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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #include <sys/zfs_context.h> 27 #include <sys/dmu_objset.h> 28 #include <sys/dmu_traverse.h> 29 #include <sys/dsl_dataset.h> 30 #include <sys/dsl_dir.h> 31 #include <sys/dsl_pool.h> 32 #include <sys/dnode.h> 33 #include <sys/spa.h> 34 #include <sys/zio.h> 35 #include <sys/dmu_impl.h> 36 #include <sys/callb.h> 37 38 #define SET_BOOKMARK(zb, objset, object, level, blkid) \ 39 { \ 40 (zb)->zb_objset = objset; \ 41 (zb)->zb_object = object; \ 42 (zb)->zb_level = level; \ 43 (zb)->zb_blkid = blkid; \ 44 } 45 46 struct prefetch_data { 47 kmutex_t pd_mtx; 48 kcondvar_t pd_cv; 49 int pd_blks_max; 50 int pd_blks_fetched; 51 int pd_flags; 52 boolean_t pd_cancel; 53 boolean_t pd_exited; 54 }; 55 56 struct traverse_data { 57 spa_t *td_spa; 58 uint64_t td_objset; 59 blkptr_t *td_rootbp; 60 uint64_t td_min_txg; 61 int td_flags; 62 struct prefetch_data *td_pfd; 63 blkptr_cb_t *td_func; 64 void *td_arg; 65 }; 66 67 /* ARGSUSED */ 68 static void 69 traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) 70 { 71 struct traverse_data *td = arg; 72 zbookmark_t zb; 73 74 if (bp->blk_birth == 0) 75 return; 76 77 if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa)) 78 return; 79 80 zb.zb_objset = td->td_objset; 81 zb.zb_object = 0; 82 zb.zb_level = -1; 83 zb.zb_blkid = bp->blk_cksum.zc_word[ZIL_ZC_SEQ]; 84 VERIFY(0 == td->td_func(td->td_spa, bp, &zb, NULL, td->td_arg)); 85 } 86 87 /* ARGSUSED */ 88 static void 89 traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg) 90 { 91 struct traverse_data *td = arg; 92 93 if (lrc->lrc_txtype == TX_WRITE) { 94 lr_write_t *lr = (lr_write_t *)lrc; 95 blkptr_t *bp = &lr->lr_blkptr; 96 zbookmark_t zb; 97 98 if (bp->blk_birth == 0) 99 return; 100 101 if (claim_txg == 0 || bp->blk_birth < claim_txg) 102 return; 103 104 zb.zb_objset = td->td_objset; 105 zb.zb_object = lr->lr_foid; 106 zb.zb_level = BP_GET_LEVEL(bp); 107 zb.zb_blkid = lr->lr_offset / BP_GET_LSIZE(bp); 108 VERIFY(0 == td->td_func(td->td_spa, bp, &zb, NULL, td->td_arg)); 109 } 110 } 111 112 static void 113 traverse_zil(struct traverse_data *td, zil_header_t *zh) 114 { 115 uint64_t claim_txg = zh->zh_claim_txg; 116 zilog_t *zilog; 117 118 /* 119 * We only want to visit blocks that have been claimed but not yet 120 * replayed (or, in read-only mode, blocks that *would* be claimed). 121 */ 122 if (claim_txg == 0 && (spa_mode & FWRITE)) 123 return; 124 125 zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh); 126 127 (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td, 128 claim_txg); 129 130 zil_free(zilog); 131 } 132 133 static int 134 traverse_visitbp(struct traverse_data *td, const dnode_phys_t *dnp, 135 arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb) 136 { 137 int err = 0; 138 arc_buf_t *buf = NULL; 139 struct prefetch_data *pd = td->td_pfd; 140 141 if (bp->blk_birth == 0) { 142 err = td->td_func(td->td_spa, NULL, zb, dnp, td->td_arg); 143 return (err); 144 } 145 146 if (bp->blk_birth <= td->td_min_txg) 147 return (0); 148 149 if (pd && !pd->pd_exited && 150 ((pd->pd_flags & TRAVERSE_PREFETCH_DATA) || 151 BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) { 152 mutex_enter(&pd->pd_mtx); 153 ASSERT(pd->pd_blks_fetched >= 0); 154 while (pd->pd_blks_fetched == 0 && !pd->pd_exited) 155 cv_wait(&pd->pd_cv, &pd->pd_mtx); 