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  */
27 
28 #include <sys/spa.h>
29 #include <sys/spa_impl.h>
30 #include <sys/zap.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>
38 #include <sys/cred.h>
39 #include "zfs_comutil.h"
40 #ifdef _KERNEL
41 #include <sys/cmn_err.h>
42 #include <sys/zone.h>
43 #endif
44 
45 /*
46  * Routines to manage the on-disk history log.
47  *
48  * The history log is stored as a dmu object containing
49  * <packed record length, record nvlist> tuples.
50  *
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.
54  *
55  * The log is implemented as a ring buffer, though the original creation
56  * of the pool ('zpool create') is never overwritten.
57  *
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.
69  *
70  * 'sh_records_lost' keeps track of how many records have been overwritten
71  * and permanently lost.
72  */
73 
74 /* convert a logical offset to physical */
75 static uint64_t
spa_history_log_to_phys(uint64_t log_off,spa_history_phys_t * shpp)76 spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
77 {
78 	uint64_t phys_len;
79 
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);
83 }
84 
85 void
spa_history_create_obj(spa_t * spa,dmu_tx_t * tx)86 spa_history_create_obj(spa_t *spa, dmu_tx_t *tx)
87 {
88 	dmu_buf_t *dbp;
89 	spa_history_phys_t *shpp;
90 	objset_t *mos = spa->spa_meta_objset;
91 
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);
96 
97 	VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
98 	    DMU_POOL_HISTORY, sizeof (uint64_t), 1,
99 	    &spa->spa_history, tx) == 0);
100 
101 	VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
102 	ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));
103 
104 	shpp = dbp->db_data;
105 	dmu_buf_will_dirty(dbp, tx);
106 
107 	/*
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.
110 	 */
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);
115 
116 	dmu_buf_rele(dbp, FTAG);
117 }
118 
119 /*
120  * Change 'sh_bof' to the beginning of the next record.
121  */
122 static int
spa_history_advance_bof(spa_t * spa,spa_history_phys_t * shpp)123 spa_history_advance_bof(spa_t *spa, spa_history_phys_t *shpp)
124 {
125 	objset_t *mos = spa->spa_meta_objset;
126 	uint64_t firstread, reclen, phys_bof;
127 	char buf[sizeof (reclen)];
128 	int err;
129 
130 	phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
131 	firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);
132 
133 	if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
134 	    buf, DMU_READ_PREFETCH)) != 0)
135 		return (err);
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)
140 			return (err);
141 	}
142 
143 	reclen = LE_64(*((uint64_t *)buf));
144 	shpp->sh_bof += reclen + sizeof (reclen);
145 	shpp->sh_records_lost++;
146 	return (0);
147 }
148 
149 static int
spa_history_write(spa_t * spa,void * buf,uint64_t len,spa_history_phys_t * shpp,dmu_tx_t * tx)150 spa_history_write(spa_t *spa, void *buf, uint64_t len, spa_history_phys_t *shpp,
151     dmu_tx_t *tx)
152 {
153 	uint64_t firstwrite, phys_eof;
154 	objset_t *mos = spa->spa_meta_objset;
155 	int err;
156 
157 	ASSERT(MUTEX_HELD(&spa->spa_history_lock));
158 
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) {
163 			return (err);
164 		}
165 	}
166 
167 	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
168 	firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
169 	shpp->sh_eof += len;
170 	dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);
171 
172 	len -= firstwrite;
173 	if (len > 0) {
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);
177 	}
178 
179 	return (0);
180 }
181 
182 static char *
spa_history_zone(void)183 spa_history_zone(void)
184 {
185 #ifdef _KERNEL
186 #ifdef __FreeBSD__
187 	/* XXX: pr_hostname can be changed by default from within a jail! */
188 	if (jailed(curthread->td_ucred))
189 		return (curthread->td_ucred->cr_prison->pr_hostname);
190 #endif
191 #endif
192 	return (NULL);
193 }
194 
195 /*
196  * Write out a history event.
