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 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/utsname.h>
34#include <sys/sunddi.h>
35#include "zfs_comutil.h"
36#ifdef _KERNEL
37#include <sys/cmn_err.h>
38#include <sys/zone.h>
39#endif
40
41/*
42 * Routines to manage the on-disk history log.
43 *
44 * The history log is stored as a dmu object containing
45 * <packed record length, record nvlist> tuples.
46 *
47 * Where "record nvlist" is a nvlist containing uint64_ts and strings, and
48 * "packed record length" is the packed length of the "record nvlist" stored
49 * as a little endian uint64_t.
50 *
51 * The log is implemented as a ring buffer, though the original creation
52 * of the pool ('zpool create') is never overwritten.
53 *
54 * The history log is tracked as object 'spa_t::spa_history'. The bonus buffer
55 * of 'spa_history' stores the offsets for logging/retrieving history as
56 * 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of
57 * where the 'zpool create' record is stored. This allows us to never
58 * overwrite the original creation of the pool. 'sh_phys_max_off' is the
59 * physical ending offset in bytes of the log. This tells you the length of
60 * the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record
61 * is added, 'sh_eof' is incremented by the the size of the record.
62 * 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes).
63 * This is where the consumer should start reading from after reading in
64 * the 'zpool create' portion of the log.
65 *
66 * 'sh_records_lost' keeps track of how many records have been overwritten
67 * and permanently lost.
68 */
69
70/* convert a logical offset to physical */
71static uint64_t
72spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
73{
74 uint64_t phys_len;
75
76 phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len;
77 return ((log_off - shpp->sh_pool_create_len) % phys_len
78 + shpp->sh_pool_create_len);
79}
80
81void
82spa_history_create_obj(spa_t *spa, dmu_tx_t *tx)
83{
84 dmu_buf_t *dbp;
85 spa_history_phys_t *shpp;
86 objset_t *mos = spa->spa_meta_objset;
87
88 ASSERT(spa->spa_history == 0);
89 spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY,
90 SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS,
91 sizeof (spa_history_phys_t), tx);
92
93 VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
94 DMU_POOL_HISTORY, sizeof (uint64_t), 1,
95 &spa->spa_history, tx) == 0);
96
97 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
98 ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));
99
100 shpp = dbp->db_data;
101 dmu_buf_will_dirty(dbp, tx);
102
103 /*
104 * Figure out maximum size of history log. We set it at
105 * 0.1% of pool size, with a max of 1G and min of 128KB.
106 */
107 shpp->sh_phys_max_off =
108 metaslab_class_get_dspace(spa_normal_class(spa)) / 1000;
109 shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 1<<30);
110 shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10);
111
112 dmu_buf_rele(dbp, FTAG);
113}
114
115/*
116 * Change 'sh_bof' to the beginning of the next record.
117 */
118static int
119spa_history_advance_bof(spa_t *spa, spa_history_phys_t *shpp)
120{
121 objset_t *mos = spa->spa_meta_objset;
122 uint64_t firstread, reclen, phys_bof;
123 char buf[sizeof (reclen)];
124 int err;
125
126 phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
127 firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);
128
129 if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
130 buf, DMU_READ_PREFETCH)) != 0)
131 return (err);
132 if (firstread != sizeof (reclen)) {
133 if ((err = dmu_read(mos, spa->spa_history,
134 shpp->sh_pool_create_len, sizeof (reclen) - firstread,
135 buf + firstread, DMU_READ_PREFETCH)) != 0)
136 return (err);
137 }
138
139 reclen = LE_64(*((uint64_t *)buf));
140 shpp->sh_bof += reclen + sizeof (reclen);
141 shpp->sh_records_lost++;
142 return (0);
143}
144
145static int
146spa_history_write(spa_t *spa, void *buf, uint64_t len, spa_history_phys_t *shpp,
147 dmu_tx_t *tx)
148{
149 uint64_t firstwrite, phys_eof;
150 objset_t *mos = spa->spa_meta_objset;
151 int err;
152
153 ASSERT(MUTEX_HELD(&spa->spa_history_lock));
154
155 /* see if we need to reset logical BOF */
156 while (shpp->sh_phys_max_off - shpp->sh_pool_create_len -
157 (shpp->sh_eof - shpp->sh_bof) <= len) {
158 if ((err = spa_history_advance_bof(spa, shpp)) != 0) {
159 return (err);
160 }
161 }
162
163 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
164 firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
165 shpp->sh_eof += len;
166 dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);
167
168 len -= firstwrite;
169 if (len > 0) {
170 /* write out the rest at the beginning of physical file */
171 dmu_write(mos, spa->spa_history, shpp->sh_pool_create_len,
172 len, (char *)buf + firstwrite, tx);
173 }
174
175 return (0);
176}
177
178static char *
179spa_history_zone()
180{
181#ifdef _KERNEL
182 /* XXX: pr_hostname can be changed by default from within a jail! */
183 if (jailed(curthread->td_ucred))
184 return (curthread->td_ucred->cr_prison->pr_hostname);
185#endif
186 return ("global");
187}
188
189/*
190 * Write out a history event.
