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 if ((fp
= fopen(MNTTAB
, "r")) == NULL
) {
119 (void) fprintf(stderr
, "cannot open /etc/mnttab\n");
124 while (getextmntent(fp
, &mp
, sizeof (mp
)) == 0) {
125 if (makedev(mp
.mnt_major
, mp
.mnt_minor
) == statbuf
->st_dev
) {
132 (void) fprintf(stderr
, "cannot find mountpoint for '%s'\n",
137 if (strcmp(mp
.mnt_fstype
, MNTTYPE_ZFS
) != 0) {
138 (void) fprintf(stderr
, "invalid path '%s': not a ZFS "
139 "filesystem\n", fullpath
);
143 if (strncmp(fullpath
, mp
.mnt_mountp
, strlen(mp
.mnt_mountp
)) != 0) {
144 (void) fprintf(stderr
, "invalid path '%s': mountpoint "
145 "doesn't match path\n", fullpath
);
149 (void) strcpy(dataset
, mp
.mnt_special
);
151 rel
= fullpath
+ strlen(mp
.mnt_mountp
);
154 (void) strcpy(relpath
, rel
);
160 * Convert from a (dataset, path) pair into a (objset, object) pair. Note that
161 * we grab the object number from the inode number, since looking this up via
162 * libzpool is a real pain.
166 object_from_path(const char *dataset
, const char *path
, struct stat64
*statbuf
,
167 zinject_record_t
*record
)
173 * Before doing any libzpool operations, call sync() to ensure that the
174 * on-disk state is consistent with the in-core state.
178 err
= dmu_objset_own(dataset
, DMU_OST_ZFS
, B_TRUE
, FTAG
, &os
);
180 (void) fprintf(stderr
, "cannot open dataset '%s': %s\n",
181 dataset
, strerror(err
));
185 record
->zi_objset
= dmu_objset_id(os
);
186 record
->zi_object
= statbuf
->st_ino
;
188 dmu_objset_disown(os
, FTAG
);
194 * Calculate the real range based on the type, level, and range given.
197 calculate_range(const char *dataset
, err_type_t type
, int level
, char *range
,
198 zinject_record_t
*record
)
206 * Determine the numeric range from the string.
210 * If range is unspecified, set the range to [0,-1], which
211 * indicates that the whole object should be treated as an
214 record
->zi_start
= 0;
215 record
->zi_end
= -1ULL;
219 /* XXX add support for suffixes */
220 record
->zi_start
= strtoull(range
, &end
, 10);
224 record
->zi_end
= record
->zi_start
+ 1;
225 else if (*end
== ',')
226 record
->zi_end
= strtoull(end
+ 1, &end
, 10);
229 (void) fprintf(stderr
, "invalid range '%s': must be "
230 "a numeric range of the form 'start[,end]'\n",
242 * If this is a request to inject faults into the dnode, then we
243 * must translate the current (objset,object) pair into an
244 * offset within the metadnode for the objset. Specifying any
245 * kind of range with type 'dnode' is illegal.
248 (void) fprintf(stderr
, "range cannot be specified when "
249 "type is 'dnode'\n");
253 record
->zi_start
= record
->zi_object
* sizeof (dnode_phys_t
);
254 record
->zi_end
= record
->zi_start
+ sizeof (dnode_phys_t
);
255 record
->zi_object
= 0;
260 * Get the dnode associated with object, so we can calculate the block
263 if ((err
= dmu_objset_own(dataset
, DMU_OST_ANY
,
264 B_TRUE
, FTAG
, &os
)) != 0) {
265 (void) fprintf(stderr
, "cannot open dataset '%s': %s\n",
266 dataset
, strerror(err
));
270 if (record
->zi_object
== 0) {
271 dn
= DMU_META_DNODE(os
);
273 err
= dnode_hold(os
, record
->zi_object
, FTAG
, &dn
);
275 (void) fprintf(stderr
, "failed to hold dnode "
277 (u_longlong_t
)record
->zi_object
);
283 ziprintf("data shift: %d\n", (int)dn
->dn_datablkshift
);
284 ziprintf(" ind shift: %d\n", (int)dn
->dn_indblkshift
);
287 * Translate range into block IDs.
289 if (record
->zi_start
!= 0 || record
->zi_end
!= -1ULL) {
290 record
->zi_start
>>= dn
->dn_datablkshift
;
291 record
->zi_end
>>= dn
->dn_datablkshift
;
295 * Check level, and then translate level 0 blkids into ranges
296 * appropriate for level of indirection.
298 record
->zi_level
= level
;
300 ziprintf("level 0 blkid range: [%llu, %llu]\n",
301 record
->zi_start
, record
->zi_end
);
303 if (level
>= dn
->dn_nlevels
) {
304 (void) fprintf(stderr
, "level %d exceeds max level "
305 "of object (%d)\n", level
, dn
->dn_nlevels
- 1);
309 if (record
->zi_start
!= 0 || record
->zi_end
!= 0) {
310 int shift
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
312 for (; level
> 0; level
--) {
313 record
->zi_start
>>= shift
;
314 record
->zi_end
>>= shift
;
322 if (dn
!= DMU_META_DNODE(os
))
323 dnode_rele(dn
, FTAG
);
326 dmu_objset_disown(os
, FTAG
);
332 translate_record(err_type_t type
, const char *object
, const char *range
,
333 int level
, zinject_record_t
*record
, char *poolname
, char *dataset
)
335 char path
[MAXPATHLEN
];
337 struct stat64 statbuf
;
342 debug
= (getenv("ZINJECT_DEBUG") != NULL
);
344 ziprintf("translating: %s\n", object
);
346 if (MOS_TYPE(type
)) {
348 * MOS objects are treated specially.
