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/cmn_err.h>
38 #include <sys/sunddi.h>
39 #include <sys/cred.h>
40 #include "zfs_comutil.h"
41 #ifdef _KERNEL
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 if (INGLOBALZONE(curproc))
187 return (NULL);
188 return (curproc->p_zone->zone_name);
189 #else
190 return (NULL);
191 #endif
192 }
193
194 /*
195 * Write out a history event.
196 */
197 /*ARGSUSED*/
198 static void
spa_history_log_sync(void * arg,dmu_tx_t * tx)199 spa_history_log_sync(void *arg, dmu_tx_t *tx)
200 {
201 nvlist_t *nvl = arg;
202 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
203 objset_t *mos = spa->spa_meta_objset;
204 dmu_buf_t *dbp;
205 spa_history_phys_t *shpp;
206 size_t reclen;
207 uint64_t le_len;
208 char *record_packed = NULL;
209 int ret;
210
211 /*
212 * If we have an older pool that doesn't have a command
213 * history object, create it now.
214 */
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);
219
220 /*
221 * Get the offset of where we need to write via the bonus buffer.
222 * Update the offset when the write completes.
223 */
224 VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
225 shpp = dbp->db_data;
226
227 dmu_buf_will_dirty(dbp, tx);
228
229 #ifdef ZFS_DEBUG
230 {
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);
234 }
235 #endif
236
237 fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
238 #ifdef _KERNEL
239 fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname.nodename);
240 #endif
241 if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
242 zfs_dbgmsg("command: %s",
243 fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
244 } else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
245 if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
246 zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
247 fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
248 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
249 fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
250 fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
251 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
252 } else {
253 zfs_dbgmsg("txg %lld %s %s",
254 fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
255 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
256 fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
257 }
258 } else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
259 zfs_dbgmsg("ioctl %s",
260 fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
261 }
262
263 record_packed = fnvlist_pack(nvl, &reclen);
264
265 mutex_enter(&spa->spa_history_lock);
266
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);
270 if (!ret)
271 ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
272
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;
277 }
278
279 mutex_exit(&spa->spa_history_lock);
280 fnvlist_pack_free(record_packed, reclen);
281 dmu_buf_rele(dbp, FTAG);
282 fnvlist_free(nvl);
283 }
284
285 /*
286 * Write out a history event.
287 */
288 int
spa_history_log(spa_t * spa,const char * msg)289 spa_history_log(spa_t *spa, const char *msg)
290 {
291 int err;
292 nvlist_t *nvl = fnvlist_alloc();
293
294 fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg);
295 err = spa_history_log_nvl(spa, nvl);
296 fnvlist_free(nvl);
297 return (err);
298 }
299
300 int
spa_history_log_nvl(spa_t * spa,nvlist_t * nvl)301 spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
302 {
303 int err = 0;
304 dmu_tx_t *tx;
305 nvlist_t *nvarg;
306
307 if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY || !spa_writeable(spa))
308 return (SET_ERROR(EINVAL));
309
310 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
311 err = dmu_tx_assign(tx, TXG_WAIT);
312 if (err) {
313 dmu_tx_abort(tx);
314 return (err);
315 }
316
317 nvarg = fnvlist_dup(nvl);
318 if (spa_history_zone() != NULL) {
319 fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
320 spa_history_zone());
321 }
322 fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
323
324 /* Kick this off asynchronously; errors are ignored. */
325 dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
326 nvarg, 0, ZFS_SPACE_CHECK_NONE, tx);
327 dmu_tx_commit(tx);
328
329 /* spa_history_log_sync will free nvl */
330 return (err);
331
332 }
333
334 /*
335 * Read out the command history.
336 */
337 int
spa_history_get(spa_t * spa,uint64_t * offp,uint64_t * len,char * buf)338 spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
339 {
340 objset_t *mos = spa->spa_meta_objset;
341 dmu_buf_t *dbp;
342 uint64_t read_len, phys_read_off, phys_eof;
343 uint64_t leftover = 0;
344 spa_history_phys_t *shpp;
345 int err;
346
347 /*
348 * If the command history doesn't exist (older pool),
349 * that's ok, just return ENOENT.
350 */
351 if (!spa->spa_history)
352 return (SET_ERROR(ENOENT));
353
354 /*
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.
358 */
359 if (*offp == 0 && spa_writeable(spa))
360 txg_wait_synced(spa_get_dsl(spa), 0);
361
362 if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
363 return (err);
364 shpp = dbp->db_data;
365
366 #ifdef ZFS_DEBUG
367 {
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);
371 }
372 #endif
373
374 mutex_enter(&spa->spa_history_lock);
375 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
376
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 -
381 phys_read_off);
382 } else {
383 /*
384 * Need to reset passed in offset to BOF if the passed in
385 * offset has since been overwritten.
