xref: /titanic_41/usr/src/cmd/mdb/common/modules/zfs/zfs.c (revision ea394cb00fd96864e34d2841b4a22357b621c78f)
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 #include <sys/zfs_acl.h>
42 #include <sys/sa_impl.h>
43 
44 #ifndef _KERNEL
45 #include "../genunix/list.h"
46 #endif
47 
48 #ifdef _KERNEL
49 #define	ZFS_OBJ_NAME	"zfs"
50 #else
51 #define	ZFS_OBJ_NAME	"libzpool.so.1"
52 #endif
53 
54 #ifndef _KERNEL
55 int aok;
56 #endif
57 
58 static int
59 getmember(uintptr_t addr, const char *type, mdb_ctf_id_t *idp,
60     const char *member, int len, void *buf)
61 {
62 	mdb_ctf_id_t id;
63 	ulong_t off;
64 	char name[64];
65 
66 	if (idp == NULL) {
67 		if (mdb_ctf_lookup_by_name(type, &id) == -1) {
68 			mdb_warn("couldn't find type %s", type);
69 			return (DCMD_ERR);
70 		}
71 		idp = &id;
72 	} else {
73 		type = name;
74 		mdb_ctf_type_name(*idp, name, sizeof (name));
75 	}
76 
77 	if (mdb_ctf_offsetof(*idp, member, &off) == -1) {
78 		mdb_warn("couldn't find member %s of type %s\n", member, type);
79 		return (DCMD_ERR);
80 	}
81 	if (off % 8 != 0) {
82 		mdb_warn("member %s of type %s is unsupported bitfield",
83 		    member, type);
84 		return (DCMD_ERR);
85 	}
86 	off /= 8;
87 
88 	if (mdb_vread(buf, len, addr + off) == -1) {
89 		mdb_warn("failed to read %s from %s at %p",
90 		    member, type, addr + off);
91 		return (DCMD_ERR);
92 	}
93 	/* mdb_warn("read %s from %s at %p+%llx\n", member, type, addr, off); */
94 
95 	return (0);
96 }
97 
98 #define	GETMEMB(addr, type, member, dest) \
99 	getmember(addr, #type, NULL, #member, sizeof (dest), &(dest))
100 
101 #define	GETMEMBID(addr, ctfid, member, dest) \
102 	getmember(addr, NULL, ctfid, #member, sizeof (dest), &(dest))
103 
104 static int
105 getrefcount(uintptr_t addr, mdb_ctf_id_t *id,
106     const char *member, uint64_t *rc)
107 {
108 	static int gotid;
109 	static mdb_ctf_id_t rc_id;
110 	ulong_t off;
111 
112 	if (!gotid) {
113 		if (mdb_ctf_lookup_by_name("struct refcount", &rc_id) == -1) {
114 			mdb_warn("couldn't find struct refcount");
115 			return (DCMD_ERR);
116 		}
117 		gotid = TRUE;
118 	}
119 
120 	if (mdb_ctf_offsetof(*id, member, &off) == -1) {
121 		char name[64];
122 		mdb_ctf_type_name(*id, name, sizeof (name));
123 		mdb_warn("couldn't find member %s of type %s\n", member, name);
124 		return (DCMD_ERR);
125 	}
126 	off /= 8;
127 
128 	return (GETMEMBID(addr + off, &rc_id, rc_count, *rc));
129 }
130 
131 static int verbose;
132 
133 static int
134 freelist_walk_init(mdb_walk_state_t *wsp)
135 {
136 	if (wsp->walk_addr == NULL) {
137 		mdb_warn("must supply starting address\n");
138 		return (WALK_ERR);
139 	}
140 
141 	wsp->walk_data = 0;  /* Index into the freelist */
142 	return (WALK_NEXT);
143 }
144 
145 static int
146 freelist_walk_step(mdb_walk_state_t *wsp)
147 {
148 	uint64_t entry;
149 	uintptr_t number = (uintptr_t)wsp->walk_data;
150 	char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
151 			    "INVALID", "INVALID", "INVALID", "INVALID" };
152 	int mapshift = SPA_MINBLOCKSHIFT;
153 
154 	if (mdb_vread(&entry, sizeof (entry), wsp->walk_addr) == -1) {
155 		mdb_warn("failed to read freelist entry %p", wsp->walk_addr);
156 		return (WALK_DONE);
157 	}
158 	wsp->walk_addr += sizeof (entry);
159 	wsp->walk_data = (void *)(number + 1);
160 
161 	if (SM_DEBUG_DECODE(entry)) {
162 		mdb_printf("DEBUG: %3u  %10s: txg=%llu  pass=%llu\n",
163 		    number,
164 		    ddata[SM_DEBUG_ACTION_DECODE(entry)],
165 		    SM_DEBUG_TXG_DECODE(entry),
166 		    SM_DEBUG_SYNCPASS_DECODE(entry));
167 	} else {
168 		mdb_printf("Entry: %3u  offsets=%08llx-%08llx  type=%c  "
169 		    "size=%06llx", number,
170 		    SM_OFFSET_DECODE(entry) << mapshift,
171 		    (SM_OFFSET_DECODE(entry) + SM_RUN_DECODE(entry)) <<
172 		    mapshift,
173 		    SM_TYPE_DECODE(entry) == SM_ALLOC ? 'A' : 'F',
174 		    SM_RUN_DECODE(entry) << mapshift);
175 		if (verbose)
176 			mdb_printf("      (raw=%012llx)\n", entry);
177 		mdb_printf("\n");
178 	}
179 	return (WALK_NEXT);
180 }
181 
182 
183 static int
184 dataset_name(uintptr_t addr, char *buf)
185 {
186 	static int gotid;
187 	static mdb_ctf_id_t dd_id;
188 	uintptr_t dd_parent;
189 	char dd_myname[MAXNAMELEN];
190 
191 	if (!gotid) {
192 		if (mdb_ctf_lookup_by_name("struct dsl_dir",
193 		    &dd_id) == -1) {
194 			mdb_warn("couldn't find struct dsl_dir");
195 			return (DCMD_ERR);
196 		}
197 		gotid = TRUE;
198 	}
199 	if (GETMEMBID(addr, &dd_id, dd_parent, dd_parent) ||
200 	    GETMEMBID(addr, &dd_id, dd_myname, dd_myname)) {
201 		return (DCMD_ERR);
202 	}
203 
204 	if (dd_parent) {
205 		if (dataset_name(dd_parent, buf))
206 			return (DCMD_ERR);
207 		strcat(buf, "/");
208 	}
209 
210 	if (dd_myname[0])
211 		strcat(buf, dd_myname);
212 	else
213 		strcat(buf, "???");
214 
215 	return (0);
216 }
217 
218 static int
219 objset_name(uintptr_t addr, char *buf)
220 {
221 	static int gotid;
222 	static mdb_ctf_id_t osi_id, ds_id;
223 	uintptr_t os_dsl_dataset;
224 	char ds_snapname[MAXNAMELEN];
225 	uintptr_t ds_dir;
226 
227 	buf[0] = '\0';
228 
229 	if (!gotid) {
230 		if (mdb_ctf_lookup_by_name("struct objset_impl",
231 		    &osi_id) == -1) {
232 			mdb_warn("couldn't find struct objset_impl");
233 			return (DCMD_ERR);
234 		}
235 		if (mdb_ctf_lookup_by_name("struct dsl_dataset",
236 		    &ds_id) == -1) {
237 			mdb_warn("couldn't find struct dsl_dataset");
238 			return (DCMD_ERR);
239 		}
240 
241 		gotid = TRUE;
242 	}
243 
244 	if (GETMEMBID(addr, &osi_id, os_dsl_dataset, os_dsl_dataset))
245 		return (DCMD_ERR);
246 
247 	if (os_dsl_dataset == 0) {
248 		strcat(buf, "mos");
249 		return (0);
250 	}
251 
252 	if (GETMEMBID(os_dsl_dataset, &ds_id, ds_snapname, ds_snapname) ||
253 	    GETMEMBID(os_dsl_dataset, &ds_id, ds_dir, ds_dir)) {
254 		return (DCMD_ERR);
255 	}
256 
257 	if (ds_dir && dataset_name(ds_dir, buf))
258 		return (DCMD_ERR);
259 
260 	if (ds_snapname[0]) {
261 		strcat(buf, "@");
262 		strcat(buf, ds_snapname);
263 	}
264 	return (0);
265 }
266 
267 static void
268 enum_lookup(char *out, size_t size, mdb_ctf_id_t id, int val,
269     const char *prefix)
270 {
271 	const char *cp;
272 	size_t len = strlen(prefix);
273 
274 	if ((cp = mdb_ctf_enum_name(id, val)) != NULL) {
275 		if (strncmp(cp, prefix, len) == 0)
276 			cp += len;
277 		(void) strncpy(out, cp, size);
278 	} else {
279 		mdb_snprintf(out, size, "? (%d)", val);
280 	}
281 }
282 
283 /* ARGSUSED */
284 static int
285 zfs_params(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
286 {
287 	/*
288 	 * This table can be approximately generated by running:
289 	 * egrep "^[a-z0-9_]+ [a-z0-9_]+( =.*)?;" *.c | cut -d ' ' -f 2
290 	 */
291 	static const char *params[] = {
292 		"arc_reduce_dnlc_percent",
293 		"zfs_arc_max",
294 		"zfs_arc_min",
295 		"arc_shrink_shift",
296 		"zfs_mdcomp_disable",
297 		"zfs_prefetch_disable",
298 		"zfetch_max_streams",
299 		"zfetch_min_sec_reap",
300 		"zfetch_block_cap",
301 		"zfetch_array_rd_sz",
302 		"zfs_default_bs",
303 		"zfs_default_ibs",
304 		"metaslab_aliquot",
305 		"reference_tracking_enable",
306 		"reference_history",
307 		"spa_max_replication_override",
308 		"spa_mode_global",
309 		"zfs_flags",
310 		"zfs_txg_synctime",
311 		"zfs_txg_timeout",
312 		"zfs_write_limit_min",
313 		"zfs_write_limit_max",
314 		"zfs_write_limit_shift",
315 		"zfs_write_limit_override",
316 		"zfs_no_write_throttle",
317 		"zfs_vdev_cache_max",
318 		"zfs_vdev_cache_size",
319 		"zfs_vdev_cache_bshift",
320 		"vdev_mirror_shift",
321 		"zfs_vdev_max_pending",
322 		"zfs_vdev_min_pending",
323 		"zfs_scrub_limit",
324 		"zfs_no_scrub_io",
325 		"zfs_no_scrub_prefetch",
326 		"zfs_vdev_time_shift",
327 		"zfs_vdev_ramp_rate",
328 		"zfs_vdev_aggregation_limit",
329 		"fzap_default_block_shift",
330 		"zfs_immediate_write_sz",
331 		"zfs_read_chunk_size",
332 		"zil_disable",
333 		"zfs_nocacheflush",
334 		"metaslab_gang_bang",
335 		"metaslab_df_alloc_threshold",
336 		"metaslab_df_free_pct",
337 		"zio_injection_enabled",
338 		"zvol_immediate_write_sz",
339 	};
340 
341 	for (int i = 0; i < sizeof (params) / sizeof (params[0]); i++) {
342 		int sz;
343 		uint64_t val64;
344 		uint32_t *val32p = (uint32_t *)&val64;
345 
346 		sz = mdb_readvar(&val64, params[i]);
347 		if (sz == 4) {
348 			mdb_printf("%s = 0x%x\n", params[i], *val32p);
349 		} else if (sz == 8) {
350 			mdb_printf("%s = 0x%llx\n", params[i], val64);
351 		} else {
352 			mdb_warn("variable %s not found", params[i]);
353 		}
354 	}
355 
356 	return (DCMD_OK);
357 }
358 
359 /* ARGSUSED */
360 static int
361 blkptr(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
362 {
363 	mdb_ctf_id_t type_enum, checksum_enum, compress_enum;
364 	char type[80], checksum[80], compress[80];
365 	blkptr_t blk, *bp = &blk;
366 	char buf[BP_SPRINTF_LEN];
367 
368 	if (mdb_vread(&blk, sizeof (blkptr_t), addr) == -1) {
369 		mdb_warn("failed to read blkptr_t");
370 		return (DCMD_ERR);
371 	}
372 
373 	if (mdb_ctf_lookup_by_name("enum dmu_object_type", &type_enum) == -1 ||
374 	    mdb_ctf_lookup_by_name("enum zio_checksum", &checksum_enum) == -1 ||
375 	    mdb_ctf_lookup_by_name("enum zio_compress", &compress_enum) == -1) {
376 		mdb_warn("Could not find blkptr enumerated types");
377 		return (DCMD_ERR);
378 	}
379 
380 	enum_lookup(type, sizeof (type), type_enum,
381 	    BP_GET_TYPE(bp), "DMU_OT_");
382 	enum_lookup(checksum, sizeof (checksum), checksum_enum,
383 	    BP_GET_CHECKSUM(bp), "ZIO_CHECKSUM_");
384 	enum_lookup(compress, sizeof (compress), compress_enum,
385 	    BP_GET_COMPRESS(bp), "ZIO_COMPRESS_");
386 
387 	SPRINTF_BLKPTR(mdb_snprintf, '\n', buf, bp, type, checksum, compress);
388 
389 	mdb_printf("%s\n", buf);
390 
391 	return (DCMD_OK);
392 }
393 
394 /* ARGSUSED */
395 static int
396 dbuf(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
397 {
398 	mdb_ctf_id_t id;
399 	dmu_buf_t db;
400 	uintptr_t objset;
401 	uint8_t level;
402 	uint64_t blkid;
403 	uint64_t holds;
404 	char objectname[32];
405 	char blkidname[32];
406 	char path[MAXNAMELEN];
407 
408 	if (DCMD_HDRSPEC(flags)) {
409 		mdb_printf("        addr object lvl blkid holds os\n");
410 	}
411 
412 	if (mdb_ctf_lookup_by_name("struct dmu_buf_impl", &id) == -1) {
413 		mdb_warn("couldn't find struct dmu_buf_impl_t");
414 		return (DCMD_ERR);
415 	}
416 
417 	if (GETMEMBID(addr, &id, db_objset, objset) ||
418 	    GETMEMBID(addr, &id, db, db) ||
419 	    GETMEMBID(addr, &id, db_level, level) ||
420 	    GETMEMBID(addr, &id, db_blkid, blkid)) {
421 		return (WALK_ERR);
422 	}
423 
424 	if (getrefcount(addr, &id, "db_holds", &holds)) {
425 		return (WALK_ERR);
426 	}
427 
428 	if (db.db_object == DMU_META_DNODE_OBJECT)
429 		(void) strcpy(objectname, "mdn");
430 	else
431 		(void) mdb_snprintf(objectname, sizeof (objectname), "%llx",
432 		    (u_longlong_t)db.db_object);
433 
434 	if (blkid == DMU_BONUS_BLKID)
435 		(void) strcpy(blkidname, "bonus");
436 	else
437 		(void) mdb_snprintf(blkidname, sizeof (blkidname), "%llx",
438 		    (u_longlong_t)blkid);
439 
440 	if (objset_name(objset, path)) {
441 		return (WALK_ERR);
442 	}
443 
444 	mdb_printf("%p %8s %1u %9s %2llu %s\n",
445 	    addr, objectname, level, blkidname, holds, path);
446 
447 	return (DCMD_OK);
448 }
449 
450 /* ARGSUSED */
451 static int
452 dbuf_stats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
453 {
454 #define	HISTOSZ 32
455 	uintptr_t dbp;
456 	dmu_buf_impl_t db;
457 	dbuf_hash_table_t ht;
458 	uint64_t bucket, ndbufs;
459 	uint64_t histo[HISTOSZ];
460 	uint64_t histo2[HISTOSZ];
461 	int i, maxidx;
462 
463 	if (mdb_readvar(&ht, "dbuf_hash_table") == -1) {
464 		mdb_warn("failed to read 'dbuf_hash_table'");
465 		return (DCMD_ERR);
466 	}
467 
468 	for (i = 0; i < HISTOSZ; i++) {
469 		histo[i] = 0;
470 		histo2[i] = 0;
471 	}
472 
473 	ndbufs = 0;
474 	for (bucket = 0; bucket < ht.hash_table_mask+1; bucket++) {
475 		int len;
476 
477 		if (mdb_vread(&dbp, sizeof (void *),
478 		    (uintptr_t)(ht.hash_table+bucket)) == -1) {
479 			mdb_warn("failed to read hash bucket %u at %p",
480 			    bucket, ht.hash_table+bucket);
481 			return (DCMD_ERR);
482 		}
483 
484 		len = 0;
485 		while (dbp != 0) {
486 			if (mdb_vread(&db, sizeof (dmu_buf_impl_t),
487 			    dbp) == -1) {
488 				mdb_warn("failed to read dbuf at %p", dbp);
489 				return (DCMD_ERR);
