]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
8d35c149 | 23 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. |
e6d3a843 | 24 | * Copyright (c) 2011, 2015 by Delphix. All rights reserved. |
788eb90c | 25 | * Copyright (c) 2014, Joyent, Inc. All rights reserved. |
47dfff3b | 26 | * Copyright 2014 HybridCluster. All rights reserved. |
b607405f | 27 | * Copyright 2016 RackTop Systems. |
a0bd735a | 28 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. |
8d35c149 | 29 | */ |
34dc7c2f | 30 | |
34dc7c2f BB |
31 | #include <sys/dmu.h> |
32 | #include <sys/dmu_impl.h> | |
33 | #include <sys/dmu_tx.h> | |
34 | #include <sys/dbuf.h> | |
35 | #include <sys/dnode.h> | |
36 | #include <sys/zfs_context.h> | |
37 | #include <sys/dmu_objset.h> | |
38 | #include <sys/dmu_traverse.h> | |
39 | #include <sys/dsl_dataset.h> | |
40 | #include <sys/dsl_dir.h> | |
428870ff | 41 | #include <sys/dsl_prop.h> |
34dc7c2f BB |
42 | #include <sys/dsl_pool.h> |
43 | #include <sys/dsl_synctask.h> | |
044baf00 | 44 | #include <sys/spa_impl.h> |
34dc7c2f BB |
45 | #include <sys/zfs_ioctl.h> |
46 | #include <sys/zap.h> | |
47 | #include <sys/zio_checksum.h> | |
428870ff BB |
48 | #include <sys/zfs_znode.h> |
49 | #include <zfs_fletcher.h> | |
50 | #include <sys/avl.h> | |
51 | #include <sys/ddt.h> | |
572e2857 | 52 | #include <sys/zfs_onexit.h> |
13fe0198 MA |
53 | #include <sys/dmu_send.h> |
54 | #include <sys/dsl_destroy.h> | |
9b67f605 | 55 | #include <sys/blkptr.h> |
da536844 | 56 | #include <sys/dsl_bookmark.h> |
9b67f605 | 57 | #include <sys/zfeature.h> |
fcff0f35 | 58 | #include <sys/bqueue.h> |
a0bd735a | 59 | #include <sys/zvol.h> |
f74b821a | 60 | #include <sys/policy.h> |
34dc7c2f | 61 | |
330d06f9 MA |
62 | /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */ |
63 | int zfs_send_corrupt_data = B_FALSE; | |
3b0d9928 BB |
64 | int zfs_send_queue_length = SPA_MAXBLOCKSIZE; |
65 | int zfs_recv_queue_length = SPA_MAXBLOCKSIZE; | |
b607405f AS |
66 | /* Set this tunable to FALSE to disable setting of DRR_FLAG_FREERECORDS */ |
67 | int zfs_send_set_freerecords_bit = B_TRUE; | |
330d06f9 | 68 | |
34dc7c2f | 69 | static char *dmu_recv_tag = "dmu_recv_tag"; |
47dfff3b | 70 | const char *recv_clone_name = "%recv"; |
34dc7c2f | 71 | |
ca0845d5 PD |
72 | /* |
73 | * Use this to override the recordsize calculation for fast zfs send estimates. | |
74 | */ | |
75 | unsigned long zfs_override_estimate_recordsize = 0; | |
76 | ||
fcff0f35 PD |
77 | #define BP_SPAN(datablkszsec, indblkshift, level) \ |
78 | (((uint64_t)datablkszsec) << (SPA_MINBLOCKSHIFT + \ | |
79 | (level) * (indblkshift - SPA_BLKPTRSHIFT))) | |
80 | ||
47dfff3b MA |
81 | static void byteswap_record(dmu_replay_record_t *drr); |
82 | ||
fcff0f35 PD |
83 | struct send_thread_arg { |
84 | bqueue_t q; | |
85 | dsl_dataset_t *ds; /* Dataset to traverse */ | |
86 | uint64_t fromtxg; /* Traverse from this txg */ | |
87 | int flags; /* flags to pass to traverse_dataset */ | |
88 | int error_code; | |
89 | boolean_t cancel; | |
47dfff3b | 90 | zbookmark_phys_t resume; |
fcff0f35 PD |
91 | }; |
92 | ||
93 | struct send_block_record { | |
94 | boolean_t eos_marker; /* Marks the end of the stream */ | |
95 | blkptr_t bp; | |
96 | zbookmark_phys_t zb; | |
97 | uint8_t indblkshift; | |
98 | uint16_t datablkszsec; | |
99 | bqueue_node_t ln; | |
100 | }; | |
101 | ||
044baf00 BB |
102 | typedef struct dump_bytes_io { |
103 | dmu_sendarg_t *dbi_dsp; | |
104 | void *dbi_buf; | |
105 | int dbi_len; | |
106 | } dump_bytes_io_t; | |
107 | ||
108 | static void | |
b58986ee | 109 | dump_bytes_cb(void *arg) |
34dc7c2f | 110 | { |
044baf00 BB |
111 | dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg; |
112 | dmu_sendarg_t *dsp = dbi->dbi_dsp; | |
47dfff3b | 113 | dsl_dataset_t *ds = dmu_objset_ds(dsp->dsa_os); |
34dc7c2f | 114 | ssize_t resid; /* have to get resid to get detailed errno */ |
f8866f8a ER |
115 | |
116 | /* | |
b5256303 | 117 | * The code does not rely on len being a multiple of 8. We keep |
f8866f8a ER |
118 | * this assertion because of the corresponding assertion in |
119 | * receive_read(). Keeping this assertion ensures that we do not | |
120 | * inadvertently break backwards compatibility (causing the assertion | |
b5256303 TC |
121 | * in receive_read() to trigger on old software). Newer feature flags |
122 | * (such as raw send) may break this assertion since they were | |
123 | * introduced after the requirement was made obsolete. | |
f8866f8a ER |
124 | */ |
125 | ||
b5256303 TC |
126 | ASSERT(dbi->dbi_len % 8 == 0 || |
127 | (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) != 0); | |
34dc7c2f | 128 | |
37abac6d | 129 | dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp, |
044baf00 | 130 | (caddr_t)dbi->dbi_buf, dbi->dbi_len, |
34dc7c2f | 131 | 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid); |
37abac6d BP |
132 | |
133 | mutex_enter(&ds->ds_sendstream_lock); | |
044baf00 | 134 | *dsp->dsa_off += dbi->dbi_len; |
37abac6d | 135 | mutex_exit(&ds->ds_sendstream_lock); |
044baf00 BB |
136 | } |
137 | ||
138 | static int | |
139 | dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) | |
140 | { | |
141 | dump_bytes_io_t dbi; | |
142 | ||
143 | dbi.dbi_dsp = dsp; | |
144 | dbi.dbi_buf = buf; | |
145 | dbi.dbi_len = len; | |
146 | ||
b58986ee BB |
147 | #if defined(HAVE_LARGE_STACKS) |
148 | dump_bytes_cb(&dbi); | |
149 | #else | |
044baf00 BB |
150 | /* |
151 | * The vn_rdwr() call is performed in a taskq to ensure that there is | |
152 | * always enough stack space to write safely to the target filesystem. | |
153 | * The ZIO_TYPE_FREE threads are used because there can be a lot of | |
154 | * them and they are used in vdev_file.c for a similar purpose. | |
155 | */ | |
156 | spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE, | |
b58986ee BB |
157 | ZIO_TASKQ_ISSUE, dump_bytes_cb, &dbi, TQ_SLEEP); |
158 | #endif /* HAVE_LARGE_STACKS */ | |
37abac6d BP |
159 | |
160 | return (dsp->dsa_err); | |
34dc7c2f BB |
161 | } |
162 | ||
37f8a883 MA |
163 | /* |
164 | * For all record types except BEGIN, fill in the checksum (overlaid in | |
165 | * drr_u.drr_checksum.drr_checksum). The checksum verifies everything | |
166 | * up to the start of the checksum itself. | |
167 | */ | |
168 | static int | |
169 | dump_record(dmu_sendarg_t *dsp, void *payload, int payload_len) | |
170 | { | |
171 | ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), | |
172 | ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); | |
a6255b7f | 173 | (void) fletcher_4_incremental_native(dsp->dsa_drr, |
37f8a883 MA |
174 | offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), |
175 | &dsp->dsa_zc); | |
51907a31 K |
176 | if (dsp->dsa_drr->drr_type == DRR_BEGIN) { |
177 | dsp->dsa_sent_begin = B_TRUE; | |
178 | } else { | |
37f8a883 MA |
179 | ASSERT(ZIO_CHECKSUM_IS_ZERO(&dsp->dsa_drr->drr_u. |
180 | drr_checksum.drr_checksum)); | |
181 | dsp->dsa_drr->drr_u.drr_checksum.drr_checksum = dsp->dsa_zc; | |
182 | } | |
51907a31 K |
183 | if (dsp->dsa_drr->drr_type == DRR_END) { |
184 | dsp->dsa_sent_end = B_TRUE; | |
185 | } | |
a6255b7f | 186 | (void) fletcher_4_incremental_native(&dsp->dsa_drr-> |
37f8a883 MA |
187 | drr_u.drr_checksum.drr_checksum, |
188 | sizeof (zio_cksum_t), &dsp->dsa_zc); | |
189 | if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) | |
190 | return (SET_ERROR(EINTR)); | |
191 | if (payload_len != 0) { | |
a6255b7f | 192 | (void) fletcher_4_incremental_native(payload, payload_len, |
37f8a883 MA |
193 | &dsp->dsa_zc); |
194 | if (dump_bytes(dsp, payload, payload_len) != 0) | |
195 | return (SET_ERROR(EINTR)); | |
196 | } | |
197 | return (0); | |
198 | } | |
199 | ||
e6d3a843 PD |
200 | /* |
201 | * Fill in the drr_free struct, or perform aggregation if the previous record is | |
202 | * also a free record, and the two are adjacent. | |
203 | * | |
204 | * Note that we send free records even for a full send, because we want to be | |
205 | * able to receive a full send as a clone, which requires a list of all the free | |
206 | * and freeobject records that were generated on the source. | |
207 | */ | |
34dc7c2f | 208 | static int |
37abac6d | 209 | dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, |
34dc7c2f BB |
210 | uint64_t length) |
211 | { | |
37abac6d | 212 | struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free); |
428870ff | 213 | |
ea97f8ce MA |
214 | /* |
215 | * When we receive a free record, dbuf_free_range() assumes | |
216 | * that the receiving system doesn't have any dbufs in the range | |
217 | * being freed. This is always true because there is a one-record | |
218 | * constraint: we only send one WRITE record for any given | |
47dfff3b | 219 | * object,offset. We know that the one-record constraint is |
ea97f8ce MA |
220 | * true because we always send data in increasing order by |
221 | * object,offset. | |
222 | * | |
223 | * If the increasing-order constraint ever changes, we should find | |
224 | * another way to assert that the one-record constraint is still | |
225 | * satisfied. | |
226 | */ | |
227 | ASSERT(object > dsp->dsa_last_data_object || | |
228 | (object == dsp->dsa_last_data_object && | |
229 | offset > dsp->dsa_last_data_offset)); | |
230 | ||
428870ff BB |
231 | /* |
232 | * If there is a pending op, but it's not PENDING_FREE, push it out, | |
233 | * since free block aggregation can only be done for blocks of the | |
234 | * same type (i.e., DRR_FREE records can only be aggregated with | |
235 | * other DRR_FREE records. DRR_FREEOBJECTS records can only be | |
236 | * aggregated with other DRR_FREEOBJECTS records. | |
237 | */ | |
37abac6d BP |
238 | if (dsp->dsa_pending_op != PENDING_NONE && |
239 | dsp->dsa_pending_op != PENDING_FREE) { | |
37f8a883 | 240 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 241 | return (SET_ERROR(EINTR)); |
37abac6d | 242 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff BB |
243 | } |
244 | ||
37abac6d | 245 | if (dsp->dsa_pending_op == PENDING_FREE) { |
428870ff | 246 | /* |
ee45fbd8 | 247 | * There should never be a PENDING_FREE if length is |
248 | * DMU_OBJECT_END (because dump_dnode is the only place where | |
249 | * this function is called with a DMU_OBJECT_END, and only after | |
250 | * flushing any pending record). | |
428870ff | 251 | */ |
ee45fbd8 | 252 | ASSERT(length != DMU_OBJECT_END); |
428870ff BB |
253 | /* |
254 | * Check to see whether this free block can be aggregated | |
255 | * with pending one. | |
256 | */ | |
257 | if (drrf->drr_object == object && drrf->drr_offset + | |
258 | drrf->drr_length == offset) { | |
ee45fbd8 | 259 | if (offset + length < offset) |
260 | drrf->drr_length = DMU_OBJECT_END; | |
261 | else | |
262 | drrf->drr_length += length; | |
428870ff BB |
263 | return (0); |
264 | } else { | |
265 | /* not a continuation. Push out pending record */ | |
37f8a883 | 266 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 267 | return (SET_ERROR(EINTR)); |
37abac6d | 268 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff BB |
269 | } |
270 | } | |
271 | /* create a FREE record and make it pending */ | |
37abac6d BP |
272 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); |
273 | dsp->dsa_drr->drr_type = DRR_FREE; | |
428870ff BB |
274 | drrf->drr_object = object; |
275 | drrf->drr_offset = offset; | |
ee45fbd8 | 276 | if (offset + length < offset) |
277 | drrf->drr_length = DMU_OBJECT_END; | |
278 | else | |
279 | drrf->drr_length = length; | |
37abac6d | 280 | drrf->drr_toguid = dsp->dsa_toguid; |
ee45fbd8 | 281 | if (length == DMU_OBJECT_END) { |
37f8a883 | 282 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 283 | return (SET_ERROR(EINTR)); |
428870ff | 284 | } else { |
37abac6d | 285 | dsp->dsa_pending_op = PENDING_FREE; |
428870ff | 286 | } |
34dc7c2f | 287 | |
34dc7c2f BB |
288 | return (0); |
289 | } | |
290 | ||
291 | static int | |
b5256303 TC |
292 | dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type, uint64_t object, |
293 | uint64_t offset, int lsize, int psize, const blkptr_t *bp, void *data) | |
34dc7c2f | 294 | { |
2aa34383 | 295 | uint64_t payload_size; |
b5256303 | 296 | boolean_t raw = (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_RAW); |
37abac6d | 297 | struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write); |
428870ff | 298 | |
ea97f8ce MA |
299 | /* |
300 | * We send data in increasing object, offset order. | |
301 | * See comment in dump_free() for details. | |
302 | */ | |
303 | ASSERT(object > dsp->dsa_last_data_object || | |
304 | (object == dsp->dsa_last_data_object && | |
305 | offset > dsp->dsa_last_data_offset)); | |
306 | dsp->dsa_last_data_object = object; | |
2aa34383 | 307 | dsp->dsa_last_data_offset = offset + lsize - 1; |
428870ff BB |
308 | |
309 | /* | |
310 | * If there is any kind of pending aggregation (currently either | |
311 | * a grouping of free objects or free blocks), push it out to | |
312 | * the stream, since aggregation can't be done across operations | |
313 | * of different types. | |
314 | */ | |
37abac6d | 315 | if (dsp->dsa_pending_op != PENDING_NONE) { |
37f8a883 | 316 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 317 | return (SET_ERROR(EINTR)); |
37abac6d | 318 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff | 319 | } |
37f8a883 | 320 | /* write a WRITE record */ |
37abac6d BP |
321 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); |
322 | dsp->dsa_drr->drr_type = DRR_WRITE; | |
428870ff BB |
323 | drrw->drr_object = object; |
324 | drrw->drr_type = type; | |
325 | drrw->drr_offset = offset; | |
37abac6d | 326 | drrw->drr_toguid = dsp->dsa_toguid; |
2aa34383 DK |
327 | drrw->drr_logical_size = lsize; |
328 | ||
b5256303 TC |
329 | /* only set the compression fields if the buf is compressed or raw */ |
330 | if (raw || lsize != psize) { | |
2aa34383 | 331 | ASSERT(!BP_IS_EMBEDDED(bp)); |
2aa34383 | 332 | ASSERT3S(psize, >, 0); |
2aa34383 | 333 | |
b5256303 TC |
334 | if (raw) { |
335 | ASSERT(BP_IS_PROTECTED(bp)); | |
336 | ||
337 | /* | |
9b840763 TC |
338 | * This is a raw protected block so we need to pass |
339 | * along everything the receiving side will need to | |
340 | * interpret this block, including the byteswap, salt, | |
341 | * IV, and MAC. | |
b5256303 | 342 | */ |
b5256303 TC |
343 | if (BP_SHOULD_BYTESWAP(bp)) |
344 | drrw->drr_flags |= DRR_RAW_BYTESWAP; | |
345 | zio_crypt_decode_params_bp(bp, drrw->drr_salt, | |
346 | drrw->drr_iv); | |
347 | zio_crypt_decode_mac_bp(bp, drrw->drr_mac); | |
348 | } else { | |
349 | /* this is a compressed block */ | |
350 | ASSERT(dsp->dsa_featureflags & | |
351 | DMU_BACKUP_FEATURE_COMPRESSED); | |
352 | ASSERT(!BP_SHOULD_BYTESWAP(bp)); | |
353 | ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp))); | |
354 | ASSERT3U(BP_GET_COMPRESS(bp), !=, ZIO_COMPRESS_OFF); | |
355 | ASSERT3S(lsize, >=, psize); | |
356 | } | |
357 | ||
358 | /* set fields common to compressed and raw sends */ | |
2aa34383 DK |
359 | drrw->drr_compressiontype = BP_GET_COMPRESS(bp); |
360 | drrw->drr_compressed_size = psize; | |
361 | payload_size = drrw->drr_compressed_size; | |
362 | } else { | |
363 | payload_size = drrw->drr_logical_size; | |
364 | } | |
365 | ||
b5256303 | 366 | if (bp == NULL || BP_IS_EMBEDDED(bp) || (BP_IS_PROTECTED(bp) && !raw)) { |
9b67f605 | 367 | /* |
b5256303 TC |
368 | * There's no pre-computed checksum for partial-block writes, |
369 | * embedded BP's, or encrypted BP's that are being sent as | |
370 | * plaintext, so (like fletcher4-checkummed blocks) userland | |
371 | * will have to compute a dedup-capable checksum itself. | |
9b67f605 MA |
372 | */ |
373 | drrw->drr_checksumtype = ZIO_CHECKSUM_OFF; | |
374 | } else { | |
375 | drrw->drr_checksumtype = BP_GET_CHECKSUM(bp); | |
3c67d83a TH |
376 | if (zio_checksum_table[drrw->drr_checksumtype].ci_flags & |
377 | ZCHECKSUM_FLAG_DEDUP) | |
b5256303 | 378 | drrw->drr_flags |= DRR_CHECKSUM_DEDUP; |
9b67f605 MA |
379 | DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp)); |
380 | DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp)); | |
381 | DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp)); | |
b5256303 | 382 | DDK_SET_CRYPT(&drrw->drr_key, BP_IS_PROTECTED(bp)); |
9b67f605 MA |
383 | drrw->drr_key.ddk_cksum = bp->blk_cksum; |
384 | } | |
428870ff | 385 | |
2aa34383 | 386 | if (dump_record(dsp, data, payload_size) != 0) |
2e528b49 | 387 | return (SET_ERROR(EINTR)); |
428870ff BB |
388 | return (0); |
389 | } | |
390 | ||
9b67f605 MA |
391 | static int |
392 | dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, | |
393 | int blksz, const blkptr_t *bp) | |
394 | { | |
395 | char buf[BPE_PAYLOAD_SIZE]; | |
396 | struct drr_write_embedded *drrw = | |
397 | &(dsp->dsa_drr->drr_u.drr_write_embedded); | |
398 | ||
399 | if (dsp->dsa_pending_op != PENDING_NONE) { | |
37f8a883 | 400 | if (dump_record(dsp, NULL, 0) != 0) |
ecb2b7dc | 401 | return (SET_ERROR(EINTR)); |
9b67f605 MA |
402 | dsp->dsa_pending_op = PENDING_NONE; |
403 | } | |
404 | ||
405 | ASSERT(BP_IS_EMBEDDED(bp)); | |
406 | ||
407 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); | |
408 | dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED; | |
409 | drrw->drr_object = object; | |
410 | drrw->drr_offset = offset; | |
411 | drrw->drr_length = blksz; | |
412 | drrw->drr_toguid = dsp->dsa_toguid; | |
413 | drrw->drr_compression = BP_GET_COMPRESS(bp); | |
414 | drrw->drr_etype = BPE_GET_ETYPE(bp); | |
415 | drrw->drr_lsize = BPE_GET_LSIZE(bp); | |
416 | drrw->drr_psize = BPE_GET_PSIZE(bp); | |
417 | ||
418 | decode_embedded_bp_compressed(bp, buf); | |
419 | ||
37f8a883 | 420 | if (dump_record(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0) |
ecb2b7dc | 421 | return (SET_ERROR(EINTR)); |
9b67f605 MA |
422 | return (0); |
423 | } | |
424 | ||
428870ff | 425 | static int |
b5256303 | 426 | dump_spill(dmu_sendarg_t *dsp, const blkptr_t *bp, uint64_t object, void *data) |
428870ff | 427 | { |
37abac6d | 428 | struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill); |
b5256303 | 429 | uint64_t blksz = BP_GET_LSIZE(bp); |
b0ee5946 | 430 | uint64_t payload_size = blksz; |
428870ff | 431 | |
37abac6d | 432 | if (dsp->dsa_pending_op != PENDING_NONE) { |
37f8a883 | 433 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 434 | return (SET_ERROR(EINTR)); |
37abac6d | 435 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff BB |
436 | } |
437 | ||
438 | /* write a SPILL record */ | |
37abac6d BP |
439 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); |
440 | dsp->dsa_drr->drr_type = DRR_SPILL; | |
428870ff BB |
441 | drrs->drr_object = object; |
442 | drrs->drr_length = blksz; | |
37abac6d | 443 | drrs->drr_toguid = dsp->dsa_toguid; |
34dc7c2f | 444 | |
b5256303 | 445 | /* handle raw send fields */ |
9b840763 TC |
446 | if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) { |
447 | ASSERT(BP_IS_PROTECTED(bp)); | |
448 | ||
b5256303 TC |
449 | if (BP_SHOULD_BYTESWAP(bp)) |
450 | drrs->drr_flags |= DRR_RAW_BYTESWAP; | |
451 | drrs->drr_compressiontype = BP_GET_COMPRESS(bp); | |
452 | drrs->drr_compressed_size = BP_GET_PSIZE(bp); | |
453 | zio_crypt_decode_params_bp(bp, drrs->drr_salt, drrs->drr_iv); | |
454 | zio_crypt_decode_mac_bp(bp, drrs->drr_mac); | |
b0ee5946 | 455 | payload_size = drrs->drr_compressed_size; |
b5256303 TC |
456 | } |
457 | ||
b0ee5946 | 458 | if (dump_record(dsp, data, payload_size) != 0) |
2e528b49 | 459 | return (SET_ERROR(EINTR)); |
34dc7c2f BB |
460 | return (0); |
461 | } | |
462 | ||
463 | static int | |
37abac6d | 464 | dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs) |
34dc7c2f | 465 | { |
37abac6d | 466 | struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects); |
829e95c4 FG |
467 | uint64_t maxobj = DNODES_PER_BLOCK * |
468 | (DMU_META_DNODE(dsp->dsa_os)->dn_maxblkid + 1); | |
469 | ||
470 | /* | |
471 | * ZoL < 0.7 does not handle large FREEOBJECTS records correctly, | |
472 | * leading to zfs recv never completing. to avoid this issue, don't | |
473 | * send FREEOBJECTS records for object IDs which cannot exist on the | |
474 | * receiving side. | |
475 | */ | |
476 | if (maxobj > 0) { | |
477 | if (maxobj < firstobj) | |
478 | return (0); | |
479 | ||
480 | if (maxobj < firstobj + numobjs) | |
481 | numobjs = maxobj - firstobj; | |
482 | } | |
428870ff BB |
483 | |
484 | /* | |
485 | * If there is a pending op, but it's not PENDING_FREEOBJECTS, | |
486 | * push it out, since free block aggregation can only be done for | |
487 | * blocks of the same type (i.e., DRR_FREE records can only be | |
488 | * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records | |
489 | * can only be aggregated with other DRR_FREEOBJECTS records. | |
490 | */ | |
37abac6d BP |
491 | if (dsp->dsa_pending_op != PENDING_NONE && |
492 | dsp->dsa_pending_op != PENDING_FREEOBJECTS) { | |
37f8a883 | 493 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 494 | return (SET_ERROR(EINTR)); |
37abac6d | 495 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff | 496 | } |
37abac6d | 497 | if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) { |
428870ff BB |
498 | /* |
499 | * See whether this free object array can be aggregated | |
500 | * with pending one | |
501 | */ | |
502 | if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) { | |
503 | drrfo->drr_numobjs += numobjs; | |
504 | return (0); | |
505 | } else { | |
506 | /* can't be aggregated. Push out pending record */ | |
37f8a883 | 507 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 508 | return (SET_ERROR(EINTR)); |
37abac6d | 509 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff BB |
510 | } |
511 | } | |
512 | ||
34dc7c2f | 513 | /* write a FREEOBJECTS record */ |
37abac6d BP |
514 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); |
515 | dsp->dsa_drr->drr_type = DRR_FREEOBJECTS; | |
428870ff BB |
516 | drrfo->drr_firstobj = firstobj; |
517 | drrfo->drr_numobjs = numobjs; | |
37abac6d | 518 | drrfo->drr_toguid = dsp->dsa_toguid; |
428870ff | 519 | |
37abac6d | 520 | dsp->dsa_pending_op = PENDING_FREEOBJECTS; |
34dc7c2f | 521 | |
34dc7c2f BB |
522 | return (0); |
523 | } | |
524 | ||
525 | static int | |
b5256303 TC |
526 | dump_dnode(dmu_sendarg_t *dsp, const blkptr_t *bp, uint64_t object, |
527 | dnode_phys_t *dnp) | |
34dc7c2f | 528 | { |
37abac6d | 529 | struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object); |
4807c0ba | 530 | int bonuslen; |
428870ff | 531 | |
47dfff3b MA |
532 | if (object < dsp->dsa_resume_object) { |
533 | /* | |
534 | * Note: when resuming, we will visit all the dnodes in | |
535 | * the block of dnodes that we are resuming from. In | |
536 | * this case it's unnecessary to send the dnodes prior to | |
537 | * the one we are resuming from. We should be at most one | |
538 | * block's worth of dnodes behind the resume point. | |
539 | */ | |
540 | ASSERT3U(dsp->dsa_resume_object - object, <, | |
541 | 1 << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)); | |
542 | return (0); | |
543 | } | |
544 | ||
34dc7c2f | 545 | if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) |
