]>
Commit | Line | Data |
---|---|---|
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/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 an 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 | |
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 | |
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 | ASSERT0(spa->spa_history); | |
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 | VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, | |
98 | DMU_POOL_HISTORY, sizeof (uint64_t), 1, | |
99 | &spa->spa_history, tx)); | |
100 | ||
101 | VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)); | |
102 | ASSERT3U(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 | |
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 | |
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 * | |
183 | spa_history_zone(void) | |
184 | { | |
185 | #ifdef _KERNEL | |
186 | #ifdef HAVE_SPL | |
187 | return ("linux"); | |
188 | #else | |
189 | return (curproc->p_zone->zone_name); | |
190 | #endif | |
191 | #else | |
192 | return (NULL); | |
193 | #endif | |
194 | } | |
195 | ||
196 | /* | |
197 | * Post a history sysevent. | |
198 | * | |
199 | * The nvlist_t* passed into this function will be transformed into a new | |
200 | * nvlist where: | |
201 | * | |
202 | * 1. Nested nvlists will be flattened to a single level | |
203 | * 2. Keys will have their names normalized (to remove any problematic | |
204 | * characters, such as whitespace) | |
205 | * | |
206 | * The nvlist_t passed into this function will duplicated and should be freed | |
207 | * by caller. | |
208 | * | |
209 | */ | |
210 | static void | |
211 | spa_history_log_notify(spa_t *spa, nvlist_t *nvl) | |
212 | { | |
213 | nvlist_t *hist_nvl = fnvlist_alloc(); | |
214 | uint64_t uint64; | |
215 | char *string; | |
216 | ||
217 | if (nvlist_lookup_string(nvl, ZPOOL_HIST_CMD, &string) == 0) | |
218 | fnvlist_add_string(hist_nvl, ZFS_EV_HIST_CMD, string); | |
219 | ||
220 | if (nvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME, &string) == 0) | |
221 | fnvlist_add_string(hist_nvl, ZFS_EV_HIST_INT_NAME, string); | |
222 | ||
223 | if (nvlist_lookup_string(nvl, ZPOOL_HIST_ZONE, &string) == 0) | |
224 | fnvlist_add_string(hist_nvl, ZFS_EV_HIST_ZONE, string); | |
225 | ||
226 | if (nvlist_lookup_string(nvl, ZPOOL_HIST_HOST, &string) == 0) | |
227 | fnvlist_add_string(hist_nvl, ZFS_EV_HIST_HOST, string); | |
228 | ||
229 | if (nvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME, &string) == 0) | |
230 | fnvlist_add_string(hist_nvl, ZFS_EV_HIST_DSNAME, string); | |
231 | ||
232 | if (nvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR, &string) == 0) | |
233 | fnvlist_add_string(hist_nvl, ZFS_EV_HIST_INT_STR, string); | |
234 | ||
235 | if (nvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL, &string) == 0) | |
236 | fnvlist_add_string(hist_nvl, ZFS_EV_HIST_IOCTL, string); | |
237 | ||
238 | if (nvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME, &string) == 0) | |
239 | fnvlist_add_string(hist_nvl, ZFS_EV_HIST_INT_NAME, string); | |
240 | ||
241 | if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID, &uint64) == 0) | |
242 | fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_DSID, uint64); | |
243 | ||
244 | if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG, &uint64) == 0) | |
245 | fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_TXG, uint64); | |
246 | ||
247 | if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_TIME, &uint64) == 0) | |
248 | fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_TIME, uint64); | |
249 | ||
250 | if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_WHO, &uint64) == 0) | |
251 | fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_WHO, uint64); | |
252 | ||
253 | if (nvlist_lookup_uint64(nvl, ZPOOL_HIST_INT_EVENT, &uint64) == 0) | |
254 | fnvlist_add_uint64(hist_nvl, ZFS_EV_HIST_INT_EVENT, uint64); | |
255 | ||
256 | spa_event_notify(spa, NULL, hist_nvl, ESC_ZFS_HISTORY_EVENT); | |
257 | ||
258 | nvlist_free(hist_nvl); | |
259 | } | |
260 | ||
261 | /* | |
262 | * Write out a history event. | |
263 | */ | |
264 | /*ARGSUSED*/ | |
265 | static void | |
266 | spa_history_log_sync(void *arg, dmu_tx_t *tx) | |
267 | { | |
268 | nvlist_t *nvl = arg; | |
269 | spa_t *spa = dmu_tx_pool(tx)->dp_spa; | |
270 | objset_t *mos = spa->spa_meta_objset; | |
271 | dmu_buf_t *dbp; | |
