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.
27 #include <sys/spa_impl.h>
29 #include <sys/dsl_synctask.h>
30 #include <sys/dmu_tx.h>
31 #include <sys/dmu_objset.h>
32 #include <sys/utsname.h>
33 #include <sys/cmn_err.h>
34 #include <sys/sunddi.h>
35 #include "zfs_comutil.h"
41 * Routines to manage the on-disk history log.
43 * The history log is stored as a dmu object containing
44 * <packed record length, record nvlist> tuples.
46 * Where "record nvlist" is a nvlist containing uint64_ts and strings, and
47 * "packed record length" is the packed length of the "record nvlist" stored
48 * as a little endian uint64_t.
50 * The log is implemented as a ring buffer, though the original creation
51 * of the pool ('zpool create') is never overwritten.
53 * The history log is tracked as object 'spa_t::spa_history'. The bonus buffer
54 * of 'spa_history' stores the offsets for logging/retrieving history as
55 * 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of
56 * where the 'zpool create' record is stored. This allows us to never
57 * overwrite the original creation of the pool. 'sh_phys_max_off' is the
58 * physical ending offset in bytes of the log. This tells you the length of
59 * the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record
60 * is added, 'sh_eof' is incremented by the the size of the record.
61 * 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes).
62 * This is where the consumer should start reading from after reading in
63 * the 'zpool create' portion of the log.
65 * 'sh_records_lost' keeps track of how many records have been overwritten
66 * and permanently lost.
69 /* convert a logical offset to physical */
71 spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
75 phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len;
76 return ((log_off - shpp->sh_pool_create_len) % phys_len
77 + shpp->sh_pool_create_len);
81 spa_history_create_obj(spa_t *spa, dmu_tx_t *tx)
84 spa_history_phys_t *shpp;
85 objset_t *mos = spa->spa_meta_objset;
87 ASSERT(spa->spa_history == 0);
88 spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY,
89 SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS,
90 sizeof (spa_history_phys_t), tx);
92 VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
93 DMU_POOL_HISTORY, sizeof (uint64_t), 1,
94 &spa->spa_history, tx) == 0);
96 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
97 ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));
100 dmu_buf_will_dirty(dbp, tx);
103 * Figure out maximum size of history log. We set it at
104 * 1% of pool size, with a max of 32MB and min of 128KB.
106 shpp->sh_phys_max_off =
107 metaslab_class_get_dspace(spa_normal_class(spa)) / 100;
108 shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 32<<20);
109 shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10);
111 dmu_buf_rele(dbp, FTAG);
115 * Change 'sh_bof' to the beginning of the next record.
118 spa_history_advance_bof(spa_t *spa, spa_history_phys_t *shpp)
120 objset_t *mos = spa->spa_meta_objset;
121 uint64_t firstread, reclen, phys_bof;
122 char buf[sizeof (reclen)];
125 phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
126 firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);
128 if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
129 buf, DMU_READ_PREFETCH)) != 0)
131 if (firstread != sizeof (reclen)) {
132 if ((err = dmu_read(mos, spa->spa_history,
133 shpp->sh_pool_create_len, sizeof (reclen) - firstread,
134 buf + firstread, DMU_READ_PREFETCH)) != 0)
138 reclen = LE_64(*((uint64_t *)buf));
139 shpp->sh_bof += reclen + sizeof (reclen);
140 shpp->sh_records_lost++;
145 spa_history_write(spa_t *spa, void *buf, uint64_t len, spa_history_phys_t *shpp,
148 uint64_t firstwrite, phys_eof;
149 objset_t *mos = spa->spa_meta_objset;
152 ASSERT(MUTEX_HELD(&spa->spa_history_lock));
154 /* see if we need to reset logical BOF */
155 while (shpp->sh_phys_max_off - shpp->sh_pool_create_len -
156 (shpp->sh_eof - shpp->sh_bof) <= len) {
157 if ((err = spa_history_advance_bof(spa, shpp)) != 0) {
162 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
163 firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
165 dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);
169 /* write out the rest at the beginning of physical file */
170 dmu_write(mos, spa->spa_history, shpp->sh_pool_create_len,
171 len, (char *)buf + firstwrite, tx);
178 spa_history_zone(void)
181 return (curproc->p_zone->zone_name);
188 * Write out a history event.
