xref: /illumos-gate/usr/src/cmd/mdb/common/modules/zfs/zfs.c (revision 60405de4d8688d96dd05157c28db3ade5c9bc234)
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 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <mdb/mdb_ctf.h>
29 #include <sys/zfs_context.h>
30 #include <sys/mdb_modapi.h>
31 #include <sys/dbuf.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_pool.h>
35 #include <sys/metaslab_impl.h>
36 #include <sys/space_map.h>
37 #include <sys/list.h>
38 #include <sys/spa_impl.h>
39 #include <sys/vdev_impl.h>
40 #include <sys/zio_compress.h>
41 
42 #ifndef _KERNEL
43 #include "../genunix/list.h"
44 #endif
45 
46 #ifdef _KERNEL
47 #define	ZFS_OBJ_NAME	"zfs"
48 #else
49 #define	ZFS_OBJ_NAME	"libzpool.so.1"
50 #endif
51 
52 static char *
53 local_strdup(const char *s)
54 {
55 	char *s1 = mdb_alloc(strlen(s) + 1, UM_SLEEP);
56 
57 	(void) strcpy(s1, s);
58 	return (s1);
59 }
60 
61 static int
62 getmember(uintptr_t addr, const char *type, mdb_ctf_id_t *idp,
63     const char *member, int len, void *buf)
64 {
65 	mdb_ctf_id_t id;
66 	ulong_t off;
67 	char name[64];
68 
69 	if (idp == NULL) {
70 		if (mdb_ctf_lookup_by_name(type, &id) == -1) {
71 			mdb_warn("couldn't find type %s", type);
72 			return (DCMD_ERR);
73 		}
74 		idp = &id;
75 	} else {
76 		type = name;
77 		mdb_ctf_type_name(*idp, name, sizeof (name));
78 	}
79 
80 	if (mdb_ctf_offsetof(*idp, member, &off) == -1) {
81 		mdb_warn("couldn't find member %s of type %s\n", member, type);
82 		return (DCMD_ERR);
83 	}
84 	if (off % 8 != 0) {
85 		mdb_warn("member %s of type %s is unsupported bitfield",
86 		    member, type);
87 		return (DCMD_ERR);
88 	}
89 	off /= 8;
90 
91 	if (mdb_vread(buf, len, addr + off) == -1) {
92 		mdb_warn("failed to read %s from %s at %p",
93 		    member, type, addr + off);
94 		return (DCMD_ERR);
95 	}
96 	/* mdb_warn("read %s from %s at %p+%llx\n", member, type, addr, off); */
97 
98 	return (0);
99 }
100 
101 #define	GETMEMB(addr, type, member, dest) \
102 	getmember(addr, #type, NULL, #member, sizeof (dest), &(dest))
103 
104 #define	GETMEMBID(addr, ctfid, member, dest) \
105 	getmember(addr, NULL, ctfid, #member, sizeof (dest), &(dest))
106 
107 static int
108 getrefcount(uintptr_t addr, mdb_ctf_id_t *id,
109     const char *member, uint64_t *rc)
110 {
111 	static int gotid;
112 	static mdb_ctf_id_t rc_id;
113 	ulong_t off;
114 
115 	if (!gotid) {
116 		if (mdb_ctf_lookup_by_name("struct refcount", &rc_id) == -1) {
117 			mdb_warn("couldn't find struct refcount");
118 			return (DCMD_ERR);
119 		}
120 		gotid = TRUE;
121 	}
122 
123 	if (mdb_ctf_offsetof(*id, member, &off) == -1) {
124 		char name[64];
125 		mdb_ctf_type_name(*id, name, sizeof (name));
126 		mdb_warn("couldn't find member %s of type %s\n", member, name);
127 		return (DCMD_ERR);
128 	}
129 	off /= 8;
130 
131 	return (GETMEMBID(addr + off, &rc_id, rc_count, *rc));
132 }
133 
134 static int
135 read_symbol(char *sym_name, void **bufp)
136 {
137 	GElf_Sym sym;
138 
139 	if (mdb_lookup_by_obj(MDB_TGT_OBJ_EVERY, sym_name, &sym)) {
140 		mdb_warn("can't find symbol %s", sym_name);
141 		return (DCMD_ERR);
142 	}
143 
144 	*bufp = mdb_alloc(sym.st_size, UM_SLEEP);
145 
146 	if (mdb_vread(*bufp, sym.st_size, sym.st_value) == -1) {
147 		mdb_warn("can't read data for symbol %s", sym_name);
148 		mdb_free(*bufp, sym.st_size);
149 		return (DCMD_ERR);
150 	}
151 
152 	return (DCMD_OK);
153 }
154 
155 static int verbose;
156 
157 static int
158 freelist_walk_init(mdb_walk_state_t *wsp)
159 {
160 	if (wsp->walk_addr == NULL) {
161 		mdb_warn("must supply starting address\n");
162 		return (WALK_ERR);
163 	}
164 
165 	wsp->walk_data = 0;  /* Index into the freelist */
166 	return (WALK_NEXT);
167 }
168 
169 static int
170 freelist_walk_step(mdb_walk_state_t *wsp)
171 {
172 	uint64_t entry;
173 	uintptr_t number = (uintptr_t)wsp->walk_data;
174 	char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "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 osi_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_impl",
254 		    &osi_id) == -1) {
255 			mdb_warn("couldn't find struct objset_impl");
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, &osi_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 blkptr(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
333 {
334 	blkptr_t bp;
335 	dmu_object_type_info_t *doti;
336 	zio_compress_info_t *zct;
337 	zio_checksum_info_t *zci;
338 	int i;
339 	char buf[MAXPATHLEN];
340 
341 	if (mdb_vread(&bp, sizeof (blkptr_t), addr) == -1) {
342 		mdb_warn("failed to read blkptr_t");
343 		return (DCMD_ERR);
344 	}
345 
346 	if (read_symbol("dmu_ot", (void **)&doti) != DCMD_OK)
347 		return (DCMD_ERR);
348 	for (i = 0; i < DMU_OT_NUMTYPES; i++) {
349 		mdb_readstr(buf, sizeof (buf), (uintptr_t)doti[i].ot_name);
350 		doti[i].ot_name = local_strdup(buf);
351 	}
352 
353 	if (read_symbol("zio_checksum_table", (void **)&zci) != DCMD_OK)
354 		return (DCMD_ERR);
355 	for (i = 0; i < ZIO_CHECKSUM_FUNCTIONS; i++) {
356 		mdb_readstr(buf, sizeof (buf), (uintptr_t)zci[i].ci_name);
357 		zci[i].ci_name = local_strdup(buf);
358 	}
359 
360 	if (read_symbol("zio_compress_table", (void **)&zct) != DCMD_OK)
361 		return (DCMD_ERR);
362 	for (i = 0; i < ZIO_COMPRESS_FUNCTIONS; i++) {
363 		mdb_readstr(buf, sizeof (buf), (uintptr_t)zct[i].ci_name);
364 		zct[i].ci_name = local_strdup(buf);
365 	}
366 
367 	/*
368 	 * Super-ick warning:  This code is also duplicated in
369 	 * cmd/zdb.c .   Yeah, I hate code replication, too.
