xref: /freebsd/sys/contrib/openzfs/cmd/zpool/os/linux/zpool_vdev_os.c (revision 071ab5a1f3cbfd29c8fbec27f7e619418adaf074)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or https://opensource.org/licenses/CDDL-1.0.
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 /*
24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2013, 2018 by Delphix. All rights reserved.
26  * Copyright (c) 2016, 2017 Intel Corporation.
27  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>.
28  */
29 
30 /*
31  * Functions to convert between a list of vdevs and an nvlist representing the
32  * configuration.  Each entry in the list can be one of:
33  *
34  * 	Device vdevs
35  * 		disk=(path=..., devid=...)
36  * 		file=(path=...)
37  *
38  * 	Group vdevs
39  * 		raidz[1|2]=(...)
40  * 		mirror=(...)
41  *
42  * 	Hot spares
43  *
44  * While the underlying implementation supports it, group vdevs cannot contain
45  * other group vdevs.  All userland verification of devices is contained within
46  * this file.  If successful, the nvlist returned can be passed directly to the
47  * kernel; we've done as much verification as possible in userland.
48  *
49  * Hot spares are a special case, and passed down as an array of disk vdevs, at
50  * the same level as the root of the vdev tree.
51  *
52  * The only function exported by this file is 'make_root_vdev'.  The
53  * function performs several passes:
54  *
55  * 	1. Construct the vdev specification.  Performs syntax validation and
56  *         makes sure each device is valid.
57  * 	2. Check for devices in use.  Using libblkid to make sure that no
58  *         devices are also in use.  Some can be overridden using the 'force'
59  *         flag, others cannot.
60  * 	3. Check for replication errors if the 'force' flag is not specified.
61  *         validates that the replication level is consistent across the
62  *         entire pool.
63  * 	4. Call libzfs to label any whole disks with an EFI label.
64  */
65 
66 #include <assert.h>
67 #include <ctype.h>
68 #include <errno.h>
69 #include <fcntl.h>
70 #include <libintl.h>
71 #include <libnvpair.h>
72 #include <libzutil.h>
73 #include <limits.h>
74 #include <sys/spa.h>
75 #include <stdio.h>
76 #include <string.h>
77 #include <unistd.h>
78 #include "zpool_util.h"
79 #include <sys/zfs_context.h>
80 
81 #include <scsi/scsi.h>
82 #include <scsi/sg.h>
83 #include <sys/efi_partition.h>
84 #include <sys/stat.h>
85 #include <sys/mntent.h>
86 #include <uuid/uuid.h>
87 #include <blkid/blkid.h>
88 
89 typedef struct vdev_disk_db_entry
90 {
91 	/* 24 byte name + 1 byte NULL terminator to make GCC happy */
92 	char id[25];
93 	int sector_size;
94 } vdev_disk_db_entry_t;
95 
96 /*
97  * Database of block devices that lie about physical sector sizes.  The
98  * identification string must be precisely 24 characters to avoid false
99  * negatives
100  */
101 static vdev_disk_db_entry_t vdev_disk_database[] = {
102 	{"ATA     ADATA SSD S396 3", 8192},
103 	{"ATA     APPLE SSD SM128E", 8192},
104 	{"ATA     APPLE SSD SM256E", 8192},
105 	{"ATA     APPLE SSD SM512E", 8192},
106 	{"ATA     APPLE SSD SM768E", 8192},
107 	{"ATA     C400-MTFDDAC064M", 8192},