192 spa_history_log_sync(void *arg1, void *arg2, dmu_tx_t *tx)
195 history_arg_t *hap = arg2;
196 const char *history_str = hap->ha_history_str;
197 objset_t *mos = spa->spa_meta_objset;
199 spa_history_phys_t *shpp;
203 char *record_packed = NULL;
207 * If we have an older pool that doesn't have a command
208 * history object, create it now.
210 mutex_enter(&spa->spa_history_lock);
211 if (!spa->spa_history)
212 spa_history_create_obj(spa, tx);
213 mutex_exit(&spa->spa_history_lock);
216 * Get the offset of where we need to write via the bonus buffer.
217 * Update the offset when the write completes.
219 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
222 dmu_buf_will_dirty(dbp, tx);
226 dmu_object_info_t doi;
227 dmu_object_info_from_db(dbp, &doi);
228 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
232 VERIFY(nvlist_alloc(&nvrecord, NV_UNIQUE_NAME, KM_SLEEP) == 0);
233 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TIME,
234 gethrestime_sec()) == 0);
235 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_WHO, hap->ha_uid) == 0);
236 if (hap->ha_zone != NULL)
237 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_ZONE,
240 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_HOST,
241 utsname.nodename) == 0);
243 if (hap->ha_log_type == LOG_CMD_POOL_CREATE ||
244 hap->ha_log_type == LOG_CMD_NORMAL) {
245 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_CMD,
248 zfs_dbgmsg("command: %s", history_str);
250 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_INT_EVENT,
251 hap->ha_event) == 0);
252 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TXG,
254 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_INT_STR,
257 zfs_dbgmsg("internal %s pool:%s txg:%llu %s",
258 zfs_history_event_names[hap->ha_event], spa_name(spa),
259 (longlong_t)tx->tx_txg, history_str);
263 VERIFY(nvlist_size(nvrecord, &reclen, NV_ENCODE_XDR) == 0);
264 record_packed = kmem_alloc(reclen, KM_SLEEP);
266 VERIFY(nvlist_pack(nvrecord, &record_packed, &reclen,
267 NV_ENCODE_XDR, KM_SLEEP) == 0);
269 mutex_enter(&spa->spa_history_lock);
270 if (hap->ha_log_type == LOG_CMD_POOL_CREATE)
271 VERIFY(shpp->sh_eof == shpp->sh_pool_create_len);
273 /* write out the packed length as little endian */
274 le_len = LE_64((uint64_t)reclen);
275 ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
277 ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
279 if (!ret && hap->ha_log_type == LOG_CMD_POOL_CREATE) {
280 shpp->sh_pool_create_len += sizeof (le_len) + reclen;
281 shpp->sh_bof = shpp->sh_pool_create_len;
284 mutex_exit(&spa->spa_history_lock);
285 nvlist_free(nvrecord);
286 kmem_free(record_packed, reclen);
287 dmu_buf_rele(dbp, FTAG);
289 strfree(hap->ha_history_str);
290 if (hap->ha_zone != NULL)
291 strfree(hap->ha_zone);
292 kmem_free(hap, sizeof (history_arg_t));
296 * Write out a history event.
299 spa_history_log(spa_t *spa, const char *history_str, history_log_type_t what)
305 ASSERT(what != LOG_INTERNAL);
307 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
308 err = dmu_tx_assign(tx, TXG_WAIT);
314 ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
315 ha->ha_history_str = strdup(history_str);
316 ha->ha_zone = strdup(spa_history_zone());
317 ha->ha_log_type = what;
318 ha->ha_uid = crgetuid(CRED());
320 /* Kick this off asynchronously; errors are ignored. */
321 dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
322 spa_history_log_sync, spa, ha, 0, tx);
325 /* spa_history_log_sync will free ha and strings */
330 * Read out the command history.
