]>
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; | |
fcff0f35 PD |
64 | int zfs_send_queue_length = 16 * 1024 * 1024; |
65 | int zfs_recv_queue_length = 16 * 1024 * 1024; | |
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 | |
fcff0f35 PD |
72 | #define BP_SPAN(datablkszsec, indblkshift, level) \ |
73 | (((uint64_t)datablkszsec) << (SPA_MINBLOCKSHIFT + \ | |
74 | (level) * (indblkshift - SPA_BLKPTRSHIFT))) | |
75 | ||
47dfff3b MA |
76 | static void byteswap_record(dmu_replay_record_t *drr); |
77 | ||
fcff0f35 PD |
78 | struct send_thread_arg { |
79 | bqueue_t q; | |
80 | dsl_dataset_t *ds; /* Dataset to traverse */ | |
81 | uint64_t fromtxg; /* Traverse from this txg */ | |
82 | int flags; /* flags to pass to traverse_dataset */ | |
83 | int error_code; | |
84 | boolean_t cancel; | |
47dfff3b | 85 | zbookmark_phys_t resume; |
fcff0f35 PD |
86 | }; |
87 | ||
88 | struct send_block_record { | |
89 | boolean_t eos_marker; /* Marks the end of the stream */ | |
90 | blkptr_t bp; | |
91 | zbookmark_phys_t zb; | |
92 | uint8_t indblkshift; | |
93 | uint16_t datablkszsec; | |
94 | bqueue_node_t ln; | |
95 | }; | |
96 | ||
044baf00 BB |
97 | typedef struct dump_bytes_io { |
98 | dmu_sendarg_t *dbi_dsp; | |
99 | void *dbi_buf; | |
100 | int dbi_len; | |
101 | } dump_bytes_io_t; | |
102 | ||
103 | static void | |
b58986ee | 104 | dump_bytes_cb(void *arg) |
34dc7c2f | 105 | { |
044baf00 BB |
106 | dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg; |
107 | dmu_sendarg_t *dsp = dbi->dbi_dsp; | |
47dfff3b | 108 | dsl_dataset_t *ds = dmu_objset_ds(dsp->dsa_os); |
34dc7c2f | 109 | ssize_t resid; /* have to get resid to get detailed errno */ |
f8866f8a ER |
110 | |
111 | /* | |
112 | * The code does not rely on this (len being a multiple of 8). We keep | |
113 | * this assertion because of the corresponding assertion in | |
114 | * receive_read(). Keeping this assertion ensures that we do not | |
115 | * inadvertently break backwards compatibility (causing the assertion | |
116 | * in receive_read() to trigger on old software). | |
117 | * | |
118 | * Removing the assertions could be rolled into a new feature that uses | |
119 | * data that isn't 8-byte aligned; if the assertions were removed, a | |
120 | * feature flag would have to be added. | |
121 | */ | |
122 | ||
c99c9001 | 123 | ASSERT0(dbi->dbi_len % 8); |
34dc7c2f | 124 | |
37abac6d | 125 | dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp, |
044baf00 | 126 | (caddr_t)dbi->dbi_buf, dbi->dbi_len, |
34dc7c2f | 127 | 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid); |
37abac6d BP |
128 | |
129 | mutex_enter(&ds->ds_sendstream_lock); | |
044baf00 | 130 | *dsp->dsa_off += dbi->dbi_len; |
37abac6d | 131 | mutex_exit(&ds->ds_sendstream_lock); |
044baf00 BB |
132 | } |
133 | ||
134 | static int | |
135 | dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) | |
136 | { | |
137 | dump_bytes_io_t dbi; | |
138 | ||
139 | dbi.dbi_dsp = dsp; | |
140 | dbi.dbi_buf = buf; | |
141 | dbi.dbi_len = len; | |
142 | ||
b58986ee BB |
143 | #if defined(HAVE_LARGE_STACKS) |
144 | dump_bytes_cb(&dbi); | |
145 | #else | |
044baf00 BB |
146 | /* |
147 | * The vn_rdwr() call is performed in a taskq to ensure that there is | |
148 | * always enough stack space to write safely to the target filesystem. | |
149 | * The ZIO_TYPE_FREE threads are used because there can be a lot of | |
150 | * them and they are used in vdev_file.c for a similar purpose. | |
151 | */ | |
152 | spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE, | |
b58986ee BB |
153 | ZIO_TASKQ_ISSUE, dump_bytes_cb, &dbi, TQ_SLEEP); |
154 | #endif /* HAVE_LARGE_STACKS */ | |
37abac6d BP |
155 | |
156 | return (dsp->dsa_err); | |
34dc7c2f BB |
157 | } |
158 | ||
37f8a883 MA |
159 | /* |
160 | * For all record types except BEGIN, fill in the checksum (overlaid in | |
161 | * drr_u.drr_checksum.drr_checksum). The checksum verifies everything | |
162 | * up to the start of the checksum itself. | |
163 | */ | |
164 | static int | |
165 | dump_record(dmu_sendarg_t *dsp, void *payload, int payload_len) | |
166 | { | |
167 | ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), | |
168 | ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); | |
a6255b7f | 169 | (void) fletcher_4_incremental_native(dsp->dsa_drr, |
37f8a883 MA |
170 | offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), |
171 | &dsp->dsa_zc); | |
51907a31 K |
172 | if (dsp->dsa_drr->drr_type == DRR_BEGIN) { |
173 | dsp->dsa_sent_begin = B_TRUE; | |
174 | } else { | |
37f8a883 MA |
175 | ASSERT(ZIO_CHECKSUM_IS_ZERO(&dsp->dsa_drr->drr_u. |
176 | drr_checksum.drr_checksum)); | |
177 | dsp->dsa_drr->drr_u.drr_checksum.drr_checksum = dsp->dsa_zc; | |
178 | } | |
51907a31 K |
179 | if (dsp->dsa_drr->drr_type == DRR_END) { |
180 | dsp->dsa_sent_end = B_TRUE; | |
181 | } | |
a6255b7f | 182 | (void) fletcher_4_incremental_native(&dsp->dsa_drr-> |
37f8a883 MA |
183 | drr_u.drr_checksum.drr_checksum, |
184 | sizeof (zio_cksum_t), &dsp->dsa_zc); | |
185 | if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) | |
186 | return (SET_ERROR(EINTR)); | |
187 | if (payload_len != 0) { | |
a6255b7f | 188 | (void) fletcher_4_incremental_native(payload, payload_len, |
37f8a883 MA |
189 | &dsp->dsa_zc); |
190 | if (dump_bytes(dsp, payload, payload_len) != 0) | |
191 | return (SET_ERROR(EINTR)); | |
192 | } | |
193 | return (0); | |
194 | } | |
195 | ||
e6d3a843 PD |
196 | /* |
197 | * Fill in the drr_free struct, or perform aggregation if the previous record is | |
198 | * also a free record, and the two are adjacent. | |
199 | * | |
