1/*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24 */
25
26/*
27 * ZFS Fault Injector
28 *
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.
32 *
33 *
34 * DEVICE FAULTS
35 *
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.
47 *
48 * This form of the command looks like:
49 *
50 * 	zinject -d device [-e errno] [-L <uber | nvlist | pad1 | pad2>] pool
51 *
52 *
53 * DATA FAULTS
54 *
55 * We begin with a tuple of the form:
56 *
57 * 	<type,level,range,object>
58 *
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:
62 *
63 * 		data		User data for a file
64 * 		dnode		Dnode for a file or directory
65 *
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.
69 *
70 * 		mos		Any data in the MOS
71 * 		mosdir		object directory
72 * 		config		pool configuration
73 * 		bpobj		blkptr list
74 * 		spacemap	spacemap
75 * 		metaslab	metaslab
76 * 		errlog		persistent error log
77 *
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.
81 *
82 *	range	A numerical range [start,end) within the object.  Defaults to
83 *		the full size of the file.
84 *
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.
88 *
89 * This is translated, via libzpool, into the following internal representation:
90 *
91 * 	<type,objset,object,level,range>
92 *
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.
97 *
98 *
99 * The command itself takes one of the forms:
100 *
101 * 	zinject
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
107 *
108 * With no arguments, the command prints all currently registered injection
109 * handlers, with their numeric identifiers.
110 *
111 * The '-c' option will clear the given handler, or all handlers if 'all' is
112 * specified.
113 *
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
117 * situation.
118 *
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
124 * cache.
125 *
126 * The '-f' flag controls the frequency of errors injected, expressed as a
127 * integer percentage between 1 and 100.  The default is 100.
128 *
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
132 * handler.
133 *
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.
137 */
138
139#include <errno.h>
140#include <fcntl.h>
141#include <stdio.h>
142#include <stdlib.h>
143#include <strings.h>
144#include <unistd.h>
145
146#include <sys/fs/zfs.h>
147#include <sys/param.h>
148#include <sys/mount.h>
149
150#include <libzfs.h>
151#include <libzfs_compat.h>
152
153#undef verify	/* both libzfs.h and zfs_context.h want to define this */
154
155#include "zinject.h"
156
157libzfs_handle_t *g_zfs;
158int zfs_fd;
159
160#ifndef ECKSUM
161#define	ECKSUM	EBADE
162#endif
163
164static const char *errtable[TYPE_INVAL] = {
165	"data",
166	"dnode",
167	"mos",
168	"mosdir",
169	"metaslab",
170	"config",
171	"bpobj",
172	"spacemap",
173	"errlog",
174	"uber",
175	"nvlist",
176	"pad1",
177	"pad2"
178};
179
180static err_type_t
181name_to_type(const char *arg)
182{
183	int i;
184	for (i = 0; i < TYPE_INVAL; i++)
185		if (strcmp(errtable[i], arg) == 0)
186			return (i);
187
188	return (TYPE_INVAL);
189}
190
191static const char *
192type_to_name(uint64_t type)
193{
194	switch (type) {
195	case DMU_OT_OBJECT_DIRECTORY:
196		return ("mosdir");
197	case DMU_OT_OBJECT_ARRAY:
198		return ("metaslab");
199	case DMU_OT_PACKED_NVLIST:
200		return ("config");
201	case DMU_OT_BPOBJ:
202		return ("bpobj");
203	case DMU_OT_SPACE_MAP:
204		return ("spacemap");
205	case DMU_OT_ERROR_LOG:
206		return ("errlog");
207	default:
208		return ("-");
209	}
210}
211
212
213/*
214 * Print usage message.