108 	{"ATA     C400-MTFDDAC128M", 8192},
109 	{"ATA     C400-MTFDDAC256M", 8192},
110 	{"ATA     C400-MTFDDAC512M", 8192},
111 	{"ATA     Corsair Force 3 ", 8192},
112 	{"ATA     Corsair Force GS", 8192},
113 	{"ATA     INTEL SSDSA2CT04", 8192},
114 	{"ATA     INTEL SSDSA2BZ10", 8192},
115 	{"ATA     INTEL SSDSA2BZ20", 8192},
116 	{"ATA     INTEL SSDSA2BZ30", 8192},
117 	{"ATA     INTEL SSDSA2CW04", 8192},
118 	{"ATA     INTEL SSDSA2CW08", 8192},
119 	{"ATA     INTEL SSDSA2CW12", 8192},
120 	{"ATA     INTEL SSDSA2CW16", 8192},
121 	{"ATA     INTEL SSDSA2CW30", 8192},
122 	{"ATA     INTEL SSDSA2CW60", 8192},
123 	{"ATA     INTEL SSDSC2CT06", 8192},
124 	{"ATA     INTEL SSDSC2CT12", 8192},
125 	{"ATA     INTEL SSDSC2CT18", 8192},
126 	{"ATA     INTEL SSDSC2CT24", 8192},
127 	{"ATA     INTEL SSDSC2CW06", 8192},
128 	{"ATA     INTEL SSDSC2CW12", 8192},
129 	{"ATA     INTEL SSDSC2CW18", 8192},
130 	{"ATA     INTEL SSDSC2CW24", 8192},
131 	{"ATA     INTEL SSDSC2CW48", 8192},
132 	{"ATA     KINGSTON SH100S3", 8192},
133 	{"ATA     KINGSTON SH103S3", 8192},
134 	{"ATA     M4-CT064M4SSD2  ", 8192},
135 	{"ATA     M4-CT128M4SSD2  ", 8192},
136 	{"ATA     M4-CT256M4SSD2  ", 8192},
137 	{"ATA     M4-CT512M4SSD2  ", 8192},
138 	{"ATA     OCZ-AGILITY2    ", 8192},
139 	{"ATA     OCZ-AGILITY3    ", 8192},
140 	{"ATA     OCZ-VERTEX2 3.5 ", 8192},
141 	{"ATA     OCZ-VERTEX3     ", 8192},
142 	{"ATA     OCZ-VERTEX3 LT  ", 8192},
143 	{"ATA     OCZ-VERTEX3 MI  ", 8192},
144 	{"ATA     OCZ-VERTEX4     ", 8192},
145 	{"ATA     SAMSUNG MZ7WD120", 8192},
146 	{"ATA     SAMSUNG MZ7WD240", 8192},
147 	{"ATA     SAMSUNG MZ7WD480", 8192},
148 	{"ATA     SAMSUNG MZ7WD960", 8192},
149 	{"ATA     SAMSUNG SSD 830 ", 8192},
150 	{"ATA     Samsung SSD 840 ", 8192},
151 	{"ATA     SanDisk SSD U100", 8192},
152 	{"ATA     TOSHIBA THNSNH06", 8192},
153 	{"ATA     TOSHIBA THNSNH12", 8192},
154 	{"ATA     TOSHIBA THNSNH25", 8192},
155 	{"ATA     TOSHIBA THNSNH51", 8192},
156 	{"ATA     APPLE SSD TS064C", 4096},
157 	{"ATA     APPLE SSD TS128C", 4096},
158 	{"ATA     APPLE SSD TS256C", 4096},
159 	{"ATA     APPLE SSD TS512C", 4096},
160 	{"ATA     INTEL SSDSA2M040", 4096},
161 	{"ATA     INTEL SSDSA2M080", 4096},
162 	{"ATA     INTEL SSDSA2M160", 4096},
163 	{"ATA     INTEL SSDSC2MH12", 4096},
164 	{"ATA     INTEL SSDSC2MH25", 4096},
165 	{"ATA     OCZ CORE_SSD    ", 4096},
166 	{"ATA     OCZ-VERTEX      ", 4096},
167 	{"ATA     SAMSUNG MCCOE32G", 4096},
168 	{"ATA     SAMSUNG MCCOE64G", 4096},
169 	{"ATA     SAMSUNG SSD PM80", 4096},
170 	/* Flash drives optimized for 4KB IOs on larger pages */
171 	{"ATA     INTEL SSDSC2BA10", 4096},
172 	{"ATA     INTEL SSDSC2BA20", 4096},
173 	{"ATA     INTEL SSDSC2BA40", 4096},
174 	{"ATA     INTEL SSDSC2BA80", 4096},
175 	{"ATA     INTEL SSDSC2BB08", 4096},
176 	{"ATA     INTEL SSDSC2BB12", 4096},
177 	{"ATA     INTEL SSDSC2BB16", 4096},
178 	{"ATA     INTEL SSDSC2BB24", 4096},
179 	{"ATA     INTEL SSDSC2BB30", 4096},
180 	{"ATA     INTEL SSDSC2BB40", 4096},
181 	{"ATA     INTEL SSDSC2BB48", 4096},
182 	{"ATA     INTEL SSDSC2BB60", 4096},
183 	{"ATA     INTEL SSDSC2BB80", 4096},
184 	{"ATA     INTEL SSDSC2BW24", 4096},