191 */
192/*ARGSUSED*/
193static void
194spa_history_log_sync(void *arg1, void *arg2, dmu_tx_t *tx)
195{
196 spa_t *spa = arg1;
197 history_arg_t *hap = arg2;
198 const char *history_str = hap->ha_history_str;
199 objset_t *mos = spa->spa_meta_objset;
200 dmu_buf_t *dbp;
201 spa_history_phys_t *shpp;
202 size_t reclen;
203 uint64_t le_len;
204 nvlist_t *nvrecord;
205 char *record_packed = NULL;
206 int ret;
207
208 /*
209 * If we have an older pool that doesn't have a command
210 * history object, create it now.
211 */
212 mutex_enter(&spa->spa_history_lock);
213 if (!spa->spa_history)
214 spa_history_create_obj(spa, tx);
215 mutex_exit(&spa->spa_history_lock);
216
217 /*
218 * Get the offset of where we need to write via the bonus buffer.
219 * Update the offset when the write completes.
220 */
221 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
222 shpp = dbp->db_data;
223
224 dmu_buf_will_dirty(dbp, tx);
225
226#ifdef ZFS_DEBUG
227 {
228 dmu_object_info_t doi;
229 dmu_object_info_from_db(dbp, &doi);
230 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
231 }
232#endif
233
234 VERIFY(nvlist_alloc(&nvrecord, NV_UNIQUE_NAME, KM_SLEEP) == 0);
235 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TIME,
236 gethrestime_sec()) == 0);
237 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_WHO, hap->ha_uid) == 0);
238 if (hap->ha_zone != NULL)
239 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_ZONE,
240 hap->ha_zone) == 0);
241#ifdef _KERNEL
242 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_HOST,
243 utsname.nodename) == 0);
244#endif
245 if (hap->ha_log_type == LOG_CMD_POOL_CREATE ||
246 hap->ha_log_type == LOG_CMD_NORMAL) {
247 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_CMD,
248 history_str) == 0);
249
250 zfs_dbgmsg("command: %s", history_str);
251 } else {
252 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_INT_EVENT,
253 hap->ha_event) == 0);
254 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TXG,
255 tx->tx_txg) == 0);
256 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_INT_STR,
257 history_str) == 0);
258
259 zfs_dbgmsg("internal %s pool:%s txg:%llu %s",
260 zfs_history_event_names[hap->ha_event], spa_name(spa),
261 (longlong_t)tx->tx_txg, history_str);
262
263 }
264
265 VERIFY(nvlist_size(nvrecord, &reclen, NV_ENCODE_XDR) == 0);
266 record_packed = kmem_alloc(reclen, KM_SLEEP);
267
268 VERIFY(nvlist_pack(nvrecord, &record_packed, &reclen,
269 NV_ENCODE_XDR, KM_SLEEP) == 0);
270
271 mutex_enter(&spa->spa_history_lock);
272 if (hap->ha_log_type == LOG_CMD_POOL_CREATE)
273 VERIFY(shpp->sh_eof == shpp->sh_pool_create_len);
274
275 /* write out the packed length as little endian */
276 le_len = LE_64((uint64_t)reclen);
277 ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
278 if (!ret)
279 ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
280
281 if (!ret && hap->ha_log_type == LOG_CMD_POOL_CREATE) {
282 shpp->sh_pool_create_len += sizeof (le_len) + reclen;
283 shpp->sh_bof = shpp->sh_pool_create_len;
284 }
285
286 mutex_exit(&spa->spa_history_lock);
287 nvlist_free(nvrecord);
288 kmem_free(record_packed, reclen);
289 dmu_buf_rele(dbp, FTAG);
290
291 strfree(hap->ha_history_str);
292 if (hap->ha_zone != NULL)
293 strfree(hap->ha_zone);
294 kmem_free(hap, sizeof (history_arg_t));
295}
296
297/*
298 * Write out a history event.
