]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
34dc7c2f BB |
23 | */ |
24 | ||
34dc7c2f BB |
25 | /* |
26 | * ZFS Fault Injector | |
27 | * | |
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. | |
31 | * | |
32 | * | |
33 | * DEVICE FAULTS | |
34 | * | |
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, | |
b128c09f | 38 | * that can be controlled through the '-e' option. The default is ENXIO. For |
34dc7c2f BB |
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. | |
b128c09f | 43 | * For label faults, the -L option must be specified. This allows faults |
428870ff BB |
44 | * to be injected into either the nvlist, uberblock, pad1, or pad2 region |
45 | * of all the labels for the specified device. | |
34dc7c2f BB |
46 | * |
47 | * This form of the command looks like: | |
48 | * | |
428870ff | 49 | * zinject -d device [-e errno] [-L <uber | nvlist | pad1 | pad2>] pool |
34dc7c2f BB |
50 | * |
51 | * | |
52 | * DATA FAULTS | |
53 | * | |
54 | * We begin with a tuple of the form: | |
55 | * | |
56 | * <type,level,range,object> | |
57 | * | |
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: | |
61 | * | |
62 | * data User data for a file | |
63 | * dnode Dnode for a file or directory | |
64 | * | |
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. | |
68 | * | |
69 | * mos Any data in the MOS | |
70 | * mosdir object directory | |
71 | * config pool configuration | |
428870ff | 72 | * bpobj blkptr list |
34dc7c2f BB |
73 | * spacemap spacemap |
74 | * metaslab metaslab | |
75 | * errlog persistent error log | |
76 | * | |
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. | |
80 | * | |
81 | * range A numerical range [start,end) within the object. Defaults to | |
82 | * the full size of the file. | |
83 | * | |
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. | |
87 | * | |
88 | * This is translated, via libzpool, into the following internal representation: | |
89 | * | |
90 | * <type,objset,object,level,range> | |
91 | * | |
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. | |
96 | * | |
97 | * | |
98 | * The command itself takes one of the forms: | |
99 | * | |
100 | * zinject | |
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 | |
106 | * | |
107 | * With no arguments, the command prints all currently registered injection | |
108 | * handlers, with their numeric identifiers. | |
109 | * | |
110 | * The '-c' option will clear the given handler, or all handlers if 'all' is | |
111 | * specified. | |
112 | * | |
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 | |
116 | * situation. | |
117 | * | |
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 | |
123 | * cache. | |
124 | * | |
125 | * The '-f' flag controls the frequency of errors injected, expressed as a | |
126 | * integer percentage between 1 and 100. The default is 100. | |
127 | * | |
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 | |
131 | * handler. | |
132 | * | |
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. | |
136 | */ | |
137 | ||
138 | #include <errno.h> | |
139 | #include <fcntl.h> | |
140 | #include <stdio.h> | |
141 | #include <stdlib.h> | |
142 | #include <strings.h> | |
143 | #include <unistd.h> | |
144 | ||
145 | #include <sys/fs/zfs.h> | |
146 | #include <sys/mount.h> | |
147 | ||
148 | #include <libzfs.h> | |
149 | ||
150 | #undef verify /* both libzfs.h and zfs_context.h want to define this */ | |
151 | ||
152 | #include "zinject.h" | |
153 | ||
