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.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
29 * This userland component takes a set of options and uses libzpool to translate
30 * from a user-visible object type and name to an internal representation.
31 * There are two basic types of faults: device faults and data faults.
36 * Errors can be injected into a particular vdev using the '-d' option. This
37 * option takes a path or vdev GUID to uniquely identify the device within a
38 * pool. There are two types of errors that can be injected, EIO and ENXIO,
39 * that can be controlled through the '-e' option. The default is ENXIO. For
40 * EIO failures, any attempt to read data from the device will return EIO, but
41 * subsequent attempt to reopen the device will succeed. For ENXIO failures,
42 * any attempt to read from the device will return EIO, but any attempt to
43 * reopen the device will also return ENXIO.
44 * For label faults, the -L option must be specified. This allows faults
45 * to be injected into either the nvlist, uberblock, pad1, or pad2 region
46 * of all the labels for the specified device.
48 * This form of the command looks like:
50 * zinject -d device [-e errno] [-L <uber | nvlist | pad1 | pad2>] pool
55 * We begin with a tuple of the form:
57 * <type,level,range,object>
59 * type A string describing the type of data to target. Each type
60 * implicitly describes how to interpret 'object'. Currently,
61 * the following values are supported:
63 * data User data for a file
64 * dnode Dnode for a file or directory
66 * The following MOS objects are special. Instead of injecting
67 * errors on a particular object or blkid, we inject errors across
68 * all objects of the given type.
70 * mos Any data in the MOS
71 * mosdir object directory
72 * config pool configuration
76 * errlog persistent error log
78 * level Object level. Defaults to '0', not applicable to all types. If
79 * a range is given, this corresponds to the indirect block
80 * corresponding to the specific range.
82 * range A numerical range [start,end) within the object. Defaults to
83 * the full size of the file.
85 * object A string describing the logical location of the object. For
86 * files and directories (currently the only supported types),
87 * this is the path of the object on disk.
89 * This is translated, via libzpool, into the following internal representation:
91 * <type,objset,object,level,range>
93 * These types should be self-explanatory. This tuple is then passed to the
94 * kernel via a special ioctl() to initiate fault injection for the given
95 * object. Note that 'type' is not strictly necessary for fault injection, but
96 * is used when translating existing faults into a human-readable string.
99 * The command itself takes one of the forms:
102 * zinject <-a | -u pool>
103 * zinject -c <id|all>
104 * zinject [-q] <-t type> [-f freq] [-u] [-a] [-m] [-e errno] [-l level]
105 * [-r range] <object>
106 * zinject [-f freq] [-a] [-m] [-u] -b objset:object:level:start:end pool
108 * With no arguments, the command prints all currently registered injection
109 * handlers, with their numeric identifiers.
111 * The '-c' option will clear the given handler, or all handlers if 'all' is
114 * The '-e' option takes a string describing the errno to simulate. This must
115 * be either 'io' or 'checksum'. In most cases this will result in the same
116 * behavior, but RAID-Z will produce a different set of ereports for this
119 * The '-a', '-u', and '-m' flags toggle internal flush behavior. If '-a' is
120 * specified, then the ARC cache is flushed appropriately. If '-u' is
121 * specified, then the underlying SPA is unloaded. Either of these flags can be
122 * specified independently of any other handlers. The '-m' flag automatically
123 * does an unmount and remount of the underlying dataset to aid in flushing the
126 * The '-f' flag controls the frequency of errors injected, expressed as a
127 * integer percentage between 1 and 100. The default is 100.
129 * The this form is responsible for actually injecting the handler into the
130 * framework. It takes the arguments described above, translates them to the
131 * internal tuple using libzpool, and then issues an ioctl() to register the
134 * The final form can target a specific bookmark, regardless of whether a
135 * human-readable interface has been designed. It allows developers to specify
136 * a particular block by number.
