xref: /illumos-gate/usr/src/cmd/fs.d/ufs/newfs/newfs.c (revision 060ca14c)
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  * newfs: friendly front end to mkfs
24  *
25  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
26  */
27 
28 #include <sys/param.h>
29 #include <sys/types.h>
30 #include <locale.h>
31 #include <sys/stat.h>
32 #include <sys/buf.h>
33 #include <sys/fs/ufs_fs.h>
34 #include <sys/vnode.h>
35 #include <sys/fs/ufs_inode.h>
36 #include <sys/sysmacros.h>
37 
38 #include <errno.h>
39 #include <stdio.h>
40 #include <string.h>
41 #include <stdlib.h>
42 #include <stdarg.h>
43 #include <stdio.h>
44 #include <fcntl.h>
45 #include <unistd.h>
46 #include <limits.h>
47 #include <libintl.h>
48 #include <sys/dkio.h>
49 #include <sys/vtoc.h>
50 #include <sys/mkdev.h>
51 #include <sys/efi_partition.h>
52 
53 #include <fslib.h>
54 
55 static unsigned int number(char *, char *, int, int);
56 static int64_t number64(char *, char *, int, int64_t);
57 static diskaddr_t getdiskbydev(char *);
58 static int  yes(void);
59 static int  notrand(char *);
60 static void usage();
61 static diskaddr_t get_device_size(int, char *);
62 static diskaddr_t brute_force_get_device_size(int);
63 static int validate_size(char *disk, diskaddr_t size);
64 static void exenv(void);
65 static struct fs *read_sb(char *);
66 /*PRINTFLIKE1*/
67 static void fatal(char *fmt, ...);
68 
69 #define	EPATH "PATH=/usr/sbin:/sbin:"
70 #define	CPATH "/sbin"					/* an EPATH element */
71 #define	MB (1024 * 1024)
72 #define	GBSEC ((1024 * 1024 * 1024) / DEV_BSIZE)	/* sectors in a GB */
73 #define	MINFREESEC ((64 * 1024 * 1024) / DEV_BSIZE)	/* sectors in 64 MB */
74 #define	MINCPG (16)	/* traditional */
75 #define	MAXDEFDENSITY (8 * 1024)	/* arbitrary */
76 #define	MINDENSITY (2 * 1024)	/* traditional */
77 #define	MIN_MTB_DENSITY (1024 * 1024)
78 #define	POWEROF2(num)	(((num) & ((num) - 1)) == 0)
79 #define	SECTORS_PER_TERABYTE	(1LL << 31)
80 /*
81  * The following constant specifies an upper limit for file system size
82  * that is actually a lot bigger than we expect to support with UFS. (Since
83  * it's specified in sectors, the file system size would be 2**44 * 512,
84  * which is 2**53, which is 8192 Terabytes.)  However, it's useful
85  * for checking the basic sanity of a size value that is input on the
86  * command line.
87  */
88 #define	FS_SIZE_UPPER_LIMIT	0x100000000000LL
89 
90 /* For use with number() */
91 #define	NR_NONE		0
92 #define	NR_PERCENT	0x01
93 
94 /*
95  * The following two constants set the default block and fragment sizes.
96  * Both constants must be a power of 2 and meet the following constraints:
97  *	MINBSIZE <= DESBLKSIZE <= MAXBSIZE
98  *	DEV_BSIZE <= DESFRAGSIZE <= DESBLKSIZE
99  *	DESBLKSIZE / DESFRAGSIZE <= 8
100  */
101 #define	DESBLKSIZE	8192
102 #define	DESFRAGSIZE	1024
103 
104 #ifdef DEBUG
105 #define	dprintf(x)	printf x
106 #else
107 #define	dprintf(x)
108 #endif
109 
110 static int	Nflag;		/* run mkfs without writing file system */
111 static int	Tflag;		/* set up file system for growth to over 1 TB */
112 static int	verbose;	/* show mkfs line before exec */
113 static int	fsize = 0;		/* fragment size */
114 static int	fsize_flag = 0;	/* fragment size was specified on cmd line */
115 static int	bsize;		/* block size */
116 static int	ntracks;	/* # tracks/cylinder */
117 static int	ntracks_set = 0; /* true if the user specified ntracks */
118 static int	optim = FS_OPTTIME;	/* optimization, t(ime) or s(pace) */
119 static int	nsectors;	/* # sectors/track */
120 static int	cpg;		/* cylinders/cylinder group */
121 static int	cpg_set = 0;	/* true if the user specified cpg */
122 static int	minfree = -1;	/* free space threshold */
123 static int	rpm;		/* revolutions/minute of drive */
124 static int	rpm_set = 0;	/* true if the user specified rpm */
125 static int	nrpos = 8;	/* # of distinguished rotational positions */
126 				/* 8 is the historical default */
127 static int	nrpos_set = 0;	/* true if the user specified nrpos */
128 static int	density = 0;	/* number of bytes per inode */
129 static int	apc;		/* alternates per cylinder */
130 static int	apc_set = 0;	/* true if the user specified apc */
131 static int 	rot = -1;	/* rotational delay (msecs) */
132 static int	rot_set = 0;	/* true if the user specified rot */
133 static int 	maxcontig = -1;	/* maximum number of contig blocks */
134 static int	text_sb = 0;	/* no disk changes; just final sb text dump */
135 static int	binary_sb = 0;	/* no disk changes; just final sb binary dump */
136 static int	label_type;	/* see types below */
137 
138 /*
139  * The variable use_efi_dflts is an indicator of whether to use EFI logic
140  * or the geometry logic in laying out the filesystem. This is decided
141  * based on the size/type of the disk and is used only for non-EFI labeled
142  * disks and removable media.
