?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
9166 zfs storage pool checkpoint
[unleashed.git] / usr / src / cmd / mdb / common / modules / zfs / zfs.c
blob0c99391147811efe9e5fb6b9f4615ed8a437f272
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) 2017, Joyent, Inc. All rights reserved.
25 * Copyright (c) 2011, 2017 by Delphix. 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 == 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_allocating[TXG_SIZE];
1320 uintptr_t ms_checkpointing;
1321 uintptr_t ms_freeing;
1322 uintptr_t ms_freed;
1323 uintptr_t ms_allocatable;
1324 uintptr_t ms_sm;
1325 } mdb_metaslab_t;
1326
1327 typedef struct mdb_space_map_phys_t {
1328 int64_t smp_alloc;
1329 uint64_t smp_histogram[SPACE_MAP_HISTOGRAM_SIZE];
1330 } mdb_space_map_phys_t;
1331
1332 typedef struct mdb_space_map {
1333 uint64_t sm_size;
1334 uint8_t sm_shift;
1335 int64_t sm_alloc;
1336 uintptr_t sm_phys;
1337 } mdb_space_map_t;
1338
1339 typedef struct mdb_vdev {
1340 uintptr_t vdev_path;
1341 uintptr_t vdev_ms;
1342 uintptr_t vdev_ops;
1343 uint64_t vdev_ms_count;
1344 uint64_t vdev_id;
1345 vdev_stat_t vdev_stat;
1346 } mdb_vdev_t;
1347
1348 typedef struct mdb_vdev_ops {
1349 char vdev_op_type[16];
1350 } mdb_vdev_ops_t;
1351
1352 static int
1353 metaslab_stats(uintptr_t addr, int spa_flags)
1354 {
1355 mdb_vdev_t vdev;
1356 uintptr_t *vdev_ms;
1357
1358 if (mdb_ctf_vread(&vdev, "vdev_t", "mdb_vdev_t",
1359 (uintptr_t)addr, 0) == -1) {
1360 mdb_warn("failed to read vdev at %p\n", addr);
1361 return (DCMD_ERR);
1362 }
1363
1364 mdb_inc_indent(4);
1365 mdb_printf("%<u>%-?s %6s %20s %10s %9s%</u>\n", "ADDR", "ID",
1366 "OFFSET", "FREE", "FRAGMENTATION");
1367
1368 vdev_ms = mdb_alloc(vdev.vdev_ms_count * sizeof (void *),
1369 UM_SLEEP | UM_GC);
1370 if (mdb_vread(vdev_ms, vdev.vdev_ms_count * sizeof (void *),
1371 (uintptr_t)vdev.vdev_ms) == -1) {
1372 mdb_warn("failed to read vdev_ms at %p\n", vdev.vdev_ms);
1373 return (DCMD_ERR);
1374 }
1375
1376 for (int m = 0; m < vdev.vdev_ms_count; m++) {
1377 mdb_metaslab_t ms;
1378 mdb_space_map_t sm = { 0 };
1379 char free[NICENUM_BUFLEN];
1380
1381 if (mdb_ctf_vread(&ms, "metaslab_t", "mdb_metaslab_t",
1382 (uintptr_t)vdev_ms[m], 0) == -1)
1383 return (DCMD_ERR);
1384
1385 if (ms.ms_sm != NULL &&
1386 mdb_ctf_vread(&sm, "space_map_t", "mdb_space_map_t",
1387 ms.ms_sm, 0) == -1)
1388 return (DCMD_ERR);
1389
1390 mdb_nicenum(ms.ms_size - sm.sm_alloc, free);
1391
1392 mdb_printf("%0?p %6llu %20llx %10s ", vdev_ms[m], ms.ms_id,
1393 ms.ms_start, free);
1394 if (ms.ms_fragmentation == ZFS_FRAG_INVALID)
1395 mdb_printf("%9s\n", "-");
1396 else
1397 mdb_printf("%9llu%%\n", ms.ms_fragmentation);
1398
1399 if ((spa_flags & SPA_FLAG_HISTOGRAMS) && ms.ms_sm != NULL) {
1400 mdb_space_map_phys_t smp;
1401
1402 if (sm.sm_phys == NULL)
1403 continue;
1404
1405 (void) mdb_ctf_vread(&smp, "space_map_phys_t",
1406 "mdb_space_map_phys_t", sm.sm_phys, 0);
1407
1408 dump_histogram(smp.smp_histogram,
1409 SPACE_MAP_HISTOGRAM_SIZE, sm.sm_shift);
1410 }
1411 }
1412 mdb_dec_indent(4);
1413 return (DCMD_OK);
1414 }
1415
1416 static int
1417 metaslab_group_stats(uintptr_t addr, int spa_flags)
1418 {
1419 mdb_metaslab_group_t mg;
1420 if (mdb_ctf_vread(&mg, "metaslab_group_t", "mdb_metaslab_group_t",
1421 (uintptr_t)addr, 0) == -1) {
1422 mdb_warn("failed to read vdev_mg at %p\n", addr);
1423 return (DCMD_ERR);
1424 }
1425
1426 mdb_inc_indent(4);
1427 mdb_printf("%<u>%-?s %15s%</u>\n", "ADDR", "FRAGMENTATION");
1428 if (mg.mg_fragmentation == ZFS_FRAG_INVALID)
1429 mdb_printf("%0?p %15s\n", addr, "-");
1430 else
1431 mdb_printf("%0?p %15llu%%\n", addr, mg.mg_fragmentation);
1432
1433 if (spa_flags & SPA_FLAG_HISTOGRAMS)
1434 dump_histogram(mg.mg_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1435 mdb_dec_indent(4);
1436 return (DCMD_OK);
1437 }
1438
1439 /*
1440 * ::vdev
1441 *
1442 * Print out a summarized vdev_t, in the following form:
1443 *
1444 * ADDR STATE AUX DESC
1445 * fffffffbcde23df0 HEALTHY - /dev/dsk/c0t0d0
1446 *
1447 * If '-r' is specified, recursively visit all children.
1448 *
1449 * With '-e', the statistics associated with the vdev are printed as well.
1450 */
1451 static int
1452 do_print_vdev(uintptr_t addr, int flags, int depth, boolean_t recursive,
1453 int spa_flags)
1454 {
1455 vdev_t vdev;
1456 char desc[MAXNAMELEN];
1457 int c, children;
1458 uintptr_t *child;
1459 const char *state, *aux;
1460
1461 if (mdb_vread(&vdev, sizeof (vdev), (uintptr_t)addr) == -1) {
1462 mdb_warn("failed to read vdev_t at %p\n", (uintptr_t)addr);
1463 return (DCMD_ERR);
1464 }
1465
1466 if (flags & DCMD_PIPE_OUT) {
1467 mdb_printf("%#lr\n", addr);
1468 } else {
1469 if (vdev.vdev_path != NULL) {
1470 if (mdb_readstr(desc, sizeof (desc),
1471 (uintptr_t)vdev.vdev_path) == -1) {
1472 mdb_warn("failed to read vdev_path at %p\n",
1473 vdev.vdev_path);
1474 return (DCMD_ERR);
1475 }
1476 } else if (vdev.vdev_ops != NULL) {
1477 vdev_ops_t ops;
1478 if (mdb_vread(&ops, sizeof (ops),
1479 (uintptr_t)vdev.vdev_ops) == -1) {
1480 mdb_warn("failed to read vdev_ops at %p\n",
1481 vdev.vdev_ops);
1482 return (DCMD_ERR);
1483 }
1484 (void) strcpy(desc, ops.vdev_op_type);
1485 } else {
1486 (void) strcpy(desc, "<unknown>");
1487 }
1488
1489 if (depth == 0 && DCMD_HDRSPEC(flags))
1490 mdb_printf("%<u>%-?s %-9s %-12s %-*s%</u>\n",
1491 "ADDR", "STATE", "AUX",
1492 sizeof (uintptr_t) == 4 ? 43 : 35,
1493 "DESCRIPTION");
1494
1495 mdb_printf("%0?p ", addr);
1496
1497 switch (vdev.vdev_state) {
1498 case VDEV_STATE_CLOSED:
1499 state = "CLOSED";
1500 break;
1501 case VDEV_STATE_OFFLINE:
1502 state = "OFFLINE";
1503 break;
1504 case VDEV_STATE_CANT_OPEN:
1505 state = "CANT_OPEN";
1506 break;
1507 case VDEV_STATE_DEGRADED:
1508 state = "DEGRADED";
1509 break;
1510 case VDEV_STATE_HEALTHY:
1511 state = "HEALTHY";
1512 break;
1513 case VDEV_STATE_REMOVED:
1514 state = "REMOVED";
1515 break;
1516 case VDEV_STATE_FAULTED:
1517 state = "FAULTED";
1518 break;
1519 default:
1520 state = "UNKNOWN";
1521 break;
1522 }
1523
1524 switch (vdev.vdev_stat.vs_aux) {
1525 case VDEV_AUX_NONE:
1526 aux = "-";
1527 break;
1528 case VDEV_AUX_OPEN_FAILED:
1529 aux = "OPEN_FAILED";
1530 break;
1531 case VDEV_AUX_CORRUPT_DATA:
1532 aux = "CORRUPT_DATA";
1533 break;
1534 case VDEV_AUX_NO_REPLICAS:
1535 aux = "NO_REPLICAS";
1536 break;
1537 case VDEV_AUX_BAD_GUID_SUM:
1538 aux = "BAD_GUID_SUM";
1539 break;
1540 case VDEV_AUX_TOO_SMALL:
1541 aux = "TOO_SMALL";
1542 break;
1543 case VDEV_AUX_BAD_LABEL:
1544 aux = "BAD_LABEL";
1545 break;
1546 case VDEV_AUX_VERSION_NEWER:
1547 aux = "VERS_NEWER";
1548 break;
1549 case VDEV_AUX_VERSION_OLDER:
1550 aux = "VERS_OLDER";
1551 break;
1552 case VDEV_AUX_UNSUP_FEAT:
1553 aux = "UNSUP_FEAT";
1554 break;
1555 case VDEV_AUX_SPARED:
1556 aux = "SPARED";
1557 break;
1558 case VDEV_AUX_ERR_EXCEEDED:
1559 aux = "ERR_EXCEEDED";
1560 break;
1561 case VDEV_AUX_IO_FAILURE:
1562 aux = "IO_FAILURE";
1563 break;
1564 case VDEV_AUX_BAD_LOG:
1565 aux = "BAD_LOG";
1566 break;
1567 case VDEV_AUX_EXTERNAL:
1568 aux = "EXTERNAL";
1569 break;
1570 case VDEV_AUX_SPLIT_POOL:
1571 aux = "SPLIT_POOL";
1572 break;
1573 case VDEV_AUX_CHILDREN_OFFLINE:
1574 aux = "CHILDREN_OFFLINE";
1575 break;
1576 default:
1577 aux = "UNKNOWN";
1578 break;
1579 }
1580
1581 mdb_printf("%-9s %-12s %*s%s\n", state, aux, depth, "", desc);
1582
1583 if (spa_flags & SPA_FLAG_ERRORS) {
1584 vdev_stat_t *vs = &vdev.vdev_stat;
1585 int i;
1586
1587 mdb_inc_indent(4);
1588 mdb_printf("\n");
1589 mdb_printf("%<u> %12s %12s %12s %12s "
1590 "%12s%</u>\n", "READ", "WRITE", "FREE", "CLAIM",
1591 "IOCTL");
1592 mdb_printf("OPS ");
1593 for (i = 1; i < ZIO_TYPES; i++)
1594 mdb_printf("%11#llx%s", vs->vs_ops[i],
1595 i == ZIO_TYPES - 1 ? "" : " ");
1596 mdb_printf("\n");
1597 mdb_printf("BYTES ");
1598 for (i = 1; i < ZIO_TYPES; i++)
1599 mdb_printf("%11#llx%s", vs->vs_bytes[i],
1600 i == ZIO_TYPES - 1 ? "" : " ");
1601
1602
1603 mdb_printf("\n");
1604 mdb_printf("EREAD %10#llx\n", vs->vs_read_errors);
1605 mdb_printf("EWRITE %10#llx\n", vs->vs_write_errors);
1606 mdb_printf("ECKSUM %10#llx\n",
1607 vs->vs_checksum_errors);
1608 mdb_dec_indent(4);
1609 mdb_printf("\n");
1610 }
1611
1612 if (spa_flags & SPA_FLAG_METASLAB_GROUPS &&
1613 vdev.vdev_mg != NULL) {
1614 metaslab_group_stats((uintptr_t)vdev.vdev_mg,
1615 spa_flags);
1616 }
1617 if (spa_flags & SPA_FLAG_METASLABS && vdev.vdev_ms != NULL) {
1618 metaslab_stats((uintptr_t)addr, spa_flags);
1619 }
1620 }
1621
1622 children = vdev.vdev_children;
1623
1624 if (children == 0 || !recursive)
1625 return (DCMD_OK);
1626
1627 child = mdb_alloc(children * sizeof (void *), UM_SLEEP | UM_GC);
1628 if (mdb_vread(child, children * sizeof (void *),
1629 (uintptr_t)vdev.vdev_child) == -1) {
1630 mdb_warn("failed to read vdev children at %p", vdev.vdev_child);
1631 return (DCMD_ERR);
1632 }
1633
1634 for (c = 0; c < children; c++) {
1635 if (do_print_vdev(child[c], flags, depth + 2, recursive,
1636 spa_flags)) {
1637 return (DCMD_ERR);
1638 }
1639 }
1640
1641 return (DCMD_OK);
1642 }
1643
1644 static int
1645 vdev_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1646 {
1647 uint64_t depth = 0;
1648 boolean_t recursive = B_FALSE;
1649 int spa_flags = 0;
1650
1651 if (mdb_getopts(argc, argv,
1652 'e', MDB_OPT_SETBITS, SPA_FLAG_ERRORS, &spa_flags,
1653 'm', MDB_OPT_SETBITS, SPA_FLAG_METASLABS, &spa_flags,
1654 'M', MDB_OPT_SETBITS, SPA_FLAG_METASLAB_GROUPS, &spa_flags,
1655 'h', MDB_OPT_SETBITS, SPA_FLAG_HISTOGRAMS, &spa_flags,
1656 'r', MDB_OPT_SETBITS, TRUE, &recursive,
1657 'd', MDB_OPT_UINT64, &depth, NULL) != argc)
1658 return (DCMD_USAGE);
1659
1660 if (!(flags & DCMD_ADDRSPEC)) {
1661 mdb_warn("no vdev_t address given\n");
1662 return (DCMD_ERR);
1663 }
1664
1665 return (do_print_vdev(addr, flags, (int)depth, recursive, spa_flags));
1666 }
1667
1668 typedef struct mdb_metaslab_alloc_trace {
1669 uintptr_t mat_mg;
1670 uintptr_t mat_msp;
1671 uint64_t mat_size;
1672 uint64_t mat_weight;
1673 uint64_t mat_offset;
1674 uint32_t mat_dva_id;
1675 } mdb_metaslab_alloc_trace_t;
1676
1677 static void
1678 metaslab_print_weight(uint64_t weight)
1679 {
1680 char buf[100];
1681
1682 if (WEIGHT_IS_SPACEBASED(weight)) {
1683 mdb_nicenum(
1684 weight & ~(METASLAB_ACTIVE_MASK | METASLAB_WEIGHT_TYPE),
1685 buf);
1686 } else {
1687 char size[NICENUM_BUFLEN];
1688 mdb_nicenum(1ULL << WEIGHT_GET_INDEX(weight), size);
1689 (void) mdb_snprintf(buf, sizeof (buf), "%llu x %s",
1690 WEIGHT_GET_COUNT(weight), size);
1691 }
1692 mdb_printf("%11s ", buf);
1693 }
1694
1695 /* ARGSUSED */
1696 static int
1697 metaslab_weight(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1698 {
1699 uint64_t weight = 0;
1700 char active;
1701
1702 if (argc == 0 && (flags & DCMD_ADDRSPEC)) {
1703 if (mdb_vread(&weight, sizeof (uint64_t), addr) == -1) {
1704 mdb_warn("failed to read weight at %p\n", addr);
1705 return (DCMD_ERR);
1706 }
1707 } else if (argc == 1 && !(flags & DCMD_ADDRSPEC)) {
1708 weight = (argv[0].a_type == MDB_TYPE_IMMEDIATE) ?