146 #include <sys/fs/zfs.h>
147 #include <sys/param.h>
148 #include <sys/mount.h>
152 #undef verify /* both libzfs.h and zfs_context.h want to define this */
156 libzfs_handle_t *g_zfs;
163 static const char *errtable[TYPE_INVAL] = {
180 name_to_type(const char *arg)
183 for (i = 0; i < TYPE_INVAL; i++)
184 if (strcmp(errtable[i], arg) == 0)
191 type_to_name(uint64_t type)
194 case DMU_OT_OBJECT_DIRECTORY:
196 case DMU_OT_OBJECT_ARRAY:
198 case DMU_OT_PACKED_NVLIST:
202 case DMU_OT_SPACE_MAP:
204 case DMU_OT_ERROR_LOG:
213 * Print usage message.
223 "\t\tList all active injection records.\n"
225 "\tzinject -c <id|all>\n"
227 "\t\tClear the particular record (if given a numeric ID), or\n"
228 "\t\tall records if 'all' is specificed.\n"
230 "\tzinject -p <function name> pool\n"
231 "\t\tInject a panic fault at the specified function. Only \n"
232 "\t\tfunctions which call spa_vdev_config_exit(), or \n"
233 "\t\tspa_vdev_exit() will trigger a panic.\n"
235 "\tzinject -d device [-e errno] [-L <nvlist|uber|pad1|pad2>] [-F]\n"
236 "\t [-T <read|write|free|claim|all> pool\n"
237 "\t\tInject a fault into a particular device or the device's\n"
238 "\t\tlabel. Label injection can either be 'nvlist', 'uber',\n "
239 "\t\t'pad1', or 'pad2'.\n"
240 "\t\t'errno' can be 'nxio' (the default), 'io', or 'dtl'.\n"
242 "\tzinject -d device -A <degrade|fault> pool\n"
243 "\t\tPerform a specific action on a particular device\n"
245 "\tzinject -I [-s <seconds> | -g <txgs>] pool\n"
246 "\t\tCause the pool to stop writing blocks yet not\n"
247 "\t\treport errors for a duration. Simulates buggy hardware\n"
248 "\t\tthat fails to honor cache flush requests.\n"
249 "\t\tDefault duration is 30 seconds. The machine is panicked\n"
250 "\t\tat the end of the duration.\n"
252 "\tzinject -b objset:object:level:blkid pool\n"
254 "\t\tInject an error into pool 'pool' with the numeric bookmark\n"
255 "\t\tspecified by the remaining tuple. Each number is in\n"
256 "\t\thexidecimal, and only one block can be specified.\n"
258 "\tzinject [-q] <-t type> [-e errno] [-l level] [-r range]\n"
259 "\t [-a] [-m] [-u] [-f freq] <object>\n"
261 "\t\tInject an error into the object specified by the '-t' option\n"
262 "\t\tand the object descriptor. The 'object' parameter is\n"
263 "\t\tinterperted depending on the '-t' option.\n"
265 "\t\t-q\tQuiet mode. Only print out the handler number added.\n"
266 "\t\t-e\tInject a specific error. Must be either 'io' or\n"
267 "\t\t\t'checksum'. Default is 'io'.\n"
268 "\t\t-l\tInject error at a particular block level. Default is "
270 "\t\t-m\tAutomatically remount underlying filesystem.\n"
271 "\t\t-r\tInject error over a particular logical range of an\n"
272 "\t\t\tobject. Will be translated to the appropriate blkid\n"
273 "\t\t\trange according to the object's properties.\n"
274 "\t\t-a\tFlush the ARC cache. Can be specified without any\n"
275 "\t\t\tassociated object.\n"
276 "\t\t-u\tUnload the associated pool. Can be specified with only\n"
277 "\t\t\ta pool object.\n"
278 "\t\t-f\tOnly inject errors a fraction of the time. Expressed as\n"
279 "\t\t\ta percentage between 1 and 100.\n"
281 "\t-t data\t\tInject an error into the plain file contents of a\n"
282 "\t\t\tfile. The object must be specified as a complete path\n"
283 "\t\t\tto a file on a ZFS filesystem.\n"
285 "\t-t dnode\tInject an error into the metadnode in the block\n"
