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 <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
70static void
71compress_slashes(const char *src, char *dest)
72{
73	while (*src != '\0') {
74		*dest = *src++;
75		while (*dest == '/' && *src == '/')
76			++src;
77		++dest;
78	}
79	*dest = '\0';
80}
81
82/*
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.
87 */
88static int
89parse_pathname(const char *inpath, char *dataset, char *relpath,
90    struct stat64 *statbuf)
91{
92	struct extmnttab mp;
93	FILE *fp;
94	int match;
95	const char *rel;
96	char fullpath[MAXPATHLEN];
97
98	compress_slashes(inpath, fullpath);
99
100	if (fullpath[0] != '/') {
101		(void) fprintf(stderr, "invalid object '%s': must be full "
102		    "path\n", fullpath);
103		usage();
104		return (-1);
105	}
106
107	if (strlen(fullpath) >= MAXPATHLEN) {
108		(void) fprintf(stderr, "invalid object; pathname too long\n");
109		return (-1);
110	}
111
112	if (stat64(fullpath, statbuf) != 0) {
113		(void) fprintf(stderr, "cannot open '%s': %s\n",
114		    fullpath, strerror(errno));
115		return (-1);
116	}
117
118	if ((fp = fopen(MNTTAB, "r")) == NULL) {
119		(void) fprintf(stderr, "cannot open /etc/mnttab\n");
120		return (-1);
121	}
122
123	match = 0;
124	while (getextmntent(fp, &mp, sizeof (mp)) == 0) {
125		if (makedev(mp.mnt_major, mp.mnt_minor) == statbuf->st_dev) {
126			match = 1;
127			break;
128		}
129	}
130
131	if (!match) {
132		(void) fprintf(stderr, "cannot find mountpoint for '%s'\n",
133		    fullpath);
134		return (-1);
135	}
136
137	if (strcmp(mp.mnt_fstype, MNTTYPE_ZFS) != 0) {
138		(void) fprintf(stderr, "invalid path '%s': not a ZFS "
139		    "filesystem\n", fullpath);
140		return (-1);
141	}
142
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);
146		return (-1);
147	}
148
149	(void) strcpy(dataset, mp.mnt_special);
150
151	rel = fullpath + strlen(mp.mnt_mountp);
152	if (rel[0] == '/')
153		rel++;
154	(void) strcpy(relpath, rel);
155
156	return (0);
157}
158
159/*
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.
163 */
164/* ARGSUSED */
165static int
166object_from_path(const char *dataset, const char *path, struct stat64 *statbuf,
167    zinject_record_t *record)
168{
169	objset_t *os;
170	int err;
171
172	/*
173	 * Before doing any libzpool operations, call sync() to ensure that the
174	 * on-disk state is consistent with the in-core state.
175	 */
176	sync();
177
178	err = dmu_objset_own(dataset, DMU_OST_ZFS, B_TRUE, B_FALSE, FTAG, &os);
179	if (err != 0) {
180		(void) fprintf(stderr, "cannot open dataset '%s': %s\n",
181		    dataset, strerror(err));
182		return (-1);
183	}
184
185	record->zi_objset = dmu_objset_id(os);
186	record->zi_object = statbuf->st_ino;
187
188	dmu_objset_disown(os, B_FALSE, FTAG);
189
190	return (0);
191}
192
193/*
194 * Calculate the real range based on the type, level, and range given.
195 */
196static int
197calculate_range(const char *dataset, err_type_t type, int level, char *range,
198    zinject_record_t *record)
199{
200	objset_t *os = NULL;
201	dnode_t *dn = NULL;
202	int err;
203	int ret = -1;
204
205	/*
206	 * Determine the numeric range from the string.
207	 */
208	if (range == NULL) {
209		/*
210		 * If range is unspecified, set the range to [0,-1], which
211		 * indicates that the whole object should be treated as an
212		 * error.
213		 */
214		record->zi_start = 0;
215		record->zi_end = -1ULL;
216	} else {
217		char *end;
218
219		/* XXX add support for suffixes */
220		record->zi_start = strtoull(range, &end, 10);
221
222
223		if (*end == '\0')
224			record->zi_end = record->zi_start + 1;
225		else if (*end == ',')
226			record->zi_end = strtoull(end + 1, &end, 10);
227
228		if (*end != '\0') {
229			(void) fprintf(stderr, "invalid range '%s': must be "
230			    "a numeric range of the form 'start[,end]'\n",
231			    range);
232			goto out;
233		}
234	}
235
236	switch (type) {
237	case TYPE_DATA:
238		break;
239
240	case TYPE_DNODE:
241		/*
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.
246		 */
247		if (range != NULL) {
248			(void) fprintf(stderr, "range cannot be specified when "
249			    "type is 'dnode'\n");
250			goto out;
251		}
252
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;
256		break;
257	}
258
259	/*
260	 * Get the dnode associated with object, so we can calculate the block
261	 * size.
262	 */
263	if ((err = dmu_objset_own(dataset, DMU_OST_ANY,
264	    B_TRUE, B_FALSE, FTAG, &os)) != 0) {
265		(void) fprintf(stderr, "cannot open dataset '%s': %s\n",
266		    dataset, strerror(err));
267		goto out;
268	}
269
270	if (record->zi_object == 0) {
271		dn = DMU_META_DNODE(os);
272	} else {
273		err = dnode_hold(os, record->zi_object, FTAG, &dn);
274		if (err != 0) {
275			(void) fprintf(stderr, "failed to hold dnode "
276			    "for object %llu\n",
277			    (u_longlong_t)record->zi_object);
278			goto out;
279		}
280	}
281
282
283	ziprintf("data shift: %d\n", (int)dn->dn_datablkshift);
284	ziprintf(" ind shift: %d\n", (int)dn->dn_indblkshift);
285
286	/*
287	 * Translate range into block IDs.