1709 argv[0].a_un.a_val : mdb_strtoull(argv[0].a_un.a_str);
1710 } else {
1711 return (DCMD_USAGE);
1712 }
1713
1714 if (DCMD_HDRSPEC(flags)) {
1715 mdb_printf("%<u>%-6s %9s %9s%</u>\n",
1716 "ACTIVE", "ALGORITHM", "WEIGHT");
1717 }
1718
1719 if (weight & METASLAB_WEIGHT_PRIMARY)
1720 active = 'P';
1721 else if (weight & METASLAB_WEIGHT_SECONDARY)
1722 active = 'S';
1723 else
1724 active = '-';
1725 mdb_printf("%6c %8s ", active,
1726 WEIGHT_IS_SPACEBASED(weight) ? "SPACE" : "SEGMENT");
1727 metaslab_print_weight(weight);
1728 mdb_printf("\n");
1729
1730 return (DCMD_OK);
1731 }
1732
1733 /* ARGSUSED */
1734 static int
1735 metaslab_trace(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1736 {
1737 mdb_metaslab_alloc_trace_t mat;
1738 mdb_metaslab_group_t mg = { 0 };
1739 char result_type[100];
1740
1741 if (mdb_ctf_vread(&mat, "metaslab_alloc_trace_t",
1742 "mdb_metaslab_alloc_trace_t", addr, 0) == -1) {
1743 return (DCMD_ERR);
1744 }
1745
1746 if (!(flags & DCMD_PIPE_OUT) && DCMD_HDRSPEC(flags)) {
1747 mdb_printf("%<u>%6s %6s %8s %11s %18s %18s%</u>\n",
1748 "MSID", "DVA", "ASIZE", "WEIGHT", "RESULT", "VDEV");
1749 }
1750
1751 if (mat.mat_msp != NULL) {
1752 mdb_metaslab_t ms;
1753
1754 if (mdb_ctf_vread(&ms, "metaslab_t", "mdb_metaslab_t",
1755 mat.mat_msp, 0) == -1) {
1756 return (DCMD_ERR);
1757 }
1758 mdb_printf("%6llu ", ms.ms_id);
1759 } else {
1760 mdb_printf("%6s ", "-");
1761 }
1762
1763 mdb_printf("%6d %8llx ", mat.mat_dva_id, mat.mat_size);
1764
1765 metaslab_print_weight(mat.mat_weight);
1766
1767 if ((int64_t)mat.mat_offset < 0) {
1768 if (enum_lookup("enum trace_alloc_type", mat.mat_offset,
1769 "TRACE_", sizeof (result_type), result_type) == -1) {
1770 mdb_warn("Could not find enum for trace_alloc_type");
1771 return (DCMD_ERR);
1772 }
1773 mdb_printf("%18s ", result_type);
1774 } else {
1775 mdb_printf("%<b>%18llx%</b> ", mat.mat_offset);
1776 }
1777
1778 if (mat.mat_mg != NULL &&
1779 mdb_ctf_vread(&mg, "metaslab_group_t", "mdb_metaslab_group_t",
1780 mat.mat_mg, 0) == -1) {
1781 return (DCMD_ERR);
1782 }
1783
1784 if (mg.mg_vd != NULL) {
1785 mdb_vdev_t vdev;
1786 char desc[MAXNAMELEN];
1787
1788 if (mdb_ctf_vread(&vdev, "vdev_t", "mdb_vdev_t",
1789 mg.mg_vd, 0) == -1) {
1790 return (DCMD_ERR);
1791 }
1792
1793 if (vdev.vdev_path != NULL) {
1794 char path[MAXNAMELEN];
1795
1796 if (mdb_readstr(path, sizeof (path),
1797 vdev.vdev_path) == -1) {
1798 mdb_warn("failed to read vdev_path at %p\n",
1799 vdev.vdev_path);
1800 return (DCMD_ERR);
1801 }
1802 char *slash;
1803 if ((slash = strrchr(path, '/')) != NULL) {
1804 strcpy(desc, slash + 1);
1805 } else {
1806 strcpy(desc, path);
1807 }
1808 } else if (vdev.vdev_ops != NULL) {
1809 mdb_vdev_ops_t ops;
1810 if (mdb_ctf_vread(&ops, "vdev_ops_t", "mdb_vdev_ops_t",
1811 vdev.vdev_ops, 0) == -1) {
1812 mdb_warn("failed to read vdev_ops at %p\n",
1813 vdev.vdev_ops);
1814 return (DCMD_ERR);
1815 }
1816 (void) mdb_snprintf(desc, sizeof (desc),
1817 "%s-%llu", ops.vdev_op_type, vdev.vdev_id);
1818 } else {
1819 (void) strcpy(desc, "<unknown>");
1820 }
1821 mdb_printf("%18s\n", desc);
1822 }
1823
1824 return (DCMD_OK);
1825 }
1826
1827 typedef struct metaslab_walk_data {
1828 uint64_t mw_numvdevs;
1829 uintptr_t *mw_vdevs;
1830 int mw_curvdev;
1831 uint64_t mw_nummss;
1832 uintptr_t *mw_mss;
1833 int mw_curms;
1834 } metaslab_walk_data_t;
1835
1836 static int
1837 metaslab_walk_step(mdb_walk_state_t *wsp)
1838 {
1839 metaslab_walk_data_t *mw = wsp->walk_data;
1840 metaslab_t ms;
1841 uintptr_t msp;
1842
1843 if (mw->mw_curvdev >= mw->mw_numvdevs)
1844 return (WALK_DONE);
1845
1846 if (mw->mw_mss == NULL) {
1847 uintptr_t mssp;
1848 uintptr_t vdevp;
1849
1850 ASSERT(mw->mw_curms == 0);
1851 ASSERT(mw->mw_nummss == 0);
1852
1853 vdevp = mw->mw_vdevs[mw->mw_curvdev];
1854 if (GETMEMB(vdevp, "vdev", vdev_ms, mssp) ||
1855 GETMEMB(vdevp, "vdev", vdev_ms_count, mw->mw_nummss)) {
1856 return (WALK_ERR);
1857 }
1858
1859 mw->mw_mss = mdb_alloc(mw->mw_nummss * sizeof (void*),
1860 UM_SLEEP | UM_GC);
1861 if (mdb_vread(mw->mw_mss, mw->mw_nummss * sizeof (void*),
1862 mssp) == -1) {
1863 mdb_warn("failed to read vdev_ms at %p", mssp);
1864 return (WALK_ERR);
1865 }
1866 }
1867
1868 if (mw->mw_curms >= mw->mw_nummss) {
1869 mw->mw_mss = NULL;
1870 mw->mw_curms = 0;
1871 mw->mw_nummss = 0;
1872 mw->mw_curvdev++;
1873 return (WALK_NEXT);
1874 }
1875
1876 msp = mw->mw_mss[mw->mw_curms];
1877 if (mdb_vread(&ms, sizeof (metaslab_t), msp) == -1) {
1878 mdb_warn("failed to read metaslab_t at %p", msp);
1879 return (WALK_ERR);
1880 }
1881
1882 mw->mw_curms++;
1883
1884 return (wsp->walk_callback(msp, &ms, wsp->walk_cbdata));
1885 }
1886
1887 static int
1888 metaslab_walk_init(mdb_walk_state_t *wsp)
1889 {
1890 metaslab_walk_data_t *mw;
1891 uintptr_t root_vdevp;
1892 uintptr_t childp;
1893
1894 if (wsp->walk_addr == NULL) {
1895 mdb_warn("must supply address of spa_t\n");
1896 return (WALK_ERR);
1897 }
1898
1899 mw = mdb_zalloc(sizeof (metaslab_walk_data_t), UM_SLEEP | UM_GC);
1900
1901 if (GETMEMB(wsp->walk_addr, "spa", spa_root_vdev, root_vdevp) ||
1902 GETMEMB(root_vdevp, "vdev", vdev_children, mw->mw_numvdevs) ||
1903 GETMEMB(root_vdevp, "vdev", vdev_child, childp)) {
1904 return (DCMD_ERR);
1905 }
1906
1907 mw->mw_vdevs = mdb_alloc(mw->mw_numvdevs * sizeof (void *),
1908 UM_SLEEP | UM_GC);
1909 if (mdb_vread(mw->mw_vdevs, mw->mw_numvdevs * sizeof (void *),
1910 childp) == -1) {
1911 mdb_warn("failed to read root vdev children at %p", childp);
1912 return (DCMD_ERR);
1913 }
1914
1915 wsp->walk_data = mw;
1916
1917 return (WALK_NEXT);
1918 }
1919
1920 typedef struct mdb_spa {
1921 uintptr_t spa_dsl_pool;
1922 uintptr_t spa_root_vdev;
1923 } mdb_spa_t;
1924
1925 typedef struct mdb_dsl_pool {
1926 uintptr_t dp_root_dir;
1927 } mdb_dsl_pool_t;
1928
1929 typedef struct mdb_dsl_dir {
1930 uintptr_t dd_dbuf;
1931 int64_t dd_space_towrite[TXG_SIZE];
1932 } mdb_dsl_dir_t;
1933
1934 typedef struct mdb_dsl_dir_phys {
1935 uint64_t dd_used_bytes;
1936 uint64_t dd_compressed_bytes;
1937 uint64_t dd_uncompressed_bytes;
1938 } mdb_dsl_dir_phys_t;
1939
1940 typedef struct space_data {
1941 uint64_t ms_allocating[TXG_SIZE];
1942 uint64_t ms_checkpointing;
1943 uint64_t ms_freeing;
1944 uint64_t ms_freed;
1945 uint64_t ms_allocatable;
1946 int64_t ms_deferspace;
1947 uint64_t avail;
1948 uint64_t nowavail;
1949 } space_data_t;
1950
1951 /* ARGSUSED */
1952 static int
1953 space_cb(uintptr_t addr, const void *unknown, void *arg)
1954 {
1955 space_data_t *sd = arg;
1956 mdb_metaslab_t ms;
1957 mdb_range_tree_t rt;
1958 mdb_space_map_t sm = { 0 };
1959 mdb_space_map_phys_t smp = { 0 };
1960 int i;
1961
1962 if (mdb_ctf_vread(&ms, "metaslab_t", "mdb_metaslab_t",
1963 addr, 0) == -1)
1964 return (WALK_ERR);
1965
1966 for (i = 0; i < TXG_SIZE; i++) {
1967 if (mdb_ctf_vread(&rt, "range_tree_t",
1968 "mdb_range_tree_t", ms.ms_allocating[i], 0) == -1)
1969 return (WALK_ERR);
1970
1971 sd->ms_allocating[i] += rt.rt_space;
1972
1973 }
1974
1975 if (mdb_ctf_vread(&rt, "range_tree_t",
1976 "mdb_range_tree_t", ms.ms_checkpointing, 0) == -1)
1977 return (WALK_ERR);
1978 sd->ms_checkpointing += rt.rt_space;
1979
1980 if (mdb_ctf_vread(&rt, "range_tree_t",
1981 "mdb_range_tree_t", ms.ms_freeing, 0) == -1)
1982 return (WALK_ERR);
1983 sd->ms_freeing += rt.rt_space;
1984
1985 if (mdb_ctf_vread(&rt, "range_tree_t",
1986 "mdb_range_tree_t", ms.ms_freed, 0) == -1)
1987 return (WALK_ERR);
1988 sd->ms_freed += rt.rt_space;
1989
1990 if (mdb_ctf_vread(&rt, "range_tree_t",
1991 "mdb_range_tree_t", ms.ms_allocatable, 0) == -1)
1992 return (WALK_ERR);
1993 sd->ms_allocatable += rt.rt_space;
1994
1995 if (ms.ms_sm != NULL &&
1996 mdb_ctf_vread(&sm, "space_map_t",
1997 "mdb_space_map_t", ms.ms_sm, 0) == -1)
1998 return (WALK_ERR);
1999
2000 if (sm.sm_phys != NULL) {
2001 (void) mdb_ctf_vread(&smp, "space_map_phys_t",
2002 "mdb_space_map_phys_t", sm.sm_phys, 0);
2003 }
2004
2005 sd->ms_deferspace += ms.ms_deferspace;
2006 sd->avail += sm.sm_size - sm.sm_alloc;
2007 sd->nowavail += sm.sm_size - smp.smp_alloc;
2008
2009 return (WALK_NEXT);
2010 }
2011
2012 /*
2013 * ::spa_space [-b]
2014 *
2015 * Given a spa_t, print out it's on-disk space usage and in-core
2016 * estimates of future usage. If -b is given, print space in bytes.
