?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
7614 zfs device evacuation/removal
[unleashed.git] / usr / src / cmd / mdb / common / modules / zfs / zfs.c
blob75d9fbd102f3065d0ce5384c7a93e1890e7a846b
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
25 * Copyright (c) 2017, Joyent, Inc. All rights reserved.
26 */
27
28 /* Portions Copyright 2010 Robert Milkowski */
29
30 #include <mdb/mdb_ctf.h>
31 #include <sys/zfs_context.h>
32 #include <sys/mdb_modapi.h>
33 #include <sys/dbuf.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/metaslab_impl.h>
38 #include <sys/space_map.h>
39 #include <sys/list.h>
40 #include <sys/vdev_impl.h>
41 #include <sys/zap_leaf.h>
42 #include <sys/zap_impl.h>
43 #include <ctype.h>
44 #include <sys/zfs_acl.h>
45 #include <sys/sa_impl.h>
46 #include <sys/multilist.h>
47
48 #ifdef _KERNEL
49 #define ZFS_OBJ_NAME "zfs"
50 extern int64_t mdb_gethrtime(void);
51 #else
52 #define ZFS_OBJ_NAME "libzpool.so.1"
53 #endif
54
55 #define ZFS_STRUCT "struct " ZFS_OBJ_NAME "`"
56
57 #ifndef _KERNEL
58 int aok;
59 #endif
60
61 enum spa_flags {
62 SPA_FLAG_CONFIG = 1 << 0,
63 SPA_FLAG_VDEVS = 1 << 1,
64 SPA_FLAG_ERRORS = 1 << 2,
65 SPA_FLAG_METASLAB_GROUPS = 1 << 3,
66 SPA_FLAG_METASLABS = 1 << 4,
67 SPA_FLAG_HISTOGRAMS = 1 << 5
68 };
69
70 /*
71 * If any of these flags are set, call spa_vdevs in spa_print
72 */
73 #define SPA_FLAG_ALL_VDEV \
74 (SPA_FLAG_VDEVS | SPA_FLAG_ERRORS | SPA_FLAG_METASLAB_GROUPS | \
75 SPA_FLAG_METASLABS)
76
77 static int
78 getmember(uintptr_t addr, const char *type, mdb_ctf_id_t *idp,
79 const char *member, int len, void *buf)
80 {
81 mdb_ctf_id_t id;
82 ulong_t off;
83 char name[64];
84
85 if (idp == NULL) {
86 if (mdb_ctf_lookup_by_name(type, &id) == -1) {
87 mdb_warn("couldn't find type %s", type);
88 return (DCMD_ERR);
89 }
90 idp = &id;
91 } else {
92 type = name;
93 mdb_ctf_type_name(*idp, name, sizeof (name));
94 }
95
96 if (mdb_ctf_offsetof(*idp, member, &off) == -1) {
97 mdb_warn("couldn't find member %s of type %s\n", member, type);
98 return (DCMD_ERR);
99 }
100 if (off % 8 != 0) {
101 mdb_warn("member %s of type %s is unsupported bitfield",
102 member, type);
103 return (DCMD_ERR);
104 }
105 off /= 8;
106
107 if (mdb_vread(buf, len, addr + off) == -1) {
108 mdb_warn("failed to read %s from %s at %p",
109 member, type, addr + off);
110 return (DCMD_ERR);
111 }
112 /* mdb_warn("read %s from %s at %p+%llx\n", member, type, addr, off); */
113
114 return (0);
115 }
116
117 #define GETMEMB(addr, structname, member, dest) \
118 getmember(addr, ZFS_STRUCT structname, NULL, #member, \
119 sizeof (dest), &(dest))
120
121 #define GETMEMBID(addr, ctfid, member, dest) \
122 getmember(addr, NULL, ctfid, #member, sizeof (dest), &(dest))
123
124 static boolean_t
125 strisprint(const char *cp)
126 {
127 for (; *cp; cp++) {
128 if (!isprint(*cp))
129 return (B_FALSE);
130 }
131 return (B_TRUE);
132 }
133
134 #define NICENUM_BUFLEN 6
135
136 static int
137 snprintfrac(char *buf, int len,
138 uint64_t numerator, uint64_t denom, int frac_digits)
139 {
140 int mul = 1;
141 int whole, frac, i;
142
143 for (i = frac_digits; i; i--)
144 mul *= 10;
145 whole = numerator / denom;
146 frac = mul * numerator / denom - mul * whole;
147 return (mdb_snprintf(buf, len, "%u.%0*u", whole, frac_digits, frac));
148 }
149
150 static void
151 mdb_nicenum(uint64_t num, char *buf)
152 {
153 uint64_t n = num;
154 int index = 0;
155 char *u;
156
157 while (n >= 1024) {
158 n = (n + (1024 / 2)) / 1024; /* Round up or down */
159 index++;
160 }
161
162 u = &" \0K\0M\0G\0T\0P\0E\0"[index*2];
163
164 if (index == 0) {
165 (void) mdb_snprintf(buf, NICENUM_BUFLEN, "%llu",
166 (u_longlong_t)n);
167 } else if (n < 10 && (num & (num - 1)) != 0) {
168 (void) snprintfrac(buf, NICENUM_BUFLEN,
169 num, 1ULL << 10 * index, 2);
170 strcat(buf, u);
171 } else if (n < 100 && (num & (num - 1)) != 0) {
172 (void) snprintfrac(buf, NICENUM_BUFLEN,
173 num, 1ULL << 10 * index, 1);
174 strcat(buf, u);
175 } else {
176 (void) mdb_snprintf(buf, NICENUM_BUFLEN, "%llu%s",
177 (u_longlong_t)n, u);
178 }
179 }
180
181 static int verbose;
182
183 static int
184 freelist_walk_init(mdb_walk_state_t *wsp)
185 {
186 if (wsp->walk_addr == NULL) {
187 mdb_warn("must supply starting address\n");
188 return (WALK_ERR);
189 }
190
191 wsp->walk_data = 0; /* Index into the freelist */
192 return (WALK_NEXT);
193 }
194
195 static int
196 freelist_walk_step(mdb_walk_state_t *wsp)
197 {
198 uint64_t entry;
199 uintptr_t number = (uintptr_t)wsp->walk_data;
200 char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
201 "INVALID", "INVALID", "INVALID", "INVALID" };
202 int mapshift = SPA_MINBLOCKSHIFT;
203
204 if (mdb_vread(&entry, sizeof (entry), wsp->walk_addr) == -1) {
205 mdb_warn("failed to read freelist entry %p", wsp->walk_addr);
206 return (WALK_DONE);
207 }
208 wsp->walk_addr += sizeof (entry);
209 wsp->walk_data = (void *)(number + 1);
210
211 if (SM_DEBUG_DECODE(entry)) {
212 mdb_printf("DEBUG: %3u %10s: txg=%llu pass=%llu\n",
213 number,
214 ddata[SM_DEBUG_ACTION_DECODE(entry)],
215 SM_DEBUG_TXG_DECODE(entry),
216 SM_DEBUG_SYNCPASS_DECODE(entry));
217 } else {
218 mdb_printf("Entry: %3u offsets=%08llx-%08llx type=%c "
219 "size=%06llx", number,
220 SM_OFFSET_DECODE(entry) << mapshift,
221 (SM_OFFSET_DECODE(entry) + SM_RUN_DECODE(entry)) <<
222 mapshift,
223 SM_TYPE_DECODE(entry) == SM_ALLOC ? 'A' : 'F',
224 SM_RUN_DECODE(entry) << mapshift);
225 if (verbose)
226 mdb_printf(" (raw=%012llx)\n", entry);
227 mdb_printf("\n");
228 }
229 return (WALK_NEXT);
230 }
231
232 static int
233 mdb_dsl_dir_name(uintptr_t addr, char *buf)
234 {
235 static int gotid;
236 static mdb_ctf_id_t dd_id;
237 uintptr_t dd_parent;
238 char dd_myname[ZFS_MAX_DATASET_NAME_LEN];
239
240 if (!gotid) {
241 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "dsl_dir",
242 &dd_id) == -1) {
243 mdb_warn("couldn't find struct dsl_dir");
244 return (DCMD_ERR);
245 }
246 gotid = TRUE;
247 }
248 if (GETMEMBID(addr, &dd_id, dd_parent, dd_parent) ||
249 GETMEMBID(addr, &dd_id, dd_myname, dd_myname)) {
250 return (DCMD_ERR);
251 }
252
253 if (dd_parent) {
254 if (mdb_dsl_dir_name(dd_parent, buf))
255 return (DCMD_ERR);
256 strcat(buf, "/");
257 }
258
259 if (dd_myname[0])
260 strcat(buf, dd_myname);
261 else
262 strcat(buf, "???");
263
264 return (0);
265 }
266
267 static int
268 objset_name(uintptr_t addr, char *buf)
269 {
270 static int gotid;
271 static mdb_ctf_id_t os_id, ds_id;
272 uintptr_t os_dsl_dataset;
273 char ds_snapname[ZFS_MAX_DATASET_NAME_LEN];
274 uintptr_t ds_dir;
275
276 buf[0] = '\0';
277
278 if (!gotid) {
279 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "objset",
280 &os_id) == -1) {
281 mdb_warn("couldn't find struct objset");
282 return (DCMD_ERR);
283 }
284 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "dsl_dataset",
285 &ds_id) == -1) {
286 mdb_warn("couldn't find struct dsl_dataset");
287 return (DCMD_ERR);
288 }
289
290 gotid = TRUE;
291 }
292
293 if (GETMEMBID(addr, &os_id, os_dsl_dataset, os_dsl_dataset))
294 return (DCMD_ERR);
295
296 if (os_dsl_dataset == 0) {
297 strcat(buf, "mos");
298 return (0);
299 }
300
301 if (GETMEMBID(os_dsl_dataset, &ds_id, ds_snapname, ds_snapname) ||
302 GETMEMBID(os_dsl_dataset, &ds_id, ds_dir, ds_dir)) {
303 return (DCMD_ERR);
304 }
305
306 if (ds_dir && mdb_dsl_dir_name(ds_dir, buf))
307 return (DCMD_ERR);
308
309 if (ds_snapname[0]) {
310 strcat(buf, "@");
311 strcat(buf, ds_snapname);
312 }
313 return (0);
314 }
315
316 static int
317 enum_lookup(char *type, int val, const char *prefix, size_t size, char *out)
318 {
319 const char *cp;
320 size_t len = strlen(prefix);
321 mdb_ctf_id_t enum_type;
322
323 if (mdb_ctf_lookup_by_name(type, &enum_type) != 0) {
324 mdb_warn("Could not find enum for %s", type);
325 return (-1);
326 }
327
328 if ((cp = mdb_ctf_enum_name(enum_type, val)) != NULL) {
329 if (strncmp(cp, prefix, len) == 0)
330 cp += len;
331 (void) strncpy(out, cp, size);
332 } else {
333 mdb_snprintf(out, size, "? (%d)", val);
334 }
335 return (0);
336 }
337
338 /* ARGSUSED */
339 static int
340 zfs_params(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
341 {
342 /*
343 * This table can be approximately generated by running:
344 * egrep "^[a-z0-9_]+ [a-z0-9_]+( =.*)?;" *.c | cut -d ' ' -f 2
345 */
346 static const char *params[] = {
347 "arc_reduce_dnlc_percent",
348 "arc_lotsfree_percent",
349 "zfs_dirty_data_max",
350 "zfs_dirty_data_sync",
351 "zfs_delay_max_ns",
352 "zfs_delay_min_dirty_percent",
353 "zfs_delay_scale",
354 "zfs_vdev_max_active",
355 "zfs_vdev_sync_read_min_active",
356 "zfs_vdev_sync_read_max_active",
357 "zfs_vdev_sync_write_min_active",
358 "zfs_vdev_sync_write_max_active",
359 "zfs_vdev_async_read_min_active",
360 "zfs_vdev_async_read_max_active",
361 "zfs_vdev_async_write_min_active",
362 "zfs_vdev_async_write_max_active",
363 "zfs_vdev_scrub_min_active",
364 "zfs_vdev_scrub_max_active",
365 "zfs_vdev_async_write_active_min_dirty_percent",
366 "zfs_vdev_async_write_active_max_dirty_percent",
367 "spa_asize_inflation",
368 "zfs_arc_max",
369 "zfs_arc_min",
370 "arc_shrink_shift",
371 "zfs_mdcomp_disable",
372 "zfs_prefetch_disable",
373 "zfetch_max_streams",
374 "zfetch_min_sec_reap",
375 "zfetch_block_cap",
376 "zfetch_array_rd_sz",
377 "zfs_default_bs",
378 "zfs_default_ibs",
379 "metaslab_aliquot",
380 "reference_tracking_enable",
381 "reference_history",
382 "spa_max_replication_override",
383 "spa_mode_global",
384 "zfs_flags",
385 "zfs_txg_timeout",
386 "zfs_vdev_cache_max",
387 "zfs_vdev_cache_size",
388 "zfs_vdev_cache_bshift",
389 "vdev_mirror_shift",
390 "zfs_scrub_limit",
391 "zfs_no_scrub_io",
392 "zfs_no_scrub_prefetch",
393 "zfs_vdev_aggregation_limit",
394 "fzap_default_block_shift",
395 "zfs_immediate_write_sz",
396 "zfs_read_chunk_size",
397 "zfs_nocacheflush",
398 "zil_replay_disable",
399 "metaslab_gang_bang",
400 "metaslab_df_alloc_threshold",
401 "metaslab_df_free_pct",
402 "zio_injection_enabled",
403 "zvol_immediate_write_sz",
404 };
405
406 for (int i = 0; i < sizeof (params) / sizeof (params[0]); i++) {
407 int sz;
408 uint64_t val64;
409 uint32_t *val32p = (uint32_t *)&val64;
410
411 sz = mdb_readvar(&val64, params[i]);
412 if (sz == 4) {
413 mdb_printf("%s = 0x%x\n", params[i], *val32p);
414 } else if (sz == 8) {
415 mdb_printf("%s = 0x%llx\n", params[i], val64);
416 } else {
417 mdb_warn("variable %s not found", params[i]);
418 }
419 }
420
421 return (DCMD_OK);
422 }
423
424 /* ARGSUSED */
425 static int
426 dva(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
427 {
428 dva_t dva;
429 if (mdb_vread(&dva, sizeof (dva_t), addr) == -1) {
430 mdb_warn("failed to read dva_t");
431 return (DCMD_ERR);
432 }
433 mdb_printf("<%llu:%llx:%llx>\n",
434 (u_longlong_t)DVA_GET_VDEV(&dva),
435 (u_longlong_t)DVA_GET_OFFSET(&dva),
436 (u_longlong_t)DVA_GET_ASIZE(&dva));
437
438 return (DCMD_OK);
439 }
440
441 /* ARGSUSED */
442 static int
443 blkptr(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
444 {
445 char type[80], checksum[80], compress[80];
446 blkptr_t blk, *bp = &blk;
447 char buf[BP_SPRINTF_LEN];
448
449 if (mdb_vread(&blk, sizeof (blkptr_t), addr) == -1) {
450 mdb_warn("failed to read blkptr_t");
451 return (DCMD_ERR);
452 }
453
454 if (enum_lookup("enum dmu_object_type", BP_GET_TYPE(bp), "DMU_OT_",
455 sizeof (type), type) == -1 ||
456 enum_lookup("enum zio_checksum", BP_GET_CHECKSUM(bp),
457 "ZIO_CHECKSUM_", sizeof (checksum), checksum) == -1 ||
458 enum_lookup("enum zio_compress", BP_GET_COMPRESS(bp),
459 "ZIO_COMPRESS_", sizeof (compress), compress) == -1) {
460 mdb_warn("Could not find blkptr enumerated types");
461 return (DCMD_ERR);
462 }
463
464 SNPRINTF_BLKPTR(mdb_snprintf, '\n', buf, sizeof (buf), bp, type,
465 checksum, compress);
466
467 mdb_printf("%s\n", buf);
468
469 return (DCMD_OK);
470 }
471
472 typedef struct mdb_dmu_buf_impl {
473 struct {
474 uint64_t db_object;
475 uintptr_t db_data;
476 } db;
477 uintptr_t db_objset;
478 uint64_t db_level;
479 uint64_t db_blkid;
480 struct {
481 uint64_t rc_count;
482 } db_holds;
483 } mdb_dmu_buf_impl_t;
484
485 /* ARGSUSED */
486 static int
487 dbuf(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
488 {
489 mdb_dmu_buf_impl_t db;
490 char objectname[32];
491 char blkidname[32];
492 char path[ZFS_MAX_DATASET_NAME_LEN];
493 int ptr_width = (int)(sizeof (void *)) * 2;
494
495 if (DCMD_HDRSPEC(flags))
496 mdb_printf("%*s %8s %3s %9s %5s %s\n",
497 ptr_width, "addr", "object", "lvl", "blkid", "holds", "os");
498
499 if (mdb_ctf_vread(&db, ZFS_STRUCT "dmu_buf_impl", "mdb_dmu_buf_impl_t",
500 addr, 0) == -1)
501 return (DCMD_ERR);
502
503 if (db.db.db_object == DMU_META_DNODE_OBJECT)
504 (void) strcpy(objectname, "mdn");
505 else
506 (void) mdb_snprintf(objectname, sizeof (objectname), "%llx",
507 (u_longlong_t)db.db.db_object);
508
509 if (db.db_blkid == DMU_BONUS_BLKID)
510 (void) strcpy(blkidname, "bonus");
511 else
512 (void) mdb_snprintf(blkidname, sizeof (blkidname), "%llx",
513 (u_longlong_t)db.db_blkid);
514
515 if (objset_name(db.db_objset, path)) {
516 return (DCMD_ERR);
517 }
518
519 mdb_printf("%*p %8s %3u %9s %5llu %s\n", ptr_width, addr,
520 objectname, (int)db.db_level, blkidname,
521 db.db_holds.rc_count, path);
522
523 return (DCMD_OK);
524 }
525
526 /* ARGSUSED */
527 static int
528 dbuf_stats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
529 {
530 #define HISTOSZ 32
531 uintptr_t dbp;
532 dmu_buf_impl_t db;
533 dbuf_hash_table_t ht;
534 uint64_t bucket, ndbufs;
535 uint64_t histo[HISTOSZ];
536 uint64_t histo2[HISTOSZ];
537 int i, maxidx;
538
539 if (mdb_readvar(&ht, "dbuf_hash_table") == -1) {
540 mdb_warn("failed to read 'dbuf_hash_table'");
541 return (DCMD_ERR);
542 }
543
544 for (i = 0; i < HISTOSZ; i++) {
545 histo[i] = 0;
546 histo2[i] = 0;
547 }
548
549 ndbufs = 0;
550 for (bucket = 0; bucket < ht.hash_table_mask+1; bucket++) {
551 int len;
552
553 if (mdb_vread(&dbp, sizeof (void *),
554 (uintptr_t)(ht.hash_table+bucket)) == -1) {
555 mdb_warn("failed to read hash bucket %u at %p",
556 bucket, ht.hash_table+bucket);
557 return (DCMD_ERR);
558 }
559
560 len = 0;
561 while (dbp != 0) {
562 if (mdb_vread(&db, sizeof (dmu_buf_impl_t),
563 dbp) == -1) {
564 mdb_warn("failed to read dbuf at %p", dbp);
565 return (DCMD_ERR);
566 }
567 dbp = (uintptr_t)db.db_hash_next;
568 for (i = MIN(len, HISTOSZ - 1); i >= 0; i--)
569 histo2[i]++;
570 len++;
571 ndbufs++;
572 }
573
574 if (len >= HISTOSZ)
575 len = HISTOSZ-1;
576 histo[len]++;
577 }
578
579 mdb_printf("hash table has %llu buckets, %llu dbufs "
580 "(avg %llu buckets/dbuf)\n",
581 ht.hash_table_mask+1, ndbufs,
582 (ht.hash_table_mask+1)/ndbufs);
583
584 mdb_printf("\n");
585 maxidx = 0;
586 for (i = 0; i < HISTOSZ; i++)
587 if (histo[i] > 0)
588 maxidx = i;
589 mdb_printf("hash chain length number of buckets\n");
590 for (i = 0; i <= maxidx; i++)
591 mdb_printf("%u %llu\n", i, histo[i]);
592
593 mdb_printf("\n");
594 maxidx = 0;
595 for (i = 0; i < HISTOSZ; i++)
596 if (histo2[i] > 0)
597 maxidx = i;
598 mdb_printf("hash chain depth number of dbufs\n");
599 for (i = 0; i <= maxidx; i++)
600 mdb_printf("%u or more %llu %llu%%\n",
601 i, histo2[i], histo2[i]*100/ndbufs);
602
603
604 return (DCMD_OK);
605 }
606
607 #define CHAIN_END 0xffff
608 /*
609 * ::zap_leaf [-v]
610 *
611 * Print a zap_leaf_phys_t, assumed to be 16k
612 */
613 /* ARGSUSED */
614 static int
615 zap_leaf(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
616 {
617 char buf[16*1024];
618 int verbose = B_FALSE;
619 int four = B_FALSE;
620 dmu_buf_t l_dbuf;
621 zap_leaf_t l;
622 zap_leaf_phys_t *zlp = (void *)buf;
623 int i;
624
625 if (mdb_getopts(argc, argv,
626 'v', MDB_OPT_SETBITS, TRUE, &verbose,
627 '4', MDB_OPT_SETBITS, TRUE, &four,
628 NULL) != argc)
629 return (DCMD_USAGE);
630
631 l_dbuf.db_data = zlp;
632 l.l_dbuf = &l_dbuf;
633 l.l_bs = 14; /* assume 16k blocks */
634 if (four)
635 l.l_bs = 12;
636
637 if (!(flags & DCMD_ADDRSPEC)) {
638 return (DCMD_USAGE);
639 }
640
641 if (mdb_vread(buf, sizeof (buf), addr) == -1) {
642 mdb_warn("failed to read zap_leaf_phys_t at %p", addr);
643 return (DCMD_ERR);
644 }
645
646 if (zlp->l_hdr.lh_block_type != ZBT_LEAF ||
647 zlp->l_hdr.lh_magic != ZAP_LEAF_MAGIC) {
648 mdb_warn("This does not appear to be a zap_leaf_phys_t");
649 return (DCMD_ERR);
650 }
651
652 mdb_printf("zap_leaf_phys_t at %p:\n", addr);
653 mdb_printf(" lh_prefix_len = %u\n", zlp->l_hdr.lh_prefix_len);
654 mdb_printf(" lh_prefix = %llx\n", zlp->l_hdr.lh_prefix);
655 mdb_printf(" lh_nentries = %u\n", zlp->l_hdr.lh_nentries);
656 mdb_printf(" lh_nfree = %u\n", zlp->l_hdr.lh_nfree,
657 zlp->l_hdr.lh_nfree * 100 / (ZAP_LEAF_NUMCHUNKS(&l)));
658 mdb_printf(" lh_freelist = %u\n", zlp->l_hdr.lh_freelist);
659 mdb_printf(" lh_flags = %x (%s)\n", zlp->l_hdr.lh_flags,
660 zlp->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED ?