370 	 */
371 	for (i = 0; i < BP_GET_NDVAS(&bp); i++) {
372 		dva_t *dva = &bp.blk_dva[i];
373 
374 		mdb_printf("DVA[%d]: vdev_id %lld / %llx\n", i,
375 		    DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva));
376 		mdb_printf("DVA[%d]:       GANG: %-5s  GRID:  %04x\t"
377 		    "ASIZE: %llx\n", i, DVA_GET_GANG(dva) ? "TRUE" : "FALSE",
378 		    DVA_GET_GRID(dva), DVA_GET_ASIZE(dva));
379 		mdb_printf("DVA[%d]: :%llu:%llx:%llx:%s%s%s%s\n", i,
380 		    DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), BP_GET_PSIZE(&bp),
381 		    BP_SHOULD_BYTESWAP(&bp) ? "e" : "",
382 		    !DVA_GET_GANG(dva) && BP_GET_LEVEL(&bp) != 0 ? "i" : "",
383 		    DVA_GET_GANG(dva) ? "g" : "",
384 		    BP_GET_COMPRESS(&bp) != 0 ? "d" : "");
385 	}
386 	mdb_printf("LSIZE:  %-16llx\t\tPSIZE: %llx\n",
387 	    BP_GET_LSIZE(&bp), BP_GET_PSIZE(&bp));
388 	mdb_printf("ENDIAN: %6s\t\t\t\t\tTYPE:  %s\n",
389 	    BP_GET_BYTEORDER(&bp) ? "LITTLE" : "BIG",
390 	    doti[BP_GET_TYPE(&bp)].ot_name);
391 	mdb_printf("BIRTH:  %-16llx   LEVEL: %-2d\tFILL:  %llx\n",
392 	    bp.blk_birth, BP_GET_LEVEL(&bp), bp.blk_fill);
393 	mdb_printf("CKFUNC: %-16s\t\tCOMP:  %s\n",
394 	    zci[BP_GET_CHECKSUM(&bp)].ci_name,
395 	    zct[BP_GET_COMPRESS(&bp)].ci_name);
396 	mdb_printf("CKSUM:  %llx:%llx:%llx:%llx\n",
397 	    bp.blk_cksum.zc_word[0],
398 	    bp.blk_cksum.zc_word[1],
399 	    bp.blk_cksum.zc_word[2],
400 	    bp.blk_cksum.zc_word[3]);
401 
402 	return (DCMD_OK);
403 }
404 
405 /* ARGSUSED */
406 static int
407 dbuf(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
408 {
409 	mdb_ctf_id_t id;
410 	dmu_buf_t db;
411 	uintptr_t objset;
412 	uint8_t level;
413 	uint64_t blkid;
414 	uint64_t holds;
415 	char objectname[32];
416 	char blkidname[32];
417 	char path[MAXNAMELEN];
418 
419 	if (DCMD_HDRSPEC(flags)) {
420 		mdb_printf("        addr object lvl blkid holds os\n");
421 	}
422 
423 	if (mdb_ctf_lookup_by_name("struct dmu_buf_impl", &id) == -1) {
424 		mdb_warn("couldn't find struct dmu_buf_impl_t");
425 		return (DCMD_ERR);
426 	}
427 
428 	if (GETMEMBID(addr, &id, db_objset, objset) ||
429 	    GETMEMBID(addr, &id, db, db) ||
430 	    GETMEMBID(addr, &id, db_level, level) ||
431 	    GETMEMBID(addr, &id, db_blkid, blkid)) {
432 		return (WALK_ERR);
433 	}
434 
435 	if (getrefcount(addr, &id, "db_holds", &holds)) {
436 		return (WALK_ERR);
437 	}
438 
439 	if (db.db_object == DMU_META_DNODE_OBJECT)
440 		(void) strcpy(objectname, "mdn");
441 	else
442 		(void) mdb_snprintf(objectname, sizeof (objectname), "%llx",
443 		    (u_longlong_t)db.db_object);
444 
445 	if (blkid == DB_BONUS_BLKID)
446 		(void) strcpy(blkidname, "bonus");
447 	else
448 		(void) mdb_snprintf(blkidname, sizeof (blkidname), "%llx",
449 		    (u_longlong_t)blkid);
450 
451 	if (objset_name(objset, path)) {
452 		return (WALK_ERR);
453 	}
454 
455 	mdb_printf("%p %8s %1u %9s %2llu %s\n",
456 	    addr, objectname, level, blkidname, holds, path);
457 
458 	return (DCMD_OK);
459 }
460 
461 /* ARGSUSED */
462 static int
463 dbuf_stats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
464 {
465 #define	HISTOSZ 32
466 	uintptr_t dbp;
467 	dmu_buf_impl_t db;
468 	dbuf_hash_table_t ht;
469 	uint64_t bucket, ndbufs;
470 	uint64_t histo[HISTOSZ];
471 	uint64_t histo2[HISTOSZ];
472 	int i, maxidx;
473 
474 	if (mdb_readvar(&ht, "dbuf_hash_table") == -1) {
475 		mdb_warn("failed to read 'dbuf_hash_table'");
476 		return (DCMD_ERR);
477 	}
478 
479 	for (i = 0; i < HISTOSZ; i++) {
480 		histo[i] = 0;
481 		histo2[i] = 0;
482 	}
483 
484 	ndbufs = 0;
485 	for (bucket = 0; bucket < ht.hash_table_mask+1; bucket++) {
486 		int len;
487 
488 		if (mdb_vread(&dbp, sizeof (void *),
489 		    (uintptr_t)(ht.hash_table+bucket)) == -1) {
490 			mdb_warn("failed to read hash bucket %u at %p",
491 			    bucket, ht.hash_table+bucket);
492 			return (DCMD_ERR);
493 		}
494 
495 		len = 0;
496 		while (dbp != 0) {
497 			if (mdb_vread(&db, sizeof (dmu_buf_impl_t),
498 			    dbp) == -1) {
499 				mdb_warn("failed to read dbuf at %p", dbp);
500 				return (DCMD_ERR);
501 			}
502 			dbp = (uintptr_t)db.db_hash_next;
503 			for (i = MIN(len, HISTOSZ - 1); i >= 0; i--)
504 				histo2[i]++;
505 			len++;
506 			ndbufs++;
507 		}
508 
509 		if (len >= HISTOSZ)
510 			len = HISTOSZ-1;
511 		histo[len]++;
512 	}
513 
514 	mdb_printf("hash table has %llu buckets, %llu dbufs "
515 	    "(avg %llu buckets/dbuf)\n",
516 	    ht.hash_table_mask+1, ndbufs,
517 	    (ht.hash_table_mask+1)/ndbufs);
518 
519 	mdb_printf("\n");
520 	maxidx = 0;
521 	for (i = 0; i < HISTOSZ; i++)
522 		if (histo[i] > 0)
523 			maxidx = i;
524 	mdb_printf("hash chain length	number of buckets\n");
525 	for (i = 0; i <= maxidx; i++)
526 		mdb_printf("%u			%llu\n", i, histo[i]);
527 
528 	mdb_printf("\n");
529 	maxidx = 0;
530 	for (i = 0; i < HISTOSZ; i++)
