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 2011 Nexenta Systems, Inc. All rights reserved.
24  * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
25  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26  * Copyright 2014 HybridCluster. All rights reserved.
27  * Copyright 2016 RackTop Systems.
28  * Copyright (c) 2016 Actifio, Inc. All rights reserved.
29  * Copyright (c) 2019, Klara Inc.
30  * Copyright (c) 2019, Allan Jude
31  */
32 
33 #include <sys/dmu.h>
34 #include <sys/dmu_impl.h>
35 #include <sys/dmu_tx.h>
36 #include <sys/dbuf.h>
37 #include <sys/dnode.h>
38 #include <sys/zfs_context.h>
39 #include <sys/dmu_objset.h>
40 #include <sys/dmu_traverse.h>
41 #include <sys/dsl_dataset.h>
42 #include <sys/dsl_dir.h>
43 #include <sys/dsl_prop.h>
44 #include <sys/dsl_pool.h>
45 #include <sys/dsl_synctask.h>
46 #include <sys/spa_impl.h>
47 #include <sys/zfs_ioctl.h>
48 #include <sys/zap.h>
49 #include <sys/zio_checksum.h>
50 #include <sys/zfs_znode.h>
51 #include <zfs_fletcher.h>
52 #include <sys/avl.h>
53 #include <sys/ddt.h>
54 #include <sys/zfs_onexit.h>
55 #include <sys/dmu_send.h>
56 #include <sys/dmu_recv.h>
57 #include <sys/dsl_destroy.h>
58 #include <sys/blkptr.h>
59 #include <sys/dsl_bookmark.h>
60 #include <sys/zfeature.h>
61 #include <sys/bqueue.h>
62 #include <sys/zvol.h>
63 #include <sys/policy.h>
64 #include <sys/objlist.h>
65 #ifdef _KERNEL
66 #include <sys/zfs_vfsops.h>
67 #endif
68 
69 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
70 static int zfs_send_corrupt_data = B_FALSE;
71 /*
72  * This tunable controls the amount of data (measured in bytes) that will be
73  * prefetched by zfs send.  If the main thread is blocking on reads that haven't
74  * completed, this variable might need to be increased.  If instead the main
75  * thread is issuing new reads because the prefetches have fallen out of the
76  * cache, this may need to be decreased.
77  */
78 static int zfs_send_queue_length = SPA_MAXBLOCKSIZE;
79 /*
80  * This tunable controls the length of the queues that zfs send worker threads
81  * use to communicate.  If the send_main_thread is blocking on these queues,
82  * this variable may need to be increased.  If there is a significant slowdown
83  * at the start of a send as these threads consume all the available IO
84  * resources, this variable may need to be decreased.
85  */
86 static int zfs_send_no_prefetch_queue_length = 1024 * 1024;
87 /*
88  * These tunables control the fill fraction of the queues by zfs send.  The fill
89  * fraction controls the frequency with which threads have to be cv_signaled.
90  * If a lot of cpu time is being spent on cv_signal, then these should be tuned
91  * down.  If the queues empty before the signalled thread can catch up, then
92  * these should be tuned up.
93  */
94 static int zfs_send_queue_ff = 20;
95 static int zfs_send_no_prefetch_queue_ff = 20;
96 
97 /*
98  * Use this to override the recordsize calculation for fast zfs send estimates.
99  */
100 static int zfs_override_estimate_recordsize = 0;
101 
102 /* Set this tunable to FALSE to disable setting of DRR_FLAG_FREERECORDS */
103 static const boolean_t zfs_send_set_freerecords_bit = B_TRUE;
104 
105 /* Set this tunable to FALSE is disable sending unmodified spill blocks. */
106 static int zfs_send_unmodified_spill_blocks = B_TRUE;
107 
108 static inline boolean_t
109 overflow_multiply(uint64_t a, uint64_t b, uint64_t *c)
110 {
111 	uint64_t temp = a * b;
112 	if (b != 0 && temp / b != a)
113 		return (B_FALSE);
114 	*c = temp;
115 	return (B_TRUE);
116 }
117 
118 struct send_thread_arg {
119 	bqueue_t	q;
120 	objset_t	*os;		/* Objset to traverse */
121 	uint64_t	fromtxg;	/* Traverse from this txg */
122 	int		flags;		/* flags to pass to traverse_dataset */
123 	int		error_code;
124 	boolean_t	cancel;
125 	zbookmark_phys_t resume;
126 	uint64_t	*num_blocks_visited;
127 };
128 
129 struct redact_list_thread_arg {
130 	boolean_t		cancel;
131 	bqueue_t		q;
132 	zbookmark_phys_t	resume;
133 	redaction_list_t	*rl;
134 	boolean_t		mark_redact;
135 	int			error_code;
136 	uint64_t		*num_blocks_visited;
137 };
138 
139 struct send_merge_thread_arg {
140 	bqueue_t			q;
141 	objset_t			*os;
142 	struct redact_list_thread_arg	*from_arg;
143 	struct send_thread_arg		*to_arg;
144 	struct redact_list_thread_arg	*redact_arg;
145 	int				error;
146 	boolean_t			cancel;
147 };
148 
149 struct send_range {
150 	boolean_t		eos_marker; /* Marks the end of the stream */
151 	uint64_t		object;
152 	uint64_t		start_blkid;
153 	uint64_t		end_blkid;
154 	bqueue_node_t		ln;
155 	enum type {DATA, HOLE, OBJECT, OBJECT_RANGE, REDACT,
156 	    PREVIOUSLY_REDACTED} type;
157 	union {
158 		struct srd {
159 			dmu_object_type_t	obj_type;
160 			uint32_t		datablksz; // logical size
161 			uint32_t		datasz; // payload size
162 			blkptr_t		bp;
163 			arc_buf_t		*abuf;
164 			abd_t			*abd;
165 			kmutex_t		lock;
166 			kcondvar_t		cv;
167 			boolean_t		io_outstanding;
168 			boolean_t		io_compressed;
169 			int			io_err;
170 		} data;
171 		struct srh {
172 			uint32_t		datablksz;
173 		} hole;
174 		struct sro {
175 			/*
176 			 * This is a pointer because embedding it in the
177 			 * struct causes these structures to be massively larger
178 			 * for all range types; this makes the code much less
179 			 * memory efficient.
180 			 */
181 			dnode_phys_t		*dnp;
182 			blkptr_t		bp;
183 		} object;
184 		struct srr {
185 			uint32_t		datablksz;
186 		} redact;
187 		struct sror {
188 			blkptr_t		bp;
189 		} object_range;
190 	} sru;
191 };
192 
193 /*
194  * The list of data whose inclusion in a send stream can be pending from
195  * one call to backup_cb to another.  Multiple calls to dump_free(),
196  * dump_freeobjects(), and dump_redact() can be aggregated into a single
197  * DRR_FREE, DRR_FREEOBJECTS, or DRR_REDACT replay record.
198  */
199 typedef enum {
200 	PENDING_NONE,
201 	PENDING_FREE,
202 	PENDING_FREEOBJECTS,
203 	PENDING_REDACT
204 } dmu_pendop_t;
205 
206 typedef struct dmu_send_cookie {
207 	dmu_replay_record_t *dsc_drr;
208 	dmu_send_outparams_t *dsc_dso;
209 	offset_t *dsc_off;
210 	objset_t *dsc_os;
211 	zio_cksum_t dsc_zc;
212 	uint64_t dsc_toguid;
213 	uint64_t dsc_fromtxg;
214 	int dsc_err;
215 	dmu_pendop_t dsc_pending_op;
216 	uint64_t dsc_featureflags;
217 	uint64_t dsc_last_data_object;
218 	uint64_t dsc_last_data_offset;
219 	uint64_t dsc_resume_object;
220 	uint64_t dsc_resume_offset;
221 	boolean_t dsc_sent_begin;
222 	boolean_t dsc_sent_end;
223 } dmu_send_cookie_t;
224 
225 static int do_dump(dmu_send_cookie_t *dscp, struct send_range *range);
226 
227 static void
228 range_free(struct send_range *range)
229 {
230 	if (range->type == OBJECT) {
231 		size_t size = sizeof (dnode_phys_t) *
232 		    (range->sru.object.dnp->dn_extra_slots + 1);
233 		kmem_free(range->sru.object.dnp, size);
234 	} else if (range->type == DATA) {
235 		mutex_enter(&range->sru.data.lock);
236 		while (range->sru.data.io_outstanding)
237 			cv_wait(&range->sru.data.cv, &range->sru.data.lock);
238 		if (range->sru.data.abd != NULL)
239 			abd_free(range->sru.data.abd);
240 		if (range->sru.data.abuf != NULL) {
241 			arc_buf_destroy(range->sru.data.abuf,
242 			    &range->sru.data.abuf);
243 		}
244 		mutex_exit(&range->sru.data.lock);
245 
246 		cv_destroy(&range->sru.data.cv);
247 		mutex_destroy(&range->sru.data.lock);
248 	}
249 	kmem_free(range, sizeof (*range));
250 }
251 
252 /*
253  * For all record types except BEGIN, fill in the checksum (overlaid in
254  * drr_u.drr_checksum.drr_checksum).  The checksum verifies everything
255  * up to the start of the checksum itself.
256  */
257 static int
258 dump_record(dmu_send_cookie_t *dscp, void *payload, int payload_len)
259 {
260 	dmu_send_outparams_t *dso = dscp->dsc_dso;
261 	ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum),
262 	    ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t));
263 	(void) fletcher_4_incremental_native(dscp->dsc_drr,
264 	    offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum),
265 	    &dscp->dsc_zc);
266 	if (dscp->dsc_drr->drr_type == DRR_BEGIN) {
267 		dscp->dsc_sent_begin = B_TRUE;
268 	} else {
269 		ASSERT(ZIO_CHECKSUM_IS_ZERO(&dscp->dsc_drr->drr_u.
270 		    drr_checksum.drr_checksum));
271 		dscp->dsc_drr->drr_u.drr_checksum.drr_checksum = dscp->dsc_zc;
272 	}
273 	if (dscp->dsc_drr->drr_type == DRR_END) {
274 		dscp->dsc_sent_end = B_TRUE;
275 	}
276 	(void) fletcher_4_incremental_native(&dscp->dsc_drr->
277 	    drr_u.drr_checksum.drr_checksum,
278 	    sizeof (zio_cksum_t), &dscp->dsc_zc);
279 	*dscp->dsc_off += sizeof (dmu_replay_record_t);
280 	dscp->dsc_err = dso->dso_outfunc(dscp->dsc_os, dscp->dsc_drr,
281 	    sizeof (dmu_replay_record_t), dso->dso_arg);
282 	if (dscp->dsc_err != 0)
283 		return (SET_ERROR(EINTR));
284 	if (payload_len != 0) {
285 		*dscp->dsc_off += payload_len;
286 		/*
287 		 * payload is null when dso_dryrun == B_TRUE (i.e. when we're
288 		 * doing a send size calculation)
289 		 */
290 		if (payload != NULL) {
291 			(void) fletcher_4_incremental_native(
292 			    payload, payload_len, &dscp->dsc_zc);
293 		}
294 
295 		/*
296 		 * The code does not rely on this (len being a multiple of 8).
297 		 * We keep this assertion because of the corresponding assertion
298 		 * in receive_read().  Keeping this assertion ensures that we do
299 		 * not inadvertently break backwards compatibility (causing the
300 		 * assertion in receive_read() to trigger on old software).
301 		 *
302 		 * Raw sends cannot be received on old software, and so can
303 		 * bypass this assertion.
304 		 */
305 
306 		ASSERT((payload_len % 8 == 0) ||
307 		    (dscp->dsc_featureflags & DMU_BACKUP_FEATURE_RAW));
308 
309 		dscp->dsc_err = dso->dso_outfunc(dscp->dsc_os, payload,
310 		    payload_len, dso->dso_arg);
311 		if (dscp->dsc_err != 0)
312 			return (SET_ERROR(EINTR));
313 	}
314 	return (0);
315 }
316 
317 /*
318  * Fill in the drr_free struct, or perform aggregation if the previous record is
319  * also a free record, and the two are adjacent.
320  *
321  * Note that we send free records even for a full send, because we want to be
322  * able to receive a full send as a clone, which requires a list of all the free
323  * and freeobject records that were generated on the source.
324  */
325 static int
326 dump_free(dmu_send_cookie_t *dscp, uint64_t object, uint64_t offset,
327     uint64_t length)
328 {
329 	struct drr_free *drrf = &(dscp->dsc_drr->drr_u.drr_free);
330 
331 	/*
332 	 * When we receive a free record, dbuf_free_range() assumes
333 	 * that the receiving system doesn't have any dbufs in the range
334 	 * being freed.  This is always true because there is a one-record
335 	 * constraint: we only send one WRITE record for any given
336 	 * object,offset.  We know that the one-record constraint is
337 	 * true because we always send data in increasing order by
338 	 * object,offset.
339 	 *
340 	 * If the increasing-order constraint ever changes, we should find
341 	 * another way to assert that the one-record constraint is still
342 	 * satisfied.
343 	 */
344 	ASSERT(object > dscp->dsc_last_data_object ||
345 	    (object == dscp->dsc_last_data_object &&
346 	    offset > dscp->dsc_last_data_offset));
347 
348 	/*
349 	 * If there is a pending op, but it's not PENDING_FREE, push it out,
350 	 * since free block aggregation can only be done for blocks of the
351 	 * same type (i.e., DRR_FREE records can only be aggregated with
352 	 * other DRR_FREE records.  DRR_FREEOBJECTS records can only be
353 	 * aggregated with other DRR_FREEOBJECTS records).
354 	 */
355 	if (dscp->dsc_pending_op != PENDING_NONE &&
356 	    dscp->dsc_pending_op != PENDING_FREE) {
357 		if (dump_record(dscp, NULL, 0) != 0)
358 			return (SET_ERROR(EINTR));
359 		dscp->dsc_pending_op = PENDING_NONE;
360 	}
361 
362 	if (dscp->dsc_pending_op == PENDING_FREE) {
363 		/*
364 		 * Check to see whether this free block can be aggregated
365 		 * with pending one.
366 		 */
367 		if (drrf->drr_object == object && drrf->drr_offset +
368 		    drrf->drr_length == offset) {
369 			if (offset + length < offset || length == UINT64_MAX)
370 				drrf->drr_length = UINT64_MAX;
371 			else
372 				drrf->drr_length += length;
373 			return (0);
374 		} else {
375 			/* not a continuation.  Push out pending record */
376 			if (dump_record(dscp, NULL, 0) != 0)
377 				return (SET_ERROR(EINTR));
378 			dscp->dsc_pending_op = PENDING_NONE;
379 		}
380 	}
381 	/* create a FREE record and make it pending */
382 	memset(dscp->dsc_drr, 0, sizeof (dmu_replay_record_t));
383 	dscp->dsc_drr->drr_type = DRR_FREE;
384 	drrf->drr_object = object;
385 	drrf->drr_offset = offset;
386 	if (offset + length < offset)
387 		drrf->drr_length = DMU_OBJECT_END;
388 	else
389 		drrf->drr_length = length;
390 	drrf->drr_toguid = dscp->dsc_toguid;
391 	if (length == DMU_OBJECT_END) {
392 		if (dump_record(dscp, NULL, 0) != 0)
393 			return (SET_ERROR(EINTR));
394 	} else {
395 		dscp->dsc_pending_op = PENDING_FREE;
396 	}
397 
398 	return (0);
399 }
400 
401 /*
402  * Fill in the drr_redact struct, or perform aggregation if the previous record
403  * is also a redaction record, and the two are adjacent.
404  */
405 static int
406 dump_redact(dmu_send_cookie_t *dscp, uint64_t object, uint64_t offset,
407     uint64_t length)
408 {
409 	struct drr_redact *drrr = &dscp->dsc_drr->drr_u.drr_redact;
410 
411 	/*
412 	 * If there is a pending op, but it's not PENDING_REDACT, push it out,
413 	 * since free block aggregation can only be done for blocks of the
414 	 * same type (i.e., DRR_REDACT records can only be aggregated with
415 	 * other DRR_REDACT records).
416 	 */
417 	if (dscp->dsc_pending_op != PENDING_NONE &&
418 	    dscp->dsc_pending_op != PENDING_REDACT) {
419 		if (dump_record(dscp, NULL, 0) != 0)
420 			return (SET_ERROR(EINTR));
421 		dscp->dsc_pending_op = PENDING_NONE;
422 	}
423 
424 	if (dscp->dsc_pending_op == PENDING_REDACT) {
425 		/*
426 		 * Check to see whether this redacted block can be aggregated
427 		 * with pending one.