143  */
144 static int	use_efi_dflts = 0;
145 static int	isremovable = 0;
146 static int	ishotpluggable = 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 	struct fs *sbp;	/* Pointer to superblock (if present) */
161 	diskaddr_t actual_fssize;
162 	diskaddr_t max_possible_fssize;
163 	diskaddr_t req_fssize = 0;
164 	diskaddr_t fssize = 0;
165 	char	*req_fssize_str = NULL; /* requested size argument */
166 
167 	(void) setlocale(LC_ALL, "");
168 
169 #if !defined(TEXT_DOMAIN)
170 #define	TEXT_DOMAIN	"SYS_TEST"
171 #endif
172 	(void) textdomain(TEXT_DOMAIN);
173 
174 	opterr = 0;	/* We print our own errors, disable getopt's message */
175 	while ((option = getopt(argc, argv,
176 	    "vNBSs:C:d:t:o:a:b:f:c:m:n:r:i:T")) != EOF) {
177 		switch (option) {
178 		case 'S':
179 			text_sb++;
180 			break;
181 		case 'B':
182 			binary_sb++;
183 			break;
184 		case 'v':
185 			verbose++;
186 			break;
187 
188 		case 'N':
189 			Nflag++;
190 			break;
191 
192 		case 's':
193 			/*
194 			 * The maximum file system size is a lot smaller
195 			 * than FS_SIZE_UPPER_LIMIT, but until we find out
196 			 * the device size and block size, we don't know
197 			 * what it is.  So save the requested size in a
198 			 * string so that we can print it out later if we
199 			 * determine it's too big.
200 			 */
201 			req_fssize = number64("fssize", optarg, NR_NONE,
202 			    FS_SIZE_UPPER_LIMIT);
203 			if (req_fssize < 1024)
204 				fatal(gettext(
205 				    "%s: fssize must be at least 1024"),
206 				    optarg);
207 			req_fssize_str = strdup(optarg);
208 			if (req_fssize_str == NULL)
209 				fatal(gettext(
210 				    "Insufficient memory for string copy."));
211 			break;
212 
213 		case 'C':
214 			maxcontig = number("maxcontig", optarg, NR_NONE, -1);
215 			if (maxcontig < 0)
216 				fatal(gettext("%s: bad maxcontig"), optarg);
217 			break;
218 
219 		case 'd':
220 			rot = number("rotdelay", optarg, NR_NONE, 0);
221 			rot_set = 1;
222 			if (rot < 0 || rot > 1000)
223 				fatal(gettext(
224 				    "%s: bad rotational delay"), optarg);
225 			break;
226 
227 		case 't':
228 			ntracks = number("ntrack", optarg, NR_NONE, 16);
229 			ntracks_set = 1;
230 			if ((ntracks < 0) ||
231 			    (ntracks > INT_MAX))
232 				fatal(gettext("%s: bad total tracks"), optarg);
233 			break;
234 
235 		case 'o':
236 			if (strcmp(optarg, "space") == 0)
237 				optim = FS_OPTSPACE;
238 			else if (strcmp(optarg, "time") == 0)
239 				optim = FS_OPTTIME;
240 			else
241 				fatal(gettext(
242 "%s: bad optimization preference (options are `space' or `time')"), optarg);
243 			break;
244 
245 		case 'a':
246 			apc = number("apc", optarg, NR_NONE, 0);
247 			apc_set = 1;
248 			if (apc < 0 || apc > 32768) /* see mkfs.c */
249 				fatal(gettext(
250 				    "%s: bad alternates per cyl"), optarg);
251 			break;
252 
253 		case 'b':
254 			bsize = number("bsize", optarg, NR_NONE, DESBLKSIZE);
255 			if (bsize < MINBSIZE || bsize > MAXBSIZE)
256 				fatal(gettext(
257 				    "%s: bad block size"), optarg);
258 			break;
259 
260 		case 'f':
261 			fsize = number("fragsize", optarg, NR_NONE,
262 			    DESFRAGSIZE);
263 			fsize_flag++;
264 			/* xxx ought to test against bsize for upper limit */
265 			if (fsize < DEV_BSIZE)
266 				fatal(gettext("%s: bad frag size"), optarg);
267 			break;
268 
269 		case 'c':
270 			cpg = number("cpg", optarg, NR_NONE, 16);
271 			cpg_set = 1;
272 			if (cpg < 1)
273 				fatal(gettext("%s: bad cylinders/group"),
274 				    optarg);
275 			break;
276 
277 		case 'm':
278 			minfree = number("minfree", optarg, NR_PERCENT, 10);
279 			if (minfree < 0 || minfree > 99)
280 				fatal(gettext("%s: bad free space %%"), optarg);
281 			break;
282 
283 		case 'n':
284 			nrpos = number("nrpos", optarg, NR_NONE, 8);
285 			nrpos_set = 1;
286 			if (nrpos <= 0)
287 				fatal(gettext(
288 				    "%s: bad number of rotational positions"),
289 				    optarg);
290 			break;
291 
292 		case 'r':
293 			rpm = number("rpm", optarg, NR_NONE, 3600);
294 			rpm_set = 1;
295 			if (rpm < 0)
296 				fatal(gettext("%s: bad revs/minute"), optarg);
297 			break;
298 
299 		case 'i':
300 			/* xxx ought to test against fsize */
301 			density = number("nbpi", optarg, NR_NONE, 2048);
302 			if (density < DEV_BSIZE)
303 				fatal(gettext("%s: bad bytes per inode"),
304 				    optarg);
305 			break;
306 
307 		case 'T':
308 			Tflag++;
309 			break;
310 
311 		default:
312 			usage();
313 			fatal(gettext("-%c: unknown flag"), optopt);
314 		}
315 	}
316 
317 	/* At this point, there should only be one argument left:	*/
318 	/* The raw-special-device itself. If not, print usage message.	*/
319 	if ((argc - optind) != 1) {
320 		usage();
321 		exit(1);
322 	}
323 
324 	name = argv[optind];
325 
326 	special = getfullrawname(name);
327 	if (special == NULL) {
328 		(void) fprintf(stderr, gettext("newfs: malloc failed\n"));
329 		exit(1);
330 	}
331 
332 	if (*special == '\0') {
333 		if (strchr(name, '/') != NULL) {
334 			if (stat64(name, &st) < 0) {
335 				(void) fprintf(stderr,