197  */
198 /*ARGSUSED*/
199 static void
spa_history_log_sync(void * arg,dmu_tx_t * tx)200 spa_history_log_sync(void *arg, dmu_tx_t *tx)
201 {
202 	nvlist_t	*nvl = arg;
203 	spa_t		*spa = dmu_tx_pool(tx)->dp_spa;
204 	objset_t	*mos = spa->spa_meta_objset;
205 	dmu_buf_t	*dbp;
206 	spa_history_phys_t *shpp;
207 	size_t		reclen;
208 	uint64_t	le_len;
209 	char		*record_packed = NULL;
210 	int		ret;
211 
212 	/*
213 	 * If we have an older pool that doesn't have a command
214 	 * history object, create it now.
215 	 */
216 	mutex_enter(&spa->spa_history_lock);
217 	if (!spa->spa_history)
218 		spa_history_create_obj(spa, tx);
219 	mutex_exit(&spa->spa_history_lock);
220 
221 	/*
222 	 * Get the offset of where we need to write via the bonus buffer.
223 	 * Update the offset when the write completes.
224 	 */
225 	VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
226 	shpp = dbp->db_data;
227 
228 	dmu_buf_will_dirty(dbp, tx);
229 
230 #ifdef ZFS_DEBUG
231 	{
232 		dmu_object_info_t doi;
233 		dmu_object_info_from_db(dbp, &doi);
234 		ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
235 	}
236 #endif
237 
238 	fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
239 #ifdef _KERNEL
240 	fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname.nodename);
241 #endif
242 	if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
243 		zfs_dbgmsg("command: %s",
244 		    fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
245 	} else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
246 		if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
247 			zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
248 			    fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
249 			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
250 			    fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
251 			    fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
252 			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
253 		} else {
254 			zfs_dbgmsg("txg %lld %s %s",
255 			    fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
256 			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
257 			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
258 		}
259 	} else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
260 		zfs_dbgmsg("ioctl %s",
261 		    fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
262 	}
263 
264 	record_packed = fnvlist_pack(nvl, &reclen);
265 
266 	mutex_enter(&spa->spa_history_lock);
267 
268 	/* write out the packed length as little endian */
269 	le_len = LE_64((uint64_t)reclen);
270 	ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
271 	if (!ret)
272 		ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
273 
274 	/* The first command is the create, which we keep forever */
275 	if (ret == 0 && shpp->sh_pool_create_len == 0 &&
276 	    nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
277 		shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
278 	}
279 
280 	mutex_exit(&spa->spa_history_lock);
281 	fnvlist_pack_free(record_packed, reclen);
282 	dmu_buf_rele(dbp, FTAG);
283 	fnvlist_free(nvl);
284 }
285 
286 /*
287  * Write out a history event.
288  */
289 int
spa_history_log(spa_t * spa,const char * msg)290 spa_history_log(spa_t *spa, const char *msg)
291 {
292 	int err;
293 	nvlist_t *nvl = fnvlist_alloc();
294 
295 	fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg);
296 	err = spa_history_log_nvl(spa, nvl);
297 	fnvlist_free(nvl);
298 	return (err);
299 }
300 
301 int
spa_history_log_nvl(spa_t * spa,nvlist_t * nvl)302 spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
303 {
304 	int err = 0;
305 	dmu_tx_t *tx;
306 	nvlist_t *nvarg;
307 
308 	if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY)
309 		return (EINVAL);
310 
311 	if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY || !spa_writeable(spa))
312 		return (SET_ERROR(EINVAL));
313 
314 	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
315 	err = dmu_tx_assign(tx, TXG_WAIT);
316 	if (err) {
317 		dmu_tx_abort(tx);
318 		return (err);
319 	}
320 
321 	nvarg = fnvlist_dup(nvl);
322 	if (spa_history_zone() != NULL) {
323 		fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
324 		    spa_history_zone());
325 	}
326 	fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
327 
328 	/* Kick this off asynchronously; errors are ignored. */
329 	dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
330 	    nvarg, 0, ZFS_SPACE_CHECK_NONE, tx);
331 	dmu_tx_commit(tx);
332 
333 	/* spa_history_log_sync will free nvl */
334 	return (err);
335 
336 }
337 
338 /*
339  * Read out the command history.
