]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/dmu_traverse.c
Project Quota on ZFS
[mirror_zfs.git] / module / zfs / dmu_traverse.c
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, 2016 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 int32_t send_holes_without_birth_time = 1;
43
44 typedef struct prefetch_data {
45 kmutex_t pd_mtx;
46 kcondvar_t pd_cv;
47 int32_t pd_bytes_fetched;
48 int pd_flags;
49 boolean_t pd_cancel;
50 boolean_t pd_exited;
51 zbookmark_phys_t pd_resume;
52 } prefetch_data_t;
53
54 typedef struct traverse_data {
55 spa_t *td_spa;
56 uint64_t td_objset;
57 blkptr_t *td_rootbp;
58 uint64_t td_min_txg;
59 zbookmark_phys_t *td_resume;
60 int td_flags;
61 prefetch_data_t *td_pfd;
62 boolean_t td_paused;
63 uint64_t td_hole_birth_enabled_txg;
64 blkptr_cb_t *td_func;
65 void *td_arg;
66 boolean_t td_realloc_possible;
67 } traverse_data_t;
68
69 static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
70 uint64_t objset, uint64_t object);
71 static void prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *,
72 uint64_t objset, uint64_t object);
73
74 static int
75 traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
76 {
77 traverse_data_t *td = arg;
78 zbookmark_phys_t zb;
79
80 if (BP_IS_HOLE(bp))
81 return (0);
82
83 if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
84 return (0);
85
86 SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
87 bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
88
89 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg);
90
91 return (0);
92 }
93
94 static int
95 traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
96 {
97 traverse_data_t *td = arg;
98
99 if (lrc->lrc_txtype == TX_WRITE) {
100 lr_write_t *lr = (lr_write_t *)lrc;
101 blkptr_t *bp = &lr->lr_blkptr;
102 zbookmark_phys_t zb;
103
104 if (BP_IS_HOLE(bp))
105 return (0);
106
107 if (claim_txg == 0 || bp->blk_birth < claim_txg)
108 return (0);
109
110 SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid,
111 ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
112
113 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL,
114 td->td_arg);
115 }
116 return (0);
117 }
118
119 static void
120 traverse_zil(traverse_data_t *td, zil_header_t *zh)
121 {
122 uint64_t claim_txg = zh->zh_claim_txg;
123 zilog_t *zilog;
124
125 /*
126 * We only want to visit blocks that have been claimed but not yet
127 * replayed; plus, in read-only mode, blocks that are already stable.
128 */
129 if (claim_txg == 0 && spa_writeable(td->td_spa))
130 return;
131
132 zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);
133
134 (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
135 claim_txg, !(td->td_flags & TRAVERSE_NO_DECRYPT));
136
137 zil_free(zilog);
138 }
139
140 typedef enum resume_skip {
141 RESUME_SKIP_ALL,
142 RESUME_SKIP_NONE,
143 RESUME_SKIP_CHILDREN
144 } resume_skip_t;
145
146 /*
147 * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and
148 * the block indicated by zb does not need to be visited at all. Returns
149 * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the
150 * resume point. This indicates that this block should be visited but not its
151 * children (since they must have been visited in a previous traversal).
152 * Otherwise returns RESUME_SKIP_NONE.
153 */
154 static resume_skip_t
155 resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp,
156 const zbookmark_phys_t *zb)
157 {
158 if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) {
159 /*
160 * If we already visited this bp & everything below,
161 * don't bother doing it again.
162 */
163 if (zbookmark_subtree_completed(dnp, zb, td->td_resume))
164 return (RESUME_SKIP_ALL);
165
166 /*
167 * If we found the block we're trying to resume from, zero
168 * the bookmark out to indicate that we have resumed.
169 */
170 if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) {
171 bzero(td->td_resume, sizeof (*zb));
172 if (td->td_flags & TRAVERSE_POST)
173 return (RESUME_SKIP_CHILDREN);
174 }
175 }
176 return (RESUME_SKIP_NONE);
177 }
178
179 static void
180 traverse_prefetch_metadata(traverse_data_t *td,
181 const blkptr_t *bp, const zbookmark_phys_t *zb)
182 {
183 arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
184 int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE;
185
186 if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA))
187 return;
188 /*
189 * If we are in the process of resuming, don't prefetch, because
190 * some children will not be needed (and in fact may have already
191 * been freed).
