4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
28 #include <sys/zfs_context.h>
38 #include <sys/mntent.h>
39 #include <sys/mnttab.h>
40 #include <sys/param.h>
44 #include <sys/dmu_objset.h>
45 #include <sys/dnode.h>
46 #include <sys/vdev_impl.h>
48 #include <sys/mkdev.h>
52 extern void kernel_init(int);
53 extern void kernel_fini(void);
58 ziprintf(const char *fmt, ...)
66 (void) vprintf(fmt, ap);
71 compress_slashes(const char *src, char *dest)
73 while (*src != '\0') {
75 while (*dest == '/' && *src == '/')
83 * Given a full path to a file, translate into a dataset name and a relative
84 * path within the dataset. 'dataset' must be at least MAXNAMELEN characters,
85 * and 'relpath' must be at least MAXPATHLEN characters. We also pass a stat64
86 * buffer, which we need later to get the object ID.
89 parse_pathname(const char *inpath, char *dataset, char *relpath,
90 struct stat64 *statbuf)
96 char fullpath[MAXPATHLEN];
98 compress_slashes(inpath, fullpath);
100 if (fullpath[0] != '/') {
101 (void) fprintf(stderr, "invalid object '%s': must be full "
107 if (strlen(fullpath) >= MAXPATHLEN) {
108 (void) fprintf(stderr, "invalid object; pathname too long\n");
112 if (stat64(fullpath, statbuf) != 0) {
113 (void) fprintf(stderr, "cannot open '%s': %s\n",
114 fullpath, strerror(errno));
118 #ifdef HAVE_SETMNTENT
119 if ((fp = setmntent(MNTTAB, "r")) == NULL) {
121 if ((fp = fopen(MNTTAB, "r")) == NULL) {
123 (void) fprintf(stderr, "cannot open /etc/mtab\n");
128 while (getextmntent(fp, &mp, sizeof (mp)) == 0) {
129 if (makedev(mp.mnt_major, mp.mnt_minor) == statbuf->st_dev) {
136 (void) fprintf(stderr, "cannot find mountpoint for '%s'\n",
141 if (strcmp(mp.mnt_fstype, MNTTYPE_ZFS) != 0) {
142 (void) fprintf(stderr, "invalid path '%s': not a ZFS "
143 "filesystem\n", fullpath);
147 if (strncmp(fullpath, mp.mnt_mountp, strlen(mp.mnt_mountp)) != 0) {
148 (void) fprintf(stderr, "invalid path '%s': mountpoint "
149 "doesn't match path\n", fullpath);
153 (void) strcpy(dataset, mp.mnt_special);
155 rel = fullpath + strlen(mp.mnt_mountp);
158 (void) strcpy(relpath, rel);
164 * Convert from a (dataset, path) pair into a (objset, object) pair. Note that
165 * we grab the object number from the inode number, since looking this up via
166 * libzpool is a real pain.
170 object_from_path(const char *dataset, const char *path, struct stat64 *statbuf,
171 zinject_record_t *record)
177 * Before doing any libzpool operations, call sync() to ensure that the
178 * on-disk state is consistent with the in-core state.
182 err = dmu_objset_own(dataset, DMU_OST_ZFS, B_TRUE, FTAG, &os);
184 (void) fprintf(stderr, "cannot open dataset '%s': %s\n",
185 dataset, strerror(err));
189 record->zi_objset = dmu_objset_id(os);
190 record->zi_object = statbuf->st_ino;
192 dmu_objset_disown(os, FTAG);
198 * Calculate the real range based on the type, level, and range given.
201 calculate_range(const char *dataset, err_type_t type, int level, char *range,
202 zinject_record_t *record)
210 * Determine the numeric range from the string.
214 * If range is unspecified, set the range to [0,-1], which
215 * indicates that the whole object should be treated as an
218 record->zi_start = 0;
219 record->zi_end = -1ULL;
223 /* XXX add support for suffixes */
224 record->zi_start = strtoull(range, &end, 10);
228 record->zi_end = record->zi_start + 1;
229 else if (*end == ',')
230 record->zi_end = strtoull(end + 1, &end, 10);
233 (void) fprintf(stderr, "invalid range '%s': must be "
234 "a numeric range of the form 'start[,end]'\n",
248 * If this is a request to inject faults into the dnode, then we
249 * must translate the current (objset,object) pair into an
250 * offset within the metadnode for the objset. Specifying any
251 * kind of range with type 'dnode' is illegal.
254 (void) fprintf(stderr, "range cannot be specified when "
255 "type is 'dnode'\n");
259 record->zi_start = record->zi_object * sizeof (dnode_phys_t);
260 record->zi_end = record->zi_start + sizeof (dnode_phys_t);
261 record->zi_object = 0;
266 * Get the dnode associated with object, so we can calculate the block
269 if ((err = dmu_objset_own(dataset, DMU_OST_ANY,
270 B_TRUE, FTAG, &os)) != 0) {
271 (void) fprintf(stderr, "cannot open dataset '%s': %s\n",
272 dataset, strerror(err));
276 if (record->zi_object == 0) {
277 dn = DMU_META_DNODE(os);
279 err = dnode_hold(os, record->zi_object, FTAG, &dn);
281 (void) fprintf(stderr, "failed to hold dnode "
283 (u_longlong_t)record->zi_object);
289 ziprintf("data shift: %d\n", (int)dn->dn_datablkshift);
290 ziprintf(" ind shift: %d\n", (int)dn->dn_indblkshift);
293 * Translate range into block IDs.