200 | * Note that we send free records even for a full send, because we want to be | |
201 | * able to receive a full send as a clone, which requires a list of all the free | |
202 | * and freeobject records that were generated on the source. | |
203 | */ | |
34dc7c2f | 204 | static int |
37abac6d | 205 | dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, |
34dc7c2f BB |
206 | uint64_t length) |
207 | { | |
37abac6d | 208 | struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free); |
428870ff | 209 | |
ea97f8ce MA |
210 | /* |
211 | * When we receive a free record, dbuf_free_range() assumes | |
212 | * that the receiving system doesn't have any dbufs in the range | |
213 | * being freed. This is always true because there is a one-record | |
214 | * constraint: we only send one WRITE record for any given | |
47dfff3b | 215 | * object,offset. We know that the one-record constraint is |
ea97f8ce MA |
216 | * true because we always send data in increasing order by |
217 | * object,offset. | |
218 | * | |
219 | * If the increasing-order constraint ever changes, we should find | |
220 | * another way to assert that the one-record constraint is still | |
221 | * satisfied. | |
222 | */ | |
223 | ASSERT(object > dsp->dsa_last_data_object || | |
224 | (object == dsp->dsa_last_data_object && | |
225 | offset > dsp->dsa_last_data_offset)); | |
226 | ||
c578f007 SK |
227 | if (length != -1ULL && offset + length < offset) |
228 | length = -1ULL; | |
229 | ||
428870ff BB |
230 | /* |
231 | * If there is a pending op, but it's not PENDING_FREE, push it out, | |
232 | * since free block aggregation can only be done for blocks of the | |
233 | * same type (i.e., DRR_FREE records can only be aggregated with | |
234 | * other DRR_FREE records. DRR_FREEOBJECTS records can only be | |
235 | * aggregated with other DRR_FREEOBJECTS records. | |
236 | */ | |
37abac6d BP |
237 | if (dsp->dsa_pending_op != PENDING_NONE && |
238 | dsp->dsa_pending_op != PENDING_FREE) { | |
37f8a883 | 239 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 240 | return (SET_ERROR(EINTR)); |
37abac6d | 241 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff BB |
242 | } |
243 | ||
37abac6d | 244 | if (dsp->dsa_pending_op == PENDING_FREE) { |
428870ff BB |
245 | /* |
246 | * There should never be a PENDING_FREE if length is -1 | |
247 | * (because dump_dnode is the only place where this | |
248 | * function is called with a -1, and only after flushing | |
249 | * any pending record). | |
250 | */ | |
251 | ASSERT(length != -1ULL); | |
252 | /* | |
253 | * Check to see whether this free block can be aggregated | |
254 | * with pending one. | |
255 | */ | |
256 | if (drrf->drr_object == object && drrf->drr_offset + | |
257 | drrf->drr_length == offset) { | |
258 | drrf->drr_length += length; | |
259 | return (0); | |
260 | } else { | |
261 | /* not a continuation. Push out pending record */ | |
37f8a883 | 262 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 263 | return (SET_ERROR(EINTR)); |
37abac6d | 264 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff BB |
265 | } |
266 | } | |
267 | /* create a FREE record and make it pending */ | |
37abac6d BP |
268 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); |
269 | dsp->dsa_drr->drr_type = DRR_FREE; | |
428870ff BB |
270 | drrf->drr_object = object; |
271 | drrf->drr_offset = offset; | |
272 | drrf->drr_length = length; | |
37abac6d | 273 | drrf->drr_toguid = dsp->dsa_toguid; |
428870ff | 274 | if (length == -1ULL) { |
37f8a883 | 275 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 276 | return (SET_ERROR(EINTR)); |
428870ff | 277 | } else { |
37abac6d | 278 | dsp->dsa_pending_op = PENDING_FREE; |
428870ff | 279 | } |
34dc7c2f | 280 | |
34dc7c2f BB |
281 | return (0); |
282 | } | |
283 | ||
284 | static int | |
9b67f605 | 285 | dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type, |
2aa34383 DK |
286 | uint64_t object, uint64_t offset, int lsize, int psize, const blkptr_t *bp, |
287 | void *data) | |
34dc7c2f | 288 | { |
2aa34383 | 289 | uint64_t payload_size; |
37abac6d | 290 | struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write); |
428870ff | 291 | |
ea97f8ce MA |
292 | /* |
293 | * We send data in increasing object, offset order. | |
294 | * See comment in dump_free() for details. | |
295 | */ | |
296 | ASSERT(object > dsp->dsa_last_data_object || | |
297 | (object == dsp->dsa_last_data_object && | |
298 | offset > dsp->dsa_last_data_offset)); | |
299 | dsp->dsa_last_data_object = object; | |
2aa34383 | 300 | dsp->dsa_last_data_offset = offset + lsize - 1; |
428870ff BB |
301 | |
302 | /* | |
303 | * If there is any kind of pending aggregation (currently either | |
304 | * a grouping of free objects or free blocks), push it out to | |
305 | * the stream, since aggregation can't be done across operations | |
306 | * of different types. | |
307 | */ | |
37abac6d | 308 | if (dsp->dsa_pending_op != PENDING_NONE) { |
37f8a883 | 309 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 310 | return (SET_ERROR(EINTR)); |
37abac6d | 311 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff | 312 | } |
37f8a883 | 313 | /* write a WRITE record */ |
37abac6d BP |
314 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); |
315 | dsp->dsa_drr->drr_type = DRR_WRITE; | |
428870ff BB |
316 | drrw->drr_object = object; |
317 | drrw->drr_type = type; | |
318 | drrw->drr_offset = offset; | |
37abac6d | 319 | drrw->drr_toguid = dsp->dsa_toguid; |
2aa34383 DK |
320 | drrw->drr_logical_size = lsize; |
321 | ||
322 | /* only set the compression fields if the buf is compressed */ | |
