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 2009 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 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/cmn_err.h> 35 #include <sys/sunddi.h> 36 #ifdef _KERNEL 37 #include <sys/zone.h> 38 #endif 39 40 /* 41 * Routines to manage the on-disk history log. 42 * 43 * The history log is stored as a dmu object containing 44 * <packed record length, record nvlist> tuples. 45 * 46 * Where "record nvlist" is a nvlist containing uint64_ts and strings, and 47 * "packed record length" is the packed length of the "record nvlist" stored 48 * as a little endian uint64_t. 49 * 50 * The log is implemented as a ring buffer, though the original creation 51 * of the pool ('zpool create') is never overwritten. 52 * 53 * The history log is tracked as object 'spa_t::spa_history'. The bonus buffer 54 * of 'spa_history' stores the offsets for logging/retrieving history as 55 * 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of 56 * where the 'zpool create' record is stored. This allows us to never 57 * overwrite the original creation of the pool. 'sh_phys_max_off' is the 58 * physical ending offset in bytes of the log. This tells you the length of 59 * the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record 60 * is added, 'sh_eof' is incremented by the the size of the record. 61 * 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes). 62 * This is where the consumer should start reading from after reading in 63 * the 'zpool create' portion of the log. 64 * 65 * 'sh_records_lost' keeps track of how many records have been overwritten 66 * and permanently lost. 67 */ 68 69 /* convert a logical offset to physical */ 70 static uint64_t 71 spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp) 72 { 73 uint64_t phys_len; 74 75 phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len; 76 return ((log_off - shpp->sh_pool_create_len) % phys_len 77 + shpp->sh_pool_create_len); 78 } 79 80 void 81 spa_history_create_obj(spa_t *spa, dmu_tx_t *tx) 82 { 83 dmu_buf_t *dbp; 84 spa_history_phys_t *shpp; 85 objset_t *mos = spa->spa_meta_objset; 86 87 ASSERT(spa->spa_history == 0); 88 spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY, 89 SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS, 90 sizeof (spa_history_phys_t), tx); 91 92 VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, 93 DMU_POOL_HISTORY, sizeof (uint64_t), 1, 94 &spa->spa_history, tx) == 0); 95 96 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)); 97 ASSERT(dbp->db_size >= sizeof (spa_history_phys_t)); 98 99 shpp = dbp->db_data; 100 dmu_buf_will_dirty(dbp, tx); 101 102 /* 103 * Figure out maximum size of history log. We set it at 104 * 1% of pool size, with a max of 32MB and min of 128KB. 105 */ 106 shpp->sh_phys_max_off = spa_get_dspace(spa) / 100; 107 shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 32<<20); 108 shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10); 109 110 dmu_buf_rele(dbp, FTAG); 111 } 112 113 /* 114 * Change 'sh_bof' to the beginning of the next record. 115 */ 116 static int 117 spa_history_advance_bof(spa_t *spa, spa_history_phys_t *shpp) 118 { 119 objset_t *mos = spa->spa_meta_objset; 120 uint64_t firstread, reclen, phys_bof; 121 char buf[sizeof (reclen)]; 122 int err; 123 124 phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp); 125 firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof); 126 127 if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread, 128 buf, DMU_READ_PREFETCH)) != 0) 129 return (err); 130 if (firstread != sizeof (reclen)) { 131 if ((err = dmu_read(mos, spa->spa_history, 132 shpp->sh_pool_create_len, sizeof (reclen) - firstread, 133 buf + firstread, DMU_READ_PREFETCH)) != 0) 134 return (err); 135 } 136 137 reclen = LE_64(*((uint64_t *)buf)); 138 shpp->sh_bof += reclen + sizeof (reclen); 139 shpp->sh_records_lost++; 140 return (0); 141 } 142 143 static int 144 spa_history_write(spa_t *spa, void *buf, uint64_t len, spa_history_phys_t *shpp, 145 dmu_tx_t *tx) 146 { 147 uint64_t