531 		if (histo2[i] > 0)
532 			maxidx = i;
533 	mdb_printf("hash chain depth	number of dbufs\n");
534 	for (i = 0; i <= maxidx; i++)
535 		mdb_printf("%u or more		%llu	%llu%%\n",
536 		    i, histo2[i], histo2[i]*100/ndbufs);
537 
538 
539 	return (DCMD_OK);
540 }
541 
542 typedef struct dbufs_data {
543 	mdb_ctf_id_t id;
544 	uint64_t objset;
545 	uint64_t object;
546 	uint64_t level;
547 	uint64_t blkid;
548 	char *osname;
549 } dbufs_data_t;
550 
551 #define	DBUFS_UNSET	(0xbaddcafedeadbeefULL)
552 
553 /* ARGSUSED */
554 static int
555 dbufs_cb(uintptr_t addr, const void *unknown, void *arg)
556 {
557 	dbufs_data_t *data = arg;
558 	uintptr_t objset;
559 	dmu_buf_t db;
560 	uint8_t level;
561 	uint64_t blkid;
562 	char osname[MAXNAMELEN];
563 
564 	if (GETMEMBID(addr, &data->id, db_objset, objset) ||
565 	    GETMEMBID(addr, &data->id, db, db) ||
566 	    GETMEMBID(addr, &data->id, db_level, level) ||
567 	    GETMEMBID(addr, &data->id, db_blkid, blkid)) {
568 		return (WALK_ERR);
569 	}
570 
571 	if ((data->objset == DBUFS_UNSET || data->objset == objset) &&
572 	    (data->osname == NULL || (objset_name(objset, osname) == 0 &&
573 		strcmp(data->osname, osname) == 0)) &&
574 	    (data->object == DBUFS_UNSET || data->object == db.db_object) &&
575 	    (data->level == DBUFS_UNSET || data->level == level) &&
576 	    (data->blkid == DBUFS_UNSET || data->blkid == blkid)) {
577 		mdb_printf("%#lr\n", addr);
578 	}
579 	return (WALK_NEXT);
580 }
581 
582 /* ARGSUSED */
583 static int
584 dbufs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
585 {
586 	dbufs_data_t data;
587 	char *object = NULL;
588 	char *blkid = NULL;
589 
590 	data.objset = data.object = data.level = data.blkid = DBUFS_UNSET;
591 	data.osname = NULL;
592 
593 	if (mdb_getopts(argc, argv,
594 	    'O', MDB_OPT_UINT64, &data.objset,
595 	    'n', MDB_OPT_STR, &data.osname,
596 	    'o', MDB_OPT_STR, &object,
597 	    'l', MDB_OPT_UINT64, &data.level,
598 	    'b', MDB_OPT_STR, &blkid) != argc) {
599 		return (DCMD_USAGE);
600 	}
601 
602 	if (object) {
603 		if (strcmp(object, "mdn") == 0) {
604 			data.object = DMU_META_DNODE_OBJECT;
605 		} else {
606 			data.object = mdb_strtoull(object);
607 		}
608 	}
609 
610 	if (blkid) {
611 		if (strcmp(blkid, "bonus") == 0) {
612 			data.blkid = DB_BONUS_BLKID;
613 		} else {
614 			data.blkid = mdb_strtoull(blkid);
615 		}
616 	}
617 
618 	if (mdb_ctf_lookup_by_name("struct dmu_buf_impl", &data.id) == -1) {
619 		mdb_warn("couldn't find struct dmu_buf_impl_t");
620 		return (DCMD_ERR);
621 	}
622 
623 	if (mdb_walk("dmu_buf_impl_t", dbufs_cb, &data) != 0) {
624 		mdb_warn("can't walk dbufs");
625 		return (DCMD_ERR);
626 	}
627 
628 	return (DCMD_OK);
629 }
630 
631 typedef struct abuf_find_data {
632 	dva_t dva;
633 	mdb_ctf_id_t id;
634 } abuf_find_data_t;
635 
636 /* ARGSUSED */
637 static int
638 abuf_find_cb(uintptr_t addr, const void *unknown, void *arg)
639 {
640 	abuf_find_data_t *data = arg;
641 	dva_t dva;
642 
643 	if (GETMEMBID(addr, &data->id, b_dva, dva)) {
644 		return (WALK_ERR);
645 	}
646 
647 	if (dva.dva_word[0] == data->dva.dva_word[0] &&
648 	    dva.dva_word[1] == data->dva.dva_word[1]) {
649 		mdb_printf("%#lr\n", addr);
650 	}
651 	return (WALK_NEXT);
652 }
653 
654 /* ARGSUSED */
655 static int
656 abuf_find(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
657 {
658 	abuf_find_data_t data;
659 	GElf_Sym sym;
660 	int i;
661 	const char *syms[] = {
662 		"ARC_mru_top",
663 		"ARC_mru_bot",
664 		"ARC_mfu_top",
665 		"ARC_mfu_bot",
666 	};
667 
668 	if (argc != 2)
669 		return (DCMD_USAGE);
670 
671 	for (i = 0; i < 2; i ++) {
672 		switch (argv[i].a_type) {
673 		case MDB_TYPE_STRING:
674 			data.dva.dva_word[i] = mdb_strtoull(argv[i].a_un.a_str);
675 			break;
676 		case MDB_TYPE_IMMEDIATE:
677 			data.dva.dva_word[i] = argv[i].a_un.a_val;
678 			break;
679 		default:
680 			return (DCMD_USAGE);
681 		}
682 	}
683 
684 	if (mdb_ctf_lookup_by_name("struct arc_buf_hdr", &data.id) == -1) {
685 		mdb_warn("couldn't find struct arc_buf_hdr");
686 		return (DCMD_ERR);
687 	}
688 
689 	for (i = 0; i < sizeof (syms) / sizeof (syms[0]); i++) {
690 		if (mdb_lookup_by_name(syms[i], &sym)) {
691 			mdb_warn("can't find symbol %s", syms[i]);
692 			return (DCMD_ERR);
693 		}
694 
695 		if (mdb_pwalk("list", abuf_find_cb, &data, sym.st_value) != 0) {
696 			mdb_warn("can't walk %s", syms[i]);
697 			return (DCMD_ERR);
698 		}
699 	}
700 
701 	return (DCMD_OK);
702 }
703 
704 void
705 abuf_help(void)
706 {
707 	mdb_printf("::abuf_find dva_word[0] dva_word[1]\n");
708 }
709 
710 /*
711  * ::spa
712  *
713  * 	-c	Print configuration information as well
714  * 	-v	Print vdev state
715  * 	-e	Print vdev error stats
716  *
717  * Print a summarized spa_t.  When given no arguments, prints out a table of all
718  * active pools on the system.
