xref: /illumos-gate/usr/src/uts/common/fs/zfs/spa_history.c (revision bbf215553c7233fbab8a0afdf1fac74c44781867)
1 /*
2  * CDDL HEADER START
3  *
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.
7  *
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.
12  *
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]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
25  * Copyright (c) 2014 Integros [integros.com]
26  * Copyright 2017 Joyent, Inc.
27  */
28 
29 #include <sys/spa.h>
30 #include <sys/spa_impl.h>
31 #include <sys/zap.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/utsname.h>
38 #include <sys/cmn_err.h>
39 #include <sys/sunddi.h>
40 #include <sys/cred.h>
41 #include "zfs_comutil.h"
42 #ifdef _KERNEL
43 #include <sys/zone.h>
44 #endif
45 
46 /*
47  * Routines to manage the on-disk history log.
48  *
49  * The history log is stored as a dmu object containing
50  * <packed record length, record nvlist> tuples.
51  *
52  * Where "record nvlist" is a 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.
55  *
56  * The log is implemented as a ring buffer, though the original creation
57  * of the pool ('zpool create') is never overwritten.
58  *
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 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.
70  *
71  * 'sh_records_lost' keeps track of how many records have been overwritten
72  * and permanently lost.
73  */
74 
75 /* convert a logical offset to physical */
76 static uint64_t
spa_history_log_to_phys(uint64_t log_off,spa_history_phys_t * shpp)77 spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
78 {
79 	uint64_t phys_len;
80 
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);
84 }
85 
86 void
spa_history_create_obj(spa_t * spa,dmu_tx_t * tx)87 spa_history_create_obj(spa_t *spa, dmu_tx_t *tx)
88 {
89 	dmu_buf_t *dbp;
90 	spa_history_phys_t *shpp;
91 	objset_t *mos = spa->spa_meta_objset;
92 
93 	ASSERT(spa->spa_history == 0);
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);
97 
98 	VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
99 	    DMU_POOL_HISTORY, sizeof (uint64_t), 1,
100 	    &spa->spa_history, tx) == 0);
101 
102 	VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
103 	ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));
104 
105 	shpp = dbp->db_data;
106 	dmu_buf_will_dirty(dbp, tx);
107 
108 	/*
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.
111 	 */
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);
116 
117 	dmu_buf_rele(dbp, FTAG);
118 }
119 
120 /*
121  * Change 'sh_bof' to the beginning of the next record.
122  */
123 static int
spa_history_advance_bof(spa_t * spa,spa_history_phys_t * shpp)124 spa_history_advance_bof(spa_t *spa, spa_history_phys_t *shpp)
125 {
126 	objset_t *mos = spa->spa_meta_objset;
127 	uint64_t firstread, reclen, phys_bof;
128 	char buf[sizeof (reclen)];
129 	int err;
130 
131 	phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
132 	firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);
133 
134 	if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
135 	    buf, DMU_READ_PREFETCH)) != 0)
136 		return (err);
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)
141 			return (err);
142 	}
143 
144 	reclen = LE_64(*((uint64_t *)buf));
145 	shpp->sh_bof += reclen + sizeof (reclen);
146 	shpp->sh_records_lost++;
147 	return (0);
148 }
149 
150 static int
spa_history_write(spa_t * spa,void * buf,uint64_t len,spa_history_phys_t * shpp,dmu_tx_t * tx)151 spa_history_write(spa_t *spa, void *buf, uint64_t len, spa_history_phys_t *shpp,
152     dmu_tx_t *tx)
153 {
154 	uint64_t firstwrite, phys_eof;
155 	objset_t *mos = spa->spa_meta_objset;
156 	int err;
157 
158 	ASSERT(MUTEX_HELD(&spa->spa_history_lock));
159 
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) {
164 			return (err);
165 		}
166 	}
167 
168 	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
169 	firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
170 	shpp->sh_eof += len;
171 	dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);
172 
173 	len -= firstwrite;
174 	if (len > 0) {
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);
178 	}
179 
180 	return (0);
181 }
182 
183 static char *
spa_history_zone(void)184 spa_history_zone(void)
185 {
186 #ifdef _KERNEL
187 	if (INGLOBALZONE(curproc))
188 		return (NULL);
189 	return (curproc->p_zone->zone_name);
190 #else
191 	return (NULL);
192 #endif
193 }
194 
195 /*
196  * Post a history sysevent.