192 */
193 if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume))
194 return;
195 if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg)
196 return;
197 if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)
198 return;
199
200 if ((td->td_flags & TRAVERSE_NO_DECRYPT) && BP_IS_PROTECTED(bp))
201 zio_flags |= ZIO_FLAG_RAW;
202
203 (void) arc_read(NULL, td->td_spa, bp, NULL, NULL,
204 ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
205 }
206
207 static boolean_t
208 prefetch_needed(prefetch_data_t *pfd, const blkptr_t *bp)
209 {
210 ASSERT(pfd->pd_flags & TRAVERSE_PREFETCH_DATA);
211 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) ||
212 BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
213 return (B_FALSE);
214 return (B_TRUE);
215 }
216
217 static int
218 traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
219 const blkptr_t *bp, const zbookmark_phys_t *zb)
220 {
221 int err = 0;
222 arc_buf_t *buf = NULL;
223 prefetch_data_t *pd = td->td_pfd;
224
225 switch (resume_skip_check(td, dnp, zb)) {
226 case RESUME_SKIP_ALL:
227 return (0);
228 case RESUME_SKIP_CHILDREN:
229 goto post;
230 case RESUME_SKIP_NONE:
231 break;
232 default:
233 ASSERT(0);
234 }
235
236 if (bp->blk_birth == 0) {
237 /*
238 * Since this block has a birth time of 0 it must be one of
239 * two things: a hole created before the
240 * SPA_FEATURE_HOLE_BIRTH feature was enabled, or a hole
241 * which has always been a hole in an object.
242 *
243 * If a file is written sparsely, then the unwritten parts of
244 * the file were "always holes" -- that is, they have been
245 * holes since this object was allocated. However, we (and
246 * our callers) can not necessarily tell when an object was
247 * allocated. Therefore, if it's possible that this object
248 * was freed and then its object number reused, we need to
249 * visit all the holes with birth==0.
250 *
251 * If it isn't possible that the object number was reused,
252 * then if SPA_FEATURE_HOLE_BIRTH was enabled before we wrote
253 * all the blocks we will visit as part of this traversal,
254 * then this hole must have always existed, so we can skip
255 * it. We visit blocks born after (exclusive) td_min_txg.
256 *
257 * Note that the meta-dnode cannot be reallocated.
258 */
259 if (!send_holes_without_birth_time &&
260 (!td->td_realloc_possible ||
261 zb->zb_object == DMU_META_DNODE_OBJECT) &&
262 td->td_hole_birth_enabled_txg <= td->td_min_txg)
263 return (0);
264 } else if (bp->blk_birth <= td->td_min_txg) {
265 return (0);
266 }
267
268 if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) {
269 uint64_t size = BP_GET_LSIZE(bp);
270 mutex_enter(&pd->pd_mtx);
271 ASSERT(pd->pd_bytes_fetched >= 0);
272 while (pd->pd_bytes_fetched < size && !pd->pd_exited)
273 cv_wait_sig(&pd->pd_cv, &pd->pd_mtx);
274 pd->pd_bytes_fetched -= size;
275 cv_broadcast(&pd->pd_cv);
276 mutex_exit(&pd->pd_mtx);
277 }
278
279 if (BP_IS_HOLE(bp)) {
280 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
281 if (err != 0)
282 goto post;
283 return (0);
284 }
285
286 if (td->td_flags & TRAVERSE_PRE) {
287 err = td->td_func(td->td_spa, NULL, bp, zb, dnp,
288 td->td_arg);
289 if (err == TRAVERSE_VISIT_NO_CHILDREN)
290 return (0);
291 if (err != 0)
292 goto post;
293 }
294
295 if (BP_GET_LEVEL(bp) > 0) {
296 uint32_t flags = ARC_FLAG_WAIT;
297 int32_t i;
298 int32_t epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
299 zbookmark_phys_t *czb;
300
301 ASSERT(!BP_IS_PROTECTED(bp));
302
303 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
304 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
305 if (err != 0)
306 goto post;
307
308 czb = kmem_alloc(sizeof (zbookmark_phys_t), KM_SLEEP);
309