154 | libzfs_handle_t *g_zfs; | |
155 | int zfs_fd; | |
156 | ||
157 | #define ECKSUM EBADE | |
158 | ||
159 | static const char *errtable[TYPE_INVAL] = { | |
160 | "data", | |
161 | "dnode", | |
162 | "mos", | |
163 | "mosdir", | |
164 | "metaslab", | |
165 | "config", | |
428870ff | 166 | "bpobj", |
34dc7c2f | 167 | "spacemap", |
b128c09f BB |
168 | "errlog", |
169 | "uber", | |
428870ff BB |
170 | "nvlist", |
171 | "pad1", | |
172 | "pad2" | |
34dc7c2f BB |
173 | }; |
174 | ||
175 | static err_type_t | |
176 | name_to_type(const char *arg) | |
177 | { | |
178 | int i; | |
179 | for (i = 0; i < TYPE_INVAL; i++) | |
180 | if (strcmp(errtable[i], arg) == 0) | |
181 | return (i); | |
182 | ||
183 | return (TYPE_INVAL); | |
184 | } | |
185 | ||
186 | static const char * | |
187 | type_to_name(uint64_t type) | |
188 | { | |
189 | switch (type) { | |
190 | case DMU_OT_OBJECT_DIRECTORY: | |
191 | return ("mosdir"); | |
192 | case DMU_OT_OBJECT_ARRAY: | |
193 | return ("metaslab"); | |
194 | case DMU_OT_PACKED_NVLIST: | |
195 | return ("config"); | |
428870ff BB |
196 | case DMU_OT_BPOBJ: |
197 | return ("bpobj"); | |
34dc7c2f BB |
198 | case DMU_OT_SPACE_MAP: |
199 | return ("spacemap"); | |
200 | case DMU_OT_ERROR_LOG: | |
201 | return ("errlog"); | |
202 | default: | |
203 | return ("-"); | |
204 | } | |
205 | } | |
206 | ||
207 | ||
208 | /* | |
209 | * Print usage message. | |
210 | */ | |
211 | void | |
212 | usage(void) | |
213 | { | |
214 | (void) printf( | |
215 | "usage:\n" | |
216 | "\n" | |
217 | "\tzinject\n" | |
218 | "\n" | |
219 | "\t\tList all active injection records.\n" | |
220 | "\n" | |
221 | "\tzinject -c <id|all>\n" | |
222 | "\n" | |
223 | "\t\tClear the particular record (if given a numeric ID), or\n" | |
224 | "\t\tall records if 'all' is specificed.\n" | |
225 | "\n" | |
428870ff BB |
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" | |
230 | "\n" | |
231 | "\tzinject -d device [-e errno] [-L <nvlist|uber|pad1|pad2>] [-F]\n" | |
232 | "\t [-T <read|write|free|claim|all> pool\n" | |
b128c09f | 233 | "\t\tInject a fault into a particular device or the device's\n" |
428870ff BB |
234 | "\t\tlabel. Label injection can either be 'nvlist', 'uber',\n " |
235 | "\t\t'pad1', or 'pad2'.\n" | |
b128c09f | 236 | "\t\t'errno' can either be 'nxio' (the default) or 'io'.\n" |
34dc7c2f | 237 | "\n" |
428870ff BB |
238 | "\tzinject -d device -A <degrade|fault> pool\n" |
239 | "\t\tPerform a specific action on a particular device\n" | |
240 | "\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" | |
247 | "\n" | |
34dc7c2f BB |
248 | "\tzinject -b objset:object:level:blkid pool\n" |
249 | "\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" | |
253 | "\n" | |
254 | "\tzinject [-q] <-t type> [-e errno] [-l level] [-r range]\n" | |
255 | "\t [-a] [-m] [-u] [-f freq] <object>\n" | |
256 | "\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" | |
260 | "\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 " | |
265 | "0.\n" | |
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" | |
276 | "\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" | |
280 | "\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" | |
286 | "\n" | |
287 | "\t-t <mos>\tInject errors into the MOS for objects of the given\n" | |
428870ff | 288 | "\t\t\ttype. Valid types are: mos, mosdir, config, bpobj,\n" |
34dc7c2f BB |
289 | "\t\t\tspacemap, metaslab, errlog. The only valid <object> is\n" |
