xref: /titanic_51/usr/src/cmd/mdb/common/modules/zfs/zfs.c (revision db94676fbd90b0affb8474a2af4e10718ba3136d)
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 2009 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/zio_compress.h>
39 #include <ctype.h>
40 
41 #ifndef _KERNEL
42 #include "../genunix/list.h"
43 #endif
44 
45 #ifdef _KERNEL
46 #define	ZFS_OBJ_NAME	"zfs"
47 #else
48 #define	ZFS_OBJ_NAME	"libzpool.so.1"
49 #endif
50 
51 static char *
52 local_strdup(const char *s)
53 {
54 	char *s1 = mdb_alloc(strlen(s) + 1, UM_SLEEP);
55 
56 	(void) strcpy(s1, s);
57 	return (s1);
58 }
59 
60 static int
61 getmember(uintptr_t addr, const char *type, mdb_ctf_id_t *idp,
62     const char *member, int len, void *buf)
63 {
64 	mdb_ctf_id_t id;
65 	ulong_t off;
66 	char name[64];
67 
68 	if (idp == NULL) {
69 		if (mdb_ctf_lookup_by_name(type, &id) == -1) {
70 			mdb_warn("couldn't find type %s", type);
71 			return (DCMD_ERR);
72 		}
73 		idp = &id;
74 	} else {
75 		type = name;
76 		mdb_ctf_type_name(*idp, name, sizeof (name));
77 	}
78 
79 	if (mdb_ctf_offsetof(*idp, member, &off) == -1) {
80 		mdb_warn("couldn't find member %s of type %s\n", member, type);
81 		return (DCMD_ERR);
82 	}
83 	if (off % 8 != 0) {
84 		mdb_warn("member %s of type %s is unsupported bitfield",
85 		    member, type);
86 		return (DCMD_ERR);
87 	}
88 	off /= 8;
89 
90 	if (mdb_vread(buf, len, addr + off) == -1) {
91 		mdb_warn("failed to read %s from %s at %p",
92 		    member, type, addr + off);
93 		return (DCMD_ERR);
94 	}
95 	/* mdb_warn("read %s from %s at %p+%llx\n", member, type, addr, off); */
96 
97 	return (0);
98 }
99 
100 #define	GETMEMB(addr, type, member, dest) \
101 	getmember(addr, #type, NULL, #member, sizeof (dest), &(dest))
102 
103 #define	GETMEMBID(addr, ctfid, member, dest) \
104 	getmember(addr, NULL, ctfid, #member, sizeof (dest), &(dest))
105 
106 static int
107 getrefcount(uintptr_t addr, mdb_ctf_id_t *id,
108     const char *member, uint64_t *rc)
109 {
110 	static int gotid;
111 	static mdb_ctf_id_t rc_id;
112 	ulong_t off;
113 
114 	if (!gotid) {
115 		if (mdb_ctf_lookup_by_name("struct refcount", &rc_id) == -1) {
116 			mdb_warn("couldn't find struct refcount");
117 			return (DCMD_ERR);
118 		}
119 		gotid = TRUE;
120 	}
121 
122 	if (mdb_ctf_offsetof(*id, member, &off) == -1) {
123 		char name[64];
124 		mdb_ctf_type_name(*id, name, sizeof (name));
125 		mdb_warn("couldn't find member %s of type %s\n", member, name);
126 		return (DCMD_ERR);
127 	}
128 	off /= 8;
129 
130 	return (GETMEMBID(addr + off, &rc_id, rc_count, *rc));
131 }
132 
133 static int
134 read_symbol(char *sym_name, void **bufp)
135 {
136 	GElf_Sym sym;
137 
138 	if (mdb_lookup_by_obj(MDB_TGT_OBJ_EVERY, sym_name, &sym)) {
139 		mdb_warn("can't find symbol %s", sym_name);
140 		return (DCMD_ERR);
141 	}
142 
143 	*bufp = mdb_alloc(sym.st_size, UM_SLEEP);
144 
145 	if (mdb_vread(*bufp, sym.st_size, sym.st_value) == -1) {
146 		mdb_warn("can't read data for symbol %s", sym_name);
147 		mdb_free(*bufp, sym.st_size);
148 		return (DCMD_ERR);
149 	}
150 
151 	return (DCMD_OK);
152 }
153 
154 static int verbose;
155 
156 static int
157 freelist_walk_init(mdb_walk_state_t *wsp)
158 {
159 	if (wsp->walk_addr == NULL) {
160 		mdb_warn("must supply starting address\n");
161 		return (WALK_ERR);
162 	}
163 
164 	wsp->walk_data = 0;  /* Index into the freelist */
165 	return (WALK_NEXT);
166 }
167 
168 static int
169 freelist_walk_step(mdb_walk_state_t *wsp)
170 {
171 	uint64_t entry;
172 	uintptr_t number = (uintptr_t)wsp->walk_data;
173 	char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
174 			    "INVALID", "INVALID", "INVALID", "INVALID" };
175 	int mapshift = SPA_MINBLOCKSHIFT;
176 
177 	if (mdb_vread(&entry, sizeof (entry), wsp->walk_addr) == -1) {
178 		mdb_warn("failed to read freelist entry %p", wsp->walk_addr);
179 		return (WALK_DONE);
180 	}
181 	wsp->walk_addr += sizeof (entry);
182 	wsp->walk_data = (void *)(number + 1);
183 
184 	if (SM_DEBUG_DECODE(entry)) {
185 		mdb_printf("DEBUG: %3u  %10s: txg=%llu  pass=%llu\n",
186 		    number,
187 		    ddata[SM_DEBUG_ACTION_DECODE(entry)],
188 		    SM_DEBUG_TXG_DECODE(entry),
189 		    SM_DEBUG_SYNCPASS_DECODE(entry));
190 	} else {
191 		mdb_printf("Entry: %3u  offsets=%08llx-%08llx  type=%c  "
192 		    "size=%06llx", number,
193 		    SM_OFFSET_DECODE(entry) << mapshift,
194 		    (SM_OFFSET_DECODE(entry) + SM_RUN_DECODE(entry)) <<
195 		    mapshift,
196 		    SM_TYPE_DECODE(entry) == SM_ALLOC ? 'A' : 'F',
197 		    SM_RUN_DECODE(entry) << mapshift);
198 		if (verbose)
199 			mdb_printf("      (raw=%012llx)\n", entry);
200 		mdb_printf("\n");
201 	}
202 	return (WALK_NEXT);
203 }
204 
205 
206 static int
207 dataset_name(uintptr_t addr, char *buf)
208 {
209 	static int gotid;
210 	static mdb_ctf_id_t dd_id;
211 	uintptr_t dd_parent;
212 	char dd_myname[MAXNAMELEN];
213 
214 	if (!gotid) {
215 		if (mdb_ctf_lookup_by_name("struct dsl_dir",
216 		    &dd_id) == -1) {
217 			mdb_warn("couldn't find struct dsl_dir");
218 			return (DCMD_ERR);
219 		}
220 		gotid = TRUE;
221 	}
222 	if (GETMEMBID(addr, &dd_id, dd_parent, dd_parent) ||
223 	    GETMEMBID(addr, &dd_id, dd_myname, dd_myname)) {
224 		return (DCMD_ERR);
225 	}
226 
227 	if (dd_parent) {
228 		if (dataset_name(dd_parent, buf))
229 			return (DCMD_ERR);
230 		strcat(buf, "/");
231 	}
232 
233 	if (dd_myname[0])
234 		strcat(buf, dd_myname);
235 	else
236 		strcat(buf, "???");
237 
238 	return (0);
239 }
240 
241 static int
242 objset_name(uintptr_t addr, char *buf)
243 {
244 	static int gotid;
245 	static mdb_ctf_id_t os_id, ds_id;
246 	uintptr_t os_dsl_dataset;
247 	char ds_snapname[MAXNAMELEN];
248 	uintptr_t ds_dir;
249 
250 	buf[0] = '\0';
251 
252 	if (!gotid) {
253 		if (mdb_ctf_lookup_by_name("struct objset",
254 		    &os_id) == -1) {
255 			mdb_warn("couldn't find struct objset");
256 			return (DCMD_ERR);
257 		}
258 		if (mdb_ctf_lookup_by_name("struct dsl_dataset",
259 		    &ds_id) == -1) {
260 			mdb_warn("couldn't find struct dsl_dataset");
261 			return (DCMD_ERR);
262 		}
263 
264 		gotid = TRUE;
265 	}
266 
267 	if (GETMEMBID(addr, &os_id, os_dsl_dataset, os_dsl_dataset))
268 		return (DCMD_ERR);
269 
270 	if (os_dsl_dataset == 0) {
271 		strcat(buf, "mos");
272 		return (0);
273 	}
274 
275 	if (GETMEMBID(os_dsl_dataset, &ds_id, ds_snapname, ds_snapname) ||
276 	    GETMEMBID(os_dsl_dataset, &ds_id, ds_dir, ds_dir)) {
277 		return (DCMD_ERR);
278 	}
279 
280 	if (ds_dir && dataset_name(ds_dir, buf))
281 		return (DCMD_ERR);
282 
283 	if (ds_snapname[0]) {
284 		strcat(buf, "@");
285 		strcat(buf, ds_snapname);
286 	}
287 	return (0);
288 }
289 
290 static void
291 enum_lookup(char *out, size_t size, mdb_ctf_id_t id, int val,
292     const char *prefix)
293 {
294 	const char *cp;
295 	size_t len = strlen(prefix);
296 
297 	if ((cp = mdb_ctf_enum_name(id, val)) != NULL) {
298 		if (strncmp(cp, prefix, len) == 0)
299 			cp += len;
300 		(void) strncpy(out, cp, size);
301 	} else {
302 		mdb_snprintf(out, size, "? (%d)", val);
303 	}
304 }
305 
306 /* ARGSUSED */
307 static int
308 zio_pipeline(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
309 {
310 	mdb_ctf_id_t pipe_enum;
311 	int i;
312 	char stage[1024];
313 
314 	if (mdb_ctf_lookup_by_name("enum zio_stage", &pipe_enum) == -1) {
315 		mdb_warn("Could not find enum zio_stage");
316 		return (DCMD_ERR);
317 	}
318 
319 	for (i = 0; i < 32; i++) {
320 		if (addr & (1U << i)) {
321 			enum_lookup(stage, sizeof (stage), pipe_enum, i,
322 			    "ZIO_STAGE_");
323 			mdb_printf("    %s\n", stage);
324 		}
325 	}
326 
327 	return (DCMD_OK);
328 }
329 
330 /* ARGSUSED */
331 static int
332 zfs_params(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
333 {
334 	/*
335 	 * This table can be approximately generated by running:
336 	 * egrep "^[a-z0-9_]+ [a-z0-9_]+( =.*)?;" *.c | cut -d ' ' -f 2
337 	 */
338 	static const char *params[] = {
339 		"arc_reduce_dnlc_percent",
340 		"zfs_arc_max",
341 		"zfs_arc_min",
342 		"arc_shrink_shift",
343 		"zfs_mdcomp_disable",
344 		"zfs_prefetch_disable",
345 		"zfetch_max_streams",
346 		"zfetch_min_sec_reap",
347 		"zfetch_block_cap",
348 		"zfetch_array_rd_sz",
349 		"zfs_default_bs",
350 		"zfs_default_ibs",
351 		"metaslab_aliquot",
352 		"reference_tracking_enable",
353 		"reference_history",
354 		"zio_taskq_threads",
355 		"spa_max_replication_override",
356 		"spa_mode",
357 		"zfs_flags",
358 		"zfs_txg_synctime",
359 		"zfs_txg_timeout",
360 		"zfs_write_limit_min",
361 		"zfs_write_limit_max",
362 		"zfs_write_limit_shift",
363 		"zfs_write_limit_override",
364 		"zfs_no_write_throttle",
365 		"zfs_vdev_cache_max",
366 		"zfs_vdev_cache_size",
367 		"zfs_vdev_cache_bshift",
368 		"vdev_mirror_shift",
369 		"zfs_vdev_max_pending",
370 		"zfs_vdev_min_pending",
371 		"zfs_scrub_limit",
372 		"zfs_vdev_time_shift",
373 		"zfs_vdev_ramp_rate",
374 		"zfs_vdev_aggregation_limit",
375 		"fzap_default_block_shift",
376 		"zfs_immediate_write_sz",
377 		"zfs_read_chunk_size",
378 		"zil_disable",
379 		"zfs_nocacheflush",
380 		"metaslab_gang_bang",
381 		"metaslab_df_alloc_threshold",
382 		"metaslab_df_free_pct",
383 		"zio_injection_enabled",
384 		"zvol_immediate_write_sz",
385 	};
386 	int i;
387 
388 	for (i = 0; i < sizeof (params) / sizeof (params[0]); i++) {
389 		int sz;
390 		uint64_t val64;
391 		uint32_t *val32p = (uint32_t *)&val64;
392 
393 		sz = mdb_readvar(&val64, params[i]);
394 		if (sz == 4) {
395 			mdb_printf("%s = 0x%x\n", params[i], *val32p);
396 		} else if (sz == 8) {
397 			mdb_printf("%s = 0x%llx\n", params[i], val64);
398 		} else {
399 			mdb_warn("variable %s not found", params[i]);
400 		}
401 	}
402 
403 	return (DCMD_OK);
404 }
405 
406 /* ARGSUSED */
407 static int
408 blkptr(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
409 {
410 	blkptr_t bp;
411 	dmu_object_type_info_t *doti;
412 	zio_compress_info_t *zct;
413 	zio_checksum_info_t *zci;
414 	int i;
415 	char buf[MAXPATHLEN];
416 
417 	if (mdb_vread(&bp, sizeof (blkptr_t), addr) == -1) {
418 		mdb_warn("failed to read blkptr_t");
419 		return (DCMD_ERR);
420 	}
421 
422 	if (read_symbol("dmu_ot", (void **)&doti) != DCMD_OK)
423 		return (DCMD_ERR);
424 	for (i = 0; i < DMU_OT_NUMTYPES; i++) {
425 		mdb_readstr(buf, sizeof (buf), (uintptr_t)doti[i].ot_name);
426 		doti[i].ot_name = local_strdup(buf);
427 	}
428 
429 	if (read_symbol("zio_checksum_table", (void **)&zci) != DCMD_OK)
430 		return (DCMD_ERR);
431 	for (i = 0; i < ZIO_CHECKSUM_FUNCTIONS; i++) {
432 		mdb_readstr(buf, sizeof (buf), (uintptr_t)zci[i].ci_name);
433 		zci[i].ci_name = local_strdup(buf);
434 	}
435 
436 	if (read_symbol("zio_compress_table", (void **)&zct) != DCMD_OK)
437 		return (DCMD_ERR);
438 	for (i = 0; i < ZIO_COMPRESS_FUNCTIONS; i++) {
439 		mdb_readstr(buf, sizeof (buf), (uintptr_t)zct[i].ci_name);
440 		zct[i].ci_name = local_strdup(buf);
441 	}
442 
443 	/*
444 	 * Super-ick warning:  This code is also duplicated in
445 	 * cmd/zdb.c .   Yeah, I hate code replication, too.
