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 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 */ 24 25 #include <sys/zfs_context.h> 26 #include <sys/dmu_objset.h> 27 #include <sys/dmu_traverse.h> 28 #include <sys/dsl_dataset.h> 29 #include <sys/dsl_dir.h> 30 #include <sys/dsl_pool.h> 31 #include <sys/dnode.h> 32 #include <sys/spa.h> 33 #include <sys/zio.h> 34 #include <sys/dmu_impl.h> 35 #include <sys/sa.h> 36 #include <sys/sa_impl.h> 37 #include <sys/callb.h> 38 39 int zfs_pd_blks_max = 100; 40 41 typedef struct prefetch_data { 42 kmutex_t pd_mtx; 43 kcondvar_t pd_cv; 44 int pd_blks_max; 45 int pd_blks_fetched; 46 int pd_flags; 47 boolean_t pd_cancel; 48 boolean_t pd_exited; 49 } prefetch_data_t; 50 51 typedef struct traverse_data { 52 spa_t *td_spa; 53 uint64_t td_objset; 54 blkptr_t *td_rootbp; 55 uint64_t td_min_txg; 56 int td_flags; 57 prefetch_data_t *td_pfd; 58 blkptr_cb_t *td_func; 59 void *td_arg; 60 } traverse_data_t; 61 62 static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, 63 arc_buf_t *buf, uint64_t objset, uint64_t object); 64 65 static int 66 traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) 67 { 68 traverse_data_t *td = arg; 69 zbookmark_t zb; 70 71 if (bp->blk_birth == 0) 72 return (0); 73 74 if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa)) 75 return (0); 76 77 SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, 78 bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); 79 80 (void) td->td_func(td->td_spa, zilog, bp, NULL, &zb, NULL, td->td_arg); 81 82 return (0); 83 } 84 85 static int 86 traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg) 87 { 88 traverse_data_t *td = arg; 89 90 if (lrc->lrc_txtype == TX_WRITE) { 91 lr_write_t *lr = (lr_write_t *)lrc; 92 blkptr_t *bp = &lr->lr_blkptr; 93 zbookmark_t zb; 94 95 if (bp->blk_birth == 0) 96 return (0); 97 98 if (claim_txg == 0 || bp->blk_birth < claim_txg) 99 return (0); 100 101 SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid, 102 ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); 103 104 (void) td->td_func(td->td_spa, zilog, bp, NULL, &zb, NULL, 105 td->td_arg); 106 } 107 return (0); 108 } 109 110 static void 111 traverse_zil(traverse_data_t *td, zil_header_t *zh) 112 { 113 uint64_t claim_txg = zh->zh_claim_txg; 114 zilog_t *zilog; 115 116 /* 117 * We only want to visit blocks that have been claimed but not yet 118 * replayed; plus, in read-only mode, blocks that are already stable. 119 */ 120 if (claim_txg == 0 && spa_writeable(td->td_spa)) 121 return; 122 123 zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh); 124 125 (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td, 126 claim_txg); 127 128 zil_free(zilog); 129 } 130 131 static int 132 traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, 133 arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb) 134 { 135 zbookmark_t czb; 136 int err = 0, lasterr = 0; 137 arc_buf_t *buf = NULL; 138 prefetch_data_t *pd = td->td_pfd; 139 boolean_t hard = td->td_flags & TRAVERSE_HARD; 140 141 if (bp->blk_birth == 0) { 142 err = td->td_func(td->td_spa, NULL, NULL, pbuf, zb, dnp, 143 td->td_arg); 144 return (err); 145 } 146 147 if (bp->blk_birth <= td->td_min_txg) 148 return (0); 149 150 if (pd && !pd->pd_exited && 151 ((pd->pd_flags & TRAVERSE_PREFETCH_DATA) || 152 BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) { 153 mutex_enter(&pd->pd_mtx); 154 ASSERT(pd->pd_blks_fetched >= 0); 155 while (pd->pd_blks_fetched == 0 && !pd->pd_exited) 156 cv_wait(&pd->pd_cv, &pd->pd_mtx); 157 pd->pd_blks_fetched--; 158 cv_broadcast(&pd->pd_cv); 159 mutex_exit(&pd->pd_mtx); 160 } 161 162 if (td->td_flags & TRAVERSE_PRE) { 163 err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp, 164 td->td_arg); 165 if (err) 166 return (err); 167 } 168 169 if (BP_GET_LEVEL(bp) > 0) { 170 uint32_t flags = ARC_WAIT; 171 int i; 172 blkptr_t *cbp; 173 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; 174 175 err = dsl_read(NULL, td->td_spa, bp, pbuf, 176 arc_getbuf_func, &buf, 177 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 