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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
28 * This userland component takes a set of options and uses libzpool to translate
29 * from a user-visible object type and name to an internal representation.
30 * There are two basic types of faults: device faults and data faults.
35 * Errors can be injected into a particular vdev using the '-d' option. This
36 * option takes a path or vdev GUID to uniquely identify the device within a
37 * pool. There are two types of errors that can be injected, EIO and ENXIO,
38 * that can be controlled through the '-e' option. The default is ENXIO. For
39 * EIO failures, any attempt to read data from the device will return EIO, but
40 * subsequent attempt to reopen the device will succeed. For ENXIO failures,
41 * any attempt to read from the device will return EIO, but any attempt to
42 * reopen the device will also return ENXIO.
43 * For label faults, the -L option must be specified. This allows faults
44 * to be injected into either the nvlist, uberblock, pad1, or pad2 region
45 * of all the labels for the specified device.
47 * This form of the command looks like:
49 * zinject -d device [-e errno] [-L <uber | nvlist | pad1 | pad2>] pool
54 * We begin with a tuple of the form:
56 * <type,level,range,object>
58 * type A string describing the type of data to target. Each type
59 * implicitly describes how to interpret 'object'. Currently,
60 * the following values are supported:
62 * data User data for a file
63 * dnode Dnode for a file or directory
65 * The following MOS objects are special. Instead of injecting
66 * errors on a particular object or blkid, we inject errors across
67 * all objects of the given type.
69 * mos Any data in the MOS
70 * mosdir object directory
71 * config pool configuration
75 * errlog persistent error log
77 * level Object level. Defaults to '0', not applicable to all types. If
78 * a range is given, this corresponds to the indirect block
79 * corresponding to the specific range.
81 * range A numerical range [start,end) within the object. Defaults to
82 * the full size of the file.
84 * object A string describing the logical location of the object. For
85 * files and directories (currently the only supported types),
86 * this is the path of the object on disk.
88 * This is translated, via libzpool, into the following internal representation:
90 * <type,objset,object,level,range>
92 * These types should be self-explanatory. This tuple is then passed to the
93 * kernel via a special ioctl() to initiate fault injection for the given
94 * object. Note that 'type' is not strictly necessary for fault injection, but
95 * is used when translating existing faults into a human-readable string.
98 * The command itself takes one of the forms:
101 * zinject <-a | -u pool>
102 * zinject -c <id|all>
103 * zinject [-q] <-t type> [-f freq] [-u] [-a] [-m] [-e errno] [-l level]
104 * [-r range] <object>
105 * zinject [-f freq] [-a] [-m] [-u] -b objset:object:level:start:end pool
107 * With no arguments, the command prints all currently registered injection
108 * handlers, with their numeric identifiers.
110 * The '-c' option will clear the given handler, or all handlers if 'all' is
113 * The '-e' option takes a string describing the errno to simulate. This must
114 * be either 'io' or 'checksum'. In most cases this will result in the same
115 * behavior, but RAID-Z will produce a different set of ereports for this
118 * The '-a', '-u', and '-m' flags toggle internal flush behavior. If '-a' is
119 * specified, then the ARC cache is flushed appropriately. If '-u' is
120 * specified, then the underlying SPA is unloaded. Either of these flags can be
121 * specified independently of any other handlers. The '-m' flag automatically
122 * does an unmount and remount of the underlying dataset to aid in flushing the
125 * The '-f' flag controls the frequency of errors injected, expressed as a
126 * integer percentage between 1 and 100. The default is 100.
128 * The this form is responsible for actually injecting the handler into the
129 * framework. It takes the arguments described above, translates them to the
130 * internal tuple using libzpool, and then issues an ioctl() to register the
133 * The final form can target a specific bookmark, regardless of whether a
134 * human-readable interface has been designed. It allows developers to specify
135 * a particular block by number.
145 #include <sys/fs/zfs.h>
146 #include <sys/mount.h>
150 #undef verify /* both libzfs.h and zfs_context.h want to define this */
154 libzfs_handle_t *g_zfs;
159 static const char *errtable[TYPE_INVAL] = {
176 name_to_type(const char *arg)
179 for (i = 0; i < TYPE_INVAL; i++)
180 if (strcmp(errtable[i], arg) == 0)
187 type_to_name(uint64_t type)
190 case DMU_OT_OBJECT_DIRECTORY:
192 case DMU_OT_OBJECT_ARRAY:
194 case DMU_OT_PACKED_NVLIST:
198 case DMU_OT_SPACE_MAP:
200 case DMU_OT_ERROR_LOG:
209 * Print usage message.