446 	 */
447 	for (i = 0; i < BP_GET_NDVAS(&bp); i++) {
448 		dva_t *dva = &bp.blk_dva[i];
449 
450 		mdb_printf("DVA[%d]: vdev_id %lld / %llx\n", i,
451 		    DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva));
452 		mdb_printf("DVA[%d]:       GANG: %-5s  GRID:  %04x\t"
453 		    "ASIZE: %llx\n", i, DVA_GET_GANG(dva) ? "TRUE" : "FALSE",
454 		    (int)DVA_GET_GRID(dva), DVA_GET_ASIZE(dva));
455 		mdb_printf("DVA[%d]: %llu:%llx:%llx:%s%s%s%s\n", i,
456 		    DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), BP_GET_PSIZE(&bp),
457 		    BP_SHOULD_BYTESWAP(&bp) ? "e" : "",
458 		    !DVA_GET_GANG(dva) && BP_GET_LEVEL(&bp) != 0 ? "i" : "",
459 		    DVA_GET_GANG(dva) ? "g" : "",
460 		    BP_GET_COMPRESS(&bp) != 0 ? "d" : "");
461 	}
462 	mdb_printf("LSIZE:  %-16llx\t\tPSIZE: %llx\n",
463 	    BP_GET_LSIZE(&bp), BP_GET_PSIZE(&bp));
464 	mdb_printf("ENDIAN: %6s\t\t\t\t\tTYPE:  %s\n",
465 	    BP_GET_BYTEORDER(&bp) ? "LITTLE" : "BIG",
466 	    BP_GET_TYPE(&bp) < DMU_OT_NUMTYPES ?
467 	    doti[BP_GET_TYPE(&bp)].ot_name : "UNKNOWN");
468 	mdb_printf("BIRTH:  %-16llx   LEVEL: %-2d\tFILL:  %llx\n",
469 	    bp.blk_birth, (int)BP_GET_LEVEL(&bp), bp.blk_fill);
470 	mdb_printf("CKFUNC: %-16s\t\tCOMP:  %s\n",
471 	    BP_GET_CHECKSUM(&bp) < ZIO_CHECKSUM_FUNCTIONS ?
472 	    zci[BP_GET_CHECKSUM(&bp)].ci_name : "UNKNOWN",
473 	    BP_GET_COMPRESS(&bp) < ZIO_COMPRESS_FUNCTIONS ?
474 	    zct[BP_GET_COMPRESS(&bp)].ci_name : "UNKNOWN");
475 	mdb_printf("CKSUM:  %llx:%llx:%llx:%llx\n",
476 	    bp.blk_cksum.zc_word[0],
477 	    bp.blk_cksum.zc_word[1],
478 	    bp.blk_cksum.zc_word[2],
479 	    bp.blk_cksum.zc_word[3]);
480 
481 	return (DCMD_OK);
482 }
483 
484 /* ARGSUSED */
485 static int
486 dbuf(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
487 {
488 	mdb_ctf_id_t id;
489 	dmu_buf_t db;
490 	uintptr_t objset;
491 	uint8_t level;
492 	uint64_t blkid;
493 	uint64_t holds;
494 	char objectname[32];
495 	char blkidname[32];
496 	char path[MAXNAMELEN];
497 
498 	if (DCMD_HDRSPEC(flags)) {
499 		mdb_printf("        addr object lvl blkid holds os\n");
500 	}
501 
502 	if (mdb_ctf_lookup_by_name("struct dmu_buf_impl", &id) == -1) {
503 		mdb_warn("couldn't find struct dmu_buf_impl_t");
504 		return (DCMD_ERR);
505 	}
506 
507 	if (GETMEMBID(addr, &id, db_objset, objset) ||
508 	    GETMEMBID(addr, &id, db, db) ||
509 	    GETMEMBID(addr, &id, db_level, level) ||
510 	    GETMEMBID(addr, &id, db_blkid, blkid)) {
511 		return (WALK_ERR);
512 	}
513 
514 	if (getrefcount(addr, &id, "db_holds", &holds)) {
515 		return (WALK_ERR);
516 	}
517 
518 	if (db.db_object == DMU_META_DNODE_OBJECT)
519 		(void) strcpy(objectname, "mdn");
520 	else
521 		(void) mdb_snprintf(objectname, sizeof (objectname), "%llx",
522 		    (u_longlong_t)db.db_object);
523 
524 	if (blkid == DB_BONUS_BLKID)
525 		(void) strcpy(blkidname, "bonus");
526 	else
527 		(void) mdb_snprintf(blkidname, sizeof (blkidname), "%llx",
528 		    (u_longlong_t)blkid);
529 
530 	if (objset_name(objset, path)) {
531 		return (WALK_ERR);
532 	}
533 
534 	mdb_printf("%p %8s %1u %9s %2llu %s\n",
535 	    addr, objectname, level, blkidname, holds, path);
536 
537 	return (DCMD_OK);
538 }
539 
540 /* ARGSUSED */
541 static int
542 dbuf_stats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
543 {
544 #define	HISTOSZ 32
545 	uintptr_t dbp;
546 	dmu_buf_impl_t db;
547 	dbuf_hash_table_t ht;
548 	uint64_t bucket, ndbufs;
549 	uint64_t histo[HISTOSZ];
550 	uint64_t histo2[HISTOSZ];
551 	int i, maxidx;
552 
553 	if (mdb_readvar(&ht, "dbuf_hash_table") == -1) {
554 		mdb_warn("failed to read 'dbuf_hash_table'");
555 		return (DCMD_ERR);
556 	}
557 
558 	for (i = 0; i < HISTOSZ; i++) {
559 		histo[i] = 0;
560 		histo2[i] = 0;
561 	}
562 
563 	ndbufs = 0;
564 	for (bucket = 0; bucket < ht.hash_table_mask+1; bucket++) {
565 		int len;
566 
567 		if (mdb_vread(&dbp, sizeof (void *),
568 		    (uintptr_t)(ht.hash_table+bucket)) == -1) {
569 			mdb_warn("failed to read hash bucket %u at %p",
570 			    bucket, ht.hash_table+bucket);
571 			return (DCMD_ERR);
572 		}
573 
574 		len = 0;
575 		while (dbp != 0) {
576 			if (mdb_vread(&db, sizeof (dmu_buf_impl_t),
577 			    dbp) == -1) {
578 				mdb_warn("failed to read dbuf at %p", dbp);
579 				return (DCMD_ERR);
580 			}
581 			dbp = (uintptr_t)db.db_hash_next;
582 			for (i = MIN(len, HISTOSZ - 1); i >= 0; i--)
583 				histo2[i]++;
584 			len++;
585 			ndbufs++;
586 		}
587 
588 		if (len >= HISTOSZ)
589 			len = HISTOSZ-1;
590 		histo[len]++;
591 	}
592 
593 	mdb_printf("hash table has %llu buckets, %llu dbufs "
594 	    "(avg %llu buckets/dbuf)\n",
595 	    ht.hash_table_mask+1, ndbufs,
596 	    (ht.hash_table_mask+1)/ndbufs);
597 
598 	mdb_printf("\n");
599 	maxidx = 0;
600 	for (i = 0; i < HISTOSZ; i++)
601 		if (histo[i] > 0)
602 			maxidx = i;
603 	mdb_printf("hash chain length	number of buckets\n");
604 	for (i = 0; i <= maxidx; i++)
605 		mdb_printf("%u			%llu\n", i, histo[i]);
606 
607 	mdb_printf("\n");
608 	maxidx = 0;
609 	for (i = 0; i < HISTOSZ; i++)
610 		if (histo2[i] > 0)
611 			maxidx = i;
612 	mdb_printf("hash chain depth	number of dbufs\n");
613 	for (i = 0; i <= maxidx; i++)
614 		mdb_printf("%u or more		%llu	%llu%%\n",
615 		    i, histo2[i], histo2[i]*100/ndbufs);
616 
617 
618 	return (DCMD_OK);
619 }
620 
621 typedef struct dbufs_data {
622 	mdb_ctf_id_t id;
623 	uint64_t objset;
624 	uint64_t object;
625 	uint64_t level;
626 	uint64_t blkid;
627 	char *osname;
628 } dbufs_data_t;
629 
630 #define	DBUFS_UNSET	(0xbaddcafedeadbeefULL)
631 
632 /* ARGSUSED */
633 static int
634 dbufs_cb(uintptr_t addr, const void *unknown, void *arg)
635 {
636 	dbufs_data_t *data = arg;
637 	uintptr_t objset;
638 	dmu_buf_t db;
639 	uint8_t level;
640 	uint64_t blkid;
641 	char osname[MAXNAMELEN];
642 
643 	if (GETMEMBID(addr, &data->id, db_objset, objset) ||
644 	    GETMEMBID(addr, &data->id, db, db) ||
645 	    GETMEMBID(addr, &data->id, db_level, level) ||
646 	    GETMEMBID(addr, &data->id, db_blkid, blkid)) {
647 		return (WALK_ERR);
648 	}
649 
650 	if ((data->objset == DBUFS_UNSET || data->objset == objset) &&
651 	    (data->osname == NULL || (objset_name(objset, osname) == 0 &&
652 	    strcmp(data->osname, osname) == 0)) &&
653 	    (data->object == DBUFS_UNSET || data->object == db.db_object) &&
654 	    (data->level == DBUFS_UNSET || data->level == level) &&
655 	    (data->blkid == DBUFS_UNSET || data->blkid == blkid)) {
656 		mdb_printf("%#lr\n", addr);
657 	}
658 	return (WALK_NEXT);
659 }
660 
661 /* ARGSUSED */
662 static int
663 dbufs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
664 {
665 	dbufs_data_t data;
666 	char *object = NULL;
667 	char *blkid = NULL;
668 
669 	data.objset = data.object = data.level = data.blkid = DBUFS_UNSET;
670 	data.osname = NULL;
671 
672 	if (mdb_getopts(argc, argv,
673 	    'O', MDB_OPT_UINT64, &data.objset,
674 	    'n', MDB_OPT_STR, &data.osname,
675 	    'o', MDB_OPT_STR, &object,
676 	    'l', MDB_OPT_UINT64, &data.level,
677 	    'b', MDB_OPT_STR, &blkid) != argc) {
678 		return (DCMD_USAGE);
679 	}
680 
681 	if (object) {
682 		if (strcmp(object, "mdn") == 0) {
683 			data.object = DMU_META_DNODE_OBJECT;
684 		} else {
685 			data.object = mdb_strtoull(object);
686 		}
687 	}
688 
689 	if (blkid) {
690 		if (strcmp(blkid, "bonus") == 0) {
691 			data.blkid = DB_BONUS_BLKID;
692 		} else {
693 			data.blkid = mdb_strtoull(blkid);
694 		}
695 	}
696 
