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.
252 spa_history_log_sync(void *arg, dmu_tx_t *tx)
255 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
256 objset_t *mos = spa->spa_meta_objset;
258 spa_history_phys_t *shpp;
261 char *record_packed = NULL;
265 * If we have an older pool that doesn't have a command
266 * history object, create it now.
268 mutex_enter(&spa->spa_history_lock);
269 if (!spa->spa_history)
270 spa_history_create_obj(spa, tx);
271 mutex_exit(&spa->spa_history_lock);
274 * Get the offset of where we need to write via the bonus buffer.
275 * Update the offset when the write completes.
277 VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
280 dmu_buf_will_dirty(dbp, tx);
284 dmu_object_info_t doi;
285 dmu_object_info_from_db(dbp, &doi);
286 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
290 fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname()->nodename);
292 if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
293 zfs_dbgmsg("command: %s",
294 fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
295 } else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
296 if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
297 zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
298 (longlong_t)fnvlist_lookup_uint64(nvl,
300 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
301 fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
302 (u_longlong_t)fnvlist_lookup_uint64(nvl,
304 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
306 zfs_dbgmsg("txg %lld %s %s",
307 (longlong_t)fnvlist_lookup_uint64(nvl,
309 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
310 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
313 * The history sysevent is posted only for internal history
314 * messages to show what has happened, not how it happened. For
315 * example, the following command:
317 * # zfs destroy -r tank/foo
319 * will result in one sysevent posted per dataset that is
320 * destroyed as a result of the command - which could be more
321 * than one event in total. By contrast, if the sysevent was
322 * posted as a result of the ZPOOL_HIST_CMD key being present
323 * it would result in only one sysevent being posted with the
324 * full command line arguments, requiring the consumer to know
325 * how to parse and understand zfs(8) command invocations.
327 spa_history_log_notify(spa, nvl);
328 } else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
329 zfs_dbgmsg("ioctl %s",
330 fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
333 VERIFY3U(nvlist_pack(nvl, &record_packed, &reclen, NV_ENCODE_NATIVE,
336 mutex_enter(&spa->spa_history_lock);
338 /* write out the packed length as little endian */
339 le_len = LE_64((uint64_t)reclen);
340 ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
342 ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
344 /* The first command is the create, which we keep forever */
345 if (ret == 0 && shpp->sh_pool_create_len == 0 &&
346 nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
347 shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
350 mutex_exit(&spa->spa_history_lock);
351 fnvlist_pack_free(record_packed, reclen);
352 dmu_buf_rele(dbp, FTAG);
357 * Write out a history event.
360 spa_history_log(spa_t *spa, const char *msg)
363 nvlist_t *nvl = fnvlist_alloc();
365 fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg);
366 err = spa_history_log_nvl(spa, nvl);
372 spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
376 nvlist_t *nvarg, *in_nvl = NULL;
378 if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY || !spa_writeable(spa))
379 return (SET_ERROR(EINVAL));
381 err = nvlist_lookup_nvlist(nvl, ZPOOL_HIST_INPUT_NVL, &in_nvl);
383 (void) nvlist_remove_all(in_nvl, ZPOOL_HIDDEN_ARGS);
386 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
387 err = dmu_tx_assign(tx, TXG_WAIT);
393 VERIFY0(nvlist_dup(nvl, &nvarg, KM_SLEEP));
394 if (spa_history_zone() != NULL) {
395 fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
398 fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
401 * Since the history is recorded asynchronously, the effective time is
402 * now, which may be considerably before the change is made on disk.
404 fnvlist_add_uint64(nvarg, ZPOOL_HIST_TIME, gethrestime_sec());
406 /* Kick this off asynchronously; errors are ignored. */
407 dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync, nvarg, tx);
410 /* spa_history_log_sync will free nvl */
415 * Read out the command history.
418 spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
420 objset_t *mos = spa->spa_meta_objset;
422 uint64_t read_len, phys_read_off, phys_eof;
423 uint64_t leftover = 0;
424 spa_history_phys_t *shpp;
428 * If the command history doesn't exist (older pool),
429 * that's ok, just return ENOENT.
431 if (!spa->spa_history)
432 return (SET_ERROR(ENOENT));
435 * The history is logged asynchronously, so when they request
436 * the first chunk of history, make sure everything has been
437 * synced to disk so that we get it.