37abac6d | 546 | return (dump_freeobjects(dsp, object, 1)); |
34dc7c2f | 547 | |
37abac6d | 548 | if (dsp->dsa_pending_op != PENDING_NONE) { |
37f8a883 | 549 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 550 | return (SET_ERROR(EINTR)); |
37abac6d | 551 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff BB |
552 | } |
553 | ||
34dc7c2f | 554 | /* write an OBJECT record */ |
37abac6d BP |
555 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); |
556 | dsp->dsa_drr->drr_type = DRR_OBJECT; | |
428870ff BB |
557 | drro->drr_object = object; |
558 | drro->drr_type = dnp->dn_type; | |
559 | drro->drr_bonustype = dnp->dn_bonustype; | |
560 | drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; | |
561 | drro->drr_bonuslen = dnp->dn_bonuslen; | |
50c957f7 | 562 | drro->drr_dn_slots = dnp->dn_extra_slots + 1; |
428870ff BB |
563 | drro->drr_checksumtype = dnp->dn_checksum; |
564 | drro->drr_compress = dnp->dn_compress; | |
37abac6d | 565 | drro->drr_toguid = dsp->dsa_toguid; |
428870ff | 566 | |
f1512ee6 MA |
567 | if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && |
568 | drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE) | |
569 | drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE; | |
570 | ||
4807c0ba TC |
571 | bonuslen = P2ROUNDUP(dnp->dn_bonuslen, 8); |
572 | ||
9b840763 TC |
573 | if ((dsp->dsa_featureflags & DMU_BACKUP_FEATURE_RAW)) { |
574 | ASSERT(BP_IS_ENCRYPTED(bp)); | |
575 | ||
b5256303 TC |
576 | if (BP_SHOULD_BYTESWAP(bp)) |
577 | drro->drr_flags |= DRR_RAW_BYTESWAP; | |
578 | ||
579 | /* needed for reconstructing dnp on recv side */ | |
ae76f45c | 580 | drro->drr_maxblkid = dnp->dn_maxblkid; |
b5256303 TC |
581 | drro->drr_indblkshift = dnp->dn_indblkshift; |
582 | drro->drr_nlevels = dnp->dn_nlevels; | |
583 | drro->drr_nblkptr = dnp->dn_nblkptr; | |
584 | ||
585 | /* | |
586 | * Since we encrypt the entire bonus area, the (raw) part | |
4807c0ba | 587 | * beyond the bonuslen is actually nonzero, so we need |
b5256303 TC |
588 | * to send it. |
589 | */ | |
590 | if (bonuslen != 0) { | |
591 | drro->drr_raw_bonuslen = DN_MAX_BONUS_LEN(dnp); | |
592 | bonuslen = drro->drr_raw_bonuslen; | |
593 | } | |
37f8a883 | 594 | } |
34dc7c2f | 595 | |
b5256303 TC |
596 | if (dump_record(dsp, DN_BONUS(dnp), bonuslen) != 0) |
597 | return (SET_ERROR(EINTR)); | |
598 | ||
ea97f8ce | 599 | /* Free anything past the end of the file. */ |
37abac6d | 600 | if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * |
ee45fbd8 | 601 | (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), DMU_OBJECT_END) != 0) |
2e528b49 | 602 | return (SET_ERROR(EINTR)); |
13fe0198 | 603 | if (dsp->dsa_err != 0) |
2e528b49 | 604 | return (SET_ERROR(EINTR)); |
34dc7c2f BB |
605 | return (0); |
606 | } | |
607 | ||
b5256303 TC |
608 | static int |
609 | dump_object_range(dmu_sendarg_t *dsp, const blkptr_t *bp, uint64_t firstobj, | |
610 | uint64_t numslots) | |
611 | { | |
612 | struct drr_object_range *drror = | |
613 | &(dsp->dsa_drr->drr_u.drr_object_range); | |
614 | ||
615 | /* we only use this record type for raw sends */ | |
616 | ASSERT(BP_IS_PROTECTED(bp)); | |
617 | ASSERT(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_RAW); | |
618 | ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); | |
619 | ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_DNODE); | |
620 | ASSERT0(BP_GET_LEVEL(bp)); | |
621 | ||
622 | if (dsp->dsa_pending_op != PENDING_NONE) { | |
623 | if (dump_record(dsp, NULL, 0) != 0) | |
624 | return (SET_ERROR(EINTR)); | |
625 | dsp->dsa_pending_op = PENDING_NONE; | |
626 | } | |
627 | ||
628 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); | |
629 | dsp->dsa_drr->drr_type = DRR_OBJECT_RANGE; | |
630 | drror->drr_firstobj = firstobj; | |
631 | drror->drr_numslots = numslots; | |
632 | drror->drr_toguid = dsp->dsa_toguid; | |
b5256303 TC |
633 | if (BP_SHOULD_BYTESWAP(bp)) |
634 | drror->drr_flags |= DRR_RAW_BYTESWAP; | |
635 | zio_crypt_decode_params_bp(bp, drror->drr_salt, drror->drr_iv); | |
636 | zio_crypt_decode_mac_bp(bp, drror->drr_mac); | |
637 | ||
638 | if (dump_record(dsp, NULL, 0) != 0) | |
639 | return (SET_ERROR(EINTR)); | |
640 | return (0); | |
641 | } | |
642 | ||
9b67f605 MA |
643 | static boolean_t |
644 | backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp) | |
645 | { | |
646 | if (!BP_IS_EMBEDDED(bp)) | |
647 | return (B_FALSE); | |
648 | ||
649 | /* | |
650 | * Compression function must be legacy, or explicitly enabled. | |
651 | */ | |
652 | if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS && | |
2aa34383 | 653 | !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LZ4))) |
9b67f605 MA |
654 | return (B_FALSE); |
655 | ||
656 | /* | |
657 | * Embed type must be explicitly enabled. | |
658 | */ | |
659 | switch (BPE_GET_ETYPE(bp)) { | |
660 | case BP_EMBEDDED_TYPE_DATA: | |
661 | if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) | |
662 | return (B_TRUE); | |
663 | break; | |
664 | default: | |
665 | return (B_FALSE); | |
666 | } | |
667 | return (B_FALSE); | |
668 | } | |
669 | ||
fcff0f35 PD |
670 | /* |
671 | * This is the callback function to traverse_dataset that acts as the worker | |
672 | * thread for dmu_send_impl. | |
673 | */ | |
674 | /*ARGSUSED*/ | |
675 | static int | |
676 | send_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
677 | const zbookmark_phys_t *zb, const struct dnode_phys *dnp, void *arg) | |
678 | { | |
679 | struct send_thread_arg *sta = arg; | |
680 | struct send_block_record *record; | |
681 | uint64_t record_size; | |
682 | int err = 0; | |
683 | ||
47dfff3b MA |
684 | ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || |
685 | zb->zb_object >= sta->resume.zb_object); | |
b5256303 | 686 | ASSERT3P(sta->ds, !=, NULL); |
47dfff3b | 687 | |
fcff0f35 PD |
688 | if (sta->cancel) |
689 | return (SET_ERROR(EINTR)); | |
34dc7c2f | 690 | |
fcff0f35 PD |
691 | if (bp == NULL) { |
692 | ASSERT3U(zb->zb_level, ==, ZB_DNODE_LEVEL); | |
693 | return (0); | |
694 | } else if (zb->zb_level < 0) { | |
695 | return (0); | |
696 | } | |
697 | ||
698 | record = kmem_zalloc(sizeof (struct send_block_record), KM_SLEEP); | |
699 | record->eos_marker = B_FALSE; | |
700 | record->bp = *bp; | |
701 | record->zb = *zb; | |
702 | record->indblkshift = dnp->dn_indblkshift; | |
703 | record->datablkszsec = dnp->dn_datablkszsec; | |
704 | record_size = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; | |
705 | bqueue_enqueue(&sta->q, record, record_size); | |
706 | ||
707 | return (err); | |
708 | } | |
709 | ||
710 | /* | |
711 | * This function kicks off the traverse_dataset. It also handles setting the | |
712 | * error code of the thread in case something goes wrong, and pushes the End of | |
713 | * Stream record when the traverse_dataset call has finished. If there is no | |
714 | * dataset to traverse, the thread immediately pushes End of Stream marker. | |
715 | */ | |
716 | static void | |
717 | send_traverse_thread(void *arg) | |
718 | { | |
719 | struct send_thread_arg *st_arg = arg; | |
720 | int err; | |
721 | struct send_block_record *data; | |
3e635ac1 | 722 | fstrans_cookie_t cookie = spl_fstrans_mark(); |
fcff0f35 PD |
723 | |
724 | if (st_arg->ds != NULL) { | |
47dfff3b MA |
725 | err = traverse_dataset_resume(st_arg->ds, |
726 | st_arg->fromtxg, &st_arg->resume, | |
727 | st_arg->flags, send_cb, st_arg); | |
728 | ||
fcff0f35 PD |
729 | if (err != EINTR) |
730 | st_arg->error_code = err; | |
731 | } | |
732 | data = kmem_zalloc(sizeof (*data), KM_SLEEP); | |
733 | data->eos_marker = B_TRUE; | |
734 | bqueue_enqueue(&st_arg->q, data, 1); | |
3e635ac1 | 735 | spl_fstrans_unmark(cookie); |
34a6b428 | 736 | thread_exit(); |
fcff0f35 PD |
737 | } |
738 | ||
739 | /* | |
740 | * This function actually handles figuring out what kind of record needs to be | |
741 | * dumped, reading the data (which has hopefully been prefetched), and calling | |
742 | * the appropriate helper function. | |
743 | */ | |
34dc7c2f | 744 | static int |
fcff0f35 | 745 | do_dump(dmu_sendarg_t *dsa, struct send_block_record *data) |
34dc7c2f | 746 | { |
fcff0f35 PD |
747 | dsl_dataset_t *ds = dmu_objset_ds(dsa->dsa_os); |
748 | const blkptr_t *bp = &data->bp; | |
749 | const zbookmark_phys_t *zb = &data->zb; | |
750 | uint8_t indblkshift = data->indblkshift; | |
751 | uint16_t dblkszsec = data->datablkszsec; | |
752 | spa_t *spa = ds->ds_dir->dd_pool->dp_spa; | |
34dc7c2f | 753 | dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; |
34dc7c2f BB |
754 | int err = 0; |
755 | ||
fcff0f35 | 756 | ASSERT3U(zb->zb_level, >=, 0); |
34dc7c2f | 757 | |
47dfff3b MA |
758 | ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || |
759 | zb->zb_object >= dsa->dsa_resume_object); | |
760 | ||
b5256303 TC |
761 | /* |
762 | * All bps of an encrypted os should have the encryption bit set. | |
763 | * If this is not true it indicates tampering and we report an error. | |
764 | */ | |
765 | if (dsa->dsa_os->os_encrypted && | |
766 | !BP_IS_HOLE(bp) && !BP_USES_CRYPT(bp)) { | |
767 | spa_log_error(spa, zb); | |
768 | zfs_panic_recover("unencrypted block in encrypted " | |
769 | "object set %llu", ds->ds_object); | |
770 | return (SET_ERROR(EIO)); | |
771 | } | |
772 | ||
428870ff BB |
773 | if (zb->zb_object != DMU_META_DNODE_OBJECT && |
774 | DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { | |
9babb374 | 775 | return (0); |
b0bc7a84 MG |
776 | } else if (BP_IS_HOLE(bp) && |
777 | zb->zb_object == DMU_META_DNODE_OBJECT) { | |
fcff0f35 | 778 | uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); |
b128c09f | 779 | uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; |
fcff0f35 | 780 | err = dump_freeobjects(dsa, dnobj, span >> DNODE_SHIFT); |
b0bc7a84 | 781 | } else if (BP_IS_HOLE(bp)) { |
fcff0f35 PD |
782 | uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); |
783 | uint64_t offset = zb->zb_blkid * span; | |
ee45fbd8 | 784 | /* Don't dump free records for offsets > DMU_OBJECT_END */ |
785 | if (zb->zb_blkid == 0 || span <= DMU_OBJECT_END / zb->zb_blkid) | |
786 | err = dump_free(dsa, zb->zb_object, offset, span); | |
b128c09f BB |
787 | } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { |
788 | return (0); | |
789 | } else if (type == DMU_OT_DNODE) { | |
50c957f7 | 790 | int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; |
2a432414 | 791 | arc_flags_t aflags = ARC_FLAG_WAIT; |
b128c09f | 792 | arc_buf_t *abuf; |
b5256303 | 793 | enum zio_flag zioflags = ZIO_FLAG_CANFAIL; |
fcff0f35 | 794 | |
b5256303 TC |
795 | if (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) { |
796 | ASSERT(BP_IS_ENCRYPTED(bp)); | |
797 | ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); | |
798 | zioflags |= ZIO_FLAG_RAW; | |
799 | } | |
800 | ||
fcff0f35 | 801 | ASSERT0(zb->zb_level); |
b128c09f | 802 | |
294f6806 | 803 | if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, |
b5256303 | 804 | ZIO_PRIORITY_ASYNC_READ, zioflags, &aflags, zb) != 0) |
2e528b49 | 805 | return (SET_ERROR(EIO)); |
34dc7c2f | 806 | |
1c27024e DB |
807 | dnode_phys_t *blk = abuf->b_data; |
808 | uint64_t dnobj = zb->zb_blkid * epb; | |
b5256303 TC |
809 | |
810 | /* | |
811 | * Raw sends require sending encryption parameters for the | |
812 | * block of dnodes. Regular sends do not need to send this | |
813 | * info. | |
814 | */ | |
815 | if (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) { | |
816 | ASSERT(arc_is_encrypted(abuf)); | |
817 | err = dump_object_range(dsa, bp, dnobj, epb); | |
818 | } | |
819 | ||
820 | if (err == 0) { | |
1c27024e DB |
821 | for (int i = 0; i < epb; |
822 | i += blk[i].dn_extra_slots + 1) { | |
b5256303 TC |
823 | err = dump_dnode(dsa, bp, dnobj + i, blk + i); |
824 | if (err != 0) | |
825 | break; | |
826 | } | |
34dc7c2f | 827 | } |
d3c2ae1c | 828 | arc_buf_destroy(abuf, &abuf); |
428870ff | 829 | } else if (type == DMU_OT_SA) { |
2a432414 | 830 | arc_flags_t aflags = ARC_FLAG_WAIT; |
b128c09f | 831 | arc_buf_t *abuf; |
b5256303 TC |
832 | enum zio_flag zioflags = ZIO_FLAG_CANFAIL; |
833 | ||
834 | if (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) { | |
835 | ASSERT(BP_IS_PROTECTED(bp)); | |
836 | zioflags |= ZIO_FLAG_RAW; | |
837 | } | |
b128c09f | 838 | |
294f6806 | 839 | if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, |
b5256303 | 840 | ZIO_PRIORITY_ASYNC_READ, zioflags, &aflags, zb) != 0) |
2e528b49 | 841 | return (SET_ERROR(EIO)); |
b128c09f | 842 | |
b5256303 | 843 | err = dump_spill(dsa, bp, zb->zb_object, abuf->b_data); |
d3c2ae1c | 844 | arc_buf_destroy(abuf, &abuf); |
fcff0f35 | 845 | } else if (backup_do_embed(dsa, bp)) { |
9b67f605 | 846 | /* it's an embedded level-0 block of a regular object */ |
fcff0f35 PD |
847 | int blksz = dblkszsec << SPA_MINBLOCKSHIFT; |
848 | ASSERT0(zb->zb_level); | |
849 | err = dump_write_embedded(dsa, zb->zb_object, | |
9b67f605 | 850 | zb->zb_blkid * blksz, blksz, bp); |
fcff0f35 PD |
851 | } else { |
852 | /* it's a level-0 block of a regular object */ | |
2a432414 | 853 | arc_flags_t aflags = ARC_FLAG_WAIT; |
428870ff | 854 | arc_buf_t *abuf; |
fcff0f35 PD |
855 | int blksz = dblkszsec << SPA_MINBLOCKSHIFT; |
856 | uint64_t offset; | |
2aa34383 DK |
857 | |
858 | /* | |
859 | * If we have large blocks stored on disk but the send flags | |
860 | * don't allow us to send large blocks, we split the data from | |
861 | * the arc buf into chunks. | |
862 | */ | |
a7004725 | 863 | boolean_t split_large_blocks = blksz > SPA_OLD_MAXBLOCKSIZE && |
2aa34383 | 864 | !(dsa->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS); |
b5256303 TC |
865 | |
866 | /* | |
867 | * Raw sends require that we always get raw data as it exists | |
868 | * on disk, so we assert that we are not splitting blocks here. | |
869 | */ | |
870 | boolean_t request_raw = | |
871 | (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) != 0; | |
872 | ||
2aa34383 DK |
873 | /* |
874 | * We should only request compressed data from the ARC if all | |
875 | * the following are true: | |
876 | * - stream compression was requested | |
877 | * - we aren't splitting large blocks into smaller chunks | |
878 | * - the data won't need to be byteswapped before sending | |
879 | * - this isn't an embedded block | |
880 | * - this isn't metadata (if receiving on a different endian | |
881 | * system it can be byteswapped more easily) | |
882 | */ | |
883 | boolean_t request_compressed = | |
884 | (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_COMPRESSED) && | |
885 | !split_large_blocks && !BP_SHOULD_BYTESWAP(bp) && | |
886 | !BP_IS_EMBEDDED(bp) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp)); | |
428870ff | 887 | |
b5256303 TC |
888 | IMPLY(request_raw, !split_large_blocks); |
889 | IMPLY(request_raw, BP_IS_PROTECTED(bp)); | |
da536844 | 890 | ASSERT0(zb->zb_level); |
47dfff3b MA |
891 | ASSERT(zb->zb_object > dsa->dsa_resume_object || |
892 | (zb->zb_object == dsa->dsa_resume_object && | |
893 | zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); | |
894 | ||
a7004725 DK |
895 | ASSERT3U(blksz, ==, BP_GET_LSIZE(bp)); |
896 | ||
897 | enum zio_flag zioflags = ZIO_FLAG_CANFAIL; | |
b5256303 | 898 | if (request_raw) |
2aa34383 | 899 | zioflags |= ZIO_FLAG_RAW; |
b5256303 TC |
900 | else if (request_compressed) |
901 | zioflags |= ZIO_FLAG_RAW_COMPRESS; | |
2aa34383 | 902 | |
294f6806 | 903 | if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, |
a7004725 | 904 | ZIO_PRIORITY_ASYNC_READ, zioflags, &aflags, zb) != 0) { |
330d06f9 | 905 | if (zfs_send_corrupt_data) { |
330d06f9 | 906 | /* Send a block filled with 0x"zfs badd bloc" */ |
2aa34383 DK |
907 | abuf = arc_alloc_buf(spa, &abuf, ARC_BUFC_DATA, |
908 | blksz); | |
a7004725 | 909 | uint64_t *ptr; |
330d06f9 MA |
910 | for (ptr = abuf->b_data; |
911 | (char *)ptr < (char *)abuf->b_data + blksz; | |
912 | ptr++) | |
dd26aa53 | 913 | *ptr = 0x2f5baddb10cULL; |
330d06f9 | 914 | } else { |
2e528b49 | 915 | return (SET_ERROR(EIO)); |
330d06f9 MA |
916 | } |
917 | } | |
428870ff | 918 | |
f1512ee6 MA |
919 | offset = zb->zb_blkid * blksz; |
920 | ||
2aa34383 | 921 | if (split_large_blocks) { |
b5256303 | 922 | ASSERT0(arc_is_encrypted(abuf)); |
2aa34383 DK |
923 | ASSERT3U(arc_get_compression(abuf), ==, |
924 | ZIO_COMPRESS_OFF); | |
a7004725 | 925 | char *buf = abuf->b_data; |
f1512ee6 MA |
926 | while (blksz > 0 && err == 0) { |
927 | int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE); | |
fcff0f35 | 928 | err = dump_write(dsa, type, zb->zb_object, |
2aa34383 | 929 | offset, n, n, NULL, buf); |
f1512ee6 MA |
930 | offset += n; |
931 | buf += n; | |
932 | blksz -= n; | |
933 | } | |
934 | } else { | |
2aa34383 | 935 | err = dump_write(dsa, type, zb->zb_object, offset, |
b5256303 | 936 | blksz, arc_buf_size(abuf), bp, abuf->b_data); |
f1512ee6 | 937 | } |
d3c2ae1c | 938 | arc_buf_destroy(abuf, &abuf); |
34dc7c2f BB |
939 | } |
940 | ||
941 | ASSERT(err == 0 || err == EINTR); | |
942 | return (err); | |
943 | } | |
944 | ||
6f1ffb06 | 945 | /* |
fcff0f35 PD |
946 | * Pop the new data off the queue, and free the old data. |
947 | */ | |
948 | static struct send_block_record * | |
949 | get_next_record(bqueue_t *bq, struct send_block_record *data) | |
950 | { | |
951 | struct send_block_record *tmp = bqueue_dequeue(bq); | |
952 | kmem_free(data, sizeof (*data)); | |
953 | return (tmp); | |
954 | } | |
955 | ||
956 | /* | |
957 | * Actually do the bulk of the work in a zfs send. | |
958 | * | |
959 | * Note: Releases dp using the specified tag. | |
6f1ffb06 | 960 | */ |
13fe0198 | 961 | static int |
fcff0f35 | 962 | dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *to_ds, |
2aa34383 DK |
963 | zfs_bookmark_phys_t *ancestor_zb, boolean_t is_clone, |
964 | boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, | |
b5256303 | 965 | boolean_t rawok, int outfd, uint64_t resumeobj, uint64_t resumeoff, |
47dfff3b | 966 | vnode_t *vp, offset_t *off) |
34dc7c2f | 967 | { |
13fe0198 | 968 | objset_t *os; |
34dc7c2f | 969 | dmu_replay_record_t *drr; |
37abac6d | 970 | dmu_sendarg_t *dsp; |
34dc7c2f BB |
971 | int err; |
972 | uint64_t fromtxg = 0; | |
9b67f605 | 973 | uint64_t featureflags = 0; |
fcff0f35 | 974 | struct send_thread_arg to_arg; |
47dfff3b MA |
975 | void *payload = NULL; |
976 | size_t payload_len = 0; | |
fcff0f35 | 977 | struct send_block_record *to_data; |
34dc7c2f | 978 | |
fcff0f35 | 979 | err = dmu_objset_from_ds(to_ds, &os); |
13fe0198 | 980 | if (err != 0) { |
13fe0198 MA |
981 | dsl_pool_rele(dp, tag); |
982 | return (err); | |
983 | } | |
34dc7c2f | 984 | |
b5256303 TC |
985 | /* |
986 | * If this is a non-raw send of an encrypted ds, we can ensure that | |
987 | * the objset_phys_t is authenticated. This is safe because this is | |
988 | * either a snapshot or we have owned the dataset, ensuring that | |
989 | * it can't be modified. | |
990 | */ | |
991 | if (!rawok && os->os_encrypted && | |
992 | arc_is_unauthenticated(os->os_phys_buf)) { | |
a2c2ed1b TC |
993 | zbookmark_phys_t zb; |
994 | ||
995 | SET_BOOKMARK(&zb, to_ds->ds_object, ZB_ROOT_OBJECT, | |
996 | ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
b5256303 | 997 | err = arc_untransform(os->os_phys_buf, os->os_spa, |
a2c2ed1b | 998 | &zb, B_FALSE); |
b5256303 TC |
999 | if (err != 0) { |
1000 | dsl_pool_rele(dp, tag); | |
1001 | return (err); | |
1002 | } | |
1003 | ||
1004 | ASSERT0(arc_is_unauthenticated(os->os_phys_buf)); | |
1005 | } | |
1006 | ||
34dc7c2f BB |
1007 | drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); |
1008 | drr->drr_type = DRR_BEGIN; | |
1009 | drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; | |
428870ff BB |
1010 | DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo, |
1011 | DMU_SUBSTREAM); | |
1012 | ||
47dfff3b MA |
1013 | bzero(&to_arg, sizeof (to_arg)); |
1014 | ||
428870ff | 1015 | #ifdef _KERNEL |
13fe0198 | 1016 | if (dmu_objset_type(os) == DMU_OST_ZFS) { |
428870ff | 1017 | uint64_t version; |
13fe0198 | 1018 | if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { |
37abac6d | 1019 | kmem_free(drr, sizeof (dmu_replay_record_t)); |
13fe0198 | 1020 | dsl_pool_rele(dp, tag); |
2e528b49 | 1021 | return (SET_ERROR(EINVAL)); |
37abac6d | 1022 | } |
13fe0198 | 1023 | if (version >= ZPL_VERSION_SA) { |
9b67f605 | 1024 | featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; |
428870ff BB |
1025 | } |
1026 | } | |
1027 | #endif | |
1028 | ||
b5256303 TC |
1029 | /* raw sends imply large_block_ok */ |
1030 | if ((large_block_ok || rawok) && | |
1031 | to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_BLOCKS]) | |
f1512ee6 | 1032 | featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS; |
50c957f7 NB |
1033 | if (to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_DNODE]) |
1034 | featureflags |= DMU_BACKUP_FEATURE_LARGE_DNODE; | |
b5256303 TC |
1035 | |
1036 | /* encrypted datasets will not have embedded blocks */ | |
1037 | if ((embedok || rawok) && !os->os_encrypted && | |
9b67f605 MA |
1038 | spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) { |
1039 | featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA; | |
2aa34383 | 1040 | } |
b5256303 TC |
1041 | |
1042 | /* raw send implies compressok */ | |
1043 | if (compressok || rawok) | |
2aa34383 | 1044 | featureflags |= DMU_BACKUP_FEATURE_COMPRESSED; |
b5256303 TC |
1045 | if (rawok && os->os_encrypted) |
1046 | featureflags |= DMU_BACKUP_FEATURE_RAW; | |
1047 | ||
2aa34383 | 1048 | if ((featureflags & |
b5256303 TC |
1049 | (DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_COMPRESSED | |
1050 | DMU_BACKUP_FEATURE_RAW)) != 0 && | |
1051 | spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) { | |
2aa34383 | 1052 | featureflags |= DMU_BACKUP_FEATURE_LZ4; |
9b67f605 MA |
1053 | } |
1054 | ||
47dfff3b MA |
1055 | if (resumeobj != 0 || resumeoff != 0) { |
1056 | featureflags |= DMU_BACKUP_FEATURE_RESUMING; | |
1057 | } | |
1058 | ||
9b67f605 MA |
1059 | DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo, |
1060 | featureflags); | |
1061 | ||
34dc7c2f | 1062 | drr->drr_u.drr_begin.drr_creation_time = |
fcff0f35 | 1063 | dsl_dataset_phys(to_ds)->ds_creation_time; |
13fe0198 | 1064 | drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); |
da536844 | 1065 | if (is_clone) |
34dc7c2f | 1066 | drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; |
fcff0f35 PD |
1067 | drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(to_ds)->ds_guid; |
1068 | if (dsl_dataset_phys(to_ds)->ds_flags & DS_FLAG_CI_DATASET) | |
34dc7c2f | 1069 | drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; |
b607405f AS |
1070 | if (zfs_send_set_freerecords_bit) |
1071 | drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_FREERECORDS; | |
34dc7c2f | 1072 | |
fcff0f35 PD |
1073 | if (ancestor_zb != NULL) { |
1074 | drr->drr_u.drr_begin.drr_fromguid = | |
1075 | ancestor_zb->zbm_guid; | |
1076 | fromtxg = ancestor_zb->zbm_creation_txg; | |
da536844 | 1077 | } |
fcff0f35 PD |
1078 | dsl_dataset_name(to_ds, drr->drr_u.drr_begin.drr_toname); |
1079 | if (!to_ds->ds_is_snapshot) { | |