295 if (record->zi_start != 0 || record->zi_end != -1ULL) {
296 record->zi_start >>= dn->dn_datablkshift;
297 record->zi_end >>= dn->dn_datablkshift;
301 * Check level, and then translate level 0 blkids into ranges
302 * appropriate for level of indirection.
304 record->zi_level = level;
306 ziprintf("level 0 blkid range: [%llu, %llu]\n",
307 record->zi_start, record->zi_end);
309 if (level >= dn->dn_nlevels) {
310 (void) fprintf(stderr, "level %d exceeds max level "
311 "of object (%d)\n", level, dn->dn_nlevels - 1);
315 if (record->zi_start != 0 || record->zi_end != 0) {
316 int shift = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
318 for (; level > 0; level--) {
319 record->zi_start >>= shift;
320 record->zi_end >>= shift;
328 if (dn != DMU_META_DNODE(os))
329 dnode_rele(dn, FTAG);
332 dmu_objset_disown(os, FTAG);
338 translate_record(err_type_t type, const char *object, const char *range,
339 int level, zinject_record_t *record, char *poolname, char *dataset)
341 char path[MAXPATHLEN];
343 struct stat64 statbuf;
348 debug = (getenv("ZINJECT_DEBUG") != NULL);
350 ziprintf("translating: %s\n", object);
352 if (MOS_TYPE(type)) {
354 * MOS objects are treated specially.
363 record->zi_type = DMU_OT_OBJECT_DIRECTORY;
366 record->zi_type = DMU_OT_OBJECT_ARRAY;
369 record->zi_type = DMU_OT_PACKED_NVLIST;
372 record->zi_type = DMU_OT_BPOBJ;
375 record->zi_type = DMU_OT_SPACE_MAP;
378 record->zi_type = DMU_OT_ERROR_LOG;
383 (void) strcpy(poolname, object);
388 * Convert a full path into a (dataset, file) pair.
390 if (parse_pathname(object, dataset, path, &statbuf) != 0)
393 ziprintf(" dataset: %s\n", dataset);
394 ziprintf(" path: %s\n", path);
397 * Convert (dataset, file) into (objset, object)
399 if (object_from_path(dataset, path, &statbuf, record) != 0)
402 ziprintf("raw objset: %llu\n", record->zi_objset);
403 ziprintf("raw object: %llu\n", record->zi_object);
406 * For the given object, calculate the real (type, level, range)
408 if (calculate_range(dataset, type, level, (char *)range, record) != 0)
411 ziprintf(" objset: %llu\n", record->zi_objset);
412 ziprintf(" object: %llu\n", record->zi_object);
413 if (record->zi_start == 0 &&
414 record->zi_end == -1ULL)
415 ziprintf(" range: all\n");
417 ziprintf(" range: [%llu, %llu]\n", record->zi_start,
423 (void) strcpy(poolname, dataset);
424 if ((slash = strchr(poolname, '/')) != NULL)
435 translate_raw(const char *str, zinject_record_t *record)
438 * A raw bookmark of the form objset:object:level:blkid, where each
439 * number is a hexidecimal value.
441 if (sscanf(str, "%llx:%llx:%x:%llx", (u_longlong_t *)&record->zi_objset,
442 (u_longlong_t *)&record->zi_object, &record->zi_level,
443 (u_longlong_t *)&record->zi_start) != 4) {
444 (void) fprintf(stderr, "bad raw spec '%s': must be of the form "
445 "'objset:object:level:blkid'\n", str);
449 record->zi_end = record->zi_start;
455 translate_device(const char *pool, const char *device, err_type_t label_type,
456 zinject_record_t *record)
461 boolean_t isspare, iscache;
464 * Given a device name or GUID, create an appropriate injection record
467 if ((zhp = zpool_open(g_zfs, pool)) == NULL)
470 record->zi_guid = strtoull(device, &end, 16);
471 if (record->zi_guid == 0 || *end != '\0') {
472 tgt = zpool_find_vdev(zhp, device, &isspare, &iscache, NULL);
475 (void) fprintf(stderr, "cannot find device '%s' in "
476 "pool '%s'\n", device, pool);
480 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
481 &record->zi_guid) == 0);
485 * Device faults can take on three different forms:
486 * 1). delayed or hanging I/O
487 * 2). zfs label faults
488 * 3). generic disk faults
490 if (record->zi_timer != 0) {
491 record->zi_cmd = ZINJECT_DELAY_IO;
492 } else if (label_type != TYPE_INVAL) {
493 record->zi_cmd = ZINJECT_LABEL_FAULT;
495 record->zi_cmd = ZINJECT_DEVICE_FAULT;
498 switch (label_type) {
501 case TYPE_LABEL_UBERBLOCK:
502 record->zi_start = offsetof(vdev_label_t, vl_uberblock[0]);
503 record->zi_end = record->zi_start + VDEV_UBERBLOCK_RING - 1;
505 case TYPE_LABEL_NVLIST:
506 record->zi_start = offsetof(vdev_label_t, vl_vdev_phys);
507 record->zi_end = record->zi_start + VDEV_PHYS_SIZE - 1;
509 case TYPE_LABEL_PAD1:
510 record->zi_start = offsetof(vdev_label_t, vl_pad1);
511 record->zi_end = record->zi_start + VDEV_PAD_SIZE - 1;
513 case TYPE_LABEL_PAD2:
514 record->zi_start = offsetof(vdev_label_t, vl_pad2);
515 record->zi_end = record->zi_start + VDEV_PAD_SIZE - 1;