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.
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.
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]
23 * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2017 Joyent, Inc.
30 #include <sys/spa_impl.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dsl_dataset.h>
36 #include <sys/dsl_dir.h>
37 #include <sys/cmn_err.h>
38 #include <sys/sunddi.h>
40 #include "zfs_comutil.h"
41 #include "zfs_gitrev.h"
47 * Routines to manage the on-disk history log.
49 * The history log is stored as a dmu object containing
50 * <packed record length, record nvlist> tuples.
52 * Where "record nvlist" is an nvlist containing uint64_ts and strings, and
53 * "packed record length" is the packed length of the "record nvlist" stored
54 * as a little endian uint64_t.
56 * The log is implemented as a ring buffer, though the original creation
57 * of the pool ('zpool create') is never overwritten.
59 * The history log is tracked as object 'spa_t::spa_history'. The bonus buffer
60 * of 'spa_history' stores the offsets for logging/retrieving history as
61 * 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of
62 * where the 'zpool create' record is stored. This allows us to never
63 * overwrite the original creation of the pool. 'sh_phys_max_off' is the
64 * physical ending offset in bytes of the log. This tells you the length of
65 * the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record
66 * is added, 'sh_eof' is incremented by the size of the record.
67 * 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes).
68 * This is where the consumer should start reading from after reading in
69 * the 'zpool create' portion of the log.
71 * 'sh_records_lost' keeps track of how many records have been overwritten
72 * and permanently lost.
75 /* convert a logical offset to physical */
77 spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
81 phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len;
82 return ((log_off - shpp->sh_pool_create_len) % phys_len
83 + shpp->sh_pool_create_len);
87 spa_history_create_obj(spa_t *spa, dmu_tx_t *tx)
90 spa_history_phys_t *shpp;
91 objset_t *mos = spa->spa_meta_objset;
93 ASSERT0(spa->spa_history);
94 spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY,
95 SPA_OLD_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS,
96 sizeof (spa_history_phys_t), tx);
98 VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
99 DMU_POOL_HISTORY, sizeof (uint64_t), 1,
100 &spa->spa_history, tx));
102 VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
103 ASSERT3U(dbp->db_size, >=, sizeof (spa_history_phys_t));
106 dmu_buf_will_dirty(dbp, tx);
109 * Figure out maximum size of history log. We set it at
110 * 0.1% of pool size, with a max of 1G and min of 128KB.
112 shpp->sh_phys_max_off =
113 metaslab_class_get_dspace(spa_normal_class(spa)) / 1000;
114 shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 1<<30);
115 shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10);
117 dmu_buf_rele(dbp, FTAG);
121 * Change 'sh_bof' to the beginning of the next record.
124 spa_history_advance_bof(spa_t *spa, spa_history_phys_t *shpp)
126 objset_t *mos = spa->spa_meta_objset;
127 uint64_t firstread, reclen, phys_bof;
128 char buf[sizeof (reclen)];
131 phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
132 firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);
134 if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
135 buf, DMU_READ_PREFETCH)) != 0)
137 if (firstread != sizeof (reclen)) {
138 if ((err = dmu_read(mos, spa->spa_history,
139 shpp->sh_pool_create_len, sizeof (reclen) - firstread,
140 buf + firstread, DMU_READ_PREFETCH)) != 0)
144 reclen = LE_64(*((uint64_t *)buf));
145 shpp->sh_bof += reclen + sizeof (reclen);
146 shpp->sh_records_lost++;
151 spa_history_write(spa_t *spa, void *buf, uint64_t len, spa_history_phys_t *shpp,
154 uint64_t firstwrite, phys_eof;
155 objset_t *mos = spa->spa_meta_objset;
158 ASSERT(MUTEX_HELD(&spa->spa_history_lock));
160 /* see if we need to reset logical BOF */
161 while (shpp->sh_phys_max_off - shpp->sh_pool_create_len -
162 (shpp->sh_eof - shpp->sh_bof) <= len) {
163 if ((err = spa_history_advance_bof(spa, shpp)) != 0) {
168 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
169 firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
171 dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);
175 /* write out the rest at the beginning of physical file */
176 dmu_write(mos, spa->spa_history, shpp->sh_pool_create_len,
177 len, (char *)buf + firstwrite, tx);
184 * Post a history sysevent.