178 if (err) 179 return (err); 180 181 /* recursively visitbp() blocks below this */ 182 cbp = buf->b_data; 183 for (i = 0; i < epb; i++, cbp++) { 184 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, 185 zb->zb_level - 1, 186 zb->zb_blkid * epb + i); 187 err = traverse_visitbp(td, dnp, buf, cbp, &czb); 188 if (err) { 189 if (!hard) 190 break; 191 lasterr = err; 192 } 193 } 194 } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { 195 uint32_t flags = ARC_WAIT; 196 int i; 197 int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; 198 199 err = dsl_read(NULL, td->td_spa, bp, pbuf, 200 arc_getbuf_func, &buf, 201 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 202 if (err) 203 return (err); 204 205 /* recursively visitbp() blocks below this */ 206 dnp = buf->b_data; 207 for (i = 0; i < epb; i++, dnp++) { 208 err = traverse_dnode(td, dnp, buf, zb->zb_objset, 209 zb->zb_blkid * epb + i); 210 if (err) { 211 if (!hard) 212 break; 213 lasterr = err; 214 } 215 } 216 } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { 217 uint32_t flags = ARC_WAIT; 218 objset_phys_t *osp; 219 dnode_phys_t *dnp; 220 221 err = dsl_read_nolock(NULL, td->td_spa, bp, 222 arc_getbuf_func, &buf, 223 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 224 if (err) 225 return (err); 226 227 osp = buf->b_data; 228 dnp = &osp->os_meta_dnode; 229 err = traverse_dnode(td, dnp, buf, zb->zb_objset, 230 DMU_META_DNODE_OBJECT); 231 if (err && hard) { 232 lasterr = err; 233 err = 0; 234 } 235 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { 236 dnp = &osp->os_userused_dnode; 237 err = traverse_dnode(td, dnp, buf, zb->zb_objset, 238 DMU_USERUSED_OBJECT); 239 } 240 if (err && hard) { 241 lasterr = err; 242 err = 0; 243 } 244 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { 245 dnp = &osp->os_groupused_dnode; 246 err = traverse_dnode(td, dnp, buf, zb->zb_objset, 247 DMU_GROUPUSED_OBJECT); 248 } 249 } 250 251 if (buf) 252 (void) arc_buf_remove_ref(buf, &buf); 253 254 if (err == 0 && lasterr == 0 && (td->td_flags & TRAVERSE_POST)) { 255 err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp, 256 td->td_arg); 257 } 258 259 return (err != 0 ? err : lasterr); 260 } 261 262 static int 263 traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, 264 arc_buf_t *buf, uint64_t objset, uint64_t object) 265 { 266 int j, err = 0, lasterr = 0; 267 zbookmark_t czb; 268 boolean_t hard = (td->td_flags & TRAVERSE_HARD); 269 270 for (j = 0; j < dnp->dn_nblkptr; j++) { 271 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j); 272 err = traverse_visitbp(td, dnp, buf, 273 (blkptr_t *)&dnp->dn_blkptr[j], &czb); 274 if (err) { 275 if (!hard) 276 break; 277 lasterr = err; 278 } 279 } 280 281 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { 282 SET_BOOKMARK(&czb, objset, 283 object, 0, DMU_SPILL_BLKID); 284 err = traverse_visitbp(td, dnp, buf, 285 (blkptr_t *)&dnp->dn_spill, &czb); 286 if (err) { 287 if (!hard) 288 return (err); 289 lasterr = err; 290 } 291 } 292 return (err != 0 ? err : lasterr); 293 } 294 295 /* ARGSUSED */ 296 static int 297 traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 298 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, 299 void *arg) 300 { 301 prefetch_data_t *pfd = arg; 302 uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH; 303 304 ASSERT(pfd->pd_blks_fetched >= 0); 305 if (pfd->pd_cancel) 306 return (EINTR); 307 308 if (bp == NULL || !((pfd->pd_flags & TRAVERSE_PREFETCH_DATA) || 309 BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0) || 310 BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) 311 return (0); 312 313 mutex_enter(&pfd->pd_mtx); 314 while (!pfd->pd_cancel && pfd->pd_blks_fetched >= pfd->pd_blks_max) 315 cv_wait(&pfd->pd_cv, &pfd->pd_mtx); 316 pfd->pd_blks_fetched++; 317 cv_broadcast(&pfd->pd_cv); 318 mutex_exit(&pfd->pd_mtx); 319 320 (void) dsl_read(NULL, spa, bp, pbuf, NULL, NULL, 321 ZIO_PRIORITY_ASYNC_READ, 322 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, 323 &aflags, zb); 324 325 return (0); 326 } 327 328 