290 | "\t\t\tthe poolname.\n"); | |
291 | } | |
292 | ||
293 | static int | |
294 | iter_handlers(int (*func)(int, const char *, zinject_record_t *, void *), | |
295 | void *data) | |
296 | { | |
297 | zfs_cmd_t zc; | |
298 | int ret; | |
299 | ||
300 | zc.zc_guid = 0; | |
301 | ||
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) | |
305 | return (ret); | |
306 | ||
428870ff BB |
307 | if (errno != ENOENT) { |
308 | (void) fprintf(stderr, "Unable to list handlers: %s\n", | |
309 | strerror(errno)); | |
310 | return (-1); | |
311 | } | |
312 | ||
34dc7c2f BB |
313 | return (0); |
314 | } | |
315 | ||
316 | static int | |
317 | print_data_handler(int id, const char *pool, zinject_record_t *record, | |
318 | void *data) | |
319 | { | |
320 | int *count = data; | |
321 | ||
428870ff | 322 | if (record->zi_guid != 0 || record->zi_func[0] != '\0') |
34dc7c2f BB |
323 | return (0); |
324 | ||
325 | if (*count == 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"); | |
330 | } | |
331 | ||
332 | *count += 1; | |
333 | ||
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); | |
337 | ||
338 | if (record->zi_start == 0 && | |
339 | record->zi_end == -1ULL) | |
340 | (void) printf("all\n"); | |
341 | else | |
342 | (void) printf("[%llu, %llu]\n", (u_longlong_t)record->zi_start, | |
343 | (u_longlong_t)record->zi_end); | |
344 | ||
345 | return (0); | |
346 | } | |
347 | ||
348 | static int | |
349 | print_device_handler(int id, const char *pool, zinject_record_t *record, | |
350 | void *data) | |
351 | { | |
352 | int *count = data; | |
353 | ||
428870ff | 354 | if (record->zi_guid == 0 || record->zi_func[0] != '\0') |
34dc7c2f BB |
355 | return (0); |
356 | ||
357 | if (*count == 0) { | |
358 | (void) printf("%3s %-15s %s\n", "ID", "POOL", "GUID"); | |
359 | (void) printf("--- --------------- ----------------\n"); | |
360 | } | |
361 | ||
362 | *count += 1; | |
363 | ||
364 | (void) printf("%3d %-15s %llx\n", id, pool, | |
365 | (u_longlong_t)record->zi_guid); | |
366 | ||
367 | return (0); | |
368 | } | |
369 | ||
428870ff BB |
370 | static int |
371 | print_panic_handler(int id, const char *pool, zinject_record_t *record, | |
372 | void *data) | |
373 | { | |
374 | int *count = data; | |
375 | ||
376 | if (record->zi_func[0] == '\0') | |
377 | return (0); | |
378 | ||
379 | if (*count == 0) { | |
380 | (void) printf("%3s %-15s %s\n", "ID", "POOL", "FUNCTION"); | |
381 | (void) printf("--- --------------- ----------------\n"); | |
382 | } | |
383 | ||
384 | *count += 1; | |
385 | ||
386 | (void) printf("%3d %-15s %s\n", id, pool, record->zi_func); | |
387 | ||
388 | return (0); | |
389 | } | |
390 | ||
34dc7c2f BB |
391 | /* |
392 | * Print all registered error handlers. Returns the number of handlers | |
393 | * registered. | |
394 | */ | |
395 | static int | |
396 | print_all_handlers(void) | |
397 | { | |
398 | int count = 0; | |
399 | ||
400 | (void) iter_handlers(print_device_handler, &count); | |
401 | (void) printf("\n"); | |
402 | count = 0; | |
403 | (void) iter_handlers(print_data_handler, &count); | |
428870ff BB |
404 | (void) printf("\n"); |
405 | count = 0; | |
406 | (void) iter_handlers(print_panic_handler, &count); | |
34dc7c2f BB |
407 | |
408 | return (count); | |
409 | } | |
410 | ||
411 | /* ARGSUSED */ | |
412 | static int | |
413 | cancel_one_handler(int id, const char *pool, zinject_record_t *record, | |
414 | void *data) | |
415 | { | |
416 | zfs_cmd_t zc; | |
417 | ||