2017 * Otherwise print in megabytes.
2018 */
2019 /* ARGSUSED */
2020 static int
2021 spa_space(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2022 {
2023 mdb_spa_t spa;
2024 mdb_dsl_pool_t dp;
2025 mdb_dsl_dir_t dd;
2026 mdb_dmu_buf_impl_t db;
2027 mdb_dsl_dir_phys_t dsp;
2028 space_data_t sd;
2029 int shift = 20;
2030 char *suffix = "M";
2031 int bytes = B_FALSE;
2032
2033 if (mdb_getopts(argc, argv, 'b', MDB_OPT_SETBITS, TRUE, &bytes, NULL) !=
2034 argc)
2035 return (DCMD_USAGE);
2036 if (!(flags & DCMD_ADDRSPEC))
2037 return (DCMD_USAGE);
2038
2039 if (bytes) {
2040 shift = 0;
2041 suffix = "";
2042 }
2043
2044 if (mdb_ctf_vread(&spa, ZFS_STRUCT "spa", "mdb_spa_t",
2045 addr, 0) == -1 ||
2046 mdb_ctf_vread(&dp, ZFS_STRUCT "dsl_pool", "mdb_dsl_pool_t",
2047 spa.spa_dsl_pool, 0) == -1 ||
2048 mdb_ctf_vread(&dd, ZFS_STRUCT "dsl_dir", "mdb_dsl_dir_t",
2049 dp.dp_root_dir, 0) == -1 ||
2050 mdb_ctf_vread(&db, ZFS_STRUCT "dmu_buf_impl", "mdb_dmu_buf_impl_t",
2051 dd.dd_dbuf, 0) == -1 ||
2052 mdb_ctf_vread(&dsp, ZFS_STRUCT "dsl_dir_phys",
2053 "mdb_dsl_dir_phys_t", db.db.db_data, 0) == -1) {
2054 return (DCMD_ERR);
2055 }
2056
2057 mdb_printf("dd_space_towrite = %llu%s %llu%s %llu%s %llu%s\n",
2058 dd.dd_space_towrite[0] >> shift, suffix,
2059 dd.dd_space_towrite[1] >> shift, suffix,
2060 dd.dd_space_towrite[2] >> shift, suffix,
2061 dd.dd_space_towrite[3] >> shift, suffix);
2062
2063 mdb_printf("dd_phys.dd_used_bytes = %llu%s\n",
2064 dsp.dd_used_bytes >> shift, suffix);
2065 mdb_printf("dd_phys.dd_compressed_bytes = %llu%s\n",
2066 dsp.dd_compressed_bytes >> shift, suffix);
2067 mdb_printf("dd_phys.dd_uncompressed_bytes = %llu%s\n",
2068 dsp.dd_uncompressed_bytes >> shift, suffix);
2069
2070 bzero(&sd, sizeof (sd));
2071 if (mdb_pwalk("metaslab", space_cb, &sd, addr) != 0) {
2072 mdb_warn("can't walk metaslabs");
2073 return (DCMD_ERR);
2074 }
2075
2076 mdb_printf("ms_allocmap = %llu%s %llu%s %llu%s %llu%s\n",
2077 sd.ms_allocating[0] >> shift, suffix,
2078 sd.ms_allocating[1] >> shift, suffix,
2079 sd.ms_allocating[2] >> shift, suffix,
2080 sd.ms_allocating[3] >> shift, suffix);
2081 mdb_printf("ms_checkpointing = %llu%s\n",
2082 sd.ms_checkpointing >> shift, suffix);
2083 mdb_printf("ms_freeing = %llu%s\n",
2084 sd.ms_freeing >> shift, suffix);
2085 mdb_printf("ms_freed = %llu%s\n",
2086 sd.ms_freed >> shift, suffix);
2087 mdb_printf("ms_allocatable = %llu%s\n",
2088 sd.ms_allocatable >> shift, suffix);
2089 mdb_printf("ms_deferspace = %llu%s\n",
2090 sd.ms_deferspace >> shift, suffix);
2091 mdb_printf("last synced avail = %llu%s\n",
2092 sd.avail >> shift, suffix);
2093 mdb_printf("current syncing avail = %llu%s\n",
2094 sd.nowavail >> shift, suffix);
2095
2096 return (DCMD_OK);
2097 }
2098
2099 typedef struct mdb_spa_aux_vdev {
2100 int sav_count;
2101 uintptr_t sav_vdevs;
2102 } mdb_spa_aux_vdev_t;
2103
2104 typedef struct mdb_spa_vdevs {
2105 uintptr_t spa_root_vdev;
2106 mdb_spa_aux_vdev_t spa_l2cache;
2107 mdb_spa_aux_vdev_t spa_spares;
2108 } mdb_spa_vdevs_t;
2109
2110 static int
2111 spa_print_aux(mdb_spa_aux_vdev_t *sav, uint_t flags, mdb_arg_t *v,
2112 const char *name)
2113 {
2114 uintptr_t *aux;
2115 size_t len;
2116 int ret, i;
2117
2118 /*
2119 * Iterate over aux vdevs and print those out as well. This is a
2120 * little annoying because we don't have a root vdev to pass to ::vdev.
2121 * Instead, we print a single line and then call it for each child
2122 * vdev.
2123 */
2124 if (sav->sav_count != 0) {
2125 v[1].a_type = MDB_TYPE_STRING;
2126 v[1].a_un.a_str = "-d";
2127 v[2].a_type = MDB_TYPE_IMMEDIATE;
2128 v[2].a_un.a_val = 2;
2129
2130 len = sav->sav_count * sizeof (uintptr_t);
2131 aux = mdb_alloc(len, UM_SLEEP);
2132 if (mdb_vread(aux, len, sav->sav_vdevs) == -1) {
2133 mdb_free(aux, len);
2134 mdb_warn("failed to read l2cache vdevs at %p",
2135 sav->sav_vdevs);
2136 return (DCMD_ERR);
2137 }
2138
2139 mdb_printf("%-?s %-9s %-12s %s\n", "-", "-", "-", name);
2140
2141 for (i = 0; i < sav->sav_count; i++) {
2142 ret = mdb_call_dcmd("vdev", aux[i], flags, 3, v);
2143 if (ret != DCMD_OK) {
2144 mdb_free(aux, len);
2145 return (ret);
2146 }
2147 }
2148
2149 mdb_free(aux, len);
2150 }
2151
2152 return (0);
2153 }
2154
2155 /*
2156 * ::spa_vdevs
2157 *
2158 * -e Include error stats
2159 * -m Include metaslab information
2160 * -M Include metaslab group information
2161 * -h Include histogram information (requires -m or -M)
2162 *
2163 * Print out a summarized list of vdevs for the given spa_t.
2164 * This is accomplished by invoking "::vdev -re" on the root vdev, as well as
2165 * iterating over the cache devices.
2166 */
2167 /* ARGSUSED */
2168 static int
2169 spa_vdevs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2170 {
2171 mdb_arg_t v[3];
2172 int ret;
2173 char opts[100] = "-r";
2174 int spa_flags = 0;
2175
2176 if (mdb_getopts(argc, argv,
2177 'e', MDB_OPT_SETBITS, SPA_FLAG_ERRORS, &spa_flags,
2178 'm', MDB_OPT_SETBITS, SPA_FLAG_METASLABS, &spa_flags,
2179 'M', MDB_OPT_SETBITS, SPA_FLAG_METASLAB_GROUPS, &spa_flags,
2180 'h', MDB_OPT_SETBITS, SPA_FLAG_HISTOGRAMS, &spa_flags,
2181 NULL) != argc)
2182 return (DCMD_USAGE);
2183
2184 if (!(flags & DCMD_ADDRSPEC))
2185 return (DCMD_USAGE);
2186
2187 mdb_spa_vdevs_t spa;
2188 if (mdb_ctf_vread(&spa, "spa_t", "mdb_spa_vdevs_t", addr, 0) == -1)
2189 return (DCMD_ERR);
2190
2191 /*
2192 * Unitialized spa_t structures can have a NULL root vdev.
2193 */
2194 if (spa.spa_root_vdev == NULL) {
2195 mdb_printf("no associated vdevs\n");
2196 return (DCMD_OK);
2197 }
2198
2199 if (spa_flags & SPA_FLAG_ERRORS)
2200 strcat(opts, "e");
2201 if (spa_flags & SPA_FLAG_METASLABS)
2202 strcat(opts, "m");
2203 if (spa_flags & SPA_FLAG_METASLAB_GROUPS)
2204 strcat(opts, "M");
2205 if (spa_flags & SPA_FLAG_HISTOGRAMS)
2206 strcat(opts, "h");
2207
2208 v[0].a_type = MDB_TYPE_STRING;
2209 v[0].a_un.a_str = opts;
2210
2211 ret = mdb_call_dcmd("vdev", (uintptr_t)spa.spa_root_vdev,
2212 flags, 1, v);
2213 if (ret != DCMD_OK)
2214 return (ret);
2215
2216 if (spa_print_aux(&spa.spa_l2cache, flags, v, "cache") != 0 ||
2217 spa_print_aux(&spa.spa_spares, flags, v, "spares") != 0)
2218 return (DCMD_ERR);
2219
2220 return (DCMD_OK);
2221 }
2222
2223 /*
2224 * ::zio
2225 *
2226 * Print a summary of zio_t and all its children. This is intended to display a
2227 * zio tree, and hence we only pick the most important pieces of information for
2228 * the main summary. More detailed information can always be found by doing a
2229 * '::print zio' on the underlying zio_t. The columns we display are:
2230 *
2231 * ADDRESS TYPE STAGE WAITER TIME_ELAPSED
2232 *
2233 * The 'address' column is indented by one space for each depth level as we
2234 * descend down the tree.
2235 */
2236
2237 #define ZIO_MAXINDENT 7
2238 #define ZIO_MAXWIDTH (sizeof (uintptr_t) * 2 + ZIO_MAXINDENT)
2239 #define ZIO_WALK_SELF 0
2240 #define ZIO_WALK_CHILD 1
2241 #define ZIO_WALK_PARENT 2
2242
2243 typedef struct zio_print_args {
2244 int zpa_current_depth;
2245 int zpa_min_depth;
2246 int zpa_max_depth;
2247 int zpa_type;
2248 uint_t zpa_flags;
2249 } zio_print_args_t;
2250
2251 typedef struct mdb_zio {
2252 enum zio_type io_type;
2253 enum zio_stage io_stage;
2254 uintptr_t io_waiter;
2255 uintptr_t io_spa;
2256 struct {
2257 struct {
2258 uintptr_t list_next;
2259 } list_head;
2260 } io_parent_list;
2261 int io_error;
2262 } mdb_zio_t;
2263
2264 typedef struct mdb_zio_timestamp {
2265 hrtime_t io_timestamp;
2266 } mdb_zio_timestamp_t;
2267
2268 static int zio_child_cb(uintptr_t addr, const void *unknown, void *arg);
2269
2270 static int
2271 zio_print_cb(uintptr_t addr, zio_print_args_t *zpa)
2272 {
2273 mdb_ctf_id_t type_enum, stage_enum;
2274 int indent = zpa->zpa_current_depth;
2275 const char *type, *stage;
2276 uintptr_t laddr;
2277 mdb_zio_t zio;
2278 mdb_zio_timestamp_t zio_timestamp = { 0 };
2279
2280 if (mdb_ctf_vread(&zio, ZFS_STRUCT "zio", "mdb_zio_t", addr, 0) == -1)
2281 return (WALK_ERR);
2282 (void) mdb_ctf_vread(&zio_timestamp, ZFS_STRUCT "zio",
2283 "mdb_zio_timestamp_t", addr, MDB_CTF_VREAD_QUIET);
2284
2285 if (indent > ZIO_MAXINDENT)
2286 indent = ZIO_MAXINDENT;
2287
2288 if (mdb_ctf_lookup_by_name("enum zio_type", &type_enum) == -1 ||
2289 mdb_ctf_lookup_by_name("enum zio_stage", &stage_enum) == -1) {
2290 mdb_warn("failed to lookup zio enums");
2291 return (WALK_ERR);
2292 }
2293
2294 if ((type = mdb_ctf_enum_name(type_enum, zio.io_type)) != NULL)
2295 type += sizeof ("ZIO_TYPE_") - 1;
2296 else
2297 type = "?";
2298
2299 if (zio.io_error == 0) {
2300 stage = mdb_ctf_enum_name(stage_enum, zio.io_stage);
2301 if (stage != NULL)
2302 stage += sizeof ("ZIO_STAGE_") - 1;
2303 else
2304 stage = "?";
2305 } else {
2306 stage = "FAILED";
2307 }
2308
2309 if (zpa->zpa_current_depth >= zpa->zpa_min_depth) {
2310 if (zpa->zpa_flags & DCMD_PIPE_OUT) {
2311 mdb_printf("%?p\n", addr);
2312 } else {
2313 mdb_printf("%*s%-*p %-5s %-16s ", indent, "",
2314 ZIO_MAXWIDTH - indent, addr, type, stage);
2315 if (zio.io_waiter != 0)
2316 mdb_printf("%-16lx ", zio.io_waiter);
2317 else
2318 mdb_printf("%-16s ", "-");
2319 #ifdef _KERNEL
2320 if (zio_timestamp.io_timestamp != 0) {
2321 mdb_printf("%llums", (mdb_gethrtime() -
2322 zio_timestamp.io_timestamp) /
2323 1000000);
2324 } else {
2325 mdb_printf("%-12s ", "-");
2326 }
2327 #else
2328 mdb_printf("%-12s ", "-");
2329 #endif
2330 mdb_printf("\n");
2331 }
2332 }
2333
2334 if (zpa->zpa_current_depth >= zpa->zpa_max_depth)
2335 return (WALK_NEXT);
2336
2337 if (zpa->zpa_type == ZIO_WALK_PARENT)
2338 laddr = addr + mdb_ctf_offsetof_by_name(ZFS_STRUCT "zio",
2339 "io_parent_list");
2340 else
2341 laddr = addr + mdb_ctf_offsetof_by_name(ZFS_STRUCT "zio",
2342 "io_child_list");
2343
2344 zpa->zpa_current_depth++;
2345 if (mdb_pwalk("list", zio_child_cb, zpa, laddr) != 0) {
2346 mdb_warn("failed to walk zio_t children at %p\n", laddr);
2347 return (WALK_ERR);
2348 }
2349 zpa->zpa_current_depth--;
2350
2351 return (WALK_NEXT);
2352 }
2353
2354 /* ARGSUSED */
2355 static int
2356 zio_child_cb(uintptr_t addr, const void *unknown, void *arg)
2357 {
2358 zio_link_t zl;
2359 uintptr_t ziop;
2360 zio_print_args_t *zpa = arg;
2361
2362 if (mdb_vread(&zl, sizeof (zl), addr) == -1) {
2363 mdb_warn("failed to read zio_link_t at %p", addr);
2364 return (WALK_ERR);
2365 }
2366
2367 if (zpa->zpa_type == ZIO_WALK_PARENT)
2368 ziop = (uintptr_t)zl.zl_parent;
2369 else
2370 ziop = (uintptr_t)zl.zl_child;
2371
2372 return (zio_print_cb(ziop, zpa));
2373 }
2374
2375 /* ARGSUSED */
2376 static int
2377 zio_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2378 {
2379 zio_print_args_t zpa = { 0 };
2380
2381 if (!(flags & DCMD_ADDRSPEC))
2382 return (DCMD_USAGE);
2383
2384 if (mdb_getopts(argc, argv,
2385 'r', MDB_OPT_SETBITS, INT_MAX, &zpa.zpa_max_depth,
2386 'c', MDB_OPT_SETBITS, ZIO_WALK_CHILD, &zpa.zpa_type,
2387 'p', MDB_OPT_SETBITS, ZIO_WALK_PARENT, &zpa.zpa_type,
2388 NULL) != argc)
2389 return (DCMD_USAGE);
2390
2391 zpa.zpa_flags = flags;
2392 if (zpa.zpa_max_depth != 0) {
2393 if (zpa.zpa_type == ZIO_WALK_SELF)
2394 zpa.zpa_type = ZIO_WALK_CHILD;
2395 } else if (zpa.zpa_type != ZIO_WALK_SELF) {
2396 zpa.zpa_min_depth = 1;
2397 zpa.zpa_max_depth = 1;
2398 }
2399
2400 if (!(flags & DCMD_PIPE_OUT) && DCMD_HDRSPEC(flags)) {
2401 mdb_printf("%<u>%-*s %-5s %-16s %-16s %-12s%</u>\n",
2402 ZIO_MAXWIDTH, "ADDRESS", "TYPE", "STAGE", "WAITER",
2403 "TIME_ELAPSED");
2404 }
2405
2406 if (zio_print_cb(addr, &zpa) != WALK_NEXT)
2407 return (DCMD_ERR);
2408
2409 return (DCMD_OK);
2410 }
2411
2412 /*
2413 * [addr]::zio_state
2414 *
2415 * Print a summary of all zio_t structures on the system, or for a particular
2416 * pool. This is equivalent to '::walk zio_root | ::zio'.
