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, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
29 #include <sys/spa_impl.h>
31 #include <sys/dsl_synctask.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/dmu_objset.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/utsname.h>
37 #include <sys/sunddi.h>
39 #include "zfs_comutil.h"
41 #include <sys/cmn_err.h>
46 * Routines to manage the on-disk history log.
48 * The history log is stored as a dmu object containing
49 * <packed record length, record nvlist> tuples.
51 * Where "record nvlist" is a nvlist containing uint64_ts and strings, and
52 * "packed record length" is the packed length of the "record nvlist" stored
53 * as a little endian uint64_t.
55 * The log is implemented as a ring buffer, though the original creation
56 * of the pool ('zpool create') is never overwritten.
58 * The history log is tracked as object 'spa_t::spa_history'. The bonus buffer
59 * of 'spa_history' stores the offsets for logging/retrieving history as
60 * 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of
61 * where the 'zpool create' record is stored. This allows us to never
62 * overwrite the original creation of the pool. 'sh_phys_max_off' is the
63 * physical ending offset in bytes of the log. This tells you the length of
64 * the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record
65 * is added, 'sh_eof' is incremented by the the size of the record.
66 * 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes).
67 * This is where the consumer should start reading from after reading in
68 * the 'zpool create' portion of the log.
70 * 'sh_records_lost' keeps track of how many records have been overwritten
71 * and permanently lost.
74 /* convert a logical offset to physical */
76 spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
80 phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len;
81 return ((log_off - shpp->sh_pool_create_len) % phys_len
82 + shpp->sh_pool_create_len);
86 spa_history_create_obj(spa_t *spa, dmu_tx_t *tx)
89 spa_history_phys_t *shpp;
90 objset_t *mos = spa->spa_meta_objset;
92 ASSERT(spa->spa_history == 0);
93 spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY,
94 SPA_OLD_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS,
95 sizeof (spa_history_phys_t), tx);
97 VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
98 DMU_POOL_HISTORY, sizeof (uint64_t), 1,
99 &spa->spa_history, tx) == 0);
101 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
102 ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));
105 dmu_buf_will_dirty(dbp, tx);
108 * Figure out maximum size of history log. We set it at
109 * 0.1% of pool size, with a max of 1G and min of 128KB.
111 shpp->sh_phys_max_off =
112 metaslab_class_get_dspace(spa_normal_class(spa)) / 1000;
113 shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 1<<30);
114 shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10);
116 dmu_buf_rele(dbp, FTAG);
120 * Change 'sh_bof' to the beginning of the next record.
123 spa_history_advance_bof(spa_t *spa, spa_history_phys_t *shpp)
125 objset_t *mos = spa->spa_meta_objset;
126 uint64_t firstread, reclen, phys_bof;
127 char buf[sizeof (reclen)];
130 phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
131 firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);
133 if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
134 buf, DMU_READ_PREFETCH)) != 0)
136 if (firstread != sizeof (reclen)) {
137 if ((err = dmu_read(mos, spa->spa_history,
138 shpp->sh_pool_create_len, sizeof (reclen) - firstread,
139 buf + firstread, DMU_READ_PREFETCH)) != 0)
143 reclen = LE_64(*((uint64_t *)buf));
144 shpp->sh_bof += reclen + sizeof (reclen);
145 shpp->sh_records_lost++;
150 spa_history_write(spa_t *spa, void *buf, uint64_t len, spa_history_phys_t *shpp,
153 uint64_t firstwrite, phys_eof;
154 objset_t *mos = spa->spa_meta_objset;
157 ASSERT(MUTEX_HELD(&spa->spa_history_lock));
159 /* see if we need to reset logical BOF */
160 while (shpp->sh_phys_max_off - shpp->sh_pool_create_len -
161 (shpp->sh_eof - shpp->sh_bof) <= len) {
162 if ((err = spa_history_advance_bof(spa, shpp)) != 0) {
167 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
168 firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
170 dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);
174 /* write out the rest at the beginning of physical file */
175 dmu_write(mos, spa->spa_history, shpp->sh_pool_create_len,
176 len, (char *)buf + firstwrite, tx);
183 spa_history_zone(void)
186 /* XXX: pr_hostname can be changed by default from within a jail! */
187 if (jailed(curthread->td_ucred))
188 return (curthread->td_ucred->cr_prison->pr_hostname);
194 * Write out a history event.
198 spa_history_log_sync(void *arg, dmu_tx_t *tx)
201 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
202 objset_t *mos = spa->spa_meta_objset;
204 spa_history_phys_t *shpp;
207 char *record_packed = NULL;
211 * If we have an older pool that doesn't have a command
212 * history object, create it now.
214 mutex_enter(&spa->spa_history_lock);
215 if (!spa->spa_history)
216 spa_history_create_obj(spa, tx);
217 mutex_exit(&spa->spa_history_lock);
220 * Get the offset of where we need to write via the bonus buffer.
221 * Update the offset when the write completes.
223 VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
226 dmu_buf_will_dirty(dbp, tx);
230 dmu_object_info_t doi;
231 dmu_object_info_from_db(dbp, &doi);
232 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
236 fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
238 fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname.nodename);
240 if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
241 zfs_dbgmsg("command: %s",
242 fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
243 } else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
244 if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
245 zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
246 fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
247 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
248 fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
249 fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
250 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
252 zfs_dbgmsg("txg %lld %s %s",
253 fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
254 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
255 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
257 } else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
258 zfs_dbgmsg("ioctl %s",
259 fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
262 record_packed = fnvlist_pack(nvl, &reclen);
264 mutex_enter(&spa->spa_history_lock);
266 /* write out the packed length as little endian */
267 le_len = LE_64((uint64_t)reclen);
268 ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
270 ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
272 /* The first command is the create, which we keep forever */
273 if (ret == 0 && shpp->sh_pool_create_len == 0 &&
274 nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
275 shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
278 mutex_exit(&spa->spa_history_lock);
279 fnvlist_pack_free(record_packed, reclen);
280 dmu_buf_rele(dbp, FTAG);
285 * Write out a history event.
