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) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
24 */
25
26 #include <libzfs.h>
27
28 #include <sys/zfs_context.h>
29
30 #include <errno.h>
31 #include <fcntl.h>
32 #include <stdarg.h>
33 #include <stddef.h>
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <strings.h>
37 #include <sys/file.h>
38 #include <sys/mntent.h>
39 #include <sys/mnttab.h>
40 #include <sys/param.h>
41 #include <sys/stat.h>
42
43 #include <sys/dmu.h>
44 #include <sys/dmu_objset.h>
45 #include <sys/dnode.h>
46 #include <sys/vdev_impl.h>
47
48 #include "zinject.h"
49
50 extern void kernel_init(int);
51 extern void kernel_fini(void);
52
53 static int debug;
54
55 static void
ziprintf(const char * fmt,...)56 ziprintf(const char *fmt, ...)
57 {
58 va_list ap;
59
60 if (!debug)
61 return;
62
63 va_start(ap, fmt);
64 (void) vprintf(fmt, ap);
65 va_end(ap);
66 }
67
68 static void
compress_slashes(const char * src,char * dest)69 compress_slashes(const char *src, char *dest)
70 {
71 while (*src != '\0') {
72 *dest = *src++;
73 while (*dest == '/' && *src == '/')
74 ++src;
75 ++dest;
76 }
77 *dest = '\0';
78 }
79
80 /*
81 * Given a full path to a file, translate into a dataset name and a relative
82 * path within the dataset. 'dataset' must be at least MAXNAMELEN characters,
83 * and 'relpath' must be at least MAXPATHLEN characters. We also pass a stat64
84 * buffer, which we need later to get the object ID.
85 */
86 static int
parse_pathname(const char * inpath,char * dataset,char * relpath,struct stat64 * statbuf)87 parse_pathname(const char *inpath, char *dataset, char *relpath,
88 struct stat64 *statbuf)
89 {
90 struct statfs sfs;
91 const char *rel;
92 char fullpath[MAXPATHLEN];
93
94 compress_slashes(inpath, fullpath);
95
96 if (fullpath[0] != '/') {
97 (void) fprintf(stderr, "invalid object '%s': must be full "
98 "path\n", fullpath);
99 usage();
100 return (-1);
101 }
102
103 if (strlen(fullpath) >= MAXPATHLEN) {
104 (void) fprintf(stderr, "invalid object; pathname too long\n");
105 return (-1);
106 }
107
108 if (stat64(fullpath, statbuf) != 0) {
109 (void) fprintf(stderr, "cannot open '%s': %s\n",
110 fullpath, strerror(errno));
111 return (-1);
112 }
113
114 if (statfs(fullpath, &sfs) == -1) {
115 (void) fprintf(stderr, "cannot find mountpoint for '%s': %s\n",
116 fullpath, strerror(errno));
117 return (-1);
118 }
119
120 if (strcmp(sfs.f_fstypename, MNTTYPE_ZFS) != 0) {
121 (void) fprintf(stderr, "invalid path '%s': not a ZFS "
122 "filesystem\n", fullpath);
123 return (-1);
124 }
125
126 if (strncmp(fullpath, sfs.f_mntonname, strlen(sfs.f_mntonname)) != 0) {
127 (void) fprintf(stderr, "invalid path '%s': mountpoint "
128 "doesn't match path\n", fullpath);
129 return (-1);
130 }
131
132 (void) strcpy(dataset, sfs.f_mntfromname);
133
134 rel = fullpath + strlen(sfs.f_mntonname);
135 if (rel[0] == '/')
136 rel++;
137 (void) strcpy(relpath, rel);
138
139 return (0);
140 }
141
142 /*
143 * Convert from a (dataset, path) pair into a (objset, object) pair. Note that
144 * we grab the object number from the inode number, since looking this up via
145 * libzpool is a real pain.
146 */
147 /* ARGSUSED */
148 static int
object_from_path(const char * dataset,const char * path,struct stat64 * statbuf,zinject_record_t * record)149 object_from_path(const char *dataset, const char *path, struct stat64 *statbuf,
150 zinject_record_t *record)
151 {
152 objset_t *os;
153 int err;
154
155 /*
156 * Before doing any libzpool operations, call sync() to ensure that the
157 * on-disk state is consistent with the in-core state.
158 */
159 sync();
160
161 err = dmu_objset_own(dataset, DMU_OST_ZFS, B_TRUE, FTAG, &os);
162 if (err != 0) {
163 (void) fprintf(stderr, "cannot open dataset '%s': %s\n",
164 dataset, strerror(err));
165 return (-1);
166 }
167
168 record->zi_objset = dmu_objset_id(os);
169 record->zi_object = statbuf->st_ino;
170
171 dmu_objset_disown(os, FTAG);
172
173 return (0);
174 }
175
176 /*
177 * Calculate the real range based on the type, level, and range given.
178 */
179 static int
calculate_range(const char * dataset,err_type_t type,int level,char * range,zinject_record_t * record)180 calculate_range(const char *dataset, err_type_t type, int level, char *range,
181 zinject_record_t *record)
182 {
183 objset_t *os = NULL;
184 dnode_t *dn = NULL;
185 int err;
186 int ret = -1;
187
188 /*
189 * Determine the numeric range from the string.