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 traverse_prefetch_metadata(td,
315 &((blkptr_t *)buf->b_data)[i], czb);
316 }
317
318 /* recursively visitbp() blocks below this */
319 for (i = 0; i < epb; i++) {
320 SET_BOOKMARK(czb, zb->zb_objset, zb->zb_object,
321 zb->zb_level - 1,
322 zb->zb_blkid * epb + i);
323 err = traverse_visitbp(td, dnp,
324 &((blkptr_t *)buf->b_data)[i], czb);
325 if (err != 0)
326 break;
327 }
328
329 kmem_free(czb, sizeof (zbookmark_phys_t));
330
331 } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
332 uint32_t flags = ARC_FLAG_WAIT;
333 uint32_t zio_flags = ZIO_FLAG_CANFAIL;
334 int32_t i;
335 int32_t epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
336 dnode_phys_t *child_dnp;
337
338 /*
339 * dnode blocks might have their bonus buffers encrypted, so
340 * we must be careful to honor TRAVERSE_NO_DECRYPT
341 */
342 if ((td->td_flags & TRAVERSE_NO_DECRYPT) && BP_IS_PROTECTED(bp))
343 zio_flags |= ZIO_FLAG_RAW;
344
345 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
346 ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
347 if (err != 0)
348 goto post;
349
350 child_dnp = buf->b_data;
351
352 for (i = 0; i < epb; i += child_dnp[i].dn_extra_slots + 1) {
353 prefetch_dnode_metadata(td, &child_dnp[i],
354 zb->zb_objset, zb->zb_blkid * epb + i);
355 }
356
357 /* recursively visitbp() blocks below this */
358 for (i = 0; i < epb; i += child_dnp[i].dn_extra_slots + 1) {
359 err = traverse_dnode(td, &child_dnp[i],
360 zb->zb_objset, zb->zb_blkid * epb + i);
361 if (err != 0)
362 break;
363 }
364 } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
365 uint32_t zio_flags = ZIO_FLAG_CANFAIL;
366 arc_flags_t flags = ARC_FLAG_WAIT;
367 objset_phys_t *osp;
368
369 if ((td->td_flags & TRAVERSE_NO_DECRYPT) && BP_IS_PROTECTED(bp))
370 zio_flags |= ZIO_FLAG_RAW;
371
372 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
373 ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
374 if (err != 0)
375 goto post;
376
377 osp = buf->b_data;
378 prefetch_dnode_metadata(td, &osp->os_meta_dnode, zb->zb_objset,
379 DMU_META_DNODE_OBJECT);
380 /*
381 * See the block comment above for the goal of this variable.
382 * If the maxblkid of the meta-dnode is 0, then we know that
383 * we've never had more than DNODES_PER_BLOCK objects in the
384 * dataset, which means we can't have reused any object ids.
385 */
386 if (osp->os_meta_dnode.dn_maxblkid == 0)
387 td->td_realloc_possible = B_FALSE;
388
389 if (OBJSET_BUF_HAS_USERUSED(buf)) {
390 if (OBJSET_BUF_HAS_PROJECTUSED(buf))
391 prefetch_dnode_metadata(td,
392 &osp->os_projectused_dnode,
393 zb->zb_objset, DMU_PROJECTUSED_OBJECT);
394 prefetch_dnode_metadata(td, &osp->os_groupused_dnode,
395 zb->zb_objset, DMU_GROUPUSED_OBJECT);
396 prefetch_dnode_metadata(td, &osp->os_userused_dnode,
397 zb->zb_objset, DMU_USERUSED_OBJECT);
398 }
399
400 err = traverse_dnode(td, &osp->os_meta_dnode, zb->zb_objset,
401 DMU_META_DNODE_OBJECT);
402 if (err == 0 && OBJSET_BUF_HAS_USERUSED(buf)) {
403 if (OBJSET_BUF_HAS_PROJECTUSED(buf))
404 err = traverse_dnode(td,
405 &osp->os_projectused_dnode, zb->zb_objset,
406 DMU_PROJECTUSED_OBJECT);
407 if (err == 0)
408 err = traverse_dnode(td,
409 &osp->os_groupused_dnode, zb->zb_objset,
410 DMU_GROUPUSED_OBJECT);
411 if (err == 0)
412 err = traverse_dnode(td,
413 &osp->os_userused_dnode, zb->zb_objset,
414 DMU_USERUSED_OBJECT);
415 }
416 }
417
418 if (buf)
419 arc_buf_destroy(buf, &buf);
420
421 post:
422 if (err == 0 && (td->td_flags & TRAVERSE_POST))
423 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
424
425 if ((td->td_flags & TRAVERSE_HARD) && (err == EIO || err == ECKSUM)) {
426 /*
427 * Ignore this disk error as requested by the HARD flag,
428 * and continue traversal.
