?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge commit 'd6e1c446d7897003fd9fd36ef5aa7da350b7f6af' into merges
[unleashed.git] / usr / src / cmd / mdb / common / modules / zfs / zfs.c
blob9c6ebbed4c6489617b39f1b8993e434f2536c740
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
25 * Copyright (c) 2017, Joyent, Inc. All rights reserved.
26 */
27
28 /* Portions Copyright 2010 Robert Milkowski */
29
30 #include <mdb/mdb_ctf.h>
31 #include <sys/zfs_context.h>
32 #include <sys/mdb_modapi.h>
33 #include <sys/dbuf.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/metaslab_impl.h>
38 #include <sys/space_map.h>
39 #include <sys/list.h>
40 #include <sys/vdev_impl.h>
41 #include <sys/zap_leaf.h>
42 #include <sys/zap_impl.h>
43 #include <ctype.h>
44 #include <sys/zfs_acl.h>
45 #include <sys/sa_impl.h>
46 #include <sys/multilist.h>
47
48 #ifdef _KERNEL
49 #define ZFS_OBJ_NAME "zfs"
50 extern int64_t mdb_gethrtime(void);
51 #else
52 #define ZFS_OBJ_NAME "libzpool.so.1"
53 #endif
54
55 #define ZFS_STRUCT "struct " ZFS_OBJ_NAME "`"
56
57 #ifndef _KERNEL
58 int aok;
59 #endif
60
61 enum spa_flags {
62 SPA_FLAG_CONFIG = 1 << 0,
63 SPA_FLAG_VDEVS = 1 << 1,
64 SPA_FLAG_ERRORS = 1 << 2,
65 SPA_FLAG_METASLAB_GROUPS = 1 << 3,
66 SPA_FLAG_METASLABS = 1 << 4,
67 SPA_FLAG_HISTOGRAMS = 1 << 5
68 };
69
70 /*
71 * If any of these flags are set, call spa_vdevs in spa_print
72 */
73 #define SPA_FLAG_ALL_VDEV \
74 (SPA_FLAG_VDEVS | SPA_FLAG_ERRORS | SPA_FLAG_METASLAB_GROUPS | \
75 SPA_FLAG_METASLABS)
76
77 static int
78 getmember(uintptr_t addr, const char *type, mdb_ctf_id_t *idp,
79 const char *member, int len, void *buf)
80 {
81 mdb_ctf_id_t id;
82 ulong_t off;
83 char name[64];
84
85 if (idp == NULL) {
86 if (mdb_ctf_lookup_by_name(type, &id) == -1) {
87 mdb_warn("couldn't find type %s", type);
88 return (DCMD_ERR);
89 }
90 idp = &id;
91 } else {
92 type = name;
93 mdb_ctf_type_name(*idp, name, sizeof (name));
94 }
95
96 if (mdb_ctf_offsetof(*idp, member, &off) == -1) {
97 mdb_warn("couldn't find member %s of type %s\n", member, type);
98 return (DCMD_ERR);
99 }
100 if (off % 8 != 0) {
101 mdb_warn("member %s of type %s is unsupported bitfield",
102 member, type);
103 return (DCMD_ERR);
104 }
105 off /= 8;
106
107 if (mdb_vread(buf, len, addr + off) == -1) {
108 mdb_warn("failed to read %s from %s at %p",
109 member, type, addr + off);
110 return (DCMD_ERR);
111 }
112 /* mdb_warn("read %s from %s at %p+%llx\n", member, type, addr, off); */
113
114 return (0);
115 }
116
117 #define GETMEMB(addr, structname, member, dest) \
118 getmember(addr, ZFS_STRUCT structname, NULL, #member, \
119 sizeof (dest), &(dest))
120
121 #define GETMEMBID(addr, ctfid, member, dest) \
122 getmember(addr, NULL, ctfid, #member, sizeof (dest), &(dest))
123
124 static boolean_t
125 strisprint(const char *cp)
126 {
127 for (; *cp; cp++) {
128 if (!isprint(*cp))
129 return (B_FALSE);
130 }
131 return (B_TRUE);
132 }
133
134 #define NICENUM_BUFLEN 6
135
136 static int
137 snprintfrac(char *buf, int len,
138 uint64_t numerator, uint64_t denom, int frac_digits)
139 {
140 int mul = 1;
141 int whole, frac, i;
142
143 for (i = frac_digits; i; i--)
144 mul *= 10;
145 whole = numerator / denom;
146 frac = mul * numerator / denom - mul * whole;
147 return (mdb_snprintf(buf, len, "%u.%0*u", whole, frac_digits, frac));
148 }
149
150 static void
151 mdb_nicenum(uint64_t num, char *buf)
152 {
153 uint64_t n = num;
154 int index = 0;
155 char *u;
156
157 while (n >= 1024) {
158 n = (n + (1024 / 2)) / 1024; /* Round up or down */
159 index++;
160 }
161
162 u = &" \0K\0M\0G\0T\0P\0E\0"[index*2];
163
164 if (index == 0) {
165 (void) mdb_snprintf(buf, NICENUM_BUFLEN, "%llu",
166 (u_longlong_t)n);
167 } else if (n < 10 && (num & (num - 1)) != 0) {
168 (void) snprintfrac(buf, NICENUM_BUFLEN,
169 num, 1ULL << 10 * index, 2);
170 strcat(buf, u);
171 } else if (n < 100 && (num & (num - 1)) != 0) {
172 (void) snprintfrac(buf, NICENUM_BUFLEN,
173 num, 1ULL << 10 * index, 1);
174 strcat(buf, u);
175 } else {
176 (void) mdb_snprintf(buf, NICENUM_BUFLEN, "%llu%s",
177 (u_longlong_t)n, u);
178 }
179 }
180
181 static int verbose;
182
183 static int
184 freelist_walk_init(mdb_walk_state_t *wsp)
185 {
186 if (wsp->walk_addr == (uintptr_t)NULL) {
187 mdb_warn("must supply starting address\n");
188 return (WALK_ERR);
189 }
190
191 wsp->walk_data = 0; /* Index into the freelist */
192 return (WALK_NEXT);
193 }
194
195 static int
196 freelist_walk_step(mdb_walk_state_t *wsp)
197 {
198 uint64_t entry;
199 uintptr_t number = (uintptr_t)wsp->walk_data;
200 char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
201 "INVALID", "INVALID", "INVALID", "INVALID" };
202 int mapshift = SPA_MINBLOCKSHIFT;
203
204 if (mdb_vread(&entry, sizeof (entry), wsp->walk_addr) == -1) {
205 mdb_warn("failed to read freelist entry %p", wsp->walk_addr);
206 return (WALK_DONE);
207 }
208 wsp->walk_addr += sizeof (entry);
209 wsp->walk_data = (void *)(number + 1);
210
211 if (SM_DEBUG_DECODE(entry)) {
212 mdb_printf("DEBUG: %3u %10s: txg=%llu pass=%llu\n",
213 number,
214 ddata[SM_DEBUG_ACTION_DECODE(entry)],
215 SM_DEBUG_TXG_DECODE(entry),
216 SM_DEBUG_SYNCPASS_DECODE(entry));
217 } else {
218 mdb_printf("Entry: %3u offsets=%08llx-%08llx type=%c "
219 "size=%06llx", number,
220 SM_OFFSET_DECODE(entry) << mapshift,
221 (SM_OFFSET_DECODE(entry) + SM_RUN_DECODE(entry)) <<
222 mapshift,
223 SM_TYPE_DECODE(entry) == SM_ALLOC ? 'A' : 'F',
224 SM_RUN_DECODE(entry) << mapshift);
225 if (verbose)
226 mdb_printf(" (raw=%012llx)\n", entry);
227 mdb_printf("\n");
228 }
229 return (WALK_NEXT);
230 }
231
232 static int
233 mdb_dsl_dir_name(uintptr_t addr, char *buf)
234 {
235 static int gotid;
236 static mdb_ctf_id_t dd_id;
237 uintptr_t dd_parent;
238 char dd_myname[ZFS_MAX_DATASET_NAME_LEN];
239
240 if (!gotid) {
241 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "dsl_dir",
242 &dd_id) == -1) {
243 mdb_warn("couldn't find struct dsl_dir");
244 return (DCMD_ERR);
245 }
246 gotid = TRUE;
247 }
248 if (GETMEMBID(addr, &dd_id, dd_parent, dd_parent) ||
249 GETMEMBID(addr, &dd_id, dd_myname, dd_myname)) {
250 return (DCMD_ERR);
251 }
252
253 if (dd_parent) {
254 if (mdb_dsl_dir_name(dd_parent, buf))
255 return (DCMD_ERR);
256 strcat(buf, "/");
257 }
258
259 if (dd_myname[0])
260 strcat(buf, dd_myname);
261 else
262 strcat(buf, "???");
263
264 return (0);
265 }
266
267 static int
268 objset_name(uintptr_t addr, char *buf)
269 {
270 static int gotid;
271 static mdb_ctf_id_t os_id, ds_id;
272 uintptr_t os_dsl_dataset;
273 char ds_snapname[ZFS_MAX_DATASET_NAME_LEN];
274 uintptr_t ds_dir;
275
276 buf[0] = '\0';
277
278 if (!gotid) {
279 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "objset",
280 &os_id) == -1) {
281 mdb_warn("couldn't find struct objset");
282 return (DCMD_ERR);
283 }
284 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "dsl_dataset",
285 &ds_id) == -1) {
286 mdb_warn("couldn't find struct dsl_dataset");
287 return (DCMD_ERR);
288 }
289
290 gotid = TRUE;
291 }
292
293 if (GETMEMBID(addr, &os_id, os_dsl_dataset, os_dsl_dataset))
294 return (DCMD_ERR);
295
296 if (os_dsl_dataset == 0) {
297 strcat(buf, "mos");
298 return (0);
299 }
300
301 if (GETMEMBID(os_dsl_dataset, &ds_id, ds_snapname, ds_snapname) ||
302 GETMEMBID(os_dsl_dataset, &ds_id, ds_dir, ds_dir)) {
303 return (DCMD_ERR);
304 }
305
306 if (ds_dir && mdb_dsl_dir_name(ds_dir, buf))
307 return (DCMD_ERR);
308
309 if (ds_snapname[0]) {
310 strcat(buf, "@");
311 strcat(buf, ds_snapname);
312 }
313 return (0);
314 }
315
316 static int
317 enum_lookup(char *type, int val, const char *prefix, size_t size, char *out)
318 {
319 const char *cp;
320 size_t len = strlen(prefix);
321 mdb_ctf_id_t enum_type;
322
323 if (mdb_ctf_lookup_by_name(type, &enum_type) != 0) {
324 mdb_warn("Could not find enum for %s", type);
325 return (-1);
326 }
327
328 if ((cp = mdb_ctf_enum_name(enum_type, val)) != NULL) {
329 if (strncmp(cp, prefix, len) == 0)
330 cp += len;
331 (void) strncpy(out, cp, size);
332 } else {
333 mdb_snprintf(out, size, "? (%d)", val);
334 }
335 return (0);
336 }
337
338 /* ARGSUSED */
339 static int
340 zfs_params(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
341 {
342 /*
343 * This table can be approximately generated by running:
344 * egrep "^[a-z0-9_]+ [a-z0-9_]+( =.*)?;" *.c | cut -d ' ' -f 2
345 */
346 static const char *params[] = {
347 "arc_reduce_dnlc_percent",
348 "arc_lotsfree_percent",
349 "zfs_dirty_data_max",
350 "zfs_dirty_data_sync",
351 "zfs_delay_max_ns",
352 "zfs_delay_min_dirty_percent",
353 "zfs_delay_scale",
354 "zfs_vdev_max_active",
355 "zfs_vdev_sync_read_min_active",
356 "zfs_vdev_sync_read_max_active",
357 "zfs_vdev_sync_write_min_active",
358 "zfs_vdev_sync_write_max_active",
359 "zfs_vdev_async_read_min_active",
360 "zfs_vdev_async_read_max_active",
361 "zfs_vdev_async_write_min_active",
362 "zfs_vdev_async_write_max_active",
363 "zfs_vdev_scrub_min_active",
364 "zfs_vdev_scrub_max_active",
365 "zfs_vdev_async_write_active_min_dirty_percent",
366 "zfs_vdev_async_write_active_max_dirty_percent",
367 "spa_asize_inflation",
368 "zfs_arc_max",
369 "zfs_arc_min",
370 "arc_shrink_shift",
371 "zfs_mdcomp_disable",
372 "zfs_prefetch_disable",
373 "zfetch_max_streams",
374 "zfetch_min_sec_reap",
375 "zfetch_block_cap",
376 "zfetch_array_rd_sz",
377 "zfs_default_bs",
378 "zfs_default_ibs",
379 "metaslab_aliquot",
380 "reference_tracking_enable",
381 "reference_history",
382 "spa_max_replication_override",
383 "spa_mode_global",
384 "zfs_flags",
385 "zfs_txg_timeout",
386 "zfs_vdev_cache_max",
387 "zfs_vdev_cache_size",
388 "zfs_vdev_cache_bshift",
389 "vdev_mirror_shift",
390 "zfs_scrub_limit",
391 "zfs_no_scrub_io",
392 "zfs_no_scrub_prefetch",
393 "zfs_vdev_aggregation_limit",
394 "fzap_default_block_shift",
395 "zfs_immediate_write_sz",
396 "zfs_read_chunk_size",
397 "zfs_nocacheflush",
398 "zil_replay_disable",
399 "metaslab_gang_bang",
400 "metaslab_df_alloc_threshold",
401 "metaslab_df_free_pct",
402 "zio_injection_enabled",
403 "zvol_immediate_write_sz",
404 };
405
406 for (int i = 0; i < sizeof (params) / sizeof (params[0]); i++) {
407 int sz;
408 uint64_t val64;
409 uint32_t *val32p = (uint32_t *)&val64;
410
411 sz = mdb_readvar(&val64, params[i]);
412 if (sz == 4) {
413 mdb_printf("%s = 0x%x\n", params[i], *val32p);
414 } else if (sz == 8) {
415 mdb_printf("%s = 0x%llx\n", params[i], val64);
416 } else {
417 mdb_warn("variable %s not found", params[i]);
418 }
419 }
420
421 return (DCMD_OK);
422 }
423
424 /* ARGSUSED */
425 static int
426 dva(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
427 {
428 dva_t dva;
429 if (mdb_vread(&dva, sizeof (dva_t), addr) == -1) {
430 mdb_warn("failed to read dva_t");
431 return (DCMD_ERR);
432 }
433 mdb_printf("<%llu:%llx:%llx>\n",
434 (u_longlong_t)DVA_GET_VDEV(&dva),
435 (u_longlong_t)DVA_GET_OFFSET(&dva),
436 (u_longlong_t)DVA_GET_ASIZE(&dva));
437
438 return (DCMD_OK);
439 }
440
441 /* ARGSUSED */
442 static int
443 blkptr(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
444 {
445 char type[80], checksum[80], compress[80];
446 blkptr_t blk, *bp = &blk;
447 char buf[BP_SPRINTF_LEN];
448
449 if (mdb_vread(&blk, sizeof (blkptr_t), addr) == -1) {
450 mdb_warn("failed to read blkptr_t");
451 return (DCMD_ERR);
452 }
453
454 if (enum_lookup("enum dmu_object_type", BP_GET_TYPE(bp), "DMU_OT_",
455 sizeof (type), type) == -1 ||
456 enum_lookup("enum zio_checksum", BP_GET_CHECKSUM(bp),
457 "ZIO_CHECKSUM_", sizeof (checksum), checksum) == -1 ||
458 enum_lookup("enum zio_compress", BP_GET_COMPRESS(bp),
459 "ZIO_COMPRESS_", sizeof (compress), compress) == -1) {
460 mdb_warn("Could not find blkptr enumerated types");
461 return (DCMD_ERR);
462 }
463
464 SNPRINTF_BLKPTR(mdb_snprintf, '\n', buf, sizeof (buf), bp, type,
465 checksum, compress);
466
467 mdb_printf("%s\n", buf);
468
469 return (DCMD_OK);
470 }
471
472 typedef struct mdb_dmu_buf_impl {
473 struct {
474 uint64_t db_object;
475 uintptr_t db_data;
476 } db;
477 uintptr_t db_objset;
478 uint64_t db_level;
479 uint64_t db_blkid;
480 struct {
481 uint64_t rc_count;
482 } db_holds;
483 } mdb_dmu_buf_impl_t;
484
485 /* ARGSUSED */
486 static int
487 dbuf(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
488 {
489 mdb_dmu_buf_impl_t db;
490 char objectname[32];
491 char blkidname[32];
492 char path[ZFS_MAX_DATASET_NAME_LEN];
493 int ptr_width = (int)(sizeof (void *)) * 2;
494
495 if (DCMD_HDRSPEC(flags))
496 mdb_printf("%*s %8s %3s %9s %5s %s\n",
497 ptr_width, "addr", "object", "lvl", "blkid", "holds", "os");
498
499 if (mdb_ctf_vread(&db, ZFS_STRUCT "dmu_buf_impl", "mdb_dmu_buf_impl_t",
500 addr, 0) == -1)
501 return (DCMD_ERR);
502
503 if (db.db.db_object == DMU_META_DNODE_OBJECT)
504 (void) strcpy(objectname, "mdn");
505 else
506 (void) mdb_snprintf(objectname, sizeof (objectname), "%llx",
507 (u_longlong_t)db.db.db_object);
508
509 if (db.db_blkid == DMU_BONUS_BLKID)
510 (void) strcpy(blkidname, "bonus");
511 else
512 (void) mdb_snprintf(blkidname, sizeof (blkidname), "%llx",
513 (u_longlong_t)db.db_blkid);
514
515 if (objset_name(db.db_objset, path)) {
516 return (DCMD_ERR);
517 }
518
519 mdb_printf("%*p %8s %3u %9s %5llu %s\n", ptr_width, addr,
520 objectname, (int)db.db_level, blkidname,
521 db.db_holds.rc_count, path);
522
523 return (DCMD_OK);
524 }
525
526 /* ARGSUSED */
527 static int
528 dbuf_stats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
529 {
530 #define HISTOSZ 32
531 uintptr_t dbp;
532 dmu_buf_impl_t db;
533 dbuf_hash_table_t ht;
534 uint64_t bucket, ndbufs;
535 uint64_t histo[HISTOSZ];
536 uint64_t histo2[HISTOSZ];
537 int i, maxidx;
538
539 if (mdb_readvar(&ht, "dbuf_hash_table") == -1) {
540 mdb_warn("failed to read 'dbuf_hash_table'");
541 return (DCMD_ERR);
542 }
543
544 for (i = 0; i < HISTOSZ; i++) {
545 histo[i] = 0;
546 histo2[i] = 0;
547 }
548
549 ndbufs = 0;
550 for (bucket = 0; bucket < ht.hash_table_mask+1; bucket++) {
551 int len;
552
553 if (mdb_vread(&dbp, sizeof (void *),
554 (uintptr_t)(ht.hash_table+bucket)) == -1) {
555 mdb_warn("failed to read hash bucket %u at %p",
556 bucket, ht.hash_table+bucket);
557 return (DCMD_ERR);
558 }
559
560 len = 0;
561 while (dbp != 0) {
562 if (mdb_vread(&db, sizeof (dmu_buf_impl_t),
563 dbp) == -1) {
564 mdb_warn("failed to read dbuf at %p", dbp);
565 return (DCMD_ERR);
566 }
567 dbp = (uintptr_t)db.db_hash_next;
568 for (i = MIN(len, HISTOSZ - 1); i >= 0; i--)
569 histo2[i]++;
570 len++;
571 ndbufs++;
572 }
573
574 if (len >= HISTOSZ)
575 len = HISTOSZ-1;
576 histo[len]++;
577 }
578
579 mdb_printf("hash table has %llu buckets, %llu dbufs "
580 "(avg %llu buckets/dbuf)\n",
581 ht.hash_table_mask+1, ndbufs,
582 (ht.hash_table_mask+1)/ndbufs);
583
584 mdb_printf("\n");
585 maxidx = 0;
586 for (i = 0; i < HISTOSZ; i++)
587 if (histo[i] > 0)
588 maxidx = i;
589 mdb_printf("hash chain length number of buckets\n");
590 for (i = 0; i <= maxidx; i++)
591 mdb_printf("%u %llu\n", i, histo[i]);
592
593 mdb_printf("\n");
594 maxidx = 0;
595 for (i = 0; i < HISTOSZ; i++)
596 if (histo2[i] > 0)
597 maxidx = i;
598 mdb_printf("hash chain depth number of dbufs\n");
599 for (i = 0; i <= maxidx; i++)
600 mdb_printf("%u or more %llu %llu%%\n",
601 i, histo2[i], histo2[i]*100/ndbufs);
602
603
604 return (DCMD_OK);
605 }
606
607 #define CHAIN_END 0xffff
608 /*
609 * ::zap_leaf [-v]
610 *
611 * Print a zap_leaf_phys_t, assumed to be 16k
612 */
613 /* ARGSUSED */
614 static int
615 zap_leaf(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
616 {
617 char buf[16*1024];
618 int verbose = B_FALSE;
619 int four = B_FALSE;
620 dmu_buf_t l_dbuf;
621 zap_leaf_t l;
622 zap_leaf_phys_t *zlp = (void *)buf;
623 int i;
624
625 if (mdb_getopts(argc, argv,
626 'v', MDB_OPT_SETBITS, TRUE, &verbose,
627 '4', MDB_OPT_SETBITS, TRUE, &four,
628 NULL) != argc)
629 return (DCMD_USAGE);
630
631 l_dbuf.db_data = zlp;
632 l.l_dbuf = &l_dbuf;
633 l.l_bs = 14; /* assume 16k blocks */
634 if (four)
635 l.l_bs = 12;
636
637 if (!(flags & DCMD_ADDRSPEC)) {
638 return (DCMD_USAGE);
639 }
640
641 if (mdb_vread(buf, sizeof (buf), addr) == -1) {
642 mdb_warn("failed to read zap_leaf_phys_t at %p", addr);
643 return (DCMD_ERR);
644 }
645
646 if (zlp->l_hdr.lh_block_type != ZBT_LEAF ||
647 zlp->l_hdr.lh_magic != ZAP_LEAF_MAGIC) {
648 mdb_warn("This does not appear to be a zap_leaf_phys_t");
649 return (DCMD_ERR);
650 }
651
652 mdb_printf("zap_leaf_phys_t at %p:\n", addr);
653 mdb_printf(" lh_prefix_len = %u\n", zlp->l_hdr.lh_prefix_len);
654 mdb_printf(" lh_prefix = %llx\n", zlp->l_hdr.lh_prefix);
655 mdb_printf(" lh_nentries = %u\n", zlp->l_hdr.lh_nentries);
656 mdb_printf(" lh_nfree = %u\n", zlp->l_hdr.lh_nfree,
657 zlp->l_hdr.lh_nfree * 100 / (ZAP_LEAF_NUMCHUNKS(&l)));
658 mdb_printf(" lh_freelist = %u\n", zlp->l_hdr.lh_freelist);
659 mdb_printf(" lh_flags = %x (%s)\n", zlp->l_hdr.lh_flags,
660 zlp->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED ?
