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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
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 or uberblock region of all the labels
46 * for the specified device.
48 * This form of the command looks like:
50 * zinject -d device [-e errno] [-L <uber | nvlist>] 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/mount.h>
151 #undef verify /* both libzfs.h and zfs_context.h want to define this */
155 libzfs_handle_t *g_zfs;
160 static const char *errtable[TYPE_INVAL] = {
175 name_to_type(const char *arg)
178 for (i = 0; i < TYPE_INVAL; i++)
179 if (strcmp(errtable[i], arg) == 0)
186 type_to_name(uint64_t type)
189 case DMU_OT_OBJECT_DIRECTORY:
191 case DMU_OT_OBJECT_ARRAY:
193 case DMU_OT_PACKED_NVLIST:
197 case DMU_OT_SPACE_MAP:
199 case DMU_OT_ERROR_LOG:
208 * Print usage message.
218 "\t\tList all active injection records.\n"
220 "\tzinject -c <id|all>\n"
222 "\t\tClear the particular record (if given a numeric ID), or\n"
223 "\t\tall records if 'all' is specificed.\n"
225 "\tzinject -d device [-e errno] [-L <nvlist|uber>] [-F] pool\n"
226 "\t\tInject a fault into a particular device or the device's\n"
227 "\t\tlabel. Label injection can either be 'nvlist' or 'uber'.\n"
228 "\t\t'errno' can either be 'nxio' (the default) or 'io'.\n"
230 "\tzinject -b objset:object:level:blkid pool\n"
232 "\t\tInject an error into pool 'pool' with the numeric bookmark\n"
233 "\t\tspecified by the remaining tuple. Each number is in\n"
234 "\t\thexidecimal, and only one block can be specified.\n"
236 "\tzinject [-q] <-t type> [-e errno] [-l level] [-r range]\n"
237 "\t [-a] [-m] [-u] [-f freq] <object>\n"
239 "\t\tInject an error into the object specified by the '-t' option\n"
240 "\t\tand the object descriptor. The 'object' parameter is\n"
241 "\t\tinterperted depending on the '-t' option.\n"
243 "\t\t-q\tQuiet mode. Only print out the handler number added.\n"
244 "\t\t-e\tInject a specific error. Must be either 'io' or\n"
245 "\t\t\t'checksum'. Default is 'io'.\n"
246 "\t\t-l\tInject error at a particular block level. Default is "
248 "\t\t-m\tAutomatically remount underlying filesystem.\n"
249 "\t\t-r\tInject error over a particular logical range of an\n"
250 "\t\t\tobject. Will be translated to the appropriate blkid\n"
251 "\t\t\trange according to the object's properties.\n"
252 "\t\t-a\tFlush the ARC cache. Can be specified without any\n"
253 "\t\t\tassociated object.\n"
254 "\t\t-u\tUnload the associated pool. Can be specified with only\n"
255 "\t\t\ta pool object.\n"
256 "\t\t-f\tOnly inject errors a fraction of the time. Expressed as\n"
257 "\t\t\ta percentage between 1 and 100.\n"
259 "\t-t data\t\tInject an error into the plain file contents of a\n"
260 "\t\t\tfile. The object must be specified as a complete path\n"
261 "\t\t\tto a file on a ZFS filesystem.\n"
263 "\t-t dnode\tInject an error into the metadnode in the block\n"
264 "\t\t\tcorresponding to the dnode for a file or directory. The\n"
265 "\t\t\t'-r' option is incompatible with this mode. The object\n"
266 "\t\t\tis specified as a complete path to a file or directory\n"
267 "\t\t\ton a ZFS filesystem.\n"
269 "\t-t <mos>\tInject errors into the MOS for objects of the given\n"
270 "\t\t\ttype. Valid types are: mos, mosdir, config, bplist,\n"
271 "\t\t\tspacemap, metaslab, errlog. The only valid <object> is\n"
272 "\t\t\tthe poolname.\n");
276 iter_handlers(int (*func)(int, const char *, zinject_record_t *, void *),
284 while (ioctl(zfs_fd, ZFS_IOC_INJECT_LIST_NEXT, &zc) == 0)
285 if ((ret = func((int)zc.zc_guid, zc.zc_name,
286 &zc.zc_inject_record, data)) != 0)
293 print_data_handler(int id, const char *pool, zinject_record_t *record,