428 		 */
429 		if (drrr->drr_object == object && drrr->drr_offset +
430 		    drrr->drr_length == offset) {
431 			drrr->drr_length += length;
432 			return (0);
433 		} else {
434 			/* not a continuation.  Push out pending record */
435 			if (dump_record(dscp, NULL, 0) != 0)
436 				return (SET_ERROR(EINTR));
437 			dscp->dsc_pending_op = PENDING_NONE;
438 		}
439 	}
440 	/* create a REDACT record and make it pending */
441 	memset(dscp->dsc_drr, 0, sizeof (dmu_replay_record_t));
442 	dscp->dsc_drr->drr_type = DRR_REDACT;
443 	drrr->drr_object = object;
444 	drrr->drr_offset = offset;
445 	drrr->drr_length = length;
446 	drrr->drr_toguid = dscp->dsc_toguid;
447 	dscp->dsc_pending_op = PENDING_REDACT;
448 
449 	return (0);
450 }
451 
452 static int
453 dmu_dump_write(dmu_send_cookie_t *dscp, dmu_object_type_t type, uint64_t object,
454     uint64_t offset, int lsize, int psize, const blkptr_t *bp,
455     boolean_t io_compressed, void *data)
456 {
457 	uint64_t payload_size;
458 	boolean_t raw = (dscp->dsc_featureflags & DMU_BACKUP_FEATURE_RAW);
459 	struct drr_write *drrw = &(dscp->dsc_drr->drr_u.drr_write);
460 
461 	/*
462 	 * We send data in increasing object, offset order.
463 	 * See comment in dump_free() for details.
464 	 */
465 	ASSERT(object > dscp->dsc_last_data_object ||
466 	    (object == dscp->dsc_last_data_object &&
467 	    offset > dscp->dsc_last_data_offset));
468 	dscp->dsc_last_data_object = object;
469 	dscp->dsc_last_data_offset = offset + lsize - 1;
470 
471 	/*
472 	 * If there is any kind of pending aggregation (currently either
473 	 * a grouping of free objects or free blocks), push it out to
474 	 * the stream, since aggregation can't be done across operations
475 	 * of different types.
476 	 */
477 	if (dscp->dsc_pending_op != PENDING_NONE) {
478 		if (dump_record(dscp, NULL, 0) != 0)
479 			return (SET_ERROR(EINTR));
480 		dscp->dsc_pending_op = PENDING_NONE;
481 	}
482 	/* write a WRITE record */
483 	memset(dscp->dsc_drr, 0, sizeof (dmu_replay_record_t));
484 	dscp->dsc_drr->drr_type = DRR_WRITE;
485 	drrw->drr_object = object;
486 	drrw->drr_type = type;
487 	drrw->drr_offset = offset;
488 	drrw->drr_toguid = dscp->dsc_toguid;
489 	drrw->drr_logical_size = lsize;
490 
491 	/* only set the compression fields if the buf is compressed or raw */
492 	boolean_t compressed =
493 	    (bp != NULL ? BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF &&
494 	    io_compressed : lsize != psize);
495 	if (raw || compressed) {
496 		ASSERT(raw || dscp->dsc_featureflags &
497 		    DMU_BACKUP_FEATURE_COMPRESSED);
498 		ASSERT(!BP_IS_EMBEDDED(bp));
499 		ASSERT3S(psize, >, 0);
500 
501 		if (raw) {
502 			ASSERT(BP_IS_PROTECTED(bp));
503 
504 			/*
505 			 * This is a raw protected block so we need to pass
506 			 * along everything the receiving side will need to
507 			 * interpret this block, including the byteswap, salt,
508 			 * IV, and MAC.
509 			 */
510 			if (BP_SHOULD_BYTESWAP(bp))
511 				drrw->drr_flags |= DRR_RAW_BYTESWAP;
512 			zio_crypt_decode_params_bp(bp, drrw->drr_salt,
513 			    drrw->drr_iv);
514 			zio_crypt_decode_mac_bp(bp, drrw->drr_mac);
515 		} else {
516 			/* this is a compressed block */
517 			ASSERT(dscp->dsc_featureflags &
518 			    DMU_BACKUP_FEATURE_COMPRESSED);
519 			ASSERT(!BP_SHOULD_BYTESWAP(bp));
520 			ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)));
521 			ASSERT3U(BP_GET_COMPRESS(bp), !=, ZIO_COMPRESS_OFF);
522 			ASSERT3S(lsize, >=, psize);
523 		}
524 
525 		/* set fields common to compressed and raw sends */
526 		drrw->drr_compressiontype = BP_GET_COMPRESS(bp);
527 		drrw->drr_compressed_size = psize;
528 		payload_size = drrw->drr_compressed_size;
529 	} else {
530 		payload_size = drrw->drr_logical_size;
531 	}
532 
533 	if (bp == NULL || BP_IS_EMBEDDED(bp) || (BP_IS_PROTECTED(bp) && !raw)) {
534 		/*
535 		 * There's no pre-computed checksum for partial-block writes,
536 		 * embedded BP's, or encrypted BP's that are being sent as
537 		 * plaintext, so (like fletcher4-checksummed blocks) userland
538 		 * will have to compute a dedup-capable checksum itself.
539 		 */
540 		drrw->drr_checksumtype = ZIO_CHECKSUM_OFF;
541 	} else {
542 		drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
543 		if (zio_checksum_table[drrw->drr_checksumtype].ci_flags &
544 		    ZCHECKSUM_FLAG_DEDUP)
545 			drrw->drr_flags |= DRR_CHECKSUM_DEDUP;
546 		DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
547 		DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
548 		DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
549 		DDK_SET_CRYPT(&drrw->drr_key, BP_IS_PROTECTED(bp));
550 		drrw->drr_key.ddk_cksum = bp->blk_cksum;
551 	}
552 
553 	if (dump_record(dscp, data, payload_size) != 0)
554 		return (SET_ERROR(EINTR));
555 	return (0);
556 }
557 
558 static int
559 dump_write_embedded(dmu_send_cookie_t *dscp, uint64_t object, uint64_t offset,
560     int blksz, const blkptr_t *bp)
561 {
562 	char buf[BPE_PAYLOAD_SIZE];
563 	struct drr_write_embedded *drrw =
564 	    &(dscp->dsc_drr->drr_u.drr_write_embedded);
565 
566 	if (dscp->dsc_pending_op != PENDING_NONE) {
567 		if (dump_record(dscp, NULL, 0) != 0)
568 			return (SET_ERROR(EINTR));
569 		dscp->dsc_pending_op = PENDING_NONE;
570 	}
571 
572 	ASSERT(BP_IS_EMBEDDED(bp));
573 
574 	memset(dscp->dsc_drr, 0, sizeof (dmu_replay_record_t));
575 	dscp->dsc_drr->drr_type = DRR_WRITE_EMBEDDED;
576 	drrw->drr_object = object;
577 	drrw->drr_offset = offset;
578 	drrw->drr_length = blksz;
579 	drrw->drr_toguid = dscp->dsc_toguid;
580 	drrw->drr_compression = BP_GET_COMPRESS(bp);
581 	drrw->drr_etype = BPE_GET_ETYPE(bp);
582 	drrw->drr_lsize = BPE_GET_LSIZE(bp);
583 	drrw->drr_psize = BPE_GET_PSIZE(bp);
584 
585 	decode_embedded_bp_compressed(bp, buf);
586 
587 	if (dump_record(dscp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0)
588 		return (SET_ERROR(EINTR));
589 	return (0);
590 }
591 
592 static int
593 dump_spill(dmu_send_cookie_t *dscp, const blkptr_t *bp, uint64_t object,
594     void *data)
595 {
596 	struct drr_spill *drrs = &(dscp->dsc_drr->drr_u.drr_spill);
597 	uint64_t blksz = BP_GET_LSIZE(bp);
598 	uint64_t payload_size = blksz;
599 
600 	if (dscp->dsc_pending_op != PENDING_NONE) {
601 		if (dump_record(dscp, NULL, 0) != 0)
602 			return (SET_ERROR(EINTR));
603 		dscp->dsc_pending_op = PENDING_NONE;
604 	}
605 
606 	/* write a SPILL record */
607 	memset(dscp->dsc_drr, 0, sizeof (dmu_replay_record_t));
608 	dscp->dsc_drr->drr_type = DRR_SPILL;
609 	drrs->drr_object = object;
610 	drrs->drr_length = blksz;
611 	drrs->drr_toguid = dscp->dsc_toguid;
612 
613 	/* See comment in dump_dnode() for full details */
614 	if (zfs_send_unmodified_spill_blocks &&
615 	    (bp->blk_birth <= dscp->dsc_fromtxg)) {
616 		drrs->drr_flags |= DRR_SPILL_UNMODIFIED;
617 	}
618 
619 	/* handle raw send fields */
620 	if (dscp->dsc_featureflags & DMU_BACKUP_FEATURE_RAW) {
621 		ASSERT(BP_IS_PROTECTED(bp));
622 
623 		if (BP_SHOULD_BYTESWAP(bp))
624 			drrs->drr_flags |= DRR_RAW_BYTESWAP;
625 		drrs->drr_compressiontype = BP_GET_COMPRESS(bp);
626 		drrs->drr_compressed_size = BP_GET_PSIZE(bp);
627 		zio_crypt_decode_params_bp(bp, drrs->drr_salt, drrs->drr_iv);
628 		zio_crypt_decode_mac_bp(bp, drrs->drr_mac);
629 		payload_size = drrs->drr_compressed_size;
630 	}
631 
632 	if (dump_record(dscp, data, payload_size) != 0)
633 		return (SET_ERROR(EINTR));
634 	return (0);
635 }
636 
637 static int
638 dump_freeobjects(dmu_send_cookie_t *dscp, uint64_t firstobj, uint64_t numobjs)
639 {
640 	struct drr_freeobjects *drrfo = &(dscp->dsc_drr->drr_u.drr_freeobjects);
641 	uint64_t maxobj = DNODES_PER_BLOCK *
642 	    (DMU_META_DNODE(dscp->dsc_os)->dn_maxblkid + 1);
643 
644 	/*
645 	 * ZoL < 0.7 does not handle large FREEOBJECTS records correctly,
646 	 * leading to zfs recv never completing. to avoid this issue, don't
647 	 * send FREEOBJECTS records for object IDs which cannot exist on the
648 	 * receiving side.
649 	 */
650 	if (maxobj > 0) {
651 		if (maxobj <= firstobj)
652 			return (0);
653 
654 		if (maxobj < firstobj + numobjs)
655 			numobjs = maxobj - firstobj;
656 	}
657 
658 	/*
659 	 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
660 	 * push it out, since free block aggregation can only be done for
661 	 * blocks of the same type (i.e., DRR_FREE records can only be
662 	 * aggregated with other DRR_FREE records.  DRR_FREEOBJECTS records
663 	 * can only be aggregated with other DRR_FREEOBJECTS records).
664 	 */
665 	if (dscp->dsc_pending_op != PENDING_NONE &&
666 	    dscp->dsc_pending_op != PENDING_FREEOBJECTS) {
667 		if (dump_record(dscp, NULL, 0) != 0)
668 			return (SET_ERROR(EINTR));
669 		dscp->dsc_pending_op = PENDING_NONE;
670 	}
671 
672 	if (dscp->dsc_pending_op == PENDING_FREEOBJECTS) {
673 		/*
674 		 * See whether this free object array can be aggregated
675 		 * with pending one
676 		 */
677 		if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
678 			drrfo->drr_numobjs += numobjs;
679 			return (0);
680 		} else {
681 			/* can't be aggregated.  Push out pending record */
682 			if (dump_record(dscp, NULL, 0) != 0)
683 				return (SET_ERROR(EINTR));
684 			dscp->dsc_pending_op = PENDING_NONE;
685 		}
686 	}
687 
688 	/* write a FREEOBJECTS record */
689 	memset(dscp->dsc_drr, 0, sizeof (dmu_replay_record_t));
690 	dscp->dsc_drr->drr_type = DRR_FREEOBJECTS;
691 	drrfo->drr_firstobj = firstobj;
692 	drrfo->drr_numobjs = numobjs;
693 	drrfo->drr_toguid = dscp->dsc_toguid;
694 
695 	dscp->dsc_pending_op = PENDING_FREEOBJECTS;
696 
697 	return (0);
698 }
699 
700 static int
701 dump_dnode(dmu_send_cookie_t *dscp, const blkptr_t *bp, uint64_t object,
702     dnode_phys_t *dnp)
703 {
704 	struct drr_object *drro = &(dscp->dsc_drr->drr_u.drr_object);
705 	int bonuslen;
706 
707 	if (object < dscp->dsc_resume_object) {
708 		/*
709 		 * Note: when resuming, we will visit all the dnodes in
710 		 * the block of dnodes that we are resuming from.  In
711 		 * this case it's unnecessary to send the dnodes prior to
712 		 * the one we are resuming from.  We should be at most one
713 		 * block's worth of dnodes behind the resume point.
714 		 */
715 		ASSERT3U(dscp->dsc_resume_object - object, <,
716 		    1 << (DNODE_BLOCK_SHIFT - DNODE_SHIFT));
717 		return (0);
718 	}
719 
720 	if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
721 		return (dump_freeobjects(dscp, object, 1));
722 
723 	if (dscp->dsc_pending_op != PENDING_NONE) {
724 		if (dump_record(dscp, NULL, 0) != 0)
725 			return (SET_ERROR(EINTR));
726 		dscp->dsc_pending_op = PENDING_NONE;
727 	}
728 
729 	/* write an OBJECT record */
730 	memset(dscp->dsc_drr, 0, sizeof (dmu_replay_record_t));
731 	dscp->dsc_drr->drr_type = DRR_OBJECT;
732 	drro->drr_object = object;
733 	drro->drr_type = dnp->dn_type;
734 	drro->drr_bonustype = dnp->dn_bonustype;
735 	drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
736 	drro->drr_bonuslen = dnp->dn_bonuslen;
737 	drro->drr_dn_slots = dnp->dn_extra_slots + 1;
738 	drro->drr_checksumtype = dnp->dn_checksum;
739 	drro->drr_compress = dnp->dn_compress;
740 	drro->drr_toguid = dscp->dsc_toguid;
741 
742 	if (!(dscp->dsc_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
743 	    drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE)
744 		drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE;
745 
746 	bonuslen = P2ROUNDUP(dnp->dn_bonuslen, 8);
747 
748 	if ((dscp->dsc_featureflags & DMU_BACKUP_FEATURE_RAW)) {
749 		ASSERT(BP_IS_ENCRYPTED(bp));
750 
751 		if (BP_SHOULD_BYTESWAP(bp))
752 			drro->drr_flags |= DRR_RAW_BYTESWAP;
753 
754 		/* needed for reconstructing dnp on recv side */
755 		drro->drr_maxblkid = dnp->dn_maxblkid;
756 		drro->drr_indblkshift = dnp->dn_indblkshift;
757 		drro->drr_nlevels = dnp->dn_nlevels;
758 		drro->drr_nblkptr = dnp->dn_nblkptr;
759 
760 		/*
761 		 * Since we encrypt the entire bonus area, the (raw) part
762 		 * beyond the bonuslen is actually nonzero, so we need
763 		 * to send it.
764 		 */
765 		if (bonuslen != 0) {
766 			if (drro->drr_bonuslen > DN_MAX_BONUS_LEN(dnp))
767 				return (SET_ERROR(EINVAL));
768 			drro->drr_raw_bonuslen = DN_MAX_BONUS_LEN(dnp);
769 			bonuslen = drro->drr_raw_bonuslen;
770 		}
771 	}
772 
773 	/*
774 	 * DRR_OBJECT_SPILL is set for every dnode which references a
775 	 * spill block.	 This allows the receiving pool to definitively
776 	 * determine when a spill block should be kept or freed.
777 	 */
778 	if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)
779 		drro->drr_flags |= DRR_OBJECT_SPILL;
780 
781 	if (dump_record(dscp, DN_BONUS(dnp), bonuslen) != 0)
782 		return (SET_ERROR(EINTR));
783 
784 	/* Free anything past the end of the file. */
785 	if (dump_free(dscp, object, (dnp->dn_maxblkid + 1) *
786 	    (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), DMU_OBJECT_END) != 0)
787 		return (SET_ERROR(EINTR));
788 
789 	/*
790 	 * Send DRR_SPILL records for unmodified spill blocks.	This is useful
791 	 * because changing certain attributes of the object (e.g. blocksize)
792 	 * can cause old versions of ZFS to incorrectly remove a spill block.
793 	 * Including these records in the stream forces an up to date version
794 	 * to always be written ensuring they're never lost.  Current versions
795 	 * of the code which understand the DRR_FLAG_SPILL_BLOCK feature can
796 	 * ignore these unmodified spill blocks.