286 "\t\t\tcorresponding to the dnode for a file or directory. The\n"
287 "\t\t\t'-r' option is incompatible with this mode. The object\n"
288 "\t\t\tis specified as a complete path to a file or directory\n"
289 "\t\t\ton a ZFS filesystem.\n"
291 "\t-t <mos>\tInject errors into the MOS for objects of the given\n"
292 "\t\t\ttype. Valid types are: mos, mosdir, config, bpobj,\n"
293 "\t\t\tspacemap, metaslab, errlog. The only valid <object> is\n"
294 "\t\t\tthe poolname.\n");
298 iter_handlers(int (*func)(int, const char *, zinject_record_t *, void *),
301 zfs_cmd_t zc = { 0 };
304 while (ioctl(zfs_fd, ZFS_IOC_INJECT_LIST_NEXT, &zc) == 0)
305 if ((ret = func((int)zc.zc_guid, zc.zc_name,
306 &zc.zc_inject_record, data)) != 0)
309 if (errno != ENOENT) {
310 (void) fprintf(stderr, "Unable to list handlers: %s\n",
319 print_data_handler(int id, const char *pool, zinject_record_t *record,
324 if (record->zi_guid != 0 || record->zi_func[0] != '\0')
328 (void) printf("%3s %-15s %-6s %-6s %-8s %3s %-15s\n",
329 "ID", "POOL", "OBJSET", "OBJECT", "TYPE", "LVL", "RANGE");
330 (void) printf("--- --------------- ------ "
331 "------ -------- --- ---------------\n");
336 (void) printf("%3d %-15s %-6llu %-6llu %-8s %3d ", id, pool,
337 (u_longlong_t)record->zi_objset, (u_longlong_t)record->zi_object,
338 type_to_name(record->zi_type), record->zi_level);
340 if (record->zi_start == 0 &&
341 record->zi_end == -1ULL)
342 (void) printf("all\n");
344 (void) printf("[%llu, %llu]\n", (u_longlong_t)record->zi_start,
345 (u_longlong_t)record->zi_end);
351 print_device_handler(int id, const char *pool, zinject_record_t *record,
356 if (record->zi_guid == 0 || record->zi_func[0] != '\0')
360 (void) printf("%3s %-15s %s\n", "ID", "POOL", "GUID");
361 (void) printf("--- --------------- ----------------\n");
366 (void) printf("%3d %-15s %llx\n", id, pool,
367 (u_longlong_t)record->zi_guid);
373 print_panic_handler(int id, const char *pool, zinject_record_t *record,
378 if (record->zi_func[0] == '\0')
382 (void) printf("%3s %-15s %s\n", "ID", "POOL", "FUNCTION");
383 (void) printf("--- --------------- ----------------\n");
388 (void) printf("%3d %-15s %s\n", id, pool, record->zi_func);
394 * Print all registered error handlers. Returns the number of handlers
398 print_all_handlers(void)
400 int count = 0, total = 0;
402 (void) iter_handlers(print_device_handler, &count);
409 (void) iter_handlers(print_data_handler, &count);
416 (void) iter_handlers(print_panic_handler, &count);
418 return (count + total);
423 cancel_one_handler(int id, const char *pool, zinject_record_t *record,
426 zfs_cmd_t zc = { 0 };
428 zc.zc_guid = (uint64_t)id;
430 if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) {
431 (void) fprintf(stderr, "failed to remove handler %d: %s\n",
432 id, strerror(errno));
440 * Remove all fault injection handlers.
443 cancel_all_handlers(void)
445 int ret = iter_handlers(cancel_one_handler, NULL);
448 (void) printf("removed all registered handlers\n");
454 * Remove a specific fault injection handler.
457 cancel_handler(int id)
459 zfs_cmd_t zc = { 0 };
461 zc.zc_guid = (uint64_t)id;
463 if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) {
464 (void) fprintf(stderr, "failed to remove handler %d: %s\n",
465 id, strerror(errno));
469 (void) printf("removed handler %d\n", id);
475 * Register a new fault injection handler.