298 if (record->zi_guid != 0)
302 (void) printf("%3s %-15s %-6s %-6s %-8s %3s %-15s\n",
303 "ID", "POOL", "OBJSET", "OBJECT", "TYPE", "LVL", "RANGE");
304 (void) printf("--- --------------- ------ "
305 "------ -------- --- ---------------\n");
310 (void) printf("%3d %-15s %-6llu %-6llu %-8s %3d ", id, pool,
311 (u_longlong_t)record->zi_objset, (u_longlong_t)record->zi_object,
312 type_to_name(record->zi_type), record->zi_level);
314 if (record->zi_start == 0 &&
315 record->zi_end == -1ULL)
316 (void) printf("all\n");
318 (void) printf("[%llu, %llu]\n", (u_longlong_t)record->zi_start,
319 (u_longlong_t)record->zi_end);
325 print_device_handler(int id, const char *pool, zinject_record_t *record,
330 if (record->zi_guid == 0)
334 (void) printf("%3s %-15s %s\n", "ID", "POOL", "GUID");
335 (void) printf("--- --------------- ----------------\n");
340 (void) printf("%3d %-15s %llx\n", id, pool,
341 (u_longlong_t)record->zi_guid);
347 * Print all registered error handlers. Returns the number of handlers
351 print_all_handlers(void)
355 (void) iter_handlers(print_device_handler, &count);
358 (void) iter_handlers(print_data_handler, &count);
365 cancel_one_handler(int id, const char *pool, zinject_record_t *record,
370 zc.zc_guid = (uint64_t)id;
372 if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) {
373 (void) fprintf(stderr, "failed to remove handler %d: %s\n",
374 id, strerror(errno));
382 * Remove all fault injection handlers.
385 cancel_all_handlers(void)
387 int ret = iter_handlers(cancel_one_handler, NULL);
389 (void) printf("removed all registered handlers\n");
395 * Remove a specific fault injection handler.
398 cancel_handler(int id)
402 zc.zc_guid = (uint64_t)id;
404 if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) {
405 (void) fprintf(stderr, "failed to remove handler %d: %s\n",
406 id, strerror(errno));
410 (void) printf("removed handler %d\n", id);
416 * Register a new fault injection handler.
419 register_handler(const char *pool, int flags, zinject_record_t *record,
424 (void) strcpy(zc.zc_name, pool);
425 zc.zc_inject_record = *record;
428 if (ioctl(zfs_fd, ZFS_IOC_INJECT_FAULT, &zc) != 0) {
429 (void) fprintf(stderr, "failed to add handler: %s\n",
434 if (flags & ZINJECT_NULL)
438 (void) printf("%llu\n", (u_longlong_t)zc.zc_guid);
440 (void) printf("Added handler %llu with the following "
441 "properties:\n", (u_longlong_t)zc.zc_guid);
442 (void) printf(" pool: %s\n", pool);
443 if (record->zi_guid) {
444 (void) printf(" vdev: %llx\n",
445 (u_longlong_t)record->zi_guid);
447 (void) printf("objset: %llu\n",
448 (u_longlong_t)record->zi_objset);
449 (void) printf("object: %llu\n",
450 (u_longlong_t)record->zi_object);
451 (void) printf(" type: %llu\n",
452 (u_longlong_t)record->zi_type);
453 (void) printf(" level: %d\n", record->zi_level);
454 if (record->zi_start == 0 &&
455 record->zi_end == -1ULL)
456 (void) printf(" range: all\n");
458 (void) printf(" range: [%llu, %llu)\n",
459 (u_longlong_t)record->zi_start,
460 (u_longlong_t)record->zi_end);
468 main(int argc, char **argv)
480 err_type_t type = TYPE_INVAL;
481 err_type_t label = TYPE_INVAL;
482 zinject_record_t record = { 0 };
483 char pool[MAXNAMELEN];
484 char dataset[MAXNAMELEN];
489 if ((g_zfs = libzfs_init()) == NULL) {
490 (void) fprintf(stderr, "internal error: failed to "
491 "initialize ZFS library\n");
495 libzfs_print_on_error(g_zfs, B_TRUE);
497 if ((zfs_fd = open(ZFS_DEV, O_RDWR)) < 0) {
498 (void) fprintf(stderr, "failed to open ZFS device\n");
504 * No arguments. Print the available handlers. If there are no
505 * available handlers, direct the user to '-h' for help
508 if (print_all_handlers() == 0) {
509 (void) printf("No handlers registered.\n");