323 | if (lsize != psize) { | |
324 | ASSERT(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_COMPRESSED); | |
325 | ASSERT(!BP_IS_EMBEDDED(bp)); | |
326 | ASSERT(!BP_SHOULD_BYTESWAP(bp)); | |
327 | ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp))); | |
328 | ASSERT3U(BP_GET_COMPRESS(bp), !=, ZIO_COMPRESS_OFF); | |
329 | ASSERT3S(psize, >, 0); | |
330 | ASSERT3S(lsize, >=, psize); | |
331 | ||
332 | drrw->drr_compressiontype = BP_GET_COMPRESS(bp); | |
333 | drrw->drr_compressed_size = psize; | |
334 | payload_size = drrw->drr_compressed_size; | |
335 | } else { | |
336 | payload_size = drrw->drr_logical_size; | |
337 | } | |
338 | ||
f1512ee6 | 339 | if (bp == NULL || BP_IS_EMBEDDED(bp)) { |
9b67f605 | 340 | /* |
f1512ee6 MA |
341 | * There's no pre-computed checksum for partial-block |
342 | * writes or embedded BP's, so (like | |
343 | * fletcher4-checkummed blocks) userland will have to | |
344 | * compute a dedup-capable checksum itself. | |
9b67f605 MA |
345 | */ |
346 | drrw->drr_checksumtype = ZIO_CHECKSUM_OFF; | |
347 | } else { | |
348 | drrw->drr_checksumtype = BP_GET_CHECKSUM(bp); | |
3c67d83a TH |
349 | if (zio_checksum_table[drrw->drr_checksumtype].ci_flags & |
350 | ZCHECKSUM_FLAG_DEDUP) | |
9b67f605 MA |
351 | drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP; |
352 | DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp)); | |
353 | DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp)); | |
354 | DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp)); | |
355 | drrw->drr_key.ddk_cksum = bp->blk_cksum; | |
356 | } | |
428870ff | 357 | |
2aa34383 | 358 | if (dump_record(dsp, data, payload_size) != 0) |
2e528b49 | 359 | return (SET_ERROR(EINTR)); |
428870ff BB |
360 | return (0); |
361 | } | |
362 | ||
9b67f605 MA |
363 | static int |
364 | dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, | |
365 | int blksz, const blkptr_t *bp) | |
366 | { | |
367 | char buf[BPE_PAYLOAD_SIZE]; | |
368 | struct drr_write_embedded *drrw = | |
369 | &(dsp->dsa_drr->drr_u.drr_write_embedded); | |
370 | ||
371 | if (dsp->dsa_pending_op != PENDING_NONE) { | |
37f8a883 | 372 | if (dump_record(dsp, NULL, 0) != 0) |
ecb2b7dc | 373 | return (SET_ERROR(EINTR)); |
9b67f605 MA |
374 | dsp->dsa_pending_op = PENDING_NONE; |
375 | } | |
376 | ||
377 | ASSERT(BP_IS_EMBEDDED(bp)); | |
378 | ||
379 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); | |
380 | dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED; | |
381 | drrw->drr_object = object; | |
382 | drrw->drr_offset = offset; | |
383 | drrw->drr_length = blksz; | |
384 | drrw->drr_toguid = dsp->dsa_toguid; | |
385 | drrw->drr_compression = BP_GET_COMPRESS(bp); | |
386 | drrw->drr_etype = BPE_GET_ETYPE(bp); | |
387 | drrw->drr_lsize = BPE_GET_LSIZE(bp); | |
388 | drrw->drr_psize = BPE_GET_PSIZE(bp); | |
389 | ||
390 | decode_embedded_bp_compressed(bp, buf); | |
391 | ||
37f8a883 | 392 | if (dump_record(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0) |
ecb2b7dc | 393 | return (SET_ERROR(EINTR)); |
9b67f605 MA |
394 | return (0); |
395 | } | |
396 | ||
428870ff | 397 | static int |
37abac6d | 398 | dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data) |
428870ff | 399 | { |
37abac6d | 400 | struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill); |
428870ff | 401 | |
37abac6d | 402 | if (dsp->dsa_pending_op != PENDING_NONE) { |
37f8a883 | 403 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 404 | return (SET_ERROR(EINTR)); |
37abac6d | 405 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff BB |
406 | } |
407 | ||
408 | /* write a SPILL record */ | |
37abac6d BP |
409 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); |
410 | dsp->dsa_drr->drr_type = DRR_SPILL; | |
428870ff BB |
411 | drrs->drr_object = object; |
412 | drrs->drr_length = blksz; | |
37abac6d | 413 | drrs->drr_toguid = dsp->dsa_toguid; |
34dc7c2f | 414 | |
37f8a883 | 415 | if (dump_record(dsp, data, blksz) != 0) |
2e528b49 | 416 | return (SET_ERROR(EINTR)); |
34dc7c2f BB |
417 | return (0); |
418 | } | |
419 | ||
420 | static int | |
37abac6d | 421 | dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs) |
34dc7c2f | 422 | { |
37abac6d | 423 | struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects); |
428870ff BB |
424 | |
425 | /* | |
426 | * If there is a pending op, but it's not PENDING_FREEOBJECTS, | |
427 | * push it out, since free block aggregation can only be done for | |
428 | * blocks of the same type (i.e., DRR_FREE records can only be | |
429 | * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records | |
430 | * can only be aggregated with other DRR_FREEOBJECTS records. | |
431 | */ | |
37abac6d BP |
432 | if (dsp->dsa_pending_op != PENDING_NONE && |
433 | dsp->dsa_pending_op != PENDING_FREEOBJECTS) { | |
37f8a883 | 434 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 435 | return (SET_ERROR(EINTR)); |
37abac6d | 436 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff | 437 | } |
37abac6d | 438 | if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) { |
428870ff BB |
439 | /* |
440 | * See whether this free object array can be aggregated | |
441 | * with pending one | |
442 | */ | |
443 | if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) { | |
444 | drrfo->drr_numobjs += numobjs; | |
445 | return (0); | |
446 | } else { | |
447 | /* can't be aggregated. Push out pending record */ | |
37f8a883 | 448 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 449 | return (SET_ERROR(EINTR)); |
37abac6d | 450 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff BB |
451 | } |
452 | } | |
453 | ||