215 */
216void
217usage(void)
218{
219	(void) printf(
220	    "usage:\n"
221	    "\n"
222	    "\tzinject\n"
223	    "\n"
224	    "\t\tList all active injection records.\n"
225	    "\n"
226	    "\tzinject -c <id|all>\n"
227	    "\n"
228	    "\t\tClear the particular record (if given a numeric ID), or\n"
229	    "\t\tall records if 'all' is specificed.\n"
230	    "\n"
231	    "\tzinject -p <function name> pool\n"
232	    "\n"
233	    "\t\tInject a panic fault at the specified function. Only \n"
234	    "\t\tfunctions which call spa_vdev_config_exit(), or \n"
235	    "\t\tspa_vdev_exit() will trigger a panic.\n"
236	    "\n"
237	    "\tzinject -d device [-e errno] [-L <nvlist|uber|pad1|pad2>] [-F]\n"
238	    "\t    [-T <read|write|free|claim|all> pool\n"
239	    "\n"
240	    "\t\tInject a fault into a particular device or the device's\n"
241	    "\t\tlabel.  Label injection can either be 'nvlist', 'uber',\n "
242	    "\t\t'pad1', or 'pad2'.\n"
243	    "\t\t'errno' can be 'nxio' (the default), 'io', or 'dtl'.\n"
244	    "\n"
245	    "\tzinject -d device -A <degrade|fault> pool\n"
246	    "\n"
247	    "\t\tPerform a specific action on a particular device\n"
248	    "\n"
249	    "\tzinject -d device -D latency:lanes pool\n"
250	    "\n"
251	    "\t\tAdd an artificial delay to IO requests on a particular\n"
252	    "\t\tdevice, such that the requests take a minimum of 'latency'\n"
253	    "\t\tmilliseconds to complete. Each delay has an associated\n"
254	    "\t\tnumber of 'lanes' which defines the number of concurrent\n"
255	    "\t\tIO requests that can be processed.\n"
256	    "\n"
257	    "\t\tFor example, with a single lane delay of 10 ms (-D 10:1),\n"
258	    "\t\tthe device will only be able to service a single IO request\n"
259	    "\t\tat a time with each request taking 10 ms to complete. So,\n"
260	    "\t\tif only a single request is submitted every 10 ms, the\n"
261	    "\t\taverage latency will be 10 ms; but if more than one request\n"
262	    "\t\tis submitted every 10 ms, the average latency will be more\n"
263	    "\t\tthan 10 ms.\n"
264	    "\n"
265	    "\t\tSimilarly, if a delay of 10 ms is specified to have two\n"
266	    "\t\tlanes (-D 10:2), then the device will be able to service\n"
267	    "\t\ttwo requests at a time, each with a minimum latency of\n"
268	    "\t\t10 ms. So, if two requests are submitted every 10 ms, then\n"
269	    "\t\tthe average latency will be 10 ms; but if more than two\n"
270	    "\t\trequests are submitted every 10 ms, the average latency\n"
271	    "\t\twill be more than 10 ms.\n"
272	    "\n"
273	    "\t\tAlso note, these delays are additive. So two invocations\n"
274	    "\t\tof '-D 10:1', is roughly equivalent to a single invocation\n"
275	    "\t\tof '-D 10:2'. This also means, one can specify multiple\n"
276	    "\t\tlanes with differing target latencies. For example, an\n"
277	    "\t\tinvocation of '-D 10:1' followed by '-D 25:2' will\n"
278	    "\t\tcreate 3 lanes on the device; one lane with a latency\n"
279	    "\t\tof 10 ms and two lanes with a 25 ms latency.\n"
280	    "\n"
281	    "\tzinject -I [-s <seconds> | -g <txgs>] pool\n"
282	    "\n"
283	    "\t\tCause the pool to stop writing blocks yet not\n"
284	    "\t\treport errors for a duration.  Simulates buggy hardware\n"
285	    "\t\tthat fails to honor cache flush requests.\n"
286	    "\t\tDefault duration is 30 seconds.  The machine is panicked\n"
287	    "\t\tat the end of the duration.\n"
288	    "\n"
289	    "\tzinject -b objset:object:level:blkid pool\n"
290	    "\n"
291	    "\t\tInject an error into pool 'pool' with the numeric bookmark\n"
292	    "\t\tspecified by the remaining tuple.  Each number is in\n"
