translate.c revision feef89cf5f5fee792c1a396bb0e48070935cf65a
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 2010 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
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 <sys/mkdev.h>
49
50#include "zinject.h"
51
52extern void kernel_init(int);
53extern void kernel_fini(void);
54
55static int debug;
56
57static void
58ziprintf(const char *fmt, ...)
59{
60	va_list ap;
61
62	if (!debug)
63		return;
64
65	va_start(ap, fmt);
66	(void) vprintf(fmt, ap);
67	va_end(ap);
68}
69
70/*
71 * Given a full path to a file, translate into a dataset name and a relative
72 * path within the dataset.  'dataset' must be at least MAXNAMELEN characters,
73 * and 'relpath' must be at least MAXPATHLEN characters.  We also pass a stat64
74 * buffer, which we need later to get the object ID.
75 */
76static int
77parse_pathname(const char *fullpath, char *dataset, char *relpath,
78    struct stat64 *statbuf)
79{
80	struct extmnttab mp;
81	FILE *fp;
82	int match;
83	const char *rel;
84
85	if (fullpath[0] != '/') {
86		(void) fprintf(stderr, "invalid object '%s': must be full "
87		    "path\n", fullpath);
88		usage();
89		return (-1);
90	}
91
92	if (strlen(fullpath) >= MAXPATHLEN) {
93		(void) fprintf(stderr, "invalid object; pathname too long\n");
94		return (-1);
95	}
96
97	if (stat64(fullpath, statbuf) != 0) {
98		(void) fprintf(stderr, "cannot open '%s': %s\n",
99		    fullpath, strerror(errno));
100		return (-1);
101	}
102
103	if ((fp = fopen(MNTTAB, "r")) == NULL) {
104		(void) fprintf(stderr, "cannot open /etc/mnttab\n");
105		return (-1);
106	}
107
108	match = 0;
109	while (getextmntent(fp, &mp, sizeof (mp)) == 0) {
110		if (makedev(mp.mnt_major, mp.mnt_minor) == statbuf->st_dev) {
111			match = 1;
112			break;
113		}
114	}
115
116	if (!match) {
117		(void) fprintf(stderr, "cannot find mountpoint for '%s'\n",
118		    fullpath);
119		return (-1);
120	}
121
122	if (strcmp(mp.mnt_fstype, MNTTYPE_ZFS) != 0) {
123		(void) fprintf(stderr, "invalid path '%s': not a ZFS "
124		    "filesystem\n", fullpath);
125		return (-1);
126	}
127
128	if (strncmp(fullpath, mp.mnt_mountp, strlen(mp.mnt_mountp)) != 0) {
129		(void) fprintf(stderr, "invalid path '%s': mountpoint "
130		    "doesn't match path\n", fullpath);
131		return (-1);
132	}
133
134	(void) strcpy(dataset, mp.mnt_special);
135
136	rel = fullpath + strlen(mp.mnt_mountp);
137	if (rel[0] == '/')
138		rel++;
139	(void) strcpy(relpath, rel);
140
141	return (0);
142}
143
144/*
145 * Convert from a (dataset, path) pair into a (objset, object) pair.  Note that
146 * we grab the object number from the inode number, since looking this up via
147 * libzpool is a real pain.
148 */
149/* ARGSUSED */
150static int
151object_from_path(const char *dataset, const char *path, struct stat64 *statbuf,
152    zinject_record_t *record)
153{
154	objset_t *os;
155	int err;
156
157	/*
158	 * Before doing any libzpool operations, call sync() to ensure that the
159	 * on-disk state is consistent with the in-core state.
160	 */
161	sync();
162
163	err = dmu_objset_own(dataset, DMU_OST_ZFS, B_TRUE, FTAG, &os);
164	if (err != 0) {
165		(void) fprintf(stderr, "cannot open dataset '%s': %s\n",
166		    dataset, strerror(err));
167		return (-1);
168	}
169
170	record->zi_objset = dmu_objset_id(os);
171	record->zi_object = statbuf->st_ino;
172
173	dmu_objset_disown(os, FTAG);
174
175	return (0);
176}
177
178/*
179 * Calculate the real range based on the type, level, and range given.
180 */
181static int
182calculate_range(const char *dataset, err_type_t type, int level, char *range,
183    zinject_record_t *record)
184{
185	objset_t *os = NULL;
186	dnode_t *dn = NULL;
187	int err;
188	int ret = -1;
189
190	/*
191	 * Determine the numeric range from the string.
