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, 2014 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 #include <sys/zfeature.h>
40
41 int32_t zfs_pd_bytes_max = 50 * 1024 * 1024; /* 50MB */
42
43 typedef struct prefetch_data {
44 kmutex_t pd_mtx;
45 kcondvar_t pd_cv;
46 int32_t pd_bytes_fetched;
47 int pd_flags;
48 boolean_t pd_cancel;
49 boolean_t pd_exited;
50 zbookmark_phys_t pd_resume;
51 } prefetch_data_t;
52
53 typedef struct traverse_data {
54 spa_t *td_spa;
55 uint64_t td_objset;
56 blkptr_t *td_rootbp;
57 uint64_t td_min_txg;
58 zbookmark_phys_t *td_resume;
59 int td_flags;
60 prefetch_data_t *td_pfd;
61 boolean_t td_paused;
62 uint64_t td_hole_birth_enabled_txg;
63 blkptr_cb_t *td_func;
64 void *td_arg;
65 boolean_t td_realloc_possible;
66 } traverse_data_t;
67
68 static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
69 uint64_t objset, uint64_t object);
70 static void prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *,
71 uint64_t objset, uint64_t object);
72
73 static int
traverse_zil_block(zilog_t * zilog,blkptr_t * bp,void * arg,uint64_t claim_txg)74 traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
75 {
76 traverse_data_t *td = arg;
77 zbookmark_phys_t zb;
78
79 if (BP_IS_HOLE(bp))
80 return (0);
81
82 if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
83 return (0);
84
85 SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
86 bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
87
88 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg);
89
90 return (0);
91 }
92
93 static int
traverse_zil_record(zilog_t * zilog,lr_t * lrc,void * arg,uint64_t claim_txg)94 traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
95 {
96 traverse_data_t *td = arg;
97
98 if (lrc->lrc_txtype == TX_WRITE) {
99 lr_write_t *lr = (lr_write_t *)lrc;
100 blkptr_t *bp = &lr->lr_blkptr;
101 zbookmark_phys_t zb;
102
103 if (BP_IS_HOLE(bp))
104 return (0);
105
106 if (claim_txg == 0 || bp->blk_birth < claim_txg)
107 return (0);
108
109 SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid,
110 ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
111
112 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL,
113 td->td_arg);
114 }
115 return (0);
116 }
117
118 static void
traverse_zil(traverse_data_t * td,zil_header_t * zh)119 traverse_zil(traverse_data_t *td, zil_header_t *zh)
120 {
121 uint64_t claim_txg = zh->zh_claim_txg;
122 zilog_t *zilog;
123
124 /*
125 * We only want to visit blocks that have been claimed but not yet
126 * replayed; plus, in read-only mode, blocks that are already stable.
127 */
128 if (claim_txg == 0 && spa_writeable(td->td_spa))
129 return;
130
131 zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);
132
133 (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
134 claim_txg);
135
136 zil_free(zilog);
137 }
138
139 typedef enum resume_skip {
140 RESUME_SKIP_ALL,
141 RESUME_SKIP_NONE,
142 RESUME_SKIP_CHILDREN
143 } resume_skip_t;
144
145 /*
146 * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and
147 * the block indicated by zb does not need to be visited at all. Returns
148 * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the
149 * resume point. This indicates that this block should be visited but not its
150 * children (since they must have been visited in a previous traversal).
151 * Otherwise returns RESUME_SKIP_NONE.
152 */
153 static resume_skip_t
resume_skip_check(traverse_data_t * td,const dnode_phys_t * dnp,const zbookmark_phys_t * zb)154 resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp,
155 const zbookmark_phys_t *zb)
156 {
157 if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) {
158 /*
159 * If we already visited this bp & everything below,
160 * don't bother doing it again.
161 */
162 if (zbookmark_subtree_completed(dnp, zb, td->td_resume))
163 return (RESUME_SKIP_ALL);
164
165 /*
166 * If we found the block we're trying to resume from, zero
167 * the bookmark out to indicate that we have resumed.