336 				    gettext("newfs: %s: %s\n"),
337 				    name, strerror(errno));
338 				exit(2);
339 			}
340 			fatal(gettext("%s: not a raw disk device"), name);
341 		}
342 		(void) snprintf(device, sizeof (device), "/dev/rdsk/%s", name);
343 		if ((special = getfullrawname(device)) == NULL) {
344 			(void) fprintf(stderr,
345 			    gettext("newfs: malloc failed\n"));
346 			exit(1);
347 		}
348 
349 		if (*special == '\0') {
350 			(void) snprintf(device, sizeof (device), "/dev/%s",
351 			    name);
352 			if ((special = getfullrawname(device)) == NULL) {
353 				(void) fprintf(stderr,
354 				    gettext("newfs: malloc failed\n"));
355 				exit(1);
356 			}
357 			if (*special == '\0')
358 				fatal(gettext(
359 				    "%s: not a raw disk device"), name);
360 		}
361 	}
362 
363 	/*
364 	 * getdiskbydev() determines the characteristics of the special
365 	 * device on which the file system will be built.  In the case
366 	 * of devices with SMI labels (that is, non-EFI labels), the
367 	 * following characteristics are set (if they were not already
368 	 * set on the command line, since the command line settings
369 	 * take precedence):
370 	 *
371 	 *	nsectors - sectors per track
372 	 *	ntracks - tracks per cylinder
373 	 *	rpm - disk revolutions per minute
374 	 *
375 	 *	apc is NOT set
376 	 *
377 	 * getdiskbydev() also sets the following quantities for all
378 	 * devices, if not already set:
379 	 *
380 	 *	bsize - file system block size
381 	 *	maxcontig
382 	 *	label_type (efi, vtoc, or other)
383 	 *
384 	 * getdiskbydev() returns the actual size of the device, in
385 	 * sectors.
386 	 */
387 
388 	actual_fssize = getdiskbydev(special);
389 
390 	if (req_fssize == 0) {
391 		fssize = actual_fssize;
392 	} else {
393 		/*
394 		 * If the user specified a size larger than what we've
395 		 * determined as the actual size of the device, see if the
396 		 * size specified by the user can be read.  If so, use it,
397 		 * since some devices and volume managers may not support
398 		 * the vtoc and EFI interfaces we use to determine device
399 		 * size.
400 		 */
401 		if (req_fssize > actual_fssize &&
402 		    validate_size(special, req_fssize)) {
403 			(void) fprintf(stderr, gettext(
404 "Warning: the requested size of this file system\n"
405 "(%lld sectors) is greater than the size of the\n"
406 "device reported by the driver (%lld sectors).\n"
407 "However, a read of the device at the requested size\n"
408 "does succeed, so the requested size will be used.\n"),
409 			    req_fssize, actual_fssize);
410 			fssize = req_fssize;
411 		} else {
412 			fssize = MIN(req_fssize, actual_fssize);
413 		}
414 	}
415 
416 	if (label_type == LABEL_TYPE_VTOC) {
417 		if (nsectors < 0)
418 			fatal(gettext("%s: no default #sectors/track"),
419 			    special);
420 		if (!use_efi_dflts) {
421 			if (ntracks < 0)
422 				fatal(gettext("%s: no default #tracks"),
423 				    special);
424 		}
425 		if (rpm < 0)
426 			fatal(gettext(
427 			    "%s: no default revolutions/minute value"),
428 			    special);
429 		if (rpm < 60) {
430 			(void) fprintf(stderr,
431 			    gettext("Warning: setting rpm to 60\n"));
432 			rpm = 60;
433 		}
434 	}
435 	if (label_type == LABEL_TYPE_EFI || label_type == LABEL_TYPE_OTHER) {
436 		if (ntracks_set)
437 			(void) fprintf(stderr, gettext(
438 "Warning: ntracks is obsolete for this device and will be ignored.\n"));
439 		if (cpg_set)
440 			(void) fprintf(stderr, gettext(
441 "Warning: cylinders/group is obsolete for this device and will be ignored.\n"));
442 		if (rpm_set)
443 			(void) fprintf(stderr, gettext(
444 "Warning: rpm is obsolete for this device and will be ignored.\n"));
445 		if (rot_set)
446 			(void) fprintf(stderr, gettext(
447 "Warning: rotational delay is obsolete for this device and"
448 " will be ignored.\n"));
449 		if (nrpos_set)
450 			(void) fprintf(stderr, gettext(
451 "Warning: number of rotational positions is obsolete for this device and\n"
452 "will be ignored.\n"));
453 		if (apc_set)
454 			(void) fprintf(stderr, gettext(
455 "Warning: number of alternate sectors per cylinder is obsolete for this\n"
456 "device and will be ignored.\n"));
457 
458 		/*
459 		 * We need these for the call to mkfs, even though they are
460 		 * meaningless.
461 		 */
462 		rpm = 60;
463 		nrpos = 1;
464 		apc = 0;
465 		rot = -1;
466 
467 		/*
468 		 * These values are set to produce a file system with
469 		 * a cylinder group size of 48MB.   For disks with
470 		 * non-EFI labels, most geometries result in cylinder
471 		 * groups of around 40 - 50 MB, so we arbitrarily choose
472 		 * 48MB for disks with EFI labels.  mkfs will reduce
473 		 * cylinders per group even further if necessary.
474 		 */
475 
476 		cpg = 16;
477 		nsectors = 128;
478 		ntracks = 48;
479 
480 		/*
481 		 * mkfs produces peculiar results for file systems
482 		 * that are smaller than one cylinder so don't allow
483 		 * them to be created (this check is only made for
484 		 * disks with EFI labels.  Eventually, it should probably
485 		 * be enforced for all disks.)