661 "ENTRIES_CDSORTED" : "");
662
663 if (verbose) {
664 mdb_printf(" hash table:\n");
665 for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++) {
666 if (zlp->l_hash[i] != CHAIN_END)
667 mdb_printf(" %u: %u\n", i, zlp->l_hash[i]);
668 }
669 }
670
671 mdb_printf(" chunks:\n");
672 for (i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) {
673 /* LINTED: alignment */
674 zap_leaf_chunk_t *zlc = &ZAP_LEAF_CHUNK(&l, i);
675 switch (zlc->l_entry.le_type) {
676 case ZAP_CHUNK_FREE:
677 if (verbose) {
678 mdb_printf(" %u: free; lf_next = %u\n",
679 i, zlc->l_free.lf_next);
680 }
681 break;
682 case ZAP_CHUNK_ENTRY:
683 mdb_printf(" %u: entry\n", i);
684 if (verbose) {
685 mdb_printf(" le_next = %u\n",
686 zlc->l_entry.le_next);
687 }
688 mdb_printf(" le_name_chunk = %u\n",
689 zlc->l_entry.le_name_chunk);
690 mdb_printf(" le_name_numints = %u\n",
691 zlc->l_entry.le_name_numints);
692 mdb_printf(" le_value_chunk = %u\n",
693 zlc->l_entry.le_value_chunk);
694 mdb_printf(" le_value_intlen = %u\n",
695 zlc->l_entry.le_value_intlen);
696 mdb_printf(" le_value_numints = %u\n",
697 zlc->l_entry.le_value_numints);
698 mdb_printf(" le_cd = %u\n",
699 zlc->l_entry.le_cd);
700 mdb_printf(" le_hash = %llx\n",
701 zlc->l_entry.le_hash);
702 break;
703 case ZAP_CHUNK_ARRAY:
704 mdb_printf(" %u: array", i);
705 if (strisprint((char *)zlc->l_array.la_array))
706 mdb_printf(" \"%s\"", zlc->l_array.la_array);
707 mdb_printf("\n");
708 if (verbose) {
709 int j;
710 mdb_printf(" ");
711 for (j = 0; j < ZAP_LEAF_ARRAY_BYTES; j++) {
712 mdb_printf("%02x ",
713 zlc->l_array.la_array[j]);
714 }
715 mdb_printf("\n");
716 }
717 if (zlc->l_array.la_next != CHAIN_END) {
718 mdb_printf(" lf_next = %u\n",
719 zlc->l_array.la_next);
720 }
721 break;
722 default:
723 mdb_printf(" %u: undefined type %u\n",
724 zlc->l_entry.le_type);
725 }
726 }
727
728 return (DCMD_OK);
729 }
730
731 typedef struct dbufs_data {
732 mdb_ctf_id_t id;
733 uint64_t objset;
734 uint64_t object;
735 uint64_t level;
736 uint64_t blkid;
737 char *osname;
738 } dbufs_data_t;
739
740 #define DBUFS_UNSET (0xbaddcafedeadbeefULL)
741
742 /* ARGSUSED */
743 static int
744 dbufs_cb(uintptr_t addr, const void *unknown, void *arg)
745 {
746 dbufs_data_t *data = arg;
747 uintptr_t objset;
748 dmu_buf_t db;
749 uint8_t level;
750 uint64_t blkid;
751 char osname[ZFS_MAX_DATASET_NAME_LEN];
752
753 if (GETMEMBID(addr, &data->id, db_objset, objset) ||
754 GETMEMBID(addr, &data->id, db, db) ||
755 GETMEMBID(addr, &data->id, db_level, level) ||
756 GETMEMBID(addr, &data->id, db_blkid, blkid)) {
757 return (WALK_ERR);
758 }
759
760 if ((data->objset == DBUFS_UNSET || data->objset == objset) &&
761 (data->osname == NULL || (objset_name(objset, osname) == 0 &&
762 strcmp(data->osname, osname) == 0)) &&
763 (data->object == DBUFS_UNSET || data->object == db.db_object) &&
764 (data->level == DBUFS_UNSET || data->level == level) &&
765 (data->blkid == DBUFS_UNSET || data->blkid == blkid)) {
766 mdb_printf("%#lr\n", addr);
767 }
768 return (WALK_NEXT);
769 }
770
771 /* ARGSUSED */
772 static int
773 dbufs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
774 {
775 dbufs_data_t data;
776 char *object = NULL;
777 char *blkid = NULL;
778
779 data.objset = data.object = data.level = data.blkid = DBUFS_UNSET;
780 data.osname = NULL;
781
782 if (mdb_getopts(argc, argv,
783 'O', MDB_OPT_UINT64, &data.objset,
784 'n', MDB_OPT_STR, &data.osname,
785 'o', MDB_OPT_STR, &object,
786 'l', MDB_OPT_UINT64, &data.level,
787 'b', MDB_OPT_STR, &blkid) != argc) {
788 return (DCMD_USAGE);
789 }
790
791 if (object) {
792 if (strcmp(object, "mdn") == 0) {
793 data.object = DMU_META_DNODE_OBJECT;
794 } else {
795 data.object = mdb_strtoull(object);
796 }
797 }
798
799 if (blkid) {
800 if (strcmp(blkid, "bonus") == 0) {
801 data.blkid = DMU_BONUS_BLKID;
802 } else {
803 data.blkid = mdb_strtoull(blkid);
804 }
805 }
806
807 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "dmu_buf_impl", &data.id) == -1) {
808 mdb_warn("couldn't find struct dmu_buf_impl_t");
809 return (DCMD_ERR);
810 }
811
812 if (mdb_walk("dmu_buf_impl_t", dbufs_cb, &data) != 0) {
813 mdb_warn("can't walk dbufs");
814 return (DCMD_ERR);
815 }
816
817 return (DCMD_OK);
818 }
819
820 typedef struct abuf_find_data {
821 dva_t dva;
822 mdb_ctf_id_t id;
823 } abuf_find_data_t;
824
825 /* ARGSUSED */
826 static int
827 abuf_find_cb(uintptr_t addr, const void *unknown, void *arg)
828 {
829 abuf_find_data_t *data = arg;
830 dva_t dva;
831
832 if (GETMEMBID(addr, &data->id, b_dva, dva)) {
833 return (WALK_ERR);
834 }
835
836 if (dva.dva_word[0] == data->dva.dva_word[0] &&
837 dva.dva_word[1] == data->dva.dva_word[1]) {
838 mdb_printf("%#lr\n", addr);
839 }
840 return (WALK_NEXT);
841 }
842
843 /* ARGSUSED */
844 static int
845 abuf_find(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
846 {
847 abuf_find_data_t data;
848 GElf_Sym sym;
849 int i;
850 const char *syms[] = {
851 "ARC_mru",
852 "ARC_mru_ghost",
853 "ARC_mfu",
854 "ARC_mfu_ghost",
855 };
856
857 if (argc != 2)
858 return (DCMD_USAGE);
859
860 for (i = 0; i < 2; i ++) {
861 switch (argv[i].a_type) {
862 case MDB_TYPE_STRING:
863 data.dva.dva_word[i] = mdb_strtoull(argv[i].a_un.a_str);
864 break;
865 case MDB_TYPE_IMMEDIATE:
866 data.dva.dva_word[i] = argv[i].a_un.a_val;
867 break;
868 default:
869 return (DCMD_USAGE);
870 }
871 }
872
873 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "arc_buf_hdr", &data.id) == -1) {
874 mdb_warn("couldn't find struct arc_buf_hdr");
875 return (DCMD_ERR);
876 }
877
878 for (i = 0; i < sizeof (syms) / sizeof (syms[0]); i++) {
879 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, syms[i], &sym)) {
880 mdb_warn("can't find symbol %s", syms[i]);
881 return (DCMD_ERR);
882 }
883
884 if (mdb_pwalk("list", abuf_find_cb, &data, sym.st_value) != 0) {
885 mdb_warn("can't walk %s", syms[i]);
886 return (DCMD_ERR);
887 }
888 }
889
890 return (DCMD_OK);
891 }
892
893
894 typedef struct dbgmsg_arg {
895 boolean_t da_verbose;
896 boolean_t da_address;
897 } dbgmsg_arg_t;
898
899 /* ARGSUSED */
900 static int
901 dbgmsg_cb(uintptr_t addr, const void *unknown, void *arg)
902 {
903 static mdb_ctf_id_t id;
904 static boolean_t gotid;
905 static ulong_t off;
906
907 dbgmsg_arg_t *da = arg;
908 time_t timestamp;
909 char buf[1024];
910
911 if (!gotid) {
912 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "zfs_dbgmsg", &id) ==
913 -1) {
914 mdb_warn("couldn't find struct zfs_dbgmsg");
915 return (WALK_ERR);
916 }
917 gotid = TRUE;
918 if (mdb_ctf_offsetof(id, "zdm_msg", &off) == -1) {
919 mdb_warn("couldn't find zdm_msg");
920 return (WALK_ERR);
921 }
922 off /= 8;
923 }
924
925
926 if (GETMEMBID(addr, &id, zdm_timestamp, timestamp)) {
927 return (WALK_ERR);
928 }
929
930 if (mdb_readstr(buf, sizeof (buf), addr + off) == -1) {
931 mdb_warn("failed to read zdm_msg at %p\n", addr + off);
932 return (DCMD_ERR);
933 }
934
935 if (da->da_address)
936 mdb_printf("%p ", addr);
937 if (da->da_verbose)
938 mdb_printf("%Y ", timestamp);
939
940 mdb_printf("%s\n", buf);
941
942 if (da->da_verbose)
943 (void) mdb_call_dcmd("whatis", addr, DCMD_ADDRSPEC, 0, NULL);
944
945 return (WALK_NEXT);
946 }
947
948 /* ARGSUSED */
949 static int
950 dbgmsg(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
951 {
952 GElf_Sym sym;
953 dbgmsg_arg_t da = { 0 };
954
955 if (mdb_getopts(argc, argv,
956 'v', MDB_OPT_SETBITS, B_TRUE, &da.da_verbose,
957 'a', MDB_OPT_SETBITS, B_TRUE, &da.da_address,
958 NULL) != argc)
959 return (DCMD_USAGE);
960
961 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "zfs_dbgmsgs", &sym)) {
962 mdb_warn("can't find zfs_dbgmsgs");
963 return (DCMD_ERR);
964 }
965
966 if (mdb_pwalk("list", dbgmsg_cb, &da, sym.st_value) != 0) {
967 mdb_warn("can't walk zfs_dbgmsgs");
968 return (DCMD_ERR);
969 }
970
971 return (DCMD_OK);
972 }
973
974 /*ARGSUSED*/
975 static int
976 arc_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
977 {
978 kstat_named_t *stats;
979 GElf_Sym sym;
980 int nstats, i;
981 uint_t opt_a = FALSE;
982 uint_t opt_b = FALSE;
983 uint_t shift = 0;
984 const char *suffix;
985
986 static const char *bytestats[] = {
987 "p", "c", "c_min", "c_max", "size", "duplicate_buffers_size",
988 "arc_meta_used", "arc_meta_limit", "arc_meta_max",
989 "arc_meta_min", "hdr_size", "data_size", "metadata_size",
990 "other_size", "anon_size", "anon_evictable_data",
991 "anon_evictable_metadata", "mru_size", "mru_evictable_data",
992 "mru_evictable_metadata", "mru_ghost_size",
993 "mru_ghost_evictable_data", "mru_ghost_evictable_metadata",
994 "mfu_size", "mfu_evictable_data", "mfu_evictable_metadata",
995 "mfu_ghost_size", "mfu_ghost_evictable_data",
996 "mfu_ghost_evictable_metadata", "evict_l2_cached",
997 "evict_l2_eligible", "evict_l2_ineligible", "l2_read_bytes",
998 "l2_write_bytes", "l2_size", "l2_asize", "l2_hdr_size",
999 "compressed_size", "uncompressed_size", "overhead_size",
1000 NULL
1001 };
1002
1003 static const char *extras[] = {
1004 "arc_no_grow", "arc_tempreserve",
1005 NULL
1006 };
1007
1008 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "arc_stats", &sym) == -1) {
1009 mdb_warn("failed to find 'arc_stats'");
1010 return (DCMD_ERR);
1011 }
1012
1013 stats = mdb_zalloc(sym.st_size, UM_SLEEP | UM_GC);
1014
1015 if (mdb_vread(stats, sym.st_size, sym.st_value) == -1) {
1016 mdb_warn("couldn't read 'arc_stats' at %p", sym.st_value);
1017 return (DCMD_ERR);
1018 }
1019
1020 nstats = sym.st_size / sizeof (kstat_named_t);
1021
1022 /* NB: -a / opt_a are ignored for backwards compatability */
1023 if (mdb_getopts(argc, argv,
1024 'a', MDB_OPT_SETBITS, TRUE, &opt_a,
1025 'b', MDB_OPT_SETBITS, TRUE, &opt_b,
1026 'k', MDB_OPT_SETBITS, 10, &shift,
1027 'm', MDB_OPT_SETBITS, 20, &shift,
1028 'g', MDB_OPT_SETBITS, 30, &shift,
1029 NULL) != argc)
1030 return (DCMD_USAGE);
1031
1032 if (!opt_b && !shift)
1033 shift = 20;
1034
1035 switch (shift) {
1036 case 0:
1037 suffix = "B";
1038 break;
1039 case 10:
1040 suffix = "KB";
1041 break;
1042 case 20:
1043 suffix = "MB";
1044 break;
1045 case 30:
1046 suffix = "GB";
1047 break;
1048 default:
1049 suffix = "XX";
1050 }
1051
1052 for (i = 0; i < nstats; i++) {
1053 int j;
1054 boolean_t bytes = B_FALSE;
1055
1056 for (j = 0; bytestats[j]; j++) {
1057 if (strcmp(stats[i].name, bytestats[j]) == 0) {
1058 bytes = B_TRUE;
1059 break;
1060 }
1061 }
1062
1063 if (bytes) {
1064 mdb_printf("%-25s = %9llu %s\n", stats[i].name,
1065 stats[i].value.ui64 >> shift, suffix);
1066 } else {
1067 mdb_printf("%-25s = %9llu\n", stats[i].name,
1068 stats[i].value.ui64);
1069 }
1070 }
1071
1072 for (i = 0; extras[i]; i++) {
1073 uint64_t buf;
1074
1075 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, extras[i], &sym) == -1) {
1076 mdb_warn("failed to find '%s'", extras[i]);
1077 return (DCMD_ERR);
1078 }
1079
1080 if (sym.st_size != sizeof (uint64_t) &&
1081 sym.st_size != sizeof (uint32_t)) {
1082 mdb_warn("expected scalar for variable '%s'\n",
1083 extras[i]);
1084 return (DCMD_ERR);
1085 }
1086
1087 if (mdb_vread(&buf, sym.st_size, sym.st_value) == -1) {
1088 mdb_warn("couldn't read '%s'", extras[i]);
1089 return (DCMD_ERR);
1090 }
1091
1092 mdb_printf("%-25s = ", extras[i]);
1093
1094 /* NB: all the 64-bit extras happen to be byte counts */
1095 if (sym.st_size == sizeof (uint64_t))
1096 mdb_printf("%9llu %s\n", buf >> shift, suffix);
1097
1098 if (sym.st_size == sizeof (uint32_t))
1099 mdb_printf("%9d\n", *((uint32_t *)&buf));
1100 }
1101 return (DCMD_OK);
1102 }
1103
1104 typedef struct mdb_spa_print {
1105 pool_state_t spa_state;
1106 char spa_name[ZFS_MAX_DATASET_NAME_LEN];
1107 uintptr_t spa_normal_class;
1108 } mdb_spa_print_t;
1109
1110
1111 const char histo_stars[] = "****************************************";
1112 const int histo_width = sizeof (histo_stars) - 1;
1113
1114 static void
1115 dump_histogram(const uint64_t *histo, int size, int offset)
1116 {
1117 int i;
1118 int minidx = size - 1;
1119 int maxidx = 0;
1120 uint64_t max = 0;
1121
1122 for (i = 0; i < size; i++) {
1123 if (histo[i] > max)
1124 max = histo[i];
1125 if (histo[i] > 0 && i > maxidx)
1126 maxidx = i;
1127 if (histo[i] > 0 && i < minidx)
1128 minidx = i;
1129 }
1130
1131 if (max < histo_width)
1132 max = histo_width;
1133
1134 for (i = minidx; i <= maxidx; i++) {
1135 mdb_printf("%3u: %6llu %s\n",
1136 i + offset, (u_longlong_t)histo[i],
1137 &histo_stars[(max - histo[i]) * histo_width / max]);
1138 }
1139 }
1140
1141 typedef struct mdb_metaslab_class {
1142 uint64_t mc_histogram[RANGE_TREE_HISTOGRAM_SIZE];
1143 } mdb_metaslab_class_t;
1144
1145 /*
1146 * spa_class_histogram(uintptr_t class_addr)
1147 *
1148 * Prints free space histogram for a device class
1149 *
1150 * Returns DCMD_OK, or DCMD_ERR.
