spa_history.c revision 194118
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 2008 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27#pragma ident "%Z%%M% %I% %E% SMI" 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/utsname.h> 36#include <sys/sunddi.h> 37#ifdef _KERNEL 38#include <sys/cmn_err.h> 39#include <sys/zone.h> 40#endif 41 42/* 43 * Routines to manage the on-disk history log. 44 * 45 * The history log is stored as a dmu object containing 46 * <packed record length, record nvlist> tuples. 47 * 48 * Where "record nvlist" is a nvlist containing uint64_ts and strings, and 49 * "packed record length" is the packed length of the "record nvlist" stored 50 * as a little endian uint64_t. 51 * 52 * The log is implemented as a ring buffer, though the original creation 53 * of the pool ('zpool create') is never overwritten. 54 * 55 * The history log is tracked as object 'spa_t::spa_history'. The bonus buffer 56 * of 'spa_history' stores the offsets for logging/retrieving history as 57 * 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of 58 * where the 'zpool create' record is stored. This allows us to never 59 * overwrite the original creation of the pool. 'sh_phys_max_off' is the 60 * physical ending offset in bytes of the log. This tells you the length of 61 * the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record 62 * is added, 'sh_eof' is incremented by the the size of the record. 63 * 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes). 64 * This is where the consumer should start reading from after reading in 65 * the 'zpool create' portion of the log. 66 * 67 * 'sh_records_lost' keeps track of how many records have been overwritten 68 * and permanently lost. 69 */ 70 71/* convert a logical offset to physical */ 72static uint64_t 73spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp) 74{ 75 uint64_t phys_len; 76 77 phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len; 78 return ((log_off - shpp->sh_pool_create_len) % phys_len 79 + shpp->sh_pool_create_len); 80} 81 82void 83spa_history_create_obj(spa_t *spa, dmu_tx_t *tx) 84{ 85 dmu_buf_t *dbp; 86 spa_history_phys_t *shpp; 87 objset_t *mos = spa->spa_meta_objset; 88 89 ASSERT(spa->spa_history == 0); 90 spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY, 91 SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS, 92 sizeof (spa_history_phys_t), tx); 93 94 VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, 95 DMU_POOL_HISTORY, sizeof (uint64_t), 1, 96 &spa->spa_history, tx) == 0); 97 98 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)); 99 ASSERT(dbp->db_size >= sizeof (spa_history_phys_t)); 100 101 shpp = dbp->db_data; 102 dmu_buf_will_dirty(dbp, tx); 103 104 /* 105 * Figure out maximum size of history log. We set it at 106 * 1% of pool size, with a max of 32MB and min of 128KB. 107 */ 108 shpp->sh_phys_max_off = spa_get_dspace(spa) / 100; 109 shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 32<<20); 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)) != 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)) != 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 */ 192static void 193spa_history_log_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 194{ 195 spa_t *spa = arg1; 196 history_arg_t *hap = arg2; 197 const char *history_str = hap->ha_history_str; 198 objset_t *mos = spa->spa_meta_objset; 199 dmu_buf_t *dbp; 200 spa_history_phys_t *shpp; 201 size_t reclen; 202 uint64_t le_len; 203 nvlist_t *nvrecord; 204 char *record_packed = NULL; 205 int ret; 206 207 /* 208 * If we have an older pool that doesn't have a command 209 * history object, create it now. 210 */ 211 mutex_enter(&spa->spa_history_lock); 212 if (!spa->spa_history) 213 spa_history_create_obj(spa, tx); 214 mutex_exit(&spa->spa_history_lock); 215 216 /* 217 * Get the offset of where we need to write via the bonus buffer. 218 * Update the offset when the write completes. 219 */ 220 VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)); 221 shpp = dbp->db_data; 222 223 dmu_buf_will_dirty(dbp, tx); 224 225#ifdef ZFS_DEBUG 226 { 227 dmu_object_info_t doi; 228 dmu_object_info_from_db(dbp, &doi); 229 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS); 230 } 231#endif 232 233 VERIFY(nvlist_alloc(&nvrecord, NV_UNIQUE_NAME, KM_SLEEP) == 0); 234 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TIME, 235 gethrestime_sec()) == 0); 236 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_WHO, 237 (uint64_t)crgetuid(cr)) == 0); 238 if (hap->ha_zone[0] != '\0') 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 } else { 250 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_INT_EVENT, 251 hap->ha_event) == 0); 252 VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TXG, 253 tx->tx_txg) == 0); 254 VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_INT_STR, 255 history_str) == 0); 256 } 257 258 