static void 329 traverse_prefetch_thread(void *arg) 330 { 331 traverse_data_t *td_main = arg; 332 traverse_data_t td = *td_main; 333 zbookmark_t czb; 334 335 td.td_func = traverse_prefetcher; 336 td.td_arg = td_main->td_pfd; 337 td.td_pfd = NULL; 338 339 SET_BOOKMARK(&czb, td.td_objset, 340 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 341 (void) traverse_visitbp(&td, NULL, NULL, td.td_rootbp, &czb); 342 343 mutex_enter(&td_main->td_pfd->pd_mtx); 344 td_main->td_pfd->pd_exited = B_TRUE; 345 cv_broadcast(&td_main->td_pfd->pd_cv); 346 mutex_exit(&td_main->td_pfd->pd_mtx); 347 } 348 349 /* 350 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are 351 * in syncing context). 352 */ 353 static int 354 traverse_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *rootbp, 355 uint64_t txg_start, int flags, blkptr_cb_t func, void *arg) 356 { 357 traverse_data_t td; 358 prefetch_data_t pd = { 0 }; 359 zbookmark_t czb; 360 int err; 361 362 td.td_spa = spa; 363 td.td_objset = ds ? ds->ds_object : 0; 364 td.td_rootbp = rootbp; 365 td.td_min_txg = txg_start; 366 td.td_func = func; 367 td.td_arg = arg; 368 td.td_pfd = &pd; 369 td.td_flags = flags; 370 371 pd.pd_blks_max = zfs_pd_blks_max; 372 pd.pd_flags = flags; 373 mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL); 374 cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL); 375 376 /* See comment on ZIL traversal in dsl_scan_visitds. */ 377 if (ds != NULL && !dsl_dataset_is_snapshot(ds)) { 378 objset_t *os; 379 380 err = dmu_objset_from_ds(ds, &os); 381 if (err) 382 return (err); 383 384 traverse_zil(&td, &os->os_zil_header); 385 } 386 387 if (!(flags & TRAVERSE_PREFETCH) || 388 0 == taskq_dispatch(system_taskq, traverse_prefetch_thread, 389 &td, TQ_NOQUEUE)) 390 pd.pd_exited = B_TRUE; 391 392 SET_BOOKMARK(&czb, td.td_objset, 393 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 394 err = traverse_visitbp(&td, NULL, NULL, rootbp, &czb); 395 396 mutex_enter(&pd.pd_mtx); 397 pd.pd_cancel = B_TRUE; 398 cv_broadcast(&pd.pd_cv); 399 while (!pd.pd_exited) 400 cv_wait(&pd.pd_cv, &pd.pd_mtx); 401 mutex_exit(&pd.pd_mtx); 402 403 mutex_destroy(&pd.pd_mtx); 404 cv_destroy(&pd.pd_cv); 405 406 return (err); 407 } 408 409 /* 410 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are 411 * in syncing context). 412 */ 413 int 414 traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags, 415 blkptr_cb_t func, void *arg) 416 { 417 return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, 418 &ds->ds_phys->ds_bp, txg_start, flags, func, arg)); 419 } 420 421 /* 422 * NB: pool must not be changing on-disk (eg, from zdb or sync context). 423 */ 424 int 425 traverse_pool(spa_t *spa, uint64_t txg_start, int flags, 426 blkptr_cb_t func, void *arg) 427 { 428 int err, lasterr = 0; 429 uint64_t obj; 430 dsl_pool_t *dp = spa_get_dsl(spa); 431 objset_t *mos = dp->dp_meta_objset; 432 boolean_t hard = (flags & TRAVERSE_HARD); 433 434 /* visit the MOS */ 435 err = traverse_impl(spa, NULL, spa_get_rootblkptr(spa), 436 txg_start, flags, func, arg); 437 if (err) 438 return (err); 439 440 /* visit each dataset */ 441 for (obj = 1; err == 0 || (err != ESRCH && hard); 442 err = dmu_object_next(mos, &obj, FALSE, txg_start)) { 443 dmu_object_info_t doi; 444 445 err = dmu_object_info(mos, obj, &doi); 446 if (err) { 447 if (!hard) 448 return (err); 449 lasterr = err; 450 continue; 451 } 452 453 if (doi.doi_type == DMU_OT_DSL_DATASET) { 454 dsl_dataset_t *ds; 455 uint64_t txg = txg_start; 456 457 rw_enter(&dp->dp_config_rwlock, RW_READER); 458 err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 459 rw_exit(&dp->dp_config_rwlock); 460 if (err) { 461 if (!hard) 462 return (err); 463 lasterr = err; 464 continue; 465 } 466 if (ds->ds_phys->ds_prev_snap_txg > txg) 467 txg = ds->ds_phys->ds_prev_snap_txg; 468 err = traverse_dataset(ds, txg, flags, func, arg); 469 dsl_dataset_rele(ds, FTAG); 470 if (err) { 471 if (!hard) 472 return (err); 473 lasterr = err; 474 } 475 } 476 } 477 if (err == ESRCH) 478 err = 0; 479 return (err != 0 ? err : lasterr); 480 } 481