2417 */
2418 /*ARGSUSED*/
2419 static int
2420 zio_state(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2421 {
2422 /*
2423 * MDB will remember the last address of the pipeline, so if we don't
2424 * zero this we'll end up trying to walk zio structures for a
2425 * non-existent spa_t.
2426 */
2427 if (!(flags & DCMD_ADDRSPEC))
2428 addr = 0;
2429
2430 return (mdb_pwalk_dcmd("zio_root", "zio", argc, argv, addr));
2431 }
2432
2433 typedef struct mdb_multilist {
2434 uint64_t ml_num_sublists;
2435 uintptr_t ml_sublists;
2436 } mdb_multilist_t;
2437
2438 typedef struct multilist_walk_data {
2439 uint64_t mwd_idx;
2440 mdb_multilist_t mwd_ml;
2441 } multilist_walk_data_t;
2442
2443 /* ARGSUSED */
2444 static int
2445 multilist_print_cb(uintptr_t addr, const void *unknown, void *arg)
2446 {
2447 mdb_printf("%#lr\n", addr);
2448 return (WALK_NEXT);
2449 }
2450
2451 static int
2452 multilist_walk_step(mdb_walk_state_t *wsp)
2453 {
2454 multilist_walk_data_t *mwd = wsp->walk_data;
2455
2456 if (mwd->mwd_idx >= mwd->mwd_ml.ml_num_sublists)
2457 return (WALK_DONE);
2458
2459 wsp->walk_addr = mwd->mwd_ml.ml_sublists +
2460 mdb_ctf_sizeof_by_name("multilist_sublist_t") * mwd->mwd_idx +
2461 mdb_ctf_offsetof_by_name("multilist_sublist_t", "mls_list");
2462
2463 mdb_pwalk("list", multilist_print_cb, (void*)NULL, wsp->walk_addr);
2464 mwd->mwd_idx++;
2465
2466 return (WALK_NEXT);
2467 }
2468
2469 static int
2470 multilist_walk_init(mdb_walk_state_t *wsp)
2471 {
2472 multilist_walk_data_t *mwd;
2473
2474 if (wsp->walk_addr == NULL) {
2475 mdb_warn("must supply address of multilist_t\n");
2476 return (WALK_ERR);
2477 }
2478
2479 mwd = mdb_zalloc(sizeof (multilist_walk_data_t), UM_SLEEP | UM_GC);
2480 if (mdb_ctf_vread(&mwd->mwd_ml, "multilist_t", "mdb_multilist_t",
2481 wsp->walk_addr, 0) == -1) {
2482 return (WALK_ERR);
2483 }
2484
2485 if (mwd->mwd_ml.ml_num_sublists == 0 ||
2486 mwd->mwd_ml.ml_sublists == NULL) {
2487 mdb_warn("invalid or uninitialized multilist at %#lx\n",
2488 wsp->walk_addr);
2489 return (WALK_ERR);
2490 }
2491
2492 wsp->walk_data = mwd;
2493 return (WALK_NEXT);
2494 }
2495
2496 typedef struct mdb_txg_list {
2497 size_t tl_offset;
2498 uintptr_t tl_head[TXG_SIZE];
2499 } mdb_txg_list_t;
2500
2501 typedef struct txg_list_walk_data {
2502 uintptr_t lw_head[TXG_SIZE];
2503 int lw_txgoff;
2504 int lw_maxoff;
2505 size_t lw_offset;
2506 void *lw_obj;
2507 } txg_list_walk_data_t;
2508
2509 static int
2510 txg_list_walk_init_common(mdb_walk_state_t *wsp, int txg, int maxoff)
2511 {
2512 txg_list_walk_data_t *lwd;
2513 mdb_txg_list_t list;
2514 int i;
2515
2516 lwd = mdb_alloc(sizeof (txg_list_walk_data_t), UM_SLEEP | UM_GC);
2517 if (mdb_ctf_vread(&list, "txg_list_t", "mdb_txg_list_t", wsp->walk_addr,
2518 0) == -1) {
2519 mdb_warn("failed to read txg_list_t at %#lx", wsp->walk_addr);
2520 return (WALK_ERR);
2521 }
2522
2523 for (i = 0; i < TXG_SIZE; i++)
2524 lwd->lw_head[i] = list.tl_head[i];
2525 lwd->lw_offset = list.tl_offset;
2526 lwd->lw_obj = mdb_alloc(lwd->lw_offset + sizeof (txg_node_t),
2527 UM_SLEEP | UM_GC);
2528 lwd->lw_txgoff = txg;
2529 lwd->lw_maxoff = maxoff;
2530
2531 wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
2532 wsp->walk_data = lwd;
2533
2534 return (WALK_NEXT);
2535 }
2536
2537 static int
2538 txg_list_walk_init(mdb_walk_state_t *wsp)
2539 {
2540 return (txg_list_walk_init_common(wsp, 0, TXG_SIZE-1));
2541 }
2542
2543 static int
2544 txg_list0_walk_init(mdb_walk_state_t *wsp)
2545 {
2546 return (txg_list_walk_init_common(wsp, 0, 0));
2547 }
2548
2549 static int
2550 txg_list1_walk_init(mdb_walk_state_t *wsp)
2551 {
2552 return (txg_list_walk_init_common(wsp, 1, 1));
2553 }
2554
2555 static int
2556 txg_list2_walk_init(mdb_walk_state_t *wsp)
2557 {
2558 return (txg_list_walk_init_common(wsp, 2, 2));
2559 }
2560
2561 static int
2562 txg_list3_walk_init(mdb_walk_state_t *wsp)
2563 {
2564 return (txg_list_walk_init_common(wsp, 3, 3));
2565 }
2566
2567 static int
2568 txg_list_walk_step(mdb_walk_state_t *wsp)
2569 {
2570 txg_list_walk_data_t *lwd = wsp->walk_data;
2571 uintptr_t addr;
2572 txg_node_t *node;
2573 int status;
2574
2575 while (wsp->walk_addr == NULL && lwd->lw_txgoff < lwd->lw_maxoff) {
2576 lwd->lw_txgoff++;
2577 wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
2578 }
2579
2580 if (wsp->walk_addr == NULL)
2581 return (WALK_DONE);
2582
2583 addr = wsp->walk_addr - lwd->lw_offset;
2584
2585 if (mdb_vread(lwd->lw_obj,
2586 lwd->lw_offset + sizeof (txg_node_t), addr) == -1) {
2587 mdb_warn("failed to read list element at %#lx", addr);
2588 return (WALK_ERR);
2589 }
2590
2591 status = wsp->walk_callback(addr, lwd->lw_obj, wsp->walk_cbdata);
2592 node = (txg_node_t *)((uintptr_t)lwd->lw_obj + lwd->lw_offset);
2593 wsp->walk_addr = (uintptr_t)node->tn_next[lwd->lw_txgoff];
2594
2595 return (status);
2596 }
2597
2598 /*
2599 * ::walk spa
2600 *
2601 * Walk all named spa_t structures in the namespace. This is nothing more than
2602 * a layered avl walk.
2603 */
2604 static int
2605 spa_walk_init(mdb_walk_state_t *wsp)
2606 {
2607 GElf_Sym sym;
2608
2609 if (wsp->walk_addr != NULL) {
2610 mdb_warn("spa walk only supports global walks\n");
2611 return (WALK_ERR);
2612 }
2613
2614 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "spa_namespace_avl", &sym) == -1) {
2615 mdb_warn("failed to find symbol 'spa_namespace_avl'");
2616 return (WALK_ERR);
2617 }
2618
2619 wsp->walk_addr = (uintptr_t)sym.st_value;
2620
2621 if (mdb_layered_walk("avl", wsp) == -1) {
2622 mdb_warn("failed to walk 'avl'\n");
2623 return (WALK_ERR);
2624 }
2625
2626 return (WALK_NEXT);
2627 }
2628
2629 static int
2630 spa_walk_step(mdb_walk_state_t *wsp)
2631 {
2632 return (wsp->walk_callback(wsp->walk_addr, NULL, wsp->walk_cbdata));
2633 }
2634
2635 /*
2636 * [addr]::walk zio
2637 *
2638 * Walk all active zio_t structures on the system. This is simply a layered
2639 * walk on top of ::walk zio_cache, with the optional ability to limit the
2640 * structures to a particular pool.
2641 */
2642 static int
2643 zio_walk_init(mdb_walk_state_t *wsp)
2644 {
2645 wsp->walk_data = (void *)wsp->walk_addr;
2646
2647 if (mdb_layered_walk("zio_cache", wsp) == -1) {
2648 mdb_warn("failed to walk 'zio_cache'\n");
2649 return (WALK_ERR);
2650 }
2651
2652 return (WALK_NEXT);
2653 }
2654
2655 static int
2656 zio_walk_step(mdb_walk_state_t *wsp)
2657 {
2658 mdb_zio_t zio;
2659 uintptr_t spa = (uintptr_t)wsp->walk_data;
2660
2661 if (mdb_ctf_vread(&zio, ZFS_STRUCT "zio", "mdb_zio_t",
2662 wsp->walk_addr, 0) == -1)
2663 return (WALK_ERR);
2664
2665 if (spa != 0 && spa != zio.io_spa)
2666 return (WALK_NEXT);
2667
2668 return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
2669 }
2670
2671 /*
2672 * [addr]::walk zio_root
2673 *
2674 * Walk only root zio_t structures, optionally for a particular spa_t.