293	    "\t\thexidecimal, and only one block can be specified.\n"
294	    "\n"
295	    "\tzinject [-q] <-t type> [-e errno] [-l level] [-r range]\n"
296	    "\t    [-a] [-m] [-u] [-f freq] <object>\n"
297	    "\n"
298	    "\t\tInject an error into the object specified by the '-t' option\n"
299	    "\t\tand the object descriptor.  The 'object' parameter is\n"
300	    "\t\tinterperted depending on the '-t' option.\n"
301	    "\n"
302	    "\t\t-q\tQuiet mode.  Only print out the handler number added.\n"
303	    "\t\t-e\tInject a specific error.  Must be either 'io' or\n"
304	    "\t\t\t'checksum'.  Default is 'io'.\n"
305	    "\t\t-l\tInject error at a particular block level. Default is "
306	    "0.\n"
307	    "\t\t-m\tAutomatically remount underlying filesystem.\n"
308	    "\t\t-r\tInject error over a particular logical range of an\n"
309	    "\t\t\tobject.  Will be translated to the appropriate blkid\n"
310	    "\t\t\trange according to the object's properties.\n"
311	    "\t\t-a\tFlush the ARC cache.  Can be specified without any\n"
312	    "\t\t\tassociated object.\n"
313	    "\t\t-u\tUnload the associated pool.  Can be specified with only\n"
314	    "\t\t\ta pool object.\n"
315	    "\t\t-f\tOnly inject errors a fraction of the time.  Expressed as\n"
316	    "\t\t\ta percentage between 1 and 100.\n"
317	    "\n"
318	    "\t-t data\t\tInject an error into the plain file contents of a\n"
319	    "\t\t\tfile.  The object must be specified as a complete path\n"
320	    "\t\t\tto a file on a ZFS filesystem.\n"
321	    "\n"
322	    "\t-t dnode\tInject an error into the metadnode in the block\n"
323	    "\t\t\tcorresponding to the dnode for a file or directory.  The\n"
324	    "\t\t\t'-r' option is incompatible with this mode.  The object\n"
325	    "\t\t\tis specified as a complete path to a file or directory\n"
326	    "\t\t\ton a ZFS filesystem.\n"
327	    "\n"
328	    "\t-t <mos>\tInject errors into the MOS for objects of the given\n"
329	    "\t\t\ttype.  Valid types are: mos, mosdir, config, bpobj,\n"
330	    "\t\t\tspacemap, metaslab, errlog.  The only valid <object> is\n"
331	    "\t\t\tthe poolname.\n");
332}
333
334static int
335iter_handlers(int (*func)(int, const char *, zinject_record_t *, void *),
336    void *data)
337{
338	zfs_cmd_t zc = { 0 };
339	int ret;
340
341	while (ioctl(zfs_fd, ZFS_IOC_INJECT_LIST_NEXT, &zc) == 0)
342		if ((ret = func((int)zc.zc_guid, zc.zc_name,
343		    &zc.zc_inject_record, data)) != 0)
344			return (ret);
345
346	if (errno != ENOENT) {
347		(void) fprintf(stderr, "Unable to list handlers: %s\n",
348		    strerror(errno));
349		return (-1);
350	}
351
352	return (0);
353}
354
355static int
356print_data_handler(int id, const char *pool, zinject_record_t *record,
357    void *data)
358{
359	int *count = data;
360
361	if (record->zi_guid != 0 || record->zi_func[0] != '\0')
362		return (0);
363
364	if (*count == 0) {
365		(void) printf("%3s  %-15s  %-6s  %-6s  %-8s  %3s  %-15s\n",
366		    "ID", "POOL", "OBJSET", "OBJECT", "TYPE", "LVL",  "RANGE");
367		(void) printf("---  ---------------  ------  "
368		    "------  --------  ---  ---------------\n");
369	}
370
371	*count += 1;
372
373	(void) printf("%3d  %-15s  %-6llu  %-6llu  %-8s  %3d  ", id, pool,
374	    (u_longlong_t)record->zi_objset, (u_longlong_t)record->zi_object,
375	    type_to_name(record->zi_type), record->zi_level);
376
377	if (record->zi_start == 0 &&
378	    record->zi_end == -1ULL)
379		(void) printf("all\n");
380	else
381		(void) printf("[%llu, %llu]\n", (u_longlong_t)record->zi_start,
382		    (u_longlong_t)record->zi_end);
383
384	return (0);
385}
386
387static int
388print_device_handler(int id, const char *pool, zinject_record_t *record,