1151 */
1152 static int
1153 spa_class_histogram(uintptr_t class_addr)
1154 {
1155 mdb_metaslab_class_t mc;
1156 if (mdb_ctf_vread(&mc, "metaslab_class_t",
1157 "mdb_metaslab_class_t", class_addr, 0) == -1)
1158 return (DCMD_ERR);
1159
1160 mdb_inc_indent(4);
1161 dump_histogram(mc.mc_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1162 mdb_dec_indent(4);
1163 return (DCMD_OK);
1164 }
1165
1166 /*
1167 * ::spa
1168 *
1169 * -c Print configuration information as well
1170 * -v Print vdev state
1171 * -e Print vdev error stats
1172 * -m Print vdev metaslab info
1173 * -M print vdev metaslab group info
1174 * -h Print histogram info (must be combined with -m or -M)
1175 *
1176 * Print a summarized spa_t. When given no arguments, prints out a table of all
1177 * active pools on the system.
1178 */
1179 /* ARGSUSED */
1180 static int
1181 spa_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1182 {
1183 const char *statetab[] = { "ACTIVE", "EXPORTED", "DESTROYED",
1184 "SPARE", "L2CACHE", "UNINIT", "UNAVAIL", "POTENTIAL" };
1185 const char *state;
1186 int spa_flags = 0;
1187
1188 if (mdb_getopts(argc, argv,
1189 'c', MDB_OPT_SETBITS, SPA_FLAG_CONFIG, &spa_flags,
1190 'v', MDB_OPT_SETBITS, SPA_FLAG_VDEVS, &spa_flags,
1191 'e', MDB_OPT_SETBITS, SPA_FLAG_ERRORS, &spa_flags,
1192 'M', MDB_OPT_SETBITS, SPA_FLAG_METASLAB_GROUPS, &spa_flags,
1193 'm', MDB_OPT_SETBITS, SPA_FLAG_METASLABS, &spa_flags,
1194 'h', MDB_OPT_SETBITS, SPA_FLAG_HISTOGRAMS, &spa_flags,
1195 NULL) != argc)
1196 return (DCMD_USAGE);
1197
1198 if (!(flags & DCMD_ADDRSPEC)) {
1199 if (mdb_walk_dcmd("spa", "spa", argc, argv) == -1) {
1200 mdb_warn("can't walk spa");
1201 return (DCMD_ERR);
1202 }
1203
1204 return (DCMD_OK);
1205 }
1206
1207 if (flags & DCMD_PIPE_OUT) {
1208 mdb_printf("%#lr\n", addr);
1209 return (DCMD_OK);
1210 }
1211
1212 if (DCMD_HDRSPEC(flags))
1213 mdb_printf("%<u>%-?s %9s %-*s%</u>\n", "ADDR", "STATE",
1214 sizeof (uintptr_t) == 4 ? 60 : 52, "NAME");
1215
1216 mdb_spa_print_t spa;
1217 if (mdb_ctf_vread(&spa, "spa_t", "mdb_spa_print_t", addr, 0) == -1)
1218 return (DCMD_ERR);
1219
1220 if (spa.spa_state < 0 || spa.spa_state > POOL_STATE_UNAVAIL)
1221 state = "UNKNOWN";
1222 else
1223 state = statetab[spa.spa_state];
1224
1225 mdb_printf("%0?p %9s %s\n", addr, state, spa.spa_name);
1226 if (spa_flags & SPA_FLAG_HISTOGRAMS)
1227 spa_class_histogram(spa.spa_normal_class);
1228
1229 if (spa_flags & SPA_FLAG_CONFIG) {
1230 mdb_printf("\n");
1231 mdb_inc_indent(4);
1232 if (mdb_call_dcmd("spa_config", addr, flags, 0,
1233 NULL) != DCMD_OK)
1234 return (DCMD_ERR);
1235 mdb_dec_indent(4);
1236 }
1237
1238 if (spa_flags & SPA_FLAG_ALL_VDEV) {
1239 mdb_arg_t v;
1240 char opts[100] = "-";
1241 int args =
1242 (spa_flags | SPA_FLAG_VDEVS) == SPA_FLAG_VDEVS ? 0 : 1;
1243
1244 if (spa_flags & SPA_FLAG_ERRORS)
1245 strcat(opts, "e");
1246 if (spa_flags & SPA_FLAG_METASLABS)
1247 strcat(opts, "m");
1248 if (spa_flags & SPA_FLAG_METASLAB_GROUPS)
1249 strcat(opts, "M");
1250 if (spa_flags & SPA_FLAG_HISTOGRAMS)
1251 strcat(opts, "h");
1252
1253 v.a_type = MDB_TYPE_STRING;
1254 v.a_un.a_str = opts;
1255
1256 mdb_printf("\n");
1257 mdb_inc_indent(4);
1258 if (mdb_call_dcmd("spa_vdevs", addr, flags, args,
1259 &v) != DCMD_OK)
1260 return (DCMD_ERR);
1261 mdb_dec_indent(4);
1262 }
1263
1264 return (DCMD_OK);
1265 }
1266
1267 typedef struct mdb_spa_config_spa {
1268 uintptr_t spa_config;
1269 } mdb_spa_config_spa_t;
1270
1271 /*
1272 * ::spa_config
1273 *
1274 * Given a spa_t, print the configuration information stored in spa_config.
1275 * Since it's just an nvlist, format it as an indented list of name=value pairs.
1276 * We simply read the value of spa_config and pass off to ::nvlist.
1277 */
1278 /* ARGSUSED */
1279 static int
1280 spa_print_config(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1281 {
1282 mdb_spa_config_spa_t spa;
1283
1284 if (argc != 0 || !(flags & DCMD_ADDRSPEC))
1285 return (DCMD_USAGE);
1286
1287 if (mdb_ctf_vread(&spa, ZFS_STRUCT "spa", "mdb_spa_config_spa_t",
1288 addr, 0) == -1)
1289 return (DCMD_ERR);
1290
1291 if (spa.spa_config == 0) {
1292 mdb_printf("(none)\n");
1293 return (DCMD_OK);
1294 }
1295
1296 return (mdb_call_dcmd("nvlist", spa.spa_config, flags,
1297 0, NULL));
1298 }
1299
1300
1301
1302 typedef struct mdb_range_tree {
1303 uint64_t rt_space;
1304 } mdb_range_tree_t;
1305
1306 typedef struct mdb_metaslab_group {
1307 uint64_t mg_fragmentation;
1308 uint64_t mg_histogram[RANGE_TREE_HISTOGRAM_SIZE];
1309 uintptr_t mg_vd;
1310 } mdb_metaslab_group_t;
1311
1312 typedef struct mdb_metaslab {
1313 uint64_t ms_id;
1314 uint64_t ms_start;
1315 uint64_t ms_size;
1316 int64_t ms_deferspace;
1317 uint64_t ms_fragmentation;
1318 uint64_t ms_weight;
1319 uintptr_t ms_alloctree[TXG_SIZE];
1320 uintptr_t ms_freeingtree;
1321 uintptr_t ms_freedtree;
1322 uintptr_t ms_tree;
1323 uintptr_t ms_sm;
1324 } mdb_metaslab_t;
1325
1326 typedef struct mdb_space_map_phys_t {
1327 uint64_t smp_alloc;
1328 uint64_t smp_histogram[SPACE_MAP_HISTOGRAM_SIZE];
1329 } mdb_space_map_phys_t;
1330
1331 typedef struct mdb_space_map {
1332 uint64_t sm_size;
1333 uint8_t sm_shift;
1334 uint64_t sm_alloc;
1335 uintptr_t sm_phys;
1336 } mdb_space_map_t;
1337
1338 typedef struct mdb_vdev {
1339 uintptr_t vdev_path;
1340 uintptr_t vdev_ms;
1341 uintptr_t vdev_ops;
1342 uint64_t vdev_ms_count;
1343 uint64_t vdev_id;
1344 vdev_stat_t vdev_stat;
1345 } mdb_vdev_t;
1346
1347 typedef struct mdb_vdev_ops {
1348 char vdev_op_type[16];
1349 } mdb_vdev_ops_t;
1350
1351 static int
1352 metaslab_stats(uintptr_t addr, int spa_flags)
1353 {
1354 mdb_vdev_t vdev;
1355 uintptr_t *vdev_ms;
1356
1357 if (mdb_ctf_vread(&vdev, "vdev_t", "mdb_vdev_t",
1358 (uintptr_t)addr, 0) == -1) {
1359 mdb_warn("failed to read vdev at %p\n", addr);
1360 return (DCMD_ERR);
1361 }
1362
1363 mdb_inc_indent(4);
1364 mdb_printf("%<u>%-?s %6s %20s %10s %9s%</u>\n", "ADDR", "ID",
1365 "OFFSET", "FREE", "FRAGMENTATION");
1366
1367 vdev_ms = mdb_alloc(vdev.vdev_ms_count * sizeof (void *),
1368 UM_SLEEP | UM_GC);
1369 if (mdb_vread(vdev_ms, vdev.vdev_ms_count * sizeof (void *),
1370 (uintptr_t)vdev.vdev_ms) == -1) {
1371 mdb_warn("failed to read vdev_ms at %p\n", vdev.vdev_ms);
1372 return (DCMD_ERR);
1373 }
1374
1375 for (int m = 0; m < vdev.vdev_ms_count; m++) {
1376 mdb_metaslab_t ms;
1377 mdb_space_map_t sm = { 0 };
1378 char free[NICENUM_BUFLEN];
1379
1380 if (mdb_ctf_vread(&ms, "metaslab_t", "mdb_metaslab_t",
1381 (uintptr_t)vdev_ms[m], 0) == -1)
1382 return (DCMD_ERR);
1383
1384 if (ms.ms_sm != (uintptr_t)NULL &&
1385 mdb_ctf_vread(&sm, "space_map_t", "mdb_space_map_t",
1386 ms.ms_sm, 0) == -1)
1387 return (DCMD_ERR);
1388
1389 mdb_nicenum(ms.ms_size - sm.sm_alloc, free);
1390
1391 mdb_printf("%0?p %6llu %20llx %10s ", vdev_ms[m], ms.ms_id,
1392 ms.ms_start, free);
1393 if (ms.ms_fragmentation == ZFS_FRAG_INVALID)
1394 mdb_printf("%9s\n", "-");
1395 else
1396 mdb_printf("%9llu%%\n", ms.ms_fragmentation);
1397
1398 if ((spa_flags & SPA_FLAG_HISTOGRAMS) && ms.ms_sm != 0) {
1399 mdb_space_map_phys_t smp;
1400
1401 if (sm.sm_phys == (uintptr_t)NULL)
1402 continue;
1403
1404 (void) mdb_ctf_vread(&smp, "space_map_phys_t",
1405 "mdb_space_map_phys_t", sm.sm_phys, 0);
1406
1407 dump_histogram(smp.smp_histogram,
1408 SPACE_MAP_HISTOGRAM_SIZE, sm.sm_shift);
1409 }
1410 }
1411 mdb_dec_indent(4);
1412 return (DCMD_OK);
1413 }
1414
1415 static int
1416 metaslab_group_stats(uintptr_t addr, int spa_flags)
1417 {
1418 mdb_metaslab_group_t mg;
1419 if (mdb_ctf_vread(&mg, "metaslab_group_t", "mdb_metaslab_group_t",
1420 (uintptr_t)addr, 0) == -1) {
1421 mdb_warn("failed to read vdev_mg at %p\n", addr);
1422 return (DCMD_ERR);
1423 }
1424
1425 mdb_inc_indent(4);
1426 mdb_printf("%<u>%-?s %15s%</u>\n", "ADDR", "FRAGMENTATION");
1427 if (mg.mg_fragmentation == ZFS_FRAG_INVALID)
1428 mdb_printf("%0?p %15s\n", addr, "-");
1429 else
1430 mdb_printf("%0?p %15llu%%\n", addr, mg.mg_fragmentation);
1431
1432 if (spa_flags & SPA_FLAG_HISTOGRAMS)
1433 dump_histogram(mg.mg_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1434 mdb_dec_indent(4);
1435 return (DCMD_OK);
1436 }
1437
1438 /*
1439 * ::vdev
1440 *
1441 * Print out a summarized vdev_t, in the following form:
1442 *
1443 * ADDR STATE AUX DESC
1444 * fffffffbcde23df0 HEALTHY - /dev/dsk/c0t0d0
1445 *
1446 * If '-r' is specified, recursively visit all children.
1447 *
1448 * With '-e', the statistics associated with the vdev are printed as well.
1449 */
1450 static int
1451 do_print_vdev(uintptr_t addr, int flags, int depth, boolean_t recursive,
1452 int spa_flags)
1453 {
1454 vdev_t vdev;
1455 char desc[MAXNAMELEN];
1456 int c, children;
1457 uintptr_t *child;
1458 const char *state, *aux;
1459
1460 if (mdb_vread(&vdev, sizeof (vdev), (uintptr_t)addr) == -1) {
1461 mdb_warn("failed to read vdev_t at %p\n", (uintptr_t)addr);
1462 return (DCMD_ERR);
1463 }
1464
1465 if (flags & DCMD_PIPE_OUT) {
1466 mdb_printf("%#lr\n", addr);
1467 } else {
1468 if (vdev.vdev_path != NULL) {
1469 if (mdb_readstr(desc, sizeof (desc),
1470 (uintptr_t)vdev.vdev_path) == -1) {
1471 mdb_warn("failed to read vdev_path at %p\n",
1472 vdev.vdev_path);
1473 return (DCMD_ERR);
1474 }
1475 } else if (vdev.vdev_ops != NULL) {
1476 vdev_ops_t ops;
1477 if (mdb_vread(&ops, sizeof (ops),
1478 (uintptr_t)vdev.vdev_ops) == -1) {
1479 mdb_warn("failed to read vdev_ops at %p\n",
1480 vdev.vdev_ops);
1481 return (DCMD_ERR);
1482 }
1483 (void) strcpy(desc, ops.vdev_op_type);
1484 } else {
1485 (void) strcpy(desc, "<unknown>");
1486 }
1487
1488 if (depth == 0 && DCMD_HDRSPEC(flags))
1489 mdb_printf("%<u>%-?s %-9s %-12s %-*s%</u>\n",
1490 "ADDR", "STATE", "AUX",
1491 sizeof (uintptr_t) == 4 ? 43 : 35,
1492 "DESCRIPTION");
1493
1494 mdb_printf("%0?p ", addr);
1495
1496 switch (vdev.vdev_state) {
1497 case VDEV_STATE_CLOSED:
1498 state = "CLOSED";
1499 break;
1500 case VDEV_STATE_OFFLINE:
1501 state = "OFFLINE";
1502 break;
1503 case VDEV_STATE_CANT_OPEN:
1504 state = "CANT_OPEN";
1505 break;
1506 case VDEV_STATE_DEGRADED:
1507 state = "DEGRADED";
1508 break;
1509 case VDEV_STATE_HEALTHY:
1510 state = "HEALTHY";
1511 break;
1512 case VDEV_STATE_REMOVED:
1513 state = "REMOVED";
1514 break;
1515 case VDEV_STATE_FAULTED:
1516 state = "FAULTED";
1517 break;
1518 default:
1519 state = "UNKNOWN";
1520 break;
1521 }
1522
1523 switch (vdev.vdev_stat.vs_aux) {
1524 case VDEV_AUX_NONE:
1525 aux = "-";
1526 break;
1527 case VDEV_AUX_OPEN_FAILED:
1528 aux = "OPEN_FAILED";
1529 break;
1530 case VDEV_AUX_CORRUPT_DATA:
1531 aux = "CORRUPT_DATA";
1532 break;
1533 case VDEV_AUX_NO_REPLICAS:
1534 aux = "NO_REPLICAS";
1535 break;
1536 case VDEV_AUX_BAD_GUID_SUM:
1537 aux = "BAD_GUID_SUM";
1538 break;
1539 case VDEV_AUX_TOO_SMALL:
1540 aux = "TOO_SMALL";
1541 break;
1542 case VDEV_AUX_BAD_LABEL:
1543 aux = "BAD_LABEL";
1544 break;
1545 case VDEV_AUX_VERSION_NEWER:
1546 aux = "VERS_NEWER";
1547 break;
1548 case VDEV_AUX_VERSION_OLDER:
1549 aux = "VERS_OLDER";
1550 break;
1551 case VDEV_AUX_UNSUP_FEAT:
1552 aux = "UNSUP_FEAT";
1553 break;
1554 case VDEV_AUX_SPARED:
1555 aux = "SPARED";
1556 break;
1557 case VDEV_AUX_ERR_EXCEEDED:
1558 aux = "ERR_EXCEEDED";
1559 break;
1560 case VDEV_AUX_IO_FAILURE:
1561 aux = "IO_FAILURE";
1562 break;
1563 case VDEV_AUX_BAD_LOG:
1564 aux = "BAD_LOG";
1565 break;
1566 case VDEV_AUX_EXTERNAL:
1567 aux = "EXTERNAL";
1568 break;
1569 case VDEV_AUX_SPLIT_POOL:
1570 aux = "SPLIT_POOL";
1571 break;
1572 case VDEV_AUX_CHILDREN_OFFLINE:
1573 aux = "CHILDREN_OFFLINE";
1574 break;
1575 default:
1576 aux = "UNKNOWN";
1577 break;
1578 }
1579
1580 mdb_printf("%-9s %-12s %*s%s\n", state, aux, depth, "", desc);
1581
1582 if (spa_flags & SPA_FLAG_ERRORS) {
1583 vdev_stat_t *vs = &vdev.vdev_stat;
1584 int i;
1585
1586 mdb_inc_indent(4);
1587 mdb_printf("\n");
1588 mdb_printf("%<u> %12s %12s %12s %12s "
1589 "%12s%</u>\n", "READ", "WRITE", "FREE", "CLAIM",
1590 "IOCTL");
1591 mdb_printf("OPS ");
1592 for (i = 1; i < ZIO_TYPES; i++)
1593 mdb_printf("%11#llx%s", vs->vs_ops[i],
1594 i == ZIO_TYPES - 1 ? "" : " ");
1595 mdb_printf("\n");
1596 mdb_printf("BYTES ");
1597 for (i = 1; i < ZIO_TYPES; i++)
1598 mdb_printf("%11#llx%s", vs->vs_bytes[i],
1599 i == ZIO_TYPES - 1 ? "" : " ");
1600
1601
1602 mdb_printf("\n");
1603 mdb_printf("EREAD %10#llx\n", vs->vs_read_errors);
1604 mdb_printf("EWRITE %10#llx\n", vs->vs_write_errors);
1605 mdb_printf("ECKSUM %10#llx\n",
1606 vs->vs_checksum_errors);
1607 mdb_dec_indent(4);
1608 mdb_printf("\n");
1609 }
1610
1611 if (spa_flags & SPA_FLAG_METASLAB_GROUPS &&
1612 vdev.vdev_mg != NULL) {
1613 metaslab_group_stats((uintptr_t)vdev.vdev_mg,
1614 spa_flags);
1615 }
1616 if (spa_flags & SPA_FLAG_METASLABS && vdev.vdev_ms != NULL) {
1617 metaslab_stats((uintptr_t)addr, spa_flags);
1618 }
1619 }
1620
1621 children = vdev.vdev_children;
1622
1623 if (children == 0 || !recursive)
1624 return (DCMD_OK);
1625
1626 child = mdb_alloc(children * sizeof (void *), UM_SLEEP | UM_GC);
1627 if (mdb_vread(child, children * sizeof (void *),
1628 (uintptr_t)vdev.vdev_child) == -1) {
1629 mdb_warn("failed to read vdev children at %p", vdev.vdev_child);
1630 return (DCMD_ERR);
1631 }
1632
1633 for (c = 0; c < children; c++) {
1634 if (do_print_vdev(child[c], flags, depth + 2, recursive,
1635 spa_flags)) {
1636 return (DCMD_ERR);
1637 }
1638 }
1639
1640 return (DCMD_OK);
1641 }
1642
1643 static int
1644 vdev_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1645 {
1646 uint64_t depth = 0;
1647 boolean_t recursive = B_FALSE;
1648 int spa_flags = 0;
1649
1650 if (mdb_getopts(argc, argv,
1651 'e', MDB_OPT_SETBITS, SPA_FLAG_ERRORS, &spa_flags,
1652 'm', MDB_OPT_SETBITS, SPA_FLAG_METASLABS, &spa_flags,
1653 'M', MDB_OPT_SETBITS, SPA_FLAG_METASLAB_GROUPS, &spa_flags,
1654 'h', MDB_OPT_SETBITS, SPA_FLAG_HISTOGRAMS, &spa_flags,
1655 'r', MDB_OPT_SETBITS, TRUE, &recursive,
1656 'd', MDB_OPT_UINT64, &depth, NULL) != argc)
1657 return (DCMD_USAGE);
1658
1659 if (!(flags & DCMD_ADDRSPEC)) {
1660 mdb_warn("no vdev_t address given\n");
1661 return (DCMD_ERR);
1662 }
1663
1664 return (do_print_vdev(addr, flags, (int)depth, recursive, spa_flags));
1665 }
1666
1667 typedef struct mdb_metaslab_alloc_trace {
1668 uintptr_t mat_mg;
1669 uintptr_t mat_msp;
1670 uint64_t mat_size;
1671 uint64_t mat_weight;
1672 uint64_t mat_offset;
1673 uint32_t mat_dva_id;
1674 } mdb_metaslab_alloc_trace_t;
1675
1676 static void
1677 metaslab_print_weight(uint64_t weight)
1678 {
1679 char buf[100];
1680
1681 if (WEIGHT_IS_SPACEBASED(weight)) {
1682 mdb_nicenum(
1683 weight & ~(METASLAB_ACTIVE_MASK | METASLAB_WEIGHT_TYPE),
1684 buf);
1685 } else {
1686 char size[NICENUM_BUFLEN];
1687 mdb_nicenum(1ULL << WEIGHT_GET_INDEX(weight), size);
1688 (void) mdb_snprintf(buf, sizeof (buf), "%llu x %s",
1689 WEIGHT_GET_COUNT(weight), size);
1690 }
1691 mdb_printf("%11s ", buf);
1692 }
1693
1694 /* ARGSUSED */
1695 static int
1696 metaslab_weight(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1697 {
1698 uint64_t weight = 0;
1699 char active;
1700
1701 if (argc == 0 && (flags & DCMD_ADDRSPEC)) {
1702 if (mdb_vread(&weight, sizeof (uint64_t), addr) == -1) {
1703 mdb_warn("failed to read weight at %p\n", addr);
1704 return (DCMD_ERR);
1705 }
1706 } else if (argc == 1 && !(flags & DCMD_ADDRSPEC)) {
1707 weight = (argv[0].a_type == MDB_TYPE_IMMEDIATE) ?