197  *
198  * The nvlist_t* passed into this function will be transformed into a new
199  * nvlist where:
200  *
201  * 1. Nested nvlists will be flattened to a single level
202  * 2. Keys will have their names normalized (to remove any problematic
203  * characters, such as whitespace)
204  *
205  * The nvlist_t passed into this function will duplicated and should be freed
206  * by caller.
207  *
208  */
209 static void
spa_history_log_notify(spa_t * spa,nvlist_t * nvl)210 spa_history_log_notify(spa_t *spa, nvlist_t *nvl)
211 {
212 	nvlist_t *hist_nvl = fnvlist_alloc();
213 	uint64_t uint64;
214 	char *string;
215 
216 	if (nvlist_lookup_string(nvl, ZPOOL_HIST_CMD, &string) == 0)
217 		fnvlist_add_string(hist_nvl, ZFS_EV_HIST_CMD, string);
218 
219 	if (nvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME, &string) == 0)
220 		fnvlist_add_string(hist_nvl, ZFS_EV_HIST_INT_NAME, string);
221 
222 	if (nvlist_lookup_string(nvl, ZPOOL_HIST_ZONE, &string) == 0)
223 		fnvlist_add_string(hist_nvl, ZFS_EV_HIST_ZONE, string);
224 
225 	if (nvlist_lookup_string(nvl, ZPOOL_HIST_HOST, &string) == 0)
226 		fnvlist_add_string(hist_nvl, ZFS_EV_HIST_HOST, string);
227 
228 	if (nvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME, &string) == 0)
229 		fnvlist_add_string(hist_nvl, ZFS_EV_HIST_DSNAME, string);
230 
231 	if (nvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR, &string) == 0)
232 		fnvlist_add_string(hist_nvl, ZFS_EV_HIST_INT_STR, string);
233 
234 	if (nvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL, &string) == 0)
235 		fnvlist_add_string(hist_nvl, ZFS_EV_HIST_IOCTL, string);
236 
237 	if (nvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME, &string) == 0)
238 		fnvlist_add_string(hist_nvl, ZFS_EV_HIST_INT_NAME, string);
239 
240 	if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID, &uint64) == 0)
241 		fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_DSID, uint64);
242 
243 	if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG, &uint64) == 0)
244 		fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_TXG, uint64);
245 
246 	if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_TIME, &uint64) == 0)
247 		fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_TIME, uint64);
248 
249 	if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_WHO, &uint64) == 0)
250 		fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_WHO, uint64);
251 
252 	if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_INT_EVENT, &uint64) == 0)
253 		fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_INT_EVENT, uint64);
254 
255 	spa_event_notify(spa, NULL, hist_nvl, ESC_ZFS_HISTORY_EVENT);
256 
257 	nvlist_free(hist_nvl);
258 }
259 
260 /*
261  * Write out a history event.
262  */
263 /*ARGSUSED*/
264 static void
spa_history_log_sync(void * arg,dmu_tx_t * tx)265 spa_history_log_sync(void *arg, dmu_tx_t *tx)
266 {
267 	nvlist_t	*nvl = arg;
268 	spa_t		*spa = dmu_tx_pool(tx)->dp_spa;
269 	objset_t	*mos = spa->spa_meta_objset;
270 	dmu_buf_t	*dbp;
271 	spa_history_phys_t *shpp;
272 	size_t		reclen;
273 	uint64_t	le_len;
274 	char		*record_packed = NULL;
275 	int		ret;
276 
277 	/*
278 	 * If we have an older pool that doesn't have a command
279 	 * history object, create it now.
