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.
27 #include <sys/zfs_context.h>
37 #include <sys/mntent.h>
38 #include <sys/mnttab.h>
39 #include <sys/param.h>
43 #include <sys/dmu_objset.h>
44 #include <sys/dnode.h>
45 #include <sys/vdev_impl.h>
47 #include <sys/mkdev.h>
51 extern void kernel_init(int);
52 extern void kernel_fini(void);
57 ziprintf(const char *fmt, ...)
65 (void) vprintf(fmt, ap);
70 compress_slashes(const char *src, char *dest)
72 while (*src != '\0') {
74 while (*dest == '/' && *src == '/')
82 * Given a full path to a file, translate into a dataset name and a relative
83 * path within the dataset. 'dataset' must be at least MAXNAMELEN characters,
84 * and 'relpath' must be at least MAXPATHLEN characters. We also pass a stat64
85 * buffer, which we need later to get the object ID.
88 parse_pathname(const char *inpath, char *dataset, char *relpath,
89 struct stat64 *statbuf)
95 char fullpath[MAXPATHLEN];
97 compress_slashes(inpath, fullpath);
99 if (fullpath[0] != '/') {
100 (void) fprintf(stderr, "invalid object '%s': must be full "
106 if (strlen(fullpath) >= MAXPATHLEN) {
107 (void) fprintf(stderr, "invalid object; pathname too long\n");
111 if (stat64(fullpath, statbuf) != 0) {
112 (void) fprintf(stderr, "cannot open '%s': %s\n",
113 fullpath, strerror(errno));
117 if ((fp = fopen(MNTTAB, "r")) == NULL) {
118 (void) fprintf(stderr, "cannot open /etc/mnttab\n");
123 while (getextmntent(fp, &mp, sizeof (mp)) == 0) {
124 if (makedev(mp.mnt_major, mp.mnt_minor) == statbuf->st_dev) {
131 (void) fprintf(stderr, "cannot find mountpoint for '%s'\n",
136 if (strcmp(mp.mnt_fstype, MNTTYPE_ZFS) != 0) {
137 (void) fprintf(stderr, "invalid path '%s': not a ZFS "
138 "filesystem\n", fullpath);
142 if (strncmp(fullpath, mp.mnt_mountp, strlen(mp.mnt_mountp)) != 0) {
143 (void) fprintf(stderr, "invalid path '%s': mountpoint "
144 "doesn't match path\n", fullpath);
148 (void) strcpy(dataset, mp.mnt_special);
150 rel = fullpath + strlen(mp.mnt_mountp);
153 (void) strcpy(relpath, rel);
159 * Convert from a (dataset, path) pair into a (objset, object) pair. Note that
160 * we grab the object number from the inode number, since looking this up via
161 * libzpool is a real pain.
165 object_from_path(const char *dataset, const char *path, struct stat64 *statbuf,
166 zinject_record_t *record)
172 * Before doing any libzpool operations, call sync() to ensure that the
173 * on-disk state is consistent with the in-core state.
177 err = dmu_objset_own(dataset, DMU_OST_ZFS, B_TRUE, FTAG, &os);
179 (void) fprintf(stderr, "cannot open dataset '%s': %s\n",
180 dataset, strerror(err));
184 record->zi_objset = dmu_objset_id(os);
185 record->zi_object = statbuf->st_ino;
187 dmu_objset_disown(os, FTAG);
193 * Calculate the real range based on the type, level, and range given.
196 calculate_range(const char *dataset, err_type_t type, int level, char *range,
197 zinject_record_t *record)
205 * Determine the numeric range from the string.
209 * If range is unspecified, set the range to [0,-1], which
210 * indicates that the whole object should be treated as an
213 record->zi_start = 0;
214 record->zi_end = -1ULL;
218 /* XXX add support for suffixes */
219 record->zi_start = strtoull(range, &end, 10);
223 record->zi_end = record->zi_start + 1;
224 else if (*end == ',')
225 record->zi_end = strtoull(end + 1, &end, 10);
228 (void) fprintf(stderr, "invalid range '%s': must be "
229 "a numeric range of the form 'start[,end]'\n",
241 * If this is a request to inject faults into the dnode, then we
242 * must translate the current (objset,object) pair into an
243 * offset within the metadnode for the objset. Specifying any
244 * kind of range with type 'dnode' is illegal.
247 (void) fprintf(stderr, "range cannot be specified when "
248 "type is 'dnode'\n");
252 record->zi_start = record->zi_object * sizeof (dnode_phys_t);
253 record->zi_end = record->zi_start + sizeof (dnode_phys_t);
254 record->zi_object = 0;
259 * Get the dnode associated with object, so we can calculate the block
262 if ((err = dmu_objset_own(dataset, DMU_OST_ANY,
263 B_TRUE, FTAG, &os)) != 0) {
264 (void) fprintf(stderr, "cannot open dataset '%s': %s\n",
265 dataset, strerror(err));
269 if (record->zi_object == 0) {
270 dn = os->os_meta_dnode;
272 err = dnode_hold(os, record->zi_object, FTAG, &dn);
274 (void) fprintf(stderr, "failed to hold dnode "
276 (u_longlong_t)record->zi_object);
282 ziprintf("data shift: %d\n", (int)dn->dn_datablkshift);
283 ziprintf(" ind shift: %d\n", (int)dn->dn_indblkshift);
286 * Translate range into block IDs.