490 			}
491 			dbp = (uintptr_t)db.db_hash_next;
492 			for (i = MIN(len, HISTOSZ - 1); i >= 0; i--)
493 				histo2[i]++;
494 			len++;
495 			ndbufs++;
496 		}
497 
498 		if (len >= HISTOSZ)
499 			len = HISTOSZ-1;
500 		histo[len]++;
501 	}
502 
503 	mdb_printf("hash table has %llu buckets, %llu dbufs "
504 	    "(avg %llu buckets/dbuf)\n",
505 	    ht.hash_table_mask+1, ndbufs,
506 	    (ht.hash_table_mask+1)/ndbufs);
507 
508 	mdb_printf("\n");
509 	maxidx = 0;
510 	for (i = 0; i < HISTOSZ; i++)
511 		if (histo[i] > 0)
512 			maxidx = i;
513 	mdb_printf("hash chain length	number of buckets\n");
514 	for (i = 0; i <= maxidx; i++)
515 		mdb_printf("%u			%llu\n", i, histo[i]);
516 
517 	mdb_printf("\n");
518 	maxidx = 0;
519 	for (i = 0; i < HISTOSZ; i++)
520 		if (histo2[i] > 0)
521 			maxidx = i;
522 	mdb_printf("hash chain depth	number of dbufs\n");
523 	for (i = 0; i <= maxidx; i++)
524 		mdb_printf("%u or more		%llu	%llu%%\n",
525 		    i, histo2[i], histo2[i]*100/ndbufs);
526 
527 
528 	return (DCMD_OK);
529 }
530 
531 #define	CHAIN_END 0xffff
532 /*
533  * ::zap_leaf [-v]
534  *
535  * Print a zap_leaf_phys_t, assumed to be 16k
536  */
537 /* ARGSUSED */
538 static int
539 zap_leaf(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
540 {
541 	char buf[16*1024];
542 	int verbose = B_FALSE;
543 	int four = B_FALSE;
544 	zap_leaf_t l;
545 	zap_leaf_phys_t *zlp = (void *)buf;
546 	int i;
547 
548 	if (mdb_getopts(argc, argv,
549 	    'v', MDB_OPT_SETBITS, TRUE, &verbose,
550 	    '4', MDB_OPT_SETBITS, TRUE, &four,
551 	    NULL) != argc)
552 		return (DCMD_USAGE);
553 
554 	l.l_phys = zlp;
555 	l.l_bs = 14; /* assume 16k blocks */
556 	if (four)
557 		l.l_bs = 12;
558 
559 	if (!(flags & DCMD_ADDRSPEC)) {
560 		return (DCMD_USAGE);
561 	}
562 
563 	if (mdb_vread(buf, sizeof (buf), addr) == -1) {
564 		mdb_warn("failed to read zap_leaf_phys_t at %p", addr);
565 		return (DCMD_ERR);
566 	}
567 
568 	if (zlp->l_hdr.lh_block_type != ZBT_LEAF ||
569 	    zlp->l_hdr.lh_magic != ZAP_LEAF_MAGIC) {
570 		mdb_warn("This does not appear to be a zap_leaf_phys_t");
571 		return (DCMD_ERR);
572 	}
573 
574 	mdb_printf("zap_leaf_phys_t at %p:\n", addr);
575 	mdb_printf("    lh_prefix_len = %u\n", zlp->l_hdr.lh_prefix_len);
576 	mdb_printf("    lh_prefix = %llx\n", zlp->l_hdr.lh_prefix);
577 	mdb_printf("    lh_nentries = %u\n", zlp->l_hdr.lh_nentries);
578 	mdb_printf("    lh_nfree = %u\n", zlp->l_hdr.lh_nfree,
579 	    zlp->l_hdr.lh_nfree * 100 / (ZAP_LEAF_NUMCHUNKS(&l)));
580 	mdb_printf("    lh_freelist = %u\n", zlp->l_hdr.lh_freelist);
581 	mdb_printf("    lh_flags = %x (%s)\n", zlp->l_hdr.lh_flags,
582 	    zlp->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED ?
583 	    "ENTRIES_CDSORTED" : "");
584 
585 	if (verbose) {
586 		mdb_printf(" hash table:\n");
587 		for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++) {
588 			if (zlp->l_hash[i] != CHAIN_END)
589 				mdb_printf("    %u: %u\n", i, zlp->l_hash[i]);
590 		}
591 	}
592 
593 	mdb_printf(" chunks:\n");
594 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) {
595 		/* LINTED: alignment */
596 		zap_leaf_chunk_t *zlc = &ZAP_LEAF_CHUNK(&l, i);
597 		switch (zlc->l_entry.le_type) {
598 		case ZAP_CHUNK_FREE:
599 			if (verbose) {
600 				mdb_printf("    %u: free; lf_next = %u\n",
601 				    i, zlc->l_free.lf_next);
602 			}
603 			break;
604 		case ZAP_CHUNK_ENTRY:
605 			mdb_printf("    %u: entry\n", i);
606 			if (verbose) {
607 				mdb_printf("        le_next = %u\n",
608 				    zlc->l_entry.le_next);
609 			}
610 			mdb_printf("        le_name_chunk = %u\n",
611 			    zlc->l_entry.le_name_chunk);
612 			mdb_printf("        le_name_numints = %u\n",
613 			    zlc->l_entry.le_name_numints);
614 			mdb_printf("        le_value_chunk = %u\n",
615 			    zlc->l_entry.le_value_chunk);
616 			mdb_printf("        le_value_intlen = %u\n",
617 			    zlc->l_entry.le_value_intlen);
618 			mdb_printf("        le_value_numints = %u\n",
619 			    zlc->l_entry.le_value_numints);
620 			mdb_printf("        le_cd = %u\n",
621 			    zlc->l_entry.le_cd);
622 			mdb_printf("        le_hash = %llx\n",
623 			    zlc->l_entry.le_hash);
624 			break;
625 		case ZAP_CHUNK_ARRAY:
626 			mdb_printf("    %u: array \"%s\"\n",
627 			    i, zlc->l_array.la_array);
628 			if (verbose) {
629 				int j;
630 				mdb_printf("        ");
631 				for (j = 0; j < ZAP_LEAF_ARRAY_BYTES; j++) {
632 					mdb_printf("%02x ",
633 					    zlc->l_array.la_array[j]);
634 				}
635 				mdb_printf("\n");
636 			}
637 			if (zlc->l_array.la_next != CHAIN_END) {
638 				mdb_printf("        lf_next = %u\n",
639 				    zlc->l_array.la_next);
640 			}
641 			break;
642 		default:
643 			mdb_printf("    %u: undefined type %u\n",
644 			    zlc->l_entry.le_type);
645 		}
646 	}
647 
648 	return (DCMD_OK);
649 }
650 
651 typedef struct dbufs_data {
652 	mdb_ctf_id_t id;
653 	uint64_t objset;
654 	uint64_t object;
655 	uint64_t level;
656 	uint64_t blkid;
657 	char *osname;
658 } dbufs_data_t;
659 
660 #define	DBUFS_UNSET	(0xbaddcafedeadbeefULL)
661 
662 /* ARGSUSED */
663 static int
664 dbufs_cb(uintptr_t addr, const void *unknown, void *arg)
665 {
666 	dbufs_data_t *data = arg;
667 	uintptr_t objset;
668 	dmu_buf_t db;
669 	uint8_t level;
670 	uint64_t blkid;
671 	char osname[MAXNAMELEN];
672 
673 	if (GETMEMBID(addr, &data->id, db_objset, objset) ||
674 	    GETMEMBID(addr, &data->id, db, db) ||
675 	    GETMEMBID(addr, &data->id, db_level, level) ||
676 	    GETMEMBID(addr, &data->id, db_blkid, blkid)) {
677 		return (WALK_ERR);
678 	}
679 
680 	if ((data->objset == DBUFS_UNSET || data->objset == objset) &&
681 	    (data->osname == NULL || (objset_name(objset, osname) == 0 &&
682 	    strcmp(data->osname, osname) == 0)) &&
683 	    (data->object == DBUFS_UNSET || data->object == db.db_object) &&
684 	    (data->level == DBUFS_UNSET || data->level == level) &&
685 	    (data->blkid == DBUFS_UNSET || data->blkid == blkid)) {
686 		mdb_printf("%#lr\n", addr);
687 	}
688 	return (WALK_NEXT);
689 }
690 
691 /* ARGSUSED */
692 static int
693 dbufs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
694 {
695 	dbufs_data_t data;
696 	char *object = NULL;
697 	char *blkid = NULL;
698 
699 	data.objset = data.object = data.level = data.blkid = DBUFS_UNSET;
700 	data.osname = NULL;
701 
702 	if (mdb_getopts(argc, argv,
703 	    'O', MDB_OPT_UINT64, &data.objset,
704 	    'n', MDB_OPT_STR, &data.osname,
705 	    'o', MDB_OPT_STR, &object,
706 	    'l', MDB_OPT_UINT64, &data.level,
707 	    'b', MDB_OPT_STR, &blkid) != argc) {
708 		return (DCMD_USAGE);
709 	}
710 
711 	if (object) {
712 		if (strcmp(object, "mdn") == 0) {
713 			data.object = DMU_META_DNODE_OBJECT;
714 		} else {
715 			data.object = mdb_strtoull(object);
716 		}
717 	}
718 
719 	if (blkid) {
720 		if (strcmp(blkid, "bonus") == 0) {
721 			data.blkid = DMU_BONUS_BLKID;
722 		} else {
723 			data.blkid = mdb_strtoull(blkid);
724 		}
725 	}
726 
727 	if (mdb_ctf_lookup_by_name("struct dmu_buf_impl", &data.id) == -1) {
728 		mdb_warn("couldn't find struct dmu_buf_impl_t");
729 		return (DCMD_ERR);
730 	}
731 
732 	if (mdb_walk("dmu_buf_impl_t", dbufs_cb, &data) != 0) {
733 		mdb_warn("can't walk dbufs");
734 		return (DCMD_ERR);
735 	}
736 
737 	return (DCMD_OK);
738 }
739 
740 typedef struct abuf_find_data {
741 	dva_t dva;
742 	mdb_ctf_id_t id;
743 } abuf_find_data_t;
744 
745 /* ARGSUSED */
746 static int
747 abuf_find_cb(uintptr_t addr, const void *unknown, void *arg)
748 {
749 	abuf_find_data_t *data = arg;
750 	dva_t dva;
751 
752 	if (GETMEMBID(addr, &data->id, b_dva, dva)) {
753 		return (WALK_ERR);
754 	}
755 
756 	if (dva.dva_word[0] == data->dva.dva_word[0] &&
757 	    dva.dva_word[1] == data->dva.dva_word[1]) {
758 		mdb_printf("%#lr\n", addr);
759 	}
760 	return (WALK_NEXT);
761 }
762 
763 /* ARGSUSED */
764 static int
765 abuf_find(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
766 {
767 	abuf_find_data_t data;
768 	GElf_Sym sym;
769 	int i;
770 	const char *syms[] = {
771 		"ARC_mru",
772 		"ARC_mru_ghost",
773 		"ARC_mfu",
774 		"ARC_mfu_ghost",
775 	};
776 
777 	if (argc != 2)
778 		return (DCMD_USAGE);
779 
780 	for (i = 0; i < 2; i ++) {
781 		switch (argv[i].a_type) {
782 		case MDB_TYPE_STRING:
783 			data.dva.dva_word[i] = mdb_strtoull(argv[i].a_un.a_str);
784 			break;
785 		case MDB_TYPE_IMMEDIATE:
786 			data.dva.dva_word[i] = argv[i].a_un.a_val;
787 			break;
788 		default:
789 			return (DCMD_USAGE);
790 		}
791 	}
792 
793 	if (mdb_ctf_lookup_by_name("struct arc_buf_hdr", &data.id) == -1) {
794 		mdb_warn("couldn't find struct arc_buf_hdr");
795 		return (DCMD_ERR);
796 	}
797 
798 	for (i = 0; i < sizeof (syms) / sizeof (syms[0]); i++) {
799 		if (mdb_lookup_by_name(syms[i], &sym)) {
800 			mdb_warn("can't find symbol %s", syms[i]);
801 			return (DCMD_ERR);
802 		}
803 
804 		if (mdb_pwalk("list", abuf_find_cb, &data, sym.st_value) != 0) {
805 			mdb_warn("can't walk %s", syms[i]);
806 			return (DCMD_ERR);
807 		}
808 	}
809 
810 	return (DCMD_OK);
811 }
812 
813 /*ARGSUSED*/
814 static int
815 arc_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
816 {
817 	kstat_named_t *stats;
818 	GElf_Sym sym;
819 	int nstats, i;
820 	uint_t opt_a = FALSE;
821 	uint_t opt_b = FALSE;
822 	uint_t shift = 0;
823 	const char *suffix;
824 
825 	static const char *bytestats[] = {
826 		"p", "c", "c_min", "c_max", "size", NULL
827 	};
828 
829 	static const char *extras[] = {
830 		"arc_no_grow", "arc_tempreserve",
831 		"arc_meta_used", "arc_meta_limit", "arc_meta_max",
832 		NULL
833 	};
834 
835 	if (mdb_lookup_by_name("arc_stats", &sym) == -1) {
836 		mdb_warn("failed to find 'arc_stats'");
837 		return (DCMD_ERR);
838 	}
839 
840 	stats = mdb_zalloc(sym.st_size, UM_SLEEP | UM_GC);
841 
842 	if (mdb_vread(stats, sym.st_size, sym.st_value) == -1) {
843 		mdb_warn("couldn't read 'arc_stats' at %p", sym.st_value);
844 		return (DCMD_ERR);
845 	}
846 
847 	nstats = sym.st_size / sizeof (kstat_named_t);
848 
849 	/* NB: -a / opt_a are ignored for backwards compatability */
850 	if (mdb_getopts(argc, argv,
851 	    'a', MDB_OPT_SETBITS, TRUE, &opt_a,
852 	    'b', MDB_OPT_SETBITS, TRUE, &opt_b,
853 	    'k', MDB_OPT_SETBITS, 10, &shift,
854 	    'm', MDB_OPT_SETBITS, 20, &shift,
855 	    'g', MDB_OPT_SETBITS, 30, &shift,
856 	    NULL) != argc)
857 		return (DCMD_USAGE);
858 
859 	if (!opt_b && !shift)
860 		shift = 20;
861 
862 	switch (shift) {
863 	case 0:
864 		suffix = "B";
865 		break;
866 	case 10:
867 		suffix = "KB";
868 		break;
869 	case 20:
870 		suffix = "MB";
871 		break;
872 	case 30:
873 		suffix = "GB";
874 		break;
875 	default:
876 		suffix = "XX";
877 	}
878 
879 	for (i = 0; i < nstats; i++) {
880 		int j;
881 		boolean_t bytes = B_FALSE;
882 
883 		for (j = 0; bytestats[j]; j++) {
884 			if (strcmp(stats[i].name, bytestats[j]) == 0) {
885 				bytes = B_TRUE;
886 				break;
887 			}
888 		}
889 
890 		if (bytes) {
891 			mdb_printf("%-25s = %9llu %s\n", stats[i].name,
892 			    stats[i].value.ui64 >> shift, suffix);
893 		} else {
894 			mdb_printf("%-25s = %9llu\n", stats[i].name,
895 			    stats[i].value.ui64);
896 		}
897 	}
898 
899 	for (i = 0; extras[i]; i++) {
900 		uint64_t buf;
901 
902 		if (mdb_lookup_by_name(extras[i], &sym) == -1) {
903 			mdb_warn("failed to find '%s'", extras[i]);
904 			return (DCMD_ERR);
905 		}
906 
907 		if (sym.st_size != sizeof (uint64_t) &&
908 		    sym.st_size != sizeof (uint32_t)) {
909 			mdb_warn("expected scalar for variable '%s'\n",
910 			    extras[i]);
911 			return (DCMD_ERR);
912 		}
913 
914 		if (mdb_vread(&buf, sym.st_size, sym.st_value) == -1) {
915 			mdb_warn("couldn't read '%s'", extras[i]);
916 			return (DCMD_ERR);
917 		}
918 
919 		mdb_printf("%-25s = ", extras[i]);
920 
921 		/* NB: all the 64-bit extras happen to be byte counts */
922 		if (sym.st_size == sizeof (uint64_t))
923 			mdb_printf("%9llu %s\n", buf >> shift, suffix);
924 
925 		if (sym.st_size == sizeof (uint32_t))
926 			mdb_printf("%9d\n", *((uint32_t *)&buf));
927 	}
928 	return (DCMD_OK);
929 }
930 
931 /*
932  * ::spa
933  *
934  * 	-c	Print configuration information as well
935  * 	-v	Print vdev state
936  * 	-e	Print vdev error stats
937  *
938  * Print a summarized spa_t.  When given no arguments, prints out a table of all
939  * active pools on the system.