697 	if (mdb_ctf_lookup_by_name("struct dmu_buf_impl", &data.id) == -1) {
698 		mdb_warn("couldn't find struct dmu_buf_impl_t");
699 		return (DCMD_ERR);
700 	}
701 
702 	if (mdb_walk("dmu_buf_impl_t", dbufs_cb, &data) != 0) {
703 		mdb_warn("can't walk dbufs");
704 		return (DCMD_ERR);
705 	}
706 
707 	return (DCMD_OK);
708 }
709 
710 typedef struct abuf_find_data {
711 	dva_t dva;
712 	mdb_ctf_id_t id;
713 } abuf_find_data_t;
714 
715 /* ARGSUSED */
716 static int
717 abuf_find_cb(uintptr_t addr, const void *unknown, void *arg)
718 {
719 	abuf_find_data_t *data = arg;
720 	dva_t dva;
721 
722 	if (GETMEMBID(addr, &data->id, b_dva, dva)) {
723 		return (WALK_ERR);
724 	}
725 
726 	if (dva.dva_word[0] == data->dva.dva_word[0] &&
727 	    dva.dva_word[1] == data->dva.dva_word[1]) {
728 		mdb_printf("%#lr\n", addr);
729 	}
730 	return (WALK_NEXT);
731 }
732 
733 /* ARGSUSED */
734 static int
735 abuf_find(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
736 {
737 	abuf_find_data_t data;
738 	GElf_Sym sym;
739 	int i;
740 	const char *syms[] = {
741 		"ARC_mru",
742 		"ARC_mru_ghost",
743 		"ARC_mfu",
744 		"ARC_mfu_ghost",
745 	};
746 
747 	if (argc != 2)
748 		return (DCMD_USAGE);
749 
750 	for (i = 0; i < 2; i ++) {
751 		switch (argv[i].a_type) {
752 		case MDB_TYPE_STRING:
753 			data.dva.dva_word[i] = mdb_strtoull(argv[i].a_un.a_str);
754 			break;
755 		case MDB_TYPE_IMMEDIATE:
756 			data.dva.dva_word[i] = argv[i].a_un.a_val;
757 			break;
758 		default:
759 			return (DCMD_USAGE);
760 		}
761 	}
762 
763 	if (mdb_ctf_lookup_by_name("struct arc_buf_hdr", &data.id) == -1) {
764 		mdb_warn("couldn't find struct arc_buf_hdr");
765 		return (DCMD_ERR);
766 	}
767 
768 	for (i = 0; i < sizeof (syms) / sizeof (syms[0]); i++) {
769 		if (mdb_lookup_by_name(syms[i], &sym)) {
770 			mdb_warn("can't find symbol %s", syms[i]);
771 			return (DCMD_ERR);
772 		}
773 
774 		if (mdb_pwalk("list", abuf_find_cb, &data, sym.st_value) != 0) {
775 			mdb_warn("can't walk %s", syms[i]);
776 			return (DCMD_ERR);
777 		}
778 	}
779 
780 	return (DCMD_OK);
781 }
782 
783 /*ARGSUSED*/
784 static int
785 arc_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
786 {
787 	kstat_named_t *stats;
788 	GElf_Sym sym;
789 	int nstats, i;
790 	uint_t opt_a = FALSE;
791 	uint_t opt_b = FALSE;
792 	uint_t shift = 0;
793 	const char *suffix;
794 
795 	static const char *bytestats[] = {
796 		"p", "c", "c_min", "c_max", "size", NULL
797 	};
798 
799 	static const char *extras[] = {
800 		"arc_no_grow", "arc_tempreserve",
801 		"arc_meta_used", "arc_meta_limit", "arc_meta_max",
802 		NULL
803 	};
804 
805 	if (mdb_lookup_by_name("arc_stats", &sym) == -1) {
806 		mdb_warn("failed to find 'arc_stats'");
807 		return (DCMD_ERR);
808 	}
809 
810 	stats = mdb_zalloc(sym.st_size, UM_SLEEP | UM_GC);
811 
812 	if (mdb_vread(stats, sym.st_size, sym.st_value) == -1) {
813 		mdb_warn("couldn't read 'arc_stats' at %p", sym.st_value);
814 		return (DCMD_ERR);
815 	}
816 
817 	nstats = sym.st_size / sizeof (kstat_named_t);
818 
819 	/* NB: -a / opt_a are ignored for backwards compatability */
820 	if (mdb_getopts(argc, argv,
821 	    'a', MDB_OPT_SETBITS, TRUE, &opt_a,
822 	    'b', MDB_OPT_SETBITS, TRUE, &opt_b,
823 	    'k', MDB_OPT_SETBITS, 10, &shift,
824 	    'm', MDB_OPT_SETBITS, 20, &shift,
825 	    'g', MDB_OPT_SETBITS, 30, &shift,
826 	    NULL) != argc)
827 		return (DCMD_USAGE);
828 
829 	if (!opt_b && !shift)
830 		shift = 20;
831 
832 	switch (shift) {
833 	case 0:
834 		suffix = "B";
835 		break;
836 	case 10:
837 		suffix = "KB";
838 		break;
839 	case 20:
840 		suffix = "MB";
841 		break;
842 	case 30:
843 		suffix = "GB";
844 		break;
845 	default:
846 		suffix = "XX";
847 	}
848 
849 	for (i = 0; i < nstats; i++) {
850 		int j;
851 		boolean_t bytes = B_FALSE;
852 
853 		for (j = 0; bytestats[j]; j++) {
854 			if (strcmp(stats[i].name, bytestats[j]) == 0) {
855 				bytes = B_TRUE;
856 				break;
857 			}
858 		}
859 
860 		if (bytes) {
861 			mdb_printf("%-25s = %9llu %s\n", stats[i].name,
862 			    stats[i].value.ui64 >> shift, suffix);
863 		} else {
864 			mdb_printf("%-25s = %9llu\n", stats[i].name,
865 			    stats[i].value.ui64);
866 		}
867 	}
868 
869 	for (i = 0; extras[i]; i++) {
870 		uint64_t buf;
871 
872 		if (mdb_lookup_by_name(extras[i], &sym) == -1) {
873 			mdb_warn("failed to find '%s'", extras[i]);
874 			return (DCMD_ERR);
875 		}
876 
877 		if (sym.st_size != sizeof (uint64_t) &&
878 		    sym.st_size != sizeof (uint32_t)) {
879 			mdb_warn("expected scalar for variable '%s'\n",
880 			    extras[i]);
881 			return (DCMD_ERR);
882 		}
883 
884 		if (mdb_vread(&buf, sym.st_size, sym.st_value) == -1) {
885 			mdb_warn("couldn't read '%s'", extras[i]);
886 			return (DCMD_ERR);
887 		}
888 
889 		mdb_printf("%-25s = ", extras[i]);
890 
891 		/* NB: all the 64-bit extras happen to be byte counts */
892 		if (sym.st_size == sizeof (uint64_t))
893 			mdb_printf("%9llu %s\n", buf >> shift, suffix);
894 
895 		if (sym.st_size == sizeof (uint32_t))
896 			mdb_printf("%9d\n", *((uint32_t *)&buf));
897 	}
898 	return (DCMD_OK);
899 }
900 
901 /*
902  * ::spa
903  *
904  * 	-c	Print configuration information as well
905  * 	-v	Print vdev state
906  * 	-e	Print vdev error stats
907  *
908  * Print a summarized spa_t.  When given no arguments, prints out a table of all
909  * active pools on the system.
910  */
911 /* ARGSUSED */
912 static int
913 spa_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
914 {
915 	spa_t spa;
916 	const char *statetab[] = { "ACTIVE", "EXPORTED", "DESTROYED",
917 		"SPARE", "L2CACHE", "UNINIT", "UNAVAIL", "POTENTIAL" };
918 	const char *state;
919 	int config = FALSE;
920 	int vdevs = FALSE;
921 	int errors = FALSE;
922 
923 	if (mdb_getopts(argc, argv,
924 	    'c', MDB_OPT_SETBITS, TRUE, &config,
925 	    'v', MDB_OPT_SETBITS, TRUE, &vdevs,
926 	    'e', MDB_OPT_SETBITS, TRUE, &errors,
927 	    NULL) != argc)
928 		return (DCMD_USAGE);
929 
930 	if (!(flags & DCMD_ADDRSPEC)) {
931 		if (mdb_walk_dcmd("spa", "spa", argc, argv) == -1) {
932 			mdb_warn("can't walk spa");
933 			return (DCMD_ERR);
934 		}
935 
936 		return (DCMD_OK);
937 	}
938 
939 	if (flags & DCMD_PIPE_OUT) {
940 		mdb_printf("%#lr\n", addr);
941 		return (DCMD_OK);
942 	}
943 
944 	if (DCMD_HDRSPEC(flags))
945 		mdb_printf("%<u>%-?s %9s %-*s%</u>\n", "ADDR", "STATE",
946 		    sizeof (uintptr_t) == 4 ? 60 : 52, "NAME");
947 
948 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
949 		mdb_warn("failed to read spa_t at %p", addr);
950 		return (DCMD_ERR);
951 	}
952 
953 	if (spa.spa_state < 0 || spa.spa_state > POOL_STATE_UNAVAIL)
954 		state = "UNKNOWN";
955 	else
956 		state = statetab[spa.spa_state];
957 
958 	mdb_printf("%0?p %9s %s\n", addr, state, spa.spa_name);
959 
960 	if (config) {
961 		mdb_printf("\n");
962 		mdb_inc_indent(4);
963 		if (mdb_call_dcmd("spa_config", addr, flags, 0,
964 		    NULL) != DCMD_OK)
965 			return (DCMD_ERR);
966 		mdb_dec_indent(4);
967 	}
968 
969 	if (vdevs || errors) {
970 		mdb_arg_t v;
971 
972 		v.a_type = MDB_TYPE_STRING;
973 		v.a_un.a_str = "-e";
974 
975 		mdb_printf("\n");
976 		mdb_inc_indent(4);
977 		if (mdb_call_dcmd("spa_vdevs", addr, flags, errors ? 1 : 0,
978 		    &v) != DCMD_OK)
979 			return (DCMD_ERR);
980 		mdb_dec_indent(4);
981 	}
982 
983 	return (DCMD_OK);
984 }
985 
986 /*
987  * ::spa_config
988  *
989  * Given a spa_t, print the configuration information stored in spa_config.