288	 */
289	if (record->zi_start != 0 || record->zi_end != -1ULL) {
290		record->zi_start >>= dn->dn_datablkshift;
291		record->zi_end >>= dn->dn_datablkshift;
292	}
293
294	/*
295	 * Check level, and then translate level 0 blkids into ranges
296	 * appropriate for level of indirection.
297	 */
298	record->zi_level = level;
299	if (level > 0) {
300		ziprintf("level 0 blkid range: [%llu, %llu]\n",
301		    record->zi_start, record->zi_end);
302
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);
306			goto out;
307		}
308
309		if (record->zi_start != 0 || record->zi_end != 0) {
310			int shift = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
311
312			for (; level > 0; level--) {
313				record->zi_start >>= shift;
314				record->zi_end >>= shift;
315			}
316		}
317	}
318
319	ret = 0;
320out:
321	if (dn) {
322		if (dn != DMU_META_DNODE(os))
323			dnode_rele(dn, FTAG);
324	}
325	if (os)
326		dmu_objset_disown(os, B_FALSE, FTAG);
327
328	return (ret);
329}
330
331int
332translate_record(err_type_t type, const char *object, const char *range,
333    int level, zinject_record_t *record, char *poolname, char *dataset)
334{
335	char path[MAXPATHLEN];
336	char *slash;
337	struct stat64 statbuf;
338	int ret = -1;
339
340	kernel_init(FREAD);
341
342	debug = (getenv("ZINJECT_DEBUG") != NULL);
343
344	ziprintf("translating: %s\n", object);
345
346	if (MOS_TYPE(type)) {
347		/*
348		 * MOS objects are treated specially.
349		 */
350		switch (type) {
351		case TYPE_MOS:
352			record->zi_type = 0;
353			break;
354		case TYPE_MOSDIR:
355			record->zi_type = DMU_OT_OBJECT_DIRECTORY;
356			break;
357		case TYPE_METASLAB:
358			record->zi_type = DMU_OT_OBJECT_ARRAY;
359			break;
360		case TYPE_CONFIG:
361			record->zi_type = DMU_OT_PACKED_NVLIST;
362			break;
363		case TYPE_BPOBJ:
364			record->zi_type = DMU_OT_BPOBJ;
365			break;
366		case TYPE_SPACEMAP:
367			record->zi_type = DMU_OT_SPACE_MAP;
368			break;
369		case TYPE_ERRLOG:
370			record->zi_type = DMU_OT_ERROR_LOG;
371			break;
372		}
373
374		dataset[0] = '\0';
375		(void) strcpy(poolname, object);
376		return (0);
377	}
378
379	/*
380	 * Convert a full path into a (dataset, file) pair.
381	 */
382	if (parse_pathname(object, dataset, path, &statbuf) != 0)
383		goto err;
384
385	ziprintf("   dataset: %s\n", dataset);
386	ziprintf("      path: %s\n", path);
387
388	/*
389	 * Convert (dataset, file) into (objset, object)
390	 */
391	if (object_from_path(dataset, path, &statbuf, record) != 0)
392		goto err;
393
394	ziprintf("raw objset: %llu\n", record->zi_objset);
395	ziprintf("raw object: %llu\n", record->zi_object);
396
397	/*
398	 * For the given object, calculate the real (type, level, range)
399	 */
400	if (calculate_range(dataset, type, level, (char *)range, record) != 0)
401		goto err;
402
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");
408	else
409		ziprintf("     range: [%llu, %llu]\n", record->zi_start,
410		    record->zi_end);
411
412	/*
413	 * Copy the pool name
414	 */
415	(void) strcpy(poolname, dataset);
416	if ((slash = strchr(poolname, '/')) != NULL)
417		*slash = '\0';
418
419	ret = 0;
420
421err:
422	kernel_fini();
423	return (ret);
424}
425
426int
427translate_raw(const char *str, zinject_record_t *record)
428{
429	/*
430	 * A raw bookmark of the form objset:object:level:blkid, where each
431	 * number is a hexidecimal value.
432	 */
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);
438		return (-1);
439	}
440
441	record->zi_end = record->zi_start;
442
443	return (0);
444}
445
446int
447translate_device(const char *pool, const char *device, err_type_t label_type,
448    zinject_record_t *record)
449{
450	char *end;
451	zpool_handle_t *zhp;
452	nvlist_t *tgt;
453	boolean_t isspare, iscache;
454
455	/*
456	 * Given a device name or GUID, create an appropriate injection record
457	 * with zi_guid set.
458	 */
459	if ((zhp = zpool_open(g_zfs, pool)) == NULL)
460		return (-1);
461
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);
465
466		if (tgt == NULL) {
467			(void) fprintf(stderr, "cannot find device '%s' in "
468			    "pool '%s'\n", device, pool);
469			return (-1);
470		}
471
472		verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
473		    &record->zi_guid) == 0);
474	}
475
476	/*
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
481	 */
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;
486	} else {
487		record->zi_cmd = ZINJECT_DEVICE_FAULT;
488	}
489
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;
494		break;
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;
498		break;
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;
502		break;
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;
506		break;
507	}
508	return (0);
509}
510