478 register_handler(const char *pool, int flags, zinject_record_t *record,
481 zfs_cmd_t zc = { 0 };
483 (void) strcpy(zc.zc_name, pool);
484 zc.zc_inject_record = *record;
487 if (ioctl(zfs_fd, ZFS_IOC_INJECT_FAULT, &zc) != 0) {
488 (void) fprintf(stderr, "failed to add handler: %s\n",
493 if (flags & ZINJECT_NULL)
497 (void) printf("%llu\n", (u_longlong_t)zc.zc_guid);
499 (void) printf("Added handler %llu with the following "
500 "properties:\n", (u_longlong_t)zc.zc_guid);
501 (void) printf(" pool: %s\n", pool);
502 if (record->zi_guid) {
503 (void) printf(" vdev: %llx\n",
504 (u_longlong_t)record->zi_guid);
505 } else if (record->zi_func[0] != '\0') {
506 (void) printf(" panic function: %s\n",
508 } else if (record->zi_duration > 0) {
509 (void) printf(" time: %lld seconds\n",
510 (u_longlong_t)record->zi_duration);
511 } else if (record->zi_duration < 0) {
512 (void) printf(" txgs: %lld \n",
513 (u_longlong_t)-record->zi_duration);
515 (void) printf("objset: %llu\n",
516 (u_longlong_t)record->zi_objset);
517 (void) printf("object: %llu\n",
518 (u_longlong_t)record->zi_object);
519 (void) printf(" type: %llu\n",
520 (u_longlong_t)record->zi_type);
521 (void) printf(" level: %d\n", record->zi_level);
522 if (record->zi_start == 0 &&
523 record->zi_end == -1ULL)
524 (void) printf(" range: all\n");
526 (void) printf(" range: [%llu, %llu)\n",
527 (u_longlong_t)record->zi_start,
528 (u_longlong_t)record->zi_end);
536 perform_action(const char *pool, zinject_record_t *record, int cmd)
538 zfs_cmd_t zc = { 0 };
540 ASSERT(cmd == VDEV_STATE_DEGRADED || cmd == VDEV_STATE_FAULTED);
541 (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
542 zc.zc_guid = record->zi_guid;
545 if (ioctl(zfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
552 main(int argc, char **argv)
564 int io_type = ZIO_TYPES;
565 int action = VDEV_STATE_UNKNOWN;
566 err_type_t type = TYPE_INVAL;
567 err_type_t label = TYPE_INVAL;
568 zinject_record_t record = { 0 };
569 char pool[MAXNAMELEN];
570 char dataset[MAXNAMELEN];
578 if ((g_zfs = libzfs_init()) == NULL) {
579 (void) fprintf(stderr, "internal error: failed to "
580 "initialize ZFS library\n");
584 libzfs_print_on_error(g_zfs, B_TRUE);
586 if ((zfs_fd = open(ZFS_DEV, O_RDWR)) < 0) {
587 (void) fprintf(stderr, "failed to open ZFS device\n");
593 * No arguments. Print the available handlers. If there are no
594 * available handlers, direct the user to '-h' for help
597 if (print_all_handlers() == 0) {
598 (void) printf("No handlers registered.\n");
599 (void) printf("Run 'zinject -h' for usage "
606 while ((c = getopt(argc, argv,
607 ":aA:b:d:D:f:Fg:qhIc:t:T:l:mr:s:e:uL:p:")) != -1) {
610 flags |= ZINJECT_FLUSH_ARC;
613 if (strcasecmp(optarg, "degrade") == 0) {
614 action = VDEV_STATE_DEGRADED;
615 } else if (strcasecmp(optarg, "fault") == 0) {
616 action = VDEV_STATE_FAULTED;
618 (void) fprintf(stderr, "invalid action '%s': "
619 "must be 'degrade' or 'fault'\n", optarg);
634 record.zi_timer = strtoull(optarg, &end, 10);
635 if (errno != 0 || *end != '\0') {
636 (void) fprintf(stderr, "invalid i/o delay "