219 "\t\tList all active injection records.\n"
221 "\tzinject -c <id|all>\n"
223 "\t\tClear the particular record (if given a numeric ID), or\n"
224 "\t\tall records if 'all' is specificed.\n"
226 "\tzinject -p <function name> pool\n"
227 "\t\tInject a panic fault at the specified function. Only \n"
228 "\t\tfunctions which call spa_vdev_config_exit(), or \n"
229 "\t\tspa_vdev_exit() will trigger a panic.\n"
231 "\tzinject -d device [-e errno] [-L <nvlist|uber|pad1|pad2>] [-F]\n"
232 "\t [-T <read|write|free|claim|all> pool\n"
233 "\t\tInject a fault into a particular device or the device's\n"
234 "\t\tlabel. Label injection can either be 'nvlist', 'uber',\n "
235 "\t\t'pad1', or 'pad2'.\n"
236 "\t\t'errno' can be 'nxio' (the default), 'io', or 'dtl'.\n"
238 "\tzinject -d device -A <degrade|fault> pool\n"
239 "\t\tPerform a specific action on a particular device\n"
241 "\tzinject -I [-s <seconds> | -g <txgs>] pool\n"
242 "\t\tCause the pool to stop writing blocks yet not\n"
243 "\t\treport errors for a duration. Simulates buggy hardware\n"
244 "\t\tthat fails to honor cache flush requests.\n"
245 "\t\tDefault duration is 30 seconds. The machine is panicked\n"
246 "\t\tat the end of the duration.\n"
248 "\tzinject -b objset:object:level:blkid pool\n"
250 "\t\tInject an error into pool 'pool' with the numeric bookmark\n"
251 "\t\tspecified by the remaining tuple. Each number is in\n"
252 "\t\thexidecimal, and only one block can be specified.\n"
254 "\tzinject [-q] <-t type> [-e errno] [-l level] [-r range]\n"
255 "\t [-a] [-m] [-u] [-f freq] <object>\n"
257 "\t\tInject an error into the object specified by the '-t' option\n"
258 "\t\tand the object descriptor. The 'object' parameter is\n"
259 "\t\tinterperted depending on the '-t' option.\n"
261 "\t\t-q\tQuiet mode. Only print out the handler number added.\n"
262 "\t\t-e\tInject a specific error. Must be either 'io' or\n"
263 "\t\t\t'checksum'. Default is 'io'.\n"
264 "\t\t-l\tInject error at a particular block level. Default is "
266 "\t\t-m\tAutomatically remount underlying filesystem.\n"
267 "\t\t-r\tInject error over a particular logical range of an\n"
268 "\t\t\tobject. Will be translated to the appropriate blkid\n"
269 "\t\t\trange according to the object's properties.\n"
270 "\t\t-a\tFlush the ARC cache. Can be specified without any\n"
271 "\t\t\tassociated object.\n"
272 "\t\t-u\tUnload the associated pool. Can be specified with only\n"
273 "\t\t\ta pool object.\n"
274 "\t\t-f\tOnly inject errors a fraction of the time. Expressed as\n"
275 "\t\t\ta percentage between 1 and 100.\n"
277 "\t-t data\t\tInject an error into the plain file contents of a\n"
278 "\t\t\tfile. The object must be specified as a complete path\n"
279 "\t\t\tto a file on a ZFS filesystem.\n"
281 "\t-t dnode\tInject an error into the metadnode in the block\n"
282 "\t\t\tcorresponding to the dnode for a file or directory. The\n"
283 "\t\t\t'-r' option is incompatible with this mode. The object\n"
284 "\t\t\tis specified as a complete path to a file or directory\n"
285 "\t\t\ton a ZFS filesystem.\n"
287 "\t-t <mos>\tInject errors into the MOS for objects of the given\n"
288 "\t\t\ttype. Valid types are: mos, mosdir, config, bpobj,\n"
289 "\t\t\tspacemap, metaslab, errlog. The only valid <object> is\n"
290 "\t\t\tthe poolname.\n");
294 iter_handlers(int (*func)(int, const char *, zinject_record_t *, void *),
302 while (ioctl(zfs_fd, ZFS_IOC_INJECT_LIST_NEXT, &zc) == 0)
303 if ((ret = func((int)zc.zc_guid, zc.zc_name,
304 &zc.zc_inject_record, data)) != 0)
307 if (errno != ENOENT) {
308 (void) fprintf(stderr, "Unable to list handlers: %s\n",
317 print_data_handler(int id, const char *pool, zinject_record_t *record,