1708 argv[0].a_un.a_val : mdb_strtoull(argv[0].a_un.a_str);
1709 } else {
1710 return (DCMD_USAGE);
1711 }
1712
1713 if (DCMD_HDRSPEC(flags)) {
1714 mdb_printf("%<u>%-6s %9s %9s%</u>\n",
1715 "ACTIVE", "ALGORITHM", "WEIGHT");
1716 }
1717
1718 if (weight & METASLAB_WEIGHT_PRIMARY)
1719 active = 'P';
1720 else if (weight & METASLAB_WEIGHT_SECONDARY)
1721 active = 'S';
1722 else
1723 active = '-';
1724 mdb_printf("%6c %8s ", active,
1725 WEIGHT_IS_SPACEBASED(weight) ? "SPACE" : "SEGMENT");
1726 metaslab_print_weight(weight);
1727 mdb_printf("\n");
1728
1729 return (DCMD_OK);
1730 }
1731
1732 /* ARGSUSED */
1733 static int
1734 metaslab_trace(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1735 {
1736 mdb_metaslab_alloc_trace_t mat;
1737 mdb_metaslab_group_t mg = { 0 };
1738 char result_type[100];
1739
1740 if (mdb_ctf_vread(&mat, "metaslab_alloc_trace_t",
1741 "mdb_metaslab_alloc_trace_t", addr, 0) == -1) {
1742 return (DCMD_ERR);
1743 }
1744
1745 if (!(flags & DCMD_PIPE_OUT) && DCMD_HDRSPEC(flags)) {
1746 mdb_printf("%<u>%6s %6s %8s %11s %18s %18s%</u>\n",
1747 "MSID", "DVA", "ASIZE", "WEIGHT", "RESULT", "VDEV");
1748 }
1749
1750 if (mat.mat_msp != 0) {
1751 mdb_metaslab_t ms;
1752
1753 if (mdb_ctf_vread(&ms, "metaslab_t", "mdb_metaslab_t",
1754 mat.mat_msp, 0) == -1) {
1755 return (DCMD_ERR);
1756 }
1757 mdb_printf("%6llu ", ms.ms_id);
1758 } else {
1759 mdb_printf("%6s ", "-");
1760 }
1761
1762 mdb_printf("%6d %8llx ", mat.mat_dva_id, mat.mat_size);
1763
1764 metaslab_print_weight(mat.mat_weight);
1765
1766 if ((int64_t)mat.mat_offset < 0) {
1767 if (enum_lookup("enum trace_alloc_type", mat.mat_offset,
1768 "TRACE_", sizeof (result_type), result_type) == -1) {
1769 mdb_warn("Could not find enum for trace_alloc_type");
1770 return (DCMD_ERR);
1771 }
1772 mdb_printf("%18s ", result_type);
1773 } else {
1774 mdb_printf("%<b>%18llx%</b> ", mat.mat_offset);
1775 }
1776
1777 if (mat.mat_mg != 0 &&
1778 mdb_ctf_vread(&mg, "metaslab_group_t", "mdb_metaslab_group_t",
1779 mat.mat_mg, 0) == -1) {
1780 return (DCMD_ERR);
1781 }
1782
1783 if (mg.mg_vd != 0) {
1784 mdb_vdev_t vdev;
1785 char desc[MAXNAMELEN];
1786
1787 if (mdb_ctf_vread(&vdev, "vdev_t", "mdb_vdev_t",
1788 mg.mg_vd, 0) == -1) {
1789 return (DCMD_ERR);
1790 }
1791
1792 if (vdev.vdev_path != 0) {
1793 char path[MAXNAMELEN];
1794
1795 if (mdb_readstr(path, sizeof (path),
1796 vdev.vdev_path) == -1) {
1797 mdb_warn("failed to read vdev_path at %p\n",
1798 vdev.vdev_path);
1799 return (DCMD_ERR);
1800 }
1801 char *slash;
1802 if ((slash = strrchr(path, '/')) != NULL) {
1803 strcpy(desc, slash + 1);
1804 } else {
1805 strcpy(desc, path);
1806 }
1807 } else if (vdev.vdev_ops != 0) {
1808 mdb_vdev_ops_t ops;
1809 if (mdb_ctf_vread(&ops, "vdev_ops_t", "mdb_vdev_ops_t",
1810 vdev.vdev_ops, 0) == -1) {
1811 mdb_warn("failed to read vdev_ops at %p\n",
1812 vdev.vdev_ops);
1813 return (DCMD_ERR);
1814 }
1815 (void) mdb_snprintf(desc, sizeof (desc),
1816 "%s-%llu", ops.vdev_op_type, vdev.vdev_id);
1817 } else {
1818 (void) strcpy(desc, "<unknown>");
1819 }
1820 mdb_printf("%18s\n", desc);
1821 }
1822
1823 return (DCMD_OK);
1824 }
1825
1826 typedef struct metaslab_walk_data {
1827 uint64_t mw_numvdevs;
1828 uintptr_t *mw_vdevs;
1829 int mw_curvdev;
1830 uint64_t mw_nummss;
1831 uintptr_t *mw_mss;
1832 int mw_curms;
1833 } metaslab_walk_data_t;
1834
1835 static int
1836 metaslab_walk_step(mdb_walk_state_t *wsp)
1837 {
1838 metaslab_walk_data_t *mw = wsp->walk_data;
1839 metaslab_t ms;
1840 uintptr_t msp;
1841
1842 if (mw->mw_curvdev >= mw->mw_numvdevs)
1843 return (WALK_DONE);
1844
1845 if (mw->mw_mss == NULL) {
1846 uintptr_t mssp;
1847 uintptr_t vdevp;
1848
1849 ASSERT(mw->mw_curms == 0);
1850 ASSERT(mw->mw_nummss == 0);
1851
1852 vdevp = mw->mw_vdevs[mw->mw_curvdev];
1853 if (GETMEMB(vdevp, "vdev", vdev_ms, mssp) ||
1854 GETMEMB(vdevp, "vdev", vdev_ms_count, mw->mw_nummss)) {
1855 return (WALK_ERR);
1856 }
1857
1858 mw->mw_mss = mdb_alloc(mw->mw_nummss * sizeof (void*),
1859 UM_SLEEP | UM_GC);
1860 if (mdb_vread(mw->mw_mss, mw->mw_nummss * sizeof (void*),
1861 mssp) == -1) {
1862 mdb_warn("failed to read vdev_ms at %p", mssp);
1863 return (WALK_ERR);
1864 }
1865 }
1866
1867 if (mw->mw_curms >= mw->mw_nummss) {
1868 mw->mw_mss = NULL;
1869 mw->mw_curms = 0;
1870 mw->mw_nummss = 0;
1871 mw->mw_curvdev++;
1872 return (WALK_NEXT);
1873 }
1874
1875 msp = mw->mw_mss[mw->mw_curms];
1876 if (mdb_vread(&ms, sizeof (metaslab_t), msp) == -1) {
1877 mdb_warn("failed to read metaslab_t at %p", msp);
1878 return (WALK_ERR);
1879 }
1880
1881 mw->mw_curms++;
1882
1883 return (wsp->walk_callback(msp, &ms, wsp->walk_cbdata));
1884 }
1885
1886 static int
1887 metaslab_walk_init(mdb_walk_state_t *wsp)
1888 {
1889 metaslab_walk_data_t *mw;
1890 uintptr_t root_vdevp;
1891 uintptr_t childp;
1892
1893 if (wsp->walk_addr == (uintptr_t)NULL) {
1894 mdb_warn("must supply address of spa_t\n");
1895 return (WALK_ERR);
1896 }
1897
1898 mw = mdb_zalloc(sizeof (metaslab_walk_data_t), UM_SLEEP | UM_GC);
1899
1900 if (GETMEMB(wsp->walk_addr, "spa", spa_root_vdev, root_vdevp) ||
1901 GETMEMB(root_vdevp, "vdev", vdev_children, mw->mw_numvdevs) ||
1902 GETMEMB(root_vdevp, "vdev", vdev_child, childp)) {
1903 return (DCMD_ERR);
1904 }
1905
1906 mw->mw_vdevs = mdb_alloc(mw->mw_numvdevs * sizeof (void *),
1907 UM_SLEEP | UM_GC);
1908 if (mdb_vread(mw->mw_vdevs, mw->mw_numvdevs * sizeof (void *),
1909 childp) == -1) {
1910 mdb_warn("failed to read root vdev children at %p", childp);
1911 return (DCMD_ERR);
1912 }
1913
1914 wsp->walk_data = mw;
1915
1916 return (WALK_NEXT);
1917 }
1918
1919 typedef struct mdb_spa {
1920 uintptr_t spa_dsl_pool;
1921 uintptr_t spa_root_vdev;
1922 } mdb_spa_t;
1923
1924 typedef struct mdb_dsl_pool {
1925 uintptr_t dp_root_dir;
1926 } mdb_dsl_pool_t;
1927
1928 typedef struct mdb_dsl_dir {
1929 uintptr_t dd_dbuf;
1930 int64_t dd_space_towrite[TXG_SIZE];
1931 } mdb_dsl_dir_t;
1932
1933 typedef struct mdb_dsl_dir_phys {
1934 uint64_t dd_used_bytes;
1935 uint64_t dd_compressed_bytes;
1936 uint64_t dd_uncompressed_bytes;
1937 } mdb_dsl_dir_phys_t;
1938
1939 typedef struct space_data {
1940 uint64_t ms_alloctree[TXG_SIZE];
1941 uint64_t ms_freeingtree;
1942 uint64_t ms_freedtree;
1943 uint64_t ms_tree;
1944 int64_t ms_deferspace;
1945 uint64_t avail;
1946 uint64_t nowavail;
1947 } space_data_t;
1948
1949 /* ARGSUSED */
1950 static int
1951 space_cb(uintptr_t addr, const void *unknown, void *arg)
1952 {
1953 space_data_t *sd = arg;
1954 mdb_metaslab_t ms;
1955 mdb_range_tree_t rt;
1956 mdb_space_map_t sm = { 0 };
1957 mdb_space_map_phys_t smp = { 0 };
1958 int i;
1959
1960 if (mdb_ctf_vread(&ms, "metaslab_t", "mdb_metaslab_t",
1961 addr, 0) == -1)
1962 return (WALK_ERR);
1963
1964 for (i = 0; i < TXG_SIZE; i++) {
1965 if (mdb_ctf_vread(&rt, "range_tree_t",
1966 "mdb_range_tree_t", ms.ms_alloctree[i], 0) == -1)
1967 return (WALK_ERR);
1968
1969 sd->ms_alloctree[i] += rt.rt_space;
1970
1971 }
1972
1973 if (mdb_ctf_vread(&rt, "range_tree_t",
1974 "mdb_range_tree_t", ms.ms_freeingtree, 0) == -1)
1975 return (WALK_ERR);
1976 sd->ms_freeingtree += rt.rt_space;
1977
1978 if (mdb_ctf_vread(&rt, "range_tree_t",
1979 "mdb_range_tree_t", ms.ms_freedtree, 0) == -1)
1980 return (WALK_ERR);
1981 sd->ms_freedtree += rt.rt_space;
1982
1983 if (mdb_ctf_vread(&rt, "range_tree_t",
1984 "mdb_range_tree_t", ms.ms_tree, 0) == -1)
1985 return (WALK_ERR);
1986 sd->ms_tree += rt.rt_space;
1987
1988 if (ms.ms_sm != (uintptr_t)NULL &&
1989 mdb_ctf_vread(&sm, "space_map_t",
1990 "mdb_space_map_t", ms.ms_sm, 0) == -1)
1991 return (WALK_ERR);
1992
1993 if (sm.sm_phys != (uintptr_t)NULL) {
1994 (void) mdb_ctf_vread(&smp, "space_map_phys_t",
1995 "mdb_space_map_phys_t", sm.sm_phys, 0);
1996 }
1997
1998 sd->ms_deferspace += ms.ms_deferspace;
1999 sd->avail += sm.sm_size - sm.sm_alloc;
2000 sd->nowavail += sm.sm_size - smp.smp_alloc;
2001
2002 return (WALK_NEXT);
2003 }
2004
2005 /*
2006 * ::spa_space [-b]
2007 *
2008 * Given a spa_t, print out it's on-disk space usage and in-core
2009 * estimates of future usage. If -b is given, print space in bytes.
2010 * Otherwise print in megabytes.
2011 */
2012 /* ARGSUSED */
2013 static int
2014 spa_space(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2015 {
2016 mdb_spa_t spa;
2017 mdb_dsl_pool_t dp;
2018 mdb_dsl_dir_t dd;
2019 mdb_dmu_buf_impl_t db;
2020 mdb_dsl_dir_phys_t dsp;
2021 space_data_t sd;
2022 int shift = 20;
2023 char *suffix = "M";
2024 int bytes = B_FALSE;
2025
2026 if (mdb_getopts(argc, argv, 'b', MDB_OPT_SETBITS, TRUE, &bytes, NULL) !=
2027 argc)
2028 return (DCMD_USAGE);
2029 if (!(flags & DCMD_ADDRSPEC))
2030 return (DCMD_USAGE);
2031
2032 if (bytes) {
2033 shift = 0;
2034 suffix = "";
2035 }
2036
2037 if (mdb_ctf_vread(&spa, ZFS_STRUCT "spa", "mdb_spa_t",
2038 addr, 0) == -1 ||
2039 mdb_ctf_vread(&dp, ZFS_STRUCT "dsl_pool", "mdb_dsl_pool_t",
2040 spa.spa_dsl_pool, 0) == -1 ||
2041 mdb_ctf_vread(&dd, ZFS_STRUCT "dsl_dir", "mdb_dsl_dir_t",
2042 dp.dp_root_dir, 0) == -1 ||
2043 mdb_ctf_vread(&db, ZFS_STRUCT "dmu_buf_impl", "mdb_dmu_buf_impl_t",
2044 dd.dd_dbuf, 0) == -1 ||
2045 mdb_ctf_vread(&dsp, ZFS_STRUCT "dsl_dir_phys",
2046 "mdb_dsl_dir_phys_t", db.db.db_data, 0) == -1) {
2047 return (DCMD_ERR);
2048 }
2049
2050 mdb_printf("dd_space_towrite = %llu%s %llu%s %llu%s %llu%s\n",
2051 dd.dd_space_towrite[0] >> shift, suffix,
2052 dd.dd_space_towrite[1] >> shift, suffix,
2053 dd.dd_space_towrite[2] >> shift, suffix,
2054 dd.dd_space_towrite[3] >> shift, suffix);
2055
2056 mdb_printf("dd_phys.dd_used_bytes = %llu%s\n",
2057 dsp.dd_used_bytes >> shift, suffix);
2058 mdb_printf("dd_phys.dd_compressed_bytes = %llu%s\n",
2059 dsp.dd_compressed_bytes >> shift, suffix);
2060 mdb_printf("dd_phys.dd_uncompressed_bytes = %llu%s\n",
2061 dsp.dd_uncompressed_bytes >> shift, suffix);
2062
2063 bzero(&sd, sizeof (sd));
2064 if (mdb_pwalk("metaslab", space_cb, &sd, addr) != 0) {
2065 mdb_warn("can't walk metaslabs");
2066 return (DCMD_ERR);
2067 }
2068
2069 mdb_printf("ms_allocmap = %llu%s %llu%s %llu%s %llu%s\n",
2070 sd.ms_alloctree[0] >> shift, suffix,
2071 sd.ms_alloctree[1] >> shift, suffix,
2072 sd.ms_alloctree[2] >> shift, suffix,
2073 sd.ms_alloctree[3] >> shift, suffix);
2074 mdb_printf("ms_freeingtree = %llu%s\n",
2075 sd.ms_freeingtree >> shift, suffix);
2076 mdb_printf("ms_freedtree = %llu%s\n",
2077 sd.ms_freedtree >> shift, suffix);
2078 mdb_printf("ms_tree = %llu%s\n",
2079 sd.ms_tree >> shift, suffix);
2080 mdb_printf("ms_deferspace = %llu%s\n",
2081 sd.ms_deferspace >> shift, suffix);
2082 mdb_printf("last synced avail = %llu%s\n",
2083 sd.avail >> shift, suffix);
2084 mdb_printf("current syncing avail = %llu%s\n",
2085 sd.nowavail >> shift, suffix);
2086
2087 return (DCMD_OK);
2088 }
2089
2090 typedef struct mdb_spa_aux_vdev {
2091 int sav_count;
2092 uintptr_t sav_vdevs;
2093 } mdb_spa_aux_vdev_t;
2094
2095 typedef struct mdb_spa_vdevs {
2096 uintptr_t spa_root_vdev;
2097 mdb_spa_aux_vdev_t spa_l2cache;
2098 mdb_spa_aux_vdev_t spa_spares;
2099 } mdb_spa_vdevs_t;
2100
2101 static int
2102 spa_print_aux(mdb_spa_aux_vdev_t *sav, uint_t flags, mdb_arg_t *v,
2103 const char *name)
2104 {
2105 uintptr_t *aux;
2106 size_t len;
2107 int ret, i;
2108
2109 /*
2110 * Iterate over aux vdevs and print those out as well. This is a
2111 * little annoying because we don't have a root vdev to pass to ::vdev.