637 "value: '%s'\n", optarg);
643 if (strcasecmp(optarg, "io") == 0) {
645 } else if (strcasecmp(optarg, "checksum") == 0) {
647 } else if (strcasecmp(optarg, "nxio") == 0) {
649 } else if (strcasecmp(optarg, "dtl") == 0) {
652 (void) fprintf(stderr, "invalid error type "
653 "'%s': must be 'io', 'checksum' or "
660 record.zi_freq = atoi(optarg);
661 if (record.zi_freq < 1 || record.zi_freq > 100) {
662 (void) fprintf(stderr, "frequency range must "
663 "be in the range (0, 100]\n");
668 record.zi_failfast = B_TRUE;
672 record.zi_duration = (int)strtol(optarg, &end, 10);
673 if (record.zi_duration <= 0 || *end != '\0') {
674 (void) fprintf(stderr, "invalid duration '%s': "
675 "must be a positive integer\n", optarg);
679 /* store duration of txgs as its negative */
680 record.zi_duration *= -1;
686 /* default duration, if one hasn't yet been defined */
688 if (dur_secs == 0 && dur_txg == 0)
689 record.zi_duration = 30;
692 level = (int)strtol(optarg, &end, 10);
694 (void) fprintf(stderr, "invalid level '%s': "
695 "must be an integer\n", optarg);
704 (void) strlcpy(record.zi_func, optarg,
705 sizeof (record.zi_func));
706 record.zi_cmd = ZINJECT_PANIC;
716 record.zi_duration = (int)strtol(optarg, &end, 10);
717 if (record.zi_duration <= 0 || *end != '\0') {
718 (void) fprintf(stderr, "invalid duration '%s': "
719 "must be a positive integer\n", optarg);
725 if (strcasecmp(optarg, "read") == 0) {
726 io_type = ZIO_TYPE_READ;
727 } else if (strcasecmp(optarg, "write") == 0) {
728 io_type = ZIO_TYPE_WRITE;
729 } else if (strcasecmp(optarg, "free") == 0) {
730 io_type = ZIO_TYPE_FREE;
731 } else if (strcasecmp(optarg, "claim") == 0) {
732 io_type = ZIO_TYPE_CLAIM;
733 } else if (strcasecmp(optarg, "all") == 0) {
736 (void) fprintf(stderr, "invalid I/O type "
737 "'%s': must be 'read', 'write', 'free', "
738 "'claim' or 'all'\n", optarg);
744 if ((type = name_to_type(optarg)) == TYPE_INVAL &&
746 (void) fprintf(stderr, "invalid type '%s'\n",
753 flags |= ZINJECT_UNLOAD_SPA;
756 if ((label = name_to_type(optarg)) == TYPE_INVAL &&
758 (void) fprintf(stderr, "invalid label type "
765 (void) fprintf(stderr, "option -%c requires an "
766 "operand\n", optopt);
770 (void) fprintf(stderr, "invalid option '%c'\n",
780 if (record.zi_duration != 0)
781 record.zi_cmd = ZINJECT_IGNORED_WRITES;
783 if (cancel != NULL) {
785 * '-c' is invalid with any other options.
787 if (raw != NULL || range != NULL || type != TYPE_INVAL ||
788 level != 0 || record.zi_cmd != ZINJECT_UNINITIALIZED) {
789 (void) fprintf(stderr, "cancel (-c) incompatible with "
790 "any other options\n");
795 (void) fprintf(stderr, "extraneous argument to '-c'\n");
800 if (strcmp(cancel, "all") == 0) {
801 return (cancel_all_handlers());
803 int id = (int)strtol(cancel, &end, 10);
805 (void) fprintf(stderr, "invalid handle id '%s':"
806 " must be an integer or 'all'\n", cancel);
810 return (cancel_handler(id));
814 if (device != NULL) {
816 * Device (-d) injection uses a completely different mechanism
817 * for doing injection, so handle it separately here.