418 | zc.zc_guid = (uint64_t)id; | |
419 | ||
420 | if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) { | |
421 | (void) fprintf(stderr, "failed to remove handler %d: %s\n", | |
422 | id, strerror(errno)); | |
423 | return (1); | |
424 | } | |
425 | ||
426 | return (0); | |
427 | } | |
428 | ||
429 | /* | |
430 | * Remove all fault injection handlers. | |
431 | */ | |
432 | static int | |
433 | cancel_all_handlers(void) | |
434 | { | |
435 | int ret = iter_handlers(cancel_one_handler, NULL); | |
436 | ||
428870ff BB |
437 | if (ret == 0) |
438 | (void) printf("removed all registered handlers\n"); | |
34dc7c2f BB |
439 | |
440 | return (ret); | |
441 | } | |
442 | ||
443 | /* | |
444 | * Remove a specific fault injection handler. | |
445 | */ | |
446 | static int | |
447 | cancel_handler(int id) | |
448 | { | |
449 | zfs_cmd_t zc; | |
450 | ||
451 | zc.zc_guid = (uint64_t)id; | |
452 | ||
453 | if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) { | |
454 | (void) fprintf(stderr, "failed to remove handler %d: %s\n", | |
455 | id, strerror(errno)); | |
456 | return (1); | |
457 | } | |
458 | ||
459 | (void) printf("removed handler %d\n", id); | |
460 | ||
461 | return (0); | |
462 | } | |
463 | ||
464 | /* | |
465 | * Register a new fault injection handler. | |
466 | */ | |
467 | static int | |
468 | register_handler(const char *pool, int flags, zinject_record_t *record, | |
469 | int quiet) | |
470 | { | |
471 | zfs_cmd_t zc; | |
472 | ||
473 | (void) strcpy(zc.zc_name, pool); | |
474 | zc.zc_inject_record = *record; | |
475 | zc.zc_guid = flags; | |
476 | ||
477 | if (ioctl(zfs_fd, ZFS_IOC_INJECT_FAULT, &zc) != 0) { | |
478 | (void) fprintf(stderr, "failed to add handler: %s\n", | |
479 | strerror(errno)); | |
480 | return (1); | |
481 | } | |
482 | ||
483 | if (flags & ZINJECT_NULL) | |
484 | return (0); | |
485 | ||
486 | if (quiet) { | |
487 | (void) printf("%llu\n", (u_longlong_t)zc.zc_guid); | |
488 | } else { | |
489 | (void) printf("Added handler %llu with the following " | |
490 | "properties:\n", (u_longlong_t)zc.zc_guid); | |
491 | (void) printf(" pool: %s\n", pool); | |
492 | if (record->zi_guid) { | |
493 | (void) printf(" vdev: %llx\n", | |
494 | (u_longlong_t)record->zi_guid); | |
428870ff BB |
495 | } else if (record->zi_func[0] != '\0') { |
496 | (void) printf(" panic function: %s\n", | |
497 | record->zi_func); | |
498 | } else if (record->zi_duration > 0) { | |
499 | (void) printf(" time: %lld seconds\n", | |
500 | (u_longlong_t)record->zi_duration); | |
501 | } else if (record->zi_duration < 0) { | |
502 | (void) printf(" txgs: %lld \n", | |
503 | (u_longlong_t)-record->zi_duration); | |
34dc7c2f BB |
504 | } else { |
505 | (void) printf("objset: %llu\n", | |
506 | (u_longlong_t)record->zi_objset); | |
507 | (void) printf("object: %llu\n", | |
508 | (u_longlong_t)record->zi_object); | |
509 | (void) printf(" type: %llu\n", | |
510 | (u_longlong_t)record->zi_type); | |
511 | (void) printf(" level: %d\n", record->zi_level); | |
512 | if (record->zi_start == 0 && | |
513 | record->zi_end == -1ULL) | |
514 | (void) printf(" range: all\n"); | |
515 | else | |
516 | (void) printf(" range: [%llu, %llu)\n", | |
517 | (u_longlong_t)record->zi_start, | |
518 | (u_longlong_t)record->zi_end); | |
519 | } | |
520 | } | |
521 | ||
522 | return (0); | |
523 | } | |
524 | ||
428870ff BB |
525 | int |
526 | perform_action(const char *pool, zinject_record_t *record, int cmd) | |
527 | { | |