797 	 */
798 	if (zfs_send_unmodified_spill_blocks &&
799 	    (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) &&
800 	    (DN_SPILL_BLKPTR(dnp)->blk_birth <= dscp->dsc_fromtxg)) {
801 		struct send_range record;
802 		blkptr_t *bp = DN_SPILL_BLKPTR(dnp);
803 
804 		memset(&record, 0, sizeof (struct send_range));
805 		record.type = DATA;
806 		record.object = object;
807 		record.eos_marker = B_FALSE;
808 		record.start_blkid = DMU_SPILL_BLKID;
809 		record.end_blkid = record.start_blkid + 1;
810 		record.sru.data.bp = *bp;
811 		record.sru.data.obj_type = dnp->dn_type;
812 		record.sru.data.datablksz = BP_GET_LSIZE(bp);
813 
814 		if (do_dump(dscp, &record) != 0)
815 			return (SET_ERROR(EINTR));
816 	}
817 
818 	if (dscp->dsc_err != 0)
819 		return (SET_ERROR(EINTR));
820 
821 	return (0);
822 }
823 
824 static int
825 dump_object_range(dmu_send_cookie_t *dscp, const blkptr_t *bp,
826     uint64_t firstobj, uint64_t numslots)
827 {
828 	struct drr_object_range *drror =
829 	    &(dscp->dsc_drr->drr_u.drr_object_range);
830 
831 	/* we only use this record type for raw sends */
832 	ASSERT(BP_IS_PROTECTED(bp));
833 	ASSERT(dscp->dsc_featureflags & DMU_BACKUP_FEATURE_RAW);
834 	ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF);
835 	ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_DNODE);
836 	ASSERT0(BP_GET_LEVEL(bp));
837 
838 	if (dscp->dsc_pending_op != PENDING_NONE) {
839 		if (dump_record(dscp, NULL, 0) != 0)
840 			return (SET_ERROR(EINTR));
841 		dscp->dsc_pending_op = PENDING_NONE;
842 	}
843 
844 	memset(dscp->dsc_drr, 0, sizeof (dmu_replay_record_t));
845 	dscp->dsc_drr->drr_type = DRR_OBJECT_RANGE;
846 	drror->drr_firstobj = firstobj;
847 	drror->drr_numslots = numslots;
848 	drror->drr_toguid = dscp->dsc_toguid;
849 	if (BP_SHOULD_BYTESWAP(bp))
850 		drror->drr_flags |= DRR_RAW_BYTESWAP;
851 	zio_crypt_decode_params_bp(bp, drror->drr_salt, drror->drr_iv);
852 	zio_crypt_decode_mac_bp(bp, drror->drr_mac);
853 
854 	if (dump_record(dscp, NULL, 0) != 0)
855 		return (SET_ERROR(EINTR));
856 	return (0);
857 }
858 
859 static boolean_t
860 send_do_embed(const blkptr_t *bp, uint64_t featureflags)
861 {
862 	if (!BP_IS_EMBEDDED(bp))
863 		return (B_FALSE);
864 
865 	/*
866 	 * Compression function must be legacy, or explicitly enabled.
867 	 */
868 	if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS &&
869 	    !(featureflags & DMU_BACKUP_FEATURE_LZ4)))
870 		return (B_FALSE);
871 
872 	/*
873 	 * If we have not set the ZSTD feature flag, we can't send ZSTD
874 	 * compressed embedded blocks, as the receiver may not support them.
875 	 */
876 	if ((BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD &&
877 	    !(featureflags & DMU_BACKUP_FEATURE_ZSTD)))
878 		return (B_FALSE);
879 
880 	/*
881 	 * Embed type must be explicitly enabled.
882 	 */
883 	switch (BPE_GET_ETYPE(bp)) {
884 	case BP_EMBEDDED_TYPE_DATA:
885 		if (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)
886 			return (B_TRUE);
887 		break;
888 	default:
889 		return (B_FALSE);
890 	}
891 	return (B_FALSE);
892 }
893 
894 /*
895  * This function actually handles figuring out what kind of record needs to be
896  * dumped, and calling the appropriate helper function.  In most cases,
897  * the data has already been read by send_reader_thread().
898  */
899 static int
900 do_dump(dmu_send_cookie_t *dscp, struct send_range *range)
901 {
902 	int err = 0;
903 	switch (range->type) {
904 	case OBJECT:
905 		err = dump_dnode(dscp, &range->sru.object.bp, range->object,
906 		    range->sru.object.dnp);
907 		return (err);
908 	case OBJECT_RANGE: {
909 		ASSERT3U(range->start_blkid + 1, ==, range->end_blkid);
910 		if (!(dscp->dsc_featureflags & DMU_BACKUP_FEATURE_RAW)) {
911 			return (0);
912 		}
913 		uint64_t epb = BP_GET_LSIZE(&range->sru.object_range.bp) >>
914 		    DNODE_SHIFT;
915 		uint64_t firstobj = range->start_blkid * epb;
916 		err = dump_object_range(dscp, &range->sru.object_range.bp,
917 		    firstobj, epb);
918 		break;
919 	}
920 	case REDACT: {
921 		struct srr *srrp = &range->sru.redact;
922 		err = dump_redact(dscp, range->object, range->start_blkid *
923 		    srrp->datablksz, (range->end_blkid - range->start_blkid) *
924 		    srrp->datablksz);
925 		return (err);
926 	}
927 	case DATA: {
928 		struct srd *srdp = &range->sru.data;
929 		blkptr_t *bp = &srdp->bp;
930 		spa_t *spa =
931 		    dmu_objset_spa(dscp->dsc_os);
932 
933 		ASSERT3U(srdp->datablksz, ==, BP_GET_LSIZE(bp));
934 		ASSERT3U(range->start_blkid + 1, ==, range->end_blkid);
935 		if (BP_GET_TYPE(bp) == DMU_OT_SA) {
936 			arc_flags_t aflags = ARC_FLAG_WAIT;
937 			enum zio_flag zioflags = ZIO_FLAG_CANFAIL;
938 
939 			if (dscp->dsc_featureflags & DMU_BACKUP_FEATURE_RAW) {
940 				ASSERT(BP_IS_PROTECTED(bp));
941 				zioflags |= ZIO_FLAG_RAW;
942 			}
943 
944 			zbookmark_phys_t zb;
945 			ASSERT3U(range->start_blkid, ==, DMU_SPILL_BLKID);
946 			zb.zb_objset = dmu_objset_id(dscp->dsc_os);
947 			zb.zb_object = range->object;
948 			zb.zb_level = 0;
949 			zb.zb_blkid = range->start_blkid;
950 
951 			arc_buf_t *abuf = NULL;
952 			if (!dscp->dsc_dso->dso_dryrun && arc_read(NULL, spa,
953 			    bp, arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
954 			    zioflags, &aflags, &zb) != 0)
955 				return (SET_ERROR(EIO));
956 
957 			err = dump_spill(dscp, bp, zb.zb_object,
958 			    (abuf == NULL ? NULL : abuf->b_data));
959 			if (abuf != NULL)
960 				arc_buf_destroy(abuf, &abuf);
961 			return (err);
962 		}
963 		if (send_do_embed(bp, dscp->dsc_featureflags)) {
964 			err = dump_write_embedded(dscp, range->object,
965 			    range->start_blkid * srdp->datablksz,
966 			    srdp->datablksz, bp);
967 			return (err);
968 		}
969 		ASSERT(range->object > dscp->dsc_resume_object ||
970 		    (range->object == dscp->dsc_resume_object &&
971 		    range->start_blkid * srdp->datablksz >=
972 		    dscp->dsc_resume_offset));
973 		/* it's a level-0 block of a regular object */
974 
975 		mutex_enter(&srdp->lock);
976 		while (srdp->io_outstanding)
977 			cv_wait(&srdp->cv, &srdp->lock);
978 		err = srdp->io_err;
979 		mutex_exit(&srdp->lock);
980 
981 		if (err != 0) {
982 			if (zfs_send_corrupt_data &&
983 			    !dscp->dsc_dso->dso_dryrun) {
984 				/*
985 				 * Send a block filled with 0x"zfs badd bloc"
986 				 */
987 				srdp->abuf = arc_alloc_buf(spa, &srdp->abuf,
988 				    ARC_BUFC_DATA, srdp->datablksz);
989 				uint64_t *ptr;
990 				for (ptr = srdp->abuf->b_data;
991 				    (char *)ptr < (char *)srdp->abuf->b_data +
992 				    srdp->datablksz; ptr++)
993 					*ptr = 0x2f5baddb10cULL;
994 			} else {
995 				return (SET_ERROR(EIO));
996 			}
997 		}
998 
999 		ASSERT(dscp->dsc_dso->dso_dryrun ||
1000 		    srdp->abuf != NULL || srdp->abd != NULL);
1001 
1002 		uint64_t offset = range->start_blkid * srdp->datablksz;
1003 
1004 		char *data = NULL;
1005 		if (srdp->abd != NULL) {
1006 			data = abd_to_buf(srdp->abd);
1007 			ASSERT3P(srdp->abuf, ==, NULL);
1008 		} else if (srdp->abuf != NULL) {
1009 			data = srdp->abuf->b_data;
1010 		}
1011 
1012 		/*
1013 		 * If we have large blocks stored on disk but the send flags
1014 		 * don't allow us to send large blocks, we split the data from
1015 		 * the arc buf into chunks.
1016 		 */
1017 		if (srdp->datablksz > SPA_OLD_MAXBLOCKSIZE &&
1018 		    !(dscp->dsc_featureflags &
1019 		    DMU_BACKUP_FEATURE_LARGE_BLOCKS)) {
1020 			while (srdp->datablksz > 0 && err == 0) {
1021 				int n = MIN(srdp->datablksz,
1022 				    SPA_OLD_MAXBLOCKSIZE);
1023 				err = dmu_dump_write(dscp, srdp->obj_type,
1024 				    range->object, offset, n, n, NULL, B_FALSE,
1025 				    data);
1026 				offset += n;
1027 				/*
1028 				 * When doing dry run, data==NULL is used as a
1029 				 * sentinel value by
1030 				 * dmu_dump_write()->dump_record().
1031 				 */
1032 				if (data != NULL)
1033 					data += n;
1034 				srdp->datablksz -= n;
1035 			}
1036 		} else {
1037 			err = dmu_dump_write(dscp, srdp->obj_type,
1038 			    range->object, offset,
1039 			    srdp->datablksz, srdp->datasz, bp,
1040 			    srdp->io_compressed, data);
1041 		}
1042 		return (err);
1043 	}
1044 	case HOLE: {
1045 		struct srh *srhp = &range->sru.hole;
1046 		if (range->object == DMU_META_DNODE_OBJECT) {
1047 			uint32_t span = srhp->datablksz >> DNODE_SHIFT;
1048 			uint64_t first_obj = range->start_blkid * span;
1049 			uint64_t numobj = range->end_blkid * span - first_obj;
1050 			return (dump_freeobjects(dscp, first_obj, numobj));
1051 		}
1052 		uint64_t offset = 0;
1053 
1054 		/*
1055 		 * If this multiply overflows, we don't need to send this block.
1056 		 * Even if it has a birth time, it can never not be a hole, so
1057 		 * we don't need to send records for it.
1058 		 */
1059 		if (!overflow_multiply(range->start_blkid, srhp->datablksz,
1060 		    &offset)) {
1061 			return (0);
1062 		}
1063 		uint64_t len = 0;
1064 
1065 		if (!overflow_multiply(range->end_blkid, srhp->datablksz, &len))
1066 			len = UINT64_MAX;
1067 		len = len - offset;
1068 		return (dump_free(dscp, range->object, offset, len));
1069 	}
1070 	default:
1071 		panic("Invalid range type in do_dump: %d", range->type);
1072 	}
1073 	return (err);
1074 }
1075 
1076 static struct send_range *
1077 range_alloc(enum type type, uint64_t object, uint64_t start_blkid,
1078     uint64_t end_blkid, boolean_t eos)
1079 {
1080 	struct send_range *range = kmem_alloc(sizeof (*range), KM_SLEEP);
1081 	range->type = type;
1082 	range->object = object;
1083 	range->start_blkid = start_blkid;
1084 	range->end_blkid = end_blkid;
1085 	range->eos_marker = eos;
1086 	if (type == DATA) {
1087 		range->sru.data.abd = NULL;
1088 		range->sru.data.abuf = NULL;
1089 		mutex_init(&range->sru.data.lock, NULL, MUTEX_DEFAULT, NULL);
1090 		cv_init(&range->sru.data.cv, NULL, CV_DEFAULT, NULL);
1091 		range->sru.data.io_outstanding = 0;
1092 		range->sru.data.io_err = 0;
1093 		range->sru.data.io_compressed = B_FALSE;
1094 	}
1095 	return (range);
1096 }
1097 
1098 /*
1099  * This is the callback function to traverse_dataset that acts as a worker
1100  * thread for dmu_send_impl.
1101  */
1102 static int
1103 send_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1104     const zbookmark_phys_t *zb, const struct dnode_phys *dnp, void *arg)
1105 {
1106 	(void) zilog;
1107 	struct send_thread_arg *sta = arg;
1108 	struct send_range *record;
1109 
1110 	ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT ||
1111 	    zb->zb_object >= sta->resume.zb_object);
1112 
1113 	/*
1114 	 * All bps of an encrypted os should have the encryption bit set.
1115 	 * If this is not true it indicates tampering and we report an error.
1116 	 */
1117 	if (sta->os->os_encrypted &&
1118 	    !BP_IS_HOLE(bp) && !BP_USES_CRYPT(bp)) {
1119 		spa_log_error(spa, zb);
1120 		zfs_panic_recover("unencrypted block in encrypted "
1121 		    "object set %llu", dmu_objset_id(sta->os));
1122 		return (SET_ERROR(EIO));
1123 	}
1124 
1125 	if (sta->cancel)
1126 		return (SET_ERROR(EINTR));
1127 	if (zb->zb_object != DMU_META_DNODE_OBJECT &&
1128 	    DMU_OBJECT_IS_SPECIAL(zb->zb_object))
1129 		return (0);
1130 	atomic_inc_64(sta->num_blocks_visited);
1131 
1132 	if (zb->zb_level == ZB_DNODE_LEVEL) {
1133 		if (zb->zb_object == DMU_META_DNODE_OBJECT)
1134 			return (0);
1135 		record = range_alloc(OBJECT, zb->zb_object, 0, 0, B_FALSE);
1136 		record->sru.object.bp = *bp;
1137 		size_t size  = sizeof (*dnp) * (dnp->dn_extra_slots + 1);
1138 		record->sru.object.dnp = kmem_alloc(size, KM_SLEEP);
1139 		memcpy(record->sru.object.dnp, dnp, size);
1140 		bqueue_enqueue(&sta->q, record, sizeof (*record));
1141 		return (0);
1142 	}
1143 	if (zb->zb_level == 0 && zb->zb_object == DMU_META_DNODE_OBJECT &&
1144 	    !BP_IS_HOLE(bp)) {
1145 		record = range_alloc(OBJECT_RANGE, 0, zb->zb_blkid,
1146 		    zb->zb_blkid + 1, B_FALSE);
1147 		record->sru.object_range.bp = *bp;
1148 		bqueue_enqueue(&sta->q, record, sizeof (*record));
1149 		return (0);
1150 	}
1151 	if (zb->zb_level < 0 || (zb->zb_level > 0 && !BP_IS_HOLE(bp)))
1152 		return (0);
1153 	if (zb->zb_object == DMU_META_DNODE_OBJECT && !BP_IS_HOLE(bp))
1154 		return (0);
1155 
1156 	uint64_t span = bp_span_in_blocks(dnp->dn_indblkshift, zb->zb_level);
1157 	uint64_t start;
1158 
1159 	/*
1160 	 * If this multiply overflows, we don't need to send this block.
1161 	 * Even if it has a birth time, it can never not be a hole, so
1162 	 * we don't need to send records for it.
1163 	 */
1164 	if (!overflow_multiply(span, zb->zb_blkid, &start) || (!(zb->zb_blkid ==
1165 	    DMU_SPILL_BLKID || DMU_OT_IS_METADATA(dnp->dn_type)) &&
1166 	    span * zb->zb_blkid > dnp->dn_maxblkid)) {
1167 		ASSERT(BP_IS_HOLE(bp));
1168 		return (0);
1169 	}
1170 
1171 	if (zb->zb_blkid == DMU_SPILL_BLKID)
1172 		ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_SA);
1173 
1174 	enum type record_type = DATA;
1175 	if (BP_IS_HOLE(bp))
1176 		record_type = HOLE;
1177 	else if (BP_IS_REDACTED(bp))
1178 		record_type = REDACT;
1179 	else
1180 		record_type = DATA;
1181 
1182 	record = range_alloc(record_type, zb->zb_object, start,
1183 	    (start + span < start ? 0 : start + span), B_FALSE);
1184 
1185 	uint64_t datablksz = (zb->zb_blkid == DMU_SPILL_BLKID ?
1186 	    BP_GET_LSIZE(bp) : dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
1187 
1188 	if (BP_IS_HOLE(bp)) {
1189 		record->sru.hole.datablksz = datablksz;
1190 	} else if (BP_IS_REDACTED(bp)) {
1191 		record->sru.redact.datablksz = datablksz;
1192 	} else {
1193 		record->sru.data.datablksz = datablksz;
1194 		record->sru.data.obj_type = dnp->dn_type;
1195 		record->sru.data.bp = *bp;
1196 	}
1197 
1198 	bqueue_enqueue(&sta->q, record, sizeof (*record));
1199 	return (0);
1200 }
1201 
1202 struct redact_list_cb_arg {
1203 	uint64_t *num_blocks_visited;
1204 	bqueue_t *q;
1205 	boolean_t *cancel;
1206 	boolean_t mark_redact;
1207 };
1208 
1209 static int
1210 redact_list_cb(redact_block_phys_t *rb, void *arg)
1211 {
1212 	struct redact_list_cb_arg *rlcap = arg;
1213 
1214 	atomic_inc_64(rlcap->num_blocks_visited);
1215 	if (*rlcap->cancel)
1216 		return (-1);
1217 
1218 	struct send_range *data = range_alloc(REDACT, rb->rbp_object,
1219 	    rb->rbp_blkid, rb->rbp_blkid + redact_block_get_count(rb), B_FALSE);
1220 	ASSERT3U(data->end_blkid, >, rb->rbp_blkid);
1221 	if (rlcap->mark_redact) {
1222 		data->type = REDACT;
1223 		data->sru.redact.datablksz = redact_block_get_size(rb);
1224 	} else {
1225 		data->type = PREVIOUSLY_REDACTED;
1226 	}
1227 	bqueue_enqueue(rlcap->q, data, sizeof (*data));
1228 
1229 	return (0);
1230 }
1231 
1232 /*
1233  * This function kicks off the traverse_dataset.  It also handles setting the
1234  * error code of the thread in case something goes wrong, and pushes the End of
1235  * Stream record when the traverse_dataset call has finished.
