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) 2012 by Delphix. All rights reserved.
25 */
26
27 #include <sys/spa.h>
28 #include <sys/spa_impl.h>
29 #include <sys/zap.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/utsname.h>
36 #include <sys/cmn_err.h>
37 #include <sys/sunddi.h>
38 #include <sys/cred.h>
39 #include "zfs_comutil.h"
40 #ifdef _KERNEL
41 #include <sys/zone.h>
42 #endif
43
44 /*
45 * Routines to manage the on-disk history log.
46 *
47 * The history log is stored as a dmu object containing
48 * <packed record length, record nvlist> tuples.
49 *
50 * Where "record nvlist" is a nvlist containing uint64_ts and strings, and
51 * "packed record length" is the packed length of the "record nvlist" stored
52 * as a little endian uint64_t.
53 *
54 * The log is implemented as a ring buffer, though the original creation
55 * of the pool ('zpool create') is never overwritten.
56 *
57 * The history log is tracked as object 'spa_t::spa_history'. The bonus buffer
58 * of 'spa_history' stores the offsets for logging/retrieving history as
59 * 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of
60 * where the 'zpool create' record is stored. This allows us to never
61 * overwrite the original creation of the pool. 'sh_phys_max_off' is the
62 * physical ending offset in bytes of the log. This tells you the length of
63 * the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record
64 * is added, 'sh_eof' is incremented by the the size of the record.
65 * 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes).
66 * This is where the consumer should start reading from after reading in
67 * the 'zpool create' portion of the log.
68 *
69 * 'sh_records_lost' keeps track of how many records have been overwritten
70 * and permanently lost.
71 */
72
73 /* convert a logical offset to physical */
74 static uint64_t
75 spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
76 {
77 uint64_t phys_len;
78
79 phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len;
80 return ((log_off - shpp->sh_pool_create_len) % phys_len
81 + shpp->sh_pool_create_len);
82 }
83
84 void
85 spa_history_create_obj(spa_t *spa, dmu_tx_t *tx)
86 {
87 dmu_buf_t *dbp;
88 spa_history_phys_t *shpp;
89 objset_t *mos = spa->spa_meta_objset;
90
91 ASSERT(spa->spa_history == 0);
92 spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY,
93 SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS,
94 sizeof (spa_history_phys_t), tx);
95
96 VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
97 DMU_POOL_HISTORY, sizeof (uint64_t), 1,
98 &spa->spa_history, tx) == 0);
99
100 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
101 ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));
102
103 shpp = dbp->db_data;
104 dmu_buf_will_dirty(dbp, tx);
105
106 /*
107 * Figure out maximum size of history log. We set it at
108 * 0.1% of pool size, with a max of 1G and min of 128KB.
109 */
110 shpp->sh_phys_max_off =
111 metaslab_class_get_dspace(spa_normal_class(spa)) / 1000;
112 shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 1<<30);
113 shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10);
114
115 dmu_buf_rele(dbp, FTAG);
116 }
117
118 /*
119 * Change 'sh_bof' to the beginning of the next record.
120 */
121 static int
122 spa_history_advance_bof(spa_t *spa, spa_history_phys_t *shpp)
123 {
124 objset_t *mos = spa->spa_meta_objset;
125 uint64_t firstread, reclen, phys_bof;
126 char buf[sizeof (reclen)];
127 int err;
128
129 phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
130 firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);
131
132 if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
133 buf, DMU_READ_PREFETCH)) != 0)
134 return (err);
135 if (firstread != sizeof (reclen)) {
136 if ((err = dmu_read(mos, spa->spa_history,
137 shpp->sh_pool_create_len, sizeof (reclen) - firstread,
138 buf + firstread, DMU_READ_PREFETCH)) != 0)
139 return (err);
140 }
141
142 reclen = LE_64(*((uint64_t *)buf));
143 shpp->sh_bof += reclen + sizeof (reclen);
144 shpp->sh_records_lost++;
145 return (0);
146 }
147
148 static int
149 spa_history_write(spa_t *spa, void *buf, uint64_t len, spa_history_phys_t *shpp,
150 dmu_tx_t *tx)
151 {
152 uint64_t firstwrite, phys_eof;
153 objset_t *mos = spa->spa_meta_objset;
154 int err;
155
156 ASSERT(MUTEX_HELD(&spa->spa_history_lock));
157
158 /* see if we need to reset logical BOF */
159 while (shpp->sh_phys_max_off - shpp->sh_pool_create_len -
160 (shpp->sh_eof - shpp->sh_bof) <= len) {
161 if ((err = spa_history_advance_bof(spa, shpp)) != 0) {
162 return (err);
163 }
164 }
165
166 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
167 firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
168 shpp->sh_eof += len;
169 dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);
170
171 len -= firstwrite;
172 if (len > 0) {
173 /* write out the rest at the beginning of physical file */
174 dmu_write(mos, spa->spa_history, shpp->sh_pool_create_len,
175 len, (char *)buf + firstwrite, tx);
176 }
177
178 return (0);
179 }
180
181 static char *
182 spa_history_zone(void)
183 {
184 #ifdef _KERNEL
185 if (INGLOBALZONE(curproc))
186 return (NULL);
187 return (curproc->p_zone->zone_name);
188 #else
189 return (NULL);
190 #endif
191 }
192
193 /*
194 * Write out a history event.