486 		 */
487 
488 		if (fssize < nsectors * ntracks) {
489 			fatal(gettext(
490 			    "file system size must be at least %d sectors"),
491 			    nsectors * ntracks);
492 		}
493 	}
494 
495 	if (fssize > INT_MAX)
496 		Tflag = 1;
497 
498 	/*
499 	 * If the user requested that the file system be set up for
500 	 * eventual growth to over a terabyte, or if it's already greater
501 	 * than a terabyte, set the inode density (nbpi) to MIN_MTB_DENSITY
502 	 * (unless the user has specified a larger nbpi), set the frag size
503 	 * equal to the block size, and set the cylinders-per-group value
504 	 * passed to mkfs to -1, which tells mkfs to make cylinder groups
505 	 * as large as possible.
506 	 */
507 	if (Tflag) {
508 		if (density < MIN_MTB_DENSITY)
509 			density = MIN_MTB_DENSITY;
510 		fsize = bsize;
511 		cpg = -1; 	/* says make cyl groups as big as possible */
512 	} else {
513 		if (fsize == 0)
514 			fsize = DESFRAGSIZE;
515 	}
516 
517 	if (!POWEROF2(fsize)) {
518 		(void) fprintf(stderr, gettext(
519 		    "newfs: fragment size must a power of 2, not %d\n"), fsize);
520 		fsize = bsize/8;
521 		(void) fprintf(stderr, gettext(
522 		    "newfs: fragsize reset to %ld\n"), fsize);
523 	}
524 
525 	/*
526 	 * The file system is limited in size by the fragment size.
527 	 * The number of fragments in the file system must fit into
528 	 * a signed 32-bit quantity, so the number of sectors in the
529 	 * file system is INT_MAX * the number of sectors in a frag.
530 	 */
531 
532 	max_possible_fssize = ((uint64_t)fsize)/DEV_BSIZE * INT_MAX;
533 	if (fssize > max_possible_fssize)
534 		fssize = max_possible_fssize;
535 
536 	/*
537 	 * Now fssize is the final size of the file system (in sectors).
538 	 * If it's less than what the user requested, print a message.
539 	 */
540 	if (fssize < req_fssize) {
541 		(void) fprintf(stderr, gettext(
542 		    "newfs: requested size of %s disk blocks is too large.\n"),
543 		    req_fssize_str);
544 		(void) fprintf(stderr, gettext(
545 		    "newfs: Resetting size to %lld\n"), fssize);
546 	}
547 
548 	/*
549 	 * fssize now equals the size (in sectors) of the file system
550 	 * that will be created.
551 	 */
552 
553 	/* XXX - following defaults are both here and in mkfs */
554 	if (density <= 0) {
555 		if (fssize < GBSEC)
556 			density = MINDENSITY;
557 		else
558 			density = (int)((((longlong_t)fssize + (GBSEC - 1)) /
559 			    GBSEC) * MINDENSITY);
560 		if (density <= 0)
561 			density = MINDENSITY;
562 		if (density > MAXDEFDENSITY)
563 			density = MAXDEFDENSITY;
564 	}
565 	if (cpg == 0) {
566 		/*
567 		 * maxcpg calculation adapted from mkfs
568 		 * In the case of disks with EFI labels, cpg has
569 		 * already been set, so we won't enter this code.
570 		 */
571 		long maxcpg, maxipg;
572 
573 		maxipg = roundup(bsize * NBBY / 3,
574 		    bsize / sizeof (struct inode));
575 		maxcpg = (bsize - sizeof (struct cg) - howmany(maxipg, NBBY)) /
576 		    (sizeof (long) + nrpos * sizeof (short) +
577 		    nsectors / (MAXFRAG * NBBY));
578 		cpg = (fssize / GBSEC) * 32;
579 		if (cpg > maxcpg)
580 			cpg = maxcpg;
581 		if (cpg <= 0)
582 			cpg = MINCPG;
583 	}
584 	if (minfree < 0) {
585 		minfree = (int)(((float)MINFREESEC / fssize) * 100);
586 		if (minfree > 10)
587 			minfree = 10;
588 		if (minfree <= 0)
589 			minfree = 1;
590 	}
591 #ifdef i386	/* Bug 1170182 */
592 	if (ntracks > 32 && (ntracks % 16) != 0) {
593 		ntracks -= (ntracks % 16);
594 	}
595 #endif
596 	/*
597 	 * Confirmation
598 	 */
599 	if (isatty(fileno(stdin)) && !Nflag) {
600 		/*
601 		 * If we can read a valid superblock, report the mount
602 		 * point on which this filesystem was last mounted.
603 		 */
604 		if (((sbp = read_sb(special)) != 0) &&
605 		    (*sbp->fs_fsmnt != '\0')) {
606 			(void) printf(gettext(
607 			    "newfs: %s last mounted as %s\n"),
608 			    special, sbp->fs_fsmnt);
609 		}
610 		(void) printf(gettext(
611 		    "newfs: construct a new file system %s: (y/n)? "),
612 		    special);
613 		(void) fflush(stdout);
614 		if (!yes())
615 			exit(0);
616 	}
617 
618 	dprintf(("DeBuG newfs : nsect=%d ntrak=%d cpg=%d\n",
619 	    nsectors, ntracks, cpg));
620 	/*
621 	 * If alternates-per-cylinder is ever implemented:
622 	 * need to get apc from dp->d_apc if no -a switch???