429 */
430 err = 0;
431 }
432
433 /*
434 * If we are stopping here, set td_resume.
435 */
436 if (td->td_resume != NULL && err != 0 && !td->td_paused) {
437 td->td_resume->zb_objset = zb->zb_objset;
438 td->td_resume->zb_object = zb->zb_object;
439 td->td_resume->zb_level = 0;
440 /*
441 * If we have stopped on an indirect block (e.g. due to
442 * i/o error), we have not visited anything below it.
443 * Set the bookmark to the first level-0 block that we need
444 * to visit. This way, the resuming code does not need to
445 * deal with resuming from indirect blocks.
446 *
447 * Note, if zb_level <= 0, dnp may be NULL, so we don't want
448 * to dereference it.
449 */
450 td->td_resume->zb_blkid = zb->zb_blkid;
451 if (zb->zb_level > 0) {
452 td->td_resume->zb_blkid <<= zb->zb_level *
453 (dnp->dn_indblkshift - SPA_BLKPTRSHIFT);
454 }
455 td->td_paused = B_TRUE;
456 }
457
458 return (err);
459 }
460
461 static void
462 prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp,
463 uint64_t objset, uint64_t object)
464 {
465 int j;
466 zbookmark_phys_t czb;
467
468 for (j = 0; j < dnp->dn_nblkptr; j++) {
469 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
470 traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb);
471 }
472
473 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
474 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
475 traverse_prefetch_metadata(td, DN_SPILL_BLKPTR(dnp), &czb);
476 }
477 }
478
479 static int
480 traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
481 uint64_t objset, uint64_t object)
482 {
483 int j, err = 0;
484 zbookmark_phys_t czb;
485
486 if (object != DMU_META_DNODE_OBJECT && td->td_resume != NULL &&
487 object < td->td_resume->zb_object)
488 return (0);
489
490 if (td->td_flags & TRAVERSE_PRE) {
491 SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL,
492 ZB_DNODE_BLKID);
493 err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp,
494 td->td_arg);
495 if (err == TRAVERSE_VISIT_NO_CHILDREN)
496 return (0);
497 if (err != 0)
498 return (err);
499 }
500
501 for (j = 0; j < dnp->dn_nblkptr; j++) {
502 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
503 err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb);
504 if (err != 0)
505 break;
506 }
507
508 if (err == 0 && (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) {
509 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
510 err = traverse_visitbp(td, dnp, DN_SPILL_BLKPTR(dnp), &czb);
511 }
512
513 if (err == 0 && (td->td_flags & TRAVERSE_POST)) {
514 SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL,
515 ZB_DNODE_BLKID);
516 err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp,
517 td->td_arg);
518 if (err == TRAVERSE_VISIT_NO_CHILDREN)
519 return (0);
520 if (err != 0)
521 return (err);
522 }
523 return (err);
524 }
525
526 /* ARGSUSED */
527 static int
528 traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
529 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
530 {
531 prefetch_data_t *pfd = arg;
532 int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE;
533 arc_flags_t aflags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH |
534 ARC_FLAG_PRESCIENT_PREFETCH;
535
536 ASSERT(pfd->pd_bytes_fetched >= 0);
537 if (bp == NULL)
538 return (0);
539 if (pfd->pd_cancel)
540 return (SET_ERROR(EINTR));
541
542 if (!prefetch_needed(pfd, bp))
543 return (0);
544
545 mutex_enter(&pfd->pd_mtx);
546 while (!pfd->pd_cancel && pfd->pd_bytes_fetched >= zfs_pd_bytes_max)
547 cv_wait_sig(&pfd->pd_cv, &pfd->pd_mtx);
548 pfd->pd_bytes_fetched += BP_GET_LSIZE(bp);
549 cv_broadcast(&pfd->pd_cv);
550 mutex_exit(&pfd->pd_mtx);
551
552 if ((pfd->pd_flags & TRAVERSE_NO_DECRYPT) && BP_IS_PROTECTED(bp))
553 zio_flags |= ZIO_FLAG_RAW;
554
555 (void) arc_read(NULL, spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
556 zio_flags, &aflags, zb);
557
558 return (0);
559 }
560
561 static void
562 traverse_prefetch_thread(void *arg)
563 {
564 traverse_data_t *td_main = arg;
565 traverse_data_t td = *td_main;
566 zbookmark_phys_t czb;
567 fstrans_cookie_t cookie = spl_fstrans_mark();
568
569 td.td_func = traverse_prefetcher;
570 td.td_arg = td_main->td_pfd;
571 td.td_pfd = NULL;
572 td.td_resume = &td_main->td_pfd->pd_resume;
573
574 SET_BOOKMARK(&czb, td.td_objset,
575 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
576 (void) traverse_visitbp(&td, NULL, td.td_rootbp, &czb);
577
578 mutex_enter(&td_main->td_pfd->pd_mtx);
579 td_main->td_pfd->pd_exited = B_TRUE;
580 cv_broadcast(&td_main->td_pfd->pd_cv);
581 mutex_exit(&td_main->td_pfd->pd_mtx);
582 spl_fstrans_unmark(cookie);
583 }
584
585 /*
586 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
587 * in syncing context).