322 if (record->zi_guid != 0 || record->zi_func[0] != '\0')
326 (void) printf("%3s %-15s %-6s %-6s %-8s %3s %-15s\n",
327 "ID", "POOL", "OBJSET", "OBJECT", "TYPE", "LVL", "RANGE");
328 (void) printf("--- --------------- ------ "
329 "------ -------- --- ---------------\n");
334 (void) printf("%3d %-15s %-6llu %-6llu %-8s %3d ", id, pool,
335 (u_longlong_t)record->zi_objset, (u_longlong_t)record->zi_object,
336 type_to_name(record->zi_type), record->zi_level);
338 if (record->zi_start == 0 &&
339 record->zi_end == -1ULL)
340 (void) printf("all\n");
342 (void) printf("[%llu, %llu]\n", (u_longlong_t)record->zi_start,
343 (u_longlong_t)record->zi_end);
349 print_device_handler(int id, const char *pool, zinject_record_t *record,
354 if (record->zi_guid == 0 || record->zi_func[0] != '\0')
358 (void) printf("%3s %-15s %s\n", "ID", "POOL", "GUID");
359 (void) printf("--- --------------- ----------------\n");
364 (void) printf("%3d %-15s %llx\n", id, pool,
365 (u_longlong_t)record->zi_guid);
371 print_panic_handler(int id, const char *pool, zinject_record_t *record,
376 if (record->zi_func[0] == '\0')
380 (void) printf("%3s %-15s %s\n", "ID", "POOL", "FUNCTION");
381 (void) printf("--- --------------- ----------------\n");
386 (void) printf("%3d %-15s %s\n", id, pool, record->zi_func);
392 * Print all registered error handlers. Returns the number of handlers
396 print_all_handlers(void)
398 int count = 0, total = 0;
400 (void) iter_handlers(print_device_handler, &count);
407 (void) iter_handlers(print_data_handler, &count);
414 (void) iter_handlers(print_panic_handler, &count);
416 return (count + total);
421 cancel_one_handler(int id, const char *pool, zinject_record_t *record,
426 zc.zc_guid = (uint64_t)id;
428 if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) {
429 (void) fprintf(stderr, "failed to remove handler %d: %s\n",
430 id, strerror(errno));
438 * Remove all fault injection handlers.
441 cancel_all_handlers(void)
443 int ret = iter_handlers(cancel_one_handler, NULL);
446 (void) printf("removed all registered handlers\n");
452 * Remove a specific fault injection handler.
455 cancel_handler(int id)
459 zc.zc_guid = (uint64_t)id;
461 if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) {
462 (void) fprintf(stderr, "failed to remove handler %d: %s\n",
463 id, strerror(errno));
467 (void) printf("removed handler %d\n", id);
473 * Register a new fault injection handler.
476 register_handler(const char *pool, int flags, zinject_record_t *record,
481 (void) strcpy(zc.zc_name, pool);
482 zc.zc_inject_record = *record;
485 if (ioctl(zfs_fd, ZFS_IOC_INJECT_FAULT, &zc) != 0) {
486 (void) fprintf(stderr, "failed to add handler: %s\n",
491 if (flags & ZINJECT_NULL)
495 (void) printf("%llu\n", (u_longlong_t)zc.zc_guid);
497 (void) printf("Added handler %llu with the following "
498 "properties:\n", (u_longlong_t)zc.zc_guid);
499 (void) printf(" pool: %s\n", pool);
500 if (record->zi_guid) {
501 (void) printf(" vdev: %llx\n",
502 (u_longlong_t)record->zi_guid);
503 } else if (record->zi_func[0] != '\0') {
504 (void) printf(" panic function: %s\n",
506 } else if (record->zi_duration > 0) {
507 (void) printf(" time: %lld seconds\n",
508 (u_longlong_t)record->zi_duration);
509 } else if (record->zi_duration < 0) {
510 (void) printf(" txgs: %lld \n",
511 (u_longlong_t)-record->zi_duration);
513 (void) printf("objset: %llu\n",
514 (u_longlong_t)record->zi_objset);
515 (void) printf("object: %llu\n",
516 (u_longlong_t)record->zi_object);
517 (void) printf(" type: %llu\n",
518 (u_longlong_t)record->zi_type);
519 (void) printf(" level: %d\n", record->zi_level);
520 if (record->zi_start == 0 &&
521 record->zi_end == -1ULL)
522 (void) printf(" range: all\n");
524 (void) printf(" range: [%llu, %llu)\n",