34dc7c2f | 454 | /* write a FREEOBJECTS record */ |
37abac6d BP |
455 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); |
456 | dsp->dsa_drr->drr_type = DRR_FREEOBJECTS; | |
428870ff BB |
457 | drrfo->drr_firstobj = firstobj; |
458 | drrfo->drr_numobjs = numobjs; | |
37abac6d | 459 | drrfo->drr_toguid = dsp->dsa_toguid; |
428870ff | 460 | |
37abac6d | 461 | dsp->dsa_pending_op = PENDING_FREEOBJECTS; |
34dc7c2f | 462 | |
34dc7c2f BB |
463 | return (0); |
464 | } | |
465 | ||
466 | static int | |
37abac6d | 467 | dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp) |
34dc7c2f | 468 | { |
37abac6d | 469 | struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object); |
428870ff | 470 | |
47dfff3b MA |
471 | if (object < dsp->dsa_resume_object) { |
472 | /* | |
473 | * Note: when resuming, we will visit all the dnodes in | |
474 | * the block of dnodes that we are resuming from. In | |
475 | * this case it's unnecessary to send the dnodes prior to | |
476 | * the one we are resuming from. We should be at most one | |
477 | * block's worth of dnodes behind the resume point. | |
478 | */ | |
479 | ASSERT3U(dsp->dsa_resume_object - object, <, | |
480 | 1 << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)); | |
481 | return (0); | |
482 | } | |
483 | ||
34dc7c2f | 484 | if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) |
37abac6d | 485 | return (dump_freeobjects(dsp, object, 1)); |
34dc7c2f | 486 | |
37abac6d | 487 | if (dsp->dsa_pending_op != PENDING_NONE) { |
37f8a883 | 488 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 489 | return (SET_ERROR(EINTR)); |
37abac6d | 490 | dsp->dsa_pending_op = PENDING_NONE; |
428870ff BB |
491 | } |
492 | ||
34dc7c2f | 493 | /* write an OBJECT record */ |
37abac6d BP |
494 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); |
495 | dsp->dsa_drr->drr_type = DRR_OBJECT; | |
428870ff BB |
496 | drro->drr_object = object; |
497 | drro->drr_type = dnp->dn_type; | |
498 | drro->drr_bonustype = dnp->dn_bonustype; | |
499 | drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; | |
500 | drro->drr_bonuslen = dnp->dn_bonuslen; | |
50c957f7 | 501 | drro->drr_dn_slots = dnp->dn_extra_slots + 1; |
428870ff BB |
502 | drro->drr_checksumtype = dnp->dn_checksum; |
503 | drro->drr_compress = dnp->dn_compress; | |
37abac6d | 504 | drro->drr_toguid = dsp->dsa_toguid; |
428870ff | 505 | |
f1512ee6 MA |
506 | if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && |
507 | drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE) | |
508 | drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE; | |
509 | ||
37f8a883 MA |
510 | if (dump_record(dsp, DN_BONUS(dnp), |
511 | P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0) { | |
2e528b49 | 512 | return (SET_ERROR(EINTR)); |
37f8a883 | 513 | } |
34dc7c2f | 514 | |
ea97f8ce | 515 | /* Free anything past the end of the file. */ |
37abac6d | 516 | if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * |
ea97f8ce | 517 | (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0) |
2e528b49 | 518 | return (SET_ERROR(EINTR)); |
13fe0198 | 519 | if (dsp->dsa_err != 0) |
2e528b49 | 520 | return (SET_ERROR(EINTR)); |
34dc7c2f BB |
521 | return (0); |
522 | } | |
523 | ||
9b67f605 MA |
524 | static boolean_t |
525 | backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp) | |
526 | { | |
527 | if (!BP_IS_EMBEDDED(bp)) | |
528 | return (B_FALSE); | |
529 | ||
530 | /* | |
531 | * Compression function must be legacy, or explicitly enabled. | |
532 | */ | |
533 | if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS && | |
2aa34383 | 534 | !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LZ4))) |
9b67f605 MA |
535 | return (B_FALSE); |
536 | ||
537 | /* | |
538 | * Embed type must be explicitly enabled. | |
539 | */ | |
540 | switch (BPE_GET_ETYPE(bp)) { | |
541 | case BP_EMBEDDED_TYPE_DATA: | |
542 | if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) | |
543 | return (B_TRUE); | |
544 | break; | |
545 | default: | |
546 | return (B_FALSE); | |
547 | } | |
548 | return (B_FALSE); | |
549 | } | |
550 | ||
fcff0f35 PD |
551 | /* |
552 | * This is the callback function to traverse_dataset that acts as the worker | |
553 | * thread for dmu_send_impl. | |
554 | */ | |
555 | /*ARGSUSED*/ | |
556 | static int | |
557 | send_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
558 | const zbookmark_phys_t *zb, const struct dnode_phys *dnp, void *arg) | |
559 | { | |
560 | struct send_thread_arg *sta = arg; | |
561 | struct send_block_record *record; | |
562 | uint64_t record_size; | |
563 | int err = 0; | |
564 | ||
47dfff3b MA |
565 | ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || |
566 | zb->zb_object >= sta->resume.zb_object); | |
567 | ||
fcff0f35 PD |
568 | if (sta->cancel) |
569 | return (SET_ERROR(EINTR)); | |
34dc7c2f | 570 | |
fcff0f35 PD |
571 | if (bp == NULL) { |
572 | ASSERT3U(zb->zb_level, ==, ZB_DNODE_LEVEL); | |
573 | return (0); | |
574 | } else if (zb->zb_level < 0) { | |
575 | return (0); | |
576 | } | |
577 | ||
578 | record = kmem_zalloc(sizeof (struct send_block_record), KM_SLEEP); | |
579 | record->eos_marker = B_FALSE; | |
580 | record->bp = *bp; | |
581 | record->zb = *zb; | |
582 | record->indblkshift = dnp->dn_indblkshift; | |
583 | record->datablkszsec = dnp->dn_datablkszsec; | |
584 | record_size = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; | |
585 | bqueue_enqueue(&sta->q, record, record_size); | |
586 | ||
587 | return (err); | |
588 | } | |
589 | ||
590 | /* | |
591 | * This function kicks off the traverse_dataset. It also handles setting the | |