185 	{"ATA     INTEL SSDSC2BW48", 4096},
186 	{"ATA     INTEL SSDSC2BP24", 4096},
187 	{"ATA     INTEL SSDSC2BP48", 4096},
188 	{"NA      SmrtStorSDLKAE9W", 4096},
189 	{"NVMe    Amazon EC2 NVMe ", 4096},
190 	/* Imported from Open Solaris */
191 	{"ATA     MARVELL SD88SA02", 4096},
192 	/* Advanced format Hard drives */
193 	{"ATA     Hitachi HDS5C303", 4096},
194 	{"ATA     SAMSUNG HD204UI ", 4096},
195 	{"ATA     ST2000DL004 HD20", 4096},
196 	{"ATA     WDC WD10EARS-00M", 4096},
197 	{"ATA     WDC WD10EARS-00S", 4096},
198 	{"ATA     WDC WD10EARS-00Z", 4096},
199 	{"ATA     WDC WD15EARS-00M", 4096},
200 	{"ATA     WDC WD15EARS-00S", 4096},
201 	{"ATA     WDC WD15EARS-00Z", 4096},
202 	{"ATA     WDC WD20EARS-00M", 4096},
203 	{"ATA     WDC WD20EARS-00S", 4096},
204 	{"ATA     WDC WD20EARS-00Z", 4096},
205 	{"ATA     WDC WD1600BEVT-0", 4096},
206 	{"ATA     WDC WD2500BEVT-0", 4096},
207 	{"ATA     WDC WD3200BEVT-0", 4096},
208 	{"ATA     WDC WD5000BEVT-0", 4096},
209 };
210 
211 
212 #define	INQ_REPLY_LEN	96
213 #define	INQ_CMD_LEN	6
214 
215 static const int vdev_disk_database_size =
216 	sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
217 
218 boolean_t
check_sector_size_database(char * path,int * sector_size)219 check_sector_size_database(char *path, int *sector_size)
220 {
221 	unsigned char inq_buff[INQ_REPLY_LEN];
222 	unsigned char sense_buffer[32];
223 	unsigned char inq_cmd_blk[INQ_CMD_LEN] =
224 	    {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
225 	sg_io_hdr_t io_hdr;
226 	int error;
227 	int fd;
228 	int i;
229 
230 	/* Prepare INQUIRY command */
231 	memset(&io_hdr, 0, sizeof (sg_io_hdr_t));
232 	io_hdr.interface_id = 'S';
233 	io_hdr.cmd_len = sizeof (inq_cmd_blk);
234 	io_hdr.mx_sb_len = sizeof (sense_buffer);
235 	io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
236 	io_hdr.dxfer_len = INQ_REPLY_LEN;
237 	io_hdr.dxferp = inq_buff;
238 	io_hdr.cmdp = inq_cmd_blk;
239 	io_hdr.sbp = sense_buffer;
240 	io_hdr.timeout = 10;		/* 10 milliseconds is ample time */
241 
242 	if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
243 		return (B_FALSE);
244 
245 	error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
246 
247 	(void) close(fd);
248 
249 	if (error < 0)
250 		return (B_FALSE);
251 
252 	if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
253 		return (B_FALSE);
254 
255 	for (i = 0; i < vdev_disk_database_size; i++) {
256 		if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
257 			continue;
258 
259 		*sector_size = vdev_disk_database[i].sector_size;
260 		return (B_TRUE);
261 	}
262 
263 	return (B_FALSE);
264 }
265 
266 static int
check_slice(const char * path,blkid_cache cache,int force,boolean_t isspare)267 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
268 {
269 	int err;
270 	char *value;
271 
272 	/* No valid type detected device is safe to use */
273 	value = blkid_get_tag_value(cache, "TYPE", path);
274 	if (value == NULL)
275 		return (0);
276 
277 	/*
278 	 * If libblkid detects a ZFS device, we check the device
279 	 * using check_file() to see if it's safe.  The one safe
280 	 * case is a spare device shared between multiple pools.