340  */
341 int
spa_history_get(spa_t * spa,uint64_t * offp,uint64_t * len,char * buf)342 spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
343 {
344 	objset_t *mos = spa->spa_meta_objset;
345 	dmu_buf_t *dbp;
346 	uint64_t read_len, phys_read_off, phys_eof;
347 	uint64_t leftover = 0;
348 	spa_history_phys_t *shpp;
349 	int err;
350 
351 	/*
352 	 * If the command history doesn't exist (older pool),
353 	 * that's ok, just return ENOENT.
354 	 */
355 	if (!spa->spa_history)
356 		return (SET_ERROR(ENOENT));
357 
358 	/*
359 	 * The history is logged asynchronously, so when they request
360 	 * the first chunk of history, make sure everything has been
361 	 * synced to disk so that we get it.
362 	 */
363 	if (*offp == 0 && spa_writeable(spa))
364 		txg_wait_synced(spa_get_dsl(spa), 0);
365 
366 	if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
367 		return (err);
368 	shpp = dbp->db_data;
369 
370 #ifdef ZFS_DEBUG
371 	{
372 		dmu_object_info_t doi;
373 		dmu_object_info_from_db(dbp, &doi);
374 		ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
375 	}
376 #endif
377 
378 	mutex_enter(&spa->spa_history_lock);
379 	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
380 
381 	if (*offp < shpp->sh_pool_create_len) {
382 		/* read in just the zpool create history */
383 		phys_read_off = *offp;
384 		read_len = MIN(*len, shpp->sh_pool_create_len -
385 		    phys_read_off);
386 	} else {
387 		/*
388 		 * Need to reset passed in offset to BOF if the passed in
389 		 * offset has since been overwritten.
390 		 */
391 		*offp = MAX(*offp, shpp->sh_bof);
392 		phys_read_off = spa_history_log_to_phys(*offp, shpp);
393 
394 		/*
395 		 * Read up to the minimum of what the user passed down or
396 		 * the EOF (physical or logical).  If we hit physical EOF,
397 		 * use 'leftover' to read from the physical BOF.
398 		 */
399 		if (phys_read_off <= phys_eof) {
400 			read_len = MIN(*len, phys_eof - phys_read_off);
401 		} else {
402 			read_len = MIN(*len,
403 			    shpp->sh_phys_max_off - phys_read_off);
404 			if (phys_read_off + *len > shpp->sh_phys_max_off) {
405 				leftover = MIN(*len - read_len,
406 				    phys_eof - shpp->sh_pool_create_len);
407 			}
408 		}
409 	}
410 
411 	/* offset for consumer to use next */
412 	*offp += read_len + leftover;
413 
414 	/* tell the consumer how much you actually read */
415 	*len = read_len + leftover;
416 
417 	if (read_len == 0) {
418 		mutex_exit(&spa->spa_history_lock);
419 		dmu_buf_rele(dbp, FTAG);
420 		return (0);
421 	}
422 
423 	err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
424 	    DMU_READ_PREFETCH);
425 	if (leftover && err == 0) {
426 		err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
427 		    leftover, buf + read_len, DMU_READ_PREFETCH);
428 	}
429 	mutex_exit(&spa->spa_history_lock);
430 
431 	dmu_buf_rele(dbp, FTAG);
432 	return (err);
433 }
434 
435 /*
436  * The nvlist will be consumed by this call.
437  */
438 static void
log_internal(nvlist_t * nvl,const char * operation,spa_t * spa,dmu_tx_t * tx,const char * fmt,va_list adx)439 log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
440     dmu_tx_t *tx, const char *fmt, va_list adx)
441 {
442 	char *msg;
443 	va_list adx2;
444 
445 	/*
446 	 * If this is part of creating a pool, not everything is
447 	 * initialized yet, so don't bother logging the internal events.