da536844 MA |
1080 | (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--", |
1081 | sizeof (drr->drr_u.drr_begin.drr_toname)); | |
13fe0198 | 1082 | } |
34dc7c2f | 1083 | |
37abac6d BP |
1084 | dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); |
1085 | ||
1086 | dsp->dsa_drr = drr; | |
1087 | dsp->dsa_vp = vp; | |
1088 | dsp->dsa_outfd = outfd; | |
1089 | dsp->dsa_proc = curproc; | |
13fe0198 | 1090 | dsp->dsa_os = os; |
37abac6d | 1091 | dsp->dsa_off = off; |
fcff0f35 | 1092 | dsp->dsa_toguid = dsl_dataset_phys(to_ds)->ds_guid; |
37abac6d | 1093 | dsp->dsa_pending_op = PENDING_NONE; |
9b67f605 | 1094 | dsp->dsa_featureflags = featureflags; |
47dfff3b MA |
1095 | dsp->dsa_resume_object = resumeobj; |
1096 | dsp->dsa_resume_offset = resumeoff; | |
37abac6d | 1097 | |
fcff0f35 PD |
1098 | mutex_enter(&to_ds->ds_sendstream_lock); |
1099 | list_insert_head(&to_ds->ds_sendstreams, dsp); | |
1100 | mutex_exit(&to_ds->ds_sendstream_lock); | |
37abac6d | 1101 | |
fcff0f35 | 1102 | dsl_dataset_long_hold(to_ds, FTAG); |
7ec09286 MA |
1103 | dsl_pool_rele(dp, tag); |
1104 | ||
b5256303 TC |
1105 | /* handle features that require a DRR_BEGIN payload */ |
1106 | if (featureflags & | |
1107 | (DMU_BACKUP_FEATURE_RESUMING | DMU_BACKUP_FEATURE_RAW)) { | |
1108 | nvlist_t *keynvl = NULL; | |
1109 | nvlist_t *nvl = fnvlist_alloc(); | |
1110 | ||
1111 | if (featureflags & DMU_BACKUP_FEATURE_RESUMING) { | |
1112 | dmu_object_info_t to_doi; | |
1113 | err = dmu_object_info(os, resumeobj, &to_doi); | |
1114 | if (err != 0) { | |
1115 | fnvlist_free(nvl); | |
1116 | goto out; | |
1117 | } | |
1118 | ||
1119 | SET_BOOKMARK(&to_arg.resume, to_ds->ds_object, | |
1120 | resumeobj, 0, | |
1121 | resumeoff / to_doi.doi_data_block_size); | |
1122 | ||
1123 | fnvlist_add_uint64(nvl, "resume_object", resumeobj); | |
1124 | fnvlist_add_uint64(nvl, "resume_offset", resumeoff); | |
1125 | } | |
1126 | ||
1127 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
1128 | ASSERT(os->os_encrypted); | |
1129 | ||
1130 | err = dsl_crypto_populate_key_nvlist(to_ds, &keynvl); | |
1131 | if (err != 0) { | |
1132 | fnvlist_free(nvl); | |
1133 | goto out; | |
1134 | } | |
1135 | ||
1136 | fnvlist_add_nvlist(nvl, "crypt_keydata", keynvl); | |
1137 | } | |
47dfff3b | 1138 | |
47dfff3b MA |
1139 | payload = fnvlist_pack(nvl, &payload_len); |
1140 | drr->drr_payloadlen = payload_len; | |
b5256303 | 1141 | fnvlist_free(keynvl); |
47dfff3b MA |
1142 | fnvlist_free(nvl); |
1143 | } | |
1144 | ||
1145 | err = dump_record(dsp, payload, payload_len); | |
1146 | fnvlist_pack_free(payload, payload_len); | |
1147 | if (err != 0) { | |
37abac6d BP |
1148 | err = dsp->dsa_err; |
1149 | goto out; | |
34dc7c2f BB |
1150 | } |
1151 | ||
3b0d9928 BB |
1152 | err = bqueue_init(&to_arg.q, |
1153 | MAX(zfs_send_queue_length, 2 * zfs_max_recordsize), | |
fcff0f35 PD |
1154 | offsetof(struct send_block_record, ln)); |
1155 | to_arg.error_code = 0; | |
1156 | to_arg.cancel = B_FALSE; | |
1157 | to_arg.ds = to_ds; | |
1158 | to_arg.fromtxg = fromtxg; | |
1159 | to_arg.flags = TRAVERSE_PRE | TRAVERSE_PREFETCH; | |
b5256303 TC |
1160 | if (rawok) |
1161 | to_arg.flags |= TRAVERSE_NO_DECRYPT; | |
fcff0f35 PD |
1162 | (void) thread_create(NULL, 0, send_traverse_thread, &to_arg, 0, curproc, |
1163 | TS_RUN, minclsyspri); | |
1164 | ||
1165 | to_data = bqueue_dequeue(&to_arg.q); | |
1166 | ||
1167 | while (!to_data->eos_marker && err == 0) { | |
1168 | err = do_dump(dsp, to_data); | |
1169 | to_data = get_next_record(&to_arg.q, to_data); | |
1170 | if (issig(JUSTLOOKING) && issig(FORREAL)) | |
1171 | err = EINTR; | |
1172 | } | |
1173 | ||
1174 | if (err != 0) { | |
1175 | to_arg.cancel = B_TRUE; | |
1176 | while (!to_data->eos_marker) { | |
1177 | to_data = get_next_record(&to_arg.q, to_data); | |
1178 | } | |
1179 | } | |
1180 | kmem_free(to_data, sizeof (*to_data)); | |
1181 | ||
1182 | bqueue_destroy(&to_arg.q); | |
1183 | ||
1184 | if (err == 0 && to_arg.error_code != 0) | |
1185 | err = to_arg.error_code; | |
1186 | ||
1187 | if (err != 0) | |
1188 | goto out; | |
34dc7c2f | 1189 | |
37abac6d | 1190 | if (dsp->dsa_pending_op != PENDING_NONE) |
37f8a883 | 1191 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 1192 | err = SET_ERROR(EINTR); |
428870ff | 1193 | |
13fe0198 MA |
1194 | if (err != 0) { |
1195 | if (err == EINTR && dsp->dsa_err != 0) | |
37abac6d BP |
1196 | err = dsp->dsa_err; |
1197 | goto out; | |
34dc7c2f BB |
1198 | } |
1199 | ||
1200 | bzero(drr, sizeof (dmu_replay_record_t)); | |
1201 | drr->drr_type = DRR_END; | |
37abac6d BP |
1202 | drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc; |
1203 | drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid; | |
34dc7c2f | 1204 | |
fcff0f35 | 1205 | if (dump_record(dsp, NULL, 0) != 0) |
37abac6d | 1206 | err = dsp->dsa_err; |
37abac6d | 1207 | out: |
fcff0f35 PD |
1208 | mutex_enter(&to_ds->ds_sendstream_lock); |
1209 | list_remove(&to_ds->ds_sendstreams, dsp); | |
1210 | mutex_exit(&to_ds->ds_sendstream_lock); | |
37abac6d | 1211 | |
51907a31 K |
1212 | VERIFY(err != 0 || (dsp->dsa_sent_begin && dsp->dsa_sent_end)); |
1213 | ||
34dc7c2f | 1214 | kmem_free(drr, sizeof (dmu_replay_record_t)); |
37abac6d | 1215 | kmem_free(dsp, sizeof (dmu_sendarg_t)); |
34dc7c2f | 1216 | |
fcff0f35 | 1217 | dsl_dataset_long_rele(to_ds, FTAG); |
13fe0198 | 1218 | |
37abac6d | 1219 | return (err); |
34dc7c2f BB |
1220 | } |
1221 | ||
330d06f9 | 1222 | int |
13fe0198 | 1223 | dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, |
2aa34383 | 1224 | boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, |
b5256303 | 1225 | boolean_t rawok, int outfd, vnode_t *vp, offset_t *off) |
13fe0198 MA |
1226 | { |
1227 | dsl_pool_t *dp; | |
1228 | dsl_dataset_t *ds; | |
1229 | dsl_dataset_t *fromds = NULL; | |
b5256303 | 1230 | ds_hold_flags_t dsflags = (rawok) ? 0 : DS_HOLD_FLAG_DECRYPT; |
13fe0198 MA |
1231 | int err; |
1232 | ||
1233 | err = dsl_pool_hold(pool, FTAG, &dp); | |
1234 | if (err != 0) | |
1235 | return (err); | |
1236 | ||
b5256303 | 1237 | err = dsl_dataset_hold_obj_flags(dp, tosnap, dsflags, FTAG, &ds); |
13fe0198 MA |
1238 | if (err != 0) { |
1239 | dsl_pool_rele(dp, FTAG); | |
1240 | return (err); | |
1241 | } | |
1242 | ||
1243 | if (fromsnap != 0) { | |
da536844 MA |
1244 | zfs_bookmark_phys_t zb; |
1245 | boolean_t is_clone; | |
1246 | ||
13fe0198 MA |
1247 | err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); |
1248 | if (err != 0) { | |
b5256303 | 1249 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
13fe0198 MA |
1250 | dsl_pool_rele(dp, FTAG); |
1251 | return (err); | |
1252 | } | |
da536844 MA |
1253 | if (!dsl_dataset_is_before(ds, fromds, 0)) |
1254 | err = SET_ERROR(EXDEV); | |
d683ddbb JG |
1255 | zb.zbm_creation_time = |
1256 | dsl_dataset_phys(fromds)->ds_creation_time; | |
1257 | zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg; | |
1258 | zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; | |
da536844 MA |
1259 | is_clone = (fromds->ds_dir != ds->ds_dir); |
1260 | dsl_dataset_rele(fromds, FTAG); | |
f1512ee6 | 1261 | err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, |
b5256303 TC |
1262 | embedok, large_block_ok, compressok, rawok, outfd, |
1263 | 0, 0, vp, off); | |
da536844 | 1264 | } else { |
f1512ee6 | 1265 | err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, |
b5256303 TC |
1266 | embedok, large_block_ok, compressok, rawok, outfd, |
1267 | 0, 0, vp, off); | |
13fe0198 | 1268 | } |
b5256303 | 1269 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
da536844 | 1270 | return (err); |
13fe0198 MA |
1271 | } |
1272 | ||
1273 | int | |
47dfff3b | 1274 | dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok, |
b5256303 TC |
1275 | boolean_t large_block_ok, boolean_t compressok, boolean_t rawok, |
1276 | int outfd, uint64_t resumeobj, uint64_t resumeoff, vnode_t *vp, | |
1277 | offset_t *off) | |
13fe0198 MA |
1278 | { |
1279 | dsl_pool_t *dp; | |
1280 | dsl_dataset_t *ds; | |
13fe0198 | 1281 | int err; |
b5256303 | 1282 | ds_hold_flags_t dsflags = (rawok) ? 0 : DS_HOLD_FLAG_DECRYPT; |
da536844 | 1283 | boolean_t owned = B_FALSE; |
13fe0198 | 1284 | |
da536844 | 1285 | if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL) |
2e528b49 | 1286 | return (SET_ERROR(EINVAL)); |
13fe0198 MA |
1287 | |
1288 | err = dsl_pool_hold(tosnap, FTAG, &dp); | |
1289 | if (err != 0) | |
1290 | return (err); | |
1291 | ||
da536844 MA |
1292 | if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) { |
1293 | /* | |
1294 | * We are sending a filesystem or volume. Ensure | |
1295 | * that it doesn't change by owning the dataset. | |
1296 | */ | |
b5256303 | 1297 | err = dsl_dataset_own(dp, tosnap, dsflags, FTAG, &ds); |
da536844 MA |
1298 | owned = B_TRUE; |
1299 | } else { | |
b5256303 | 1300 | err = dsl_dataset_hold_flags(dp, tosnap, dsflags, FTAG, &ds); |
da536844 | 1301 | } |
13fe0198 MA |
1302 | if (err != 0) { |
1303 | dsl_pool_rele(dp, FTAG); | |
1304 | return (err); | |
1305 | } | |
1306 | ||
1307 | if (fromsnap != NULL) { | |
da536844 MA |
1308 | zfs_bookmark_phys_t zb; |
1309 | boolean_t is_clone = B_FALSE; | |
1310 | int fsnamelen = strchr(tosnap, '@') - tosnap; | |
1311 | ||
1312 | /* | |
1313 | * If the fromsnap is in a different filesystem, then | |
1314 | * mark the send stream as a clone. | |
1315 | */ | |
1316 | if (strncmp(tosnap, fromsnap, fsnamelen) != 0 || | |
1317 | (fromsnap[fsnamelen] != '@' && | |
1318 | fromsnap[fsnamelen] != '#')) { | |
1319 | is_clone = B_TRUE; | |
1320 | } | |
1321 | ||
1322 | if (strchr(fromsnap, '@')) { | |
1323 | dsl_dataset_t *fromds; | |
1324 | err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); | |
1325 | if (err == 0) { | |
1326 | if (!dsl_dataset_is_before(ds, fromds, 0)) | |
1327 | err = SET_ERROR(EXDEV); | |
1328 | zb.zbm_creation_time = | |
d683ddbb | 1329 | dsl_dataset_phys(fromds)->ds_creation_time; |
da536844 | 1330 | zb.zbm_creation_txg = |
d683ddbb JG |
1331 | dsl_dataset_phys(fromds)->ds_creation_txg; |
1332 | zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; | |
da536844 MA |
1333 | is_clone = (ds->ds_dir != fromds->ds_dir); |
1334 | dsl_dataset_rele(fromds, FTAG); | |
1335 | } | |
1336 | } else { | |
1337 | err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb); | |
1338 | } | |
13fe0198 | 1339 | if (err != 0) { |
b5256303 TC |
1340 | if (owned) |
1341 | dsl_dataset_disown(ds, dsflags, FTAG); | |
1342 | else | |
1343 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1344 | ||
13fe0198 MA |
1345 | dsl_pool_rele(dp, FTAG); |
1346 | return (err); | |
1347 | } | |
f1512ee6 | 1348 | err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, |
b5256303 | 1349 | embedok, large_block_ok, compressok, rawok, |
47dfff3b | 1350 | outfd, resumeobj, resumeoff, vp, off); |
da536844 | 1351 | } else { |
f1512ee6 | 1352 | err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, |
b5256303 | 1353 | embedok, large_block_ok, compressok, rawok, |
47dfff3b | 1354 | outfd, resumeobj, resumeoff, vp, off); |
13fe0198 | 1355 | } |
da536844 | 1356 | if (owned) |
b5256303 | 1357 | dsl_dataset_disown(ds, dsflags, FTAG); |
da536844 | 1358 | else |
b5256303 TC |
1359 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
1360 | ||
da536844 | 1361 | return (err); |
13fe0198 MA |
1362 | } |
1363 | ||
5dc8b736 | 1364 | static int |
2aa34383 DK |
1365 | dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t uncompressed, |
1366 | uint64_t compressed, boolean_t stream_compressed, uint64_t *sizep) | |
5dc8b736 | 1367 | { |
ca0845d5 | 1368 | int err = 0; |
2aa34383 | 1369 | uint64_t size; |
5dc8b736 MG |
1370 | /* |
1371 | * Assume that space (both on-disk and in-stream) is dominated by | |
1372 | * data. We will adjust for indirect blocks and the copies property, | |
1373 | * but ignore per-object space used (eg, dnodes and DRR_OBJECT records). | |
1374 | */ | |
1375 | ||
2aa34383 DK |
1376 | uint64_t recordsize; |
1377 | uint64_t record_count; | |
dd429b46 PD |
1378 | objset_t *os; |
1379 | VERIFY0(dmu_objset_from_ds(ds, &os)); | |
2aa34383 DK |
1380 | |
1381 | /* Assume all (uncompressed) blocks are recordsize. */ | |
ca0845d5 PD |
1382 | if (zfs_override_estimate_recordsize != 0) { |
1383 | recordsize = zfs_override_estimate_recordsize; | |
1384 | } else if (os->os_phys->os_type == DMU_OST_ZVOL) { | |
dd429b46 PD |
1385 | err = dsl_prop_get_int_ds(ds, |
1386 | zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &recordsize); | |
1387 | } else { | |
1388 | err = dsl_prop_get_int_ds(ds, | |
1389 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), &recordsize); | |
1390 | } | |
2aa34383 DK |
1391 | if (err != 0) |
1392 | return (err); | |
1393 | record_count = uncompressed / recordsize; | |
1394 | ||
1395 | /* | |
1396 | * If we're estimating a send size for a compressed stream, use the | |
1397 | * compressed data size to estimate the stream size. Otherwise, use the | |
1398 | * uncompressed data size. | |
1399 | */ | |
1400 | size = stream_compressed ? compressed : uncompressed; | |
1401 | ||
5dc8b736 MG |
1402 | /* |
1403 | * Subtract out approximate space used by indirect blocks. | |
1404 | * Assume most space is used by data blocks (non-indirect, non-dnode). | |
2aa34383 | 1405 | * Assume no ditto blocks or internal fragmentation. |
5dc8b736 MG |
1406 | * |
1407 | * Therefore, space used by indirect blocks is sizeof(blkptr_t) per | |
2aa34383 | 1408 | * block. |
5dc8b736 | 1409 | */ |
2aa34383 | 1410 | size -= record_count * sizeof (blkptr_t); |
5dc8b736 MG |
1411 | |
1412 | /* Add in the space for the record associated with each block. */ | |
2aa34383 | 1413 | size += record_count * sizeof (dmu_replay_record_t); |
5dc8b736 MG |
1414 | |
1415 | *sizep = size; | |
1416 | ||
1417 | return (0); | |
1418 | } | |
1419 | ||
13fe0198 | 1420 | int |
2aa34383 DK |
1421 | dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, |
1422 | boolean_t stream_compressed, uint64_t *sizep) | |
330d06f9 | 1423 | { |
330d06f9 | 1424 | int err; |
2aa34383 | 1425 | uint64_t uncomp, comp; |
13fe0198 | 1426 | |
fd0fd646 | 1427 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); |
330d06f9 MA |
1428 | |
1429 | /* tosnap must be a snapshot */ | |
0c66c32d | 1430 | if (!ds->ds_is_snapshot) |
2e528b49 | 1431 | return (SET_ERROR(EINVAL)); |
330d06f9 | 1432 | |
71e2fe41 AG |
1433 | /* fromsnap, if provided, must be a snapshot */ |
1434 | if (fromds != NULL && !fromds->ds_is_snapshot) | |
1435 | return (SET_ERROR(EINVAL)); | |
1436 | ||
6f1ffb06 MA |
1437 | /* |
1438 | * fromsnap must be an earlier snapshot from the same fs as tosnap, | |
1439 | * or the origin's fs. | |
1440 | */ | |
da536844 | 1441 | if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0)) |
2e528b49 | 1442 | return (SET_ERROR(EXDEV)); |
330d06f9 | 1443 | |
2aa34383 | 1444 | /* Get compressed and uncompressed size estimates of changed data. */ |
330d06f9 | 1445 | if (fromds == NULL) { |
2aa34383 DK |
1446 | uncomp = dsl_dataset_phys(ds)->ds_uncompressed_bytes; |
1447 | comp = dsl_dataset_phys(ds)->ds_compressed_bytes; | |
330d06f9 | 1448 | } else { |
2aa34383 | 1449 | uint64_t used; |
330d06f9 | 1450 | err = dsl_dataset_space_written(fromds, ds, |
2aa34383 | 1451 | &used, &comp, &uncomp); |
13fe0198 | 1452 | if (err != 0) |
330d06f9 MA |
1453 | return (err); |
1454 | } | |
1455 | ||
2aa34383 DK |
1456 | err = dmu_adjust_send_estimate_for_indirects(ds, uncomp, comp, |
1457 | stream_compressed, sizep); | |
dd429b46 PD |
1458 | /* |
1459 | * Add the size of the BEGIN and END records to the estimate. | |
1460 | */ | |
1461 | *sizep += 2 * sizeof (dmu_replay_record_t); | |
5dc8b736 MG |
1462 | return (err); |
1463 | } | |
330d06f9 | 1464 | |
2aa34383 DK |
1465 | struct calculate_send_arg { |
1466 | uint64_t uncompressed; | |
1467 | uint64_t compressed; | |
1468 | }; | |
1469 | ||
5dc8b736 MG |
1470 | /* |
1471 | * Simple callback used to traverse the blocks of a snapshot and sum their | |
2aa34383 | 1472 | * uncompressed and compressed sizes. |
5dc8b736 MG |
1473 | */ |
1474 | /* ARGSUSED */ | |
1475 | static int | |
1476 | dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
1477 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) | |
1478 | { | |
2aa34383 | 1479 | struct calculate_send_arg *space = arg; |
5dc8b736 | 1480 | if (bp != NULL && !BP_IS_HOLE(bp)) { |
2aa34383 DK |
1481 | space->uncompressed += BP_GET_UCSIZE(bp); |
1482 | space->compressed += BP_GET_PSIZE(bp); | |
5dc8b736 MG |
1483 | } |
1484 | return (0); | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Given a desination snapshot and a TXG, calculate the approximate size of a | |
1489 | * send stream sent from that TXG. from_txg may be zero, indicating that the | |
1490 | * whole snapshot will be sent. | |
1491 | */ | |
1492 | int | |
1493 | dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg, | |
2aa34383 | 1494 | boolean_t stream_compressed, uint64_t *sizep) |
5dc8b736 | 1495 | { |
5dc8b736 | 1496 | int err; |
2aa34383 | 1497 | struct calculate_send_arg size = { 0 }; |
5dc8b736 | 1498 | |
fd0fd646 | 1499 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); |
5dc8b736 MG |
1500 | |
1501 | /* tosnap must be a snapshot */ | |
1502 | if (!dsl_dataset_is_snapshot(ds)) | |
1503 | return (SET_ERROR(EINVAL)); | |
1504 | ||
1505 | /* verify that from_txg is before the provided snapshot was taken */ | |
1506 | if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) { | |
1507 | return (SET_ERROR(EXDEV)); | |
1508 | } | |
330d06f9 | 1509 | /* |
5dc8b736 MG |
1510 | * traverse the blocks of the snapshot with birth times after |
1511 | * from_txg, summing their uncompressed size | |
330d06f9 | 1512 | */ |
b5256303 TC |
1513 | err = traverse_dataset(ds, from_txg, |
1514 | TRAVERSE_POST | TRAVERSE_NO_DECRYPT, | |
5dc8b736 | 1515 | dmu_calculate_send_traversal, &size); |
2aa34383 | 1516 | |
5dc8b736 | 1517 | if (err) |
330d06f9 | 1518 | return (err); |
330d06f9 | 1519 | |
2aa34383 DK |
1520 | err = dmu_adjust_send_estimate_for_indirects(ds, size.uncompressed, |
1521 | size.compressed, stream_compressed, sizep); | |
5dc8b736 | 1522 | return (err); |
330d06f9 MA |
1523 | } |
1524 | ||
13fe0198 MA |
1525 | typedef struct dmu_recv_begin_arg { |
1526 | const char *drba_origin; | |
1527 | dmu_recv_cookie_t *drba_cookie; | |
1528 | cred_t *drba_cred; | |
d9c460a0 | 1529 | dsl_crypto_params_t *drba_dcp; |
19580676 | 1530 | uint64_t drba_snapobj; |
13fe0198 | 1531 | } dmu_recv_begin_arg_t; |
34dc7c2f | 1532 | |
34dc7c2f | 1533 | static int |
13fe0198 | 1534 | recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, |
d9c460a0 | 1535 | uint64_t fromguid, uint64_t featureflags) |
34dc7c2f | 1536 | { |
34dc7c2f | 1537 | uint64_t val; |
13fe0198 MA |
1538 | int error; |
1539 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
4a385862 TC |
1540 | boolean_t encrypted = ds->ds_dir->dd_crypto_obj != 0; |
1541 | boolean_t raw = (featureflags & DMU_BACKUP_FEATURE_RAW) != 0; | |
1fff937a | 1542 | boolean_t embed = (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) != 0; |
34dc7c2f | 1543 | |
13fe0198 MA |
1544 | /* temporary clone name must not exist */ |
1545 | error = zap_lookup(dp->dp_meta_objset, | |
d683ddbb | 1546 | dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, |
13fe0198 MA |
1547 | 8, 1, &val); |
1548 | if (error != ENOENT) | |
1549 | return (error == 0 ? EBUSY : error); | |
1550 | ||
572e2857 | 1551 | /* new snapshot name must not exist */ |
13fe0198 | 1552 | error = zap_lookup(dp->dp_meta_objset, |
d683ddbb JG |
1553 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, |
1554 | drba->drba_cookie->drc_tosnap, 8, 1, &val); | |
13fe0198 MA |
1555 | if (error != ENOENT) |
1556 | return (error == 0 ? EEXIST : error); | |
572e2857 | 1557 | |
788eb90c JJ |
1558 | /* |
1559 | * Check snapshot limit before receiving. We'll recheck again at the | |
1560 | * end, but might as well abort before receiving if we're already over | |
1561 | * the limit. | |
1562 | * | |
1563 | * Note that we do not check the file system limit with | |
1564 | * dsl_dir_fscount_check because the temporary %clones don't count | |
1565 | * against that limit. | |
1566 | */ | |
1567 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, | |
1568 | NULL, drba->drba_cred); | |
1569 | if (error != 0) | |
1570 | return (error); | |
1571 | ||
13fe0198 | 1572 | if (fromguid != 0) { |
19580676 | 1573 | dsl_dataset_t *snap; |
d683ddbb | 1574 | uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; |
19580676 | 1575 | |
4a385862 TC |
1576 | /* Can't perform a raw receive on top of a non-raw receive */ |
1577 | if (!encrypted && raw) | |
1578 | return (SET_ERROR(EINVAL)); | |
1579 | ||
1fff937a TC |
1580 | /* Encryption is incompatible with embedded data */ |
1581 | if (encrypted && embed) | |
1582 | return (SET_ERROR(EINVAL)); | |
1583 | ||
19580676 MA |
1584 | /* Find snapshot in this dir that matches fromguid. */ |
1585 | while (obj != 0) { | |
1586 | error = dsl_dataset_hold_obj(dp, obj, FTAG, | |
1587 | &snap); | |
1588 | if (error != 0) | |
1589 | return (SET_ERROR(ENODEV)); | |
1590 | if (snap->ds_dir != ds->ds_dir) { | |
1591 | dsl_dataset_rele(snap, FTAG); | |
1592 | return (SET_ERROR(ENODEV)); | |
1593 | } | |
d683ddbb | 1594 | if (dsl_dataset_phys(snap)->ds_guid == fromguid) |
19580676 | 1595 | break; |
d683ddbb | 1596 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; |
19580676 MA |
1597 | dsl_dataset_rele(snap, FTAG); |
1598 | } | |
1599 | if (obj == 0) | |
2e528b49 | 1600 | return (SET_ERROR(ENODEV)); |
34dc7c2f | 1601 | |
19580676 MA |
1602 | if (drba->drba_cookie->drc_force) { |
1603 | drba->drba_snapobj = obj; | |
1604 | } else { | |
1605 | /* | |
1606 | * If we are not forcing, there must be no | |
1607 | * changes since fromsnap. | |
1608 | */ | |
1609 | if (dsl_dataset_modified_since_snap(ds, snap)) { | |
428870ff | 1610 | dsl_dataset_rele(snap, FTAG); |
19580676 | 1611 | return (SET_ERROR(ETXTBSY)); |
428870ff | 1612 | } |
19580676 | 1613 | drba->drba_snapobj = ds->ds_prev->ds_object; |
428870ff | 1614 | } |
19580676 MA |
1615 | |
1616 | dsl_dataset_rele(snap, FTAG); | |
428870ff | 1617 | } else { |
cf50a2b0 AG |
1618 | /* if full, then must be forced */ |
1619 | if (!drba->drba_cookie->drc_force) | |
1620 | return (SET_ERROR(EEXIST)); | |
b5256303 TC |
1621 | |
1622 | /* | |
1623 | * We don't support using zfs recv -F to blow away | |
1624 | * encrypted filesystems. This would require the | |
1625 | * dsl dir to point to the old encryption key and | |
1626 | * the new one at the same time during the receive. | |
1627 | */ | |
4a385862 | 1628 | if ((!encrypted && raw) || encrypted) |
b5256303 TC |
1629 | return (SET_ERROR(EINVAL)); |
1630 | ||
1fff937a TC |
1631 | /* |
1632 | * Perform the same encryption checks we would if | |
1633 | * we were creating a new dataset from scratch. | |
1634 | */ | |
1635 | if (!raw) { | |
1636 | boolean_t will_encrypt; | |
1637 | ||
d9c460a0 | 1638 | error = dmu_objset_create_crypt_check( |
1fff937a TC |
1639 | ds->ds_dir->dd_parent, drba->drba_dcp, |
1640 | &will_encrypt); | |
d9c460a0 TC |