719  */
720 /* ARGSUSED */
721 static int
722 spa_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
723 {
724 	spa_t spa;
725 	char poolname[MAXNAMELEN];
726 	const char *statetab[] = { "ACTIVE", "EXPORTED", "DESTROYED",
727 		"UNINIT", "UNAVAIL" };
728 	const char *state;
729 	int config = FALSE;
730 	int vdevs = FALSE;
731 	int errors = FALSE;
732 
733 	if (mdb_getopts(argc, argv,
734 	    'c', MDB_OPT_SETBITS, TRUE, &config,
735 	    'v', MDB_OPT_SETBITS, TRUE, &vdevs,
736 	    'e', MDB_OPT_SETBITS, TRUE, &errors,
737 	    NULL) != argc)
738 		return (DCMD_USAGE);
739 
740 	if (!(flags & DCMD_ADDRSPEC)) {
741 		if (mdb_walk_dcmd("spa", "spa", argc, argv) == -1) {
742 			mdb_warn("can't walk spa");
743 			return (DCMD_ERR);
744 		}
745 
746 		return (DCMD_OK);
747 	}
748 
749 	if (flags & DCMD_PIPE_OUT) {
750 		mdb_printf("%#lr\n", addr);
751 		return (DCMD_OK);
752 	}
753 
754 	if (DCMD_HDRSPEC(flags))
755 		mdb_printf("%<u>%-?s %9s %-*s%</u>\n", "ADDR", "STATE",
756 		    sizeof (uintptr_t) == 4 ? 60 : 52, "NAME");
757 
758 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
759 		mdb_warn("failed to read spa_t at %p", addr);
760 		return (DCMD_ERR);
761 	}
762 
763 	if (mdb_readstr(poolname, sizeof (poolname), (uintptr_t)spa.spa_name)
764 	    == -1) {
765 		mdb_warn("failed to read pool name at %p", spa.spa_name);
766 		return (DCMD_ERR);
767 	}
768 
769 	if (spa.spa_state < 0 || spa.spa_state > POOL_STATE_UNAVAIL)
770 		state = "UNKNOWN";
771 	else
772 		state = statetab[spa.spa_state];
773 
774 	mdb_printf("%0?p %9s %s\n", addr, state, poolname);
775 
776 	if (config) {
777 		mdb_printf("\n");
778 		mdb_inc_indent(4);
779 		if (mdb_call_dcmd("spa_config", addr, flags, 0,
780 		    NULL) != DCMD_OK)
781 			return (DCMD_ERR);
782 		mdb_dec_indent(4);
783 	}
784 
785 	if (vdevs || errors) {
786 		mdb_arg_t v;
787 
788 		v.a_type = MDB_TYPE_STRING;
789 		v.a_un.a_str = "-e";
790 
791 		mdb_printf("\n");
792 		mdb_inc_indent(4);
793 		if (mdb_call_dcmd("spa_vdevs", addr, flags, errors ? 1 : 0,
794 		    &v) != DCMD_OK)
795 			return (DCMD_ERR);
796 		mdb_dec_indent(4);
797 	}
798 
799 	return (DCMD_OK);
800 }
801 
802 /*
803  * ::spa_config
804  *
805  * Given a spa_t, print the configuration information stored in spa_config.
806  * Since it's just an nvlist, format it as an indented list of name=value pairs.
807  * We simply read the value of spa_config and pass off to ::nvlist.
808  */
809 /* ARGSUSED */
810 static int
811 spa_print_config(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
812 {
813 	spa_t spa;
814 
815 	if (argc != 0 || !(flags & DCMD_ADDRSPEC))
816 		return (DCMD_USAGE);
817 
818 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
819 		mdb_warn("failed to read spa_t at %p", addr);
820 		return (DCMD_ERR);
821 	}
822 
823 	if (spa.spa_config == NULL) {
824 		mdb_printf("(none)\n");
825 		return (DCMD_OK);
826 	}
827 
828 	return (mdb_call_dcmd("nvlist", (uintptr_t)spa.spa_config, flags,
829 	    0, NULL));
830 }
831 
832 void
833 vdev_help(void)
834 {
835 	mdb_printf("[vdev_t*]::vdev [-qr]\n"
836 		"\t-> -q display vdev_queue parameters\n"
837 		"\t-> -r recursive (visit all children)\n");
838 }
839 
840 /*
841  * ::vdev
842  *
843  * Print out a summarized vdev_t, in the following form:
844  *
845  * ADDR             STATE	AUX            DESC
846  * fffffffbcde23df0 HEALTHY	-              /dev/dsk/c0t0d0
847  *
848  * or with "-q" to print out a vdev_t's vdev_queue parameters:
849  *
850  *  vdev_t: c26ae4c0
851  *     c26ae73c min pending         0x2
852  *     c26ae744 max pending         0x23
853  *     c26ae74c agg limit           0x20000
854  *     c26ae754 time shift          0x4
855  *     c26ae75c ramp rate           0x2
856  *
857  * If '-r' is specified, recursively visit all children.
858  *
859  * With '-e', the statistics associated with the vdev are printed as well.
860  */
861 static int
862 do_print_vdev(uintptr_t addr, int flags, int depth, int queue, int stats,
863     int recursive)
864 {
865 	vdev_t vdev;
866 	char desc[MAXNAMELEN];
867 	int c, children;
868 	uintptr_t *child;
869 	const char *state, *aux;
870 
871 	if (mdb_vread(&vdev, sizeof (vdev), (uintptr_t)addr) == -1) {
872 		mdb_warn("failed to read vdev_t at %p\n", (uintptr_t)addr);
873 		return (DCMD_ERR);
874 	}
875 
876 	if (flags & DCMD_PIPE_OUT) {
877 		mdb_printf("%#lr", addr);
878 	} else {
879 		if (vdev.vdev_path != NULL) {
880 			if (mdb_readstr(desc, sizeof (desc),
881 			    (uintptr_t)vdev.vdev_path) == -1) {
882 				mdb_warn("failed to read vdev_path at %p\n",
883 				    vdev.vdev_path);
884 				return (DCMD_ERR);
885 			}
886 		} else if (vdev.vdev_ops != NULL) {
887 			vdev_ops_t ops;
888 			if (mdb_vread(&ops, sizeof (ops),
889 			    (uintptr_t)vdev.vdev_ops) == -1) {
890 				mdb_warn("failed to read vdev_ops at %p\n",
891 				    vdev.vdev_ops);
892 				return (DCMD_ERR);
893 			}