592 | * error code of the thread in case something goes wrong, and pushes the End of | |
593 | * Stream record when the traverse_dataset call has finished. If there is no | |
594 | * dataset to traverse, the thread immediately pushes End of Stream marker. | |
595 | */ | |
596 | static void | |
597 | send_traverse_thread(void *arg) | |
598 | { | |
599 | struct send_thread_arg *st_arg = arg; | |
600 | int err; | |
601 | struct send_block_record *data; | |
3e635ac1 | 602 | fstrans_cookie_t cookie = spl_fstrans_mark(); |
fcff0f35 PD |
603 | |
604 | if (st_arg->ds != NULL) { | |
47dfff3b MA |
605 | err = traverse_dataset_resume(st_arg->ds, |
606 | st_arg->fromtxg, &st_arg->resume, | |
607 | st_arg->flags, send_cb, st_arg); | |
608 | ||
fcff0f35 PD |
609 | if (err != EINTR) |
610 | st_arg->error_code = err; | |
611 | } | |
612 | data = kmem_zalloc(sizeof (*data), KM_SLEEP); | |
613 | data->eos_marker = B_TRUE; | |
614 | bqueue_enqueue(&st_arg->q, data, 1); | |
3e635ac1 | 615 | spl_fstrans_unmark(cookie); |
34a6b428 | 616 | thread_exit(); |
fcff0f35 PD |
617 | } |
618 | ||
619 | /* | |
620 | * This function actually handles figuring out what kind of record needs to be | |
621 | * dumped, reading the data (which has hopefully been prefetched), and calling | |
622 | * the appropriate helper function. | |
623 | */ | |
34dc7c2f | 624 | static int |
fcff0f35 | 625 | do_dump(dmu_sendarg_t *dsa, struct send_block_record *data) |
34dc7c2f | 626 | { |
fcff0f35 PD |
627 | dsl_dataset_t *ds = dmu_objset_ds(dsa->dsa_os); |
628 | const blkptr_t *bp = &data->bp; | |
629 | const zbookmark_phys_t *zb = &data->zb; | |
630 | uint8_t indblkshift = data->indblkshift; | |
631 | uint16_t dblkszsec = data->datablkszsec; | |
632 | spa_t *spa = ds->ds_dir->dd_pool->dp_spa; | |
34dc7c2f | 633 | dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; |
34dc7c2f | 634 | int err = 0; |
fcff0f35 | 635 | uint64_t dnobj; |
34dc7c2f | 636 | |
fcff0f35 | 637 | ASSERT3U(zb->zb_level, >=, 0); |
34dc7c2f | 638 | |
47dfff3b MA |
639 | ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || |
640 | zb->zb_object >= dsa->dsa_resume_object); | |
641 | ||
428870ff BB |
642 | if (zb->zb_object != DMU_META_DNODE_OBJECT && |
643 | DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { | |
9babb374 | 644 | return (0); |
b0bc7a84 MG |
645 | } else if (BP_IS_HOLE(bp) && |
646 | zb->zb_object == DMU_META_DNODE_OBJECT) { | |
fcff0f35 | 647 | uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); |
b128c09f | 648 | uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; |
fcff0f35 | 649 | err = dump_freeobjects(dsa, dnobj, span >> DNODE_SHIFT); |
b0bc7a84 | 650 | } else if (BP_IS_HOLE(bp)) { |
fcff0f35 PD |
651 | uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); |
652 | uint64_t offset = zb->zb_blkid * span; | |
653 | err = dump_free(dsa, zb->zb_object, offset, span); | |
b128c09f BB |
654 | } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { |
655 | return (0); | |
656 | } else if (type == DMU_OT_DNODE) { | |
50c957f7 NB |
657 | dnode_phys_t *blk; |
658 | int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; | |
2a432414 | 659 | arc_flags_t aflags = ARC_FLAG_WAIT; |
b128c09f | 660 | arc_buf_t *abuf; |
fcff0f35 PD |
661 | int i; |
662 | ||
663 | ASSERT0(zb->zb_level); | |
b128c09f | 664 | |
294f6806 GW |
665 | if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, |
666 | ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, | |
667 | &aflags, zb) != 0) | |
2e528b49 | 668 | return (SET_ERROR(EIO)); |
34dc7c2f | 669 | |
b128c09f | 670 | blk = abuf->b_data; |
50c957f7 NB |
671 | dnobj = zb->zb_blkid * epb; |
672 | for (i = 0; i < epb; i += blk[i].dn_extra_slots + 1) { | |
fcff0f35 | 673 | err = dump_dnode(dsa, dnobj + i, blk + i); |
13fe0198 | 674 | if (err != 0) |
34dc7c2f BB |
675 | break; |
676 | } | |
d3c2ae1c | 677 | arc_buf_destroy(abuf, &abuf); |
428870ff | 678 | } else if (type == DMU_OT_SA) { |
2a432414 | 679 | arc_flags_t aflags = ARC_FLAG_WAIT; |
b128c09f | 680 | arc_buf_t *abuf; |
34dc7c2f | 681 | int blksz = BP_GET_LSIZE(bp); |
b128c09f | 682 | |
294f6806 GW |
683 | if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, |
684 | ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, | |
685 | &aflags, zb) != 0) | |
2e528b49 | 686 | return (SET_ERROR(EIO)); |
b128c09f | 687 | |
fcff0f35 | 688 | err = dump_spill(dsa, zb->zb_object, blksz, abuf->b_data); |
d3c2ae1c | 689 | arc_buf_destroy(abuf, &abuf); |
fcff0f35 | 690 | } else if (backup_do_embed(dsa, bp)) { |
9b67f605 | 691 | /* it's an embedded level-0 block of a regular object */ |
fcff0f35 PD |
692 | int blksz = dblkszsec << SPA_MINBLOCKSHIFT; |
693 | ASSERT0(zb->zb_level); | |
694 | err = dump_write_embedded(dsa, zb->zb_object, | |
9b67f605 | 695 | zb->zb_blkid * blksz, blksz, bp); |
fcff0f35 PD |
696 | } else { |
697 | /* it's a level-0 block of a regular object */ | |
2a432414 | 698 | arc_flags_t aflags = ARC_FLAG_WAIT; |
428870ff | 699 | arc_buf_t *abuf; |
fcff0f35 PD |
700 | int blksz = dblkszsec << SPA_MINBLOCKSHIFT; |
701 | uint64_t offset; | |
2aa34383 DK |
702 | |
703 | /* | |
704 | * If we have large blocks stored on disk but the send flags | |
705 | * don't allow us to send large blocks, we split the data from | |
706 | * the arc buf into chunks. | |
707 | */ | |
a7004725 | 708 | boolean_t split_large_blocks = blksz > SPA_OLD_MAXBLOCKSIZE && |
2aa34383 DK |
709 | !(dsa->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS); |
710 | /* | |
711 | * We should only request compressed data from the ARC if all | |
712 | * the following are true: | |
713 | * - stream compression was requested | |