1641 | if (error != 0) |
1642 | return (error); | |
1fff937a TC |
1643 | |
1644 | if (will_encrypt && embed) | |
1645 | return (SET_ERROR(EINVAL)); | |
d9c460a0 TC |
1646 | } |
1647 | ||
b5256303 | 1648 | drba->drba_snapobj = 0; |
428870ff | 1649 | } |
34dc7c2f | 1650 | |
34dc7c2f | 1651 | return (0); |
13fe0198 MA |
1652 | |
1653 | } | |
1654 | ||
1655 | static int | |
1656 | dmu_recv_begin_check(void *arg, dmu_tx_t *tx) | |
1657 | { | |
1658 | dmu_recv_begin_arg_t *drba = arg; | |
1659 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1660 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
1661 | uint64_t fromguid = drrb->drr_fromguid; | |
1662 | int flags = drrb->drr_flags; | |
b5256303 | 1663 | ds_hold_flags_t dsflags = 0; |
13fe0198 | 1664 | int error; |
9b67f605 | 1665 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); |
13fe0198 MA |
1666 | dsl_dataset_t *ds; |
1667 | const char *tofs = drba->drba_cookie->drc_tofs; | |
1668 | ||
1669 | /* already checked */ | |
1670 | ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); | |
47dfff3b | 1671 | ASSERT(!(featureflags & DMU_BACKUP_FEATURE_RESUMING)); |
13fe0198 MA |
1672 | |
1673 | if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == | |
1674 | DMU_COMPOUNDSTREAM || | |
1675 | drrb->drr_type >= DMU_OST_NUMTYPES || | |
1676 | ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) | |
2e528b49 | 1677 | return (SET_ERROR(EINVAL)); |
13fe0198 MA |
1678 | |
1679 | /* Verify pool version supports SA if SA_SPILL feature set */ | |
9b67f605 MA |
1680 | if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && |
1681 | spa_version(dp->dp_spa) < SPA_VERSION_SA) | |
1682 | return (SET_ERROR(ENOTSUP)); | |
1683 | ||
47dfff3b MA |
1684 | if (drba->drba_cookie->drc_resumable && |
1685 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EXTENSIBLE_DATASET)) | |
1686 | return (SET_ERROR(ENOTSUP)); | |
1687 | ||
9b67f605 MA |
1688 | /* |
1689 | * The receiving code doesn't know how to translate a WRITE_EMBEDDED | |
2aa34383 | 1690 | * record to a plain WRITE record, so the pool must have the |
9b67f605 MA |
1691 | * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED |
1692 | * records. Same with WRITE_EMBEDDED records that use LZ4 compression. | |
1693 | */ | |
1694 | if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && | |
1695 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) | |
1696 | return (SET_ERROR(ENOTSUP)); | |
2aa34383 | 1697 | if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && |
9b67f605 | 1698 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) |
2e528b49 | 1699 | return (SET_ERROR(ENOTSUP)); |
13fe0198 | 1700 | |
f1512ee6 MA |
1701 | /* |
1702 | * The receiving code doesn't know how to translate large blocks | |
1703 | * to smaller ones, so the pool must have the LARGE_BLOCKS | |
73aac4aa NB |
1704 | * feature enabled if the stream has LARGE_BLOCKS. Same with |
1705 | * large dnodes. | |
f1512ee6 MA |
1706 | */ |
1707 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
1708 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) | |
1709 | return (SET_ERROR(ENOTSUP)); | |
50c957f7 NB |
1710 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_DNODE) && |
1711 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_DNODE)) | |
1712 | return (SET_ERROR(ENOTSUP)); | |
1713 | ||
d9c460a0 | 1714 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { |
b5256303 TC |
1715 | /* raw receives require the encryption feature */ |
1716 | if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ENCRYPTION)) | |
1717 | return (SET_ERROR(ENOTSUP)); | |
1fff937a TC |
1718 | |
1719 | /* embedded data is incompatible with encryption and raw recv */ | |
1720 | if (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) | |
1721 | return (SET_ERROR(EINVAL)); | |
b5256303 TC |
1722 | } else { |
1723 | dsflags |= DS_HOLD_FLAG_DECRYPT; | |
1724 | } | |
1725 | ||
1726 | error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); | |
13fe0198 MA |
1727 | if (error == 0) { |
1728 | /* target fs already exists; recv into temp clone */ | |
1729 | ||
1730 | /* Can't recv a clone into an existing fs */ | |
e6d3a843 | 1731 | if (flags & DRR_FLAG_CLONE || drba->drba_origin) { |
b5256303 | 1732 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
2e528b49 | 1733 | return (SET_ERROR(EINVAL)); |
13fe0198 MA |
1734 | } |
1735 | ||
d9c460a0 TC |
1736 | error = recv_begin_check_existing_impl(drba, ds, fromguid, |
1737 | featureflags); | |
b5256303 | 1738 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
13fe0198 MA |
1739 | } else if (error == ENOENT) { |
1740 | /* target fs does not exist; must be a full backup or clone */ | |
eca7b760 | 1741 | char buf[ZFS_MAX_DATASET_NAME_LEN]; |
13fe0198 MA |
1742 | |
1743 | /* | |
1744 | * If it's a non-clone incremental, we are missing the | |
1745 | * target fs, so fail the recv. | |
1746 | */ | |
fcff0f35 PD |
1747 | if (fromguid != 0 && !(flags & DRR_FLAG_CLONE || |
1748 | drba->drba_origin)) | |
2e528b49 | 1749 | return (SET_ERROR(ENOENT)); |
13fe0198 | 1750 | |
e6d3a843 PD |
1751 | /* |
1752 | * If we're receiving a full send as a clone, and it doesn't | |
1753 | * contain all the necessary free records and freeobject | |
1754 | * records, reject it. | |
1755 | */ | |
1756 | if (fromguid == 0 && drba->drba_origin && | |
1757 | !(flags & DRR_FLAG_FREERECORDS)) | |
1758 | return (SET_ERROR(EINVAL)); | |
1759 | ||
13fe0198 | 1760 | /* Open the parent of tofs */ |
eca7b760 | 1761 | ASSERT3U(strlen(tofs), <, sizeof (buf)); |
13fe0198 | 1762 | (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); |
b5256303 | 1763 | error = dsl_dataset_hold_flags(dp, buf, dsflags, FTAG, &ds); |
13fe0198 MA |
1764 | if (error != 0) |
1765 | return (error); | |
1766 | ||
d9c460a0 TC |
1767 | if ((featureflags & DMU_BACKUP_FEATURE_RAW) == 0 && |
1768 | drba->drba_origin == NULL) { | |
1fff937a TC |
1769 | boolean_t will_encrypt; |
1770 | ||
1771 | /* | |
1772 | * Check that we aren't breaking any encryption rules | |
1773 | * and that we have all the parameters we need to | |
1774 | * create an encrypted dataset if necessary. If we are | |
1775 | * making an encrypted dataset the stream can't have | |
1776 | * embedded data. | |
1777 | */ | |
d9c460a0 | 1778 | error = dmu_objset_create_crypt_check(ds->ds_dir, |
1fff937a | 1779 | drba->drba_dcp, &will_encrypt); |
d9c460a0 TC |
1780 | if (error != 0) { |
1781 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1782 | return (error); | |
1783 | } | |
1fff937a TC |
1784 | |
1785 | if (will_encrypt && | |
1786 | (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)) { | |
1787 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1788 | return (SET_ERROR(EINVAL)); | |
1789 | } | |
d9c460a0 TC |
1790 | } |
1791 | ||
788eb90c JJ |
1792 | /* |
1793 | * Check filesystem and snapshot limits before receiving. We'll | |
1794 | * recheck snapshot limits again at the end (we create the | |
1795 | * filesystems and increment those counts during begin_sync). | |
1796 | */ | |
1797 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, | |
1798 | ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred); | |
1799 | if (error != 0) { | |
b5256303 | 1800 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
788eb90c JJ |
1801 | return (error); |
1802 | } | |
1803 | ||
1804 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, | |
1805 | ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); | |
1806 | if (error != 0) { | |
b5256303 | 1807 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
788eb90c JJ |
1808 | return (error); |
1809 | } | |
1810 | ||
13fe0198 MA |
1811 | if (drba->drba_origin != NULL) { |
1812 | dsl_dataset_t *origin; | |
b5256303 TC |
1813 | |
1814 | error = dsl_dataset_hold_flags(dp, drba->drba_origin, | |
1815 | dsflags, FTAG, &origin); | |
13fe0198 | 1816 | if (error != 0) { |
b5256303 | 1817 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
13fe0198 MA |
1818 | return (error); |
1819 | } | |
0c66c32d | 1820 | if (!origin->ds_is_snapshot) { |
9b840763 | 1821 | dsl_dataset_rele_flags(origin, dsflags, FTAG); |
b5256303 | 1822 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
2e528b49 | 1823 | return (SET_ERROR(EINVAL)); |
13fe0198 | 1824 | } |
e6d3a843 PD |
1825 | if (dsl_dataset_phys(origin)->ds_guid != fromguid && |
1826 | fromguid != 0) { | |
9b840763 | 1827 | dsl_dataset_rele_flags(origin, dsflags, FTAG); |
b5256303 | 1828 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
2e528b49 | 1829 | return (SET_ERROR(ENODEV)); |
13fe0198 | 1830 | } |
1fff937a TC |
1831 | if (origin->ds_dir->dd_crypto_obj != 0 && |
1832 | (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)) { | |
1833 | dsl_dataset_rele_flags(origin, dsflags, FTAG); | |
1834 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1835 | return (SET_ERROR(EINVAL)); | |
1836 | } | |
b5256303 TC |
1837 | dsl_dataset_rele_flags(origin, |
1838 | dsflags, FTAG); | |
13fe0198 | 1839 | } |
b5256303 | 1840 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
13fe0198 MA |
1841 | error = 0; |
1842 | } | |
1843 | return (error); | |
34dc7c2f BB |
1844 | } |
1845 | ||
34dc7c2f | 1846 | static void |
13fe0198 | 1847 | dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1848 | { |
13fe0198 MA |
1849 | dmu_recv_begin_arg_t *drba = arg; |
1850 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
47dfff3b | 1851 | objset_t *mos = dp->dp_meta_objset; |
13fe0198 MA |
1852 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; |
1853 | const char *tofs = drba->drba_cookie->drc_tofs; | |
b5256303 | 1854 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); |
13fe0198 | 1855 | dsl_dataset_t *ds, *newds; |
b5256303 | 1856 | objset_t *os; |
34dc7c2f | 1857 | uint64_t dsobj; |
b5256303 | 1858 | ds_hold_flags_t dsflags = 0; |
13fe0198 | 1859 | int error; |
47dfff3b | 1860 | uint64_t crflags = 0; |
d9c460a0 TC |
1861 | dsl_crypto_params_t dummy_dcp = { 0 }; |
1862 | dsl_crypto_params_t *dcp = drba->drba_dcp; | |
13fe0198 | 1863 | |
47dfff3b MA |
1864 | if (drrb->drr_flags & DRR_FLAG_CI_DATA) |
1865 | crflags |= DS_FLAG_CI_DATASET; | |
d9c460a0 TC |
1866 | |
1867 | if ((featureflags & DMU_BACKUP_FEATURE_RAW) == 0) | |
b5256303 | 1868 | dsflags |= DS_HOLD_FLAG_DECRYPT; |
d9c460a0 TC |
1869 | |
1870 | /* | |
1871 | * Raw, non-incremental recvs always use a dummy dcp with | |
1872 | * the raw cmd set. Raw incremental recvs do not use a dcp | |
1873 | * since the encryption parameters are already set in stone. | |
1874 | */ | |
1875 | if (dcp == NULL && drba->drba_snapobj == 0 && | |
1876 | drba->drba_origin == NULL) { | |
1877 | ASSERT3P(dcp, ==, NULL); | |
1878 | dcp = &dummy_dcp; | |
1879 | ||
1880 | if (featureflags & DMU_BACKUP_FEATURE_RAW) | |
1881 | dcp->cp_cmd = DCP_CMD_RAW_RECV; | |
b5256303 | 1882 | } |
34dc7c2f | 1883 | |
b5256303 | 1884 | error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); |
13fe0198 MA |
1885 | if (error == 0) { |
1886 | /* create temporary clone */ | |
19580676 | 1887 | dsl_dataset_t *snap = NULL; |
b5256303 | 1888 | |
19580676 MA |
1889 | if (drba->drba_snapobj != 0) { |
1890 | VERIFY0(dsl_dataset_hold_obj(dp, | |
1891 | drba->drba_snapobj, FTAG, &snap)); | |
d9c460a0 | 1892 | ASSERT3P(dcp, ==, NULL); |
19580676 | 1893 | } |
b5256303 | 1894 | |
13fe0198 | 1895 | dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, |
d9c460a0 | 1896 | snap, crflags, drba->drba_cred, dcp, tx); |
6b42ea85 PD |
1897 | if (drba->drba_snapobj != 0) |
1898 | dsl_dataset_rele(snap, FTAG); | |
b5256303 | 1899 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
13fe0198 MA |
1900 | } else { |
1901 | dsl_dir_t *dd; | |
1902 | const char *tail; | |
1903 | dsl_dataset_t *origin = NULL; | |
1904 | ||
1905 | VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); | |
1906 | ||
1907 | if (drba->drba_origin != NULL) { | |
1908 | VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, | |
1909 | FTAG, &origin)); | |
d9c460a0 | 1910 | ASSERT3P(dcp, ==, NULL); |
13fe0198 MA |
1911 | } |
1912 | ||
1913 | /* Create new dataset. */ | |
b5256303 | 1914 | dsobj = dsl_dataset_create_sync(dd, strrchr(tofs, '/') + 1, |
d9c460a0 | 1915 | origin, crflags, drba->drba_cred, dcp, tx); |
13fe0198 MA |
1916 | if (origin != NULL) |
1917 | dsl_dataset_rele(origin, FTAG); | |
1918 | dsl_dir_rele(dd, FTAG); | |
1919 | drba->drba_cookie->drc_newfs = B_TRUE; | |
1920 | } | |
d9c460a0 | 1921 | |
b5256303 TC |
1922 | VERIFY0(dsl_dataset_own_obj(dp, dsobj, dsflags, dmu_recv_tag, &newds)); |
1923 | VERIFY0(dmu_objset_from_ds(newds, &os)); | |
13fe0198 | 1924 | |
47dfff3b | 1925 | if (drba->drba_cookie->drc_resumable) { |
47dfff3b MA |
1926 | dsl_dataset_zapify(newds, tx); |
1927 | if (drrb->drr_fromguid != 0) { | |
1928 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_FROMGUID, | |
1929 | 8, 1, &drrb->drr_fromguid, tx)); | |
1930 | } | |
1931 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TOGUID, | |
1932 | 8, 1, &drrb->drr_toguid, tx)); | |
1933 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TONAME, | |
1934 | 1, strlen(drrb->drr_toname) + 1, drrb->drr_toname, tx)); | |
1c27024e DB |
1935 | uint64_t one = 1; |
1936 | uint64_t zero = 0; | |
47dfff3b MA |
1937 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OBJECT, |
1938 | 8, 1, &one, tx)); | |
1939 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OFFSET, | |
1940 | 8, 1, &zero, tx)); | |
1941 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_BYTES, | |
1942 | 8, 1, &zero, tx)); | |
b5256303 | 1943 | if (featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) { |
2aa34383 | 1944 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_LARGEBLOCK, |
02730c33 | 1945 | 8, 1, &one, tx)); |
2aa34383 | 1946 | } |
b5256303 | 1947 | if (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) { |
47dfff3b MA |
1948 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_EMBEDOK, |
1949 | 8, 1, &one, tx)); | |
1950 | } | |
b5256303 | 1951 | if (featureflags & DMU_BACKUP_FEATURE_COMPRESSED) { |
2aa34383 DK |
1952 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_COMPRESSOK, |
1953 | 8, 1, &one, tx)); | |
1954 | } | |
b5256303 TC |
1955 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { |
1956 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_RAWOK, | |
1957 | 8, 1, &one, tx)); | |
1958 | } | |
1959 | } | |
1960 | ||
1961 | /* | |
1962 | * Usually the os->os_encrypted value is tied to the presence of a | |
1963 | * DSL Crypto Key object in the dd. However, that will not be received | |
1964 | * until dmu_recv_stream(), so we set the value manually for now. | |
1965 | */ | |
1966 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
1967 | os->os_encrypted = B_TRUE; | |
1968 | drba->drba_cookie->drc_raw = B_TRUE; | |
47dfff3b MA |
1969 | } |
1970 | ||
13fe0198 | 1971 | dmu_buf_will_dirty(newds->ds_dbuf, tx); |
d683ddbb | 1972 | dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; |
34dc7c2f | 1973 | |
428870ff | 1974 | /* |
b5256303 TC |
1975 | * If we actually created a non-clone, we need to create the objset |
1976 | * in our new dataset. If this is a raw send we postpone this until | |
1977 | * dmu_recv_stream() so that we can allocate the metadnode with the | |
1978 | * properties from the DRR_BEGIN payload. | |
428870ff | 1979 | */ |
cc9bb3e5 | 1980 | rrw_enter(&newds->ds_bp_rwlock, RW_READER, FTAG); |
b5256303 TC |
1981 | if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds)) && |
1982 | (featureflags & DMU_BACKUP_FEATURE_RAW) == 0) { | |
428870ff | 1983 | (void) dmu_objset_create_impl(dp->dp_spa, |
13fe0198 | 1984 | newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); |
428870ff | 1985 | } |
cc9bb3e5 | 1986 | rrw_exit(&newds->ds_bp_rwlock, FTAG); |
34dc7c2f | 1987 | |
13fe0198 | 1988 | drba->drba_cookie->drc_ds = newds; |
428870ff | 1989 | |
13fe0198 | 1990 | spa_history_log_internal_ds(newds, "receive", tx, ""); |
34dc7c2f BB |
1991 | } |
1992 | ||
47dfff3b MA |
1993 | static int |
1994 | dmu_recv_resume_begin_check(void *arg, dmu_tx_t *tx) | |
1995 | { | |
1996 | dmu_recv_begin_arg_t *drba = arg; | |
1997 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1998 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
1999 | int error; | |
b5256303 | 2000 | ds_hold_flags_t dsflags = 0; |
47dfff3b MA |
2001 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); |
2002 | dsl_dataset_t *ds; | |
2003 | const char *tofs = drba->drba_cookie->drc_tofs; | |
eca7b760 | 2004 | |
47dfff3b MA |
2005 | /* already checked */ |
2006 | ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); | |
2007 | ASSERT(featureflags & DMU_BACKUP_FEATURE_RESUMING); | |
2008 | ||
2009 | if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == | |
2010 | DMU_COMPOUNDSTREAM || | |
2011 | drrb->drr_type >= DMU_OST_NUMTYPES) | |
2012 | return (SET_ERROR(EINVAL)); | |
2013 | ||
2014 | /* Verify pool version supports SA if SA_SPILL feature set */ | |
2015 | if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && | |
2016 | spa_version(dp->dp_spa) < SPA_VERSION_SA) | |
2017 | return (SET_ERROR(ENOTSUP)); | |
2018 | ||
2019 | /* | |
2020 | * The receiving code doesn't know how to translate a WRITE_EMBEDDED | |
2021 | * record to a plain WRITE record, so the pool must have the | |
2022 | * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED | |
2023 | * records. Same with WRITE_EMBEDDED records that use LZ4 compression. | |
2024 | */ | |
2025 | if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && | |
2026 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) | |
2027 | return (SET_ERROR(ENOTSUP)); | |
2aa34383 | 2028 | if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && |
47dfff3b MA |
2029 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) |
2030 | return (SET_ERROR(ENOTSUP)); | |
2031 | ||
73aac4aa NB |
2032 | /* |
2033 | * The receiving code doesn't know how to translate large blocks | |
2034 | * to smaller ones, so the pool must have the LARGE_BLOCKS | |
2035 | * feature enabled if the stream has LARGE_BLOCKS. Same with | |
2036 | * large dnodes. | |
2037 | */ | |
2038 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
2039 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) | |
2040 | return (SET_ERROR(ENOTSUP)); | |
2041 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_DNODE) && | |
2042 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_DNODE)) | |
2043 | return (SET_ERROR(ENOTSUP)); | |
2044 | ||
1c27024e DB |
2045 | /* 6 extra bytes for /%recv */ |
2046 | char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; | |
47dfff3b MA |
2047 | (void) snprintf(recvname, sizeof (recvname), "%s/%s", |
2048 | tofs, recv_clone_name); | |
2049 | ||
b5256303 TC |
2050 | if ((featureflags & DMU_BACKUP_FEATURE_RAW) == 0) |
2051 | dsflags |= DS_HOLD_FLAG_DECRYPT; | |
2052 | ||
2053 | if (dsl_dataset_hold_flags(dp, recvname, dsflags, FTAG, &ds) != 0) { | |
47dfff3b | 2054 | /* %recv does not exist; continue in tofs */ |
b5256303 | 2055 | error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); |
47dfff3b MA |
2056 | if (error != 0) |
2057 | return (error); | |
2058 | } | |
2059 | ||
2060 | /* check that ds is marked inconsistent */ | |
2061 | if (!DS_IS_INCONSISTENT(ds)) { | |
b5256303 | 2062 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
47dfff3b MA |
2063 | return (SET_ERROR(EINVAL)); |
2064 | } | |
2065 | ||
2066 | /* check that there is resuming data, and that the toguid matches */ | |
2067 | if (!dsl_dataset_is_zapified(ds)) { | |
b5256303 | 2068 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
47dfff3b MA |
2069 | return (SET_ERROR(EINVAL)); |
2070 | } | |
1c27024e | 2071 | uint64_t val; |
47dfff3b MA |
2072 | error = zap_lookup(dp->dp_meta_objset, ds->ds_object, |
2073 | DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val); | |
2074 | if (error != 0 || drrb->drr_toguid != val) { | |
b5256303 | 2075 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
47dfff3b MA |
2076 | return (SET_ERROR(EINVAL)); |
2077 | } | |
2078 | ||
2079 | /* | |
2080 | * Check if the receive is still running. If so, it will be owned. | |
2081 | * Note that nothing else can own the dataset (e.g. after the receive | |
2082 | * fails) because it will be marked inconsistent. | |
2083 | */ | |
2084 | if (dsl_dataset_has_owner(ds)) { | |
b5256303 | 2085 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
47dfff3b MA |
2086 | return (SET_ERROR(EBUSY)); |
2087 | } | |
2088 | ||
2089 | /* There should not be any snapshots of this fs yet. */ | |
2090 | if (ds->ds_prev != NULL && ds->ds_prev->ds_dir == ds->ds_dir) { | |
b5256303 | 2091 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
47dfff3b MA |
2092 | return (SET_ERROR(EINVAL)); |
2093 | } | |
2094 | ||
2095 | /* | |
2096 | * Note: resume point will be checked when we process the first WRITE | |
2097 | * record. | |
2098 | */ | |
2099 | ||
2100 | /* check that the origin matches */ | |
2101 | val = 0; | |
2102 | (void) zap_lookup(dp->dp_meta_objset, ds->ds_object, | |
2103 | DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val); | |
2104 | if (drrb->drr_fromguid != val) { | |
b5256303 | 2105 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
47dfff3b MA |
2106 | return (SET_ERROR(EINVAL)); |
2107 | } | |
2108 | ||
b5256303 | 2109 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
47dfff3b MA |
2110 | return (0); |
2111 | } | |
2112 | ||
2113 | static void | |
2114 | dmu_recv_resume_begin_sync(void *arg, dmu_tx_t *tx) | |
2115 | { | |
2116 | dmu_recv_begin_arg_t *drba = arg; | |
2117 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
2118 | const char *tofs = drba->drba_cookie->drc_tofs; | |
b5256303 TC |
2119 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; |
2120 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); | |
47dfff3b | 2121 | dsl_dataset_t *ds; |
b5256303 TC |
2122 | objset_t *os; |
2123 | ds_hold_flags_t dsflags = 0; | |
47dfff3b | 2124 | uint64_t dsobj; |
eca7b760 IK |
2125 | /* 6 extra bytes for /%recv */ |
2126 | char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; | |
47dfff3b MA |
2127 | |
2128 | (void) snprintf(recvname, sizeof (recvname), "%s/%s", | |
2129 | tofs, recv_clone_name); | |
2130 | ||
b5256303 TC |
2131 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { |
2132 | drba->drba_cookie->drc_raw = B_TRUE; | |
2133 | } else { | |
2134 | dsflags |= DS_HOLD_FLAG_DECRYPT; | |
2135 | } | |
2136 | ||
2137 | if (dsl_dataset_hold_flags(dp, recvname, dsflags, FTAG, &ds) != 0) { | |
47dfff3b | 2138 | /* %recv does not exist; continue in tofs */ |
b5256303 | 2139 | VERIFY0(dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds)); |
47dfff3b MA |
2140 | drba->drba_cookie->drc_newfs = B_TRUE; |
2141 | } | |
2142 | ||
2143 | /* clear the inconsistent flag so that we can own it */ | |
2144 | ASSERT(DS_IS_INCONSISTENT(ds)); | |
2145 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
2146 | dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; | |
2147 | dsobj = ds->ds_object; | |
b5256303 | 2148 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
47dfff3b | 2149 | |
b5256303 TC |