940  */
941 /* ARGSUSED */
942 static int
943 spa_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
944 {
945 	spa_t spa;
946 	const char *statetab[] = { "ACTIVE", "EXPORTED", "DESTROYED",
947 		"SPARE", "L2CACHE", "UNINIT", "UNAVAIL", "POTENTIAL" };
948 	const char *state;
949 	int config = FALSE;
950 	int vdevs = FALSE;
951 	int errors = FALSE;
952 
953 	if (mdb_getopts(argc, argv,
954 	    'c', MDB_OPT_SETBITS, TRUE, &config,
955 	    'v', MDB_OPT_SETBITS, TRUE, &vdevs,
956 	    'e', MDB_OPT_SETBITS, TRUE, &errors,
957 	    NULL) != argc)
958 		return (DCMD_USAGE);
959 
960 	if (!(flags & DCMD_ADDRSPEC)) {
961 		if (mdb_walk_dcmd("spa", "spa", argc, argv) == -1) {
962 			mdb_warn("can't walk spa");
963 			return (DCMD_ERR);
964 		}
965 
966 		return (DCMD_OK);
967 	}
968 
969 	if (flags & DCMD_PIPE_OUT) {
970 		mdb_printf("%#lr\n", addr);
971 		return (DCMD_OK);
972 	}
973 
974 	if (DCMD_HDRSPEC(flags))
975 		mdb_printf("%<u>%-?s %9s %-*s%</u>\n", "ADDR", "STATE",
976 		    sizeof (uintptr_t) == 4 ? 60 : 52, "NAME");
977 
978 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
979 		mdb_warn("failed to read spa_t at %p", addr);
980 		return (DCMD_ERR);
981 	}
982 
983 	if (spa.spa_state < 0 || spa.spa_state > POOL_STATE_UNAVAIL)
984 		state = "UNKNOWN";
985 	else
986 		state = statetab[spa.spa_state];
987 
988 	mdb_printf("%0?p %9s %s\n", addr, state, spa.spa_name);
989 
990 	if (config) {
991 		mdb_printf("\n");
992 		mdb_inc_indent(4);
993 		if (mdb_call_dcmd("spa_config", addr, flags, 0,
994 		    NULL) != DCMD_OK)
995 			return (DCMD_ERR);
996 		mdb_dec_indent(4);
997 	}
998 
999 	if (vdevs || errors) {
1000 		mdb_arg_t v;
1001 
1002 		v.a_type = MDB_TYPE_STRING;
1003 		v.a_un.a_str = "-e";
1004 
1005 		mdb_printf("\n");
1006 		mdb_inc_indent(4);
1007 		if (mdb_call_dcmd("spa_vdevs", addr, flags, errors ? 1 : 0,
1008 		    &v) != DCMD_OK)
1009 			return (DCMD_ERR);
1010 		mdb_dec_indent(4);
1011 	}
1012 
1013 	return (DCMD_OK);
1014 }
1015 
1016 /*
1017  * ::spa_config
1018  *
1019  * Given a spa_t, print the configuration information stored in spa_config.
1020  * Since it's just an nvlist, format it as an indented list of name=value pairs.
1021  * We simply read the value of spa_config and pass off to ::nvlist.
1022  */
1023 /* ARGSUSED */
1024 static int
1025 spa_print_config(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1026 {
1027 	spa_t spa;
1028 
1029 	if (argc != 0 || !(flags & DCMD_ADDRSPEC))
1030 		return (DCMD_USAGE);
1031 
1032 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1033 		mdb_warn("failed to read spa_t at %p", addr);
1034 		return (DCMD_ERR);
1035 	}
1036 
1037 	if (spa.spa_config == NULL) {
1038 		mdb_printf("(none)\n");
1039 		return (DCMD_OK);
1040 	}
1041 
1042 	return (mdb_call_dcmd("nvlist", (uintptr_t)spa.spa_config, flags,
1043 	    0, NULL));
1044 }
1045 
1046 /*
1047  * ::vdev
1048  *
1049  * Print out a summarized vdev_t, in the following form:
1050  *
1051  * ADDR             STATE	AUX            DESC
1052  * fffffffbcde23df0 HEALTHY	-              /dev/dsk/c0t0d0
1053  *
1054  * If '-r' is specified, recursively visit all children.
1055  *
1056  * With '-e', the statistics associated with the vdev are printed as well.
1057  */
1058 static int
1059 do_print_vdev(uintptr_t addr, int flags, int depth, int stats,
1060     int recursive)
1061 {
1062 	vdev_t vdev;
1063 	char desc[MAXNAMELEN];
1064 	int c, children;
1065 	uintptr_t *child;
1066 	const char *state, *aux;
1067 
1068 	if (mdb_vread(&vdev, sizeof (vdev), (uintptr_t)addr) == -1) {
1069 		mdb_warn("failed to read vdev_t at %p\n", (uintptr_t)addr);
1070 		return (DCMD_ERR);
1071 	}
1072 
1073 	if (flags & DCMD_PIPE_OUT) {
1074 		mdb_printf("%#lr", addr);
1075 	} else {
1076 		if (vdev.vdev_path != NULL) {
1077 			if (mdb_readstr(desc, sizeof (desc),
1078 			    (uintptr_t)vdev.vdev_path) == -1) {
1079 				mdb_warn("failed to read vdev_path at %p\n",
1080 				    vdev.vdev_path);
1081 				return (DCMD_ERR);
1082 			}
1083 		} else if (vdev.vdev_ops != NULL) {
1084 			vdev_ops_t ops;
1085 			if (mdb_vread(&ops, sizeof (ops),
1086 			    (uintptr_t)vdev.vdev_ops) == -1) {
1087 				mdb_warn("failed to read vdev_ops at %p\n",
1088 				    vdev.vdev_ops);
1089 				return (DCMD_ERR);
1090 			}
1091 			(void) strcpy(desc, ops.vdev_op_type);
1092 		} else {
1093 			(void) strcpy(desc, "<unknown>");
1094 		}
1095 
1096 		if (depth == 0 && DCMD_HDRSPEC(flags))
1097 			mdb_printf("%<u>%-?s %-9s %-12s %-*s%</u>\n",
1098 			    "ADDR", "STATE", "AUX",
1099 			    sizeof (uintptr_t) == 4 ? 43 : 35,
1100 			    "DESCRIPTION");
1101 
1102 		mdb_printf("%0?p ", addr);
1103 
1104 		switch (vdev.vdev_state) {
1105 		case VDEV_STATE_CLOSED:
1106 			state = "CLOSED";
1107 			break;
1108 		case VDEV_STATE_OFFLINE:
1109 			state = "OFFLINE";
1110 			break;
1111 		case VDEV_STATE_CANT_OPEN:
1112 			state = "CANT_OPEN";
1113 			break;
1114 		case VDEV_STATE_DEGRADED:
1115 			state = "DEGRADED";
1116 			break;
1117 		case VDEV_STATE_HEALTHY:
1118 			state = "HEALTHY";
1119 			break;
1120 		case VDEV_STATE_REMOVED:
1121 			state = "REMOVED";
1122 			break;
1123 		case VDEV_STATE_FAULTED:
1124 			state = "FAULTED";
1125 			break;
1126 		default:
1127 			state = "UNKNOWN";
1128 			break;
1129 		}
1130 
1131 		switch (vdev.vdev_stat.vs_aux) {
1132 		case VDEV_AUX_NONE:
1133 			aux = "-";
1134 			break;
1135 		case VDEV_AUX_OPEN_FAILED:
1136 			aux = "OPEN_FAILED";
1137 			break;
1138 		case VDEV_AUX_CORRUPT_DATA:
1139 			aux = "CORRUPT_DATA";
1140 			break;
1141 		case VDEV_AUX_NO_REPLICAS:
1142 			aux = "NO_REPLICAS";
1143 			break;
1144 		case VDEV_AUX_BAD_GUID_SUM:
1145 			aux = "BAD_GUID_SUM";
1146 			break;
1147 		case VDEV_AUX_TOO_SMALL:
1148 			aux = "TOO_SMALL";
1149 			break;
1150 		case VDEV_AUX_BAD_LABEL:
1151 			aux = "BAD_LABEL";
1152 			break;
1153 		case VDEV_AUX_VERSION_NEWER:
1154 			aux = "VERS_NEWER";
1155 			break;
1156 		case VDEV_AUX_VERSION_OLDER:
1157 			aux = "VERS_OLDER";
1158 			break;
1159 		case VDEV_AUX_SPARED:
1160 			aux = "SPARED";
1161 			break;
1162 		case VDEV_AUX_ERR_EXCEEDED:
1163 			aux = "ERR_EXCEEDED";
1164 			break;
1165 		case VDEV_AUX_IO_FAILURE:
1166 			aux = "IO_FAILURE";
1167 			break;
1168 		case VDEV_AUX_BAD_LOG:
1169 			aux = "BAD_LOG";
1170 			break;
1171 		case VDEV_AUX_EXTERNAL:
1172 			aux = "EXTERNAL";
1173 			break;
1174 		case VDEV_AUX_SPLIT_POOL:
1175 			aux = "SPLIT_POOL";
1176 			break;
1177 		default:
1178 			aux = "UNKNOWN";
1179 			break;
1180 		}
1181 
1182 		mdb_printf("%-9s %-12s %*s%s\n", state, aux, depth, "", desc);
1183 
1184 		if (stats) {
1185 			vdev_stat_t *vs = &vdev.vdev_stat;
1186 			int i;
1187 
1188 			mdb_inc_indent(4);
1189 			mdb_printf("\n");
1190 			mdb_printf("%<u>       %12s %12s %12s %12s "
1191 			    "%12s%</u>\n", "READ", "WRITE", "FREE", "CLAIM",
1192 			    "IOCTL");
1193 			mdb_printf("OPS     ");
1194 			for (i = 1; i < ZIO_TYPES; i++)
1195 				mdb_printf("%11#llx%s", vs->vs_ops[i],
1196 				    i == ZIO_TYPES - 1 ? "" : "  ");
1197 			mdb_printf("\n");
1198 			mdb_printf("BYTES   ");
1199 			for (i = 1; i < ZIO_TYPES; i++)
1200 				mdb_printf("%11#llx%s", vs->vs_bytes[i],
1201 				    i == ZIO_TYPES - 1 ? "" : "  ");
1202 
1203 
1204 			mdb_printf("\n");
1205 			mdb_printf("EREAD    %10#llx\n", vs->vs_read_errors);
1206 			mdb_printf("EWRITE   %10#llx\n", vs->vs_write_errors);
1207 			mdb_printf("ECKSUM   %10#llx\n",
1208 			    vs->vs_checksum_errors);
1209 			mdb_dec_indent(4);
1210 		}
1211 
1212 		if (stats)
1213 			mdb_printf("\n");
1214 	}
1215 
1216 	children = vdev.vdev_children;
1217 
1218 	if (children == 0 || !recursive)
1219 		return (DCMD_OK);
1220 
1221 	child = mdb_alloc(children * sizeof (void *), UM_SLEEP | UM_GC);
1222 	if (mdb_vread(child, children * sizeof (void *),
1223 	    (uintptr_t)vdev.vdev_child) == -1) {
1224 		mdb_warn("failed to read vdev children at %p", vdev.vdev_child);
1225 		return (DCMD_ERR);
1226 	}
1227 
1228 	for (c = 0; c < children; c++) {
1229 		if (do_print_vdev(child[c], flags, depth + 2, stats,
1230 		    recursive))
1231 			return (DCMD_ERR);
1232 	}
1233 
1234 	return (DCMD_OK);
1235 }
1236 
1237 static int
1238 vdev_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1239 {
1240 	int recursive = FALSE;
1241 	int stats = FALSE;
1242 	uint64_t depth = 0;
1243 
1244 	if (mdb_getopts(argc, argv,
1245 	    'r', MDB_OPT_SETBITS, TRUE, &recursive,
1246 	    'e', MDB_OPT_SETBITS, TRUE, &stats,
1247 	    'd', MDB_OPT_UINT64, &depth,
1248 	    NULL) != argc)
1249 		return (DCMD_USAGE);
1250 
1251 	if (!(flags & DCMD_ADDRSPEC)) {
1252 		mdb_warn("no vdev_t address given\n");
1253 		return (DCMD_ERR);
1254 	}
1255 
1256 	return (do_print_vdev(addr, flags, (int)depth, stats, recursive));
1257 }
1258 
1259 typedef struct metaslab_walk_data {
1260 	uint64_t mw_numvdevs;
1261 	uintptr_t *mw_vdevs;
1262 	int mw_curvdev;
1263 	uint64_t mw_nummss;
1264 	uintptr_t *mw_mss;
1265 	int mw_curms;
1266 } metaslab_walk_data_t;
1267 
1268 static int
1269 metaslab_walk_step(mdb_walk_state_t *wsp)
1270 {
