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) 2013 by Delphix. All rights reserved.
28 #include <sys/spa_impl.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dsl_dataset.h>
34 #include <sys/dsl_dir.h>
35 #include <sys/cmn_err.h>
36 #include <sys/sunddi.h>
38 #include "zfs_comutil.h"
44 * Routines to manage the on-disk history log.
46 * The history log is stored as a dmu object containing
47 * <packed record length, record nvlist> tuples.
49 * Where "record nvlist" is a nvlist containing uint64_ts and strings, and
50 * "packed record length" is the packed length of the "record nvlist" stored
51 * as a little endian uint64_t.
53 * The log is implemented as a ring buffer, though the original creation
54 * of the pool ('zpool create') is never overwritten.
56 * The history log is tracked as object 'spa_t::spa_history'. The bonus buffer
57 * of 'spa_history' stores the offsets for logging/retrieving history as
58 * 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of
59 * where the 'zpool create' record is stored. This allows us to never
60 * overwrite the original creation of the pool. 'sh_phys_max_off' is the
61 * physical ending offset in bytes of the log. This tells you the length of
62 * the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record
63 * is added, 'sh_eof' is incremented by the the size of the record.
64 * 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes).
65 * This is where the consumer should start reading from after reading in
66 * the 'zpool create' portion of the log.
68 * 'sh_records_lost' keeps track of how many records have been overwritten
69 * and permanently lost.
72 /* convert a logical offset to physical */
74 spa_history_log_to_phys(uint64_t log_off
, spa_history_phys_t
*shpp
)
78 phys_len
= shpp
->sh_phys_max_off
- shpp
->sh_pool_create_len
;
79 return ((log_off
- shpp
->sh_pool_create_len
) % phys_len
80 + shpp
->sh_pool_create_len
);
84 spa_history_create_obj(spa_t
*spa
, dmu_tx_t
*tx
)
87 spa_history_phys_t
*shpp
;
88 objset_t
*mos
= spa
->spa_meta_objset
;
90 ASSERT(spa
->spa_history
== 0);
91 spa
->spa_history
= dmu_object_alloc(mos
, DMU_OT_SPA_HISTORY
,
92 SPA_MAXBLOCKSIZE
, DMU_OT_SPA_HISTORY_OFFSETS
,
93 sizeof (spa_history_phys_t
), tx
);
95 VERIFY(zap_add(mos
, DMU_POOL_DIRECTORY_OBJECT
,
96 DMU_POOL_HISTORY
, sizeof (uint64_t), 1,
97 &spa
->spa_history
, tx
) == 0);
99 VERIFY(0 == dmu_bonus_hold(mos
, spa
->spa_history
, FTAG
, &dbp
));
100 ASSERT(dbp
->db_size
>= sizeof (spa_history_phys_t
));
103 dmu_buf_will_dirty(dbp
, tx
);
106 * Figure out maximum size of history log. We set it at
107 * 0.1% of pool size, with a max of 1G and min of 128KB.
109 shpp
->sh_phys_max_off
=
110 metaslab_class_get_dspace(spa_normal_class(spa
)) / 1000;
111 shpp
->sh_phys_max_off
= MIN(shpp
->sh_phys_max_off
, 1<<30);
112 shpp
->sh_phys_max_off
= MAX(shpp
->sh_phys_max_off
, 128<<10);
114 dmu_buf_rele(dbp
, FTAG
);
118 * Change 'sh_bof' to the beginning of the next record.
121 spa_history_advance_bof(spa_t
*spa
, spa_history_phys_t
*shpp
)
123 objset_t
*mos
= spa
->spa_meta_objset
;
124 uint64_t firstread
, reclen
, phys_bof
;
125 char buf
[sizeof (reclen
)];
128 phys_bof
= spa_history_log_to_phys(shpp
->sh_bof
, shpp
);
129 firstread
= MIN(sizeof (reclen
), shpp
->sh_phys_max_off
- phys_bof
);
131 if ((err
= dmu_read(mos
, spa
->spa_history
, phys_bof
, firstread
,
132 buf
, DMU_READ_PREFETCH
)) != 0)
134 if (firstread
!= sizeof (reclen
)) {
135 if ((err
= dmu_read(mos
, spa
->spa_history
,
136 shpp
->sh_pool_create_len
, sizeof (reclen
) - firstread
,
137 buf
+ firstread
, DMU_READ_PREFETCH
)) != 0)
141 reclen
= LE_64(*((uint64_t *)buf
));
142 shpp
->sh_bof
+= reclen
+ sizeof (reclen
);
143 shpp
->sh_records_lost
++;
148 spa_history_write(spa_t
*spa
, void *buf
, uint64_t len
, spa_history_phys_t
*shpp
,
151 uint64_t firstwrite
, phys_eof
;
152 objset_t
*mos
= spa
->spa_meta_objset
;
155 ASSERT(MUTEX_HELD(&spa
->spa_history_lock
));
157 /* see if we need to reset logical BOF */
158 while (shpp
->sh_phys_max_off
- shpp
->sh_pool_create_len
-
159 (shpp
->sh_eof
- shpp
->sh_bof
) <= len
) {
160 if ((err
= spa_history_advance_bof(spa
, shpp
)) != 0) {
165 phys_eof
= spa_history_log_to_phys(shpp
->sh_eof
, shpp
);
166 firstwrite
= MIN(len
, shpp
->sh_phys_max_off
- phys_eof
);
168 dmu_write(mos
, spa
->spa_history
, phys_eof
, firstwrite
, buf
, tx
);
172 /* write out the rest at the beginning of physical file */
173 dmu_write(mos
, spa
->spa_history
, shpp
->sh_pool_create_len
,
174 len
, (char *)buf
+ firstwrite
, tx
);
181 spa_history_zone(void)
187 return (curproc
->p_zone
->zone_name
);
195 * Write out a history event.