168 */
169 if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) {
170 bzero(td->td_resume, sizeof (*zb));
171 if (td->td_flags & TRAVERSE_POST)
172 return (RESUME_SKIP_CHILDREN);
173 }
174 }
175 return (RESUME_SKIP_NONE);
176 }
177
178 static void
traverse_prefetch_metadata(traverse_data_t * td,const blkptr_t * bp,const zbookmark_phys_t * zb)179 traverse_prefetch_metadata(traverse_data_t *td,
180 const blkptr_t *bp, const zbookmark_phys_t *zb)
181 {
182 arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
183
184 if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA))
185 return;
186 /*
187 * If we are in the process of resuming, don't prefetch, because
188 * some children will not be needed (and in fact may have already
189 * been freed).
190 */
191 if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume))
192 return;
193 if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg)
194 return;
195 if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)
196 return;
197
198 (void) arc_read(NULL, td->td_spa, bp, NULL, NULL,
199 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
200 }
201
202 static boolean_t
prefetch_needed(prefetch_data_t * pfd,const blkptr_t * bp)203 prefetch_needed(prefetch_data_t *pfd, const blkptr_t *bp)
204 {
205 ASSERT(pfd->pd_flags & TRAVERSE_PREFETCH_DATA);
206 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) ||
207 BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
208 return (B_FALSE);
209 return (B_TRUE);
210 }
211
212 static int
traverse_visitbp(traverse_data_t * td,const dnode_phys_t * dnp,const blkptr_t * bp,const zbookmark_phys_t * zb)213 traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
214 const blkptr_t *bp, const zbookmark_phys_t *zb)
215 {
216 zbookmark_phys_t czb;
217 int err = 0;
218 arc_buf_t *buf = NULL;
219 prefetch_data_t *pd = td->td_pfd;
220 boolean_t hard = td->td_flags & TRAVERSE_HARD;
221
222 switch (resume_skip_check(td, dnp, zb)) {
223 case RESUME_SKIP_ALL:
224 return (0);
225 case RESUME_SKIP_CHILDREN:
226 goto post;
227 case RESUME_SKIP_NONE:
228 break;
229 default:
230 ASSERT(0);
231 }
232
233 if (bp->blk_birth == 0 && (!td->td_realloc_possible ||
234 zb->zb_object == DMU_META_DNODE_OBJECT)) {
235 /*
236 * Since this block has a birth time of 0 it must be one of two
237 * things: a hole created before the SPA_FEATURE_HOLE_BIRTH
238 * feature was enabled, or a hole which has always been a hole
239 * in an object. This occurs when an object is created and then
240 * a write is performed that leaves part of the file as zeroes.
241 * Normally we don't care about this case, but if this object
242 * has the same object number as an object that has been
243 * destroyed, callers of dmu_traverse may not be able to tell
244 * that the object has been recreated. Thus, we must give them
245 * all of the holes with birth == 0 in any objects that may have
246 * been recreated. Thus, we cannot skip the block if it is
247 * possible the object has been recreated. If it isn't, then if
248 * SPA_FEATURE_HOLE_BIRTH was enabled before the min_txg for
249 * this traveral we know the hole must have been created before
250 * the min_txg for this traveral, so we can skip it. If
251 * SPA_FEATURE_HOLE_BIRTH was enabled after the min_txg for this
252 * traveral we cannot tell if the hole was created before or
253 * after the min_txg for this traversal, so we cannot skip it.
254 * Note that the meta-dnode cannot be reallocated, so we needn't
255 * worry about that case.
