xref: /illumos-gate/usr/src/cmd/mdb/common/modules/zfs/zfs.c (revision df3cd224ef765c29101e4110546062199562f757)
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 <mdb/mdb_ctf.h>
27 #include <sys/zfs_context.h>
28 #include <sys/mdb_modapi.h>
29 #include <sys/dbuf.h>
30 #include <sys/dmu_objset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_pool.h>
33 #include <sys/metaslab_impl.h>
34 #include <sys/space_map.h>
35 #include <sys/list.h>
36 #include <sys/spa_impl.h>
37 #include <sys/vdev_impl.h>
38 #include <sys/zap_leaf.h>
39 #include <sys/zap_impl.h>
40 #include <ctype.h>
41 
42 #ifndef _KERNEL
43 #include "../genunix/list.h"
44 #endif
45 
46 #ifdef _KERNEL
47 #define	ZFS_OBJ_NAME	"zfs"
48 #else
49 #define	ZFS_OBJ_NAME	"libzpool.so.1"
50 #endif
51 
52 static int
53 getmember(uintptr_t addr, const char *type, mdb_ctf_id_t *idp,
54     const char *member, int len, void *buf)
55 {
56 	mdb_ctf_id_t id;
57 	ulong_t off;
58 	char name[64];
59 
60 	if (idp == NULL) {
61 		if (mdb_ctf_lookup_by_name(type, &id) == -1) {
62 			mdb_warn("couldn't find type %s", type);
63 			return (DCMD_ERR);
64 		}
65 		idp = &id;
66 	} else {
67 		type = name;
68 		mdb_ctf_type_name(*idp, name, sizeof (name));
69 	}
70 
71 	if (mdb_ctf_offsetof(*idp, member, &off) == -1) {
72 		mdb_warn("couldn't find member %s of type %s\n", member, type);
73 		return (DCMD_ERR);
74 	}
75 	if (off % 8 != 0) {
76 		mdb_warn("member %s of type %s is unsupported bitfield",
77 		    member, type);
78 		return (DCMD_ERR);
79 	}
80 	off /= 8;
81 
82 	if (mdb_vread(buf, len, addr + off) == -1) {
83 		mdb_warn("failed to read %s from %s at %p",
84 		    member, type, addr + off);
85 		return (DCMD_ERR);
86 	}
87 	/* mdb_warn("read %s from %s at %p+%llx\n", member, type, addr, off); */
88 
89 	return (0);
90 }
91 
92 #define	GETMEMB(addr, type, member, dest) \
93 	getmember(addr, #type, NULL, #member, sizeof (dest), &(dest))
94 
95 #define	GETMEMBID(addr, ctfid, member, dest) \
96 	getmember(addr, NULL, ctfid, #member, sizeof (dest), &(dest))
97 
98 static int
99 getrefcount(uintptr_t addr, mdb_ctf_id_t *id,
100     const char *member, uint64_t *rc)
101 {
102 	static int gotid;
103 	static mdb_ctf_id_t rc_id;
104 	ulong_t off;
105 
106 	if (!gotid) {
107 		if (mdb_ctf_lookup_by_name("struct refcount", &rc_id) == -1) {
108 			mdb_warn("couldn't find struct refcount");
109 			return (DCMD_ERR);
110 		}
111 		gotid = TRUE;
112 	}
113 
114 	if (mdb_ctf_offsetof(*id, member, &off) == -1) {
115 		char name[64];
116 		mdb_ctf_type_name(*id, name, sizeof (name));
117 		mdb_warn("couldn't find member %s of type %s\n", member, name);
118 		return (DCMD_ERR);
119 	}
120 	off /= 8;
121 
122 	return (GETMEMBID(addr + off, &rc_id, rc_count, *rc));
123 }
124 
125 static int verbose;
126 
127 static int
128 freelist_walk_init(mdb_walk_state_t *wsp)
129 {
130 	if (wsp->walk_addr == NULL) {
131 		mdb_warn("must supply starting address\n");
132 		return (WALK_ERR);
133 	}
134 
135 	wsp->walk_data = 0;  /* Index into the freelist */
136 	return (WALK_NEXT);
137 }
138 
139 static int
140 freelist_walk_step(mdb_walk_state_t *wsp)
141 {
142 	uint64_t entry;
143 	uintptr_t number = (uintptr_t)wsp->walk_data;
144 	char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
145 			    "INVALID", "INVALID", "INVALID", "INVALID" };
146 	int mapshift = SPA_MINBLOCKSHIFT;
147 
148 	if (mdb_vread(&entry, sizeof (entry), wsp->walk_addr) == -1) {
149 		mdb_warn("failed to read freelist entry %p", wsp->walk_addr);
150 		return (WALK_DONE);
151 	}
152 	wsp->walk_addr += sizeof (entry);
153 	wsp->walk_data = (void *)(number + 1);
154 
155 	if (SM_DEBUG_DECODE(entry)) {
156 		mdb_printf("DEBUG: %3u  %10s: txg=%llu  pass=%llu\n",
157 		    number,
158 		    ddata[SM_DEBUG_ACTION_DECODE(entry)],
159 		    SM_DEBUG_TXG_DECODE(entry),
160 		    SM_DEBUG_SYNCPASS_DECODE(entry));
161 	} else {
162 		mdb_printf("Entry: %3u  offsets=%08llx-%08llx  type=%c  "
163 		    "size=%06llx", number,
164 		    SM_OFFSET_DECODE(entry) << mapshift,
165 		    (SM_OFFSET_DECODE(entry) + SM_RUN_DECODE(entry)) <<
166 		    mapshift,
167 		    SM_TYPE_DECODE(entry) == SM_ALLOC ? 'A' : 'F',
168 		    SM_RUN_DECODE(entry) << mapshift);
169 		if (verbose)
170 			mdb_printf("      (raw=%012llx)\n", entry);
171 		mdb_printf("\n");
172 	}
173 	return (WALK_NEXT);
174 }
175 
176 
177 static int
178 dataset_name(uintptr_t addr, char *buf)
179 {
180 	static int gotid;
181 	static mdb_ctf_id_t dd_id;
182 	uintptr_t dd_parent;
183 	char dd_myname[MAXNAMELEN];
184 
185 	if (!gotid) {
186 		if (mdb_ctf_lookup_by_name("struct dsl_dir",
187 		    &dd_id) == -1) {
188 			mdb_warn("couldn't find struct dsl_dir");
189 			return (DCMD_ERR);
190 		}
191 		gotid = TRUE;
192 	}
193 	if (GETMEMBID(addr, &dd_id, dd_parent, dd_parent) ||
194 	    GETMEMBID(addr, &dd_id, dd_myname, dd_myname)) {
195 		return (DCMD_ERR);
196 	}
197 
198 	if (dd_parent) {
199 		if (dataset_name(dd_parent, buf))
200 			return (DCMD_ERR);
201 		strcat(buf, "/");
202 	}
203 
204 	if (dd_myname[0])
205 		strcat(buf, dd_myname);
206 	else
207 		strcat(buf, "???");
208 
209 	return (0);
210 }
211 
212 static int
213 objset_name(uintptr_t addr, char *buf)
214 {
215 	static int gotid;
216 	static mdb_ctf_id_t os_id, ds_id;
217 	uintptr_t os_dsl_dataset;
218 	char ds_snapname[MAXNAMELEN];
219 	uintptr_t ds_dir;
220 
221 	buf[0] = '\0';
222 
223 	if (!gotid) {
224 		if (mdb_ctf_lookup_by_name("struct objset",
225 		    &os_id) == -1) {
226 			mdb_warn("couldn't find struct objset");
227 			return (DCMD_ERR);
228 		}
229 		if (mdb_ctf_lookup_by_name("struct dsl_dataset",
230 		    &ds_id) == -1) {
231 			mdb_warn("couldn't find struct dsl_dataset");
232 			return (DCMD_ERR);
233 		}
234 
235 		gotid = TRUE;
236 	}
237 
238 	if (GETMEMBID(addr, &os_id, os_dsl_dataset, os_dsl_dataset))
239 		return (DCMD_ERR);
240 
241 	if (os_dsl_dataset == 0) {
242 		strcat(buf, "mos");
243 		return (0);
244 	}
245 
246 	if (GETMEMBID(os_dsl_dataset, &ds_id, ds_snapname, ds_snapname) ||
247 	    GETMEMBID(os_dsl_dataset, &ds_id, ds_dir, ds_dir)) {
248 		return (DCMD_ERR);
249 	}
250 
251 	if (ds_dir && dataset_name(ds_dir, buf))
252 		return (DCMD_ERR);
253 
254 	if (ds_snapname[0]) {
255 		strcat(buf, "@");
256 		strcat(buf, ds_snapname);
257 	}
258 	return (0);
259 }
260 
261 static void
262 enum_lookup(char *out, size_t size, mdb_ctf_id_t id, int val,
263     const char *prefix)
264 {
265 	const char *cp;
266 	size_t len = strlen(prefix);
267 
268 	if ((cp = mdb_ctf_enum_name(id, val)) != NULL) {
269 		if (strncmp(cp, prefix, len) == 0)
270 			cp += len;
271 		(void) strncpy(out, cp, size);
272 	} else {
273 		mdb_snprintf(out, size, "? (%d)", val);
274 	}
275 }
276 
277 /* ARGSUSED */
278 static int
279 zfs_params(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
280 {
281 	/*
282 	 * This table can be approximately generated by running:
283 	 * egrep "^[a-z0-9_]+ [a-z0-9_]+( =.*)?;" *.c | cut -d ' ' -f 2
284 	 */
285 	static const char *params[] = {
286 		"arc_reduce_dnlc_percent",
287 		"zfs_arc_max",
288 		"zfs_arc_min",
289 		"arc_shrink_shift",
290 		"zfs_mdcomp_disable",
291 		"zfs_prefetch_disable",
292 		"zfetch_max_streams",
293 		"zfetch_min_sec_reap",
294 		"zfetch_block_cap",
295 		"zfetch_array_rd_sz",
296 		"zfs_default_bs",
297 		"zfs_default_ibs",
298 		"metaslab_aliquot",
299 		"reference_tracking_enable",
300 		"reference_history",
301 		"spa_max_replication_override",
302 		"spa_mode_global",
303 		"zfs_flags",
304 		"zfs_txg_synctime",
305 		"zfs_txg_timeout",
306 		"zfs_write_limit_min",
307 		"zfs_write_limit_max",
308 		"zfs_write_limit_shift",
309 		"zfs_write_limit_override",