623 	 */
624 	(void) snprintf(cmd, sizeof (cmd), "mkfs -F ufs "
625 	    "%s%s%s%s %lld %d %d %d %d %d %d %d %d %s %d %d %d %d %s",
626 	    Nflag ? "-o N " : "", binary_sb ? "-o calcbinsb " : "",
627 	    text_sb ? "-o calcsb " : "", special,
628 	    fssize, nsectors, ntracks, bsize, fsize, cpg, minfree, rpm/60,
629 	    density, optim == FS_OPTSPACE ? "s" : "t", apc, rot, nrpos,
630 	    maxcontig, Tflag ? "y" : "n");
631 	if (verbose) {
632 		(void) printf("%s\n", cmd);
633 		(void) fflush(stdout);
634 	}
635 	exenv();
636 	if (status = system(cmd))
637 		exit(status >> 8);
638 	if (Nflag)
639 		exit(0);
640 	(void) snprintf(cmd, sizeof (cmd), "/usr/sbin/fsirand %s", special);
641 	if (notrand(special) && (status = system(cmd)) != 0)
642 		(void) fprintf(stderr,
643 		    gettext("%s: failed, status = %d\n"),
644 		    cmd, status);
645 	return (0);
646 }
647 
648 static void
649 exenv(void)
650 {
651 	char *epath;				/* executable file path */
652 	char *cpath;				/* current path */
653 
654 	if ((cpath = getenv("PATH")) == NULL) {
655 		(void) fprintf(stderr, gettext("newfs: no PATH in env\n"));
656 		/*
657 		 * Background: the Bourne shell interpolates "." into
658 		 * the path where said path starts with a colon, ends
659 		 * with a colon, or has two adjacent colons.  Thus,
660 		 * the path ":/sbin::/usr/sbin:" is equivalent to
661 		 * ".:/sbin:.:/usr/sbin:.".  Now, we have no cpath,
662 		 * and epath ends in a colon (to make for easy
663 		 * catenation in the normal case).  By the above, if
664 		 * we use "", then "." becomes part of path.  That's
665 		 * bad, so use CPATH (which is just a duplicate of some
666 		 * element in EPATH).  No point in opening ourselves
667 		 * up to a Trojan horse attack when we don't have to....
668 		 */
669 		cpath = CPATH;
670 	}
671 	if ((epath = malloc(strlen(EPATH) + strlen(cpath) + 1)) == NULL) {
672 		(void) fprintf(stderr, gettext("newfs: malloc failed\n"));
673 		exit(1);
674 	}
675 	(void) strcpy(epath, EPATH);
676 	(void) strcat(epath, cpath);
677 	if (putenv(epath) < 0) {
678 		(void) fprintf(stderr, gettext("newfs: putenv failed\n"));
679 		exit(1);
680 	}
681 }
682 
683 static int
684 yes(void)
685 {
686 	int	i, b;
687 
688 	i = b = getchar();
689 	while (b != '\n' && b != '\0' && b != EOF)
690 		b = getchar();
691 	return (i == 'y');
692 }
693 
694 /*
695  * xxx Caller must run fmt through gettext(3) for us, if we ever
696  * xxx go the i18n route....
697  */
698 static void
699 fatal(char *fmt, ...)
700 {
701 	va_list pvar;
702 
703 	(void) fprintf(stderr, "newfs: ");
704 	va_start(pvar, fmt);
705 	(void) vfprintf(stderr, fmt, pvar);
706 	va_end(pvar);
707 	(void) putc('\n', stderr);
708 	exit(10);
709 }
710 
711 static diskaddr_t
712 getdiskbydev(char *disk)
713 {
714 	struct dk_geom g;
715 	struct dk_cinfo ci;
716 	struct dk_minfo info;
717 	diskaddr_t actual_size;
718 	int fd;
719 
720 	if ((fd = open64(disk, 0)) < 0) {
721 		perror(disk);
722 		exit(1);
723 	}
724 
725 	/*
726 	 * get_device_size() determines the actual size of the
727 	 * device, and also the disk's attributes, such as geometry.
728 	 */
729 	actual_size = get_device_size(fd, disk);
730 
731 	if (label_type == LABEL_TYPE_VTOC) {
732 
733 		/*
734 		 * Geometry information does not make sense for removable or
735 		 * hotpluggable media anyway, so indicate mkfs to use EFI
736 		 * 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 		}
744 
745 		/* If removable check if a floppy disk */
746 		if (isremovable) {
747 			if (ioctl(fd, DKIOCGMEDIAINFO, &info)) {
748 				dprintf(("DeBuG newfs : Unable to get media"
749 				    " info from %s.\n", disk));
750 			} else {
751 				if (info.dki_media_type == DK_FLOPPY) {
752 					isremovable = 0;
753 				}
754 			}
755 		}
756 
757 		if (ioctl(fd, DKIOCHOTPLUGGABLE, &ishotpluggable)) {
758 			dprintf(("DeBuG newfs : Unable to determine if %s is"
759 			    " Hotpluggable Media. Proceeding with system"
760 			    " determined parameters.\n", disk));
761 			ishotpluggable = 0;
762 		}
763 
764 		if ((isremovable || ishotpluggable) && !Tflag)
765 			use_efi_dflts = 1;
766 
767 		if (ioctl(fd, DKIOCGGEOM, &g))
768 			fatal(gettext(
769 			    "%s: Unable to read Disk geometry"), disk);
770 		if ((((diskaddr_t)g.dkg_ncyl * g.dkg_nhead *
771 		    g.dkg_nsect) > CHSLIMIT) && !Tflag) {
772 			use_efi_dflts = 1;
773 		}
774 		dprintf(("DeBuG newfs : geom=%llu, CHSLIMIT=%d "
775 		    "isremovable = %d ishotpluggable = %d use_efi_dflts = %d\n",
776 		    (diskaddr_t)g.dkg_ncyl * g.dkg_nhead * g.dkg_nsect,
777 		    CHSLIMIT, isremovable, ishotpluggable, use_efi_dflts));
778 		/*
779 		 * The ntracks that is passed to mkfs is decided here based
780 		 * on 'use_efi_dflts' and whether ntracks was specified as a
781 		 * command line parameter to newfs.
782 		 * If ntracks of -1 is passed to mkfs, mkfs uses DEF_TRACKS_EFI
783 		 * and DEF_SECTORS_EFI for ntracks and nsectors respectively.
784 		 */
785 		if (nsectors == 0)
786 			nsectors = g.dkg_nsect;
787 		if (ntracks == 0)
788 			ntracks = use_efi_dflts ? -1 : g.dkg_nhead;
789 		if (rpm == 0)
790 			rpm = ((int)g.dkg_rpm <= 0) ? 3600: g.dkg_rpm;
791 	}
792 
793 	if (bsize == 0)
794 		bsize = DESBLKSIZE;
795 	/*
796 	 * Adjust maxcontig by the device's maxtransfer. If maxtransfer
797 	 * information is not available, default to the min of a MB and
798 	 * maxphys.