281 	 */
282 	if (strcmp(value, "zfs_member") == 0) {
283 		err = check_file(path, force, isspare);
284 	} else {
285 		if (force) {
286 			err = 0;
287 		} else {
288 			err = -1;
289 			vdev_error(gettext("%s contains a filesystem of "
290 			    "type '%s'\n"), path, value);
291 		}
292 	}
293 
294 	free(value);
295 
296 	return (err);
297 }
298 
299 /*
300  * Validate that a disk including all partitions are safe to use.
301  *
302  * For EFI labeled disks this can done relatively easily with the libefi
303  * library.  The partition numbers are extracted from the label and used
304  * to generate the expected /dev/ paths.  Each partition can then be
305  * checked for conflicts.
306  *
307  * For non-EFI labeled disks (MBR/EBR/etc) the same process is possible
308  * but due to the lack of a readily available libraries this scanning is
309  * not implemented.  Instead only the device path as given is checked.
310  */
311 static int
check_disk(const char * path,blkid_cache cache,int force,boolean_t isspare,boolean_t iswholedisk)312 check_disk(const char *path, blkid_cache cache, int force,
313     boolean_t isspare, boolean_t iswholedisk)
314 {
315 	struct dk_gpt *vtoc;
316 	char slice_path[MAXPATHLEN];
317 	int err = 0;
318 	int fd, i;
319 	int flags = O_RDONLY|O_DIRECT;
320 
321 	if (!iswholedisk)
322 		return (check_slice(path, cache, force, isspare));
323 
324 	/* only spares can be shared, other devices require exclusive access */
325 	if (!isspare)
326 		flags |= O_EXCL;
327 
328 	if ((fd = open(path, flags)) < 0) {
329 		char *value = blkid_get_tag_value(cache, "TYPE", path);
330 		(void) fprintf(stderr, gettext("%s is in use and contains "
331 		    "a %s filesystem.\n"), path, value ? value : "unknown");
332 		free(value);
333 		return (-1);
334 	}
335 
336 	/*
337 	 * Expected to fail for non-EFI labeled disks.  Just check the device
338 	 * as given and do not attempt to detect and scan partitions.
339 	 */
340 	err = efi_alloc_and_read(fd, &vtoc);
341 	if (err) {
342 		(void) close(fd);
343 		return (check_slice(path, cache, force, isspare));
344 	}
345 
346 	/*
347 	 * The primary efi partition label is damaged however the secondary
348 	 * label at the end of the device is intact.  Rather than use this
349 	 * label we should play it safe and treat this as a non efi device.
350 	 */
351 	if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
352 		efi_free(vtoc);
353 		(void) close(fd);
354 
355 		if (force) {
356 			/* Partitions will now be created using the backup */
357 			return (0);
358 		} else {
359 			vdev_error(gettext("%s contains a corrupt primary "
360 			    "EFI label.\n"), path);
361 			return (-1);
362 		}
363 	}
364 
365 	for (i = 0; i < vtoc->efi_nparts; i++) {
366 
367 		if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
368 		    uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
369 			continue;
370 
371 		if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
372 			(void) snprintf(slice_path, sizeof (slice_path),
373 			    "%s%s%d", path, "-part", i+1);
374 		else
375 			(void) snprintf(slice_path, sizeof (slice_path),
376 			    "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
377 			    "p" : "", i+1);
378 
379 		err = check_slice(slice_path, cache, force, isspare);
380 		if (err)
381 			break;
382 	}
383 
384 	efi_free(vtoc);
385 	(void) close(fd);
386 
387 	return (err);
388 }
389 
390 int
check_device(const char * path,boolean_t force,boolean_t isspare,boolean_t iswholedisk)391 check_device(const char *path, boolean_t force,
392     boolean_t isspare, boolean_t iswholedisk)
393 {
394 	blkid_cache cache;
395 	int error;
396 
397 	error = blkid_get_cache(&cache, NULL);
398 	if (error != 0) {
399 		(void) fprintf(stderr, gettext("unable to access the blkid "
400 		    "cache.\n"));
401 		return (-1);
402 	}
403 
404 	error = check_disk(path, cache, force, isspare, iswholedisk);
405 	blkid_put_cache(cache);
406 
407 	return (error);
408 }
409 
410 void
after_zpool_upgrade(zpool_handle_t * zhp)411 after_zpool_upgrade(zpool_handle_t *zhp)
412 {
413 	(void) zhp;
414 }
415 
416 int
check_file(const char * file,boolean_t force,boolean_t isspare)417 check_file(const char *file, boolean_t force, boolean_t isspare)
418 {
419 	return (check_file_generic(file, force, isspare));
420 }
421 
422 /*
423  * Read from a sysfs file and return an allocated string.  Removes
424  * the newline from the end of the string if there is one.