661 "ENTRIES_CDSORTED" : "");
662
663 if (verbose) {
664 mdb_printf(" hash table:\n");
665 for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++) {
666 if (zlp->l_hash[i] != CHAIN_END)
667 mdb_printf(" %u: %u\n", i, zlp->l_hash[i]);
668 }
669 }
670
671 mdb_printf(" chunks:\n");
672 for (i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) {
673 /* LINTED: alignment */
674 zap_leaf_chunk_t *zlc = &ZAP_LEAF_CHUNK(&l, i);
675 switch (zlc->l_entry.le_type) {
676 case ZAP_CHUNK_FREE:
677 if (verbose) {
678 mdb_printf(" %u: free; lf_next = %u\n",
679 i, zlc->l_free.lf_next);
680 }
681 break;
682 case ZAP_CHUNK_ENTRY:
683 mdb_printf(" %u: entry\n", i);
684 if (verbose) {
685 mdb_printf(" le_next = %u\n",
686 zlc->l_entry.le_next);
687 }
688 mdb_printf(" le_name_chunk = %u\n",
689 zlc->l_entry.le_name_chunk);
690 mdb_printf(" le_name_numints = %u\n",
691 zlc->l_entry.le_name_numints);
692 mdb_printf(" le_value_chunk = %u\n",
693 zlc->l_entry.le_value_chunk);
694 mdb_printf(" le_value_intlen = %u\n",
695 zlc->l_entry.le_value_intlen);
696 mdb_printf(" le_value_numints = %u\n",
697 zlc->l_entry.le_value_numints);
698 mdb_printf(" le_cd = %u\n",
699 zlc->l_entry.le_cd);
700 mdb_printf(" le_hash = %llx\n",
701 zlc->l_entry.le_hash);
702 break;
703 case ZAP_CHUNK_ARRAY:
704 mdb_printf(" %u: array", i);
705 if (strisprint((char *)zlc->l_array.la_array))
706 mdb_printf(" \"%s\"", zlc->l_array.la_array);
707 mdb_printf("\n");
708 if (verbose) {
709 int j;
710 mdb_printf(" ");
711 for (j = 0; j < ZAP_LEAF_ARRAY_BYTES; j++) {
712 mdb_printf("%02x ",
713 zlc->l_array.la_array[j]);
714 }
715 mdb_printf("\n");
716 }
717 if (zlc->l_array.la_next != CHAIN_END) {
718 mdb_printf(" lf_next = %u\n",
719 zlc->l_array.la_next);
720 }
721 break;
722 default:
723 mdb_printf(" %u: undefined type %u\n",
724 zlc->l_entry.le_type);
725 }
726 }
727
728 return (DCMD_OK);
729 }
730
731 typedef struct dbufs_data {
732 mdb_ctf_id_t id;
733 uint64_t objset;
734 uint64_t object;
735 uint64_t level;
736 uint64_t blkid;
737 char *osname;
738 } dbufs_data_t;
739
740 #define DBUFS_UNSET (0xbaddcafedeadbeefULL)
741
742 /* ARGSUSED */
743 static int
744 dbufs_cb(uintptr_t addr, const void *unknown, void *arg)
745 {
746 dbufs_data_t *data = arg;
747 uintptr_t objset;
748 dmu_buf_t db;
749 uint8_t level;
750 uint64_t blkid;
751 char osname[ZFS_MAX_DATASET_NAME_LEN];
752
753 if (GETMEMBID(addr, &data->id, db_objset, objset) ||
754 GETMEMBID(addr, &data->id, db, db) ||
755 GETMEMBID(addr, &data->id, db_level, level) ||
756 GETMEMBID(addr, &data->id, db_blkid, blkid)) {
757 return (WALK_ERR);
758 }
759
760 if ((data->objset == DBUFS_UNSET || data->objset == objset) &&
761 (data->osname == NULL || (objset_name(objset, osname) == 0 &&
762 strcmp(data->osname, osname) == 0)) &&
763 (data->object == DBUFS_UNSET || data->object == db.db_object) &&
764 (data->level == DBUFS_UNSET || data->level == level) &&
765 (data->blkid == DBUFS_UNSET || data->blkid == blkid)) {
766 mdb_printf("%#lr\n", addr);
767 }
768 return (WALK_NEXT);
769 }
770
771 /* ARGSUSED */
772 static int
773 dbufs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
774 {
775 dbufs_data_t data;
776 char *object = NULL;
777 char *blkid = NULL;
778
779 data.objset = data.object = data.level = data.blkid = DBUFS_UNSET;
780 data.osname = NULL;
781
782 if (mdb_getopts(argc, argv,
783 'O', MDB_OPT_UINT64, &data.objset,
784 'n', MDB_OPT_STR, &data.osname,
785 'o', MDB_OPT_STR, &object,
786 'l', MDB_OPT_UINT64, &data.level,
787 'b', MDB_OPT_STR, &blkid) != argc) {
788 return (DCMD_USAGE);
789 }
790
791 if (object) {
792 if (strcmp(object, "mdn") == 0) {
793 data.object = DMU_META_DNODE_OBJECT;
794 } else {
795 data.object = mdb_strtoull(object);
796 }
797 }
798
799 if (blkid) {
800 if (strcmp(blkid, "bonus") == 0) {
801 data.blkid = DMU_BONUS_BLKID;
802 } else {
803 data.blkid = mdb_strtoull(blkid);
804 }
805 }
806
807 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "dmu_buf_impl", &data.id) == -1) {
808 mdb_warn("couldn't find struct dmu_buf_impl_t");
809 return (DCMD_ERR);
810 }
811
812 if (mdb_walk("dmu_buf_impl_t", dbufs_cb, &data) != 0) {
813 mdb_warn("can't walk dbufs");
814 return (DCMD_ERR);
815 }
816
817 return (DCMD_OK);
818 }
819
820 typedef struct abuf_find_data {
821 dva_t dva;
822 mdb_ctf_id_t id;
823 } abuf_find_data_t;
824
825 /* ARGSUSED */
826 static int
827 abuf_find_cb(uintptr_t addr, const void *unknown, void *arg)
828 {
829 abuf_find_data_t *data = arg;
830 dva_t dva;
831
832 if (GETMEMBID(addr, &data->id, b_dva, dva)) {
833 return (WALK_ERR);
834 }
835
836 if (dva.dva_word[0] == data->dva.dva_word[0] &&
837 dva.dva_word[1] == data->dva.dva_word[1]) {
838 mdb_printf("%#lr\n", addr);
839 }
840 return (WALK_NEXT);
841 }
842
843 /* ARGSUSED */
844 static int
845 abuf_find(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
846 {
847 abuf_find_data_t data;
848 GElf_Sym sym;
849 int i;
850 const char *syms[] = {
851 "ARC_mru",
852 "ARC_mru_ghost",
853 "ARC_mfu",
854 "ARC_mfu_ghost",
855 };
856
857 if (argc != 2)
858 return (DCMD_USAGE);
859
860 for (i = 0; i < 2; i ++) {
861 switch (argv[i].a_type) {
862 case MDB_TYPE_STRING:
863 data.dva.dva_word[i] = mdb_strtoull(argv[i].a_un.a_str);
864 break;
865 case MDB_TYPE_IMMEDIATE:
866 data.dva.dva_word[i] = argv[i].a_un.a_val;
867 break;
868 default:
869 return (DCMD_USAGE);
870 }
871 }
872
873 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "arc_buf_hdr", &data.id) == -1) {
874 mdb_warn("couldn't find struct arc_buf_hdr");
875 return (DCMD_ERR);
876 }
877
878 for (i = 0; i < sizeof (syms) / sizeof (syms[0]); i++) {
879 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, syms[i], &sym)) {
880 mdb_warn("can't find symbol %s", syms[i]);
881 return (DCMD_ERR);
882 }
883
884 if (mdb_pwalk("list", abuf_find_cb, &data, sym.st_value) != 0) {
885 mdb_warn("can't walk %s", syms[i]);
886 return (DCMD_ERR);
887 }
888 }
889
890 return (DCMD_OK);
891 }
892
893
894 typedef struct dbgmsg_arg {
895 boolean_t da_verbose;
896 boolean_t da_address;
897 } dbgmsg_arg_t;
898
899 /* ARGSUSED */
900 static int
901 dbgmsg_cb(uintptr_t addr, const void *unknown, void *arg)
902 {
903 static mdb_ctf_id_t id;
904 static boolean_t gotid;
905 static ulong_t off;
906
907 dbgmsg_arg_t *da = arg;
908 time_t timestamp;
909 char buf[1024];
910
911 if (!gotid) {
912 if (mdb_ctf_lookup_by_name(ZFS_STRUCT "zfs_dbgmsg", &id) ==
913 -1) {
914 mdb_warn("couldn't find struct zfs_dbgmsg");
915 return (WALK_ERR);
916 }
917 gotid = TRUE;
918 if (mdb_ctf_offsetof(id, "zdm_msg", &off) == -1) {
919 mdb_warn("couldn't find zdm_msg");
920 return (WALK_ERR);
921 }
922 off /= 8;
923 }
924
925
926 if (GETMEMBID(addr, &id, zdm_timestamp, timestamp)) {
927 return (WALK_ERR);
928 }
929
930 if (mdb_readstr(buf, sizeof (buf), addr + off) == -1) {
931 mdb_warn("failed to read zdm_msg at %p\n", addr + off);
932 return (DCMD_ERR);
933 }
934
935 if (da->da_address)
936 mdb_printf("%p ", addr);
937 if (da->da_verbose)
938 mdb_printf("%Y ", timestamp);
939
940 mdb_printf("%s\n", buf);
941
942 if (da->da_verbose)
943 (void) mdb_call_dcmd("whatis", addr, DCMD_ADDRSPEC, 0, NULL);
944
945 return (WALK_NEXT);
946 }
947
948 /* ARGSUSED */
949 static int
950 dbgmsg(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
951 {
952 GElf_Sym sym;
953 dbgmsg_arg_t da = { 0 };
954
955 if (mdb_getopts(argc, argv,
956 'v', MDB_OPT_SETBITS, B_TRUE, &da.da_verbose,
957 'a', MDB_OPT_SETBITS, B_TRUE, &da.da_address,
958 NULL) != argc)
959 return (DCMD_USAGE);
960
961 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "zfs_dbgmsgs", &sym)) {
962 mdb_warn("can't find zfs_dbgmsgs");
963 return (DCMD_ERR);
964 }
965
966 if (mdb_pwalk("list", dbgmsg_cb, &da, sym.st_value) != 0) {
967 mdb_warn("can't walk zfs_dbgmsgs");
968 return (DCMD_ERR);
969 }
970
971 return (DCMD_OK);
972 }
973
974 /*ARGSUSED*/
975 static int
976 arc_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
977 {
978 kstat_named_t *stats;
979 GElf_Sym sym;
980 int nstats, i;
981 uint_t opt_a = FALSE;
982 uint_t opt_b = FALSE;
983 uint_t shift = 0;
984 const char *suffix;
985
986 static const char *bytestats[] = {
987 "p", "c", "c_min", "c_max", "size", "duplicate_buffers_size",
988 "arc_meta_used", "arc_meta_limit", "arc_meta_max",
989 "arc_meta_min", "hdr_size", "data_size", "metadata_size",
990 "other_size", "anon_size", "anon_evictable_data",
991 "anon_evictable_metadata", "mru_size", "mru_evictable_data",
992 "mru_evictable_metadata", "mru_ghost_size",
993 "mru_ghost_evictable_data", "mru_ghost_evictable_metadata",
994 "mfu_size", "mfu_evictable_data", "mfu_evictable_metadata",
995 "mfu_ghost_size", "mfu_ghost_evictable_data",
996 "mfu_ghost_evictable_metadata", "evict_l2_cached",
997 "evict_l2_eligible", "evict_l2_ineligible", "l2_read_bytes",
998 "l2_write_bytes", "l2_size", "l2_asize", "l2_hdr_size",
999 "compressed_size", "uncompressed_size", "overhead_size",
1000 NULL
1001 };
1002
1003 static const char *extras[] = {
1004 "arc_no_grow", "arc_tempreserve",
1005 NULL
1006 };
1007
1008 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "arc_stats", &sym) == -1) {
1009 mdb_warn("failed to find 'arc_stats'");
1010 return (DCMD_ERR);
1011 }
1012
1013 stats = mdb_zalloc(sym.st_size, UM_SLEEP | UM_GC);
1014
1015 if (mdb_vread(stats, sym.st_size, sym.st_value) == -1) {
1016 mdb_warn("couldn't read 'arc_stats' at %p", sym.st_value);
1017 return (DCMD_ERR);
1018 }
1019
1020 nstats = sym.st_size / sizeof (kstat_named_t);
1021
1022 /* NB: -a / opt_a are ignored for backwards compatability */
1023 if (mdb_getopts(argc, argv,
1024 'a', MDB_OPT_SETBITS, TRUE, &opt_a,
1025 'b', MDB_OPT_SETBITS, TRUE, &opt_b,
1026 'k', MDB_OPT_SETBITS, 10, &shift,
1027 'm', MDB_OPT_SETBITS, 20, &shift,
1028 'g', MDB_OPT_SETBITS, 30, &shift,
1029 NULL) != argc)
1030 return (DCMD_USAGE);
1031
1032 if (!opt_b && !shift)
1033 shift = 20;
1034
1035 switch (shift) {
1036 case 0:
1037 suffix = "B";
1038 break;
1039 case 10:
1040 suffix = "KB";
1041 break;
1042 case 20:
1043 suffix = "MB";
1044 break;
1045 case 30:
1046 suffix = "GB";
1047 break;
1048 default:
1049 suffix = "XX";
1050 }
1051
1052 for (i = 0; i < nstats; i++) {
1053 int j;
1054 boolean_t bytes = B_FALSE;
1055
1056 for (j = 0; bytestats[j]; j++) {
1057 if (strcmp(stats[i].name, bytestats[j]) == 0) {
1058 bytes = B_TRUE;
1059 break;
1060 }
1061 }
1062
1063 if (bytes) {
1064 mdb_printf("%-25s = %9llu %s\n", stats[i].name,
1065 stats[i].value.ui64 >> shift, suffix);
1066 } else {
1067 mdb_printf("%-25s = %9llu\n", stats[i].name,
1068 stats[i].value.ui64);
1069 }
1070 }
1071
1072 for (i = 0; extras[i]; i++) {
1073 uint64_t buf;
1074
1075 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, extras[i], &sym) == -1) {
1076 mdb_warn("failed to find '%s'", extras[i]);
1077 return (DCMD_ERR);
1078 }
1079
1080 if (sym.st_size != sizeof (uint64_t) &&
1081 sym.st_size != sizeof (uint32_t)) {
1082 mdb_warn("expected scalar for variable '%s'\n",
1083 extras[i]);
1084 return (DCMD_ERR);
1085 }
1086
1087 if (mdb_vread(&buf, sym.st_size, sym.st_value) == -1) {
1088 mdb_warn("couldn't read '%s'", extras[i]);
1089 return (DCMD_ERR);
1090 }
1091
1092 mdb_printf("%-25s = ", extras[i]);
1093
1094 /* NB: all the 64-bit extras happen to be byte counts */
1095 if (sym.st_size == sizeof (uint64_t))
1096 mdb_printf("%9llu %s\n", buf >> shift, suffix);
1097
1098 if (sym.st_size == sizeof (uint32_t))
1099 mdb_printf("%9d\n", *((uint32_t *)&buf));
1100 }
1101 return (DCMD_OK);
1102 }
1103
1104 typedef struct mdb_spa_print {
1105 pool_state_t spa_state;
1106 char spa_name[ZFS_MAX_DATASET_NAME_LEN];
1107 uintptr_t spa_normal_class;
1108 } mdb_spa_print_t;
1109
1110
1111 const char histo_stars[] = "****************************************";
1112 const int histo_width = sizeof (histo_stars) - 1;
1113
1114 static void
1115 dump_histogram(const uint64_t *histo, int size, int offset)
1116 {
1117 int i;
1118 int minidx = size - 1;
1119 int maxidx = 0;
1120 uint64_t max = 0;
1121
1122 for (i = 0; i < size; i++) {
1123 if (histo[i] > max)
1124 max = histo[i];
1125 if (histo[i] > 0 && i > maxidx)
1126 maxidx = i;
1127 if (histo[i] > 0 && i < minidx)
1128 minidx = i;
1129 }
1130
1131 if (max < histo_width)
1132 max = histo_width;
1133
1134 for (i = minidx; i <= maxidx; i++) {
1135 mdb_printf("%3u: %6llu %s\n",
1136 i + offset, (u_longlong_t)histo[i],
1137 &histo_stars[(max - histo[i]) * histo_width / max]);
1138 }
1139 }
1140
1141 typedef struct mdb_metaslab_class {
1142 uint64_t mc_histogram[RANGE_TREE_HISTOGRAM_SIZE];
1143 } mdb_metaslab_class_t;
1144
1145 /*
1146 * spa_class_histogram(uintptr_t class_addr)
1147 *
1148 * Prints free space histogram for a device class
1149 *
1150 * Returns DCMD_OK, or DCMD_ERR.