199 spa_history_log_sync(void *arg
, dmu_tx_t
*tx
)
202 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
203 objset_t
*mos
= spa
->spa_meta_objset
;
205 spa_history_phys_t
*shpp
;
208 char *record_packed
= NULL
;
212 * If we have an older pool that doesn't have a command
213 * history object, create it now.
215 mutex_enter(&spa
->spa_history_lock
);
216 if (!spa
->spa_history
)
217 spa_history_create_obj(spa
, tx
);
218 mutex_exit(&spa
->spa_history_lock
);
221 * Get the offset of where we need to write via the bonus buffer.
222 * Update the offset when the write completes.
224 VERIFY0(dmu_bonus_hold(mos
, spa
->spa_history
, FTAG
, &dbp
));
227 dmu_buf_will_dirty(dbp
, tx
);
231 dmu_object_info_t doi
;
232 dmu_object_info_from_db(dbp
, &doi
);
233 ASSERT3U(doi
.doi_bonus_type
, ==, DMU_OT_SPA_HISTORY_OFFSETS
);
237 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 VERIFY3U(nvlist_pack(nvl
, &record_packed
, &reclen
, NV_ENCODE_NATIVE
,
265 mutex_enter(&spa
->spa_history_lock
);
267 /* write out the packed length as little endian */
268 le_len
= LE_64((uint64_t)reclen
);
269 ret
= spa_history_write(spa
, &le_len
, sizeof (le_len
), shpp
, tx
);
271 ret
= spa_history_write(spa
, record_packed
, reclen
, shpp
, tx
);
273 /* The first command is the create, which we keep forever */
274 if (ret
== 0 && shpp
->sh_pool_create_len
== 0 &&
275 nvlist_exists(nvl
, ZPOOL_HIST_CMD
)) {
276 shpp
->sh_pool_create_len
= shpp
->sh_bof
= shpp
->sh_eof
;
279 mutex_exit(&spa
->spa_history_lock
);
280 fnvlist_pack_free(record_packed
, reclen
);
281 dmu_buf_rele(dbp
, FTAG
);
286 * Write out a history event.
289 spa_history_log(spa_t
*spa
, const char *msg
)
292 nvlist_t
*nvl
= fnvlist_alloc();
294 fnvlist_add_string(nvl
, ZPOOL_HIST_CMD
, msg
);
295 err
= spa_history_log_nvl(spa
, nvl
);
301 spa_history_log_nvl(spa_t
*spa
, nvlist_t
*nvl
)
307 if (spa_version(spa
) < SPA_VERSION_ZPOOL_HISTORY
|| !spa_writeable(spa
))
308 return (SET_ERROR(EINVAL
));
310 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
311 err
= dmu_tx_assign(tx
, TXG_WAIT
);
317 VERIFY0(nvlist_dup(nvl
, &nvarg
, KM_SLEEP
));
318 if (spa_history_zone() != NULL
) {
319 fnvlist_add_string(nvarg
, ZPOOL_HIST_ZONE
,
322 fnvlist_add_uint64(nvarg
, ZPOOL_HIST_WHO
, crgetruid(CRED()));
324 /* Kick this off asynchronously; errors are ignored. */
325 dsl_sync_task_nowait(spa_get_dsl(spa
), spa_history_log_sync
,
329 /* spa_history_log_sync will free nvl */
335 * Read out the command history.