280 	 */
281 	mutex_enter(&spa->spa_history_lock);
282 	if (!spa->spa_history)
283 		spa_history_create_obj(spa, tx);
284 	mutex_exit(&spa->spa_history_lock);
285 
286 	/*
287 	 * Get the offset of where we need to write via the bonus buffer.
288 	 * Update the offset when the write completes.
289 	 */
290 	VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
291 	shpp = dbp->db_data;
292 
293 	dmu_buf_will_dirty(dbp, tx);
294 
295 #ifdef ZFS_DEBUG
296 	{
297 		dmu_object_info_t doi;
298 		dmu_object_info_from_db(dbp, &doi);
299 		ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
300 	}
301 #endif
302 
303 	fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
304 #ifdef _KERNEL
305 	fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname.nodename);
306 #endif
307 	if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
308 		zfs_dbgmsg("command: %s",
309 		    fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
310 	} else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
311 		if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
312 			zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
313 			    fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
314 			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
315 			    fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
316 			    fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
317 			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
318 		} else {
319 			zfs_dbgmsg("txg %lld %s %s",
320 			    fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
321 			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
322 			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
323 		}
324 		/*
325 		 * The history sysevent is posted only for internal history
326 		 * messages to show what has happened, not how it happened. For
327 		 * example, the following command:
328 		 *
329 		 * # zfs destroy -r tank/foo
330 		 *
331 		 * will result in one sysevent posted per dataset that is
332 		 * destroyed as a result of the command - which could be more
333 		 * than one event in total.  By contrast, if the sysevent was
334 		 * posted as a result of the ZPOOL_HIST_CMD key being present
335 		 * it would result in only one sysevent being posted with the
336 		 * full command line arguments, requiring the consumer to know
337 		 * how to parse and understand zfs(8) command invocations.
338 		 */
339 		spa_history_log_notify(spa, nvl);
340 	} else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
341 		zfs_dbgmsg("ioctl %s",
342 		    fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
343 	}
344 
345 	record_packed = fnvlist_pack(nvl, &reclen);
346 
347 	mutex_enter(&spa->spa_history_lock);
348 
349 	/* write out the packed length as little endian */
350 	le_len = LE_64((uint64_t)reclen);
351 	ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
352 	if (!ret)
353 		ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
354 
355 	/* The first command is the create, which we keep forever */
356 	if (ret == 0 && shpp->sh_pool_create_len == 0 &&
357 	    nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
358 		shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
359 	}
360 
361 	mutex_exit(&spa->spa_history_lock);
362 	fnvlist_pack_free(record_packed, reclen);
363 	dmu_buf_rele(dbp, FTAG);
364 	fnvlist_free(nvl);
365 }
366 
367 /*
368  * Write out a history event.
369  */
370 int
spa_history_log(spa_t * spa,const char * msg)371 spa_history_log(spa_t *spa, const char *msg)
372 {
373 	int err;
374 	nvlist_t *nvl = fnvlist_alloc();
375 
376 	fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg);
377 	err = spa_history_log_nvl(spa, nvl);
378 	fnvlist_free(nvl);
379 	return (err);
380 }
381 
382 int
spa_history_log_nvl(spa_t * spa,nvlist_t * nvl)383 spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
384 {
385 	int err = 0;
386 	dmu_tx_t *tx;
387 	nvlist_t *nvarg, *in_nvl = NULL;
388 
389 	if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY || !spa_writeable(spa))
390 		return (SET_ERROR(EINVAL));
391 
392 	err = nvlist_lookup_nvlist(nvl, ZPOOL_HIST_INPUT_NVL, &in_nvl);
393 	if (err == 0) {
394 		(void) nvlist_remove_all(in_nvl, ZPOOL_HIDDEN_ARGS);
395 	}
396 
397 	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
398 	err = dmu_tx_assign(tx, TXG_WAIT);
399 	if (err) {
400 		dmu_tx_abort(tx);
401 		return (err);
402 	}
403 
404 	nvarg = fnvlist_dup(nvl);
405 	if (spa_history_zone() != NULL) {
406 		fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
407 		    spa_history_zone());
408 	}
409 	fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
410 
411 	/* Kick this off asynchronously; errors are ignored. */
412 	dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
413 	    nvarg, 0, ZFS_SPACE_CHECK_NONE, tx);
414 	dmu_tx_commit(tx);
415 
416 	/* spa_history_log_sync will free nvl */
417 	return (err);
418 
419 }
420 
421 /*
422  * Read out the command history.