256 */
257 if (td->td_hole_birth_enabled_txg <= td->td_min_txg)
258 return (0);
259 } else if (bp->blk_birth <= td->td_min_txg) {
260 return (0);
261 }
262
263 if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) {
264 uint64_t size = BP_GET_LSIZE(bp);
265 mutex_enter(&pd->pd_mtx);
266 ASSERT(pd->pd_bytes_fetched >= 0);
267 while (pd->pd_bytes_fetched < size && !pd->pd_exited)
268 cv_wait(&pd->pd_cv, &pd->pd_mtx);
269 pd->pd_bytes_fetched -= size;
270 cv_broadcast(&pd->pd_cv);
271 mutex_exit(&pd->pd_mtx);
272 }
273
274 if (BP_IS_HOLE(bp)) {
275 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
276 if (err != 0)
277 goto post;
278 return (0);
279 }
280
281 if (td->td_flags & TRAVERSE_PRE) {
282 err = td->td_func(td->td_spa, NULL, bp, zb, dnp,
283 td->td_arg);
284 if (err == TRAVERSE_VISIT_NO_CHILDREN)
285 return (0);
286 if (err != 0)
287 goto post;
288 }
289
290 if (BP_GET_LEVEL(bp) > 0) {
291 arc_flags_t flags = ARC_FLAG_WAIT;
292 int i;
293 blkptr_t *cbp;
294 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
295
296 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
297 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
298 if (err != 0)
299 goto post;
300 cbp = buf->b_data;
301
302 for (i = 0; i < epb; i++) {
303 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
304 zb->zb_level - 1,
305 zb->zb_blkid * epb + i);
306 traverse_prefetch_metadata(td, &cbp[i], &czb);
307 }
308
309 /* recursively visitbp() blocks below this */
310 for (i = 0; i < epb; i++) {
311 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
312 zb->zb_level - 1,
313 zb->zb_blkid * epb + i);
314 err = traverse_visitbp(td, dnp, &cbp[i], &czb);
315 if (err != 0)
316 break;
317 }
318 } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
319 arc_flags_t flags = ARC_FLAG_WAIT;
320 int i;
321 int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
322
323 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
324 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
325 if (err != 0)
326 goto post;
327 dnode_phys_t *child_dnp = buf->b_data;
328
329 for (i = 0; i < epb; i++) {
330 prefetch_dnode_metadata(td, &child_dnp[i],
331 zb->zb_objset, zb->zb_blkid * epb + i);
332 }
333
334 /* recursively visitbp() blocks below this */
335 for (i = 0; i < epb; i++) {
336 err = traverse_dnode(td, &child_dnp[i],
337 zb->zb_objset, zb->zb_blkid * epb + i);
338 if (err != 0)
339 break;
340 }
341 } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
342 arc_flags_t flags = ARC_FLAG_WAIT;
343
344 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
345 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
346 if (err != 0)
347 goto post;
348
349 objset_phys_t *osp = buf->b_data;
350 prefetch_dnode_metadata(td, &osp->os_meta_dnode, zb->zb_objset,
351 DMU_META_DNODE_OBJECT);
352 /*
353 * See the block comment above for the goal of this variable.
354 * If the maxblkid of the meta-dnode is 0, then we know that
355 * we've never had more than DNODES_PER_BLOCK objects in the
356 * dataset, which means we can't have reused any object ids.
357 */
358 if (osp->os_meta_dnode.dn_maxblkid == 0)
359 td->td_realloc_possible = B_FALSE;
360
361 if (arc_buf_size(buf) >= sizeof (objset_phys_t)) {
362 prefetch_dnode_metadata(td, &osp->os_groupused_dnode,
363 zb->zb_objset, DMU_GROUPUSED_OBJECT);
364 prefetch_dnode_metadata(td, &osp->os_userused_dnode,
365 zb->zb_objset, DMU_USERUSED_OBJECT);
366 }
367
368 err = traverse_dnode(td, &osp->os_meta_dnode, zb->zb_objset,
369 DMU_META_DNODE_OBJECT);
370 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
371 err = traverse_dnode(td, &osp->os_groupused_dnode,
372 zb->zb_objset, DMU_GROUPUSED_OBJECT);
373 }
374 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
375 err = traverse_dnode(td, &osp->os_userused_dnode,
376 zb->zb_objset, DMU_USERUSED_OBJECT);
377 }
378 }
379
380 if (buf)
381 (void) arc_buf_remove_ref(buf, &buf);
382
383 post:
384 if (err == 0 && (td->td_flags & TRAVERSE_POST))
385 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
386
387 if (hard && (err == EIO || err == ECKSUM)) {
388 /*
389 * Ignore this disk error as requested by the HARD flag,
390 * and continue traversal.