1151 */
1152 static int
1153 spa_class_histogram(uintptr_t class_addr)
1154 {
1155 mdb_metaslab_class_t mc;
1156 if (mdb_ctf_vread(&mc, "metaslab_class_t",
1157 "mdb_metaslab_class_t", class_addr, 0) == -1)
1158 return (DCMD_ERR);
1159
1160 mdb_inc_indent(4);
1161 dump_histogram(mc.mc_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1162 mdb_dec_indent(4);
1163 return (DCMD_OK);
1164 }
1165
1166 /*
1167 * ::spa
1168 *
1169 * -c Print configuration information as well
1170 * -v Print vdev state
1171 * -e Print vdev error stats
1172 * -m Print vdev metaslab info
1173 * -M print vdev metaslab group info
1174 * -h Print histogram info (must be combined with -m or -M)
1175 *
1176 * Print a summarized spa_t. When given no arguments, prints out a table of all
1177 * active pools on the system.
1178 */
1179 /* ARGSUSED */
1180 static int
1181 spa_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1182 {
1183 const char *statetab[] = { "ACTIVE", "EXPORTED", "DESTROYED",
1184 "SPARE", "L2CACHE", "UNINIT", "UNAVAIL", "POTENTIAL" };
1185 const char *state;
1186 int spa_flags = 0;
1187
1188 if (mdb_getopts(argc, argv,
1189 'c', MDB_OPT_SETBITS, SPA_FLAG_CONFIG, &spa_flags,
1190 'v', MDB_OPT_SETBITS, SPA_FLAG_VDEVS, &spa_flags,
1191 'e', MDB_OPT_SETBITS, SPA_FLAG_ERRORS, &spa_flags,
1192 'M', MDB_OPT_SETBITS, SPA_FLAG_METASLAB_GROUPS, &spa_flags,
1193 'm', MDB_OPT_SETBITS, SPA_FLAG_METASLABS, &spa_flags,
1194 'h', MDB_OPT_SETBITS, SPA_FLAG_HISTOGRAMS, &spa_flags,
1195 NULL) != argc)
1196 return (DCMD_USAGE);
1197
1198 if (!(flags & DCMD_ADDRSPEC)) {
1199 if (mdb_walk_dcmd("spa", "spa", argc, argv) == -1) {
1200 mdb_warn("can't walk spa");
1201 return (DCMD_ERR);
1202 }
1203
1204 return (DCMD_OK);
1205 }
1206
1207 if (flags & DCMD_PIPE_OUT) {
1208 mdb_printf("%#lr\n", addr);
1209 return (DCMD_OK);
1210 }
1211
1212 if (DCMD_HDRSPEC(flags))
1213 mdb_printf("%<u>%-?s %9s %-*s%</u>\n", "ADDR", "STATE",
1214 sizeof (uintptr_t) == 4 ? 60 : 52, "NAME");
1215
1216 mdb_spa_print_t spa;
1217 if (mdb_ctf_vread(&spa, "spa_t", "mdb_spa_print_t", addr, 0) == -1)
1218 return (DCMD_ERR);
1219
1220 if (spa.spa_state < 0 || spa.spa_state > POOL_STATE_UNAVAIL)
1221 state = "UNKNOWN";
1222 else
1223 state = statetab[spa.spa_state];
1224
1225 mdb_printf("%0?p %9s %s\n", addr, state, spa.spa_name);
1226 if (spa_flags & SPA_FLAG_HISTOGRAMS)
1227 spa_class_histogram(spa.spa_normal_class);
1228
1229 if (spa_flags & SPA_FLAG_CONFIG) {
1230 mdb_printf("\n");
1231 mdb_inc_indent(4);
1232 if (mdb_call_dcmd("spa_config", addr, flags, 0,
1233 NULL) != DCMD_OK)
1234 return (DCMD_ERR);
1235 mdb_dec_indent(4);
1236 }
1237
1238 if (spa_flags & SPA_FLAG_ALL_VDEV) {
1239 mdb_arg_t v;
1240 char opts[100] = "-";
1241 int args =
1242 (spa_flags | SPA_FLAG_VDEVS) == SPA_FLAG_VDEVS ? 0 : 1;
1243
1244 if (spa_flags & SPA_FLAG_ERRORS)
1245 strcat(opts, "e");
1246 if (spa_flags & SPA_FLAG_METASLABS)
1247 strcat(opts, "m");
1248 if (spa_flags & SPA_FLAG_METASLAB_GROUPS)
1249 strcat(opts, "M");
1250 if (spa_flags & SPA_FLAG_HISTOGRAMS)
1251 strcat(opts, "h");
1252
1253 v.a_type = MDB_TYPE_STRING;
1254 v.a_un.a_str = opts;
1255
1256 mdb_printf("\n");
1257 mdb_inc_indent(4);
1258 if (mdb_call_dcmd("spa_vdevs", addr, flags, args,
1259 &v) != DCMD_OK)
1260 return (DCMD_ERR);
1261 mdb_dec_indent(4);
1262 }
1263
1264 return (DCMD_OK);
1265 }
1266
1267 typedef struct mdb_spa_config_spa {
1268 uintptr_t spa_config;
1269 } mdb_spa_config_spa_t;
1270
1271 /*
1272 * ::spa_config
1273 *
1274 * Given a spa_t, print the configuration information stored in spa_config.
1275 * Since it's just an nvlist, format it as an indented list of name=value pairs.
1276 * We simply read the value of spa_config and pass off to ::nvlist.
1277 */
1278 /* ARGSUSED */
1279 static int
1280 spa_print_config(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1281 {
1282 mdb_spa_config_spa_t spa;
1283
1284 if (argc != 0 || !(flags & DCMD_ADDRSPEC))
1285 return (DCMD_USAGE);
1286
1287 if (mdb_ctf_vread(&spa, ZFS_STRUCT "spa", "mdb_spa_config_spa_t",
1288 addr, 0) == -1)
1289 return (DCMD_ERR);
1290
1291 if (spa.spa_config == 0) {
1292 mdb_printf("(none)\n");
1293 return (DCMD_OK);
1294 }
1295
1296 return (mdb_call_dcmd("nvlist", spa.spa_config, flags,
1297 0, NULL));
1298 }
1299
1300
1301
1302 typedef struct mdb_range_tree {
1303 uint64_t rt_space;
1304 } mdb_range_tree_t;
1305
1306 typedef struct mdb_metaslab_group {
1307 uint64_t mg_fragmentation;
1308 uint64_t mg_histogram[RANGE_TREE_HISTOGRAM_SIZE];
1309 uintptr_t mg_vd;
1310 } mdb_metaslab_group_t;
1311
1312 typedef struct mdb_metaslab {
1313 uint64_t ms_id;
1314 uint64_t ms_start;
1315 uint64_t ms_size;
1316 int64_t ms_deferspace;
1317 uint64_t ms_fragmentation;
1318 uint64_t ms_weight;
1319 uintptr_t ms_alloctree[TXG_SIZE];
1320 uintptr_t ms_freeingtree;
1321 uintptr_t ms_freedtree;
1322 uintptr_t ms_tree;
1323 uintptr_t ms_sm;
1324 } mdb_metaslab_t;
1325
1326 typedef struct mdb_space_map_phys_t {
1327 uint64_t smp_alloc;
1328 uint64_t smp_histogram[SPACE_MAP_HISTOGRAM_SIZE];
1329 } mdb_space_map_phys_t;
1330
1331 typedef struct mdb_space_map {
1332 uint64_t sm_size;
1333 uint8_t sm_shift;
1334 uint64_t sm_alloc;
1335 uintptr_t sm_phys;
1336 } mdb_space_map_t;
1337
1338 typedef struct mdb_vdev {
1339 uintptr_t vdev_path;
1340 uintptr_t vdev_ms;
1341 uintptr_t vdev_ops;
1342 uint64_t vdev_ms_count;
1343 uint64_t vdev_id;
1344 vdev_stat_t vdev_stat;
1345 } mdb_vdev_t;
1346
1347 typedef struct mdb_vdev_ops {
1348 char vdev_op_type[16];
1349 } mdb_vdev_ops_t;
1350
1351 static int
1352 metaslab_stats(uintptr_t addr, int spa_flags)
1353 {
1354 mdb_vdev_t vdev;
1355 uintptr_t *vdev_ms;
1356
1357 if (mdb_ctf_vread(&vdev, "vdev_t", "mdb_vdev_t",
1358 (uintptr_t)addr, 0) == -1) {
1359 mdb_warn("failed to read vdev at %p\n", addr);
1360 return (DCMD_ERR);
1361 }
1362
1363 mdb_inc_indent(4);
1364 mdb_printf("%<u>%-?s %6s %20s %10s %9s%</u>\n", "ADDR", "ID",
1365 "OFFSET", "FREE", "FRAGMENTATION");
1366
1367 vdev_ms = mdb_alloc(vdev.vdev_ms_count * sizeof (void *),
1368 UM_SLEEP | UM_GC);
1369 if (mdb_vread(vdev_ms, vdev.vdev_ms_count * sizeof (void *),
1370 (uintptr_t)vdev.vdev_ms) == -1) {
1371 mdb_warn("failed to read vdev_ms at %p\n", vdev.vdev_ms);
1372 return (DCMD_ERR);
1373 }
1374
1375 for (int m = 0; m < vdev.vdev_ms_count; m++) {
1376 mdb_metaslab_t ms;
1377 mdb_space_map_t sm = { 0 };
1378 char free[NICENUM_BUFLEN];
1379
1380 if (mdb_ctf_vread(&ms, "metaslab_t", "mdb_metaslab_t",
1381 (uintptr_t)vdev_ms[m], 0) == -1)
1382 return (DCMD_ERR);
1383
1384 if (ms.ms_sm != NULL &&
1385 mdb_ctf_vread(&sm, "space_map_t", "mdb_space_map_t",
1386 ms.ms_sm, 0) == -1)
1387 return (DCMD_ERR);
1388
1389 mdb_nicenum(ms.ms_size - sm.sm_alloc, free);
1390
1391 mdb_printf("%0?p %6llu %20llx %10s ", vdev_ms[m], ms.ms_id,
1392 ms.ms_start, free);
1393 if (ms.ms_fragmentation == ZFS_FRAG_INVALID)
1394 mdb_printf("%9s\n", "-");
1395 else
1396 mdb_printf("%9llu%%\n", ms.ms_fragmentation);
1397
1398 if ((spa_flags & SPA_FLAG_HISTOGRAMS) && ms.ms_sm != NULL) {
1399 mdb_space_map_phys_t smp;
1400
1401 if (sm.sm_phys == NULL)
1402 continue;
1403
1404 (void) mdb_ctf_vread(&smp, "space_map_phys_t",
1405 "mdb_space_map_phys_t", sm.sm_phys, 0);
1406
1407 dump_histogram(smp.smp_histogram,
1408 SPACE_MAP_HISTOGRAM_SIZE, sm.sm_shift);
1409 }
1410 }
1411 mdb_dec_indent(4);
1412 return (DCMD_OK);
1413 }
1414
1415 static int
1416 metaslab_group_stats(uintptr_t addr, int spa_flags)
1417 {
1418 mdb_metaslab_group_t mg;
1419 if (mdb_ctf_vread(&mg, "metaslab_group_t", "mdb_metaslab_group_t",
1420 (uintptr_t)addr, 0) == -1) {
1421 mdb_warn("failed to read vdev_mg at %p\n", addr);
1422 return (DCMD_ERR);
1423 }
1424
1425 mdb_inc_indent(4);
1426 mdb_printf("%<u>%-?s %15s%</u>\n", "ADDR", "FRAGMENTATION");
1427 if (mg.mg_fragmentation == ZFS_FRAG_INVALID)
1428 mdb_printf("%0?p %15s\n", addr, "-");
1429 else
1430 mdb_printf("%0?p %15llu%%\n", addr, mg.mg_fragmentation);
1431
1432 if (spa_flags & SPA_FLAG_HISTOGRAMS)
1433 dump_histogram(mg.mg_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1434 mdb_dec_indent(4);
1435 return (DCMD_OK);
1436 }
1437
1438 /*
1439 * ::vdev
1440 *
1441 * Print out a summarized vdev_t, in the following form:
1442 *
1443 * ADDR STATE AUX DESC
1444 * fffffffbcde23df0 HEALTHY - /dev/dsk/c0t0d0
1445 *
1446 * If '-r' is specified, recursively visit all children.
1447 *
1448 * With '-e', the statistics associated with the vdev are printed as well.
1449 */
1450 static int
1451 do_print_vdev(uintptr_t addr, int flags, int depth, boolean_t recursive,
1452 int spa_flags)
1453 {
1454 vdev_t vdev;
1455 char desc[MAXNAMELEN];
1456 int c, children;
1457 uintptr_t *child;
1458 const char *state, *aux;
1459
1460 if (mdb_vread(&vdev, sizeof (vdev), (uintptr_t)addr) == -1) {
1461 mdb_warn("failed to read vdev_t at %p\n", (uintptr_t)addr);
1462 return (DCMD_ERR);
1463 }
1464
1465 if (flags & DCMD_PIPE_OUT) {
1466 mdb_printf("%#lr\n", addr);
1467 } else {
1468 if (vdev.vdev_path != NULL) {
1469 if (mdb_readstr(desc, sizeof (desc),
1470 (uintptr_t)vdev.vdev_path) == -1) {
1471 mdb_warn("failed to read vdev_path at %p\n",
1472 vdev.vdev_path);
1473 return (DCMD_ERR);
1474 }
1475 } else if (vdev.vdev_ops != NULL) {
1476 vdev_ops_t ops;
1477 if (mdb_vread(&ops, sizeof (ops),
1478 (uintptr_t)vdev.vdev_ops) == -1) {
1479 mdb_warn("failed to read vdev_ops at %p\n",
1480 vdev.vdev_ops);
1481 return (DCMD_ERR);
1482 }
1483 (void) strcpy(desc, ops.vdev_op_type);
1484 } else {
1485 (void) strcpy(desc, "<unknown>");
1486 }
1487
1488 if (depth == 0 && DCMD_HDRSPEC(flags))
1489 mdb_printf("%<u>%-?s %-9s %-12s %-*s%</u>\n",
1490 "ADDR", "STATE", "AUX",
1491 sizeof (uintptr_t) == 4 ? 43 : 35,
1492 "DESCRIPTION");
1493
1494 mdb_printf("%0?p ", addr);
1495
1496 switch (vdev.vdev_state) {
1497 case VDEV_STATE_CLOSED:
1498 state = "CLOSED";
1499 break;
1500 case VDEV_STATE_OFFLINE:
1501 state = "OFFLINE";
1502 break;
1503 case VDEV_STATE_CANT_OPEN:
1504 state = "CANT_OPEN";
1505 break;
1506 case VDEV_STATE_DEGRADED:
1507 state = "DEGRADED";
1508 break;
1509 case VDEV_STATE_HEALTHY:
1510 state = "HEALTHY";
1511 break;
1512 case VDEV_STATE_REMOVED:
1513 state = "REMOVED";
1514 break;
1515 case VDEV_STATE_FAULTED:
1516 state = "FAULTED";
1517 break;
1518 default:
1519 state = "UNKNOWN";
1520 break;
1521 }
1522
1523 switch (vdev.vdev_stat.vs_aux) {
1524 case VDEV_AUX_NONE:
1525 aux = "-";
1526 break;
1527 case VDEV_AUX_OPEN_FAILED:
1528 aux = "OPEN_FAILED";
1529 break;
1530 case VDEV_AUX_CORRUPT_DATA:
1531 aux = "CORRUPT_DATA";
1532 break;
1533 case VDEV_AUX_NO_REPLICAS:
1534 aux = "NO_REPLICAS";
1535 break;
1536 case VDEV_AUX_BAD_GUID_SUM:
1537 aux = "BAD_GUID_SUM";
1538 break;
1539 case VDEV_AUX_TOO_SMALL:
1540 aux = "TOO_SMALL";
1541 break;
1542 case VDEV_AUX_BAD_LABEL:
1543 aux = "BAD_LABEL";
1544 break;
1545 case VDEV_AUX_VERSION_NEWER:
1546 aux = "VERS_NEWER";
1547 break;
1548 case VDEV_AUX_VERSION_OLDER:
1549 aux = "VERS_OLDER";
1550 break;
1551 case VDEV_AUX_UNSUP_FEAT:
1552 aux = "UNSUP_FEAT";
1553 break;
1554 case VDEV_AUX_SPARED:
1555 aux = "SPARED";
1556 break;
1557 case VDEV_AUX_ERR_EXCEEDED:
1558 aux = "ERR_EXCEEDED";
1559 break;
1560 case VDEV_AUX_IO_FAILURE:
1561 aux = "IO_FAILURE";
1562 break;
1563 case VDEV_AUX_BAD_LOG:
1564 aux = "BAD_LOG";
1565 break;
1566 case VDEV_AUX_EXTERNAL:
1567 aux = "EXTERNAL";
1568 break;
1569 case VDEV_AUX_SPLIT_POOL:
1570 aux = "SPLIT_POOL";
1571 break;
1572 default:
1573 aux = "UNKNOWN";
1574 break;
1575 }
1576
1577 mdb_printf("%-9s %-12s %*s%s\n", state, aux, depth, "", desc);
1578
1579 if (spa_flags & SPA_FLAG_ERRORS) {
1580 vdev_stat_t *vs = &vdev.vdev_stat;
1581 int i;
1582
1583 mdb_inc_indent(4);
1584 mdb_printf("\n");
1585 mdb_printf("%<u> %12s %12s %12s %12s "
1586 "%12s%</u>\n", "READ", "WRITE", "FREE", "CLAIM",
1587 "IOCTL");
1588 mdb_printf("OPS ");
1589 for (i = 1; i < ZIO_TYPES; i++)
1590 mdb_printf("%11#llx%s", vs->vs_ops[i],
1591 i == ZIO_TYPES - 1 ? "" : " ");
1592 mdb_printf("\n");
1593 mdb_printf("BYTES ");
1594 for (i = 1; i < ZIO_TYPES; i++)
1595 mdb_printf("%11#llx%s", vs->vs_bytes[i],
1596 i == ZIO_TYPES - 1 ? "" : " ");
1597
1598
1599 mdb_printf("\n");
1600 mdb_printf("EREAD %10#llx\n", vs->vs_read_errors);
1601 mdb_printf("EWRITE %10#llx\n", vs->vs_write_errors);
1602 mdb_printf("ECKSUM %10#llx\n",
1603 vs->vs_checksum_errors);
1604 mdb_dec_indent(4);
1605 mdb_printf("\n");
1606 }
1607
1608 if (spa_flags & SPA_FLAG_METASLAB_GROUPS &&
1609 vdev.vdev_mg != NULL) {
1610 metaslab_group_stats((uintptr_t)vdev.vdev_mg,
1611 spa_flags);
1612 }
1613 if (spa_flags & SPA_FLAG_METASLABS && vdev.vdev_ms != NULL) {
1614 metaslab_stats((uintptr_t)addr, spa_flags);
1615 }
1616 }
1617
1618 children = vdev.vdev_children;
1619
1620 if (children == 0 || !recursive)
1621 return (DCMD_OK);
1622
1623 child = mdb_alloc(children * sizeof (void *), UM_SLEEP | UM_GC);
1624 if (mdb_vread(child, children * sizeof (void *),
1625 (uintptr_t)vdev.vdev_child) == -1) {
1626 mdb_warn("failed to read vdev children at %p", vdev.vdev_child);
1627 return (DCMD_ERR);
1628 }
1629
1630 for (c = 0; c < children; c++) {
1631 if (do_print_vdev(child[c], flags, depth + 2, recursive,
1632 spa_flags)) {
1633 return (DCMD_ERR);
1634 }
1635 }
1636
1637 return (DCMD_OK);
1638 }
1639
1640 static int
1641 vdev_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1642 {
1643 uint64_t depth = 0;
1644 boolean_t recursive = B_FALSE;
1645 int spa_flags = 0;
1646
1647 if (mdb_getopts(argc, argv,
1648 'e', MDB_OPT_SETBITS, SPA_FLAG_ERRORS, &spa_flags,
1649 'm', MDB_OPT_SETBITS, SPA_FLAG_METASLABS, &spa_flags,
1650 'M', MDB_OPT_SETBITS, SPA_FLAG_METASLAB_GROUPS, &spa_flags,
1651 'h', MDB_OPT_SETBITS, SPA_FLAG_HISTOGRAMS, &spa_flags,
1652 'r', MDB_OPT_SETBITS, TRUE, &recursive,
1653 'd', MDB_OPT_UINT64, &depth, NULL) != argc)
1654 return (DCMD_USAGE);
1655
1656 if (!(flags & DCMD_ADDRSPEC)) {
1657 mdb_warn("no vdev_t address given\n");
1658 return (DCMD_ERR);
1659 }
1660
1661 return (do_print_vdev(addr, flags, (int)depth, recursive, spa_flags));
1662 }
1663
1664 typedef struct mdb_metaslab_alloc_trace {
1665 uintptr_t mat_mg;
1666 uintptr_t mat_msp;
1667 uint64_t mat_size;
1668 uint64_t mat_weight;
1669 uint64_t mat_offset;
1670 uint32_t mat_dva_id;
1671 } mdb_metaslab_alloc_trace_t;
1672
1673 static void
1674 metaslab_print_weight(uint64_t weight)
1675 {
1676 char buf[100];
1677
1678 if (WEIGHT_IS_SPACEBASED(weight)) {
1679 mdb_nicenum(
1680 weight & ~(METASLAB_ACTIVE_MASK | METASLAB_WEIGHT_TYPE),
1681 buf);
1682 } else {
1683 char size[NICENUM_BUFLEN];
1684 mdb_nicenum(1ULL << WEIGHT_GET_INDEX(weight), size);
1685 (void) mdb_snprintf(buf, sizeof (buf), "%llu x %s",
1686 WEIGHT_GET_COUNT(weight), size);
1687 }
1688 mdb_printf("%11s ", buf);
1689 }
1690
1691 /* ARGSUSED */
1692 static int
1693 metaslab_weight(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1694 {
1695 uint64_t weight = 0;
1696 char active;
1697
1698 if (argc == 0 && (flags & DCMD_ADDRSPEC)) {
1699 if (mdb_vread(&weight, sizeof (uint64_t), addr) == -1) {
1700 mdb_warn("failed to read weight at %p\n", addr);
1701 return (DCMD_ERR);
1702 }
1703 } else if (argc == 1 && !(flags & DCMD_ADDRSPEC)) {
1704 weight = (argv[0].a_type == MDB_TYPE_IMMEDIATE) ?