310 		"zfs_no_write_throttle",
311 		"zfs_vdev_cache_max",
312 		"zfs_vdev_cache_size",
313 		"zfs_vdev_cache_bshift",
314 		"vdev_mirror_shift",
315 		"zfs_vdev_max_pending",
316 		"zfs_vdev_min_pending",
317 		"zfs_scrub_limit",
318 		"zfs_no_scrub_io",
319 		"zfs_no_scrub_prefetch",
320 		"zfs_vdev_time_shift",
321 		"zfs_vdev_ramp_rate",
322 		"zfs_vdev_aggregation_limit",
323 		"fzap_default_block_shift",
324 		"zfs_immediate_write_sz",
325 		"zfs_read_chunk_size",
326 		"zil_disable",
327 		"zfs_nocacheflush",
328 		"metaslab_gang_bang",
329 		"metaslab_df_alloc_threshold",
330 		"metaslab_df_free_pct",
331 		"zio_injection_enabled",
332 		"zvol_immediate_write_sz",
333 	};
334 
335 	for (int i = 0; i < sizeof (params) / sizeof (params[0]); i++) {
336 		int sz;
337 		uint64_t val64;
338 		uint32_t *val32p = (uint32_t *)&val64;
339 
340 		sz = mdb_readvar(&val64, params[i]);
341 		if (sz == 4) {
342 			mdb_printf("%s = 0x%x\n", params[i], *val32p);
343 		} else if (sz == 8) {
344 			mdb_printf("%s = 0x%llx\n", params[i], val64);
345 		} else {
346 			mdb_warn("variable %s not found", params[i]);
347 		}
348 	}
349 
350 	return (DCMD_OK);
351 }
352 
353 /* ARGSUSED */
354 static int
355 blkptr(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
356 {
357 	mdb_ctf_id_t type_enum, checksum_enum, compress_enum;
358 	char type[80], checksum[80], compress[80];
359 	blkptr_t blk, *bp = &blk;
360 	char buf[BP_SPRINTF_LEN];
361 
362 	if (mdb_vread(&blk, sizeof (blkptr_t), addr) == -1) {
363 		mdb_warn("failed to read blkptr_t");
364 		return (DCMD_ERR);
365 	}
366 
367 	if (mdb_ctf_lookup_by_name("enum dmu_object_type", &type_enum) == -1 ||
368 	    mdb_ctf_lookup_by_name("enum zio_checksum", &checksum_enum) == -1 ||
369 	    mdb_ctf_lookup_by_name("enum zio_compress", &compress_enum) == -1) {
370 		mdb_warn("Could not find blkptr enumerated types");
371 		return (DCMD_ERR);
372 	}
373 
374 	enum_lookup(type, sizeof (type), type_enum,
375 	    BP_GET_TYPE(bp), "DMU_OT_");
376 	enum_lookup(checksum, sizeof (checksum), checksum_enum,
377 	    BP_GET_CHECKSUM(bp), "ZIO_CHECKSUM_");
378 	enum_lookup(compress, sizeof (compress), compress_enum,
379 	    BP_GET_COMPRESS(bp), "ZIO_COMPRESS_");
380 
381 	SPRINTF_BLKPTR(mdb_snprintf, '\n', buf, bp, type, checksum, compress);
382 
383 	mdb_printf("%s\n", buf);
384 
385 	return (DCMD_OK);
386 }
387 
388 /* ARGSUSED */
389 static int
390 dbuf(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
391 {
392 	mdb_ctf_id_t id;
393 	dmu_buf_t db;
394 	uintptr_t objset;
395 	uint8_t level;
396 	uint64_t blkid;
397 	uint64_t holds;
398 	char objectname[32];
399 	char blkidname[32];
400 	char path[MAXNAMELEN];
401 
402 	if (DCMD_HDRSPEC(flags)) {
403 		mdb_printf("        addr object lvl blkid holds os\n");
404 	}
405 
406 	if (mdb_ctf_lookup_by_name("struct dmu_buf_impl", &id) == -1) {
407 		mdb_warn("couldn't find struct dmu_buf_impl_t");
408 		return (DCMD_ERR);
409 	}
410 
411 	if (GETMEMBID(addr, &id, db_objset, objset) ||
412 	    GETMEMBID(addr, &id, db, db) ||
413 	    GETMEMBID(addr, &id, db_level, level) ||
414 	    GETMEMBID(addr, &id, db_blkid, blkid)) {
415 		return (WALK_ERR);
416 	}
417 
418 	if (getrefcount(addr, &id, "db_holds", &holds)) {
419 		return (WALK_ERR);
420 	}
421 
422 	if (db.db_object == DMU_META_DNODE_OBJECT)
423 		(void) strcpy(objectname, "mdn");
424 	else
425 		(void) mdb_snprintf(objectname, sizeof (objectname), "%llx",
426 		    (u_longlong_t)db.db_object);
427 
428 	if (blkid == DB_BONUS_BLKID)
429 		(void) strcpy(blkidname, "bonus");
430 	else
431 		(void) mdb_snprintf(blkidname, sizeof (blkidname), "%llx",
432 		    (u_longlong_t)blkid);
433 
434 	if (objset_name(objset, path)) {
435 		return (WALK_ERR);
436 	}
437 
438 	mdb_printf("%p %8s %1u %9s %2llu %s\n",
439 	    addr, objectname, level, blkidname, holds, path);
440 
441 	return (DCMD_OK);
442 }
443 
444 /* ARGSUSED */
445 static int
446 dbuf_stats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
447 {
448 #define	HISTOSZ 32
449 	uintptr_t dbp;
450 	dmu_buf_impl_t db;
451 	dbuf_hash_table_t ht;
452 	uint64_t bucket, ndbufs;
453 	uint64_t histo[HISTOSZ];
454 	uint64_t histo2[HISTOSZ];
455 	int i, maxidx;
456 
457 	if (mdb_readvar(&ht, "dbuf_hash_table") == -1) {
458 		mdb_warn("failed to read 'dbuf_hash_table'");
459 		return (DCMD_ERR);
460 	}
461 
462 	for (i = 0; i < HISTOSZ; i++) {
463 		histo[i] = 0;
464 		histo2[i] = 0;
465 	}
466 
467 	ndbufs = 0;
468 	for (bucket = 0; bucket < ht.hash_table_mask+1; bucket++) {
469 		int len;
470 
471 		if (mdb_vread(&dbp, sizeof (void *),
472 		    (uintptr_t)(ht.hash_table+bucket)) == -1) {
473 			mdb_warn("failed to read hash bucket %u at %p",
474 			    bucket, ht.hash_table+bucket);
475 			return (DCMD_ERR);
476 		}
477 
478 		len = 0;
479 		while (dbp != 0) {
480 			if (mdb_vread(&db, sizeof (dmu_buf_impl_t),
481 			    dbp) == -1) {
482 				mdb_warn("failed to read dbuf at %p", dbp);
483 				return (DCMD_ERR);
484 			}
485 			dbp = (uintptr_t)db.db_hash_next;
486 			for (i = MIN(len, HISTOSZ - 1); i >= 0; i--)
487 				histo2[i]++;
488 			len++;
489 			ndbufs++;
490 		}
491 
492 		if (len >= HISTOSZ)
493 			len = HISTOSZ-1;
494 		histo[len]++;
495 	}
496 
497 	mdb_printf("hash table has %llu buckets, %llu dbufs "
498 	    "(avg %llu buckets/dbuf)\n",
499 	    ht.hash_table_mask+1, ndbufs,
500 	    (ht.hash_table_mask+1)/ndbufs);
501 
502 	mdb_printf("\n");
503 	maxidx = 0;
504 	for (i = 0; i < HISTOSZ; i++)
505 		if (histo[i] > 0)
506 			maxidx = i;
507 	mdb_printf("hash chain length	number of buckets\n");
508 	for (i = 0; i <= maxidx; i++)
509 		mdb_printf("%u			%llu\n", i, histo[i]);
510 
511 	mdb_printf("\n");
512 	maxidx = 0;
513 	for (i = 0; i < HISTOSZ; i++)
514 		if (histo2[i] > 0)
515 			maxidx = i;
516 	mdb_printf("hash chain depth	number of dbufs\n");
517 	for (i = 0; i <= maxidx; i++)
518 		mdb_printf("%u or more		%llu	%llu%%\n",
519 		    i, histo2[i], histo2[i]*100/ndbufs);
520 
521 
522 	return (DCMD_OK);
523 }
524 
525 #define	CHAIN_END 0xffff
526 /*
527  * ::zap_leaf [-v]
528  *
529  * Print a zap_leaf_phys_t, assumed to be 16k
530  */
531 /* ARGSUSED */
532 static int
533 zap_leaf(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
534 {
535 	char buf[16*1024];
536 	int verbose = B_FALSE;
537 	int four = B_FALSE;
538 	zap_leaf_t l;
539 	zap_leaf_phys_t *zlp = (void *)buf;
540 	int i;
541 
542 	if (mdb_getopts(argc, argv,
543 	    'v', MDB_OPT_SETBITS, TRUE, &verbose,
544 	    '4', MDB_OPT_SETBITS, TRUE, &four,
545 	    NULL) != argc)
546 		return (DCMD_USAGE);
547 
548 	l.l_phys = zlp;
549 	l.l_bs = 14; /* assume 16k blocks */
550 	if (four)
551 		l.l_bs = 12;
552 
553 	if (!(flags & DCMD_ADDRSPEC)) {
554 		return (DCMD_USAGE);
555 	}
556 
557 	if (mdb_vread(buf, sizeof (buf), addr) == -1) {
558 		mdb_warn("failed to read zap_leaf_phys_t at %p", addr);
559 		return (DCMD_ERR);
560 	}
561 
562 	if (zlp->l_hdr.lh_block_type != ZBT_LEAF ||
563 	    zlp->l_hdr.lh_magic != ZAP_LEAF_MAGIC) {
564 		mdb_warn("This does not appear to be a zap_leaf_phys_t");
565 		return (DCMD_ERR);
566 	}
567 
568 	mdb_printf("zap_leaf_phys_t at %p:\n", addr);
569 	mdb_printf("    lh_prefix_len = %u\n", zlp->l_hdr.lh_prefix_len);
570 	mdb_printf("    lh_prefix = %llx\n", zlp->l_hdr.lh_prefix);
571 	mdb_printf("    lh_nentries = %u\n", zlp->l_hdr.lh_nentries);
572 	mdb_printf("    lh_nfree = %u\n", zlp->l_hdr.lh_nfree,
573 	    zlp->l_hdr.lh_nfree * 100 / (ZAP_LEAF_NUMCHUNKS(&l)));
574 	mdb_printf("    lh_freelist = %u\n", zlp->l_hdr.lh_freelist);
575 	mdb_printf("    lh_flags = %x (%s)\n", zlp->l_hdr.lh_flags,
576 	    zlp->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED ?