819 if (raw != NULL || range != NULL || type != TYPE_INVAL ||
820 level != 0 || record.zi_cmd != ZINJECT_UNINITIALIZED) {
821 (void) fprintf(stderr, "device (-d) incompatible with "
822 "data error injection\n");
828 (void) fprintf(stderr, "device (-d) injection requires "
829 "a single pool name\n");
834 (void) strcpy(pool, argv[0]);
837 if (error == ECKSUM) {
838 (void) fprintf(stderr, "device error type must be "
843 record.zi_iotype = io_type;
844 if (translate_device(pool, device, label, &record) != 0)
849 if (action != VDEV_STATE_UNKNOWN)
850 return (perform_action(pool, &record, action));
852 } else if (raw != NULL) {
853 if (range != NULL || type != TYPE_INVAL || level != 0 ||
854 record.zi_cmd != ZINJECT_UNINITIALIZED) {
855 (void) fprintf(stderr, "raw (-b) format with "
856 "any other options\n");
862 (void) fprintf(stderr, "raw (-b) format expects a "
863 "single pool name\n");
868 (void) strcpy(pool, argv[0]);
871 if (error == ENXIO) {
872 (void) fprintf(stderr, "data error type must be "
873 "'checksum' or 'io'\n");
877 record.zi_cmd = ZINJECT_DATA_FAULT;
878 if (translate_raw(raw, &record) != 0)
882 } else if (record.zi_cmd == ZINJECT_PANIC) {
883 if (raw != NULL || range != NULL || type != TYPE_INVAL ||
884 level != 0 || device != NULL) {
885 (void) fprintf(stderr, "panic (-p) incompatible with "
891 if (argc < 1 || argc > 2) {
892 (void) fprintf(stderr, "panic (-p) injection requires "
893 "a single pool name and an optional id\n");
898 (void) strcpy(pool, argv[0]);
900 record.zi_type = atoi(argv[1]);
902 } else if (record.zi_cmd == ZINJECT_IGNORED_WRITES) {
904 (void) fprintf(stderr, "-s or -g meaningless "
905 "without -I (ignore writes)\n");
908 } else if (dur_secs && dur_txg) {
909 (void) fprintf(stderr, "choose a duration either "
910 "in seconds (-s) or a number of txgs (-g) "
914 } else if (argc != 1) {
915 (void) fprintf(stderr, "ignore writes (-I) "
916 "injection requires a single pool name\n");
921 (void) strcpy(pool, argv[0]);
923 } else if (type == TYPE_INVAL) {
925 (void) fprintf(stderr, "at least one of '-b', '-d', "
926 "'-t', '-a', '-p', '-I' or '-u' "
927 "must be specified\n");
932 if (argc == 1 && (flags & ZINJECT_UNLOAD_SPA)) {
933 (void) strcpy(pool, argv[0]);
935 } else if (argc != 0) {
936 (void) fprintf(stderr, "extraneous argument for "
942 flags |= ZINJECT_NULL;
945 (void) fprintf(stderr, "missing object\n");
950 if (error == ENXIO) {
951 (void) fprintf(stderr, "data error type must be "
952 "'checksum' or 'io'\n");
956 record.zi_cmd = ZINJECT_DATA_FAULT;
957 if (translate_record(type, argv[0], range, level, &record, pool,
965 * If this is pool-wide metadata, unmount everything. The ioctl() will
966 * unload the pool, so that we trigger spa-wide reopen of metadata next
967 * time we access the pool.
969 if (dataset[0] != '\0' && domount) {
970 if ((zhp = zfs_open(g_zfs, dataset, ZFS_TYPE_DATASET)) == NULL)
973 if (zfs_unmount(zhp, NULL, 0) != 0)
977 record.zi_error = error;
979 ret = register_handler(pool, flags, &record, quiet);
981 if (dataset[0] != '\0' && domount)
982 ret = (zfs_mount(zhp, NULL, 0) != 0);