firstwrite, phys_eof; 148 objset_t *mos = spa->spa_meta_objset; 149 int err; 150 151 ASSERT(MUTEX_HELD(&spa->spa_history_lock)); 152 153 /* see if we need to reset logical BOF */ 154 while (shpp->sh_phys_max_off - shpp->sh_pool_create_len - 155 (shpp->sh_eof - shpp->sh_bof) <= len) { 156 if ((err = spa_history_advance_bof(spa, shpp)) != 0) { 157 return (err); 158 } 159 } 160 161 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp); 162 firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof); 163 shpp->sh_eof += len; 164 dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx); 165 166 len -= firstwrite; 167 if (len > 0) { 168 /* write out the rest at the beginning of physical file */ 169 dmu_write(mos, spa->spa_history, shpp->sh_pool_create_len, 170 len, (char *)buf + firstwrite, tx); 171 } 172 173 return (0); 174 } 175 176 static char * 177 spa_history_zone() 178 { 179 #ifdef _KERNEL 180 return (curproc->p_zone->zone_name); 181 #else 182 return ("global"); 183 #endif 184 } 185 186 /* 187 * Write out a history event. 188 */ 189 static void 190 spa_history_log_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 191 { 192 spa_t *spa = arg1; 193 history_arg_t *hap = arg2; 194 const char *history_str = hap->ha_history_str; 195 objset_t *mos = spa->spa_meta_objset; 196 dmu_buf_t *dbp; 197 spa_history_phys_t *shpp; 198 size_t reclen; 199 uint64_t le_len; 200 nvlist_t *nvrecord; 201 char *record_packed = NULL; 202 int ret; 203 204 /* 205 * If we have an older pool that doesn't have a command 206 * history object, create it now. 207 */ 208 mutex_enter(&spa->spa_history_lock); 209 if (!spa->spa_history) 210 spa_history_create_obj(spa, tx); 211 mutex_exit(&spa->spa_history_lock); 212 213 /* 214 * Get the offset of where we need to write via the bonus buffer. 215 * Update the offset when the write completes. 216 */ 217 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)); 218 shpp = dbp->db_data; 219 220 dmu_buf_will_dirty(dbp, tx); 221 222 #ifdef ZFS_DEBUG 223 { 224 dmu_object_info_t doi; 225 dmu_object_info_from_db(dbp, &doi); 226 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS); 227 } 228 #endif 229 230 VERIFY(nvlist_alloc(&nvrecord, NV_UNIQUE_NAME, KM_SLEEP) == 0); 231 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TIME, 232 gethrestime_sec()) == 0); 233 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_WHO, 234 (uint64_t)crgetuid(cr)) == 0); 235 if (hap->ha_zone[0] != '\0') 236 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_ZONE, 237 hap->ha_zone) == 0); 238 #ifdef _KERNEL 239 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_HOST, 240 utsname.nodename) == 0); 241 #endif 242 if (hap->ha_log_type == LOG_CMD_POOL_CREATE || 243 hap->ha_log_type == LOG_CMD_NORMAL) { 244 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_CMD, 245 history_str) == 0); 246 } else { 247 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_INT_EVENT, 248 hap->ha_event) == 0); 249 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TXG, 250 tx->tx_txg) == 0); 251 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_INT_STR, 252 history_str) == 0); 253 } 254 255 VERIFY(nvlist_size(nvrecord, &reclen, NV_ENCODE_XDR) == 0); 256 record_packed = kmem_alloc(reclen, KM_SLEEP); 257 258 VERIFY(nvlist_pack(nvrecord, &record_packed, &reclen, 259 NV_ENCODE_XDR, KM_SLEEP) == 0); 260 261 mutex_enter(&spa->spa_history_lock); 262 if (hap->ha_log_type == LOG_CMD_POOL_CREATE) 263 VERIFY(shpp->sh_eof == shpp->sh_pool_create_len); 264 265 /* write out the packed length as little endian */ 266 le_len = LE_64((uint64_t)reclen); 267 ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx); 268 if (!ret) 269 ret = spa_history_write(spa, record_packed, reclen, shpp, tx); 270 271 if (!ret && hap->ha_log_type == LOG_CMD_POOL_CREATE) { 272 shpp->sh_pool_create_len += sizeof (le_len) + reclen; 273 shpp->sh_bof = shpp->sh_pool_create_len; 274 } 275 276 mutex_exit(&spa->spa_history_lock); 277 nvlist_free(nvrecord); 278 kmem_free(record_packed, reclen); 279 dmu_buf_rele(dbp, FTAG); 280 281 if (hap->ha_log_type == LOG_INTERNAL) { 282 kmem_free((void*)hap->ha_history_str, HIS_MAX_RECORD_LEN); 283 kmem_free(hap, sizeof (history_arg_t)); 284 } 285 } 286 287 /* 288 * Write out a history event. 