156 pd->pd_blks_fetched--; 157 cv_broadcast(&pd->pd_cv); 158 mutex_exit(&pd->pd_mtx); 159 } 160 161 if (td->td_flags & TRAVERSE_PRE) { 162 err = td->td_func(td->td_spa, bp, zb, dnp, td->td_arg); 163 if (err) 164 return (err); 165 } 166 167 if (BP_GET_LEVEL(bp) > 0) { 168 uint32_t flags = ARC_WAIT; 169 int i; 170 blkptr_t *cbp; 171 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; 172 173 err = arc_read(NULL, td->td_spa, bp, pbuf, 174 arc_getbuf_func, &buf, 175 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 176 if (err) 177 return (err); 178 179 /* recursively visitbp() blocks below this */ 180 cbp = buf->b_data; 181 for (i = 0; i < epb; i++, cbp++) { 182 zbookmark_t czb; 183 184 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 break; 190 } 191 } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { 192 uint32_t flags = ARC_WAIT; 193 int i, j; 194 int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; 195 196 err = arc_read(NULL, td->td_spa, bp, pbuf, 197 arc_getbuf_func, &buf, 198 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 199 if (err) 200 return (err); 201 202 /* recursively visitbp() blocks below this */ 203 dnp = buf->b_data; 204 for (i = 0; i < epb && err == 0; i++, dnp++) { 205 for (j = 0; j < dnp->dn_nblkptr; j++) { 206 zbookmark_t czb; 207 208 SET_BOOKMARK(&czb, zb->zb_objset, 209 zb->zb_blkid * epb + i, 210 dnp->dn_nlevels - 1, j); 211 err = traverse_visitbp(td, dnp, buf, 212 (blkptr_t *)&dnp->dn_blkptr[j], &czb); 213 if (err) 214 break; 215 } 216 } 217 } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { 218 uint32_t flags = ARC_WAIT; 219 objset_phys_t *osp; 220 int j; 221 222 err = arc_read_nolock(NULL, td->td_spa, bp, 223 arc_getbuf_func, &buf, 224 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 225 if (err) 226 return (err); 227 228 osp = buf->b_data; 229 /* 230 * traverse_zil is just here for zdb's leak checking. 231 * For other consumers, there will be no ZIL blocks. 232 */ 233 traverse_zil(td, &osp->os_zil_header); 234 235 for (j = 0; j < osp->os_meta_dnode.dn_nblkptr; j++) { 236 zbookmark_t czb; 237 238 SET_BOOKMARK(&czb, zb->zb_objset, 0, 239 osp->os_meta_dnode.dn_nlevels - 1, j); 240 err = traverse_visitbp(td, &osp->os_meta_dnode, buf, 241 (blkptr_t *)&osp->os_meta_dnode.dn_blkptr[j], 242 &czb); 243 if (err) 244 break; 245 } 246 } 247 248 if (buf) 249 (void) arc_buf_remove_ref(buf, &buf); 250 251 if (err == 0 && (td->td_flags & TRAVERSE_POST)) 252 err = td->td_func(td->td_spa, bp, zb, dnp, td->td_arg); 253 254 return (err); 255 } 256 257 /* ARGSUSED */ 258 static int 259 traverse_prefetcher(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb, 260 const dnode_phys_t *dnp, void *arg) 261 { 262 struct prefetch_data *pfd = arg; 263 uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH; 264 265 ASSERT(pfd->pd_blks_fetched >= 0); 266 if (pfd->pd_cancel) 267 return (EINTR); 268 269 if (bp == NULL || !