1236  */
1237 static __attribute__((noreturn)) void
1238 send_traverse_thread(void *arg)
1239 {
1240 	struct send_thread_arg *st_arg = arg;
1241 	int err = 0;
1242 	struct send_range *data;
1243 	fstrans_cookie_t cookie = spl_fstrans_mark();
1244 
1245 	err = traverse_dataset_resume(st_arg->os->os_dsl_dataset,
1246 	    st_arg->fromtxg, &st_arg->resume,
1247 	    st_arg->flags, send_cb, st_arg);
1248 
1249 	if (err != EINTR)
1250 		st_arg->error_code = err;
1251 	data = range_alloc(DATA, 0, 0, 0, B_TRUE);
1252 	bqueue_enqueue_flush(&st_arg->q, data, sizeof (*data));
1253 	spl_fstrans_unmark(cookie);
1254 	thread_exit();
1255 }
1256 
1257 /*
1258  * Utility function that causes End of Stream records to compare after of all
1259  * others, so that other threads' comparison logic can stay simple.
1260  */
1261 static int __attribute__((unused))
1262 send_range_after(const struct send_range *from, const struct send_range *to)
1263 {
1264 	if (from->eos_marker == B_TRUE)
1265 		return (1);
1266 	if (to->eos_marker == B_TRUE)
1267 		return (-1);
1268 
1269 	uint64_t from_obj = from->object;
1270 	uint64_t from_end_obj = from->object + 1;
1271 	uint64_t to_obj = to->object;
1272 	uint64_t to_end_obj = to->object + 1;
1273 	if (from_obj == 0) {
1274 		ASSERT(from->type == HOLE || from->type == OBJECT_RANGE);
1275 		from_obj = from->start_blkid << DNODES_PER_BLOCK_SHIFT;
1276 		from_end_obj = from->end_blkid << DNODES_PER_BLOCK_SHIFT;
1277 	}
1278 	if (to_obj == 0) {
1279 		ASSERT(to->type == HOLE || to->type == OBJECT_RANGE);
1280 		to_obj = to->start_blkid << DNODES_PER_BLOCK_SHIFT;
1281 		to_end_obj = to->end_blkid << DNODES_PER_BLOCK_SHIFT;
1282 	}
1283 
1284 	if (from_end_obj <= to_obj)
1285 		return (-1);
1286 	if (from_obj >= to_end_obj)
1287 		return (1);
1288 	int64_t cmp = TREE_CMP(to->type == OBJECT_RANGE, from->type ==
1289 	    OBJECT_RANGE);
1290 	if (unlikely(cmp))
1291 		return (cmp);
1292 	cmp = TREE_CMP(to->type == OBJECT, from->type == OBJECT);
1293 	if (unlikely(cmp))
1294 		return (cmp);
1295 	if (from->end_blkid <= to->start_blkid)
1296 		return (-1);
1297 	if (from->start_blkid >= to->end_blkid)
1298 		return (1);
1299 	return (0);
1300 }
1301 
1302 /*
1303  * Pop the new data off the queue, check that the records we receive are in
1304  * the right order, but do not free the old data.  This is used so that the
1305  * records can be sent on to the main thread without copying the data.
1306  */
1307 static struct send_range *
1308 get_next_range_nofree(bqueue_t *bq, struct send_range *prev)
1309 {
1310 	struct send_range *next = bqueue_dequeue(bq);
1311 	ASSERT3S(send_range_after(prev, next), ==, -1);
1312 	return (next);
1313 }
1314 
1315 /*
1316  * Pop the new data off the queue, check that the records we receive are in
1317  * the right order, and free the old data.
1318  */
1319 static struct send_range *
1320 get_next_range(bqueue_t *bq, struct send_range *prev)
1321 {
1322 	struct send_range *next = get_next_range_nofree(bq, prev);
1323 	range_free(prev);
1324 	return (next);
1325 }
1326 
1327 static __attribute__((noreturn)) void
1328 redact_list_thread(void *arg)
1329 {
1330 	struct redact_list_thread_arg *rlt_arg = arg;
1331 	struct send_range *record;
1332 	fstrans_cookie_t cookie = spl_fstrans_mark();
1333 	if (rlt_arg->rl != NULL) {
1334 		struct redact_list_cb_arg rlcba = {0};
1335 		rlcba.cancel = &rlt_arg->cancel;
1336 		rlcba.q = &rlt_arg->q;
1337 		rlcba.num_blocks_visited = rlt_arg->num_blocks_visited;
1338 		rlcba.mark_redact = rlt_arg->mark_redact;
1339 		int err = dsl_redaction_list_traverse(rlt_arg->rl,
1340 		    &rlt_arg->resume, redact_list_cb, &rlcba);
1341 		if (err != EINTR)
1342 			rlt_arg->error_code = err;
1343 	}
1344 	record = range_alloc(DATA, 0, 0, 0, B_TRUE);
1345 	bqueue_enqueue_flush(&rlt_arg->q, record, sizeof (*record));
1346 	spl_fstrans_unmark(cookie);
1347 
1348 	thread_exit();
1349 }
1350 
1351 /*
1352  * Compare the start point of the two provided ranges. End of stream ranges
1353  * compare last, objects compare before any data or hole inside that object and
1354  * multi-object holes that start at the same object.
1355  */
1356 static int
1357 send_range_start_compare(struct send_range *r1, struct send_range *r2)
1358 {
1359 	uint64_t r1_objequiv = r1->object;
1360 	uint64_t r1_l0equiv = r1->start_blkid;
1361 	uint64_t r2_objequiv = r2->object;
1362 	uint64_t r2_l0equiv = r2->start_blkid;
1363 	int64_t cmp = TREE_CMP(r1->eos_marker, r2->eos_marker);
1364 	if (unlikely(cmp))
1365 		return (cmp);
1366 	if (r1->object == 0) {
1367 		r1_objequiv = r1->start_blkid * DNODES_PER_BLOCK;
1368 		r1_l0equiv = 0;
1369 	}
1370 	if (r2->object == 0) {
1371 		r2_objequiv = r2->start_blkid * DNODES_PER_BLOCK;
1372 		r2_l0equiv = 0;
1373 	}
1374 
1375 	cmp = TREE_CMP(r1_objequiv, r2_objequiv);
1376 	if (likely(cmp))
1377 		return (cmp);
1378 	cmp = TREE_CMP(r2->type == OBJECT_RANGE, r1->type == OBJECT_RANGE);
1379 	if (unlikely(cmp))
1380 		return (cmp);
1381 	cmp = TREE_CMP(r2->type == OBJECT, r1->type == OBJECT);
1382 	if (unlikely(cmp))
1383 		return (cmp);
1384 
1385 	return (TREE_CMP(r1_l0equiv, r2_l0equiv));
1386 }
1387 
1388 enum q_idx {
1389 	REDACT_IDX = 0,
1390 	TO_IDX,
1391 	FROM_IDX,
1392 	NUM_THREADS
1393 };
1394 
1395 /*
1396  * This function returns the next range the send_merge_thread should operate on.
1397  * The inputs are two arrays; the first one stores the range at the front of the
1398  * queues stored in the second one.  The ranges are sorted in descending
1399  * priority order; the metadata from earlier ranges overrules metadata from
1400  * later ranges.  out_mask is used to return which threads the ranges came from;
1401  * bit i is set if ranges[i] started at the same place as the returned range.
1402  *
1403  * This code is not hardcoded to compare a specific number of threads; it could
1404  * be used with any number, just by changing the q_idx enum.
1405  *
1406  * The "next range" is the one with the earliest start; if two starts are equal,
1407  * the highest-priority range is the next to operate on.  If a higher-priority
1408  * range starts in the middle of the first range, then the first range will be
1409  * truncated to end where the higher-priority range starts, and we will operate
1410  * on that one next time.   In this way, we make sure that each block covered by
1411  * some range gets covered by a returned range, and each block covered is
1412  * returned using the metadata of the highest-priority range it appears in.
1413  *
1414  * For example, if the three ranges at the front of the queues were [2,4),
1415  * [3,5), and [1,3), then the ranges returned would be [1,2) with the metadata
1416  * from the third range, [2,4) with the metadata from the first range, and then
1417  * [4,5) with the metadata from the second.
1418  */
1419 static struct send_range *
1420 find_next_range(struct send_range **ranges, bqueue_t **qs, uint64_t *out_mask)
1421 {
1422 	int idx = 0; // index of the range with the earliest start
1423 	int i;
1424 	uint64_t bmask = 0;
1425 	for (i = 1; i < NUM_THREADS; i++) {
1426 		if (send_range_start_compare(ranges[i], ranges[idx]) < 0)
1427 			idx = i;
1428 	}
1429 	if (ranges[idx]->eos_marker) {
1430 		struct send_range *ret = range_alloc(DATA, 0, 0, 0, B_TRUE);
1431 		*out_mask = 0;
1432 		return (ret);
1433 	}
1434 	/*
1435 	 * Find all the ranges that start at that same point.
1436 	 */
1437 	for (i = 0; i < NUM_THREADS; i++) {
1438 		if (send_range_start_compare(ranges[i], ranges[idx]) == 0)
1439 			bmask |= 1 << i;
1440 	}
1441 	*out_mask = bmask;
1442 	/*
1443 	 * OBJECT_RANGE records only come from the TO thread, and should always
1444 	 * be treated as overlapping with nothing and sent on immediately.  They
1445 	 * are only used in raw sends, and are never redacted.
1446 	 */
1447 	if (ranges[idx]->type == OBJECT_RANGE) {
1448 		ASSERT3U(idx, ==, TO_IDX);
1449 		ASSERT3U(*out_mask, ==, 1 << TO_IDX);
1450 		struct send_range *ret = ranges[idx];
1451 		ranges[idx] = get_next_range_nofree(qs[idx], ranges[idx]);
1452 		return (ret);
1453 	}
1454 	/*
1455 	 * Find the first start or end point after the start of the first range.
1456 	 */
1457 	uint64_t first_change = ranges[idx]->end_blkid;
1458 	for (i = 0; i < NUM_THREADS; i++) {
1459 		if (i == idx || ranges[i]->eos_marker ||
1460 		    ranges[i]->object > ranges[idx]->object ||
1461 		    ranges[i]->object == DMU_META_DNODE_OBJECT)
1462 			continue;
1463 		ASSERT3U(ranges[i]->object, ==, ranges[idx]->object);
1464 		if (first_change > ranges[i]->start_blkid &&
1465 		    (bmask & (1 << i)) == 0)
1466 			first_change = ranges[i]->start_blkid;
1467 		else if (first_change > ranges[i]->end_blkid)
1468 			first_change = ranges[i]->end_blkid;
1469 	}
1470 	/*
1471 	 * Update all ranges to no longer overlap with the range we're
1472 	 * returning. All such ranges must start at the same place as the range
1473 	 * being returned, and end at or after first_change. Thus we update
1474 	 * their start to first_change. If that makes them size 0, then free
1475 	 * them and pull a new range from that thread.
1476 	 */
1477 	for (i = 0; i < NUM_THREADS; i++) {
1478 		if (i == idx || (bmask & (1 << i)) == 0)
1479 			continue;
1480 		ASSERT3U(first_change, >, ranges[i]->start_blkid);
1481 		ranges[i]->start_blkid = first_change;
1482 		ASSERT3U(ranges[i]->start_blkid, <=, ranges[i]->end_blkid);
1483 		if (ranges[i]->start_blkid == ranges[i]->end_blkid)
1484 			ranges[i] = get_next_range(qs[i], ranges[i]);
1485 	}
1486 	/*
1487 	 * Short-circuit the simple case; if the range doesn't overlap with
1488 	 * anything else, or it only overlaps with things that start at the same
1489 	 * place and are longer, send it on.
1490 	 */
1491 	if (first_change == ranges[idx]->end_blkid) {
1492 		struct send_range *ret = ranges[idx];
1493 		ranges[idx] = get_next_range_nofree(qs[idx], ranges[idx]);
1494 		return (ret);
1495 	}
1496 
1497 	/*
1498 	 * Otherwise, return a truncated copy of ranges[idx] and move the start
1499 	 * of ranges[idx] back to first_change.
1500 	 */
1501 	struct send_range *ret = kmem_alloc(sizeof (*ret), KM_SLEEP);
1502 	*ret = *ranges[idx];
1503 	ret->end_blkid = first_change;
1504 	ranges[idx]->start_blkid = first_change;
1505 	return (ret);
1506 }
1507 
1508 #define	FROM_AND_REDACT_BITS ((1 << REDACT_IDX) | (1 << FROM_IDX))
1509 
1510 /*
1511  * Merge the results from the from thread and the to thread, and then hand the
1512  * records off to send_prefetch_thread to prefetch them.  If this is not a
1513  * send from a redaction bookmark, the from thread will push an end of stream
1514  * record and stop, and we'll just send everything that was changed in the
1515  * to_ds since the ancestor's creation txg. If it is, then since
1516  * traverse_dataset has a canonical order, we can compare each change as
1517  * they're pulled off the queues.  That will give us a stream that is
1518  * appropriately sorted, and covers all records.  In addition, we pull the
1519  * data from the redact_list_thread and use that to determine which blocks
1520  * should be redacted.
1521  */
1522 static __attribute__((noreturn)) void
1523 send_merge_thread(void *arg)
1524 {
1525 	struct send_merge_thread_arg *smt_arg = arg;
1526 	struct send_range *front_ranges[NUM_THREADS];
1527 	bqueue_t *queues[NUM_THREADS];
1528 	int err = 0;
1529 	fstrans_cookie_t cookie = spl_fstrans_mark();
1530 
1531 	if (smt_arg->redact_arg == NULL) {
1532 		front_ranges[REDACT_IDX] =
1533 		    kmem_zalloc(sizeof (struct send_range), KM_SLEEP);
1534 		front_ranges[REDACT_IDX]->eos_marker = B_TRUE;
1535 		front_ranges[REDACT_IDX]->type = REDACT;
1536 		queues[REDACT_IDX] = NULL;
1537 	} else {
1538 		front_ranges[REDACT_IDX] =
1539 		    bqueue_dequeue(&smt_arg->redact_arg->q);
1540 		queues[REDACT_IDX] = &smt_arg->redact_arg->q;
1541 	}
1542 	front_ranges[TO_IDX] = bqueue_dequeue(&smt_arg->to_arg->q);
1543 	queues[TO_IDX] = &smt_arg->to_arg->q;
1544 	front_ranges[FROM_IDX] = bqueue_dequeue(&smt_arg->from_arg->q);
1545 	queues[FROM_IDX] = &smt_arg->from_arg->q;
1546 	uint64_t mask = 0;
1547 	struct send_range *range;
1548 	for (range = find_next_range(front_ranges, queues, &mask);
1549 	    !range->eos_marker && err == 0 && !smt_arg->cancel;
1550 	    range = find_next_range(front_ranges, queues, &mask)) {
1551 		/*
1552 		 * If the range in question was in both the from redact bookmark
1553 		 * and the bookmark we're using to redact, then don't send it.
1554 		 * It's already redacted on the receiving system, so a redaction
1555 		 * record would be redundant.
1556 		 */
1557 		if ((mask & FROM_AND_REDACT_BITS) == FROM_AND_REDACT_BITS) {
1558 			ASSERT3U(range->type, ==, REDACT);
1559 			range_free(range);
1560 			continue;
1561 		}
1562 		bqueue_enqueue(&smt_arg->q, range, sizeof (*range));
1563 
1564 		if (smt_arg->to_arg->error_code != 0) {
1565 			err = smt_arg->to_arg->error_code;
1566 		} else if (smt_arg->from_arg->error_code != 0) {
1567 			err = smt_arg->from_arg->error_code;
1568 		} else if (smt_arg->redact_arg != NULL &&
1569 		    smt_arg->redact_arg->error_code != 0) {
1570 			err = smt_arg->redact_arg->error_code;
1571 		}
1572 	}
1573 	if (smt_arg->cancel && err == 0)
1574 		err = SET_ERROR(EINTR);
1575 	smt_arg->error = err;
1576 	if (smt_arg->error != 0) {
1577 		smt_arg->to_arg->cancel = B_TRUE;
1578 		smt_arg->from_arg->cancel = B_TRUE;
1579 		if (smt_arg->redact_arg != NULL)
1580 			smt_arg->redact_arg->cancel = B_TRUE;
1581 	}
1582 	for (int i = 0; i < NUM_THREADS; i++) {
1583 		while (!front_ranges[i]->eos_marker) {
1584 			front_ranges[i] = get_next_range(queues[i],
1585 			    front_ranges[i]);
1586 		}
1587 		range_free(front_ranges[i]);
1588 	}
1589 	if (range == NULL)
1590 		range = kmem_zalloc(sizeof (*range), KM_SLEEP);
1591 	range->eos_marker = B_TRUE;
1592 	bqueue_enqueue_flush(&smt_arg->q, range, 1);
1593 	spl_fstrans_unmark(cookie);
1594 	thread_exit();
1595 }
1596 
1597 struct send_reader_thread_arg {
1598 	struct send_merge_thread_arg *smta;
1599 	bqueue_t q;
1600 	boolean_t cancel;
1601 	boolean_t issue_reads;
1602 	uint64_t featureflags;
1603 	int error;
1604 };
1605 
1606 static void
1607 dmu_send_read_done(zio_t *zio)
1608 {
1609 	struct send_range *range = zio->io_private;
1610 
1611 	mutex_enter(&range->sru.data.lock);
1612 	if (zio->io_error != 0) {
1613 		abd_free(range->sru.data.abd);
1614 		range->sru.data.abd = NULL;
1615 		range->sru.data.io_err = zio->io_error;
1616 	}
1617 
1618 	ASSERT(range->sru.data.io_outstanding);
1619 	range->sru.data.io_outstanding = B_FALSE;
1620 	cv_broadcast(&range->sru.data.cv);
1621 	mutex_exit(&range->sru.data.lock);
1622 }
1623 
1624 static void
1625 issue_data_read(struct send_reader_thread_arg *srta, struct send_range *range)
1626 {
1627 	struct srd *srdp = &range->sru.data;
1628 	blkptr_t *bp = &srdp->bp;
1629 	objset_t *os = srta->smta->os;
1630 
1631 	ASSERT3U(range->type, ==, DATA);
1632 	ASSERT3U(range->start_blkid + 1, ==, range->end_blkid);
1633 	/*
1634 	 * If we have large blocks stored on disk but
1635 	 * the send flags don't allow us to send large
1636 	 * blocks, we split the data from the arc buf
1637 	 * into chunks.