525 (u_longlong_t)record->zi_start,
526 (u_longlong_t)record->zi_end);
534 perform_action(const char *pool, zinject_record_t *record, int cmd)
538 ASSERT(cmd == VDEV_STATE_DEGRADED || cmd == VDEV_STATE_FAULTED);
539 (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
540 zc.zc_guid = record->zi_guid;
543 if (ioctl(zfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
550 main(int argc, char **argv)
562 int io_type = ZIO_TYPES;
563 int action = VDEV_STATE_UNKNOWN;
564 err_type_t type = TYPE_INVAL;
565 err_type_t label = TYPE_INVAL;
566 zinject_record_t record = { 0 };
567 char pool[MAXNAMELEN];
568 char dataset[MAXNAMELEN];
578 * No arguments. Print the available handlers. If there are no
579 * available handlers, direct the user to '-h' for help
582 if (print_all_handlers() == 0) {
583 (void) printf("No handlers registered.\n");
584 (void) printf("Run 'zinject -h' for usage "
591 while ((c = getopt(argc, argv,
592 ":aA:b:d:f:Fg:qhIc:t:T:l:mr:s:e:uL:p:")) != -1) {
595 flags |= ZINJECT_FLUSH_ARC;
598 if (strcasecmp(optarg, "degrade") == 0) {
599 action = VDEV_STATE_DEGRADED;
600 } else if (strcasecmp(optarg, "fault") == 0) {
601 action = VDEV_STATE_FAULTED;
603 (void) fprintf(stderr, "invalid action '%s': "
604 "must be 'degrade' or 'fault'\n", optarg);
619 if (strcasecmp(optarg, "io") == 0) {
621 } else if (strcasecmp(optarg, "checksum") == 0) {
623 } else if (strcasecmp(optarg, "nxio") == 0) {
625 } else if (strcasecmp(optarg, "dtl") == 0) {
628 (void) fprintf(stderr, "invalid error type "
629 "'%s': must be 'io', 'checksum' or "
636 record.zi_freq = atoi(optarg);
637 if (record.zi_freq < 1 || record.zi_freq > 100) {
638 (void) fprintf(stderr, "frequency range must "
639 "be in the range (0, 100]\n");
644 record.zi_failfast = B_TRUE;
648 record.zi_duration = (int)strtol(optarg, &end, 10);
649 if (record.zi_duration <= 0 || *end != '\0') {
650 (void) fprintf(stderr, "invalid duration '%s': "
651 "must be a positive integer\n", optarg);
655 /* store duration of txgs as its negative */
656 record.zi_duration *= -1;
662 /* default duration, if one hasn't yet been defined */
664 if (dur_secs == 0 && dur_txg == 0)
665 record.zi_duration = 30;
668 level = (int)strtol(optarg, &end, 10);
670 (void) fprintf(stderr, "invalid level '%s': "
671 "must be an integer\n", optarg);
680 (void) strlcpy(record.zi_func, optarg,
681 sizeof (record.zi_func));
691 record.zi_duration = (int)strtol(optarg, &end, 10);
692 if (record.zi_duration <= 0 || *end != '\0') {
693 (void) fprintf(stderr, "invalid duration '%s': "
694 "must be a positive integer\n", optarg);
700 if (strcasecmp(optarg, "read") == 0) {
701 io_type = ZIO_TYPE_READ;
702 } else if (strcasecmp(optarg, "write") == 0) {
703 io_type = ZIO_TYPE_WRITE;
704 } else if (strcasecmp(optarg, "free") == 0) {
705 io_type = ZIO_TYPE_FREE;
706 } else if (strcasecmp(optarg, "claim") == 0) {
707 io_type = ZIO_TYPE_CLAIM;
708 } else if (strcasecmp(optarg, "all") == 0) {
711 (void) fprintf(stderr, "invalid I/O type "
712 "'%s': must be 'read', 'write', 'free', "
713 "'claim' or 'all'\n", optarg);
719 if ((type = name_to_type(optarg)) == TYPE_INVAL &&
721 (void) fprintf(stderr, "invalid type '%s'\n",
728 flags |= ZINJECT_UNLOAD_SPA;
731 if ((label = name_to_type(optarg)) == TYPE_INVAL &&
733 (void) fprintf(stderr, "invalid label type "
740 (void) fprintf(stderr, "option -%c requires an "
741 "operand\n", optopt);
745 (void) fprintf(stderr, "invalid option '%c'\n",
755 if ((g_zfs = libzfs_init()) == NULL)
758 libzfs_print_on_error(g_zfs, B_TRUE);
760 if ((zfs_fd = open(ZFS_DEV, O_RDWR)) < 0) {
761 (void) fprintf(stderr, "failed to open ZFS device\n");
765 if (cancel != NULL) {
767 * '-c' is invalid with any other options.