528 | zfs_cmd_t zc; | |
529 | ||
530 | ASSERT(cmd == VDEV_STATE_DEGRADED || cmd == VDEV_STATE_FAULTED); | |
531 | (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name)); | |
532 | zc.zc_guid = record->zi_guid; | |
533 | zc.zc_cookie = cmd; | |
534 | ||
535 | if (ioctl(zfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0) | |
536 | return (0); | |
537 | ||
538 | return (1); | |
539 | } | |
540 | ||
34dc7c2f BB |
541 | int |
542 | main(int argc, char **argv) | |
543 | { | |
544 | int c; | |
545 | char *range = NULL; | |
546 | char *cancel = NULL; | |
547 | char *end; | |
548 | char *raw = NULL; | |
549 | char *device = NULL; | |
550 | int level = 0; | |
551 | int quiet = 0; | |
552 | int error = 0; | |
553 | int domount = 0; | |
428870ff BB |
554 | int io_type = ZIO_TYPES; |
555 | int action = VDEV_STATE_UNKNOWN; | |
34dc7c2f | 556 | err_type_t type = TYPE_INVAL; |
b128c09f | 557 | err_type_t label = TYPE_INVAL; |
34dc7c2f BB |
558 | zinject_record_t record = { 0 }; |
559 | char pool[MAXNAMELEN]; | |
560 | char dataset[MAXNAMELEN]; | |
561 | zfs_handle_t *zhp; | |
428870ff BB |
562 | int nowrites = 0; |
563 | int dur_txg = 0; | |
564 | int dur_secs = 0; | |
34dc7c2f BB |
565 | int ret; |
566 | int flags = 0; | |
567 | ||
568 | if ((g_zfs = libzfs_init()) == NULL) { | |
569 | (void) fprintf(stderr, "internal error: failed to " | |
570 | "initialize ZFS library\n"); | |
571 | return (1); | |
572 | } | |
573 | ||
574 | libzfs_print_on_error(g_zfs, B_TRUE); | |
575 | ||
576 | if ((zfs_fd = open(ZFS_DEV, O_RDWR)) < 0) { | |
577 | (void) fprintf(stderr, "failed to open ZFS device\n"); | |
578 | return (1); | |
579 | } | |
580 | ||
581 | if (argc == 1) { | |
582 | /* | |
583 | * No arguments. Print the available handlers. If there are no | |
584 | * available handlers, direct the user to '-h' for help | |
585 | * information. | |
586 | */ | |
587 | if (print_all_handlers() == 0) { | |
588 | (void) printf("No handlers registered.\n"); | |
589 | (void) printf("Run 'zinject -h' for usage " | |
590 | "information.\n"); | |
591 | } | |
592 | ||
593 | return (0); | |
594 | } | |
595 | ||
428870ff BB |
596 | while ((c = getopt(argc, argv, |
597 | ":aA:b:d:f:Fg:qhIc:t:T:l:mr:s:e:uL:p:")) != -1) { | |
34dc7c2f BB |
598 | switch (c) { |
599 | case 'a': | |
600 | flags |= ZINJECT_FLUSH_ARC; | |
601 | break; | |
428870ff BB |
602 | case 'A': |
603 | if (strcasecmp(optarg, "degrade") == 0) { | |
604 | action = VDEV_STATE_DEGRADED; | |
605 | } else if (strcasecmp(optarg, "fault") == 0) { | |
606 | action = VDEV_STATE_FAULTED; | |
607 | } else { | |
608 | (void) fprintf(stderr, "invalid action '%s': " | |
609 | "must be 'degrade' or 'fault'\n", optarg); | |
610 | usage(); | |
611 | return (1); | |
612 | } | |
613 | break; | |
34dc7c2f BB |
614 | case 'b': |
615 | raw = optarg; | |
616 | break; | |
617 | case 'c': | |
618 | cancel = optarg; | |
619 | break; | |
620 | case 'd': | |
621 | device = optarg; | |
622 | break; | |
623 | case 'e': | |
624 | if (strcasecmp(optarg, "io") == 0) { | |
625 | error = EIO; | |
626 | } else if (strcasecmp(optarg, "checksum") == 0) { | |
627 | error = ECKSUM; | |
628 | } else if (strcasecmp(optarg, "nxio") == 0) { | |
629 | error = ENXIO; | |
630 | } else { | |
631 | (void) fprintf(stderr, "invalid error type " | |
632 | "'%s': must be 'io', 'checksum' or " | |
633 | "'nxio'\n", optarg); | |
634 | usage(); | |
635 | return (1); | |
636 | } | |
637 | break; | |
638 | case 'f': | |