2112 * Instead, we print a single line and then call it for each child
2113 * vdev.
2114 */
2115 if (sav->sav_count != 0) {
2116 v[1].a_type = MDB_TYPE_STRING;
2117 v[1].a_un.a_str = "-d";
2118 v[2].a_type = MDB_TYPE_IMMEDIATE;
2119 v[2].a_un.a_val = 2;
2120
2121 len = sav->sav_count * sizeof (uintptr_t);
2122 aux = mdb_alloc(len, UM_SLEEP);
2123 if (mdb_vread(aux, len, sav->sav_vdevs) == -1) {
2124 mdb_free(aux, len);
2125 mdb_warn("failed to read l2cache vdevs at %p",
2126 sav->sav_vdevs);
2127 return (DCMD_ERR);
2128 }
2129
2130 mdb_printf("%-?s %-9s %-12s %s\n", "-", "-", "-", name);
2131
2132 for (i = 0; i < sav->sav_count; i++) {
2133 ret = mdb_call_dcmd("vdev", aux[i], flags, 3, v);
2134 if (ret != DCMD_OK) {
2135 mdb_free(aux, len);
2136 return (ret);
2137 }
2138 }
2139
2140 mdb_free(aux, len);
2141 }
2142
2143 return (0);
2144 }
2145
2146 /*
2147 * ::spa_vdevs
2148 *
2149 * -e Include error stats
2150 * -m Include metaslab information
2151 * -M Include metaslab group information
2152 * -h Include histogram information (requires -m or -M)
2153 *
2154 * Print out a summarized list of vdevs for the given spa_t.
2155 * This is accomplished by invoking "::vdev -re" on the root vdev, as well as
2156 * iterating over the cache devices.
2157 */
2158 /* ARGSUSED */
2159 static int
2160 spa_vdevs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2161 {
2162 mdb_arg_t v[3];
2163 int ret;
2164 char opts[100] = "-r";
2165 int spa_flags = 0;
2166
2167 if (mdb_getopts(argc, argv,
2168 'e', MDB_OPT_SETBITS, SPA_FLAG_ERRORS, &spa_flags,
2169 'm', MDB_OPT_SETBITS, SPA_FLAG_METASLABS, &spa_flags,
2170 'M', MDB_OPT_SETBITS, SPA_FLAG_METASLAB_GROUPS, &spa_flags,
2171 'h', MDB_OPT_SETBITS, SPA_FLAG_HISTOGRAMS, &spa_flags,
2172 NULL) != argc)
2173 return (DCMD_USAGE);
2174
2175 if (!(flags & DCMD_ADDRSPEC))
2176 return (DCMD_USAGE);
2177
2178 mdb_spa_vdevs_t spa;
2179 if (mdb_ctf_vread(&spa, "spa_t", "mdb_spa_vdevs_t", addr, 0) == -1)
2180 return (DCMD_ERR);
2181
2182 /*
2183 * Unitialized spa_t structures can have a NULL root vdev.
2184 */
2185 if (spa.spa_root_vdev == (uintptr_t)NULL) {
2186 mdb_printf("no associated vdevs\n");
2187 return (DCMD_OK);
2188 }
2189
2190 if (spa_flags & SPA_FLAG_ERRORS)
2191 strcat(opts, "e");
2192 if (spa_flags & SPA_FLAG_METASLABS)
2193 strcat(opts, "m");
2194 if (spa_flags & SPA_FLAG_METASLAB_GROUPS)
2195 strcat(opts, "M");
2196 if (spa_flags & SPA_FLAG_HISTOGRAMS)
2197 strcat(opts, "h");
2198
2199 v[0].a_type = MDB_TYPE_STRING;
2200 v[0].a_un.a_str = opts;
2201
2202 ret = mdb_call_dcmd("vdev", (uintptr_t)spa.spa_root_vdev,
2203 flags, 1, v);
2204 if (ret != DCMD_OK)
2205 return (ret);
2206
2207 if (spa_print_aux(&spa.spa_l2cache, flags, v, "cache") != 0 ||
2208 spa_print_aux(&spa.spa_spares, flags, v, "spares") != 0)
2209 return (DCMD_ERR);
2210
2211 return (DCMD_OK);
2212 }
2213
2214 /*
2215 * ::zio
2216 *
2217 * Print a summary of zio_t and all its children. This is intended to display a
2218 * zio tree, and hence we only pick the most important pieces of information for
2219 * the main summary. More detailed information can always be found by doing a
2220 * '::print zio' on the underlying zio_t. The columns we display are:
2221 *
2222 * ADDRESS TYPE STAGE WAITER TIME_ELAPSED
2223 *
2224 * The 'address' column is indented by one space for each depth level as we
2225 * descend down the tree.
2226 */
2227
2228 #define ZIO_MAXINDENT 7
2229 #define ZIO_MAXWIDTH (sizeof (uintptr_t) * 2 + ZIO_MAXINDENT)
2230 #define ZIO_WALK_SELF 0
2231 #define ZIO_WALK_CHILD 1
2232 #define ZIO_WALK_PARENT 2
2233
2234 typedef struct zio_print_args {
2235 int zpa_current_depth;
2236 int zpa_min_depth;
2237 int zpa_max_depth;
2238 int zpa_type;
2239 uint_t zpa_flags;
2240 } zio_print_args_t;
2241
2242 typedef struct mdb_zio {
2243 enum zio_type io_type;
2244 enum zio_stage io_stage;
2245 uintptr_t io_waiter;
2246 uintptr_t io_spa;
2247 struct {
2248 struct {
2249 uintptr_t list_next;
2250 } list_head;
2251 } io_parent_list;
2252 int io_error;
2253 } mdb_zio_t;
2254
2255 typedef struct mdb_zio_timestamp {
2256 hrtime_t io_timestamp;
2257 } mdb_zio_timestamp_t;
2258
2259 static int zio_child_cb(uintptr_t addr, const void *unknown, void *arg);
2260
2261 static int
2262 zio_print_cb(uintptr_t addr, zio_print_args_t *zpa)
2263 {
2264 mdb_ctf_id_t type_enum, stage_enum;
2265 int indent = zpa->zpa_current_depth;
2266 const char *type, *stage;
2267 uintptr_t laddr;
2268 mdb_zio_t zio;
2269 mdb_zio_timestamp_t zio_timestamp = { 0 };
2270
2271 if (mdb_ctf_vread(&zio, ZFS_STRUCT "zio", "mdb_zio_t", addr, 0) == -1)
2272 return (WALK_ERR);
2273 (void) mdb_ctf_vread(&zio_timestamp, ZFS_STRUCT "zio",
2274 "mdb_zio_timestamp_t", addr, MDB_CTF_VREAD_QUIET);
2275
2276 if (indent > ZIO_MAXINDENT)
2277 indent = ZIO_MAXINDENT;
2278
2279 if (mdb_ctf_lookup_by_name("enum zio_type", &type_enum) == -1 ||
2280 mdb_ctf_lookup_by_name("enum zio_stage", &stage_enum) == -1) {
2281 mdb_warn("failed to lookup zio enums");
2282 return (WALK_ERR);
2283 }
2284
2285 if ((type = mdb_ctf_enum_name(type_enum, zio.io_type)) != NULL)
2286 type += sizeof ("ZIO_TYPE_") - 1;
2287 else
2288 type = "?";
2289
2290 if (zio.io_error == 0) {
2291 stage = mdb_ctf_enum_name(stage_enum, zio.io_stage);
2292 if (stage != NULL)
2293 stage += sizeof ("ZIO_STAGE_") - 1;
2294 else
2295 stage = "?";
2296 } else {
2297 stage = "FAILED";
2298 }
2299
2300 if (zpa->zpa_current_depth >= zpa->zpa_min_depth) {
2301 if (zpa->zpa_flags & DCMD_PIPE_OUT) {
2302 mdb_printf("%?p\n", addr);
2303 } else {
2304 mdb_printf("%*s%-*p %-5s %-16s ", indent, "",
2305 ZIO_MAXWIDTH - indent, addr, type, stage);
2306 if (zio.io_waiter != 0)
2307 mdb_printf("%-16lx ", zio.io_waiter);
2308 else
2309 mdb_printf("%-16s ", "-");
2310 #ifdef _KERNEL
2311 if (zio_timestamp.io_timestamp != 0) {
2312 mdb_printf("%llums", (mdb_gethrtime() -
2313 zio_timestamp.io_timestamp) /
2314 1000000);
2315 } else {
2316 mdb_printf("%-12s ", "-");
2317 }
2318 #else
2319 mdb_printf("%-12s ", "-");
2320 #endif
2321 mdb_printf("\n");
2322 }
2323 }
2324
2325 if (zpa->zpa_current_depth >= zpa->zpa_max_depth)
2326 return (WALK_NEXT);
2327
2328 if (zpa->zpa_type == ZIO_WALK_PARENT)
2329 laddr = addr + mdb_ctf_offsetof_by_name(ZFS_STRUCT "zio",
2330 "io_parent_list");
2331 else
2332 laddr = addr + mdb_ctf_offsetof_by_name(ZFS_STRUCT "zio",
2333 "io_child_list");
2334
2335 zpa->zpa_current_depth++;
2336 if (mdb_pwalk("list", zio_child_cb, zpa, laddr) != 0) {
2337 mdb_warn("failed to walk zio_t children at %p\n", laddr);
2338 return (WALK_ERR);
2339 }
2340 zpa->zpa_current_depth--;
2341
2342 return (WALK_NEXT);
2343 }
2344
2345 /* ARGSUSED */
2346 static int
2347 zio_child_cb(uintptr_t addr, const void *unknown, void *arg)
2348 {
2349 zio_link_t zl;
2350 uintptr_t ziop;
2351 zio_print_args_t *zpa = arg;
2352
2353 if (mdb_vread(&zl, sizeof (zl), addr) == -1) {
2354 mdb_warn("failed to read zio_link_t at %p", addr);
2355 return (WALK_ERR);
2356 }
2357
2358 if (zpa->zpa_type == ZIO_WALK_PARENT)
2359 ziop = (uintptr_t)zl.zl_parent;
2360 else
2361 ziop = (uintptr_t)zl.zl_child;
2362
2363 return (zio_print_cb(ziop, zpa));
2364 }
2365
2366 /* ARGSUSED */
2367 static int
2368 zio_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2369 {
2370 zio_print_args_t zpa = { 0 };
2371
2372 if (!(flags & DCMD_ADDRSPEC))
2373 return (DCMD_USAGE);
2374
2375 if (mdb_getopts(argc, argv,
2376 'r', MDB_OPT_SETBITS, INT_MAX, &zpa.zpa_max_depth,
2377 'c', MDB_OPT_SETBITS, ZIO_WALK_CHILD, &zpa.zpa_type,
2378 'p', MDB_OPT_SETBITS, ZIO_WALK_PARENT, &zpa.zpa_type,
2379 NULL) != argc)
2380 return (DCMD_USAGE);
2381
2382 zpa.zpa_flags = flags;
2383 if (zpa.zpa_max_depth != 0) {
2384 if (zpa.zpa_type == ZIO_WALK_SELF)
2385 zpa.zpa_type = ZIO_WALK_CHILD;
2386 } else if (zpa.zpa_type != ZIO_WALK_SELF) {
2387 zpa.zpa_min_depth = 1;
2388 zpa.zpa_max_depth = 1;
2389 }
2390
2391 if (!(flags & DCMD_PIPE_OUT) && DCMD_HDRSPEC(flags)) {
2392 mdb_printf("%<u>%-*s %-5s %-16s %-16s %-12s%</u>\n",
2393 ZIO_MAXWIDTH, "ADDRESS", "TYPE", "STAGE", "WAITER",
2394 "TIME_ELAPSED");
2395 }
2396
2397 if (zio_print_cb(addr, &zpa) != WALK_NEXT)
2398 return (DCMD_ERR);
2399
2400 return (DCMD_OK);
2401 }
2402
2403 /*
2404 * [addr]::zio_state
2405 *
2406 * Print a summary of all zio_t structures on the system, or for a particular
2407 * pool. This is equivalent to '::walk zio_root | ::zio'.
2408 */
2409 /*ARGSUSED*/
2410 static int
2411 zio_state(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2412 {
2413 /*
2414 * MDB will remember the last address of the pipeline, so if we don't
2415 * zero this we'll end up trying to walk zio structures for a
2416 * non-existent spa_t.
2417 */
2418 if (!(flags & DCMD_ADDRSPEC))
2419 addr = 0;
2420
2421 return (mdb_pwalk_dcmd("zio_root", "zio", argc, argv, addr));
2422 }
2423
2424 typedef struct mdb_multilist {
2425 uint64_t ml_num_sublists;
2426 uintptr_t ml_sublists;
2427 } mdb_multilist_t;
2428
2429 typedef struct multilist_walk_data {
2430 uint64_t mwd_idx;
2431 mdb_multilist_t mwd_ml;
2432 } multilist_walk_data_t;
2433
2434 /* ARGSUSED */
2435 static int
2436 multilist_print_cb(uintptr_t addr, const void *unknown, void *arg)
2437 {
2438 mdb_printf("%#lr\n", addr);
2439 return (WALK_NEXT);
2440 }
2441
2442 static int
2443 multilist_walk_step(mdb_walk_state_t *wsp)
2444 {
2445 multilist_walk_data_t *mwd = wsp->walk_data;
2446
2447 if (mwd->mwd_idx >= mwd->mwd_ml.ml_num_sublists)
2448 return (WALK_DONE);
2449
2450 wsp->walk_addr = mwd->mwd_ml.ml_sublists +
2451 mdb_ctf_sizeof_by_name("multilist_sublist_t") * mwd->mwd_idx +
2452 mdb_ctf_offsetof_by_name("multilist_sublist_t", "mls_list");
2453
2454 mdb_pwalk("list", multilist_print_cb, NULL, wsp->walk_addr);
2455 mwd->mwd_idx++;
2456
2457 return (WALK_NEXT);
2458 }
2459
2460 static int
2461 multilist_walk_init(mdb_walk_state_t *wsp)
2462 {
2463 multilist_walk_data_t *mwd;
2464
2465 if (wsp->walk_addr == 0) {
2466 mdb_warn("must supply address of multilist_t\n");
2467 return (WALK_ERR);
2468 }
2469
2470 mwd = mdb_zalloc(sizeof (multilist_walk_data_t), UM_SLEEP | UM_GC);
2471 if (mdb_ctf_vread(&mwd->mwd_ml, "multilist_t", "mdb_multilist_t",
2472 wsp->walk_addr, 0) == -1) {
2473 return (WALK_ERR);
2474 }
2475
2476 if (mwd->mwd_ml.ml_num_sublists == 0 ||
2477 mwd->mwd_ml.ml_sublists == 0) {
2478 mdb_warn("invalid or uninitialized multilist at %#lx\n",
2479 wsp->walk_addr);
2480 return (WALK_ERR);
2481 }
2482
2483 wsp->walk_data = mwd;
2484 return (WALK_NEXT);
2485 }
2486
2487 typedef struct mdb_txg_list {
2488 size_t tl_offset;
2489 uintptr_t tl_head[TXG_SIZE];
2490 } mdb_txg_list_t;
2491
2492 typedef struct txg_list_walk_data {
2493 uintptr_t lw_head[TXG_SIZE];
2494 int lw_txgoff;
2495 int lw_maxoff;
2496 size_t lw_offset;
2497 void *lw_obj;
2498 } txg_list_walk_data_t;
2499
2500 static int
2501 txg_list_walk_init_common(mdb_walk_state_t *wsp, int txg, int maxoff)
2502 {
2503 txg_list_walk_data_t *lwd;
2504 mdb_txg_list_t list;
2505 int i;
2506
2507 lwd = mdb_alloc(sizeof (txg_list_walk_data_t), UM_SLEEP | UM_GC);
2508 if (mdb_ctf_vread(&list, "txg_list_t", "mdb_txg_list_t", wsp->walk_addr,
2509 0) == -1) {
2510 mdb_warn("failed to read txg_list_t at %#lx", wsp->walk_addr);
2511 return (WALK_ERR);
2512 }
2513
2514 for (i = 0; i < TXG_SIZE; i++)
2515 lwd->lw_head[i] = list.tl_head[i];
2516 lwd->lw_offset = list.tl_offset;
2517 lwd->lw_obj = mdb_alloc(lwd->lw_offset + sizeof (txg_node_t),
2518 UM_SLEEP | UM_GC);
2519 lwd->lw_txgoff = txg;
2520 lwd->lw_maxoff = maxoff;
2521
2522 wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
2523 wsp->walk_data = lwd;
2524
2525 return (WALK_NEXT);
2526 }
2527
2528 static int
2529 txg_list_walk_init(mdb_walk_state_t *wsp)
2530 {
2531 return (txg_list_walk_init_common(wsp, 0, TXG_SIZE-1));
2532 }
2533
2534 static int
2535 txg_list0_walk_init(mdb_walk_state_t *wsp)
2536 {
2537 return (txg_list_walk_init_common(wsp, 0, 0));
2538 }
2539
2540 static int
2541 txg_list1_walk_init(mdb_walk_state_t *wsp)
2542 {
2543 return (txg_list_walk_init_common(wsp, 1, 1));
2544 }
2545
2546 static int
2547 txg_list2_walk_init(mdb_walk_state_t *wsp)
2548 {
2549 return (txg_list_walk_init_common(wsp, 2, 2));
2550 }
2551
2552 static int
2553 txg_list3_walk_init(mdb_walk_state_t *wsp)
2554 {
2555 return (txg_list_walk_init_common(wsp, 3, 3));
2556 }
2557
2558 static int
2559 txg_list_walk_step(mdb_walk_state_t *wsp)
2560 {
2561 txg_list_walk_data_t *lwd = wsp->walk_data;
2562 uintptr_t addr;
2563 txg_node_t *node;
2564 int status;
2565
2566 while (wsp->walk_addr == (uintptr_t)NULL &&
2567 lwd->lw_txgoff < lwd->lw_maxoff) {
2568 lwd->lw_txgoff++;
2569 wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
2570 }
2571
2572 if (wsp->walk_addr == (uintptr_t)NULL)
2573 return (WALK_DONE);
2574
2575 addr = wsp->walk_addr - lwd->lw_offset;
2576
2577 if (mdb_vread(lwd->lw_obj,
2578 lwd->lw_offset + sizeof (txg_node_t), addr) == -1) {
2579 mdb_warn("failed to read list element at %#lx", addr);
2580 return (WALK_ERR);
2581 }
2582
2583 status = wsp->walk_callback(addr, lwd->lw_obj, wsp->walk_cbdata);
2584 node = (txg_node_t *)((uintptr_t)lwd->lw_obj + lwd->lw_offset);
2585 wsp->walk_addr = (uintptr_t)node->tn_next[lwd->lw_txgoff];
2586
2587 return (status);
2588 }
2589
2590 /*
2591 * ::walk spa
2592 *
2593 * Walk all named spa_t structures in the namespace. This is nothing more than
2594 * a layered avl walk.