577 	    "ENTRIES_CDSORTED" : "");
578 
579 	if (verbose) {
580 		mdb_printf(" hash table:\n");
581 		for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++) {
582 			if (zlp->l_hash[i] != CHAIN_END)
583 				mdb_printf("    %u: %u\n", i, zlp->l_hash[i]);
584 		}
585 	}
586 
587 	mdb_printf(" chunks:\n");
588 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) {
589 		/* LINTED: alignment */
590 		zap_leaf_chunk_t *zlc = &ZAP_LEAF_CHUNK(&l, i);
591 		switch (zlc->l_entry.le_type) {
592 		case ZAP_CHUNK_FREE:
593 			if (verbose) {
594 				mdb_printf("    %u: free; lf_next = %u\n",
595 				    i, zlc->l_free.lf_next);
596 			}
597 			break;
598 		case ZAP_CHUNK_ENTRY:
599 			mdb_printf("    %u: entry\n", i);
600 			if (verbose) {
601 				mdb_printf("        le_next = %u\n",
602 				    zlc->l_entry.le_next);
603 			}
604 			mdb_printf("        le_name_chunk = %u\n",
605 			    zlc->l_entry.le_name_chunk);
606 			mdb_printf("        le_name_numints = %u\n",
607 			    zlc->l_entry.le_name_numints);
608 			mdb_printf("        le_value_chunk = %u\n",
609 			    zlc->l_entry.le_value_chunk);
610 			mdb_printf("        le_value_intlen = %u\n",
611 			    zlc->l_entry.le_value_intlen);
612 			mdb_printf("        le_value_numints = %u\n",
613 			    zlc->l_entry.le_value_numints);
614 			mdb_printf("        le_cd = %u\n",
615 			    zlc->l_entry.le_cd);
616 			mdb_printf("        le_hash = %llx\n",
617 			    zlc->l_entry.le_hash);
618 			break;
619 		case ZAP_CHUNK_ARRAY:
620 			mdb_printf("    %u: array \"%s\"\n",
621 			    i, zlc->l_array.la_array);
622 			if (verbose) {
623 				int j;
624 				mdb_printf("        ");
625 				for (j = 0; j < ZAP_LEAF_ARRAY_BYTES; j++) {
626 					mdb_printf("%02x ",
627 					    zlc->l_array.la_array[j]);
628 				}
629 				mdb_printf("\n");
630 			}
631 			if (zlc->l_array.la_next != CHAIN_END) {
632 				mdb_printf("        lf_next = %u\n",
633 				    zlc->l_array.la_next);
634 			}
635 			break;
636 		default:
637 			mdb_printf("    %u: undefined type %u\n",
638 			    zlc->l_entry.le_type);
639 		}
640 	}
641 
642 	return (DCMD_OK);
643 }
644 
645 typedef struct dbufs_data {
646 	mdb_ctf_id_t id;
647 	uint64_t objset;
648 	uint64_t object;
649 	uint64_t level;
650 	uint64_t blkid;
651 	char *osname;
652 } dbufs_data_t;
653 
654 #define	DBUFS_UNSET	(0xbaddcafedeadbeefULL)
655 
656 /* ARGSUSED */
657 static int
658 dbufs_cb(uintptr_t addr, const void *unknown, void *arg)
659 {
660 	dbufs_data_t *data = arg;
661 	uintptr_t objset;
662 	dmu_buf_t db;
663 	uint8_t level;
664 	uint64_t blkid;
665 	char osname[MAXNAMELEN];
666 
667 	if (GETMEMBID(addr, &data->id, db_objset, objset) ||
668 	    GETMEMBID(addr, &data->id, db, db) ||
669 	    GETMEMBID(addr, &data->id, db_level, level) ||
670 	    GETMEMBID(addr, &data->id, db_blkid, blkid)) {
671 		return (WALK_ERR);
672 	}
673 
674 	if ((data->objset == DBUFS_UNSET || data->objset == objset) &&
675 	    (data->osname == NULL || (objset_name(objset, osname) == 0 &&
676 	    strcmp(data->osname, osname) == 0)) &&
677 	    (data->object == DBUFS_UNSET || data->object == db.db_object) &&
678 	    (data->level == DBUFS_UNSET || data->level == level) &&
679 	    (data->blkid == DBUFS_UNSET || data->blkid == blkid)) {
680 		mdb_printf("%#lr\n", addr);
681 	}
682 	return (WALK_NEXT);
683 }
684 
685 /* ARGSUSED */
686 static int
687 dbufs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
688 {
689 	dbufs_data_t data;
690 	char *object = NULL;
691 	char *blkid = NULL;
692 
693 	data.objset = data.object = data.level = data.blkid = DBUFS_UNSET;
694 	data.osname = NULL;
695 
696 	if (mdb_getopts(argc, argv,
697 	    'O', MDB_OPT_UINT64, &data.objset,
698 	    'n', MDB_OPT_STR, &data.osname,
699 	    'o', MDB_OPT_STR, &object,
700 	    'l', MDB_OPT_UINT64, &data.level,
701 	    'b', MDB_OPT_STR, &blkid) != argc) {
702 		return (DCMD_USAGE);
703 	}
704 
705 	if (object) {
706 		if (strcmp(object, "mdn") == 0) {
707 			data.object = DMU_META_DNODE_OBJECT;
708 		} else {
709 			data.object = mdb_strtoull(object);
710 		}
711 	}
712 
713 	if (blkid) {
714 		if (strcmp(blkid, "bonus") == 0) {
715 			data.blkid = DB_BONUS_BLKID;
716 		} else {
717 			data.blkid = mdb_strtoull(blkid);
718 		}
719 	}
720 
721 	if (mdb_ctf_lookup_by_name("struct dmu_buf_impl", &data.id) == -1) {
722 		mdb_warn("couldn't find struct dmu_buf_impl_t");
723 		return (DCMD_ERR);
724 	}
725 
726 	if (mdb_walk("dmu_buf_impl_t", dbufs_cb, &data) != 0) {
727 		mdb_warn("can't walk dbufs");
728 		return (DCMD_ERR);
729 	}
730 
731 	return (DCMD_OK);
732 }
733 
734 typedef struct abuf_find_data {
735 	dva_t dva;
736 	mdb_ctf_id_t id;
737 } abuf_find_data_t;
738 
739 /* ARGSUSED */
740 static int
741 abuf_find_cb(uintptr_t addr, const void *unknown, void *arg)
742 {
743 	abuf_find_data_t *data = arg;
744 	dva_t dva;
745 
746 	if (GETMEMBID(addr, &data->id, b_dva, dva)) {
747 		return (WALK_ERR);
748 	}
749 
750 	if (dva.dva_word[0] == data->dva.dva_word[0] &&
751 	    dva.dva_word[1] == data->dva.dva_word[1]) {
752 		mdb_printf("%#lr\n", addr);
753 	}
754 	return (WALK_NEXT);
755 }
756 
757 /* ARGSUSED */
758 static int
759 abuf_find(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
760 {
761 	abuf_find_data_t data;
762 	GElf_Sym sym;
763 	int i;
764 	const char *syms[] = {
765 		"ARC_mru",
766 		"ARC_mru_ghost",
767 		"ARC_mfu",
768 		"ARC_mfu_ghost",
769 	};
770 
771 	if (argc != 2)
772 		return (DCMD_USAGE);
773 
774 	for (i = 0; i < 2; i ++) {
775 		switch (argv[i].a_type) {
776 		case MDB_TYPE_STRING:
777 			data.dva.dva_word[i] = mdb_strtoull(argv[i].a_un.a_str);
778 			break;
779 		case MDB_TYPE_IMMEDIATE:
780 			data.dva.dva_word[i] = argv[i].a_un.a_val;
781 			break;
782 		default:
783 			return (DCMD_USAGE);
784 		}
785 	}
786 
787 	if (mdb_ctf_lookup_by_name("struct arc_buf_hdr", &data.id) == -1) {
788 		mdb_warn("couldn't find struct arc_buf_hdr");
789 		return (DCMD_ERR);
790 	}
791 
792 	for (i = 0; i < sizeof (syms) / sizeof (syms[0]); i++) {
793 		if (mdb_lookup_by_name(syms[i], &sym)) {
794 			mdb_warn("can't find symbol %s", syms[i]);
795 			return (DCMD_ERR);
796 		}
797 
798 		if (mdb_pwalk("list", abuf_find_cb, &data, sym.st_value) != 0) {
799 			mdb_warn("can't walk %s", syms[i]);
800 			return (DCMD_ERR);
801 		}
802 	}
803 
804 	return (DCMD_OK);
805 }
806 
807 /*ARGSUSED*/
808 static int
809 arc_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
810 {
811 	kstat_named_t *stats;
812 	GElf_Sym sym;
813 	int nstats, i;
814 	uint_t opt_a = FALSE;
815 	uint_t opt_b = FALSE;
816 	uint_t shift = 0;
817 	const char *suffix;
818 
819 	static const char *bytestats[] = {
820 		"p", "c", "c_min", "c_max", "size", NULL
821 	};
822 
823 	static const char *extras[] = {
824 		"arc_no_grow", "arc_tempreserve",
825 		"arc_meta_used", "arc_meta_limit", "arc_meta_max",
826 		NULL
827 	};
828 
829 	if (mdb_lookup_by_name("arc_stats", &sym) == -1) {
830 		mdb_warn("failed to find 'arc_stats'");
831 		return (DCMD_ERR);
832 	}
833 
834 	stats = mdb_zalloc(sym.st_size, UM_SLEEP | UM_GC);
835 
836 	if (mdb_vread(stats, sym.st_size, sym.st_value) == -1) {
837 		mdb_warn("couldn't read 'arc_stats' at %p", sym.st_value);
838 		return (DCMD_ERR);
839 	}
840 
841 	nstats = sym.st_size / sizeof (kstat_named_t);
842 
843 	/* NB: -a / opt_a are ignored for backwards compatability */
844 	if (mdb_getopts(argc, argv,
845 	    'a', MDB_OPT_SETBITS, TRUE, &opt_a,
846 	    'b', MDB_OPT_SETBITS, TRUE, &opt_b,
847 	    'k', MDB_OPT_SETBITS, 10, &shift,
848 	    'm', MDB_OPT_SETBITS, 20, &shift,
849 	    'g', MDB_OPT_SETBITS, 30, &shift,
850 	    NULL) != argc)
851 		return (DCMD_USAGE);
852 
853 	if (!opt_b && !shift)
854 		shift = 20;
855 
856 	switch (shift) {
857 	case 0:
858 		suffix = "B";
859 		break;
860 	case 10:
861 		suffix = "KB";
862 		break;
863 	case 20:
864 		suffix = "MB";
865 		break;
866 	case 30:
867 		suffix = "GB";
868 		break;
869 	default:
870 		suffix = "XX";
871 	}
872 
873 	for (i = 0; i < nstats; i++) {
874 		int j;
875 		boolean_t bytes = B_FALSE;
876 
877 		for (j = 0; bytestats[j]; j++) {
878 			if (strcmp(stats[i].name, bytestats[j]) == 0) {
879 				bytes = B_TRUE;
880 				break;
881 			}
882 		}
883 
884 		if (bytes) {
885 			mdb_printf("%-25s = %9llu %s\n", stats[i].name,
886 			    stats[i].value.ui64 >> shift, suffix);
887 		} else {
888 			mdb_printf("%-25s = %9llu\n", stats[i].name,
889 			    stats[i].value.ui64);
890 		}
891 	}
892 
893 	for (i = 0; extras[i]; i++) {
894 		uint64_t buf;
895 
896 		if (mdb_lookup_by_name(extras[i], &sym) == -1) {
897 			mdb_warn("failed to find '%s'", extras[i]);
898 			return (DCMD_ERR);
899 		}
900 
901 		if (sym.st_size != sizeof (uint64_t) &&
902 		    sym.st_size != sizeof (uint32_t)) {
903 			mdb_warn("expected scalar for variable '%s'\n",
904 			    extras[i]);
905 			return (DCMD_ERR);
906 		}
907 
908 		if (mdb_vread(&buf, sym.st_size, sym.st_value) == -1) {
909 			mdb_warn("couldn't read '%s'", extras[i]);
910 			return (DCMD_ERR);
911 		}
912 
913 		mdb_printf("%-25s = ", extras[i]);
914 
915 		/* NB: all the 64-bit extras happen to be byte counts */
916 		if (sym.st_size == sizeof (uint64_t))
917 			mdb_printf("%9llu %s\n", buf >> shift, suffix);
918 
919 		if (sym.st_size == sizeof (uint32_t))
920 			mdb_printf("%9d\n", *((uint32_t *)&buf));
921 	}
922 	return (DCMD_OK);
923 }
924 
925 /*
926  * ::spa
927  *
928  * 	-c	Print configuration information as well
929  * 	-v	Print vdev state
930  * 	-e	Print vdev error stats
931  *
932  * Print a summarized spa_t.  When given no arguments, prints out a table of all
933  * active pools on the system.