588 */
589 static int
590 traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
591 uint64_t txg_start, zbookmark_phys_t *resume, int flags,
592 blkptr_cb_t func, void *arg)
593 {
594 traverse_data_t *td;
595 prefetch_data_t *pd;
596 zbookmark_phys_t *czb;
597 int err;
598
599 ASSERT(ds == NULL || objset == ds->ds_object);
600 ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST));
601
602 td = kmem_alloc(sizeof (traverse_data_t), KM_SLEEP);
603 pd = kmem_zalloc(sizeof (prefetch_data_t), KM_SLEEP);
604 czb = kmem_alloc(sizeof (zbookmark_phys_t), KM_SLEEP);
605
606 td->td_spa = spa;
607 td->td_objset = objset;
608 td->td_rootbp = rootbp;
609 td->td_min_txg = txg_start;
610 td->td_resume = resume;
611 td->td_func = func;
612 td->td_arg = arg;
613 td->td_pfd = pd;
614 td->td_flags = flags;
615 td->td_paused = B_FALSE;
616 td->td_realloc_possible = (txg_start == 0 ? B_FALSE : B_TRUE);
617
618 if (spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) {
619 VERIFY(spa_feature_enabled_txg(spa,
620 SPA_FEATURE_HOLE_BIRTH, &td->td_hole_birth_enabled_txg));
621 } else {
622 td->td_hole_birth_enabled_txg = UINT64_MAX;
623 }
624
625 pd->pd_flags = flags;
626 if (resume != NULL)
627 pd->pd_resume = *resume;
628 mutex_init(&pd->pd_mtx, NULL, MUTEX_DEFAULT, NULL);
629 cv_init(&pd->pd_cv, NULL, CV_DEFAULT, NULL);
630
631 SET_BOOKMARK(czb, td->td_objset,
632 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
633
634 /* See comment on ZIL traversal in dsl_scan_visitds. */
635 if (ds != NULL && !ds->ds_is_snapshot && !BP_IS_HOLE(rootbp)) {
636 enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
637 uint32_t flags = ARC_FLAG_WAIT;
638 objset_phys_t *osp;
639 arc_buf_t *buf;
640
641 if ((td->td_flags & TRAVERSE_NO_DECRYPT) &&
642 BP_IS_PROTECTED(rootbp))
643 zio_flags |= ZIO_FLAG_RAW;
644
645 err = arc_read(NULL, td->td_spa, rootbp, arc_getbuf_func,
646 &buf, ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, czb);
647 if (err != 0) {
648 /*
649 * If both TRAVERSE_HARD and TRAVERSE_PRE are set,
650 * continue to visitbp so that td_func can be called
651 * in pre stage, and err will reset to zero.