639 | record.zi_freq = atoi(optarg); | |
640 | if (record.zi_freq < 1 || record.zi_freq > 100) { | |
641 | (void) fprintf(stderr, "frequency range must " | |
642 | "be in the range (0, 100]\n"); | |
643 | return (1); | |
644 | } | |
645 | break; | |
9babb374 BB |
646 | case 'F': |
647 | record.zi_failfast = B_TRUE; | |
648 | break; | |
428870ff BB |
649 | case 'g': |
650 | dur_txg = 1; | |
651 | record.zi_duration = (int)strtol(optarg, &end, 10); | |
652 | if (record.zi_duration <= 0 || *end != '\0') { | |
653 | (void) fprintf(stderr, "invalid duration '%s': " | |
654 | "must be a positive integer\n", optarg); | |
655 | usage(); | |
656 | return (1); | |
657 | } | |
658 | /* store duration of txgs as its negative */ | |
659 | record.zi_duration *= -1; | |
660 | break; | |
34dc7c2f BB |
661 | case 'h': |
662 | usage(); | |
663 | return (0); | |
428870ff BB |
664 | case 'I': |
665 | /* default duration, if one hasn't yet been defined */ | |
666 | nowrites = 1; | |
667 | if (dur_secs == 0 && dur_txg == 0) | |
668 | record.zi_duration = 30; | |
669 | break; | |
34dc7c2f BB |
670 | case 'l': |
671 | level = (int)strtol(optarg, &end, 10); | |
672 | if (*end != '\0') { | |
673 | (void) fprintf(stderr, "invalid level '%s': " | |
674 | "must be an integer\n", optarg); | |
675 | usage(); | |
676 | return (1); | |
677 | } | |
678 | break; | |
679 | case 'm': | |
680 | domount = 1; | |
681 | break; | |
428870ff BB |
682 | case 'p': |
683 | (void) strlcpy(record.zi_func, optarg, | |
684 | sizeof (record.zi_func)); | |
685 | break; | |
34dc7c2f BB |
686 | case 'q': |
687 | quiet = 1; | |
688 | break; | |
689 | case 'r': | |
690 | range = optarg; | |
691 | break; | |
428870ff BB |
692 | case 's': |
693 | dur_secs = 1; | |
694 | record.zi_duration = (int)strtol(optarg, &end, 10); | |
695 | if (record.zi_duration <= 0 || *end != '\0') { | |
696 | (void) fprintf(stderr, "invalid duration '%s': " | |
697 | "must be a positive integer\n", optarg); | |
698 | usage(); | |
699 | return (1); | |
700 | } | |
701 | break; | |
702 | case 'T': | |
703 | if (strcasecmp(optarg, "read") == 0) { | |
704 | io_type = ZIO_TYPE_READ; | |
705 | } else if (strcasecmp(optarg, "write") == 0) { | |
706 | io_type = ZIO_TYPE_WRITE; | |
707 | } else if (strcasecmp(optarg, "free") == 0) { | |
708 | io_type = ZIO_TYPE_FREE; | |
709 | } else if (strcasecmp(optarg, "claim") == 0) { | |
710 | io_type = ZIO_TYPE_CLAIM; | |
711 | } else if (strcasecmp(optarg, "all") == 0) { | |
712 | io_type = ZIO_TYPES; | |
713 | } else { | |
714 | (void) fprintf(stderr, "invalid I/O type " | |
715 | "'%s': must be 'read', 'write', 'free', " | |
716 | "'claim' or 'all'\n", optarg); | |
717 | usage(); | |
718 | return (1); | |
719 | } | |
720 | break; | |
34dc7c2f | 721 | case 't': |
b128c09f BB |
722 | if ((type = name_to_type(optarg)) == TYPE_INVAL && |
723 | !MOS_TYPE(type)) { | |
34dc7c2f BB |
724 | (void) fprintf(stderr, "invalid type '%s'\n", |
725 | optarg); | |
726 | usage(); | |
727 | return (1); | |
728 | } | |
729 | break; | |
730 | case 'u': | |
731 | flags |= ZINJECT_UNLOAD_SPA; | |
732 | break; | |
b128c09f BB |
733 | case 'L': |
734 | if ((label = name_to_type(optarg)) == TYPE_INVAL && | |
735 | !LABEL_TYPE(type)) { | |
736 | (void) fprintf(stderr, "invalid label type " | |
737 | "'%s'\n", optarg); | |
738 | usage(); | |
739 | return (1); | |
740 | } | |
741 | break; | |
34dc7c2f BB |
742 | case ':': |
743 | (void) fprintf(stderr, "option -%c requires an " | |