714 | * - we aren't splitting large blocks into smaller chunks | |
715 | * - the data won't need to be byteswapped before sending | |
716 | * - this isn't an embedded block | |
717 | * - this isn't metadata (if receiving on a different endian | |
718 | * system it can be byteswapped more easily) | |
719 | */ | |
720 | boolean_t request_compressed = | |
721 | (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_COMPRESSED) && | |
722 | !split_large_blocks && !BP_SHOULD_BYTESWAP(bp) && | |
723 | !BP_IS_EMBEDDED(bp) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp)); | |
428870ff | 724 | |
da536844 | 725 | ASSERT0(zb->zb_level); |
47dfff3b MA |
726 | ASSERT(zb->zb_object > dsa->dsa_resume_object || |
727 | (zb->zb_object == dsa->dsa_resume_object && | |
728 | zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); | |
729 | ||
a7004725 DK |
730 | ASSERT3U(blksz, ==, BP_GET_LSIZE(bp)); |
731 | ||
732 | enum zio_flag zioflags = ZIO_FLAG_CANFAIL; | |
2aa34383 DK |
733 | if (request_compressed) |
734 | zioflags |= ZIO_FLAG_RAW; | |
735 | ||
294f6806 | 736 | if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, |
a7004725 | 737 | ZIO_PRIORITY_ASYNC_READ, zioflags, &aflags, zb) != 0) { |
330d06f9 | 738 | if (zfs_send_corrupt_data) { |
330d06f9 | 739 | /* Send a block filled with 0x"zfs badd bloc" */ |
2aa34383 DK |
740 | abuf = arc_alloc_buf(spa, &abuf, ARC_BUFC_DATA, |
741 | blksz); | |
a7004725 | 742 | uint64_t *ptr; |
330d06f9 MA |
743 | for (ptr = abuf->b_data; |
744 | (char *)ptr < (char *)abuf->b_data + blksz; | |
745 | ptr++) | |
dd26aa53 | 746 | *ptr = 0x2f5baddb10cULL; |
330d06f9 | 747 | } else { |
2e528b49 | 748 | return (SET_ERROR(EIO)); |
330d06f9 MA |
749 | } |
750 | } | |
428870ff | 751 | |
f1512ee6 MA |
752 | offset = zb->zb_blkid * blksz; |
753 | ||
2aa34383 | 754 | if (split_large_blocks) { |
2aa34383 DK |
755 | ASSERT3U(arc_get_compression(abuf), ==, |
756 | ZIO_COMPRESS_OFF); | |
a7004725 | 757 | char *buf = abuf->b_data; |
f1512ee6 MA |
758 | while (blksz > 0 && err == 0) { |
759 | int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE); | |
fcff0f35 | 760 | err = dump_write(dsa, type, zb->zb_object, |
2aa34383 | 761 | offset, n, n, NULL, buf); |
f1512ee6 MA |
762 | offset += n; |
763 | buf += n; | |
764 | blksz -= n; | |
765 | } | |
766 | } else { | |
2aa34383 DK |
767 | err = dump_write(dsa, type, zb->zb_object, offset, |
768 | blksz, arc_buf_size(abuf), bp, | |
769 | abuf->b_data); | |
f1512ee6 | 770 | } |
d3c2ae1c | 771 | arc_buf_destroy(abuf, &abuf); |
34dc7c2f BB |
772 | } |
773 | ||
774 | ASSERT(err == 0 || err == EINTR); | |
775 | return (err); | |
776 | } | |
777 | ||
6f1ffb06 | 778 | /* |
fcff0f35 PD |
779 | * Pop the new data off the queue, and free the old data. |
780 | */ | |
781 | static struct send_block_record * | |
782 | get_next_record(bqueue_t *bq, struct send_block_record *data) | |
783 | { | |
784 | struct send_block_record *tmp = bqueue_dequeue(bq); | |
785 | kmem_free(data, sizeof (*data)); | |
786 | return (tmp); | |
787 | } | |
788 | ||
789 | /* | |
790 | * Actually do the bulk of the work in a zfs send. | |
791 | * | |
792 | * Note: Releases dp using the specified tag. | |
6f1ffb06 | 793 | */ |
13fe0198 | 794 | static int |
fcff0f35 | 795 | dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *to_ds, |
2aa34383 DK |
796 | zfs_bookmark_phys_t *ancestor_zb, boolean_t is_clone, |
797 | boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, | |
798 | int outfd, uint64_t resumeobj, uint64_t resumeoff, | |
47dfff3b | 799 | vnode_t *vp, offset_t *off) |
34dc7c2f | 800 | { |
13fe0198 | 801 | objset_t *os; |
34dc7c2f | 802 | dmu_replay_record_t *drr; |
37abac6d | 803 | dmu_sendarg_t *dsp; |
34dc7c2f BB |
804 | int err; |
805 | uint64_t fromtxg = 0; | |
9b67f605 | 806 | uint64_t featureflags = 0; |
fcff0f35 | 807 | struct send_thread_arg to_arg; |
47dfff3b MA |
808 | void *payload = NULL; |
809 | size_t payload_len = 0; | |
fcff0f35 | 810 | struct send_block_record *to_data; |
34dc7c2f | 811 | |
fcff0f35 | 812 | err = dmu_objset_from_ds(to_ds, &os); |
13fe0198 | 813 | if (err != 0) { |
13fe0198 MA |
814 | dsl_pool_rele(dp, tag); |
815 | return (err); | |
816 | } | |
34dc7c2f | 817 | |
34dc7c2f BB |
818 | drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); |
819 | drr->drr_type = DRR_BEGIN; | |
820 | drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; | |
428870ff BB |
821 | DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo, |
822 | DMU_SUBSTREAM); | |
823 | ||
47dfff3b MA |
824 | bzero(&to_arg, sizeof (to_arg)); |
825 | ||
428870ff | 826 | #ifdef _KERNEL |
13fe0198 | 827 | if (dmu_objset_type(os) == DMU_OST_ZFS) { |
428870ff | 828 | uint64_t version; |
13fe0198 | 829 | if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { |
37abac6d | 830 | kmem_free(drr, sizeof (dmu_replay_record_t)); |
13fe0198 | 831 | dsl_pool_rele(dp, tag); |
2e528b49 | 832 | return (SET_ERROR(EINVAL)); |
37abac6d | 833 | } |
13fe0198 | 834 | if (version >= ZPL_VERSION_SA) { |
9b67f605 | 835 | featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; |
428870ff BB |
836 | } |
837 | } | |
838 | #endif | |
839 | ||
fcff0f35 | 840 | if (large_block_ok && to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_BLOCKS]) |
f1512ee6 | 841 | featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS; |
50c957f7 NB |
842 | if (to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_DNODE]) |
843 | featureflags |= DMU_BACKUP_FEATURE_LARGE_DNODE; | |
9b67f605 MA |
844 | if (embedok && |
845 | spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) { | |