338 spa_history_get(spa_t
*spa
, uint64_t *offp
, uint64_t *len
, char *buf
)
340 objset_t
*mos
= spa
->spa_meta_objset
;
342 uint64_t read_len
, phys_read_off
, phys_eof
;
343 uint64_t leftover
= 0;
344 spa_history_phys_t
*shpp
;
348 * If the command history doesn't exist (older pool),
349 * that's ok, just return ENOENT.
351 if (!spa
->spa_history
)
352 return (SET_ERROR(ENOENT
));
355 * The history is logged asynchronously, so when they request
356 * the first chunk of history, make sure everything has been
357 * synced to disk so that we get it.
359 if (*offp
== 0 && spa_writeable(spa
))
360 txg_wait_synced(spa_get_dsl(spa
), 0);
362 if ((err
= dmu_bonus_hold(mos
, spa
->spa_history
, FTAG
, &dbp
)) != 0)
368 dmu_object_info_t doi
;
369 dmu_object_info_from_db(dbp
, &doi
);
370 ASSERT3U(doi
.doi_bonus_type
, ==, DMU_OT_SPA_HISTORY_OFFSETS
);
374 mutex_enter(&spa
->spa_history_lock
);
375 phys_eof
= spa_history_log_to_phys(shpp
->sh_eof
, shpp
);
377 if (*offp
< shpp
->sh_pool_create_len
) {
378 /* read in just the zpool create history */
379 phys_read_off
= *offp
;
380 read_len
= MIN(*len
, shpp
->sh_pool_create_len
-
384 * Need to reset passed in offset to BOF if the passed in
385 * offset has since been overwritten.
387 *offp
= MAX(*offp
, shpp
->sh_bof
);
388 phys_read_off
= spa_history_log_to_phys(*offp
, shpp
);
391 * Read up to the minimum of what the user passed down or
392 * the EOF (physical or logical). If we hit physical EOF,
393 * use 'leftover' to read from the physical BOF.
395 if (phys_read_off
<= phys_eof
) {
396 read_len
= MIN(*len
, phys_eof
- phys_read_off
);
399 shpp
->sh_phys_max_off
- phys_read_off
);
400 if (phys_read_off
+ *len
> shpp
->sh_phys_max_off
) {
401 leftover
= MIN(*len
- read_len
,
402 phys_eof
- shpp
->sh_pool_create_len
);
407 /* offset for consumer to use next */
408 *offp
+= read_len
+ leftover
;
410 /* tell the consumer how much you actually read */
411 *len
= read_len
+ leftover
;
414 mutex_exit(&spa
->spa_history_lock
);
415 dmu_buf_rele(dbp
, FTAG
);
419 err
= dmu_read(mos
, spa
->spa_history
, phys_read_off
, read_len
, buf
,
421 if (leftover
&& err
== 0) {
422 err
= dmu_read(mos
, spa
->spa_history
, shpp
->sh_pool_create_len
,
423 leftover
, buf
+ read_len
, DMU_READ_PREFETCH
);
425 mutex_exit(&spa
->spa_history_lock
);
427 dmu_buf_rele(dbp
, FTAG
);
432 * The nvlist will be consumed by this call.
435 log_internal(nvlist_t
*nvl
, const char *operation
, spa_t
*spa
,
436 dmu_tx_t
*tx
, const char *fmt
, va_list adx
)
443 * If this is part of creating a pool, not everything is
444 * initialized yet, so don't bother logging the internal events.
445 * Likewise if the pool is not writeable.
447 if (tx
->tx_txg
== TXG_INITIAL
|| !spa_writeable(spa
)) {
453 size
= vsnprintf(NULL
, 0, fmt
, adx1
) + 1;
454 msg
= kmem_alloc(size
, KM_SLEEP
);
457 (void) vsprintf(msg
, fmt
, adx1
);
459 fnvlist_add_string(nvl
, ZPOOL_HIST_INT_STR
, msg
);
460 kmem_free(msg
, size
);
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, 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 dd
->dd_phys
->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 utsname_t
*u
= utsname();
543 spa_history_log_internal(spa
, operation
, NULL
,
544 "pool version %llu; software version %llu/%d; uts %s %s %s %s",
545 (u_longlong_t
)spa_version(spa
), SPA_VERSION
, ZPL_VERSION
,
546 u
->nodename
, u
->release
, u
->version
, u
->machine
);
549 #if defined(_KERNEL) && defined(HAVE_SPL)
550 EXPORT_SYMBOL(spa_history_create_obj
);
551 EXPORT_SYMBOL(spa_history_get
);
552 EXPORT_SYMBOL(spa_history_log
);
553 EXPORT_SYMBOL(spa_history_log_internal
);
554 EXPORT_SYMBOL(spa_history_log_version
);