((pfd->pd_flags & TRAVERSE_PREFETCH_DATA) || 270 BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) 271 return (0); 272 273 mutex_enter(&pfd->pd_mtx); 274 while (!pfd->pd_cancel && pfd->pd_blks_fetched >= pfd->pd_blks_max) 275 cv_wait(&pfd->pd_cv, &pfd->pd_mtx); 276 pfd->pd_blks_fetched++; 277 cv_broadcast(&pfd->pd_cv); 278 mutex_exit(&pfd->pd_mtx); 279 280 (void) arc_read_nolock(NULL, spa, bp, NULL, NULL, 281 ZIO_PRIORITY_ASYNC_READ, 282 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, 283 &aflags, zb); 284 285 return (0); 286 } 287 288 static void 289 traverse_prefetch_thread(void *arg) 290 { 291 struct traverse_data *td_main = arg; 292 struct traverse_data td = *td_main; 293 zbookmark_t czb; 294 295 td.td_func = traverse_prefetcher; 296 td.td_arg = td_main->td_pfd; 297 td.td_pfd = NULL; 298 299 SET_BOOKMARK(&czb, td.td_objset, 0, -1, 0); 300 (void) traverse_visitbp(&td, NULL, NULL, td.td_rootbp, &czb); 301 302 mutex_enter(&td_main->td_pfd->pd_mtx); 303 td_main->td_pfd->pd_exited = B_TRUE; 304 cv_broadcast(&td_main->td_pfd->pd_cv); 305 mutex_exit(&td_main->td_pfd->pd_mtx); 306 } 307 308 /* 309 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are 310 * in syncing context). 311 */ 312 static int 313 traverse_impl(spa_t *spa, uint64_t objset, blkptr_t *rootbp, 314 uint64_t txg_start, int flags, blkptr_cb_t func, void *arg) 315 { 316 struct traverse_data td; 317 struct prefetch_data pd = { 0 }; 318 zbookmark_t czb; 319 int err; 320 321 td.td_spa = spa; 322 td.td_objset = objset; 323 td.td_rootbp = rootbp; 324 td.td_min_txg = txg_start; 325 td.td_func = func; 326 td.td_arg = arg; 327 td.td_pfd = &pd; 328 td.td_flags = flags; 329 330 pd.pd_blks_max = 100; 331 pd.pd_flags = flags; 332 mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL); 333 cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL); 334 335 if (!(flags & TRAVERSE_PREFETCH) || 336 0 == taskq_dispatch(system_taskq, traverse_prefetch_thread, 337 &td, TQ_NOQUEUE)) 338 pd.pd_exited = B_TRUE; 339 340 SET_BOOKMARK(&czb, objset, 0, -1, 0); 341 err = traverse_visitbp(&td, NULL, NULL, rootbp, &czb); 342 343 mutex_enter(&pd.pd_mtx); 344 pd.pd_cancel = B_TRUE; 345 cv_broadcast(&pd.pd_cv); 346 while (!pd.pd_exited) 347 cv_wait(&pd.pd_cv, &pd.pd_mtx); 348 mutex_exit(&pd.pd_mtx); 349 350 mutex_destroy(&pd.pd_mtx); 351 cv_destroy(&pd.pd_cv); 352 353 return (err); 354 } 355 356 /* 357 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are 358 * in syncing context). 359 */ 360 int 361 traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags, 362 blkptr_cb_t func, void *arg) 363 { 364 return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds->ds_object, 365 &ds->ds_phys->ds_bp, txg_start, flags, func, arg)); 366 } 367 368 /* 369 * NB: pool must not be changing on-disk (eg, from zdb or sync context). 370 */ 371 int 372 traverse_pool(spa_t *spa, blkptr_cb_t func, void *arg) 373 { 374 int err; 375 uint64_t obj; 376 dsl_pool_t *dp = spa_get_dsl(spa); 377 objset_t *mos = dp->dp_meta_objset; 378 379 /* visit the MOS */ 380 err = traverse_impl(spa, 0, spa_get_rootblkptr(spa), 381 0, TRAVERSE_PRE, func, arg); 382 if (err) 383 return (err); 384 385 /* visit each dataset */ 386 for (obj = 1; err == 0; err = dmu_object_next(mos, &obj, FALSE, 0)) { 387 dmu_object_info_t doi; 388 389 err = dmu_object_info(mos, obj, &doi); 390 if (err) 391 return (err); 392 393 if (doi.doi_type == DMU_OT_DSL_DATASET) { 394 dsl_dataset_t *ds; 395 rw_enter(&dp->dp_config_rwlock, RW_READER); 396 err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 397 rw_exit(&dp->dp_config_rwlock); 398 if (err) 399 return (err); 400 err = traverse_dataset(ds, 401 ds->ds_phys->ds_prev_snap_txg, TRAVERSE_PRE, 402 func, arg); 403 dsl_dataset_rele(ds, FTAG); 404 if (err) 405 return (err); 406 } 407 } 408 if (err == ESRCH) 409 err = 0; 410 return (err); 411 } 412