894 			(void) strcpy(desc, ops.vdev_op_type);
895 		} else {
896 			(void) strcpy(desc, "<unknown>");
897 		}
898 
899 		if (depth == 0 && DCMD_HDRSPEC(flags))
900 			mdb_printf("%<u>%-?s %-9s %-12s %-*s%</u>\n",
901 			    "ADDR", "STATE", "AUX",
902 			    sizeof (uintptr_t) == 4 ? 43 : 35,
903 			    "DESCRIPTION");
904 
905 		mdb_printf("%0?p ", addr);
906 
907 		switch (vdev.vdev_state) {
908 		case VDEV_STATE_CLOSED:
909 		    state = "CLOSED";
910 		    break;
911 		case VDEV_STATE_OFFLINE:
912 		    state = "OFFLINE";
913 		    break;
914 		case VDEV_STATE_CANT_OPEN:
915 		    state = "CANT_OPEN";
916 		    break;
917 		case VDEV_STATE_DEGRADED:
918 		    state = "DEGRADED";
919 		    break;
920 		case VDEV_STATE_HEALTHY:
921 		    state = "HEALTHY";
922 		    break;
923 		default:
924 		    state = "UNKNOWN";
925 		    break;
926 		}
927 
928 		switch (vdev.vdev_stat.vs_aux) {
929 		case VDEV_AUX_NONE:
930 			aux = "-";
931 			break;
932 		case VDEV_AUX_OPEN_FAILED:
933 			aux = "OPEN_FAILED";
934 			break;
935 		case VDEV_AUX_CORRUPT_DATA:
936 			aux = "CORRUPT_DATA";
937 			break;
938 		case VDEV_AUX_NO_REPLICAS:
939 			aux = "NO_REPLICAS";
940 			break;
941 		case VDEV_AUX_BAD_GUID_SUM:
942 			aux = "BAD_GUID_SUM";
943 			break;
944 		case VDEV_AUX_TOO_SMALL:
945 			aux = "TOO_SMALL";
946 			break;
947 		case VDEV_AUX_BAD_LABEL:
948 			aux = "BAD_LABEL";
949 			break;
950 		default:
951 			aux = "UNKNOWN";
952 			break;
953 		}
954 
955 		mdb_printf("%-9s %-12s %*s%s\n", state, aux, depth, "", desc);
956 
957 		if (queue) {
958 			mdb_inc_indent(4);
959 			mdb_printf("\n");
960 			mdb_printf("%p min pending		0x%llx\n",
961 			    (uintptr_t)(addr + offsetof(vdev_t,
962 			    vdev_queue.vq_min_pending)),
963 			    vdev.vdev_queue.vq_min_pending);
964 			mdb_printf("%p max pending		0x%llx\n",
965 			    (uintptr_t)(addr + offsetof(vdev_t,
966 			    vdev_queue.vq_max_pending)),
967 			    vdev.vdev_queue.vq_max_pending);
968 			mdb_printf("%p agg limit		0x%llx\n",
969 			    (uintptr_t)(addr + offsetof(vdev_t,
970 			    vdev_queue.vq_agg_limit)),
971 			    vdev.vdev_queue.vq_agg_limit);
972 			mdb_printf("%p time shift		0x%llx\n",
973 			    (uintptr_t)(addr + offsetof(vdev_t,
974 			    vdev_queue.vq_time_shift)),
975 			    vdev.vdev_queue.vq_time_shift);
976 			mdb_printf("%p ramp rate 		0x%llx\n",
977 			    (uintptr_t)(addr + offsetof(vdev_t,
978 			    vdev_queue.vq_ramp_rate)),
979 			    vdev.vdev_queue.vq_ramp_rate);
980 			mdb_dec_indent(4);
981 		}
982 
983 		if (stats) {
984 			vdev_stat_t *vs = &vdev.vdev_stat;
985 			int i;
986 
987 			mdb_inc_indent(4);
988 			mdb_printf("\n");
989 			mdb_printf("%<u>       %12s %12s %12s %12s "
990 			    "%12s%</u>\n", "READ", "WRITE", "FREE", "CLAIM",
991 			    "IOCTL");
992 			mdb_printf("OPS     ");
993 			for (i = 1; i < ZIO_TYPES; i++)
994 				mdb_printf("%11#llx%s", vs->vs_ops[i],
995 				    i == ZIO_TYPES - 1 ? "" : "  ");
996 			mdb_printf("\n");
997 			mdb_printf("BYTES   ");
998 			for (i = 1; i < ZIO_TYPES; i++)
999 				mdb_printf("%11#llx%s", vs->vs_bytes[i],
1000 				    i == ZIO_TYPES - 1 ? "" : "  ");
1001 
1002 
1003 			mdb_printf("\n");
1004 			mdb_printf("EREAD    %10#llx\n", vs->vs_read_errors);
1005 			mdb_printf("EWRITE   %10#llx\n", vs->vs_write_errors);
1006 			mdb_printf("ECKSUM   %10#llx\n",
1007 			    vs->vs_checksum_errors);
1008 			mdb_dec_indent(4);
1009 		}
1010 
1011 		if (queue || stats)
1012 			mdb_printf("\n");
1013 	}
1014 
1015 	children = vdev.vdev_children;
1016 
1017 	if (children == 0 || !recursive)
1018 		return (DCMD_OK);
1019 
1020 	child = mdb_alloc(children * sizeof (void *), UM_SLEEP | UM_GC);
1021 	if (mdb_vread(child, children * sizeof (void *),
1022 	    (uintptr_t)vdev.vdev_child) == -1) {
1023 		mdb_warn("failed to read vdev children at %p", vdev.vdev_child);
1024 		return (DCMD_ERR);
1025 	}
1026 
1027 	for (c = 0; c < children; c++) {
1028 		if (do_print_vdev(child[c], flags, depth + 2, queue, stats,
1029 		    recursive))
1030 			return (DCMD_ERR);
1031 	}
1032 
1033 	return (DCMD_OK);
1034 }
1035 
1036 static int
1037 vdev_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1038 {
1039 	int print_queue = FALSE;
1040 	int recursive = FALSE;
1041 	int stats = FALSE;
1042 
1043 	if (mdb_getopts(argc, argv,
1044 	    'q', MDB_OPT_SETBITS, TRUE, &print_queue,
1045 	    'r', MDB_OPT_SETBITS, TRUE, &recursive,
1046 	    'e', MDB_OPT_SETBITS, TRUE, &stats,
1047 	    NULL) != argc)
1048 		return (DCMD_USAGE);
1049 
1050 	if (!(flags & DCMD_ADDRSPEC)) {
1051 		mdb_warn("no vdev_t address given\n");
1052 		return (DCMD_ERR);
1053 	}
1054 
1055 	return (do_print_vdev(addr, flags, 0, print_queue, stats, recursive));
1056 }
1057 
1058 typedef struct metaslab_walk_data {
1059 	uint64_t mw_numvdevs;
1060 	uintptr_t *mw_vdevs;
1061 	int mw_curvdev;
1062 	uint64_t mw_nummss;
1063 	uintptr_t *mw_mss;
1064 	int mw_curms;
1065 } metaslab_walk_data_t;