2595 */
2596 static int
2597 spa_walk_init(mdb_walk_state_t *wsp)
2598 {
2599 GElf_Sym sym;
2600
2601 if (wsp->walk_addr != (uintptr_t)NULL) {
2602 mdb_warn("spa walk only supports global walks\n");
2603 return (WALK_ERR);
2604 }
2605
2606 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "spa_namespace_avl", &sym) == -1) {
2607 mdb_warn("failed to find symbol 'spa_namespace_avl'");
2608 return (WALK_ERR);
2609 }
2610
2611 wsp->walk_addr = (uintptr_t)sym.st_value;
2612
2613 if (mdb_layered_walk("avl", wsp) == -1) {
2614 mdb_warn("failed to walk 'avl'\n");
2615 return (WALK_ERR);
2616 }
2617
2618 return (WALK_NEXT);
2619 }
2620
2621 static int
2622 spa_walk_step(mdb_walk_state_t *wsp)
2623 {
2624 return (wsp->walk_callback(wsp->walk_addr, NULL, wsp->walk_cbdata));
2625 }
2626
2627 /*
2628 * [addr]::walk zio
2629 *
2630 * Walk all active zio_t structures on the system. This is simply a layered
2631 * walk on top of ::walk zio_cache, with the optional ability to limit the
2632 * structures to a particular pool.
2633 */
2634 static int
2635 zio_walk_init(mdb_walk_state_t *wsp)
2636 {
2637 wsp->walk_data = (void *)wsp->walk_addr;
2638
2639 if (mdb_layered_walk("zio_cache", wsp) == -1) {
2640 mdb_warn("failed to walk 'zio_cache'\n");
2641 return (WALK_ERR);
2642 }
2643
2644 return (WALK_NEXT);
2645 }
2646
2647 static int
2648 zio_walk_step(mdb_walk_state_t *wsp)
2649 {
2650 mdb_zio_t zio;
2651 uintptr_t spa = (uintptr_t)wsp->walk_data;
2652
2653 if (mdb_ctf_vread(&zio, ZFS_STRUCT "zio", "mdb_zio_t",
2654 wsp->walk_addr, 0) == -1)
2655 return (WALK_ERR);
2656
2657 if (spa != 0 && spa != zio.io_spa)
2658 return (WALK_NEXT);
2659
2660 return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
2661 }
2662
2663 /*
2664 * [addr]::walk zio_root
2665 *
2666 * Walk only root zio_t structures, optionally for a particular spa_t.
2667 */
2668 static int
2669 zio_walk_root_step(mdb_walk_state_t *wsp)
2670 {
2671 mdb_zio_t zio;
2672 uintptr_t spa = (uintptr_t)wsp->walk_data;
2673
2674 if (mdb_ctf_vread(&zio, ZFS_STRUCT "zio", "mdb_zio_t",
2675 wsp->walk_addr, 0) == -1)
2676 return (WALK_ERR);
2677
2678 if (spa != 0 && spa != zio.io_spa)
2679 return (WALK_NEXT);
2680
2681 /* If the parent list is not empty, ignore */
2682 if (zio.io_parent_list.list_head.list_next !=
2683 wsp->walk_addr +
2684 mdb_ctf_offsetof_by_name(ZFS_STRUCT "zio", "io_parent_list") +
2685 mdb_ctf_offsetof_by_name("struct list", "list_head"))
2686 return (WALK_NEXT);
2687
2688 return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
2689 }
2690
2691 /*
2692 * ::zfs_blkstats
2693 *
2694 * -v print verbose per-level information
2695 *
2696 */
2697 static int
2698 zfs_blkstats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2699 {
2700 boolean_t verbose = B_FALSE;
2701 zfs_all_blkstats_t stats;
2702 dmu_object_type_t t;
2703 zfs_blkstat_t *tzb;
2704 uint64_t ditto;
2705 dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES + 10];
2706 /* +10 in case it grew */
2707
2708 if (mdb_readvar(&dmu_ot, "dmu_ot") == -1) {
2709 mdb_warn("failed to read 'dmu_ot'");
2710 return (DCMD_ERR);
2711 }
2712
2713 if (mdb_getopts(argc, argv,
2714 'v', MDB_OPT_SETBITS, TRUE, &verbose,
2715 NULL) != argc)
2716 return (DCMD_USAGE);
2717
2718 if (!(flags & DCMD_ADDRSPEC))
2719 return (DCMD_USAGE);
2720
2721 if (GETMEMB(addr, "spa", spa_dsl_pool, addr) ||
2722 GETMEMB(addr, "dsl_pool", dp_blkstats, addr) ||
2723 mdb_vread(&stats, sizeof (zfs_all_blkstats_t), addr) == -1) {
2724 mdb_warn("failed to read data at %p;", addr);
2725 mdb_printf("maybe no stats? run \"zpool scrub\" first.");
2726 return (DCMD_ERR);
2727 }
2728
2729 tzb = &stats.zab_type[DN_MAX_LEVELS][DMU_OT_TOTAL];
2730 if (tzb->zb_gangs != 0) {
2731 mdb_printf("Ganged blocks: %llu\n",
2732 (longlong_t)tzb->zb_gangs);
2733 }
2734
2735 ditto = tzb->zb_ditto_2_of_2_samevdev + tzb->zb_ditto_2_of_3_samevdev +
2736 tzb->zb_ditto_3_of_3_samevdev;
2737 if (ditto != 0) {
2738 mdb_printf("Dittoed blocks on same vdev: %llu\n",
2739 (longlong_t)ditto);
2740 }
2741
2742 mdb_printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
2743 "\t avg\t comp\t%%Total\tType\n");
2744
2745 for (t = 0; t <= DMU_OT_TOTAL; t++) {
2746 char csize[NICENUM_BUFLEN], lsize[NICENUM_BUFLEN];
2747 char psize[NICENUM_BUFLEN], asize[NICENUM_BUFLEN];
2748 char avg[NICENUM_BUFLEN];
2749 char comp[NICENUM_BUFLEN], pct[NICENUM_BUFLEN];
2750 char typename[64];
2751 int l;
2752
2753
2754 if (t == DMU_OT_DEFERRED)
2755 strcpy(typename, "deferred free");
2756 else if (t == DMU_OT_OTHER)
2757 strcpy(typename, "other");
2758 else if (t == DMU_OT_TOTAL)
2759 strcpy(typename, "Total");
2760 else if (mdb_readstr(typename, sizeof (typename),
2761 (uintptr_t)dmu_ot[t].ot_name) == -1) {
2762 mdb_warn("failed to read type name");
2763 return (DCMD_ERR);
2764 }
2765
2766 if (stats.zab_type[DN_MAX_LEVELS][t].zb_asize == 0)
2767 continue;
2768
2769 for (l = -1; l < DN_MAX_LEVELS; l++) {
2770 int level = (l == -1 ? DN_MAX_LEVELS : l);
2771 zfs_blkstat_t *zb = &stats.zab_type[level][t];
2772
2773 if (zb->zb_asize == 0)
2774 continue;
2775
2776 /*
2777 * Don't print each level unless requested.
2778 */
2779 if (!verbose && level != DN_MAX_LEVELS)
2780 continue;
2781
2782 /*
2783 * If all the space is level 0, don't print the
2784 * level 0 separately.
2785 */
2786 if (level == 0 && zb->zb_asize ==
2787 stats.zab_type[DN_MAX_LEVELS][t].zb_asize)
2788 continue;
2789
2790 mdb_nicenum(zb->zb_count, csize);
2791 mdb_nicenum(zb->zb_lsize, lsize);
2792 mdb_nicenum(zb->zb_psize, psize);
2793 mdb_nicenum(zb->zb_asize, asize);
2794 mdb_nicenum(zb->zb_asize / zb->zb_count, avg);
2795 (void) snprintfrac(comp, NICENUM_BUFLEN,
2796 zb->zb_lsize, zb->zb_psize, 2);
2797 (void) snprintfrac(pct, NICENUM_BUFLEN,
2798 100 * zb->zb_asize, tzb->zb_asize, 2);
2799
2800 mdb_printf("%6s\t%5s\t%5s\t%5s\t%5s"
2801 "\t%5s\t%6s\t",
2802 csize, lsize, psize, asize, avg, comp, pct);
2803
2804 if (level == DN_MAX_LEVELS)
2805 mdb_printf("%s\n", typename);
2806 else
2807 mdb_printf(" L%d %s\n",
2808 level, typename);
2809 }
2810 }
2811
2812 return (DCMD_OK);
2813 }
2814
2815 typedef struct mdb_reference {
2816 uintptr_t ref_holder;
2817 uintptr_t ref_removed;
2818 uint64_t ref_number;
2819 } mdb_reference_t;
2820
2821 /* ARGSUSED */
2822 static int
2823 reference_cb(uintptr_t addr, const void *ignored, void *arg)
2824 {
2825 mdb_reference_t ref;
2826 boolean_t holder_is_str = B_FALSE;
2827 char holder_str[128];
2828 boolean_t removed = (boolean_t)arg;
2829
2830 if (mdb_ctf_vread(&ref, "reference_t", "mdb_reference_t", addr,
2831 0) == -1)
2832 return (DCMD_ERR);
2833
2834 if (mdb_readstr(holder_str, sizeof (holder_str),
2835 ref.ref_holder) != -1)
2836 holder_is_str = strisprint(holder_str);
2837
2838 if (removed)
2839 mdb_printf("removed ");
2840 mdb_printf("reference ");
2841 if (ref.ref_number != 1)
2842 mdb_printf("with count=%llu ", ref.ref_number);
2843 mdb_printf("with tag %lx", ref.ref_holder);
2844 if (holder_is_str)
2845 mdb_printf(" \"%s\"", holder_str);
2846 mdb_printf(", held at:\n");
2847
2848 (void) mdb_call_dcmd("whatis", addr, DCMD_ADDRSPEC, 0, NULL);
2849
2850 if (removed) {
2851 mdb_printf("removed at:\n");
2852 (void) mdb_call_dcmd("whatis", ref.ref_removed,
2853 DCMD_ADDRSPEC, 0, NULL);
2854 }
2855
2856 mdb_printf("\n");
2857
2858 return (WALK_NEXT);
2859 }
2860
2861 typedef struct mdb_refcount {
2862 uint64_t rc_count;
2863 } mdb_refcount_t;
2864
2865 typedef struct mdb_refcount_removed {
2866 uint64_t rc_removed_count;
2867 } mdb_refcount_removed_t;
2868
2869 typedef struct mdb_refcount_tracked {
2870 boolean_t rc_tracked;
2871 } mdb_refcount_tracked_t;
2872
2873 /* ARGSUSED */
2874 static int
2875 refcount(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2876 {
2877 mdb_refcount_t rc;
2878 mdb_refcount_removed_t rcr;
2879 mdb_refcount_tracked_t rct;
2880 int off;
2881 boolean_t released = B_FALSE;
2882
2883 if (!(flags & DCMD_ADDRSPEC))
2884 return (DCMD_USAGE);
2885
2886 if (mdb_getopts(argc, argv,
2887 'r', MDB_OPT_SETBITS, B_TRUE, &released,
2888 NULL) != argc)
2889 return (DCMD_USAGE);
2890
2891 if (mdb_ctf_vread(&rc, "refcount_t", "mdb_refcount_t", addr,
2892 0) == -1)
2893 return (DCMD_ERR);
2894
2895 if (mdb_ctf_vread(&rcr, "refcount_t", "mdb_refcount_removed_t", addr,
2896 MDB_CTF_VREAD_QUIET) == -1) {
2897 mdb_printf("refcount_t at %p has %llu holds (untracked)\n",
2898 addr, (longlong_t)rc.rc_count);
2899 return (DCMD_OK);
2900 }
2901
2902 if (mdb_ctf_vread(&rct, "refcount_t", "mdb_refcount_tracked_t", addr,
2903 MDB_CTF_VREAD_QUIET) == -1) {
2904 /* If this is an old target, it might be tracked. */
2905 rct.rc_tracked = B_TRUE;
2906 }
2907
2908 mdb_printf("refcount_t at %p has %llu current holds, "
2909 "%llu recently released holds\n",
2910 addr, (longlong_t)rc.rc_count, (longlong_t)rcr.rc_removed_count);
2911
2912 if (rct.rc_tracked && rc.rc_count > 0)
2913 mdb_printf("current holds:\n");
2914 off = mdb_ctf_offsetof_by_name("refcount_t", "rc_list");
2915 if (off == -1)
2916 return (DCMD_ERR);
2917 mdb_pwalk("list", reference_cb, (void*)B_FALSE, addr + off);
2918
2919 if (released && rcr.rc_removed_count > 0) {
2920 mdb_printf("released holds:\n");
2921
2922 off = mdb_ctf_offsetof_by_name("refcount_t", "rc_removed");
2923 if (off == -1)
2924 return (DCMD_ERR);
2925 mdb_pwalk("list", reference_cb, (void*)B_TRUE, addr + off);
2926 }
2927
2928 return (DCMD_OK);
2929 }
2930
2931 /* ARGSUSED */
2932 static int
2933 sa_attr_table(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2934 {
2935 sa_attr_table_t *table;
2936 sa_os_t sa_os;
2937 char *name;
2938 int i;
2939
2940 if (mdb_vread(&sa_os, sizeof (sa_os_t), addr) == -1) {
2941 mdb_warn("failed to read sa_os at %p", addr);
2942 return (DCMD_ERR);
2943 }
2944
2945 table = mdb_alloc(sizeof (sa_attr_table_t) * sa_os.sa_num_attrs,
2946 UM_SLEEP | UM_GC);
2947 name = mdb_alloc(MAXPATHLEN, UM_SLEEP | UM_GC);
2948
2949 if (mdb_vread(table, sizeof (sa_attr_table_t) * sa_os.sa_num_attrs,
2950 (uintptr_t)sa_os.sa_attr_table) == -1) {
2951 mdb_warn("failed to read sa_os at %p", addr);
2952 return (DCMD_ERR);
2953 }
2954
2955 mdb_printf("%<u>%-10s %-10s %-10s %-10s %s%</u>\n",
2956 "ATTR ID", "REGISTERED", "LENGTH", "BSWAP", "NAME");
2957 for (i = 0; i != sa_os.sa_num_attrs; i++) {
2958 mdb_readstr(name, MAXPATHLEN, (uintptr_t)table[i].sa_name);
2959 mdb_printf("%5x %8x %8x %8x %-s\n",
2960 (int)table[i].sa_attr, (int)table[i].sa_registered,
2961 (int)table[i].sa_length, table[i].sa_byteswap, name);
2962 }
2963
2964 return (DCMD_OK);
2965 }
2966
2967 static int
2968 sa_get_off_table(uintptr_t addr, uint32_t **off_tab, int attr_count)
2969 {
2970 uintptr_t idx_table;
2971
2972 if (GETMEMB(addr, "sa_idx_tab", sa_idx_tab, idx_table)) {
2973 mdb_printf("can't find offset table in sa_idx_tab\n");
2974 return (-1);
2975 }
2976
2977 *off_tab = mdb_alloc(attr_count * sizeof (uint32_t),
2978 UM_SLEEP | UM_GC);
2979
2980 if (mdb_vread(*off_tab,
2981 attr_count * sizeof (uint32_t), idx_table) == -1) {
2982 mdb_warn("failed to attribute offset table %p", idx_table);
2983 return (-1);
2984 }
2985
2986 return (DCMD_OK);
2987 }
2988
2989 /*ARGSUSED*/
2990 static int
2991 sa_attr_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2992 {
2993 uint32_t *offset_tab;
2994 int attr_count;
2995 uint64_t attr_id;
2996 uintptr_t attr_addr;
2997 uintptr_t bonus_tab, spill_tab;
2998 uintptr_t db_bonus, db_spill;
2999 uintptr_t os, os_sa;
3000 uintptr_t db_data;
3001
3002 if (argc != 1)
3003 return (DCMD_USAGE);
3004
3005 if (argv[0].a_type == MDB_TYPE_STRING)
3006 attr_id = mdb_strtoull(argv[0].a_un.a_str);
3007 else
3008 return (DCMD_USAGE);
3009
3010 if (GETMEMB(addr, "sa_handle", sa_bonus_tab, bonus_tab) ||
3011 GETMEMB(addr, "sa_handle", sa_spill_tab, spill_tab) ||
3012 GETMEMB(addr, "sa_handle", sa_os, os) ||
3013 GETMEMB(addr, "sa_handle", sa_bonus, db_bonus) ||
3014 GETMEMB(addr, "sa_handle", sa_spill, db_spill)) {
3015 mdb_printf("Can't find necessary information in sa_handle "
3016 "in sa_handle\n");
3017 return (DCMD_ERR);
3018 }
3019
3020 if (GETMEMB(os, "objset", os_sa, os_sa)) {
3021 mdb_printf("Can't find os_sa in objset\n");
3022 return (DCMD_ERR);
3023 }
3024
3025 if (GETMEMB(os_sa, "sa_os", sa_num_attrs, attr_count)) {
3026 mdb_printf("Can't find sa_num_attrs\n");
3027 return (DCMD_ERR);
3028 }
3029
3030 if (attr_id > attr_count) {
3031 mdb_printf("attribute id number is out of range\n");
3032 return (DCMD_ERR);
3033 }
3034
3035 if (bonus_tab) {
3036 if (sa_get_off_table(bonus_tab, &offset_tab,
3037 attr_count) == -1) {
3038 return (DCMD_ERR);
3039 }
3040
3041 if (GETMEMB(db_bonus, "dmu_buf", db_data, db_data)) {
3042 mdb_printf("can't find db_data in bonus dbuf\n");
3043 return (DCMD_ERR);
3044 }
3045 }
3046
3047 if (bonus_tab && !TOC_ATTR_PRESENT(offset_tab[attr_id]) &&
3048 spill_tab == (uintptr_t)NULL) {
3049 mdb_printf("Attribute does not exist\n");
3050 return (DCMD_ERR);
3051 } else if (!TOC_ATTR_PRESENT(offset_tab[attr_id]) && spill_tab) {
3052 if (sa_get_off_table(spill_tab, &offset_tab,
3053 attr_count) == -1) {
3054 return (DCMD_ERR);
3055 }
3056 if (GETMEMB(db_spill, "dmu_buf", db_data, db_data)) {
3057 mdb_printf("can't find db_data in spill dbuf\n");
3058 return (DCMD_ERR);
3059 }
3060 if (!TOC_ATTR_PRESENT(offset_tab[attr_id])) {
3061 mdb_printf("Attribute does not exist\n");
3062 return (DCMD_ERR);
3063 }
3064 }
3065 attr_addr = db_data + TOC_OFF(offset_tab[attr_id]);
3066 mdb_printf("%p\n", attr_addr);
3067 return (DCMD_OK);
3068 }
3069
3070 /* ARGSUSED */
3071 static int
3072 zfs_ace_print_common(uintptr_t addr, uint_t flags,
3073 uint64_t id, uint32_t access_mask, uint16_t ace_flags,
3074 uint16_t ace_type, int verbose)
3075 {
3076 if (DCMD_HDRSPEC(flags) && !verbose)
3077 mdb_printf("%<u>%-?s %-8s %-8s %-8s %s%</u>\n",
3078 "ADDR", "FLAGS", "MASK", "TYPE", "ID");
3079
3080 if (!verbose) {
3081 mdb_printf("%0?p %-8x %-8x %-8x %-llx\n", addr,
3082 ace_flags, access_mask, ace_type, id);
3083 return (DCMD_OK);
3084 }
3085
3086 switch (ace_flags & ACE_TYPE_FLAGS) {
3087 case ACE_OWNER:
3088 mdb_printf("owner@:");
3089 break;
3090 case (ACE_IDENTIFIER_GROUP | ACE_GROUP):
3091 mdb_printf("group@:");
3092 break;
3093 case ACE_EVERYONE:
3094 mdb_printf("everyone@:");
3095 break;
3096 case ACE_IDENTIFIER_GROUP:
3097 mdb_printf("group:%llx:", (u_longlong_t)id);
3098 break;
3099 case 0: /* User entry */
3100 mdb_printf("user:%llx:", (u_longlong_t)id);
3101 break;
3102 }
3103
3104 /* print out permission mask */
3105 if (access_mask & ACE_READ_DATA)
3106 mdb_printf("r");
3107 else
3108 mdb_printf("-");
3109 if (access_mask & ACE_WRITE_DATA)
3110 mdb_printf("w");
3111 else
3112 mdb_printf("-");
3113 if (access_mask & ACE_EXECUTE)
3114 mdb_printf("x");
3115 else
3116 mdb_printf("-");
