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 */
26
27 #include <ctype.h>
28 #include <dirent.h>
29 #include <fcntl.h>
30 #include <stdio.h>
31 #include <stdlib.h>
32 #include <string.h>
33 #include <sys/efi_partition.h>
34
35 #ifdef HAVE_LIBUDEV
36 #include <libudev.h>
37 #endif
38
39 #include <libzutil.h>
40
41 /*
42 * Append partition suffix to an otherwise fully qualified device path.
43 * This is used to generate the name the full path as its stored in
44 * ZPOOL_CONFIG_PATH for whole disk devices. On success the new length
45 * of 'path' will be returned on error a negative value is returned.
46 */
47 int
zfs_append_partition(char * path,size_t max_len)48 zfs_append_partition(char *path, size_t max_len)
49 {
50 int len = strlen(path);
51
52 if ((strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0) ||
53 (strncmp(path, ZVOL_ROOT, strlen(ZVOL_ROOT)) == 0)) {
54 if (len + 6 >= max_len)
55 return (-1);
56
57 (void) strcat(path, "-part1");
58 len += 6;
59 } else {
60 if (len + 2 >= max_len)
61 return (-1);
62
63 if (isdigit(path[len-1])) {
64 (void) strcat(path, "p1");
65 len += 2;
66 } else {
67 (void) strcat(path, "1");
68 len += 1;
69 }
70 }
71
72 return (len);
73 }
74
75 /*
76 * Remove partition suffix from a vdev path. Partition suffixes may take three
77 * forms: "-partX", "pX", or "X", where X is a string of digits. The second
78 * case only occurs when the suffix is preceded by a digit, i.e. "md0p0" The
79 * third case only occurs when preceded by a string matching the regular
80 * expression "^([hsv]|xv)d[a-z]+", i.e. a scsi, ide, virtio or xen disk.
81 *
82 * caller must free the returned string
83 */
84 char *
zfs_strip_partition(const char * path)85 zfs_strip_partition(const char *path)
86 {
87 char *tmp = strdup(path);
88 char *part = NULL, *d = NULL;
89 if (!tmp)
90 return (NULL);
91
92 if ((part = strstr(tmp, "-part")) && part != tmp) {
93 d = part + 5;
94 } else if ((part = strrchr(tmp, 'p')) &&
95 part > tmp + 1 && isdigit(*(part-1))) {
96 d = part + 1;
97 } else if ((tmp[0] == 'h' || tmp[0] == 's' || tmp[0] == 'v') &&
98 tmp[1] == 'd') {
99 for (d = &tmp[2]; isalpha(*d); part = ++d) { }
100 } else if (strncmp("xvd", tmp, 3) == 0) {
101 for (d = &tmp[3]; isalpha(*d); part = ++d) { }
102 }
103 if (part && d && *d != '\0') {
104 for (; isdigit(*d); d++) { }
105 if (*d == '\0')
106 *part = '\0';
107 }
108
109 return (tmp);
110 }
111
112 /*
113 * Same as zfs_strip_partition, but allows "/dev/" to be in the pathname
114 *
115 * path: /dev/sda1
116 * returns: /dev/sda
117 *
118 * Returned string must be freed.
119 */
120 static char *
zfs_strip_partition_path(const char * path)121 zfs_strip_partition_path(const char *path)
122 {
123 char *newpath = strdup(path);
124 char *sd_offset;
125 char *new_sd;
126
127 if (!newpath)
128 return (NULL);
129
130 /* Point to "sda1" part of "/dev/sda1" */
131 sd_offset = strrchr(newpath, '/') + 1;
132
133 /* Get our new name "sda" */
134 new_sd = zfs_strip_partition(sd_offset);
135 if (!new_sd) {
136 free(newpath);
137 return (NULL);
138 }
139
140 /* Paste the "sda" where "sda1" was */
141 strlcpy(sd_offset, new_sd, strlen(sd_offset) + 1);
142
143 /* Free temporary "sda" */
144 free(new_sd);
145
146 return (newpath);
147 }
148
149 /*
150 * Strip the unwanted portion of a device path.
151 */
152 const char *
zfs_strip_path(const char * path)153 zfs_strip_path(const char *path)
154 {
155 size_t spath_count;
156 const char *const *spaths = zpool_default_search_paths(&spath_count);
157
158 for (size_t i = 0; i < spath_count; ++i)
159 if (strncmp(path, spaths[i], strlen(spaths[i])) == 0 &&
160 path[strlen(spaths[i])] == '/')
161 return (path + strlen(spaths[i]) + 1);
162
163 return (path);
164 }
165
166 /*
167 * Read the contents of a sysfs file into an allocated buffer and remove the
168 * last newline.