990  * Since it's just an nvlist, format it as an indented list of name=value pairs.
991  * We simply read the value of spa_config and pass off to ::nvlist.
992  */
993 /* ARGSUSED */
994 static int
995 spa_print_config(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
996 {
997 	spa_t spa;
998 
999 	if (argc != 0 || !(flags & DCMD_ADDRSPEC))
1000 		return (DCMD_USAGE);
1001 
1002 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1003 		mdb_warn("failed to read spa_t at %p", addr);
1004 		return (DCMD_ERR);
1005 	}
1006 
1007 	if (spa.spa_config == NULL) {
1008 		mdb_printf("(none)\n");
1009 		return (DCMD_OK);
1010 	}
1011 
1012 	return (mdb_call_dcmd("nvlist", (uintptr_t)spa.spa_config, flags,
1013 	    0, NULL));
1014 }
1015 
1016 /*
1017  * ::vdev
1018  *
1019  * Print out a summarized vdev_t, in the following form:
1020  *
1021  * ADDR             STATE	AUX            DESC
1022  * fffffffbcde23df0 HEALTHY	-              /dev/dsk/c0t0d0
1023  *
1024  * If '-r' is specified, recursively visit all children.
1025  *
1026  * With '-e', the statistics associated with the vdev are printed as well.
1027  */
1028 static int
1029 do_print_vdev(uintptr_t addr, int flags, int depth, int stats,
1030     int recursive)
1031 {
1032 	vdev_t vdev;
1033 	char desc[MAXNAMELEN];
1034 	int c, children;
1035 	uintptr_t *child;
1036 	const char *state, *aux;
1037 
1038 	if (mdb_vread(&vdev, sizeof (vdev), (uintptr_t)addr) == -1) {
1039 		mdb_warn("failed to read vdev_t at %p\n", (uintptr_t)addr);
1040 		return (DCMD_ERR);
1041 	}
1042 
1043 	if (flags & DCMD_PIPE_OUT) {
1044 		mdb_printf("%#lr", addr);
1045 	} else {
1046 		if (vdev.vdev_path != NULL) {
1047 			if (mdb_readstr(desc, sizeof (desc),
1048 			    (uintptr_t)vdev.vdev_path) == -1) {
1049 				mdb_warn("failed to read vdev_path at %p\n",
1050 				    vdev.vdev_path);
1051 				return (DCMD_ERR);
1052 			}
1053 		} else if (vdev.vdev_ops != NULL) {
1054 			vdev_ops_t ops;
1055 			if (mdb_vread(&ops, sizeof (ops),
1056 			    (uintptr_t)vdev.vdev_ops) == -1) {
1057 				mdb_warn("failed to read vdev_ops at %p\n",
1058 				    vdev.vdev_ops);
1059 				return (DCMD_ERR);
1060 			}
1061 			(void) strcpy(desc, ops.vdev_op_type);
1062 		} else {
1063 			(void) strcpy(desc, "<unknown>");
1064 		}
1065 
1066 		if (depth == 0 && DCMD_HDRSPEC(flags))
1067 			mdb_printf("%<u>%-?s %-9s %-12s %-*s%</u>\n",
1068 			    "ADDR", "STATE", "AUX",
1069 			    sizeof (uintptr_t) == 4 ? 43 : 35,
1070 			    "DESCRIPTION");
1071 
1072 		mdb_printf("%0?p ", addr);
1073 
1074 		switch (vdev.vdev_state) {
1075 		case VDEV_STATE_CLOSED:
1076 			state = "CLOSED";
1077 			break;
1078 		case VDEV_STATE_OFFLINE:
1079 			state = "OFFLINE";
1080 			break;
1081 		case VDEV_STATE_CANT_OPEN:
1082 			state = "CANT_OPEN";
1083 			break;
1084 		case VDEV_STATE_DEGRADED:
1085 			state = "DEGRADED";
1086 			break;
1087 		case VDEV_STATE_HEALTHY:
1088 			state = "HEALTHY";
1089 			break;
1090 		case VDEV_STATE_REMOVED:
1091 			state = "REMOVED";
1092 			break;
1093 		case VDEV_STATE_FAULTED:
1094 			state = "FAULTED";
1095 			break;
1096 		default:
1097 			state = "UNKNOWN";
1098 			break;
1099 		}
1100 
1101 		switch (vdev.vdev_stat.vs_aux) {
1102 		case VDEV_AUX_NONE:
1103 			aux = "-";
1104 			break;
1105 		case VDEV_AUX_OPEN_FAILED:
1106 			aux = "OPEN_FAILED";
1107 			break;
1108 		case VDEV_AUX_CORRUPT_DATA:
1109 			aux = "CORRUPT_DATA";
1110 			break;
1111 		case VDEV_AUX_NO_REPLICAS:
1112 			aux = "NO_REPLICAS";
1113 			break;
1114 		case VDEV_AUX_BAD_GUID_SUM:
1115 			aux = "BAD_GUID_SUM";
1116 			break;
1117 		case VDEV_AUX_TOO_SMALL:
1118 			aux = "TOO_SMALL";
1119 			break;
1120 		case VDEV_AUX_BAD_LABEL:
1121 			aux = "BAD_LABEL";
1122 			break;
1123 		case VDEV_AUX_VERSION_NEWER:
1124 			aux = "VERS_NEWER";
1125 			break;
1126 		case VDEV_AUX_VERSION_OLDER:
1127 			aux = "VERS_OLDER";
1128 			break;
1129 		case VDEV_AUX_SPARED:
1130 			aux = "SPARED";
1131 			break;
1132 		case VDEV_AUX_ERR_EXCEEDED:
1133 			aux = "ERR_EXCEEDED";
1134 			break;
1135 		case VDEV_AUX_IO_FAILURE:
1136 			aux = "IO_FAILURE";
1137 			break;
1138 		case VDEV_AUX_BAD_LOG:
1139 			aux = "BAD_LOG";
1140 			break;
1141 		default:
1142 			aux = "UNKNOWN";
1143 			break;
1144 		}
1145 
1146 		mdb_printf("%-9s %-12s %*s%s\n", state, aux, depth, "", desc);
1147 
1148 		if (stats) {
1149 			vdev_stat_t *vs = &vdev.vdev_stat;
1150 			int i;
1151 
1152 			mdb_inc_indent(4);
1153 			mdb_printf("\n");
1154 			mdb_printf("%<u>       %12s %12s %12s %12s "
1155 			    "%12s%</u>\n", "READ", "WRITE", "FREE", "CLAIM",
1156 			    "IOCTL");
1157 			mdb_printf("OPS     ");
1158 			for (i = 1; i < ZIO_TYPES; i++)
1159 				mdb_printf("%11#llx%s", vs->vs_ops[i],
1160 				    i == ZIO_TYPES - 1 ? "" : "  ");
1161 			mdb_printf("\n");
1162 			mdb_printf("BYTES   ");
1163 			for (i = 1; i < ZIO_TYPES; i++)
1164 				mdb_printf("%11#llx%s", vs->vs_bytes[i],
1165 				    i == ZIO_TYPES - 1 ? "" : "  ");
1166 
1167 
1168 			mdb_printf("\n");
1169 			mdb_printf("EREAD    %10#llx\n", vs->vs_read_errors);
1170 			mdb_printf("EWRITE   %10#llx\n", vs->vs_write_errors);
1171 			mdb_printf("ECKSUM   %10#llx\n",
1172 			    vs->vs_checksum_errors);
1173 			mdb_dec_indent(4);
1174 		}
1175 
1176 		if (stats)
1177 			mdb_printf("\n");
1178 	}
1179 
1180 	children = vdev.vdev_children;
1181 
1182 	if (children == 0 || !recursive)
1183 		return (DCMD_OK);
1184 
1185 	child = mdb_alloc(children * sizeof (void *), UM_SLEEP | UM_GC);
1186 	if (mdb_vread(child, children * sizeof (void *),
1187 	    (uintptr_t)vdev.vdev_child) == -1) {
1188 		mdb_warn("failed to read vdev children at %p", vdev.vdev_child);
1189 		return (DCMD_ERR);
1190 	}
1191 
1192 	for (c = 0; c < children; c++) {
1193 		if (do_print_vdev(child[c], flags, depth + 2, stats,
1194 		    recursive))
1195 			return (DCMD_ERR);
1196 	}
1197 
1198 	return (DCMD_OK);
1199 }
1200 
1201 static int
1202 vdev_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1203 {
1204 	int recursive = FALSE;
1205 	int stats = FALSE;
1206 	uint64_t depth = 0;
1207 
1208 	if (mdb_getopts(argc, argv,
1209 	    'r', MDB_OPT_SETBITS, TRUE, &recursive,
1210 	    'e', MDB_OPT_SETBITS, TRUE, &stats,
1211 	    'd', MDB_OPT_UINT64, &depth,
1212 	    NULL) != argc)
1213 		return (DCMD_USAGE);
1214 
1215 	if (!(flags & DCMD_ADDRSPEC)) {
1216 		mdb_warn("no vdev_t address given\n");
1217 		return (DCMD_ERR);
1218 	}
1219 
1220 	return (do_print_vdev(addr, flags, (int)depth, stats, recursive));
1221 }
1222 
1223 typedef struct metaslab_walk_data {
1224 	uint64_t mw_numvdevs;
1225 	uintptr_t *mw_vdevs;
1226 	int mw_curvdev;
1227 	uint64_t mw_nummss;
1228 	uintptr_t *mw_mss;
1229 	int mw_curms;
1230 } metaslab_walk_data_t;
1231 
1232 static int
1233 metaslab_walk_step(mdb_walk_state_t *wsp)
1234 {
1235 	metaslab_walk_data_t *mw = wsp->walk_data;
1236 	metaslab_t ms;
1237 	uintptr_t msp;
1238 
1239 	if (mw->mw_curvdev >= mw->mw_numvdevs)
1240 		return (WALK_DONE);
1241 
1242 	if (mw->mw_mss == NULL) {
1243 		uintptr_t mssp;
1244 		uintptr_t vdevp;
1245 
1246 		ASSERT(mw->mw_curms == 0);
1247 		ASSERT(mw->mw_nummss == 0);
1248 
1249 		vdevp = mw->mw_vdevs[mw->mw_curvdev];
1250 		if (GETMEMB(vdevp, struct vdev, vdev_ms, mssp) ||
1251 		    GETMEMB(vdevp, struct vdev, vdev_ms_count, mw->mw_nummss)) {
1252 			return (WALK_ERR);
1253 		}
1254 
1255 		mw->mw_mss = mdb_alloc(mw->mw_nummss * sizeof (void*),
1256 		    UM_SLEEP | UM_GC);
1257 		if (mdb_vread(mw->mw_mss, mw->mw_nummss * sizeof (void*),
1258 		    mssp) == -1) {
1259 			mdb_warn("failed to read vdev_ms at %p", mssp);
1260 			return (WALK_ERR);
1261 		}
1262 	}
1263 
1264 	if (mw->mw_curms >= mw->mw_nummss) {
1265 		mw->mw_mss = NULL;
1266 		mw->mw_curms = 0;
1267 		mw->mw_nummss = 0;
1268 		mw->mw_curvdev++;
1269 		return (WALK_NEXT);
1270 	}
1271 
1272 	msp = mw->mw_mss[mw->mw_curms];
1273 	if (mdb_vread(&ms, sizeof (metaslab_t), msp) == -1) {