195 */
196 /*ARGSUSED*/
197 static void
198 spa_history_log_sync(void *arg1, void *arg2, dmu_tx_t *tx)
199 {
200 spa_t *spa = arg1;
201 nvlist_t *nvl = arg2;
202 objset_t *mos = spa->spa_meta_objset;
203 dmu_buf_t *dbp;
204 spa_history_phys_t *shpp;
205 size_t reclen;
206 uint64_t le_len;
207 char *record_packed = NULL;
208 int ret;
209
210 /*
211 * If we have an older pool that doesn't have a command
212 * history object, create it now.
213 */
214 mutex_enter(&spa->spa_history_lock);
215 if (!spa->spa_history)
216 spa_history_create_obj(spa, tx);
217 mutex_exit(&spa->spa_history_lock);
218
219 /*
220 * Get the offset of where we need to write via the bonus buffer.
221 * Update the offset when the write completes.
222 */
223 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
224 shpp = dbp->db_data;
225
226 dmu_buf_will_dirty(dbp, tx);
227
228 #ifdef ZFS_DEBUG
229 {
230 dmu_object_info_t doi;
231 dmu_object_info_from_db(dbp, &doi);
232 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
233 }
234 #endif
235
236 fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
237 #ifdef _KERNEL
238 fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname.nodename);
239 #endif
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));
251 } else {
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));
256 }
257 } else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
258 zfs_dbgmsg("ioctl %s",
259 fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
260 }
261
262 record_packed = fnvlist_pack(nvl, &reclen);
263
264 mutex_enter(&spa->spa_history_lock);
265
266 /* write out the packed length as little endian */
267 le_len = LE_64((uint64_t)reclen);
268 ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
269 if (!ret)
270 ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
271
272 /* The first command is the create, which we keep forever */
273 if (ret == 0 && shpp->sh_pool_create_len == 0 &&
274 nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
275 shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
276 }
277
278 mutex_exit(&spa->spa_history_lock);
279 fnvlist_pack_free(record_packed, reclen);
280 dmu_buf_rele(dbp, FTAG);
281 fnvlist_free(nvl);
282 }
283
284 /*
285 * Write out a history event.
286 */
287 int
288 spa_history_log(spa_t *spa, const char *msg)
289 {
290 int err;
291 nvlist_t *nvl = fnvlist_alloc();
292
293 fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg);
294 err = spa_history_log_nvl(spa, nvl);
295 fnvlist_free(nvl);
296 return (err);
297 }
298
299 int
300 spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
301 {
302 int err = 0;
303 dmu_tx_t *tx;
304 nvlist_t *nvarg;
305
306 if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY)
307 return (EINVAL);
308
309 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
310 err = dmu_tx_assign(tx, TXG_WAIT);
311 if (err) {
312 dmu_tx_abort(tx);
313 return (err);
314 }
315
316 nvarg = fnvlist_dup(nvl);
317 if (spa_history_zone() != NULL) {
318 fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
319 spa_history_zone());
320 }
321 fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
322
323 /* Kick this off asynchronously; errors are ignored. */
324 dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
325 spa_history_log_sync, spa, nvarg, 0, tx);
326 dmu_tx_commit(tx);
327
328 /* spa_history_log_sync will free nvl */
329 return (err);
330
331 }
332
333 /*
334 * Read out the command history.
335 */
336 int
337 spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
338 {
339 objset_t *mos = spa->spa_meta_objset;
340 dmu_buf_t *dbp;
341 uint64_t read_len, phys_read_off, phys_eof;
342 uint64_t leftover = 0;
343 spa_history_phys_t *shpp;
344 int err;
345
346 /*
347 * If the command history doesn't exist (older pool),
348 * that's ok, just return ENOENT.
349 */
350 if (!spa->spa_history)
351 return (ENOENT);
352
353 /*
354 * The history is logged asynchronously, so when they request
355 * the first chunk of history, make sure everything has been
356 * synced to disk so that we get it.
357 */
358 if (*offp == 0 && spa_writeable(spa))
359 txg_wait_synced(spa_get_dsl(spa), 0);
360
361 if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
362 return (err);
363 shpp = dbp->db_data;
364
365 #ifdef ZFS_DEBUG
366 {
367 dmu_object_info_t doi;
368 dmu_object_info_from_db(dbp, &doi);
369 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
370 }
371 #endif
372
373 mutex_enter(&spa->spa_history_lock);
374 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
375
376 if (*offp < shpp->sh_pool_create_len) {
377 /* read in just the zpool create history */
378 phys_read_off = *offp;
379 read_len = MIN(*len, shpp->sh_pool_create_len -
380 phys_read_off);
381 } else {
382 /*
383 * Need to reset passed in offset to BOF if the passed in
384 * offset has since been overwritten.