389    void *data)
390{
391	int *count = data;
392
393	if (record->zi_guid == 0 || record->zi_func[0] != '\0')
394		return (0);
395
396	if (record->zi_cmd == ZINJECT_DELAY_IO)
397		return (0);
398
399	if (*count == 0) {
400		(void) printf("%3s  %-15s  %s\n", "ID", "POOL", "GUID");
401		(void) printf("---  ---------------  ----------------\n");
402	}
403
404	*count += 1;
405
406	(void) printf("%3d  %-15s  %llx\n", id, pool,
407	    (u_longlong_t)record->zi_guid);
408
409	return (0);
410}
411
412static int
413print_delay_handler(int id, const char *pool, zinject_record_t *record,
414    void *data)
415{
416	int *count = data;
417
418	if (record->zi_guid == 0 || record->zi_func[0] != '\0')
419		return (0);
420
421	if (record->zi_cmd != ZINJECT_DELAY_IO)
422		return (0);
423
424	if (*count == 0) {
425		(void) printf("%3s  %-15s  %-15s  %-15s  %s\n",
426		    "ID", "POOL", "DELAY (ms)", "LANES", "GUID");
427		(void) printf("---  ---------------  ---------------  "
428		    "---------------  ----------------\n");
429	}
430
431	*count += 1;
432
433	(void) printf("%3d  %-15s  %-15llu  %-15llu  %llx\n", id, pool,
434	    (u_longlong_t)NSEC2MSEC(record->zi_timer),
435	    (u_longlong_t)record->zi_nlanes,
436	    (u_longlong_t)record->zi_guid);
437
438	return (0);
439}
440
441static int
442print_panic_handler(int id, const char *pool, zinject_record_t *record,
443    void *data)
444{
445	int *count = data;
446
447	if (record->zi_func[0] == '\0')
448		return (0);
449
450	if (*count == 0) {
451		(void) printf("%3s  %-15s  %s\n", "ID", "POOL", "FUNCTION");
452		(void) printf("---  ---------------  ----------------\n");
453	}
454
455	*count += 1;
456
457	(void) printf("%3d  %-15s  %s\n", id, pool, record->zi_func);
458
459	return (0);
460}
461
462/*
463 * Print all registered error handlers.  Returns the number of handlers
464 * registered.
465 */
466static int
467print_all_handlers(void)
468{
469	int count = 0, total = 0;
470
471	(void) iter_handlers(print_device_handler, &count);
472	if (count > 0) {
473		total += count;
474		(void) printf("\n");
475		count = 0;
476	}
477
478	(void) iter_handlers(print_delay_handler, &count);
479	if (count > 0) {
480		total += count;
481		(void) printf("\n");
482		count = 0;
483	}
484
485	(void) iter_handlers(print_data_handler, &count);
486	if (count > 0) {
487		total += count;
488		(void) printf("\n");
489		count = 0;
490	}
491
492	(void) iter_handlers(print_panic_handler, &count);
493
494	return (count + total);
495}
496
497/* ARGSUSED */
498static int
499cancel_one_handler(int id, const char *pool, zinject_record_t *record,
500    void *data)
501{
502	zfs_cmd_t zc = { 0 };
503
504	zc.zc_guid = (uint64_t)id;
505
506	if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) {
507		(void) fprintf(stderr, "failed to remove handler %d: %s\n",
508		    id, strerror(errno));
509		return (1);
510	}
511
512	return (0);
513}
514
515/*
516 * Remove all fault injection handlers.
517 */
518static int
519cancel_all_handlers(void)
520{
521	int ret = iter_handlers(cancel_one_handler, NULL);
522
523	if (ret == 0)
524		(void) printf("removed all registered handlers\n");
525
526	return (ret);
527}
528
529/*
530 * Remove a specific fault injection handler.
531 */
532static int
533cancel_handler(int id)
534{
535	zfs_cmd_t zc = { 0 };
536
537	zc.zc_guid = (uint64_t)id;
538
539	if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) {
540		(void) fprintf(stderr, "failed to remove handler %d: %s\n",
541		    id, strerror(errno));
542		return (1);
543	}
544
545	(void) printf("removed handler %d\n", id);
546
547	return (0);
548}
549
550/*
551 * Register a new fault injection handler.
