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  * Copyright (c) 2012 by Delphix. All rights reserved.
  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/sa.h>
  37 #include <sys/sa_impl.h>
  38 #include <sys/callb.h>
  39 
  40 int zfs_pd_blks_max = 100;
  41 
  42 typedef struct prefetch_data {
  43         kmutex_t pd_mtx;
  44         kcondvar_t pd_cv;
  45         int pd_blks_max;
  46         int pd_blks_fetched;
  47         int pd_flags;
  48         boolean_t pd_cancel;
  49         boolean_t pd_exited;
  50 } prefetch_data_t;
  51 
  52 typedef struct traverse_data {
  53         spa_t *td_spa;
  54         uint64_t td_objset;
  55         blkptr_t *td_rootbp;
  56         uint64_t td_min_txg;
  57         zbookmark_t *td_resume;
  58         int td_flags;
  59         prefetch_data_t *td_pfd;
  60         blkptr_cb_t *td_func;
  61         void *td_arg;
  62 } traverse_data_t;
  63 
  64 static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
  65     arc_buf_t *buf, uint64_t objset, uint64_t object);
  66 
  67 static int
  68 traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
  69 {
  70         traverse_data_t *td = arg;
  71         zbookmark_t zb;
  72 
  73         if (bp->blk_birth == 0)
  74                 return (0);
  75 
  76         if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
  77                 return (0);
  78 
  79         SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
  80             bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
  81 
  82         (void) td->td_func(td->td_spa, zilog, bp, NULL, &zb, NULL, td->td_arg);
  83 
  84         return (0);
  85 }
  86 
  87 static int
  88 traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
  89 {
  90         traverse_data_t *td = arg;
  91 
  92         if (lrc->lrc_txtype == TX_WRITE) {
  93                 lr_write_t *lr = (lr_write_t *)lrc;
  94                 blkptr_t *bp = &lr->lr_blkptr;
  95                 zbookmark_t zb;
  96 
  97                 if (bp->blk_birth == 0)
  98                         return (0);
  99 
 100                 if (claim_txg == 0 || bp->blk_birth < claim_txg)
 101                         return (0);
 102 
 103                 SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid,
 104                     ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
 105 
 106                 (void) td->td_func(td->td_spa, zilog, bp, NULL, &zb, NULL,
 107                     td->td_arg);
 108         }
 109         return (0);
 110 }
 111 
 112 static void
 113 traverse_zil(traverse_data_t *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; plus, in read-only mode, blocks that are already stable.
 121          */
 122         if (claim_txg == 0 && spa_writeable(td->td_spa))
 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 typedef enum resume_skip {
 134         RESUME_SKIP_ALL,
 135         RESUME_SKIP_NONE,
 136         RESUME_SKIP_CHILDREN
 137 } resume_skip_t;
 138 
 139 /*
 140  * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and
 141  * the block indicated by zb does not need to be visited at all. Returns
 142  * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the
 143  * resume point. This indicates that this block should be visited but not its
 144  * children (since they must have been visited in a previous traversal).
 145  * Otherwise returns RESUME_SKIP_NONE.
 146  */
 147 static resume_skip_t
 148 resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp,
 149     const zbookmark_t *zb)
 150 {
 151         if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) {
 152                 /*
 153                  * If we already visited this bp & everything below,
 154                  * don't bother doing it again.
 155                  */
 156                 if (zbookmark_is_before(dnp, zb, td->td_resume))
 157                         return (RESUME_SKIP_ALL);
 158 
 159                 /*
 160                  * If we found the block we're trying to resume from, zero
 161                  * the bookmark out to indicate that we have resumed.