2150 | VERIFY0(dsl_dataset_own_obj(dp, dsobj, dsflags, dmu_recv_tag, &ds)); |
2151 | VERIFY0(dmu_objset_from_ds(ds, &os)); | |
47dfff3b MA |
2152 | |
2153 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
2154 | dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT; | |
2155 | ||
cc9bb3e5 | 2156 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
2637dda8 TC |
2157 | ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds)) || |
2158 | drba->drba_cookie->drc_raw); | |
cc9bb3e5 | 2159 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
47dfff3b MA |
2160 | |
2161 | drba->drba_cookie->drc_ds = ds; | |
2162 | ||
2163 | spa_history_log_internal_ds(ds, "resume receive", tx, ""); | |
2164 | } | |
2165 | ||
34dc7c2f BB |
2166 | /* |
2167 | * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() | |
2168 | * succeeds; otherwise we will leak the holds on the datasets. | |
2169 | */ | |
2170 | int | |
47dfff3b | 2171 | dmu_recv_begin(char *tofs, char *tosnap, dmu_replay_record_t *drr_begin, |
d9c460a0 TC |
2172 | boolean_t force, boolean_t resumable, nvlist_t *localprops, |
2173 | nvlist_t *hidden_args, char *origin, dmu_recv_cookie_t *drc) | |
34dc7c2f | 2174 | { |
13fe0198 | 2175 | dmu_recv_begin_arg_t drba = { 0 }; |
34dc7c2f BB |
2176 | |
2177 | bzero(drc, sizeof (dmu_recv_cookie_t)); | |
47dfff3b MA |
2178 | drc->drc_drr_begin = drr_begin; |
2179 | drc->drc_drrb = &drr_begin->drr_u.drr_begin; | |
34dc7c2f | 2180 | drc->drc_tosnap = tosnap; |
13fe0198 | 2181 | drc->drc_tofs = tofs; |
34dc7c2f | 2182 | drc->drc_force = force; |
47dfff3b | 2183 | drc->drc_resumable = resumable; |
788eb90c | 2184 | drc->drc_cred = CRED(); |
48fbb9dd | 2185 | drc->drc_clone = (origin != NULL); |
34dc7c2f | 2186 | |
47dfff3b | 2187 | if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { |
13fe0198 | 2188 | drc->drc_byteswap = B_TRUE; |
a6255b7f | 2189 | (void) fletcher_4_incremental_byteswap(drr_begin, |
13fe0198 | 2190 | sizeof (dmu_replay_record_t), &drc->drc_cksum); |
47dfff3b MA |
2191 | byteswap_record(drr_begin); |
2192 | } else if (drc->drc_drrb->drr_magic == DMU_BACKUP_MAGIC) { | |
a6255b7f | 2193 | (void) fletcher_4_incremental_native(drr_begin, |
13fe0198 | 2194 | sizeof (dmu_replay_record_t), &drc->drc_cksum); |
47dfff3b MA |
2195 | } else { |
2196 | return (SET_ERROR(EINVAL)); | |
34dc7c2f BB |
2197 | } |
2198 | ||
13fe0198 MA |
2199 | drba.drba_origin = origin; |
2200 | drba.drba_cookie = drc; | |
2201 | drba.drba_cred = CRED(); | |
2202 | ||
47dfff3b MA |
2203 | if (DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & |
2204 | DMU_BACKUP_FEATURE_RESUMING) { | |
2205 | return (dsl_sync_task(tofs, | |
2206 | dmu_recv_resume_begin_check, dmu_recv_resume_begin_sync, | |
2207 | &drba, 5, ZFS_SPACE_CHECK_NORMAL)); | |
2208 | } else { | |
d9c460a0 TC |
2209 | int err; |
2210 | ||
2211 | /* | |
2212 | * For non-raw, non-incremental, non-resuming receives the | |
2213 | * user can specify encryption parameters on the command line | |
2214 | * with "zfs recv -o". For these receives we create a dcp and | |
2215 | * pass it to the sync task. Creating the dcp will implicitly | |
2216 | * remove the encryption params from the localprops nvlist, | |
2217 | * which avoids errors when trying to set these normally | |
2218 | * read-only properties. Any other kind of receive that | |
2219 | * attempts to set these properties will fail as a result. | |
2220 | */ | |
2221 | if ((DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & | |
2222 | DMU_BACKUP_FEATURE_RAW) == 0 && | |
2223 | origin == NULL && drc->drc_drrb->drr_fromguid == 0) { | |
2224 | err = dsl_crypto_params_create_nvlist(DCP_CMD_NONE, | |
2225 | localprops, hidden_args, &drba.drba_dcp); | |
2226 | if (err != 0) | |
2227 | return (err); | |
2228 | } | |
2229 | ||
2230 | err = dsl_sync_task(tofs, | |
47dfff3b | 2231 | dmu_recv_begin_check, dmu_recv_begin_sync, |
d9c460a0 TC |
2232 | &drba, 5, ZFS_SPACE_CHECK_NORMAL); |
2233 | dsl_crypto_params_free(drba.drba_dcp, !!err); | |
2234 | ||
2235 | return (err); | |
47dfff3b | 2236 | } |
34dc7c2f BB |
2237 | } |
2238 | ||
fcff0f35 PD |
2239 | struct receive_record_arg { |
2240 | dmu_replay_record_t header; | |
2241 | void *payload; /* Pointer to a buffer containing the payload */ | |
2242 | /* | |
2243 | * If the record is a write, pointer to the arc_buf_t containing the | |
2244 | * payload. | |
2245 | */ | |
b5256303 | 2246 | arc_buf_t *arc_buf; |
fcff0f35 | 2247 | int payload_size; |
47dfff3b | 2248 | uint64_t bytes_read; /* bytes read from stream when record created */ |
fcff0f35 PD |
2249 | boolean_t eos_marker; /* Marks the end of the stream */ |
2250 | bqueue_node_t node; | |
2251 | }; | |
2252 | ||
2253 | struct receive_writer_arg { | |
37f8a883 | 2254 | objset_t *os; |
13fe0198 | 2255 | boolean_t byteswap; |
fcff0f35 | 2256 | bqueue_t q; |
47dfff3b | 2257 | |
fcff0f35 PD |
2258 | /* |
2259 | * These three args are used to signal to the main thread that we're | |
2260 | * done. | |
2261 | */ | |
2262 | kmutex_t mutex; | |
2263 | kcondvar_t cv; | |
2264 | boolean_t done; | |
47dfff3b | 2265 | |
fcff0f35 PD |
2266 | int err; |
2267 | /* A map from guid to dataset to help handle dedup'd streams. */ | |
2268 | avl_tree_t *guid_to_ds_map; | |
47dfff3b | 2269 | boolean_t resumable; |
9b840763 | 2270 | boolean_t raw; |
48fbb9dd FG |
2271 | uint64_t last_object; |
2272 | uint64_t last_offset; | |
2273 | uint64_t max_object; /* highest object ID referenced in stream */ | |
47dfff3b | 2274 | uint64_t bytes_read; /* bytes read when current record created */ |
095495e0 TC |
2275 | |
2276 | /* Encryption parameters for the last received DRR_OBJECT_RANGE */ | |
0c03d21a | 2277 | boolean_t or_crypt_params_present; |
095495e0 TC |
2278 | uint64_t or_firstobj; |
2279 | uint64_t or_numslots; | |
2280 | uint8_t or_salt[ZIO_DATA_SALT_LEN]; | |
2281 | uint8_t or_iv[ZIO_DATA_IV_LEN]; | |
2282 | uint8_t or_mac[ZIO_DATA_MAC_LEN]; | |
2283 | boolean_t or_byteorder; | |
fcff0f35 | 2284 | }; |
37f8a883 | 2285 | |
e6d3a843 PD |
2286 | struct objlist { |
2287 | list_t list; /* List of struct receive_objnode. */ | |
2288 | /* | |
2289 | * Last object looked up. Used to assert that objects are being looked | |
2290 | * up in ascending order. | |
2291 | */ | |
2292 | uint64_t last_lookup; | |
2293 | }; | |
2294 | ||
2295 | struct receive_objnode { | |
2296 | list_node_t node; | |
2297 | uint64_t object; | |
2298 | }; | |
2299 | ||
fcff0f35 PD |
2300 | struct receive_arg { |
2301 | objset_t *os; | |
2302 | vnode_t *vp; /* The vnode to read the stream from */ | |
2303 | uint64_t voff; /* The current offset in the stream */ | |
47dfff3b | 2304 | uint64_t bytes_read; |
fcff0f35 PD |
2305 | /* |
2306 | * A record that has had its payload read in, but hasn't yet been handed | |
2307 | * off to the worker thread. | |
2308 | */ | |
2309 | struct receive_record_arg *rrd; | |
2310 | /* A record that has had its header read in, but not its payload. */ | |
2311 | struct receive_record_arg *next_rrd; | |
34dc7c2f | 2312 | zio_cksum_t cksum; |
37f8a883 | 2313 | zio_cksum_t prev_cksum; |
fcff0f35 PD |
2314 | int err; |
2315 | boolean_t byteswap; | |
9b840763 | 2316 | boolean_t raw; |
b5256303 | 2317 | uint64_t featureflags; |
fcff0f35 | 2318 | /* Sorted list of objects not to issue prefetches for. */ |
e6d3a843 | 2319 | struct objlist ignore_objlist; |
34dc7c2f BB |
2320 | }; |
2321 | ||
428870ff BB |
2322 | typedef struct guid_map_entry { |
2323 | uint64_t guid; | |
b5256303 | 2324 | boolean_t raw; |
428870ff BB |
2325 | dsl_dataset_t *gme_ds; |
2326 | avl_node_t avlnode; | |
2327 | } guid_map_entry_t; | |
2328 | ||
2329 | static int | |
2330 | guid_compare(const void *arg1, const void *arg2) | |
2331 | { | |
ee36c709 GN |
2332 | const guid_map_entry_t *gmep1 = (const guid_map_entry_t *)arg1; |
2333 | const guid_map_entry_t *gmep2 = (const guid_map_entry_t *)arg2; | |
428870ff | 2334 | |
ee36c709 | 2335 | return (AVL_CMP(gmep1->guid, gmep2->guid)); |
428870ff BB |
2336 | } |
2337 | ||
572e2857 BB |
2338 | static void |
2339 | free_guid_map_onexit(void *arg) | |
2340 | { | |
2341 | avl_tree_t *ca = arg; | |
2342 | void *cookie = NULL; | |
2343 | guid_map_entry_t *gmep; | |
2344 | ||
2345 | while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { | |
149ce888 TC |
2346 | ds_hold_flags_t dsflags = DS_HOLD_FLAG_DECRYPT; |
2347 | ||
2348 | if (gmep->raw) { | |
2349 | gmep->gme_ds->ds_objset->os_raw_receive = B_FALSE; | |
2350 | dsflags &= ~DS_HOLD_FLAG_DECRYPT; | |
2351 | } | |
2352 | ||
2353 | dsl_dataset_disown(gmep->gme_ds, dsflags, gmep); | |
572e2857 BB |
2354 | kmem_free(gmep, sizeof (guid_map_entry_t)); |
2355 | } | |
2356 | avl_destroy(ca); | |
2357 | kmem_free(ca, sizeof (avl_tree_t)); | |
2358 | } | |
2359 | ||
37f8a883 | 2360 | static int |
fcff0f35 | 2361 | receive_read(struct receive_arg *ra, int len, void *buf) |
34dc7c2f | 2362 | { |
34dc7c2f BB |
2363 | int done = 0; |
2364 | ||
f8866f8a ER |
2365 | /* |
2366 | * The code doesn't rely on this (lengths being multiples of 8). See | |
2367 | * comment in dump_bytes. | |
2368 | */ | |
b5256303 TC |
2369 | ASSERT(len % 8 == 0 || |
2370 | (ra->featureflags & DMU_BACKUP_FEATURE_RAW) != 0); | |
34dc7c2f BB |
2371 | |
2372 | while (done < len) { | |
2373 | ssize_t resid; | |
2374 | ||
2375 | ra->err = vn_rdwr(UIO_READ, ra->vp, | |
37f8a883 | 2376 | (char *)buf + done, len - done, |
34dc7c2f BB |
2377 | ra->voff, UIO_SYSSPACE, FAPPEND, |
2378 | RLIM64_INFINITY, CRED(), &resid); | |
2379 | ||
47dfff3b MA |
2380 | if (resid == len - done) { |
2381 | /* | |
2382 | * Note: ECKSUM indicates that the receive | |
2383 | * was interrupted and can potentially be resumed. | |
2384 | */ | |
2385 | ra->err = SET_ERROR(ECKSUM); | |
2386 | } | |
34dc7c2f BB |
2387 | ra->voff += len - done - resid; |
2388 | done = len - resid; | |
13fe0198 | 2389 | if (ra->err != 0) |
37f8a883 | 2390 | return (ra->err); |
34dc7c2f BB |
2391 | } |
2392 | ||
47dfff3b MA |
2393 | ra->bytes_read += len; |
2394 | ||
34dc7c2f | 2395 | ASSERT3U(done, ==, len); |
37f8a883 | 2396 | return (0); |
34dc7c2f BB |
2397 | } |
2398 | ||
60948de1 | 2399 | noinline static void |
37f8a883 | 2400 | byteswap_record(dmu_replay_record_t *drr) |
34dc7c2f BB |
2401 | { |
2402 | #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) | |
2403 | #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) | |
2404 | drr->drr_type = BSWAP_32(drr->drr_type); | |
2405 | drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); | |
37f8a883 | 2406 | |
34dc7c2f BB |
2407 | switch (drr->drr_type) { |
2408 | case DRR_BEGIN: | |
2409 | DO64(drr_begin.drr_magic); | |
428870ff | 2410 | DO64(drr_begin.drr_versioninfo); |
34dc7c2f BB |
2411 | DO64(drr_begin.drr_creation_time); |
2412 | DO32(drr_begin.drr_type); | |
2413 | DO32(drr_begin.drr_flags); | |
2414 | DO64(drr_begin.drr_toguid); | |
2415 | DO64(drr_begin.drr_fromguid); | |
2416 | break; | |
2417 | case DRR_OBJECT: | |
2418 | DO64(drr_object.drr_object); | |
34dc7c2f BB |
2419 | DO32(drr_object.drr_type); |
2420 | DO32(drr_object.drr_bonustype); | |
2421 | DO32(drr_object.drr_blksz); | |
2422 | DO32(drr_object.drr_bonuslen); | |
b5256303 | 2423 | DO32(drr_object.drr_raw_bonuslen); |
428870ff | 2424 | DO64(drr_object.drr_toguid); |
ae76f45c | 2425 | DO64(drr_object.drr_maxblkid); |
34dc7c2f BB |
2426 | break; |
2427 | case DRR_FREEOBJECTS: | |
2428 | DO64(drr_freeobjects.drr_firstobj); | |
2429 | DO64(drr_freeobjects.drr_numobjs); | |
428870ff | 2430 | DO64(drr_freeobjects.drr_toguid); |
34dc7c2f BB |
2431 | break; |
2432 | case DRR_WRITE: | |
2433 | DO64(drr_write.drr_object); | |
2434 | DO32(drr_write.drr_type); | |
2435 | DO64(drr_write.drr_offset); | |
2aa34383 | 2436 | DO64(drr_write.drr_logical_size); |
428870ff | 2437 | DO64(drr_write.drr_toguid); |
37f8a883 | 2438 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write.drr_key.ddk_cksum); |
428870ff | 2439 | DO64(drr_write.drr_key.ddk_prop); |
2aa34383 | 2440 | DO64(drr_write.drr_compressed_size); |
428870ff BB |
2441 | break; |
2442 | case DRR_WRITE_BYREF: | |
2443 | DO64(drr_write_byref.drr_object); | |
2444 | DO64(drr_write_byref.drr_offset); | |
2445 | DO64(drr_write_byref.drr_length); | |
2446 | DO64(drr_write_byref.drr_toguid); | |
2447 | DO64(drr_write_byref.drr_refguid); | |
2448 | DO64(drr_write_byref.drr_refobject); | |
2449 | DO64(drr_write_byref.drr_refoffset); | |
37f8a883 MA |
2450 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write_byref. |
2451 | drr_key.ddk_cksum); | |
428870ff | 2452 | DO64(drr_write_byref.drr_key.ddk_prop); |
34dc7c2f | 2453 | break; |
9b67f605 MA |
2454 | case DRR_WRITE_EMBEDDED: |
2455 | DO64(drr_write_embedded.drr_object); | |
2456 | DO64(drr_write_embedded.drr_offset); | |
2457 | DO64(drr_write_embedded.drr_length); | |
2458 | DO64(drr_write_embedded.drr_toguid); | |
2459 | DO32(drr_write_embedded.drr_lsize); | |
2460 | DO32(drr_write_embedded.drr_psize); | |
2461 | break; | |
34dc7c2f BB |
2462 | case DRR_FREE: |
2463 | DO64(drr_free.drr_object); | |
2464 | DO64(drr_free.drr_offset); | |
2465 | DO64(drr_free.drr_length); | |
428870ff BB |
2466 | DO64(drr_free.drr_toguid); |
2467 | break; | |
2468 | case DRR_SPILL: | |
2469 | DO64(drr_spill.drr_object); | |
2470 | DO64(drr_spill.drr_length); | |
2471 | DO64(drr_spill.drr_toguid); | |
b5256303 TC |
2472 | DO64(drr_spill.drr_compressed_size); |
2473 | DO32(drr_spill.drr_type); | |
2474 | break; | |
2475 | case DRR_OBJECT_RANGE: | |
2476 | DO64(drr_object_range.drr_firstobj); | |
2477 | DO64(drr_object_range.drr_numslots); | |
2478 | DO64(drr_object_range.drr_toguid); | |
34dc7c2f BB |
2479 | break; |
2480 | case DRR_END: | |
428870ff | 2481 | DO64(drr_end.drr_toguid); |
37f8a883 | 2482 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_end.drr_checksum); |
34dc7c2f | 2483 | break; |
e75c13c3 BB |
2484 | default: |
2485 | break; | |
34dc7c2f | 2486 | } |
37f8a883 MA |
2487 | |
2488 | if (drr->drr_type != DRR_BEGIN) { | |
2489 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_checksum.drr_checksum); | |
2490 | } | |
2491 | ||
34dc7c2f BB |
2492 | #undef DO64 |
2493 | #undef DO32 | |
2494 | } | |
2495 | ||
6c59307a MA |
2496 | static inline uint8_t |
2497 | deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) | |
2498 | { | |
2499 | if (bonus_type == DMU_OT_SA) { | |
2500 | return (1); | |
2501 | } else { | |
2502 | return (1 + | |
50c957f7 NB |
2503 | ((DN_OLD_MAX_BONUSLEN - |
2504 | MIN(DN_OLD_MAX_BONUSLEN, bonus_size)) >> SPA_BLKPTRSHIFT)); | |
6c59307a MA |
2505 | } |
2506 | } | |
2507 | ||
47dfff3b MA |
2508 | static void |
2509 | save_resume_state(struct receive_writer_arg *rwa, | |
2510 | uint64_t object, uint64_t offset, dmu_tx_t *tx) | |
2511 | { | |
2512 | int txgoff = dmu_tx_get_txg(tx) & TXG_MASK; | |
2513 | ||
2514 | if (!rwa->resumable) | |
2515 | return; | |
2516 | ||
2517 | /* | |
2518 | * We use ds_resume_bytes[] != 0 to indicate that we need to | |
2519 | * update this on disk, so it must not be 0. | |
2520 | */ | |
2521 | ASSERT(rwa->bytes_read != 0); | |
2522 | ||
2523 | /* | |
2524 | * We only resume from write records, which have a valid | |
2525 | * (non-meta-dnode) object number. | |
2526 | */ | |
2527 | ASSERT(object != 0); | |
2528 | ||
2529 | /* | |
2530 | * For resuming to work correctly, we must receive records in order, | |
2531 | * sorted by object,offset. This is checked by the callers, but | |
2532 | * assert it here for good measure. | |
2533 | */ | |
2534 | ASSERT3U(object, >=, rwa->os->os_dsl_dataset->ds_resume_object[txgoff]); | |
2535 | ASSERT(object != rwa->os->os_dsl_dataset->ds_resume_object[txgoff] || | |
2536 | offset >= rwa->os->os_dsl_dataset->ds_resume_offset[txgoff]); | |
2537 | ASSERT3U(rwa->bytes_read, >=, | |
2538 | rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff]); | |
2539 | ||
2540 | rwa->os->os_dsl_dataset->ds_resume_object[txgoff] = object; | |
2541 | rwa->os->os_dsl_dataset->ds_resume_offset[txgoff] = offset; | |
2542 | rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff] = rwa->bytes_read; | |
2543 | } | |
2544 | ||
60948de1 | 2545 | noinline static int |
fcff0f35 | 2546 | receive_object(struct receive_writer_arg *rwa, struct drr_object *drro, |
4ea3f864 | 2547 | void *data) |
34dc7c2f | 2548 | { |
6c59307a | 2549 | dmu_object_info_t doi; |
34dc7c2f | 2550 | dmu_tx_t *tx; |
6c59307a MA |
2551 | uint64_t object; |
2552 | int err; | |
da2feb42 TC |
2553 | uint8_t dn_slots = drro->drr_dn_slots != 0 ? |
2554 | drro->drr_dn_slots : DNODE_MIN_SLOTS; | |
34dc7c2f | 2555 | |
34dc7c2f | 2556 | if (drro->drr_type == DMU_OT_NONE || |
9ae529ec CS |
2557 | !DMU_OT_IS_VALID(drro->drr_type) || |
2558 | !DMU_OT_IS_VALID(drro->drr_bonustype) || | |
428870ff | 2559 | drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || |
34dc7c2f BB |
2560 | drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || |
2561 | P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || | |
2562 | drro->drr_blksz < SPA_MINBLOCKSIZE || | |
fcff0f35 | 2563 | drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(rwa->os)) || |
50c957f7 | 2564 | drro->drr_bonuslen > |
73aac4aa | 2565 | DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa->os))) || |
da2feb42 | 2566 | dn_slots > |
73aac4aa | 2567 | (spa_maxdnodesize(dmu_objset_spa(rwa->os)) >> DNODE_SHIFT)) { |
2e528b49 | 2568 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
2569 | } |
2570 | ||
9b840763 | 2571 | if (rwa->raw) { |
095495e0 TC |
2572 | /* |
2573 | * We should have received a DRR_OBJECT_RANGE record | |
2574 | * containing this block and stored it in rwa. | |
2575 | */ | |
2576 | if (drro->drr_object < rwa->or_firstobj || | |
2577 | drro->drr_object >= rwa->or_firstobj + rwa->or_numslots || | |
2578 | drro->drr_raw_bonuslen < drro->drr_bonuslen || | |
b5256303 TC |
2579 | drro->drr_indblkshift > SPA_MAXBLOCKSHIFT || |
2580 | drro->drr_nlevels > DN_MAX_LEVELS || | |
2581 | drro->drr_nblkptr > DN_MAX_NBLKPTR || | |
da2feb42 | 2582 | DN_SLOTS_TO_BONUSLEN(dn_slots) < |
b5256303 TC |
2583 | drro->drr_raw_bonuslen) |
2584 | return (SET_ERROR(EINVAL)); | |
2585 | } else { | |
2586 | if (drro->drr_flags != 0 || drro->drr_raw_bonuslen != 0 || | |
2587 | drro->drr_indblkshift != 0 || drro->drr_nlevels != 0 || | |
2588 | drro->drr_nblkptr != 0) | |
2589 | return (SET_ERROR(EINVAL)); | |
2590 | } | |
2591 | ||
fcff0f35 | 2592 | err = dmu_object_info(rwa->os, drro->drr_object, &doi); |
047116ac | 2593 | if (err != 0 && err != ENOENT && err != EEXIST) |
2e528b49 | 2594 | return (SET_ERROR(EINVAL)); |
9babb374 | 2595 | |
48fbb9dd FG |
2596 | if (drro->drr_object > rwa->max_object) |
2597 | rwa->max_object = drro->drr_object; | |
2598 | ||
6c59307a MA |
2599 | /* |
2600 | * If we are losing blkptrs or changing the block size this must | |
2601 | * be a new file instance. We must clear out the previous file | |
2602 | * contents before we can change this type of metadata in the dnode. | |
b5256303 TC |
2603 | * Raw receives will also check that the indirect structure of the |
2604 | * dnode hasn't changed. | |
6c59307a MA |
2605 | */ |
2606 | if (err == 0) { | |
b5256303 TC |
2607 | uint32_t indblksz = drro->drr_indblkshift ? |
2608 | 1ULL << drro->drr_indblkshift : 0; | |
2609 | int nblkptr = deduce_nblkptr(drro->drr_bonustype, | |
6c59307a MA |
2610 | drro->drr_bonuslen); |
2611 | ||
047116ac TC |
2612 | object = drro->drr_object; |
2613 | ||
b5256303 | 2614 | /* nblkptr will be bounded by the bonus size and type */ |
9b840763 | 2615 | if (rwa->raw && nblkptr != drro->drr_nblkptr) |
b5256303 TC |
2616 | return (SET_ERROR(EINVAL)); |
2617 | ||
0c03d21a | 2618 | if (drro->drr_blksz != doi.doi_data_block_size || |
b5256303 | 2619 | nblkptr < doi.doi_nblkptr || |
da2feb42 | 2620 | dn_slots != doi.doi_dnodesize >> DNODE_SHIFT || |
0c03d21a MA |
2621 | (rwa->raw && |
2622 | (indblksz != doi.doi_metadata_block_size || | |
2623 | drro->drr_nlevels < doi.doi_indirection))) { | |
2624 | err = dmu_free_long_range(rwa->os, | |
ae76f45c TC |
2625 | drro->drr_object, 0, DMU_OBJECT_END); |
2626 | if (err != 0) | |
2627 | return (SET_ERROR(EINVAL)); | |
34dc7c2f | 2628 | } |
047116ac TC |
2629 | |
2630 | /* | |
2631 | * The dmu does not currently support decreasing nlevels | |
2632 | * on an object. For non-raw sends, this does not matter | |
2633 | * and the new object can just use the previous one's nlevels. | |
2634 | * For raw sends, however, the structure of the received dnode | |
2635 | * (including nlevels) must match that of the send side. | |
2636 | * Therefore, instead of using dmu_object_reclaim(), we must | |
2637 | * free the object completely and call dmu_object_claim_dnsize() | |
2638 | * instead. | |
2639 | */ | |
2640 | if ((rwa->raw && drro->drr_nlevels < doi.doi_indirection) || | |
da2feb42 | 2641 | dn_slots != doi.doi_dnodesize >> DNODE_SHIFT) { |
0c03d21a | 2642 | err = dmu_free_long_object(rwa->os, drro->drr_object); |
047116ac TC |
2643 | if (err != 0) |
2644 | return (SET_ERROR(EINVAL)); | |
2645 | ||
2646 | txg_wait_synced(dmu_objset_pool(rwa->os), 0); | |
2647 | object = DMU_NEW_OBJECT; | |
2648 | } | |
2649 | } else if (err == EEXIST) { | |
2650 | /* | |
2651 | * The object requested is currently an interior slot of a | |
2652 | * multi-slot dnode. This will be resolved when the next txg | |
2653 | * is synced out, since the send stream will have told us | |
2654 | * to free this slot when we freed the associated dnode | |
2655 | * earlier in the stream. | |
2656 | */ | |
2657 | txg_wait_synced(dmu_objset_pool(rwa->os), 0); | |
2658 | object = drro->drr_object; | |
2659 | } else { | |
2660 | /* object is free and we are about to allocate a new one */ | |
2661 | object = DMU_NEW_OBJECT; | |
2662 | } | |
2663 | ||
2664 | /* | |
2665 | * If this is a multi-slot dnode there is a chance that this | |
2666 | * object will expand into a slot that is already used by | |
2667 | * another object from the previous snapshot. We must free | |
2668 | * these objects before we attempt to allocate the new dnode. | |
2669 | */ | |
da2feb42 | 2670 | if (dn_slots > 1) { |
5121c4fb TC |
2671 | boolean_t need_sync = B_FALSE; |
2672 | ||
047116ac | 2673 | for (uint64_t slot = drro->drr_object + 1; |
da2feb42 | 2674 | slot < drro->drr_object + dn_slots; |
047116ac TC |
2675 | slot++) { |
2676 | dmu_object_info_t slot_doi; | |
2677 | ||
2678 | err = dmu_object_info(rwa->os, slot, &slot_doi); | |
2679 | if (err == ENOENT || err == EEXIST) | |
2680 | continue; | |
2681 | else if (err != 0) | |
2682 | return (err); | |
2683 | ||
0c03d21a | 2684 | err = dmu_free_long_object(rwa->os, slot); |
047116ac TC |
2685 | |
2686 | if (err != 0) | |
2687 | return (err); | |
5121c4fb TC |
2688 | |
2689 | need_sync = B_TRUE; | |
047116ac TC |
2690 | } |
2691 | ||
5121c4fb TC |
2692 | if (need_sync) |
2693 | txg_wait_synced(dmu_objset_pool(rwa->os), 0); | |
6c59307a MA |
2694 | } |
2695 | ||
fcff0f35 | 2696 | tx = dmu_tx_create(rwa->os); |
6c59307a | 2697 | dmu_tx_hold_bonus(tx, object); |
b5256303 | 2698 | dmu_tx_hold_write(tx, object, 0, 0); |
6c59307a MA |
2699 | err = dmu_tx_assign(tx, TXG_WAIT); |
2700 | if (err != 0) { | |