769 if (raw != NULL || range != NULL || type != TYPE_INVAL ||
770 level != 0 || record.zi_func[0] != '\0' ||
771 record.zi_duration != 0) {
772 (void) fprintf(stderr, "cancel (-c) incompatible with "
773 "any other options\n");
778 (void) fprintf(stderr, "extraneous argument to '-c'\n");
783 if (strcmp(cancel, "all") == 0) {
784 return (cancel_all_handlers());
786 int id = (int)strtol(cancel, &end, 10);
788 (void) fprintf(stderr, "invalid handle id '%s':"
789 " must be an integer or 'all'\n", cancel);
793 return (cancel_handler(id));
797 if (device != NULL) {
799 * Device (-d) injection uses a completely different mechanism
800 * for doing injection, so handle it separately here.
802 if (raw != NULL || range != NULL || type != TYPE_INVAL ||
803 level != 0 || record.zi_func[0] != '\0' ||
804 record.zi_duration != 0) {
805 (void) fprintf(stderr, "device (-d) incompatible with "
806 "data error injection\n");
812 (void) fprintf(stderr, "device (-d) injection requires "
813 "a single pool name\n");
818 (void) strcpy(pool, argv[0]);
821 if (error == ECKSUM) {
822 (void) fprintf(stderr, "device error type must be "
827 record.zi_iotype = io_type;
828 if (translate_device(pool, device, label, &record) != 0)
833 if (action != VDEV_STATE_UNKNOWN)
834 return (perform_action(pool, &record, action));
836 } else if (raw != NULL) {
837 if (range != NULL || type != TYPE_INVAL || level != 0 ||
838 record.zi_func[0] != '\0' || record.zi_duration != 0) {
839 (void) fprintf(stderr, "raw (-b) format with "
840 "any other options\n");
846 (void) fprintf(stderr, "raw (-b) format expects a "
847 "single pool name\n");
852 (void) strcpy(pool, argv[0]);
855 if (error == ENXIO) {
856 (void) fprintf(stderr, "data error type must be "
857 "'checksum' or 'io'\n");
861 if (translate_raw(raw, &record) != 0)
865 } else if (record.zi_func[0] != '\0') {
866 if (raw != NULL || range != NULL || type != TYPE_INVAL ||
867 level != 0 || device != NULL || record.zi_duration != 0) {
868 (void) fprintf(stderr, "panic (-p) incompatible with "
874 if (argc < 1 || argc > 2) {
875 (void) fprintf(stderr, "panic (-p) injection requires "
876 "a single pool name and an optional id\n");
881 (void) strcpy(pool, argv[0]);
883 record.zi_type = atoi(argv[1]);
885 } else if (record.zi_duration != 0) {
887 (void) fprintf(stderr, "-s or -g meaningless "
888 "without -I (ignore writes)\n");
891 } else if (dur_secs && dur_txg) {
892 (void) fprintf(stderr, "choose a duration either "
893 "in seconds (-s) or a number of txgs (-g) "
897 } else if (argc != 1) {
898 (void) fprintf(stderr, "ignore writes (-I) "
899 "injection requires a single pool name\n");
904 (void) strcpy(pool, argv[0]);
906 } else if (type == TYPE_INVAL) {
908 (void) fprintf(stderr, "at least one of '-b', '-d', "
909 "'-t', '-a', '-p', '-I' or '-u' "
910 "must be specified\n");
915 if (argc == 1 && (flags & ZINJECT_UNLOAD_SPA)) {
916 (void) strcpy(pool, argv[0]);
918 } else if (argc != 0) {
919 (void) fprintf(stderr, "extraneous argument for "
925 flags |= ZINJECT_NULL;
928 (void) fprintf(stderr, "missing object\n");
933 if (error == ENXIO) {
934 (void) fprintf(stderr, "data error type must be "
935 "'checksum' or 'io'\n");
939 if (translate_record(type, argv[0], range, level, &record, pool,
947 * If this is pool-wide metadata, unmount everything. The ioctl() will
948 * unload the pool, so that we trigger spa-wide reopen of metadata next
949 * time we access the pool.
951 if (dataset[0] != '\0' && domount) {
952 if ((zhp = zfs_open(g_zfs, dataset, ZFS_TYPE_DATASET)) == NULL)
955 if (zfs_unmount(zhp, NULL, 0) != 0)
957 #endif /* HAVE_ZPL */
960 record.zi_error = error;
962 ret = register_handler(pool, flags, &record, quiet);
965 if (dataset[0] != '\0' && domount)
966 ret = (zfs_mount(zhp, NULL, 0) != 0);
967 #endif /* HAVE_ZPL */