 162                  */
 163                 ASSERT3U(zb->zb_object, <=, td->td_resume->zb_object);
 164                 if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) {
 165                         bzero(td->td_resume, sizeof (*zb));
 166                         if (td->td_flags & TRAVERSE_POST)
 167                                 return (RESUME_SKIP_CHILDREN);
 168                 }
 169         }
 170         return (RESUME_SKIP_NONE);
 171 }
 172 
 173 static void
 174 traverse_pause(traverse_data_t *td, const zbookmark_t *zb)
 175 {
 176         ASSERT(td->td_resume != NULL);
 177         ASSERT3U(zb->zb_level, ==, 0);
 178         bcopy(zb, td->td_resume, sizeof (*td->td_resume));
 179 }
 180 
 181 static int
 182 traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
 183     arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb)
 184 {
 185         zbookmark_t czb;
 186         int err = 0, lasterr = 0;
 187         arc_buf_t *buf = NULL;
 188         prefetch_data_t *pd = td->td_pfd;
 189         boolean_t hard = td->td_flags & TRAVERSE_HARD;
 190         boolean_t pause = B_FALSE;
 191 
 192         switch (resume_skip_check(td, dnp, zb)) {
 193         case RESUME_SKIP_ALL:
 194                 return (0);
 195         case RESUME_SKIP_CHILDREN:
 196                 goto post;
 197         case RESUME_SKIP_NONE:
 198                 break;
 199         default:
 200                 ASSERT(0);
 201         }
 202 
 203         if (BP_IS_HOLE(bp)) {
 204                 err = td->td_func(td->td_spa, NULL, NULL, pbuf, zb, dnp,
 205                     td->td_arg);
 206                 return (err);
 207         }
 208 
 209         if (bp->blk_birth <= td->td_min_txg)
 210                 return (0);
 211 
 212         if (pd && !pd->pd_exited &&
 213             ((pd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
 214             BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) {
 215                 mutex_enter(&pd->pd_mtx);
 216                 ASSERT(pd->pd_blks_fetched >= 0);
 217                 while (pd->pd_blks_fetched == 0 && !pd->pd_exited)
 218                         cv_wait(&pd->pd_cv, &pd->pd_mtx);
 219                 pd->pd_blks_fetched--;
 220                 cv_broadcast(&pd->pd_cv);
 221                 mutex_exit(&pd->pd_mtx);
 222         }
 223 
 224         if (td->td_flags & TRAVERSE_PRE) {
 225                 err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp,
 226                     td->td_arg);
 227                 if (err == TRAVERSE_VISIT_NO_CHILDREN)
 228                         return (0);
 229                 if (err == ERESTART)
 230                         pause = B_TRUE; /* handle pausing at a common point */
 231                 if (err != 0)
 232                         goto post;
 233         }
 234 
 235         if (BP_GET_LEVEL(bp) > 0) {
 236                 uint32_t flags = ARC_WAIT;
 237                 int i;
 238                 blkptr_t *cbp;
 239                 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
 240 
 241                 err = dsl_read(NULL, td->td_spa, bp, pbuf,
 242                     arc_getbuf_func, &buf,
 243                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 244                 if (err)
 245                         return (err);
 246 
 247                 /* recursively visitbp() blocks below this */
 248                 cbp = buf->b_data;
 249                 for (i = 0; i < epb; i++, cbp++) {
 250                         SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
 251                             zb->zb_level - 1,
 252                             zb->zb_blkid * epb + i);
 253                         err = traverse_visitbp(td, dnp, buf, cbp, &czb);
 254                         if (err) {
 255                                 if (!hard)
 256                                         break;
 257                                 lasterr = err;
 258                         }
 259                 }
 260         } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
 261                 uint32_t flags = ARC_WAIT;
 262                 int i;
 263                 int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
 264 
 265                 err = dsl_read(NULL, td->td_spa, bp, pbuf,
 266                     arc_getbuf_func, &buf,
 267                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 268                 if (err)
 269                         return (err);
 270 
 271                 /* recursively visitbp() blocks below this */
 272                 dnp = buf->b_data;
 273                 for (i = 0; i < epb; i++, dnp++) {
 274                         err = traverse_dnode(td, dnp, buf, zb->zb_objset,
 275                             zb->zb_blkid * epb + i);
 276                         if (err) {
 277                                 if (!hard)
 278                                         break;
 279                                 lasterr = err;
 280                         }
 281                 }
 282         } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
 283                 uint32_t flags = ARC_WAIT;
 284                 objset_phys_t *osp;
 285                 dnode_phys_t *dnp;
 286 
 287                 err = dsl_read_nolock(NULL, td->td_spa, bp,