425  *
426  * Returns a string on success (which must be freed), or NULL on error.
427  */
zpool_sysfs_gets(char * path)428 static char *zpool_sysfs_gets(char *path)
429 {
430 	int fd;
431 	struct stat statbuf;
432 	char *buf = NULL;
433 	ssize_t count = 0;
434 	fd = open(path, O_RDONLY);
435 	if (fd < 0)
436 		return (NULL);
437 
438 	if (fstat(fd, &statbuf) != 0) {
439 		close(fd);
440 		return (NULL);
441 	}
442 
443 	buf = calloc(statbuf.st_size + 1, sizeof (*buf));
444 	if (buf == NULL) {
445 		close(fd);
446 		return (NULL);
447 	}
448 
449 	/*
450 	 * Note, we can read less bytes than st_size, and that's ok.  Sysfs
451 	 * files will report their size is 4k even if they only return a small
452 	 * string.
453 	 */
454 	count = read(fd, buf, statbuf.st_size);
455 	if (count < 0) {
456 		/* Error doing read() or we overran the buffer */
457 		close(fd);
458 		free(buf);
459 		return (NULL);
460 	}
461 
462 	/* Remove trailing newline */
463 	if (count > 0 && buf[count - 1] == '\n')
464 		buf[count - 1] = 0;
465 
466 	close(fd);
467 
468 	return (buf);
469 }
470 
471 /*
472  * Write a string to a sysfs file.
473  *
474  * Returns 0 on success, non-zero otherwise.
475  */
zpool_sysfs_puts(char * path,char * str)476 static int zpool_sysfs_puts(char *path, char *str)
477 {
478 	FILE *file;
479 
480 	file = fopen(path, "w");
481 	if (!file) {
482 		return (-1);
483 	}
484 
485 	if (fputs(str, file) < 0) {
486 		fclose(file);
487 		return (-2);
488 	}
489 	fclose(file);
490 	return (0);
491 }
492 
493 /* Given a vdev nvlist_t, rescan its enclosure sysfs path */
494 static void
rescan_vdev_config_dev_sysfs_path(nvlist_t * vdev_nv)495 rescan_vdev_config_dev_sysfs_path(nvlist_t *vdev_nv)
496 {
497 	update_vdev_config_dev_sysfs_path(vdev_nv,
498 	    fnvlist_lookup_string(vdev_nv, ZPOOL_CONFIG_PATH),
499 	    ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH);
500 }
501 
502 /*
503  * Given a power string: "on", "off", "1", or "0", return 0 if it's an
504  * off value, 1 if it's an on value, and -1 if the value is unrecognized.
505  */
zpool_power_parse_value(char * str)506 static int zpool_power_parse_value(char *str)
507 {
508 	if ((strcmp(str, "off") == 0) || (strcmp(str, "0") == 0))
509 		return (0);
510 
511 	if ((strcmp(str, "on") == 0) || (strcmp(str, "1") == 0))
512 		return (1);
513 
514 	return (-1);
515 }
516 
517 /*
518  * Given a vdev string return an allocated string containing the sysfs path to
519  * its power control file.  Also do a check if the power control file really
520  * exists and has correct permissions.
521  *
522  * Example returned strings:
523  *
524  * /sys/class/enclosure/0:0:122:0/10/power_status
525  * /sys/bus/pci/slots/10/power
526  *
527  * Returns allocated string on success (which must be freed), NULL on failure.