1271 	metaslab_walk_data_t *mw = wsp->walk_data;
1272 	metaslab_t ms;
1273 	uintptr_t msp;
1274 
1275 	if (mw->mw_curvdev >= mw->mw_numvdevs)
1276 		return (WALK_DONE);
1277 
1278 	if (mw->mw_mss == NULL) {
1279 		uintptr_t mssp;
1280 		uintptr_t vdevp;
1281 
1282 		ASSERT(mw->mw_curms == 0);
1283 		ASSERT(mw->mw_nummss == 0);
1284 
1285 		vdevp = mw->mw_vdevs[mw->mw_curvdev];
1286 		if (GETMEMB(vdevp, struct vdev, vdev_ms, mssp) ||
1287 		    GETMEMB(vdevp, struct vdev, vdev_ms_count, mw->mw_nummss)) {
1288 			return (WALK_ERR);
1289 		}
1290 
1291 		mw->mw_mss = mdb_alloc(mw->mw_nummss * sizeof (void*),
1292 		    UM_SLEEP | UM_GC);
1293 		if (mdb_vread(mw->mw_mss, mw->mw_nummss * sizeof (void*),
1294 		    mssp) == -1) {
1295 			mdb_warn("failed to read vdev_ms at %p", mssp);
1296 			return (WALK_ERR);
1297 		}
1298 	}
1299 
1300 	if (mw->mw_curms >= mw->mw_nummss) {
1301 		mw->mw_mss = NULL;
1302 		mw->mw_curms = 0;
1303 		mw->mw_nummss = 0;
1304 		mw->mw_curvdev++;
1305 		return (WALK_NEXT);
1306 	}
1307 
1308 	msp = mw->mw_mss[mw->mw_curms];
1309 	if (mdb_vread(&ms, sizeof (metaslab_t), msp) == -1) {
1310 		mdb_warn("failed to read metaslab_t at %p", msp);
1311 		return (WALK_ERR);
1312 	}
1313 
1314 	mw->mw_curms++;
1315 
1316 	return (wsp->walk_callback(msp, &ms, wsp->walk_cbdata));
1317 }
1318 
1319 /* ARGSUSED */
1320 static int
1321 metaslab_walk_init(mdb_walk_state_t *wsp)
1322 {
1323 	metaslab_walk_data_t *mw;
1324 	uintptr_t root_vdevp;
1325 	uintptr_t childp;
1326 
1327 	if (wsp->walk_addr == NULL) {
1328 		mdb_warn("must supply address of spa_t\n");
1329 		return (WALK_ERR);
1330 	}
1331 
1332 	mw = mdb_zalloc(sizeof (metaslab_walk_data_t), UM_SLEEP | UM_GC);
1333 
1334 	if (GETMEMB(wsp->walk_addr, struct spa, spa_root_vdev, root_vdevp) ||
1335 	    GETMEMB(root_vdevp, struct vdev, vdev_children, mw->mw_numvdevs) ||
1336 	    GETMEMB(root_vdevp, struct vdev, vdev_child, childp)) {
1337 		return (DCMD_ERR);
1338 	}
1339 
1340 	mw->mw_vdevs = mdb_alloc(mw->mw_numvdevs * sizeof (void *),
1341 	    UM_SLEEP | UM_GC);
1342 	if (mdb_vread(mw->mw_vdevs, mw->mw_numvdevs * sizeof (void *),
1343 	    childp) == -1) {
1344 		mdb_warn("failed to read root vdev children at %p", childp);
1345 		return (DCMD_ERR);
1346 	}
1347 
1348 	wsp->walk_data = mw;
1349 
1350 	return (WALK_NEXT);
1351 }
1352 
1353 typedef struct mdb_spa {
1354 	uintptr_t spa_dsl_pool;
1355 	uintptr_t spa_root_vdev;
1356 } mdb_spa_t;
1357 
1358 typedef struct mdb_dsl_dir {
1359 	uintptr_t dd_phys;
1360 	int64_t dd_space_towrite[TXG_SIZE];
1361 } mdb_dsl_dir_t;
1362 
1363 typedef struct mdb_dsl_dir_phys {
1364 	uint64_t dd_used_bytes;
1365 	uint64_t dd_compressed_bytes;
1366 	uint64_t dd_uncompressed_bytes;
1367 } mdb_dsl_dir_phys_t;
1368 
1369 typedef struct mdb_vdev {
1370 	uintptr_t vdev_parent;
1371 	uintptr_t vdev_ms;
1372 	uint64_t vdev_ms_count;
1373 	vdev_stat_t vdev_stat;
1374 } mdb_vdev_t;
1375 
1376 typedef struct mdb_metaslab {
1377 	space_map_t ms_allocmap[TXG_SIZE];
1378 	space_map_t ms_freemap[TXG_SIZE];
1379 	space_map_t ms_map;
1380 	space_map_obj_t ms_smo;
1381 	space_map_obj_t ms_smo_syncing;
1382 } mdb_metaslab_t;
1383 
1384 typedef struct space_data {
1385 	uint64_t ms_allocmap[TXG_SIZE];
1386 	uint64_t ms_freemap[TXG_SIZE];
1387 	uint64_t ms_map;
1388 	uint64_t avail;
1389 	uint64_t nowavail;
1390 } space_data_t;
1391 
1392 /* ARGSUSED */
1393 static int
1394 space_cb(uintptr_t addr, const void *unknown, void *arg)
1395 {
1396 	space_data_t *sd = arg;
1397 	mdb_metaslab_t ms;
1398 
1399 	if (GETMEMB(addr, struct metaslab, ms_allocmap, ms.ms_allocmap) ||
1400 	    GETMEMB(addr, struct metaslab, ms_freemap, ms.ms_freemap) ||
1401 	    GETMEMB(addr, struct metaslab, ms_map, ms.ms_map) ||
1402 	    GETMEMB(addr, struct metaslab, ms_smo, ms.ms_smo) ||
1403 	    GETMEMB(addr, struct metaslab, ms_smo_syncing, ms.ms_smo_syncing)) {
1404 		return (WALK_ERR);
1405 	}
1406 
1407 	sd->ms_allocmap[0] += ms.ms_allocmap[0].sm_space;
1408 	sd->ms_allocmap[1] += ms.ms_allocmap[1].sm_space;
1409 	sd->ms_allocmap[2] += ms.ms_allocmap[2].sm_space;
1410 	sd->ms_allocmap[3] += ms.ms_allocmap[3].sm_space;
1411 	sd->ms_freemap[0] += ms.ms_freemap[0].sm_space;
1412 	sd->ms_freemap[1] += ms.ms_freemap[1].sm_space;
1413 	sd->ms_freemap[2] += ms.ms_freemap[2].sm_space;
1414 	sd->ms_freemap[3] += ms.ms_freemap[3].sm_space;
1415 	sd->ms_map += ms.ms_map.sm_space;
1416 	sd->avail += ms.ms_map.sm_size - ms.ms_smo.smo_alloc;
1417 	sd->nowavail += ms.ms_map.sm_size - ms.ms_smo_syncing.smo_alloc;
1418 
1419 	return (WALK_NEXT);
1420 }
1421 
1422 /*
1423  * ::spa_space [-b]
1424  *
1425  * Given a spa_t, print out it's on-disk space usage and in-core
1426  * estimates of future usage.  If -b is given, print space in bytes.
1427  * Otherwise print in megabytes.
1428  */
1429 /* ARGSUSED */
1430 static int
1431 spa_space(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1432 {
1433 	mdb_spa_t spa;
1434 	uintptr_t dp_root_dir;
1435 	mdb_dsl_dir_t dd;
1436 	mdb_dsl_dir_phys_t dsp;
1437 	uint64_t children;
1438 	uintptr_t childaddr;
1439 	space_data_t sd;
1440 	int shift = 20;
1441 	char *suffix = "M";
1442 	int bits = FALSE;
1443 
1444 	if (mdb_getopts(argc, argv, 'b', MDB_OPT_SETBITS, TRUE, &bits, NULL) !=
1445 	    argc)
1446 		return (DCMD_USAGE);
1447 	if (!(flags & DCMD_ADDRSPEC))
1448 		return (DCMD_USAGE);
1449 
1450 	if (bits) {
1451 		shift = 0;
1452 		suffix = "";
1453 	}
1454 
1455 	if (GETMEMB(addr, struct spa, spa_dsl_pool, spa.spa_dsl_pool) ||
1456 	    GETMEMB(addr, struct spa, spa_root_vdev, spa.spa_root_vdev) ||
1457 	    GETMEMB(spa.spa_root_vdev, struct vdev, vdev_children, children) ||
1458 	    GETMEMB(spa.spa_root_vdev, struct vdev, vdev_child, childaddr) ||
1459 	    GETMEMB(spa.spa_dsl_pool, struct dsl_pool,
1460 	    dp_root_dir, dp_root_dir) ||
1461 	    GETMEMB(dp_root_dir, struct dsl_dir, dd_phys, dd.dd_phys) ||
1462 	    GETMEMB(dp_root_dir, struct dsl_dir,
1463 	    dd_space_towrite, dd.dd_space_towrite) ||
1464 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1465 	    dd_used_bytes, dsp.dd_used_bytes) ||
1466 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1467 	    dd_compressed_bytes, dsp.dd_compressed_bytes) ||
1468 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1469 	    dd_uncompressed_bytes, dsp.dd_uncompressed_bytes)) {
1470 		return (DCMD_ERR);
1471 	}
1472 
1473 	mdb_printf("dd_space_towrite = %llu%s %llu%s %llu%s %llu%s\n",
1474 	    dd.dd_space_towrite[0] >> shift, suffix,
1475 	    dd.dd_space_towrite[1] >> shift, suffix,
1476 	    dd.dd_space_towrite[2] >> shift, suffix,
1477 	    dd.dd_space_towrite[3] >> shift, suffix);
1478 
1479 	mdb_printf("dd_phys.dd_used_bytes = %llu%s\n",
1480 	    dsp.dd_used_bytes >> shift, suffix);
1481 	mdb_printf("dd_phys.dd_compressed_bytes = %llu%s\n",
1482 	    dsp.dd_compressed_bytes >> shift, suffix);
1483 	mdb_printf("dd_phys.dd_uncompressed_bytes = %llu%s\n",
1484 	    dsp.dd_uncompressed_bytes >> shift, suffix);
1485 
1486 	bzero(&sd, sizeof (sd));
1487 	if (mdb_pwalk("metaslab", space_cb, &sd, addr) != 0) {
1488 		mdb_warn("can't walk metaslabs");
1489 		return (DCMD_ERR);
1490 	}
1491 
1492 	mdb_printf("ms_allocmap = %llu%s %llu%s %llu%s %llu%s\n",
1493 	    sd.ms_allocmap[0] >> shift, suffix,
1494 	    sd.ms_allocmap[1] >> shift, suffix,
1495 	    sd.ms_allocmap[2] >> shift, suffix,
1496 	    sd.ms_allocmap[3] >> shift, suffix);
1497 	mdb_printf("ms_freemap = %llu%s %llu%s %llu%s %llu%s\n",
1498 	    sd.ms_freemap[0] >> shift, suffix,
1499 	    sd.ms_freemap[1] >> shift, suffix,
1500 	    sd.ms_freemap[2] >> shift, suffix,
1501 	    sd.ms_freemap[3] >> shift, suffix);
1502 	mdb_printf("ms_map = %llu%s\n", sd.ms_map >> shift, suffix);
1503 	mdb_printf("last synced avail = %llu%s\n", sd.avail >> shift, suffix);
1504 	mdb_printf("current syncing avail = %llu%s\n",
1505 	    sd.nowavail >> shift, suffix);
1506 
1507 	return (DCMD_OK);
1508 }
1509 
1510 /*
1511  * ::spa_verify
1512  *
1513  * Given a spa_t, verify that that the pool is self-consistent.
1514  * Currently, it only checks to make sure that the vdev tree exists.
1515  */
1516 /* ARGSUSED */
1517 static int
1518 spa_verify(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1519 {
1520 	spa_t spa;
1521 
1522 	if (argc != 0 || !(flags & DCMD_ADDRSPEC))
1523 		return (DCMD_USAGE);
1524 
1525 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1526 		mdb_warn("failed to read spa_t at %p", addr);
1527 		return (DCMD_ERR);
1528 	}
1529 
1530 	if (spa.spa_root_vdev == NULL) {
1531 		mdb_printf("no vdev tree present\n");
1532 		return (DCMD_OK);
1533 	}
1534 
1535 	return (DCMD_OK);
1536 }
1537 
1538 static int
1539 spa_print_aux(spa_aux_vdev_t *sav, uint_t flags, mdb_arg_t *v,
1540     const char *name)
1541 {
1542 	uintptr_t *aux;
1543 	size_t len;
1544 	int ret, i;
1545 
1546 	/*
1547 	 * Iterate over aux vdevs and print those out as well.  This is a
1548 	 * little annoying because we don't have a root vdev to pass to ::vdev.
1549 	 * Instead, we print a single line and then call it for each child
1550 	 * vdev.