2675 */
2676 static int
2677 zio_walk_root_step(mdb_walk_state_t *wsp)
2678 {
2679 mdb_zio_t zio;
2680 uintptr_t spa = (uintptr_t)wsp->walk_data;
2681
2682 if (mdb_ctf_vread(&zio, ZFS_STRUCT "zio", "mdb_zio_t",
2683 wsp->walk_addr, 0) == -1)
2684 return (WALK_ERR);
2685
2686 if (spa != 0 && spa != zio.io_spa)
2687 return (WALK_NEXT);
2688
2689 /* If the parent list is not empty, ignore */
2690 if (zio.io_parent_list.list_head.list_next !=
2691 wsp->walk_addr +
2692 mdb_ctf_offsetof_by_name(ZFS_STRUCT "zio", "io_parent_list") +
2693 mdb_ctf_offsetof_by_name("struct list", "list_head"))
2694 return (WALK_NEXT);
2695
2696 return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
2697 }
2698
2699 /*
2700 * ::zfs_blkstats
2701 *
2702 * -v print verbose per-level information
2703 *
2704 */
2705 static int
2706 zfs_blkstats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2707 {
2708 boolean_t verbose = B_FALSE;
2709 zfs_all_blkstats_t stats;
2710 dmu_object_type_t t;
2711 zfs_blkstat_t *tzb;
2712 uint64_t ditto;
2713 dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES + 10];
2714 /* +10 in case it grew */
2715
2716 if (mdb_readvar(&dmu_ot, "dmu_ot") == -1) {
2717 mdb_warn("failed to read 'dmu_ot'");
2718 return (DCMD_ERR);
2719 }
2720
2721 if (mdb_getopts(argc, argv,
2722 'v', MDB_OPT_SETBITS, TRUE, &verbose,
2723 NULL) != argc)
2724 return (DCMD_USAGE);
2725
2726 if (!(flags & DCMD_ADDRSPEC))
2727 return (DCMD_USAGE);
2728
2729 if (GETMEMB(addr, "spa", spa_dsl_pool, addr) ||
2730 GETMEMB(addr, "dsl_pool", dp_blkstats, addr) ||
2731 mdb_vread(&stats, sizeof (zfs_all_blkstats_t), addr) == -1) {
2732 mdb_warn("failed to read data at %p;", addr);
2733 mdb_printf("maybe no stats? run \"zpool scrub\" first.");
2734 return (DCMD_ERR);
2735 }
2736
2737 tzb = &stats.zab_type[DN_MAX_LEVELS][DMU_OT_TOTAL];
2738 if (tzb->zb_gangs != 0) {
2739 mdb_printf("Ganged blocks: %llu\n",
2740 (longlong_t)tzb->zb_gangs);
2741 }
2742
2743 ditto = tzb->zb_ditto_2_of_2_samevdev + tzb->zb_ditto_2_of_3_samevdev +
2744 tzb->zb_ditto_3_of_3_samevdev;
2745 if (ditto != 0) {
2746 mdb_printf("Dittoed blocks on same vdev: %llu\n",
2747 (longlong_t)ditto);
2748 }
2749
2750 mdb_printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
2751 "\t avg\t comp\t%%Total\tType\n");
2752
2753 for (t = 0; t <= DMU_OT_TOTAL; t++) {
2754 char csize[NICENUM_BUFLEN], lsize[NICENUM_BUFLEN];
2755 char psize[NICENUM_BUFLEN], asize[NICENUM_BUFLEN];
2756 char avg[NICENUM_BUFLEN];
2757 char comp[NICENUM_BUFLEN], pct[NICENUM_BUFLEN];
2758 char typename[64];
2759 int l;
2760
2761
2762 if (t == DMU_OT_DEFERRED)
2763 strcpy(typename, "deferred free");
2764 else if (t == DMU_OT_OTHER)
2765 strcpy(typename, "other");
2766 else if (t == DMU_OT_TOTAL)
2767 strcpy(typename, "Total");
2768 else if (mdb_readstr(typename, sizeof (typename),
2769 (uintptr_t)dmu_ot[t].ot_name) == -1) {
2770 mdb_warn("failed to read type name");
2771 return (DCMD_ERR);
2772 }
2773
2774 if (stats.zab_type[DN_MAX_LEVELS][t].zb_asize == 0)
2775 continue;
2776
2777 for (l = -1; l < DN_MAX_LEVELS; l++) {
2778 int level = (l == -1 ? DN_MAX_LEVELS : l);
2779 zfs_blkstat_t *zb = &stats.zab_type[level][t];
2780
2781 if (zb->zb_asize == 0)
2782 continue;
2783
2784 /*
2785 * Don't print each level unless requested.
2786 */
2787 if (!verbose && level != DN_MAX_LEVELS)
2788 continue;
2789
2790 /*
2791 * If all the space is level 0, don't print the
2792 * level 0 separately.
2793 */
2794 if (level == 0 && zb->zb_asize ==
2795 stats.zab_type[DN_MAX_LEVELS][t].zb_asize)
2796 continue;
2797
2798 mdb_nicenum(zb->zb_count, csize);
2799 mdb_nicenum(zb->zb_lsize, lsize);
2800 mdb_nicenum(zb->zb_psize, psize);
2801 mdb_nicenum(zb->zb_asize, asize);
2802 mdb_nicenum(zb->zb_asize / zb->zb_count, avg);
2803 (void) snprintfrac(comp, NICENUM_BUFLEN,
2804 zb->zb_lsize, zb->zb_psize, 2);
2805 (void) snprintfrac(pct, NICENUM_BUFLEN,
2806 100 * zb->zb_asize, tzb->zb_asize, 2);
2807
2808 mdb_printf("%6s\t%5s\t%5s\t%5s\t%5s"
2809 "\t%5s\t%6s\t",
2810 csize, lsize, psize, asize, avg, comp, pct);
2811
2812 if (level == DN_MAX_LEVELS)
2813 mdb_printf("%s\n", typename);
2814 else
2815 mdb_printf(" L%d %s\n",
2816 level, typename);
2817 }
2818 }
2819
2820 return (DCMD_OK);
2821 }
2822
2823 typedef struct mdb_reference {
2824 uintptr_t ref_holder;
2825 uintptr_t ref_removed;
2826 uint64_t ref_number;
2827 } mdb_reference_t;
2828
2829 /* ARGSUSED */
2830 static int
2831 reference_cb(uintptr_t addr, const void *ignored, void *arg)
2832 {
2833 mdb_reference_t ref;
2834 boolean_t holder_is_str = B_FALSE;
2835 char holder_str[128];
2836 boolean_t removed = (boolean_t)arg;
2837
2838 if (mdb_ctf_vread(&ref, "reference_t", "mdb_reference_t", addr,
2839 0) == -1)
2840 return (DCMD_ERR);
2841
2842 if (mdb_readstr(holder_str, sizeof (holder_str),
2843 ref.ref_holder) != -1)
2844 holder_is_str = strisprint(holder_str);
2845
2846 if (removed)
2847 mdb_printf("removed ");
2848 mdb_printf("reference ");
2849 if (ref.ref_number != 1)
2850 mdb_printf("with count=%llu ", ref.ref_number);
2851 mdb_printf("with tag %lx", ref.ref_holder);
2852 if (holder_is_str)
2853 mdb_printf(" \"%s\"", holder_str);
2854 mdb_printf(", held at:\n");
2855
2856 (void) mdb_call_dcmd("whatis", addr, DCMD_ADDRSPEC, 0, NULL);
2857
2858 if (removed) {
2859 mdb_printf("removed at:\n");
2860 (void) mdb_call_dcmd("whatis", ref.ref_removed,
2861 DCMD_ADDRSPEC, 0, NULL);
2862 }
2863
2864 mdb_printf("\n");
2865
2866 return (WALK_NEXT);
2867 }
2868
2869 typedef struct mdb_refcount {
2870 uint64_t rc_count;
2871 } mdb_refcount_t;
2872
2873 typedef struct mdb_refcount_removed {
2874 uint64_t rc_removed_count;
2875 } mdb_refcount_removed_t;
2876
2877 typedef struct mdb_refcount_tracked {
2878 boolean_t rc_tracked;
2879 } mdb_refcount_tracked_t;
2880
2881 /* ARGSUSED */
2882 static int
2883 refcount(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2884 {
2885 mdb_refcount_t rc;
2886 mdb_refcount_removed_t rcr;
2887 mdb_refcount_tracked_t rct;
2888 int off;
2889 boolean_t released = B_FALSE;
2890
2891 if (!(flags & DCMD_ADDRSPEC))
2892 return (DCMD_USAGE);
2893
2894 if (mdb_getopts(argc, argv,
2895 'r', MDB_OPT_SETBITS, B_TRUE, &released,
2896 NULL) != argc)
2897 return (DCMD_USAGE);
2898
2899 if (mdb_ctf_vread(&rc, "refcount_t", "mdb_refcount_t", addr,
2900 0) == -1)
2901 return (DCMD_ERR);
2902
2903 if (mdb_ctf_vread(&rcr, "refcount_t", "mdb_refcount_removed_t", addr,
2904 MDB_CTF_VREAD_QUIET) == -1) {
2905 mdb_printf("refcount_t at %p has %llu holds (untracked)\n",
2906 addr, (longlong_t)rc.rc_count);
2907 return (DCMD_OK);
2908 }
2909
2910 if (mdb_ctf_vread(&rct, "refcount_t", "mdb_refcount_tracked_t", addr,
2911 MDB_CTF_VREAD_QUIET) == -1) {
2912 /* If this is an old target, it might be tracked. */
2913 rct.rc_tracked = B_TRUE;
2914 }
2915
2916 mdb_printf("refcount_t at %p has %llu current holds, "
2917 "%llu recently released holds\n",
2918 addr, (longlong_t)rc.rc_count, (longlong_t)rcr.rc_removed_count);
2919
2920 if (rct.rc_tracked && rc.rc_count > 0)
2921 mdb_printf("current holds:\n");
2922 off = mdb_ctf_offsetof_by_name("refcount_t", "rc_list");
2923 if (off == -1)
2924 return (DCMD_ERR);
2925 mdb_pwalk("list", reference_cb, (void*)B_FALSE, addr + off);
2926
2927 if (released && rcr.rc_removed_count > 0) {
2928 mdb_printf("released holds:\n");
2929
2930 off = mdb_ctf_offsetof_by_name("refcount_t", "rc_removed");
2931 if (off == -1)
2932 return (DCMD_ERR);
2933 mdb_pwalk("list", reference_cb, (void*)B_TRUE, addr + off);
2934 }
2935
2936 return (DCMD_OK);
2937 }
2938
2939 /* ARGSUSED */
2940 static int
2941 sa_attr_table(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2942 {
2943 sa_attr_table_t *table;
2944 sa_os_t sa_os;
2945 char *name;
2946 int i;
2947
2948 if (mdb_vread(&sa_os, sizeof (sa_os_t), addr) == -1) {
2949 mdb_warn("failed to read sa_os at %p", addr);
2950 return (DCMD_ERR);
2951 }
2952
2953 table = mdb_alloc(sizeof (sa_attr_table_t) * sa_os.sa_num_attrs,
2954 UM_SLEEP | UM_GC);
2955 name = mdb_alloc(MAXPATHLEN, UM_SLEEP | UM_GC);
2956
2957 if (mdb_vread(table, sizeof (sa_attr_table_t) * sa_os.sa_num_attrs,
2958 (uintptr_t)sa_os.sa_attr_table) == -1) {
2959 mdb_warn("failed to read sa_os at %p", addr);
2960 return (DCMD_ERR);
2961 }
2962
2963 mdb_printf("%<u>%-10s %-10s %-10s %-10s %s%</u>\n",
2964 "ATTR ID", "REGISTERED", "LENGTH", "BSWAP", "NAME");
2965 for (i = 0; i != sa_os.sa_num_attrs; i++) {
2966 mdb_readstr(name, MAXPATHLEN, (uintptr_t)table[i].sa_name);
2967 mdb_printf("%5x %8x %8x %8x %-s\n",
2968 (int)table[i].sa_attr, (int)table[i].sa_registered,
2969 (int)table[i].sa_length, table[i].sa_byteswap, name);
2970 }
2971
2972 return (DCMD_OK);
2973 }
2974
2975 static int
2976 sa_get_off_table(uintptr_t addr, uint32_t **off_tab, int attr_count)
2977 {
2978 uintptr_t idx_table;
2979
2980 if (GETMEMB(addr, "sa_idx_tab", sa_idx_tab, idx_table)) {
2981 mdb_printf("can't find offset table in sa_idx_tab\n");
2982 return (-1);
2983 }
2984
2985 *off_tab = mdb_alloc(attr_count * sizeof (uint32_t),
2986 UM_SLEEP | UM_GC);
2987
2988 if (mdb_vread(*off_tab,
2989 attr_count * sizeof (uint32_t), idx_table) == -1) {
2990 mdb_warn("failed to attribute offset table %p", idx_table);
2991 return (-1);
2992 }
2993
2994 return (DCMD_OK);
2995 }
2996
2997 /*ARGSUSED*/
2998 static int
2999 sa_attr_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3000 {
3001 uint32_t *offset_tab;
3002 int attr_count;
3003 uint64_t attr_id;
3004 uintptr_t attr_addr;
3005 uintptr_t bonus_tab, spill_tab;
3006 uintptr_t db_bonus, db_spill;
3007 uintptr_t os, os_sa;
3008 uintptr_t db_data;
3009
3010 if (argc != 1)
3011 return (DCMD_USAGE);
3012
3013 if (argv[0].a_type == MDB_TYPE_STRING)
3014 attr_id = mdb_strtoull(argv[0].a_un.a_str);
3015 else
3016 return (DCMD_USAGE);
3017
3018 if (GETMEMB(addr, "sa_handle", sa_bonus_tab, bonus_tab) ||
3019 GETMEMB(addr, "sa_handle", sa_spill_tab, spill_tab) ||
3020 GETMEMB(addr, "sa_handle", sa_os, os) ||
3021 GETMEMB(addr, "sa_handle", sa_bonus, db_bonus) ||
3022 GETMEMB(addr, "sa_handle", sa_spill, db_spill)) {
3023 mdb_printf("Can't find necessary information in sa_handle "
3024 "in sa_handle\n");
3025 return (DCMD_ERR);
3026 }
3027
3028 if (GETMEMB(os, "objset", os_sa, os_sa)) {
3029 mdb_printf("Can't find os_sa in objset\n");
3030 return (DCMD_ERR);
3031 }
3032
3033 if (GETMEMB(os_sa, "sa_os", sa_num_attrs, attr_count)) {
3034 mdb_printf("Can't find sa_num_attrs\n");
3035 return (DCMD_ERR);
3036 }
3037
3038 if (attr_id > attr_count) {
3039 mdb_printf("attribute id number is out of range\n");
3040 return (DCMD_ERR);
3041 }
3042
3043 if (bonus_tab) {
3044 if (sa_get_off_table(bonus_tab, &offset_tab,
3045 attr_count) == -1) {
3046 return (DCMD_ERR);
3047 }
3048
3049 if (GETMEMB(db_bonus, "dmu_buf", db_data, db_data)) {
3050 mdb_printf("can't find db_data in bonus dbuf\n");
3051 return (DCMD_ERR);
3052 }
3053 }
3054
3055 if (bonus_tab && !TOC_ATTR_PRESENT(offset_tab[attr_id]) &&
3056 spill_tab == NULL) {
3057 mdb_printf("Attribute does not exist\n");
3058 return (DCMD_ERR);
3059 } else if (!TOC_ATTR_PRESENT(offset_tab[attr_id]) && spill_tab) {
3060 if (sa_get_off_table(spill_tab, &offset_tab,
3061 attr_count) == -1) {
3062 return (DCMD_ERR);
3063 }
3064 if (GETMEMB(db_spill, "dmu_buf", db_data, db_data)) {