 288                     arc_getbuf_func, &buf,
 289                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 290                 if (err)
 291                         return (err);
 292 
 293                 osp = buf->b_data;
 294                 dnp = &osp->os_meta_dnode;
 295                 err = traverse_dnode(td, dnp, buf, zb->zb_objset,
 296                     DMU_META_DNODE_OBJECT);
 297                 if (err && hard) {
 298                         lasterr = err;
 299                         err = 0;
 300                 }
 301                 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
 302                         dnp = &osp->os_userused_dnode;
 303                         err = traverse_dnode(td, dnp, buf, zb->zb_objset,
 304                             DMU_USERUSED_OBJECT);
 305                 }
 306                 if (err && hard) {
 307                         lasterr = err;
 308                         err = 0;
 309                 }
 310                 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
 311                         dnp = &osp->os_groupused_dnode;
 312                         err = traverse_dnode(td, dnp, buf, zb->zb_objset,
 313                             DMU_GROUPUSED_OBJECT);
 314                 }
 315         }
 316 
 317         if (buf)
 318                 (void) arc_buf_remove_ref(buf, &buf);
 319 
 320 post:
 321         if (err == 0 && lasterr == 0 && (td->td_flags & TRAVERSE_POST)) {
 322                 err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp,
 323                     td->td_arg);
 324                 if (err == ERESTART)
 325                         pause = B_TRUE;
 326         }
 327 
 328         if (pause && td->td_resume != NULL) {
 329                 ASSERT3U(err, ==, ERESTART);
 330                 ASSERT(!hard);
 331                 traverse_pause(td, zb);
 332         }
 333 
 334         return (err != 0 ? err : lasterr);
 335 }
 336 
 337 static int
 338 traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
 339     arc_buf_t *buf, uint64_t objset, uint64_t object)
 340 {
 341         int j, err = 0, lasterr = 0;
 342         zbookmark_t czb;
 343         boolean_t hard = (td->td_flags & TRAVERSE_HARD);
 344 
 345         for (j = 0; j < dnp->dn_nblkptr; j++) {
 346                 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
 347                 err = traverse_visitbp(td, dnp, buf,
 348                     (blkptr_t *)&dnp->dn_blkptr[j], &czb);
 349                 if (err) {
 350                         if (!hard)
 351                                 break;
 352                         lasterr = err;
 353                 }
 354         }
 355 
 356         if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
 357                 SET_BOOKMARK(&czb, objset,
 358                     object, 0, DMU_SPILL_BLKID);
 359                 err = traverse_visitbp(td, dnp, buf,
 360                     (blkptr_t *)&dnp->dn_spill, &czb);
 361                 if (err) {
 362                         if (!hard)
 363                                 return (err);
 364                         lasterr = err;
 365                 }
 366         }
 367         return (err != 0 ? err : lasterr);
 368 }
 369 
 370 /* ARGSUSED */
 371 static int
 372 traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 373     arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp,
 374     void *arg)
 375 {
 376         prefetch_data_t *pfd = arg;
 377         uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
 378 
 379         ASSERT(pfd->pd_blks_fetched >= 0);
 380         if (pfd->pd_cancel)
 381                 return (EINTR);
 382 
 383         if (bp == NULL || !((pfd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
 384             BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0) ||
 385             BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
 386                 return (0);
 387 
 388         mutex_enter(&pfd->pd_mtx);
 389         while (!pfd->pd_cancel && pfd->pd_blks_fetched >= pfd->pd_blks_max)
 390                 cv_wait(&pfd->pd_cv, &pfd->pd_mtx);
 391         pfd->pd_blks_fetched++;
 392         cv_broadcast(&pfd->pd_cv);
 393         mutex_exit(&pfd->pd_mtx);
 394 
 395         (void) dsl_read(NULL, spa, bp, pbuf, NULL, NULL,
 396             ZIO_PRIORITY_ASYNC_READ,
 397             ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
 398             &aflags, zb);
 399 
 400         return (0);
 401 }
 402 
 403 static void
 404 traverse_prefetch_thread(void *arg)
 405 {
 406         traverse_data_t *td_main = arg;
 407         traverse_data_t td = *td_main;
 408         zbookmark_t czb;
 409 
 410         td.td_func = traverse_prefetcher;
 411         td.td_arg = td_main->td_pfd;
 412         td.td_pfd = NULL;
 413 
 414         SET_BOOKMARK(&czb, td.td_objset,
 415             ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
 416         (void) traverse_visitbp(&td, NULL, NULL, td.td_rootbp, &czb);
 417 
 418         mutex_enter(&td_main->td_pfd->pd_mtx);
 419         td_main->td_pfd->pd_exited = B_TRUE;
 420         cv_broadcast(&td_main->td_pfd->pd_cv);
 421         mutex_exit(&td_main->td_pfd->pd_mtx);
 422 }
 423 
 424 /*
 425  * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
 426  * in syncing context).