288 if (record->zi_start != 0 || record->zi_end != -1ULL) {
289 record->zi_start >>= dn->dn_datablkshift;
290 record->zi_end >>= dn->dn_datablkshift;
294 * Check level, and then translate level 0 blkids into ranges
295 * appropriate for level of indirection.
297 record->zi_level = level;
299 ziprintf("level 0 blkid range: [%llu, %llu]\n",
300 record->zi_start, record->zi_end);
302 if (level >= dn->dn_nlevels) {
303 (void) fprintf(stderr, "level %d exceeds max level "
304 "of object (%d)\n", level, dn->dn_nlevels - 1);
308 if (record->zi_start != 0 || record->zi_end != 0) {
309 int shift = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
311 for (; level > 0; level--) {
312 record->zi_start >>= shift;
313 record->zi_end >>= shift;
321 if (dn != os->os_meta_dnode)
322 dnode_rele(dn, FTAG);
325 dmu_objset_disown(os, FTAG);
331 translate_record(err_type_t type, const char *object, const char *range,
332 int level, zinject_record_t *record, char *poolname, char *dataset)
334 char path[MAXPATHLEN];
336 struct stat64 statbuf;
341 debug = (getenv("ZINJECT_DEBUG") != NULL);
343 ziprintf("translating: %s\n", object);
345 if (MOS_TYPE(type)) {
347 * MOS objects are treated specially.
354 record->zi_type = DMU_OT_OBJECT_DIRECTORY;
357 record->zi_type = DMU_OT_OBJECT_ARRAY;
360 record->zi_type = DMU_OT_PACKED_NVLIST;
363 record->zi_type = DMU_OT_BPOBJ;
366 record->zi_type = DMU_OT_SPACE_MAP;
369 record->zi_type = DMU_OT_ERROR_LOG;
374 (void) strcpy(poolname, object);
379 * Convert a full path into a (dataset, file) pair.
381 if (parse_pathname(object, dataset, path, &statbuf) != 0)
384 ziprintf(" dataset: %s\n", dataset);
385 ziprintf(" path: %s\n", path);
388 * Convert (dataset, file) into (objset, object)
390 if (object_from_path(dataset, path, &statbuf, record) != 0)
393 ziprintf("raw objset: %llu\n", record->zi_objset);
394 ziprintf("raw object: %llu\n", record->zi_object);
397 * For the given object, calculate the real (type, level, range)
399 if (calculate_range(dataset, type, level, (char *)range, record) != 0)
402 ziprintf(" objset: %llu\n", record->zi_objset);
403 ziprintf(" object: %llu\n", record->zi_object);
404 if (record->zi_start == 0 &&
405 record->zi_end == -1ULL)
406 ziprintf(" range: all\n");
408 ziprintf(" range: [%llu, %llu]\n", record->zi_start,
414 (void) strcpy(poolname, dataset);
415 if ((slash = strchr(poolname, '/')) != NULL)
426 translate_raw(const char *str, zinject_record_t *record)
429 * A raw bookmark of the form objset:object:level:blkid, where each
430 * number is a hexidecimal value.
432 if (sscanf(str, "%llx:%llx:%x:%llx", (u_longlong_t *)&record->zi_objset,
433 (u_longlong_t *)&record->zi_object, &record->zi_level,
434 (u_longlong_t *)&record->zi_start) != 4) {
435 (void) fprintf(stderr, "bad raw spec '%s': must be of the form "
436 "'objset:object:level:blkid'\n", str);
440 record->zi_end = record->zi_start;
446 translate_device(const char *pool, const char *device, err_type_t label_type,
447 zinject_record_t *record)
452 boolean_t isspare, iscache;
455 * Given a device name or GUID, create an appropriate injection record
458 if ((zhp = zpool_open(g_zfs, pool)) == NULL)
461 record->zi_guid = strtoull(device, &end, 16);
462 if (record->zi_guid == 0 || *end != '\0') {
463 tgt = zpool_find_vdev(zhp, device, &isspare, &iscache, NULL);
466 (void) fprintf(stderr, "cannot find device '%s' in "
467 "pool '%s'\n", device, pool);
471 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
472 &record->zi_guid) == 0);
475 switch (label_type) {
476 case TYPE_LABEL_UBERBLOCK:
477 record->zi_start = offsetof(vdev_label_t, vl_uberblock[0]);
478 record->zi_end = record->zi_start + VDEV_UBERBLOCK_RING - 1;
480 case TYPE_LABEL_NVLIST:
481 record->zi_start = offsetof(vdev_label_t, vl_vdev_phys);
482 record->zi_end = record->zi_start + VDEV_PHYS_SIZE - 1;
484 case TYPE_LABEL_PAD1:
485 record->zi_start = offsetof(vdev_label_t, vl_pad1);
486 record->zi_end = record->zi_start + VDEV_PAD_SIZE - 1;
488 case TYPE_LABEL_PAD2:
489 record->zi_start = offsetof(vdev_label_t, vl_pad2);
490 record->zi_end = record->zi_start + VDEV_PAD_SIZE - 1;