3065 mdb_printf("can't find db_data in spill dbuf\n");
3066 return (DCMD_ERR);
3067 }
3068 if (!TOC_ATTR_PRESENT(offset_tab[attr_id])) {
3069 mdb_printf("Attribute does not exist\n");
3070 return (DCMD_ERR);
3071 }
3072 }
3073 attr_addr = db_data + TOC_OFF(offset_tab[attr_id]);
3074 mdb_printf("%p\n", attr_addr);
3075 return (DCMD_OK);
3076 }
3077
3078 /* ARGSUSED */
3079 static int
3080 zfs_ace_print_common(uintptr_t addr, uint_t flags,
3081 uint64_t id, uint32_t access_mask, uint16_t ace_flags,
3082 uint16_t ace_type, int verbose)
3083 {
3084 if (DCMD_HDRSPEC(flags) && !verbose)
3085 mdb_printf("%<u>%-?s %-8s %-8s %-8s %s%</u>\n",
3086 "ADDR", "FLAGS", "MASK", "TYPE", "ID");
3087
3088 if (!verbose) {
3089 mdb_printf("%0?p %-8x %-8x %-8x %-llx\n", addr,
3090 ace_flags, access_mask, ace_type, id);
3091 return (DCMD_OK);
3092 }
3093
3094 switch (ace_flags & ACE_TYPE_FLAGS) {
3095 case ACE_OWNER:
3096 mdb_printf("owner@:");
3097 break;
3098 case (ACE_IDENTIFIER_GROUP | ACE_GROUP):
3099 mdb_printf("group@:");
3100 break;
3101 case ACE_EVERYONE:
3102 mdb_printf("everyone@:");
3103 break;
3104 case ACE_IDENTIFIER_GROUP:
3105 mdb_printf("group:%llx:", (u_longlong_t)id);
3106 break;
3107 case 0: /* User entry */
3108 mdb_printf("user:%llx:", (u_longlong_t)id);
3109 break;
3110 }
3111
3112 /* print out permission mask */
3113 if (access_mask & ACE_READ_DATA)
3114 mdb_printf("r");
3115 else
3116 mdb_printf("-");
3117 if (access_mask & ACE_WRITE_DATA)
3118 mdb_printf("w");
3119 else
3120 mdb_printf("-");
3121 if (access_mask & ACE_EXECUTE)
3122 mdb_printf("x");
3123 else
3124 mdb_printf("-");
3125 if (access_mask & ACE_APPEND_DATA)
3126 mdb_printf("p");
3127 else
3128 mdb_printf("-");
3129 if (access_mask & ACE_DELETE)
3130 mdb_printf("d");
3131 else
3132 mdb_printf("-");
3133 if (access_mask & ACE_DELETE_CHILD)
3134 mdb_printf("D");
3135 else
3136 mdb_printf("-");
3137 if (access_mask & ACE_READ_ATTRIBUTES)
3138 mdb_printf("a");
3139 else
3140 mdb_printf("-");
3141 if (access_mask & ACE_WRITE_ATTRIBUTES)
3142 mdb_printf("A");
3143 else
3144 mdb_printf("-");
3145 if (access_mask & ACE_READ_NAMED_ATTRS)
3146 mdb_printf("R");
3147 else
3148 mdb_printf("-");
3149 if (access_mask & ACE_WRITE_NAMED_ATTRS)
3150 mdb_printf("W");
3151 else
3152 mdb_printf("-");
3153 if (access_mask & ACE_READ_ACL)
3154 mdb_printf("c");
3155 else
3156 mdb_printf("-");
3157 if (access_mask & ACE_WRITE_ACL)
3158 mdb_printf("C");
3159 else
3160 mdb_printf("-");
3161 if (access_mask & ACE_WRITE_OWNER)
3162 mdb_printf("o");
3163 else
3164 mdb_printf("-");
3165 if (access_mask & ACE_SYNCHRONIZE)
3166 mdb_printf("s");
3167 else
3168 mdb_printf("-");
3169
3170 mdb_printf(":");
3171
3172 /* Print out inheritance flags */
3173 if (ace_flags & ACE_FILE_INHERIT_ACE)
3174 mdb_printf("f");
3175 else
3176 mdb_printf("-");
3177 if (ace_flags & ACE_DIRECTORY_INHERIT_ACE)
3178 mdb_printf("d");
3179 else
3180 mdb_printf("-");
3181 if (ace_flags & ACE_INHERIT_ONLY_ACE)
3182 mdb_printf("i");
3183 else
3184 mdb_printf("-");
3185 if (ace_flags & ACE_NO_PROPAGATE_INHERIT_ACE)
3186 mdb_printf("n");
3187 else
3188 mdb_printf("-");
3189 if (ace_flags & ACE_SUCCESSFUL_ACCESS_ACE_FLAG)
3190 mdb_printf("S");
3191 else
3192 mdb_printf("-");
3193 if (ace_flags & ACE_FAILED_ACCESS_ACE_FLAG)
3194 mdb_printf("F");
3195 else
3196 mdb_printf("-");
3197 if (ace_flags & ACE_INHERITED_ACE)
3198 mdb_printf("I");
3199 else
3200 mdb_printf("-");
3201
3202 switch (ace_type) {
3203 case ACE_ACCESS_ALLOWED_ACE_TYPE:
3204 mdb_printf(":allow\n");
3205 break;
3206 case ACE_ACCESS_DENIED_ACE_TYPE:
3207 mdb_printf(":deny\n");
3208 break;
3209 case ACE_SYSTEM_AUDIT_ACE_TYPE:
3210 mdb_printf(":audit\n");
3211 break;
3212 case ACE_SYSTEM_ALARM_ACE_TYPE:
3213 mdb_printf(":alarm\n");
3214 break;
3215 default:
3216 mdb_printf(":?\n");
3217 }
3218 return (DCMD_OK);
3219 }
3220
3221 /* ARGSUSED */
3222 static int
3223 zfs_ace_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3224 {
3225 zfs_ace_t zace;
3226 int verbose = FALSE;
3227 uint64_t id;
3228
3229 if (!(flags & DCMD_ADDRSPEC))
3230 return (DCMD_USAGE);
3231
3232 if (mdb_getopts(argc, argv,
3233 'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
3234 return (DCMD_USAGE);
3235
3236 if (mdb_vread(&zace, sizeof (zfs_ace_t), addr) == -1) {
3237 mdb_warn("failed to read zfs_ace_t");
3238 return (DCMD_ERR);
3239 }
3240
3241 if ((zace.z_hdr.z_flags & ACE_TYPE_FLAGS) == 0 ||
3242 (zace.z_hdr.z_flags & ACE_TYPE_FLAGS) == ACE_IDENTIFIER_GROUP)
3243 id = zace.z_fuid;
3244 else
3245 id = -1;
3246
3247 return (zfs_ace_print_common(addr, flags, id, zace.z_hdr.z_access_mask,
3248 zace.z_hdr.z_flags, zace.z_hdr.z_type, verbose));
3249 }
3250
3251 /* ARGSUSED */
3252 static int
3253 zfs_ace0_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3254 {
3255 ace_t ace;
3256 uint64_t id;
3257 int verbose = FALSE;
3258
3259 if (!(flags & DCMD_ADDRSPEC))
3260 return (DCMD_USAGE);
3261
3262 if (mdb_getopts(argc, argv,
3263 'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
3264 return (DCMD_USAGE);
3265
3266 if (mdb_vread(&ace, sizeof (ace_t), addr) == -1) {
3267 mdb_warn("failed to read ace_t");
3268 return (DCMD_ERR);
3269 }
3270
3271 if ((ace.a_flags & ACE_TYPE_FLAGS) == 0 ||
3272 (ace.a_flags & ACE_TYPE_FLAGS) == ACE_IDENTIFIER_GROUP)
3273 id = ace.a_who;
3274 else
3275 id = -1;
3276
3277 return (zfs_ace_print_common(addr, flags, id, ace.a_access_mask,
3278 ace.a_flags, ace.a_type, verbose));
3279 }
3280
3281 typedef struct acl_dump_args {
3282 int a_argc;
3283 const mdb_arg_t *a_argv;
3284 uint16_t a_version;
3285 int a_flags;
3286 } acl_dump_args_t;
3287
3288 /* ARGSUSED */
3289 static int
3290 acl_aces_cb(uintptr_t addr, const void *unknown, void *arg)
3291 {
3292 acl_dump_args_t *acl_args = (acl_dump_args_t *)arg;
3293
3294 if (acl_args->a_version == 1) {
3295 if (mdb_call_dcmd("zfs_ace", addr,
3296 DCMD_ADDRSPEC|acl_args->a_flags, acl_args->a_argc,
3297 acl_args->a_argv) != DCMD_OK) {
3298 return (WALK_ERR);
3299 }
3300 } else {
3301 if (mdb_call_dcmd("zfs_ace0", addr,
3302 DCMD_ADDRSPEC|acl_args->a_flags, acl_args->a_argc,
3303 acl_args->a_argv) != DCMD_OK) {
3304 return (WALK_ERR);
3305 }
3306 }
3307 acl_args->a_flags = DCMD_LOOP;
3308 return (WALK_NEXT);
3309 }
3310
3311 /* ARGSUSED */
3312 static int
3313 acl_cb(uintptr_t addr, const void *unknown, void *arg)
3314 {
3315 acl_dump_args_t *acl_args = (acl_dump_args_t *)arg;
3316
3317 if (acl_args->a_version == 1) {
3318 if (mdb_pwalk("zfs_acl_node_aces", acl_aces_cb,
3319 arg, addr) != 0) {
3320 mdb_warn("can't walk ACEs");
3321 return (DCMD_ERR);
3322 }
3323 } else {
3324 if (mdb_pwalk("zfs_acl_node_aces0", acl_aces_cb,
3325 arg, addr) != 0) {
3326 mdb_warn("can't walk ACEs");
3327 return (DCMD_ERR);
3328 }
3329 }
3330 return (WALK_NEXT);
3331 }
3332
3333 /* ARGSUSED */
3334 static int
3335 zfs_acl_dump(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3336 {
3337 zfs_acl_t zacl;
3338 int verbose = FALSE;
3339 acl_dump_args_t acl_args;
3340
3341 if (!(flags & DCMD_ADDRSPEC))
3342 return (DCMD_USAGE);
3343
3344 if (mdb_getopts(argc, argv,
3345 'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
3346 return (DCMD_USAGE);
3347
3348 if (mdb_vread(&zacl, sizeof (zfs_acl_t), addr) == -1) {
3349 mdb_warn("failed to read zfs_acl_t");
3350 return (DCMD_ERR);
3351 }
3352
3353 acl_args.a_argc = argc;
3354 acl_args.a_argv = argv;
3355 acl_args.a_version = zacl.z_version;
3356 acl_args.a_flags = DCMD_LOOPFIRST;
3357
3358 if (mdb_pwalk("zfs_acl_node", acl_cb, &acl_args, addr) != 0) {
3359 mdb_warn("can't walk ACL");
3360 return (DCMD_ERR);
3361 }
3362
3363 return (DCMD_OK);
3364 }
3365
3366 /* ARGSUSED */
3367 static int
3368 zfs_acl_node_walk_init(mdb_walk_state_t *wsp)
3369 {
3370 if (wsp->walk_addr == NULL) {
3371 mdb_warn("must supply address of zfs_acl_node_t\n");
3372 return (WALK_ERR);
3373 }
3374
3375 wsp->walk_addr +=
3376 mdb_ctf_offsetof_by_name(ZFS_STRUCT "zfs_acl", "z_acl");
3377
3378 if (mdb_layered_walk("list", wsp) == -1) {
3379 mdb_warn("failed to walk 'list'\n");
3380 return (WALK_ERR);
3381 }
3382
3383 return (WALK_NEXT);
3384 }
3385
3386 static int
3387 zfs_acl_node_walk_step(mdb_walk_state_t *wsp)
3388 {
3389 zfs_acl_node_t aclnode;
3390
3391 if (mdb_vread(&aclnode, sizeof (zfs_acl_node_t),
3392 wsp->walk_addr) == -1) {
3393 mdb_warn("failed to read zfs_acl_node at %p", wsp->walk_addr);
3394 return (WALK_ERR);
3395 }
3396
3397 return (wsp->walk_callback(wsp->walk_addr, &aclnode, wsp->walk_cbdata));
3398 }
3399
3400 typedef struct ace_walk_data {
3401 int ace_count;
3402 int ace_version;
3403 } ace_walk_data_t;
3404
3405 static int
3406 zfs_aces_walk_init_common(mdb_walk_state_t *wsp, int version,
3407 int ace_count, uintptr_t ace_data)
3408 {
3409 ace_walk_data_t *ace_walk_data;
3410
3411 if (wsp->walk_addr == NULL) {
3412 mdb_warn("must supply address of zfs_acl_node_t\n");
3413 return (WALK_ERR);
3414 }
3415
3416 ace_walk_data = mdb_alloc(sizeof (ace_walk_data_t), UM_SLEEP | UM_GC);
3417
3418 ace_walk_data->ace_count = ace_count;
3419 ace_walk_data->ace_version = version;
3420
3421 wsp->walk_addr = ace_data;
3422 wsp->walk_data = ace_walk_data;
3423
3424 return (WALK_NEXT);
3425 }
3426
3427 static int
3428 zfs_acl_node_aces_walk_init_common(mdb_walk_state_t *wsp, int version)
3429 {
3430 static int gotid;
3431 static mdb_ctf_id_t acl_id;
3432 int z_ace_count;
3433 uintptr_t z_acldata;
3434
3435 if (!gotid) {
3436 if (mdb_ctf_lookup_by_name("struct zfs_acl_node",
3437 &acl_id) == -1) {
3438 mdb_warn("couldn't find struct zfs_acl_node");
3439 return (DCMD_ERR);
3440 }
3441 gotid = TRUE;
3442 }
3443
3444 if (GETMEMBID(wsp->walk_addr, &acl_id, z_ace_count, z_ace_count)) {
3445 return (DCMD_ERR);
3446 }
3447 if (GETMEMBID(wsp->walk_addr, &acl_id, z_acldata, z_acldata)) {
3448 return (DCMD_ERR);
3449 }
3450
3451 return (zfs_aces_walk_init_common(wsp, version,
3452 z_ace_count, z_acldata));
3453 }
3454
3455 /* ARGSUSED */
3456 static int
3457 zfs_acl_node_aces_walk_init(mdb_walk_state_t *wsp)
3458 {
3459 return (zfs_acl_node_aces_walk_init_common(wsp, 1));
3460 }
3461
3462 /* ARGSUSED */
3463 static int
3464 zfs_acl_node_aces0_walk_init(mdb_walk_state_t *wsp)
3465 {
3466 return (zfs_acl_node_aces_walk_init_common(wsp, 0));
3467 }
3468
3469 static int
3470 zfs_aces_walk_step(mdb_walk_state_t *wsp)
3471 {
3472 ace_walk_data_t *ace_data = wsp->walk_data;
3473 zfs_ace_t zace;
3474 ace_t *acep;
3475 int status;
3476 int entry_type;
3477 int allow_type;
3478 uintptr_t ptr;
3479
3480 if (ace_data->ace_count == 0)
3481 return (WALK_DONE);
3482
3483 if (mdb_vread(&zace, sizeof (zfs_ace_t), wsp->walk_addr) == -1) {
3484 mdb_warn("failed to read zfs_ace_t at %#lx",
3485 wsp->walk_addr);
3486 return (WALK_ERR);
3487 }
3488
3489 switch (ace_data->ace_version) {
3490 case 0:
3491 acep = (ace_t *)&zace;
3492 entry_type = acep->a_flags & ACE_TYPE_FLAGS;
3493 allow_type = acep->a_type;
3494 break;
3495 case 1:
3496 entry_type = zace.z_hdr.z_flags & ACE_TYPE_FLAGS;
3497 allow_type = zace.z_hdr.z_type;
3498 break;
3499 default:
3500 return (WALK_ERR);
3501 }
3502
3503 ptr = (uintptr_t)wsp->walk_addr;
3504 switch (entry_type) {
3505 case ACE_OWNER:
3506 case ACE_EVERYONE:
3507 case (ACE_IDENTIFIER_GROUP | ACE_GROUP):
3508 ptr += ace_data->ace_version == 0 ?