391 */
392 err = 0;
393 }
394
395 /*
396 * If we are stopping here, set td_resume.
397 */
398 if (td->td_resume != NULL && err != 0 && !td->td_paused) {
399 td->td_resume->zb_objset = zb->zb_objset;
400 td->td_resume->zb_object = zb->zb_object;
401 td->td_resume->zb_level = 0;
402 /*
403 * If we have stopped on an indirect block (e.g. due to
404 * i/o error), we have not visited anything below it.
405 * Set the bookmark to the first level-0 block that we need
406 * to visit. This way, the resuming code does not need to
407 * deal with resuming from indirect blocks.
408 *
409 * Note, if zb_level <= 0, dnp may be NULL, so we don't want
410 * to dereference it.
411 */
412 td->td_resume->zb_blkid = zb->zb_blkid;
413 if (zb->zb_level > 0) {
414 td->td_resume->zb_blkid <<= zb->zb_level *
415 (dnp->dn_indblkshift - SPA_BLKPTRSHIFT);
416 }
417 td->td_paused = B_TRUE;
418 }
419
420 return (err);
421 }
422
423 static void
prefetch_dnode_metadata(traverse_data_t * td,const dnode_phys_t * dnp,uint64_t objset,uint64_t object)424 prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp,
425 uint64_t objset, uint64_t object)
426 {
427 int j;
428 zbookmark_phys_t czb;
429
430 for (j = 0; j < dnp->dn_nblkptr; j++) {
431 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
432 traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb);
433 }
434
435 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
436 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
437 traverse_prefetch_metadata(td, &dnp->dn_spill, &czb);
438 }
439 }
440
441 static int
traverse_dnode(traverse_data_t * td,const dnode_phys_t * dnp,uint64_t objset,uint64_t object)442 traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
443 uint64_t objset, uint64_t object)
444 {
445 int j, err = 0;
446 zbookmark_phys_t czb;
447
448 if (object != DMU_META_DNODE_OBJECT && td->td_resume != NULL &&
449 object < td->td_resume->zb_object)
450 return (0);
451
452 if (td->td_flags & TRAVERSE_PRE) {
453 SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL,
454 ZB_DNODE_BLKID);
455 err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp,
456 td->td_arg);
457 if (err == TRAVERSE_VISIT_NO_CHILDREN)
458 return (0);
459 if (err != 0)
460 return (err);
461 }
462
463 for (j = 0; j < dnp->dn_nblkptr; j++) {
464 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
465 err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb);
466 if (err != 0)
467 break;
468 }
469
470 if (err == 0 && (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) {
471 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
472 err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb);
473 }
474
475 if (err == 0 && (td->td_flags & TRAVERSE_POST)) {
476 SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL,
477 ZB_DNODE_BLKID);
478 err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp,
479 td->td_arg);
480 if (err == TRAVERSE_VISIT_NO_CHILDREN)
481 return (0);
482 if (err != 0)
483 return (err);
484 }
485 return (err);
486 }
487
488 /* ARGSUSED */
489 static int
traverse_prefetcher(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)490 traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
491 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
492 {
493 prefetch_data_t *pfd = arg;
494 arc_flags_t aflags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
495
496 ASSERT(pfd->pd_bytes_fetched >= 0);
497 if (bp == NULL)
498 return (0);
499 if (pfd->pd_cancel)
500 return (SET_ERROR(EINTR));
501
502 if (!prefetch_needed(pfd, bp))
503 return (0);
504
505 mutex_enter(&pfd->pd_mtx);
506 while (!pfd->pd_cancel && pfd->pd_bytes_fetched >= zfs_pd_bytes_max)