1705 argv[0].a_un.a_val : mdb_strtoull(argv[0].a_un.a_str);
1706 } else {
1707 return (DCMD_USAGE);
1708 }
1709
1710 if (DCMD_HDRSPEC(flags)) {
1711 mdb_printf("%<u>%-6s %9s %9s%</u>\n",
1712 "ACTIVE", "ALGORITHM", "WEIGHT");
1713 }
1714
1715 if (weight & METASLAB_WEIGHT_PRIMARY)
1716 active = 'P';
1717 else if (weight & METASLAB_WEIGHT_SECONDARY)
1718 active = 'S';
1719 else
1720 active = '-';
1721 mdb_printf("%6c %8s ", active,
1722 WEIGHT_IS_SPACEBASED(weight) ? "SPACE" : "SEGMENT");
1723 metaslab_print_weight(weight);
1724 mdb_printf("\n");
1725
1726 return (DCMD_OK);
1727 }
1728
1729 /* ARGSUSED */
1730 static int
1731 metaslab_trace(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1732 {
1733 mdb_metaslab_alloc_trace_t mat;
1734 mdb_metaslab_group_t mg = { 0 };
1735 char result_type[100];
1736
1737 if (mdb_ctf_vread(&mat, "metaslab_alloc_trace_t",
1738 "mdb_metaslab_alloc_trace_t", addr, 0) == -1) {
1739 return (DCMD_ERR);
1740 }
1741
1742 if (!(flags & DCMD_PIPE_OUT) && DCMD_HDRSPEC(flags)) {
1743 mdb_printf("%<u>%6s %6s %8s %11s %18s %18s%</u>\n",
1744 "MSID", "DVA", "ASIZE", "WEIGHT", "RESULT", "VDEV");
1745 }
1746
1747 if (mat.mat_msp != NULL) {
1748 mdb_metaslab_t ms;
1749
1750 if (mdb_ctf_vread(&ms, "metaslab_t", "mdb_metaslab_t",
1751 mat.mat_msp, 0) == -1) {
1752 return (DCMD_ERR);
1753 }
1754 mdb_printf("%6llu ", ms.ms_id);
1755 } else {
1756 mdb_printf("%6s ", "-");
1757 }
1758
1759 mdb_printf("%6d %8llx ", mat.mat_dva_id, mat.mat_size);
1760
1761 metaslab_print_weight(mat.mat_weight);
1762
1763 if ((int64_t)mat.mat_offset < 0) {
1764 if (enum_lookup("enum trace_alloc_type", mat.mat_offset,
1765 "TRACE_", sizeof (result_type), result_type) == -1) {
1766 mdb_warn("Could not find enum for trace_alloc_type");
1767 return (DCMD_ERR);
1768 }
1769 mdb_printf("%18s ", result_type);
1770 } else {
1771 mdb_printf("%<b>%18llx%</b> ", mat.mat_offset);
1772 }
1773
1774 if (mat.mat_mg != NULL &&
1775 mdb_ctf_vread(&mg, "metaslab_group_t", "mdb_metaslab_group_t",
1776 mat.mat_mg, 0) == -1) {
1777 return (DCMD_ERR);
1778 }
1779
1780 if (mg.mg_vd != NULL) {
1781 mdb_vdev_t vdev;
1782 char desc[MAXNAMELEN];
1783
1784 if (mdb_ctf_vread(&vdev, "vdev_t", "mdb_vdev_t",
1785 mg.mg_vd, 0) == -1) {
1786 return (DCMD_ERR);
1787 }
1788
1789 if (vdev.vdev_path != NULL) {
1790 char path[MAXNAMELEN];
1791
1792 if (mdb_readstr(path, sizeof (path),
1793 vdev.vdev_path) == -1) {
1794 mdb_warn("failed to read vdev_path at %p\n",
1795 vdev.vdev_path);
1796 return (DCMD_ERR);
1797 }
1798 char *slash;
1799 if ((slash = strrchr(path, '/')) != NULL) {
1800 strcpy(desc, slash + 1);
1801 } else {
1802 strcpy(desc, path);
1803 }
1804 } else if (vdev.vdev_ops != NULL) {
1805 mdb_vdev_ops_t ops;
1806 if (mdb_ctf_vread(&ops, "vdev_ops_t", "mdb_vdev_ops_t",
1807 vdev.vdev_ops, 0) == -1) {
1808 mdb_warn("failed to read vdev_ops at %p\n",
1809 vdev.vdev_ops);
1810 return (DCMD_ERR);
1811 }
1812 (void) mdb_snprintf(desc, sizeof (desc),
1813 "%s-%llu", ops.vdev_op_type, vdev.vdev_id);
1814 } else {
1815 (void) strcpy(desc, "<unknown>");
1816 }
1817 mdb_printf("%18s\n", desc);
1818 }
1819
1820 return (DCMD_OK);
1821 }
1822
1823 typedef struct metaslab_walk_data {
1824 uint64_t mw_numvdevs;
1825 uintptr_t *mw_vdevs;
1826 int mw_curvdev;
1827 uint64_t mw_nummss;
1828 uintptr_t *mw_mss;
1829 int mw_curms;
1830 } metaslab_walk_data_t;
1831
1832 static int
1833 metaslab_walk_step(mdb_walk_state_t *wsp)
1834 {
1835 metaslab_walk_data_t *mw = wsp->walk_data;
1836 metaslab_t ms;
1837 uintptr_t msp;
1838
1839 if (mw->mw_curvdev >= mw->mw_numvdevs)
1840 return (WALK_DONE);
1841
1842 if (mw->mw_mss == NULL) {
1843 uintptr_t mssp;
1844 uintptr_t vdevp;
1845
1846 ASSERT(mw->mw_curms == 0);
1847 ASSERT(mw->mw_nummss == 0);
1848
1849 vdevp = mw->mw_vdevs[mw->mw_curvdev];
1850 if (GETMEMB(vdevp, "vdev", vdev_ms, mssp) ||
1851 GETMEMB(vdevp, "vdev", vdev_ms_count, mw->mw_nummss)) {
1852 return (WALK_ERR);
1853 }
1854
1855 mw->mw_mss = mdb_alloc(mw->mw_nummss * sizeof (void*),
1856 UM_SLEEP | UM_GC);
1857 if (mdb_vread(mw->mw_mss, mw->mw_nummss * sizeof (void*),
1858 mssp) == -1) {
1859 mdb_warn("failed to read vdev_ms at %p", mssp);
1860 return (WALK_ERR);
1861 }
1862 }
1863
1864 if (mw->mw_curms >= mw->mw_nummss) {
1865 mw->mw_mss = NULL;
1866 mw->mw_curms = 0;
1867 mw->mw_nummss = 0;
1868 mw->mw_curvdev++;
1869 return (WALK_NEXT);
1870 }
1871
1872 msp = mw->mw_mss[mw->mw_curms];
1873 if (mdb_vread(&ms, sizeof (metaslab_t), msp) == -1) {
1874 mdb_warn("failed to read metaslab_t at %p", msp);
1875 return (WALK_ERR);
1876 }
1877
1878 mw->mw_curms++;
1879
1880 return (wsp->walk_callback(msp, &ms, wsp->walk_cbdata));
1881 }
1882
1883 static int
1884 metaslab_walk_init(mdb_walk_state_t *wsp)
1885 {
1886 metaslab_walk_data_t *mw;
1887 uintptr_t root_vdevp;
1888 uintptr_t childp;
1889
1890 if (wsp->walk_addr == NULL) {
1891 mdb_warn("must supply address of spa_t\n");
1892 return (WALK_ERR);
1893 }
1894
1895 mw = mdb_zalloc(sizeof (metaslab_walk_data_t), UM_SLEEP | UM_GC);
1896
1897 if (GETMEMB(wsp->walk_addr, "spa", spa_root_vdev, root_vdevp) ||
1898 GETMEMB(root_vdevp, "vdev", vdev_children, mw->mw_numvdevs) ||
1899 GETMEMB(root_vdevp, "vdev", vdev_child, childp)) {
1900 return (DCMD_ERR);
1901 }
1902
1903 mw->mw_vdevs = mdb_alloc(mw->mw_numvdevs * sizeof (void *),
1904 UM_SLEEP | UM_GC);
1905 if (mdb_vread(mw->mw_vdevs, mw->mw_numvdevs * sizeof (void *),
1906 childp) == -1) {
1907 mdb_warn("failed to read root vdev children at %p", childp);
1908 return (DCMD_ERR);
1909 }
1910
1911 wsp->walk_data = mw;
1912
1913 return (WALK_NEXT);
1914 }
1915
1916 typedef struct mdb_spa {
1917 uintptr_t spa_dsl_pool;
1918 uintptr_t spa_root_vdev;
1919 } mdb_spa_t;
1920
1921 typedef struct mdb_dsl_pool {
1922 uintptr_t dp_root_dir;
1923 } mdb_dsl_pool_t;
1924
1925 typedef struct mdb_dsl_dir {
1926 uintptr_t dd_dbuf;
1927 int64_t dd_space_towrite[TXG_SIZE];
1928 } mdb_dsl_dir_t;
1929
1930 typedef struct mdb_dsl_dir_phys {
1931 uint64_t dd_used_bytes;
1932 uint64_t dd_compressed_bytes;
1933 uint64_t dd_uncompressed_bytes;
1934 } mdb_dsl_dir_phys_t;
1935
1936 typedef struct space_data {
1937 uint64_t ms_alloctree[TXG_SIZE];
1938 uint64_t ms_freeingtree;
1939 uint64_t ms_freedtree;
1940 uint64_t ms_tree;
1941 int64_t ms_deferspace;
1942 uint64_t avail;
1943 uint64_t nowavail;
1944 } space_data_t;
1945
1946 /* ARGSUSED */
1947 static int
1948 space_cb(uintptr_t addr, const void *unknown, void *arg)
1949 {
1950 space_data_t *sd = arg;
1951 mdb_metaslab_t ms;
1952 mdb_range_tree_t rt;
1953 mdb_space_map_t sm = { 0 };
1954 mdb_space_map_phys_t smp = { 0 };
1955 int i;
1956
1957 if (mdb_ctf_vread(&ms, "metaslab_t", "mdb_metaslab_t",
1958 addr, 0) == -1)
1959 return (WALK_ERR);
1960
1961 for (i = 0; i < TXG_SIZE; i++) {
1962 if (mdb_ctf_vread(&rt, "range_tree_t",
1963 "mdb_range_tree_t", ms.ms_alloctree[i], 0) == -1)
1964 return (WALK_ERR);
1965
1966 sd->ms_alloctree[i] += rt.rt_space;
1967
1968 }
1969
1970 if (mdb_ctf_vread(&rt, "range_tree_t",
1971 "mdb_range_tree_t", ms.ms_freeingtree, 0) == -1)
1972 return (WALK_ERR);
1973 sd->ms_freeingtree += rt.rt_space;
1974
1975 if (mdb_ctf_vread(&rt, "range_tree_t",
1976 "mdb_range_tree_t", ms.ms_freedtree, 0) == -1)
1977 return (WALK_ERR);
1978 sd->ms_freedtree += rt.rt_space;
1979
1980 if (mdb_ctf_vread(&rt, "range_tree_t",
1981 "mdb_range_tree_t", ms.ms_tree, 0) == -1)
1982 return (WALK_ERR);
1983 sd->ms_tree += rt.rt_space;
1984
1985 if (ms.ms_sm != NULL &&
1986 mdb_ctf_vread(&sm, "space_map_t",
1987 "mdb_space_map_t", ms.ms_sm, 0) == -1)
1988 return (WALK_ERR);
1989
1990 if (sm.sm_phys != NULL) {
1991 (void) mdb_ctf_vread(&smp, "space_map_phys_t",
1992 "mdb_space_map_phys_t", sm.sm_phys, 0);
1993 }
1994
1995 sd->ms_deferspace += ms.ms_deferspace;
1996 sd->avail += sm.sm_size - sm.sm_alloc;
1997 sd->nowavail += sm.sm_size - smp.smp_alloc;
1998
1999 return (WALK_NEXT);
2000 }
2001
2002 /*
2003 * ::spa_space [-b]
2004 *
2005 * Given a spa_t, print out it's on-disk space usage and in-core
2006 * estimates of future usage. If -b is given, print space in bytes.
2007 * Otherwise print in megabytes.
2008 */
2009 /* ARGSUSED */
2010 static int
2011 spa_space(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2012 {
2013 mdb_spa_t spa;
2014 mdb_dsl_pool_t dp;
2015 mdb_dsl_dir_t dd;
2016 mdb_dmu_buf_impl_t db;
2017 mdb_dsl_dir_phys_t dsp;
2018 space_data_t sd;
2019 int shift = 20;
2020 char *suffix = "M";
2021 int bytes = B_FALSE;
2022
2023 if (mdb_getopts(argc, argv, 'b', MDB_OPT_SETBITS, TRUE, &bytes, NULL) !=
2024 argc)
2025 return (DCMD_USAGE);
2026 if (!(flags & DCMD_ADDRSPEC))
2027 return (DCMD_USAGE);
2028
2029 if (bytes) {
2030 shift = 0;
2031 suffix = "";
2032 }
2033
2034 if (mdb_ctf_vread(&spa, ZFS_STRUCT "spa", "mdb_spa_t",
2035 addr, 0) == -1 ||
2036 mdb_ctf_vread(&dp, ZFS_STRUCT "dsl_pool", "mdb_dsl_pool_t",
2037 spa.spa_dsl_pool, 0) == -1 ||
2038 mdb_ctf_vread(&dd, ZFS_STRUCT "dsl_dir", "mdb_dsl_dir_t",
2039 dp.dp_root_dir, 0) == -1 ||
2040 mdb_ctf_vread(&db, ZFS_STRUCT "dmu_buf_impl", "mdb_dmu_buf_impl_t",
2041 dd.dd_dbuf, 0) == -1 ||
2042 mdb_ctf_vread(&dsp, ZFS_STRUCT "dsl_dir_phys",
2043 "mdb_dsl_dir_phys_t", db.db.db_data, 0) == -1) {
2044 return (DCMD_ERR);
2045 }
2046
2047 mdb_printf("dd_space_towrite = %llu%s %llu%s %llu%s %llu%s\n",
2048 dd.dd_space_towrite[0] >> shift, suffix,
2049 dd.dd_space_towrite[1] >> shift, suffix,
2050 dd.dd_space_towrite[2] >> shift, suffix,
2051 dd.dd_space_towrite[3] >> shift, suffix);
2052
2053 mdb_printf("dd_phys.dd_used_bytes = %llu%s\n",
2054 dsp.dd_used_bytes >> shift, suffix);
2055 mdb_printf("dd_phys.dd_compressed_bytes = %llu%s\n",
2056 dsp.dd_compressed_bytes >> shift, suffix);
2057 mdb_printf("dd_phys.dd_uncompressed_bytes = %llu%s\n",
2058 dsp.dd_uncompressed_bytes >> shift, suffix);
2059
2060 bzero(&sd, sizeof (sd));
2061 if (mdb_pwalk("metaslab", space_cb, &sd, addr) != 0) {
2062 mdb_warn("can't walk metaslabs");
2063 return (DCMD_ERR);
2064 }
2065
2066 mdb_printf("ms_allocmap = %llu%s %llu%s %llu%s %llu%s\n",
2067 sd.ms_alloctree[0] >> shift, suffix,
2068 sd.ms_alloctree[1] >> shift, suffix,
2069 sd.ms_alloctree[2] >> shift, suffix,
2070 sd.ms_alloctree[3] >> shift, suffix);
2071 mdb_printf("ms_freeingtree = %llu%s\n",
2072 sd.ms_freeingtree >> shift, suffix);
2073 mdb_printf("ms_freedtree = %llu%s\n",
2074 sd.ms_freedtree >> shift, suffix);
2075 mdb_printf("ms_tree = %llu%s\n",
2076 sd.ms_tree >> shift, suffix);
2077 mdb_printf("ms_deferspace = %llu%s\n",
2078 sd.ms_deferspace >> shift, suffix);
2079 mdb_printf("last synced avail = %llu%s\n",
2080 sd.avail >> shift, suffix);
2081 mdb_printf("current syncing avail = %llu%s\n",
2082 sd.nowavail >> shift, suffix);
2083
2084 return (DCMD_OK);
2085 }
2086
2087 typedef struct mdb_spa_aux_vdev {
2088 int sav_count;
2089 uintptr_t sav_vdevs;
2090 } mdb_spa_aux_vdev_t;
2091
2092 typedef struct mdb_spa_vdevs {
2093 uintptr_t spa_root_vdev;
2094 mdb_spa_aux_vdev_t spa_l2cache;
2095 mdb_spa_aux_vdev_t spa_spares;
2096 } mdb_spa_vdevs_t;
2097
2098 static int
2099 spa_print_aux(mdb_spa_aux_vdev_t *sav, uint_t flags, mdb_arg_t *v,
2100 const char *name)
2101 {
2102 uintptr_t *aux;
2103 size_t len;
2104 int ret, i;
2105
2106 /*
2107 * Iterate over aux vdevs and print those out as well. This is a
2108 * little annoying because we don't have a root vdev to pass to ::vdev.