652 */
653 if (!(td->td_flags & TRAVERSE_HARD) ||
654 !(td->td_flags & TRAVERSE_PRE))
655 return (err);
656 } else {
657 osp = buf->b_data;
658 traverse_zil(td, &osp->os_zil_header);
659 arc_buf_destroy(buf, &buf);
660 }
661 }
662
663 if (!(flags & TRAVERSE_PREFETCH_DATA) ||
664 taskq_dispatch(system_taskq, traverse_prefetch_thread,
665 td, TQ_NOQUEUE) == TASKQID_INVALID)
666 pd->pd_exited = B_TRUE;
667
668 err = traverse_visitbp(td, NULL, rootbp, czb);
669
670 mutex_enter(&pd->pd_mtx);
671 pd->pd_cancel = B_TRUE;
672 cv_broadcast(&pd->pd_cv);
673 while (!pd->pd_exited)
674 cv_wait_sig(&pd->pd_cv, &pd->pd_mtx);
675 mutex_exit(&pd->pd_mtx);
676
677 mutex_destroy(&pd->pd_mtx);
678 cv_destroy(&pd->pd_cv);
679
680 kmem_free(czb, sizeof (zbookmark_phys_t));
681 kmem_free(pd, sizeof (struct prefetch_data));
682 kmem_free(td, sizeof (struct traverse_data));
683
684 return (err);
685 }
686
687 /*
688 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
689 * in syncing context).
690 */
691 int
692 traverse_dataset_resume(dsl_dataset_t *ds, uint64_t txg_start,
693 zbookmark_phys_t *resume,
694 int flags, blkptr_cb_t func, void *arg)
695 {
696 return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object,
697 &dsl_dataset_phys(ds)->ds_bp, txg_start, resume, flags, func, arg));
698 }
699
700 int
701 traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start,
702 int flags, blkptr_cb_t func, void *arg)
703 {
704 return (traverse_dataset_resume(ds, txg_start, NULL, flags, func, arg));
705 }
706
707 int
708 traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr,
709 uint64_t txg_start, zbookmark_phys_t *resume, int flags,
710 blkptr_cb_t func, void *arg)
711 {
712 return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET,
713 blkptr, txg_start, resume, flags, func, arg));
714 }
715
716 /*
717 * NB: pool must not be changing on-disk (eg, from zdb or sync context).
718 */
719 int
720 traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
721 blkptr_cb_t func, void *arg)
722 {
723 int err;
724 dsl_pool_t *dp = spa_get_dsl(spa);
725 objset_t *mos = dp->dp_meta_objset;
726 boolean_t hard = (flags & TRAVERSE_HARD);
727
728 /* visit the MOS */
729 err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa),
730 txg_start, NULL, flags, func, arg);
731 if (err != 0)
732 return (err);
733
734 /* visit each dataset */
735 for (uint64_t obj = 1; err == 0;
736 err = dmu_object_next(mos, &obj, B_FALSE, txg_start)) {
737 dmu_object_info_t doi;
738
739 err = dmu_object_info(mos, obj, &doi);
740 if (err != 0) {
741 if (hard)
742 continue;
743 break;
744 }
745
746 if (doi.doi_bonus_type == DMU_OT_DSL_DATASET) {
747 dsl_dataset_t *ds;
748 uint64_t txg = txg_start;
749
750 dsl_pool_config_enter(dp, FTAG);
751 err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
752 dsl_pool_config_exit(dp, FTAG);
753 if (err != 0) {
754 if (hard)
755 continue;
756 break;
757 }
758 if (dsl_dataset_phys(ds)->ds_prev_snap_txg > txg)
759 txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
760 err = traverse_dataset(ds, txg, flags, func, arg);
761 dsl_dataset_rele(ds, FTAG);
762 if (err != 0)
763 break;
764 }
765 }
766 if (err == ESRCH)
767 err = 0;
768 return (err);
769 }
770
771 #if defined(_KERNEL) && defined(HAVE_SPL)
772 EXPORT_SYMBOL(traverse_dataset);
773 EXPORT_SYMBOL(traverse_pool);
774
775 module_param(zfs_pd_bytes_max, int, 0644);
776 MODULE_PARM_DESC(zfs_pd_bytes_max, "Max number of bytes to prefetch");
777
778 module_param_named(ignore_hole_birth, send_holes_without_birth_time, int, 0644);
779 MODULE_PARM_DESC(ignore_hole_birth, "Alias for send_holes_without_birth_time");
780
781 module_param_named(send_holes_without_birth_time,
782 send_holes_without_birth_time, int, 0644);
783 MODULE_PARM_DESC(send_holes_without_birth_time,
784 "Ignore hole_birth txg for zfs send");
785 #endif