1551 	 */
1552 	if (sav->sav_count != 0) {
1553 		v[1].a_type = MDB_TYPE_STRING;
1554 		v[1].a_un.a_str = "-d";
1555 		v[2].a_type = MDB_TYPE_IMMEDIATE;
1556 		v[2].a_un.a_val = 2;
1557 
1558 		len = sav->sav_count * sizeof (uintptr_t);
1559 		aux = mdb_alloc(len, UM_SLEEP);
1560 		if (mdb_vread(aux, len,
1561 		    (uintptr_t)sav->sav_vdevs) == -1) {
1562 			mdb_free(aux, len);
1563 			mdb_warn("failed to read l2cache vdevs at %p",
1564 			    sav->sav_vdevs);
1565 			return (DCMD_ERR);
1566 		}
1567 
1568 		mdb_printf("%-?s %-9s %-12s %s\n", "-", "-", "-", name);
1569 
1570 		for (i = 0; i < sav->sav_count; i++) {
1571 			ret = mdb_call_dcmd("vdev", aux[i], flags, 3, v);
1572 			if (ret != DCMD_OK) {
1573 				mdb_free(aux, len);
1574 				return (ret);
1575 			}
1576 		}
1577 
1578 		mdb_free(aux, len);
1579 	}
1580 
1581 	return (0);
1582 }
1583 
1584 /*
1585  * ::spa_vdevs
1586  *
1587  * 	-e	Include error stats
1588  *
1589  * Print out a summarized list of vdevs for the given spa_t.
1590  * This is accomplished by invoking "::vdev -re" on the root vdev, as well as
1591  * iterating over the cache devices.
1592  */
1593 /* ARGSUSED */
1594 static int
1595 spa_vdevs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1596 {
1597 	spa_t spa;
1598 	mdb_arg_t v[3];
1599 	int errors = FALSE;
1600 	int ret;
1601 
1602 	if (mdb_getopts(argc, argv,
1603 	    'e', MDB_OPT_SETBITS, TRUE, &errors,
1604 	    NULL) != argc)
1605 		return (DCMD_USAGE);
1606 
1607 	if (!(flags & DCMD_ADDRSPEC))
1608 		return (DCMD_USAGE);
1609 
1610 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1611 		mdb_warn("failed to read spa_t at %p", addr);
1612 		return (DCMD_ERR);
1613 	}
1614 
1615 	/*
1616 	 * Unitialized spa_t structures can have a NULL root vdev.
1617 	 */
1618 	if (spa.spa_root_vdev == NULL) {
1619 		mdb_printf("no associated vdevs\n");
1620 		return (DCMD_OK);
1621 	}
1622 
1623 	v[0].a_type = MDB_TYPE_STRING;
1624 	v[0].a_un.a_str = errors ? "-re" : "-r";
1625 
1626 	ret = mdb_call_dcmd("vdev", (uintptr_t)spa.spa_root_vdev,
1627 	    flags, 1, v);
1628 	if (ret != DCMD_OK)
1629 		return (ret);
1630 
1631 	if (spa_print_aux(&spa.spa_l2cache, flags, v, "cache") != 0 ||
1632 	    spa_print_aux(&spa.spa_spares, flags, v, "spares") != 0)
1633 		return (DCMD_ERR);
1634 
1635 	return (DCMD_OK);
1636 }
1637 
1638 /*
1639  * ::zio
1640  *
1641  * Print a summary of zio_t and all its children.  This is intended to display a
1642  * zio tree, and hence we only pick the most important pieces of information for
1643  * the main summary.  More detailed information can always be found by doing a
1644  * '::print zio' on the underlying zio_t.  The columns we display are:
1645  *
1646  *	ADDRESS		TYPE	STAGE		WAITER
1647  *
1648  * The 'address' column is indented by one space for each depth level as we
1649  * descend down the tree.
1650  */
1651 
1652 #define	ZIO_MAXINDENT	24
1653 #define	ZIO_MAXWIDTH	(sizeof (uintptr_t) * 2 + ZIO_MAXINDENT)
1654 #define	ZIO_WALK_SELF	0
1655 #define	ZIO_WALK_CHILD	1
1656 #define	ZIO_WALK_PARENT	2
1657 
1658 typedef struct zio_print_args {
1659 	int	zpa_current_depth;
1660 	int	zpa_min_depth;
1661 	int	zpa_max_depth;
1662 	int	zpa_type;
1663 	uint_t	zpa_flags;
1664 } zio_print_args_t;
1665 
1666 static int zio_child_cb(uintptr_t addr, const void *unknown, void *arg);
1667 
1668 static int
1669 zio_print_cb(uintptr_t addr, const void *data, void *priv)
1670 {
1671 	const zio_t *zio = data;
1672 	zio_print_args_t *zpa = priv;
1673 	mdb_ctf_id_t type_enum, stage_enum;
1674 	int indent = zpa->zpa_current_depth;
1675 	const char *type, *stage;
1676 	uintptr_t laddr;
1677 
1678 	if (indent > ZIO_MAXINDENT)
1679 		indent = ZIO_MAXINDENT;
1680 
1681 	if (mdb_ctf_lookup_by_name("enum zio_type", &type_enum) == -1 ||
1682 	    mdb_ctf_lookup_by_name("enum zio_stage", &stage_enum) == -1) {
1683 		mdb_warn("failed to lookup zio enums");
1684 		return (WALK_ERR);
1685 	}
1686 
1687 	if ((type = mdb_ctf_enum_name(type_enum, zio->io_type)) != NULL)
1688 		type += sizeof ("ZIO_TYPE_") - 1;
1689 	else
1690 		type = "?";
1691 
1692 	if ((stage = mdb_ctf_enum_name(stage_enum, zio->io_stage)) != NULL)
1693 		stage += sizeof ("ZIO_STAGE_") - 1;
1694 	else
1695 		stage = "?";
1696 
1697 	if (zpa->zpa_current_depth >= zpa->zpa_min_depth) {
1698 		if (zpa->zpa_flags & DCMD_PIPE_OUT) {
1699 			mdb_printf("%?p\n", addr);
1700 		} else {
1701 			mdb_printf("%*s%-*p %-5s %-16s ", indent, "",
1702 			    ZIO_MAXWIDTH - indent, addr, type, stage);
1703 			if (zio->io_waiter)
1704 				mdb_printf("%?p\n", zio->io_waiter);
1705 			else
1706 				mdb_printf("-\n");
1707 		}
1708 	}
1709 
1710 	if (zpa->zpa_current_depth >= zpa->zpa_max_depth)
1711 		return (WALK_NEXT);
1712 
1713 	if (zpa->zpa_type == ZIO_WALK_PARENT)
1714 		laddr = addr + OFFSETOF(zio_t, io_parent_list);
1715 	else
1716 		laddr = addr + OFFSETOF(zio_t, io_child_list);
1717 
1718 	zpa->zpa_current_depth++;
1719 	if (mdb_pwalk("list", zio_child_cb, zpa, laddr) != 0) {
1720 		mdb_warn("failed to walk zio_t children at %p\n", laddr);
1721 		return (WALK_ERR);
1722 	}
1723 	zpa->zpa_current_depth--;
1724 
1725 	return (WALK_NEXT);
1726 }
1727 
1728 /* ARGSUSED */
1729 static int
1730 zio_child_cb(uintptr_t addr, const void *unknown, void *arg)
1731 {
1732 	zio_link_t zl;
1733 	zio_t zio;
1734 	uintptr_t ziop;
1735 	zio_print_args_t *zpa = arg;
1736 
1737 	if (mdb_vread(&zl, sizeof (zl), addr) == -1) {
1738 		mdb_warn("failed to read zio_link_t at %p", addr);
1739 		return (WALK_ERR);
1740 	}
1741 
1742 	if (zpa->zpa_type == ZIO_WALK_PARENT)
1743 		ziop = (uintptr_t)zl.zl_parent;
1744 	else
1745 		ziop = (uintptr_t)zl.zl_child;
1746 
1747 	if (mdb_vread(&zio, sizeof (zio_t), ziop) == -1) {
1748 		mdb_warn("failed to read zio_t at %p", ziop);
1749 		return (WALK_ERR);
1750 	}
1751 
1752 	return (zio_print_cb(ziop, &zio, arg));
1753 }
1754 
1755 /* ARGSUSED */
1756 static int
1757 zio_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1758 {
1759 	zio_t zio;
1760 	zio_print_args_t zpa = { 0 };
1761 
1762 	if (!(flags & DCMD_ADDRSPEC))
1763 		return (DCMD_USAGE);
1764 
1765 	if (mdb_getopts(argc, argv,
1766 	    'r', MDB_OPT_SETBITS, INT_MAX, &zpa.zpa_max_depth,
1767 	    'c', MDB_OPT_SETBITS, ZIO_WALK_CHILD, &zpa.zpa_type,
1768 	    'p', MDB_OPT_SETBITS, ZIO_WALK_PARENT, &zpa.zpa_type,
1769 	    NULL) != argc)
1770 		return (DCMD_USAGE);
1771 
1772 	zpa.zpa_flags = flags;
1773 	if (zpa.zpa_max_depth != 0) {
1774 		if (zpa.zpa_type == ZIO_WALK_SELF)
1775 			zpa.zpa_type = ZIO_WALK_CHILD;
1776 	} else if (zpa.zpa_type != ZIO_WALK_SELF) {
1777 		zpa.zpa_min_depth = 1;
1778 		zpa.zpa_max_depth = 1;
1779 	}
1780 
1781 	if (mdb_vread(&zio, sizeof (zio_t), addr) == -1) {
1782 		mdb_warn("failed to read zio_t at %p", addr);
1783 		return (DCMD_ERR);
1784 	}
1785 
1786 	if (!(flags & DCMD_PIPE_OUT) && DCMD_HDRSPEC(flags))
1787 		mdb_printf("%<u>%-*s %-5s %-16s %-?s%</u>\n", ZIO_MAXWIDTH,
1788 		    "ADDRESS", "TYPE", "STAGE", "WAITER");
1789 
1790 	if (zio_print_cb(addr, &zio, &zpa) != WALK_NEXT)
1791 		return (DCMD_ERR);
1792 
1793 	return (DCMD_OK);
1794 }
1795 
1796 /*
1797  * [addr]::zio_state
1798  *
1799  * Print a summary of all zio_t structures on the system, or for a particular
1800  * pool.  This is equivalent to '::walk zio_root | ::zio'.
1801  */
1802 /*ARGSUSED*/
1803 static int
1804 zio_state(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1805 {
1806 	/*
1807 	 * MDB will remember the last address of the pipeline, so if we don't
1808 	 * zero this we'll end up trying to walk zio structures for a
1809 	 * non-existent spa_t.
1810 	 */
1811 	if (!(flags & DCMD_ADDRSPEC))
1812 		addr = 0;
1813 
1814 	return (mdb_pwalk_dcmd("zio_root", "zio", argc, argv, addr));
1815 }
1816 
1817 typedef struct txg_list_walk_data {
1818 	uintptr_t lw_head[TXG_SIZE];
1819 	int	lw_txgoff;
1820 	int	lw_maxoff;
1821 	size_t	lw_offset;
1822 	void	*lw_obj;
1823 } txg_list_walk_data_t;
1824 
1825 static int
1826 txg_list_walk_init_common(mdb_walk_state_t *wsp, int txg, int maxoff)
1827 {
1828 	txg_list_walk_data_t *lwd;
1829 	txg_list_t list;
1830 	int i;
1831 
1832 	lwd = mdb_alloc(sizeof (txg_list_walk_data_t), UM_SLEEP | UM_GC);
1833 	if (mdb_vread(&list, sizeof (txg_list_t), wsp->walk_addr) == -1) {
1834 		mdb_warn("failed to read txg_list_t at %#lx", wsp->walk_addr);
1835 		return (WALK_ERR);
1836 	}
1837 
1838 	for (i = 0; i < TXG_SIZE; i++)
1839 		lwd->lw_head[i] = (uintptr_t)list.tl_head[i];
1840 	lwd->lw_offset = list.tl_offset;
1841 	lwd->lw_obj = mdb_alloc(lwd->lw_offset + sizeof (txg_node_t),
1842 	    UM_SLEEP | UM_GC);
1843 	lwd->lw_txgoff = txg;
1844 	lwd->lw_maxoff = maxoff;
1845 
1846 	wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
1847 	wsp->walk_data = lwd;
1848 
1849 	return (WALK_NEXT);
1850 }
1851 
1852 static int
1853 txg_list_walk_init(mdb_walk_state_t *wsp)
1854 {
1855 	return (txg_list_walk_init_common(wsp, 0, TXG_SIZE-1));
1856 }
1857 
1858 static int
1859 txg_list0_walk_init(mdb_walk_state_t *wsp)
1860 {
1861 	return (txg_list_walk_init_common(wsp, 0, 0));
1862 }
1863 
1864 static int
1865 txg_list1_walk_init(mdb_walk_state_t *wsp)
1866 {
1867 	return (txg_list_walk_init_common(wsp, 1, 1));
1868 }
1869 
1870 static int
1871 txg_list2_walk_init(mdb_walk_state_t *wsp)
1872 {
1873 	return (txg_list_walk_init_common(wsp, 2, 2));
1874 }
1875 
1876 static int
1877 txg_list3_walk_init(mdb_walk_state_t *wsp)
1878 {
1879 	return (txg_list_walk_init_common(wsp, 3, 3));
1880 }
1881 
1882 static int
1883 txg_list_walk_step(mdb_walk_state_t *wsp)
1884 {
1885 	txg_list_walk_data_t *lwd = wsp->walk_data;
1886 	uintptr_t addr;
1887 	txg_node_t *node;
1888 	int status;
1889 
1890 	while (wsp->walk_addr == NULL && lwd->lw_txgoff < lwd->lw_maxoff) {
1891 		lwd->lw_txgoff++;
1892 		wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
1893 	}
1894 
1895 	if (wsp->walk_addr == NULL)
1896 		return (WALK_DONE);
1897 
1898 	addr = wsp->walk_addr - lwd->lw_offset;
1899 
1900 	if (mdb_vread(lwd->lw_obj,
1901 	    lwd->lw_offset + sizeof (txg_node_t), addr) == -1) {
1902 		mdb_warn("failed to read list element at %#lx", addr);
1903 		return (WALK_ERR);
1904 	}
1905 
1906 	status = wsp->walk_callback(addr, lwd->lw_obj, wsp->walk_cbdata);
1907 	node = (txg_node_t *)((uintptr_t)lwd->lw_obj + lwd->lw_offset);
1908 	wsp->walk_addr = (uintptr_t)node->tn_next[lwd->lw_txgoff];
1909 
1910 	return (status);
1911 }
1912 
1913 /*
1914  * ::walk spa
1915  *
1916  * Walk all named spa_t structures in the namespace.  This is nothing more than
1917  * a layered avl walk.