2109 * Instead, we print a single line and then call it for each child
2110 * vdev.
2111 */
2112 if (sav->sav_count != 0) {
2113 v[1].a_type = MDB_TYPE_STRING;
2114 v[1].a_un.a_str = "-d";
2115 v[2].a_type = MDB_TYPE_IMMEDIATE;
2116 v[2].a_un.a_val = 2;
2117
2118 len = sav->sav_count * sizeof (uintptr_t);
2119 aux = mdb_alloc(len, UM_SLEEP);
2120 if (mdb_vread(aux, len, sav->sav_vdevs) == -1) {
2121 mdb_free(aux, len);
2122 mdb_warn("failed to read l2cache vdevs at %p",
2123 sav->sav_vdevs);
2124 return (DCMD_ERR);
2125 }
2126
2127 mdb_printf("%-?s %-9s %-12s %s\n", "-", "-", "-", name);
2128
2129 for (i = 0; i < sav->sav_count; i++) {
2130 ret = mdb_call_dcmd("vdev", aux[i], flags, 3, v);
2131 if (ret != DCMD_OK) {
2132 mdb_free(aux, len);
2133 return (ret);
2134 }
2135 }
2136
2137 mdb_free(aux, len);
2138 }
2139
2140 return (0);
2141 }
2142
2143 /*
2144 * ::spa_vdevs
2145 *
2146 * -e Include error stats
2147 * -m Include metaslab information
2148 * -M Include metaslab group information
2149 * -h Include histogram information (requires -m or -M)
2150 *
2151 * Print out a summarized list of vdevs for the given spa_t.
2152 * This is accomplished by invoking "::vdev -re" on the root vdev, as well as
2153 * iterating over the cache devices.
2154 */
2155 /* ARGSUSED */
2156 static int
2157 spa_vdevs(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2158 {
2159 mdb_arg_t v[3];
2160 int ret;
2161 char opts[100] = "-r";
2162 int spa_flags = 0;
2163
2164 if (mdb_getopts(argc, argv,
2165 'e', MDB_OPT_SETBITS, SPA_FLAG_ERRORS, &spa_flags,
2166 'm', MDB_OPT_SETBITS, SPA_FLAG_METASLABS, &spa_flags,
2167 'M', MDB_OPT_SETBITS, SPA_FLAG_METASLAB_GROUPS, &spa_flags,
2168 'h', MDB_OPT_SETBITS, SPA_FLAG_HISTOGRAMS, &spa_flags,
2169 NULL) != argc)
2170 return (DCMD_USAGE);
2171
2172 if (!(flags & DCMD_ADDRSPEC))
2173 return (DCMD_USAGE);
2174
2175 mdb_spa_vdevs_t spa;
2176 if (mdb_ctf_vread(&spa, "spa_t", "mdb_spa_vdevs_t", addr, 0) == -1)
2177 return (DCMD_ERR);
2178
2179 /*
2180 * Unitialized spa_t structures can have a NULL root vdev.
2181 */
2182 if (spa.spa_root_vdev == NULL) {
2183 mdb_printf("no associated vdevs\n");
2184 return (DCMD_OK);
2185 }
2186
2187 if (spa_flags & SPA_FLAG_ERRORS)
2188 strcat(opts, "e");
2189 if (spa_flags & SPA_FLAG_METASLABS)
2190 strcat(opts, "m");
2191 if (spa_flags & SPA_FLAG_METASLAB_GROUPS)
2192 strcat(opts, "M");
2193 if (spa_flags & SPA_FLAG_HISTOGRAMS)
2194 strcat(opts, "h");
2195
2196 v[0].a_type = MDB_TYPE_STRING;
2197 v[0].a_un.a_str = opts;
2198
2199 ret = mdb_call_dcmd("vdev", (uintptr_t)spa.spa_root_vdev,
2200 flags, 1, v);
2201 if (ret != DCMD_OK)
2202 return (ret);
2203
2204 if (spa_print_aux(&spa.spa_l2cache, flags, v, "cache") != 0 ||
2205 spa_print_aux(&spa.spa_spares, flags, v, "spares") != 0)
2206 return (DCMD_ERR);
2207
2208 return (DCMD_OK);
2209 }
2210
2211 /*
2212 * ::zio
2213 *
2214 * Print a summary of zio_t and all its children. This is intended to display a
2215 * zio tree, and hence we only pick the most important pieces of information for
2216 * the main summary. More detailed information can always be found by doing a
2217 * '::print zio' on the underlying zio_t. The columns we display are:
2218 *
2219 * ADDRESS TYPE STAGE WAITER TIME_ELAPSED
2220 *
2221 * The 'address' column is indented by one space for each depth level as we
2222 * descend down the tree.
2223 */
2224
2225 #define ZIO_MAXINDENT 7
2226 #define ZIO_MAXWIDTH (sizeof (uintptr_t) * 2 + ZIO_MAXINDENT)
2227 #define ZIO_WALK_SELF 0
2228 #define ZIO_WALK_CHILD 1
2229 #define ZIO_WALK_PARENT 2
2230
2231 typedef struct zio_print_args {
2232 int zpa_current_depth;
2233 int zpa_min_depth;
2234 int zpa_max_depth;
2235 int zpa_type;
2236 uint_t zpa_flags;
2237 } zio_print_args_t;
2238
2239 typedef struct mdb_zio {
2240 enum zio_type io_type;
2241 enum zio_stage io_stage;
2242 uintptr_t io_waiter;
2243 uintptr_t io_spa;
2244 struct {
2245 struct {
2246 uintptr_t list_next;
2247 } list_head;
2248 } io_parent_list;
2249 int io_error;
2250 } mdb_zio_t;
2251
2252 typedef struct mdb_zio_timestamp {
2253 hrtime_t io_timestamp;
2254 } mdb_zio_timestamp_t;
2255
2256 static int zio_child_cb(uintptr_t addr, const void *unknown, void *arg);
2257
2258 static int
2259 zio_print_cb(uintptr_t addr, zio_print_args_t *zpa)
2260 {
2261 mdb_ctf_id_t type_enum, stage_enum;
2262 int indent = zpa->zpa_current_depth;
2263 const char *type, *stage;
2264 uintptr_t laddr;
2265 mdb_zio_t zio;
2266 mdb_zio_timestamp_t zio_timestamp = { 0 };
2267
2268 if (mdb_ctf_vread(&zio, ZFS_STRUCT "zio", "mdb_zio_t", addr, 0) == -1)
2269 return (WALK_ERR);
2270 (void) mdb_ctf_vread(&zio_timestamp, ZFS_STRUCT "zio",
2271 "mdb_zio_timestamp_t", addr, MDB_CTF_VREAD_QUIET);
2272
2273 if (indent > ZIO_MAXINDENT)
2274 indent = ZIO_MAXINDENT;
2275
2276 if (mdb_ctf_lookup_by_name("enum zio_type", &type_enum) == -1 ||
2277 mdb_ctf_lookup_by_name("enum zio_stage", &stage_enum) == -1) {
2278 mdb_warn("failed to lookup zio enums");
2279 return (WALK_ERR);
2280 }
2281
2282 if ((type = mdb_ctf_enum_name(type_enum, zio.io_type)) != NULL)
2283 type += sizeof ("ZIO_TYPE_") - 1;
2284 else
2285 type = "?";
2286
2287 if (zio.io_error == 0) {
2288 stage = mdb_ctf_enum_name(stage_enum, zio.io_stage);
2289 if (stage != NULL)
2290 stage += sizeof ("ZIO_STAGE_") - 1;
2291 else
2292 stage = "?";
2293 } else {
2294 stage = "FAILED";
2295 }
2296
2297 if (zpa->zpa_current_depth >= zpa->zpa_min_depth) {
2298 if (zpa->zpa_flags & DCMD_PIPE_OUT) {
2299 mdb_printf("%?p\n", addr);
2300 } else {
2301 mdb_printf("%*s%-*p %-5s %-16s ", indent, "",
2302 ZIO_MAXWIDTH - indent, addr, type, stage);
2303 if (zio.io_waiter != 0)
2304 mdb_printf("%-16lx ", zio.io_waiter);
2305 else
2306 mdb_printf("%-16s ", "-");
2307 #ifdef _KERNEL
2308 if (zio_timestamp.io_timestamp != 0) {
2309 mdb_printf("%llums", (mdb_gethrtime() -
2310 zio_timestamp.io_timestamp) /
2311 1000000);
2312 } else {
2313 mdb_printf("%-12s ", "-");
2314 }
2315 #else
2316 mdb_printf("%-12s ", "-");
2317 #endif
2318 mdb_printf("\n");
2319 }
2320 }
2321
2322 if (zpa->zpa_current_depth >= zpa->zpa_max_depth)
2323 return (WALK_NEXT);
2324
2325 if (zpa->zpa_type == ZIO_WALK_PARENT)
2326 laddr = addr + mdb_ctf_offsetof_by_name(ZFS_STRUCT "zio",
2327 "io_parent_list");
2328 else
2329 laddr = addr + mdb_ctf_offsetof_by_name(ZFS_STRUCT "zio",
2330 "io_child_list");
2331
2332 zpa->zpa_current_depth++;
2333 if (mdb_pwalk("list", zio_child_cb, zpa, laddr) != 0) {
2334 mdb_warn("failed to walk zio_t children at %p\n", laddr);
2335 return (WALK_ERR);
2336 }
2337 zpa->zpa_current_depth--;
2338
2339 return (WALK_NEXT);
2340 }
2341
2342 /* ARGSUSED */
2343 static int
2344 zio_child_cb(uintptr_t addr, const void *unknown, void *arg)
2345 {
2346 zio_link_t zl;
2347 uintptr_t ziop;
2348 zio_print_args_t *zpa = arg;
2349
2350 if (mdb_vread(&zl, sizeof (zl), addr) == -1) {
2351 mdb_warn("failed to read zio_link_t at %p", addr);
2352 return (WALK_ERR);
2353 }
2354
2355 if (zpa->zpa_type == ZIO_WALK_PARENT)
2356 ziop = (uintptr_t)zl.zl_parent;
2357 else
2358 ziop = (uintptr_t)zl.zl_child;
2359
2360 return (zio_print_cb(ziop, zpa));
2361 }
2362
2363 /* ARGSUSED */
2364 static int
2365 zio_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2366 {
2367 zio_print_args_t zpa = { 0 };
2368
2369 if (!(flags & DCMD_ADDRSPEC))
2370 return (DCMD_USAGE);
2371
2372 if (mdb_getopts(argc, argv,
2373 'r', MDB_OPT_SETBITS, INT_MAX, &zpa.zpa_max_depth,
2374 'c', MDB_OPT_SETBITS, ZIO_WALK_CHILD, &zpa.zpa_type,
2375 'p', MDB_OPT_SETBITS, ZIO_WALK_PARENT, &zpa.zpa_type,
2376 NULL) != argc)
2377 return (DCMD_USAGE);
2378
2379 zpa.zpa_flags = flags;
2380 if (zpa.zpa_max_depth != 0) {
2381 if (zpa.zpa_type == ZIO_WALK_SELF)
2382 zpa.zpa_type = ZIO_WALK_CHILD;
2383 } else if (zpa.zpa_type != ZIO_WALK_SELF) {
2384 zpa.zpa_min_depth = 1;
2385 zpa.zpa_max_depth = 1;
2386 }
2387
2388 if (!(flags & DCMD_PIPE_OUT) && DCMD_HDRSPEC(flags)) {
2389 mdb_printf("%<u>%-*s %-5s %-16s %-16s %-12s%</u>\n",
2390 ZIO_MAXWIDTH, "ADDRESS", "TYPE", "STAGE", "WAITER",
2391 "TIME_ELAPSED");
2392 }
2393
2394 if (zio_print_cb(addr, &zpa) != WALK_NEXT)
2395 return (DCMD_ERR);
2396
2397 return (DCMD_OK);
2398 }
2399
2400 /*
2401 * [addr]::zio_state
2402 *
2403 * Print a summary of all zio_t structures on the system, or for a particular
2404 * pool. This is equivalent to '::walk zio_root | ::zio'.
2405 */
2406 /*ARGSUSED*/
2407 static int
2408 zio_state(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2409 {
2410 /*
2411 * MDB will remember the last address of the pipeline, so if we don't
2412 * zero this we'll end up trying to walk zio structures for a
2413 * non-existent spa_t.
2414 */
2415 if (!(flags & DCMD_ADDRSPEC))
2416 addr = 0;
2417
2418 return (mdb_pwalk_dcmd("zio_root", "zio", argc, argv, addr));
2419 }
2420
2421 typedef struct mdb_multilist {
2422 uint64_t ml_num_sublists;
2423 uintptr_t ml_sublists;
2424 } mdb_multilist_t;
2425
2426 typedef struct multilist_walk_data {
2427 uint64_t mwd_idx;
2428 mdb_multilist_t mwd_ml;
2429 } multilist_walk_data_t;
2430
2431 /* ARGSUSED */
2432 static int
2433 multilist_print_cb(uintptr_t addr, const void *unknown, void *arg)
2434 {
2435 mdb_printf("%#lr\n", addr);
2436 return (WALK_NEXT);
2437 }
2438
2439 static int
2440 multilist_walk_step(mdb_walk_state_t *wsp)
2441 {
2442 multilist_walk_data_t *mwd = wsp->walk_data;
2443
2444 if (mwd->mwd_idx >= mwd->mwd_ml.ml_num_sublists)
2445 return (WALK_DONE);
2446
2447 wsp->walk_addr = mwd->mwd_ml.ml_sublists +
2448 mdb_ctf_sizeof_by_name("multilist_sublist_t") * mwd->mwd_idx +
2449 mdb_ctf_offsetof_by_name("multilist_sublist_t", "mls_list");
2450
2451 mdb_pwalk("list", multilist_print_cb, (void*)NULL, wsp->walk_addr);
2452 mwd->mwd_idx++;
2453
2454 return (WALK_NEXT);
2455 }
2456
2457 static int
2458 multilist_walk_init(mdb_walk_state_t *wsp)
2459 {
2460 multilist_walk_data_t *mwd;
2461
2462 if (wsp->walk_addr == NULL) {
2463 mdb_warn("must supply address of multilist_t\n");
2464 return (WALK_ERR);
2465 }
2466
2467 mwd = mdb_zalloc(sizeof (multilist_walk_data_t), UM_SLEEP | UM_GC);
2468 if (mdb_ctf_vread(&mwd->mwd_ml, "multilist_t", "mdb_multilist_t",
2469 wsp->walk_addr, 0) == -1) {
2470 return (WALK_ERR);
2471 }
2472
2473 if (mwd->mwd_ml.ml_num_sublists == 0 ||
2474 mwd->mwd_ml.ml_sublists == NULL) {
2475 mdb_warn("invalid or uninitialized multilist at %#lx\n",
2476 wsp->walk_addr);
2477 return (WALK_ERR);
2478 }
2479
2480 wsp->walk_data = mwd;
2481 return (WALK_NEXT);
2482 }
2483
2484 typedef struct mdb_txg_list {
2485 size_t tl_offset;
2486 uintptr_t tl_head[TXG_SIZE];
2487 } mdb_txg_list_t;
2488
2489 typedef struct txg_list_walk_data {
2490 uintptr_t lw_head[TXG_SIZE];
2491 int lw_txgoff;
2492 int lw_maxoff;
2493 size_t lw_offset;
2494 void *lw_obj;
2495 } txg_list_walk_data_t;
2496
2497 static int
2498 txg_list_walk_init_common(mdb_walk_state_t *wsp, int txg, int maxoff)
2499 {
2500 txg_list_walk_data_t *lwd;
2501 mdb_txg_list_t list;
2502 int i;
2503
2504 lwd = mdb_alloc(sizeof (txg_list_walk_data_t), UM_SLEEP | UM_GC);
2505 if (mdb_ctf_vread(&list, "txg_list_t", "mdb_txg_list_t", wsp->walk_addr,
2506 0) == -1) {
2507 mdb_warn("failed to read txg_list_t at %#lx", wsp->walk_addr);
2508 return (WALK_ERR);
2509 }
2510
2511 for (i = 0; i < TXG_SIZE; i++)
2512 lwd->lw_head[i] = list.tl_head[i];
2513 lwd->lw_offset = list.tl_offset;
2514 lwd->lw_obj = mdb_alloc(lwd->lw_offset + sizeof (txg_node_t),
2515 UM_SLEEP | UM_GC);
2516 lwd->lw_txgoff = txg;
2517 lwd->lw_maxoff = maxoff;
2518
2519 wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
2520 wsp->walk_data = lwd;
2521
2522 return (WALK_NEXT);
2523 }
2524
2525 static int
2526 txg_list_walk_init(mdb_walk_state_t *wsp)
2527 {
2528 return (txg_list_walk_init_common(wsp, 0, TXG_SIZE-1));
2529 }
2530
2531 static int
2532 txg_list0_walk_init(mdb_walk_state_t *wsp)
2533 {
2534 return (txg_list_walk_init_common(wsp, 0, 0));
2535 }
2536
2537 static int
2538 txg_list1_walk_init(mdb_walk_state_t *wsp)
2539 {
2540 return (txg_list_walk_init_common(wsp, 1, 1));
2541 }
2542
2543 static int
2544 txg_list2_walk_init(mdb_walk_state_t *wsp)
2545 {
2546 return (txg_list_walk_init_common(wsp, 2, 2));
2547 }
2548
2549 static int
2550 txg_list3_walk_init(mdb_walk_state_t *wsp)
2551 {
2552 return (txg_list_walk_init_common(wsp, 3, 3));
2553 }
2554
2555 static int
2556 txg_list_walk_step(mdb_walk_state_t *wsp)
2557 {
2558 txg_list_walk_data_t *lwd = wsp->walk_data;
2559 uintptr_t addr;
2560 txg_node_t *node;
2561 int status;
2562
2563 while (wsp->walk_addr == NULL && lwd->lw_txgoff < lwd->lw_maxoff) {
2564 lwd->lw_txgoff++;
2565 wsp->walk_addr = lwd->lw_head[lwd->lw_txgoff];
2566 }
2567
2568 if (wsp->walk_addr == NULL)
2569 return (WALK_DONE);
2570
2571 addr = wsp->walk_addr - lwd->lw_offset;
2572
2573 if (mdb_vread(lwd->lw_obj,
2574 lwd->lw_offset + sizeof (txg_node_t), addr) == -1) {
2575 mdb_warn("failed to read list element at %#lx", addr);
2576 return (WALK_ERR);
2577 }
2578
2579 status = wsp->walk_callback(addr, lwd->lw_obj, wsp->walk_cbdata);
2580 node = (txg_node_t *)((uintptr_t)lwd->lw_obj + lwd->lw_offset);
2581 wsp->walk_addr = (uintptr_t)node->tn_next[lwd->lw_txgoff];
2582
2583 return (status);
2584 }
2585
2586 /*
2587 * ::walk spa
2588 *
2589 * Walk all named spa_t structures in the namespace. This is nothing more than
2590 * a layered avl walk.
2591 */
2592 static int
2593 spa_walk_init(mdb_walk_state_t *wsp)
2594 {
2595 GElf_Sym sym;
2596
2597 if (wsp->walk_addr != NULL) {
2598 mdb_warn("spa walk only supports global walks\n");
2599 return (WALK_ERR);
2600 }
2601
2602 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "spa_namespace_avl", &sym) == -1) {
2603 mdb_warn("failed to find symbol 'spa_namespace_avl'");
2604 return (WALK_ERR);
2605 }
2606
2607 wsp->walk_addr = (uintptr_t)sym.st_value;
2608
2609 if (mdb_layered_walk("avl", wsp) == -1) {
2610 mdb_warn("failed to walk 'avl'\n");
2611 return (WALK_ERR);
2612 }
2613
2614 return (WALK_NEXT);
2615 }
2616
2617 static int
2618 spa_walk_step(mdb_walk_state_t *wsp)
2619 {
2620 return (wsp->walk_callback(wsp->walk_addr, NULL, wsp->walk_cbdata));
2621 }
2622
2623 /*
2624 * [addr]::walk zio
2625 *
2626 * Walk all active zio_t structures on the system. This is simply a layered
2627 * walk on top of ::walk zio_cache, with the optional ability to limit the
2628 * structures to a particular pool.