1274 		mdb_warn("failed to read metaslab_t at %p", msp);
1275 		return (WALK_ERR);
1276 	}
1277 
1278 	mw->mw_curms++;
1279 
1280 	return (wsp->walk_callback(msp, &ms, wsp->walk_cbdata));
1281 }
1282 
1283 /* ARGSUSED */
1284 static int
1285 metaslab_walk_init(mdb_walk_state_t *wsp)
1286 {
1287 	metaslab_walk_data_t *mw;
1288 	uintptr_t root_vdevp;
1289 	uintptr_t childp;
1290 
1291 	if (wsp->walk_addr == NULL) {
1292 		mdb_warn("must supply address of spa_t\n");
1293 		return (WALK_ERR);
1294 	}
1295 
1296 	mw = mdb_zalloc(sizeof (metaslab_walk_data_t), UM_SLEEP | UM_GC);
1297 
1298 	if (GETMEMB(wsp->walk_addr, struct spa, spa_root_vdev, root_vdevp) ||
1299 	    GETMEMB(root_vdevp, struct vdev, vdev_children, mw->mw_numvdevs) ||
1300 	    GETMEMB(root_vdevp, struct vdev, vdev_child, childp)) {
1301 		return (DCMD_ERR);
1302 	}
1303 
1304 	mw->mw_vdevs = mdb_alloc(mw->mw_numvdevs * sizeof (void *),
1305 	    UM_SLEEP | UM_GC);
1306 	if (mdb_vread(mw->mw_vdevs, mw->mw_numvdevs * sizeof (void *),
1307 	    childp) == -1) {
1308 		mdb_warn("failed to read root vdev children at %p", childp);
1309 		return (DCMD_ERR);
1310 	}
1311 
1312 	wsp->walk_data = mw;
1313 
1314 	return (WALK_NEXT);
1315 }
1316 
1317 typedef struct mdb_spa {
1318 	uintptr_t spa_dsl_pool;
1319 	uintptr_t spa_root_vdev;
1320 } mdb_spa_t;
1321 
1322 typedef struct mdb_dsl_dir {
1323 	uintptr_t dd_phys;
1324 	int64_t dd_space_towrite[TXG_SIZE];
1325 } mdb_dsl_dir_t;
1326 
1327 typedef struct mdb_dsl_dir_phys {
1328 	uint64_t dd_used_bytes;
1329 	uint64_t dd_compressed_bytes;
1330 	uint64_t dd_uncompressed_bytes;
1331 } mdb_dsl_dir_phys_t;
1332 
1333 typedef struct mdb_vdev {
1334 	uintptr_t vdev_parent;
1335 	uintptr_t vdev_ms;
1336 	uint64_t vdev_ms_count;
1337 	vdev_stat_t vdev_stat;
1338 } mdb_vdev_t;
1339 
1340 typedef struct mdb_metaslab {
1341 	space_map_t ms_allocmap[TXG_SIZE];
1342 	space_map_t ms_freemap[TXG_SIZE];
1343 	space_map_t ms_map;
1344 	space_map_obj_t ms_smo;
1345 	space_map_obj_t ms_smo_syncing;
1346 } mdb_metaslab_t;
1347 
1348 typedef struct space_data {
1349 	uint64_t ms_allocmap[TXG_SIZE];
1350 	uint64_t ms_freemap[TXG_SIZE];
1351 	uint64_t ms_map;
1352 	uint64_t avail;
1353 	uint64_t nowavail;
1354 } space_data_t;
1355 
1356 /* ARGSUSED */
1357 static int
1358 space_cb(uintptr_t addr, const void *unknown, void *arg)
1359 {
1360 	space_data_t *sd = arg;
1361 	mdb_metaslab_t ms;
1362 
1363 	if (GETMEMB(addr, struct metaslab, ms_allocmap, ms.ms_allocmap) ||
1364 	    GETMEMB(addr, struct metaslab, ms_freemap, ms.ms_freemap) ||
1365 	    GETMEMB(addr, struct metaslab, ms_map, ms.ms_map) ||
1366 	    GETMEMB(addr, struct metaslab, ms_smo, ms.ms_smo) ||
1367 	    GETMEMB(addr, struct metaslab, ms_smo_syncing, ms.ms_smo_syncing)) {
1368 		return (WALK_ERR);
1369 	}
1370 
1371 	sd->ms_allocmap[0] += ms.ms_allocmap[0].sm_space;
1372 	sd->ms_allocmap[1] += ms.ms_allocmap[1].sm_space;
1373 	sd->ms_allocmap[2] += ms.ms_allocmap[2].sm_space;
1374 	sd->ms_allocmap[3] += ms.ms_allocmap[3].sm_space;
1375 	sd->ms_freemap[0] += ms.ms_freemap[0].sm_space;
1376 	sd->ms_freemap[1] += ms.ms_freemap[1].sm_space;
1377 	sd->ms_freemap[2] += ms.ms_freemap[2].sm_space;
1378 	sd->ms_freemap[3] += ms.ms_freemap[3].sm_space;
1379 	sd->ms_map += ms.ms_map.sm_space;
1380 	sd->avail += ms.ms_map.sm_size - ms.ms_smo.smo_alloc;
1381 	sd->nowavail += ms.ms_map.sm_size - ms.ms_smo_syncing.smo_alloc;
1382 
1383 	return (WALK_NEXT);
1384 }
1385 
1386 /*
1387  * ::spa_space [-b]
1388  *
1389  * Given a spa_t, print out it's on-disk space usage and in-core
1390  * estimates of future usage.  If -b is given, print space in bytes.
1391  * Otherwise print in megabytes.
1392  */
1393 /* ARGSUSED */
1394 static int
1395 spa_space(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1396 {
1397 	mdb_spa_t spa;
1398 	uintptr_t dp_root_dir;
1399 	mdb_dsl_dir_t dd;
1400 	mdb_dsl_dir_phys_t dsp;
1401 	uint64_t children;
1402 	uintptr_t childaddr;
1403 	space_data_t sd;
1404 	int shift = 20;
1405 	char *suffix = "M";
1406 	int bits = FALSE;
1407 
1408 	if (mdb_getopts(argc, argv, 'b', MDB_OPT_SETBITS, TRUE, &bits, NULL) !=
1409 	    argc)
1410 		return (DCMD_USAGE);
1411 	if (!(flags & DCMD_ADDRSPEC))
1412 		return (DCMD_USAGE);
1413 
1414 	if (bits) {
1415 		shift = 0;
1416 		suffix = "";
1417 	}
1418 
1419 	if (GETMEMB(addr, struct spa, spa_dsl_pool, spa.spa_dsl_pool) ||
1420 	    GETMEMB(addr, struct spa, spa_root_vdev, spa.spa_root_vdev) ||
1421 	    GETMEMB(spa.spa_root_vdev, struct vdev, vdev_children, children) ||
1422 	    GETMEMB(spa.spa_root_vdev, struct vdev, vdev_child, childaddr) ||
1423 	    GETMEMB(spa.spa_dsl_pool, struct dsl_pool,
1424 	    dp_root_dir, dp_root_dir) ||
1425 	    GETMEMB(dp_root_dir, struct dsl_dir, dd_phys, dd.dd_phys) ||
1426 	    GETMEMB(dp_root_dir, struct dsl_dir,
1427 	    dd_space_towrite, dd.dd_space_towrite) ||
1428 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1429 	    dd_used_bytes, dsp.dd_used_bytes) ||
1430 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1431 	    dd_compressed_bytes, dsp.dd_compressed_bytes) ||
1432 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1433 	    dd_uncompressed_bytes, dsp.dd_uncompressed_bytes)) {
1434 		return (DCMD_ERR);
1435 	}
1436 
1437 	mdb_printf("dd_space_towrite = %llu%s %llu%s %llu%s %llu%s\n",
1438 	    dd.dd_space_towrite[0] >> shift, suffix,
1439 	    dd.dd_space_towrite[1] >> shift, suffix,
1440 	    dd.dd_space_towrite[2] >> shift, suffix,
1441 	    dd.dd_space_towrite[3] >> shift, suffix);
1442 
1443 	mdb_printf("dd_phys.dd_used_bytes = %llu%s\n",
1444 	    dsp.dd_used_bytes >> shift, suffix);
1445 	mdb_printf("dd_phys.dd_compressed_bytes = %llu%s\n",
1446 	    dsp.dd_compressed_bytes >> shift, suffix);
1447 	mdb_printf("dd_phys.dd_uncompressed_bytes = %llu%s\n",
1448 	    dsp.dd_uncompressed_bytes >> shift, suffix);
1449 
1450 	bzero(&sd, sizeof (sd));
1451 	if (mdb_pwalk("metaslab", space_cb, &sd, addr) != 0) {
1452 		mdb_warn("can't walk metaslabs");
1453 		return (DCMD_ERR);
1454 	}
1455 
1456 	mdb_printf("ms_allocmap = %llu%s %llu%s %llu%s %llu%s\n",
1457 	    sd.ms_allocmap[0] >> shift, suffix,
1458 	    sd.ms_allocmap[1] >> shift, suffix,
1459 	    sd.ms_allocmap[2] >> shift, suffix,
1460 	    sd.ms_allocmap[3] >> shift, suffix);
1461 	mdb_printf("ms_freemap = %llu%s %llu%s %llu%s %llu%s\n",
1462 	    sd.ms_freemap[0] >> shift, suffix,
1463 	    sd.ms_freemap[1] >> shift, suffix,
1464 	    sd.ms_freemap[2] >> shift, suffix,
1465 	    sd.ms_freemap[3] >> shift, suffix);
1466 	mdb_printf("ms_map = %llu%s\n", sd.ms_map >> shift, suffix);
1467 	mdb_printf("last synced avail = %llu%s\n", sd.avail >> shift, suffix);
1468 	mdb_printf("current syncing avail = %llu%s\n",
1469 	    sd.nowavail >> shift, suffix);
1470 
1471 	return (DCMD_OK);
1472 }
1473 
1474 /*
1475  * ::spa_verify
1476  *
1477  * Given a spa_t, verify that that the pool is self-consistent.
1478  * Currently, it only checks to make sure that the vdev tree exists.
1479  */
1480 /* ARGSUSED */
1481 static int
1482 spa_verify(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1483 {
1484 	spa_t spa;
1485 
1486 	if (argc != 0 || !(flags & DCMD_ADDRSPEC))
1487 		return (DCMD_USAGE);
1488 
1489 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1490 		mdb_warn("failed to read spa_t at %p", addr);
1491 		return (DCMD_ERR);
1492 	}
1493 
1494 	if (spa.spa_root_vdev == NULL) {
1495 		mdb_printf("no vdev tree present\n");
1496 		return (DCMD_OK);
1497 	}
1498 
1499 	return (DCMD_OK);
1500 }
1501 
1502 static int
1503 spa_print_aux(spa_aux_vdev_t *sav, uint_t flags, mdb_arg_t *v,
1504     const char *name)
1505 {
1506 	uintptr_t *aux;
1507 	size_t len;
1508 	int ret, i;
1509 
1510 	/*
1511 	 * Iterate over aux vdevs and print those out as well.  This is a
1512 	 * little annoying because we don't have a root vdev to pass to ::vdev.