1638 	 */
1639 	boolean_t split_large_blocks =
1640 	    srdp->datablksz > SPA_OLD_MAXBLOCKSIZE &&
1641 	    !(srta->featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS);
1642 	/*
1643 	 * We should only request compressed data from the ARC if all
1644 	 * the following are true:
1645 	 *  - stream compression was requested
1646 	 *  - we aren't splitting large blocks into smaller chunks
1647 	 *  - the data won't need to be byteswapped before sending
1648 	 *  - this isn't an embedded block
1649 	 *  - this isn't metadata (if receiving on a different endian
1650 	 *    system it can be byteswapped more easily)
1651 	 */
1652 	boolean_t request_compressed =
1653 	    (srta->featureflags & DMU_BACKUP_FEATURE_COMPRESSED) &&
1654 	    !split_large_blocks && !BP_SHOULD_BYTESWAP(bp) &&
1655 	    !BP_IS_EMBEDDED(bp) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp));
1656 
1657 	enum zio_flag zioflags = ZIO_FLAG_CANFAIL;
1658 
1659 	if (srta->featureflags & DMU_BACKUP_FEATURE_RAW) {
1660 		zioflags |= ZIO_FLAG_RAW;
1661 		srdp->io_compressed = B_TRUE;
1662 	} else if (request_compressed) {
1663 		zioflags |= ZIO_FLAG_RAW_COMPRESS;
1664 		srdp->io_compressed = B_TRUE;
1665 	}
1666 
1667 	srdp->datasz = (zioflags & ZIO_FLAG_RAW_COMPRESS) ?
1668 	    BP_GET_PSIZE(bp) : BP_GET_LSIZE(bp);
1669 
1670 	if (!srta->issue_reads)
1671 		return;
1672 	if (BP_IS_REDACTED(bp))
1673 		return;
1674 	if (send_do_embed(bp, srta->featureflags))
1675 		return;
1676 
1677 	zbookmark_phys_t zb = {
1678 	    .zb_objset = dmu_objset_id(os),
1679 	    .zb_object = range->object,
1680 	    .zb_level = 0,
1681 	    .zb_blkid = range->start_blkid,
1682 	};
1683 
1684 	arc_flags_t aflags = ARC_FLAG_CACHED_ONLY;
1685 
1686 	int arc_err = arc_read(NULL, os->os_spa, bp,
1687 	    arc_getbuf_func, &srdp->abuf, ZIO_PRIORITY_ASYNC_READ,
1688 	    zioflags, &aflags, &zb);
1689 	/*
1690 	 * If the data is not already cached in the ARC, we read directly
1691 	 * from zio.  This avoids the performance overhead of adding a new
1692 	 * entry to the ARC, and we also avoid polluting the ARC cache with
1693 	 * data that is not likely to be used in the future.
1694 	 */
1695 	if (arc_err != 0) {
1696 		srdp->abd = abd_alloc_linear(srdp->datasz, B_FALSE);
1697 		srdp->io_outstanding = B_TRUE;
1698 		zio_nowait(zio_read(NULL, os->os_spa, bp, srdp->abd,
1699 		    srdp->datasz, dmu_send_read_done, range,
1700 		    ZIO_PRIORITY_ASYNC_READ, zioflags, &zb));
1701 	}
1702 }
1703 
1704 /*
1705  * Create a new record with the given values.
1706  */
1707 static void
1708 enqueue_range(struct send_reader_thread_arg *srta, bqueue_t *q, dnode_t *dn,
1709     uint64_t blkid, uint64_t count, const blkptr_t *bp, uint32_t datablksz)
1710 {
1711 	enum type range_type = (bp == NULL || BP_IS_HOLE(bp) ? HOLE :
1712 	    (BP_IS_REDACTED(bp) ? REDACT : DATA));
1713 
1714 	struct send_range *range = range_alloc(range_type, dn->dn_object,
1715 	    blkid, blkid + count, B_FALSE);
1716 
1717 	if (blkid == DMU_SPILL_BLKID)
1718 		ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_SA);
1719 
1720 	switch (range_type) {
1721 	case HOLE:
1722 		range->sru.hole.datablksz = datablksz;
1723 		break;
1724 	case DATA:
1725 		ASSERT3U(count, ==, 1);
1726 		range->sru.data.datablksz = datablksz;
1727 		range->sru.data.obj_type = dn->dn_type;
1728 		range->sru.data.bp = *bp;
1729 		issue_data_read(srta, range);
1730 		break;
1731 	case REDACT:
1732 		range->sru.redact.datablksz = datablksz;
1733 		break;
1734 	default:
1735 		break;
1736 	}
1737 	bqueue_enqueue(q, range, datablksz);
1738 }
1739 
1740 /*
1741  * This thread is responsible for two things: First, it retrieves the correct
1742  * blkptr in the to ds if we need to send the data because of something from
1743  * the from thread.  As a result of this, we're the first ones to discover that
1744  * some indirect blocks can be discarded because they're not holes. Second,
1745  * it issues prefetches for the data we need to send.
1746  */
1747 static __attribute__((noreturn)) void
1748 send_reader_thread(void *arg)
1749 {
1750 	struct send_reader_thread_arg *srta = arg;
1751 	struct send_merge_thread_arg *smta = srta->smta;
1752 	bqueue_t *inq = &smta->q;
1753 	bqueue_t *outq = &srta->q;
1754 	objset_t *os = smta->os;
1755 	fstrans_cookie_t cookie = spl_fstrans_mark();
1756 	struct send_range *range = bqueue_dequeue(inq);
1757 	int err = 0;
1758 
1759 	/*
1760 	 * If the record we're analyzing is from a redaction bookmark from the
1761 	 * fromds, then we need to know whether or not it exists in the tods so
1762 	 * we know whether to create records for it or not. If it does, we need
1763 	 * the datablksz so we can generate an appropriate record for it.
1764 	 * Finally, if it isn't redacted, we need the blkptr so that we can send
1765 	 * a WRITE record containing the actual data.
1766 	 */
1767 	uint64_t last_obj = UINT64_MAX;
1768 	uint64_t last_obj_exists = B_TRUE;
1769 	while (!range->eos_marker && !srta->cancel && smta->error == 0 &&
1770 	    err == 0) {
1771 		switch (range->type) {
1772 		case DATA:
1773 			issue_data_read(srta, range);
1774 			bqueue_enqueue(outq, range, range->sru.data.datablksz);
1775 			range = get_next_range_nofree(inq, range);
1776 			break;
1777 		case HOLE:
1778 		case OBJECT:
1779 		case OBJECT_RANGE:
1780 		case REDACT: // Redacted blocks must exist
1781 			bqueue_enqueue(outq, range, sizeof (*range));
1782 			range = get_next_range_nofree(inq, range);
1783 			break;
1784 		case PREVIOUSLY_REDACTED: {
1785 			/*
1786 			 * This entry came from the "from bookmark" when
1787 			 * sending from a bookmark that has a redaction
1788 			 * list.  We need to check if this object/blkid
1789 			 * exists in the target ("to") dataset, and if
1790 			 * not then we drop this entry.  We also need
1791 			 * to fill in the block pointer so that we know
1792 			 * what to prefetch.
1793 			 *
1794 			 * To accomplish the above, we first cache whether or
1795 			 * not the last object we examined exists.  If it
1796 			 * doesn't, we can drop this record. If it does, we hold
1797 			 * the dnode and use it to call dbuf_dnode_findbp. We do
1798 			 * this instead of dbuf_bookmark_findbp because we will
1799 			 * often operate on large ranges, and holding the dnode
1800 			 * once is more efficient.
1801 			 */
1802 			boolean_t object_exists = B_TRUE;
1803 			/*
1804 			 * If the data is redacted, we only care if it exists,
1805 			 * so that we don't send records for objects that have
1806 			 * been deleted.
1807 			 */
1808 			dnode_t *dn;
1809 			if (range->object == last_obj && !last_obj_exists) {
1810 				/*
1811 				 * If we're still examining the same object as
1812 				 * previously, and it doesn't exist, we don't
1813 				 * need to call dbuf_bookmark_findbp.
1814 				 */
1815 				object_exists = B_FALSE;
1816 			} else {
1817 				err = dnode_hold(os, range->object, FTAG, &dn);
1818 				if (err == ENOENT) {
1819 					object_exists = B_FALSE;
1820 					err = 0;
1821 				}
1822 				last_obj = range->object;
1823 				last_obj_exists = object_exists;
1824 			}
1825 
1826 			if (err != 0) {
1827 				break;
1828 			} else if (!object_exists) {
1829 				/*
1830 				 * The block was modified, but doesn't
1831 				 * exist in the to dataset; if it was
1832 				 * deleted in the to dataset, then we'll
1833 				 * visit the hole bp for it at some point.
1834 				 */
1835 				range = get_next_range(inq, range);
1836 				continue;
1837 			}
1838 			uint64_t file_max =
1839 			    (dn->dn_maxblkid < range->end_blkid ?
1840 			    dn->dn_maxblkid : range->end_blkid);
1841 			/*
1842 			 * The object exists, so we need to try to find the
1843 			 * blkptr for each block in the range we're processing.
1844 			 */
1845 			rw_enter(&dn->dn_struct_rwlock, RW_READER);
1846 			for (uint64_t blkid = range->start_blkid;
1847 			    blkid < file_max; blkid++) {
1848 				blkptr_t bp;
1849 				uint32_t datablksz =
1850 				    dn->dn_phys->dn_datablkszsec <<
1851 				    SPA_MINBLOCKSHIFT;
1852 				uint64_t offset = blkid * datablksz;
1853 				/*
1854 				 * This call finds the next non-hole block in
1855 				 * the object. This is to prevent a
1856 				 * performance problem where we're unredacting
1857 				 * a large hole. Using dnode_next_offset to
1858 				 * skip over the large hole avoids iterating
1859 				 * over every block in it.
1860 				 */
1861 				err = dnode_next_offset(dn, DNODE_FIND_HAVELOCK,
1862 				    &offset, 1, 1, 0);
1863 				if (err == ESRCH) {
1864 					offset = UINT64_MAX;
1865 					err = 0;
1866 				} else if (err != 0) {
1867 					break;
1868 				}
1869 				if (offset != blkid * datablksz) {
1870 					/*
1871 					 * if there is a hole from here
1872 					 * (blkid) to offset
1873 					 */
1874 					offset = MIN(offset, file_max *
1875 					    datablksz);
1876 					uint64_t nblks = (offset / datablksz) -
1877 					    blkid;
1878 					enqueue_range(srta, outq, dn, blkid,
1879 					    nblks, NULL, datablksz);
1880 					blkid += nblks;
1881 				}
1882 				if (blkid >= file_max)
1883 					break;
1884 				err = dbuf_dnode_findbp(dn, 0, blkid, &bp,
1885 				    NULL, NULL);
1886 				if (err != 0)
1887 					break;
1888 				ASSERT(!BP_IS_HOLE(&bp));
1889 				enqueue_range(srta, outq, dn, blkid, 1, &bp,
1890 				    datablksz);
1891 			}
1892 			rw_exit(&dn->dn_struct_rwlock);
1893 			dnode_rele(dn, FTAG);
1894 			range = get_next_range(inq, range);
1895 		}
1896 		}
1897 	}
1898 	if (srta->cancel || err != 0) {
1899 		smta->cancel = B_TRUE;
1900 		srta->error = err;
1901 	} else if (smta->error != 0) {
1902 		srta->error = smta->error;
1903 	}
1904 	while (!range->eos_marker)
1905 		range = get_next_range(inq, range);
1906 
1907 	bqueue_enqueue_flush(outq, range, 1);
1908 	spl_fstrans_unmark(cookie);
1909 	thread_exit();
1910 }
1911 
1912 #define	NUM_SNAPS_NOT_REDACTED UINT64_MAX
1913 
1914 struct dmu_send_params {
1915 	/* Pool args */
1916 	void *tag; // Tag that dp was held with, will be used to release dp.