3117 if (access_mask & ACE_APPEND_DATA)
3118 mdb_printf("p");
3119 else
3120 mdb_printf("-");
3121 if (access_mask & ACE_DELETE)
3122 mdb_printf("d");
3123 else
3124 mdb_printf("-");
3125 if (access_mask & ACE_DELETE_CHILD)
3126 mdb_printf("D");
3127 else
3128 mdb_printf("-");
3129 if (access_mask & ACE_READ_ATTRIBUTES)
3130 mdb_printf("a");
3131 else
3132 mdb_printf("-");
3133 if (access_mask & ACE_WRITE_ATTRIBUTES)
3134 mdb_printf("A");
3135 else
3136 mdb_printf("-");
3137 if (access_mask & ACE_READ_NAMED_ATTRS)
3138 mdb_printf("R");
3139 else
3140 mdb_printf("-");
3141 if (access_mask & ACE_WRITE_NAMED_ATTRS)
3142 mdb_printf("W");
3143 else
3144 mdb_printf("-");
3145 if (access_mask & ACE_READ_ACL)
3146 mdb_printf("c");
3147 else
3148 mdb_printf("-");
3149 if (access_mask & ACE_WRITE_ACL)
3150 mdb_printf("C");
3151 else
3152 mdb_printf("-");
3153 if (access_mask & ACE_WRITE_OWNER)
3154 mdb_printf("o");
3155 else
3156 mdb_printf("-");
3157 if (access_mask & ACE_SYNCHRONIZE)
3158 mdb_printf("s");
3159 else
3160 mdb_printf("-");
3161
3162 mdb_printf(":");
3163
3164 /* Print out inheritance flags */
3165 if (ace_flags & ACE_FILE_INHERIT_ACE)
3166 mdb_printf("f");
3167 else
3168 mdb_printf("-");
3169 if (ace_flags & ACE_DIRECTORY_INHERIT_ACE)
3170 mdb_printf("d");
3171 else
3172 mdb_printf("-");
3173 if (ace_flags & ACE_INHERIT_ONLY_ACE)
3174 mdb_printf("i");
3175 else
3176 mdb_printf("-");
3177 if (ace_flags & ACE_NO_PROPAGATE_INHERIT_ACE)
3178 mdb_printf("n");
3179 else
3180 mdb_printf("-");
3181 if (ace_flags & ACE_SUCCESSFUL_ACCESS_ACE_FLAG)
3182 mdb_printf("S");
3183 else
3184 mdb_printf("-");
3185 if (ace_flags & ACE_FAILED_ACCESS_ACE_FLAG)
3186 mdb_printf("F");
3187 else
3188 mdb_printf("-");
3189 if (ace_flags & ACE_INHERITED_ACE)
3190 mdb_printf("I");
3191 else
3192 mdb_printf("-");
3193
3194 switch (ace_type) {
3195 case ACE_ACCESS_ALLOWED_ACE_TYPE:
3196 mdb_printf(":allow\n");
3197 break;
3198 case ACE_ACCESS_DENIED_ACE_TYPE:
3199 mdb_printf(":deny\n");
3200 break;
3201 case ACE_SYSTEM_AUDIT_ACE_TYPE:
3202 mdb_printf(":audit\n");
3203 break;
3204 case ACE_SYSTEM_ALARM_ACE_TYPE:
3205 mdb_printf(":alarm\n");
3206 break;
3207 default:
3208 mdb_printf(":?\n");
3209 }
3210 return (DCMD_OK);
3211 }
3212
3213 /* ARGSUSED */
3214 static int
3215 zfs_ace_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3216 {
3217 zfs_ace_t zace;
3218 int verbose = FALSE;
3219 uint64_t id;
3220
3221 if (!(flags & DCMD_ADDRSPEC))
3222 return (DCMD_USAGE);
3223
3224 if (mdb_getopts(argc, argv,
3225 'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
3226 return (DCMD_USAGE);
3227
3228 if (mdb_vread(&zace, sizeof (zfs_ace_t), addr) == -1) {
3229 mdb_warn("failed to read zfs_ace_t");
3230 return (DCMD_ERR);
3231 }
3232
3233 if ((zace.z_hdr.z_flags & ACE_TYPE_FLAGS) == 0 ||
3234 (zace.z_hdr.z_flags & ACE_TYPE_FLAGS) == ACE_IDENTIFIER_GROUP)
3235 id = zace.z_fuid;
3236 else
3237 id = -1;
3238
3239 return (zfs_ace_print_common(addr, flags, id, zace.z_hdr.z_access_mask,
3240 zace.z_hdr.z_flags, zace.z_hdr.z_type, verbose));
3241 }
3242
3243 /* ARGSUSED */
3244 static int
3245 zfs_ace0_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3246 {
3247 ace_t ace;
3248 uint64_t id;
3249 int verbose = FALSE;
3250
3251 if (!(flags & DCMD_ADDRSPEC))
3252 return (DCMD_USAGE);
3253
3254 if (mdb_getopts(argc, argv,
3255 'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
3256 return (DCMD_USAGE);
3257
3258 if (mdb_vread(&ace, sizeof (ace_t), addr) == -1) {
3259 mdb_warn("failed to read ace_t");
3260 return (DCMD_ERR);
3261 }
3262
3263 if ((ace.a_flags & ACE_TYPE_FLAGS) == 0 ||
3264 (ace.a_flags & ACE_TYPE_FLAGS) == ACE_IDENTIFIER_GROUP)
3265 id = ace.a_who;
3266 else
3267 id = -1;
3268
3269 return (zfs_ace_print_common(addr, flags, id, ace.a_access_mask,
3270 ace.a_flags, ace.a_type, verbose));
3271 }
3272
3273 typedef struct acl_dump_args {
3274 int a_argc;
3275 const mdb_arg_t *a_argv;
3276 uint16_t a_version;
3277 int a_flags;
3278 } acl_dump_args_t;
3279
3280 /* ARGSUSED */
3281 static int
3282 acl_aces_cb(uintptr_t addr, const void *unknown, void *arg)
3283 {
3284 acl_dump_args_t *acl_args = (acl_dump_args_t *)arg;
3285
3286 if (acl_args->a_version == 1) {
3287 if (mdb_call_dcmd("zfs_ace", addr,
3288 DCMD_ADDRSPEC|acl_args->a_flags, acl_args->a_argc,
3289 acl_args->a_argv) != DCMD_OK) {
3290 return (WALK_ERR);
3291 }
3292 } else {
3293 if (mdb_call_dcmd("zfs_ace0", addr,
3294 DCMD_ADDRSPEC|acl_args->a_flags, acl_args->a_argc,
3295 acl_args->a_argv) != DCMD_OK) {
3296 return (WALK_ERR);
3297 }
3298 }
3299 acl_args->a_flags = DCMD_LOOP;
3300 return (WALK_NEXT);
3301 }
3302
3303 /* ARGSUSED */
3304 static int
3305 acl_cb(uintptr_t addr, const void *unknown, void *arg)
3306 {
3307 acl_dump_args_t *acl_args = (acl_dump_args_t *)arg;
3308
3309 if (acl_args->a_version == 1) {
3310 if (mdb_pwalk("zfs_acl_node_aces", acl_aces_cb,
3311 arg, addr) != 0) {
3312 mdb_warn("can't walk ACEs");
3313 return (DCMD_ERR);
3314 }
3315 } else {
3316 if (mdb_pwalk("zfs_acl_node_aces0", acl_aces_cb,
3317 arg, addr) != 0) {
3318 mdb_warn("can't walk ACEs");
3319 return (DCMD_ERR);
3320 }
3321 }
3322 return (WALK_NEXT);
3323 }
3324
3325 /* ARGSUSED */
3326 static int
3327 zfs_acl_dump(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3328 {
3329 zfs_acl_t zacl;
3330 int verbose = FALSE;
3331 acl_dump_args_t acl_args;
3332
3333 if (!(flags & DCMD_ADDRSPEC))
3334 return (DCMD_USAGE);
3335
3336 if (mdb_getopts(argc, argv,
3337 'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
3338 return (DCMD_USAGE);
3339
3340 if (mdb_vread(&zacl, sizeof (zfs_acl_t), addr) == -1) {
3341 mdb_warn("failed to read zfs_acl_t");
3342 return (DCMD_ERR);
3343 }
3344
3345 acl_args.a_argc = argc;
3346 acl_args.a_argv = argv;
3347 acl_args.a_version = zacl.z_version;
3348 acl_args.a_flags = DCMD_LOOPFIRST;
3349
3350 if (mdb_pwalk("zfs_acl_node", acl_cb, &acl_args, addr) != 0) {
3351 mdb_warn("can't walk ACL");
3352 return (DCMD_ERR);
3353 }
3354
3355 return (DCMD_OK);
3356 }
3357
3358 /* ARGSUSED */
3359 static int
3360 zfs_acl_node_walk_init(mdb_walk_state_t *wsp)
3361 {
3362 if (wsp->walk_addr == (uintptr_t)NULL) {
3363 mdb_warn("must supply address of zfs_acl_node_t\n");
3364 return (WALK_ERR);
3365 }
3366
3367 wsp->walk_addr +=
3368 mdb_ctf_offsetof_by_name(ZFS_STRUCT "zfs_acl", "z_acl");
3369
3370 if (mdb_layered_walk("list", wsp) == -1) {
3371 mdb_warn("failed to walk 'list'\n");
3372 return (WALK_ERR);
3373 }
3374
3375 return (WALK_NEXT);
3376 }
3377
3378 static int
3379 zfs_acl_node_walk_step(mdb_walk_state_t *wsp)
3380 {
3381 zfs_acl_node_t aclnode;
3382
3383 if (mdb_vread(&aclnode, sizeof (zfs_acl_node_t),
3384 wsp->walk_addr) == -1) {
3385 mdb_warn("failed to read zfs_acl_node at %p", wsp->walk_addr);
3386 return (WALK_ERR);
3387 }
3388
3389 return (wsp->walk_callback(wsp->walk_addr, &aclnode, wsp->walk_cbdata));
3390 }
3391
3392 typedef struct ace_walk_data {
3393 int ace_count;
3394 int ace_version;
3395 } ace_walk_data_t;
3396
3397 static int
3398 zfs_aces_walk_init_common(mdb_walk_state_t *wsp, int version,
3399 int ace_count, uintptr_t ace_data)
3400 {
3401 ace_walk_data_t *ace_walk_data;
3402
3403 if (wsp->walk_addr == (uintptr_t)NULL) {
3404 mdb_warn("must supply address of zfs_acl_node_t\n");
3405 return (WALK_ERR);
3406 }
3407
3408 ace_walk_data = mdb_alloc(sizeof (ace_walk_data_t), UM_SLEEP | UM_GC);
3409
3410 ace_walk_data->ace_count = ace_count;
3411 ace_walk_data->ace_version = version;
3412
3413 wsp->walk_addr = ace_data;
3414 wsp->walk_data = ace_walk_data;
3415
3416 return (WALK_NEXT);
3417 }
3418
3419 static int
3420 zfs_acl_node_aces_walk_init_common(mdb_walk_state_t *wsp, int version)
3421 {
3422 static int gotid;
3423 static mdb_ctf_id_t acl_id;
3424 int z_ace_count;
3425 uintptr_t z_acldata;
3426
3427 if (!gotid) {
3428 if (mdb_ctf_lookup_by_name("struct zfs_acl_node",
3429 &acl_id) == -1) {
3430 mdb_warn("couldn't find struct zfs_acl_node");
3431 return (DCMD_ERR);
3432 }
3433 gotid = TRUE;
3434 }
3435
3436 if (GETMEMBID(wsp->walk_addr, &acl_id, z_ace_count, z_ace_count)) {
3437 return (DCMD_ERR);
3438 }
3439 if (GETMEMBID(wsp->walk_addr, &acl_id, z_acldata, z_acldata)) {
3440 return (DCMD_ERR);
3441 }
3442
3443 return (zfs_aces_walk_init_common(wsp, version,
3444 z_ace_count, z_acldata));
3445 }
3446
3447 /* ARGSUSED */
3448 static int
3449 zfs_acl_node_aces_walk_init(mdb_walk_state_t *wsp)
3450 {
3451 return (zfs_acl_node_aces_walk_init_common(wsp, 1));
3452 }
3453
3454 /* ARGSUSED */
3455 static int
3456 zfs_acl_node_aces0_walk_init(mdb_walk_state_t *wsp)
3457 {
3458 return (zfs_acl_node_aces_walk_init_common(wsp, 0));
3459 }
3460
3461 static int
3462 zfs_aces_walk_step(mdb_walk_state_t *wsp)
3463 {
3464 ace_walk_data_t *ace_data = wsp->walk_data;
3465 zfs_ace_t zace;
3466 ace_t *acep;
3467 int status;
3468 int entry_type;
3469 int allow_type;
3470 uintptr_t ptr;
3471
3472 if (ace_data->ace_count == 0)
3473 return (WALK_DONE);
3474
3475 if (mdb_vread(&zace, sizeof (zfs_ace_t), wsp->walk_addr) == -1) {
3476 mdb_warn("failed to read zfs_ace_t at %#lx",
3477 wsp->walk_addr);
3478 return (WALK_ERR);
3479 }
3480
3481 switch (ace_data->ace_version) {
3482 case 0:
3483 acep = (ace_t *)&zace;
3484 entry_type = acep->a_flags & ACE_TYPE_FLAGS;
3485 allow_type = acep->a_type;
3486 break;
3487 case 1:
3488 entry_type = zace.z_hdr.z_flags & ACE_TYPE_FLAGS;
3489 allow_type = zace.z_hdr.z_type;
3490 break;
3491 default:
3492 return (WALK_ERR);
3493 }
3494
3495 ptr = (uintptr_t)wsp->walk_addr;
3496 switch (entry_type) {
3497 case ACE_OWNER:
3498 case ACE_EVERYONE:
3499 case (ACE_IDENTIFIER_GROUP | ACE_GROUP):
3500 ptr += ace_data->ace_version == 0 ?
3501 sizeof (ace_t) : sizeof (zfs_ace_hdr_t);
3502 break;
3503 case ACE_IDENTIFIER_GROUP:
3504 default:
3505 switch (allow_type) {
3506 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
3507 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
3508 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
3509 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
3510 ptr += ace_data->ace_version == 0 ?
3511 sizeof (ace_t) : sizeof (zfs_object_ace_t);
3512 break;
3513 default:
3514 ptr += ace_data->ace_version == 0 ?
3515 sizeof (ace_t) : sizeof (zfs_ace_t);
3516 break;
3517 }
3518 }
3519
3520 ace_data->ace_count--;
3521 status = wsp->walk_callback(wsp->walk_addr,
3522 (void *)(uintptr_t)&zace, wsp->walk_cbdata);
3523
3524 wsp->walk_addr = ptr;
3525 return (status);
3526 }
3527
3528 typedef struct mdb_zfs_rrwlock {
3529 uintptr_t rr_writer;
3530 boolean_t rr_writer_wanted;
3531 } mdb_zfs_rrwlock_t;
3532
3533 static uint_t rrw_key;
3534
3535 /* ARGSUSED */
3536 static int
3537 rrwlock(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3538 {
3539 mdb_zfs_rrwlock_t rrw;
3540
3541 if (rrw_key == 0) {
3542 if (mdb_ctf_readsym(&rrw_key, "uint_t", "rrw_tsd_key", 0) == -1)
3543 return (DCMD_ERR);
3544 }
3545
3546 if (mdb_ctf_vread(&rrw, "rrwlock_t", "mdb_zfs_rrwlock_t", addr,
3547 0) == -1)
3548 return (DCMD_ERR);
3549
3550 if (rrw.rr_writer != 0) {
3551 mdb_printf("write lock held by thread %lx\n", rrw.rr_writer);
3552 return (DCMD_OK);
3553 }
3554
3555 if (rrw.rr_writer_wanted) {
3556 mdb_printf("writer wanted\n");
3557 }
3558
3559 mdb_printf("anonymous references:\n");
3560 (void) mdb_call_dcmd("refcount", addr +
3561 mdb_ctf_offsetof_by_name(ZFS_STRUCT "rrwlock", "rr_anon_rcount"),
3562 DCMD_ADDRSPEC, 0, NULL);
3563
3564 mdb_printf("linked references:\n");
3565 (void) mdb_call_dcmd("refcount", addr +
3566 mdb_ctf_offsetof_by_name(ZFS_STRUCT "rrwlock", "rr_linked_rcount"),
3567 DCMD_ADDRSPEC, 0, NULL);
3568
3569 /*
3570 * XXX This should find references from
3571 * "::walk thread | ::tsd -v <rrw_key>", but there is no support
3572 * for programmatic consumption of dcmds, so this would be
3573 * difficult, potentially requiring reimplementing ::tsd (both
3574 * user and kernel versions) in this MDB module.
3575 */
3576
3577 return (DCMD_OK);
3578 }
3579
3580 typedef struct mdb_arc_buf_hdr_t {
3581 uint16_t b_psize;
3582 uint16_t b_lsize;
3583 struct {
3584 uint32_t b_bufcnt;
3585 uintptr_t b_state;
3586 } b_l1hdr;
3587 } mdb_arc_buf_hdr_t;
3588
3589 enum arc_cflags {
3590 ARC_CFLAG_VERBOSE = 1 << 0,
3591 ARC_CFLAG_ANON = 1 << 1,
3592 ARC_CFLAG_MRU = 1 << 2,
3593 ARC_CFLAG_MFU = 1 << 3,
3594 ARC_CFLAG_BUFS = 1 << 4,
3595 };
3596
3597 typedef struct arc_compression_stats_data {
3598 GElf_Sym anon_sym; /* ARC_anon symbol */
3599 GElf_Sym mru_sym; /* ARC_mru symbol */
3600 GElf_Sym mrug_sym; /* ARC_mru_ghost symbol */
3601 GElf_Sym mfu_sym; /* ARC_mfu symbol */
3602 GElf_Sym mfug_sym; /* ARC_mfu_ghost symbol */
3603 GElf_Sym l2c_sym; /* ARC_l2c_only symbol */
3604 uint64_t *anon_c_hist; /* histogram of compressed sizes in anon */
3605 uint64_t *anon_u_hist; /* histogram of uncompressed sizes in anon */
3606 uint64_t *anon_bufs; /* histogram of buffer counts in anon state */
3607 uint64_t *mru_c_hist; /* histogram of compressed sizes in mru */
3608 uint64_t *mru_u_hist; /* histogram of uncompressed sizes in mru */
3609 uint64_t *mru_bufs; /* histogram of buffer counts in mru */
3610 uint64_t *mfu_c_hist; /* histogram of compressed sizes in mfu */
3611 uint64_t *mfu_u_hist; /* histogram of uncompressed sizes in mfu */
3612 uint64_t *mfu_bufs; /* histogram of buffer counts in mfu */
3613 uint64_t *all_c_hist; /* histogram of compressed anon + mru + mfu */
3614 uint64_t *all_u_hist; /* histogram of uncompressed anon + mru + mfu */
3615 uint64_t *all_bufs; /* histogram of buffer counts in all states */
3616 int arc_cflags; /* arc compression flags, specified by user */
3617 int hist_nbuckets; /* number of buckets in each histogram */
3618 } arc_compression_stats_data_t;
3619
3620 int
3621 highbit64(uint64_t i)
3622 {
3623 int h = 1;
3624
3625 if (i == 0)
3626 return (0);
3627 if (i & 0xffffffff00000000ULL) {
3628 h += 32; i >>= 32;
3629 }
3630 if (i & 0xffff0000) {
3631 h += 16; i >>= 16;
3632 }
3633 if (i & 0xff00) {
3634 h += 8; i >>= 8;
3635 }
3636 if (i & 0xf0) {
3637 h += 4; i >>= 4;
3638 }
3639 if (i & 0xc) {
3640 h += 2; i >>= 2;
3641 }
3642 if (i & 0x2) {
3643 h += 1;
3644 }
3645 return (h);
3646 }
3647
3648 /* ARGSUSED */
3649 static int
3650 arc_compression_stats_cb(uintptr_t addr, const void *unknown, void *arg)
3651 {
3652 arc_compression_stats_data_t *data = arg;
3653 mdb_arc_buf_hdr_t hdr;
3654 int cbucket, ubucket, bufcnt;
3655
3656 if (mdb_ctf_vread(&hdr, "arc_buf_hdr_t", "mdb_arc_buf_hdr_t",
3657 addr, 0) == -1) {
3658 return (WALK_ERR);
3659 }
3660
3661 /*
3662 * Headers in the ghost states, or the l2c_only state don't have
3663 * arc buffers linked off of them. Thus, their compressed size
3664 * is meaningless, so we skip these from the stats.