299 */
300int
301spa_history_log(spa_t *spa, const char *history_str, history_log_type_t what)
302{
303 history_arg_t *ha;
304 int err = 0;
305 dmu_tx_t *tx;
306
307 ASSERT(what != LOG_INTERNAL);
308
|
312 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
313 err = dmu_tx_assign(tx, TXG_WAIT);
314 if (err) {
315 dmu_tx_abort(tx);
316 return (err);
317 }
318
319 ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
320 ha->ha_history_str = strdup(history_str);
321 ha->ha_zone = strdup(spa_history_zone());
322 ha->ha_log_type = what;
323 ha->ha_uid = crgetuid(CRED());
324
325 /* Kick this off asynchronously; errors are ignored. */
326 dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
327 spa_history_log_sync, spa, ha, 0, tx);
328 dmu_tx_commit(tx);
329
330 /* spa_history_log_sync will free ha and strings */
331 return (err);
332}
333
334/*
335 * Read out the command history.
336 */
337int
338spa_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 (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
431static void
432log_internal(history_internal_events_t event, spa_t *spa,
433 dmu_tx_t *tx, const char *fmt, va_list adx)
434{
435 history_arg_t *ha;
436 va_list adx2;
437
438 /*
439 * If this is part of creating a pool, not everything is
440 * initialized yet, so don't bother logging the internal events.
|
444 return;
445
446 va_copy(adx2, adx);
447
448 ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
449 ha->ha_history_str = kmem_alloc(vsnprintf(NULL, 0, fmt, adx2) + 1,
450 KM_SLEEP);
451
452 va_end(adx2);
453
454 (void) vsprintf(ha->ha_history_str, fmt, adx);
455
456 ha->ha_log_type = LOG_INTERNAL;
457 ha->ha_event = event;
458 ha->ha_zone = NULL;
459 ha->ha_uid = 0;
460
461 if (dmu_tx_is_syncing(tx)) {
462 spa_history_log_sync(spa, ha, tx);
463 } else {
464 dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
465 spa_history_log_sync, spa, ha, 0, tx);
466 }
467 /* spa_history_log_sync() will free ha and strings */
468}
469
470void
471spa_history_log_internal(history_internal_events_t event, spa_t *spa,
472 dmu_tx_t *tx, const char *fmt, ...)
473{
474 dmu_tx_t *htx = tx;
475 va_list adx;
476
477 /* create a tx if we didn't get one */
478 if (tx == NULL) {
479 htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
480 if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
481 dmu_tx_abort(htx);
482 return;
483 }
484 }
485
486 va_start(adx, fmt);
487 log_internal(event, spa, htx, fmt, adx);
488 va_end(adx);
489
490 /* if we didn't get a tx from the caller, commit the one we made */
491 if (tx == NULL)
492 dmu_tx_commit(htx);
493}
494
495void
496spa_history_log_version(spa_t *spa, history_internal_events_t event)
497{
498#ifdef _KERNEL
499 uint64_t current_vers = spa_version(spa);
500
501 if (current_vers >= SPA_VERSION_ZPOOL_HISTORY) {
502 spa_history_log_internal(event, spa, NULL,
503 "pool spa %llu; zfs spa %llu; zpl %d; uts %s %s %s %s",
504 (u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
505 utsname.nodename, utsname.release, utsname.version,
506 utsname.machine);
507 }
508#if 0
509 cmn_err(CE_CONT, "!%s version %llu pool %s using %llu",
510 event == LOG_POOL_IMPORT ? "imported" :
511 event == LOG_POOL_CREATE ? "created" : "accessed",
512 (u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);
513#endif
514#endif
515}
|