744 | "operand\n", optopt); | |
745 | usage(); | |
746 | return (1); | |
747 | case '?': | |
748 | (void) fprintf(stderr, "invalid option '%c'\n", | |
749 | optopt); | |
750 | usage(); | |
751 | return (2); | |
752 | } | |
753 | } | |
754 | ||
755 | argc -= optind; | |
756 | argv += optind; | |
757 | ||
758 | if (cancel != NULL) { | |
759 | /* | |
760 | * '-c' is invalid with any other options. | |
761 | */ | |
762 | if (raw != NULL || range != NULL || type != TYPE_INVAL || | |
428870ff BB |
763 | level != 0 || record.zi_func[0] != '\0' || |
764 | record.zi_duration != 0) { | |
34dc7c2f BB |
765 | (void) fprintf(stderr, "cancel (-c) incompatible with " |
766 | "any other options\n"); | |
767 | usage(); | |
768 | return (2); | |
769 | } | |
770 | if (argc != 0) { | |
771 | (void) fprintf(stderr, "extraneous argument to '-c'\n"); | |
772 | usage(); | |
773 | return (2); | |
774 | } | |
775 | ||
776 | if (strcmp(cancel, "all") == 0) { | |
777 | return (cancel_all_handlers()); | |
778 | } else { | |
779 | int id = (int)strtol(cancel, &end, 10); | |
780 | if (*end != '\0') { | |
781 | (void) fprintf(stderr, "invalid handle id '%s':" | |
782 | " must be an integer or 'all'\n", cancel); | |
783 | usage(); | |
784 | return (1); | |
785 | } | |
786 | return (cancel_handler(id)); | |
787 | } | |
788 | } | |
789 | ||
790 | if (device != NULL) { | |
791 | /* | |
792 | * Device (-d) injection uses a completely different mechanism | |
793 | * for doing injection, so handle it separately here. | |
794 | */ | |
795 | if (raw != NULL || range != NULL || type != TYPE_INVAL || | |
428870ff BB |
796 | level != 0 || record.zi_func[0] != '\0' || |
797 | record.zi_duration != 0) { | |
34dc7c2f BB |
798 | (void) fprintf(stderr, "device (-d) incompatible with " |
799 | "data error injection\n"); | |
800 | usage(); | |
801 | return (2); | |
802 | } | |
803 | ||
804 | if (argc != 1) { | |
805 | (void) fprintf(stderr, "device (-d) injection requires " | |
806 | "a single pool name\n"); | |
807 | usage(); | |
808 | return (2); | |
809 | } | |
810 | ||
811 | (void) strcpy(pool, argv[0]); | |
812 | dataset[0] = '\0'; | |
813 | ||
814 | if (error == ECKSUM) { | |
815 | (void) fprintf(stderr, "device error type must be " | |
816 | "'io' or 'nxio'\n"); | |
817 | return (1); | |
818 | } | |
819 | ||
428870ff | 820 | record.zi_iotype = io_type; |
b128c09f | 821 | if (translate_device(pool, device, label, &record) != 0) |
34dc7c2f BB |
822 | return (1); |
823 | if (!error) | |
824 | error = ENXIO; | |
428870ff BB |
825 | |
826 | if (action != VDEV_STATE_UNKNOWN) | |
827 | return (perform_action(pool, &record, action)); | |
828 | ||
34dc7c2f | 829 | } else if (raw != NULL) { |
428870ff BB |
830 | if (range != NULL || type != TYPE_INVAL || level != 0 || |
831 | record.zi_func[0] != '\0' || record.zi_duration != 0) { | |
34dc7c2f BB |
832 | (void) fprintf(stderr, "raw (-b) format with " |
833 | "any other options\n"); | |
834 | usage(); | |
835 | return (2); | |
836 | } | |
837 | ||
838 | if (argc != 1) { | |
839 | (void) fprintf(stderr, "raw (-b) format expects a " | |
840 | "single pool name\n"); | |
841 | usage(); | |
842 | return (2); | |
843 | } | |
844 | ||
845 | (void) strcpy(pool, argv[0]); | |
846 | dataset[0] = '\0'; | |
847 | ||
848 | if (error == ENXIO) { | |
849 | (void) fprintf(stderr, "data error type must be " | |
850 | "'checksum' or 'io'\n"); | |
851 | return (1); | |
852 | } | |
853 | ||