552 */
553static int
554register_handler(const char *pool, int flags, zinject_record_t *record,
555    int quiet)
556{
557	zfs_cmd_t zc = { 0 };
558
559	(void) strcpy(zc.zc_name, pool);
560	zc.zc_inject_record = *record;
561	zc.zc_guid = flags;
562
563	if (ioctl(zfs_fd, ZFS_IOC_INJECT_FAULT, &zc) != 0) {
564		(void) fprintf(stderr, "failed to add handler: %s\n",
565		    strerror(errno));
566		return (1);
567	}
568
569	if (flags & ZINJECT_NULL)
570		return (0);
571
572	if (quiet) {
573		(void) printf("%llu\n", (u_longlong_t)zc.zc_guid);
574	} else {
575		(void) printf("Added handler %llu with the following "
576		    "properties:\n", (u_longlong_t)zc.zc_guid);
577		(void) printf("  pool: %s\n", pool);
578		if (record->zi_guid) {
579			(void) printf("  vdev: %llx\n",
580			    (u_longlong_t)record->zi_guid);
581		} else if (record->zi_func[0] != '\0') {
582			(void) printf("  panic function: %s\n",
583			    record->zi_func);
584		} else if (record->zi_duration > 0) {
585			(void) printf(" time: %lld seconds\n",
586			    (u_longlong_t)record->zi_duration);
587		} else if (record->zi_duration < 0) {
588			(void) printf(" txgs: %lld \n",
589			    (u_longlong_t)-record->zi_duration);
590		} else {
591			(void) printf("objset: %llu\n",
592			    (u_longlong_t)record->zi_objset);
593			(void) printf("object: %llu\n",
594			    (u_longlong_t)record->zi_object);
595			(void) printf("  type: %llu\n",
596			    (u_longlong_t)record->zi_type);
597			(void) printf(" level: %d\n", record->zi_level);
598			if (record->zi_start == 0 &&
599			    record->zi_end == -1ULL)
600				(void) printf(" range: all\n");
601			else
602				(void) printf(" range: [%llu, %llu)\n",
603				    (u_longlong_t)record->zi_start,
604				    (u_longlong_t)record->zi_end);
605		}
606	}
607
608	return (0);
609}
610
611int
612perform_action(const char *pool, zinject_record_t *record, int cmd)
613{
614	zfs_cmd_t zc = { 0 };
615
616	ASSERT(cmd == VDEV_STATE_DEGRADED || cmd == VDEV_STATE_FAULTED);
617	(void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
618	zc.zc_guid = record->zi_guid;
619	zc.zc_cookie = cmd;
620
621	if (ioctl(zfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
622		return (0);
623
624	return (1);
625}
626
627static int
628parse_delay(char *str, uint64_t *delay, uint64_t *nlanes)
629{
630	unsigned long scan_delay;
631	unsigned long scan_nlanes;
632
633	if (sscanf(str, "%lu:%lu", &scan_delay, &scan_nlanes) != 2)
634		return (1);
635
636	/*
637	 * We explicitly disallow a delay of zero here, because we key
638	 * off this value being non-zero in translate_device(), to
639	 * determine if the fault is a ZINJECT_DELAY_IO fault or not.
640	 */
641	if (scan_delay == 0)
642		return (1);
643
644	/*
645	 * The units for the CLI delay parameter is milliseconds, but
646	 * the data passed to the kernel is interpreted as nanoseconds.
647	 * Thus we scale the milliseconds to nanoseconds here, and this
648	 * nanosecond value is used to pass the delay to the kernel.
649	 */
650	*delay = MSEC2NSEC(scan_delay);
651	*nlanes = scan_nlanes;
652
653	return (0);
654}
655
656int
657main(int argc, char **argv)
658{
659	int c;
660	char *range = NULL;
661	char *cancel = NULL;
662	char *end;
663	char *raw = NULL;
664	char *device = NULL;
665	int level = 0;
666	int quiet = 0;
667	int error = 0;
668	int domount = 0;
669	int io_type = ZIO_TYPES;
670	int action = VDEV_STATE_UNKNOWN;
671	err_type_t type = TYPE_INVAL;
672	err_type_t label = TYPE_INVAL;
673	zinject_record_t record = { 0 };
674	char pool[MAXNAMELEN];
675	char dataset[MAXNAMELEN];
676	zfs_handle_t *zhp;
677	int nowrites = 0;
678	int dur_txg = 0;
679	int dur_secs = 0;
680	int ret;
681	int flags = 0;
682
683	if ((g_zfs = libzfs_init()) == NULL) {
684		(void) fprintf(stderr, "internal error: failed to "
685		    "initialize ZFS library\n");
686		return (1);
687	}
688
689	libzfs_print_on_error(g_zfs, B_TRUE);
690
691	if ((zfs_fd = open(ZFS_DEV, O_RDWR)) < 0) {
692		(void) fprintf(stderr, "failed to open ZFS device\n");
693		return (1);
694	}
695
696	if (argc == 1) {
697		/*
698		 * No arguments.  Print the available handlers.  If there are no
699		 * available handlers, direct the user to '-h' for help
700		 * information.