333 spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
335 objset_t *mos = spa->spa_meta_objset;
337 uint64_t read_len, phys_read_off, phys_eof;
338 uint64_t leftover = 0;
339 spa_history_phys_t *shpp;
343 * If the command history doesn't exist (older pool),
344 * that's ok, just return ENOENT.
346 if (!spa->spa_history)
350 * The history is logged asynchronously, so when they request
351 * the first chunk of history, make sure everything has been
352 * synced to disk so that we get it.
354 if (*offp == 0 && spa_writeable(spa))
355 txg_wait_synced(spa_get_dsl(spa), 0);
357 if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
363 dmu_object_info_t doi;
364 dmu_object_info_from_db(dbp, &doi);
365 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
369 mutex_enter(&spa->spa_history_lock);
370 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
372 if (*offp < shpp->sh_pool_create_len) {
373 /* read in just the zpool create history */
374 phys_read_off = *offp;
375 read_len = MIN(*len, shpp->sh_pool_create_len -
379 * Need to reset passed in offset to BOF if the passed in
380 * offset has since been overwritten.
382 *offp = MAX(*offp, shpp->sh_bof);
383 phys_read_off = spa_history_log_to_phys(*offp, shpp);
386 * Read up to the minimum of what the user passed down or
387 * the EOF (physical or logical). If we hit physical EOF,
388 * use 'leftover' to read from the physical BOF.
390 if (phys_read_off <= phys_eof) {
391 read_len = MIN(*len, phys_eof - phys_read_off);
394 shpp->sh_phys_max_off - phys_read_off);
395 if (phys_read_off + *len > shpp->sh_phys_max_off) {
396 leftover = MIN(*len - read_len,
397 phys_eof - shpp->sh_pool_create_len);
402 /* offset for consumer to use next */
403 *offp += read_len + leftover;
405 /* tell the consumer how much you actually read */
406 *len = read_len + leftover;
409 mutex_exit(&spa->spa_history_lock);
410 dmu_buf_rele(dbp, FTAG);
414 err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
416 if (leftover && err == 0) {
417 err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
418 leftover, buf + read_len, DMU_READ_PREFETCH);
420 mutex_exit(&spa->spa_history_lock);
422 dmu_buf_rele(dbp, FTAG);
427 log_internal(history_internal_events_t event, spa_t *spa,
428 dmu_tx_t *tx, const char *fmt, va_list adx)
433 * If this is part of creating a pool, not everything is
434 * initialized yet, so don't bother logging the internal events.
436 if (tx->tx_txg == TXG_INITIAL)
439 ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
440 ha->ha_history_str = kmem_alloc(vsnprintf(NULL, 0, fmt, adx) + 1,
443 (void) vsprintf(ha->ha_history_str, fmt, adx);
445 ha->ha_log_type = LOG_INTERNAL;
446 ha->ha_event = event;
450 if (dmu_tx_is_syncing(tx)) {
451 spa_history_log_sync(spa, ha, tx);
453 dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
454 spa_history_log_sync, spa, ha, 0, tx);
456 /* spa_history_log_sync() will free ha and strings */
460 spa_history_log_internal(history_internal_events_t event, spa_t *spa,
461 dmu_tx_t *tx, const char *fmt, ...)
466 /* create a tx if we didn't get one */
468 htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
469 if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
476 log_internal(event, spa, htx, fmt, adx);
479 /* if we didn't get a tx from the caller, commit the one we made */
485 spa_history_log_version(spa_t *spa, history_internal_events_t event)
488 uint64_t current_vers = spa_version(spa);
490 if (current_vers >= SPA_VERSION_ZPOOL_HISTORY) {
491 spa_history_log_internal(event, spa, NULL,
492 "pool spa %llu; zfs spa %llu; zpl %d; uts %s %s %s %s",
493 (u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
494 utsname.nodename, utsname.release, utsname.version,
497 cmn_err(CE_CONT, "!%s version %llu pool %s using %llu",
498 event == LOG_POOL_IMPORT ? "imported" :
499 event == LOG_POOL_CREATE ? "created" : "accessed",
500 (u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);