288 spa_history_log(spa_t *spa, const char *msg)
291 nvlist_t *nvl = fnvlist_alloc();
293 fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg);
294 err = spa_history_log_nvl(spa, nvl);
300 spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
306 if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY)
309 if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY || !spa_writeable(spa))
310 return (SET_ERROR(EINVAL));
312 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
313 err = dmu_tx_assign(tx, TXG_WAIT);
319 nvarg = fnvlist_dup(nvl);
320 if (spa_history_zone() != NULL) {
321 fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
324 fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
326 /* Kick this off asynchronously; errors are ignored. */
327 dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
328 nvarg, 0, ZFS_SPACE_CHECK_NONE, tx);
331 /* spa_history_log_sync will free nvl */
337 * Read out the command history.
340 spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
342 objset_t *mos = spa->spa_meta_objset;
344 uint64_t read_len, phys_read_off, phys_eof;
345 uint64_t leftover = 0;
346 spa_history_phys_t *shpp;
350 * If the command history doesn't exist (older pool),
351 * that's ok, just return ENOENT.
353 if (!spa->spa_history)
354 return (SET_ERROR(ENOENT));
357 * The history is logged asynchronously, so when they request
358 * the first chunk of history, make sure everything has been
359 * synced to disk so that we get it.
361 if (*offp == 0 && spa_writeable(spa))
362 txg_wait_synced(spa_get_dsl(spa), 0);
364 if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
370 dmu_object_info_t doi;
371 dmu_object_info_from_db(dbp, &doi);
372 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
376 mutex_enter(&spa->spa_history_lock);
377 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
379 if (*offp < shpp->sh_pool_create_len) {
380 /* read in just the zpool create history */
381 phys_read_off = *offp;
382 read_len = MIN(*len, shpp->sh_pool_create_len -
386 * Need to reset passed in offset to BOF if the passed in
387 * offset has since been overwritten.
389 *offp = MAX(*offp, shpp->sh_bof);
390 phys_read_off = spa_history_log_to_phys(*offp, shpp);
393 * Read up to the minimum of what the user passed down or
394 * the EOF (physical or logical). If we hit physical EOF,
395 * use 'leftover' to read from the physical BOF.
397 if (phys_read_off <= phys_eof) {
398 read_len = MIN(*len, phys_eof - phys_read_off);
401 shpp->sh_phys_max_off - phys_read_off);
402 if (phys_read_off + *len > shpp->sh_phys_max_off) {
403 leftover = MIN(*len - read_len,
404 phys_eof - shpp->sh_pool_create_len);
409 /* offset for consumer to use next */
410 *offp += read_len + leftover;
412 /* tell the consumer how much you actually read */
413 *len = read_len + leftover;
416 mutex_exit(&spa->spa_history_lock);
417 dmu_buf_rele(dbp, FTAG);
421 err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
423 if (leftover && err == 0) {
424 err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
425 leftover, buf + read_len, DMU_READ_PREFETCH);
427 mutex_exit(&spa->spa_history_lock);
429 dmu_buf_rele(dbp, FTAG);
434 * The nvlist will be consumed by this call.
437 log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
438 dmu_tx_t *tx, const char *fmt, va_list adx)
444 * If this is part of creating a pool, not everything is
445 * initialized yet, so don't bother logging the internal events.
446 * Likewise if the pool is not writeable.
448 if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
455 msg = kmem_alloc(vsnprintf(NULL, 0, fmt, adx) + 1, KM_SLEEP);
456 (void) vsprintf(msg, fmt, adx2);
457 fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
462 fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
463 fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
465 if (dmu_tx_is_syncing(tx)) {
466 spa_history_log_sync(nvl, tx);
468 dsl_sync_task_nowait(spa_get_dsl(spa),
469 spa_history_log_sync, nvl, 0, ZFS_SPACE_CHECK_NONE, tx);
471 /* spa_history_log_sync() will free nvl */
475 spa_history_log_internal(spa_t *spa, const char *operation,
476 dmu_tx_t *tx, const char *fmt, ...)
481 /* create a tx if we didn't get one */
483 htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
484 if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
491 log_internal(fnvlist_alloc(), operation, spa, htx, fmt, adx);
494 /* if we didn't get a tx from the caller, commit the one we made */
500 spa_history_log_internal_ds(dsl_dataset_t *ds, const char *operation,
501 dmu_tx_t *tx, const char *fmt, ...)
504 char namebuf[MAXNAMELEN];
505 nvlist_t *nvl = fnvlist_alloc();
509 dsl_dataset_name(ds, namebuf);
510 fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
511 fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID, ds->ds_object);
514 log_internal(nvl, operation, dsl_dataset_get_spa(ds), tx, fmt, adx);
519 spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
520 dmu_tx_t *tx, const char *fmt, ...)
523 char namebuf[MAXNAMELEN];
524 nvlist_t *nvl = fnvlist_alloc();
528 dsl_dir_name(dd, namebuf);
529 fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
530 fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
531 dsl_dir_phys(dd)->dd_head_dataset_obj);
534 log_internal(nvl, operation, dd->dd_pool->dp_spa, tx, fmt, adx);
539 spa_history_log_version(spa_t *spa, const char *operation)
541 spa_history_log_internal(spa, operation, NULL,
542 "pool version %llu; software version %llu/%d; uts %s %s %s %s",
543 (u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
544 utsname.nodename, utsname.release, utsname.version,