355 record
->zi_type
= DMU_OT_OBJECT_DIRECTORY
;
358 record
->zi_type
= DMU_OT_OBJECT_ARRAY
;
361 record
->zi_type
= DMU_OT_PACKED_NVLIST
;
364 record
->zi_type
= DMU_OT_BPOBJ
;
367 record
->zi_type
= DMU_OT_SPACE_MAP
;
370 record
->zi_type
= DMU_OT_ERROR_LOG
;
375 (void) strcpy(poolname
, object
);
380 * Convert a full path into a (dataset, file) pair.
382 if (parse_pathname(object
, dataset
, path
, &statbuf
) != 0)
385 ziprintf(" dataset: %s\n", dataset
);
386 ziprintf(" path: %s\n", path
);
389 * Convert (dataset, file) into (objset, object)
391 if (object_from_path(dataset
, path
, &statbuf
, record
) != 0)
394 ziprintf("raw objset: %llu\n", record
->zi_objset
);
395 ziprintf("raw object: %llu\n", record
->zi_object
);
398 * For the given object, calculate the real (type, level, range)
400 if (calculate_range(dataset
, type
, level
, (char *)range
, record
) != 0)
403 ziprintf(" objset: %llu\n", record
->zi_objset
);
404 ziprintf(" object: %llu\n", record
->zi_object
);
405 if (record
->zi_start
== 0 &&
406 record
->zi_end
== -1ULL)
407 ziprintf(" range: all\n");
409 ziprintf(" range: [%llu, %llu]\n", record
->zi_start
,
415 (void) strcpy(poolname
, dataset
);
416 if ((slash
= strchr(poolname
, '/')) != NULL
)
427 translate_raw(const char *str
, zinject_record_t
*record
)
430 * A raw bookmark of the form objset:object:level:blkid, where each
431 * number is a hexidecimal value.
433 if (sscanf(str
, "%llx:%llx:%x:%llx", (u_longlong_t
*)&record
->zi_objset
,
434 (u_longlong_t
*)&record
->zi_object
, &record
->zi_level
,
435 (u_longlong_t
*)&record
->zi_start
) != 4) {
436 (void) fprintf(stderr
, "bad raw spec '%s': must be of the form "
437 "'objset:object:level:blkid'\n", str
);
441 record
->zi_end
= record
->zi_start
;
447 translate_device(const char *pool
, const char *device
, err_type_t label_type
,
448 zinject_record_t
*record
)
453 boolean_t isspare
, iscache
;
456 * Given a device name or GUID, create an appropriate injection record
459 if ((zhp
= zpool_open(g_zfs
, pool
)) == NULL
)
462 record
->zi_guid
= strtoull(device
, &end
, 16);
463 if (record
->zi_guid
== 0 || *end
!= '\0') {
464 tgt
= zpool_find_vdev(zhp
, device
, &isspare
, &iscache
, NULL
);
467 (void) fprintf(stderr
, "cannot find device '%s' in "
468 "pool '%s'\n", device
, pool
);
472 verify(nvlist_lookup_uint64(tgt
, ZPOOL_CONFIG_GUID
,
473 &record
->zi_guid
) == 0);
477 * Device faults can take on three different forms:
478 * 1). delayed or hanging I/O
479 * 2). zfs label faults
480 * 3). generic disk faults
482 if (record
->zi_timer
!= 0) {
483 record
->zi_cmd
= ZINJECT_DELAY_IO
;
484 } else if (label_type
!= TYPE_INVAL
) {
485 record
->zi_cmd
= ZINJECT_LABEL_FAULT
;
487 record
->zi_cmd
= ZINJECT_DEVICE_FAULT
;
490 switch (label_type
) {
491 case TYPE_LABEL_UBERBLOCK
:
492 record
->zi_start
= offsetof(vdev_label_t
, vl_uberblock
[0]);
493 record
->zi_end
= record
->zi_start
+ VDEV_UBERBLOCK_RING
- 1;
495 case TYPE_LABEL_NVLIST
:
496 record
->zi_start
= offsetof(vdev_label_t
, vl_vdev_phys
);
497 record
->zi_end
= record
->zi_start
+ VDEV_PHYS_SIZE
- 1;
499 case TYPE_LABEL_PAD1
:
500 record
->zi_start
= offsetof(vdev_label_t
, vl_pad1
);
501 record
->zi_end
= record
->zi_start
+ VDEV_PAD_SIZE
- 1;
503 case TYPE_LABEL_PAD2
:
504 record
->zi_start
= offsetof(vdev_label_t
, vl_pad2
);
505 record
->zi_end
= record
->zi_start
+ VDEV_PAD_SIZE
- 1;