1917 	dsl_pool_t *dp;
1918 	/* To snapshot args */
1919 	const char *tosnap;
1920 	dsl_dataset_t *to_ds;
1921 	/* From snapshot args */
1922 	zfs_bookmark_phys_t ancestor_zb;
1923 	uint64_t *fromredactsnaps;
1924 	/* NUM_SNAPS_NOT_REDACTED if not sending from redaction bookmark */
1925 	uint64_t numfromredactsnaps;
1926 	/* Stream params */
1927 	boolean_t is_clone;
1928 	boolean_t embedok;
1929 	boolean_t large_block_ok;
1930 	boolean_t compressok;
1931 	boolean_t rawok;
1932 	boolean_t savedok;
1933 	uint64_t resumeobj;
1934 	uint64_t resumeoff;
1935 	uint64_t saved_guid;
1936 	zfs_bookmark_phys_t *redactbook;
1937 	/* Stream output params */
1938 	dmu_send_outparams_t *dso;
1939 
1940 	/* Stream progress params */
1941 	offset_t *off;
1942 	int outfd;
1943 	char saved_toname[MAXNAMELEN];
1944 };
1945 
1946 static int
1947 setup_featureflags(struct dmu_send_params *dspp, objset_t *os,
1948     uint64_t *featureflags)
1949 {
1950 	dsl_dataset_t *to_ds = dspp->to_ds;
1951 	dsl_pool_t *dp = dspp->dp;
1952 #ifdef _KERNEL
1953 	if (dmu_objset_type(os) == DMU_OST_ZFS) {
1954 		uint64_t version;
1955 		if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0)
1956 			return (SET_ERROR(EINVAL));
1957 
1958 		if (version >= ZPL_VERSION_SA)
1959 			*featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1960 	}
1961 #endif
1962 
1963 	/* raw sends imply large_block_ok */
1964 	if ((dspp->rawok || dspp->large_block_ok) &&
1965 	    dsl_dataset_feature_is_active(to_ds, SPA_FEATURE_LARGE_BLOCKS)) {
1966 		*featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS;
1967 	}
1968 
1969 	/* encrypted datasets will not have embedded blocks */
1970 	if ((dspp->embedok || dspp->rawok) && !os->os_encrypted &&
1971 	    spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) {
1972 		*featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA;
1973 	}
1974 
1975 	/* raw send implies compressok */
1976 	if (dspp->compressok || dspp->rawok)
1977 		*featureflags |= DMU_BACKUP_FEATURE_COMPRESSED;
1978 
1979 	if (dspp->rawok && os->os_encrypted)
1980 		*featureflags |= DMU_BACKUP_FEATURE_RAW;
1981 
1982 	if ((*featureflags &
1983 	    (DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_COMPRESSED |
1984 	    DMU_BACKUP_FEATURE_RAW)) != 0 &&
1985 	    spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) {
1986 		*featureflags |= DMU_BACKUP_FEATURE_LZ4;
1987 	}
1988 
1989 	/*
1990 	 * We specifically do not include DMU_BACKUP_FEATURE_EMBED_DATA here to
1991 	 * allow sending ZSTD compressed datasets to a receiver that does not
1992 	 * support ZSTD
1993 	 */
1994 	if ((*featureflags &
1995 	    (DMU_BACKUP_FEATURE_COMPRESSED | DMU_BACKUP_FEATURE_RAW)) != 0 &&
1996 	    dsl_dataset_feature_is_active(to_ds, SPA_FEATURE_ZSTD_COMPRESS)) {
1997 		*featureflags |= DMU_BACKUP_FEATURE_ZSTD;
1998 	}
1999 
2000 	if (dspp->resumeobj != 0 || dspp->resumeoff != 0) {
2001 		*featureflags |= DMU_BACKUP_FEATURE_RESUMING;
2002 	}
2003 
2004 	if (dspp->redactbook != NULL) {
2005 		*featureflags |= DMU_BACKUP_FEATURE_REDACTED;
2006 	}
2007 
2008 	if (dsl_dataset_feature_is_active(to_ds, SPA_FEATURE_LARGE_DNODE)) {
2009 		*featureflags |= DMU_BACKUP_FEATURE_LARGE_DNODE;
2010 	}
2011 	return (0);
2012 }
2013 
2014 static dmu_replay_record_t *
2015 create_begin_record(struct dmu_send_params *dspp, objset_t *os,
2016     uint64_t featureflags)
2017 {
2018 	dmu_replay_record_t *drr = kmem_zalloc(sizeof (dmu_replay_record_t),
2019 	    KM_SLEEP);
2020 	drr->drr_type = DRR_BEGIN;
2021 
2022 	struct drr_begin *drrb = &drr->drr_u.drr_begin;
2023 	dsl_dataset_t *to_ds = dspp->to_ds;
2024 
2025 	drrb->drr_magic = DMU_BACKUP_MAGIC;
2026 	drrb->drr_creation_time = dsl_dataset_phys(to_ds)->ds_creation_time;
2027 	drrb->drr_type = dmu_objset_type(os);
2028 	drrb->drr_toguid = dsl_dataset_phys(to_ds)->ds_guid;
2029 	drrb->drr_fromguid = dspp->ancestor_zb.zbm_guid;
2030 
2031 	DMU_SET_STREAM_HDRTYPE(drrb->drr_versioninfo, DMU_SUBSTREAM);
2032 	DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, featureflags);
2033 
2034 	if (dspp->is_clone)
2035 		drrb->drr_flags |= DRR_FLAG_CLONE;
2036 	if (dsl_dataset_phys(dspp->to_ds)->ds_flags & DS_FLAG_CI_DATASET)
2037 		drrb->drr_flags |= DRR_FLAG_CI_DATA;
2038 	if (zfs_send_set_freerecords_bit)
2039 		drrb->drr_flags |= DRR_FLAG_FREERECORDS;
2040 	drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_SPILL_BLOCK;
2041 
2042 	if (dspp->savedok) {
2043 		drrb->drr_toguid = dspp->saved_guid;
2044 		strlcpy(drrb->drr_toname, dspp->saved_toname,
2045 		    sizeof (drrb->drr_toname));
2046 	} else {
2047 		dsl_dataset_name(to_ds, drrb->drr_toname);
2048 		if (!to_ds->ds_is_snapshot) {
2049 			(void) strlcat(drrb->drr_toname, "@--head--",
2050 			    sizeof (drrb->drr_toname));
2051 		}
2052 	}
2053 	return (drr);
2054 }
2055 
2056 static void
2057 setup_to_thread(struct send_thread_arg *to_arg, objset_t *to_os,
2058     dmu_sendstatus_t *dssp, uint64_t fromtxg, boolean_t rawok)
2059 {
2060 	VERIFY0(bqueue_init(&to_arg->q, zfs_send_no_prefetch_queue_ff,
2061 	    MAX(zfs_send_no_prefetch_queue_length, 2 * zfs_max_recordsize),
2062 	    offsetof(struct send_range, ln)));
2063 	to_arg->error_code = 0;
2064 	to_arg->cancel = B_FALSE;
2065 	to_arg->os = to_os;
2066 	to_arg->fromtxg = fromtxg;
2067 	to_arg->flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA;
2068 	if (rawok)
2069 		to_arg->flags |= TRAVERSE_NO_DECRYPT;
2070 	if (zfs_send_corrupt_data)
2071 		to_arg->flags |= TRAVERSE_HARD;
2072 	to_arg->num_blocks_visited = &dssp->dss_blocks;
2073 	(void) thread_create(NULL, 0, send_traverse_thread, to_arg, 0,
2074 	    curproc, TS_RUN, minclsyspri);
2075 }
2076 
2077 static void
2078 setup_from_thread(struct redact_list_thread_arg *from_arg,
2079     redaction_list_t *from_rl, dmu_sendstatus_t *dssp)
2080 {
2081 	VERIFY0(bqueue_init(&from_arg->q, zfs_send_no_prefetch_queue_ff,
2082 	    MAX(zfs_send_no_prefetch_queue_length, 2 * zfs_max_recordsize),
2083 	    offsetof(struct send_range, ln)));
2084 	from_arg->error_code = 0;
2085 	from_arg->cancel = B_FALSE;
2086 	from_arg->rl = from_rl;
2087 	from_arg->mark_redact = B_FALSE;
2088 	from_arg->num_blocks_visited = &dssp->dss_blocks;
2089 	/*
2090 	 * If from_ds is null, send_traverse_thread just returns success and
2091 	 * enqueues an eos marker.
2092 	 */
2093 	(void) thread_create(NULL, 0, redact_list_thread, from_arg, 0,
2094 	    curproc, TS_RUN, minclsyspri);
2095 }
2096 
2097 static void
2098 setup_redact_list_thread(struct redact_list_thread_arg *rlt_arg,
2099     struct dmu_send_params *dspp, redaction_list_t *rl, dmu_sendstatus_t *dssp)
2100 {
2101 	if (dspp->redactbook == NULL)
2102 		return;
2103 
2104 	rlt_arg->cancel = B_FALSE;
2105 	VERIFY0(bqueue_init(&rlt_arg->q, zfs_send_no_prefetch_queue_ff,
2106 	    MAX(zfs_send_no_prefetch_queue_length, 2 * zfs_max_recordsize),
2107 	    offsetof(struct send_range, ln)));
2108 	rlt_arg->error_code = 0;
2109 	rlt_arg->mark_redact = B_TRUE;
2110 	rlt_arg->rl = rl;
2111 	rlt_arg->num_blocks_visited = &dssp->dss_blocks;
2112 
2113 	(void) thread_create(NULL, 0, redact_list_thread, rlt_arg, 0,
2114 	    curproc, TS_RUN, minclsyspri);
2115 }
2116 
2117 static void
2118 setup_merge_thread(struct send_merge_thread_arg *smt_arg,
2119     struct dmu_send_params *dspp, struct redact_list_thread_arg *from_arg,
2120     struct send_thread_arg *to_arg, struct redact_list_thread_arg *rlt_arg,
2121     objset_t *os)
2122 {
2123 	VERIFY0(bqueue_init(&smt_arg->q, zfs_send_no_prefetch_queue_ff,
2124 	    MAX(zfs_send_no_prefetch_queue_length, 2 * zfs_max_recordsize),
2125 	    offsetof(struct send_range, ln)));
2126 	smt_arg->cancel = B_FALSE;
2127 	smt_arg->error = 0;
2128 	smt_arg->from_arg = from_arg;
2129 	smt_arg->to_arg = to_arg;
2130 	if (dspp->redactbook != NULL)
2131 		smt_arg->redact_arg = rlt_arg;
2132 
2133 	smt_arg->os = os;
2134 	(void) thread_create(NULL, 0, send_merge_thread, smt_arg, 0, curproc,
2135 	    TS_RUN, minclsyspri);
2136 }
2137 
2138 static void
2139 setup_reader_thread(struct send_reader_thread_arg *srt_arg,
2140     struct dmu_send_params *dspp, struct send_merge_thread_arg *smt_arg,
2141     uint64_t featureflags)
2142 {
2143 	VERIFY0(bqueue_init(&srt_arg->q, zfs_send_queue_ff,
2144 	    MAX(zfs_send_queue_length, 2 * zfs_max_recordsize),
2145 	    offsetof(struct send_range, ln)));
2146 	srt_arg->smta = smt_arg;
2147 	srt_arg->issue_reads = !dspp->dso->dso_dryrun;
2148 	srt_arg->featureflags = featureflags;
2149 	(void) thread_create(NULL, 0, send_reader_thread, srt_arg, 0,
2150 	    curproc, TS_RUN, minclsyspri);
2151 }
2152 
2153 static int
2154 setup_resume_points(struct dmu_send_params *dspp,
2155     struct send_thread_arg *to_arg, struct redact_list_thread_arg *from_arg,
2156     struct redact_list_thread_arg *rlt_arg,
2157     struct send_merge_thread_arg *smt_arg, boolean_t resuming, objset_t *os,
2158     redaction_list_t *redact_rl, nvlist_t *nvl)
2159 {
2160 	(void) smt_arg;
2161 	dsl_dataset_t *to_ds = dspp->to_ds;
2162 	int err = 0;
2163 
2164 	uint64_t obj = 0;
2165 	uint64_t blkid = 0;
2166 	if (resuming) {
2167 		obj = dspp->resumeobj;
2168 		dmu_object_info_t to_doi;
2169 		err = dmu_object_info(os, obj, &to_doi);
2170 		if (err != 0)
2171 			return (err);
2172 
2173 		blkid = dspp->resumeoff / to_doi.doi_data_block_size;
2174 	}
2175 	/*
2176 	 * If we're resuming a redacted send, we can skip to the appropriate
2177 	 * point in the redaction bookmark by binary searching through it.
2178 	 */
2179 	if (redact_rl != NULL) {
2180 		SET_BOOKMARK(&rlt_arg->resume, to_ds->ds_object, obj, 0, blkid);
2181 	}
2182 
2183 	SET_BOOKMARK(&to_arg->resume, to_ds->ds_object, obj, 0, blkid);
2184 	if (nvlist_exists(nvl, BEGINNV_REDACT_FROM_SNAPS)) {
2185 		uint64_t objset = dspp->ancestor_zb.zbm_redaction_obj;
2186 		/*
2187 		 * Note: If the resume point is in an object whose
2188 		 * blocksize is different in the from vs to snapshots,
2189 		 * we will have divided by the "wrong" blocksize.
2190 		 * However, in this case fromsnap's send_cb() will
2191 		 * detect that the blocksize has changed and therefore
2192 		 * ignore this object.
2193 		 *
2194 		 * If we're resuming a send from a redaction bookmark,
2195 		 * we still cannot accidentally suggest blocks behind
2196 		 * the to_ds.  In addition, we know that any blocks in
2197 		 * the object in the to_ds will have to be sent, since
2198 		 * the size changed.  Therefore, we can't cause any harm
2199 		 * this way either.
2200 		 */
2201 		SET_BOOKMARK(&from_arg->resume, objset, obj, 0, blkid);
2202 	}
2203 	if (resuming) {
2204 		fnvlist_add_uint64(nvl, BEGINNV_RESUME_OBJECT, dspp->resumeobj);
2205 		fnvlist_add_uint64(nvl, BEGINNV_RESUME_OFFSET, dspp->resumeoff);
2206 	}
2207 	return (0);
2208 }
2209 
2210 static dmu_sendstatus_t *
2211 setup_send_progress(struct dmu_send_params *dspp)
2212 {
2213 	dmu_sendstatus_t *dssp = kmem_zalloc(sizeof (*dssp), KM_SLEEP);
2214 	dssp->dss_outfd = dspp->outfd;
2215 	dssp->dss_off = dspp->off;
2216 	dssp->dss_proc = curproc;
2217 	mutex_enter(&dspp->to_ds->ds_sendstream_lock);
2218 	list_insert_head(&dspp->to_ds->ds_sendstreams, dssp);
2219 	mutex_exit(&dspp->to_ds->ds_sendstream_lock);
2220 	return (dssp);
2221 }
2222 
2223 /*
2224  * Actually do the bulk of the work in a zfs send.
2225  *
2226  * The idea is that we want to do a send from ancestor_zb to to_ds.  We also
2227  * want to not send any data that has been modified by all the datasets in
2228  * redactsnaparr, and store the list of blocks that are redacted in this way in
2229  * a bookmark named redactbook, created on the to_ds.  We do this by creating
2230  * several worker threads, whose function is described below.
2231  *
2232  * There are three cases.
2233  * The first case is a redacted zfs send.  In this case there are 5 threads.
2234  * The first thread is the to_ds traversal thread: it calls dataset_traverse on
2235  * the to_ds and finds all the blocks that have changed since ancestor_zb (if
2236  * it's a full send, that's all blocks in the dataset).  It then sends those
2237  * blocks on to the send merge thread. The redact list thread takes the data
2238  * from the redaction bookmark and sends those blocks on to the send merge
2239  * thread.  The send merge thread takes the data from the to_ds traversal
2240  * thread, and combines it with the redaction records from the redact list
2241  * thread.  If a block appears in both the to_ds's data and the redaction data,
2242  * the send merge thread will mark it as redacted and send it on to the prefetch
2243  * thread.  Otherwise, the send merge thread will send the block on to the
2244  * prefetch thread unchanged. The prefetch thread will issue prefetch reads for
2245  * any data that isn't redacted, and then send the data on to the main thread.
2246  * The main thread behaves the same as in a normal send case, issuing demand
2247  * reads for data blocks and sending out records over the network
2248  *
2249  * The graphic below diagrams the flow of data in the case of a redacted zfs
2250  * send.  Each box represents a thread, and each line represents the flow of
2251  * data.
2252  *
2253  *             Records from the |
2254  *           redaction bookmark |
2255  * +--------------------+       |  +---------------------------+
2256  * |                    |       v  | Send Merge Thread         |
2257  * | Redact List Thread +----------> Apply redaction marks to  |
2258  * |                    |          | records as specified by   |
2259  * +--------------------+          | redaction ranges          |
2260  *                                 +----^---------------+------+
2261  *                                      |               | Merged data
2262  *                                      |               |
2263  *                                      |  +------------v--------+
2264  *                                      |  | Prefetch Thread     |
2265  * +--------------------+               |  | Issues prefetch     |
2266  * | to_ds Traversal    |               |  | reads of data blocks|
2267  * | Thread (finds      +---------------+  +------------+--------+
2268  * | candidate blocks)  |  Blocks modified              | Prefetched data
2269  * +--------------------+  by to_ds since               |
2270  *                         ancestor_zb     +------------v----+
2271  *                                         | Main Thread     |  File Descriptor
2272  *                                         | Sends data over +->(to zfs receive)
2273  *                                         | wire            |
2274  *                                         +-----------------+
2275  *
2276  * The second case is an incremental send from a redaction bookmark.  The to_ds
2277  * traversal thread and the main thread behave the same as in the redacted
2278  * send case.  The new thread is the from bookmark traversal thread.  It
2279  * iterates over the redaction list in the redaction bookmark, and enqueues
2280  * records for each block that was redacted in the original send.  The send
2281  * merge thread now has to merge the data from the two threads.  For details
2282  * about that process, see the header comment of send_merge_thread().  Any data
2283  * it decides to send on will be prefetched by the prefetch thread.  Note that
2284  * you can perform a redacted send from a redaction bookmark; in that case,
2285  * the data flow behaves very similarly to the flow in the redacted send case,
2286  * except with the addition of the bookmark traversal thread iterating over the
2287  * redaction bookmark.  The send_merge_thread also has to take on the
2288  * responsibility of merging the redact list thread's records, the bookmark
2289  * traversal thread's records, and the to_ds records.
2290  *
2291  * +---------------------+
2292  * |                     |
2293  * | Redact List Thread  +--------------+
2294  * |                     |              |
2295  * +---------------------+              |
2296  *        Blocks in redaction list      | Ranges modified by every secure snap
2297  *        of from bookmark              | (or EOS if not readcted)
2298  *                                      |
2299  * +---------------------+   |     +----v----------------------+
2300  * | bookmark Traversal  |   v     | Send Merge Thread         |
2301  * | Thread (finds       +---------> Merges bookmark, rlt, and |
2302  * | candidate blocks)   |         | to_ds send records        |
2303  * +---------------------+         +----^---------------+------+
2304  *                                      |               | Merged data
2305  *                                      |  +------------v--------+
2306  *                                      |  | Prefetch Thread     |
2307  * +--------------------+               |  | Issues prefetch     |
2308  * | to_ds Traversal    |               |  | reads of data blocks|
2309  * | Thread (finds      +---------------+  +------------+--------+
2310  * | candidate blocks)  |  Blocks modified              | Prefetched data
2311  * +--------------------+  by to_ds since  +------------v----+
2312  *                         ancestor_zb     | Main Thread     |  File Descriptor
2313  *                                         | Sends data over +->(to zfs receive)
2314  *                                         | wire            |
2315  *                                         +-----------------+
2316  *
2317  * The final case is a simple zfs full or incremental send.  The to_ds traversal
2318  * thread behaves the same as always. The redact list thread is never started.
2319  * The send merge thread takes all the blocks that the to_ds traversal thread
2320  * sends it, prefetches the data, and sends the blocks on to the main thread.
2321  * The main thread sends the data over the wire.
2322  *
2323  * To keep performance acceptable, we want to prefetch the data in the worker
2324  * threads.  While the to_ds thread could simply use the TRAVERSE_PREFETCH
2325  * feature built into traverse_dataset, the combining and deletion of records
2326  * due to redaction and sends from redaction bookmarks mean that we could
2327  * issue many unnecessary prefetches.  As a result, we only prefetch data
2328  * after we've determined that the record is not going to be redacted.  To
2329  * prevent the prefetching from getting too far ahead of the main thread, the
2330  * blocking queues that are used for communication are capped not by the
2331  * number of entries in the queue, but by the sum of the size of the
2332  * prefetches associated with them.  The limit on the amount of data that the
2333  * thread can prefetch beyond what the main thread has reached is controlled
2334  * by the global variable zfs_send_queue_length.  In addition, to prevent poor
2335  * performance in the beginning of a send, we also limit the distance ahead
2336  * that the traversal threads can be.  That distance is controlled by the
2337  * zfs_send_no_prefetch_queue_length tunable.