272 | spa_history_phys_t *shpp; | |
273 | size_t reclen; | |
274 | uint64_t le_len; | |
275 | char *record_packed = NULL; | |
276 | int ret; | |
277 | ||
278 | /* | |
279 | * If we have an older pool that doesn't have a command | |
280 | * history object, create it now. | |
281 | */ | |
282 | mutex_enter(&spa->spa_history_lock); | |
283 | if (!spa->spa_history) | |
284 | spa_history_create_obj(spa, tx); | |
285 | mutex_exit(&spa->spa_history_lock); | |
286 | ||
287 | /* | |
288 | * Get the offset of where we need to write via the bonus buffer. | |
289 | * Update the offset when the write completes. | |
290 | */ | |
291 | VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)); | |
292 | shpp = dbp->db_data; | |
293 | ||
294 | dmu_buf_will_dirty(dbp, tx); | |
295 | ||
296 | #ifdef ZFS_DEBUG | |
297 | { | |
298 | dmu_object_info_t doi; | |
299 | dmu_object_info_from_db(dbp, &doi); | |
300 | ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS); | |
301 | } | |
302 | #endif | |
303 | ||
304 | fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec()); | |
305 | fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname()->nodename); | |
306 | ||
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(1M) 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 | VERIFY3U(nvlist_pack(nvl, &record_packed, &reclen, NV_ENCODE_NATIVE, | |
346 | KM_SLEEP), ==, 0); | |
347 | ||
348 | mutex_enter(&spa->spa_history_lock); | |
349 | ||
350 | /* write out the packed length as little endian */ | |
351 | le_len = LE_64((uint64_t)reclen); | |
352 | ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx); | |
353 | if (!ret) | |
354 | ret = spa_history_write(spa, record_packed, reclen, shpp, tx); | |
355 | ||
356 | /* The first command is the create, which we keep forever */ | |
357 | if (ret == 0 && shpp->sh_pool_create_len == 0 && | |
358 | nvlist_exists(nvl, ZPOOL_HIST_CMD)) { | |
359 | shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof; | |
360 | } | |
361 | ||
362 | mutex_exit(&spa->spa_history_lock); | |
363 | fnvlist_pack_free(record_packed, reclen); | |
364 | dmu_buf_rele(dbp, FTAG); | |
365 | fnvlist_free(nvl); | |
366 | } | |
367 | ||
368 | /* | |
369 | * Write out a history event. | |
370 | */ | |
371 | int | |
372 | spa_history_log(spa_t *spa, const char *msg) | |
373 | { | |
374 | int err; | |
375 | nvlist_t *nvl = fnvlist_alloc(); | |
376 | ||
377 | fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg); | |
378 | err = spa_history_log_nvl(spa, nvl); | |
379 | fnvlist_free(nvl); | |
380 | return (err); | |
381 | } | |
382 | ||
383 | int | |
384 | spa_history_log_nvl(spa_t *spa, nvlist_t *nvl) | |
385 | { | |
386 | int err = 0; | |
387 | dmu_tx_t *tx; | |
388 | nvlist_t *nvarg, *in_nvl = NULL; | |
389 | ||
390 | if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY || !spa_writeable(spa)) | |
391 | return (SET_ERROR(EINVAL)); | |
392 | ||
393 | err = nvlist_lookup_nvlist(nvl, ZPOOL_HIST_INPUT_NVL, &in_nvl); | |
394 | if (err == 0) { | |
395 | (void) nvlist_remove_all(in_nvl, ZPOOL_HIDDEN_ARGS); | |
396 | } | |
397 | ||
398 | tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); | |
399 | err = dmu_tx_assign(tx, TXG_WAIT); | |
400 | if (err) { | |
401 | dmu_tx_abort(tx); | |
402 | return (err); | |
403 | } | |
404 | ||
405 | VERIFY0(nvlist_dup(nvl, &nvarg, KM_SLEEP)); | |
406 | if (spa_history_zone() != NULL) { | |
407 | fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE, | |
408 | spa_history_zone()); | |
409 | } | |
410 | fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED())); | |
411 | ||
412 | /* Kick this off asynchronously; errors are ignored. */ | |
413 | dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync, | |
414 | nvarg, 0, ZFS_SPACE_CHECK_NONE, tx); | |
415 | dmu_tx_commit(tx); | |
416 | ||
417 | /* spa_history_log_sync will free nvl */ | |
418 | return (err); | |
419 | ||
420 | } | |
421 | ||
422 | /* | |
423 | * Read out the command history. | |
424 | */ | |
425 | int | |
426 | spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf) | |
427 | { | |
428 | objset_t *mos = spa->spa_meta_objset; | |
429 | dmu_buf_t *dbp; | |
430 | uint64_t read_len, phys_read_off, phys_eof; | |
431 | uint64_t leftover = 0; | |
432 | spa_history_phys_t *shpp; | |
433 | int err; | |
434 | ||
435 | /* | |
436 | * If the command history doesn't exist (older pool), | |
437 | * that's ok, just return ENOENT. | |