799 	 */
800 	if (maxcontig == -1 && ioctl(fd, DKIOCINFO, &ci) == 0) {
801 		maxcontig = ci.dki_maxtransfer * DEV_BSIZE;
802 		if (maxcontig < 0) {
803 			int	error, gotit, maxphys;
804 			gotit = fsgetmaxphys(&maxphys, &error);
805 
806 			/*
807 			 * If we cannot get the maxphys value, default
808 			 * to ufs_maxmaxphys (MB).
809 			 */
810 			if (gotit) {
811 				maxcontig = MIN(maxphys, MB);
812 			} else {
813 				(void) fprintf(stderr, gettext(
814 "Warning: Could not get system value for maxphys. The value for maxcontig\n"
815 "will default to 1MB.\n"));
816 			maxcontig = MB;
817 			}
818 		}
819 		maxcontig /= bsize;
820 	}
821 	(void) close(fd);
822 	return (actual_size);
823 }
824 
825 /*
826  * Figure out how big the partition we're dealing with is.
827  */
828 static diskaddr_t
829 get_device_size(int fd, char *name)
830 {
831 	struct extvtoc vtoc;
832 	dk_gpt_t *efi_vtoc;
833 	diskaddr_t	slicesize;
834 
835 	int index = read_extvtoc(fd, &vtoc);
836 
837 	if (index >= 0) {
838 		label_type = LABEL_TYPE_VTOC;
839 	} else {
840 		if (index == VT_ENOTSUP || index == VT_ERROR) {
841 			/* it might be an EFI label */
842 			index = efi_alloc_and_read(fd, &efi_vtoc);
843 			if (index >= 0)
844 				label_type = LABEL_TYPE_EFI;
845 		}
846 	}
847 
848 	if (index < 0) {
849 		/*
850 		 * Since both attempts to read the label failed, we're
851 		 * going to fall back to a brute force approach to
852 		 * determining the device's size:  see how far out we can
853 		 * perform reads on the device.
854 		 */
855 
856 		slicesize = brute_force_get_device_size(fd);
857 		if (slicesize == 0) {
858 			switch (index) {
859 			case VT_ERROR:
860 				(void) fprintf(stderr, gettext(
861 				    "newfs: %s: %s\n"), name, strerror(errno));
862 				exit(10);
863 				/*NOTREACHED*/
864 			case VT_EIO:
865 				fatal(gettext(
866 				    "%s: I/O error accessing VTOC"), name);
867 				/*NOTREACHED*/
868 			case VT_EINVAL:
869 				fatal(gettext(
870 				    "%s: Invalid field in VTOC"), name);
871 				/*NOTREACHED*/
872 			default:
873 				fatal(gettext(
874 				    "%s: unknown error accessing VTOC"),
875 				    name);
876 				/*NOTREACHED*/
877 			}
878 		} else {
879 			label_type = LABEL_TYPE_OTHER;
880 		}
881 	}
882 
883 	if (label_type == LABEL_TYPE_EFI) {
884 		slicesize = efi_vtoc->efi_parts[index].p_size;
885 		efi_free(efi_vtoc);
886 	} else if (label_type == LABEL_TYPE_VTOC) {
887 		slicesize = vtoc.v_part[index].p_size;
888 	}
889 
890 	return (slicesize);
891 }
892 
893 /*
894  * brute_force_get_device_size
895  *
896  * Determine the size of the device by seeing how far we can
897  * read.  Doing an llseek( , , SEEK_END) would probably work
898  * in most cases, but we've seen at least one third-party driver
899  * which doesn't correctly support the SEEK_END option when the
900  * the device is greater than a terabyte.
901  */
902 
903 static diskaddr_t
904 brute_force_get_device_size(int fd)
905 {
906 	diskaddr_t	min_fail = 0;
907 	diskaddr_t	max_succeed = 0;
908 	diskaddr_t	cur_db_off;
909 	char 		buf[DEV_BSIZE];
910 
911 	/*
912 	 * First, see if we can read the device at all, just to
913 	 * eliminate errors that have nothing to do with the
914 	 * device's size.
915 	 */
916 
917 	if (((llseek(fd, (offset_t)0, SEEK_SET)) == -1) ||
918 	    ((read(fd, buf, DEV_BSIZE)) == -1))
919 		return (0);  /* can't determine size */
920 
921 	/*
922 	 * Now, go sequentially through the multiples of 4TB
923 	 * to find the first read that fails (this isn't strictly
924 	 * the most efficient way to find the actual size if the
925 	 * size really could be anything between 0 and 2**64 bytes.
926 	 * We expect the sizes to be less than 16 TB for some time,
927 	 * so why do a bunch of reads that are larger than that?
928 	 * However, this algorithm *will* work for sizes of greater
929 	 * than 16 TB.  We're just not optimizing for those sizes.)
930 	 */
931 
932 	for (cur_db_off = SECTORS_PER_TERABYTE * 4;
933 	    min_fail == 0 && cur_db_off < FS_SIZE_UPPER_LIMIT;
934 	    cur_db_off += 4 * SECTORS_PER_TERABYTE) {
935 		if (((llseek(fd, (offset_t)(cur_db_off * DEV_BSIZE),
936 		    SEEK_SET)) == -1) ||
937 		    ((read(fd, buf, DEV_BSIZE)) != DEV_BSIZE))
938 			min_fail = cur_db_off;
939 		else
940 			max_succeed = cur_db_off;
941 	}
942 
943 	if (min_fail == 0)
944 		return (0);
945 
946 	/*
947 	 * We now know that the size of the device is less than
948 	 * min_fail and greater than or equal to max_succeed.  Now
949 	 * keep splitting the difference until the actual size in
950 	 * sectors in known.  We also know that the difference
951 	 * between max_succeed and min_fail at this time is
952 	 * 4 * SECTORS_PER_TERABYTE, which is a power of two, which
953 	 * simplifies the math below.