190 */
191 if (range == NULL) {
192 /*
193 * If range is unspecified, set the range to [0,-1], which
194 * indicates that the whole object should be treated as an
195 * error.
196 */
197 record->zi_start = 0;
198 record->zi_end = -1ULL;
199 } else {
200 char *end;
201
202 /* XXX add support for suffixes */
203 record->zi_start = strtoull(range, &end, 10);
204
205
206 if (*end == '\0')
207 record->zi_end = record->zi_start + 1;
208 else if (*end == ',')
209 record->zi_end = strtoull(end + 1, &end, 10);
210
211 if (*end != '\0') {
212 (void) fprintf(stderr, "invalid range '%s': must be "
213 "a numeric range of the form 'start[,end]'\n",
214 range);
215 goto out;
216 }
217 }
218
219 switch (type) {
220 case TYPE_DATA:
221 break;
222
223 case TYPE_DNODE:
224 /*
225 * If this is a request to inject faults into the dnode, then we
226 * must translate the current (objset,object) pair into an
227 * offset within the metadnode for the objset. Specifying any
228 * kind of range with type 'dnode' is illegal.
229 */
230 if (range != NULL) {
231 (void) fprintf(stderr, "range cannot be specified when "
232 "type is 'dnode'\n");
233 goto out;
234 }
235
236 record->zi_start = record->zi_object * sizeof (dnode_phys_t);
237 record->zi_end = record->zi_start + sizeof (dnode_phys_t);
238 record->zi_object = 0;
239 break;
240 }
241
242 /*
243 * Get the dnode associated with object, so we can calculate the block
244 * size.
245 */
246 if ((err = dmu_objset_own(dataset, DMU_OST_ANY,
247 B_TRUE, FTAG, &os)) != 0) {
248 (void) fprintf(stderr, "cannot open dataset '%s': %s\n",
249 dataset, strerror(err));
250 goto out;
251 }
252
253 if (record->zi_object == 0) {
254 dn = DMU_META_DNODE(os);
255 } else {
256 err = dnode_hold(os, record->zi_object, FTAG, &dn);
257 if (err != 0) {
258 (void) fprintf(stderr, "failed to hold dnode "
259 "for object %llu\n",
260 (u_longlong_t)record->zi_object);
261 goto out;
262 }
263 }
264
265
266 ziprintf("data shift: %d\n", (int)dn->dn_datablkshift);
267 ziprintf(" ind shift: %d\n", (int)dn->dn_indblkshift);
268
269 /*
270 * Translate range into block IDs.
271 */
272 if (record->zi_start != 0 || record->zi_end != -1ULL) {
273 record->zi_start >>= dn->dn_datablkshift;
274 record->zi_end >>= dn->dn_datablkshift;
275 }
276
277 /*
278 * Check level, and then translate level 0 blkids into ranges
279 * appropriate for level of indirection.
280 */
281 record->zi_level = level;
282 if (level > 0) {
283 ziprintf("level 0 blkid range: [%llu, %llu]\n",
284 record->zi_start, record->zi_end);
285
286 if (level >= dn->dn_nlevels) {
287 (void) fprintf(stderr, "level %d exceeds max level "
288 "of object (%d)\n", level, dn->dn_nlevels - 1);
289 goto out;
290 }
291
292 if (record->zi_start != 0 || record->zi_end != 0) {
293 int shift = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
294
295 for (; level > 0; level--) {
296 record->zi_start >>= shift;
297 record->zi_end >>= shift;
298 }
299 }
300 }
301
302 ret = 0;
303 out:
304 if (dn) {
305 if (dn != DMU_META_DNODE(os))
306 dnode_rele(dn, FTAG);
307 }
308 if (os)
309 dmu_objset_disown(os, FTAG);
310
311 return (ret);
312 }
313
314 int
translate_record(err_type_t type,const char * object,const char * range,int level,zinject_record_t * record,char * poolname,char * dataset)315 translate_record(err_type_t type, const char *object, const char *range,
316 int level, zinject_record_t *record, char *poolname, char *dataset)
317 {
318 char path[MAXPATHLEN];
319 char *slash;
320 struct stat64 statbuf;
321 int ret = -1;
322
323 kernel_init(FREAD);
324
325 debug = (getenv("ZINJECT_DEBUG") != NULL);
326
327 ziprintf("translating: %s\n", object);
328
329 if (MOS_TYPE(type)) {
330 /*
331 * MOS objects are treated specially.
332 */
333 switch (type) {
334 case TYPE_MOS:
335 record->zi_type = 0;
336 break;
337 case TYPE_MOSDIR:
338 record->zi_type = DMU_OT_OBJECT_DIRECTORY;
339 break;
340 case TYPE_METASLAB:
341 record->zi_type = DMU_OT_OBJECT_ARRAY;
342 break;
343 case TYPE_CONFIG:
344 record->zi_type = DMU_OT_PACKED_NVLIST;
345 break;
346 case TYPE_BPOBJ:
347 record->zi_type = DMU_OT_BPOBJ;
348 break;
349 case TYPE_SPACEMAP:
350 record->zi_type = DMU_OT_SPACE_MAP;
351 break;
352 case TYPE_ERRLOG:
353 record->zi_type = DMU_OT_ERROR_LOG;
354 break;
355 }
356
357 dataset[0] = '\0';
358 (void) strcpy(poolname, object);
359 return (0);
360 }
361
362 /*
363 * Convert a full path into a (dataset, file) pair.