1918  */
1919 static int
1920 spa_walk_init(mdb_walk_state_t *wsp)
1921 {
1922 	GElf_Sym sym;
1923 
1924 	if (wsp->walk_addr != NULL) {
1925 		mdb_warn("spa walk only supports global walks\n");
1926 		return (WALK_ERR);
1927 	}
1928 
1929 	if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "spa_namespace_avl", &sym) == -1) {
1930 		mdb_warn("failed to find symbol 'spa_namespace_avl'");
1931 		return (WALK_ERR);
1932 	}
1933 
1934 	wsp->walk_addr = (uintptr_t)sym.st_value;
1935 
1936 	if (mdb_layered_walk("avl", wsp) == -1) {
1937 		mdb_warn("failed to walk 'avl'\n");
1938 		return (WALK_ERR);
1939 	}
1940 
1941 	return (WALK_NEXT);
1942 }
1943 
1944 static int
1945 spa_walk_step(mdb_walk_state_t *wsp)
1946 {
1947 	spa_t	spa;
1948 
1949 	if (mdb_vread(&spa, sizeof (spa), wsp->walk_addr) == -1) {
1950 		mdb_warn("failed to read spa_t at %p", wsp->walk_addr);
1951 		return (WALK_ERR);
1952 	}
1953 
1954 	return (wsp->walk_callback(wsp->walk_addr, &spa, wsp->walk_cbdata));
1955 }
1956 
1957 /*
1958  * [addr]::walk zio
1959  *
1960  * Walk all active zio_t structures on the system.  This is simply a layered
1961  * walk on top of ::walk zio_cache, with the optional ability to limit the
1962  * structures to a particular pool.
1963  */
1964 static int
1965 zio_walk_init(mdb_walk_state_t *wsp)
1966 {
1967 	wsp->walk_data = (void *)wsp->walk_addr;
1968 
1969 	if (mdb_layered_walk("zio_cache", wsp) == -1) {
1970 		mdb_warn("failed to walk 'zio_cache'\n");
1971 		return (WALK_ERR);
1972 	}
1973 
1974 	return (WALK_NEXT);
1975 }
1976 
1977 static int
1978 zio_walk_step(mdb_walk_state_t *wsp)
1979 {
1980 	zio_t zio;
1981 
1982 	if (mdb_vread(&zio, sizeof (zio), wsp->walk_addr) == -1) {
1983 		mdb_warn("failed to read zio_t at %p", wsp->walk_addr);
1984 		return (WALK_ERR);
1985 	}
1986 
1987 	if (wsp->walk_data != NULL && wsp->walk_data != zio.io_spa)
1988 		return (WALK_NEXT);
1989 
1990 	return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
1991 }
1992 
1993 /*
1994  * [addr]::walk zio_root
1995  *
1996  * Walk only root zio_t structures, optionally for a particular spa_t.
1997  */
1998 static int
1999 zio_walk_root_step(mdb_walk_state_t *wsp)
2000 {
2001 	zio_t zio;
2002 
2003 	if (mdb_vread(&zio, sizeof (zio), wsp->walk_addr) == -1) {
2004 		mdb_warn("failed to read zio_t at %p", wsp->walk_addr);
2005 		return (WALK_ERR);
2006 	}
2007 
2008 	if (wsp->walk_data != NULL && wsp->walk_data != zio.io_spa)
2009 		return (WALK_NEXT);
2010 
2011 	/* If the parent list is not empty, ignore */
2012 	if (zio.io_parent_list.list_head.list_next !=
2013 	    &((zio_t *)wsp->walk_addr)->io_parent_list.list_head)
2014 		return (WALK_NEXT);
2015 
2016 	return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
2017 }
2018 
2019 #define	NICENUM_BUFLEN 6
2020 
2021 static int
2022 snprintfrac(char *buf, int len,
2023     uint64_t numerator, uint64_t denom, int frac_digits)
2024 {
2025 	int mul = 1;
2026 	int whole, frac, i;
2027 
2028 	for (i = frac_digits; i; i--)
2029 		mul *= 10;
2030 	whole = numerator / denom;
2031 	frac = mul * numerator / denom - mul * whole;
2032 	return (mdb_snprintf(buf, len, "%u.%0*u", whole, frac_digits, frac));
2033 }
2034 
2035 static void
2036 mdb_nicenum(uint64_t num, char *buf)
2037 {
2038 	uint64_t n = num;
2039 	int index = 0;
2040 	char *u;
2041 
2042 	while (n >= 1024) {
2043 		n = (n + (1024 / 2)) / 1024; /* Round up or down */
2044 		index++;
2045 	}
2046 
2047 	u = &" \0K\0M\0G\0T\0P\0E\0"[index*2];
2048 
2049 	if (index == 0) {
2050 		(void) mdb_snprintf(buf, NICENUM_BUFLEN, "%llu",
2051 		    (u_longlong_t)n);
2052 	} else if (n < 10 && (num & (num - 1)) != 0) {
2053 		(void) snprintfrac(buf, NICENUM_BUFLEN,
2054 		    num, 1ULL << 10 * index, 2);
2055 		strcat(buf, u);
2056 	} else if (n < 100 && (num & (num - 1)) != 0) {
2057 		(void) snprintfrac(buf, NICENUM_BUFLEN,
2058 		    num, 1ULL << 10 * index, 1);
2059 		strcat(buf, u);
2060 	} else {
2061 		(void) mdb_snprintf(buf, NICENUM_BUFLEN, "%llu%s",
2062 		    (u_longlong_t)n, u);
2063 	}
2064 }
2065 
2066 /*
2067  * ::zfs_blkstats
2068  *
2069  * 	-v	print verbose per-level information
2070  *
2071  */
2072 static int
2073 zfs_blkstats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2074 {
2075 	boolean_t verbose = B_FALSE;
2076 	zfs_all_blkstats_t stats;
2077 	dmu_object_type_t t;
2078 	zfs_blkstat_t *tzb;
2079 	uint64_t ditto;
2080 	dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES + 10];
2081 	/* +10 in case it grew */
2082 
2083 	if (mdb_readvar(&dmu_ot, "dmu_ot") == -1) {
2084 		mdb_warn("failed to read 'dmu_ot'");
2085 		return (DCMD_ERR);
2086 	}
2087 
2088 	if (mdb_getopts(argc, argv,
2089 	    'v', MDB_OPT_SETBITS, TRUE, &verbose,
2090 	    NULL) != argc)
2091 		return (DCMD_USAGE);
2092 
2093 	if (!(flags & DCMD_ADDRSPEC))
2094 		return (DCMD_USAGE);
2095 
2096 	if (GETMEMB(addr, struct spa, spa_dsl_pool, addr) ||
2097 	    GETMEMB(addr, struct dsl_pool, dp_blkstats, addr) ||
2098 	    mdb_vread(&stats, sizeof (zfs_all_blkstats_t), addr) == -1) {
2099 		mdb_warn("failed to read data at %p;", addr);
2100 		mdb_printf("maybe no stats? run \"zpool scrub\" first.");
2101 		return (DCMD_ERR);
2102 	}
2103 
2104 	tzb = &stats.zab_type[DN_MAX_LEVELS][DMU_OT_NUMTYPES];
2105 	if (tzb->zb_gangs != 0) {
2106 		mdb_printf("Ganged blocks: %llu\n",
2107 		    (longlong_t)tzb->zb_gangs);
2108 	}
2109 
2110 	ditto = tzb->zb_ditto_2_of_2_samevdev + tzb->zb_ditto_2_of_3_samevdev +
2111 	    tzb->zb_ditto_3_of_3_samevdev;
2112 	if (ditto != 0) {
2113 		mdb_printf("Dittoed blocks on same vdev: %llu\n",
2114 		    (longlong_t)ditto);
2115 	}
2116 
2117 	mdb_printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
2118 	    "\t  avg\t comp\t%%Total\tType\n");
2119 
2120 	for (t = 0; t <= DMU_OT_NUMTYPES; t++) {
2121 		char csize[NICENUM_BUFLEN], lsize[NICENUM_BUFLEN];
2122 		char psize[NICENUM_BUFLEN], asize[NICENUM_BUFLEN];
2123 		char avg[NICENUM_BUFLEN];
2124 		char comp[NICENUM_BUFLEN], pct[NICENUM_BUFLEN];
2125 		char typename[64];
2126 		int l;
2127 
2128 
2129 		if (t == DMU_OT_DEFERRED)
2130 			strcpy(typename, "deferred free");
2131 		else if (t == DMU_OT_TOTAL)
2132 			strcpy(typename, "Total");
2133 		else if (mdb_readstr(typename, sizeof (typename),
2134 		    (uintptr_t)dmu_ot[t].ot_name) == -1) {
2135 			mdb_warn("failed to read type name");
2136 			return (DCMD_ERR);
2137 		}
2138 
2139 		if (stats.zab_type[DN_MAX_LEVELS][t].zb_asize == 0)
2140 			continue;
2141 
2142 		for (l = -1; l < DN_MAX_LEVELS; l++) {
2143 			int level = (l == -1 ? DN_MAX_LEVELS : l);
2144 			zfs_blkstat_t *zb = &stats.zab_type[level][t];
2145 
2146 			if (zb->zb_asize == 0)
2147 				continue;
2148 
2149 			/*
2150 			 * Don't print each level unless requested.
2151 			 */
2152 			if (!verbose && level != DN_MAX_LEVELS)
2153 				continue;
2154 
2155 			/*
2156 			 * If all the space is level 0, don't print the
2157 			 * level 0 separately.
2158 			 */
2159 			if (level == 0 && zb->zb_asize ==
2160 			    stats.zab_type[DN_MAX_LEVELS][t].zb_asize)
2161 				continue;
2162 
2163 			mdb_nicenum(zb->zb_count, csize);
2164 			mdb_nicenum(zb->zb_lsize, lsize);
2165 			mdb_nicenum(zb->zb_psize, psize);
2166 			mdb_nicenum(zb->zb_asize, asize);
2167 			mdb_nicenum(zb->zb_asize / zb->zb_count, avg);
2168 			(void) snprintfrac(comp, NICENUM_BUFLEN,
2169 			    zb->zb_lsize, zb->zb_psize, 2);
2170 			(void) snprintfrac(pct, NICENUM_BUFLEN,
2171 			    100 * zb->zb_asize, tzb->zb_asize, 2);
2172 
2173 			mdb_printf("%6s\t%5s\t%5s\t%5s\t%5s"
2174 			    "\t%5s\t%6s\t",
2175 			    csize, lsize, psize, asize, avg, comp, pct);
2176 
2177 			if (level == DN_MAX_LEVELS)
2178 				mdb_printf("%s\n", typename);
2179 			else
2180 				mdb_printf("  L%d %s\n",
2181 				    level, typename);
2182 		}
2183 	}
2184 
2185 	return (DCMD_OK);
2186 }
2187 
2188 /* ARGSUSED */
2189 static int
2190 reference_cb(uintptr_t addr, const void *ignored, void *arg)
2191 {
2192 	static int gotid;
2193 	static mdb_ctf_id_t ref_id;
2194 	uintptr_t ref_holder;
2195 	uintptr_t ref_removed;
2196 	uint64_t ref_number;
2197 	boolean_t holder_is_str;
2198 	char holder_str[128];
2199 	boolean_t removed = (boolean_t)arg;
2200 
2201 	if (!gotid) {
2202 		if (mdb_ctf_lookup_by_name("struct reference", &ref_id) == -1) {
2203 			mdb_warn("couldn't find struct reference");
2204 			return (WALK_ERR);
2205 		}
2206 		gotid = TRUE;
2207 	}
2208 
2209 	if (GETMEMBID(addr, &ref_id, ref_holder, ref_holder) ||
2210 	    GETMEMBID(addr, &ref_id, ref_removed, ref_removed) ||
2211 	    GETMEMBID(addr, &ref_id, ref_number, ref_number))
2212 		return (WALK_ERR);
2213 
2214 	if (mdb_readstr(holder_str, sizeof (holder_str), ref_holder) != -1) {
2215 		char *cp;
2216 		holder_is_str = B_TRUE;
2217 		for (cp = holder_str; *cp; cp++) {
2218 			if (!isprint(*cp)) {
2219 				holder_is_str = B_FALSE;
2220 				break;
2221 			}
2222 		}
2223 	} else {
2224 		holder_is_str = B_FALSE;
2225 	}
2226 
2227 	if (removed)
2228 		mdb_printf("removed ");
2229 	mdb_printf("reference ");
2230 	if (ref_number != 1)
2231 		mdb_printf("with count=%llu ", ref_number);
2232 	mdb_printf("with tag %p", (void*)ref_holder);
2233 	if (holder_is_str)
2234 		mdb_printf(" \"%s\"", holder_str);
2235 	mdb_printf(", held at:\n");
2236 
2237 	(void) mdb_call_dcmd("whatis", addr, DCMD_ADDRSPEC, 0, NULL);
2238 
2239 	if (removed) {
2240 		mdb_printf("removed at:\n");
2241 		(void) mdb_call_dcmd("whatis", ref_removed,
2242 		    DCMD_ADDRSPEC, 0, NULL);
2243 	}
2244 
2245 	mdb_printf("\n");
2246 
2247 	return (WALK_NEXT);
2248 }
2249 
2250 /* ARGSUSED */
2251 static int
2252 refcount(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2253 {
2254 	uint64_t rc_count, rc_removed_count;
2255 	uintptr_t rc_list, rc_removed;
2256 	static int gotid;
2257 	static mdb_ctf_id_t rc_id;
2258 	ulong_t off;
2259 
2260 	if (!(flags & DCMD_ADDRSPEC))
2261 		return (DCMD_USAGE);
2262 
2263 	if (!gotid) {
2264 		if (mdb_ctf_lookup_by_name("struct refcount", &rc_id) == -1) {
2265 			mdb_warn("couldn't find struct refcount");
2266 			return (DCMD_ERR);
2267 		}
2268 		gotid = TRUE;
2269 	}
2270 
2271 	if (GETMEMBID(addr, &rc_id, rc_count, rc_count) ||
2272 	    GETMEMBID(addr, &rc_id, rc_removed_count, rc_removed_count))
2273 		return (DCMD_ERR);
2274 
2275 	mdb_printf("refcount_t at %p has %llu current holds, "
2276 	    "%llu recently released holds\n",
2277 	    addr, (longlong_t)rc_count, (longlong_t)rc_removed_count);
2278 
2279 	if (rc_count > 0)
2280 		mdb_printf("current holds:\n");
2281 	if (mdb_ctf_offsetof(rc_id, "rc_list", &off) == -1)
2282 		return (DCMD_ERR);
2283 	rc_list = addr + off/NBBY;
2284 	mdb_pwalk("list", reference_cb, (void*)B_FALSE, rc_list);
2285 
2286 	if (rc_removed_count > 0)
2287 		mdb_printf("released holds:\n");