507 cv_wait(&pfd->pd_cv, &pfd->pd_mtx);
508 pfd->pd_bytes_fetched += BP_GET_LSIZE(bp);
509 cv_broadcast(&pfd->pd_cv);
510 mutex_exit(&pfd->pd_mtx);
511
512 (void) arc_read(NULL, spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
513 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, zb);
514
515 return (0);
516 }
517
518 static void
traverse_prefetch_thread(void * arg)519 traverse_prefetch_thread(void *arg)
520 {
521 traverse_data_t *td_main = arg;
522 traverse_data_t td = *td_main;
523 zbookmark_phys_t czb;
524
525 td.td_func = traverse_prefetcher;
526 td.td_arg = td_main->td_pfd;
527 td.td_pfd = NULL;
528 td.td_resume = &td_main->td_pfd->pd_resume;
529
530 SET_BOOKMARK(&czb, td.td_objset,
531 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
532 (void) traverse_visitbp(&td, NULL, td.td_rootbp, &czb);
533
534 mutex_enter(&td_main->td_pfd->pd_mtx);
535 td_main->td_pfd->pd_exited = B_TRUE;
536 cv_broadcast(&td_main->td_pfd->pd_cv);
537 mutex_exit(&td_main->td_pfd->pd_mtx);
538 }
539
540 /*
541 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
542 * in syncing context).
543 */
544 static int
traverse_impl(spa_t * spa,dsl_dataset_t * ds,uint64_t objset,blkptr_t * rootbp,uint64_t txg_start,zbookmark_phys_t * resume,int flags,blkptr_cb_t func,void * arg)545 traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
546 uint64_t txg_start, zbookmark_phys_t *resume, int flags,
547 blkptr_cb_t func, void *arg)
548 {
549 traverse_data_t td;
550 prefetch_data_t pd = { 0 };
551 zbookmark_phys_t czb;
552 int err;
553
554 ASSERT(ds == NULL || objset == ds->ds_object);
555 ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST));
556
557 td.td_spa = spa;
558 td.td_objset = objset;
559 td.td_rootbp = rootbp;
560 td.td_min_txg = txg_start;
561 td.td_resume = resume;
562 td.td_func = func;
563 td.td_arg = arg;
564 td.td_pfd = &pd;
565 td.td_flags = flags;
566 td.td_paused = B_FALSE;
567 td.td_realloc_possible = (txg_start == 0 ? B_FALSE : B_TRUE);
568
569 if (spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) {
570 VERIFY(spa_feature_enabled_txg(spa,
571 SPA_FEATURE_HOLE_BIRTH, &td.td_hole_birth_enabled_txg));
572 } else {
573 td.td_hole_birth_enabled_txg = UINT64_MAX;
574 }
575
576 pd.pd_flags = flags;
577 if (resume != NULL)
578 pd.pd_resume = *resume;
579 mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL);
580 cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL);
581
582 /* See comment on ZIL traversal in dsl_scan_visitds. */
583 if (ds != NULL && !ds->ds_is_snapshot && !BP_IS_HOLE(rootbp)) {
584 arc_flags_t flags = ARC_FLAG_WAIT;
585 objset_phys_t *osp;
586 arc_buf_t *buf;
587
588 err = arc_read(NULL, td.td_spa, rootbp,
589 arc_getbuf_func, &buf,
590 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, NULL);
591 if (err != 0)
592 return (err);
593
594 osp = buf->b_data;
595 traverse_zil(&td, &osp->os_zil_header);
596 (void) arc_buf_remove_ref(buf, &buf);
597 }
598
599 if (!(flags & TRAVERSE_PREFETCH_DATA) ||
600 0 == taskq_dispatch(system_taskq, traverse_prefetch_thread,
601 &td, TQ_NOQUEUE))
602 pd.pd_exited = B_TRUE;
603
604 SET_BOOKMARK(&czb, td.td_objset,
605 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
606 err = traverse_visitbp(&td, NULL, rootbp, &czb);
607
608 mutex_enter(&pd.pd_mtx);
609 pd.pd_cancel = B_TRUE;
610 cv_broadcast(&pd.pd_cv);
611 while (!pd.pd_exited)
612 cv_wait(&pd.pd_cv, &pd.pd_mtx);
613 mutex_exit(&pd.pd_mtx);
614
615 mutex_destroy(&pd.pd_mtx);
616 cv_destroy(&pd.pd_cv);
617
618 return (err);
619 }
620
621 /*
622 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
623 * in syncing context).