3665 */
3666 if (hdr.b_l1hdr.b_state == data->mrug_sym.st_value ||
3667 hdr.b_l1hdr.b_state == data->mfug_sym.st_value ||
3668 hdr.b_l1hdr.b_state == data->l2c_sym.st_value) {
3669 return (WALK_NEXT);
3670 }
3671
3672 /*
3673 * The physical size (compressed) and logical size
3674 * (uncompressed) are in units of SPA_MINBLOCKSIZE. By default,
3675 * we use the log2 of this value (rounded down to the nearest
3676 * integer) to determine the bucket to assign this header to.
3677 * Thus, the histogram is logarithmic with respect to the size
3678 * of the header. For example, the following is a mapping of the
3679 * bucket numbers and the range of header sizes they correspond to:
3680 *
3681 * 0: 0 byte headers
3682 * 1: 512 byte headers
3683 * 2: [1024 - 2048) byte headers
3684 * 3: [2048 - 4096) byte headers
3685 * 4: [4096 - 8192) byte headers
3686 * 5: [8192 - 16394) byte headers
3687 * 6: [16384 - 32768) byte headers
3688 * 7: [32768 - 65536) byte headers
3689 * 8: [65536 - 131072) byte headers
3690 * 9: 131072 byte headers
3691 *
3692 * If the ARC_CFLAG_VERBOSE flag was specified, we use the
3693 * physical and logical sizes directly. Thus, the histogram will
3694 * no longer be logarithmic; instead it will be linear with
3695 * respect to the size of the header. The following is a mapping
3696 * of the first many bucket numbers and the header size they
3697 * correspond to:
3698 *
3699 * 0: 0 byte headers
3700 * 1: 512 byte headers
3701 * 2: 1024 byte headers
3702 * 3: 1536 byte headers
3703 * 4: 2048 byte headers
3704 * 5: 2560 byte headers
3705 * 6: 3072 byte headers
3706 *
3707 * And so on. Keep in mind that a range of sizes isn't used in
3708 * the case of linear scale because the headers can only
3709 * increment or decrement in sizes of 512 bytes. So, it's not
3710 * possible for a header to be sized in between whats listed
3711 * above.
3712 *
3713 * Also, the above mapping values were calculated assuming a
3714 * SPA_MINBLOCKSHIFT of 512 bytes and a SPA_MAXBLOCKSIZE of 128K.
3715 */
3716
3717 if (data->arc_cflags & ARC_CFLAG_VERBOSE) {
3718 cbucket = hdr.b_psize;
3719 ubucket = hdr.b_lsize;
3720 } else {
3721 cbucket = highbit64(hdr.b_psize);
3722 ubucket = highbit64(hdr.b_lsize);
3723 }
3724
3725 bufcnt = hdr.b_l1hdr.b_bufcnt;
3726 if (bufcnt >= data->hist_nbuckets)
3727 bufcnt = data->hist_nbuckets - 1;
3728
3729 /* Ensure we stay within the bounds of the histogram array */
3730 ASSERT3U(cbucket, <, data->hist_nbuckets);
3731 ASSERT3U(ubucket, <, data->hist_nbuckets);
3732
3733 if (hdr.b_l1hdr.b_state == data->anon_sym.st_value) {
3734 data->anon_c_hist[cbucket]++;
3735 data->anon_u_hist[ubucket]++;
3736 data->anon_bufs[bufcnt]++;
3737 } else if (hdr.b_l1hdr.b_state == data->mru_sym.st_value) {
3738 data->mru_c_hist[cbucket]++;
3739 data->mru_u_hist[ubucket]++;
3740 data->mru_bufs[bufcnt]++;
3741 } else if (hdr.b_l1hdr.b_state == data->mfu_sym.st_value) {
3742 data->mfu_c_hist[cbucket]++;
3743 data->mfu_u_hist[ubucket]++;
3744 data->mfu_bufs[bufcnt]++;
3745 }
3746
3747 data->all_c_hist[cbucket]++;
3748 data->all_u_hist[ubucket]++;
3749 data->all_bufs[bufcnt]++;
3750
3751 return (WALK_NEXT);
3752 }
3753
3754 /* ARGSUSED */
3755 static int
3756 arc_compression_stats(uintptr_t addr, uint_t flags, int argc,
3757 const mdb_arg_t *argv)
3758 {
3759 arc_compression_stats_data_t data = { 0 };
3760 unsigned int max_shifted = SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT;
3761 unsigned int hist_size;
3762 char range[32];
3763 int rc = DCMD_OK;
3764
3765 if (mdb_getopts(argc, argv,
3766 'v', MDB_OPT_SETBITS, ARC_CFLAG_VERBOSE, &data.arc_cflags,
3767 'a', MDB_OPT_SETBITS, ARC_CFLAG_ANON, &data.arc_cflags,
3768 'b', MDB_OPT_SETBITS, ARC_CFLAG_BUFS, &data.arc_cflags,
3769 'r', MDB_OPT_SETBITS, ARC_CFLAG_MRU, &data.arc_cflags,
3770 'f', MDB_OPT_SETBITS, ARC_CFLAG_MFU, &data.arc_cflags) != argc)
3771 return (DCMD_USAGE);
3772
3773 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_anon", &data.anon_sym) ||
3774 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mru", &data.mru_sym) ||
3775 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mru_ghost", &data.mrug_sym) ||
3776 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mfu", &data.mfu_sym) ||
3777 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mfu_ghost", &data.mfug_sym) ||
3778 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_l2c_only", &data.l2c_sym)) {
3779 mdb_warn("can't find arc state symbol");
3780 return (DCMD_ERR);
3781 }
3782
3783 /*
3784 * Determine the maximum expected size for any header, and use
3785 * this to determine the number of buckets needed for each
3786 * histogram. If ARC_CFLAG_VERBOSE is specified, this value is
3787 * used directly; otherwise the log2 of the maximum size is
3788 * used. Thus, if using a log2 scale there's a maximum of 10
3789 * possible buckets, while the linear scale (when using
3790 * ARC_CFLAG_VERBOSE) has a maximum of 257 buckets.
3791 */
3792 if (data.arc_cflags & ARC_CFLAG_VERBOSE)
3793 data.hist_nbuckets = max_shifted + 1;
3794 else
3795 data.hist_nbuckets = highbit64(max_shifted) + 1;
3796
3797 hist_size = sizeof (uint64_t) * data.hist_nbuckets;
3798
3799 data.anon_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3800 data.anon_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3801 data.anon_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3802
3803 data.mru_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3804 data.mru_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3805 data.mru_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3806
3807 data.mfu_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3808 data.mfu_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3809 data.mfu_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3810
3811 data.all_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3812 data.all_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3813 data.all_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3814
3815 if (mdb_walk("arc_buf_hdr_t_full", arc_compression_stats_cb,
3816 &data) != 0) {
3817 mdb_warn("can't walk arc_buf_hdr's");
3818 rc = DCMD_ERR;
3819 goto out;
3820 }
3821
3822 if (data.arc_cflags & ARC_CFLAG_VERBOSE) {
3823 rc = mdb_snprintf(range, sizeof (range),
3824 "[n*%llu, (n+1)*%llu)", SPA_MINBLOCKSIZE,
3825 SPA_MINBLOCKSIZE);
3826 } else {
3827 rc = mdb_snprintf(range, sizeof (range),
3828 "[2^(n-1)*%llu, 2^n*%llu)", SPA_MINBLOCKSIZE,
3829 SPA_MINBLOCKSIZE);
3830 }
3831
3832 if (rc < 0) {
3833 /* snprintf failed, abort the dcmd */
3834 rc = DCMD_ERR;
3835 goto out;
3836 } else {
3837 /* snprintf succeeded above, reset return code */
3838 rc = DCMD_OK;
3839 }
3840
3841 if (data.arc_cflags & ARC_CFLAG_ANON) {
3842 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3843 mdb_printf("Histogram of the number of anon buffers "
3844 "that are associated with an arc hdr.\n");
3845 dump_histogram(data.anon_bufs, data.hist_nbuckets, 0);
3846 mdb_printf("\n");
3847 }
3848 mdb_printf("Histogram of compressed anon buffers.\n"
3849 "Each bucket represents buffers of size: %s.\n", range);
3850 dump_histogram(data.anon_c_hist, data.hist_nbuckets, 0);
3851 mdb_printf("\n");
3852
3853 mdb_printf("Histogram of uncompressed anon buffers.\n"
3854 "Each bucket represents buffers of size: %s.\n", range);
3855 dump_histogram(data.anon_u_hist, data.hist_nbuckets, 0);
3856 mdb_printf("\n");
3857 }
3858
3859 if (data.arc_cflags & ARC_CFLAG_MRU) {
3860 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3861 mdb_printf("Histogram of the number of mru buffers "
3862 "that are associated with an arc hdr.\n");
3863 dump_histogram(data.mru_bufs, data.hist_nbuckets, 0);
3864 mdb_printf("\n");
3865 }
3866 mdb_printf("Histogram of compressed mru buffers.\n"
3867 "Each bucket represents buffers of size: %s.\n", range);
3868 dump_histogram(data.mru_c_hist, data.hist_nbuckets, 0);
3869 mdb_printf("\n");
3870
3871 mdb_printf("Histogram of uncompressed mru buffers.\n"
3872 "Each bucket represents buffers of size: %s.\n", range);
3873 dump_histogram(data.mru_u_hist, data.hist_nbuckets, 0);
3874 mdb_printf("\n");
3875 }
3876
3877 if (data.arc_cflags & ARC_CFLAG_MFU) {
3878 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3879 mdb_printf("Histogram of the number of mfu buffers "
3880 "that are associated with an arc hdr.\n");
3881 dump_histogram(data.mfu_bufs, data.hist_nbuckets, 0);
3882 mdb_printf("\n");
3883 }
3884
3885 mdb_printf("Histogram of compressed mfu buffers.\n"
3886 "Each bucket represents buffers of size: %s.\n", range);
3887 dump_histogram(data.mfu_c_hist, data.hist_nbuckets, 0);
3888 mdb_printf("\n");
3889
3890 mdb_printf("Histogram of uncompressed mfu buffers.\n"
3891 "Each bucket represents buffers of size: %s.\n", range);
3892 dump_histogram(data.mfu_u_hist, data.hist_nbuckets, 0);
3893 mdb_printf("\n");
3894 }
3895
3896 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3897 mdb_printf("Histogram of all buffers that "
3898 "are associated with an arc hdr.\n");
3899 dump_histogram(data.all_bufs, data.hist_nbuckets, 0);
3900 mdb_printf("\n");
3901 }
3902
3903 mdb_printf("Histogram of all compressed buffers.\n"
3904 "Each bucket represents buffers of size: %s.\n", range);
3905 dump_histogram(data.all_c_hist, data.hist_nbuckets, 0);
3906 mdb_printf("\n");
3907
3908 mdb_printf("Histogram of all uncompressed buffers.\n"
3909 "Each bucket represents buffers of size: %s.\n", range);
3910 dump_histogram(data.all_u_hist, data.hist_nbuckets, 0);
3911
3912 out:
3913 mdb_free(data.anon_c_hist, hist_size);
3914 mdb_free(data.anon_u_hist, hist_size);
3915 mdb_free(data.anon_bufs, hist_size);
3916
3917 mdb_free(data.mru_c_hist, hist_size);
3918 mdb_free(data.mru_u_hist, hist_size);
3919 mdb_free(data.mru_bufs, hist_size);
3920
3921 mdb_free(data.mfu_c_hist, hist_size);
3922 mdb_free(data.mfu_u_hist, hist_size);
3923 mdb_free(data.mfu_bufs, hist_size);
3924
3925 mdb_free(data.all_c_hist, hist_size);
3926 mdb_free(data.all_u_hist, hist_size);
3927 mdb_free(data.all_bufs, hist_size);
3928
3929 return (rc);
3930 }
3931
3932 /*
3933 * MDB module linkage information:
3934 *
3935 * We declare a list of structures describing our dcmds, and a function
3936 * named _mdb_init to return a pointer to our module information.
3937 */
3938
3939 static const mdb_dcmd_t dcmds[] = {
3940 { "arc", "[-bkmg]", "print ARC variables", arc_print },
3941 { "blkptr", ":", "print blkptr_t", blkptr },
3942 { "dva", ":", "print dva_t", dva },
3943 { "dbuf", ":", "print dmu_buf_impl_t", dbuf },
3944 { "dbuf_stats", ":", "dbuf stats", dbuf_stats },
3945 { "dbufs",
3946 "\t[-O objset_t*] [-n objset_name | \"mos\"] "
3947 "[-o object | \"mdn\"] \n"
3948 "\t[-l level] [-b blkid | \"bonus\"]",
3949 "find dmu_buf_impl_t's that match specified criteria", dbufs },
3950 { "abuf_find", "dva_word[0] dva_word[1]",
3951 "find arc_buf_hdr_t of a specified DVA",
3952 abuf_find },
3953 { "spa", "?[-cevmMh]\n"
3954 "\t-c display spa config\n"
3955 "\t-e display vdev statistics\n"
3956 "\t-v display vdev information\n"
3957 "\t-m display metaslab statistics\n"
3958 "\t-M display metaslab group statistics\n"
3959 "\t-h display histogram (requires -m or -M)\n",
3960 "spa_t summary", spa_print },
3961 { "spa_config", ":", "print spa_t configuration", spa_print_config },
3962 { "spa_space", ":[-b]", "print spa_t on-disk space usage", spa_space },
3963 { "spa_vdevs", ":[-emMh]\n"
3964 "\t-e display vdev statistics\n"
3965 "\t-m dispaly metaslab statistics\n"
3966 "\t-M display metaslab group statistic\n"
3967 "\t-h display histogram (requires -m or -M)\n",
3968 "given a spa_t, print vdev summary", spa_vdevs },
3969 { "vdev", ":[-remMh]\n"
3970 "\t-r display recursively\n"
3971 "\t-e display statistics\n"
3972 "\t-m display metaslab statistics (top level vdev only)\n"
3973 "\t-M display metaslab group statistics (top level vdev only)\n"
3974 "\t-h display histogram (requires -m or -M)\n",
3975 "vdev_t summary", vdev_print },
3976 { "zio", ":[-cpr]\n"
3977 "\t-c display children\n"
3978 "\t-p display parents\n"
3979 "\t-r display recursively",
3980 "zio_t summary", zio_print },
3981 { "zio_state", "?", "print out all zio_t structures on system or "
3982 "for a particular pool", zio_state },
3983 { "zfs_blkstats", ":[-v]",
3984 "given a spa_t, print block type stats from last scrub",
3985 zfs_blkstats },
3986 { "zfs_params", "", "print zfs tunable parameters", zfs_params },
3987 { "refcount", ":[-r]\n"
3988 "\t-r display recently removed references",
3989 "print refcount_t holders", refcount },
3990 { "zap_leaf", "", "print zap_leaf_phys_t", zap_leaf },
3991 { "zfs_aces", ":[-v]", "print all ACEs from a zfs_acl_t",
3992 zfs_acl_dump },
3993 { "zfs_ace", ":[-v]", "print zfs_ace", zfs_ace_print },
3994 { "zfs_ace0", ":[-v]", "print zfs_ace0", zfs_ace0_print },
3995 { "sa_attr_table", ":", "print SA attribute table from sa_os_t",
3996 sa_attr_table},
3997 { "sa_attr", ": attr_id",
3998 "print SA attribute address when given sa_handle_t", sa_attr_print},
3999 { "zfs_dbgmsg", ":[-va]",
4000 "print zfs debug log", dbgmsg},
4001 { "rrwlock", ":",
4002 "print rrwlock_t, including readers", rrwlock},
4003 { "metaslab_weight", "weight",
4004 "print metaslab weight", metaslab_weight},
4005 { "metaslab_trace", ":",
4006 "print metaslab allocation trace records", metaslab_trace},
4007 { "arc_compression_stats", ":[-vabrf]\n"
4008 "\t-v verbose, display a linearly scaled histogram\n"
4009 "\t-a display ARC_anon state statistics individually\n"
4010 "\t-r display ARC_mru state statistics individually\n"
4011 "\t-f display ARC_mfu state statistics individually\n"
4012 "\t-b display histogram of buffer counts\n",
4013 "print a histogram of compressed arc buffer sizes",
4014 arc_compression_stats},
4015 { NULL }
4016 };
4017
4018 static const mdb_walker_t walkers[] = {
4019 { "zms_freelist", "walk ZFS metaslab freelist",
4020 freelist_walk_init, freelist_walk_step, NULL },
4021 { "txg_list", "given any txg_list_t *, walk all entries in all txgs",
4022 txg_list_walk_init, txg_list_walk_step, NULL },
4023 { "txg_list0", "given any txg_list_t *, walk all entries in txg 0",
4024 txg_list0_walk_init, txg_list_walk_step, NULL },
4025 { "txg_list1", "given any txg_list_t *, walk all entries in txg 1",
4026 txg_list1_walk_init, txg_list_walk_step, NULL },
4027 { "txg_list2", "given any txg_list_t *, walk all entries in txg 2",
4028 txg_list2_walk_init, txg_list_walk_step, NULL },
4029 { "txg_list3", "given any txg_list_t *, walk all entries in txg 3",
4030 txg_list3_walk_init, txg_list_walk_step, NULL },
4031 { "zio", "walk all zio structures, optionally for a particular spa_t",
4032 zio_walk_init, zio_walk_step, NULL },
4033 { "zio_root",
4034 "walk all root zio_t structures, optionally for a particular spa_t",
4035 zio_walk_init, zio_walk_root_step, NULL },
4036 { "spa", "walk all spa_t entries in the namespace",
4037 spa_walk_init, spa_walk_step, NULL },
4038 { "metaslab", "given a spa_t *, walk all metaslab_t structures",
4039 metaslab_walk_init, metaslab_walk_step, NULL },
4040 { "multilist", "given a multilist_t *, walk all list_t structures",
4041 multilist_walk_init, multilist_walk_step, NULL },
4042 { "zfs_acl_node", "given a zfs_acl_t, walk all zfs_acl_nodes",
4043 zfs_acl_node_walk_init, zfs_acl_node_walk_step, NULL },
4044 { "zfs_acl_node_aces", "given a zfs_acl_node_t, walk all ACEs",
4045 zfs_acl_node_aces_walk_init, zfs_aces_walk_step, NULL },
4046 { "zfs_acl_node_aces0",
4047 "given a zfs_acl_node_t, walk all ACEs as ace_t",
4048 zfs_acl_node_aces0_walk_init, zfs_aces_walk_step, NULL },
4049 { NULL }
4050 };
4051
4052 static const mdb_modinfo_t modinfo = {
4053 MDB_API_VERSION, dcmds, walkers
4054 };
4055
4056 const mdb_modinfo_t *
4057 _mdb_init(void)
4058 {
4059 return (&modinfo);
4060 }
Unleashed OS