2701 | dmu_tx_abort(tx); | |
2702 | return (err); | |
2703 | } | |
2704 | ||
2705 | if (object == DMU_NEW_OBJECT) { | |
2706 | /* currently free, want to be allocated */ | |
50c957f7 | 2707 | err = dmu_object_claim_dnsize(rwa->os, drro->drr_object, |
34dc7c2f | 2708 | drro->drr_type, drro->drr_blksz, |
50c957f7 | 2709 | drro->drr_bonustype, drro->drr_bonuslen, |
da2feb42 | 2710 | dn_slots << DNODE_SHIFT, tx); |
6c59307a MA |
2711 | } else if (drro->drr_type != doi.doi_type || |
2712 | drro->drr_blksz != doi.doi_data_block_size || | |
2713 | drro->drr_bonustype != doi.doi_bonus_type || | |
e0dd0a32 | 2714 | drro->drr_bonuslen != doi.doi_bonus_size) { |
6c59307a | 2715 | /* currently allocated, but with different properties */ |
e14a32b1 | 2716 | err = dmu_object_reclaim_dnsize(rwa->os, drro->drr_object, |
34dc7c2f | 2717 | drro->drr_type, drro->drr_blksz, |
e14a32b1 | 2718 | drro->drr_bonustype, drro->drr_bonuslen, |
da2feb42 | 2719 | dn_slots << DNODE_SHIFT, tx); |
34dc7c2f | 2720 | } |
13fe0198 | 2721 | if (err != 0) { |
9b840763 | 2722 | dmu_tx_commit(tx); |
2e528b49 | 2723 | return (SET_ERROR(EINVAL)); |
428870ff | 2724 | } |
9babb374 | 2725 | |
0c03d21a | 2726 | if (rwa->or_crypt_params_present) { |
095495e0 | 2727 | /* |
0c03d21a MA |
2728 | * Set the crypt params for the buffer associated with this |
2729 | * range of dnodes. This causes the blkptr_t to have the | |
2730 | * same crypt params (byteorder, salt, iv, mac) as on the | |
2731 | * sending side. | |
2732 | * | |
2733 | * Since we are committing this tx now, it is possible for | |
2734 | * the dnode block to end up on-disk with the incorrect MAC, | |
2735 | * if subsequent objects in this block are received in a | |
2736 | * different txg. However, since the dataset is marked as | |
2737 | * inconsistent, no code paths will do a non-raw read (or | |
2738 | * decrypt the block / verify the MAC). The receive code and | |
2739 | * scrub code can safely do raw reads and verify the | |
2740 | * checksum. They don't need to verify the MAC. | |
095495e0 | 2741 | */ |
0c03d21a MA |
2742 | dmu_buf_t *db = NULL; |
2743 | uint64_t offset = rwa->or_firstobj * DNODE_MIN_SIZE; | |
2744 | ||
2745 | err = dmu_buf_hold_by_dnode(DMU_META_DNODE(rwa->os), | |
2746 | offset, FTAG, &db, DMU_READ_PREFETCH | DMU_READ_NO_DECRYPT); | |
095495e0 TC |
2747 | if (err != 0) { |
2748 | dmu_tx_commit(tx); | |
2749 | return (SET_ERROR(EINVAL)); | |
2750 | } | |
0c03d21a MA |
2751 | |
2752 | dmu_buf_set_crypt_params(db, rwa->or_byteorder, | |
2753 | rwa->or_salt, rwa->or_iv, rwa->or_mac, tx); | |
2754 | ||
2755 | dmu_buf_rele(db, FTAG); | |
2756 | ||
2757 | rwa->or_crypt_params_present = B_FALSE; | |
095495e0 | 2758 | } |
9b840763 | 2759 | |
fcff0f35 | 2760 | dmu_object_set_checksum(rwa->os, drro->drr_object, |
37f8a883 | 2761 | drro->drr_checksumtype, tx); |
fcff0f35 | 2762 | dmu_object_set_compress(rwa->os, drro->drr_object, |
37f8a883 | 2763 | drro->drr_compress, tx); |
34dc7c2f | 2764 | |
b5256303 | 2765 | /* handle more restrictive dnode structuring for raw recvs */ |
9b840763 | 2766 | if (rwa->raw) { |
b5256303 TC |
2767 | /* |
2768 | * Set the indirect block shift and nlevels. This will not fail | |
2769 | * because we ensured all of the blocks were free earlier if | |
2770 | * this is a new object. | |
2771 | */ | |
2772 | VERIFY0(dmu_object_set_blocksize(rwa->os, drro->drr_object, | |
2773 | drro->drr_blksz, drro->drr_indblkshift, tx)); | |
2774 | VERIFY0(dmu_object_set_nlevels(rwa->os, drro->drr_object, | |
2775 | drro->drr_nlevels, tx)); | |
ae76f45c TC |
2776 | VERIFY0(dmu_object_set_maxblkid(rwa->os, drro->drr_object, |
2777 | drro->drr_maxblkid, tx)); | |
b5256303 TC |
2778 | } |
2779 | ||
b128c09f | 2780 | if (data != NULL) { |
34dc7c2f | 2781 | dmu_buf_t *db; |
b5256303 | 2782 | uint32_t flags = DMU_READ_NO_PREFETCH; |
b128c09f | 2783 | |
9b840763 | 2784 | if (rwa->raw) |
b5256303 TC |
2785 | flags |= DMU_READ_NO_DECRYPT; |
2786 | ||
2787 | VERIFY0(dmu_bonus_hold_impl(rwa->os, drro->drr_object, | |
2788 | FTAG, flags, &db)); | |
34dc7c2f BB |
2789 | dmu_buf_will_dirty(db, tx); |
2790 | ||
2791 | ASSERT3U(db->db_size, >=, drro->drr_bonuslen); | |
b5256303 TC |
2792 | bcopy(data, db->db_data, DRR_OBJECT_PAYLOAD_SIZE(drro)); |
2793 | ||
2794 | /* | |
2795 | * Raw bonus buffers have their byteorder determined by the | |
2796 | * DRR_OBJECT_RANGE record. | |
2797 | */ | |
9b840763 | 2798 | if (rwa->byteswap && !rwa->raw) { |
9ae529ec CS |
2799 | dmu_object_byteswap_t byteswap = |
2800 | DMU_OT_BYTESWAP(drro->drr_bonustype); | |
2801 | dmu_ot_byteswap[byteswap].ob_func(db->db_data, | |
b5256303 | 2802 | DRR_OBJECT_PAYLOAD_SIZE(drro)); |
34dc7c2f BB |
2803 | } |
2804 | dmu_buf_rele(db, FTAG); | |
2805 | } | |
2806 | dmu_tx_commit(tx); | |
47dfff3b | 2807 | |
34dc7c2f BB |
2808 | return (0); |
2809 | } | |
2810 | ||
2811 | /* ARGSUSED */ | |
60948de1 | 2812 | noinline static int |
fcff0f35 | 2813 | receive_freeobjects(struct receive_writer_arg *rwa, |
34dc7c2f BB |
2814 | struct drr_freeobjects *drrfo) |
2815 | { | |
2816 | uint64_t obj; | |
e6d3a843 | 2817 | int next_err = 0; |
34dc7c2f BB |
2818 | |
2819 | if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) | |
2e528b49 | 2820 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 2821 | |
50c957f7 | 2822 | for (obj = drrfo->drr_firstobj == 0 ? 1 : drrfo->drr_firstobj; |
e6d3a843 PD |
2823 | obj < drrfo->drr_firstobj + drrfo->drr_numobjs && next_err == 0; |
2824 | next_err = dmu_object_next(rwa->os, &obj, FALSE, 0)) { | |
50c957f7 | 2825 | dmu_object_info_t doi; |
34dc7c2f BB |
2826 | int err; |
2827 | ||
50c957f7 | 2828 | err = dmu_object_info(rwa->os, obj, &doi); |
39f56627 | 2829 | if (err == ENOENT) |
34dc7c2f | 2830 | continue; |
39f56627 | 2831 | else if (err != 0) |
50c957f7 | 2832 | return (err); |
34dc7c2f | 2833 | |
0c03d21a | 2834 | err = dmu_free_long_object(rwa->os, obj); |
440a3eb9 | 2835 | |
13fe0198 | 2836 | if (err != 0) |
34dc7c2f | 2837 | return (err); |
48fbb9dd FG |
2838 | |
2839 | if (obj > rwa->max_object) | |
2840 | rwa->max_object = obj; | |
34dc7c2f | 2841 | } |
e6d3a843 PD |
2842 | if (next_err != ESRCH) |
2843 | return (next_err); | |
34dc7c2f BB |
2844 | return (0); |
2845 | } | |
2846 | ||
60948de1 | 2847 | noinline static int |
fcff0f35 | 2848 | receive_write(struct receive_writer_arg *rwa, struct drr_write *drrw, |
4ea3f864 | 2849 | arc_buf_t *abuf) |
34dc7c2f | 2850 | { |
34dc7c2f | 2851 | int err; |
440a3eb9 TC |
2852 | dmu_tx_t *tx; |
2853 | dnode_t *dn; | |
34dc7c2f | 2854 | |
2aa34383 | 2855 | if (drrw->drr_offset + drrw->drr_logical_size < drrw->drr_offset || |
9ae529ec | 2856 | !DMU_OT_IS_VALID(drrw->drr_type)) |
2e528b49 | 2857 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 2858 | |
47dfff3b MA |
2859 | /* |
2860 | * For resuming to work, records must be in increasing order | |
2861 | * by (object, offset). | |
2862 | */ | |
2863 | if (drrw->drr_object < rwa->last_object || | |
2864 | (drrw->drr_object == rwa->last_object && | |
2865 | drrw->drr_offset < rwa->last_offset)) { | |
2866 | return (SET_ERROR(EINVAL)); | |
2867 | } | |
2868 | rwa->last_object = drrw->drr_object; | |
2869 | rwa->last_offset = drrw->drr_offset; | |
2870 | ||
48fbb9dd FG |
2871 | if (rwa->last_object > rwa->max_object) |
2872 | rwa->max_object = rwa->last_object; | |
2873 | ||
fcff0f35 | 2874 | if (dmu_object_info(rwa->os, drrw->drr_object, NULL) != 0) |
88904bb3 MA |
2875 | return (SET_ERROR(EINVAL)); |
2876 | ||
fcff0f35 | 2877 | tx = dmu_tx_create(rwa->os); |
34dc7c2f | 2878 | dmu_tx_hold_write(tx, drrw->drr_object, |
2aa34383 | 2879 | drrw->drr_offset, drrw->drr_logical_size); |
34dc7c2f | 2880 | err = dmu_tx_assign(tx, TXG_WAIT); |
13fe0198 | 2881 | if (err != 0) { |
34dc7c2f BB |
2882 | dmu_tx_abort(tx); |
2883 | return (err); | |
2884 | } | |
9b840763 | 2885 | |
b5256303 TC |
2886 | if (rwa->byteswap && !arc_is_encrypted(abuf) && |
2887 | arc_get_compression(abuf) == ZIO_COMPRESS_OFF) { | |
9ae529ec CS |
2888 | dmu_object_byteswap_t byteswap = |
2889 | DMU_OT_BYTESWAP(drrw->drr_type); | |
37f8a883 | 2890 | dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, |
2aa34383 | 2891 | DRR_WRITE_PAYLOAD_SIZE(drrw)); |
9ae529ec | 2892 | } |
37f8a883 | 2893 | |
440a3eb9 TC |
2894 | VERIFY0(dnode_hold(rwa->os, drrw->drr_object, FTAG, &dn)); |
2895 | dmu_assign_arcbuf_by_dnode(dn, drrw->drr_offset, abuf, tx); | |
2896 | dnode_rele(dn, FTAG); | |
47dfff3b MA |
2897 | |
2898 | /* | |
2899 | * Note: If the receive fails, we want the resume stream to start | |
2900 | * with the same record that we last successfully received (as opposed | |
2901 | * to the next record), so that we can verify that we are | |
2902 | * resuming from the correct location. | |
2903 | */ | |
2904 | save_resume_state(rwa, drrw->drr_object, drrw->drr_offset, tx); | |
34dc7c2f | 2905 | dmu_tx_commit(tx); |
47dfff3b | 2906 | |
34dc7c2f BB |
2907 | return (0); |
2908 | } | |
2909 | ||
428870ff BB |
2910 | /* |
2911 | * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed | |
2912 | * streams to refer to a copy of the data that is already on the | |
2913 | * system because it came in earlier in the stream. This function | |
2914 | * finds the earlier copy of the data, and uses that copy instead of | |
2915 | * data from the stream to fulfill this write. | |
2916 | */ | |
2917 | static int | |
fcff0f35 PD |
2918 | receive_write_byref(struct receive_writer_arg *rwa, |
2919 | struct drr_write_byref *drrwbr) | |
428870ff BB |
2920 | { |
2921 | dmu_tx_t *tx; | |
2922 | int err; | |
2923 | guid_map_entry_t gmesrch; | |
2924 | guid_map_entry_t *gmep; | |
9b67f605 | 2925 | avl_index_t where; |
428870ff | 2926 | objset_t *ref_os = NULL; |
b5256303 | 2927 | int flags = DMU_READ_PREFETCH; |
428870ff BB |
2928 | dmu_buf_t *dbp; |
2929 | ||
2930 | if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) | |
2e528b49 | 2931 | return (SET_ERROR(EINVAL)); |
428870ff BB |
2932 | |
2933 | /* | |
2934 | * If the GUID of the referenced dataset is different from the | |
2935 | * GUID of the target dataset, find the referenced dataset. | |
2936 | */ | |
2937 | if (drrwbr->drr_toguid != drrwbr->drr_refguid) { | |
2938 | gmesrch.guid = drrwbr->drr_refguid; | |
fcff0f35 | 2939 | if ((gmep = avl_find(rwa->guid_to_ds_map, &gmesrch, |
428870ff | 2940 | &where)) == NULL) { |
2e528b49 | 2941 | return (SET_ERROR(EINVAL)); |
428870ff BB |
2942 | } |
2943 | if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) | |
2e528b49 | 2944 | return (SET_ERROR(EINVAL)); |
428870ff | 2945 | } else { |
fcff0f35 | 2946 | ref_os = rwa->os; |
428870ff BB |
2947 | } |
2948 | ||
48fbb9dd FG |
2949 | if (drrwbr->drr_object > rwa->max_object) |
2950 | rwa->max_object = drrwbr->drr_object; | |
2951 | ||
9b840763 | 2952 | if (rwa->raw) |
b5256303 | 2953 | flags |= DMU_READ_NO_DECRYPT; |
b5256303 TC |
2954 | |
2955 | /* may return either a regular db or an encrypted one */ | |
c65aa5b2 | 2956 | err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, |
b5256303 | 2957 | drrwbr->drr_refoffset, FTAG, &dbp, flags); |
9b67f605 | 2958 | if (err != 0) |
428870ff BB |
2959 | return (err); |
2960 | ||
fcff0f35 | 2961 | tx = dmu_tx_create(rwa->os); |
428870ff BB |
2962 | |
2963 | dmu_tx_hold_write(tx, drrwbr->drr_object, | |
2964 | drrwbr->drr_offset, drrwbr->drr_length); | |
2965 | err = dmu_tx_assign(tx, TXG_WAIT); | |
13fe0198 | 2966 | if (err != 0) { |
428870ff BB |
2967 | dmu_tx_abort(tx); |
2968 | return (err); | |
2969 | } | |
b5256303 | 2970 | |
9b840763 | 2971 | if (rwa->raw) { |
b5256303 TC |
2972 | dmu_copy_from_buf(rwa->os, drrwbr->drr_object, |
2973 | drrwbr->drr_offset, dbp, tx); | |
2974 | } else { | |
2975 | dmu_write(rwa->os, drrwbr->drr_object, | |
2976 | drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); | |
2977 | } | |
428870ff | 2978 | dmu_buf_rele(dbp, FTAG); |
47dfff3b MA |
2979 | |
2980 | /* See comment in restore_write. */ | |
2981 | save_resume_state(rwa, drrwbr->drr_object, drrwbr->drr_offset, tx); | |
428870ff BB |
2982 | dmu_tx_commit(tx); |
2983 | return (0); | |
2984 | } | |
2985 | ||
9b67f605 | 2986 | static int |
fcff0f35 | 2987 | receive_write_embedded(struct receive_writer_arg *rwa, |
47dfff3b | 2988 | struct drr_write_embedded *drrwe, void *data) |
9b67f605 MA |
2989 | { |
2990 | dmu_tx_t *tx; | |
2991 | int err; | |
9b67f605 | 2992 | |
47dfff3b | 2993 | if (drrwe->drr_offset + drrwe->drr_length < drrwe->drr_offset) |
ecb2b7dc | 2994 | return (SET_ERROR(EINVAL)); |
9b67f605 | 2995 | |
47dfff3b | 2996 | if (drrwe->drr_psize > BPE_PAYLOAD_SIZE) |
ecb2b7dc | 2997 | return (SET_ERROR(EINVAL)); |
9b67f605 | 2998 | |
47dfff3b | 2999 | if (drrwe->drr_etype >= NUM_BP_EMBEDDED_TYPES) |
ecb2b7dc | 3000 | return (SET_ERROR(EINVAL)); |
47dfff3b | 3001 | if (drrwe->drr_compression >= ZIO_COMPRESS_FUNCTIONS) |
ecb2b7dc | 3002 | return (SET_ERROR(EINVAL)); |
440a3eb9 TC |
3003 | if (rwa->raw) |
3004 | return (SET_ERROR(EINVAL)); | |
9b67f605 | 3005 | |
48fbb9dd FG |
3006 | if (drrwe->drr_object > rwa->max_object) |
3007 | rwa->max_object = drrwe->drr_object; | |
3008 | ||
fcff0f35 | 3009 | tx = dmu_tx_create(rwa->os); |
9b67f605 | 3010 | |
47dfff3b MA |
3011 | dmu_tx_hold_write(tx, drrwe->drr_object, |
3012 | drrwe->drr_offset, drrwe->drr_length); | |
9b67f605 MA |
3013 | err = dmu_tx_assign(tx, TXG_WAIT); |
3014 | if (err != 0) { | |
3015 | dmu_tx_abort(tx); | |
3016 | return (err); | |
3017 | } | |
3018 | ||
47dfff3b MA |
3019 | dmu_write_embedded(rwa->os, drrwe->drr_object, |
3020 | drrwe->drr_offset, data, drrwe->drr_etype, | |
3021 | drrwe->drr_compression, drrwe->drr_lsize, drrwe->drr_psize, | |
fcff0f35 | 3022 | rwa->byteswap ^ ZFS_HOST_BYTEORDER, tx); |
9b67f605 | 3023 | |
47dfff3b MA |
3024 | /* See comment in restore_write. */ |
3025 | save_resume_state(rwa, drrwe->drr_object, drrwe->drr_offset, tx); | |
9b67f605 MA |
3026 | dmu_tx_commit(tx); |
3027 | return (0); | |
3028 | } | |
3029 | ||
428870ff | 3030 | static int |
fcff0f35 | 3031 | receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs, |
b5256303 | 3032 | arc_buf_t *abuf) |
428870ff BB |
3033 | { |
3034 | dmu_tx_t *tx; | |
428870ff BB |
3035 | dmu_buf_t *db, *db_spill; |
3036 | int err; | |
e7504d7a | 3037 | uint32_t flags = 0; |
428870ff BB |
3038 | |
3039 | if (drrs->drr_length < SPA_MINBLOCKSIZE || | |
fcff0f35 | 3040 | drrs->drr_length > spa_maxblocksize(dmu_objset_spa(rwa->os))) |
2e528b49 | 3041 | return (SET_ERROR(EINVAL)); |
428870ff | 3042 | |
9b840763 | 3043 | if (rwa->raw) { |
b5256303 TC |
3044 | if (!DMU_OT_IS_VALID(drrs->drr_type) || |
3045 | drrs->drr_compressiontype >= ZIO_COMPRESS_FUNCTIONS || | |
3046 | drrs->drr_compressed_size == 0) | |
3047 | return (SET_ERROR(EINVAL)); | |
e7504d7a TC |
3048 | |
3049 | flags |= DMU_READ_NO_DECRYPT; | |
b5256303 TC |
3050 | } |
3051 | ||
fcff0f35 | 3052 | if (dmu_object_info(rwa->os, drrs->drr_object, NULL) != 0) |
2e528b49 | 3053 | return (SET_ERROR(EINVAL)); |
428870ff | 3054 | |
48fbb9dd FG |
3055 | if (drrs->drr_object > rwa->max_object) |
3056 | rwa->max_object = drrs->drr_object; | |
3057 | ||
fcff0f35 | 3058 | VERIFY0(dmu_bonus_hold(rwa->os, drrs->drr_object, FTAG, &db)); |
e7504d7a TC |
3059 | if ((err = dmu_spill_hold_by_bonus(db, DMU_READ_NO_DECRYPT, FTAG, |
3060 | &db_spill)) != 0) { | |
428870ff BB |
3061 | dmu_buf_rele(db, FTAG); |
3062 | return (err); | |
3063 | } | |
3064 | ||
fcff0f35 | 3065 | tx = dmu_tx_create(rwa->os); |
428870ff BB |
3066 | |
3067 | dmu_tx_hold_spill(tx, db->db_object); | |
3068 | ||
3069 | err = dmu_tx_assign(tx, TXG_WAIT); | |
13fe0198 | 3070 | if (err != 0) { |
428870ff BB |
3071 | dmu_buf_rele(db, FTAG); |
3072 | dmu_buf_rele(db_spill, FTAG); | |
3073 | dmu_tx_abort(tx); | |
3074 | return (err); | |
3075 | } | |
428870ff BB |
3076 | |
3077 | if (db_spill->db_size < drrs->drr_length) | |
3078 | VERIFY(0 == dbuf_spill_set_blksz(db_spill, | |
3079 | drrs->drr_length, tx)); | |
1a234278 PZ |
3080 | |
3081 | if (rwa->byteswap && !arc_is_encrypted(abuf) && | |
3082 | arc_get_compression(abuf) == ZIO_COMPRESS_OFF) { | |
3083 | dmu_object_byteswap_t byteswap = | |
3084 | DMU_OT_BYTESWAP(drrs->drr_type); | |
3085 | dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, | |
3086 | DRR_SPILL_PAYLOAD_SIZE(drrs)); | |
3087 | } | |
3088 | ||
440a3eb9 | 3089 | dbuf_assign_arcbuf((dmu_buf_impl_t *)db_spill, abuf, tx); |
428870ff BB |
3090 | |
3091 | dmu_buf_rele(db, FTAG); | |
3092 | dmu_buf_rele(db_spill, FTAG); | |
3093 | ||
3094 | dmu_tx_commit(tx); | |
3095 | return (0); | |
3096 | } | |
3097 | ||
34dc7c2f | 3098 | /* ARGSUSED */ |
60948de1 | 3099 | noinline static int |
fcff0f35 | 3100 | receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf) |
34dc7c2f | 3101 | { |
34dc7c2f BB |
3102 | int err; |
3103 | ||
ee45fbd8 | 3104 | if (drrf->drr_length != DMU_OBJECT_END && |
34dc7c2f | 3105 | drrf->drr_offset + drrf->drr_length < drrf->drr_offset) |
2e528b49 | 3106 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 3107 | |
fcff0f35 | 3108 | if (dmu_object_info(rwa->os, drrf->drr_object, NULL) != 0) |
2e528b49 | 3109 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 3110 | |
48fbb9dd FG |
3111 | if (drrf->drr_object > rwa->max_object) |
3112 | rwa->max_object = drrf->drr_object; | |
3113 | ||
0c03d21a MA |
3114 | err = dmu_free_long_range(rwa->os, drrf->drr_object, |
3115 | drrf->drr_offset, drrf->drr_length); | |
fcff0f35 | 3116 | |
34dc7c2f BB |
3117 | return (err); |
3118 | } | |
3119 | ||
b5256303 TC |
3120 | static int |
3121 | receive_object_range(struct receive_writer_arg *rwa, | |
3122 | struct drr_object_range *drror) | |
3123 | { | |
b5256303 TC |
3124 | /* |
3125 | * By default, we assume this block is in our native format | |
3126 | * (ZFS_HOST_BYTEORDER). We then take into account whether | |
3127 | * the send stream is byteswapped (rwa->byteswap). Finally, | |
3128 | * we need to byteswap again if this particular block was | |
3129 | * in non-native format on the send side. | |
3130 | */ | |
3131 | boolean_t byteorder = ZFS_HOST_BYTEORDER ^ rwa->byteswap ^ | |
3132 | !!DRR_IS_RAW_BYTESWAPPED(drror->drr_flags); | |
3133 | ||
3134 | /* | |
3135 | * Since dnode block sizes are constant, we should not need to worry | |
3136 | * about making sure that the dnode block size is the same on the | |
3137 | * sending and receiving sides for the time being. For non-raw sends, | |
3138 | * this does not matter (and in fact we do not send a DRR_OBJECT_RANGE | |
3139 | * record at all). Raw sends require this record type because the | |
3140 | * encryption parameters are used to protect an entire block of bonus | |
3141 | * buffers. If the size of dnode blocks ever becomes variable, | |
3142 | * handling will need to be added to ensure that dnode block sizes | |
3143 | * match on the sending and receiving side. | |
3144 | */ | |
3145 | if (drror->drr_numslots != DNODES_PER_BLOCK || | |
3146 | P2PHASE(drror->drr_firstobj, DNODES_PER_BLOCK) != 0 || | |
9b840763 | 3147 | !rwa->raw) |
b5256303 TC |
3148 | return (SET_ERROR(EINVAL)); |
3149 | ||
48fbb9dd FG |
3150 | if (drror->drr_firstobj > rwa->max_object) |
3151 | rwa->max_object = drror->drr_firstobj; | |
3152 | ||
b5256303 | 3153 | /* |
095495e0 | 3154 | * The DRR_OBJECT_RANGE handling must be deferred to receive_object() |
0c03d21a MA |
3155 | * so that the block of dnodes is not written out when it's empty, |
3156 | * and converted to a HOLE BP. | |
b5256303 | 3157 | */ |
0c03d21a | 3158 | rwa->or_crypt_params_present = B_TRUE; |
095495e0 TC |
3159 | rwa->or_firstobj = drror->drr_firstobj; |
3160 | rwa->or_numslots = drror->drr_numslots; | |
3161 | bcopy(drror->drr_salt, rwa->or_salt, ZIO_DATA_SALT_LEN); | |
3162 | bcopy(drror->drr_iv, rwa->or_iv, ZIO_DATA_IV_LEN); | |
3163 | bcopy(drror->drr_mac, rwa->or_mac, ZIO_DATA_MAC_LEN); | |
3164 | rwa->or_byteorder = byteorder; | |
3165 | ||
b5256303 TC |
3166 | return (0); |
3167 | } | |
3168 | ||
13fe0198 MA |
3169 | /* used to destroy the drc_ds on error */ |
3170 | static void | |
3171 | dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) | |
3172 | { | |
2637dda8 | 3173 | dsl_dataset_t *ds = drc->drc_ds; |
b5256303 TC |
3174 | ds_hold_flags_t dsflags = (drc->drc_raw) ? 0 : DS_HOLD_FLAG_DECRYPT; |
3175 | ||
3176 | /* | |
3177 | * Wait for the txg sync before cleaning up the receive. For | |
3178 | * resumable receives, this ensures that our resume state has | |
3179 | * been written out to disk. For raw receives, this ensures | |
3180 | * that the user accounting code will not attempt to do anything | |
3181 | * after we stopped receiving the dataset. | |
3182 | */ | |
2637dda8 | 3183 | txg_wait_synced(ds->ds_dir->dd_pool, 0); |
0c03d21a | 3184 | ds->ds_objset->os_raw_receive = B_FALSE; |
b5256303 | 3185 | |
2637dda8 TC |
3186 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
3187 | if (drc->drc_resumable && !BP_IS_HOLE(dsl_dataset_get_blkptr(ds))) { | |
3188 | rrw_exit(&ds->ds_bp_rwlock, FTAG); | |
3189 | dsl_dataset_disown(ds, dsflags, dmu_recv_tag); | |
47dfff3b | 3190 | } else { |
eca7b760 | 3191 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
2637dda8 TC |
3192 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
3193 | dsl_dataset_name(ds, name); | |
3194 | dsl_dataset_disown(ds, dsflags, dmu_recv_tag); | |
47dfff3b MA |
3195 | (void) dsl_destroy_head(name); |
3196 | } | |
13fe0198 MA |
3197 | } |
3198 | ||
37f8a883 | 3199 | static void |
fcff0f35 | 3200 | receive_cksum(struct receive_arg *ra, int len, void *buf) |
37f8a883 MA |
3201 | { |
3202 | if (ra->byteswap) { | |
a6255b7f | 3203 | (void) fletcher_4_incremental_byteswap(buf, len, &ra->cksum); |
37f8a883 | 3204 | } else { |
a6255b7f | 3205 | (void) fletcher_4_incremental_native(buf, len, &ra->cksum); |
37f8a883 MA |
3206 | } |
3207 | } | |
3208 | ||
3209 | /* | |
fcff0f35 PD |
3210 | * Read the payload into a buffer of size len, and update the current record's |
3211 | * payload field. | |
3212 | * Allocate ra->next_rrd and read the next record's header into | |
3213 | * ra->next_rrd->header. | |
37f8a883 MA |
3214 | * Verify checksum of payload and next record. |
3215 | */ | |
3216 | static int | |
fcff0f35 | 3217 | receive_read_payload_and_next_header(struct receive_arg *ra, int len, void *buf) |
37f8a883 MA |
3218 | { |
3219 | int err; | |
3220 | zio_cksum_t cksum_orig; | |
3221 | zio_cksum_t *cksump; | |
3222 | ||
3223 | if (len != 0) { | |
fcff0f35 | 3224 | ASSERT3U(len, <=, SPA_MAXBLOCKSIZE); |
47dfff3b | 3225 | err = receive_read(ra, len, buf); |
37f8a883 MA |
3226 | if (err != 0) |
3227 | return (err); | |
47dfff3b MA |
3228 | receive_cksum(ra, len, buf); |
3229 | ||
3230 | /* note: rrd is NULL when reading the begin record's payload */ | |
3231 | if (ra->rrd != NULL) { | |
3232 | ra->rrd->payload = buf; | |
3233 | ra->rrd->payload_size = len; | |
3234 | ra->rrd->bytes_read = ra->bytes_read; | |
3235 | } | |
37f8a883 MA |
3236 | } |
3237 | ||
3238 | ra->prev_cksum = ra->cksum; | |
3239 | ||
fcff0f35 PD |
3240 | ra->next_rrd = kmem_zalloc(sizeof (*ra->next_rrd), KM_SLEEP); |
3241 | err = receive_read(ra, sizeof (ra->next_rrd->header), | |