934  */
935 /* ARGSUSED */
936 static int
937 spa_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
938 {
939 	spa_t spa;
940 	const char *statetab[] = { "ACTIVE", "EXPORTED", "DESTROYED",
941 		"SPARE", "L2CACHE", "UNINIT", "UNAVAIL", "POTENTIAL" };
942 	const char *state;
943 	int config = FALSE;
944 	int vdevs = FALSE;
945 	int errors = FALSE;
946 
947 	if (mdb_getopts(argc, argv,
948 	    'c', MDB_OPT_SETBITS, TRUE, &config,
949 	    'v', MDB_OPT_SETBITS, TRUE, &vdevs,
950 	    'e', MDB_OPT_SETBITS, TRUE, &errors,
951 	    NULL) != argc)
952 		return (DCMD_USAGE);
953 
954 	if (!(flags & DCMD_ADDRSPEC)) {
955 		if (mdb_walk_dcmd("spa", "spa", argc, argv) == -1) {
956 			mdb_warn("can't walk spa");
957 			return (DCMD_ERR);
958 		}
959 
960 		return (DCMD_OK);
961 	}
962 
963 	if (flags & DCMD_PIPE_OUT) {
964 		mdb_printf("%#lr\n", addr);
965 		return (DCMD_OK);
966 	}
967 
968 	if (DCMD_HDRSPEC(flags))
969 		mdb_printf("%<u>%-?s %9s %-*s%</u>\n", "ADDR", "STATE",
970 		    sizeof (uintptr_t) == 4 ? 60 : 52, "NAME");
971 
972 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
973 		mdb_warn("failed to read spa_t at %p", addr);
974 		return (DCMD_ERR);
975 	}
976 
977 	if (spa.spa_state < 0 || spa.spa_state > POOL_STATE_UNAVAIL)
978 		state = "UNKNOWN";
979 	else
980 		state = statetab[spa.spa_state];
981 
982 	mdb_printf("%0?p %9s %s\n", addr, state, spa.spa_name);
983 
984 	if (config) {
985 		mdb_printf("\n");
986 		mdb_inc_indent(4);
987 		if (mdb_call_dcmd("spa_config", addr, flags, 0,
988 		    NULL) != DCMD_OK)
989 			return (DCMD_ERR);
990 		mdb_dec_indent(4);
991 	}
992 
993 	if (vdevs || errors) {
994 		mdb_arg_t v;
995 
996 		v.a_type = MDB_TYPE_STRING;
997 		v.a_un.a_str = "-e";
998 
999 		mdb_printf("\n");
1000 		mdb_inc_indent(4);
1001 		if (mdb_call_dcmd("spa_vdevs", addr, flags, errors ? 1 : 0,
1002 		    &v) != DCMD_OK)
1003 			return (DCMD_ERR);
1004 		mdb_dec_indent(4);
1005 	}
1006 
1007 	return (DCMD_OK);
1008 }
1009 
1010 /*
1011  * ::spa_config
1012  *
1013  * Given a spa_t, print the configuration information stored in spa_config.
1014  * Since it's just an nvlist, format it as an indented list of name=value pairs.
1015  * We simply read the value of spa_config and pass off to ::nvlist.
1016  */
1017 /* ARGSUSED */
1018 static int
1019 spa_print_config(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1020 {
1021 	spa_t spa;
1022 
1023 	if (argc != 0 || !(flags & DCMD_ADDRSPEC))
1024 		return (DCMD_USAGE);
1025 
1026 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1027 		mdb_warn("failed to read spa_t at %p", addr);
1028 		return (DCMD_ERR);
1029 	}
1030 
1031 	if (spa.spa_config == NULL) {
1032 		mdb_printf("(none)\n");
1033 		return (DCMD_OK);
1034 	}
1035 
1036 	return (mdb_call_dcmd("nvlist", (uintptr_t)spa.spa_config, flags,
1037 	    0, NULL));
1038 }
1039 
1040 /*
1041  * ::vdev
1042  *
1043  * Print out a summarized vdev_t, in the following form:
1044  *
1045  * ADDR             STATE	AUX            DESC
1046  * fffffffbcde23df0 HEALTHY	-              /dev/dsk/c0t0d0
1047  *
1048  * If '-r' is specified, recursively visit all children.
1049  *
1050  * With '-e', the statistics associated with the vdev are printed as well.
1051  */
1052 static int
1053 do_print_vdev(uintptr_t addr, int flags, int depth, int stats,
1054     int recursive)
1055 {
1056 	vdev_t vdev;
1057 	char desc[MAXNAMELEN];
1058 	int c, children;
1059 	uintptr_t *child;
1060 	const char *state, *aux;
1061 
1062 	if (mdb_vread(&vdev, sizeof (vdev), (uintptr_t)addr) == -1) {
1063 		mdb_warn("failed to read vdev_t at %p\n", (uintptr_t)addr);
1064 		return (DCMD_ERR);
1065 	}
1066 
1067 	if (flags & DCMD_PIPE_OUT) {
1068 		mdb_printf("%#lr", addr);
1069 	} else {
1070 		if (vdev.vdev_path != NULL) {
1071 			if (mdb_readstr(desc, sizeof (desc),
1072 			    (uintptr_t)vdev.vdev_path) == -1) {
1073 				mdb_warn("failed to read vdev_path at %p\n",
1074 				    vdev.vdev_path);
1075 				return (DCMD_ERR);
1076 			}
1077 		} else if (vdev.vdev_ops != NULL) {
1078 			vdev_ops_t ops;
1079 			if (mdb_vread(&ops, sizeof (ops),
1080 			    (uintptr_t)vdev.vdev_ops) == -1) {
1081 				mdb_warn("failed to read vdev_ops at %p\n",
1082 				    vdev.vdev_ops);
1083 				return (DCMD_ERR);
1084 			}
1085 			(void) strcpy(desc, ops.vdev_op_type);
1086 		} else {
1087 			(void) strcpy(desc, "<unknown>");
1088 		}
1089 
1090 		if (depth == 0 && DCMD_HDRSPEC(flags))
1091 			mdb_printf("%<u>%-?s %-9s %-12s %-*s%</u>\n",
1092 			    "ADDR", "STATE", "AUX",
1093 			    sizeof (uintptr_t) == 4 ? 43 : 35,
1094 			    "DESCRIPTION");
1095 
1096 		mdb_printf("%0?p ", addr);
1097 
1098 		switch (vdev.vdev_state) {
1099 		case VDEV_STATE_CLOSED:
1100 			state = "CLOSED";
1101 			break;
1102 		case VDEV_STATE_OFFLINE:
1103 			state = "OFFLINE";
1104 			break;
1105 		case VDEV_STATE_CANT_OPEN:
1106 			state = "CANT_OPEN";
1107 			break;
1108 		case VDEV_STATE_DEGRADED:
1109 			state = "DEGRADED";
1110 			break;
1111 		case VDEV_STATE_HEALTHY:
1112 			state = "HEALTHY";
1113 			break;
1114 		case VDEV_STATE_REMOVED:
1115 			state = "REMOVED";
1116 			break;
1117 		case VDEV_STATE_FAULTED:
1118 			state = "FAULTED";
1119 			break;
1120 		default:
1121 			state = "UNKNOWN";
1122 			break;
1123 		}
1124 
1125 		switch (vdev.vdev_stat.vs_aux) {
1126 		case VDEV_AUX_NONE:
1127 			aux = "-";
1128 			break;
1129 		case VDEV_AUX_OPEN_FAILED:
1130 			aux = "OPEN_FAILED";
1131 			break;
1132 		case VDEV_AUX_CORRUPT_DATA:
1133 			aux = "CORRUPT_DATA";
1134 			break;
1135 		case VDEV_AUX_NO_REPLICAS:
1136 			aux = "NO_REPLICAS";
1137 			break;
1138 		case VDEV_AUX_BAD_GUID_SUM:
1139 			aux = "BAD_GUID_SUM";
1140 			break;
1141 		case VDEV_AUX_TOO_SMALL:
1142 			aux = "TOO_SMALL";
1143 			break;
1144 		case VDEV_AUX_BAD_LABEL:
1145 			aux = "BAD_LABEL";
1146 			break;
1147 		case VDEV_AUX_VERSION_NEWER:
1148 			aux = "VERS_NEWER";
1149 			break;
1150 		case VDEV_AUX_VERSION_OLDER:
1151 			aux = "VERS_OLDER";
1152 			break;
1153 		case VDEV_AUX_SPARED:
1154 			aux = "SPARED";
1155 			break;
1156 		case VDEV_AUX_ERR_EXCEEDED:
1157 			aux = "ERR_EXCEEDED";
1158 			break;
1159 		case VDEV_AUX_IO_FAILURE:
1160 			aux = "IO_FAILURE";
1161 			break;
1162 		case VDEV_AUX_BAD_LOG:
1163 			aux = "BAD_LOG";
1164 			break;
1165 		case VDEV_AUX_EXTERNAL:
1166 			aux = "EXTERNAL";
1167 			break;
1168 		case VDEV_AUX_SPLIT_POOL:
1169 			aux = "SPLIT_POOL";
1170 			break;
1171 		default:
1172 			aux = "UNKNOWN";
1173 			break;
1174 		}
1175 
1176 		mdb_printf("%-9s %-12s %*s%s\n", state, aux, depth, "", desc);
1177 
1178 		if (stats) {
1179 			vdev_stat_t *vs = &vdev.vdev_stat;
1180 			int i;
1181 
1182 			mdb_inc_indent(4);
1183 			mdb_printf("\n");
1184 			mdb_printf("%<u>       %12s %12s %12s %12s "
1185 			    "%12s%</u>\n", "READ", "WRITE", "FREE", "CLAIM",
1186 			    "IOCTL");
1187 			mdb_printf("OPS     ");
1188 			for (i = 1; i < ZIO_TYPES; i++)
1189 				mdb_printf("%11#llx%s", vs->vs_ops[i],
1190 				    i == ZIO_TYPES - 1 ? "" : "  ");
1191 			mdb_printf("\n");
1192 			mdb_printf("BYTES   ");
1193 			for (i = 1; i < ZIO_TYPES; i++)
1194 				mdb_printf("%11#llx%s", vs->vs_bytes[i],
1195 				    i == ZIO_TYPES - 1 ? "" : "  ");
1196 
1197 
1198 			mdb_printf("\n");
1199 			mdb_printf("EREAD    %10#llx\n", vs->vs_read_errors);
1200 			mdb_printf("EWRITE   %10#llx\n", vs->vs_write_errors);
1201 			mdb_printf("ECKSUM   %10#llx\n",
1202 			    vs->vs_checksum_errors);
1203 			mdb_dec_indent(4);
1204 		}
1205 
1206 		if (stats)
1207 			mdb_printf("\n");
1208 	}
1209 
1210 	children = vdev.vdev_children;
1211 
1212 	if (children == 0 || !recursive)
1213 		return (DCMD_OK);
1214 
1215 	child = mdb_alloc(children * sizeof (void *), UM_SLEEP | UM_GC);
1216 	if (mdb_vread(child, children * sizeof (void *),
1217 	    (uintptr_t)vdev.vdev_child) == -1) {
1218 		mdb_warn("failed to read vdev children at %p", vdev.vdev_child);
1219 		return (DCMD_ERR);
1220 	}
1221 
1222 	for (c = 0; c < children; c++) {
1223 		if (do_print_vdev(child[c], flags, depth + 2, stats,
1224 		    recursive))
1225 			return (DCMD_ERR);
1226 	}
1227 
1228 	return (DCMD_OK);
1229 }
1230 
1231 static int
1232 vdev_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1233 {
1234 	int recursive = FALSE;
1235 	int stats = FALSE;
1236 	uint64_t depth = 0;
1237 
1238 	if (mdb_getopts(argc, argv,
1239 	    'r', MDB_OPT_SETBITS, TRUE, &recursive,
1240 	    'e', MDB_OPT_SETBITS, TRUE, &stats,
1241 	    'd', MDB_OPT_UINT64, &depth,
1242 	    NULL) != argc)
1243 		return (DCMD_USAGE);
1244 
1245 	if (!(flags & DCMD_ADDRSPEC)) {