2629 */
2630 static int
2631 zio_walk_init(mdb_walk_state_t *wsp)
2632 {
2633 wsp->walk_data = (void *)wsp->walk_addr;
2634
2635 if (mdb_layered_walk("zio_cache", wsp) == -1) {
2636 mdb_warn("failed to walk 'zio_cache'\n");
2637 return (WALK_ERR);
2638 }
2639
2640 return (WALK_NEXT);
2641 }
2642
2643 static int
2644 zio_walk_step(mdb_walk_state_t *wsp)
2645 {
2646 mdb_zio_t zio;
2647 uintptr_t spa = (uintptr_t)wsp->walk_data;
2648
2649 if (mdb_ctf_vread(&zio, ZFS_STRUCT "zio", "mdb_zio_t",
2650 wsp->walk_addr, 0) == -1)
2651 return (WALK_ERR);
2652
2653 if (spa != 0 && spa != zio.io_spa)
2654 return (WALK_NEXT);
2655
2656 return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
2657 }
2658
2659 /*
2660 * [addr]::walk zio_root
2661 *
2662 * Walk only root zio_t structures, optionally for a particular spa_t.
2663 */
2664 static int
2665 zio_walk_root_step(mdb_walk_state_t *wsp)
2666 {
2667 mdb_zio_t zio;
2668 uintptr_t spa = (uintptr_t)wsp->walk_data;
2669
2670 if (mdb_ctf_vread(&zio, ZFS_STRUCT "zio", "mdb_zio_t",
2671 wsp->walk_addr, 0) == -1)
2672 return (WALK_ERR);
2673
2674 if (spa != 0 && spa != zio.io_spa)
2675 return (WALK_NEXT);
2676
2677 /* If the parent list is not empty, ignore */
2678 if (zio.io_parent_list.list_head.list_next !=
2679 wsp->walk_addr +
2680 mdb_ctf_offsetof_by_name(ZFS_STRUCT "zio", "io_parent_list") +
2681 mdb_ctf_offsetof_by_name("struct list", "list_head"))
2682 return (WALK_NEXT);
2683
2684 return (wsp->walk_callback(wsp->walk_addr, &zio, wsp->walk_cbdata));
2685 }
2686
2687 /*
2688 * ::zfs_blkstats
2689 *
2690 * -v print verbose per-level information
2691 *
2692 */
2693 static int
2694 zfs_blkstats(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2695 {
2696 boolean_t verbose = B_FALSE;
2697 zfs_all_blkstats_t stats;
2698 dmu_object_type_t t;
2699 zfs_blkstat_t *tzb;
2700 uint64_t ditto;
2701 dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES + 10];
2702 /* +10 in case it grew */
2703
2704 if (mdb_readvar(&dmu_ot, "dmu_ot") == -1) {
2705 mdb_warn("failed to read 'dmu_ot'");
2706 return (DCMD_ERR);
2707 }
2708
2709 if (mdb_getopts(argc, argv,
2710 'v', MDB_OPT_SETBITS, TRUE, &verbose,
2711 NULL) != argc)
2712 return (DCMD_USAGE);
2713
2714 if (!(flags & DCMD_ADDRSPEC))
2715 return (DCMD_USAGE);
2716
2717 if (GETMEMB(addr, "spa", spa_dsl_pool, addr) ||
2718 GETMEMB(addr, "dsl_pool", dp_blkstats, addr) ||
2719 mdb_vread(&stats, sizeof (zfs_all_blkstats_t), addr) == -1) {
2720 mdb_warn("failed to read data at %p;", addr);
2721 mdb_printf("maybe no stats? run \"zpool scrub\" first.");
2722 return (DCMD_ERR);
2723 }
2724
2725 tzb = &stats.zab_type[DN_MAX_LEVELS][DMU_OT_TOTAL];
2726 if (tzb->zb_gangs != 0) {
2727 mdb_printf("Ganged blocks: %llu\n",
2728 (longlong_t)tzb->zb_gangs);
2729 }
2730
2731 ditto = tzb->zb_ditto_2_of_2_samevdev + tzb->zb_ditto_2_of_3_samevdev +
2732 tzb->zb_ditto_3_of_3_samevdev;
2733 if (ditto != 0) {
2734 mdb_printf("Dittoed blocks on same vdev: %llu\n",
2735 (longlong_t)ditto);
2736 }
2737
2738 mdb_printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
2739 "\t avg\t comp\t%%Total\tType\n");
2740
2741 for (t = 0; t <= DMU_OT_TOTAL; t++) {
2742 char csize[NICENUM_BUFLEN], lsize[NICENUM_BUFLEN];
2743 char psize[NICENUM_BUFLEN], asize[NICENUM_BUFLEN];
2744 char avg[NICENUM_BUFLEN];
2745 char comp[NICENUM_BUFLEN], pct[NICENUM_BUFLEN];
2746 char typename[64];
2747 int l;
2748
2749
2750 if (t == DMU_OT_DEFERRED)
2751 strcpy(typename, "deferred free");
2752 else if (t == DMU_OT_OTHER)
2753 strcpy(typename, "other");
2754 else if (t == DMU_OT_TOTAL)
2755 strcpy(typename, "Total");
2756 else if (mdb_readstr(typename, sizeof (typename),
2757 (uintptr_t)dmu_ot[t].ot_name) == -1) {
2758 mdb_warn("failed to read type name");
2759 return (DCMD_ERR);
2760 }
2761
2762 if (stats.zab_type[DN_MAX_LEVELS][t].zb_asize == 0)
2763 continue;
2764
2765 for (l = -1; l < DN_MAX_LEVELS; l++) {
2766 int level = (l == -1 ? DN_MAX_LEVELS : l);
2767 zfs_blkstat_t *zb = &stats.zab_type[level][t];
2768
2769 if (zb->zb_asize == 0)
2770 continue;
2771
2772 /*
2773 * Don't print each level unless requested.
2774 */
2775 if (!verbose && level != DN_MAX_LEVELS)
2776 continue;
2777
2778 /*
2779 * If all the space is level 0, don't print the
2780 * level 0 separately.
2781 */
2782 if (level == 0 && zb->zb_asize ==
2783 stats.zab_type[DN_MAX_LEVELS][t].zb_asize)
2784 continue;
2785
2786 mdb_nicenum(zb->zb_count, csize);
2787 mdb_nicenum(zb->zb_lsize, lsize);
2788 mdb_nicenum(zb->zb_psize, psize);
2789 mdb_nicenum(zb->zb_asize, asize);
2790 mdb_nicenum(zb->zb_asize / zb->zb_count, avg);
2791 (void) snprintfrac(comp, NICENUM_BUFLEN,
2792 zb->zb_lsize, zb->zb_psize, 2);
2793 (void) snprintfrac(pct, NICENUM_BUFLEN,
2794 100 * zb->zb_asize, tzb->zb_asize, 2);
2795
2796 mdb_printf("%6s\t%5s\t%5s\t%5s\t%5s"
2797 "\t%5s\t%6s\t",
2798 csize, lsize, psize, asize, avg, comp, pct);
2799
2800 if (level == DN_MAX_LEVELS)
2801 mdb_printf("%s\n", typename);
2802 else
2803 mdb_printf(" L%d %s\n",
2804 level, typename);
2805 }
2806 }
2807
2808 return (DCMD_OK);
2809 }
2810
2811 typedef struct mdb_reference {
2812 uintptr_t ref_holder;
2813 uintptr_t ref_removed;
2814 uint64_t ref_number;
2815 } mdb_reference_t;
2816
2817 /* ARGSUSED */
2818 static int
2819 reference_cb(uintptr_t addr, const void *ignored, void *arg)
2820 {
2821 mdb_reference_t ref;
2822 boolean_t holder_is_str = B_FALSE;
2823 char holder_str[128];
2824 boolean_t removed = (boolean_t)arg;
2825
2826 if (mdb_ctf_vread(&ref, "reference_t", "mdb_reference_t", addr,
2827 0) == -1)
2828 return (DCMD_ERR);
2829
2830 if (mdb_readstr(holder_str, sizeof (holder_str),
2831 ref.ref_holder) != -1)
2832 holder_is_str = strisprint(holder_str);
2833
2834 if (removed)
2835 mdb_printf("removed ");
2836 mdb_printf("reference ");
2837 if (ref.ref_number != 1)
2838 mdb_printf("with count=%llu ", ref.ref_number);
2839 mdb_printf("with tag %lx", ref.ref_holder);
2840 if (holder_is_str)
2841 mdb_printf(" \"%s\"", holder_str);
2842 mdb_printf(", held at:\n");
2843
2844 (void) mdb_call_dcmd("whatis", addr, DCMD_ADDRSPEC, 0, NULL);
2845
2846 if (removed) {
2847 mdb_printf("removed at:\n");
2848 (void) mdb_call_dcmd("whatis", ref.ref_removed,
2849 DCMD_ADDRSPEC, 0, NULL);
2850 }
2851
2852 mdb_printf("\n");
2853
2854 return (WALK_NEXT);
2855 }
2856
2857 typedef struct mdb_refcount {
2858 uint64_t rc_count;
2859 } mdb_refcount_t;
2860
2861 typedef struct mdb_refcount_removed {
2862 uint64_t rc_removed_count;
2863 } mdb_refcount_removed_t;
2864
2865 typedef struct mdb_refcount_tracked {
2866 boolean_t rc_tracked;
2867 } mdb_refcount_tracked_t;
2868
2869 /* ARGSUSED */
2870 static int
2871 refcount(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2872 {
2873 mdb_refcount_t rc;
2874 mdb_refcount_removed_t rcr;
2875 mdb_refcount_tracked_t rct;
2876 int off;
2877 boolean_t released = B_FALSE;
2878
2879 if (!(flags & DCMD_ADDRSPEC))
2880 return (DCMD_USAGE);
2881
2882 if (mdb_getopts(argc, argv,
2883 'r', MDB_OPT_SETBITS, B_TRUE, &released,
2884 NULL) != argc)
2885 return (DCMD_USAGE);
2886
2887 if (mdb_ctf_vread(&rc, "refcount_t", "mdb_refcount_t", addr,
2888 0) == -1)
2889 return (DCMD_ERR);
2890
2891 if (mdb_ctf_vread(&rcr, "refcount_t", "mdb_refcount_removed_t", addr,
2892 MDB_CTF_VREAD_QUIET) == -1) {
2893 mdb_printf("refcount_t at %p has %llu holds (untracked)\n",
2894 addr, (longlong_t)rc.rc_count);
2895 return (DCMD_OK);
2896 }
2897
2898 if (mdb_ctf_vread(&rct, "refcount_t", "mdb_refcount_tracked_t", addr,
2899 MDB_CTF_VREAD_QUIET) == -1) {
2900 /* If this is an old target, it might be tracked. */
2901 rct.rc_tracked = B_TRUE;
2902 }
2903
2904 mdb_printf("refcount_t at %p has %llu current holds, "
2905 "%llu recently released holds\n",
2906 addr, (longlong_t)rc.rc_count, (longlong_t)rcr.rc_removed_count);
2907
2908 if (rct.rc_tracked && rc.rc_count > 0)
2909 mdb_printf("current holds:\n");
2910 off = mdb_ctf_offsetof_by_name("refcount_t", "rc_list");
2911 if (off == -1)
2912 return (DCMD_ERR);
2913 mdb_pwalk("list", reference_cb, (void*)B_FALSE, addr + off);
2914
2915 if (released && rcr.rc_removed_count > 0) {
2916 mdb_printf("released holds:\n");
2917
2918 off = mdb_ctf_offsetof_by_name("refcount_t", "rc_removed");
2919 if (off == -1)
2920 return (DCMD_ERR);
2921 mdb_pwalk("list", reference_cb, (void*)B_TRUE, addr + off);
2922 }
2923
2924 return (DCMD_OK);
2925 }
2926
2927 /* ARGSUSED */
2928 static int
2929 sa_attr_table(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2930 {
2931 sa_attr_table_t *table;
2932 sa_os_t sa_os;
2933 char *name;
2934 int i;
2935
2936 if (mdb_vread(&sa_os, sizeof (sa_os_t), addr) == -1) {
2937 mdb_warn("failed to read sa_os at %p", addr);
2938 return (DCMD_ERR);
2939 }
2940
2941 table = mdb_alloc(sizeof (sa_attr_table_t) * sa_os.sa_num_attrs,
2942 UM_SLEEP | UM_GC);
2943 name = mdb_alloc(MAXPATHLEN, UM_SLEEP | UM_GC);
2944
2945 if (mdb_vread(table, sizeof (sa_attr_table_t) * sa_os.sa_num_attrs,
2946 (uintptr_t)sa_os.sa_attr_table) == -1) {
2947 mdb_warn("failed to read sa_os at %p", addr);
2948 return (DCMD_ERR);
2949 }
2950
2951 mdb_printf("%<u>%-10s %-10s %-10s %-10s %s%</u>\n",
2952 "ATTR ID", "REGISTERED", "LENGTH", "BSWAP", "NAME");
2953 for (i = 0; i != sa_os.sa_num_attrs; i++) {
2954 mdb_readstr(name, MAXPATHLEN, (uintptr_t)table[i].sa_name);
2955 mdb_printf("%5x %8x %8x %8x %-s\n",
2956 (int)table[i].sa_attr, (int)table[i].sa_registered,
2957 (int)table[i].sa_length, table[i].sa_byteswap, name);
2958 }
2959
2960 return (DCMD_OK);
2961 }
2962
2963 static int
2964 sa_get_off_table(uintptr_t addr, uint32_t **off_tab, int attr_count)
2965 {
2966 uintptr_t idx_table;
2967
2968 if (GETMEMB(addr, "sa_idx_tab", sa_idx_tab, idx_table)) {
2969 mdb_printf("can't find offset table in sa_idx_tab\n");
2970 return (-1);
2971 }
2972
2973 *off_tab = mdb_alloc(attr_count * sizeof (uint32_t),
2974 UM_SLEEP | UM_GC);
2975
2976 if (mdb_vread(*off_tab,
2977 attr_count * sizeof (uint32_t), idx_table) == -1) {
2978 mdb_warn("failed to attribute offset table %p", idx_table);
2979 return (-1);
2980 }
2981
2982 return (DCMD_OK);
2983 }
2984
2985 /*ARGSUSED*/
2986 static int
2987 sa_attr_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2988 {
2989 uint32_t *offset_tab;
2990 int attr_count;
2991 uint64_t attr_id;
2992 uintptr_t attr_addr;
2993 uintptr_t bonus_tab, spill_tab;
2994 uintptr_t db_bonus, db_spill;
2995 uintptr_t os, os_sa;
2996 uintptr_t db_data;
2997
2998 if (argc != 1)
2999 return (DCMD_USAGE);
3000
3001 if (argv[0].a_type == MDB_TYPE_STRING)
3002 attr_id = mdb_strtoull(argv[0].a_un.a_str);
3003 else
3004 return (DCMD_USAGE);
3005
3006 if (GETMEMB(addr, "sa_handle", sa_bonus_tab, bonus_tab) ||
3007 GETMEMB(addr, "sa_handle", sa_spill_tab, spill_tab) ||
3008 GETMEMB(addr, "sa_handle", sa_os, os) ||
3009 GETMEMB(addr, "sa_handle", sa_bonus, db_bonus) ||
3010 GETMEMB(addr, "sa_handle", sa_spill, db_spill)) {
3011 mdb_printf("Can't find necessary information in sa_handle "
3012 "in sa_handle\n");
3013 return (DCMD_ERR);
3014 }
3015
3016 if (GETMEMB(os, "objset", os_sa, os_sa)) {
3017 mdb_printf("Can't find os_sa in objset\n");
3018 return (DCMD_ERR);
3019 }
3020
3021 if (GETMEMB(os_sa, "sa_os", sa_num_attrs, attr_count)) {
3022 mdb_printf("Can't find sa_num_attrs\n");
3023 return (DCMD_ERR);
3024 }
3025
3026 if (attr_id > attr_count) {
3027 mdb_printf("attribute id number is out of range\n");
3028 return (DCMD_ERR);
3029 }
3030
3031 if (bonus_tab) {
3032 if (sa_get_off_table(bonus_tab, &offset_tab,
3033 attr_count) == -1) {
3034 return (DCMD_ERR);
3035 }
3036
3037 if (GETMEMB(db_bonus, "dmu_buf", db_data, db_data)) {
3038 mdb_printf("can't find db_data in bonus dbuf\n");
3039 return (DCMD_ERR);
3040 }
3041 }
3042
3043 if (bonus_tab && !TOC_ATTR_PRESENT(offset_tab[attr_id]) &&
3044 spill_tab == NULL) {
3045 mdb_printf("Attribute does not exist\n");
3046 return (DCMD_ERR);
3047 } else if (!TOC_ATTR_PRESENT(offset_tab[attr_id]) && spill_tab) {
3048 if (sa_get_off_table(spill_tab, &offset_tab,
3049 attr_count) == -1) {
3050 return (DCMD_ERR);
3051 }
3052 if (GETMEMB(db_spill, "dmu_buf", db_data, db_data)) {
3053 mdb_printf("can't find db_data in spill dbuf\n");
3054 return (DCMD_ERR);
3055 }
3056 if (!TOC_ATTR_PRESENT(offset_tab[attr_id])) {
3057 mdb_printf("Attribute does not exist\n");
3058 return (DCMD_ERR);
3059 }
3060 }
3061 attr_addr = db_data + TOC_OFF(offset_tab[attr_id]);
3062 mdb_printf("%p\n", attr_addr);
3063 return (DCMD_OK);
3064 }
3065
3066 /* ARGSUSED */
3067 static int
3068 zfs_ace_print_common(uintptr_t addr, uint_t flags,
3069 uint64_t id, uint32_t access_mask, uint16_t ace_flags,
3070 uint16_t ace_type, int verbose)
3071 {
3072 if (DCMD_HDRSPEC(flags) && !verbose)
3073 mdb_printf("%<u>%-?s %-8s %-8s %-8s %s%</u>\n",
3074 "ADDR", "FLAGS", "MASK", "TYPE", "ID");
3075
3076 if (!verbose) {
3077 mdb_printf("%0?p %-8x %-8x %-8x %-llx\n", addr,
3078 ace_flags, access_mask, ace_type, id);
3079 return (DCMD_OK);
3080 }
3081
3082 switch (ace_flags & ACE_TYPE_FLAGS) {
3083 case ACE_OWNER:
3084 mdb_printf("owner@:");
3085 break;
3086 case (ACE_IDENTIFIER_GROUP | ACE_GROUP):
3087 mdb_printf("group@:");
3088 break;
3089 case ACE_EVERYONE:
3090 mdb_printf("everyone@:");
3091 break;
3092 case ACE_IDENTIFIER_GROUP:
3093 mdb_printf("group:%llx:", (u_longlong_t)id);
3094 break;
3095 case 0: /* User entry */
3096 mdb_printf("user:%llx:", (u_longlong_t)id);
3097 break;
3098 }
3099
3100 /* print out permission mask */
3101 if (access_mask & ACE_READ_DATA)
3102 mdb_printf("r");
3103 else
3104 mdb_printf("-");
3105 if (access_mask & ACE_WRITE_DATA)
3106 mdb_printf("w");
3107 else
3108 mdb_printf("-");
3109 if (access_mask & ACE_EXECUTE)
3110 mdb_printf("x");
3111 else
3112 mdb_printf("-");
3113 if (access_mask & ACE_APPEND_DATA)
3114 mdb_printf("p");
3115 else
3116 mdb_printf("-");
3117 if (access_mask & ACE_DELETE)
3118 mdb_printf("d");
3119 else
3120 mdb_printf("-");
3121 if (access_mask & ACE_DELETE_CHILD)
3122 mdb_printf("D");
3123 else
3124 mdb_printf("-");
3125 if (access_mask & ACE_READ_ATTRIBUTES)
3126 mdb_printf("a");
3127 else
3128 mdb_printf("-");
3129 if (access_mask & ACE_WRITE_ATTRIBUTES)
3130 mdb_printf("A");
3131 else
3132 mdb_printf("-");
3133 if (access_mask & ACE_READ_NAMED_ATTRS)