192	 */
193	if (range == NULL) {
194		/*
195		 * If range is unspecified, set the range to [0,-1], which
196		 * indicates that the whole object should be treated as an
197		 * error.
198		 */
199		record->zi_start = 0;
200		record->zi_end = -1ULL;
201	} else {
202		char *end;
203
204		/* XXX add support for suffixes */
205		record->zi_start = strtoull(range, &end, 10);
206
207
208		if (*end == '\0')
209			record->zi_end = record->zi_start + 1;
210		else if (*end == ',')
211			record->zi_end = strtoull(end + 1, &end, 10);
212
213		if (*end != '\0') {
214			(void) fprintf(stderr, "invalid range '%s': must be "
215			    "a numeric range of the form 'start[,end]'\n",
216			    range);
217			goto out;
218		}
219	}
220
221	switch (type) {
222	case TYPE_DATA:
223		break;
224
225	case TYPE_DNODE:
226		/*
227		 * If this is a request to inject faults into the dnode, then we
228		 * must translate the current (objset,object) pair into an
229		 * offset within the metadnode for the objset.  Specifying any
230		 * kind of range with type 'dnode' is illegal.
231		 */
232		if (range != NULL) {
233			(void) fprintf(stderr, "range cannot be specified when "
234			    "type is 'dnode'\n");
235			goto out;
236		}
237
238		record->zi_start = record->zi_object * sizeof (dnode_phys_t);
239		record->zi_end = record->zi_start + sizeof (dnode_phys_t);
240		record->zi_object = 0;
241		break;
242	}
243
244	/*
245	 * Get the dnode associated with object, so we can calculate the block
246	 * size.
247	 */
248	if ((err = dmu_objset_own(dataset, DMU_OST_ANY,
249	    B_TRUE, FTAG, &os)) != 0) {
250		(void) fprintf(stderr, "cannot open dataset '%s': %s\n",
251		    dataset, strerror(err));
252		goto out;
253	}
254
255	if (record->zi_object == 0) {
256		dn = os->os_meta_dnode;
257	} else {
258		err = dnode_hold(os, record->zi_object, FTAG, &dn);
259		if (err != 0) {
260			(void) fprintf(stderr, "failed to hold dnode "
261			    "for object %llu\n",
262			    (u_longlong_t)record->zi_object);
263			goto out;
264		}
265	}
266
267
268	ziprintf("data shift: %d\n", (int)dn->dn_datablkshift);
269	ziprintf(" ind shift: %d\n", (int)dn->dn_indblkshift);
270
271	/*
272	 * Translate range into block IDs.
273	 */
274	if (record->zi_start != 0 || record->zi_end != -1ULL) {
275		record->zi_start >>= dn->dn_datablkshift;
276		record->zi_end >>= dn->dn_datablkshift;
277	}
278
279	/*
280	 * Check level, and then translate level 0 blkids into ranges
281	 * appropriate for level of indirection.
282	 */
283	record->zi_level = level;
284	if (level > 0) {
285		ziprintf("level 0 blkid range: [%llu, %llu]\n",
286		    record->zi_start, record->zi_end);
287
288		if (level >= dn->dn_nlevels) {
289			(void) fprintf(stderr, "level %d exceeds max level "
290			    "of object (%d)\n", level, dn->dn_nlevels - 1);
291			goto out;
292		}
293
294		if (record->zi_start != 0 || record->zi_end != 0) {
295			int shift = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
296
297			for (; level > 0; level--) {
298				record->zi_start >>= shift;
299				record->zi_end >>= shift;
300			}
301		}
302	}
303
304	ret = 0;
305out:
306	if (dn) {
307		if (dn != os->os_meta_dnode)
308			dnode_rele(dn, FTAG);
309	}
310	if (os)
311		dmu_objset_disown(os, FTAG);
312
313	return (ret);
314}
315
316int
317translate_record(err_type_t type, const char *object, const char *range,
318    int level, zinject_record_t *record, char *poolname, char *dataset)
319{
320	char path[MAXPATHLEN];
321	char *slash;
322	struct stat64 statbuf;
323	int ret = -1;
324
325	kernel_init(FREAD);
326
327	debug = (getenv("ZINJECT_DEBUG") != NULL);
328
329	ziprintf("translating: %s\n", object);
330
331	if (MOS_TYPE(type)) {
332		/*
333		 * MOS objects are treated specially.