1513 	 * Instead, we print a single line and then call it for each child
1514 	 * vdev.
1515 	 */
1516 	if (sav->sav_count != 0) {
1517 		v[1].a_type = MDB_TYPE_STRING;
1518 		v[1].a_un.a_str = "-d";
1519 		v[2].a_type = MDB_TYPE_IMMEDIATE;
1520 		v[2].a_un.a_val = 2;
1521 
1522 		len = sav->sav_count * sizeof (uintptr_t);
1523 		aux = mdb_alloc(len, UM_SLEEP);
1524 		if (mdb_vread(aux, len,
1525 		    (uintptr_t)sav->sav_vdevs) == -1) {
1526 			mdb_free(aux, len);
1527 			mdb_warn("failed to read l2cache vdevs at %p",
1528 			    sav->sav_vdevs);
1529 			return (DCMD_ERR);
1530 		}
1531 
1532 		mdb_printf("%-?s %-9s %-12s %s\n", "-", "-", "-", name);
1533 
1534 		for (i = 0; i < sav->sav_count; i++) {
1535 			ret = mdb_call_dcmd("vdev", aux[i], flags, 3, v);
1536 			if (ret != DCMD_OK) {
1537 				mdb_free(aux, len);
1538 				return (ret);
1539 			}
1540 		}
1541 
1542 		mdb_free(aux, len);
1543 	}
1544 
1545 	return (0);
1546 }
1547 
1548 /*
1549  * ::spa_vdevs
1550  *
1551  * 	-e	Include error stats
1552  *
1553  * Print out a summarized list of vdevs for the given spa_t.
1554  * This is accomplished by invoking "::vdev -re" on the root vdev, as well as
1555  * iterating over the cache devices.
1556  */
1557 /* ARGSUSED */
1558 static int
1559 spa_vdevs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1560 {
1561 	spa_t spa;
1562 	mdb_arg_t v[3];
1563 	int errors = FALSE;
1564 	int ret;
1565 
1566 	if (mdb_getopts(argc, argv,
1567 	    'e', MDB_OPT_SETBITS, TRUE, &errors,
1568 	    NULL) != argc)
1569 		return (DCMD_USAGE);
1570 
1571 	if (!(flags & DCMD_ADDRSPEC))
1572 		return (DCMD_USAGE);
1573 
1574 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1575 		mdb_warn("failed to read spa_t at %p", addr);
1576 		return (DCMD_ERR);
1577 	}
1578 
1579 	/*
1580 	 * Unitialized spa_t structures can have a NULL root vdev.
1581 	 */
1582 	if (spa.spa_root_vdev == NULL) {
1583 		mdb_printf("no associated vdevs\n");
1584 		return (DCMD_OK);
1585 	}
1586 
1587 	v[0].a_type = MDB_TYPE_STRING;
1588 	v[0].a_un.a_str = errors ? "-re" : "-r";
1589 
1590 	ret = mdb_call_dcmd("vdev", (uintptr_t)spa.spa_root_vdev,
1591 	    flags, 1, v);
1592 	if (ret != DCMD_OK)
1593 		return (ret);
1594 
1595 	if (spa_print_aux(&spa.spa_l2cache, flags, v, "cache") != 0 ||
1596 	    spa_print_aux(&spa.spa_spares, flags, v, "spares") != 0)
1597 		return (DCMD_ERR);
1598 
1599 	return (DCMD_OK);
1600 }
1601 
1602 /*
1603  * ::zio
1604  *
1605  * Print a summary of zio_t and all its children.  This is intended to display a
1606  * zio tree, and hence we only pick the most important pieces of information for
1607  * the main summary.  More detailed information can always be found by doing a
1608  * '::print zio' on the underlying zio_t.  The columns we display are:
1609  *
1610  *	ADDRESS		TYPE	STAGE		WAITER
1611  *
1612  * The 'address' column is indented by one space for each depth level as we
1613  * descend down the tree.
1614  */
1615 
1616 #define	ZIO_MAXINDENT	24
1617 #define	ZIO_MAXWIDTH	(sizeof (uintptr_t) * 2 + ZIO_MAXINDENT)
1618 #define	ZIO_WALK_SELF	0
1619 #define	ZIO_WALK_CHILD	1
1620 #define	ZIO_WALK_PARENT	2
1621 
1622 typedef struct zio_print_args {
1623 	int	zpa_current_depth;
1624 	int	zpa_min_depth;
1625 	int	zpa_max_depth;
1626 	int	zpa_type;
1627 	uint_t	zpa_flags;
1628 } zio_print_args_t;
1629 
1630 static int zio_child_cb(uintptr_t addr, const void *unknown, void *arg);
1631 
1632 static int
1633 zio_print_cb(uintptr_t addr, const void *data, void *priv)
1634 {
1635 	const zio_t *zio = data;
1636 	zio_print_args_t *zpa = priv;
1637 	mdb_ctf_id_t type_enum, stage_enum;
1638 	int indent = zpa->zpa_current_depth;
1639 	const char *type, *stage;
1640 	uintptr_t laddr;
1641 
1642 	if (indent > ZIO_MAXINDENT)
1643 		indent = ZIO_MAXINDENT;
1644 
1645 	if (mdb_ctf_lookup_by_name("enum zio_type", &type_enum) == -1 ||
1646 	    mdb_ctf_lookup_by_name("enum zio_stage", &stage_enum) == -1) {
1647 		mdb_warn("failed to lookup zio enums");
1648 		return (WALK_ERR);
1649 	}
1650 
1651 	if ((type = mdb_ctf_enum_name(type_enum, zio->io_type)) != NULL)
1652 		type += sizeof ("ZIO_TYPE_") - 1;
1653 	else
1654 		type = "?";
1655 
1656 	if ((stage = mdb_ctf_enum_name(stage_enum, zio->io_stage)) != NULL)
1657 		stage += sizeof ("ZIO_STAGE_") - 1;
1658 	else
1659 		stage = "?";
1660 
1661 	if (zpa->zpa_current_depth >= zpa->zpa_min_depth) {
1662 		if (zpa->zpa_flags & DCMD_PIPE_OUT) {
1663 			mdb_printf("%?p\n", addr);
1664 		} else {
1665 			mdb_printf("%*s%-*p %-5s %-16s ", indent, "",
1666 			    ZIO_MAXWIDTH - indent, addr, type, stage);
1667 			if (zio->io_waiter)
1668 				mdb_printf("%?p\n", zio->io_waiter);
1669 			else
1670 				mdb_printf("-\n");
1671 		}
1672 	}
1673 
1674 	if (zpa->zpa_current_depth >= zpa->zpa_max_depth)
1675 		return (WALK_NEXT);
1676 
1677 	if (zpa->zpa_type == ZIO_WALK_PARENT)
1678 		laddr = addr + OFFSETOF(zio_t, io_parent_list);
1679 	else
1680 		laddr = addr + OFFSETOF(zio_t, io_child_list);
1681 
1682 	zpa->zpa_current_depth++;
1683 	if (mdb_pwalk("list", zio_child_cb, zpa, laddr) != 0) {
1684 		mdb_warn("failed to walk zio_t children at %p\n", laddr);
1685 		return (WALK_ERR);
1686 	}
1687 	zpa->zpa_current_depth--;
1688 
1689 	return (WALK_NEXT);
1690 }
1691 
1692 /* ARGSUSED */
1693 static int
1694 zio_child_cb(uintptr_t addr, const void *unknown, void *arg)
1695 {
1696 	zio_link_t zl;
1697 	zio_t zio;
1698 	uintptr_t ziop;
1699 	zio_print_args_t *zpa = arg;
1700 
1701 	if (mdb_vread(&zl, sizeof (zl), addr) == -1) {
1702 		mdb_warn("failed to read zio_link_t at %p", addr);
1703 		return (WALK_ERR);
1704 	}
1705 
1706 	if (zpa->zpa_type == ZIO_WALK_PARENT)
1707 		ziop = (uintptr_t)zl.zl_parent;
1708 	else
1709 		ziop = (uintptr_t)zl.zl_child;
1710 
1711 	if (mdb_vread(&zio, sizeof (zio_t), ziop) == -1) {
1712 		mdb_warn("failed to read zio_t at %p", ziop);
1713 		return (WALK_ERR);
1714 	}
1715 
1716 	return (zio_print_cb(ziop, &zio, arg));
1717 }
1718 
1719 /* ARGSUSED */
1720 static int
1721 zio_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1722 {
1723 	zio_t zio;
1724 	zio_print_args_t zpa = { 0 };
1725 
1726 	if (!(flags & DCMD_ADDRSPEC))
1727 		return (DCMD_USAGE);
1728 
1729 	if (mdb_getopts(argc, argv,
1730 	    'r', MDB_OPT_SETBITS, INT_MAX, &zpa.zpa_max_depth,
1731 	    'c', MDB_OPT_SETBITS, ZIO_WALK_CHILD, &zpa.zpa_type,
1732 	    'p', MDB_OPT_SETBITS, ZIO_WALK_PARENT, &zpa.zpa_type,
1733 	    NULL) != argc)
1734 		return (DCMD_USAGE);
1735 
1736 	zpa.zpa_flags = flags;
1737 	if (zpa.zpa_max_depth != 0) {
1738 		if (zpa.zpa_type == ZIO_WALK_SELF)
1739 			zpa.zpa_type = ZIO_WALK_CHILD;
1740 	} else if (zpa.zpa_type != ZIO_WALK_SELF) {
1741 		zpa.zpa_min_depth = 1;
1742 		zpa.zpa_max_depth = 1;
1743 	}
1744 
1745 	if (mdb_vread(&zio, sizeof (zio_t), addr) == -1) {
1746 		mdb_warn("failed to read zio_t at %p", addr);
1747 		return (DCMD_ERR);
1748 	}
1749 
1750 	if (!(flags & DCMD_PIPE_OUT) && DCMD_HDRSPEC(flags))
1751 		mdb_printf("%<u>%-*s %-5s %-16s %-?s%</u>\n", ZIO_MAXWIDTH,
1752 		    "ADDRESS", "TYPE", "STAGE", "WAITER");
1753 
1754 	if (zio_print_cb(addr, &zio, &zpa) != WALK_NEXT)
1755 		return (DCMD_ERR);
1756 
1757 	return (DCMD_OK);
1758 }
1759 
1760 /*
1761  * [addr]::zio_state
1762  *
1763  * Print a summary of all zio_t structures on the system, or for a particular
1764  * pool.  This is equivalent to '::walk zio_root | ::zio'.
1765  */
1766 /*ARGSUSED*/
1767 static int
1768 zio_state(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1769 {
1770 	/*
1771 	 * MDB will remember the last address of the pipeline, so if we don't
1772 	 * zero this we'll end up trying to walk zio structures for a
1773 	 * non-existent spa_t.