3242 | &ra->next_rrd->header); | |
47dfff3b | 3243 | ra->next_rrd->bytes_read = ra->bytes_read; |
b5256303 | 3244 | |
fcff0f35 PD |
3245 | if (err != 0) { |
3246 | kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); | |
3247 | ra->next_rrd = NULL; | |
37f8a883 | 3248 | return (err); |
fcff0f35 PD |
3249 | } |
3250 | if (ra->next_rrd->header.drr_type == DRR_BEGIN) { | |
3251 | kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); | |
3252 | ra->next_rrd = NULL; | |
37f8a883 | 3253 | return (SET_ERROR(EINVAL)); |
fcff0f35 | 3254 | } |
37f8a883 MA |
3255 | |
3256 | /* | |
3257 | * Note: checksum is of everything up to but not including the | |
3258 | * checksum itself. | |
3259 | */ | |
3260 | ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), | |
3261 | ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); | |
fcff0f35 | 3262 | receive_cksum(ra, |
37f8a883 | 3263 | offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), |
fcff0f35 | 3264 | &ra->next_rrd->header); |
37f8a883 | 3265 | |
fcff0f35 PD |
3266 | cksum_orig = ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; |
3267 | cksump = &ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; | |
37f8a883 MA |
3268 | |
3269 | if (ra->byteswap) | |
fcff0f35 | 3270 | byteswap_record(&ra->next_rrd->header); |
37f8a883 MA |
3271 | |
3272 | if ((!ZIO_CHECKSUM_IS_ZERO(cksump)) && | |
fcff0f35 PD |
3273 | !ZIO_CHECKSUM_EQUAL(ra->cksum, *cksump)) { |
3274 | kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); | |
3275 | ra->next_rrd = NULL; | |
37f8a883 | 3276 | return (SET_ERROR(ECKSUM)); |
fcff0f35 | 3277 | } |
37f8a883 | 3278 | |
fcff0f35 | 3279 | receive_cksum(ra, sizeof (cksum_orig), &cksum_orig); |
37f8a883 MA |
3280 | |
3281 | return (0); | |
3282 | } | |
3283 | ||
e6d3a843 PD |
3284 | static void |
3285 | objlist_create(struct objlist *list) | |
3286 | { | |
3287 | list_create(&list->list, sizeof (struct receive_objnode), | |
3288 | offsetof(struct receive_objnode, node)); | |
3289 | list->last_lookup = 0; | |
3290 | } | |
3291 | ||
3292 | static void | |
3293 | objlist_destroy(struct objlist *list) | |
3294 | { | |
1c27024e | 3295 | for (struct receive_objnode *n = list_remove_head(&list->list); |
e6d3a843 PD |
3296 | n != NULL; n = list_remove_head(&list->list)) { |
3297 | kmem_free(n, sizeof (*n)); | |
3298 | } | |
3299 | list_destroy(&list->list); | |
3300 | } | |
3301 | ||
3302 | /* | |
3303 | * This function looks through the objlist to see if the specified object number | |
3304 | * is contained in the objlist. In the process, it will remove all object | |
3305 | * numbers in the list that are smaller than the specified object number. Thus, | |
3306 | * any lookup of an object number smaller than a previously looked up object | |
3307 | * number will always return false; therefore, all lookups should be done in | |
3308 | * ascending order. | |
3309 | */ | |
3310 | static boolean_t | |
3311 | objlist_exists(struct objlist *list, uint64_t object) | |
3312 | { | |
3313 | struct receive_objnode *node = list_head(&list->list); | |
3314 | ASSERT3U(object, >=, list->last_lookup); | |
3315 | list->last_lookup = object; | |
3316 | while (node != NULL && node->object < object) { | |
3317 | VERIFY3P(node, ==, list_remove_head(&list->list)); | |
3318 | kmem_free(node, sizeof (*node)); | |
3319 | node = list_head(&list->list); | |
3320 | } | |
3321 | return (node != NULL && node->object == object); | |
3322 | } | |
3323 | ||
3324 | /* | |
3325 | * The objlist is a list of object numbers stored in ascending order. However, | |
3326 | * the insertion of new object numbers does not seek out the correct location to | |
3327 | * store a new object number; instead, it appends it to the list for simplicity. | |
3328 | * Thus, any users must take care to only insert new object numbers in ascending | |
3329 | * order. | |
3330 | */ | |
3331 | static void | |
3332 | objlist_insert(struct objlist *list, uint64_t object) | |
3333 | { | |
3334 | struct receive_objnode *node = kmem_zalloc(sizeof (*node), KM_SLEEP); | |
3335 | node->object = object; | |
3336 | #ifdef ZFS_DEBUG | |
3337 | { | |
3338 | struct receive_objnode *last_object = list_tail(&list->list); | |
3339 | uint64_t last_objnum = (last_object != NULL ? last_object->object : 0); | |
3340 | ASSERT3U(node->object, >, last_objnum); | |
3341 | } | |
3342 | #endif | |
3343 | list_insert_tail(&list->list, node); | |
3344 | } | |
3345 | ||
fcff0f35 PD |
3346 | /* |
3347 | * Issue the prefetch reads for any necessary indirect blocks. | |
3348 | * | |
3349 | * We use the object ignore list to tell us whether or not to issue prefetches | |
3350 | * for a given object. We do this for both correctness (in case the blocksize | |
3351 | * of an object has changed) and performance (if the object doesn't exist, don't | |
3352 | * needlessly try to issue prefetches). We also trim the list as we go through | |
3353 | * the stream to prevent it from growing to an unbounded size. | |
3354 | * | |
3355 | * The object numbers within will always be in sorted order, and any write | |
3356 | * records we see will also be in sorted order, but they're not sorted with | |
3357 | * respect to each other (i.e. we can get several object records before | |
3358 | * receiving each object's write records). As a result, once we've reached a | |
3359 | * given object number, we can safely remove any reference to lower object | |
3360 | * numbers in the ignore list. In practice, we receive up to 32 object records | |
3361 | * before receiving write records, so the list can have up to 32 nodes in it. | |
3362 | */ | |
3363 | /* ARGSUSED */ | |
3364 | static void | |
3365 | receive_read_prefetch(struct receive_arg *ra, | |
3366 | uint64_t object, uint64_t offset, uint64_t length) | |
3367 | { | |
e6d3a843 | 3368 | if (!objlist_exists(&ra->ignore_objlist, object)) { |
fcff0f35 PD |
3369 | dmu_prefetch(ra->os, object, 1, offset, length, |
3370 | ZIO_PRIORITY_SYNC_READ); | |
3371 | } | |
3372 | } | |
3373 | ||
3374 | /* | |
3375 | * Read records off the stream, issuing any necessary prefetches. | |
3376 | */ | |
37f8a883 | 3377 | static int |
fcff0f35 | 3378 | receive_read_record(struct receive_arg *ra) |
37f8a883 MA |
3379 | { |
3380 | int err; | |
3381 | ||
fcff0f35 | 3382 | switch (ra->rrd->header.drr_type) { |
37f8a883 MA |
3383 | case DRR_OBJECT: |
3384 | { | |
fcff0f35 | 3385 | struct drr_object *drro = &ra->rrd->header.drr_u.drr_object; |
b5256303 | 3386 | uint32_t size = DRR_OBJECT_PAYLOAD_SIZE(drro); |
fcff0f35 PD |
3387 | void *buf = kmem_zalloc(size, KM_SLEEP); |
3388 | dmu_object_info_t doi; | |
b5256303 | 3389 | |
fcff0f35 PD |
3390 | err = receive_read_payload_and_next_header(ra, size, buf); |
3391 | if (err != 0) { | |
3392 | kmem_free(buf, size); | |
37f8a883 | 3393 | return (err); |
fcff0f35 PD |
3394 | } |
3395 | err = dmu_object_info(ra->os, drro->drr_object, &doi); | |
3396 | /* | |
3397 | * See receive_read_prefetch for an explanation why we're | |
3398 | * storing this object in the ignore_obj_list. | |
3399 | */ | |
047116ac | 3400 | if (err == ENOENT || err == EEXIST || |
fcff0f35 | 3401 | (err == 0 && doi.doi_data_block_size != drro->drr_blksz)) { |
e6d3a843 | 3402 | objlist_insert(&ra->ignore_objlist, drro->drr_object); |
fcff0f35 PD |
3403 | err = 0; |
3404 | } | |
3405 | return (err); | |
37f8a883 MA |
3406 | } |
3407 | case DRR_FREEOBJECTS: | |
3408 | { | |
fcff0f35 PD |
3409 | err = receive_read_payload_and_next_header(ra, 0, NULL); |
3410 | return (err); | |
37f8a883 MA |
3411 | } |
3412 | case DRR_WRITE: | |
3413 | { | |
fcff0f35 | 3414 | struct drr_write *drrw = &ra->rrd->header.drr_u.drr_write; |
2aa34383 DK |
3415 | arc_buf_t *abuf; |
3416 | boolean_t is_meta = DMU_OT_IS_METADATA(drrw->drr_type); | |
b5256303 | 3417 | |
9b840763 | 3418 | if (ra->raw) { |
b5256303 TC |
3419 | boolean_t byteorder = ZFS_HOST_BYTEORDER ^ |
3420 | !!DRR_IS_RAW_BYTESWAPPED(drrw->drr_flags) ^ | |
3421 | ra->byteswap; | |
3422 | ||
3423 | abuf = arc_loan_raw_buf(dmu_objset_spa(ra->os), | |
3424 | drrw->drr_object, byteorder, drrw->drr_salt, | |
3425 | drrw->drr_iv, drrw->drr_mac, drrw->drr_type, | |
3426 | drrw->drr_compressed_size, drrw->drr_logical_size, | |
3427 | drrw->drr_compressiontype); | |
3428 | } else if (DRR_WRITE_COMPRESSED(drrw)) { | |
2aa34383 DK |
3429 | ASSERT3U(drrw->drr_compressed_size, >, 0); |
3430 | ASSERT3U(drrw->drr_logical_size, >=, | |
3431 | drrw->drr_compressed_size); | |
3432 | ASSERT(!is_meta); | |
3433 | abuf = arc_loan_compressed_buf( | |
3434 | dmu_objset_spa(ra->os), | |
3435 | drrw->drr_compressed_size, drrw->drr_logical_size, | |
3436 | drrw->drr_compressiontype); | |
3437 | } else { | |
3438 | abuf = arc_loan_buf(dmu_objset_spa(ra->os), | |
3439 | is_meta, drrw->drr_logical_size); | |
3440 | } | |
37f8a883 | 3441 | |
fcff0f35 | 3442 | err = receive_read_payload_and_next_header(ra, |
2aa34383 | 3443 | DRR_WRITE_PAYLOAD_SIZE(drrw), abuf->b_data); |
fcff0f35 | 3444 | if (err != 0) { |
37f8a883 | 3445 | dmu_return_arcbuf(abuf); |
fcff0f35 PD |
3446 | return (err); |
3447 | } | |
b5256303 | 3448 | ra->rrd->arc_buf = abuf; |
fcff0f35 | 3449 | receive_read_prefetch(ra, drrw->drr_object, drrw->drr_offset, |
2aa34383 | 3450 | drrw->drr_logical_size); |
37f8a883 MA |
3451 | return (err); |
3452 | } | |
3453 | case DRR_WRITE_BYREF: | |
3454 | { | |
fcff0f35 PD |
3455 | struct drr_write_byref *drrwb = |
3456 | &ra->rrd->header.drr_u.drr_write_byref; | |
3457 | err = receive_read_payload_and_next_header(ra, 0, NULL); | |
3458 | receive_read_prefetch(ra, drrwb->drr_object, drrwb->drr_offset, | |
3459 | drrwb->drr_length); | |
3460 | return (err); | |
37f8a883 MA |
3461 | } |
3462 | case DRR_WRITE_EMBEDDED: | |
3463 | { | |
3464 | struct drr_write_embedded *drrwe = | |
fcff0f35 PD |
3465 | &ra->rrd->header.drr_u.drr_write_embedded; |
3466 | uint32_t size = P2ROUNDUP(drrwe->drr_psize, 8); | |
3467 | void *buf = kmem_zalloc(size, KM_SLEEP); | |
3468 | ||
3469 | err = receive_read_payload_and_next_header(ra, size, buf); | |
3470 | if (err != 0) { | |
3471 | kmem_free(buf, size); | |
37f8a883 | 3472 | return (err); |
fcff0f35 PD |
3473 | } |
3474 | ||
3475 | receive_read_prefetch(ra, drrwe->drr_object, drrwe->drr_offset, | |
3476 | drrwe->drr_length); | |
3477 | return (err); | |
37f8a883 MA |
3478 | } |
3479 | case DRR_FREE: | |
3480 | { | |
fcff0f35 PD |
3481 | /* |
3482 | * It might be beneficial to prefetch indirect blocks here, but | |
3483 | * we don't really have the data to decide for sure. | |
3484 | */ | |
3485 | err = receive_read_payload_and_next_header(ra, 0, NULL); | |
3486 | return (err); | |
37f8a883 MA |
3487 | } |
3488 | case DRR_END: | |
3489 | { | |
fcff0f35 | 3490 | struct drr_end *drre = &ra->rrd->header.drr_u.drr_end; |
37f8a883 | 3491 | if (!ZIO_CHECKSUM_EQUAL(ra->prev_cksum, drre->drr_checksum)) |
47dfff3b | 3492 | return (SET_ERROR(ECKSUM)); |
37f8a883 MA |
3493 | return (0); |
3494 | } | |
3495 | case DRR_SPILL: | |
3496 | { | |
fcff0f35 | 3497 | struct drr_spill *drrs = &ra->rrd->header.drr_u.drr_spill; |
b5256303 TC |
3498 | arc_buf_t *abuf; |
3499 | int len = DRR_SPILL_PAYLOAD_SIZE(drrs); | |
3500 | ||
3501 | /* DRR_SPILL records are either raw or uncompressed */ | |
9b840763 | 3502 | if (ra->raw) { |
b5256303 TC |
3503 | boolean_t byteorder = ZFS_HOST_BYTEORDER ^ |
3504 | !!DRR_IS_RAW_BYTESWAPPED(drrs->drr_flags) ^ | |
3505 | ra->byteswap; | |
3506 | ||
3507 | abuf = arc_loan_raw_buf(dmu_objset_spa(ra->os), | |
b0ee5946 | 3508 | dmu_objset_id(ra->os), byteorder, drrs->drr_salt, |
b5256303 TC |
3509 | drrs->drr_iv, drrs->drr_mac, drrs->drr_type, |
3510 | drrs->drr_compressed_size, drrs->drr_length, | |
3511 | drrs->drr_compressiontype); | |
3512 | } else { | |
3513 | abuf = arc_loan_buf(dmu_objset_spa(ra->os), | |
3514 | DMU_OT_IS_METADATA(drrs->drr_type), | |
3515 | drrs->drr_length); | |
3516 | } | |
3517 | ||
3518 | err = receive_read_payload_and_next_header(ra, len, | |
3519 | abuf->b_data); | |
3520 | if (err != 0) { | |
3521 | dmu_return_arcbuf(abuf); | |
3522 | return (err); | |
3523 | } | |
3524 | ra->rrd->arc_buf = abuf; | |
3525 | return (err); | |
3526 | } | |
3527 | case DRR_OBJECT_RANGE: | |
3528 | { | |
3529 | err = receive_read_payload_and_next_header(ra, 0, NULL); | |
fcff0f35 PD |
3530 | return (err); |
3531 | } | |
3532 | default: | |
3533 | return (SET_ERROR(EINVAL)); | |
3534 | } | |
3535 | } | |
3536 | ||
6a8ee4f7 NB |
3537 | static void |
3538 | dprintf_drr(struct receive_record_arg *rrd, int err) | |
3539 | { | |
2fd92c3d | 3540 | #ifdef ZFS_DEBUG |
6a8ee4f7 NB |
3541 | switch (rrd->header.drr_type) { |
3542 | case DRR_OBJECT: | |
3543 | { | |
3544 | struct drr_object *drro = &rrd->header.drr_u.drr_object; | |
3545 | dprintf("drr_type = OBJECT obj = %llu type = %u " | |
3546 | "bonustype = %u blksz = %u bonuslen = %u cksumtype = %u " | |
3547 | "compress = %u dn_slots = %u err = %d\n", | |
3548 | drro->drr_object, drro->drr_type, drro->drr_bonustype, | |
3549 | drro->drr_blksz, drro->drr_bonuslen, | |
3550 | drro->drr_checksumtype, drro->drr_compress, | |
3551 | drro->drr_dn_slots, err); | |
3552 | break; | |
3553 | } | |
3554 | case DRR_FREEOBJECTS: | |
3555 | { | |
3556 | struct drr_freeobjects *drrfo = | |
3557 | &rrd->header.drr_u.drr_freeobjects; | |
3558 | dprintf("drr_type = FREEOBJECTS firstobj = %llu " | |
3559 | "numobjs = %llu err = %d\n", | |
3560 | drrfo->drr_firstobj, drrfo->drr_numobjs, err); | |
3561 | break; | |
3562 | } | |
3563 | case DRR_WRITE: | |
3564 | { | |
3565 | struct drr_write *drrw = &rrd->header.drr_u.drr_write; | |
3566 | dprintf("drr_type = WRITE obj = %llu type = %u offset = %llu " | |
3567 | "lsize = %llu cksumtype = %u cksumflags = %u " | |
3568 | "compress = %u psize = %llu err = %d\n", | |
3569 | drrw->drr_object, drrw->drr_type, drrw->drr_offset, | |
3570 | drrw->drr_logical_size, drrw->drr_checksumtype, | |
b5256303 | 3571 | drrw->drr_flags, drrw->drr_compressiontype, |
6a8ee4f7 NB |
3572 | drrw->drr_compressed_size, err); |
3573 | break; | |
3574 | } | |
3575 | case DRR_WRITE_BYREF: | |
3576 | { | |
3577 | struct drr_write_byref *drrwbr = | |
3578 | &rrd->header.drr_u.drr_write_byref; | |
3579 | dprintf("drr_type = WRITE_BYREF obj = %llu offset = %llu " | |
3580 | "length = %llu toguid = %llx refguid = %llx " | |
3581 | "refobject = %llu refoffset = %llu cksumtype = %u " | |
3582 | "cksumflags = %u err = %d\n", | |
3583 | drrwbr->drr_object, drrwbr->drr_offset, | |
3584 | drrwbr->drr_length, drrwbr->drr_toguid, | |
3585 | drrwbr->drr_refguid, drrwbr->drr_refobject, | |
3586 | drrwbr->drr_refoffset, drrwbr->drr_checksumtype, | |
b5256303 | 3587 | drrwbr->drr_flags, err); |
6a8ee4f7 NB |
3588 | break; |
3589 | } | |
3590 | case DRR_WRITE_EMBEDDED: | |
3591 | { | |
3592 | struct drr_write_embedded *drrwe = | |
3593 | &rrd->header.drr_u.drr_write_embedded; | |
3594 | dprintf("drr_type = WRITE_EMBEDDED obj = %llu offset = %llu " | |
3595 | "length = %llu compress = %u etype = %u lsize = %u " | |
3596 | "psize = %u err = %d\n", | |
3597 | drrwe->drr_object, drrwe->drr_offset, drrwe->drr_length, | |
3598 | drrwe->drr_compression, drrwe->drr_etype, | |
3599 | drrwe->drr_lsize, drrwe->drr_psize, err); | |
3600 | break; | |
3601 | } | |
3602 | case DRR_FREE: | |
3603 | { | |
3604 | struct drr_free *drrf = &rrd->header.drr_u.drr_free; | |
3605 | dprintf("drr_type = FREE obj = %llu offset = %llu " | |
3606 | "length = %lld err = %d\n", | |
3607 | drrf->drr_object, drrf->drr_offset, drrf->drr_length, | |
3608 | err); | |
3609 | break; | |
3610 | } | |
3611 | case DRR_SPILL: | |
3612 | { | |
3613 | struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; | |
3614 | dprintf("drr_type = SPILL obj = %llu length = %llu " | |
3615 | "err = %d\n", drrs->drr_object, drrs->drr_length, err); | |
3616 | break; | |
3617 | } | |
3618 | default: | |
3619 | return; | |
3620 | } | |
2fd92c3d | 3621 | #endif |
6a8ee4f7 NB |
3622 | } |
3623 | ||
fcff0f35 PD |
3624 | /* |
3625 | * Commit the records to the pool. | |
3626 | */ | |
3627 | static int | |
3628 | receive_process_record(struct receive_writer_arg *rwa, | |
3629 | struct receive_record_arg *rrd) | |
3630 | { | |
3631 | int err; | |
3632 | ||
47dfff3b MA |
3633 | /* Processing in order, therefore bytes_read should be increasing. */ |
3634 | ASSERT3U(rrd->bytes_read, >=, rwa->bytes_read); | |
3635 | rwa->bytes_read = rrd->bytes_read; | |
3636 | ||
fcff0f35 PD |
3637 | switch (rrd->header.drr_type) { |
3638 | case DRR_OBJECT: | |
3639 | { | |
3640 | struct drr_object *drro = &rrd->header.drr_u.drr_object; | |
3641 | err = receive_object(rwa, drro, rrd->payload); | |
3642 | kmem_free(rrd->payload, rrd->payload_size); | |
3643 | rrd->payload = NULL; | |
6a8ee4f7 | 3644 | break; |
fcff0f35 PD |
3645 | } |
3646 | case DRR_FREEOBJECTS: | |
3647 | { | |
3648 | struct drr_freeobjects *drrfo = | |
3649 | &rrd->header.drr_u.drr_freeobjects; | |
6a8ee4f7 NB |
3650 | err = receive_freeobjects(rwa, drrfo); |
3651 | break; | |
fcff0f35 PD |
3652 | } |
3653 | case DRR_WRITE: | |
3654 | { | |
3655 | struct drr_write *drrw = &rrd->header.drr_u.drr_write; | |
b5256303 | 3656 | err = receive_write(rwa, drrw, rrd->arc_buf); |
fcff0f35 PD |
3657 | /* if receive_write() is successful, it consumes the arc_buf */ |
3658 | if (err != 0) | |
b5256303 TC |
3659 | dmu_return_arcbuf(rrd->arc_buf); |
3660 | rrd->arc_buf = NULL; | |
fcff0f35 | 3661 | rrd->payload = NULL; |
6a8ee4f7 | 3662 | break; |
fcff0f35 PD |
3663 | } |
3664 | case DRR_WRITE_BYREF: | |
3665 | { | |
3666 | struct drr_write_byref *drrwbr = | |
3667 | &rrd->header.drr_u.drr_write_byref; | |
6a8ee4f7 NB |
3668 | err = receive_write_byref(rwa, drrwbr); |
3669 | break; | |
fcff0f35 PD |
3670 | } |
3671 | case DRR_WRITE_EMBEDDED: | |
3672 | { | |
3673 | struct drr_write_embedded *drrwe = | |
3674 | &rrd->header.drr_u.drr_write_embedded; | |
3675 | err = receive_write_embedded(rwa, drrwe, rrd->payload); | |
3676 | kmem_free(rrd->payload, rrd->payload_size); | |
3677 | rrd->payload = NULL; | |
6a8ee4f7 | 3678 | break; |
fcff0f35 PD |
3679 | } |
3680 | case DRR_FREE: | |
3681 | { | |
3682 | struct drr_free *drrf = &rrd->header.drr_u.drr_free; | |
6a8ee4f7 NB |
3683 | err = receive_free(rwa, drrf); |
3684 | break; | |
fcff0f35 PD |
3685 | } |
3686 | case DRR_SPILL: | |
3687 | { | |
3688 | struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; | |
b5256303 TC |
3689 | err = receive_spill(rwa, drrs, rrd->arc_buf); |
3690 | /* if receive_spill() is successful, it consumes the arc_buf */ | |
3691 | if (err != 0) | |
3692 | dmu_return_arcbuf(rrd->arc_buf); | |
3693 | rrd->arc_buf = NULL; | |
fcff0f35 | 3694 | rrd->payload = NULL; |
6a8ee4f7 | 3695 | break; |
37f8a883 | 3696 | } |
b5256303 TC |
3697 | case DRR_OBJECT_RANGE: |
3698 | { | |
3699 | struct drr_object_range *drror = | |
3700 | &rrd->header.drr_u.drr_object_range; | |
3701 | return (receive_object_range(rwa, drror)); | |
3702 | } | |
37f8a883 MA |
3703 | default: |
3704 | return (SET_ERROR(EINVAL)); | |
3705 | } | |
6a8ee4f7 NB |
3706 | |
3707 | if (err != 0) | |
3708 | dprintf_drr(rrd, err); | |
3709 | ||
3710 | return (err); | |
37f8a883 MA |
3711 | } |
3712 | ||
34dc7c2f | 3713 | /* |
fcff0f35 PD |
3714 | * dmu_recv_stream's worker thread; pull records off the queue, and then call |
3715 | * receive_process_record When we're done, signal the main thread and exit. | |
3716 | */ | |
3717 | static void | |
3718 | receive_writer_thread(void *arg) | |
3719 | { | |
3720 | struct receive_writer_arg *rwa = arg; | |
3721 | struct receive_record_arg *rrd; | |
3e635ac1 TC |
3722 | fstrans_cookie_t cookie = spl_fstrans_mark(); |
3723 | ||
fcff0f35 PD |
3724 | for (rrd = bqueue_dequeue(&rwa->q); !rrd->eos_marker; |
3725 | rrd = bqueue_dequeue(&rwa->q)) { | |
3726 | /* | |
3727 | * If there's an error, the main thread will stop putting things | |
3728 | * on the queue, but we need to clear everything in it before we | |
3729 | * can exit. | |
3730 | */ | |
3731 | if (rwa->err == 0) { | |
3732 | rwa->err = receive_process_record(rwa, rrd); | |
b5256303 TC |
3733 | } else if (rrd->arc_buf != NULL) { |
3734 | dmu_return_arcbuf(rrd->arc_buf); | |
3735 | rrd->arc_buf = NULL; | |
fcff0f35 PD |
3736 | rrd->payload = NULL; |
3737 | } else if (rrd->payload != NULL) { | |
3738 | kmem_free(rrd->payload, rrd->payload_size); | |
3739 | rrd->payload = NULL; | |
3740 | } | |
3741 | kmem_free(rrd, sizeof (*rrd)); | |
3742 | } | |
3743 | kmem_free(rrd, sizeof (*rrd)); | |
3744 | mutex_enter(&rwa->mutex); | |
3745 | rwa->done = B_TRUE; | |
3746 | cv_signal(&rwa->cv); | |
3747 | mutex_exit(&rwa->mutex); | |
3e635ac1 | 3748 | spl_fstrans_unmark(cookie); |
34a6b428 | 3749 | thread_exit(); |
fcff0f35 PD |
3750 | } |
3751 | ||
47dfff3b MA |
3752 | static int |
3753 | resume_check(struct receive_arg *ra, nvlist_t *begin_nvl) | |
3754 | { | |
3755 | uint64_t val; | |
3756 | objset_t *mos = dmu_objset_pool(ra->os)->dp_meta_objset; | |
3757 | uint64_t dsobj = dmu_objset_id(ra->os); | |
3758 | uint64_t resume_obj, resume_off; | |
3759 | ||
3760 | if (nvlist_lookup_uint64(begin_nvl, | |
3761 | "resume_object", &resume_obj) != 0 || | |
3762 | nvlist_lookup_uint64(begin_nvl, | |
3763 | "resume_offset", &resume_off) != 0) { | |
3764 | return (SET_ERROR(EINVAL)); | |
3765 | } | |
3766 | VERIFY0(zap_lookup(mos, dsobj, | |
3767 | DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val)); | |
3768 | if (resume_obj != val) | |
3769 | return (SET_ERROR(EINVAL)); | |
3770 | VERIFY0(zap_lookup(mos, dsobj, | |
3771 | DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val)); | |
3772 | if (resume_off != val) | |
3773 | return (SET_ERROR(EINVAL)); | |
3774 | ||
3775 | return (0); | |
3776 | } | |
3777 | ||
fcff0f35 PD |
3778 | /* |
3779 | * Read in the stream's records, one by one, and apply them to the pool. There | |
3780 | * are two threads involved; the thread that calls this function will spin up a | |
3781 | * worker thread, read the records off the stream one by one, and issue | |
3782 | * prefetches for any necessary indirect blocks. It will then push the records | |
3783 | * onto an internal blocking queue. The worker thread will pull the records off | |
3784 | * the queue, and actually write the data into the DMU. This way, the worker | |
3785 | * thread doesn't have to wait for reads to complete, since everything it needs | |
3786 | * (the indirect blocks) will be prefetched. | |
3787 | * | |
34dc7c2f BB |
3788 | * NB: callers *must* call dmu_recv_end() if this succeeds. |
3789 | */ | |
3790 | int | |
572e2857 BB |
3791 | dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, |
3792 | int cleanup_fd, uint64_t *action_handlep) | |
34dc7c2f | 3793 | { |
37f8a883 | 3794 | int err = 0; |
04bc4610 NB |
3795 | struct receive_arg *ra; |
3796 | struct receive_writer_arg *rwa; | |
428870ff | 3797 | int featureflags; |
47dfff3b MA |
3798 | uint32_t payloadlen; |
3799 | void *payload; | |
3800 | nvlist_t *begin_nvl = NULL; | |
34dc7c2f | 3801 | |
04bc4610 NB |
3802 | ra = kmem_zalloc(sizeof (*ra), KM_SLEEP); |
3803 | rwa = kmem_zalloc(sizeof (*rwa), KM_SLEEP); | |
3804 | ||