1246 		mdb_warn("no vdev_t address given\n");
1247 		return (DCMD_ERR);
1248 	}
1249 
1250 	return (do_print_vdev(addr, flags, (int)depth, stats, recursive));
1251 }
1252 
1253 typedef struct metaslab_walk_data {
1254 	uint64_t mw_numvdevs;
1255 	uintptr_t *mw_vdevs;
1256 	int mw_curvdev;
1257 	uint64_t mw_nummss;
1258 	uintptr_t *mw_mss;
1259 	int mw_curms;
1260 } metaslab_walk_data_t;
1261 
1262 static int
1263 metaslab_walk_step(mdb_walk_state_t *wsp)
1264 {
1265 	metaslab_walk_data_t *mw = wsp->walk_data;
1266 	metaslab_t ms;
1267 	uintptr_t msp;
1268 
1269 	if (mw->mw_curvdev >= mw->mw_numvdevs)
1270 		return (WALK_DONE);
1271 
1272 	if (mw->mw_mss == NULL) {
1273 		uintptr_t mssp;
1274 		uintptr_t vdevp;
1275 
1276 		ASSERT(mw->mw_curms == 0);
1277 		ASSERT(mw->mw_nummss == 0);
1278 
1279 		vdevp = mw->mw_vdevs[mw->mw_curvdev];
1280 		if (GETMEMB(vdevp, struct vdev, vdev_ms, mssp) ||
1281 		    GETMEMB(vdevp, struct vdev, vdev_ms_count, mw->mw_nummss)) {
1282 			return (WALK_ERR);
1283 		}
1284 
1285 		mw->mw_mss = mdb_alloc(mw->mw_nummss * sizeof (void*),
1286 		    UM_SLEEP | UM_GC);
1287 		if (mdb_vread(mw->mw_mss, mw->mw_nummss * sizeof (void*),
1288 		    mssp) == -1) {
1289 			mdb_warn("failed to read vdev_ms at %p", mssp);
1290 			return (WALK_ERR);
1291 		}
1292 	}
1293 
1294 	if (mw->mw_curms >= mw->mw_nummss) {
1295 		mw->mw_mss = NULL;
1296 		mw->mw_curms = 0;
1297 		mw->mw_nummss = 0;
1298 		mw->mw_curvdev++;
1299 		return (WALK_NEXT);
1300 	}
1301 
1302 	msp = mw->mw_mss[mw->mw_curms];
1303 	if (mdb_vread(&ms, sizeof (metaslab_t), msp) == -1) {
1304 		mdb_warn("failed to read metaslab_t at %p", msp);
1305 		return (WALK_ERR);
1306 	}
1307 
1308 	mw->mw_curms++;
1309 
1310 	return (wsp->walk_callback(msp, &ms, wsp->walk_cbdata));
1311 }
1312 
1313 /* ARGSUSED */
1314 static int
1315 metaslab_walk_init(mdb_walk_state_t *wsp)
1316 {
1317 	metaslab_walk_data_t *mw;
1318 	uintptr_t root_vdevp;
1319 	uintptr_t childp;
1320 
1321 	if (wsp->walk_addr == NULL) {
1322 		mdb_warn("must supply address of spa_t\n");
1323 		return (WALK_ERR);
1324 	}
1325 
1326 	mw = mdb_zalloc(sizeof (metaslab_walk_data_t), UM_SLEEP | UM_GC);
1327 
1328 	if (GETMEMB(wsp->walk_addr, struct spa, spa_root_vdev, root_vdevp) ||
1329 	    GETMEMB(root_vdevp, struct vdev, vdev_children, mw->mw_numvdevs) ||
1330 	    GETMEMB(root_vdevp, struct vdev, vdev_child, childp)) {
1331 		return (DCMD_ERR);
1332 	}
1333 
1334 	mw->mw_vdevs = mdb_alloc(mw->mw_numvdevs * sizeof (void *),
1335 	    UM_SLEEP | UM_GC);
1336 	if (mdb_vread(mw->mw_vdevs, mw->mw_numvdevs * sizeof (void *),
1337 	    childp) == -1) {
1338 		mdb_warn("failed to read root vdev children at %p", childp);
1339 		return (DCMD_ERR);
1340 	}
1341 
1342 	wsp->walk_data = mw;
1343 
1344 	return (WALK_NEXT);
1345 }
1346 
1347 typedef struct mdb_spa {
1348 	uintptr_t spa_dsl_pool;
1349 	uintptr_t spa_root_vdev;
1350 } mdb_spa_t;
1351 
1352 typedef struct mdb_dsl_dir {
1353 	uintptr_t dd_phys;
1354 	int64_t dd_space_towrite[TXG_SIZE];
1355 } mdb_dsl_dir_t;
1356 
1357 typedef struct mdb_dsl_dir_phys {
1358 	uint64_t dd_used_bytes;
1359 	uint64_t dd_compressed_bytes;
1360 	uint64_t dd_uncompressed_bytes;
1361 } mdb_dsl_dir_phys_t;
1362 
1363 typedef struct mdb_vdev {
1364 	uintptr_t vdev_parent;
1365 	uintptr_t vdev_ms;
1366 	uint64_t vdev_ms_count;
1367 	vdev_stat_t vdev_stat;
1368 } mdb_vdev_t;
1369 
1370 typedef struct mdb_metaslab {
1371 	space_map_t ms_allocmap[TXG_SIZE];
1372 	space_map_t ms_freemap[TXG_SIZE];
1373 	space_map_t ms_map;
1374 	space_map_obj_t ms_smo;
1375 	space_map_obj_t ms_smo_syncing;
1376 } mdb_metaslab_t;
1377 
1378 typedef struct space_data {
1379 	uint64_t ms_allocmap[TXG_SIZE];
1380 	uint64_t ms_freemap[TXG_SIZE];
1381 	uint64_t ms_map;
1382 	uint64_t avail;
1383 	uint64_t nowavail;
1384 } space_data_t;
1385 
1386 /* ARGSUSED */
1387 static int
1388 space_cb(uintptr_t addr, const void *unknown, void *arg)
1389 {
1390 	space_data_t *sd = arg;
1391 	mdb_metaslab_t ms;
1392 
1393 	if (GETMEMB(addr, struct metaslab, ms_allocmap, ms.ms_allocmap) ||
1394 	    GETMEMB(addr, struct metaslab, ms_freemap, ms.ms_freemap) ||
1395 	    GETMEMB(addr, struct metaslab, ms_map, ms.ms_map) ||
1396 	    GETMEMB(addr, struct metaslab, ms_smo, ms.ms_smo) ||
1397 	    GETMEMB(addr, struct metaslab, ms_smo_syncing, ms.ms_smo_syncing)) {
1398 		return (WALK_ERR);
1399 	}
1400 
1401 	sd->ms_allocmap[0] += ms.ms_allocmap[0].sm_space;
1402 	sd->ms_allocmap[1] += ms.ms_allocmap[1].sm_space;
1403 	sd->ms_allocmap[2] += ms.ms_allocmap[2].sm_space;
1404 	sd->ms_allocmap[3] += ms.ms_allocmap[3].sm_space;
1405 	sd->ms_freemap[0] += ms.ms_freemap[0].sm_space;
1406 	sd->ms_freemap[1] += ms.ms_freemap[1].sm_space;
1407 	sd->ms_freemap[2] += ms.ms_freemap[2].sm_space;
1408 	sd->ms_freemap[3] += ms.ms_freemap[3].sm_space;
1409 	sd->ms_map += ms.ms_map.sm_space;
1410 	sd->avail += ms.ms_map.sm_size - ms.ms_smo.smo_alloc;
1411 	sd->nowavail += ms.ms_map.sm_size - ms.ms_smo_syncing.smo_alloc;
1412 
1413 	return (WALK_NEXT);
1414 }
1415 
1416 /*
1417  * ::spa_space [-b]
1418  *
1419  * Given a spa_t, print out it's on-disk space usage and in-core
1420  * estimates of future usage.  If -b is given, print space in bytes.
1421  * Otherwise print in megabytes.
1422  */
1423 /* ARGSUSED */
1424 static int
1425 spa_space(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1426 {
1427 	mdb_spa_t spa;
1428 	uintptr_t dp_root_dir;
1429 	mdb_dsl_dir_t dd;
1430 	mdb_dsl_dir_phys_t dsp;
1431 	uint64_t children;
1432 	uintptr_t childaddr;
1433 	space_data_t sd;
1434 	int shift = 20;
1435 	char *suffix = "M";
1436 	int bits = FALSE;
1437 
1438 	if (mdb_getopts(argc, argv, 'b', MDB_OPT_SETBITS, TRUE, &bits, NULL) !=
1439 	    argc)
1440 		return (DCMD_USAGE);
1441 	if (!(flags & DCMD_ADDRSPEC))
1442 		return (DCMD_USAGE);
1443 
1444 	if (bits) {
1445 		shift = 0;
1446 		suffix = "";
1447 	}
1448 
1449 	if (GETMEMB(addr, struct spa, spa_dsl_pool, spa.spa_dsl_pool) ||
1450 	    GETMEMB(addr, struct spa, spa_root_vdev, spa.spa_root_vdev) ||
1451 	    GETMEMB(spa.spa_root_vdev, struct vdev, vdev_children, children) ||
1452 	    GETMEMB(spa.spa_root_vdev, struct vdev, vdev_child, childaddr) ||
1453 	    GETMEMB(spa.spa_dsl_pool, struct dsl_pool,
1454 	    dp_root_dir, dp_root_dir) ||
1455 	    GETMEMB(dp_root_dir, struct dsl_dir, dd_phys, dd.dd_phys) ||
1456 	    GETMEMB(dp_root_dir, struct dsl_dir,
1457 	    dd_space_towrite, dd.dd_space_towrite) ||
1458 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1459 	    dd_used_bytes, dsp.dd_used_bytes) ||
1460 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1461 	    dd_compressed_bytes, dsp.dd_compressed_bytes) ||
1462 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1463 	    dd_uncompressed_bytes, dsp.dd_uncompressed_bytes)) {
1464 		return (DCMD_ERR);
1465 	}
1466 
1467 	mdb_printf("dd_space_towrite = %llu%s %llu%s %llu%s %llu%s\n",
1468 	    dd.dd_space_towrite[0] >> shift, suffix,
1469 	    dd.dd_space_towrite[1] >> shift, suffix,
1470 	    dd.dd_space_towrite[2] >> shift, suffix,
1471 	    dd.dd_space_towrite[3] >> shift, suffix);
1472 
1473 	mdb_printf("dd_phys.dd_used_bytes = %llu%s\n",
1474 	    dsp.dd_used_bytes >> shift, suffix);
1475 	mdb_printf("dd_phys.dd_compressed_bytes = %llu%s\n",
1476 	    dsp.dd_compressed_bytes >> shift, suffix);
1477 	mdb_printf("dd_phys.dd_uncompressed_bytes = %llu%s\n",
1478 	    dsp.dd_uncompressed_bytes >> shift, suffix);
1479 
1480 	bzero(&sd, sizeof (sd));
1481 	if (mdb_pwalk("metaslab", space_cb, &sd, addr) != 0) {
1482 		mdb_warn("can't walk metaslabs");
1483 		return (DCMD_ERR);
1484 	}
1485 
1486 	mdb_printf("ms_allocmap = %llu%s %llu%s %llu%s %llu%s\n",
1487 	    sd.ms_allocmap[0] >> shift, suffix,
1488 	    sd.ms_allocmap[1] >> shift, suffix,
1489 	    sd.ms_allocmap[2] >> shift, suffix,
1490 	    sd.ms_allocmap[3] >> shift, suffix);
1491 	mdb_printf("ms_freemap = %llu%s %llu%s %llu%s %llu%s\n",
1492 	    sd.ms_freemap[0] >> shift, suffix,
1493 	    sd.ms_freemap[1] >> shift, suffix,
1494 	    sd.ms_freemap[2] >> shift, suffix,
1495 	    sd.ms_freemap[3] >> shift, suffix);
1496 	mdb_printf("ms_map = %llu%s\n", sd.ms_map >> shift, suffix);
1497 	mdb_printf("last synced avail = %llu%s\n", sd.avail >> shift, suffix);
1498 	mdb_printf("current syncing avail = %llu%s\n",
1499 	    sd.nowavail >> shift, suffix);
1500 
1501 	return (DCMD_OK);
1502 }
1503 
1504 /*
1505  * ::spa_verify
1506  *
1507  * Given a spa_t, verify that that the pool is self-consistent.
1508  * Currently, it only checks to make sure that the vdev tree exists.