438 | */ | |
439 | if (!spa->spa_history) | |
440 | return (SET_ERROR(ENOENT)); | |
441 | ||
442 | /* | |
443 | * The history is logged asynchronously, so when they request | |
444 | * the first chunk of history, make sure everything has been | |
445 | * synced to disk so that we get it. | |
446 | */ | |
447 | if (*offp == 0 && spa_writeable(spa)) | |
448 | txg_wait_synced(spa_get_dsl(spa), 0); | |
449 | ||
450 | if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0) | |
451 | return (err); | |
452 | shpp = dbp->db_data; | |
453 | ||
454 | #ifdef ZFS_DEBUG | |
455 | { | |
456 | dmu_object_info_t doi; | |
457 | dmu_object_info_from_db(dbp, &doi); | |
458 | ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS); | |
459 | } | |
460 | #endif | |
461 | ||
462 | mutex_enter(&spa->spa_history_lock); | |
463 | phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp); | |
464 | ||
465 | if (*offp < shpp->sh_pool_create_len) { | |
466 | /* read in just the zpool create history */ | |
467 | phys_read_off = *offp; | |
468 | read_len = MIN(*len, shpp->sh_pool_create_len - | |
469 | phys_read_off); | |
470 | } else { | |
471 | /* | |
472 | * Need to reset passed in offset to BOF if the passed in | |
473 | * offset has since been overwritten. | |
474 | */ | |
475 | *offp = MAX(*offp, shpp->sh_bof); | |
476 | phys_read_off = spa_history_log_to_phys(*offp, shpp); | |
477 | ||
478 | /* | |
479 | * Read up to the minimum of what the user passed down or | |
480 | * the EOF (physical or logical). If we hit physical EOF, | |
481 | * use 'leftover' to read from the physical BOF. | |
482 | */ | |
483 | if (phys_read_off <= phys_eof) { | |
484 | read_len = MIN(*len, phys_eof - phys_read_off); | |
485 | } else { | |
486 | read_len = MIN(*len, | |
487 | shpp->sh_phys_max_off - phys_read_off); | |
488 | if (phys_read_off + *len > shpp->sh_phys_max_off) { | |
489 | leftover = MIN(*len - read_len, | |
490 | phys_eof - shpp->sh_pool_create_len); | |
491 | } | |
492 | } | |
493 | } | |
494 | ||
495 | /* offset for consumer to use next */ | |
496 | *offp += read_len + leftover; | |
497 | ||
498 | /* tell the consumer how much you actually read */ | |
499 | *len = read_len + leftover; | |
500 | ||
501 | if (read_len == 0) { | |
502 | mutex_exit(&spa->spa_history_lock); | |
503 | dmu_buf_rele(dbp, FTAG); | |
504 | return (0); | |
505 | } | |
506 | ||
507 | err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf, | |
508 | DMU_READ_PREFETCH); | |
509 | if (leftover && err == 0) { | |
510 | err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len, | |
511 | leftover, buf + read_len, DMU_READ_PREFETCH); | |
512 | } | |
513 | mutex_exit(&spa->spa_history_lock); | |
514 | ||
515 | dmu_buf_rele(dbp, FTAG); | |
516 | return (err); | |
517 | } | |
518 | ||
519 | /* | |
520 | * The nvlist will be consumed by this call. | |
521 | */ | |
522 | static void | |
523 | log_internal(nvlist_t *nvl, const char *operation, spa_t *spa, | |
524 | dmu_tx_t *tx, const char *fmt, va_list adx) | |
525 | { | |
526 | char *msg; | |
527 | ||
528 | /* | |
529 | * If this is part of creating a pool, not everything is | |
530 | * initialized yet, so don't bother logging the internal events. | |
531 | * Likewise if the pool is not writeable. | |
532 | */ | |
533 | if (spa_is_initializing(spa) || !spa_writeable(spa)) { | |
534 | fnvlist_free(nvl); | |
535 | return; | |
536 | } | |
537 | ||
538 | msg = kmem_vasprintf(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 | |
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 | |
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 | |
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 | |
619 | spa_history_log_version(spa_t *spa, const char *operation, dmu_tx_t *tx) | |
620 | { | |
621 | utsname_t *u = utsname(); | |
622 | ||
623 | spa_history_log_internal(spa, operation, tx, | |
624 | "pool version %llu; software version %llu/%llu; uts %s %s %s %s", | |
625 | (u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION, | |
626 | u->nodename, u->release, u->version, u->machine); | |
627 | } | |
628 | ||
629 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
630 | EXPORT_SYMBOL(spa_history_create_obj); | |
631 | EXPORT_SYMBOL(spa_history_get); | |
632 | EXPORT_SYMBOL(spa_history_log); | |
633 | EXPORT_SYMBOL(spa_history_log_internal); | |
634 | EXPORT_SYMBOL(spa_history_log_version); | |
635 | #endif |