386 */
387 *offp = MAX(*offp, shpp->sh_bof);
388 phys_read_off = spa_history_log_to_phys(*offp, shpp);
389
390 /*
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.
394 */
395 if (phys_read_off <= phys_eof) {
396 read_len = MIN(*len, phys_eof - phys_read_off);
397 } else {
398 read_len = MIN(*len,
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);
403 }
404 }
405 }
406
407 /* offset for consumer to use next */
408 *offp += read_len + leftover;
409
410 /* tell the consumer how much you actually read */
411 *len = read_len + leftover;
412
413 if (read_len == 0) {
414 mutex_exit(&spa->spa_history_lock);
415 dmu_buf_rele(dbp, FTAG);
416 return (0);
417 }
418
419 err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
420 DMU_READ_PREFETCH);
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);
424 }
425 mutex_exit(&spa->spa_history_lock);
426
427 dmu_buf_rele(dbp, FTAG);
428 return (err);
429 }
430
431 /*
432 * The nvlist will be consumed by this call.
433 */
434 static void
log_internal(nvlist_t * nvl,const char * operation,spa_t * spa,dmu_tx_t * tx,const char * fmt,va_list adx)435 log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
436 dmu_tx_t *tx, const char *fmt, va_list adx)
437 {
438 char *msg;
439
440 /*
441 * If this is part of creating a pool, not everything is
442 * initialized yet, so don't bother logging the internal events.
443 * Likewise if the pool is not writeable.
444 */
445 if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
446 fnvlist_free(nvl);
447 return;
448 }
449
450 msg = kmem_alloc(vsnprintf(NULL, 0, fmt, adx) + 1, KM_SLEEP);
451 (void) vsprintf(msg, fmt, adx);
452 fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
453 strfree(msg);
454
455 fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
456 fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
457
458 if (dmu_tx_is_syncing(tx)) {
459 spa_history_log_sync(nvl, tx);
460 } else {
461 dsl_sync_task_nowait(spa_get_dsl(spa),
462 spa_history_log_sync, nvl, 0, ZFS_SPACE_CHECK_NONE, tx);
463 }
464 /* spa_history_log_sync() will free nvl */
465 }
466
467 void
spa_history_log_internal(spa_t * spa,const char * operation,dmu_tx_t * tx,const char * fmt,...)468 spa_history_log_internal(spa_t *spa, const char *operation,
469 dmu_tx_t *tx, const char *fmt, ...)
470 {
471 dmu_tx_t *htx = tx;
472 va_list adx;
473
474 /* create a tx if we didn't get one */
475 if (tx == NULL) {
476 htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
477 if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
478 dmu_tx_abort(htx);
479 return;
480 }
481 }
482
483 va_start(adx, fmt);
484 log_internal(fnvlist_alloc(), operation, spa, htx, fmt, adx);
485 va_end(adx);
486
487 /* if we didn't get a tx from the caller, commit the one we made */
488 if (tx == NULL)
489 dmu_tx_commit(htx);
490 }
491
492 void
spa_history_log_internal_ds(dsl_dataset_t * ds,const char * operation,dmu_tx_t * tx,const char * fmt,...)493 spa_history_log_internal_ds(dsl_dataset_t *ds, const char *operation,
494 dmu_tx_t *tx, const char *fmt, ...)
495 {
496 va_list adx;
497 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
498 nvlist_t *nvl = fnvlist_alloc();
499
500 ASSERT(tx != NULL);
501
502 dsl_dataset_name(ds, namebuf);
503 fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
504 fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID, ds->ds_object);
505
506 va_start(adx, fmt);
507 log_internal(nvl, operation, dsl_dataset_get_spa(ds), tx, fmt, adx);
508 va_end(adx);
509 }
510
511 void
spa_history_log_internal_dd(dsl_dir_t * dd,const char * operation,dmu_tx_t * tx,const char * fmt,...)512 spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
513 dmu_tx_t *tx, const char *fmt, ...)
514 {
515 va_list adx;
516 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
517 nvlist_t *nvl = fnvlist_alloc();
518
519 ASSERT(tx != NULL);
520
521 dsl_dir_name(dd, namebuf);
522 fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
523 fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
524 dsl_dir_phys(dd)->dd_head_dataset_obj);
525
526 va_start(adx, fmt);
527 log_internal(nvl, operation, dd->dd_pool->dp_spa, tx, fmt, adx);
528 va_end(adx);
529 }
530
531 void
spa_history_log_version(spa_t * spa,const char * operation)532 spa_history_log_version(spa_t *spa, const char *operation)
533 {
534 spa_history_log_internal(spa, operation, NULL,
535 "pool version %llu; software version %llu/%d; uts %s %s %s %s",
536 (u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
537 utsname.nodename, utsname.release, utsname.version,
538 utsname.machine);
539 }
540