2338  *
2339  * Note: Releases dp using the specified tag.
2340  */
2341 static int
2342 dmu_send_impl(struct dmu_send_params *dspp)
2343 {
2344 	objset_t *os;
2345 	dmu_replay_record_t *drr;
2346 	dmu_sendstatus_t *dssp;
2347 	dmu_send_cookie_t dsc = {0};
2348 	int err;
2349 	uint64_t fromtxg = dspp->ancestor_zb.zbm_creation_txg;
2350 	uint64_t featureflags = 0;
2351 	struct redact_list_thread_arg *from_arg;
2352 	struct send_thread_arg *to_arg;
2353 	struct redact_list_thread_arg *rlt_arg;
2354 	struct send_merge_thread_arg *smt_arg;
2355 	struct send_reader_thread_arg *srt_arg;
2356 	struct send_range *range;
2357 	redaction_list_t *from_rl = NULL;
2358 	redaction_list_t *redact_rl = NULL;
2359 	boolean_t resuming = (dspp->resumeobj != 0 || dspp->resumeoff != 0);
2360 	boolean_t book_resuming = resuming;
2361 
2362 	dsl_dataset_t *to_ds = dspp->to_ds;
2363 	zfs_bookmark_phys_t *ancestor_zb = &dspp->ancestor_zb;
2364 	dsl_pool_t *dp = dspp->dp;
2365 	void *tag = dspp->tag;
2366 
2367 	err = dmu_objset_from_ds(to_ds, &os);
2368 	if (err != 0) {
2369 		dsl_pool_rele(dp, tag);
2370 		return (err);
2371 	}
2372 
2373 	/*
2374 	 * If this is a non-raw send of an encrypted ds, we can ensure that
2375 	 * the objset_phys_t is authenticated. This is safe because this is
2376 	 * either a snapshot or we have owned the dataset, ensuring that
2377 	 * it can't be modified.
2378 	 */
2379 	if (!dspp->rawok && os->os_encrypted &&
2380 	    arc_is_unauthenticated(os->os_phys_buf)) {
2381 		zbookmark_phys_t zb;
2382 
2383 		SET_BOOKMARK(&zb, to_ds->ds_object, ZB_ROOT_OBJECT,
2384 		    ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
2385 		err = arc_untransform(os->os_phys_buf, os->os_spa,
2386 		    &zb, B_FALSE);
2387 		if (err != 0) {
2388 			dsl_pool_rele(dp, tag);
2389 			return (err);
2390 		}
2391 
2392 		ASSERT0(arc_is_unauthenticated(os->os_phys_buf));
2393 	}
2394 
2395 	if ((err = setup_featureflags(dspp, os, &featureflags)) != 0) {
2396 		dsl_pool_rele(dp, tag);
2397 		return (err);
2398 	}
2399 
2400 	/*
2401 	 * If we're doing a redacted send, hold the bookmark's redaction list.
2402 	 */
2403 	if (dspp->redactbook != NULL) {
2404 		err = dsl_redaction_list_hold_obj(dp,
2405 		    dspp->redactbook->zbm_redaction_obj, FTAG,
2406 		    &redact_rl);
2407 		if (err != 0) {
2408 			dsl_pool_rele(dp, tag);
2409 			return (SET_ERROR(EINVAL));
2410 		}
2411 		dsl_redaction_list_long_hold(dp, redact_rl, FTAG);
2412 	}
2413 
2414 	/*
2415 	 * If we're sending from a redaction bookmark, hold the redaction list
2416 	 * so that we can consider sending the redacted blocks.
2417 	 */
2418 	if (ancestor_zb->zbm_redaction_obj != 0) {
2419 		err = dsl_redaction_list_hold_obj(dp,
2420 		    ancestor_zb->zbm_redaction_obj, FTAG, &from_rl);
2421 		if (err != 0) {
2422 			if (redact_rl != NULL) {
2423 				dsl_redaction_list_long_rele(redact_rl, FTAG);
2424 				dsl_redaction_list_rele(redact_rl, FTAG);
2425 			}
2426 			dsl_pool_rele(dp, tag);
2427 			return (SET_ERROR(EINVAL));
2428 		}
2429 		dsl_redaction_list_long_hold(dp, from_rl, FTAG);
2430 	}
2431 
2432 	dsl_dataset_long_hold(to_ds, FTAG);
2433 
2434 	from_arg = kmem_zalloc(sizeof (*from_arg), KM_SLEEP);
2435 	to_arg = kmem_zalloc(sizeof (*to_arg), KM_SLEEP);
2436 	rlt_arg = kmem_zalloc(sizeof (*rlt_arg), KM_SLEEP);
2437 	smt_arg = kmem_zalloc(sizeof (*smt_arg), KM_SLEEP);
2438 	srt_arg = kmem_zalloc(sizeof (*srt_arg), KM_SLEEP);
2439 
2440 	drr = create_begin_record(dspp, os, featureflags);
2441 	dssp = setup_send_progress(dspp);
2442 
2443 	dsc.dsc_drr = drr;
2444 	dsc.dsc_dso = dspp->dso;
2445 	dsc.dsc_os = os;
2446 	dsc.dsc_off = dspp->off;
2447 	dsc.dsc_toguid = dsl_dataset_phys(to_ds)->ds_guid;
2448 	dsc.dsc_fromtxg = fromtxg;
2449 	dsc.dsc_pending_op = PENDING_NONE;
2450 	dsc.dsc_featureflags = featureflags;
2451 	dsc.dsc_resume_object = dspp->resumeobj;
2452 	dsc.dsc_resume_offset = dspp->resumeoff;
2453 
2454 	dsl_pool_rele(dp, tag);
2455 
2456 	void *payload = NULL;
2457 	size_t payload_len = 0;
2458 	nvlist_t *nvl = fnvlist_alloc();
2459 
2460 	/*
2461 	 * If we're doing a redacted send, we include the snapshots we're
2462 	 * redacted with respect to so that the target system knows what send
2463 	 * streams can be correctly received on top of this dataset. If we're
2464 	 * instead sending a redacted dataset, we include the snapshots that the
2465 	 * dataset was created with respect to.
2466 	 */
2467 	if (dspp->redactbook != NULL) {
2468 		fnvlist_add_uint64_array(nvl, BEGINNV_REDACT_SNAPS,
2469 		    redact_rl->rl_phys->rlp_snaps,
2470 		    redact_rl->rl_phys->rlp_num_snaps);
2471 	} else if (dsl_dataset_feature_is_active(to_ds,
2472 	    SPA_FEATURE_REDACTED_DATASETS)) {
2473 		uint64_t *tods_guids;
2474 		uint64_t length;
2475 		VERIFY(dsl_dataset_get_uint64_array_feature(to_ds,
2476 		    SPA_FEATURE_REDACTED_DATASETS, &length, &tods_guids));
2477 		fnvlist_add_uint64_array(nvl, BEGINNV_REDACT_SNAPS, tods_guids,
2478 		    length);
2479 	}
2480 
2481 	/*
2482 	 * If we're sending from a redaction bookmark, then we should retrieve
2483 	 * the guids of that bookmark so we can send them over the wire.
2484 	 */
2485 	if (from_rl != NULL) {
2486 		fnvlist_add_uint64_array(nvl, BEGINNV_REDACT_FROM_SNAPS,
2487 		    from_rl->rl_phys->rlp_snaps,
2488 		    from_rl->rl_phys->rlp_num_snaps);
2489 	}
2490 
2491 	/*
2492 	 * If the snapshot we're sending from is redacted, include the redaction
2493 	 * list in the stream.
2494 	 */
2495 	if (dspp->numfromredactsnaps != NUM_SNAPS_NOT_REDACTED) {
2496 		ASSERT3P(from_rl, ==, NULL);
2497 		fnvlist_add_uint64_array(nvl, BEGINNV_REDACT_FROM_SNAPS,
2498 		    dspp->fromredactsnaps, (uint_t)dspp->numfromredactsnaps);
2499 		if (dspp->numfromredactsnaps > 0) {
2500 			kmem_free(dspp->fromredactsnaps,
2501 			    dspp->numfromredactsnaps * sizeof (uint64_t));
2502 			dspp->fromredactsnaps = NULL;
2503 		}
2504 	}
2505 
2506 	if (resuming || book_resuming) {
2507 		err = setup_resume_points(dspp, to_arg, from_arg,
2508 		    rlt_arg, smt_arg, resuming, os, redact_rl, nvl);
2509 		if (err != 0)
2510 			goto out;
2511 	}
2512 
2513 	if (featureflags & DMU_BACKUP_FEATURE_RAW) {
2514 		uint64_t ivset_guid = (ancestor_zb != NULL) ?
2515 		    ancestor_zb->zbm_ivset_guid : 0;
2516 		nvlist_t *keynvl = NULL;
2517 		ASSERT(os->os_encrypted);
2518 
2519 		err = dsl_crypto_populate_key_nvlist(os, ivset_guid,
2520 		    &keynvl);
2521 		if (err != 0) {
2522 			fnvlist_free(nvl);
2523 			goto out;
2524 		}
2525 
2526 		fnvlist_add_nvlist(nvl, "crypt_keydata", keynvl);
2527 		fnvlist_free(keynvl);
2528 	}
2529 
2530 	if (!nvlist_empty(nvl)) {
2531 		payload = fnvlist_pack(nvl, &payload_len);
2532 		drr->drr_payloadlen = payload_len;
2533 	}
2534 
2535 	fnvlist_free(nvl);
2536 	err = dump_record(&dsc, payload, payload_len);
2537 	fnvlist_pack_free(payload, payload_len);
2538 	if (err != 0) {
2539 		err = dsc.dsc_err;
2540 		goto out;
2541 	}
2542 
2543 	setup_to_thread(to_arg, os, dssp, fromtxg, dspp->rawok);
2544 	setup_from_thread(from_arg, from_rl, dssp);
2545 	setup_redact_list_thread(rlt_arg, dspp, redact_rl, dssp);
2546 	setup_merge_thread(smt_arg, dspp, from_arg, to_arg, rlt_arg, os);
2547 	setup_reader_thread(srt_arg, dspp, smt_arg, featureflags);
2548 
2549 	range = bqueue_dequeue(&srt_arg->q);
2550 	while (err == 0 && !range->eos_marker) {
2551 		err = do_dump(&dsc, range);
2552 		range = get_next_range(&srt_arg->q, range);
2553 		if (issig(JUSTLOOKING) && issig(FORREAL))
2554 			err = SET_ERROR(EINTR);
2555 	}
2556 
2557 	/*
2558 	 * If we hit an error or are interrupted, cancel our worker threads and
2559 	 * clear the queue of any pending records.  The threads will pass the
2560 	 * cancel up the tree of worker threads, and each one will clean up any
2561 	 * pending records before exiting.
2562 	 */
2563 	if (err != 0) {
2564 		srt_arg->cancel = B_TRUE;
2565 		while (!range->eos_marker) {
2566 			range = get_next_range(&srt_arg->q, range);
2567 		}
2568 	}
2569 	range_free(range);
2570 
2571 	bqueue_destroy(&srt_arg->q);
2572 	bqueue_destroy(&smt_arg->q);
2573 	if (dspp->redactbook != NULL)
2574 		bqueue_destroy(&rlt_arg->q);
2575 	bqueue_destroy(&to_arg->q);
2576 	bqueue_destroy(&from_arg->q);
2577 
2578 	if (err == 0 && srt_arg->error != 0)
2579 		err = srt_arg->error;
2580 
2581 	if (err != 0)
2582 		goto out;
2583 
2584 	if (dsc.dsc_pending_op != PENDING_NONE)
2585 		if (dump_record(&dsc, NULL, 0) != 0)
2586 			err = SET_ERROR(EINTR);
2587 
2588 	if (err != 0) {
2589 		if (err == EINTR && dsc.dsc_err != 0)
2590 			err = dsc.dsc_err;
2591 		goto out;
2592 	}
2593 
2594 	/*
2595 	 * Send the DRR_END record if this is not a saved stream.
2596 	 * Otherwise, the omitted DRR_END record will signal to
2597 	 * the receive side that the stream is incomplete.
2598 	 */
2599 	if (!dspp->savedok) {
2600 		memset(drr, 0, sizeof (dmu_replay_record_t));
2601 		drr->drr_type = DRR_END;
2602 		drr->drr_u.drr_end.drr_checksum = dsc.dsc_zc;
2603 		drr->drr_u.drr_end.drr_toguid = dsc.dsc_toguid;
2604 
2605 		if (dump_record(&dsc, NULL, 0) != 0)
2606 			err = dsc.dsc_err;
2607 	}
2608 out:
2609 	mutex_enter(&to_ds->ds_sendstream_lock);
2610 	list_remove(&to_ds->ds_sendstreams, dssp);
2611 	mutex_exit(&to_ds->ds_sendstream_lock);
2612 
2613 	VERIFY(err != 0 || (dsc.dsc_sent_begin &&
2614 	    (dsc.dsc_sent_end || dspp->savedok)));
2615 
2616 	kmem_free(drr, sizeof (dmu_replay_record_t));
2617 	kmem_free(dssp, sizeof (dmu_sendstatus_t));
2618 	kmem_free(from_arg, sizeof (*from_arg));
2619 	kmem_free(to_arg, sizeof (*to_arg));
2620 	kmem_free(rlt_arg, sizeof (*rlt_arg));
2621 	kmem_free(smt_arg, sizeof (*smt_arg));
2622 	kmem_free(srt_arg, sizeof (*srt_arg));
2623 
2624 	dsl_dataset_long_rele(to_ds, FTAG);
2625 	if (from_rl != NULL) {
2626 		dsl_redaction_list_long_rele(from_rl, FTAG);
2627 		dsl_redaction_list_rele(from_rl, FTAG);
2628 	}
2629 	if (redact_rl != NULL) {
2630 		dsl_redaction_list_long_rele(redact_rl, FTAG);
2631 		dsl_redaction_list_rele(redact_rl, FTAG);
2632 	}
2633 
2634 	return (err);
2635 }
2636 
2637 int
2638 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
2639     boolean_t embedok, boolean_t large_block_ok, boolean_t compressok,
2640     boolean_t rawok, boolean_t savedok, int outfd, offset_t *off,
2641     dmu_send_outparams_t *dsop)
2642 {
2643 	int err;
2644 	dsl_dataset_t *fromds;
2645 	ds_hold_flags_t dsflags;
2646 	struct dmu_send_params dspp = {0};
2647 	dspp.embedok = embedok;
2648 	dspp.large_block_ok = large_block_ok;
2649 	dspp.compressok = compressok;
2650 	dspp.outfd = outfd;
2651 	dspp.off = off;
2652 	dspp.dso = dsop;
2653 	dspp.tag = FTAG;
2654 	dspp.rawok = rawok;
2655 	dspp.savedok = savedok;
2656 
2657 	dsflags = (rawok) ? DS_HOLD_FLAG_NONE : DS_HOLD_FLAG_DECRYPT;
2658 	err = dsl_pool_hold(pool, FTAG, &dspp.dp);
2659 	if (err != 0)
2660 		return (err);
2661 
2662 	err = dsl_dataset_hold_obj_flags(dspp.dp, tosnap, dsflags, FTAG,
2663 	    &dspp.to_ds);
2664 	if (err != 0) {
2665 		dsl_pool_rele(dspp.dp, FTAG);
2666 		return (err);
2667 	}
2668 
2669 	if (fromsnap != 0) {
2670 		err = dsl_dataset_hold_obj_flags(dspp.dp, fromsnap, dsflags,
2671 		    FTAG, &fromds);
2672 		if (err != 0) {
2673 			dsl_dataset_rele_flags(dspp.to_ds, dsflags, FTAG);
2674 			dsl_pool_rele(dspp.dp, FTAG);
2675 			return (err);
2676 		}
2677 		dspp.ancestor_zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid;
2678 		dspp.ancestor_zb.zbm_creation_txg =
2679 		    dsl_dataset_phys(fromds)->ds_creation_txg;
2680 		dspp.ancestor_zb.zbm_creation_time =
2681 		    dsl_dataset_phys(fromds)->ds_creation_time;
2682 
2683 		if (dsl_dataset_is_zapified(fromds)) {
2684 			(void) zap_lookup(dspp.dp->dp_meta_objset,
2685 			    fromds->ds_object, DS_FIELD_IVSET_GUID, 8, 1,
2686 			    &dspp.ancestor_zb.zbm_ivset_guid);
2687 		}
2688 
2689 		/* See dmu_send for the reasons behind this. */
2690 		uint64_t *fromredact;
2691 
2692 		if (!dsl_dataset_get_uint64_array_feature(fromds,
2693 		    SPA_FEATURE_REDACTED_DATASETS,
2694 		    &dspp.numfromredactsnaps,
2695 		    &fromredact)) {
2696 			dspp.numfromredactsnaps = NUM_SNAPS_NOT_REDACTED;
2697 		} else if (dspp.numfromredactsnaps > 0) {
2698 			uint64_t size = dspp.numfromredactsnaps *
2699 			    sizeof (uint64_t);
2700 			dspp.fromredactsnaps = kmem_zalloc(size, KM_SLEEP);
2701 			memcpy(dspp.fromredactsnaps, fromredact, size);
2702 		}
2703 
2704 		boolean_t is_before =
2705 		    dsl_dataset_is_before(dspp.to_ds, fromds, 0);
2706 		dspp.is_clone = (dspp.to_ds->ds_dir !=
2707 		    fromds->ds_dir);
2708 		dsl_dataset_rele(fromds, FTAG);
2709 		if (!is_before) {
2710 			dsl_pool_rele(dspp.dp, FTAG);
2711 			err = SET_ERROR(EXDEV);
2712 		} else {
2713 			err = dmu_send_impl(&dspp);
2714 		}
2715 	} else {
2716 		dspp.numfromredactsnaps = NUM_SNAPS_NOT_REDACTED;
2717 		err = dmu_send_impl(&dspp);
2718 	}
2719 	dsl_dataset_rele(dspp.to_ds, FTAG);
2720 	return (err);
2721 }
2722 
2723 int
2724 dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok,
2725     boolean_t large_block_ok, boolean_t compressok, boolean_t rawok,
2726     boolean_t savedok, uint64_t resumeobj, uint64_t resumeoff,
2727     const char *redactbook, int outfd, offset_t *off,
2728     dmu_send_outparams_t *dsop)
2729 {
2730 	int err = 0;
2731 	ds_hold_flags_t dsflags;
2732 	boolean_t owned = B_FALSE;
2733 	dsl_dataset_t *fromds = NULL;
2734 	zfs_bookmark_phys_t book = {0};
2735 	struct dmu_send_params dspp = {0};
2736 
2737 	dsflags = (rawok) ? DS_HOLD_FLAG_NONE : DS_HOLD_FLAG_DECRYPT;
2738 	dspp.tosnap = tosnap;
2739 	dspp.embedok = embedok;
2740 	dspp.large_block_ok = large_block_ok;
2741 	dspp.compressok = compressok;
2742 	dspp.outfd = outfd;
2743 	dspp.off = off;
2744 	dspp.dso = dsop;
2745 	dspp.tag = FTAG;
2746 	dspp.resumeobj = resumeobj;
2747 	dspp.resumeoff = resumeoff;
2748 	dspp.rawok = rawok;
2749 	dspp.savedok = savedok;
2750 
2751 	if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL)
2752 		return (SET_ERROR(EINVAL));
2753 
2754 	err = dsl_pool_hold(tosnap, FTAG, &dspp.dp);
2755 	if (err != 0)
2756 		return (err);
2757 
2758 	if (strchr(tosnap, '@') == NULL && spa_writeable(dspp.dp->dp_spa)) {
2759 		/*
2760 		 * We are sending a filesystem or volume.  Ensure
2761 		 * that it doesn't change by owning the dataset.