439 if (*offp == 0 && spa_writeable(spa))
440 txg_wait_synced(spa_get_dsl(spa), 0);
442 if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
448 dmu_object_info_t doi;
449 dmu_object_info_from_db(dbp, &doi);
450 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
454 mutex_enter(&spa->spa_history_lock);
455 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
457 if (*offp < shpp->sh_pool_create_len) {
458 /* read in just the zpool create history */
459 phys_read_off = *offp;
460 read_len = MIN(*len, shpp->sh_pool_create_len -
464 * Need to reset passed in offset to BOF if the passed in
465 * offset has since been overwritten.
467 *offp = MAX(*offp, shpp->sh_bof);
468 phys_read_off = spa_history_log_to_phys(*offp, shpp);
471 * Read up to the minimum of what the user passed down or
472 * the EOF (physical or logical). If we hit physical EOF,
473 * use 'leftover' to read from the physical BOF.
475 if (phys_read_off <= phys_eof) {
476 read_len = MIN(*len, phys_eof - phys_read_off);
479 shpp->sh_phys_max_off - phys_read_off);
480 if (phys_read_off + *len > shpp->sh_phys_max_off) {
481 leftover = MIN(*len - read_len,
482 phys_eof - shpp->sh_pool_create_len);
487 /* offset for consumer to use next */
488 *offp += read_len + leftover;
490 /* tell the consumer how much you actually read */
491 *len = read_len + leftover;
494 mutex_exit(&spa->spa_history_lock);
495 dmu_buf_rele(dbp, FTAG);
499 err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
501 if (leftover && err == 0) {
502 err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
503 leftover, buf + read_len, DMU_READ_PREFETCH);
505 mutex_exit(&spa->spa_history_lock);
507 dmu_buf_rele(dbp, FTAG);
512 * The nvlist will be consumed by this call.
515 log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
516 dmu_tx_t *tx, const char *fmt, va_list adx)
521 * If this is part of creating a pool, not everything is
522 * initialized yet, so don't bother logging the internal events.
523 * Likewise if the pool is not writeable.
525 if (spa_is_initializing(spa) || !spa_writeable(spa)) {
530 msg = kmem_vasprintf(fmt, adx);
531 fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
534 fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
535 fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
536 fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
538 if (dmu_tx_is_syncing(tx)) {
539 spa_history_log_sync(nvl, tx);
541 dsl_sync_task_nowait(spa_get_dsl(spa),
542 spa_history_log_sync, nvl, tx);
544 /* spa_history_log_sync() will free nvl */
548 spa_history_log_internal(spa_t *spa, const char *operation,
549 dmu_tx_t *tx, const char *fmt, ...)
554 /* create a tx if we didn't get one */
556 htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
557 if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
564 log_internal(fnvlist_alloc(), operation, spa, htx, fmt, adx);
567 /* if we didn't get a tx from the caller, commit the one we made */
573 spa_history_log_internal_ds(dsl_dataset_t *ds, const char *operation,
574 dmu_tx_t *tx, const char *fmt, ...)
577 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
578 nvlist_t *nvl = fnvlist_alloc();
582 dsl_dataset_name(ds, namebuf);
583 fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
584 fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID, ds->ds_object);
587 log_internal(nvl, operation, dsl_dataset_get_spa(ds), tx, fmt, adx);
592 spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
593 dmu_tx_t *tx, const char *fmt, ...)
596 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
597 nvlist_t *nvl = fnvlist_alloc();
601 dsl_dir_name(dd, namebuf);
602 fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
603 fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
604 dsl_dir_phys(dd)->dd_head_dataset_obj);
607 log_internal(nvl, operation, dd->dd_pool->dp_spa, tx, fmt, adx);
612 spa_history_log_version(spa_t *spa, const char *operation, dmu_tx_t *tx)
614 utsname_t *u = utsname();
616 spa_history_log_internal(spa, operation, tx,
617 "pool version %llu; software version %s; uts %s %s %s %s",
618 (u_longlong_t)spa_version(spa), ZFS_META_GITREV,
619 u->nodename, u->release, u->version, u->machine);
624 spa_history_zone(void)
631 EXPORT_SYMBOL(spa_history_create_obj);
632 EXPORT_SYMBOL(spa_history_get);
633 EXPORT_SYMBOL(spa_history_log);
634 EXPORT_SYMBOL(spa_history_log_internal);
635 EXPORT_SYMBOL(spa_history_log_version);