701		 */
702		if (print_all_handlers() == 0) {
703			(void) printf("No handlers registered.\n");
704			(void) printf("Run 'zinject -h' for usage "
705			    "information.\n");
706		}
707
708		return (0);
709	}
710
711	while ((c = getopt(argc, argv,
712	    ":aA:b:d:D:f:Fg:qhIc:t:T:l:mr:s:e:uL:p:")) != -1) {
713		switch (c) {
714		case 'a':
715			flags |= ZINJECT_FLUSH_ARC;
716			break;
717		case 'A':
718			if (strcasecmp(optarg, "degrade") == 0) {
719				action = VDEV_STATE_DEGRADED;
720			} else if (strcasecmp(optarg, "fault") == 0) {
721				action = VDEV_STATE_FAULTED;
722			} else {
723				(void) fprintf(stderr, "invalid action '%s': "
724				    "must be 'degrade' or 'fault'\n", optarg);
725				usage();
726				return (1);
727			}
728			break;
729		case 'b':
730			raw = optarg;
731			break;
732		case 'c':
733			cancel = optarg;
734			break;
735		case 'd':
736			device = optarg;
737			break;
738		case 'D':
739			ret = parse_delay(optarg, &record.zi_timer,
740			    &record.zi_nlanes);
741			if (ret != 0) {
742				(void) fprintf(stderr, "invalid i/o delay "
743				    "value: '%s'\n", optarg);
744				usage();
745				return (1);
746			}
747			break;
748		case 'e':
749			if (strcasecmp(optarg, "io") == 0) {
750				error = EIO;
751			} else if (strcasecmp(optarg, "checksum") == 0) {
752				error = ECKSUM;
753			} else if (strcasecmp(optarg, "nxio") == 0) {
754				error = ENXIO;
755			} else if (strcasecmp(optarg, "dtl") == 0) {
756				error = ECHILD;
757			} else {
758				(void) fprintf(stderr, "invalid error type "
759				    "'%s': must be 'io', 'checksum' or "
760				    "'nxio'\n", optarg);
761				usage();
762				return (1);
763			}
764			break;
765		case 'f':
766			record.zi_freq = atoi(optarg);
767			if (record.zi_freq < 1 || record.zi_freq > 100) {
768				(void) fprintf(stderr, "frequency range must "
769				    "be in the range (0, 100]\n");
770				return (1);
771			}
772			break;
773		case 'F':
774			record.zi_failfast = B_TRUE;
775			break;
776		case 'g':
777			dur_txg = 1;
778			record.zi_duration = (int)strtol(optarg, &end, 10);
779			if (record.zi_duration <= 0 || *end != '\0') {
780				(void) fprintf(stderr, "invalid duration '%s': "
781				    "must be a positive integer\n", optarg);
782				usage();
783				return (1);
784			}
785			/* store duration of txgs as its negative */
786			record.zi_duration *= -1;
787			break;
788		case 'h':
789			usage();
790			return (0);
791		case 'I':
792			/* default duration, if one hasn't yet been defined */
793			nowrites = 1;
794			if (dur_secs == 0 && dur_txg == 0)
795				record.zi_duration = 30;
796			break;
797		case 'l':
798			level = (int)strtol(optarg, &end, 10);
799			if (*end != '\0') {
800				(void) fprintf(stderr, "invalid level '%s': "
801				    "must be an integer\n", optarg);
802				usage();
803				return (1);
804			}
805			break;
806		case 'm':
807			domount = 1;
808			break;
809		case 'p':
810			(void) strlcpy(record.zi_func, optarg,
811			    sizeof (record.zi_func));
812			record.zi_cmd = ZINJECT_PANIC;
813			break;
814		case 'q':
815			quiet = 1;
816			break;
817		case 'r':
818			range = optarg;
819			break;
820		case 's':
821			dur_secs = 1;
822			record.zi_duration = (int)strtol(optarg, &end, 10);
823			if (record.zi_duration <= 0 || *end != '\0') {
824				(void) fprintf(stderr, "invalid duration '%s': "
825				    "must be a positive integer\n", optarg);
826				usage();
827				return (1);
828			}
829			break;
830		case 'T':
831			if (strcasecmp(optarg, "read") == 0) {
832				io_type = ZIO_TYPE_READ;
833			} else if (strcasecmp(optarg, "write") == 0) {