423  */
424 int
spa_history_get(spa_t * spa,uint64_t * offp,uint64_t * len,char * buf)425 spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
426 {
427 	objset_t *mos = spa->spa_meta_objset;
428 	dmu_buf_t *dbp;
429 	uint64_t read_len, phys_read_off, phys_eof;
430 	uint64_t leftover = 0;
431 	spa_history_phys_t *shpp;
432 	int err;
433 
434 	/*
435 	 * If the command history doesn't exist (older pool),
436 	 * that's ok, just return ENOENT.
437 	 */
438 	if (!spa->spa_history)
439 		return (SET_ERROR(ENOENT));
440 
441 	/*
442 	 * The history is logged asynchronously, so when they request
443 	 * the first chunk of history, make sure everything has been
444 	 * synced to disk so that we get it.
445 	 */
446 	if (*offp == 0 && spa_writeable(spa))
447 		txg_wait_synced(spa_get_dsl(spa), 0);
448 
449 	if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
450 		return (err);
451 	shpp = dbp->db_data;
452 
453 #ifdef ZFS_DEBUG
454 	{
455 		dmu_object_info_t doi;
456 		dmu_object_info_from_db(dbp, &doi);
457 		ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
458 	}
459 #endif
460 
461 	mutex_enter(&spa->spa_history_lock);
462 	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
463 
464 	if (*offp < shpp->sh_pool_create_len) {
465 		/* read in just the zpool create history */
466 		phys_read_off = *offp;
467 		read_len = MIN(*len, shpp->sh_pool_create_len -
468 		    phys_read_off);
469 	} else {
470 		/*
471 		 * Need to reset passed in offset to BOF if the passed in
472 		 * offset has since been overwritten.
473 		 */
474 		*offp = MAX(*offp, shpp->sh_bof);
475 		phys_read_off = spa_history_log_to_phys(*offp, shpp);
476 
477 		/*
478 		 * Read up to the minimum of what the user passed down or
479 		 * the EOF (physical or logical).  If we hit physical EOF,
480 		 * use 'leftover' to read from the physical BOF.
481 		 */
482 		if (phys_read_off <= phys_eof) {
483 			read_len = MIN(*len, phys_eof - phys_read_off);
484 		} else {
485 			read_len = MIN(*len,
486 			    shpp->sh_phys_max_off - phys_read_off);
487 			if (phys_read_off + *len > shpp->sh_phys_max_off) {
488 				leftover = MIN(*len - read_len,
489 				    phys_eof - shpp->sh_pool_create_len);
490 			}
491 		}
492 	}
493 
494 	/* offset for consumer to use next */
495 	*offp += read_len + leftover;
496 
497 	/* tell the consumer how much you actually read */
498 	*len = read_len + leftover;
499 
500 	if (read_len == 0) {
501 		mutex_exit(&spa->spa_history_lock);
502 		dmu_buf_rele(dbp, FTAG);
503 		return (0);
504 	}
505 
506 	err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
507 	    DMU_READ_PREFETCH);
508 	if (leftover && err == 0) {
509 		err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
510 		    leftover, buf + read_len, DMU_READ_PREFETCH);
511 	}
512 	mutex_exit(&spa->spa_history_lock);
513 
514 	dmu_buf_rele(dbp, FTAG);
515 	return (err);
516 }
517 
518 /*
519  * The nvlist will be consumed by this call.