186 * The nvlist_t* passed into this function will be transformed into a new
189 * 1. Nested nvlists will be flattened to a single level
190 * 2. Keys will have their names normalized (to remove any problematic
191 * characters, such as whitespace)
193 * The nvlist_t passed into this function will duplicated and should be freed
198 spa_history_log_notify(spa_t *spa, nvlist_t *nvl)
200 nvlist_t *hist_nvl = fnvlist_alloc();
204 if (nvlist_lookup_string(nvl, ZPOOL_HIST_CMD, &string) == 0)
205 fnvlist_add_string(hist_nvl, ZFS_EV_HIST_CMD, string);
207 if (nvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME, &string) == 0)
208 fnvlist_add_string(hist_nvl, ZFS_EV_HIST_INT_NAME, string);
210 if (nvlist_lookup_string(nvl, ZPOOL_HIST_ZONE, &string) == 0)
211 fnvlist_add_string(hist_nvl, ZFS_EV_HIST_ZONE, string);
213 if (nvlist_lookup_string(nvl, ZPOOL_HIST_HOST, &string) == 0)
214 fnvlist_add_string(hist_nvl, ZFS_EV_HIST_HOST, string);
216 if (nvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME, &string) == 0)
217 fnvlist_add_string(hist_nvl, ZFS_EV_HIST_DSNAME, string);
219 if (nvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR, &string) == 0)
220 fnvlist_add_string(hist_nvl, ZFS_EV_HIST_INT_STR, string);
222 if (nvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL, &string) == 0)
223 fnvlist_add_string(hist_nvl, ZFS_EV_HIST_IOCTL, string);
225 if (nvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME, &string) == 0)
226 fnvlist_add_string(hist_nvl, ZFS_EV_HIST_INT_NAME, string);
228 if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID, &uint64) == 0)
229 fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_DSID, uint64);
231 if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG, &uint64) == 0)
232 fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_TXG, uint64);
234 if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_TIME, &uint64) == 0)
235 fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_TIME, uint64);
237 if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_WHO, &uint64) == 0)
238 fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_WHO, uint64);
240 if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_INT_EVENT, &uint64) == 0)
241 fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_INT_EVENT, uint64);
243 spa_event_notify(spa, NULL, hist_nvl, ESC_ZFS_HISTORY_EVENT);
245 nvlist_free(hist_nvl);
249 * Write out a history event.
253 spa_history_log_sync(void *arg, dmu_tx_t *tx)
256 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
257 objset_t *mos = spa->spa_meta_objset;
259 spa_history_phys_t *shpp;
262 char *record_packed = NULL;
266 * If we have an older pool that doesn't have a command
267 * history object, create it now.
269 mutex_enter(&spa->spa_history_lock);
270 if (!spa->spa_history)
271 spa_history_create_obj(spa, tx);
272 mutex_exit(&spa->spa_history_lock);
275 * Get the offset of where we need to write via the bonus buffer.
276 * Update the offset when the write completes.
278 VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
281 dmu_buf_will_dirty(dbp, tx);
285 dmu_object_info_t doi;
286 dmu_object_info_from_db(dbp, &doi);
287 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
291 fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
292 fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname()->nodename);
294 if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
295 zfs_dbgmsg("command: %s",
296 fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
297 } else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
298 if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
299 zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
300 fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
301 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
302 fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
303 fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
304 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
306 zfs_dbgmsg("txg %lld %s %s",
307 fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
308 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
309 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
312 * The history sysevent is posted only for internal history
313 * messages to show what has happened, not how it happened. For
314 * example, the following command:
316 * # zfs destroy -r tank/foo
318 * will result in one sysevent posted per dataset that is
319 * destroyed as a result of the command - which could be more
320 * than one event in total. By contrast, if the sysevent was
321 * posted as a result of the ZPOOL_HIST_CMD key being present
322 * it would result in only one sysevent being posted with the
323 * full command line arguments, requiring the consumer to know
324 * how to parse and understand zfs(1M) command invocations.
326 spa_history_log_notify(spa, nvl);
327 } else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
328 zfs_dbgmsg("ioctl %s",
329 fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
332 VERIFY3U(nvlist_pack(nvl, &record_packed, &reclen, NV_ENCODE_NATIVE,
335 mutex_enter(&spa->spa_history_lock);
337 /* write out the packed length as little endian */
338 le_len = LE_64((uint64_t)reclen);
339 ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
341 ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
343 /* The first command is the create, which we keep forever */
344 if (ret == 0 && shpp->sh_pool_create_len == 0 &&
345 nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
346 shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
349 mutex_exit(&spa->spa_history_lock);
350 fnvlist_pack_free(record_packed, reclen);
351 dmu_buf_rele(dbp, FTAG);
356 * Write out a history event.
359 spa_history_log(spa_t *spa, const char *msg)
362 nvlist_t *nvl = fnvlist_alloc();
364 fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg);
365 err = spa_history_log_nvl(spa, nvl);
371 spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
375 nvlist_t *nvarg, *in_nvl = NULL;
377 if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY || !spa_writeable(spa))
378 return (SET_ERROR(EINVAL));
380 err = nvlist_lookup_nvlist(nvl, ZPOOL_HIST_INPUT_NVL, &in_nvl);
382 (void) nvlist_remove_all(in_nvl, ZPOOL_HIDDEN_ARGS);
385 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
386 err = dmu_tx_assign(tx, TXG_WAIT);
392 VERIFY0(nvlist_dup(nvl, &nvarg, KM_SLEEP));
393 if (spa_history_zone() != NULL) {
394 fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
397 fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
399 /* Kick this off asynchronously; errors are ignored. */
400 dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
401 nvarg, 0, ZFS_SPACE_CHECK_NONE, tx);
404 /* spa_history_log_sync will free nvl */
409 * Read out the command history.