3805 | ra->byteswap = drc->drc_byteswap; | |
9b840763 | 3806 | ra->raw = drc->drc_raw; |
04bc4610 NB |
3807 | ra->cksum = drc->drc_cksum; |
3808 | ra->vp = vp; | |
3809 | ra->voff = *voffp; | |
47dfff3b MA |
3810 | |
3811 | if (dsl_dataset_is_zapified(drc->drc_ds)) { | |
3812 | (void) zap_lookup(drc->drc_ds->ds_dir->dd_pool->dp_meta_objset, | |
3813 | drc->drc_ds->ds_object, DS_FIELD_RESUME_BYTES, | |
3814 | sizeof (ra->bytes_read), 1, &ra->bytes_read); | |
3815 | } | |
3816 | ||
e6d3a843 | 3817 | objlist_create(&ra->ignore_objlist); |
34dc7c2f BB |
3818 | |
3819 | /* these were verified in dmu_recv_begin */ | |
13fe0198 | 3820 | ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, |
428870ff | 3821 | DMU_SUBSTREAM); |
13fe0198 | 3822 | ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); |
34dc7c2f BB |
3823 | |
3824 | /* | |
3825 | * Open the objset we are modifying. | |
3826 | */ | |
04bc4610 | 3827 | VERIFY0(dmu_objset_from_ds(drc->drc_ds, &ra->os)); |
34dc7c2f | 3828 | |
d683ddbb | 3829 | ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); |
34dc7c2f | 3830 | |
428870ff | 3831 | featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); |
b5256303 | 3832 | ra->featureflags = featureflags; |
428870ff | 3833 | |
1fff937a TC |
3834 | ASSERT0(ra->os->os_encrypted && |
3835 | (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)); | |
9b840763 | 3836 | |
428870ff BB |
3837 | /* if this stream is dedup'ed, set up the avl tree for guid mapping */ |
3838 | if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { | |
572e2857 BB |
3839 | minor_t minor; |
3840 | ||
3841 | if (cleanup_fd == -1) { | |
9b840763 | 3842 | err = SET_ERROR(EBADF); |
572e2857 BB |
3843 | goto out; |
3844 | } | |
9b840763 TC |
3845 | err = zfs_onexit_fd_hold(cleanup_fd, &minor); |
3846 | if (err != 0) { | |
572e2857 BB |
3847 | cleanup_fd = -1; |
3848 | goto out; | |
3849 | } | |
3850 | ||
3851 | if (*action_handlep == 0) { | |
04bc4610 | 3852 | rwa->guid_to_ds_map = |
572e2857 | 3853 | kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); |
04bc4610 | 3854 | avl_create(rwa->guid_to_ds_map, guid_compare, |
572e2857 BB |
3855 | sizeof (guid_map_entry_t), |
3856 | offsetof(guid_map_entry_t, avlnode)); | |
37f8a883 | 3857 | err = zfs_onexit_add_cb(minor, |
04bc4610 | 3858 | free_guid_map_onexit, rwa->guid_to_ds_map, |
572e2857 | 3859 | action_handlep); |
9b840763 | 3860 | if (err != 0) |
572e2857 BB |
3861 | goto out; |
3862 | } else { | |
37f8a883 | 3863 | err = zfs_onexit_cb_data(minor, *action_handlep, |
04bc4610 | 3864 | (void **)&rwa->guid_to_ds_map); |
9b840763 | 3865 | if (err != 0) |
572e2857 BB |
3866 | goto out; |
3867 | } | |
8d35c149 | 3868 | |
04bc4610 | 3869 | drc->drc_guid_to_ds_map = rwa->guid_to_ds_map; |
428870ff BB |
3870 | } |
3871 | ||
47dfff3b MA |
3872 | payloadlen = drc->drc_drr_begin->drr_payloadlen; |
3873 | payload = NULL; | |
3874 | if (payloadlen != 0) | |
3875 | payload = kmem_alloc(payloadlen, KM_SLEEP); | |
3876 | ||
3877 | err = receive_read_payload_and_next_header(ra, payloadlen, payload); | |
3878 | if (err != 0) { | |
3879 | if (payloadlen != 0) | |
3880 | kmem_free(payload, payloadlen); | |
37f8a883 | 3881 | goto out; |
47dfff3b MA |
3882 | } |
3883 | if (payloadlen != 0) { | |
3884 | err = nvlist_unpack(payload, payloadlen, &begin_nvl, KM_SLEEP); | |
3885 | kmem_free(payload, payloadlen); | |
3886 | if (err != 0) | |
3887 | goto out; | |
3888 | } | |
3889 | ||
b5256303 TC |
3890 | /* handle DSL encryption key payload */ |
3891 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
3892 | nvlist_t *keynvl = NULL; | |
3893 | ||
3894 | ASSERT(ra->os->os_encrypted); | |
3895 | ASSERT(drc->drc_raw); | |
3896 | ||
3897 | err = nvlist_lookup_nvlist(begin_nvl, "crypt_keydata", &keynvl); | |
3898 | if (err != 0) | |
3899 | goto out; | |
3900 | ||
b0918402 TC |
3901 | /* |
3902 | * If this is a new dataset we set the key immediately. | |
3903 | * Otherwise we don't want to change the key until we | |
3904 | * are sure the rest of the receive succeeded so we stash | |
3905 | * the keynvl away until then. | |
3906 | */ | |
3907 | err = dsl_crypto_recv_raw(spa_name(ra->os->os_spa), | |
b5256303 | 3908 | drc->drc_ds->ds_object, drc->drc_drrb->drr_type, |
b0918402 | 3909 | keynvl, drc->drc_newfs); |
b5256303 TC |
3910 | if (err != 0) |
3911 | goto out; | |
b0918402 TC |
3912 | |
3913 | if (!drc->drc_newfs) | |
3914 | drc->drc_keynvl = fnvlist_dup(keynvl); | |
b5256303 TC |
3915 | } |
3916 | ||
47dfff3b MA |
3917 | if (featureflags & DMU_BACKUP_FEATURE_RESUMING) { |
3918 | err = resume_check(ra, begin_nvl); | |
3919 | if (err != 0) | |
3920 | goto out; | |
3921 | } | |
37f8a883 | 3922 | |
3b0d9928 BB |
3923 | (void) bqueue_init(&rwa->q, |
3924 | MAX(zfs_recv_queue_length, 2 * zfs_max_recordsize), | |
fcff0f35 | 3925 | offsetof(struct receive_record_arg, node)); |
04bc4610 NB |
3926 | cv_init(&rwa->cv, NULL, CV_DEFAULT, NULL); |
3927 | mutex_init(&rwa->mutex, NULL, MUTEX_DEFAULT, NULL); | |
3928 | rwa->os = ra->os; | |
3929 | rwa->byteswap = drc->drc_byteswap; | |
47dfff3b | 3930 | rwa->resumable = drc->drc_resumable; |
9b840763 | 3931 | rwa->raw = drc->drc_raw; |
0c03d21a | 3932 | rwa->os->os_raw_receive = drc->drc_raw; |
fcff0f35 | 3933 | |
04bc4610 | 3934 | (void) thread_create(NULL, 0, receive_writer_thread, rwa, 0, curproc, |
fcff0f35 PD |
3935 | TS_RUN, minclsyspri); |
3936 | /* | |
04bc4610 | 3937 | * We're reading rwa->err without locks, which is safe since we are the |
fcff0f35 PD |
3938 | * only reader, and the worker thread is the only writer. It's ok if we |
3939 | * miss a write for an iteration or two of the loop, since the writer | |
3940 | * thread will keep freeing records we send it until we send it an eos | |
3941 | * marker. | |
3942 | * | |
04bc4610 | 3943 | * We can leave this loop in 3 ways: First, if rwa->err is |
fcff0f35 PD |
3944 | * non-zero. In that case, the writer thread will free the rrd we just |
3945 | * pushed. Second, if we're interrupted; in that case, either it's the | |
47dfff3b | 3946 | * first loop and ra->rrd was never allocated, or it's later and ra->rrd |
fcff0f35 PD |
3947 | * has been handed off to the writer thread who will free it. Finally, |
3948 | * if receive_read_record fails or we're at the end of the stream, then | |
04bc4610 | 3949 | * we free ra->rrd and exit. |
fcff0f35 | 3950 | */ |
04bc4610 | 3951 | while (rwa->err == 0) { |
34dc7c2f | 3952 | if (issig(JUSTLOOKING) && issig(FORREAL)) { |
37f8a883 MA |
3953 | err = SET_ERROR(EINTR); |
3954 | break; | |
34dc7c2f BB |
3955 | } |
3956 | ||
04bc4610 NB |
3957 | ASSERT3P(ra->rrd, ==, NULL); |
3958 | ra->rrd = ra->next_rrd; | |
3959 | ra->next_rrd = NULL; | |
3960 | /* Allocates and loads header into ra->next_rrd */ | |
3961 | err = receive_read_record(ra); | |
34dc7c2f | 3962 | |
04bc4610 NB |
3963 | if (ra->rrd->header.drr_type == DRR_END || err != 0) { |
3964 | kmem_free(ra->rrd, sizeof (*ra->rrd)); | |
3965 | ra->rrd = NULL; | |
428870ff | 3966 | break; |
fcff0f35 PD |
3967 | } |
3968 | ||
04bc4610 NB |
3969 | bqueue_enqueue(&rwa->q, ra->rrd, |
3970 | sizeof (struct receive_record_arg) + ra->rrd->payload_size); | |
3971 | ra->rrd = NULL; | |
fcff0f35 | 3972 | } |
04bc4610 NB |
3973 | if (ra->next_rrd == NULL) |
3974 | ra->next_rrd = kmem_zalloc(sizeof (*ra->next_rrd), KM_SLEEP); | |
3975 | ra->next_rrd->eos_marker = B_TRUE; | |
3976 | bqueue_enqueue(&rwa->q, ra->next_rrd, 1); | |
fcff0f35 | 3977 | |
04bc4610 NB |
3978 | mutex_enter(&rwa->mutex); |
3979 | while (!rwa->done) { | |
3980 | cv_wait(&rwa->cv, &rwa->mutex); | |
34dc7c2f | 3981 | } |
04bc4610 | 3982 | mutex_exit(&rwa->mutex); |
fcff0f35 | 3983 | |
48fbb9dd FG |
3984 | /* |
3985 | * If we are receiving a full stream as a clone, all object IDs which | |
3986 | * are greater than the maximum ID referenced in the stream are | |
3987 | * by definition unused and must be freed. | |
3988 | */ | |
3989 | if (drc->drc_clone && drc->drc_drrb->drr_fromguid == 0) { | |
3990 | uint64_t obj = rwa->max_object + 1; | |
3991 | int free_err = 0; | |
3992 | int next_err = 0; | |
3993 | ||
3994 | while (next_err == 0) { | |
0c03d21a | 3995 | free_err = dmu_free_long_object(rwa->os, obj); |
48fbb9dd FG |
3996 | if (free_err != 0 && free_err != ENOENT) |
3997 | break; | |
3998 | ||
3999 | next_err = dmu_object_next(rwa->os, &obj, FALSE, 0); | |
4000 | } | |
4001 | ||
4002 | if (err == 0) { | |
4003 | if (free_err != 0 && free_err != ENOENT) | |
4004 | err = free_err; | |
4005 | else if (next_err != ESRCH) | |
4006 | err = next_err; | |
4007 | } | |
4008 | } | |
4009 | ||
04bc4610 NB |
4010 | cv_destroy(&rwa->cv); |
4011 | mutex_destroy(&rwa->mutex); | |
4012 | bqueue_destroy(&rwa->q); | |
fcff0f35 | 4013 | if (err == 0) |
04bc4610 | 4014 | err = rwa->err; |
34dc7c2f BB |
4015 | |
4016 | out: | |
47dfff3b | 4017 | nvlist_free(begin_nvl); |
572e2857 BB |
4018 | if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) |
4019 | zfs_onexit_fd_rele(cleanup_fd); | |
4020 | ||
37f8a883 | 4021 | if (err != 0) { |
34dc7c2f | 4022 | /* |
47dfff3b MA |
4023 | * Clean up references. If receive is not resumable, |
4024 | * destroy what we created, so we don't leave it in | |
4025 | * the inconsistent state. | |
34dc7c2f | 4026 | */ |
13fe0198 | 4027 | dmu_recv_cleanup_ds(drc); |
b0918402 | 4028 | nvlist_free(drc->drc_keynvl); |
34dc7c2f BB |
4029 | } |
4030 | ||
04bc4610 | 4031 | *voffp = ra->voff; |
e6d3a843 | 4032 | objlist_destroy(&ra->ignore_objlist); |
04bc4610 NB |
4033 | kmem_free(ra, sizeof (*ra)); |
4034 | kmem_free(rwa, sizeof (*rwa)); | |
37f8a883 | 4035 | return (err); |
34dc7c2f BB |
4036 | } |
4037 | ||
34dc7c2f | 4038 | static int |
13fe0198 | 4039 | dmu_recv_end_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 4040 | { |
13fe0198 MA |
4041 | dmu_recv_cookie_t *drc = arg; |
4042 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
4043 | int error; | |
4044 | ||
4045 | ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); | |
4046 | ||
4047 | if (!drc->drc_newfs) { | |
4048 | dsl_dataset_t *origin_head; | |
4049 | ||
4050 | error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); | |
4051 | if (error != 0) | |
4052 | return (error); | |
19580676 MA |
4053 | if (drc->drc_force) { |
4054 | /* | |
4055 | * We will destroy any snapshots in tofs (i.e. before | |
4056 | * origin_head) that are after the origin (which is | |
4057 | * the snap before drc_ds, because drc_ds can not | |
4058 | * have any snaps of its own). | |
4059 | */ | |
d683ddbb JG |
4060 | uint64_t obj; |
4061 | ||
4062 | obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
4063 | while (obj != | |
4064 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { | |
19580676 MA |
4065 | dsl_dataset_t *snap; |
4066 | error = dsl_dataset_hold_obj(dp, obj, FTAG, | |
4067 | &snap); | |
4068 | if (error != 0) | |
b6640117 | 4069 | break; |
19580676 MA |
4070 | if (snap->ds_dir != origin_head->ds_dir) |
4071 | error = SET_ERROR(EINVAL); | |
4072 | if (error == 0) { | |
4073 | error = dsl_destroy_snapshot_check_impl( | |
4074 | snap, B_FALSE); | |
4075 | } | |
d683ddbb | 4076 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; |
19580676 MA |
4077 | dsl_dataset_rele(snap, FTAG); |
4078 | if (error != 0) | |
b6640117 AG |
4079 | break; |
4080 | } | |
4081 | if (error != 0) { | |
4082 | dsl_dataset_rele(origin_head, FTAG); | |
4083 | return (error); | |
19580676 MA |
4084 | } |
4085 | } | |
b0918402 TC |
4086 | if (drc->drc_keynvl != NULL) { |
4087 | error = dsl_crypto_recv_raw_key_check(drc->drc_ds, | |
4088 | drc->drc_keynvl, tx); | |
4089 | if (error != 0) { | |
4090 | dsl_dataset_rele(origin_head, FTAG); | |
4091 | return (error); | |
4092 | } | |
4093 | } | |
4094 | ||
13fe0198 | 4095 | error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, |
831baf06 | 4096 | origin_head, drc->drc_force, drc->drc_owner, tx); |
13fe0198 MA |
4097 | if (error != 0) { |
4098 | dsl_dataset_rele(origin_head, FTAG); | |
4099 | return (error); | |
4100 | } | |
4101 | error = dsl_dataset_snapshot_check_impl(origin_head, | |
788eb90c | 4102 | drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); |
13fe0198 MA |
4103 | dsl_dataset_rele(origin_head, FTAG); |
4104 | if (error != 0) | |
4105 | return (error); | |
34dc7c2f | 4106 | |
13fe0198 MA |
4107 | error = dsl_destroy_head_check_impl(drc->drc_ds, 1); |
4108 | } else { | |
4109 | error = dsl_dataset_snapshot_check_impl(drc->drc_ds, | |
788eb90c | 4110 | drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); |
13fe0198 MA |
4111 | } |
4112 | return (error); | |
34dc7c2f BB |
4113 | } |
4114 | ||
4115 | static void | |
13fe0198 | 4116 | dmu_recv_end_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 4117 | { |
13fe0198 MA |
4118 | dmu_recv_cookie_t *drc = arg; |
4119 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
b5256303 | 4120 | boolean_t encrypted = drc->drc_ds->ds_dir->dd_crypto_obj != 0; |
13fe0198 MA |
4121 | |
4122 | spa_history_log_internal_ds(drc->drc_ds, "finish receiving", | |
4123 | tx, "snap=%s", drc->drc_tosnap); | |
0c03d21a | 4124 | drc->drc_ds->ds_objset->os_raw_receive = B_FALSE; |
13fe0198 MA |
4125 | |
4126 | if (!drc->drc_newfs) { | |
4127 | dsl_dataset_t *origin_head; | |
4128 | ||
4129 | VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, | |
4130 | &origin_head)); | |
19580676 MA |
4131 | |
4132 | if (drc->drc_force) { | |
4133 | /* | |
4134 | * Destroy any snapshots of drc_tofs (origin_head) | |
4135 | * after the origin (the snap before drc_ds). | |
4136 | */ | |
d683ddbb JG |
4137 | uint64_t obj; |
4138 | ||
4139 | obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
4140 | while (obj != | |
4141 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { | |
19580676 MA |
4142 | dsl_dataset_t *snap; |
4143 | VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, | |
4144 | &snap)); | |
4145 | ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); | |
d683ddbb | 4146 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; |
19580676 MA |
4147 | dsl_destroy_snapshot_sync_impl(snap, |
4148 | B_FALSE, tx); | |
4149 | dsl_dataset_rele(snap, FTAG); | |
4150 | } | |
4151 | } | |
b0918402 TC |
4152 | if (drc->drc_keynvl != NULL) { |
4153 | dsl_crypto_recv_raw_key_sync(drc->drc_ds, | |
4154 | drc->drc_keynvl, tx); | |
4155 | nvlist_free(drc->drc_keynvl); | |
4156 | drc->drc_keynvl = NULL; | |
4157 | } | |
4158 | ||
4159 | VERIFY3P(drc->drc_ds->ds_prev, ==, origin_head->ds_prev); | |
19580676 | 4160 | |
13fe0198 MA |
4161 | dsl_dataset_clone_swap_sync_impl(drc->drc_ds, |
4162 | origin_head, tx); | |
4163 | dsl_dataset_snapshot_sync_impl(origin_head, | |
4164 | drc->drc_tosnap, tx); | |
4165 | ||
4166 | /* set snapshot's creation time and guid */ | |
4167 | dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); | |
d683ddbb | 4168 | dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = |
13fe0198 | 4169 | drc->drc_drrb->drr_creation_time; |
d683ddbb | 4170 | dsl_dataset_phys(origin_head->ds_prev)->ds_guid = |
13fe0198 | 4171 | drc->drc_drrb->drr_toguid; |
d683ddbb | 4172 | dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= |
13fe0198 MA |
4173 | ~DS_FLAG_INCONSISTENT; |
4174 | ||
4175 | dmu_buf_will_dirty(origin_head->ds_dbuf, tx); | |
d683ddbb JG |
4176 | dsl_dataset_phys(origin_head)->ds_flags &= |
4177 | ~DS_FLAG_INCONSISTENT; | |
13fe0198 | 4178 | |
2e0e443a GM |
4179 | drc->drc_newsnapobj = |
4180 | dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
4181 | ||
13fe0198 MA |
4182 | dsl_dataset_rele(origin_head, FTAG); |
4183 | dsl_destroy_head_sync_impl(drc->drc_ds, tx); | |
831baf06 KW |
4184 | |
4185 | if (drc->drc_owner != NULL) | |
4186 | VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); | |
13fe0198 MA |
4187 | } else { |
4188 | dsl_dataset_t *ds = drc->drc_ds; | |
34dc7c2f | 4189 | |
13fe0198 | 4190 | dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); |
34dc7c2f | 4191 | |
13fe0198 MA |
4192 | /* set snapshot's creation time and guid */ |
4193 | dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
d683ddbb | 4194 | dsl_dataset_phys(ds->ds_prev)->ds_creation_time = |
13fe0198 | 4195 | drc->drc_drrb->drr_creation_time; |
d683ddbb JG |
4196 | dsl_dataset_phys(ds->ds_prev)->ds_guid = |
4197 | drc->drc_drrb->drr_toguid; | |
4198 | dsl_dataset_phys(ds->ds_prev)->ds_flags &= | |
4199 | ~DS_FLAG_INCONSISTENT; | |
34dc7c2f | 4200 | |
13fe0198 | 4201 | dmu_buf_will_dirty(ds->ds_dbuf, tx); |
d683ddbb | 4202 | dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; |
47dfff3b MA |
4203 | if (dsl_dataset_has_resume_receive_state(ds)) { |
4204 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
4205 | DS_FIELD_RESUME_FROMGUID, tx); | |
4206 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
4207 | DS_FIELD_RESUME_OBJECT, tx); | |
4208 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
4209 | DS_FIELD_RESUME_OFFSET, tx); | |
4210 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
4211 | DS_FIELD_RESUME_BYTES, tx); | |
4212 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
4213 | DS_FIELD_RESUME_TOGUID, tx); | |
4214 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
4215 | DS_FIELD_RESUME_TONAME, tx); | |
4216 | } | |
2e0e443a GM |
4217 | drc->drc_newsnapobj = |
4218 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; | |
13fe0198 | 4219 | } |
a0bd735a | 4220 | zvol_create_minors(dp->dp_spa, drc->drc_tofs, B_TRUE); |
b5256303 | 4221 | |
13fe0198 MA |
4222 | /* |
4223 | * Release the hold from dmu_recv_begin. This must be done before | |
b5256303 TC |
4224 | * we return to open context, so that when we free the dataset's dnode |
4225 | * we can evict its bonus buffer. Since the dataset may be destroyed | |
4226 | * at this point (and therefore won't have a valid pointer to the spa) | |
4227 | * we release the key mapping manually here while we do have a valid | |
4228 | * pointer, if it exists. | |
13fe0198 | 4229 | */ |
b5256303 TC |
4230 | if (!drc->drc_raw && encrypted) { |
4231 | (void) spa_keystore_remove_mapping(dmu_tx_pool(tx)->dp_spa, | |
4232 | drc->drc_ds->ds_object, drc->drc_ds); | |
4233 | } | |
4234 | dsl_dataset_disown(drc->drc_ds, 0, dmu_recv_tag); | |
13fe0198 | 4235 | drc->drc_ds = NULL; |
34dc7c2f BB |
4236 | } |
4237 | ||
8d35c149 | 4238 | static int |
b5256303 TC |
4239 | add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj, |
4240 | boolean_t raw) | |
8d35c149 | 4241 | { |
13fe0198 | 4242 | dsl_pool_t *dp; |
8d35c149 AS |
4243 | dsl_dataset_t *snapds; |
4244 | guid_map_entry_t *gmep; | |
149ce888 | 4245 | objset_t *os; |
b5256303 | 4246 | ds_hold_flags_t dsflags = (raw) ? 0 : DS_HOLD_FLAG_DECRYPT; |
8d35c149 AS |
4247 | int err; |
4248 | ||
4249 | ASSERT(guid_map != NULL); | |
4250 | ||
13fe0198 MA |
4251 | err = dsl_pool_hold(name, FTAG, &dp); |
4252 | if (err != 0) | |
4253 | return (err); | |
7ec09286 | 4254 | gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP); |
149ce888 | 4255 | err = dsl_dataset_own_obj(dp, snapobj, dsflags, gmep, &snapds); |
8d35c149 | 4256 | if (err == 0) { |
149ce888 TC |
4257 | /* |
4258 | * If this is a deduplicated raw send stream, we need | |
4259 | * to make sure that we can still read raw blocks from | |
4260 | * earlier datasets in the stream, so we set the | |
4261 | * os_raw_receive flag now. | |
4262 | */ | |
4263 | if (raw) { | |
4264 | err = dmu_objset_from_ds(snapds, &os); | |
4265 | if (err != 0) { | |
4266 | dsl_dataset_disown(snapds, dsflags, FTAG); | |
4267 | dsl_pool_rele(dp, FTAG); | |
4268 | kmem_free(gmep, sizeof (*gmep)); | |
4269 | return (err); | |
4270 | } | |
4271 | os->os_raw_receive = B_TRUE; | |
4272 | } | |
4273 | ||
b5256303 | 4274 | gmep->raw = raw; |
149ce888 | 4275 | gmep->guid = dsl_dataset_phys(snapds)->ds_guid; |
8d35c149 AS |
4276 | gmep->gme_ds = snapds; |
4277 | avl_add(guid_map, gmep); | |
7ec09286 MA |
4278 | } else { |
4279 | kmem_free(gmep, sizeof (*gmep)); | |
8d35c149 AS |
4280 | } |
4281 | ||
13fe0198 | 4282 | dsl_pool_rele(dp, FTAG); |
8d35c149 AS |
4283 | return (err); |
4284 | } | |
4285 | ||
13fe0198 MA |
4286 | static int dmu_recv_end_modified_blocks = 3; |
4287 | ||
428870ff BB |
4288 | static int |
4289 | dmu_recv_existing_end(dmu_recv_cookie_t *drc) | |
34dc7c2f | 4290 | { |
13fe0198 | 4291 | #ifdef _KERNEL |
13fe0198 MA |
4292 | /* |
4293 | * We will be destroying the ds; make sure its origin is unmounted if | |
4294 | * necessary. | |
4295 | */ | |
eca7b760 | 4296 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
13fe0198 MA |
4297 | dsl_dataset_name(drc->drc_ds, name); |
4298 | zfs_destroy_unmount_origin(name); | |
13fe0198 | 4299 | #endif |
34dc7c2f | 4300 | |
2e0e443a | 4301 | return (dsl_sync_task(drc->drc_tofs, |
13fe0198 | 4302 | dmu_recv_end_check, dmu_recv_end_sync, drc, |
2e0e443a | 4303 | dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); |
34dc7c2f | 4304 | } |
428870ff BB |
4305 | |
4306 | static int | |
4307 | dmu_recv_new_end(dmu_recv_cookie_t *drc) | |
2e0e443a GM |
4308 | { |
4309 | return (dsl_sync_task(drc->drc_tofs, | |
4310 | dmu_recv_end_check, dmu_recv_end_sync, drc, | |
4311 | dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); | |
4312 | } | |
4313 | ||
4314 | int | |
4315 | dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) | |
428870ff | 4316 | { |
13fe0198 | 4317 | int error; |
428870ff | 4318 | |
2e0e443a GM |
4319 | drc->drc_owner = owner; |
4320 | ||
4321 | if (drc->drc_newfs) | |
4322 | error = dmu_recv_new_end(drc); | |
4323 | else | |
4324 | error = dmu_recv_existing_end(drc); | |
428870ff | 4325 | |
13fe0198 MA |
4326 | if (error != 0) { |
4327 | dmu_recv_cleanup_ds(drc); | |
b0918402 | 4328 | nvlist_free(drc->drc_keynvl); |
13fe0198 | 4329 | } else if (drc->drc_guid_to_ds_map != NULL) { |
b5256303 TC |
4330 | (void) add_ds_to_guidmap(drc->drc_tofs, drc->drc_guid_to_ds_map, |
4331 | drc->drc_newsnapobj, drc->drc_raw); | |
428870ff | 4332 | } |
13fe0198 | 4333 | return (error); |
428870ff BB |
4334 | } |
4335 | ||
ea97f8ce MA |
4336 | /* |
4337 | * Return TRUE if this objset is currently being received into. | |
4338 | */ | |
4339 | boolean_t | |
4340 | dmu_objset_is_receiving(objset_t *os) | |
4341 | { | |
4342 | return (os->os_dsl_dataset != NULL && | |
4343 | os->os_dsl_dataset->ds_owner == dmu_recv_tag); | |
4344 | } | |
fd8febbd TF |
4345 | |
4346 | #if defined(_KERNEL) | |
ca0845d5 PD |
4347 | /* BEGIN CSTYLED */ |
4348 | module_param(zfs_override_estimate_recordsize, ulong, 0644); | |
4349 | MODULE_PARM_DESC(zfs_override_estimate_recordsize, | |
4350 | "Record size calculation override for zfs send estimates"); | |
4351 | /* END CSTYLED */ | |
4352 | ||
fd8febbd TF |
4353 | module_param(zfs_send_corrupt_data, int, 0644); |
4354 | MODULE_PARM_DESC(zfs_send_corrupt_data, "Allow sending corrupt data"); | |
3b0d9928 BB |
4355 | |
4356 | module_param(zfs_send_queue_length, int, 0644); | |
4357 | MODULE_PARM_DESC(zfs_send_queue_length, "Maximum send queue length"); | |
4358 | ||
4359 | module_param(zfs_recv_queue_length, int, 0644); | |
4360 | MODULE_PARM_DESC(zfs_recv_queue_length, "Maximum receive queue length"); | |
fd8febbd | 4361 | #endif |