385 */
386 *offp = MAX(*offp, shpp->sh_bof);
387 phys_read_off = spa_history_log_to_phys(*offp, shpp);
388
389 /*
390 * Read up to the minimum of what the user passed down or
391 * the EOF (physical or logical). If we hit physical EOF,
392 * use 'leftover' to read from the physical BOF.
393 */
394 if (phys_read_off <= phys_eof) {
395 read_len = MIN(*len, phys_eof - phys_read_off);
396 } else {
397 read_len = MIN(*len,
398 shpp->sh_phys_max_off - phys_read_off);
399 if (phys_read_off + *len > shpp->sh_phys_max_off) {
400 leftover = MIN(*len - read_len,
401 phys_eof - shpp->sh_pool_create_len);
402 }
403 }
404 }
405
406 /* offset for consumer to use next */
407 *offp += read_len + leftover;
408
409 /* tell the consumer how much you actually read */
410 *len = read_len + leftover;
411
412 if (read_len == 0) {
413 mutex_exit(&spa->spa_history_lock);
414 dmu_buf_rele(dbp, FTAG);
415 return (0);
416 }
417
418 err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
419 DMU_READ_PREFETCH);
420 if (leftover && err == 0) {
421 err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
422 leftover, buf + read_len, DMU_READ_PREFETCH);
423 }
424 mutex_exit(&spa->spa_history_lock);
425
426 dmu_buf_rele(dbp, FTAG);
427 return (err);
428 }
429
430 /*
431 * The nvlist will be consumed by this call.
432 */
433 static void
434 log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
435 dmu_tx_t *tx, const char *fmt, va_list adx)
436 {
437 char *msg;
438
439 /*
440 * If this is part of creating a pool, not everything is
441 * initialized yet, so don't bother logging the internal events.
442 */
443 if (tx->tx_txg == TXG_INITIAL)
444 return;
445
446 msg = kmem_alloc(vsnprintf(NULL, 0, fmt, adx) + 1, KM_SLEEP);
447 (void) vsprintf(msg, fmt, adx);
448 fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
449 strfree(msg);
450
451 fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
452 fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
453
454 if (dmu_tx_is_syncing(tx)) {
455 spa_history_log_sync(spa, nvl, tx);
456 } else {
457 dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
458 spa_history_log_sync, spa, nvl, 0, tx);
459 }
460 /* spa_history_log_sync() will free nvl */
461 }
462
463 void
464 spa_history_log_internal(spa_t *spa, const char *operation,
465 dmu_tx_t *tx, const char *fmt, ...)
466 {
467 dmu_tx_t *htx = tx;
468 va_list adx;
469
470 /* create a tx if we didn't get one */
471 if (tx == NULL) {
472 htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
473 if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
474 dmu_tx_abort(htx);
475 return;
476 }
477 }
478
479 va_start(adx, fmt);
480 log_internal(fnvlist_alloc(), operation, spa, htx, fmt, adx);
481 va_end(adx);
482
483 /* if we didn't get a tx from the caller, commit the one we made */
484 if (tx == NULL)
485 dmu_tx_commit(htx);
486 }
487
488 void
489 spa_history_log_internal_ds(dsl_dataset_t *ds, const char *operation,
490 dmu_tx_t *tx, const char *fmt, ...)
491 {
492 va_list adx;
493 char namebuf[MAXNAMELEN];
494 nvlist_t *nvl = fnvlist_alloc();
495
496 ASSERT(tx != NULL);
497
498 dsl_dataset_name(ds, namebuf);
499 fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
500 fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID, ds->ds_object);
501
502 va_start(adx, fmt);
503 log_internal(nvl, operation, dsl_dataset_get_spa(ds), tx, fmt, adx);
504 va_end(adx);
505 }
506
507 void
508 spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
509 dmu_tx_t *tx, const char *fmt, ...)
510 {
511 va_list adx;
512 char namebuf[MAXNAMELEN];
513 nvlist_t *nvl = fnvlist_alloc();
514
515 ASSERT(tx != NULL);
516
517 dsl_dir_name(dd, namebuf);
518 fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
519 fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
520 dd->dd_phys->dd_head_dataset_obj);
521
522 va_start(adx, fmt);
523 log_internal(nvl, operation, dd->dd_pool->dp_spa, tx, fmt, adx);
524 va_end(adx);
525 }
526
527 void
528 spa_history_log_version(spa_t *spa, const char *operation)
529 {
530 #ifdef _KERNEL
531 uint64_t current_vers = spa_version(spa);
532
533 spa_history_log_internal(spa, operation, NULL,
534 "pool version %llu; software version %llu/%d; uts %s %s %s %s",
535 (u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
536 utsname.nodename, utsname.release, utsname.version,
537 utsname.machine);
538 cmn_err(CE_CONT, "!%s version %llu pool %s using %llu", operation,
539 (u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);
540 #endif
541 }