1066 
1067 static int
1068 metaslab_walk_step(mdb_walk_state_t *wsp)
1069 {
1070 	metaslab_walk_data_t *mw = wsp->walk_data;
1071 	metaslab_t ms;
1072 	uintptr_t msp;
1073 
1074 	if (mw->mw_curvdev >= mw->mw_numvdevs)
1075 		return (WALK_DONE);
1076 
1077 	if (mw->mw_mss == NULL) {
1078 		uintptr_t mssp;
1079 		uintptr_t vdevp;
1080 
1081 		ASSERT(mw->mw_curms == 0);
1082 		ASSERT(mw->mw_nummss == 0);
1083 
1084 		vdevp = mw->mw_vdevs[mw->mw_curvdev];
1085 		if (GETMEMB(vdevp, struct vdev, vdev_ms, mssp) ||
1086 		    GETMEMB(vdevp, struct vdev, vdev_ms_count, mw->mw_nummss)) {
1087 			return (WALK_ERR);
1088 		}
1089 
1090 		mw->mw_mss = mdb_alloc(mw->mw_nummss * sizeof (void*),
1091 		    UM_SLEEP | UM_GC);
1092 		if (mdb_vread(mw->mw_mss, mw->mw_nummss * sizeof (void*),
1093 		    mssp) == -1) {
1094 			mdb_warn("failed to read vdev_ms at %p", mssp);
1095 			return (WALK_ERR);
1096 		}
1097 	}
1098 
1099 	if (mw->mw_curms >= mw->mw_nummss) {
1100 		mw->mw_mss = NULL;
1101 		mw->mw_curms = 0;
1102 		mw->mw_nummss = 0;
1103 		mw->mw_curvdev++;
1104 		return (WALK_NEXT);
1105 	}
1106 
1107 	msp = mw->mw_mss[mw->mw_curms];
1108 	if (mdb_vread(&ms, sizeof (metaslab_t), msp) == -1) {
1109 		mdb_warn("failed to read metaslab_t at %p", msp);
1110 		return (WALK_ERR);
1111 	}
1112 
1113 	mw->mw_curms++;
1114 
1115 	return (wsp->walk_callback(msp, &ms, wsp->walk_cbdata));
1116 }
1117 
1118 /* ARGSUSED */
1119 static int
1120 metaslab_walk_init(mdb_walk_state_t *wsp)
1121 {
1122 	metaslab_walk_data_t *mw;
1123 	uintptr_t root_vdevp;
1124 	uintptr_t childp;
1125 
1126 	if (wsp->walk_addr == NULL) {
1127 		mdb_warn("must supply address of spa_t\n");
1128 		return (WALK_ERR);
1129 	}
1130 
1131 	mw = mdb_zalloc(sizeof (metaslab_walk_data_t), UM_SLEEP | UM_GC);
1132 
1133 	if (GETMEMB(wsp->walk_addr, struct spa, spa_root_vdev, root_vdevp) ||
1134 	    GETMEMB(root_vdevp, struct vdev, vdev_children, mw->mw_numvdevs) ||
1135 	    GETMEMB(root_vdevp, struct vdev, vdev_child, childp)) {
1136 		return (DCMD_ERR);
1137 	}
1138 
1139 	mw->mw_vdevs = mdb_alloc(mw->mw_numvdevs * sizeof (void *),
1140 	    UM_SLEEP | UM_GC);
1141 	if (mdb_vread(mw->mw_vdevs, mw->mw_numvdevs * sizeof (void *),
1142 	    childp) == -1) {
1143 		mdb_warn("failed to read root vdev children at %p", childp);
1144 		return (DCMD_ERR);
1145 	}
1146 
1147 	wsp->walk_data = mw;
1148 
1149 	return (WALK_NEXT);
1150 }
1151 
1152 typedef struct mdb_spa {
1153 	uintptr_t spa_dsl_pool;
1154 	uintptr_t spa_root_vdev;
1155 } mdb_spa_t;
1156 
1157 typedef struct mdb_dsl_dir {
1158 	uintptr_t dd_phys;
1159 	uint64_t dd_used_bytes;
1160 	int64_t dd_space_towrite[TXG_SIZE];
1161 } mdb_dsl_dir_t;
1162 
1163 typedef struct mdb_dsl_dir_phys {
1164 	uint64_t dd_used_bytes;
1165 	uint64_t dd_compressed_bytes;
1166 	uint64_t dd_uncompressed_bytes;
1167 } mdb_dsl_dir_phys_t;
1168 
1169 typedef struct mdb_vdev {
1170 	uintptr_t vdev_parent;
1171 	uintptr_t vdev_ms;
1172 	uint64_t vdev_ms_count;
1173 	vdev_stat_t vdev_stat;
1174 } mdb_vdev_t;
1175 
1176 typedef struct mdb_metaslab {
1177 	space_map_t ms_allocmap[TXG_SIZE];
1178 	space_map_t ms_freemap[TXG_SIZE];
1179 	space_map_t ms_map;
1180 	space_map_obj_t ms_smo;
1181 	space_map_obj_t ms_smo_syncing;
1182 } mdb_metaslab_t;
1183 
1184 typedef struct space_data {
1185 	uint64_t ms_allocmap[TXG_SIZE];
1186 	uint64_t ms_freemap[TXG_SIZE];
1187 	uint64_t ms_map;
1188 	uint64_t avail;
1189 	uint64_t nowavail;
1190 } space_data_t;
1191 
1192 /* ARGSUSED */
1193 static int
1194 space_cb(uintptr_t addr, const void *unknown, void *arg)
1195 {
1196 	space_data_t *sd = arg;
1197 	mdb_metaslab_t ms;
1198 
1199 	if (GETMEMB(addr, struct metaslab, ms_allocmap, ms.ms_allocmap) ||
1200 	    GETMEMB(addr, struct metaslab, ms_freemap, ms.ms_freemap) ||
1201 	    GETMEMB(addr, struct metaslab, ms_map, ms.ms_map) ||
1202 	    GETMEMB(addr, struct metaslab, ms_smo, ms.ms_smo) ||
1203 	    GETMEMB(addr, struct metaslab, ms_smo_syncing, ms.ms_smo_syncing)) {
1204 		return (WALK_ERR);
1205 	}
1206 
1207 	sd->ms_allocmap[0] += ms.ms_allocmap[0].sm_space;
1208 	sd->ms_allocmap[1] += ms.ms_allocmap[1].sm_space;
1209 	sd->ms_allocmap[2] += ms.ms_allocmap[2].sm_space;
1210 	sd->ms_allocmap[3] += ms.ms_allocmap[3].sm_space;
1211 	sd->ms_freemap[0] += ms.ms_freemap[0].sm_space;
1212 	sd->ms_freemap[1] += ms.ms_freemap[1].sm_space;
1213 	sd->ms_freemap[2] += ms.ms_freemap[2].sm_space;
1214 	sd->ms_freemap[3] += ms.ms_freemap[3].sm_space;
1215 	sd->ms_map += ms.ms_map.sm_space;
1216 	sd->avail += ms.ms_map.sm_size - ms.ms_smo.smo_alloc;
1217 	sd->nowavail += ms.ms_map.sm_size - ms.ms_smo_syncing.smo_alloc;
1218 
1219 	return (WALK_NEXT);
1220 }
1221 
1222 /*
1223  * ::spa_space [-b]
1224  *
1225  * Given a spa_t, print out it's on-disk space usage and in-core
1226  * estimates of future usage.  If -b is given, print space in bytes.
1227  * Otherwise print in megabytes.