520  */
521 static void
log_internal(nvlist_t * nvl,const char * operation,spa_t * spa,dmu_tx_t * tx,const char * fmt,va_list adx)522 log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
523     dmu_tx_t *tx, const char *fmt, va_list adx)
524 {
525 	char *msg;
526 
527 	/*
528 	 * If this is part of creating a pool, not everything is
529 	 * initialized yet, so don't bother logging the internal events.
530 	 * Likewise if the pool is not writeable.
531 	 */
532 	if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
533 		fnvlist_free(nvl);
534 		return;
535 	}
536 
537 	msg = kmem_alloc(vsnprintf(NULL, 0, fmt, adx) + 1, KM_SLEEP);
538 	(void) vsprintf(msg, fmt, adx);
539 	fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
540 	strfree(msg);
541 
542 	fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
543 	fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
544 
545 	if (dmu_tx_is_syncing(tx)) {
546 		spa_history_log_sync(nvl, tx);
547 	} else {
548 		dsl_sync_task_nowait(spa_get_dsl(spa),
549 		    spa_history_log_sync, nvl, 0, ZFS_SPACE_CHECK_NONE, tx);
550 	}
551 	/* spa_history_log_sync() will free nvl */
552 }
553 
554 void
spa_history_log_internal(spa_t * spa,const char * operation,dmu_tx_t * tx,const char * fmt,...)555 spa_history_log_internal(spa_t *spa, const char *operation,
556     dmu_tx_t *tx, const char *fmt, ...)
557 {
558 	dmu_tx_t *htx = tx;
559 	va_list adx;
560 
561 	/* create a tx if we didn't get one */
562 	if (tx == NULL) {
563 		htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
564 		if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
565 			dmu_tx_abort(htx);
566 			return;
567 		}
568 	}
569 
570 	va_start(adx, fmt);
571 	log_internal(fnvlist_alloc(), operation, spa, htx, fmt, adx);
572 	va_end(adx);
573 
574 	/* if we didn't get a tx from the caller, commit the one we made */
575 	if (tx == NULL)
576 		dmu_tx_commit(htx);
577 }
578 
579 void
spa_history_log_internal_ds(dsl_dataset_t * ds,const char * operation,dmu_tx_t * tx,const char * fmt,...)580 spa_history_log_internal_ds(dsl_dataset_t *ds, const char *operation,
581     dmu_tx_t *tx, const char *fmt, ...)
582 {
583 	va_list adx;
584 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
585 	nvlist_t *nvl = fnvlist_alloc();
586 
587 	ASSERT(tx != NULL);
588 
589 	dsl_dataset_name(ds, namebuf);
590 	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
591 	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID, ds->ds_object);
592 
593 	va_start(adx, fmt);
594 	log_internal(nvl, operation, dsl_dataset_get_spa(ds), tx, fmt, adx);
595 	va_end(adx);
596 }
597 
598 void
spa_history_log_internal_dd(dsl_dir_t * dd,const char * operation,dmu_tx_t * tx,const char * fmt,...)599 spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
600     dmu_tx_t *tx, const char *fmt, ...)
601 {
602 	va_list adx;
603 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
604 	nvlist_t *nvl = fnvlist_alloc();
605 
606 	ASSERT(tx != NULL);
607 
608 	dsl_dir_name(dd, namebuf);
609 	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
610 	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
611 	    dsl_dir_phys(dd)->dd_head_dataset_obj);
612 
613 	va_start(adx, fmt);
614 	log_internal(nvl, operation, dd->dd_pool->dp_spa, tx, fmt, adx);
615 	va_end(adx);
616 }
617 
618 void
spa_history_log_version(spa_t * spa,const char * operation)619 spa_history_log_version(spa_t *spa, const char *operation)
620 {
621 	spa_history_log_internal(spa, operation, NULL,
622 	    "pool version %llu; software version %llu/%llu; uts %s %s %s %s",
623 	    (u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
624 	    utsname.nodename, utsname.release, utsname.version,
625 	    utsname.machine);
626 }
627