2288 	if (mdb_ctf_offsetof(rc_id, "rc_removed", &off) == -1)
2289 		return (DCMD_ERR);
2290 	rc_removed = addr + off/NBBY;
2291 	mdb_pwalk("list", reference_cb, (void*)B_TRUE, rc_removed);
2292 
2293 	return (DCMD_OK);
2294 }
2295 
2296 /* ARGSUSED */
2297 static int
2298 sa_attr_table(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2299 {
2300 	sa_attr_table_t *table;
2301 	sa_os_t sa_os;
2302 	char *name;
2303 	int i;
2304 
2305 	if (mdb_vread(&sa_os, sizeof (sa_os_t), addr) == -1) {
2306 		mdb_warn("failed to read sa_os at %p", addr);
2307 		return (DCMD_ERR);
2308 	}
2309 
2310 	table = mdb_alloc(sizeof (sa_attr_table_t) * sa_os.sa_num_attrs,
2311 	    UM_SLEEP | UM_GC);
2312 	name = mdb_alloc(MAXPATHLEN, UM_SLEEP | UM_GC);
2313 
2314 	if (mdb_vread(table, sizeof (sa_attr_table_t) * sa_os.sa_num_attrs,
2315 	    (uintptr_t)sa_os.sa_attr_table) == -1) {
2316 		mdb_warn("failed to read sa_os at %p", addr);
2317 		return (DCMD_ERR);
2318 	}
2319 
2320 	mdb_printf("%<u>%-10s %-10s %-10s %-10s %s%</u>\n",
2321 	    "ATTR ID", "REGISTERED", "LENGTH", "BSWAP", "NAME");
2322 	for (i = 0; i != sa_os.sa_num_attrs; i++) {
2323 		mdb_readstr(name, MAXPATHLEN, (uintptr_t)table[i].sa_name);
2324 		mdb_printf("%5x   %8x %8x %8x          %-s\n",
2325 		    (int)table[i].sa_attr, (int)table[i].sa_registered,
2326 		    (int)table[i].sa_length, table[i].sa_byteswap, name);
2327 	}
2328 
2329 	return (DCMD_OK);
2330 }
2331 
2332 static int
2333 sa_get_off_table(uintptr_t addr, uint32_t **off_tab, int attr_count)
2334 {
2335 	uintptr_t idx_table;
2336 
2337 	if (GETMEMB(addr, struct sa_idx_tab, sa_idx_tab, idx_table)) {
2338 		mdb_printf("can't find offset table in sa_idx_tab\n");
2339 		return (-1);
2340 	}
2341 
2342 	*off_tab = mdb_alloc(attr_count * sizeof (uint32_t),
2343 	    UM_SLEEP | UM_GC);
2344 
2345 	if (mdb_vread(*off_tab,
2346 	    attr_count * sizeof (uint32_t), idx_table) == -1) {
2347 		mdb_warn("failed to attribute offset table %p", idx_table);
2348 		return (-1);
2349 	}
2350 
2351 	return (DCMD_OK);
2352 }
2353 
2354 /*ARGSUSED*/
2355 static int
2356 sa_attr_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2357 {
2358 	uint32_t *offset_tab;
2359 	int attr_count;
2360 	uint64_t attr_id;
2361 	uintptr_t attr_addr;
2362 	uintptr_t bonus_tab, spill_tab;
2363 	uintptr_t db_bonus, db_spill;
2364 	uintptr_t os, os_sa;
2365 	uintptr_t db_data;
2366 
2367 	if (argc != 1)
2368 		return (DCMD_USAGE);
2369 
2370 	if (argv[0].a_type == MDB_TYPE_STRING)
2371 		attr_id = mdb_strtoull(argv[0].a_un.a_str);
2372 	else
2373 		return (DCMD_USAGE);
2374 
2375 	if (GETMEMB(addr, struct sa_handle, sa_bonus_tab, bonus_tab) ||
2376 	    GETMEMB(addr, struct sa_handle, sa_spill_tab, spill_tab) ||
2377 	    GETMEMB(addr, struct sa_handle, sa_os, os) ||
2378 	    GETMEMB(addr, struct sa_handle, sa_bonus, db_bonus) ||
2379 	    GETMEMB(addr, struct sa_handle, sa_spill, db_spill)) {
2380 		mdb_printf("Can't find necessary information in sa_handle "
2381 		    "in sa_handle\n");
2382 		return (DCMD_ERR);
2383 	}
2384 
2385 	if (GETMEMB(os, struct objset, os_sa, os_sa)) {
2386 		mdb_printf("Can't find os_sa in objset\n");
2387 		return (DCMD_ERR);
2388 	}
2389 
2390 	if (GETMEMB(os_sa, struct sa_os, sa_num_attrs, attr_count)) {
2391 		mdb_printf("Can't find sa_num_attrs\n");
2392 		return (DCMD_ERR);
2393 	}
2394 
2395 	if (attr_id > attr_count) {
2396 		mdb_printf("attribute id number is out of range\n");
2397 		return (DCMD_ERR);
2398 	}
2399 
2400 	if (bonus_tab) {
2401 		if (sa_get_off_table(bonus_tab, &offset_tab,
2402 		    attr_count) == -1) {
2403 			return (DCMD_ERR);
2404 		}
2405 
2406 		if (GETMEMB(db_bonus, struct dmu_buf, db_data, db_data)) {
2407 			mdb_printf("can't find db_data in bonus dbuf\n");
2408 			return (DCMD_ERR);
2409 		}
2410 	}
2411 
2412 	if (bonus_tab && !TOC_ATTR_PRESENT(offset_tab[attr_id]) &&
2413 	    spill_tab == NULL) {
2414 		mdb_printf("Attribute does not exist\n");
2415 		return (DCMD_ERR);
2416 	} else if (!TOC_ATTR_PRESENT(offset_tab[attr_id]) && spill_tab) {
2417 		if (sa_get_off_table(spill_tab, &offset_tab,
2418 		    attr_count) == -1) {
2419 			return (DCMD_ERR);
2420 		}
2421 		if (GETMEMB(db_spill, struct dmu_buf, db_data, db_data)) {
2422 			mdb_printf("can't find db_data in spill dbuf\n");
2423 			return (DCMD_ERR);
2424 		}
2425 		if (!TOC_ATTR_PRESENT(offset_tab[attr_id])) {
2426 			mdb_printf("Attribute does not exist\n");
2427 			return (DCMD_ERR);
2428 		}
2429 	}
2430 	attr_addr = db_data + TOC_OFF(offset_tab[attr_id]);
2431 	mdb_printf("%p\n", attr_addr);
2432 	return (DCMD_OK);
2433 }
2434 
2435 /* ARGSUSED */
2436 static int
2437 zfs_ace_print_common(uintptr_t addr, uint_t flags,
2438     uint64_t id, uint32_t access_mask, uint16_t ace_flags,
2439     uint16_t ace_type, int verbose)
2440 {
2441 	if (DCMD_HDRSPEC(flags) && !verbose)
2442 		mdb_printf("%<u>%-?s %-8s %-8s %-8s %s%</u>\n",
2443 		    "ADDR", "FLAGS", "MASK", "TYPE", "ID");
2444 
2445 	if (!verbose) {
2446 		mdb_printf("%0?p %-8x %-8x %-8x %-llx\n", addr,
2447 		    ace_flags, access_mask, ace_type, id);
2448 		return (DCMD_OK);
2449 	}
2450 
2451 	switch (ace_flags & ACE_TYPE_FLAGS) {
2452 	case ACE_OWNER:
2453 		mdb_printf("owner@:");
2454 		break;
2455 	case (ACE_IDENTIFIER_GROUP | ACE_GROUP):
2456 		mdb_printf("group@:");
2457 		break;
2458 	case ACE_EVERYONE:
2459 		mdb_printf("everyone@:");
2460 		break;
2461 	case ACE_IDENTIFIER_GROUP:
2462 		mdb_printf("group:%llx:", (u_longlong_t)id);
2463 		break;
2464 	case 0: /* User entry */
2465 		mdb_printf("user:%llx:", (u_longlong_t)id);
2466 		break;
2467 	}
2468 
2469 	/* print out permission mask */
2470 	if (access_mask & ACE_READ_DATA)
2471 		mdb_printf("r");
2472 	else
2473 		mdb_printf("-");
2474 	if (access_mask & ACE_WRITE_DATA)
2475 		mdb_printf("w");
2476 	else
2477 		mdb_printf("-");
2478 	if (access_mask & ACE_EXECUTE)
2479 		mdb_printf("x");
2480 	else
2481 		mdb_printf("-");
2482 	if (access_mask & ACE_APPEND_DATA)
2483 		mdb_printf("p");
2484 	else
2485 		mdb_printf("-");
2486 	if (access_mask & ACE_DELETE)
2487 		mdb_printf("d");
2488 	else
2489 		mdb_printf("-");
2490 	if (access_mask & ACE_DELETE_CHILD)
2491 		mdb_printf("D");
2492 	else
2493 		mdb_printf("-");
2494 	if (access_mask & ACE_READ_ATTRIBUTES)
2495 		mdb_printf("a");
2496 	else
2497 		mdb_printf("-");
2498 	if (access_mask & ACE_WRITE_ATTRIBUTES)
2499 		mdb_printf("A");
2500 	else
2501 		mdb_printf("-");
2502 	if (access_mask & ACE_READ_NAMED_ATTRS)
2503 		mdb_printf("R");
2504 	else
2505 		mdb_printf("-");
2506 	if (access_mask & ACE_WRITE_NAMED_ATTRS)
2507 		mdb_printf("W");
2508 	else
2509 		mdb_printf("-");
2510 	if (access_mask & ACE_READ_ACL)
2511 		mdb_printf("c");
2512 	else
2513 		mdb_printf("-");
2514 	if (access_mask & ACE_WRITE_ACL)
2515 		mdb_printf("C");
2516 	else
2517 		mdb_printf("-");
2518 	if (access_mask & ACE_WRITE_OWNER)
2519 		mdb_printf("o");
2520 	else
2521 		mdb_printf("-");
2522 	if (access_mask & ACE_SYNCHRONIZE)
2523 		mdb_printf("s");
2524 	else
2525 		mdb_printf("-");
2526 
2527 	mdb_printf(":");
2528 
2529 	/* Print out inheritance flags */
2530 	if (ace_flags & ACE_FILE_INHERIT_ACE)
2531 		mdb_printf("f");
2532 	else
2533 		mdb_printf("-");
2534 	if (ace_flags & ACE_DIRECTORY_INHERIT_ACE)
2535 		mdb_printf("d");
2536 	else
2537 		mdb_printf("-");
2538 	if (ace_flags & ACE_INHERIT_ONLY_ACE)
2539 		mdb_printf("i");
2540 	else
2541 		mdb_printf("-");
2542 	if (ace_flags & ACE_NO_PROPAGATE_INHERIT_ACE)
2543 		mdb_printf("n");
2544 	else
2545 		mdb_printf("-");
2546 	if (ace_flags & ACE_SUCCESSFUL_ACCESS_ACE_FLAG)
2547 		mdb_printf("S");
2548 	else
2549 		mdb_printf("-");
2550 	if (ace_flags & ACE_FAILED_ACCESS_ACE_FLAG)
2551 		mdb_printf("F");
2552 	else
2553 		mdb_printf("-");
2554 	if (ace_flags & ACE_INHERITED_ACE)
2555 		mdb_printf("I");
2556 	else
2557 		mdb_printf("-");
2558 
2559 	switch (ace_type) {
2560 	case ACE_ACCESS_ALLOWED_ACE_TYPE:
2561 		mdb_printf(":allow\n");
2562 		break;
2563 	case ACE_ACCESS_DENIED_ACE_TYPE:
2564 		mdb_printf(":deny\n");
2565 		break;
2566 	case ACE_SYSTEM_AUDIT_ACE_TYPE:
2567 		mdb_printf(":audit\n");
2568 		break;
2569 	case ACE_SYSTEM_ALARM_ACE_TYPE:
2570 		mdb_printf(":alarm\n");
2571 		break;
2572 	default:
2573 		mdb_printf(":?\n");
2574 	}
2575 	return (DCMD_OK);
2576 }
2577 
2578 /* ARGSUSED */
2579 static int
2580 zfs_ace_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2581 {
2582 	zfs_ace_t zace;
2583 	int verbose = FALSE;
2584 	uint64_t id;
2585 
2586 	if (!(flags & DCMD_ADDRSPEC))
2587 		return (DCMD_USAGE);
2588 
2589 	if (mdb_getopts(argc, argv,
2590 	    'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
2591 		return (DCMD_USAGE);
2592 
2593 	if (mdb_vread(&zace, sizeof (zfs_ace_t), addr) == -1) {
2594 		mdb_warn("failed to read zfs_ace_t");
2595 		return (DCMD_ERR);
2596 	}
2597 
2598 	if ((zace.z_hdr.z_flags & ACE_TYPE_FLAGS) == 0 ||
2599 	    (zace.z_hdr.z_flags & ACE_TYPE_FLAGS) == ACE_IDENTIFIER_GROUP)
2600 		id = zace.z_fuid;
2601 	else
2602 		id = -1;
2603 
2604 	return (zfs_ace_print_common(addr, flags, id, zace.z_hdr.z_access_mask,
2605 	    zace.z_hdr.z_flags, zace.z_hdr.z_type, verbose));
2606 }
2607 
2608 /* ARGSUSED */
2609 static int
2610 zfs_ace0_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2611 {
2612 	ace_t ace;
2613 	uint64_t id;
2614 	int verbose = FALSE;
2615 
2616 	if (!(flags & DCMD_ADDRSPEC))
2617 		return (DCMD_USAGE);
2618 
2619 	if (mdb_getopts(argc, argv,
2620 	    'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
2621 		return (DCMD_USAGE);
2622 
2623 	if (mdb_vread(&ace, sizeof (ace_t), addr) == -1) {
2624 		mdb_warn("failed to read ace_t");
2625 		return (DCMD_ERR);
2626 	}
2627 
2628 	if ((ace.a_flags & ACE_TYPE_FLAGS) == 0 ||
2629 	    (ace.a_flags & ACE_TYPE_FLAGS) == ACE_IDENTIFIER_GROUP)
2630 		id = ace.a_who;
2631 	else
2632 		id = -1;
2633 
2634 	return (zfs_ace_print_common(addr, flags, id, ace.a_access_mask,
2635 	    ace.a_flags, ace.a_type, verbose));
2636 }
2637 
2638 typedef struct acl_dump_args {
2639 	int a_argc;
2640 	const mdb_arg_t *a_argv;
2641 	uint16_t a_version;
2642 	int a_flags;
2643 } acl_dump_args_t;
2644 
2645 /* ARGSUSED */
2646 static int
2647 acl_aces_cb(uintptr_t addr, const void *unknown, void *arg)
2648 {
2649 	acl_dump_args_t *acl_args = (acl_dump_args_t *)arg;
2650 
2651 	if (acl_args->a_version == 1) {
2652 		if (mdb_call_dcmd("zfs_ace", addr,
2653 		    DCMD_ADDRSPEC|acl_args->a_flags, acl_args->a_argc,
2654 		    acl_args->a_argv) != DCMD_OK) {
2655 			return (WALK_ERR);
2656 		}
2657 	} else {
2658 		if (mdb_call_dcmd("zfs_ace0", addr,
2659 		    DCMD_ADDRSPEC|acl_args->a_flags, acl_args->a_argc,
2660 		    acl_args->a_argv) != DCMD_OK) {
2661 			return (WALK_ERR);