954 	 */
955 
956 	while (min_fail - max_succeed > 1) {
957 		cur_db_off = max_succeed + (min_fail - max_succeed)/2;
958 		if (((llseek(fd, (offset_t)(cur_db_off * DEV_BSIZE),
959 		    SEEK_SET)) == -1) ||
960 		    ((read(fd, buf, DEV_BSIZE)) != DEV_BSIZE))
961 			min_fail = cur_db_off;
962 		else
963 			max_succeed = cur_db_off;
964 	}
965 
966 	/* the size is the last successfully read sector offset plus one */
967 	return (max_succeed + 1);
968 }
969 
970 /*
971  * validate_size
972  *
973  * Return 1 if the device appears to be at least "size" sectors long.
974  * Return 0 if it's shorter or we can't read it.
975  */
976 
977 static int
978 validate_size(char *disk, diskaddr_t size)
979 {
980 	char 		buf[DEV_BSIZE];
981 	int fd, rc;
982 
983 	if ((fd = open64(disk, O_RDONLY)) < 0) {
984 		perror(disk);
985 		exit(1);
986 	}
987 
988 	if ((llseek(fd, (offset_t)((size - 1) * DEV_BSIZE), SEEK_SET) == -1) ||
989 	    (read(fd, buf, DEV_BSIZE)) != DEV_BSIZE)
990 		rc = 0;
991 	else
992 		rc = 1;
993 	(void) close(fd);
994 	return (rc);
995 }
996 
997 /*
998  * read_sb(char * rawdev) - Attempt to read the superblock from a raw device
999  *
1000  * Returns:
1001  *	0 :
1002  *		Could not read a valid superblock for a variety of reasons.
1003  *		Since 'newfs' handles any fatal conditions, we're not going
1004  *		to make any guesses as to why this is failing or what should
1005  *		be done about it.
1006  *
1007  *	struct fs *:
1008  *		A pointer to (what we think is) a valid superblock. The
1009  *		space for the superblock is static (inside the function)
1010  *		since we will only be reading the values from it.
1011  */
1012 
1013 struct fs *
1014 read_sb(char *fsdev)
1015 {
1016 	static struct fs	sblock;
1017 	struct stat64		statb;
1018 	int			dskfd;
1019 	char			*bufp = NULL;
1020 	int			bufsz = 0;
1021 
1022 	if (stat64(fsdev, &statb) < 0)
1023 		return (0);
1024 
1025 	if ((dskfd = open64(fsdev, O_RDONLY)) < 0)
1026 		return (0);
1027 
1028 	/*
1029 	 * We need a buffer whose size is a multiple of DEV_BSIZE in order
1030 	 * to read from a raw device (which we were probably passed).
1031 	 */
1032 	bufsz = ((sizeof (sblock) / DEV_BSIZE) + 1) * DEV_BSIZE;
1033 	if ((bufp = malloc(bufsz)) == NULL) {
1034 		(void) close(dskfd);
1035 		return (0);
1036 	}
1037 
1038 	if (llseek(dskfd, (offset_t)SBOFF, SEEK_SET) < 0 ||
1039 	    read(dskfd, bufp, bufsz) < 0) {
1040 		(void) close(dskfd);
1041 		free(bufp);
1042 		return (0);
1043 	}
1044 	(void) close(dskfd);	/* Done with the file */
1045 
1046 	(void) memcpy(&sblock, bufp, sizeof (sblock));
1047 	free(bufp);	/* Don't need this anymore */
1048 
1049 	if (((sblock.fs_magic != FS_MAGIC) &&
1050 	    (sblock.fs_magic != MTB_UFS_MAGIC)) ||
1051 	    sblock.fs_ncg < 1 || sblock.fs_cpg < 1)
1052 		return (0);
1053 
1054 	if (sblock.fs_ncg * sblock.fs_cpg < sblock.fs_ncyl ||
1055 	    (sblock.fs_ncg - 1) * sblock.fs_cpg >= sblock.fs_ncyl)
1056 		return (0);
1057 
1058 	if (sblock.fs_sbsize < 0 || sblock.fs_sbsize > SBSIZE)
1059 		return (0);
1060 
1061 	return (&sblock);
1062 }
1063 
1064 /*
1065  * Read the UFS file system on the raw device SPECIAL.  If it does not
1066  * appear to be a UFS file system, return non-zero, indicating that
1067  * fsirand should be called (and it will spit out an error message).
1068  * If it is a UFS file system, take a look at the inodes in the first
1069  * cylinder group.  If they appear to be randomized (non-zero), return
1070  * zero, which will cause fsirand to not be called.  If the inode generation
1071  * counts are all zero, then we must call fsirand, so return non-zero.