169 *
170 * This is useful for reading sysfs files that return a single string. Return
171 * an allocated string pointer on success, NULL otherwise. Returned buffer
172 * must be freed by the user.
173 */
174 static char *
zfs_read_sysfs_file(char * filepath)175 zfs_read_sysfs_file(char *filepath)
176 {
177 char buf[4096]; /* all sysfs files report 4k size */
178 char *str = NULL;
179
180 FILE *fp = fopen(filepath, "r");
181 if (fp == NULL) {
182 return (NULL);
183 }
184 if (fgets(buf, sizeof (buf), fp) == buf) {
185 /* success */
186
187 /* Remove the last newline (if any) */
188 size_t len = strlen(buf);
189 if (buf[len - 1] == '\n') {
190 buf[len - 1] = '\0';
191 }
192 str = strdup(buf);
193 }
194
195 fclose(fp);
196
197 return (str);
198 }
199
200 /*
201 * Given a dev name like "nvme0n1", return the full PCI slot sysfs path to
202 * the drive (in /sys/bus/pci/slots).
203 *
204 * For example:
205 * dev: "nvme0n1"
206 * returns: "/sys/bus/pci/slots/0"
207 *
208 * 'dev' must be an NVMe device.
209 *
210 * Returned string must be freed. Returns NULL on error or no sysfs path.
211 */
212 static char *
zfs_get_pci_slots_sys_path(const char * dev_name)213 zfs_get_pci_slots_sys_path(const char *dev_name)
214 {
215 DIR *dp = NULL;
216 struct dirent *ep;
217 char *address1 = NULL;
218 char *address2 = NULL;
219 char *path = NULL;
220 char buf[MAXPATHLEN];
221 char *tmp;
222
223 /* If they preface 'dev' with a path (like "/dev") then strip it off */
224 tmp = strrchr(dev_name, '/');
225 if (tmp != NULL)
226 dev_name = tmp + 1; /* +1 since we want the chr after '/' */
227
228 if (strncmp("nvme", dev_name, 4) != 0)
229 return (NULL);
230
231 (void) snprintf(buf, sizeof (buf), "/sys/block/%s/device/address",
232 dev_name);
233
234 address1 = zfs_read_sysfs_file(buf);
235 if (!address1)
236 return (NULL);
237
238 /*
239 * /sys/block/nvme0n1/device/address format will
240 * be "0000:01:00.0" while /sys/bus/pci/slots/0/address will be
241 * "0000:01:00". Just NULL terminate at the '.' so they match.
242 */
243 tmp = strrchr(address1, '.');
244 if (tmp != NULL)
245 *tmp = '\0';
246
247 dp = opendir("/sys/bus/pci/slots/");
248 if (dp == NULL) {
249 free(address1);
250 return (NULL);
251 }
252
253 /*
254 * Look through all the /sys/bus/pci/slots/ subdirs
255 */
256 while ((ep = readdir(dp))) {
257 /*
258 * We only care about directory names that are a single number.
259 * Sometimes there's other directories like
260 * "/sys/bus/pci/slots/0-3/" in there - skip those.
261 */
262 if (!zfs_isnumber(ep->d_name))
263 continue;
264
265 (void) snprintf(buf, sizeof (buf),
266 "/sys/bus/pci/slots/%s/address", ep->d_name);
267
268 address2 = zfs_read_sysfs_file(buf);
269 if (!address2)
270 continue;
271
272 if (strcmp(address1, address2) == 0) {
273 /* Addresses match, we're all done */
274 free(address2);
275 if (asprintf(&path, "/sys/bus/pci/slots/%s",
276 ep->d_name) == -1) {
277 continue;
278 }
279 break;
280 }
281 free(address2);
282 }
283
284 closedir(dp);
285 free(address1);
286
287 return (path);
288 }
289
290 /*
291 * Given a dev name like "sda", return the full enclosure sysfs path to
292 * the disk. You can also pass in the name with "/dev" prepended
293 * to it (like /dev/sda). This works for both JBODs and NVMe PCI devices.
294 *
295 * For example, disk "sda" in enclosure slot 1:
296 * dev_name: "sda"
297 * returns: "/sys/class/enclosure/1:0:3:0/Slot 1"
298 *
299 * Or:
300 *
301 * dev_name: "nvme0n1"
302 * returns: "/sys/bus/pci/slots/0"
303 *
304 * 'dev' must be a non-devicemapper device.