834				io_type = ZIO_TYPE_WRITE;
835			} else if (strcasecmp(optarg, "free") == 0) {
836				io_type = ZIO_TYPE_FREE;
837			} else if (strcasecmp(optarg, "claim") == 0) {
838				io_type = ZIO_TYPE_CLAIM;
839			} else if (strcasecmp(optarg, "all") == 0) {
840				io_type = ZIO_TYPES;
841			} else {
842				(void) fprintf(stderr, "invalid I/O type "
843				    "'%s': must be 'read', 'write', 'free', "
844				    "'claim' or 'all'\n", optarg);
845				usage();
846				return (1);
847			}
848			break;
849		case 't':
850			if ((type = name_to_type(optarg)) == TYPE_INVAL &&
851			    !MOS_TYPE(type)) {
852				(void) fprintf(stderr, "invalid type '%s'\n",
853				    optarg);
854				usage();
855				return (1);
856			}
857			break;
858		case 'u':
859			flags |= ZINJECT_UNLOAD_SPA;
860			break;
861		case 'L':
862			if ((label = name_to_type(optarg)) == TYPE_INVAL &&
863			    !LABEL_TYPE(type)) {
864				(void) fprintf(stderr, "invalid label type "
865				    "'%s'\n", optarg);
866				usage();
867				return (1);
868			}
869			break;
870		case ':':
871			(void) fprintf(stderr, "option -%c requires an "
872			    "operand\n", optopt);
873			usage();
874			return (1);
875		case '?':
876			(void) fprintf(stderr, "invalid option '%c'\n",
877			    optopt);
878			usage();
879			return (2);
880		}
881	}
882
883	argc -= optind;
884	argv += optind;
885
886	if (record.zi_duration != 0)
887		record.zi_cmd = ZINJECT_IGNORED_WRITES;
888
889	if (cancel != NULL) {
890		/*
891		 * '-c' is invalid with any other options.
892		 */
893		if (raw != NULL || range != NULL || type != TYPE_INVAL ||
894		    level != 0 || record.zi_cmd != ZINJECT_UNINITIALIZED) {
895			(void) fprintf(stderr, "cancel (-c) incompatible with "
896			    "any other options\n");
897			usage();
898			return (2);
899		}
900		if (argc != 0) {
901			(void) fprintf(stderr, "extraneous argument to '-c'\n");
902			usage();
903			return (2);
904		}
905
906		if (strcmp(cancel, "all") == 0) {
907			return (cancel_all_handlers());
908		} else {
909			int id = (int)strtol(cancel, &end, 10);
910			if (*end != '\0') {
911				(void) fprintf(stderr, "invalid handle id '%s':"
912				    " must be an integer or 'all'\n", cancel);
913				usage();
914				return (1);
915			}
916			return (cancel_handler(id));
917		}
918	}
919
920	if (device != NULL) {
921		/*
922		 * Device (-d) injection uses a completely different mechanism
923		 * for doing injection, so handle it separately here.
924		 */
925		if (raw != NULL || range != NULL || type != TYPE_INVAL ||
926		    level != 0 || record.zi_cmd != ZINJECT_UNINITIALIZED) {
927			(void) fprintf(stderr, "device (-d) incompatible with "
928			    "data error injection\n");
929			usage();
930			return (2);
931		}
932
933		if (argc != 1) {
934			(void) fprintf(stderr, "device (-d) injection requires "
935			    "a single pool name\n");
936			usage();
937			return (2);
938		}
939
940		(void) strcpy(pool, argv[0]);
941		dataset[0] = '\0';
942
943		if (error == ECKSUM) {
944			(void) fprintf(stderr, "device error type must be "
945			    "'io' or 'nxio'\n");
946			return (1);
947		}
948
949		record.zi_iotype = io_type;
950		if (translate_device(pool, device, label, &record) != 0)
951			return (1);
952		if (!error)
953			error = ENXIO;
954
955		if (action != VDEV_STATE_UNKNOWN)
956			return (perform_action(pool, &record, action));
957
958	} else if (raw != NULL) {
959		if (range != NULL || type != TYPE_INVAL || level != 0 ||
960		    record.zi_cmd != ZINJECT_UNINITIALIZED) {
961			(void) fprintf(stderr, "raw (-b) format with "