3134 mdb_printf("R");
3135 else
3136 mdb_printf("-");
3137 if (access_mask & ACE_WRITE_NAMED_ATTRS)
3138 mdb_printf("W");
3139 else
3140 mdb_printf("-");
3141 if (access_mask & ACE_READ_ACL)
3142 mdb_printf("c");
3143 else
3144 mdb_printf("-");
3145 if (access_mask & ACE_WRITE_ACL)
3146 mdb_printf("C");
3147 else
3148 mdb_printf("-");
3149 if (access_mask & ACE_WRITE_OWNER)
3150 mdb_printf("o");
3151 else
3152 mdb_printf("-");
3153 if (access_mask & ACE_SYNCHRONIZE)
3154 mdb_printf("s");
3155 else
3156 mdb_printf("-");
3157
3158 mdb_printf(":");
3159
3160 /* Print out inheritance flags */
3161 if (ace_flags & ACE_FILE_INHERIT_ACE)
3162 mdb_printf("f");
3163 else
3164 mdb_printf("-");
3165 if (ace_flags & ACE_DIRECTORY_INHERIT_ACE)
3166 mdb_printf("d");
3167 else
3168 mdb_printf("-");
3169 if (ace_flags & ACE_INHERIT_ONLY_ACE)
3170 mdb_printf("i");
3171 else
3172 mdb_printf("-");
3173 if (ace_flags & ACE_NO_PROPAGATE_INHERIT_ACE)
3174 mdb_printf("n");
3175 else
3176 mdb_printf("-");
3177 if (ace_flags & ACE_SUCCESSFUL_ACCESS_ACE_FLAG)
3178 mdb_printf("S");
3179 else
3180 mdb_printf("-");
3181 if (ace_flags & ACE_FAILED_ACCESS_ACE_FLAG)
3182 mdb_printf("F");
3183 else
3184 mdb_printf("-");
3185 if (ace_flags & ACE_INHERITED_ACE)
3186 mdb_printf("I");
3187 else
3188 mdb_printf("-");
3189
3190 switch (ace_type) {
3191 case ACE_ACCESS_ALLOWED_ACE_TYPE:
3192 mdb_printf(":allow\n");
3193 break;
3194 case ACE_ACCESS_DENIED_ACE_TYPE:
3195 mdb_printf(":deny\n");
3196 break;
3197 case ACE_SYSTEM_AUDIT_ACE_TYPE:
3198 mdb_printf(":audit\n");
3199 break;
3200 case ACE_SYSTEM_ALARM_ACE_TYPE:
3201 mdb_printf(":alarm\n");
3202 break;
3203 default:
3204 mdb_printf(":?\n");
3205 }
3206 return (DCMD_OK);
3207 }
3208
3209 /* ARGSUSED */
3210 static int
3211 zfs_ace_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3212 {
3213 zfs_ace_t zace;
3214 int verbose = FALSE;
3215 uint64_t id;
3216
3217 if (!(flags & DCMD_ADDRSPEC))
3218 return (DCMD_USAGE);
3219
3220 if (mdb_getopts(argc, argv,
3221 'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
3222 return (DCMD_USAGE);
3223
3224 if (mdb_vread(&zace, sizeof (zfs_ace_t), addr) == -1) {
3225 mdb_warn("failed to read zfs_ace_t");
3226 return (DCMD_ERR);
3227 }
3228
3229 if ((zace.z_hdr.z_flags & ACE_TYPE_FLAGS) == 0 ||
3230 (zace.z_hdr.z_flags & ACE_TYPE_FLAGS) == ACE_IDENTIFIER_GROUP)
3231 id = zace.z_fuid;
3232 else
3233 id = -1;
3234
3235 return (zfs_ace_print_common(addr, flags, id, zace.z_hdr.z_access_mask,
3236 zace.z_hdr.z_flags, zace.z_hdr.z_type, verbose));
3237 }
3238
3239 /* ARGSUSED */
3240 static int
3241 zfs_ace0_print(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3242 {
3243 ace_t ace;
3244 uint64_t id;
3245 int verbose = FALSE;
3246
3247 if (!(flags & DCMD_ADDRSPEC))
3248 return (DCMD_USAGE);
3249
3250 if (mdb_getopts(argc, argv,
3251 'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
3252 return (DCMD_USAGE);
3253
3254 if (mdb_vread(&ace, sizeof (ace_t), addr) == -1) {
3255 mdb_warn("failed to read ace_t");
3256 return (DCMD_ERR);
3257 }
3258
3259 if ((ace.a_flags & ACE_TYPE_FLAGS) == 0 ||
3260 (ace.a_flags & ACE_TYPE_FLAGS) == ACE_IDENTIFIER_GROUP)
3261 id = ace.a_who;
3262 else
3263 id = -1;
3264
3265 return (zfs_ace_print_common(addr, flags, id, ace.a_access_mask,
3266 ace.a_flags, ace.a_type, verbose));
3267 }
3268
3269 typedef struct acl_dump_args {
3270 int a_argc;
3271 const mdb_arg_t *a_argv;
3272 uint16_t a_version;
3273 int a_flags;
3274 } acl_dump_args_t;
3275
3276 /* ARGSUSED */
3277 static int
3278 acl_aces_cb(uintptr_t addr, const void *unknown, void *arg)
3279 {
3280 acl_dump_args_t *acl_args = (acl_dump_args_t *)arg;
3281
3282 if (acl_args->a_version == 1) {
3283 if (mdb_call_dcmd("zfs_ace", addr,
3284 DCMD_ADDRSPEC|acl_args->a_flags, acl_args->a_argc,
3285 acl_args->a_argv) != DCMD_OK) {
3286 return (WALK_ERR);
3287 }
3288 } else {
3289 if (mdb_call_dcmd("zfs_ace0", addr,
3290 DCMD_ADDRSPEC|acl_args->a_flags, acl_args->a_argc,
3291 acl_args->a_argv) != DCMD_OK) {
3292 return (WALK_ERR);
3293 }
3294 }
3295 acl_args->a_flags = DCMD_LOOP;
3296 return (WALK_NEXT);
3297 }
3298
3299 /* ARGSUSED */
3300 static int
3301 acl_cb(uintptr_t addr, const void *unknown, void *arg)
3302 {
3303 acl_dump_args_t *acl_args = (acl_dump_args_t *)arg;
3304
3305 if (acl_args->a_version == 1) {
3306 if (mdb_pwalk("zfs_acl_node_aces", acl_aces_cb,
3307 arg, addr) != 0) {
3308 mdb_warn("can't walk ACEs");
3309 return (DCMD_ERR);
3310 }
3311 } else {
3312 if (mdb_pwalk("zfs_acl_node_aces0", acl_aces_cb,
3313 arg, addr) != 0) {
3314 mdb_warn("can't walk ACEs");
3315 return (DCMD_ERR);
3316 }
3317 }
3318 return (WALK_NEXT);
3319 }
3320
3321 /* ARGSUSED */
3322 static int
3323 zfs_acl_dump(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3324 {
3325 zfs_acl_t zacl;
3326 int verbose = FALSE;
3327 acl_dump_args_t acl_args;
3328
3329 if (!(flags & DCMD_ADDRSPEC))
3330 return (DCMD_USAGE);
3331
3332 if (mdb_getopts(argc, argv,
3333 'v', MDB_OPT_SETBITS, TRUE, &verbose, TRUE, NULL) != argc)
3334 return (DCMD_USAGE);
3335
3336 if (mdb_vread(&zacl, sizeof (zfs_acl_t), addr) == -1) {
3337 mdb_warn("failed to read zfs_acl_t");
3338 return (DCMD_ERR);
3339 }
3340
3341 acl_args.a_argc = argc;
3342 acl_args.a_argv = argv;
3343 acl_args.a_version = zacl.z_version;
3344 acl_args.a_flags = DCMD_LOOPFIRST;
3345
3346 if (mdb_pwalk("zfs_acl_node", acl_cb, &acl_args, addr) != 0) {
3347 mdb_warn("can't walk ACL");
3348 return (DCMD_ERR);
3349 }
3350
3351 return (DCMD_OK);
3352 }
3353
3354 /* ARGSUSED */
3355 static int
3356 zfs_acl_node_walk_init(mdb_walk_state_t *wsp)
3357 {
3358 if (wsp->walk_addr == NULL) {
3359 mdb_warn("must supply address of zfs_acl_node_t\n");
3360 return (WALK_ERR);
3361 }
3362
3363 wsp->walk_addr +=
3364 mdb_ctf_offsetof_by_name(ZFS_STRUCT "zfs_acl", "z_acl");
3365
3366 if (mdb_layered_walk("list", wsp) == -1) {
3367 mdb_warn("failed to walk 'list'\n");
3368 return (WALK_ERR);
3369 }
3370
3371 return (WALK_NEXT);
3372 }
3373
3374 static int
3375 zfs_acl_node_walk_step(mdb_walk_state_t *wsp)
3376 {
3377 zfs_acl_node_t aclnode;
3378
3379 if (mdb_vread(&aclnode, sizeof (zfs_acl_node_t),
3380 wsp->walk_addr) == -1) {
3381 mdb_warn("failed to read zfs_acl_node at %p", wsp->walk_addr);
3382 return (WALK_ERR);
3383 }
3384
3385 return (wsp->walk_callback(wsp->walk_addr, &aclnode, wsp->walk_cbdata));
3386 }
3387
3388 typedef struct ace_walk_data {
3389 int ace_count;
3390 int ace_version;
3391 } ace_walk_data_t;
3392
3393 static int
3394 zfs_aces_walk_init_common(mdb_walk_state_t *wsp, int version,
3395 int ace_count, uintptr_t ace_data)
3396 {
3397 ace_walk_data_t *ace_walk_data;
3398
3399 if (wsp->walk_addr == NULL) {
3400 mdb_warn("must supply address of zfs_acl_node_t\n");
3401 return (WALK_ERR);
3402 }
3403
3404 ace_walk_data = mdb_alloc(sizeof (ace_walk_data_t), UM_SLEEP | UM_GC);
3405
3406 ace_walk_data->ace_count = ace_count;
3407 ace_walk_data->ace_version = version;
3408
3409 wsp->walk_addr = ace_data;
3410 wsp->walk_data = ace_walk_data;
3411
3412 return (WALK_NEXT);
3413 }
3414
3415 static int
3416 zfs_acl_node_aces_walk_init_common(mdb_walk_state_t *wsp, int version)
3417 {
3418 static int gotid;
3419 static mdb_ctf_id_t acl_id;
3420 int z_ace_count;
3421 uintptr_t z_acldata;
3422
3423 if (!gotid) {
3424 if (mdb_ctf_lookup_by_name("struct zfs_acl_node",
3425 &acl_id) == -1) {
3426 mdb_warn("couldn't find struct zfs_acl_node");
3427 return (DCMD_ERR);
3428 }
3429 gotid = TRUE;
3430 }
3431
3432 if (GETMEMBID(wsp->walk_addr, &acl_id, z_ace_count, z_ace_count)) {
3433 return (DCMD_ERR);
3434 }
3435 if (GETMEMBID(wsp->walk_addr, &acl_id, z_acldata, z_acldata)) {
3436 return (DCMD_ERR);
3437 }
3438
3439 return (zfs_aces_walk_init_common(wsp, version,
3440 z_ace_count, z_acldata));
3441 }
3442
3443 /* ARGSUSED */
3444 static int
3445 zfs_acl_node_aces_walk_init(mdb_walk_state_t *wsp)
3446 {
3447 return (zfs_acl_node_aces_walk_init_common(wsp, 1));
3448 }
3449
3450 /* ARGSUSED */
3451 static int
3452 zfs_acl_node_aces0_walk_init(mdb_walk_state_t *wsp)
3453 {
3454 return (zfs_acl_node_aces_walk_init_common(wsp, 0));
3455 }
3456
3457 static int
3458 zfs_aces_walk_step(mdb_walk_state_t *wsp)
3459 {
3460 ace_walk_data_t *ace_data = wsp->walk_data;
3461 zfs_ace_t zace;
3462 ace_t *acep;
3463 int status;
3464 int entry_type;
3465 int allow_type;
3466 uintptr_t ptr;
3467
3468 if (ace_data->ace_count == 0)
3469 return (WALK_DONE);
3470
3471 if (mdb_vread(&zace, sizeof (zfs_ace_t), wsp->walk_addr) == -1) {
3472 mdb_warn("failed to read zfs_ace_t at %#lx",
3473 wsp->walk_addr);
3474 return (WALK_ERR);
3475 }
3476
3477 switch (ace_data->ace_version) {
3478 case 0:
3479 acep = (ace_t *)&zace;
3480 entry_type = acep->a_flags & ACE_TYPE_FLAGS;
3481 allow_type = acep->a_type;
3482 break;
3483 case 1:
3484 entry_type = zace.z_hdr.z_flags & ACE_TYPE_FLAGS;
3485 allow_type = zace.z_hdr.z_type;
3486 break;
3487 default:
3488 return (WALK_ERR);
3489 }
3490
3491 ptr = (uintptr_t)wsp->walk_addr;
3492 switch (entry_type) {
3493 case ACE_OWNER:
3494 case ACE_EVERYONE:
3495 case (ACE_IDENTIFIER_GROUP | ACE_GROUP):
3496 ptr += ace_data->ace_version == 0 ?
3497 sizeof (ace_t) : sizeof (zfs_ace_hdr_t);
3498 break;
3499 case ACE_IDENTIFIER_GROUP:
3500 default:
3501 switch (allow_type) {
3502 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
3503 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
3504 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
3505 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
3506 ptr += ace_data->ace_version == 0 ?
3507 sizeof (ace_t) : sizeof (zfs_object_ace_t);
3508 break;
3509 default:
3510 ptr += ace_data->ace_version == 0 ?
3511 sizeof (ace_t) : sizeof (zfs_ace_t);
3512 break;
3513 }
3514 }
3515
3516 ace_data->ace_count--;
3517 status = wsp->walk_callback(wsp->walk_addr,
3518 (void *)(uintptr_t)&zace, wsp->walk_cbdata);
3519
3520 wsp->walk_addr = ptr;
3521 return (status);
3522 }
3523
3524 typedef struct mdb_zfs_rrwlock {
3525 uintptr_t rr_writer;
3526 boolean_t rr_writer_wanted;
3527 } mdb_zfs_rrwlock_t;
3528
3529 static uint_t rrw_key;
3530
3531 /* ARGSUSED */
3532 static int
3533 rrwlock(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
3534 {
3535 mdb_zfs_rrwlock_t rrw;
3536
3537 if (rrw_key == 0) {
3538 if (mdb_ctf_readsym(&rrw_key, "uint_t", "rrw_tsd_key", 0) == -1)
3539 return (DCMD_ERR);
3540 }
3541
3542 if (mdb_ctf_vread(&rrw, "rrwlock_t", "mdb_zfs_rrwlock_t", addr,
3543 0) == -1)
3544 return (DCMD_ERR);
3545
3546 if (rrw.rr_writer != 0) {
3547 mdb_printf("write lock held by thread %lx\n", rrw.rr_writer);
3548 return (DCMD_OK);
3549 }
3550
3551 if (rrw.rr_writer_wanted) {
3552 mdb_printf("writer wanted\n");
3553 }
3554
3555 mdb_printf("anonymous references:\n");
3556 (void) mdb_call_dcmd("refcount", addr +
3557 mdb_ctf_offsetof_by_name(ZFS_STRUCT "rrwlock", "rr_anon_rcount"),
3558 DCMD_ADDRSPEC, 0, NULL);
3559
3560 mdb_printf("linked references:\n");
3561 (void) mdb_call_dcmd("refcount", addr +
3562 mdb_ctf_offsetof_by_name(ZFS_STRUCT "rrwlock", "rr_linked_rcount"),
3563 DCMD_ADDRSPEC, 0, NULL);
3564
3565 /*
3566 * XXX This should find references from
3567 * "::walk thread | ::tsd -v <rrw_key>", but there is no support
3568 * for programmatic consumption of dcmds, so this would be
3569 * difficult, potentially requiring reimplementing ::tsd (both
3570 * user and kernel versions) in this MDB module.
3571 */
3572
3573 return (DCMD_OK);
3574 }
3575
3576 typedef struct mdb_arc_buf_hdr_t {
3577 uint16_t b_psize;
3578 uint16_t b_lsize;
3579 struct {
3580 uint32_t b_bufcnt;
3581 uintptr_t b_state;
3582 } b_l1hdr;
3583 } mdb_arc_buf_hdr_t;
3584
3585 enum arc_cflags {
3586 ARC_CFLAG_VERBOSE = 1 << 0,
3587 ARC_CFLAG_ANON = 1 << 1,
3588 ARC_CFLAG_MRU = 1 << 2,
3589 ARC_CFLAG_MFU = 1 << 3,
3590 ARC_CFLAG_BUFS = 1 << 4,
3591 };
3592
3593 typedef struct arc_compression_stats_data {
3594 GElf_Sym anon_sym; /* ARC_anon symbol */
3595 GElf_Sym mru_sym; /* ARC_mru symbol */
3596 GElf_Sym mrug_sym; /* ARC_mru_ghost symbol */
3597 GElf_Sym mfu_sym; /* ARC_mfu symbol */
3598 GElf_Sym mfug_sym; /* ARC_mfu_ghost symbol */
3599 GElf_Sym l2c_sym; /* ARC_l2c_only symbol */
3600 uint64_t *anon_c_hist; /* histogram of compressed sizes in anon */
3601 uint64_t *anon_u_hist; /* histogram of uncompressed sizes in anon */
3602 uint64_t *anon_bufs; /* histogram of buffer counts in anon state */
3603 uint64_t *mru_c_hist; /* histogram of compressed sizes in mru */
3604 uint64_t *mru_u_hist; /* histogram of uncompressed sizes in mru */
3605 uint64_t *mru_bufs; /* histogram of buffer counts in mru */
3606 uint64_t *mfu_c_hist; /* histogram of compressed sizes in mfu */
3607 uint64_t *mfu_u_hist; /* histogram of uncompressed sizes in mfu */
3608 uint64_t *mfu_bufs; /* histogram of buffer counts in mfu */
3609 uint64_t *all_c_hist; /* histogram of compressed anon + mru + mfu */
3610 uint64_t *all_u_hist; /* histogram of uncompressed anon + mru + mfu */
3611 uint64_t *all_bufs; /* histogram of buffer counts in all states */
3612 int arc_cflags; /* arc compression flags, specified by user */
3613 int hist_nbuckets; /* number of buckets in each histogram */
3614 } arc_compression_stats_data_t;
3615
3616 int
3617 highbit64(uint64_t i)
3618 {
3619 int h = 1;
3620
3621 if (i == 0)
3622 return (0);
3623 if (i & 0xffffffff00000000ULL) {
3624 h += 32; i >>= 32;
3625 }
3626 if (i & 0xffff0000) {
3627 h += 16; i >>= 16;
3628 }
3629 if (i & 0xff00) {
3630 h += 8; i >>= 8;
3631 }
3632 if (i & 0xf0) {
3633 h += 4; i >>= 4;
3634 }
3635 if (i & 0xc) {
3636 h += 2; i >>= 2;
3637 }
3638 if (i & 0x2) {
3639 h += 1;
3640 }
3641 return (h);
3642 }
3643
3644 /* ARGSUSED */
3645 static int
3646 arc_compression_stats_cb(uintptr_t addr, const void *unknown, void *arg)
3647 {
3648 arc_compression_stats_data_t *data = arg;
3649 mdb_arc_buf_hdr_t hdr;
3650 int cbucket, ubucket, bufcnt;
3651
3652 if (mdb_ctf_vread(&hdr, "arc_buf_hdr_t", "mdb_arc_buf_hdr_t",
3653 addr, 0) == -1) {
3654 return (WALK_ERR);
3655 }
3656
3657 /*
3658 * Headers in the ghost states, or the l2c_only state don't have
3659 * arc buffers linked off of them. Thus, their compressed size
3660 * is meaningless, so we skip these from the stats.