510 (void) printf("Run 'zinject -h' for usage "
517 while ((c = getopt(argc, argv, ":ab:d:f:Fqhc:t:l:mr:e:uL:")) != -1) {
520 flags |= ZINJECT_FLUSH_ARC;
532 if (strcasecmp(optarg, "io") == 0) {
534 } else if (strcasecmp(optarg, "checksum") == 0) {
536 } else if (strcasecmp(optarg, "nxio") == 0) {
539 (void) fprintf(stderr, "invalid error type "
540 "'%s': must be 'io', 'checksum' or "
547 record.zi_freq = atoi(optarg);
548 if (record.zi_freq < 1 || record.zi_freq > 100) {
549 (void) fprintf(stderr, "frequency range must "
550 "be in the range (0, 100]\n");
555 record.zi_failfast = B_TRUE;
561 level = (int)strtol(optarg, &end, 10);
563 (void) fprintf(stderr, "invalid level '%s': "
564 "must be an integer\n", optarg);
579 if ((type = name_to_type(optarg)) == TYPE_INVAL &&
581 (void) fprintf(stderr, "invalid type '%s'\n",
588 flags |= ZINJECT_UNLOAD_SPA;
591 if ((label = name_to_type(optarg)) == TYPE_INVAL &&
593 (void) fprintf(stderr, "invalid label type "
600 (void) fprintf(stderr, "option -%c requires an "
601 "operand\n", optopt);
605 (void) fprintf(stderr, "invalid option '%c'\n",
615 if (cancel != NULL) {
617 * '-c' is invalid with any other options.
619 if (raw != NULL || range != NULL || type != TYPE_INVAL ||
621 (void) fprintf(stderr, "cancel (-c) incompatible with "
622 "any other options\n");
627 (void) fprintf(stderr, "extraneous argument to '-c'\n");
632 if (strcmp(cancel, "all") == 0) {
633 return (cancel_all_handlers());
635 int id = (int)strtol(cancel, &end, 10);
637 (void) fprintf(stderr, "invalid handle id '%s':"
638 " must be an integer or 'all'\n", cancel);
642 return (cancel_handler(id));
646 if (device != NULL) {
648 * Device (-d) injection uses a completely different mechanism
649 * for doing injection, so handle it separately here.
651 if (raw != NULL || range != NULL || type != TYPE_INVAL ||
653 (void) fprintf(stderr, "device (-d) incompatible with "
654 "data error injection\n");
660 (void) fprintf(stderr, "device (-d) injection requires "
661 "a single pool name\n");
666 (void) strcpy(pool, argv[0]);
669 if (error == ECKSUM) {
670 (void) fprintf(stderr, "device error type must be "
675 if (translate_device(pool, device, label, &record) != 0)
679 } else if (raw != NULL) {
680 if (range != NULL || type != TYPE_INVAL || level != 0) {
681 (void) fprintf(stderr, "raw (-b) format with "
682 "any other options\n");
688 (void) fprintf(stderr, "raw (-b) format expects a "
689 "single pool name\n");
694 (void) strcpy(pool, argv[0]);
697 if (error == ENXIO) {
698 (void) fprintf(stderr, "data error type must be "
699 "'checksum' or 'io'\n");
703 if (translate_raw(raw, &record) != 0)
707 } else if (type == TYPE_INVAL) {
709 (void) fprintf(stderr, "at least one of '-b', '-d', "
710 "'-t', '-a', or '-u' must be specified\n");
715 if (argc == 1 && (flags & ZINJECT_UNLOAD_SPA)) {
716 (void) strcpy(pool, argv[0]);
718 } else if (argc != 0) {
719 (void) fprintf(stderr, "extraneous argument for "
725 flags |= ZINJECT_NULL;
728 (void) fprintf(stderr, "missing object\n");
733 if (error == ENXIO) {
734 (void) fprintf(stderr, "data error type must be "
735 "'checksum' or 'io'\n");
739 if (translate_record(type, argv[0], range, level, &record, pool,
747 * If this is pool-wide metadata, unmount everything. The ioctl() will
748 * unload the pool, so that we trigger spa-wide reopen of metadata next
749 * time we access the pool.
751 if (dataset[0] != '\0' && domount) {
752 if ((zhp = zfs_open(g_zfs, dataset, ZFS_TYPE_DATASET)) == NULL)
755 if (zfs_unmount(zhp, NULL, 0) != 0)
759 record.zi_error = error;
761 ret = register_handler(pool, flags, &record, quiet);
763 if (dataset[0] != '\0' && domount)
764 ret = (zfs_mount(zhp, NULL, 0) != 0);