846 | featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA; | |
2aa34383 DK |
847 | } |
848 | if (compressok) { | |
849 | featureflags |= DMU_BACKUP_FEATURE_COMPRESSED; | |
850 | } | |
851 | if ((featureflags & | |
852 | (DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_COMPRESSED)) != | |
853 | 0 && spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) { | |
854 | featureflags |= DMU_BACKUP_FEATURE_LZ4; | |
9b67f605 MA |
855 | } |
856 | ||
47dfff3b MA |
857 | if (resumeobj != 0 || resumeoff != 0) { |
858 | featureflags |= DMU_BACKUP_FEATURE_RESUMING; | |
859 | } | |
860 | ||
9b67f605 MA |
861 | DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo, |
862 | featureflags); | |
863 | ||
34dc7c2f | 864 | drr->drr_u.drr_begin.drr_creation_time = |
fcff0f35 | 865 | dsl_dataset_phys(to_ds)->ds_creation_time; |
13fe0198 | 866 | drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); |
da536844 | 867 | if (is_clone) |
34dc7c2f | 868 | drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; |
fcff0f35 PD |
869 | drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(to_ds)->ds_guid; |
870 | if (dsl_dataset_phys(to_ds)->ds_flags & DS_FLAG_CI_DATASET) | |
34dc7c2f | 871 | drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; |
b607405f AS |
872 | if (zfs_send_set_freerecords_bit) |
873 | drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_FREERECORDS; | |
34dc7c2f | 874 | |
fcff0f35 PD |
875 | if (ancestor_zb != NULL) { |
876 | drr->drr_u.drr_begin.drr_fromguid = | |
877 | ancestor_zb->zbm_guid; | |
878 | fromtxg = ancestor_zb->zbm_creation_txg; | |
da536844 | 879 | } |
fcff0f35 PD |
880 | dsl_dataset_name(to_ds, drr->drr_u.drr_begin.drr_toname); |
881 | if (!to_ds->ds_is_snapshot) { | |
da536844 MA |
882 | (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--", |
883 | sizeof (drr->drr_u.drr_begin.drr_toname)); | |
13fe0198 | 884 | } |
34dc7c2f | 885 | |
37abac6d BP |
886 | dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); |
887 | ||
888 | dsp->dsa_drr = drr; | |
889 | dsp->dsa_vp = vp; | |
890 | dsp->dsa_outfd = outfd; | |
891 | dsp->dsa_proc = curproc; | |
13fe0198 | 892 | dsp->dsa_os = os; |
37abac6d | 893 | dsp->dsa_off = off; |
fcff0f35 | 894 | dsp->dsa_toguid = dsl_dataset_phys(to_ds)->ds_guid; |
37abac6d | 895 | dsp->dsa_pending_op = PENDING_NONE; |
9b67f605 | 896 | dsp->dsa_featureflags = featureflags; |
47dfff3b MA |
897 | dsp->dsa_resume_object = resumeobj; |
898 | dsp->dsa_resume_offset = resumeoff; | |
37abac6d | 899 | |
fcff0f35 PD |
900 | mutex_enter(&to_ds->ds_sendstream_lock); |
901 | list_insert_head(&to_ds->ds_sendstreams, dsp); | |
902 | mutex_exit(&to_ds->ds_sendstream_lock); | |
37abac6d | 903 | |
fcff0f35 | 904 | dsl_dataset_long_hold(to_ds, FTAG); |
7ec09286 MA |
905 | dsl_pool_rele(dp, tag); |
906 | ||
47dfff3b MA |
907 | if (resumeobj != 0 || resumeoff != 0) { |
908 | dmu_object_info_t to_doi; | |
909 | nvlist_t *nvl; | |
910 | err = dmu_object_info(os, resumeobj, &to_doi); | |
911 | if (err != 0) | |
912 | goto out; | |
913 | SET_BOOKMARK(&to_arg.resume, to_ds->ds_object, resumeobj, 0, | |
914 | resumeoff / to_doi.doi_data_block_size); | |
915 | ||
916 | nvl = fnvlist_alloc(); | |
917 | fnvlist_add_uint64(nvl, "resume_object", resumeobj); | |
918 | fnvlist_add_uint64(nvl, "resume_offset", resumeoff); | |
919 | payload = fnvlist_pack(nvl, &payload_len); | |
920 | drr->drr_payloadlen = payload_len; | |
921 | fnvlist_free(nvl); | |
922 | } | |
923 | ||
924 | err = dump_record(dsp, payload, payload_len); | |
925 | fnvlist_pack_free(payload, payload_len); | |
926 | if (err != 0) { | |
37abac6d BP |
927 | err = dsp->dsa_err; |
928 | goto out; | |
34dc7c2f BB |
929 | } |
930 | ||
fcff0f35 PD |
931 | err = bqueue_init(&to_arg.q, zfs_send_queue_length, |
932 | offsetof(struct send_block_record, ln)); | |
933 | to_arg.error_code = 0; | |
934 | to_arg.cancel = B_FALSE; | |
935 | to_arg.ds = to_ds; | |
936 | to_arg.fromtxg = fromtxg; | |
937 | to_arg.flags = TRAVERSE_PRE | TRAVERSE_PREFETCH; | |
938 | (void) thread_create(NULL, 0, send_traverse_thread, &to_arg, 0, curproc, | |
939 | TS_RUN, minclsyspri); | |
940 | ||
941 | to_data = bqueue_dequeue(&to_arg.q); | |
942 | ||
943 | while (!to_data->eos_marker && err == 0) { | |
944 | err = do_dump(dsp, to_data); | |
945 | to_data = get_next_record(&to_arg.q, to_data); | |
946 | if (issig(JUSTLOOKING) && issig(FORREAL)) | |
947 | err = EINTR; | |
948 | } | |
949 | ||
950 | if (err != 0) { | |
951 | to_arg.cancel = B_TRUE; | |
952 | while (!to_data->eos_marker) { | |
953 | to_data = get_next_record(&to_arg.q, to_data); | |
954 | } | |
955 | } | |
956 | kmem_free(to_data, sizeof (*to_data)); | |
957 | ||
958 | bqueue_destroy(&to_arg.q); | |
959 | ||
960 | if (err == 0 && to_arg.error_code != 0) | |
961 | err = to_arg.error_code; | |
962 | ||
963 | if (err != 0) | |
964 | goto out; | |
34dc7c2f | 965 | |
37abac6d | 966 | if (dsp->dsa_pending_op != PENDING_NONE) |
37f8a883 | 967 | if (dump_record(dsp, NULL, 0) != 0) |
2e528b49 | 968 | err = SET_ERROR(EINTR); |
428870ff | 969 | |
13fe0198 MA |
970 | if (err != 0) { |
971 | if (err == EINTR && dsp->dsa_err != 0) | |
37abac6d BP |
972 | err = dsp->dsa_err; |
973 | goto out; | |
34dc7c2f BB |
974 | } |
975 | ||
976 | bzero(drr, sizeof (dmu_replay_record_t)); | |
977 | drr->drr_type = DRR_END; | |
37abac6d BP |
978 | drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc; |
979 | drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid; | |
34dc7c2f | 980 | |
fcff0f35 | 981 | if (dump_record(dsp, NULL, 0) != 0) |
37abac6d | 982 | err = dsp->dsa_err; |
34dc7c2f | 983 | |
37abac6d | 984 | out: |