3661 */
3662 if (hdr.b_l1hdr.b_state == data->mrug_sym.st_value ||
3663 hdr.b_l1hdr.b_state == data->mfug_sym.st_value ||
3664 hdr.b_l1hdr.b_state == data->l2c_sym.st_value) {
3665 return (WALK_NEXT);
3666 }
3667
3668 /*
3669 * The physical size (compressed) and logical size
3670 * (uncompressed) are in units of SPA_MINBLOCKSIZE. By default,
3671 * we use the log2 of this value (rounded down to the nearest
3672 * integer) to determine the bucket to assign this header to.
3673 * Thus, the histogram is logarithmic with respect to the size
3674 * of the header. For example, the following is a mapping of the
3675 * bucket numbers and the range of header sizes they correspond to:
3676 *
3677 * 0: 0 byte headers
3678 * 1: 512 byte headers
3679 * 2: [1024 - 2048) byte headers
3680 * 3: [2048 - 4096) byte headers
3681 * 4: [4096 - 8192) byte headers
3682 * 5: [8192 - 16394) byte headers
3683 * 6: [16384 - 32768) byte headers
3684 * 7: [32768 - 65536) byte headers
3685 * 8: [65536 - 131072) byte headers
3686 * 9: 131072 byte headers
3687 *
3688 * If the ARC_CFLAG_VERBOSE flag was specified, we use the
3689 * physical and logical sizes directly. Thus, the histogram will
3690 * no longer be logarithmic; instead it will be linear with
3691 * respect to the size of the header. The following is a mapping
3692 * of the first many bucket numbers and the header size they
3693 * correspond to:
3694 *
3695 * 0: 0 byte headers
3696 * 1: 512 byte headers
3697 * 2: 1024 byte headers
3698 * 3: 1536 byte headers
3699 * 4: 2048 byte headers
3700 * 5: 2560 byte headers
3701 * 6: 3072 byte headers
3702 *
3703 * And so on. Keep in mind that a range of sizes isn't used in
3704 * the case of linear scale because the headers can only
3705 * increment or decrement in sizes of 512 bytes. So, it's not
3706 * possible for a header to be sized in between whats listed
3707 * above.
3708 *
3709 * Also, the above mapping values were calculated assuming a
3710 * SPA_MINBLOCKSHIFT of 512 bytes and a SPA_MAXBLOCKSIZE of 128K.
3711 */
3712
3713 if (data->arc_cflags & ARC_CFLAG_VERBOSE) {
3714 cbucket = hdr.b_psize;
3715 ubucket = hdr.b_lsize;
3716 } else {
3717 cbucket = highbit64(hdr.b_psize);
3718 ubucket = highbit64(hdr.b_lsize);
3719 }
3720
3721 bufcnt = hdr.b_l1hdr.b_bufcnt;
3722 if (bufcnt >= data->hist_nbuckets)
3723 bufcnt = data->hist_nbuckets - 1;
3724
3725 /* Ensure we stay within the bounds of the histogram array */
3726 ASSERT3U(cbucket, <, data->hist_nbuckets);
3727 ASSERT3U(ubucket, <, data->hist_nbuckets);
3728
3729 if (hdr.b_l1hdr.b_state == data->anon_sym.st_value) {
3730 data->anon_c_hist[cbucket]++;
3731 data->anon_u_hist[ubucket]++;
3732 data->anon_bufs[bufcnt]++;
3733 } else if (hdr.b_l1hdr.b_state == data->mru_sym.st_value) {
3734 data->mru_c_hist[cbucket]++;
3735 data->mru_u_hist[ubucket]++;
3736 data->mru_bufs[bufcnt]++;
3737 } else if (hdr.b_l1hdr.b_state == data->mfu_sym.st_value) {
3738 data->mfu_c_hist[cbucket]++;
3739 data->mfu_u_hist[ubucket]++;
3740 data->mfu_bufs[bufcnt]++;
3741 }
3742
3743 data->all_c_hist[cbucket]++;
3744 data->all_u_hist[ubucket]++;
3745 data->all_bufs[bufcnt]++;
3746
3747 return (WALK_NEXT);
3748 }
3749
3750 /* ARGSUSED */
3751 static int
3752 arc_compression_stats(uintptr_t addr, uint_t flags, int argc,
3753 const mdb_arg_t *argv)
3754 {
3755 arc_compression_stats_data_t data = { 0 };
3756 unsigned int max_shifted = SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT;
3757 unsigned int hist_size;
3758 char range[32];
3759 int rc = DCMD_OK;
3760
3761 if (mdb_getopts(argc, argv,
3762 'v', MDB_OPT_SETBITS, ARC_CFLAG_VERBOSE, &data.arc_cflags,
3763 'a', MDB_OPT_SETBITS, ARC_CFLAG_ANON, &data.arc_cflags,
3764 'b', MDB_OPT_SETBITS, ARC_CFLAG_BUFS, &data.arc_cflags,
3765 'r', MDB_OPT_SETBITS, ARC_CFLAG_MRU, &data.arc_cflags,
3766 'f', MDB_OPT_SETBITS, ARC_CFLAG_MFU, &data.arc_cflags) != argc)
3767 return (DCMD_USAGE);
3768
3769 if (mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_anon", &data.anon_sym) ||
3770 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mru", &data.mru_sym) ||
3771 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mru_ghost", &data.mrug_sym) ||
3772 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mfu", &data.mfu_sym) ||
3773 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_mfu_ghost", &data.mfug_sym) ||
3774 mdb_lookup_by_obj(ZFS_OBJ_NAME, "ARC_l2c_only", &data.l2c_sym)) {
3775 mdb_warn("can't find arc state symbol");
3776 return (DCMD_ERR);
3777 }
3778
3779 /*
3780 * Determine the maximum expected size for any header, and use
3781 * this to determine the number of buckets needed for each
3782 * histogram. If ARC_CFLAG_VERBOSE is specified, this value is
3783 * used directly; otherwise the log2 of the maximum size is
3784 * used. Thus, if using a log2 scale there's a maximum of 10
3785 * possible buckets, while the linear scale (when using
3786 * ARC_CFLAG_VERBOSE) has a maximum of 257 buckets.
3787 */
3788 if (data.arc_cflags & ARC_CFLAG_VERBOSE)
3789 data.hist_nbuckets = max_shifted + 1;
3790 else
3791 data.hist_nbuckets = highbit64(max_shifted) + 1;
3792
3793 hist_size = sizeof (uint64_t) * data.hist_nbuckets;
3794
3795 data.anon_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3796 data.anon_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3797 data.anon_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3798
3799 data.mru_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3800 data.mru_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3801 data.mru_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3802
3803 data.mfu_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3804 data.mfu_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3805 data.mfu_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3806
3807 data.all_c_hist = mdb_zalloc(hist_size, UM_SLEEP);
3808 data.all_u_hist = mdb_zalloc(hist_size, UM_SLEEP);
3809 data.all_bufs = mdb_zalloc(hist_size, UM_SLEEP);
3810
3811 if (mdb_walk("arc_buf_hdr_t_full", arc_compression_stats_cb,
3812 &data) != 0) {
3813 mdb_warn("can't walk arc_buf_hdr's");
3814 rc = DCMD_ERR;
3815 goto out;
3816 }
3817
3818 if (data.arc_cflags & ARC_CFLAG_VERBOSE) {
3819 rc = mdb_snprintf(range, sizeof (range),
3820 "[n*%llu, (n+1)*%llu)", SPA_MINBLOCKSIZE,
3821 SPA_MINBLOCKSIZE);
3822 } else {
3823 rc = mdb_snprintf(range, sizeof (range),
3824 "[2^(n-1)*%llu, 2^n*%llu)", SPA_MINBLOCKSIZE,
3825 SPA_MINBLOCKSIZE);
3826 }
3827
3828 if (rc < 0) {
3829 /* snprintf failed, abort the dcmd */
3830 rc = DCMD_ERR;
3831 goto out;
3832 } else {
3833 /* snprintf succeeded above, reset return code */
3834 rc = DCMD_OK;
3835 }
3836
3837 if (data.arc_cflags & ARC_CFLAG_ANON) {
3838 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3839 mdb_printf("Histogram of the number of anon buffers "
3840 "that are associated with an arc hdr.\n");
3841 dump_histogram(data.anon_bufs, data.hist_nbuckets, 0);
3842 mdb_printf("\n");
3843 }
3844 mdb_printf("Histogram of compressed anon buffers.\n"
3845 "Each bucket represents buffers of size: %s.\n", range);
3846 dump_histogram(data.anon_c_hist, data.hist_nbuckets, 0);
3847 mdb_printf("\n");
3848
3849 mdb_printf("Histogram of uncompressed anon buffers.\n"
3850 "Each bucket represents buffers of size: %s.\n", range);
3851 dump_histogram(data.anon_u_hist, data.hist_nbuckets, 0);
3852 mdb_printf("\n");
3853 }
3854
3855 if (data.arc_cflags & ARC_CFLAG_MRU) {
3856 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3857 mdb_printf("Histogram of the number of mru buffers "
3858 "that are associated with an arc hdr.\n");
3859 dump_histogram(data.mru_bufs, data.hist_nbuckets, 0);
3860 mdb_printf("\n");
3861 }
3862 mdb_printf("Histogram of compressed mru buffers.\n"
3863 "Each bucket represents buffers of size: %s.\n", range);
3864 dump_histogram(data.mru_c_hist, data.hist_nbuckets, 0);
3865 mdb_printf("\n");
3866
3867 mdb_printf("Histogram of uncompressed mru buffers.\n"
3868 "Each bucket represents buffers of size: %s.\n", range);
3869 dump_histogram(data.mru_u_hist, data.hist_nbuckets, 0);
3870 mdb_printf("\n");
3871 }
3872
3873 if (data.arc_cflags & ARC_CFLAG_MFU) {
3874 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3875 mdb_printf("Histogram of the number of mfu buffers "
3876 "that are associated with an arc hdr.\n");
3877 dump_histogram(data.mfu_bufs, data.hist_nbuckets, 0);
3878 mdb_printf("\n");
3879 }
3880
3881 mdb_printf("Histogram of compressed mfu buffers.\n"
3882 "Each bucket represents buffers of size: %s.\n", range);
3883 dump_histogram(data.mfu_c_hist, data.hist_nbuckets, 0);
3884 mdb_printf("\n");
3885
3886 mdb_printf("Histogram of uncompressed mfu buffers.\n"
3887 "Each bucket represents buffers of size: %s.\n", range);
3888 dump_histogram(data.mfu_u_hist, data.hist_nbuckets, 0);
3889 mdb_printf("\n");
3890 }
3891
3892 if (data.arc_cflags & ARC_CFLAG_BUFS) {
3893 mdb_printf("Histogram of all buffers that "
3894 "are associated with an arc hdr.\n");
3895 dump_histogram(data.all_bufs, data.hist_nbuckets, 0);
3896 mdb_printf("\n");
3897 }
3898
3899 mdb_printf("Histogram of all compressed buffers.\n"
3900 "Each bucket represents buffers of size: %s.\n", range);
3901 dump_histogram(data.all_c_hist, data.hist_nbuckets, 0);
3902 mdb_printf("\n");
3903
3904 mdb_printf("Histogram of all uncompressed buffers.\n"
3905 "Each bucket represents buffers of size: %s.\n", range);
3906 dump_histogram(data.all_u_hist, data.hist_nbuckets, 0);
3907
3908 out:
3909 mdb_free(data.anon_c_hist, hist_size);
3910 mdb_free(data.anon_u_hist, hist_size);
3911 mdb_free(data.anon_bufs, hist_size);
3912
3913 mdb_free(data.mru_c_hist, hist_size);
3914 mdb_free(data.mru_u_hist, hist_size);
3915 mdb_free(data.mru_bufs, hist_size);
3916
3917 mdb_free(data.mfu_c_hist, hist_size);
3918 mdb_free(data.mfu_u_hist, hist_size);
3919 mdb_free(data.mfu_bufs, hist_size);
3920
3921 mdb_free(data.all_c_hist, hist_size);
3922 mdb_free(data.all_u_hist, hist_size);
3923 mdb_free(data.all_bufs, hist_size);
3924
3925 return (rc);
3926 }
3927
3928 /*
3929 * MDB module linkage information:
3930 *
3931 * We declare a list of structures describing our dcmds, and a function
3932 * named _mdb_init to return a pointer to our module information.
3933 */
3934
3935 static const mdb_dcmd_t dcmds[] = {
3936 { "arc", "[-bkmg]", "print ARC variables", arc_print },
3937 { "blkptr", ":", "print blkptr_t", blkptr },
3938 { "dva", ":", "print dva_t", dva },
3939 { "dbuf", ":", "print dmu_buf_impl_t", dbuf },
3940 { "dbuf_stats", ":", "dbuf stats", dbuf_stats },
3941 { "dbufs",
3942 "\t[-O objset_t*] [-n objset_name | \"mos\"] "
3943 "[-o object | \"mdn\"] \n"
3944 "\t[-l level] [-b blkid | \"bonus\"]",
3945 "find dmu_buf_impl_t's that match specified criteria", dbufs },
3946 { "abuf_find", "dva_word[0] dva_word[1]",
3947 "find arc_buf_hdr_t of a specified DVA",
3948 abuf_find },
3949 { "spa", "?[-cevmMh]\n"
3950 "\t-c display spa config\n"
3951 "\t-e display vdev statistics\n"
3952 "\t-v display vdev information\n"
3953 "\t-m display metaslab statistics\n"
3954 "\t-M display metaslab group statistics\n"
3955 "\t-h display histogram (requires -m or -M)\n",
3956 "spa_t summary", spa_print },
3957 { "spa_config", ":", "print spa_t configuration", spa_print_config },
3958 { "spa_space", ":[-b]", "print spa_t on-disk space usage", spa_space },
3959 { "spa_vdevs", ":[-emMh]\n"
3960 "\t-e display vdev statistics\n"
3961 "\t-m dispaly metaslab statistics\n"
3962 "\t-M display metaslab group statistic\n"
3963 "\t-h display histogram (requires -m or -M)\n",
3964 "given a spa_t, print vdev summary", spa_vdevs },
3965 { "vdev", ":[-remMh]\n"
3966 "\t-r display recursively\n"
3967 "\t-e display statistics\n"
3968 "\t-m display metaslab statistics (top level vdev only)\n"
3969 "\t-M display metaslab group statistics (top level vdev only)\n"
3970 "\t-h display histogram (requires -m or -M)\n",
3971 "vdev_t summary", vdev_print },
3972 { "zio", ":[-cpr]\n"
3973 "\t-c display children\n"
3974 "\t-p display parents\n"
3975 "\t-r display recursively",
3976 "zio_t summary", zio_print },
3977 { "zio_state", "?", "print out all zio_t structures on system or "
3978 "for a particular pool", zio_state },
3979 { "zfs_blkstats", ":[-v]",
3980 "given a spa_t, print block type stats from last scrub",
3981 zfs_blkstats },
3982 { "zfs_params", "", "print zfs tunable parameters", zfs_params },
3983 { "refcount", ":[-r]\n"
3984 "\t-r display recently removed references",
3985 "print refcount_t holders", refcount },
3986 { "zap_leaf", "", "print zap_leaf_phys_t", zap_leaf },
3987 { "zfs_aces", ":[-v]", "print all ACEs from a zfs_acl_t",
3988 zfs_acl_dump },
3989 { "zfs_ace", ":[-v]", "print zfs_ace", zfs_ace_print },
3990 { "zfs_ace0", ":[-v]", "print zfs_ace0", zfs_ace0_print },
3991 { "sa_attr_table", ":", "print SA attribute table from sa_os_t",
3992 sa_attr_table},
3993 { "sa_attr", ": attr_id",
3994 "print SA attribute address when given sa_handle_t", sa_attr_print},
3995 { "zfs_dbgmsg", ":[-va]",
3996 "print zfs debug log", dbgmsg},
3997 { "rrwlock", ":",
3998 "print rrwlock_t, including readers", rrwlock},
3999 { "metaslab_weight", "weight",
4000 "print metaslab weight", metaslab_weight},
4001 { "metaslab_trace", ":",
4002 "print metaslab allocation trace records", metaslab_trace},
4003 { "arc_compression_stats", ":[-vabrf]\n"
4004 "\t-v verbose, display a linearly scaled histogram\n"
4005 "\t-a display ARC_anon state statistics individually\n"
4006 "\t-r display ARC_mru state statistics individually\n"
4007 "\t-f display ARC_mfu state statistics individually\n"
4008 "\t-b display histogram of buffer counts\n",
4009 "print a histogram of compressed arc buffer sizes",
4010 arc_compression_stats},
4011 { NULL }
4012 };
4013
4014 static const mdb_walker_t walkers[] = {
4015 { "zms_freelist", "walk ZFS metaslab freelist",
4016 freelist_walk_init, freelist_walk_step, NULL },
4017 { "txg_list", "given any txg_list_t *, walk all entries in all txgs",
4018 txg_list_walk_init, txg_list_walk_step, NULL },
4019 { "txg_list0", "given any txg_list_t *, walk all entries in txg 0",
4020 txg_list0_walk_init, txg_list_walk_step, NULL },
4021 { "txg_list1", "given any txg_list_t *, walk all entries in txg 1",
4022 txg_list1_walk_init, txg_list_walk_step, NULL },
4023 { "txg_list2", "given any txg_list_t *, walk all entries in txg 2",
4024 txg_list2_walk_init, txg_list_walk_step, NULL },
4025 { "txg_list3", "given any txg_list_t *, walk all entries in txg 3",
4026 txg_list3_walk_init, txg_list_walk_step, NULL },
4027 { "zio", "walk all zio structures, optionally for a particular spa_t",
4028 zio_walk_init, zio_walk_step, NULL },
4029 { "zio_root",
4030 "walk all root zio_t structures, optionally for a particular spa_t",
4031 zio_walk_init, zio_walk_root_step, NULL },
4032 { "spa", "walk all spa_t entries in the namespace",
4033 spa_walk_init, spa_walk_step, NULL },
4034 { "metaslab", "given a spa_t *, walk all metaslab_t structures",
4035 metaslab_walk_init, metaslab_walk_step, NULL },
4036 { "multilist", "given a multilist_t *, walk all list_t structures",
4037 multilist_walk_init, multilist_walk_step, NULL },
4038 { "zfs_acl_node", "given a zfs_acl_t, walk all zfs_acl_nodes",
4039 zfs_acl_node_walk_init, zfs_acl_node_walk_step, NULL },
4040 { "zfs_acl_node_aces", "given a zfs_acl_node_t, walk all ACEs",
4041 zfs_acl_node_aces_walk_init, zfs_aces_walk_step, NULL },
4042 { "zfs_acl_node_aces0",
4043 "given a zfs_acl_node_t, walk all ACEs as ace_t",
4044 zfs_acl_node_aces0_walk_init, zfs_aces_walk_step, NULL },
4045 { NULL }
4046 };
4047
4048 static const mdb_modinfo_t modinfo = {
4049 MDB_API_VERSION, dcmds, walkers
4050 };
4051
4052 const mdb_modinfo_t *
4053 _mdb_init(void)
4054 {
4055 return (&modinfo);
4056 }
Unleashed OS