305 *
306 * Returned string must be freed. Returns NULL on error.
307 */
308 char *
zfs_get_enclosure_sysfs_path(const char * dev_name)309 zfs_get_enclosure_sysfs_path(const char *dev_name)
310 {
311 DIR *dp = NULL;
312 struct dirent *ep;
313 char buf[MAXPATHLEN];
314 char *tmp1 = NULL;
315 char *tmp2 = NULL;
316 char *tmp3 = NULL;
317 char *path = NULL;
318 size_t size;
319 int tmpsize;
320
321 if (dev_name == NULL)
322 return (NULL);
323
324 /* If they preface 'dev' with a path (like "/dev") then strip it off */
325 tmp1 = strrchr(dev_name, '/');
326 if (tmp1 != NULL)
327 dev_name = tmp1 + 1; /* +1 since we want the chr after '/' */
328
329 tmpsize = asprintf(&tmp1, "/sys/block/%s/device", dev_name);
330 if (tmpsize == -1 || tmp1 == NULL) {
331 tmp1 = NULL;
332 goto end;
333 }
334
335 dp = opendir(tmp1);
336 if (dp == NULL)
337 goto end;
338
339 /*
340 * Look though all sysfs entries in /sys/block/<dev>/device for
341 * the enclosure symlink.
342 */
343 while ((ep = readdir(dp))) {
344 /* Ignore everything that's not our enclosure_device link */
345 if (strstr(ep->d_name, "enclosure_device") == NULL)
346 continue;
347
348 if (tmp2 != NULL)
349 free(tmp2);
350 if (asprintf(&tmp2, "%s/%s", tmp1, ep->d_name) == -1) {
351 tmp2 = NULL;
352 break;
353 }
354
355 size = readlink(tmp2, buf, sizeof (buf));
356
357 /* Did readlink fail or crop the link name? */
358 if (size == -1 || size >= sizeof (buf))
359 break;
360
361 /*
362 * We got a valid link. readlink() doesn't terminate strings
363 * so we have to do it.
364 */
365 buf[size] = '\0';
366
367 /*
368 * Our link will look like:
369 *
370 * "../../../../port-11:1:2/..STUFF../enclosure/1:0:3:0/SLOT 1"
371 *
372 * We want to grab the "enclosure/1:0:3:0/SLOT 1" part
373 */
374 tmp3 = strstr(buf, "enclosure");
375 if (tmp3 == NULL)
376 break;
377
378 if (path != NULL)
379 free(path);
380 if (asprintf(&path, "/sys/class/%s", tmp3) == -1) {
381 /* If asprintf() fails, 'path' is undefined */
382 path = NULL;
383 break;
384 }
385 }
386
387 end:
388 free(tmp2);
389 free(tmp1);
390
391 if (dp != NULL)
392 closedir(dp);
393
394 if (!path) {
395 /*
396 * This particular disk isn't in a JBOD. It could be an NVMe
397 * drive. If so, look up the NVMe device's path in
398 * /sys/bus/pci/slots/. Within that directory is a 'attention'
399 * file which controls the NVMe fault LED.
400 */
401 path = zfs_get_pci_slots_sys_path(dev_name);
402 }
403
404 return (path);
405 }
406
407 /*
408 * Allocate and return the underlying device name for a device mapper device.
409 *
410 * For example, dm_name = "/dev/dm-0" could return "/dev/sda". Symlinks to a
411 * DM device (like /dev/disk/by-vdev/A0) are also allowed.
412 *
413 * If the DM device has multiple underlying devices (like with multipath
414 * DM devices), then favor underlying devices that have a symlink back to their
415 * back to their enclosure device in sysfs. This will be useful for the
416 * zedlet scripts that toggle the fault LED.
417 *
418 * Returns an underlying device name, or NULL on error or no match. If dm_name
419 * is not a DM device then return NULL.
420 *
421 * NOTE: The returned name string must be *freed*.
422 */
423 static char *
dm_get_underlying_path(const char * dm_name)424 dm_get_underlying_path(const char *dm_name)
425 {
426 DIR *dp = NULL;
427 struct dirent *ep;
428 char *realp;
429 char *tmp = NULL;
430 char *path = NULL;
431 char *dev_str;
432 char *first_path = NULL;
433 char *enclosure_path;
434
435 if (dm_name == NULL)
436 return (NULL);
437
438 /* dm name may be a symlink (like /dev/disk/by-vdev/A0) */
439 realp = realpath(dm_name, NULL);
440 if (realp == NULL)
441 return (NULL);
442
443 /*
444 * If they preface 'dev' with a path (like "/dev") then strip it off.