962			    "any other options\n");
963			usage();
964			return (2);
965		}
966
967		if (argc != 1) {
968			(void) fprintf(stderr, "raw (-b) format expects a "
969			    "single pool name\n");
970			usage();
971			return (2);
972		}
973
974		(void) strcpy(pool, argv[0]);
975		dataset[0] = '\0';
976
977		if (error == ENXIO) {
978			(void) fprintf(stderr, "data error type must be "
979			    "'checksum' or 'io'\n");
980			return (1);
981		}
982
983		record.zi_cmd = ZINJECT_DATA_FAULT;
984		if (translate_raw(raw, &record) != 0)
985			return (1);
986		if (!error)
987			error = EIO;
988	} else if (record.zi_cmd == ZINJECT_PANIC) {
989		if (raw != NULL || range != NULL || type != TYPE_INVAL ||
990		    level != 0 || device != NULL) {
991			(void) fprintf(stderr, "panic (-p) incompatible with "
992			    "other options\n");
993			usage();
994			return (2);
995		}
996
997		if (argc < 1 || argc > 2) {
998			(void) fprintf(stderr, "panic (-p) injection requires "
999			    "a single pool name and an optional id\n");
1000			usage();
1001			return (2);
1002		}
1003
1004		(void) strcpy(pool, argv[0]);
1005		if (argv[1] != NULL)
1006			record.zi_type = atoi(argv[1]);
1007		dataset[0] = '\0';
1008	} else if (record.zi_cmd == ZINJECT_IGNORED_WRITES) {
1009		if (nowrites == 0) {
1010			(void) fprintf(stderr, "-s or -g meaningless "
1011			    "without -I (ignore writes)\n");
1012			usage();
1013			return (2);
1014		} else if (dur_secs && dur_txg) {
1015			(void) fprintf(stderr, "choose a duration either "
1016			    "in seconds (-s) or a number of txgs (-g) "
1017			    "but not both\n");
1018			usage();
1019			return (2);
1020		} else if (argc != 1) {
1021			(void) fprintf(stderr, "ignore writes (-I) "
1022			    "injection requires a single pool name\n");
1023			usage();
1024			return (2);
1025		}
1026
1027		(void) strcpy(pool, argv[0]);
1028		dataset[0] = '\0';
1029	} else if (type == TYPE_INVAL) {
1030		if (flags == 0) {
1031			(void) fprintf(stderr, "at least one of '-b', '-d', "
1032			    "'-t', '-a', '-p', '-I' or '-u' "
1033			    "must be specified\n");
1034			usage();
1035			return (2);
1036		}
1037
1038		if (argc == 1 && (flags & ZINJECT_UNLOAD_SPA)) {
1039			(void) strcpy(pool, argv[0]);
1040			dataset[0] = '\0';
1041		} else if (argc != 0) {
1042			(void) fprintf(stderr, "extraneous argument for "
1043			    "'-f'\n");
1044			usage();
1045			return (2);
1046		}
1047
1048		flags |= ZINJECT_NULL;
1049	} else {
1050		if (argc != 1) {
1051			(void) fprintf(stderr, "missing object\n");
1052			usage();
1053			return (2);
1054		}
1055
1056		if (error == ENXIO) {
1057			(void) fprintf(stderr, "data error type must be "
1058			    "'checksum' or 'io'\n");
1059			return (1);
1060		}
1061
1062		record.zi_cmd = ZINJECT_DATA_FAULT;
1063		if (translate_record(type, argv[0], range, level, &record, pool,
1064		    dataset) != 0)
1065			return (1);
1066		if (!error)
1067			error = EIO;
1068	}
1069
1070	/*
1071	 * If this is pool-wide metadata, unmount everything.  The ioctl() will
1072	 * unload the pool, so that we trigger spa-wide reopen of metadata next
1073	 * time we access the pool.
1074	 */
1075	if (dataset[0] != '\0' && domount) {
1076		if ((zhp = zfs_open(g_zfs, dataset, ZFS_TYPE_DATASET)) == NULL)
1077			return (1);
1078
1079		if (zfs_unmount(zhp, NULL, 0) != 0)
1080			return (1);
1081	}
1082
1083	record.zi_error = error;
1084
1085	ret = register_handler(pool, flags, &record, quiet);
1086
1087	if (dataset[0] != '\0' && domount)
1088		ret = (zfs_mount(zhp, NULL, 0) != 0);
1089
1090	libzfs_fini(g_zfs);
1091
1092	return (ret);
1093}
1094