1228  */
1229 /* ARGSUSED */
1230 static int
1231 spa_space(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1232 {
1233 	mdb_spa_t spa;
1234 	uintptr_t dp_root_dir;
1235 	mdb_dsl_dir_t dd;
1236 	mdb_dsl_dir_phys_t dsp;
1237 	uint64_t children;
1238 	uintptr_t childaddr;
1239 	space_data_t sd;
1240 	int shift = 20;
1241 	char *suffix = "M";
1242 	int bits = FALSE;
1243 
1244 	if (mdb_getopts(argc, argv, 'b', MDB_OPT_SETBITS, TRUE, &bits, NULL) !=
1245 	    argc)
1246 		return (DCMD_USAGE);
1247 	if (!(flags & DCMD_ADDRSPEC))
1248 		return (DCMD_USAGE);
1249 
1250 	if (bits) {
1251 		shift = 0;
1252 		suffix = "";
1253 	}
1254 
1255 	if (GETMEMB(addr, struct spa, spa_dsl_pool, spa.spa_dsl_pool) ||
1256 	    GETMEMB(addr, struct spa, spa_root_vdev, spa.spa_root_vdev) ||
1257 	    GETMEMB(spa.spa_root_vdev, struct vdev, vdev_children, children) ||
1258 	    GETMEMB(spa.spa_root_vdev, struct vdev, vdev_child, childaddr) ||
1259 	    GETMEMB(spa.spa_dsl_pool, struct dsl_pool,
1260 	    dp_root_dir, dp_root_dir) ||
1261 	    GETMEMB(dp_root_dir, struct dsl_dir, dd_phys, dd.dd_phys) ||
1262 	    GETMEMB(dp_root_dir, struct dsl_dir,
1263 	    dd_used_bytes, dd.dd_used_bytes) ||
1264 	    GETMEMB(dp_root_dir, struct dsl_dir,
1265 	    dd_space_towrite, dd.dd_space_towrite) ||
1266 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1267 	    dd_used_bytes, dsp.dd_used_bytes) ||
1268 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1269 	    dd_compressed_bytes, dsp.dd_compressed_bytes) ||
1270 	    GETMEMB(dd.dd_phys, struct dsl_dir_phys,
1271 	    dd_uncompressed_bytes, dsp.dd_uncompressed_bytes)) {
1272 		return (DCMD_ERR);
1273 	}
1274 
1275 	mdb_printf("dd_space_towrite = %llu%s %llu%s %llu%s %llu%s\n",
1276 	    dd.dd_space_towrite[0] >> shift, suffix,
1277 	    dd.dd_space_towrite[1] >> shift, suffix,
1278 	    dd.dd_space_towrite[2] >> shift, suffix,
1279 	    dd.dd_space_towrite[3] >> shift, suffix);
1280 	mdb_printf("dd_used_bytes = %llu%s\n",
1281 	    dd.dd_used_bytes >> shift, suffix);
1282 
1283 	mdb_printf("dd_phys.dd_used_bytes = %llu%s\n",
1284 	    dsp.dd_used_bytes >> shift, suffix);
1285 	mdb_printf("dd_phys.dd_compressed_bytes = %llu%s\n",
1286 	    dsp.dd_compressed_bytes >> shift, suffix);
1287 	mdb_printf("dd_phys.dd_uncompressed_bytes = %llu%s\n",
1288 	    dsp.dd_uncompressed_bytes >> shift, suffix);
1289 
1290 	bzero(&sd, sizeof (sd));
1291 	if (mdb_pwalk("metaslab", space_cb, &sd, addr) != 0) {
1292 		mdb_warn("can't walk metaslabs");
1293 		return (DCMD_ERR);
1294 	}
1295 
1296 	mdb_printf("ms_allocmap = %llu%s %llu%s %llu%s %llu%s\n",
1297 	    sd.ms_allocmap[0] >> shift, suffix,
1298 	    sd.ms_allocmap[1] >> shift, suffix,
1299 	    sd.ms_allocmap[2] >> shift, suffix,
1300 	    sd.ms_allocmap[3] >> shift, suffix);
1301 	mdb_printf("ms_freemap = %llu%s %llu%s %llu%s %llu%s\n",
1302 	    sd.ms_freemap[0] >> shift, suffix,
1303 	    sd.ms_freemap[1] >> shift, suffix,
1304 	    sd.ms_freemap[2] >> shift, suffix,
1305 	    sd.ms_freemap[3] >> shift, suffix);
1306 	mdb_printf("ms_map = %llu%s\n", sd.ms_map >> shift, suffix);
1307 	mdb_printf("last synced avail = %llu%s\n", sd.avail >> shift, suffix);
1308 	mdb_printf("current syncing avail = %llu%s\n",
1309 	    sd.nowavail >> shift, suffix);
1310 
1311 	return (DCMD_OK);
1312 }
1313 
1314 /*
1315  * ::spa_verify
1316  *
1317  * Given a spa_t, verify that that the pool is self-consistent.
1318  * Currently, it only checks to make sure that the vdev tree exists.
1319  */
1320 /* ARGSUSED */
1321 static int
1322 spa_verify(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1323 {
1324 	spa_t spa;
1325 
1326 	if (argc != 0 || !(flags & DCMD_ADDRSPEC))
1327 		return (DCMD_USAGE);
1328 
1329 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1330 		mdb_warn("failed to read spa_t at %p", addr);
1331 		return (DCMD_ERR);
1332 	}
1333 
1334 	if (spa.spa_root_vdev == NULL) {
1335 		mdb_printf("no vdev tree present\n");
1336 		return (DCMD_OK);
1337 	}
1338 
1339 	return (DCMD_OK);
1340 }
1341 
1342 /*
1343  * ::spa_vdevs
1344  *
1345  * 	-e	Include error stats
1346  *
1347  * Print out a summarized list of vdevs for the given spa_t.
1348  * This is accomplished by invoking "::vdev -re" on the root vdev.
1349  */
1350 /* ARGSUSED */
1351 static int
1352 spa_vdevs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1353 {
1354 	spa_t spa;
1355 	mdb_arg_t v;
1356 	int errors = FALSE;
1357 
1358 	if (mdb_getopts(argc, argv,
1359 	    'e', MDB_OPT_SETBITS, TRUE, &errors,
1360 	    NULL) != argc)
1361 		return (DCMD_USAGE);
1362 
1363 	if (!(flags & DCMD_ADDRSPEC))
1364 		return (DCMD_USAGE);
1365 
1366 	if (mdb_vread(&spa, sizeof (spa), addr) == -1) {
1367 		mdb_warn("failed to read spa_t at %p", addr);
1368 		return (DCMD_ERR);
1369 	}
1370 
1371 	/*
1372 	 * Unitialized spa_t structures can have a NULL root vdev.
1373 	 */
1374 	if (spa.spa_root_vdev == NULL) {
1375 		mdb_printf("no associated vdevs\n");
1376 		return (DCMD_OK);
1377 	}
1378 
1379 	v.a_type = MDB_TYPE_STRING;
1380 	v.a_un.a_str = errors ? "-re" : "-r";
1381 
1382 	return (mdb_call_dcmd("vdev", (uintptr_t)spa.spa_root_vdev,
1383 	    flags, 1, &v));
1384 }
1385 
1386 typedef struct txg_list_walk_data {
1387 	uintptr_t lw_head[TXG_SIZE];
1388 	int	lw_txgoff;
1389 	int	lw_maxoff;
1390 	size_t	lw_offset;
1391 	void	*lw_obj;
1392 } txg_list_walk_data_t;
1393 
1394 static int
1395 txg_list_walk_init_common(mdb_walk_state_t *wsp, int txg, int maxoff)
1396 {
1397 	txg_list_walk_data_t *lwd;
1398 	txg_list_t list;
1399 	int i;
1400 
1401 	lwd = mdb_alloc(sizeof (txg_list_walk_data_t), UM_SLEEP | UM_GC);
1402 	if (mdb_vread(&list, sizeof (txg_list_t), wsp->walk_addr) == -1) {
1403 		mdb_warn("failed to read txg_list_t at %#lx", wsp->walk_addr);
1404 		return (WALK_ERR);
1405 	}
1406 
1407 	for (i = 0; i < TXG_SIZE; i++)