2762 		 */
2763 
2764 		if (savedok) {
2765 			/*
2766 			 * We are looking for the dataset that represents the
2767 			 * partially received send stream. If this stream was
2768 			 * received as a new snapshot of an existing dataset,
2769 			 * this will be saved in a hidden clone named
2770 			 * "<pool>/<dataset>/%recv". Otherwise, the stream
2771 			 * will be saved in the live dataset itself. In
2772 			 * either case we need to use dsl_dataset_own_force()
2773 			 * because the stream is marked as inconsistent,
2774 			 * which would normally make it unavailable to be
2775 			 * owned.
2776 			 */
2777 			char *name = kmem_asprintf("%s/%s", tosnap,
2778 			    recv_clone_name);
2779 			err = dsl_dataset_own_force(dspp.dp, name, dsflags,
2780 			    FTAG, &dspp.to_ds);
2781 			if (err == ENOENT) {
2782 				err = dsl_dataset_own_force(dspp.dp, tosnap,
2783 				    dsflags, FTAG, &dspp.to_ds);
2784 			}
2785 
2786 			if (err == 0) {
2787 				err = zap_lookup(dspp.dp->dp_meta_objset,
2788 				    dspp.to_ds->ds_object,
2789 				    DS_FIELD_RESUME_TOGUID, 8, 1,
2790 				    &dspp.saved_guid);
2791 			}
2792 
2793 			if (err == 0) {
2794 				err = zap_lookup(dspp.dp->dp_meta_objset,
2795 				    dspp.to_ds->ds_object,
2796 				    DS_FIELD_RESUME_TONAME, 1,
2797 				    sizeof (dspp.saved_toname),
2798 				    dspp.saved_toname);
2799 			}
2800 			if (err != 0)
2801 				dsl_dataset_disown(dspp.to_ds, dsflags, FTAG);
2802 
2803 			kmem_strfree(name);
2804 		} else {
2805 			err = dsl_dataset_own(dspp.dp, tosnap, dsflags,
2806 			    FTAG, &dspp.to_ds);
2807 		}
2808 		owned = B_TRUE;
2809 	} else {
2810 		err = dsl_dataset_hold_flags(dspp.dp, tosnap, dsflags, FTAG,
2811 		    &dspp.to_ds);
2812 	}
2813 
2814 	if (err != 0) {
2815 		dsl_pool_rele(dspp.dp, FTAG);
2816 		return (err);
2817 	}
2818 
2819 	if (redactbook != NULL) {
2820 		char path[ZFS_MAX_DATASET_NAME_LEN];
2821 		(void) strlcpy(path, tosnap, sizeof (path));
2822 		char *at = strchr(path, '@');
2823 		if (at == NULL) {
2824 			err = EINVAL;
2825 		} else {
2826 			(void) snprintf(at, sizeof (path) - (at - path), "#%s",
2827 			    redactbook);
2828 			err = dsl_bookmark_lookup(dspp.dp, path,
2829 			    NULL, &book);
2830 			dspp.redactbook = &book;
2831 		}
2832 	}
2833 
2834 	if (err != 0) {
2835 		dsl_pool_rele(dspp.dp, FTAG);
2836 		if (owned)
2837 			dsl_dataset_disown(dspp.to_ds, dsflags, FTAG);
2838 		else
2839 			dsl_dataset_rele_flags(dspp.to_ds, dsflags, FTAG);
2840 		return (err);
2841 	}
2842 
2843 	if (fromsnap != NULL) {
2844 		zfs_bookmark_phys_t *zb = &dspp.ancestor_zb;
2845 		int fsnamelen;
2846 		if (strpbrk(tosnap, "@#") != NULL)
2847 			fsnamelen = strpbrk(tosnap, "@#") - tosnap;
2848 		else
2849 			fsnamelen = strlen(tosnap);
2850 
2851 		/*
2852 		 * If the fromsnap is in a different filesystem, then
2853 		 * mark the send stream as a clone.
2854 		 */
2855 		if (strncmp(tosnap, fromsnap, fsnamelen) != 0 ||
2856 		    (fromsnap[fsnamelen] != '@' &&
2857 		    fromsnap[fsnamelen] != '#')) {
2858 			dspp.is_clone = B_TRUE;
2859 		}
2860 
2861 		if (strchr(fromsnap, '@') != NULL) {
2862 			err = dsl_dataset_hold(dspp.dp, fromsnap, FTAG,
2863 			    &fromds);
2864 
2865 			if (err != 0) {
2866 				ASSERT3P(fromds, ==, NULL);
2867 			} else {
2868 				/*
2869 				 * We need to make a deep copy of the redact
2870 				 * snapshots of the from snapshot, because the
2871 				 * array will be freed when we evict from_ds.
2872 				 */
2873 				uint64_t *fromredact;
2874 				if (!dsl_dataset_get_uint64_array_feature(
2875 				    fromds, SPA_FEATURE_REDACTED_DATASETS,
2876 				    &dspp.numfromredactsnaps,
2877 				    &fromredact)) {
2878 					dspp.numfromredactsnaps =
2879 					    NUM_SNAPS_NOT_REDACTED;
2880 				} else if (dspp.numfromredactsnaps > 0) {
2881 					uint64_t size =
2882 					    dspp.numfromredactsnaps *
2883 					    sizeof (uint64_t);
2884 					dspp.fromredactsnaps = kmem_zalloc(size,
2885 					    KM_SLEEP);
2886 					memcpy(dspp.fromredactsnaps, fromredact,
2887 					    size);
2888 				}
2889 				if (!dsl_dataset_is_before(dspp.to_ds, fromds,
2890 				    0)) {
2891 					err = SET_ERROR(EXDEV);
2892 				} else {
2893 					zb->zbm_creation_txg =
2894 					    dsl_dataset_phys(fromds)->
2895 					    ds_creation_txg;
2896 					zb->zbm_creation_time =
2897 					    dsl_dataset_phys(fromds)->
2898 					    ds_creation_time;
2899 					zb->zbm_guid =
2900 					    dsl_dataset_phys(fromds)->ds_guid;
2901 					zb->zbm_redaction_obj = 0;
2902 
2903 					if (dsl_dataset_is_zapified(fromds)) {
2904 						(void) zap_lookup(
2905 						    dspp.dp->dp_meta_objset,
2906 						    fromds->ds_object,
2907 						    DS_FIELD_IVSET_GUID, 8, 1,
2908 						    &zb->zbm_ivset_guid);
2909 					}
2910 				}
2911 				dsl_dataset_rele(fromds, FTAG);
2912 			}
2913 		} else {
2914 			dspp.numfromredactsnaps = NUM_SNAPS_NOT_REDACTED;
2915 			err = dsl_bookmark_lookup(dspp.dp, fromsnap, dspp.to_ds,
2916 			    zb);
2917 			if (err == EXDEV && zb->zbm_redaction_obj != 0 &&
2918 			    zb->zbm_guid ==
2919 			    dsl_dataset_phys(dspp.to_ds)->ds_guid)
2920 				err = 0;
2921 		}
2922 
2923 		if (err == 0) {
2924 			/* dmu_send_impl will call dsl_pool_rele for us. */
2925 			err = dmu_send_impl(&dspp);
2926 		} else {
2927 			dsl_pool_rele(dspp.dp, FTAG);
2928 		}
2929 	} else {
2930 		dspp.numfromredactsnaps = NUM_SNAPS_NOT_REDACTED;
2931 		err = dmu_send_impl(&dspp);
2932 	}
2933 	if (owned)
2934 		dsl_dataset_disown(dspp.to_ds, dsflags, FTAG);
2935 	else
2936 		dsl_dataset_rele_flags(dspp.to_ds, dsflags, FTAG);
2937 	return (err);
2938 }
2939 
2940 static int
2941 dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t uncompressed,
2942     uint64_t compressed, boolean_t stream_compressed, uint64_t *sizep)
2943 {
2944 	int err = 0;
2945 	uint64_t size;
2946 	/*
2947 	 * Assume that space (both on-disk and in-stream) is dominated by
2948 	 * data.  We will adjust for indirect blocks and the copies property,
2949 	 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
2950 	 */
2951 
2952 	uint64_t recordsize;
2953 	uint64_t record_count;
2954 	objset_t *os;
2955 	VERIFY0(dmu_objset_from_ds(ds, &os));
2956 
2957 	/* Assume all (uncompressed) blocks are recordsize. */
2958 	if (zfs_override_estimate_recordsize != 0) {
2959 		recordsize = zfs_override_estimate_recordsize;
2960 	} else if (os->os_phys->os_type == DMU_OST_ZVOL) {
2961 		err = dsl_prop_get_int_ds(ds,
2962 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &recordsize);
2963 	} else {
2964 		err = dsl_prop_get_int_ds(ds,
2965 		    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), &recordsize);
2966 	}
2967 	if (err != 0)
2968 		return (err);
2969 	record_count = uncompressed / recordsize;
2970 
2971 	/*
2972 	 * If we're estimating a send size for a compressed stream, use the
2973 	 * compressed data size to estimate the stream size. Otherwise, use the
2974 	 * uncompressed data size.
2975 	 */
2976 	size = stream_compressed ? compressed : uncompressed;
2977 
2978 	/*
2979 	 * Subtract out approximate space used by indirect blocks.
2980 	 * Assume most space is used by data blocks (non-indirect, non-dnode).
2981 	 * Assume no ditto blocks or internal fragmentation.
2982 	 *
2983 	 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
2984 	 * block.
2985 	 */
2986 	size -= record_count * sizeof (blkptr_t);
2987 
2988 	/* Add in the space for the record associated with each block. */
2989 	size += record_count * sizeof (dmu_replay_record_t);
2990 
2991 	*sizep = size;
2992 
2993 	return (0);
2994 }
2995 
2996 int
2997 dmu_send_estimate_fast(dsl_dataset_t *origds, dsl_dataset_t *fromds,
2998     zfs_bookmark_phys_t *frombook, boolean_t stream_compressed,
2999     boolean_t saved, uint64_t *sizep)
3000 {
3001 	int err;
3002 	dsl_dataset_t *ds = origds;
3003 	uint64_t uncomp, comp;
3004 
3005 	ASSERT(dsl_pool_config_held(origds->ds_dir->dd_pool));
3006 	ASSERT(fromds == NULL || frombook == NULL);
3007 
3008 	/*
3009 	 * If this is a saved send we may actually be sending
3010 	 * from the %recv clone used for resuming.
3011 	 */
3012 	if (saved) {
3013 		objset_t *mos = origds->ds_dir->dd_pool->dp_meta_objset;
3014 		uint64_t guid;
3015 		char dsname[ZFS_MAX_DATASET_NAME_LEN + 6];
3016 
3017 		dsl_dataset_name(origds, dsname);
3018 		(void) strcat(dsname, "/");
3019 		(void) strcat(dsname, recv_clone_name);
3020 
3021 		err = dsl_dataset_hold(origds->ds_dir->dd_pool,
3022 		    dsname, FTAG, &ds);
3023 		if (err != ENOENT && err != 0) {
3024 			return (err);
3025 		} else if (err == ENOENT) {
3026 			ds = origds;
3027 		}
3028 
3029 		/* check that this dataset has partially received data */
3030 		err = zap_lookup(mos, ds->ds_object,
3031 		    DS_FIELD_RESUME_TOGUID, 8, 1, &guid);
3032 		if (err != 0) {
3033 			err = SET_ERROR(err == ENOENT ? EINVAL : err);
3034 			goto out;
3035 		}
3036 
3037 		err = zap_lookup(mos, ds->ds_object,
3038 		    DS_FIELD_RESUME_TONAME, 1, sizeof (dsname), dsname);
3039 		if (err != 0) {
3040 			err = SET_ERROR(err == ENOENT ? EINVAL : err);
3041 			goto out;
3042 		}
3043 	}
3044 
3045 	/* tosnap must be a snapshot or the target of a saved send */
3046 	if (!ds->ds_is_snapshot && ds == origds)
3047 		return (SET_ERROR(EINVAL));
3048 
3049 	if (fromds != NULL) {
3050 		uint64_t used;
3051 		if (!fromds->ds_is_snapshot) {
3052 			err = SET_ERROR(EINVAL);
3053 			goto out;
3054 		}
3055 
3056 		if (!dsl_dataset_is_before(ds, fromds, 0)) {
3057 			err = SET_ERROR(EXDEV);
3058 			goto out;
3059 		}
3060 
3061 		err = dsl_dataset_space_written(fromds, ds, &used, &comp,
3062 		    &uncomp);
3063 		if (err != 0)
3064 			goto out;
3065 	} else if (frombook != NULL) {
3066 		uint64_t used;
3067 		err = dsl_dataset_space_written_bookmark(frombook, ds, &used,
3068 		    &comp, &uncomp);
3069 		if (err != 0)
3070 			goto out;
3071 	} else {
3072 		uncomp = dsl_dataset_phys(ds)->ds_uncompressed_bytes;
3073 		comp = dsl_dataset_phys(ds)->ds_compressed_bytes;
3074 	}
3075 
3076 	err = dmu_adjust_send_estimate_for_indirects(ds, uncomp, comp,
3077 	    stream_compressed, sizep);
3078 	/*
3079 	 * Add the size of the BEGIN and END records to the estimate.
3080 	 */
3081 	*sizep += 2 * sizeof (dmu_replay_record_t);
3082 
3083 out:
3084 	if (ds != origds)
3085 		dsl_dataset_rele(ds, FTAG);
3086 	return (err);
3087 }
3088 
3089 ZFS_MODULE_PARAM(zfs_send, zfs_send_, corrupt_data, INT, ZMOD_RW,
3090 	"Allow sending corrupt data");
3091 
3092 ZFS_MODULE_PARAM(zfs_send, zfs_send_, queue_length, INT, ZMOD_RW,
3093 	"Maximum send queue length");
3094 
3095 ZFS_MODULE_PARAM(zfs_send, zfs_send_, unmodified_spill_blocks, INT, ZMOD_RW,
3096 	"Send unmodified spill blocks");
3097 
3098 ZFS_MODULE_PARAM(zfs_send, zfs_send_, no_prefetch_queue_length, INT, ZMOD_RW,
3099 	"Maximum send queue length for non-prefetch queues");
3100 
3101 ZFS_MODULE_PARAM(zfs_send, zfs_send_, queue_ff, INT, ZMOD_RW,
3102 	"Send queue fill fraction");
3103 
3104 ZFS_MODULE_PARAM(zfs_send, zfs_send_, no_prefetch_queue_ff, INT, ZMOD_RW,
3105 	"Send queue fill fraction for non-prefetch queues");
3106 
3107 ZFS_MODULE_PARAM(zfs_send, zfs_, override_estimate_recordsize, INT, ZMOD_RW,
3108 	"Override block size estimate with fixed size");
3109