334		 */
335		switch (type) {
336		case TYPE_MOS:
337			record->zi_type = 0;
338			break;
339		case TYPE_MOSDIR:
340			record->zi_type = DMU_OT_OBJECT_DIRECTORY;
341			break;
342		case TYPE_METASLAB:
343			record->zi_type = DMU_OT_OBJECT_ARRAY;
344			break;
345		case TYPE_CONFIG:
346			record->zi_type = DMU_OT_PACKED_NVLIST;
347			break;
348		case TYPE_BPLIST:
349			record->zi_type = DMU_OT_BPLIST;
350			break;
351		case TYPE_SPACEMAP:
352			record->zi_type = DMU_OT_SPACE_MAP;
353			break;
354		case TYPE_ERRLOG:
355			record->zi_type = DMU_OT_ERROR_LOG;
356			break;
357		}
358
359		dataset[0] = '\0';
360		(void) strcpy(poolname, object);
361		return (0);
362	}
363
364	/*
365	 * Convert a full path into a (dataset, file) pair.
366	 */
367	if (parse_pathname(object, dataset, path, &statbuf) != 0)
368		goto err;
369
370	ziprintf("   dataset: %s\n", dataset);
371	ziprintf("      path: %s\n", path);
372
373	/*
374	 * Convert (dataset, file) into (objset, object)
375	 */
376	if (object_from_path(dataset, path, &statbuf, record) != 0)
377		goto err;
378
379	ziprintf("raw objset: %llu\n", record->zi_objset);
380	ziprintf("raw object: %llu\n", record->zi_object);
381
382	/*
383	 * For the given object, calculate the real (type, level, range)
384	 */
385	if (calculate_range(dataset, type, level, (char *)range, record) != 0)
386		goto err;
387
388	ziprintf("    objset: %llu\n", record->zi_objset);
389	ziprintf("    object: %llu\n", record->zi_object);
390	if (record->zi_start == 0 &&
391	    record->zi_end == -1ULL)
392		ziprintf("     range: all\n");
393	else
394		ziprintf("     range: [%llu, %llu]\n", record->zi_start,
395		    record->zi_end);
396
397	/*
398	 * Copy the pool name
399	 */
400	(void) strcpy(poolname, dataset);
401	if ((slash = strchr(poolname, '/')) != NULL)
402		*slash = '\0';
403
404	ret = 0;
405
406err:
407	kernel_fini();
408	return (ret);
409}
410
411int
412translate_raw(const char *str, zinject_record_t *record)
413{
414	/*
415	 * A raw bookmark of the form objset:object:level:blkid, where each
416	 * number is a hexidecimal value.
417	 */
418	if (sscanf(str, "%llx:%llx:%x:%llx", (u_longlong_t *)&record->zi_objset,
419	    (u_longlong_t *)&record->zi_object, &record->zi_level,
420	    (u_longlong_t *)&record->zi_start) != 4) {
421		(void) fprintf(stderr, "bad raw spec '%s': must be of the form "
422		    "'objset:object:level:blkid'\n", str);
423		return (-1);
424	}
425
426	record->zi_end = record->zi_start;
427
428	return (0);
429}
430
431int
432translate_device(const char *pool, const char *device, err_type_t label_type,
433    zinject_record_t *record)
434{
435	char *end;
436	zpool_handle_t *zhp;
437	nvlist_t *tgt;
438	boolean_t isspare, iscache;
439
440	/*
441	 * Given a device name or GUID, create an appropriate injection record
442	 * with zi_guid set.
443	 */
444	if ((zhp = zpool_open(g_zfs, pool)) == NULL)
445		return (-1);
446
447	record->zi_guid = strtoull(device, &end, 16);
448	if (record->zi_guid == 0 || *end != '\0') {
449		tgt = zpool_find_vdev(zhp, device, &isspare, &iscache, NULL);
450
451		if (tgt == NULL) {
452			(void) fprintf(stderr, "cannot find device '%s' in "
453			    "pool '%s'\n", device, pool);
454			return (-1);
455		}
456
457		verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
458		    &record->zi_guid) == 0);
459	}
460
461	switch (label_type) {
462	case TYPE_LABEL_UBERBLOCK:
463		record->zi_start = offsetof(vdev_label_t, vl_uberblock[0]);
464		record->zi_end = record->zi_start + VDEV_UBERBLOCK_RING - 1;
465		break;
466	case TYPE_LABEL_NVLIST:
467		record->zi_start = offsetof(vdev_label_t, vl_vdev_phys);
468		record->zi_end = record->zi_start + VDEV_PHYS_SIZE - 1;
469		break;
470	}
471	return (0);
472}
473