445 * We just want the 'dm-N' part.
446 */
447 tmp = strrchr(realp, '/');
448 if (tmp != NULL)
449 dev_str = tmp + 1; /* +1 since we want the chr after '/' */
450 else
451 dev_str = tmp;
452
453 if (asprintf(&tmp, "/sys/block/%s/slaves/", dev_str) == -1) {
454 tmp = NULL;
455 goto end;
456 }
457
458 dp = opendir(tmp);
459 if (dp == NULL)
460 goto end;
461
462 /*
463 * A device-mapper device can have multiple paths to it (multipath).
464 * Favor paths that have a symlink back to their enclosure device.
465 * We have to do this since some enclosures may only provide a symlink
466 * back for one underlying path to a disk and not the other.
467 *
468 * If no paths have links back to their enclosure, then just return the
469 * first path.
470 */
471 while ((ep = readdir(dp))) {
472 if (ep->d_type != DT_DIR) { /* skip "." and ".." dirs */
473 if (!first_path)
474 first_path = strdup(ep->d_name);
475
476 enclosure_path =
477 zfs_get_enclosure_sysfs_path(ep->d_name);
478
479 if (!enclosure_path)
480 continue;
481
482 if (asprintf(&path, "/dev/%s", ep->d_name) == -1)
483 path = NULL;
484 free(enclosure_path);
485 break;
486 }
487 }
488
489 end:
490 if (dp != NULL)
491 closedir(dp);
492 free(tmp);
493 free(realp);
494
495 if (!path && first_path) {
496 /*
497 * None of the underlying paths had a link back to their
498 * enclosure devices. Throw up out hands and return the first
499 * underlying path.
500 */
501 if (asprintf(&path, "/dev/%s", first_path) == -1)
502 path = NULL;
503 }
504
505 free(first_path);
506 return (path);
507 }
508
509 /*
510 * Return B_TRUE if device is a device mapper or multipath device.
511 * Return B_FALSE if not.
512 */
513 boolean_t
zfs_dev_is_dm(const char * dev_name)514 zfs_dev_is_dm(const char *dev_name)
515 {
516
517 char *tmp;
518 tmp = dm_get_underlying_path(dev_name);
519 if (tmp == NULL)
520 return (B_FALSE);
521
522 free(tmp);
523 return (B_TRUE);
524 }
525
526 /*
527 * By "whole disk" we mean an entire physical disk (something we can
528 * label, toggle the write cache on, etc.) as opposed to the full
529 * capacity of a pseudo-device such as lofi or did. We act as if we
530 * are labeling the disk, which should be a pretty good test of whether
531 * it's a viable device or not. Returns B_TRUE if it is and B_FALSE if
532 * it isn't.
533 */
534 boolean_t
zfs_dev_is_whole_disk(const char * dev_name)535 zfs_dev_is_whole_disk(const char *dev_name)
536 {
537 struct dk_gpt *label = NULL;
538 int fd;
539
540 if ((fd = open(dev_name, O_RDONLY | O_DIRECT | O_CLOEXEC)) < 0)
541 return (B_FALSE);
542
543 if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
544 (void) close(fd);
545 return (B_FALSE);
546 }
547
548 efi_free(label);
549 (void) close(fd);
550
551 return (B_TRUE);
552 }
553
554 /*
555 * Lookup the underlying device for a device name
556 *
557 * Often you'll have a symlink to a device, a partition device,
558 * or a multipath device, and want to look up the underlying device.
559 * This function returns the underlying device name. If the device
560 * name is already the underlying device, then just return the same
561 * name. If the device is a DM device with multiple underlying devices
562 * then return the first one.