1774 	 */
1775 	if (!(flags & DCMD_ADDRSPEC))
1776 		addr = 0;
1777 
1778 	return (mdb_pwalk_dcmd("zio_root", "zio", argc, argv, addr));
1779 }
1780 
1781 typedef struct txg_list_walk_data {
1782 	uintptr_t lw_head[TXG_SIZE];
1783 	int	lw_txgoff;
1784 	int	lw_maxoff;
1785 	size_t	lw_offset;
1786 	void	*lw_obj;
1787 } txg_list_walk_data_t;
1788 
1789 static int
1790 txg_list_walk_init_common(mdb_walk_state_t *wsp, int txg, int maxoff)
1791 {
1792 	txg_list_walk_data_t *lwd;
1793 	txg_list_t list;
1794 	int i;
1795 
1796 	lwd = mdb_alloc(sizeof (txg_list_walk_data_t), UM_SLEEP | UM_GC);
1797 	if (mdb_vread(&list, sizeof (txg_list_t), wsp->walk_addr) == -1) {
1798 		mdb_warn("failed to read txg_list_t at %#lx", wsp->walk_addr);
1799 		return (WALK_ERR);
1800 	}
1801 
1802 	for (i = 0; i < TXG_SIZE; i++)
1803 		lwd->lw_head[i] = (uintptr_t)list.tl_head[i];
1804 	lwd->lw_offset = list.tl_offset;
1805 	lwd->lw_obj = mdb_alloc(lwd->lw_offset + sizeof (txg_node_t),
1806 	    UM_SLEEP | UM_GC);
1807 	lwd->lw_txgoff = txg;
1808 	lwd->lw_maxoff = maxoff;
1809 
1810 	wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
1811 	wsp->walk_data = lwd;
1812 
1813 	return (WALK_NEXT);
1814 }
1815 
1816 static int
1817 txg_list_walk_init(mdb_walk_state_t *wsp)
1818 {
1819 	return (txg_list_walk_init_common(wsp, 0, TXG_SIZE-1));
1820 }
1821 
1822 static int
1823 txg_list0_walk_init(mdb_walk_state_t *wsp)
1824 {
1825 	return (txg_list_walk_init_common(wsp, 0, 0));
1826 }
1827 
1828 static int
1829 txg_list1_walk_init(mdb_walk_state_t *wsp)
1830 {
1831 	return (txg_list_walk_init_common(wsp, 1, 1));
1832 }
1833 
1834 static int
1835 txg_list2_walk_init(mdb_walk_state_t *wsp)
1836 {
1837 	return (txg_list_walk_init_common(wsp, 2, 2));
1838 }
1839 
1840 static int
1841 txg_list3_walk_init(mdb_walk_state_t *wsp)
1842 {
1843 	return (txg_list_walk_init_common(wsp, 3, 3));
1844 }
1845 
1846 static int
1847 txg_list_walk_step(mdb_walk_state_t *wsp)
1848 {
1849 	txg_list_walk_data_t *lwd = wsp->walk_data;
1850 	uintptr_t addr;
1851 	txg_node_t *node;
1852 	int status;
1853 
1854 	while (wsp->walk_addr == NULL && lwd->lw_txgoff < lwd->lw_maxoff) {
1855 		lwd->lw_txgoff++;
1856 		wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
1857 	}
1858 
1859 	if (wsp->walk_addr == NULL)
1860 		return (WALK_DONE);
1861 
1862 	addr = wsp->walk_addr - lwd->lw_offset;
1863 
1864 	if (mdb_vread(lwd->lw_obj,
1865 	    lwd->lw_offset + sizeof (txg_node_t), addr) == -1) {
1866 		mdb_warn("failed to read list element at %#lx", addr);
1867 		return (WALK_ERR);
1868 	}
1869 
1870 	status = wsp->walk_callback(addr, lwd->lw_obj, wsp->walk_cbdata);
1871 	node = (txg_node_t *)((uintptr_t)lwd->lw_obj + lwd->lw_offset);
1872 	wsp->walk_addr = (uintptr_t)node->tn_next[lwd->lw_txgoff];
1873 
1874 	return (status);
1875 }
1876 
1877 /*
1878  * ::walk spa
1879  *
1880  * Walk all named spa_t structures in the namespace.  This is nothing more than
1881  * a layered avl walk.
1882  */
1883 static int
1884 spa_walk_init(mdb_walk_state_t *wsp)
1885 {
1886 	GElf_Sym sym;
1887 
1888 	if (wsp->walk_addr != NULL) {
1889 		mdb_warn("spa walk only supports global walks\n");
1890 		return (WALK_ERR);
1891 	}
1892 
1893 	if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "spa_namespace_avl", &sym) == -1) {
1894 		mdb_warn("failed to find symbol 'spa_namespace_avl'");
1895 		return (WALK_ERR);
1896 	}
1897 
1898 	wsp->walk_addr = (uintptr_t)sym.st_value;
1899 
1900 	if (mdb_layered_walk("avl", wsp) == -1) {
1901 		mdb_warn("failed to walk 'avl'\n");
1902 		return (WALK_ERR);
1903 	}
1904 
1905 	return (WALK_NEXT);
1906 }
1907 
1908 static int
1909 spa_walk_step(mdb_walk_state_t *wsp)
1910 {
1911 	spa_t	spa;
1912 
1913 	if (mdb_vread(&spa, sizeof (spa), wsp->walk_addr) == -1) {
1914 		mdb_warn("failed to read spa_t at %p", wsp->walk_addr);
1915 		return (WALK_ERR);
1916 	}
1917 
1918 	return (wsp->walk_callback(wsp->walk_addr, &spa, wsp->walk_cbdata));
1919 }
1920 
1921 /*
1922  * [addr]::walk zio
1923  *
1924  * Walk all active zio_t structures on the system.  This is simply a layered
1925  * walk on top of ::walk zio_cache, with the optional ability to limit the
1926  * structures to a particular pool.
1927  */
1928 static int
1929 zio_walk_init(mdb_walk_state_t *wsp)
1930 {
1931 	wsp->walk_data = (void *)wsp->walk_addr;
1932 
1933 	if (mdb_layered_walk("zio_cache", wsp) == -1) {
1934 		mdb_warn("failed to walk 'zio_cache'\n");
1935 		return (WALK_ERR);
1936 	}
1937 
1938 	return (WALK_NEXT);
1939 }
1940 
1941 static int
1942 zio_walk_step(mdb_walk_state_t *wsp)
1943 {
1944 	zio_t zio;
1945 
1946 	if (mdb_vread(&zio, sizeof (zio), wsp->walk_addr) == -1) {
1947 		mdb_warn("failed to read zio_t at %p", wsp->walk_addr);
1948 		return (WALK_ERR);
1949 	}
1950 
1951 	if (wsp->walk_data != NULL && wsp->walk_data != zio.io_spa)
1952 		return (WALK_NEXT);
1953 
1954 	return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
1955 }
1956 
1957 /*
1958  * [addr]::walk zio_root
1959  *
1960  * Walk only root zio_t structures, optionally for a particular spa_t.
1961  */
1962 static int
1963 zio_walk_root_step(mdb_walk_state_t *wsp)
1964 {
1965 	zio_t zio;
1966 
1967 	if (mdb_vread(&zio, sizeof (zio), wsp->walk_addr) == -1) {
1968 		mdb_warn("failed to read zio_t at %p", wsp->walk_addr);
1969 		return (WALK_ERR);
1970 	}
1971 
1972 	if (wsp->walk_data != NULL && wsp->walk_data != zio.io_spa)
1973 		return (WALK_NEXT);
1974 
1975 	/* If the parent list is not empty, ignore */
1976 	if (zio.io_parent_list.list_head.list_next !=
1977 	    &((zio_t *)wsp->walk_addr)->io_parent_list.list_head)
1978 		return (WALK_NEXT);
1979 
1980 	return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
1981 }
1982 
1983 #define	NICENUM_BUFLEN 6
1984 
1985 static int
1986 snprintfrac(char *buf, int len,
1987     uint64_t numerator, uint64_t denom, int frac_digits)
1988 {
1989 	int mul = 1;
1990 	int whole, frac, i;
1991 
1992 	for (i = frac_digits; i; i--)
1993 		mul *= 10;
1994 	whole = numerator / denom;
1995 	frac = mul * numerator / denom - mul * whole;
1996 	return (mdb_snprintf(buf, len, "%u.%0*u", whole, frac_digits, frac));
1997 }
1998 
1999 static void
2000 mdb_nicenum(uint64_t num, char *buf)
2001 {
2002 	uint64_t n = num;
2003 	int index = 0;
2004 	char *u;
2005 
2006 	while (n >= 1024) {
2007 		n = (n + (1024 / 2)) / 1024; /* Round up or down */
2008 		index++;
2009 	}
2010 
2011 	u = &" \0K\0M\0G\0T\0P\0E\0"[index*2];
2012 
2013 	if (index == 0) {
2014 		(void) mdb_snprintf(buf, NICENUM_BUFLEN, "%llu",
2015 		    (u_longlong_t)n);
2016 	} else if (n < 10 && (num & (num - 1)) != 0) {
2017 		(void) snprintfrac(buf, NICENUM_BUFLEN,
2018 		    num, 1ULL << 10 * index, 2);
2019 		strcat(buf, u);
2020 	} else if (n < 100 && (num & (num - 1)) != 0) {
2021 		(void) snprintfrac(buf, NICENUM_BUFLEN,
2022 		    num, 1ULL << 10 * index, 1);
2023 		strcat(buf, u);
2024 	} else {
2025 		(void) mdb_snprintf(buf, NICENUM_BUFLEN, "%llu%s",
2026 		    (u_longlong_t)n, u);
2027 	}
2028 }
2029 
2030 /*
2031  * ::zfs_blkstats
2032  *
2033  * 	-v	print verbose per-level information
2034  *
2035  */
2036 static int
2037 zfs_blkstats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2038 {
2039 	boolean_t verbose = B_FALSE;
2040 	zfs_all_blkstats_t stats;
2041 	dmu_object_type_t t;
2042 	zfs_blkstat_t *tzb;
2043 	uint64_t ditto;
2044 	dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES + 10];
2045 	/* +10 in case it grew */
2046 
2047 	if (mdb_readvar(&dmu_ot, "dmu_ot") == -1) {
2048 		mdb_warn("failed to read 'dmu_ot'");
2049 		return (DCMD_ERR);
2050 	}
2051 
2052 	if (mdb_getopts(argc, argv,
2053 	    'v', MDB_OPT_SETBITS, TRUE, &verbose,
2054 	    NULL) != argc)
2055 		return (DCMD_USAGE);
2056 
2057 	if (!(flags & DCMD_ADDRSPEC))
2058 		return (DCMD_USAGE);
2059 
2060 	if (GETMEMB(addr, struct spa, spa_dsl_pool, addr) ||
2061 	    GETMEMB(addr, struct dsl_pool, dp_blkstats, addr) ||
2062 	    mdb_vread(&stats, sizeof (zfs_all_blkstats_t), addr) == -1) {
2063 		mdb_warn("failed to read data at %p;", addr);
2064 		mdb_printf("maybe no stats? run \"zpool scrub\" first.");
2065 		return (DCMD_ERR);
2066 	}
2067 
2068 	tzb = &stats.zab_type[DN_MAX_LEVELS][DMU_OT_NUMTYPES];
2069 	if (tzb->zb_gangs != 0) {
2070 		mdb_printf("Ganged blocks: %llu\n",
2071 		    (longlong_t)tzb->zb_gangs);
2072 	}
2073 
2074 	ditto = tzb->zb_ditto_2_of_2_samevdev + tzb->zb_ditto_2_of_3_samevdev +
2075 	    tzb->zb_ditto_3_of_3_samevdev;
2076 	if (ditto != 0) {
2077 		mdb_printf("Dittoed blocks on same vdev: %llu\n",
2078 		    (longlong_t)ditto);
2079 	}
2080 
2081 	mdb_printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
2082 	    "\t  avg\t comp\t%%Total\tType\n");
2083 
2084 	for (t = 0; t <= DMU_OT_NUMTYPES; t++) {
2085 		char csize[NICENUM_BUFLEN], lsize[NICENUM_BUFLEN];
2086 		char psize[NICENUM_BUFLEN], asize[NICENUM_BUFLEN];
2087 		char avg[NICENUM_BUFLEN];
2088 		char comp[NICENUM_BUFLEN], pct[NICENUM_BUFLEN];
2089 		char typename[64];
2090 		int l;
2091 
2092 
2093 		if (t == DMU_OT_DEFERRED)
2094 			strcpy(typename, "deferred free");
2095 		else if (t == DMU_OT_TOTAL)
2096 			strcpy(typename, "Total");
2097 		else if (mdb_readstr(typename, sizeof (typename),
2098 		    (uintptr_t)dmu_ot[t].ot_name) == -1) {
2099 			mdb_warn("failed to read type name");
2100 			return (DCMD_ERR);
2101 		}
2102 
2103 		if (stats.zab_type[DN_MAX_LEVELS][t].zb_asize == 0)
2104 			continue;
2105 
2106 		for (l = -1; l < DN_MAX_LEVELS; l++) {
2107 			int level = (l == -1 ? DN_MAX_LEVELS : l);
2108 			zfs_blkstat_t *zb = &stats.zab_type[level][t];
2109 
2110 			if (zb->zb_asize == 0)
2111 				continue;
2112 
2113 			/*
2114 			 * Don't print each level unless requested.