854 | if (translate_raw(raw, &record) != 0) | |
855 | return (1); | |
856 | if (!error) | |
857 | error = EIO; | |
428870ff BB |
858 | } else if (record.zi_func[0] != '\0') { |
859 | if (raw != NULL || range != NULL || type != TYPE_INVAL || | |
860 | level != 0 || device != NULL || record.zi_duration != 0) { | |
861 | (void) fprintf(stderr, "panic (-p) incompatible with " | |
862 | "other options\n"); | |
863 | usage(); | |
864 | return (2); | |
865 | } | |
866 | ||
867 | if (argc < 1 || argc > 2) { | |
868 | (void) fprintf(stderr, "panic (-p) injection requires " | |
869 | "a single pool name and an optional id\n"); | |
870 | usage(); | |
871 | return (2); | |
872 | } | |
873 | ||
874 | (void) strcpy(pool, argv[0]); | |
875 | if (argv[1] != NULL) | |
876 | record.zi_type = atoi(argv[1]); | |
877 | dataset[0] = '\0'; | |
878 | } else if (record.zi_duration != 0) { | |
879 | if (nowrites == 0) { | |
880 | (void) fprintf(stderr, "-s or -g meaningless " | |
881 | "without -I (ignore writes)\n"); | |
882 | usage(); | |
883 | return (2); | |
884 | } else if (dur_secs && dur_txg) { | |
885 | (void) fprintf(stderr, "choose a duration either " | |
886 | "in seconds (-s) or a number of txgs (-g) " | |
887 | "but not both\n"); | |
888 | usage(); | |
889 | return (2); | |
890 | } else if (argc != 1) { | |
891 | (void) fprintf(stderr, "ignore writes (-I) " | |
892 | "injection requires a single pool name\n"); | |
893 | usage(); | |
894 | return (2); | |
895 | } | |
896 | ||
897 | (void) strcpy(pool, argv[0]); | |
898 | dataset[0] = '\0'; | |
34dc7c2f BB |
899 | } else if (type == TYPE_INVAL) { |
900 | if (flags == 0) { | |
901 | (void) fprintf(stderr, "at least one of '-b', '-d', " | |
428870ff BB |
902 | "'-t', '-a', '-p', '-I' or '-u' " |
903 | "must be specified\n"); | |
34dc7c2f BB |
904 | usage(); |
905 | return (2); | |
906 | } | |
907 | ||
908 | if (argc == 1 && (flags & ZINJECT_UNLOAD_SPA)) { | |
909 | (void) strcpy(pool, argv[0]); | |
910 | dataset[0] = '\0'; | |
911 | } else if (argc != 0) { | |
912 | (void) fprintf(stderr, "extraneous argument for " | |
913 | "'-f'\n"); | |
914 | usage(); | |
915 | return (2); | |
916 | } | |
917 | ||
918 | flags |= ZINJECT_NULL; | |
919 | } else { | |
920 | if (argc != 1) { | |
921 | (void) fprintf(stderr, "missing object\n"); | |
922 | usage(); | |
923 | return (2); | |
924 | } | |
925 | ||
926 | if (error == ENXIO) { | |
927 | (void) fprintf(stderr, "data error type must be " | |
928 | "'checksum' or 'io'\n"); | |
929 | return (1); | |
930 | } | |
931 | ||
932 | if (translate_record(type, argv[0], range, level, &record, pool, | |
933 | dataset) != 0) | |
934 | return (1); | |
935 | if (!error) | |
936 | error = EIO; | |
937 | } | |
938 | ||
939 | /* | |
940 | * If this is pool-wide metadata, unmount everything. The ioctl() will | |
941 | * unload the pool, so that we trigger spa-wide reopen of metadata next | |
942 | * time we access the pool. | |
943 | */ | |
944 | if (dataset[0] != '\0' && domount) { | |
945 | if ((zhp = zfs_open(g_zfs, dataset, ZFS_TYPE_DATASET)) == NULL) | |
946 | return (1); | |
947 | ||
948 | if (zfs_unmount(zhp, NULL, 0) != 0) | |
949 | return (1); | |
950 | } | |
951 | ||
952 | record.zi_error = error; | |
953 | ||
954 | ret = register_handler(pool, flags, &record, quiet); | |
955 | ||
956 | if (dataset[0] != '\0' && domount) | |
957 | ret = (zfs_mount(zhp, NULL, 0) != 0); | |
958 | ||
959 | libzfs_fini(g_zfs); | |
960 | ||
961 | return (ret); | |
962 | } |