563 *
564 * For example:
565 *
566 * 1. /dev/disk/by-id/ata-QEMU_HARDDISK_QM00001 -> ../../sda
567 * dev_name: /dev/disk/by-id/ata-QEMU_HARDDISK_QM00001
568 * returns: /dev/sda
569 *
570 * 2. /dev/mapper/mpatha (made up of /dev/sda and /dev/sdb)
571 * dev_name: /dev/mapper/mpatha
572 * returns: /dev/sda (first device)
573 *
574 * 3. /dev/sda (already the underlying device)
575 * dev_name: /dev/sda
576 * returns: /dev/sda
577 *
578 * 4. /dev/dm-3 (mapped to /dev/sda)
579 * dev_name: /dev/dm-3
580 * returns: /dev/sda
581 *
582 * 5. /dev/disk/by-id/scsi-0QEMU_drive-scsi0-0-0-0-part9 -> ../../sdb9
583 * dev_name: /dev/disk/by-id/scsi-0QEMU_drive-scsi0-0-0-0-part9
584 * returns: /dev/sdb
585 *
586 * 6. /dev/disk/by-uuid/5df030cf-3cd9-46e4-8e99-3ccb462a4e9a -> ../dev/sda2
587 * dev_name: /dev/disk/by-uuid/5df030cf-3cd9-46e4-8e99-3ccb462a4e9a
588 * returns: /dev/sda
589 *
590 * Returns underlying device name, or NULL on error or no match.
591 *
592 * NOTE: The returned name string must be *freed*.
593 */
594 char *
zfs_get_underlying_path(const char * dev_name)595 zfs_get_underlying_path(const char *dev_name)
596 {
597 char *name = NULL;
598 char *tmp;
599
600 if (dev_name == NULL)
601 return (NULL);
602
603 tmp = dm_get_underlying_path(dev_name);
604
605 /* dev_name not a DM device, so just un-symlinkize it */
606 if (tmp == NULL)
607 tmp = realpath(dev_name, NULL);
608
609 if (tmp != NULL) {
610 name = zfs_strip_partition_path(tmp);
611 free(tmp);
612 }
613
614 return (name);
615 }
616
617
618 #ifdef HAVE_LIBUDEV
619
620 /*
621 * A disk is considered a multipath whole disk when:
622 * DEVNAME key value has "dm-"
623 * DM_UUID key exists and starts with 'mpath-'
624 * ID_PART_TABLE_TYPE key does not exist or is not gpt
625 * ID_FS_LABEL key does not exist (disk isn't labeled)
626 */
627 static boolean_t
is_mpath_udev_sane(struct udev_device * dev)628 is_mpath_udev_sane(struct udev_device *dev)
629 {
630 const char *devname, *type, *uuid, *label;
631
632 devname = udev_device_get_property_value(dev, "DEVNAME");
633 type = udev_device_get_property_value(dev, "ID_PART_TABLE_TYPE");
634 uuid = udev_device_get_property_value(dev, "DM_UUID");
635 label = udev_device_get_property_value(dev, "ID_FS_LABEL");
636
637 if ((devname != NULL && strncmp(devname, "/dev/dm-", 8) == 0) &&
638 ((type == NULL) || (strcmp(type, "gpt") != 0)) &&
639 ((uuid != NULL) && (strncmp(uuid, "mpath-", 6) == 0)) &&
640 (label == NULL)) {
641 return (B_TRUE);
642 }
643
644 return (B_FALSE);
645 }
646
647 /*
648 * Check if a disk is a multipath "blank" disk:
649 *
650 * 1. The disk has udev values that suggest it's a multipath disk
651 * 2. The disk is not currently labeled with a filesystem of any type
652 * 3. There are no partitions on the disk
653 */
654 boolean_t
is_mpath_whole_disk(const char * path)655 is_mpath_whole_disk(const char *path)
656 {
657 struct udev *udev;
658 struct udev_device *dev = NULL;
659 char nodepath[MAXPATHLEN];
660 char *sysname;
661
662 if (realpath(path, nodepath) == NULL)
663 return (B_FALSE);
664 sysname = strrchr(nodepath, '/') + 1;
665 if (strncmp(sysname, "dm-", 3) != 0)
666 return (B_FALSE);
667 if ((udev = udev_new()) == NULL)
668 return (B_FALSE);
669 if ((dev = udev_device_new_from_subsystem_sysname(udev, "block",
670 sysname)) == NULL) {
671 udev_device_unref(dev);
672 return (B_FALSE);
673 }
674
675 /* Sanity check some udev values */
676 boolean_t is_sane = is_mpath_udev_sane(dev);
677 udev_device_unref(dev);
678
679 return (is_sane);
680 }
681
682 #else /* HAVE_LIBUDEV */
683
684 boolean_t
is_mpath_whole_disk(const char * path)685 is_mpath_whole_disk(const char *path)
686 {
687 (void) path;
688 return (B_FALSE);
689 }
690
691 #endif /* HAVE_LIBUDEV */
692