VERIFY(nvlist_size(nvrecord, &reclen, NV_ENCODE_XDR) == 0); 259 record_packed = kmem_alloc(reclen, KM_SLEEP); 260 261 VERIFY(nvlist_pack(nvrecord, &record_packed, &reclen, 262 NV_ENCODE_XDR, KM_SLEEP) == 0); 263 264 mutex_enter(&spa->spa_history_lock); 265 if (hap->ha_log_type == LOG_CMD_POOL_CREATE) 266 VERIFY(shpp->sh_eof == shpp->sh_pool_create_len); 267 268 /* write out the packed length as little endian */ 269 le_len = LE_64((uint64_t)reclen); 270 ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx); 271 if (!ret) 272 ret = spa_history_write(spa, record_packed, reclen, shpp, tx); 273 274 if (!ret && hap->ha_log_type == LOG_CMD_POOL_CREATE) { 275 shpp->sh_pool_create_len += sizeof (le_len) + reclen; 276 shpp->sh_bof = shpp->sh_pool_create_len; 277 } 278 279 mutex_exit(&spa->spa_history_lock); 280 nvlist_free(nvrecord); 281 kmem_free(record_packed, reclen); 282 dmu_buf_rele(dbp, FTAG); 283 284 if (hap->ha_log_type == LOG_INTERNAL) { 285 kmem_free((void*)hap->ha_history_str, HIS_MAX_RECORD_LEN); 286 kmem_free(hap, sizeof (history_arg_t)); 287 } 288} 289 290/* 291 * Write out a history event. 292 */ 293int 294spa_history_log(spa_t *spa, const char *history_str, history_log_type_t what) 295{ 296 history_arg_t ha; 297 298 ASSERT(what != LOG_INTERNAL); 299 300 ha.ha_history_str = history_str; 301 ha.ha_log_type = what; 302 (void) strlcpy(ha.ha_zone, spa_history_zone(), sizeof (ha.ha_zone)); 303 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_history_log_sync, 304 spa, &ha, 0)); 305} 306 307/* 308 * Read out the command history. 309 */ 310int 311spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf) 312{ 313 objset_t *mos = spa->spa_meta_objset; 314 dmu_buf_t *dbp; 315 uint64_t read_len, phys_read_off, phys_eof; 316 uint64_t leftover = 0; 317 spa_history_phys_t *shpp; 318 int err; 319 320 /* 321 * If the command history doesn't exist (older pool), 322 * that's ok, just return ENOENT. 323 */ 324 if (!spa->spa_history) 325 return (ENOENT); 326 327 if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0) 328 return (err); 329 shpp = dbp->db_data; 330 331#ifdef ZFS_DEBUG 332 { 333 dmu_object_info_t doi; 334 dmu_object_info_from_db(dbp, &doi); 335 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS); 336 } 337#endif 338 339 mutex_enter(&spa->spa_history_lock); 340 phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp); 341 342 if (*offp < shpp->sh_pool_create_len) { 343 /* read in just the zpool create history */ 344 phys_read_off = *offp; 345 read_len = MIN(*len, shpp->sh_pool_create_len - 346 phys_read_off); 347 } else { 348 /* 349 * Need to reset passed in offset to BOF if the passed in 350 * offset has since been overwritten. 351 */ 352 *offp = MAX(*offp, shpp->sh_bof); 353 phys_read_off = spa_history_log_to_phys(*offp, shpp); 354 355 /* 356 * Read up to the minimum of what the user passed down or 357 * the EOF (physical or logical). If we hit physical EOF, 358 * use 'leftover' to read from the physical BOF. 359 */ 360 if (phys_read_off <= phys_eof) { 361 read_len = MIN(*len, phys_eof - phys_read_off); 362 } else { 363 read_len = MIN(*len, 364 shpp->sh_phys_max_off - phys_read_off); 365 if (phys_read_off + *len > shpp->sh_phys_max_off) { 366 leftover = MIN(*len - read_len, 367 phys_eof - shpp->sh_pool_create_len); 368 } 369 } 370 } 371 372 /* offset for consumer to use next */ 373 *offp += read_len + leftover; 374 375 /* tell the consumer how much you actually read */ 376 *len = read_len + leftover; 377 378 if (read_len == 0) { 379 mutex_exit(&spa->spa_history_lock); 380 dmu_buf_rele(dbp, FTAG); 381 return (0); 382 } 383 384 err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf); 385 if (leftover && err == 0) { 386 err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len, 387 leftover, buf + read_len); 388 } 389 mutex_exit(&spa->spa_history_lock); 390 391 dmu_buf_rele(dbp, FTAG); 392 return (err); 393} 394 395void 396spa_history_internal_log(history_internal_events_t event, spa_t *spa, 397 dmu_tx_t *tx, cred_t *cr, const char *fmt, ...) 398{ 399 history_arg_t *hap; 400 char *str; 401 va_list adx; 402 403 /* 404 * If this is part of creating a pool, not everything is 405 * initialized yet, so don't bother logging the internal events. 406 */ 407 if (tx->tx_txg == TXG_INITIAL) 408 return; 409 410 hap = kmem_alloc(sizeof (history_arg_t), KM_SLEEP); 411 str = kmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP); 412 413 va_start(adx, fmt); 414 (void) vsnprintf(str, HIS_MAX_RECORD_LEN, fmt, adx); 415 va_end(adx); 416 417 hap->ha_log_type = LOG_INTERNAL; 418 hap->ha_history_str = str; 419 hap->ha_event = event; 420 hap->ha_zone[0] = '\0'; 421 422 if (dmu_tx_is_syncing(tx)) { 423 spa_history_log_sync(spa, hap, cr, tx); 424 } else { 425 dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL, 426 spa_history_log_sync, spa, hap, 0, tx); 427 } 428 /* spa_history_log_sync() will free hap and str */ 429} 430