1509  */
1510 /* ARGSUSED */
1511 static int
1512 spa_verify(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1513 {
1514 	spa_t spa;
1515 
1516 	if (argc != 0 || !(flags & DCMD_ADDRSPEC))
1517 		return (DCMD_USAGE);
1518 
1519 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1520 		mdb_warn("failed to read spa_t at %p", addr);
1521 		return (DCMD_ERR);
1522 	}
1523 
1524 	if (spa.spa_root_vdev == NULL) {
1525 		mdb_printf("no vdev tree present\n");
1526 		return (DCMD_OK);
1527 	}
1528 
1529 	return (DCMD_OK);
1530 }
1531 
1532 static int
1533 spa_print_aux(spa_aux_vdev_t *sav, uint_t flags, mdb_arg_t *v,
1534     const char *name)
1535 {
1536 	uintptr_t *aux;
1537 	size_t len;
1538 	int ret, i;
1539 
1540 	/*
1541 	 * Iterate over aux vdevs and print those out as well.  This is a
1542 	 * little annoying because we don't have a root vdev to pass to ::vdev.
1543 	 * Instead, we print a single line and then call it for each child
1544 	 * vdev.
1545 	 */
1546 	if (sav->sav_count != 0) {
1547 		v[1].a_type = MDB_TYPE_STRING;
1548 		v[1].a_un.a_str = "-d";
1549 		v[2].a_type = MDB_TYPE_IMMEDIATE;
1550 		v[2].a_un.a_val = 2;
1551 
1552 		len = sav->sav_count * sizeof (uintptr_t);
1553 		aux = mdb_alloc(len, UM_SLEEP);
1554 		if (mdb_vread(aux, len,
1555 		    (uintptr_t)sav->sav_vdevs) == -1) {
1556 			mdb_free(aux, len);
1557 			mdb_warn("failed to read l2cache vdevs at %p",
1558 			    sav->sav_vdevs);
1559 			return (DCMD_ERR);
1560 		}
1561 
1562 		mdb_printf("%-?s %-9s %-12s %s\n", "-", "-", "-", name);
1563 
1564 		for (i = 0; i < sav->sav_count; i++) {
1565 			ret = mdb_call_dcmd("vdev", aux[i], flags, 3, v);
1566 			if (ret != DCMD_OK) {
1567 				mdb_free(aux, len);
1568 				return (ret);
1569 			}
1570 		}
1571 
1572 		mdb_free(aux, len);
1573 	}
1574 
1575 	return (0);
1576 }
1577 
1578 /*
1579  * ::spa_vdevs
1580  *
1581  * 	-e	Include error stats
1582  *
1583  * Print out a summarized list of vdevs for the given spa_t.
1584  * This is accomplished by invoking "::vdev -re" on the root vdev, as well as
1585  * iterating over the cache devices.
1586  */
1587 /* ARGSUSED */
1588 static int
1589 spa_vdevs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1590 {
1591 	spa_t spa;
1592 	mdb_arg_t v[3];
1593 	int errors = FALSE;
1594 	int ret;
1595 
1596 	if (mdb_getopts(argc, argv,
1597 	    'e', MDB_OPT_SETBITS, TRUE, &errors,
1598 	    NULL) != argc)
1599 		return (DCMD_USAGE);
1600 
1601 	if (!(flags & DCMD_ADDRSPEC))
1602 		return (DCMD_USAGE);
1603 
1604 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1605 		mdb_warn("failed to read spa_t at %p", addr);
1606 		return (DCMD_ERR);
1607 	}
1608 
1609 	/*
1610 	 * Unitialized spa_t structures can have a NULL root vdev.
1611 	 */
1612 	if (spa.spa_root_vdev == NULL) {
1613 		mdb_printf("no associated vdevs\n");
1614 		return (DCMD_OK);
1615 	}
1616 
1617 	v[0].a_type = MDB_TYPE_STRING;
1618 	v[0].a_un.a_str = errors ? "-re" : "-r";
1619 
1620 	ret = mdb_call_dcmd("vdev", (uintptr_t)spa.spa_root_vdev,
1621 	    flags, 1, v);
1622 	if (ret != DCMD_OK)
1623 		return (ret);
1624 
1625 	if (spa_print_aux(&spa.spa_l2cache, flags, v, "cache") != 0 ||
1626 	    spa_print_aux(&spa.spa_spares, flags, v, "spares") != 0)
1627 		return (DCMD_ERR);
1628 
1629 	return (DCMD_OK);
1630 }
1631 
1632 /*
1633  * ::zio
1634  *
1635  * Print a summary of zio_t and all its children.  This is intended to display a
1636  * zio tree, and hence we only pick the most important pieces of information for
1637  * the main summary.  More detailed information can always be found by doing a
1638  * '::print zio' on the underlying zio_t.  The columns we display are:
1639  *
1640  *	ADDRESS		TYPE	STAGE		WAITER
1641  *
1642  * The 'address' column is indented by one space for each depth level as we
1643  * descend down the tree.
1644  */
1645 
1646 #define	ZIO_MAXINDENT	24
1647 #define	ZIO_MAXWIDTH	(sizeof (uintptr_t) * 2 + ZIO_MAXINDENT)
1648 #define	ZIO_WALK_SELF	0
1649 #define	ZIO_WALK_CHILD	1
1650 #define	ZIO_WALK_PARENT	2
1651 
1652 typedef struct zio_print_args {
1653 	int	zpa_current_depth;
1654 	int	zpa_min_depth;
1655 	int	zpa_max_depth;
1656 	int	zpa_type;
1657 	uint_t	zpa_flags;
1658 } zio_print_args_t;
1659 
1660 static int zio_child_cb(uintptr_t addr, const void *unknown, void *arg);
1661 
1662 static int
1663 zio_print_cb(uintptr_t addr, const void *data, void *priv)
1664 {
1665 	const zio_t *zio = data;
1666 	zio_print_args_t *zpa = priv;
1667 	mdb_ctf_id_t type_enum, stage_enum;
1668 	int indent = zpa->zpa_current_depth;
1669 	const char *type, *stage;
1670 	uintptr_t laddr;
1671 
1672 	if (indent > ZIO_MAXINDENT)
1673 		indent = ZIO_MAXINDENT;
1674 
1675 	if (mdb_ctf_lookup_by_name("enum zio_type", &type_enum) == -1 ||
1676 	    mdb_ctf_lookup_by_name("enum zio_stage", &stage_enum) == -1) {
1677 		mdb_warn("failed to lookup zio enums");
1678 		return (WALK_ERR);
1679 	}
1680 
1681 	if ((type = mdb_ctf_enum_name(type_enum, zio->io_type)) != NULL)
1682 		type += sizeof ("ZIO_TYPE_") - 1;
1683 	else
1684 		type = "?";
1685 
1686 	if ((stage = mdb_ctf_enum_name(stage_enum, zio->io_stage)) != NULL)
1687 		stage += sizeof ("ZIO_STAGE_") - 1;
1688 	else
1689 		stage = "?";
1690 
1691 	if (zpa->zpa_current_depth >= zpa->zpa_min_depth) {
1692 		if (zpa->zpa_flags & DCMD_PIPE_OUT) {
1693 			mdb_printf("%?p\n", addr);
1694 		} else {
1695 			mdb_printf("%*s%-*p %-5s %-16s ", indent, "",
1696 			    ZIO_MAXWIDTH - indent, addr, type, stage);
1697 			if (zio->io_waiter)
1698 				mdb_printf("%?p\n", zio->io_waiter);
1699 			else
1700 				mdb_printf("-\n");
1701 		}
1702 	}
1703 
1704 	if (zpa->zpa_current_depth >= zpa->zpa_max_depth)
1705 		return (WALK_NEXT);
1706 
1707 	if (zpa->zpa_type == ZIO_WALK_PARENT)
1708 		laddr = addr + OFFSETOF(zio_t, io_parent_list);
1709 	else
1710 		laddr = addr + OFFSETOF(zio_t, io_child_list);
1711 
1712 	zpa->zpa_current_depth++;
1713 	if (mdb_pwalk("list", zio_child_cb, zpa, laddr) != 0) {
1714 		mdb_warn("failed to walk zio_t children at %p\n", laddr);
1715 		return (WALK_ERR);
1716 	}
1717 	zpa->zpa_current_depth--;
1718 
1719 	return (WALK_NEXT);
1720 }
1721 
1722 /* ARGSUSED */
1723 static int
1724 zio_child_cb(uintptr_t addr, const void *unknown, void *arg)
1725 {
1726 	zio_link_t zl;
1727 	zio_t zio;
1728 	uintptr_t ziop;
1729 	zio_print_args_t *zpa = arg;
1730 
1731 	if (mdb_vread(&zl, sizeof (zl), addr) == -1) {
1732 		mdb_warn("failed to read zio_link_t at %p", addr);
1733 		return (WALK_ERR);
1734 	}
1735 
1736 	if (zpa->zpa_type == ZIO_WALK_PARENT)
1737 		ziop = (uintptr_t)zl.zl_parent;
1738 	else
1739 		ziop = (uintptr_t)zl.zl_child;
1740 
1741 	if (mdb_vread(&zio, sizeof (zio_t), ziop) == -1) {
1742 		mdb_warn("failed to read zio_t at %p", ziop);
1743 		return (WALK_ERR);
1744 	}
1745 
1746 	return (zio_print_cb(ziop, &zio, arg));
1747 }
1748 
1749 /* ARGSUSED */
1750 static int
1751 zio_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1752 {
1753 	zio_t zio;
1754 	zio_print_args_t zpa = { 0 };
1755 
1756 	if (!(flags & DCMD_ADDRSPEC))
1757 		return (DCMD_USAGE);
1758 
1759 	if (mdb_getopts(argc, argv,
1760 	    'r', MDB_OPT_SETBITS, INT_MAX, &zpa.zpa_max_depth,
1761 	    'c', MDB_OPT_SETBITS, ZIO_WALK_CHILD, &zpa.zpa_type,
1762 	    'p', MDB_OPT_SETBITS, ZIO_WALK_PARENT, &zpa.zpa_type,
1763 	    NULL) != argc)
1764 		return (DCMD_USAGE);
1765 
1766 	zpa.zpa_flags = flags;
1767 	if (zpa.zpa_max_depth != 0) {
1768 		if (zpa.zpa_type == ZIO_WALK_SELF)
1769 			zpa.zpa_type = ZIO_WALK_CHILD;
1770 	} else if (zpa.zpa_type != ZIO_WALK_SELF) {
1771 		zpa.zpa_min_depth = 1;
1772 		zpa.zpa_max_depth = 1;
1773 	}
1774 
1775 	if (mdb_vread(&zio, sizeof (zio_t), addr) == -1) {
1776 		mdb_warn("failed to read zio_t at %p", addr);
1777 		return (DCMD_ERR);
1778 	}
1779 
1780 	if (!(flags & DCMD_PIPE_OUT) && DCMD_HDRSPEC(flags))
1781 		mdb_printf("%<u>%-*s %-5s %-16s %-?s%</u>\n", ZIO_MAXWIDTH,
1782 		    "ADDRESS", "TYPE", "STAGE", "WAITER");
1783 
1784 	if (zio_print_cb(addr, &zio, &zpa) != WALK_NEXT)
1785 		return (DCMD_ERR);
1786 
1787 	return (DCMD_OK);
1788 }
1789 
1790 /*
1791  * [addr]::zio_state
1792  *
1793  * Print a summary of all zio_t structures on the system, or for a particular
1794  * pool.  This is equivalent to '::walk zio_root | ::zio'.
1795  */
1796 /*ARGSUSED*/
1797 static int
1798 zio_state(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1799 {
1800 	/*
1801 	 * MDB will remember the last address of the pipeline, so if we don't
1802 	 * zero this we'll end up trying to walk zio structures for a
1803 	 * non-existent spa_t.