1072  */
1073 
1074 #define	RANDOMIZED	0
1075 #define	NOT_RANDOMIZED	1
1076 
1077 static int
1078 notrand(char *special)
1079 {
1080 	long fsbuf[SBSIZE / sizeof (long)];
1081 	struct dinode dibuf[MAXBSIZE/sizeof (struct dinode)];
1082 	struct fs *fs;
1083 	struct dinode *dip;
1084 	offset_t seekaddr;
1085 	int bno, inum;
1086 	int fd;
1087 
1088 	fs = (struct fs *)fsbuf;
1089 	if ((fd = open64(special, 0)) == -1)
1090 		return (NOT_RANDOMIZED);
1091 	if (llseek(fd, (offset_t)SBLOCK * DEV_BSIZE, 0) == -1 ||
1092 	    read(fd, (char *)fs, SBSIZE) != SBSIZE ||
1093 	    ((fs->fs_magic != FS_MAGIC) && (fs->fs_magic != MTB_UFS_MAGIC))) {
1094 		(void) close(fd);
1095 		return (NOT_RANDOMIZED);
1096 	}
1097 
1098 	/* looks like a UFS file system; read the first cylinder group */
1099 	bsize = INOPB(fs) * sizeof (struct dinode);
1100 	inum = 0;
1101 	while (inum < fs->fs_ipg) {
1102 		bno = itod(fs, inum);
1103 		seekaddr = (offset_t)fsbtodb(fs, bno) * DEV_BSIZE;
1104 		if (llseek(fd, seekaddr, 0) == -1 ||
1105 		    read(fd, (char *)dibuf, bsize) != bsize) {
1106 			(void) close(fd);
1107 			return (NOT_RANDOMIZED);
1108 		}
1109 		for (dip = dibuf; dip < &dibuf[INOPB(fs)]; dip++) {
1110 			if (dip->di_gen != 0) {
1111 				(void) close(fd);
1112 				return (RANDOMIZED);
1113 			}
1114 			inum++;
1115 		}
1116 	}
1117 	(void) close(fd);
1118 	return (NOT_RANDOMIZED);
1119 }
1120 
1121 static void
1122 usage(void)
1123 {
1124 	(void) fprintf(stderr, gettext(
1125 	    "usage: newfs [ -v ] [ mkfs-options ] raw-special-device\n"));
1126 	(void) fprintf(stderr, gettext("where mkfs-options are:\n"));
1127 	(void) fprintf(stderr, gettext(
1128 	    "\t-N do not create file system, just print out parameters\n"));
1129 	(void) fprintf(stderr, gettext(
1130 "\t-T configure file system for eventual growth to over a terabyte\n"));
1131 	(void) fprintf(stderr, gettext("\t-s file system size (sectors)\n"));
1132 	(void) fprintf(stderr, gettext("\t-b block size\n"));
1133 	(void) fprintf(stderr, gettext("\t-f frag size\n"));
1134 	(void) fprintf(stderr, gettext("\t-t tracks/cylinder\n"));
1135 	(void) fprintf(stderr, gettext("\t-c cylinders/group\n"));
1136 	(void) fprintf(stderr, gettext("\t-m minimum free space %%\n"));
1137 	(void) fprintf(stderr, gettext(
1138 	    "\t-o optimization preference (`space' or `time')\n"));
1139 	(void) fprintf(stderr, gettext("\t-r revolutions/minute\n"));
1140 	(void) fprintf(stderr, gettext("\t-i number of bytes per inode\n"));
1141 	(void) fprintf(stderr, gettext(
1142 	    "\t-a number of alternates per cylinder\n"));
1143 	(void) fprintf(stderr, gettext("\t-C maxcontig\n"));
1144 	(void) fprintf(stderr, gettext("\t-d rotational delay\n"));
1145 	(void) fprintf(stderr, gettext(
1146 	    "\t-n number of rotational positions\n"));
1147 	(void) fprintf(stderr, gettext(
1148 "\t-S print a textual version of the calculated superblock to stdout\n"));
1149 	(void) fprintf(stderr, gettext(
1150 "\t-B dump a binary version of the calculated superblock to stdout\n"));
1151 }
1152 
1153 /*
1154  * Error-detecting version of atoi(3).  Adapted from mkfs' number().
1155  */
1156 static unsigned int
1157 number(char *param, char *value, int flags, int def_value)
1158 {
1159 	char *cs;
1160 	int n;
1161 	int cut = INT_MAX / 10;    /* limit to avoid overflow */
1162 	int minus = 0;
1163 
1164 	cs = value;
1165 	if (*cs == '-') {
1166 		minus = 1;
1167 		cs += 1;
1168 	}
1169 	if ((*cs < '0') || (*cs > '9')) {
1170 		goto bail_out;
1171 	}
1172 	n = 0;
1173 	while ((*cs >= '0') && (*cs <= '9') && (n <= cut)) {
1174 		n = n*10 + *cs++ - '0';
1175 	}
1176 	if (minus)
1177 		n = -n;
1178 	for (;;) {
1179 		switch (*cs++) {
1180 		case '\0':
1181 			return (n);
1182 
1183 		case '0': case '1': case '2': case '3': case '4':
1184 		case '5': case '6': case '7': case '8': case '9':
1185 			(void) fprintf(stderr, gettext(
1186 			    "newfs: value for %s overflowed, using %d\n"),
1187 			    param, def_value);
1188 			return (def_value);
1189 
1190 		case '%':
1191 			if (flags & NR_PERCENT)
1192 				break;
1193 			/* FALLTHROUGH */
1194 
1195 		default:
1196 bail_out:
1197 			fatal(gettext("bad numeric arg for %s: \"%s\""),
1198 			    param, value);
1199 
1200 		}
1201 	}
1202 	/* NOTREACHED */
1203 }
1204 
1205 /*
1206  * Error-detecting version of atoi(3).  Adapted from mkfs' number().
1207  */
1208 static int64_t
1209 number64(char *param, char *value, int flags, int64_t def_value)
1210 {
1211 	char *cs;
1212 	int64_t n;
1213 	int64_t cut = FS_SIZE_UPPER_LIMIT/ 10;    /* limit to avoid overflow */
1214 	int minus = 0;
1215 
1216 	cs = value;
1217 	if (*cs == '-') {
1218 		minus = 1;
1219 		cs += 1;
1220 	}
1221 	if ((*cs < '0') || (*cs > '9')) {
1222 		goto bail_out;
1223 	}
1224 	n = 0;
1225 	while ((*cs >= '0') && (*cs <= '9') && (n <= cut)) {
1226 		n = n*10 + *cs++ - '0';
1227 	}
1228 	if (minus)
1229 		n = -n;
1230 	for (;;) {
1231 		switch (*cs++) {
1232 		case '\0':
1233 			return (n);
1234 
1235 		case '0': case '1': case '2': case '3': case '4':
1236 		case '5': case '6': case '7': case '8': case '9':
1237 			(void) fprintf(stderr, gettext(
1238 			    "newfs: value for %s overflowed, using %d\n"),
1239 			    param, def_value);
1240 			return (def_value);
1241 
1242 		case '%':
1243 			if (flags & NR_PERCENT)
1244 				break;
1245 			/* FALLTHROUGH */
1246 
1247 		default:
1248 bail_out:
1249 			fatal(gettext("bad numeric arg for %s: \"%s\""),
1250 			    param, value);
1251 
1252 		}
1253 	}
1254 	/* NOTREACHED */
1255 }
1256