448 	 * Likewise if the pool is not writeable.
449 	 */
450 	if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
451 		fnvlist_free(nvl);
452 		return;
453 	}
454 
455 	va_copy(adx2, adx);
456 
457 	msg = kmem_alloc(vsnprintf(NULL, 0, fmt, adx) + 1, KM_SLEEP);
458 	(void) vsprintf(msg, fmt, adx2);
459 	fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
460 	strfree(msg);
461 
462 	va_end(adx2);
463 
464 	fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
465 	fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
466 
467 	if (dmu_tx_is_syncing(tx)) {
468 		spa_history_log_sync(nvl, tx);
469 	} else {
470 		dsl_sync_task_nowait(spa_get_dsl(spa),
471 		    spa_history_log_sync, nvl, 0, ZFS_SPACE_CHECK_NONE, tx);
472 	}
473 	/* spa_history_log_sync() will free nvl */
474 }
475 
476 void
spa_history_log_internal(spa_t * spa,const char * operation,dmu_tx_t * tx,const char * fmt,...)477 spa_history_log_internal(spa_t *spa, const char *operation,
478     dmu_tx_t *tx, const char *fmt, ...)
479 {
480 	dmu_tx_t *htx = tx;
481 	va_list adx;
482 
483 	/* create a tx if we didn't get one */
484 	if (tx == NULL) {
485 		htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
486 		if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
487 			dmu_tx_abort(htx);
488 			return;
489 		}
490 	}
491 
492 	va_start(adx, fmt);
493 	log_internal(fnvlist_alloc(), operation, spa, htx, fmt, adx);
494 	va_end(adx);
495 
496 	/* if we didn't get a tx from the caller, commit the one we made */
497 	if (tx == NULL)
498 		dmu_tx_commit(htx);
499 }
500 
501 void
spa_history_log_internal_ds(dsl_dataset_t * ds,const char * operation,dmu_tx_t * tx,const char * fmt,...)502 spa_history_log_internal_ds(dsl_dataset_t *ds, const char *operation,
503     dmu_tx_t *tx, const char *fmt, ...)
504 {
505 	va_list adx;
506 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
507 	nvlist_t *nvl = fnvlist_alloc();
508 
509 	ASSERT(tx != NULL);
510 
511 	dsl_dataset_name(ds, namebuf);
512 	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
513 	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID, ds->ds_object);
514 
515 	va_start(adx, fmt);
516 	log_internal(nvl, operation, dsl_dataset_get_spa(ds), tx, fmt, adx);
517 	va_end(adx);
518 }
519 
520 void
spa_history_log_internal_dd(dsl_dir_t * dd,const char * operation,dmu_tx_t * tx,const char * fmt,...)521 spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
522     dmu_tx_t *tx, const char *fmt, ...)
523 {
524 	va_list adx;
525 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
526 	nvlist_t *nvl = fnvlist_alloc();
527 
528 	ASSERT(tx != NULL);
529 
530 	dsl_dir_name(dd, namebuf);
531 	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
532 	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
533 	    dsl_dir_phys(dd)->dd_head_dataset_obj);
534 
535 	va_start(adx, fmt);
536 	log_internal(nvl, operation, dd->dd_pool->dp_spa, tx, fmt, adx);
537 	va_end(adx);
538 }
539 
540 void
spa_history_log_version(spa_t * spa,const char * operation)541 spa_history_log_version(spa_t *spa, const char *operation)
542 {
543 #ifdef __NetBSD__
544 	spa_history_log_internal(spa, operation, NULL,
545 	    "pool version %llu; software version %llu/%d; %s %s %s %s %s",
546 	    (u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
547 	    utsname.sysname, utsname.nodename, utsname.release, utsname.version,
548 	    utsname.machine);
549 #else
550 	spa_history_log_internal(spa, operation, NULL,
551 	    "pool version %llu; software version %llu/%d; uts %s %s %s %s",
552 	    (u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
553 	    utsname.nodename, utsname.release, utsname.version,
554 	    utsname.machine);
555 #endif
556 }
557