3509 sizeof (ace_t) : sizeof (zfs_ace_hdr_t);
3510 break;
3511 case ACE_IDENTIFIER_GROUP:
3512 default:
3513 switch (allow_type) {
3514 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
3515 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
3516 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
3517 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
3518 ptr += ace_data->ace_version == 0 ?
3519 sizeof (ace_t) : sizeof (zfs_object_ace_t);
3520 break;
3521 default:
3522 ptr += ace_data->ace_version == 0 ?
3523 sizeof (ace_t) : sizeof (zfs_ace_t);
3524 break;
3525 }
3526 }
3527
3528 ace_data->ace_count--;
3529 status = wsp->walk_callback(wsp->walk_addr,
3530 (void *)(uintptr_t)&zace, wsp->walk_cbdata);
3531
3532 wsp->walk_addr = ptr;
3533 return (status);
3534 }
3535
3536 typedef struct mdb_zfs_rrwlock {
3537 uintptr_t rr_writer;
3538 boolean_t rr_writer_wanted;
3539 } mdb_zfs_rrwlock_t;
3540
3541 static uint_t rrw_key;
3542
3543 /* ARGSUSED */
3544 static int
3545 rrwlock(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3546 {
3547 mdb_zfs_rrwlock_t rrw;
3548
3549 if (rrw_key == 0) {
3550 if (mdb_ctf_readsym(&rrw_key, "uint_t", "rrw_tsd_key", 0) == -1)
3551 return (DCMD_ERR);
3552 }
3553
3554 if (mdb_ctf_vread(&rrw, "rrwlock_t", "mdb_zfs_rrwlock_t", addr,
3555 0) == -1)
3556 return (DCMD_ERR);
3557
3558 if (rrw.rr_writer != 0) {
3559 mdb_printf("write lock held by thread %lx\n", rrw.rr_writer);
3560 return (DCMD_OK);
3561 }
3562
3563 if (rrw.rr_writer_wanted) {
3564 mdb_printf("writer wanted\n");
3565 }
3566
3567 mdb_printf("anonymous references:\n");
3568 (void) mdb_call_dcmd("refcount", addr +
3569 mdb_ctf_offsetof_by_name(ZFS_STRUCT "rrwlock", "rr_anon_rcount"),
3570 DCMD_ADDRSPEC, 0, NULL);
3571
3572 mdb_printf("linked references:\n");
3573 (void) mdb_call_dcmd("refcount", addr +
3574 mdb_ctf_offsetof_by_name(ZFS_STRUCT "rrwlock", "rr_linked_rcount"),
3575 DCMD_ADDRSPEC, 0, NULL);
3576
3577 /*
3578 * XXX This should find references from
3579 * "::walk thread | ::tsd -v <rrw_key>", but there is no support
3580 * for programmatic consumption of dcmds, so this would be
3581 * difficult, potentially requiring reimplementing ::tsd (both
3582 * user and kernel versions) in this MDB module.
3583 */
3584
3585 return (DCMD_OK);
3586 }
3587
3588 typedef struct mdb_arc_buf_hdr_t {
3589 uint16_t b_psize;
3590 uint16_t b_lsize;
3591 struct {
3592 uint32_t b_bufcnt;
3593 uintptr_t b_state;
3594 } b_l1hdr;
3595 } mdb_arc_buf_hdr_t;
3596
3597 enum arc_cflags {
3598 ARC_CFLAG_VERBOSE = 1 << 0,
3599 ARC_CFLAG_ANON = 1 << 1,
3600 ARC_CFLAG_MRU = 1 << 2,
3601 ARC_CFLAG_MFU = 1 << 3,
3602 ARC_CFLAG_BUFS = 1 << 4,
3603 };
3604
3605 typedef struct arc_compression_stats_data {
3606 GElf_Sym anon_sym; /* ARC_anon symbol */
3607 GElf_Sym mru_sym; /* ARC_mru symbol */
3608 GElf_Sym mrug_sym; /* ARC_mru_ghost symbol */
3609 GElf_Sym mfu_sym; /* ARC_mfu symbol */
3610 GElf_Sym mfug_sym; /* ARC_mfu_ghost symbol */
3611 GElf_Sym l2c_sym; /* ARC_l2c_only symbol */
3612 uint64_t *anon_c_hist; /* histogram of compressed sizes in anon */
3613 uint64_t *anon_u_hist; /* histogram of uncompressed sizes in anon */
3614 uint64_t *anon_bufs; /* histogram of buffer counts in anon state */
3615 uint64_t *mru_c_hist; /* histogram of compressed sizes in mru */
3616 uint64_t *mru_u_hist; /* histogram of uncompressed sizes in mru */
3617 uint64_t *mru_bufs; /* histogram of buffer counts in mru */
3618 uint64_t *mfu_c_hist; /* histogram of compressed sizes in mfu */
3619 uint64_t *mfu_u_hist; /* histogram of uncompressed sizes in mfu */
3620 uint64_t *mfu_bufs; /* histogram of buffer counts in mfu */
3621 uint64_t *all_c_hist; /* histogram of compressed anon + mru + mfu */
3622 uint64_t *all_u_hist; /* histogram of uncompressed anon + mru + mfu */
3623 uint64_t *all_bufs; /* histogram of buffer counts in all states */
3624 int arc_cflags; /* arc compression flags, specified by user */
3625 int hist_nbuckets; /* number of buckets in each histogram */
3626 } arc_compression_stats_data_t;
3627
3628 int
3629 highbit64(uint64_t i)
3630 {
3631 int h = 1;
3632
3633 if (i == 0)
3634 return (0);
3635 if (i & 0xffffffff00000000ULL) {
3636 h += 32; i >>= 32;
3637 }
3638 if (i & 0xffff0000) {
3639 h += 16; i >>= 16;
3640 }
3641 if (i & 0xff00) {
3642 h += 8; i >>= 8;
3643 }
3644 if (i & 0xf0) {
3645 h += 4; i >>= 4;
3646 }
3647 if (i & 0xc) {
3648 h += 2; i >>= 2;
3649 }
3650 if (i & 0x2) {
3651 h += 1;
3652 }
3653 return (h);
3654 }
3655
3656 /* ARGSUSED */
3657 static int
3658 arc_compression_stats_cb(uintptr_t addr, const void *unknown, void *arg)
3659 {
3660 arc_compression_stats_data_t *data = arg;
3661 mdb_arc_buf_hdr_t hdr;
3662 int cbucket, ubucket, bufcnt;
3663
3664 if (mdb_ctf_vread(&hdr, "arc_buf_hdr_t", "mdb_arc_buf_hdr_t",
3665 addr, 0) == -1) {
3666 return (WALK_ERR);
3667 }
3668
3669 /*
3670 * Headers in the ghost states, or the l2c_only state don't have
3671 * arc buffers linked off of them. Thus, their compressed size
3672 * is meaningless, so we skip these from the stats.
3673 */
3674 if (hdr.b_l1hdr.b_state == data->mrug_sym.st_value ||
3675 hdr.b_l1hdr.b_state == data->mfug_sym.st_value ||
3676 hdr.b_l1hdr.b_state == data->l2c_sym.st_value) {
3677 return (WALK_NEXT);
3678 }
3679
3680 /*
3681 * The physical size (compressed) and logical size
3682 * (uncompressed) are in units of SPA_MINBLOCKSIZE. By default,
3683 * we use the log2 of this value (rounded down to the nearest
3684 * integer) to determine the bucket to assign this header to.
3685 * Thus, the histogram is logarithmic with respect to the size
3686 * of the header. For example, the following is a mapping of the
3687 * bucket numbers and the range of header sizes they correspond to:
3688 *
3689 * 0: 0 byte headers
3690 * 1: 512 byte headers
3691 * 2: [1024 - 2048) byte headers
3692 * 3: [2048 - 4096) byte headers
3693 * 4: [4096 - 8192) byte headers
3694 * 5: [8192 - 16394) byte headers
3695 * 6: [16384 - 32768) byte headers
3696 * 7: [32768 - 65536) byte headers
3697 * 8: [65536 - 131072) byte headers
3698 * 9: 131072 byte headers
3699 *
3700 * If the ARC_CFLAG_VERBOSE flag was specified, we use the
3701 * physical and logical sizes directly. Thus, the histogram will
3702 * no longer be logarithmic; instead it will be linear with
3703 * respect to the size of the header. The following is a mapping
3704 * of the first many bucket numbers and the header size they
3705 * correspond to:
3706 *
3707 * 0: 0 byte headers
3708 * 1: 512 byte headers
3709 * 2: 1024 byte headers
3710 * 3: 1536 byte headers
3711 * 4: 2048 byte headers
3712 * 5: 2560 byte headers
3713 * 6: 3072 byte headers
3714 *
3715 * And so on. Keep in mind that a range of sizes isn't used in
3716 * the case of linear scale because the headers can only
3717 * increment or decrement in sizes of 512 bytes. So, it's not
3718 * possible for a header to be sized in between whats listed
3719 * above.
3720 *
3721 * Also, the above mapping values were calculated assuming a
3722 * SPA_MINBLOCKSHIFT of 512 bytes and a SPA_MAXBLOCKSIZE of 128K.
3723 */
3724
3725 if (data->arc_cflags & ARC_CFLAG_VERBOSE) {
3726 cbucket = hdr.b_psize;
3727 ubucket = hdr.b_lsize;
3728 } else {
3729 cbucket = highbit64(hdr.b_psize);
3730 ubucket = highbit64(hdr.b_lsize);
3731 }
3732
3733 bufcnt = hdr.b_l1hdr.b_bufcnt;
3734 if (bufcnt >= data->hist_nbuckets)
3735 bufcnt = data->hist_nbuckets - 1;
3736
3737 /* Ensure we stay within the bounds of the histogram array */
3738 ASSERT3U(cbucket, <, data->hist_nbuckets);
3739 ASSERT3U(ubucket, <, data->hist_nbuckets);
3740
3741 if (hdr.b_l1hdr.b_state == data->anon_sym.st_value) {
3742 data->anon_c_hist[cbucket]++;
3743 data->anon_u_hist[ubucket]++;
3744 data->anon_bufs[bufcnt]++;
3745 } else if (hdr.b_l1hdr.b_state == data->mru_sym.st_value) {
3746 data->mru_c_hist[cbucket]++;
3747 data->mru_u_hist[ubucket]++;
3748 data->mru_bufs[bufcnt]++;
3749 } else if (hdr.b_l1hdr.b_state == data->mfu_sym.st_value) {
3750 data->mfu_c_hist[cbucket]++;
3751 data->mfu_u_hist[ubucket]++;
3752 data->mfu_bufs[bufcnt]++;
3753 }
3754
3755 data->all_c_hist[cbucket]++;
3756 data->all_u_hist[ubucket]++;
3757 data->all_bufs[bufcnt]++;
3758
3759 return (WALK_NEXT);
3760 }
3761
3762 /* ARGSUSED */
3763 static int
3764 arc_compression_stats(uintptr_t addr, uint_t flags, int argc,
3765 const mdb_arg_t *argv)
3766 {
3767 arc_compression_stats_data_t data = { 0 };
3768 unsigned int max_shifted = SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT;
3769 unsigned int hist_size;
3770 char range[32];
3771 int rc = DCMD_OK;
3772
3773 if (mdb_getopts(argc, argv,
3774 'v', MDB_OPT_SETBITS, ARC_CFLAG_VERBOSE, &data.arc_cflags,
3775 'a', MDB_OPT_SETBITS, ARC_CFLAG_ANON, &data.arc_cflags,
3776 'b', MDB_OPT_SETBITS, ARC_CFLAG_BUFS, &data.arc_cflags,
3777 'r', MDB_OPT_SETBITS, ARC_CFLAG_MRU, &data.arc_cflags,
3778 'f', MDB_OPT_SETBITS, ARC_CFLAG_MFU, &data.arc_cflags) != argc)
3779 return (DCMD_USAGE);
3780
3781 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_anon", &data.anon_sym) ||
3782 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mru", &data.mru_sym) ||
3783 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mru_ghost", &data.mrug_sym) ||
3784 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mfu", &data.mfu_sym) ||
3785 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mfu_ghost", &data.mfug_sym) ||
3786 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_l2c_only", &data.l2c_sym)) {
3787 mdb_warn("can't find arc state symbol");
3788 return (DCMD_ERR);
3789 }
3790
3791 /*
3792 * Determine the maximum expected size for any header, and use
3793 * this to determine the number of buckets needed for each
3794 * histogram. If ARC_CFLAG_VERBOSE is specified, this value is
3795 * used directly; otherwise the log2 of the maximum size is
3796 * used. Thus, if using a log2 scale there's a maximum of 10
3797 * possible buckets, while the linear scale (when using
3798 * ARC_CFLAG_VERBOSE) has a maximum of 257 buckets.