528  */
529 static char *
zpool_power_sysfs_path(zpool_handle_t * zhp,char * vdev)530 zpool_power_sysfs_path(zpool_handle_t *zhp, char *vdev)
531 {
532 	const char *enc_sysfs_dir = NULL;
533 	char *path = NULL;
534 	nvlist_t *vdev_nv = zpool_find_vdev(zhp, vdev, NULL, NULL, NULL);
535 
536 	if (vdev_nv == NULL) {
537 		return (NULL);
538 	}
539 
540 	/* Make sure we're getting the updated enclosure sysfs path */
541 	rescan_vdev_config_dev_sysfs_path(vdev_nv);
542 
543 	if (nvlist_lookup_string(vdev_nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
544 	    &enc_sysfs_dir) != 0) {
545 		return (NULL);
546 	}
547 
548 	if (asprintf(&path, "%s/power_status", enc_sysfs_dir) == -1)
549 		return (NULL);
550 
551 	if (access(path, W_OK) != 0) {
552 		free(path);
553 		path = NULL;
554 		/* No HDD 'power_control' file, maybe it's NVMe? */
555 		if (asprintf(&path, "%s/power", enc_sysfs_dir) == -1) {
556 			return (NULL);
557 		}
558 
559 		if (access(path, R_OK | W_OK) != 0) {
560 			/* Not NVMe either */
561 			free(path);
562 			return (NULL);
563 		}
564 	}
565 
566 	return (path);
567 }
568 
569 /*
570  * Given a path to a sysfs power control file, return B_TRUE if you should use
571  * "on/off" words to control it, or B_FALSE otherwise ("0/1" to control).
572  */
573 static boolean_t
zpool_power_use_word(char * sysfs_path)574 zpool_power_use_word(char *sysfs_path)
575 {
576 	if (strcmp(&sysfs_path[strlen(sysfs_path) - strlen("power_status")],
577 	    "power_status") == 0) {
578 		return (B_TRUE);
579 	}
580 	return (B_FALSE);
581 }
582 
583 /*
584  * Check the sysfs power control value for a vdev.
585  *
586  * Returns:
587  *  0 - Power is off
588  *  1 - Power is on
589  * -1 - Error or unsupported
590  */
591 int
zpool_power_current_state(zpool_handle_t * zhp,char * vdev)592 zpool_power_current_state(zpool_handle_t *zhp, char *vdev)
593 {
594 	char *val;
595 	int rc;
596 
597 	char *path = zpool_power_sysfs_path(zhp, vdev);
598 	if (path == NULL)
599 		return (-1);
600 
601 	val = zpool_sysfs_gets(path);
602 	if (val == NULL) {
603 		free(path);
604 		return (-1);
605 	}
606 
607 	rc = zpool_power_parse_value(val);
608 	free(val);
609 	free(path);
610 	return (rc);
611 }
612 
613 /*
614  * Turn on or off the slot to a device
615  *
616  * Device path is the full path to the device (like /dev/sda or /dev/sda1).
617  *
618  * Return code:
619  * 0:		Success
620  * ENOTSUP:	Power control not supported for OS
621  * EBADSLT:	Couldn't read current power state
622  * ENOENT:	No sysfs path to power control
623  * EIO:	Couldn't write sysfs power value
624  * EBADE:	Sysfs power value didn't change
625  */
626 int
zpool_power(zpool_handle_t * zhp,char * vdev,boolean_t turn_on)627 zpool_power(zpool_handle_t *zhp, char *vdev, boolean_t turn_on)
628 {
629 	char *sysfs_path;
630 	const char *val;
631 	int rc;
632 	int timeout_ms;
633 
634 	rc = zpool_power_current_state(zhp, vdev);
635 	if (rc == -1) {
636 		return (EBADSLT);
637 	}
638 
639 	/* Already correct value? */
640 	if (rc == (int)turn_on)
641 		return (0);
642 
643 	sysfs_path = zpool_power_sysfs_path(zhp, vdev);
644 	if (sysfs_path == NULL)
645 		return (ENOENT);
646 
647 	if (zpool_power_use_word(sysfs_path)) {
648 		val = turn_on ? "on" : "off";
649 	} else {
650 		val = turn_on ? "1" : "0";
651 	}
652 
653 	rc = zpool_sysfs_puts(sysfs_path, (char *)val);
654 
655 	free(sysfs_path);
656 	if (rc != 0) {
657 		return (EIO);
658 	}
659 
660 	/*
661 	 * Wait up to 30 seconds for sysfs power value to change after
662 	 * writing it.
663 	 */
664 	timeout_ms = zpool_getenv_int("ZPOOL_POWER_ON_SLOT_TIMEOUT_MS", 30000);
665 	for (int i = 0; i < MAX(1, timeout_ms / 200); i++) {
666 		rc = zpool_power_current_state(zhp, vdev);
667 		if (rc == (int)turn_on)
668 			return (0);	/* success */
669 
670 		fsleep(0.200);	/* 200ms */
671 	}
672 
673 	/* sysfs value never changed */
674 	return (EBADE);
675 }
676