624 */
625 int
traverse_dataset_resume(dsl_dataset_t * ds,uint64_t txg_start,zbookmark_phys_t * resume,int flags,blkptr_cb_t func,void * arg)626 traverse_dataset_resume(dsl_dataset_t *ds, uint64_t txg_start,
627 zbookmark_phys_t *resume,
628 int flags, blkptr_cb_t func, void *arg)
629 {
630 return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object,
631 &dsl_dataset_phys(ds)->ds_bp, txg_start, resume, flags, func, arg));
632 }
633
634 int
traverse_dataset(dsl_dataset_t * ds,uint64_t txg_start,int flags,blkptr_cb_t func,void * arg)635 traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start,
636 int flags, blkptr_cb_t func, void *arg)
637 {
638 return (traverse_dataset_resume(ds, txg_start, NULL, flags, func, arg));
639 }
640
641 int
traverse_dataset_destroyed(spa_t * spa,blkptr_t * blkptr,uint64_t txg_start,zbookmark_phys_t * resume,int flags,blkptr_cb_t func,void * arg)642 traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr,
643 uint64_t txg_start, zbookmark_phys_t *resume, int flags,
644 blkptr_cb_t func, void *arg)
645 {
646 return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET,
647 blkptr, txg_start, resume, flags, func, arg));
648 }
649
650 /*
651 * NB: pool must not be changing on-disk (eg, from zdb or sync context).
652 */
653 int
traverse_pool(spa_t * spa,uint64_t txg_start,int flags,blkptr_cb_t func,void * arg)654 traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
655 blkptr_cb_t func, void *arg)
656 {
657 int err;
658 dsl_pool_t *dp = spa_get_dsl(spa);
659 objset_t *mos = dp->dp_meta_objset;
660 boolean_t hard = (flags & TRAVERSE_HARD);
661
662 /* visit the MOS */
663 err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa),
664 txg_start, NULL, flags, func, arg);
665 if (err != 0)
666 return (err);
667
668 /* visit each dataset */
669 for (uint64_t obj = 1; err == 0;
670 err = dmu_object_next(mos, &obj, B_FALSE, txg_start)) {
671 dmu_object_info_t doi;
672
673 err = dmu_object_info(mos, obj, &doi);
674 if (err != 0) {
675 if (hard)
676 continue;
677 break;
678 }
679
680 if (doi.doi_bonus_type == DMU_OT_DSL_DATASET) {
681 dsl_dataset_t *ds;
682 uint64_t txg = txg_start;
683
684 dsl_pool_config_enter(dp, FTAG);
685 err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
686 dsl_pool_config_exit(dp, FTAG);
687 if (err != 0) {
688 if (hard)
689 continue;
690 break;
691 }
692 if (dsl_dataset_phys(ds)->ds_prev_snap_txg > txg)
693 txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
694 err = traverse_dataset(ds, txg, flags, func, arg);
695 dsl_dataset_rele(ds, FTAG);
696 if (err != 0)
697 break;
698 }
699 }
700 if (err == ESRCH)
701 err = 0;
702 return (err);
703 }
704