2115 			 */
2116 			if (!verbose && level != DN_MAX_LEVELS)
2117 				continue;
2118 
2119 			/*
2120 			 * If all the space is level 0, don't print the
2121 			 * level 0 separately.
2122 			 */
2123 			if (level == 0 && zb->zb_asize ==
2124 			    stats.zab_type[DN_MAX_LEVELS][t].zb_asize)
2125 				continue;
2126 
2127 			mdb_nicenum(zb->zb_count, csize);
2128 			mdb_nicenum(zb->zb_lsize, lsize);
2129 			mdb_nicenum(zb->zb_psize, psize);
2130 			mdb_nicenum(zb->zb_asize, asize);
2131 			mdb_nicenum(zb->zb_asize / zb->zb_count, avg);
2132 			(void) snprintfrac(comp, NICENUM_BUFLEN,
2133 			    zb->zb_lsize, zb->zb_psize, 2);
2134 			(void) snprintfrac(pct, NICENUM_BUFLEN,
2135 			    100 * zb->zb_asize, tzb->zb_asize, 2);
2136 
2137 			mdb_printf("%6s\t%5s\t%5s\t%5s\t%5s"
2138 			    "\t%5s\t%6s\t",
2139 			    csize, lsize, psize, asize, avg, comp, pct);
2140 
2141 			if (level == DN_MAX_LEVELS)
2142 				mdb_printf("%s\n", typename);
2143 			else
2144 				mdb_printf("  L%d %s\n",
2145 				    level, typename);
2146 		}
2147 	}
2148 
2149 	return (DCMD_OK);
2150 }
2151 
2152 /* ARGSUSED */
2153 static int
2154 reference_cb(uintptr_t addr, const void *ignored, void *arg)
2155 {
2156 	static int gotid;
2157 	static mdb_ctf_id_t ref_id;
2158 	uintptr_t ref_holder;
2159 	uintptr_t ref_removed;
2160 	uint64_t ref_number;
2161 	boolean_t holder_is_str;
2162 	char holder_str[128];
2163 	boolean_t removed = (boolean_t)arg;
2164 
2165 	if (!gotid) {
2166 		if (mdb_ctf_lookup_by_name("struct reference", &ref_id) == -1) {
2167 			mdb_warn("couldn't find struct reference");
2168 			return (WALK_ERR);
2169 		}
2170 		gotid = TRUE;
2171 	}
2172 
2173 	if (GETMEMBID(addr, &ref_id, ref_holder, ref_holder) ||
2174 	    GETMEMBID(addr, &ref_id, ref_removed, ref_removed) ||
2175 	    GETMEMBID(addr, &ref_id, ref_number, ref_number))
2176 		return (WALK_ERR);
2177 
2178 	if (mdb_readstr(holder_str, sizeof (holder_str), ref_holder) != -1) {
2179 		char *cp;
2180 		holder_is_str = B_TRUE;
2181 		for (cp = holder_str; *cp; cp++) {
2182 			if (!isprint(*cp)) {
2183 				holder_is_str = B_FALSE;
2184 				break;
2185 			}
2186 		}
2187 	} else {
2188 		holder_is_str = B_FALSE;
2189 	}
2190 
2191 	if (removed)
2192 		mdb_printf("removed ");
2193 	mdb_printf("reference ");
2194 	if (ref_number != 1)
2195 		mdb_printf("with count=%llu ", ref_number);
2196 	mdb_printf("with tag %p", (void*)ref_holder);
2197 	if (holder_is_str)
2198 		mdb_printf(" \"%s\"", holder_str);
2199 	mdb_printf(", held at:\n");
2200 
2201 	(void) mdb_call_dcmd("whatis", addr, DCMD_ADDRSPEC, 0, NULL);
2202 
2203 	if (removed) {
2204 		mdb_printf("removed at:\n");
2205 		(void) mdb_call_dcmd("whatis", ref_removed,
2206 		    DCMD_ADDRSPEC, 0, NULL);
2207 	}
2208 
2209 	mdb_printf("\n");
2210 
2211 	return (WALK_NEXT);
2212 }
2213 
2214 /* ARGSUSED */
2215 static int
2216 refcount(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2217 {
2218 	uint64_t rc_count, rc_removed_count;
2219 	uintptr_t rc_list, rc_removed;
2220 	static int gotid;
2221 	static mdb_ctf_id_t rc_id;
2222 	ulong_t off;
2223 
2224 	if (!(flags & DCMD_ADDRSPEC))
2225 		return (DCMD_USAGE);
2226 
2227 	if (!gotid) {
2228 		if (mdb_ctf_lookup_by_name("struct refcount", &rc_id) == -1) {
2229 			mdb_warn("couldn't find struct refcount");
2230 			return (DCMD_ERR);
2231 		}
2232 		gotid = TRUE;
2233 	}
2234 
2235 	if (GETMEMBID(addr, &rc_id, rc_count, rc_count) ||
2236 	    GETMEMBID(addr, &rc_id, rc_removed_count, rc_removed_count))
2237 		return (DCMD_ERR);
2238 
2239 	mdb_printf("refcount_t at %p has %llu current holds, "
2240 	    "%llu recently released holds\n",
2241 	    addr, (longlong_t)rc_count, (longlong_t)rc_removed_count);
2242 
2243 	if (rc_count > 0)
2244 		mdb_printf("current holds:\n");
2245 	if (mdb_ctf_offsetof(rc_id, "rc_list", &off) == -1)
2246 		return (DCMD_ERR);
2247 	rc_list = addr + off/NBBY;
2248 	mdb_pwalk("list", reference_cb, (void*)B_FALSE, rc_list);
2249 
2250 	if (rc_removed_count > 0)
2251 		mdb_printf("released holds:\n");
2252 	if (mdb_ctf_offsetof(rc_id, "rc_removed", &off) == -1)
2253 		return (DCMD_ERR);
2254 	rc_removed = addr + off/NBBY;
2255 	mdb_pwalk("list", reference_cb, (void*)B_TRUE, rc_removed);
2256 
2257 	return (DCMD_OK);
2258 }
2259 
2260 /*
2261  * MDB module linkage information:
2262  *
2263  * We declare a list of structures describing our dcmds, and a function
2264  * named _mdb_init to return a pointer to our module information.
2265  */
2266 
2267 static const mdb_dcmd_t dcmds[] = {
2268 	{ "arc", "[-bkmg]", "print ARC variables", arc_print },
2269 	{ "blkptr", ":", "print blkptr_t", blkptr },
2270 	{ "dbuf", ":", "print dmu_buf_impl_t", dbuf },
2271 	{ "dbuf_stats", ":", "dbuf stats", dbuf_stats },
2272 	{ "dbufs",
2273 	    "\t[-O objset_t*] [-n objset_name | \"mos\"] "
2274 	    "[-o object | \"mdn\"] \n"
2275 	    "\t[-l level] [-b blkid | \"bonus\"]",
2276 	    "find dmu_buf_impl_t's that match specified criteria", dbufs },
2277 	{ "abuf_find", "dva_word[0] dva_word[1]",
2278 	    "find arc_buf_hdr_t of a specified DVA",
2279 	    abuf_find },
2280 	{ "spa", "?[-cv]", "spa_t summary", spa_print },
2281 	{ "spa_config", ":", "print spa_t configuration", spa_print_config },
2282 	{ "spa_verify", ":", "verify spa_t consistency", spa_verify },
2283 	{ "spa_space", ":[-b]", "print spa_t on-disk space usage", spa_space },
2284 	{ "spa_vdevs", ":", "given a spa_t, print vdev summary", spa_vdevs },
2285 	{ "vdev", ":[-re]\n"
2286 	    "\t-r display recursively\n"
2287 	    "\t-e print statistics",
2288 	    "vdev_t summary", vdev_print },
2289 	{ "zio", ":[cpr]\n"
2290 	    "\t-c display children\n"
2291 	    "\t-p display parents\n"
2292 	    "\t-r display recursively",
2293 	    "zio_t summary", zio_print },
2294 	{ "zio_state", "?", "print out all zio_t structures on system or "
2295 	    "for a particular pool", zio_state },
2296 	{ "zio_pipeline", ":", "decode a zio pipeline", zio_pipeline },
2297 	{ "zfs_blkstats", ":[-v]",
2298 	    "given a spa_t, print block type stats from last scrub",
2299 	    zfs_blkstats },
2300 	{ "zfs_params", "", "print zfs tunable parameters", zfs_params },
2301 	{ "refcount", "", "print refcount_t holders", refcount },
2302 	{ NULL }
2303 };
2304 
2305 static const mdb_walker_t walkers[] = {
2306 	/*
2307 	 * In userland, there is no generic provider of list_t walkers, so we
2308 	 * need to add it.
2309 	 */
2310 #ifndef _KERNEL
2311 	{ LIST_WALK_NAME, LIST_WALK_DESC,
2312 		list_walk_init, list_walk_step, list_walk_fini },
2313 #endif
2314 	{ "zms_freelist", "walk ZFS metaslab freelist",
2315 		freelist_walk_init, freelist_walk_step, NULL },
2316 	{ "txg_list", "given any txg_list_t *, walk all entries in all txgs",
2317 		txg_list_walk_init, txg_list_walk_step, NULL },
2318 	{ "txg_list0", "given any txg_list_t *, walk all entries in txg 0",
2319 		txg_list0_walk_init, txg_list_walk_step, NULL },
2320 	{ "txg_list1", "given any txg_list_t *, walk all entries in txg 1",
2321 		txg_list1_walk_init, txg_list_walk_step, NULL },
2322 	{ "txg_list2", "given any txg_list_t *, walk all entries in txg 2",
2323 		txg_list2_walk_init, txg_list_walk_step, NULL },
2324 	{ "txg_list3", "given any txg_list_t *, walk all entries in txg 3",
2325 		txg_list3_walk_init, txg_list_walk_step, NULL },
2326 	{ "zio", "walk all zio structures, optionally for a particular spa_t",
2327 		zio_walk_init, zio_walk_step, NULL },
2328 	{ "zio_root", "walk all root zio_t structures, optionally for a "
2329 	    "particular spa_t",
2330 		zio_walk_init, zio_walk_root_step, NULL },
2331 	{ "spa", "walk all spa_t entries in the namespace",
2332 		spa_walk_init, spa_walk_step, NULL },
2333 	{ "metaslab", "given a spa_t *, walk all metaslab_t structures",
2334 		metaslab_walk_init, metaslab_walk_step, NULL },
2335 	{ NULL }
2336 };
2337 
2338 static const mdb_modinfo_t modinfo = {
2339 	MDB_API_VERSION, dcmds, walkers
2340 };
2341 
2342 const mdb_modinfo_t *
2343 _mdb_init(void)
2344 {
2345 	return (&modinfo);
2346 }
2347