2662 		}
2663 	}
2664 	acl_args->a_flags = DCMD_LOOP;
2665 	return (WALK_NEXT);
2666 }
2667 
2668 /* ARGSUSED */
2669 static int
2670 acl_cb(uintptr_t addr, const void *unknown, void *arg)
2671 {
2672 	acl_dump_args_t *acl_args = (acl_dump_args_t *)arg;
2673 
2674 	if (acl_args->a_version == 1) {
2675 		if (mdb_pwalk("zfs_acl_node_aces", acl_aces_cb,
2676 		    arg, addr) != 0) {
2677 			mdb_warn("can't walk ACEs");
2678 			return (DCMD_ERR);
2679 		}
2680 	} else {
2681 		if (mdb_pwalk("zfs_acl_node_aces0", acl_aces_cb,
2682 		    arg, addr) != 0) {
2683 			mdb_warn("can't walk ACEs");
2684 			return (DCMD_ERR);
2685 		}
2686 	}
2687 	return (WALK_NEXT);
2688 }
2689 
2690 /* ARGSUSED */
2691 static int
2692 zfs_acl_dump(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2693 {
2694 	zfs_acl_t zacl;
2695 	int verbose = FALSE;
2696 	acl_dump_args_t acl_args;
2697 
2698 	if (!(flags & DCMD_ADDRSPEC))
2699 		return (DCMD_USAGE);
2700 
2701 	if (mdb_getopts(argc, argv,
2702 	    'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
2703 		return (DCMD_USAGE);
2704 
2705 	if (mdb_vread(&zacl, sizeof (zfs_acl_t), addr) == -1) {
2706 		mdb_warn("failed to read zfs_acl_t");
2707 		return (DCMD_ERR);
2708 	}
2709 
2710 	acl_args.a_argc = argc;
2711 	acl_args.a_argv = argv;
2712 	acl_args.a_version = zacl.z_version;
2713 	acl_args.a_flags = DCMD_LOOPFIRST;
2714 
2715 	if (mdb_pwalk("zfs_acl_node", acl_cb, &acl_args, addr) != 0) {
2716 		mdb_warn("can't walk ACL");
2717 		return (DCMD_ERR);
2718 	}
2719 
2720 	return (DCMD_OK);
2721 }
2722 
2723 /* ARGSUSED */
2724 static int
2725 zfs_acl_node_walk_init(mdb_walk_state_t *wsp)
2726 {
2727 	if (wsp->walk_addr == NULL) {
2728 		mdb_warn("must supply address of zfs_acl_node_t\n");
2729 		return (WALK_ERR);
2730 	}
2731 
2732 	wsp->walk_addr += OFFSETOF(zfs_acl_t, z_acl);
2733 
2734 	if (mdb_layered_walk("list", wsp) == -1) {
2735 		mdb_warn("failed to walk 'list'\n");
2736 		return (WALK_ERR);
2737 	}
2738 
2739 	return (WALK_NEXT);
2740 }
2741 
2742 static int
2743 zfs_acl_node_walk_step(mdb_walk_state_t *wsp)
2744 {
2745 	zfs_acl_node_t	aclnode;
2746 
2747 	if (mdb_vread(&aclnode, sizeof (zfs_acl_node_t),
2748 	    wsp->walk_addr) == -1) {
2749 		mdb_warn("failed to read zfs_acl_node at %p", wsp->walk_addr);
2750 		return (WALK_ERR);
2751 	}
2752 
2753 	return (wsp->walk_callback(wsp->walk_addr, &aclnode, wsp->walk_cbdata));
2754 }
2755 
2756 typedef struct ace_walk_data {
2757 	int		ace_count;
2758 	int		ace_version;
2759 } ace_walk_data_t;
2760 
2761 static int
2762 zfs_aces_walk_init_common(mdb_walk_state_t *wsp, int version,
2763     int ace_count, uintptr_t ace_data)
2764 {
2765 	ace_walk_data_t *ace_walk_data;
2766 
2767 	if (wsp->walk_addr == NULL) {
2768 		mdb_warn("must supply address of zfs_acl_node_t\n");
2769 		return (WALK_ERR);
2770 	}
2771 
2772 	ace_walk_data = mdb_alloc(sizeof (ace_walk_data_t), UM_SLEEP | UM_GC);
2773 
2774 	ace_walk_data->ace_count = ace_count;
2775 	ace_walk_data->ace_version = version;
2776 
2777 	wsp->walk_addr = ace_data;
2778 	wsp->walk_data = ace_walk_data;
2779 
2780 	return (WALK_NEXT);
2781 }
2782 
2783 static int
2784 zfs_acl_node_aces_walk_init_common(mdb_walk_state_t *wsp, int version)
2785 {
2786 	static int gotid;
2787 	static mdb_ctf_id_t acl_id;
2788 	int z_ace_count;
2789 	uintptr_t z_acldata;
2790 
2791 	if (!gotid) {
2792 		if (mdb_ctf_lookup_by_name("struct zfs_acl_node",
2793 		    &acl_id) == -1) {
2794 			mdb_warn("couldn't find struct zfs_acl_node");
2795 			return (DCMD_ERR);
2796 		}
2797 		gotid = TRUE;
2798 	}
2799 
2800 	if (GETMEMBID(wsp->walk_addr, &acl_id, z_ace_count, z_ace_count)) {
2801 		return (DCMD_ERR);
2802 	}
2803 	if (GETMEMBID(wsp->walk_addr, &acl_id, z_acldata, z_acldata)) {
2804 		return (DCMD_ERR);
2805 	}
2806 
2807 	return (zfs_aces_walk_init_common(wsp, version,
2808 	    z_ace_count, z_acldata));
2809 }
2810 
2811 /* ARGSUSED */
2812 static int
2813 zfs_acl_node_aces_walk_init(mdb_walk_state_t *wsp)
2814 {
2815 	return (zfs_acl_node_aces_walk_init_common(wsp, 1));
2816 }
2817 
2818 /* ARGSUSED */
2819 static int
2820 zfs_acl_node_aces0_walk_init(mdb_walk_state_t *wsp)
2821 {
2822 	return (zfs_acl_node_aces_walk_init_common(wsp, 0));
2823 }
2824 
2825 static int
2826 zfs_aces_walk_step(mdb_walk_state_t *wsp)
2827 {
2828 	ace_walk_data_t *ace_data = wsp->walk_data;
2829 	zfs_ace_t zace;
2830 	ace_t *acep;
2831 	int status;
2832 	int entry_type;
2833 	int allow_type;
2834 	uintptr_t ptr;
2835 
2836 	if (ace_data->ace_count == 0)
2837 		return (WALK_DONE);
2838 
2839 	if (mdb_vread(&zace, sizeof (zfs_ace_t), wsp->walk_addr) == -1) {
2840 		mdb_warn("failed to read zfs_ace_t at %#lx",
2841 		    wsp->walk_addr);
2842 		return (WALK_ERR);
2843 	}
2844 
2845 	switch (ace_data->ace_version) {
2846 	case 0:
2847 		acep = (ace_t *)&zace;
2848 		entry_type = acep->a_flags & ACE_TYPE_FLAGS;
2849 		allow_type = acep->a_type;
2850 		break;
2851 	case 1:
2852 		entry_type = zace.z_hdr.z_flags & ACE_TYPE_FLAGS;
2853 		allow_type = zace.z_hdr.z_type;
2854 		break;
2855 	default:
2856 		return (WALK_ERR);
2857 	}
2858 
2859 	ptr = (uintptr_t)wsp->walk_addr;
2860 	switch (entry_type) {
2861 	case ACE_OWNER:
2862 	case ACE_EVERYONE:
2863 	case (ACE_IDENTIFIER_GROUP | ACE_GROUP):
2864 		ptr += ace_data->ace_version == 0 ?
2865 		    sizeof (ace_t) : sizeof (zfs_ace_hdr_t);
2866 		break;
2867 	case ACE_IDENTIFIER_GROUP:
2868 	default:
2869 		switch (allow_type) {
2870 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
2871 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
2872 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
2873 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
2874 			ptr += ace_data->ace_version == 0 ?
2875 			    sizeof (ace_t) : sizeof (zfs_object_ace_t);
2876 			break;
2877 		default:
2878 			ptr += ace_data->ace_version == 0 ?
2879 			    sizeof (ace_t) : sizeof (zfs_ace_t);
2880 			break;
2881 		}
2882 	}
2883 
2884 	ace_data->ace_count--;
2885 	status = wsp->walk_callback(wsp->walk_addr,
2886 	    (void *)(uintptr_t)&zace, wsp->walk_cbdata);
2887 
2888 	wsp->walk_addr = ptr;
2889 	return (status);
2890 }
2891 
2892 /*
2893  * MDB module linkage information:
2894  *
2895  * We declare a list of structures describing our dcmds, and a function
2896  * named _mdb_init to return a pointer to our module information.
2897  */
2898 
2899 static const mdb_dcmd_t dcmds[] = {
2900 	{ "arc", "[-bkmg]", "print ARC variables", arc_print },
2901 	{ "blkptr", ":", "print blkptr_t", blkptr },
2902 	{ "dbuf", ":", "print dmu_buf_impl_t", dbuf },
2903 	{ "dbuf_stats", ":", "dbuf stats", dbuf_stats },
2904 	{ "dbufs",
2905 	    "\t[-O objset_impl_t*] [-n objset_name | \"mos\"] "
2906 	    "[-o object | \"mdn\"] \n"
2907 	    "\t[-l level] [-b blkid | \"bonus\"]",
2908 	    "find dmu_buf_impl_t's that match specified criteria", dbufs },
2909 	{ "abuf_find", "dva_word[0] dva_word[1]",
2910 	    "find arc_buf_hdr_t of a specified DVA",
2911 	    abuf_find },
2912 	{ "spa", "?[-cv]", "spa_t summary", spa_print },
2913 	{ "spa_config", ":", "print spa_t configuration", spa_print_config },
2914 	{ "spa_verify", ":", "verify spa_t consistency", spa_verify },
2915 	{ "spa_space", ":[-b]", "print spa_t on-disk space usage", spa_space },
2916 	{ "spa_vdevs", ":", "given a spa_t, print vdev summary", spa_vdevs },
2917 	{ "vdev", ":[-re]\n"
2918 	    "\t-r display recursively\n"
2919 	    "\t-e print statistics",
2920 	    "vdev_t summary", vdev_print },
2921 	{ "zio", ":[cpr]\n"
2922 	    "\t-c display children\n"
2923 	    "\t-p display parents\n"
2924 	    "\t-r display recursively",
2925 	    "zio_t summary", zio_print },
2926 	{ "zio_state", "?", "print out all zio_t structures on system or "
2927 	    "for a particular pool", zio_state },
2928 	{ "zfs_blkstats", ":[-v]",
2929 	    "given a spa_t, print block type stats from last scrub",
2930 	    zfs_blkstats },
2931 	{ "zfs_params", "", "print zfs tunable parameters", zfs_params },
2932 	{ "refcount", "", "print refcount_t holders", refcount },
2933 	{ "zap_leaf", "", "print zap_leaf_phys_t", zap_leaf },
2934 	{ "zfs_aces", ":[-v]", "print all ACEs from a zfs_acl_t",
2935 	    zfs_acl_dump },
2936 	{ "zfs_ace", ":[-v]", "print zfs_ace", zfs_ace_print },
2937 	{ "zfs_ace0", ":[-v]", "print zfs_ace0", zfs_ace0_print },
2938 	{ "sa_attr_table", ":", "print SA attribute table from sa_os_t",
2939 	    sa_attr_table},
2940 	{ "sa_attr", ": attr_id",
2941 	    "print SA attribute address when given sa_handle_t", sa_attr_print},
2942 	{ NULL }
2943 };
2944 
2945 static const mdb_walker_t walkers[] = {
2946 	/*
2947 	 * In userland, there is no generic provider of list_t walkers, so we
2948 	 * need to add it.
2949 	 */
2950 #ifndef _KERNEL
2951 	{ LIST_WALK_NAME, LIST_WALK_DESC,
2952 		list_walk_init, list_walk_step, list_walk_fini },
2953 #endif
2954 	{ "zms_freelist", "walk ZFS metaslab freelist",
2955 		freelist_walk_init, freelist_walk_step, NULL },
2956 	{ "txg_list", "given any txg_list_t *, walk all entries in all txgs",
2957 		txg_list_walk_init, txg_list_walk_step, NULL },
2958 	{ "txg_list0", "given any txg_list_t *, walk all entries in txg 0",
2959 		txg_list0_walk_init, txg_list_walk_step, NULL },
2960 	{ "txg_list1", "given any txg_list_t *, walk all entries in txg 1",
2961 		txg_list1_walk_init, txg_list_walk_step, NULL },
2962 	{ "txg_list2", "given any txg_list_t *, walk all entries in txg 2",
2963 		txg_list2_walk_init, txg_list_walk_step, NULL },
2964 	{ "txg_list3", "given any txg_list_t *, walk all entries in txg 3",
2965 		txg_list3_walk_init, txg_list_walk_step, NULL },
2966 	{ "zio", "walk all zio structures, optionally for a particular spa_t",
2967 		zio_walk_init, zio_walk_step, NULL },
2968 	{ "zio_root", "walk all root zio_t structures, optionally for a "
2969 	    "particular spa_t",
2970 		zio_walk_init, zio_walk_root_step, NULL },
2971 	{ "spa", "walk all spa_t entries in the namespace",
2972 		spa_walk_init, spa_walk_step, NULL },
2973 	{ "metaslab", "given a spa_t *, walk all metaslab_t structures",
2974 		metaslab_walk_init, metaslab_walk_step, NULL },
2975 	{ "zfs_acl_node", "given a zfs_acl_t, walk all zfs_acl_nodes",
2976 	    zfs_acl_node_walk_init, zfs_acl_node_walk_step, NULL },
2977 	{ "zfs_acl_node_aces", "given a zfs_acl_node_t, walk all ACEs",
2978 	    zfs_acl_node_aces_walk_init, zfs_aces_walk_step, NULL },
2979 	{ "zfs_acl_node_aces0",
2980 	    "given a zfs_acl_node_t, walk all ACEs as ace_t",
2981 	    zfs_acl_node_aces0_walk_init, zfs_aces_walk_step, NULL },
2982 	{ NULL }
2983 };
2984 
2985 static const mdb_modinfo_t modinfo = {
2986 	MDB_API_VERSION, dcmds, walkers
2987 };
2988 
2989 const mdb_modinfo_t *
2990 _mdb_init(void)
2991 {
2992 	return (&modinfo);
2993 }
2994