364 */
365 if (parse_pathname(object, dataset, path, &statbuf) != 0)
366 goto err;
367
368 ziprintf(" dataset: %s\n", dataset);
369 ziprintf(" path: %s\n", path);
370
371 /*
372 * Convert (dataset, file) into (objset, object)
373 */
374 if (object_from_path(dataset, path, &statbuf, record) != 0)
375 goto err;
376
377 ziprintf("raw objset: %llu\n", record->zi_objset);
378 ziprintf("raw object: %llu\n", record->zi_object);
379
380 /*
381 * For the given object, calculate the real (type, level, range)
382 */
383 if (calculate_range(dataset, type, level, (char *)range, record) != 0)
384 goto err;
385
386 ziprintf(" objset: %llu\n", record->zi_objset);
387 ziprintf(" object: %llu\n", record->zi_object);
388 if (record->zi_start == 0 &&
389 record->zi_end == -1ULL)
390 ziprintf(" range: all\n");
391 else
392 ziprintf(" range: [%llu, %llu]\n", record->zi_start,
393 record->zi_end);
394
395 /*
396 * Copy the pool name
397 */
398 (void) strcpy(poolname, dataset);
399 if ((slash = strchr(poolname, '/')) != NULL)
400 *slash = '\0';
401
402 ret = 0;
403
404 err:
405 kernel_fini();
406 return (ret);
407 }
408
409 int
translate_raw(const char * str,zinject_record_t * record)410 translate_raw(const char *str, zinject_record_t *record)
411 {
412 /*
413 * A raw bookmark of the form objset:object:level:blkid, where each
414 * number is a hexidecimal value.
415 */
416 if (sscanf(str, "%llx:%llx:%x:%llx", (u_longlong_t *)&record->zi_objset,
417 (u_longlong_t *)&record->zi_object, &record->zi_level,
418 (u_longlong_t *)&record->zi_start) != 4) {
419 (void) fprintf(stderr, "bad raw spec '%s': must be of the form "
420 "'objset:object:level:blkid'\n", str);
421 return (-1);
422 }
423
424 record->zi_end = record->zi_start;
425
426 return (0);
427 }
428
429 int
translate_device(const char * pool,const char * device,err_type_t label_type,zinject_record_t * record)430 translate_device(const char *pool, const char *device, err_type_t label_type,
431 zinject_record_t *record)
432 {
433 char *end;
434 zpool_handle_t *zhp;
435 nvlist_t *tgt;
436 boolean_t isspare, iscache;
437
438 /*
439 * Given a device name or GUID, create an appropriate injection record
440 * with zi_guid set.
441 */
442 if ((zhp = zpool_open(g_zfs, pool)) == NULL)
443 return (-1);
444
445 record->zi_guid = strtoull(device, &end, 16);
446 if (record->zi_guid == 0 || *end != '\0') {
447 tgt = zpool_find_vdev(zhp, device, &isspare, &iscache, NULL);
448
449 if (tgt == NULL) {
450 (void) fprintf(stderr, "cannot find device '%s' in "
451 "pool '%s'\n", device, pool);
452 return (-1);
453 }
454
455 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
456 &record->zi_guid) == 0);
457 }
458
459 /*
460 * Device faults can take on three different forms:
461 * 1). delayed or hanging I/O
462 * 2). zfs label faults
463 * 3). generic disk faults
464 */
465 if (record->zi_timer != 0) {
466 record->zi_cmd = ZINJECT_DELAY_IO;
467 } else if (label_type != TYPE_INVAL) {
468 record->zi_cmd = ZINJECT_LABEL_FAULT;
469 } else {
470 record->zi_cmd = ZINJECT_DEVICE_FAULT;
471 }
472
473 switch (label_type) {
474 case TYPE_LABEL_UBERBLOCK:
475 record->zi_start = offsetof(vdev_label_t, vl_uberblock[0]);
476 record->zi_end = record->zi_start + VDEV_UBERBLOCK_RING - 1;
477 break;
478 case TYPE_LABEL_NVLIST:
479 record->zi_start = offsetof(vdev_label_t, vl_vdev_phys);
480 record->zi_end = record->zi_start + VDEV_PHYS_SIZE - 1;
481 break;
482 case TYPE_LABEL_PAD1:
483 record->zi_start = offsetof(vdev_label_t, vl_pad1);
484 record->zi_end = record->zi_start + VDEV_PAD_SIZE - 1;
485 break;
486 case TYPE_LABEL_PAD2:
487 record->zi_start = offsetof(vdev_label_t, vl_pad2);
488 record->zi_end = record->zi_start + VDEV_PAD_SIZE - 1;
489 break;
490 }
491 return (0);
492 }
493