1408 		lwd->lw_head[i] = (uintptr_t)list.tl_head[i];
1409 	lwd->lw_offset = list.tl_offset;
1410 	lwd->lw_obj = mdb_alloc(lwd->lw_offset + sizeof (txg_node_t),
1411 	    UM_SLEEP | UM_GC);
1412 	lwd->lw_txgoff = txg;
1413 	lwd->lw_maxoff = maxoff;
1414 
1415 	wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
1416 	wsp->walk_data = lwd;
1417 
1418 	return (WALK_NEXT);
1419 }
1420 
1421 static int
1422 txg_list_walk_init(mdb_walk_state_t *wsp)
1423 {
1424 	return (txg_list_walk_init_common(wsp, 0, TXG_SIZE-1));
1425 }
1426 
1427 static int
1428 txg_list0_walk_init(mdb_walk_state_t *wsp)
1429 {
1430 	return (txg_list_walk_init_common(wsp, 0, 0));
1431 }
1432 
1433 static int
1434 txg_list1_walk_init(mdb_walk_state_t *wsp)
1435 {
1436 	return (txg_list_walk_init_common(wsp, 1, 1));
1437 }
1438 
1439 static int
1440 txg_list2_walk_init(mdb_walk_state_t *wsp)
1441 {
1442 	return (txg_list_walk_init_common(wsp, 2, 2));
1443 }
1444 
1445 static int
1446 txg_list3_walk_init(mdb_walk_state_t *wsp)
1447 {
1448 	return (txg_list_walk_init_common(wsp, 3, 3));
1449 }
1450 
1451 static int
1452 txg_list_walk_step(mdb_walk_state_t *wsp)
1453 {
1454 	txg_list_walk_data_t *lwd = wsp->walk_data;
1455 	uintptr_t addr;
1456 	txg_node_t *node;
1457 	int status;
1458 
1459 	while (wsp->walk_addr == NULL && lwd->lw_txgoff < lwd->lw_maxoff) {
1460 		lwd->lw_txgoff++;
1461 		wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
1462 	}
1463 
1464 	if (wsp->walk_addr == NULL)
1465 		return (WALK_DONE);
1466 
1467 	addr = wsp->walk_addr - lwd->lw_offset;
1468 
1469 	if (mdb_vread(lwd->lw_obj,
1470 	    lwd->lw_offset + sizeof (txg_node_t), addr) == -1) {
1471 		mdb_warn("failed to read list element at %#lx", addr);
1472 		return (WALK_ERR);
1473 	}
1474 
1475 	status = wsp->walk_callback(addr, lwd->lw_obj, wsp->walk_cbdata);
1476 	node = (txg_node_t *)((uintptr_t)lwd->lw_obj + lwd->lw_offset);
1477 	wsp->walk_addr = (uintptr_t)node->tn_next[lwd->lw_txgoff];
1478 
1479 	return (status);
1480 }
1481 
1482 /*
1483  * ::walk spa
1484  *
1485  * Walk all named spa_t structures in the namespace.  This is nothing more than
1486  * a layered avl walk.
1487  */
1488 static int
1489 spa_walk_init(mdb_walk_state_t *wsp)
1490 {
1491 	GElf_Sym sym;
1492 
1493 	if (wsp->walk_addr != NULL) {
1494 		mdb_warn("spa walk only supports global walks\n");
1495 		return (WALK_ERR);
1496 	}
1497 
1498 	if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "spa_namespace_avl", &sym) == -1) {
1499 		mdb_warn("failed to find symbol 'spa_namespace_avl'");
1500 		return (WALK_ERR);
1501 	}
1502 
1503 	wsp->walk_addr = (uintptr_t)sym.st_value;
1504 
1505 	if (mdb_layered_walk("avl", wsp) == -1) {
1506 		mdb_warn("failed to walk 'avl'\n");
1507 		return (WALK_ERR);
1508 	}
1509 
1510 	return (WALK_NEXT);
1511 }
1512 
1513 static int
1514 spa_walk_step(mdb_walk_state_t *wsp)
1515 {
1516 	spa_t	spa;
1517 
1518 	if (mdb_vread(&spa, sizeof (spa), wsp->walk_addr) == -1) {
1519 		mdb_warn("failed to read spa_t at %p", wsp->walk_addr);
1520 		return (WALK_ERR);
1521 	}
1522 
1523 	return (wsp->walk_callback(wsp->walk_addr, &spa, wsp->walk_cbdata));
1524 }
1525 
1526 /*
1527  * MDB module linkage information:
1528  *
1529  * We declare a list of structures describing our dcmds, and a function
1530  * named _mdb_init to return a pointer to our module information.
1531  */
1532 
1533 static const mdb_dcmd_t dcmds[] = {
1534 	{ "blkptr", ":", "print blkptr_t", blkptr },
1535 	{ "dbuf", ":", "print dmu_buf_impl_t", dbuf },
1536 	{ "dbuf_stats", ":", "dbuf stats", dbuf_stats },
1537 	{ "dbufs",
1538 	"\t[-O objset_t*] [-n objset_name | \"mos\"] [-o object | \"mdn\"] \n"
1539 	"\t[-l level] [-b blkid | \"bonus\"]",
1540 	"find dmu_buf_impl_t's that meet criterion", dbufs },
1541 	{ "abuf_find", "dva_word[0] dva_word[1]",
1542 	"find arc_buf_hdr_t of a specified DVA",
1543 	abuf_find },
1544 	{ "spa", "?[-cv]", "spa_t summary", spa_print },
1545 	{ "spa_config", ":", "print spa_t configuration", spa_print_config },
1546 	{ "spa_verify", ":", "verify spa_t consistency", spa_verify },
1547 	{ "spa_space", ":[-b]", "print spa_t on-disk space usage", spa_space },
1548 	{ "spa_vdevs", ":", "given a spa_t, print vdev summary", spa_vdevs },
1549 	{ "vdev", ":[-qre]", "vdev_t summary", vdev_print },
1550 	{ "zio_pipeline", ":", "decode a zio pipeline", zio_pipeline },
1551 	{ NULL }
1552 };
1553 
1554 static const mdb_walker_t walkers[] = {
1555 	/*
1556 	 * In userland, there is no generic provider of list_t walkers, so we
1557 	 * need to add it.
1558 	 */
1559 #ifndef _KERNEL
1560 	{ LIST_WALK_NAME, LIST_WALK_DESC,
1561 		list_walk_init, list_walk_step, list_walk_fini },
1562 #endif
1563 	{ "zms_freelist", "walk ZFS metaslab freelist",
1564 		freelist_walk_init, freelist_walk_step, NULL },
1565 	{ "txg_list", "given any txg_list_t *, walk all entries in all txgs",
1566 		txg_list_walk_init, txg_list_walk_step, NULL },
1567 	{ "txg_list0", "given any txg_list_t *, walk all entries in txg 0",
1568 		txg_list0_walk_init, txg_list_walk_step, NULL },
1569 	{ "txg_list1", "given any txg_list_t *, walk all entries in txg 1",
1570 		txg_list1_walk_init, txg_list_walk_step, NULL },
1571 	{ "txg_list2", "given any txg_list_t *, walk all entries in txg 2",
1572 		txg_list2_walk_init, txg_list_walk_step, NULL },
1573 	{ "txg_list3", "given any txg_list_t *, walk all entries in txg 3",
1574 		txg_list3_walk_init, txg_list_walk_step, NULL },
1575 	{ "spa", "walk all spa_t entries in the namespace",
1576 		spa_walk_init, spa_walk_step, NULL },
1577 	{ "metaslab", "given a spa_t *, walk all metaslab_t structures",
1578 		metaslab_walk_init, metaslab_walk_step, NULL },
1579 	{ NULL }
1580 };
1581 
1582 static const mdb_modinfo_t modinfo = {
1583 	MDB_API_VERSION, dcmds, walkers
1584 };
1585 
1586 const mdb_modinfo_t *
1587 _mdb_init(void)
1588 {
1589 	return (&modinfo);
1590 }
1591