 427  */
 428 static int
 429 traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
 430     uint64_t txg_start, zbookmark_t *resume, int flags,
 431     blkptr_cb_t func, void *arg)
 432 {
 433         traverse_data_t td;
 434         prefetch_data_t pd = { 0 };
 435         zbookmark_t czb;
 436         int err;
 437 
 438         ASSERT(ds == NULL || objset == ds->ds_object);
 439         ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST));
 440 
 441         td.td_spa = spa;
 442         td.td_objset = objset;
 443         td.td_rootbp = rootbp;
 444         td.td_min_txg = txg_start;
 445         td.td_resume = resume;
 446         td.td_func = func;
 447         td.td_arg = arg;
 448         td.td_pfd = &pd;
 449         td.td_flags = flags;
 450 
 451         pd.pd_blks_max = zfs_pd_blks_max;
 452         pd.pd_flags = flags;
 453         mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL);
 454         cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL);
 455 
 456         /* See comment on ZIL traversal in dsl_scan_visitds. */
 457         if (ds != NULL && !dsl_dataset_is_snapshot(ds)) {
 458                 objset_t *os;
 459 
 460                 err = dmu_objset_from_ds(ds, &os);
 461                 if (err)
 462                         return (err);
 463 
 464                 traverse_zil(&td, &os->os_zil_header);
 465         }
 466 
 467         if (!(flags & TRAVERSE_PREFETCH) ||
 468             0 == taskq_dispatch(system_taskq, traverse_prefetch_thread,
 469             &td, TQ_NOQUEUE))
 470                 pd.pd_exited = B_TRUE;
 471 
 472         SET_BOOKMARK(&czb, td.td_objset,
 473             ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
 474         err = traverse_visitbp(&td, NULL, NULL, rootbp, &czb);
 475 
 476         mutex_enter(&pd.pd_mtx);
 477         pd.pd_cancel = B_TRUE;
 478         cv_broadcast(&pd.pd_cv);
 479         while (!pd.pd_exited)
 480                 cv_wait(&pd.pd_cv, &pd.pd_mtx);
 481         mutex_exit(&pd.pd_mtx);
 482 
 483         mutex_destroy(&pd.pd_mtx);
 484         cv_destroy(&pd.pd_cv);
 485 
 486         return (err);
 487 }
 488 
 489 /*
 490  * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
 491  * in syncing context).
 492  */
 493 int
 494 traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags,
 495     blkptr_cb_t func, void *arg)
 496 {
 497         return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object,
 498             &ds->ds_phys->ds_bp, txg_start, NULL, flags, func, arg));
 499 }
 500 
 501 int
 502 traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr,
 503     uint64_t txg_start, zbookmark_t *resume, int flags,
 504     blkptr_cb_t func, void *arg)
 505 {
 506         return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET,
 507             blkptr, txg_start, resume, flags, func, arg));
 508 }
 509 
 510 /*
 511  * NB: pool must not be changing on-disk (eg, from zdb or sync context).
 512  */
 513 int
 514 traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
 515     blkptr_cb_t func, void *arg)
 516 {
 517         int err, lasterr = 0;
 518         uint64_t obj;
 519         dsl_pool_t *dp = spa_get_dsl(spa);
 520         objset_t *mos = dp->dp_meta_objset;
 521         boolean_t hard = (flags & TRAVERSE_HARD);
 522 
 523         /* visit the MOS */
 524         err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa),
 525             txg_start, NULL, flags, func, arg);
 526         if (err)
 527                 return (err);
 528 
 529         /* visit each dataset */
 530         for (obj = 1; err == 0 || (err != ESRCH && hard);
 531             err = dmu_object_next(mos, &obj, FALSE, txg_start)) {
 532                 dmu_object_info_t doi;
 533 
 534                 err = dmu_object_info(mos, obj, &doi);
 535                 if (err) {
 536                         if (!hard)
 537                                 return (err);
 538                         lasterr = err;
 539                         continue;
 540                 }
 541 
 542                 if (doi.doi_type == DMU_OT_DSL_DATASET) {
 543                         dsl_dataset_t *ds;
 544                         uint64_t txg = txg_start;
 545 
 546                         rw_enter(&dp->dp_config_rwlock, RW_READER);
 547                         err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
 548                         rw_exit(&dp->dp_config_rwlock);
 549                         if (err) {
 550                                 if (!hard)
 551                                         return (err);
 552                                 lasterr = err;
 553                                 continue;
 554                         }
 555                         if (ds->ds_phys->ds_prev_snap_txg > txg)
 556                                 txg = ds->ds_phys->ds_prev_snap_txg;
 557                         err = traverse_dataset(ds, txg, flags, func, arg);
 558                         dsl_dataset_rele(ds, FTAG);
 559                         if (err) {
 560                                 if (!hard)
 561                                         return (err);
 562                                 lasterr = err;
 563                         }
 564                 }
 565         }
 566         if (err == ESRCH)
 567                 err = 0;
 568         return (err != 0 ? err : lasterr);
 569 }