1804 	 */
1805 	if (!(flags & DCMD_ADDRSPEC))
1806 		addr = 0;
1807 
1808 	return (mdb_pwalk_dcmd("zio_root", "zio", argc, argv, addr));
1809 }
1810 
1811 typedef struct txg_list_walk_data {
1812 	uintptr_t lw_head[TXG_SIZE];
1813 	int	lw_txgoff;
1814 	int	lw_maxoff;
1815 	size_t	lw_offset;
1816 	void	*lw_obj;
1817 } txg_list_walk_data_t;
1818 
1819 static int
1820 txg_list_walk_init_common(mdb_walk_state_t *wsp, int txg, int maxoff)
1821 {
1822 	txg_list_walk_data_t *lwd;
1823 	txg_list_t list;
1824 	int i;
1825 
1826 	lwd = mdb_alloc(sizeof (txg_list_walk_data_t), UM_SLEEP | UM_GC);
1827 	if (mdb_vread(&list, sizeof (txg_list_t), wsp->walk_addr) == -1) {
1828 		mdb_warn("failed to read txg_list_t at %#lx", wsp->walk_addr);
1829 		return (WALK_ERR);
1830 	}
1831 
1832 	for (i = 0; i < TXG_SIZE; i++)
1833 		lwd->lw_head[i] = (uintptr_t)list.tl_head[i];
1834 	lwd->lw_offset = list.tl_offset;
1835 	lwd->lw_obj = mdb_alloc(lwd->lw_offset + sizeof (txg_node_t),
1836 	    UM_SLEEP | UM_GC);
1837 	lwd->lw_txgoff = txg;
1838 	lwd->lw_maxoff = maxoff;
1839 
1840 	wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
1841 	wsp->walk_data = lwd;
1842 
1843 	return (WALK_NEXT);
1844 }
1845 
1846 static int
1847 txg_list_walk_init(mdb_walk_state_t *wsp)
1848 {
1849 	return (txg_list_walk_init_common(wsp, 0, TXG_SIZE-1));
1850 }
1851 
1852 static int
1853 txg_list0_walk_init(mdb_walk_state_t *wsp)
1854 {
1855 	return (txg_list_walk_init_common(wsp, 0, 0));
1856 }
1857 
1858 static int
1859 txg_list1_walk_init(mdb_walk_state_t *wsp)
1860 {
1861 	return (txg_list_walk_init_common(wsp, 1, 1));
1862 }
1863 
1864 static int
1865 txg_list2_walk_init(mdb_walk_state_t *wsp)
1866 {
1867 	return (txg_list_walk_init_common(wsp, 2, 2));
1868 }
1869 
1870 static int
1871 txg_list3_walk_init(mdb_walk_state_t *wsp)
1872 {
1873 	return (txg_list_walk_init_common(wsp, 3, 3));
1874 }
1875 
1876 static int
1877 txg_list_walk_step(mdb_walk_state_t *wsp)
1878 {
1879 	txg_list_walk_data_t *lwd = wsp->walk_data;
1880 	uintptr_t addr;
1881 	txg_node_t *node;
1882 	int status;
1883 
1884 	while (wsp->walk_addr == NULL && lwd->lw_txgoff < lwd->lw_maxoff) {
1885 		lwd->lw_txgoff++;
1886 		wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
1887 	}
1888 
1889 	if (wsp->walk_addr == NULL)
1890 		return (WALK_DONE);
1891 
1892 	addr = wsp->walk_addr - lwd->lw_offset;
1893 
1894 	if (mdb_vread(lwd->lw_obj,
1895 	    lwd->lw_offset + sizeof (txg_node_t), addr) == -1) {
1896 		mdb_warn("failed to read list element at %#lx", addr);
1897 		return (WALK_ERR);
1898 	}
1899 
1900 	status = wsp->walk_callback(addr, lwd->lw_obj, wsp->walk_cbdata);
1901 	node = (txg_node_t *)((uintptr_t)lwd->lw_obj + lwd->lw_offset);
1902 	wsp->walk_addr = (uintptr_t)node->tn_next[lwd->lw_txgoff];
1903 
1904 	return (status);
1905 }
1906 
1907 /*
1908  * ::walk spa
1909  *
1910  * Walk all named spa_t structures in the namespace.  This is nothing more than
1911  * a layered avl walk.
1912  */
1913 static int
1914 spa_walk_init(mdb_walk_state_t *wsp)
1915 {
1916 	GElf_Sym sym;
1917 
1918 	if (wsp->walk_addr != NULL) {
1919 		mdb_warn("spa walk only supports global walks\n");
1920 		return (WALK_ERR);
1921 	}
1922 
1923 	if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "spa_namespace_avl", &sym) == -1) {
1924 		mdb_warn("failed to find symbol 'spa_namespace_avl'");
1925 		return (WALK_ERR);
1926 	}
1927 
1928 	wsp->walk_addr = (uintptr_t)sym.st_value;
1929 
1930 	if (mdb_layered_walk("avl", wsp) == -1) {
1931 		mdb_warn("failed to walk 'avl'\n");
1932 		return (WALK_ERR);
1933 	}
1934 
1935 	return (WALK_NEXT);
1936 }
1937 
1938 static int
1939 spa_walk_step(mdb_walk_state_t *wsp)
1940 {
1941 	spa_t	spa;
1942 
1943 	if (mdb_vread(&spa, sizeof (spa), wsp->walk_addr) == -1) {
1944 		mdb_warn("failed to read spa_t at %p", wsp->walk_addr);
1945 		return (WALK_ERR);
1946 	}
1947 
1948 	return (wsp->walk_callback(wsp->walk_addr, &spa, wsp->walk_cbdata));
1949 }
1950 
1951 /*
1952  * [addr]::walk zio
1953  *
1954  * Walk all active zio_t structures on the system.  This is simply a layered
1955  * walk on top of ::walk zio_cache, with the optional ability to limit the
1956  * structures to a particular pool.
1957  */
1958 static int
1959 zio_walk_init(mdb_walk_state_t *wsp)
1960 {
1961 	wsp->walk_data = (void *)wsp->walk_addr;
1962 
1963 	if (mdb_layered_walk("zio_cache", wsp) == -1) {
1964 		mdb_warn("failed to walk 'zio_cache'\n");
1965 		return (WALK_ERR);
1966 	}
1967 
1968 	return (WALK_NEXT);
1969 }
1970 
1971 static int
1972 zio_walk_step(mdb_walk_state_t *wsp)
1973 {
1974 	zio_t zio;
1975 
1976 	if (mdb_vread(&zio, sizeof (zio), wsp->walk_addr) == -1) {
1977 		mdb_warn("failed to read zio_t at %p", wsp->walk_addr);
1978 		return (WALK_ERR);
1979 	}
1980 
1981 	if (wsp->walk_data != NULL && wsp->walk_data != zio.io_spa)
1982 		return (WALK_NEXT);
1983 
1984 	return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
1985 }
1986 
1987 /*
1988  * [addr]::walk zio_root
1989  *
1990  * Walk only root zio_t structures, optionally for a particular spa_t.
1991  */
1992 static int
1993 zio_walk_root_step(mdb_walk_state_t *wsp)
1994 {
1995 	zio_t zio;
1996 
1997 	if (mdb_vread(&zio, sizeof (zio), wsp->walk_addr) == -1) {
1998 		mdb_warn("failed to read zio_t at %p", wsp->walk_addr);
1999 		return (WALK_ERR);
2000 	}
2001 
2002 	if (wsp->walk_data != NULL && wsp->walk_data != zio.io_spa)
2003 		return (WALK_NEXT);
2004 
2005 	/* If the parent list is not empty, ignore */
2006 	if (zio.io_parent_list.list_head.list_next !=
2007 	    &((zio_t *)wsp->walk_addr)->io_parent_list.list_head)
2008 		return (WALK_NEXT);
2009 
2010 	return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
2011 }
2012 
2013 #define	NICENUM_BUFLEN 6
2014 
2015 static int
2016 snprintfrac(char *buf, int len,
2017     uint64_t numerator, uint64_t denom, int frac_digits)
2018 {
2019 	int mul = 1;
2020 	int whole, frac, i;
2021 
2022 	for (i = frac_digits; i; i--)
2023 		mul *= 10;
2024 	whole = numerator / denom;
2025 	frac = mul * numerator / denom - mul * whole;
2026 	return (mdb_snprintf(buf, len, "%u.%0*u", whole, frac_digits, frac));
2027 }
2028 
2029 static void
2030 mdb_nicenum(uint64_t num, char *buf)
2031 {
2032 	uint64_t n = num;
2033 	int index = 0;
2034 	char *u;
2035 
2036 	while (n >= 1024) {
2037 		n = (n + (1024 / 2)) / 1024; /* Round up or down */
2038 		index++;
2039 	}
2040 
2041 	u = &" \0K\0M\0G\0T\0P\0E\0"[index*2];
2042 
2043 	if (index == 0) {
2044 		(void) mdb_snprintf(buf, NICENUM_BUFLEN, "%llu",
2045 		    (u_longlong_t)n);
2046 	} else if (n < 10 && (num & (num - 1)) != 0) {
2047 		(void) snprintfrac(buf, NICENUM_BUFLEN,
2048 		    num, 1ULL << 10 * index, 2);
2049 		strcat(buf, u);
2050 	} else if (n < 100 && (num & (num - 1)) != 0) {
2051 		(void) snprintfrac(buf, NICENUM_BUFLEN,
2052 		    num, 1ULL << 10 * index, 1);
2053 		strcat(buf, u);
2054 	} else {
2055 		(void) mdb_snprintf(buf, NICENUM_BUFLEN, "%llu%s",
2056 		    (u_longlong_t)n, u);
2057 	}
2058 }
2059 
2060 /*
2061  * ::zfs_blkstats
2062  *
2063  * 	-v	print verbose per-level information
2064  *
2065  */
2066 static int
2067 zfs_blkstats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2068 {
2069 	boolean_t verbose = B_FALSE;
2070 	zfs_all_blkstats_t stats;
2071 	dmu_object_type_t t;
2072 	zfs_blkstat_t *tzb;
2073 	uint64_t ditto;
2074 	dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES + 10];
2075 	/* +10 in case it grew */
2076 
2077 	if (mdb_readvar(&dmu_ot, "dmu_ot") == -1) {
2078 		mdb_warn("failed to read 'dmu_ot'");
2079 		return (DCMD_ERR);
2080 	}
2081 
2082 	if (mdb_getopts(argc, argv,
2083 	    'v', MDB_OPT_SETBITS, TRUE, &verbose,
2084 	    NULL) != argc)
2085 		return (DCMD_USAGE);
2086 
2087 	if (!(flags & DCMD_ADDRSPEC))
2088 		return (DCMD_USAGE);
2089 
2090 	if (GETMEMB(addr, struct spa, spa_dsl_pool, addr) ||
2091 	    GETMEMB(addr, struct dsl_pool, dp_blkstats, addr) ||
2092 	    mdb_vread(&stats, sizeof (zfs_all_blkstats_t), addr) == -1) {
2093 		mdb_warn("failed to read data at %p;", addr);
2094 		mdb_printf("maybe no stats? run \"zpool scrub\" first.");
2095 		return (DCMD_ERR);
2096 	}
2097 
2098 	tzb = &stats.zab_type[DN_MAX_LEVELS][DMU_OT_NUMTYPES];
2099 	if (tzb->zb_gangs != 0) {
2100 		mdb_printf("Ganged blocks: %llu\n",
2101 		    (longlong_t)tzb->zb_gangs);
2102 	}
2103 
2104 	ditto = tzb->zb_ditto_2_of_2_samevdev + tzb->zb_ditto_2_of_3_samevdev +
2105 	    tzb->zb_ditto_3_of_3_samevdev;
2106 	if (ditto != 0) {
2107 		mdb_printf("Dittoed blocks on same vdev: %llu\n",
2108 		    (longlong_t)ditto);
2109 	}
2110 
2111 	mdb_printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
2112 	    "\t  avg\t comp\t%%Total\tType\n");
2113 
2114 	for (t = 0; t <= DMU_OT_NUMTYPES; t++) {
2115 		char csize[NICENUM_BUFLEN], lsize[NICENUM_BUFLEN];
2116 		char psize[NICENUM_BUFLEN], asize[NICENUM_BUFLEN];
2117 		char avg[NICENUM_BUFLEN];
2118 		char comp[NICENUM_BUFLEN], pct[NICENUM_BUFLEN];
2119 		char typename[64];
2120 		int l;
2121 
2122 
2123 		if (t == DMU_OT_DEFERRED)
2124 			strcpy(typename, "deferred free");
2125 		else if (t == DMU_OT_TOTAL)
2126 			strcpy(typename, "Total");
2127 		else if (mdb_readstr(typename, sizeof (typename),
2128 		    (uintptr_t)dmu_ot[t].ot_name) == -1) {
2129 			mdb_warn("failed to read type name");
2130 			return (DCMD_ERR);
2131 		}
2132 
2133 		if (stats.zab_type[DN_MAX_LEVELS][t].zb_asize == 0)
2134 			continue;
2135 
2136 		for (l = -1; l < DN_MAX_LEVELS; l++) {
2137 			int level = (l == -1 ? DN_MAX_LEVELS : l);
2138 			zfs_blkstat_t *zb = &stats.zab_type[level][t];
2139 
2140 			if (zb->zb_asize == 0)
2141 				continue;
2142 
2143 			/*
2144 			 * Don't print each level unless requested.
