1139 }
1140 
1141 /*
1142  * Error-detecting version of atoi(3).  Adapted from mkfs' number().
1143  */
1144 static unsigned int
1145 number(char *param, char *value, int flags, int def_value)
1146 {
1147 	char *cs;
1148 	int n;
1149 	int cut = INT_MAX / 10;    /* limit to avoid overflow */
1150 	int minus = 0;
1151 
1152 	cs = value;
1153 	if (*cs == '-') {
1154 		minus = 1;
1155 		cs += 1;
1156 	}
1157 	if ((*cs < '0') || (*cs > '9')) {
1158 		goto bail_out;
1159 	}
1160 	n = 0;
1161 	while ((*cs >= '0') && (*cs <= '9') && (n <= cut)) {
1162 		n = n*10 + *cs++ - '0';
1163 	}
1164 	if (minus)
1165 		n = -n;
1166 	for (;;) {
1167 		switch (*cs++) {
1168 		case '\0':
1169 			return (n);
1170 
1171 		case '0': case '1': case '2': case '3': case '4':
1172 		case '5': case '6': case '7': case '8': case '9':
1173 			(void) fprintf(stderr, gettext(
1174 			    "newfs: value for %s overflowed, using %d\n"),
1175 			    param, def_value);
1176 			return (def_value);
1177 
1178 		case '%':
1179 			if (flags & NR_PERCENT)
1180 				break;
1181 			/* FALLTHROUGH */
1182 
1183 		default:
1184 bail_out:
1185 			fatal(gettext("bad numeric arg for %s: \"%s\""),
1186 			    param, value);
1187 
1188 		}
1189 	}
1190 	/* NOTREACHED */
1191 }
1192 
1193 /*
1194  * Error-detecting version of atoi(3).  Adapted from mkfs' number().
1195  */
1196 static int64_t
1197 number64(char *param, char *value, int flags, int64_t def_value)
1198 {
1199 	char *cs;
1200 	int64_t n;
1201 	int64_t cut = FS_SIZE_UPPER_LIMIT/ 10;    /* limit to avoid overflow */
1202 	int minus = 0;
1203 
1204 	cs = value;
1205 	if (*cs == '-') {
1206 		minus = 1;
1207 		cs += 1;
1208 	}
1209 	if ((*cs < '0') || (*cs > '9')) {
1210 		goto bail_out;
1211 	}
1212 	n = 0;
1213 	while ((*cs >= '0') && (*cs <= '9') && (n <= cut)) {
1214 		n = n*10 + *cs++ - '0';
1215 	}
1216 	if (minus)
1217 		n = -n;
1218 	for (;;) {
1219 		switch (*cs++) {
1220 		case '\0':
1221 			return (n);
1222 
1223 		case '0': case '1': case '2': case '3': case '4':
1224 		case '5': case '6': case '7': case '8': case '9':
1225 			(void) fprintf(stderr, gettext(
1226 			    "newfs: value for %s overflowed, using %d\n"),
1227 			    param, def_value);
1228 			return (def_value);
1229 
1230 		case '%':
1231 			if (flags & NR_PERCENT)
1232 				break;
1233 			/* FALLTHROUGH */
1234 
1235 		default:
1236 bail_out:
1237 			fatal(gettext("bad numeric arg for %s: \"%s\""),
1238 			    param, value);
1239 
1240 		}
1241 	}
1242 	/* NOTREACHED */
1243 }
1244