3799 */
3800 if (data.arc_cflags & ARC_CFLAG_VERBOSE)
3801 data.hist_nbuckets = max_shifted + 1;
3802 else
3803 data.hist_nbuckets = highbit64(max_shifted) + 1;
3804
3805 hist_size = sizeof (uint64_t) * data.hist_nbuckets;
3806
3807 data.anon_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3808 data.anon_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3809 data.anon_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3810
3811 data.mru_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3812 data.mru_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3813 data.mru_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3814
3815 data.mfu_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3816 data.mfu_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3817 data.mfu_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3818
3819 data.all_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3820 data.all_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3821 data.all_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3822
3823 if (mdb_walk("arc_buf_hdr_t_full", arc_compression_stats_cb,
3824 &data) != 0) {
3825 mdb_warn("can't walk arc_buf_hdr's");
3826 rc = DCMD_ERR;
3827 goto out;
3828 }
3829
3830 if (data.arc_cflags & ARC_CFLAG_VERBOSE) {
3831 rc = mdb_snprintf(range, sizeof (range),
3832 "[n*%llu, (n+1)*%llu)", SPA_MINBLOCKSIZE,
3833 SPA_MINBLOCKSIZE);
3834 } else {
3835 rc = mdb_snprintf(range, sizeof (range),
3836 "[2^(n-1)*%llu, 2^n*%llu)", SPA_MINBLOCKSIZE,
3837 SPA_MINBLOCKSIZE);
3838 }
3839
3840 if (rc < 0) {
3841 /* snprintf failed, abort the dcmd */
3842 rc = DCMD_ERR;
3843 goto out;
3844 } else {
3845 /* snprintf succeeded above, reset return code */
3846 rc = DCMD_OK;
3847 }
3848
3849 if (data.arc_cflags & ARC_CFLAG_ANON) {
3850 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3851 mdb_printf("Histogram of the number of anon buffers "
3852 "that are associated with an arc hdr.\n");
3853 dump_histogram(data.anon_bufs, data.hist_nbuckets, 0);
3854 mdb_printf("\n");
3855 }
3856 mdb_printf("Histogram of compressed anon buffers.\n"
3857 "Each bucket represents buffers of size: %s.\n", range);
3858 dump_histogram(data.anon_c_hist, data.hist_nbuckets, 0);
3859 mdb_printf("\n");
3860
3861 mdb_printf("Histogram of uncompressed anon buffers.\n"
3862 "Each bucket represents buffers of size: %s.\n", range);
3863 dump_histogram(data.anon_u_hist, data.hist_nbuckets, 0);
3864 mdb_printf("\n");
3865 }
3866
3867 if (data.arc_cflags & ARC_CFLAG_MRU) {
3868 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3869 mdb_printf("Histogram of the number of mru buffers "
3870 "that are associated with an arc hdr.\n");
3871 dump_histogram(data.mru_bufs, data.hist_nbuckets, 0);
3872 mdb_printf("\n");
3873 }
3874 mdb_printf("Histogram of compressed mru buffers.\n"
3875 "Each bucket represents buffers of size: %s.\n", range);
3876 dump_histogram(data.mru_c_hist, data.hist_nbuckets, 0);
3877 mdb_printf("\n");
3878
3879 mdb_printf("Histogram of uncompressed mru buffers.\n"
3880 "Each bucket represents buffers of size: %s.\n", range);
3881 dump_histogram(data.mru_u_hist, data.hist_nbuckets, 0);
3882 mdb_printf("\n");
3883 }
3884
3885 if (data.arc_cflags & ARC_CFLAG_MFU) {
3886 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3887 mdb_printf("Histogram of the number of mfu buffers "
3888 "that are associated with an arc hdr.\n");
3889 dump_histogram(data.mfu_bufs, data.hist_nbuckets, 0);
3890 mdb_printf("\n");
3891 }
3892
3893 mdb_printf("Histogram of compressed mfu buffers.\n"
3894 "Each bucket represents buffers of size: %s.\n", range);
3895 dump_histogram(data.mfu_c_hist, data.hist_nbuckets, 0);
3896 mdb_printf("\n");
3897
3898 mdb_printf("Histogram of uncompressed mfu buffers.\n"
3899 "Each bucket represents buffers of size: %s.\n", range);
3900 dump_histogram(data.mfu_u_hist, data.hist_nbuckets, 0);
3901 mdb_printf("\n");
3902 }
3903
3904 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3905 mdb_printf("Histogram of all buffers that "
3906 "are associated with an arc hdr.\n");
3907 dump_histogram(data.all_bufs, data.hist_nbuckets, 0);
3908 mdb_printf("\n");
3909 }
3910
3911 mdb_printf("Histogram of all compressed buffers.\n"
3912 "Each bucket represents buffers of size: %s.\n", range);
3913 dump_histogram(data.all_c_hist, data.hist_nbuckets, 0);
3914 mdb_printf("\n");
3915
3916 mdb_printf("Histogram of all uncompressed buffers.\n"
3917 "Each bucket represents buffers of size: %s.\n", range);
3918 dump_histogram(data.all_u_hist, data.hist_nbuckets, 0);
3919
3920 out:
3921 mdb_free(data.anon_c_hist, hist_size);
3922 mdb_free(data.anon_u_hist, hist_size);
3923 mdb_free(data.anon_bufs, hist_size);
3924
3925 mdb_free(data.mru_c_hist, hist_size);
3926 mdb_free(data.mru_u_hist, hist_size);
3927 mdb_free(data.mru_bufs, hist_size);
3928
3929 mdb_free(data.mfu_c_hist, hist_size);
3930 mdb_free(data.mfu_u_hist, hist_size);
3931 mdb_free(data.mfu_bufs, hist_size);
3932
3933 mdb_free(data.all_c_hist, hist_size);
3934 mdb_free(data.all_u_hist, hist_size);
3935 mdb_free(data.all_bufs, hist_size);
3936
3937 return (rc);
3938 }
3939
3940 /*
3941 * MDB module linkage information:
3942 *
3943 * We declare a list of structures describing our dcmds, and a function
3944 * named _mdb_init to return a pointer to our module information.
3945 */
3946
3947 static const mdb_dcmd_t dcmds[] = {
3948 { "arc", "[-bkmg]", "print ARC variables", arc_print },
3949 { "blkptr", ":", "print blkptr_t", blkptr },
3950 { "dva", ":", "print dva_t", dva },
3951 { "dbuf", ":", "print dmu_buf_impl_t", dbuf },
3952 { "dbuf_stats", ":", "dbuf stats", dbuf_stats },
3953 { "dbufs",
3954 "\t[-O objset_t*] [-n objset_name | \"mos\"] "
3955 "[-o object | \"mdn\"] \n"
3956 "\t[-l level] [-b blkid | \"bonus\"]",
3957 "find dmu_buf_impl_t's that match specified criteria", dbufs },
3958 { "abuf_find", "dva_word[0] dva_word[1]",
3959 "find arc_buf_hdr_t of a specified DVA",
3960 abuf_find },
3961 { "spa", "?[-cevmMh]\n"
3962 "\t-c display spa config\n"
3963 "\t-e display vdev statistics\n"
3964 "\t-v display vdev information\n"
3965 "\t-m display metaslab statistics\n"
3966 "\t-M display metaslab group statistics\n"
3967 "\t-h display histogram (requires -m or -M)\n",
3968 "spa_t summary", spa_print },
3969 { "spa_config", ":", "print spa_t configuration", spa_print_config },
3970 { "spa_space", ":[-b]", "print spa_t on-disk space usage", spa_space },
3971 { "spa_vdevs", ":[-emMh]\n"
3972 "\t-e display vdev statistics\n"
3973 "\t-m dispaly metaslab statistics\n"
3974 "\t-M display metaslab group statistic\n"
3975 "\t-h display histogram (requires -m or -M)\n",
3976 "given a spa_t, print vdev summary", spa_vdevs },
3977 { "vdev", ":[-remMh]\n"
3978 "\t-r display recursively\n"
3979 "\t-e display statistics\n"
3980 "\t-m display metaslab statistics (top level vdev only)\n"
3981 "\t-M display metaslab group statistics (top level vdev only)\n"
3982 "\t-h display histogram (requires -m or -M)\n",
3983 "vdev_t summary", vdev_print },
3984 { "zio", ":[-cpr]\n"
3985 "\t-c display children\n"
3986 "\t-p display parents\n"
3987 "\t-r display recursively",
3988 "zio_t summary", zio_print },
3989 { "zio_state", "?", "print out all zio_t structures on system or "
3990 "for a particular pool", zio_state },
3991 { "zfs_blkstats", ":[-v]",
3992 "given a spa_t, print block type stats from last scrub",
3993 zfs_blkstats },
3994 { "zfs_params", "", "print zfs tunable parameters", zfs_params },
3995 { "refcount", ":[-r]\n"
3996 "\t-r display recently removed references",
3997 "print refcount_t holders", refcount },
3998 { "zap_leaf", "", "print zap_leaf_phys_t", zap_leaf },
3999 { "zfs_aces", ":[-v]", "print all ACEs from a zfs_acl_t",
4000 zfs_acl_dump },
4001 { "zfs_ace", ":[-v]", "print zfs_ace", zfs_ace_print },
4002 { "zfs_ace0", ":[-v]", "print zfs_ace0", zfs_ace0_print },
4003 { "sa_attr_table", ":", "print SA attribute table from sa_os_t",
4004 sa_attr_table},
4005 { "sa_attr", ": attr_id",
4006 "print SA attribute address when given sa_handle_t", sa_attr_print},
4007 { "zfs_dbgmsg", ":[-va]",
4008 "print zfs debug log", dbgmsg},
4009 { "rrwlock", ":",
4010 "print rrwlock_t, including readers", rrwlock},
4011 { "metaslab_weight", "weight",
4012 "print metaslab weight", metaslab_weight},
4013 { "metaslab_trace", ":",
4014 "print metaslab allocation trace records", metaslab_trace},
4015 { "arc_compression_stats", ":[-vabrf]\n"
4016 "\t-v verbose, display a linearly scaled histogram\n"
4017 "\t-a display ARC_anon state statistics individually\n"
4018 "\t-r display ARC_mru state statistics individually\n"
4019 "\t-f display ARC_mfu state statistics individually\n"
4020 "\t-b display histogram of buffer counts\n",
4021 "print a histogram of compressed arc buffer sizes",
4022 arc_compression_stats},
4023 { NULL }
4024 };
4025
4026 static const mdb_walker_t walkers[] = {
4027 { "zms_freelist", "walk ZFS metaslab freelist",
4028 freelist_walk_init, freelist_walk_step, NULL },
4029 { "txg_list", "given any txg_list_t *, walk all entries in all txgs",
4030 txg_list_walk_init, txg_list_walk_step, NULL },
4031 { "txg_list0", "given any txg_list_t *, walk all entries in txg 0",
4032 txg_list0_walk_init, txg_list_walk_step, NULL },
4033 { "txg_list1", "given any txg_list_t *, walk all entries in txg 1",
4034 txg_list1_walk_init, txg_list_walk_step, NULL },
4035 { "txg_list2", "given any txg_list_t *, walk all entries in txg 2",
4036 txg_list2_walk_init, txg_list_walk_step, NULL },
4037 { "txg_list3", "given any txg_list_t *, walk all entries in txg 3",
4038 txg_list3_walk_init, txg_list_walk_step, NULL },
4039 { "zio", "walk all zio structures, optionally for a particular spa_t",
4040 zio_walk_init, zio_walk_step, NULL },
4041 { "zio_root",
4042 "walk all root zio_t structures, optionally for a particular spa_t",
4043 zio_walk_init, zio_walk_root_step, NULL },
4044 { "spa", "walk all spa_t entries in the namespace",
4045 spa_walk_init, spa_walk_step, NULL },
4046 { "metaslab", "given a spa_t *, walk all metaslab_t structures",
4047 metaslab_walk_init, metaslab_walk_step, NULL },
4048 { "multilist", "given a multilist_t *, walk all list_t structures",
4049 multilist_walk_init, multilist_walk_step, NULL },
4050 { "zfs_acl_node", "given a zfs_acl_t, walk all zfs_acl_nodes",
4051 zfs_acl_node_walk_init, zfs_acl_node_walk_step, NULL },
4052 { "zfs_acl_node_aces", "given a zfs_acl_node_t, walk all ACEs",
4053 zfs_acl_node_aces_walk_init, zfs_aces_walk_step, NULL },
4054 { "zfs_acl_node_aces0",
4055 "given a zfs_acl_node_t, walk all ACEs as ace_t",
4056 zfs_acl_node_aces0_walk_init, zfs_aces_walk_step, NULL },
4057 { NULL }
4058 };
4059
4060 static const mdb_modinfo_t modinfo = {
4061 MDB_API_VERSION, dcmds, walkers
4062 };
4063
4064 const mdb_modinfo_t *
4065 _mdb_init(void)
4066 {
4067 return (&modinfo);
4068 }
Unleashed OS