2145 			 */
2146 			if (!verbose && level != DN_MAX_LEVELS)
2147 				continue;
2148 
2149 			/*
2150 			 * If all the space is level 0, don't print the
2151 			 * level 0 separately.
2152 			 */
2153 			if (level == 0 && zb->zb_asize ==
2154 			    stats.zab_type[DN_MAX_LEVELS][t].zb_asize)
2155 				continue;
2156 
2157 			mdb_nicenum(zb->zb_count, csize);
2158 			mdb_nicenum(zb->zb_lsize, lsize);
2159 			mdb_nicenum(zb->zb_psize, psize);
2160 			mdb_nicenum(zb->zb_asize, asize);
2161 			mdb_nicenum(zb->zb_asize / zb->zb_count, avg);
2162 			(void) snprintfrac(comp, NICENUM_BUFLEN,
2163 			    zb->zb_lsize, zb->zb_psize, 2);
2164 			(void) snprintfrac(pct, NICENUM_BUFLEN,
2165 			    100 * zb->zb_asize, tzb->zb_asize, 2);
2166 
2167 			mdb_printf("%6s\t%5s\t%5s\t%5s\t%5s"
2168 			    "\t%5s\t%6s\t",
2169 			    csize, lsize, psize, asize, avg, comp, pct);
2170 
2171 			if (level == DN_MAX_LEVELS)
2172 				mdb_printf("%s\n", typename);
2173 			else
2174 				mdb_printf("  L%d %s\n",
2175 				    level, typename);
2176 		}
2177 	}
2178 
2179 	return (DCMD_OK);
2180 }
2181 
2182 /* ARGSUSED */
2183 static int
2184 reference_cb(uintptr_t addr, const void *ignored, void *arg)
2185 {
2186 	static int gotid;
2187 	static mdb_ctf_id_t ref_id;
2188 	uintptr_t ref_holder;
2189 	uintptr_t ref_removed;
2190 	uint64_t ref_number;
2191 	boolean_t holder_is_str;
2192 	char holder_str[128];
2193 	boolean_t removed = (boolean_t)arg;
2194 
2195 	if (!gotid) {
2196 		if (mdb_ctf_lookup_by_name("struct reference", &ref_id) == -1) {
2197 			mdb_warn("couldn't find struct reference");
2198 			return (WALK_ERR);
2199 		}
2200 		gotid = TRUE;
2201 	}
2202 
2203 	if (GETMEMBID(addr, &ref_id, ref_holder, ref_holder) ||
2204 	    GETMEMBID(addr, &ref_id, ref_removed, ref_removed) ||
2205 	    GETMEMBID(addr, &ref_id, ref_number, ref_number))
2206 		return (WALK_ERR);
2207 
2208 	if (mdb_readstr(holder_str, sizeof (holder_str), ref_holder) != -1) {
2209 		char *cp;
2210 		holder_is_str = B_TRUE;
2211 		for (cp = holder_str; *cp; cp++) {
2212 			if (!isprint(*cp)) {
2213 				holder_is_str = B_FALSE;
2214 				break;
2215 			}
2216 		}
2217 	} else {
2218 		holder_is_str = B_FALSE;
2219 	}
2220 
2221 	if (removed)
2222 		mdb_printf("removed ");
2223 	mdb_printf("reference ");
2224 	if (ref_number != 1)
2225 		mdb_printf("with count=%llu ", ref_number);
2226 	mdb_printf("with tag %p", (void*)ref_holder);
2227 	if (holder_is_str)
2228 		mdb_printf(" \"%s\"", holder_str);
2229 	mdb_printf(", held at:\n");
2230 
2231 	(void) mdb_call_dcmd("whatis", addr, DCMD_ADDRSPEC, 0, NULL);
2232 
2233 	if (removed) {
2234 		mdb_printf("removed at:\n");
2235 		(void) mdb_call_dcmd("whatis", ref_removed,
2236 		    DCMD_ADDRSPEC, 0, NULL);
2237 	}
2238 
2239 	mdb_printf("\n");
2240 
2241 	return (WALK_NEXT);
2242 }
2243 
2244 /* ARGSUSED */
2245 static int
2246 refcount(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2247 {
2248 	uint64_t rc_count, rc_removed_count;
2249 	uintptr_t rc_list, rc_removed;
2250 	static int gotid;
2251 	static mdb_ctf_id_t rc_id;
2252 	ulong_t off;
2253 
2254 	if (!(flags & DCMD_ADDRSPEC))
2255 		return (DCMD_USAGE);
2256 
2257 	if (!gotid) {
2258 		if (mdb_ctf_lookup_by_name("struct refcount", &rc_id) == -1) {
2259 			mdb_warn("couldn't find struct refcount");
2260 			return (DCMD_ERR);
2261 		}
2262 		gotid = TRUE;
2263 	}
2264 
2265 	if (GETMEMBID(addr, &rc_id, rc_count, rc_count) ||
2266 	    GETMEMBID(addr, &rc_id, rc_removed_count, rc_removed_count))
2267 		return (DCMD_ERR);
2268 
2269 	mdb_printf("refcount_t at %p has %llu current holds, "
2270 	    "%llu recently released holds\n",
2271 	    addr, (longlong_t)rc_count, (longlong_t)rc_removed_count);
2272 
2273 	if (rc_count > 0)
2274 		mdb_printf("current holds:\n");
2275 	if (mdb_ctf_offsetof(rc_id, "rc_list", &off) == -1)
2276 		return (DCMD_ERR);
2277 	rc_list = addr + off/NBBY;
2278 	mdb_pwalk("list", reference_cb, (void*)B_FALSE, rc_list);
2279 
2280 	if (rc_removed_count > 0)
2281 		mdb_printf("released holds:\n");
2282 	if (mdb_ctf_offsetof(rc_id, "rc_removed", &off) == -1)
2283 		return (DCMD_ERR);
2284 	rc_removed = addr + off/NBBY;
2285 	mdb_pwalk("list", reference_cb, (void*)B_TRUE, rc_removed);
2286 
2287 	return (DCMD_OK);
2288 }
2289 
2290 /*
2291  * MDB module linkage information:
2292  *
2293  * We declare a list of structures describing our dcmds, and a function
2294  * named _mdb_init to return a pointer to our module information.
2295  */
2296 
2297 static const mdb_dcmd_t dcmds[] = {
2298 	{ "arc", "[-bkmg]", "print ARC variables", arc_print },
2299 	{ "blkptr", ":", "print blkptr_t", blkptr },
2300 	{ "dbuf", ":", "print dmu_buf_impl_t", dbuf },
2301 	{ "dbuf_stats", ":", "dbuf stats", dbuf_stats },
2302 	{ "dbufs",
2303 	    "\t[-O objset_t*] [-n objset_name | \"mos\"] "
2304 	    "[-o object | \"mdn\"] \n"
2305 	    "\t[-l level] [-b blkid | \"bonus\"]",
2306 	    "find dmu_buf_impl_t's that match specified criteria", dbufs },
2307 	{ "abuf_find", "dva_word[0] dva_word[1]",
2308 	    "find arc_buf_hdr_t of a specified DVA",
2309 	    abuf_find },
2310 	{ "spa", "?[-cv]", "spa_t summary", spa_print },
2311 	{ "spa_config", ":", "print spa_t configuration", spa_print_config },
2312 	{ "spa_verify", ":", "verify spa_t consistency", spa_verify },
2313 	{ "spa_space", ":[-b]", "print spa_t on-disk space usage", spa_space },
2314 	{ "spa_vdevs", ":", "given a spa_t, print vdev summary", spa_vdevs },
2315 	{ "vdev", ":[-re]\n"
2316 	    "\t-r display recursively\n"
2317 	    "\t-e print statistics",
2318 	    "vdev_t summary", vdev_print },
2319 	{ "zio", ":[cpr]\n"
2320 	    "\t-c display children\n"
2321 	    "\t-p display parents\n"
2322 	    "\t-r display recursively",
2323 	    "zio_t summary", zio_print },
2324 	{ "zio_state", "?", "print out all zio_t structures on system or "
2325 	    "for a particular pool", zio_state },
2326 	{ "zfs_blkstats", ":[-v]",
2327 	    "given a spa_t, print block type stats from last scrub",
2328 	    zfs_blkstats },
2329 	{ "zfs_params", "", "print zfs tunable parameters", zfs_params },
2330 	{ "refcount", "", "print refcount_t holders", refcount },
2331 	{ "zap_leaf", "", "print zap_leaf_phys_t", zap_leaf },
2332 	{ NULL }
2333 };
2334 
2335 static const mdb_walker_t walkers[] = {
2336 	/*
2337 	 * In userland, there is no generic provider of list_t walkers, so we
2338 	 * need to add it.
2339 	 */
2340 #ifndef _KERNEL
2341 	{ LIST_WALK_NAME, LIST_WALK_DESC,
2342 		list_walk_init, list_walk_step, list_walk_fini },
2343 #endif
2344 	{ "zms_freelist", "walk ZFS metaslab freelist",
2345 		freelist_walk_init, freelist_walk_step, NULL },
2346 	{ "txg_list", "given any txg_list_t *, walk all entries in all txgs",
2347 		txg_list_walk_init, txg_list_walk_step, NULL },
2348 	{ "txg_list0", "given any txg_list_t *, walk all entries in txg 0",
2349 		txg_list0_walk_init, txg_list_walk_step, NULL },
2350 	{ "txg_list1", "given any txg_list_t *, walk all entries in txg 1",
2351 		txg_list1_walk_init, txg_list_walk_step, NULL },
2352 	{ "txg_list2", "given any txg_list_t *, walk all entries in txg 2",
2353 		txg_list2_walk_init, txg_list_walk_step, NULL },
2354 	{ "txg_list3", "given any txg_list_t *, walk all entries in txg 3",
2355 		txg_list3_walk_init, txg_list_walk_step, NULL },
2356 	{ "zio", "walk all zio structures, optionally for a particular spa_t",
2357 		zio_walk_init, zio_walk_step, NULL },
2358 	{ "zio_root", "walk all root zio_t structures, optionally for a "
2359 	    "particular spa_t",
2360 		zio_walk_init, zio_walk_root_step, NULL },
2361 	{ "spa", "walk all spa_t entries in the namespace",
2362 		spa_walk_init, spa_walk_step, NULL },
2363 	{ "metaslab", "given a spa_t *, walk all metaslab_t structures",
2364 		metaslab_walk_init, metaslab_walk_step, NULL },
2365 	{ NULL }
2366 };
2367 
2368 static const mdb_modinfo_t modinfo = {
2369 	MDB_API_VERSION, dcmds, walkers
2370 };
2371 
2372 const mdb_modinfo_t *
2373 _mdb_init(void)
2374 {
2375 	return (&modinfo);
2376 }
2377