289 */ 290 int 291 spa_history_log(spa_t *spa, const char *history_str, history_log_type_t what) 292 { 293 history_arg_t ha; 294 295 ASSERT(what != LOG_INTERNAL); 296 297 ha.ha_history_str = history_str; 298 ha.ha_log_type = what; 299 (void) strlcpy(ha.ha_zone, spa_history_zone(), sizeof (ha.ha_zone)); 300 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_history_log_sync, 301 spa, &ha, 0)); 302 } 303 304 /* 305 * Read out the command history. 306 */ 307 int 308 spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf) 309 { 310 objset_t *mos = spa->spa_meta_objset; 311 dmu_buf_t *dbp; 312 uint64_t read_len, phys_read_off, phys_eof; 313 uint64_t leftover = 0; 314 spa_history_phys_t *shpp; 315 int err; 316 317 /* 318 * If the command history doesn't exist (older pool), 319 * that's ok, just return ENOENT. 320 */ 321 if (!spa->spa_history) 322 return (ENOENT); 323 324 if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0) 325 return (err); 326 shpp = dbp->db_data; 327 328 #ifdef ZFS_DEBUG 329 { 330 dmu_object_info_t doi; 331 dmu_object_info_from_db(dbp, &doi); 332 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS); 333 } 334 #endif 335 336 mutex_enter(&spa->spa_history_lock); 337 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp); 338 339 if (*offp < shpp->sh_pool_create_len) { 340 /* read in just the zpool create history */ 341 phys_read_off = *offp; 342 read_len = MIN(*len, shpp->sh_pool_create_len - 343 phys_read_off); 344 } else { 345 /* 346 * Need to reset passed in offset to BOF if the passed in 347 * offset has since been overwritten. 348 */ 349 *offp = MAX(*offp, shpp->sh_bof); 350 phys_read_off = spa_history_log_to_phys(*offp, shpp); 351 352 /* 353 * Read up to the minimum of what the user passed down or 354 * the EOF (physical or logical). If we hit physical EOF, 355 * use 'leftover' to read from the physical BOF. 356 */ 357 if (phys_read_off <= phys_eof) { 358 read_len = MIN(*len, phys_eof - phys_read_off); 359 } else { 360 read_len = MIN(*len, 361 shpp->sh_phys_max_off - phys_read_off); 362 if (phys_read_off + *len > shpp->sh_phys_max_off) { 363 leftover = MIN(*len - read_len, 364 phys_eof - shpp->sh_pool_create_len); 365 } 366 } 367 } 368 369 /* offset for consumer to use next */ 370 *offp += read_len + leftover; 371 372 /* tell the consumer how much you actually read */ 373 *len = read_len + leftover; 374 375 if (read_len == 0) { 376 mutex_exit(&spa->spa_history_lock); 377 dmu_buf_rele(dbp, FTAG); 378 return (0); 379 } 380 381 err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf, 382 DMU_READ_PREFETCH); 383 if (leftover && err == 0) { 384 err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len, 385 leftover, buf + read_len, DMU_READ_PREFETCH); 386 } 387 mutex_exit(&spa->spa_history_lock); 388 389 dmu_buf_rele(dbp, FTAG); 390 return (err); 391 } 392 393 void 394 spa_history_internal_log(history_internal_events_t event, spa_t *spa, 395 dmu_tx_t *tx, cred_t *cr, const char *fmt, ...) 396 { 397 history_arg_t *hap; 398 char *str; 399 va_list adx; 400 401 /* 402 * If this is part of creating a pool, not everything is 403 * initialized yet, so don't bother logging the internal events. 404 */ 405 if (tx->tx_txg == TXG_INITIAL) 406 return; 407 408 hap = kmem_alloc(sizeof (history_arg_t), KM_SLEEP); 409 str = kmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP); 410 411 va_start(adx, fmt); 412 (void) vsnprintf(str, HIS_MAX_RECORD_LEN, fmt, adx); 413 va_end(adx); 414 415 hap->ha_log_type = LOG_INTERNAL; 416 hap->ha_history_str = str; 417 hap->ha_event = event; 418 hap->ha_zone[0] = '\0'; 419 420 if (dmu_tx_is_syncing(tx)) { 421 spa_history_log_sync(spa, hap, cr, tx); 422 } else { 423 dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL, 424 spa_history_log_sync, spa, hap, 0, tx); 425 } 426 /* spa_history_log_sync() will free hap and str */ 427 } 428