412 spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
414 objset_t *mos = spa->spa_meta_objset;
416 uint64_t read_len, phys_read_off, phys_eof;
417 uint64_t leftover = 0;
418 spa_history_phys_t *shpp;
422 * If the command history doesn't exist (older pool),
423 * that's ok, just return ENOENT.
425 if (!spa->spa_history)
426 return (SET_ERROR(ENOENT));
429 * The history is logged asynchronously, so when they request
430 * the first chunk of history, make sure everything has been
431 * synced to disk so that we get it.
433 if (*offp == 0 && spa_writeable(spa))
434 txg_wait_synced(spa_get_dsl(spa), 0);
436 if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
442 dmu_object_info_t doi;
443 dmu_object_info_from_db(dbp, &doi);
444 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
448 mutex_enter(&spa->spa_history_lock);
449 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
451 if (*offp < shpp->sh_pool_create_len) {
452 /* read in just the zpool create history */
453 phys_read_off = *offp;
454 read_len = MIN(*len, shpp->sh_pool_create_len -
458 * Need to reset passed in offset to BOF if the passed in
459 * offset has since been overwritten.
461 *offp = MAX(*offp, shpp->sh_bof);
462 phys_read_off = spa_history_log_to_phys(*offp, shpp);
465 * Read up to the minimum of what the user passed down or
466 * the EOF (physical or logical). If we hit physical EOF,
467 * use 'leftover' to read from the physical BOF.
469 if (phys_read_off <= phys_eof) {
470 read_len = MIN(*len, phys_eof - phys_read_off);
473 shpp->sh_phys_max_off - phys_read_off);
474 if (phys_read_off + *len > shpp->sh_phys_max_off) {
475 leftover = MIN(*len - read_len,
476 phys_eof - shpp->sh_pool_create_len);
481 /* offset for consumer to use next */
482 *offp += read_len + leftover;
484 /* tell the consumer how much you actually read */
485 *len = read_len + leftover;
488 mutex_exit(&spa->spa_history_lock);
489 dmu_buf_rele(dbp, FTAG);
493 err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
495 if (leftover && err == 0) {
496 err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
497 leftover, buf + read_len, DMU_READ_PREFETCH);
499 mutex_exit(&spa->spa_history_lock);
501 dmu_buf_rele(dbp, FTAG);
506 * The nvlist will be consumed by this call.
509 log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
510 dmu_tx_t *tx, const char *fmt, va_list adx)
515 * If this is part of creating a pool, not everything is
516 * initialized yet, so don't bother logging the internal events.
517 * Likewise if the pool is not writeable.
519 if (spa_is_initializing(spa) || !spa_writeable(spa)) {
524 msg = kmem_vasprintf(fmt, adx);
525 fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
528 fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
529 fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
531 if (dmu_tx_is_syncing(tx)) {
532 spa_history_log_sync(nvl, tx);
534 dsl_sync_task_nowait(spa_get_dsl(spa),
535 spa_history_log_sync, nvl, 0, ZFS_SPACE_CHECK_NONE, tx);
537 /* spa_history_log_sync() will free nvl */
541 spa_history_log_internal(spa_t *spa, const char *operation,
542 dmu_tx_t *tx, const char *fmt, ...)
547 /* create a tx if we didn't get one */
549 htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
550 if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
557 log_internal(fnvlist_alloc(), operation, spa, htx, fmt, adx);
560 /* if we didn't get a tx from the caller, commit the one we made */
566 spa_history_log_internal_ds(dsl_dataset_t *ds, const char *operation,
567 dmu_tx_t *tx, const char *fmt, ...)
570 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
571 nvlist_t *nvl = fnvlist_alloc();
575 dsl_dataset_name(ds, namebuf);
576 fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
577 fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID, ds->ds_object);
580 log_internal(nvl, operation, dsl_dataset_get_spa(ds), tx, fmt, adx);
585 spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
586 dmu_tx_t *tx, const char *fmt, ...)
589 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
590 nvlist_t *nvl = fnvlist_alloc();
594 dsl_dir_name(dd, namebuf);
595 fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
596 fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
597 dsl_dir_phys(dd)->dd_head_dataset_obj);
600 log_internal(nvl, operation, dd->dd_pool->dp_spa, tx, fmt, adx);
605 spa_history_log_version(spa_t *spa, const char *operation, dmu_tx_t *tx)
607 utsname_t *u = utsname();
609 spa_history_log_internal(spa, operation, tx,
610 "pool version %llu; software version %s; uts %s %s %s %s",
611 (u_longlong_t)spa_version(spa), ZFS_META_GITREV,
612 u->nodename, u->release, u->version, u->machine);
617 spa_history_zone(void)
624 EXPORT_SYMBOL(spa_history_create_obj);
625 EXPORT_SYMBOL(spa_history_get);
626 EXPORT_SYMBOL(spa_history_log);
627 EXPORT_SYMBOL(spa_history_log_internal);
628 EXPORT_SYMBOL(spa_history_log_version);