fcff0f35 PD |
985 | mutex_enter(&to_ds->ds_sendstream_lock); |
986 | list_remove(&to_ds->ds_sendstreams, dsp); | |
987 | mutex_exit(&to_ds->ds_sendstream_lock); | |
37abac6d | 988 | |
51907a31 K |
989 | VERIFY(err != 0 || (dsp->dsa_sent_begin && dsp->dsa_sent_end)); |
990 | ||
34dc7c2f | 991 | kmem_free(drr, sizeof (dmu_replay_record_t)); |
37abac6d | 992 | kmem_free(dsp, sizeof (dmu_sendarg_t)); |
34dc7c2f | 993 | |
fcff0f35 | 994 | dsl_dataset_long_rele(to_ds, FTAG); |
13fe0198 | 995 | |
37abac6d | 996 | return (err); |
34dc7c2f BB |
997 | } |
998 | ||
330d06f9 | 999 | int |
13fe0198 | 1000 | dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, |
2aa34383 | 1001 | boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, |
f1512ee6 | 1002 | int outfd, vnode_t *vp, offset_t *off) |
13fe0198 MA |
1003 | { |
1004 | dsl_pool_t *dp; | |
1005 | dsl_dataset_t *ds; | |
1006 | dsl_dataset_t *fromds = NULL; | |
1007 | int err; | |
1008 | ||
1009 | err = dsl_pool_hold(pool, FTAG, &dp); | |
1010 | if (err != 0) | |
1011 | return (err); | |
1012 | ||
1013 | err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds); | |
1014 | if (err != 0) { | |
1015 | dsl_pool_rele(dp, FTAG); | |
1016 | return (err); | |
1017 | } | |
1018 | ||
1019 | if (fromsnap != 0) { | |
da536844 MA |
1020 | zfs_bookmark_phys_t zb; |
1021 | boolean_t is_clone; | |
1022 | ||
13fe0198 MA |
1023 | err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); |
1024 | if (err != 0) { | |
1025 | dsl_dataset_rele(ds, FTAG); | |
1026 | dsl_pool_rele(dp, FTAG); | |
1027 | return (err); | |
1028 | } | |
da536844 MA |
1029 | if (!dsl_dataset_is_before(ds, fromds, 0)) |
1030 | err = SET_ERROR(EXDEV); | |
d683ddbb JG |
1031 | zb.zbm_creation_time = |
1032 | dsl_dataset_phys(fromds)->ds_creation_time; | |
1033 | zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg; | |
1034 | zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; | |
da536844 MA |
1035 | is_clone = (fromds->ds_dir != ds->ds_dir); |
1036 | dsl_dataset_rele(fromds, FTAG); | |
f1512ee6 | 1037 | err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, |
2aa34383 | 1038 | embedok, large_block_ok, compressok, outfd, 0, 0, vp, off); |
da536844 | 1039 | } else { |
f1512ee6 | 1040 | err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, |
2aa34383 | 1041 | embedok, large_block_ok, compressok, outfd, 0, 0, vp, off); |
13fe0198 | 1042 | } |
da536844 MA |
1043 | dsl_dataset_rele(ds, FTAG); |
1044 | return (err); | |
13fe0198 MA |
1045 | } |
1046 | ||
1047 | int | |
47dfff3b | 1048 | dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok, |
2aa34383 DK |
1049 | boolean_t large_block_ok, boolean_t compressok, int outfd, |
1050 | uint64_t resumeobj, uint64_t resumeoff, | |
47dfff3b | 1051 | vnode_t *vp, offset_t *off) |
13fe0198 MA |
1052 | { |
1053 | dsl_pool_t *dp; | |
1054 | dsl_dataset_t *ds; | |
13fe0198 | 1055 | int err; |
da536844 | 1056 | boolean_t owned = B_FALSE; |
13fe0198 | 1057 | |
da536844 | 1058 | if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL) |
2e528b49 | 1059 | return (SET_ERROR(EINVAL)); |
13fe0198 MA |
1060 | |
1061 | err = dsl_pool_hold(tosnap, FTAG, &dp); | |
1062 | if (err != 0) | |
1063 | return (err); | |
1064 | ||
da536844 MA |
1065 | if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) { |
1066 | /* | |
1067 | * We are sending a filesystem or volume. Ensure | |
1068 | * that it doesn't change by owning the dataset. | |
1069 | */ | |
1070 | err = dsl_dataset_own(dp, tosnap, FTAG, &ds); | |
1071 | owned = B_TRUE; | |
1072 | } else { | |
1073 | err = dsl_dataset_hold(dp, tosnap, FTAG, &ds); | |
1074 | } | |
13fe0198 MA |
1075 | if (err != 0) { |
1076 | dsl_pool_rele(dp, FTAG); | |
1077 | return (err); | |
1078 | } | |
1079 | ||
1080 | if (fromsnap != NULL) { | |
da536844 MA |
1081 | zfs_bookmark_phys_t zb; |
1082 | boolean_t is_clone = B_FALSE; | |
1083 | int fsnamelen = strchr(tosnap, '@') - tosnap; | |
1084 | ||
1085 | /* | |
1086 | * If the fromsnap is in a different filesystem, then | |
1087 | * mark the send stream as a clone. | |
1088 | */ | |
1089 | if (strncmp(tosnap, fromsnap, fsnamelen) != 0 || | |
1090 | (fromsnap[fsnamelen] != '@' && | |
1091 | fromsnap[fsnamelen] != '#')) { | |
1092 | is_clone = B_TRUE; | |
1093 | } | |
1094 | ||
1095 | if (strchr(fromsnap, '@')) { | |
1096 | dsl_dataset_t *fromds; | |
1097 | err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); | |
1098 | if (err == 0) { | |
1099 | if (!dsl_dataset_is_before(ds, fromds, 0)) | |
1100 | err = SET_ERROR(EXDEV); | |
1101 | zb.zbm_creation_time = | |
d683ddbb | 1102 | dsl_dataset_phys(fromds)->ds_creation_time; |
da536844 | 1103 | zb.zbm_creation_txg = |
d683ddbb JG |
1104 | dsl_dataset_phys(fromds)->ds_creation_txg; |
1105 | zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; | |
da536844 MA |
1106 | is_clone = (ds->ds_dir != fromds->ds_dir); |
1107 | dsl_dataset_rele(fromds, FTAG); | |
1108 | } | |
1109 | } else { | |
1110 | err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb); | |
1111 | } | |
13fe0198 MA |
1112 | if (err != 0) { |
1113 | dsl_dataset_rele(ds, FTAG); | |
1114 | dsl_pool_rele(dp, FTAG); | |
1115 | return (err); | |
1116 | } | |
f1512ee6 | 1117 | err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, |
2aa34383 | 1118 | embedok, large_block_ok, compressok, |
47dfff3b | 1119 | outfd, resumeobj, resumeoff, vp, off); |
da536844 | 1120 | } else { |
f1512ee6 | 1121 | err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, |
2aa34383 | 1122 | embedok, large_block_ok, compressok, |
47dfff3b | 1123 | outfd, resumeobj, resumeoff, vp, off); |
13fe0198 | 1124 | } |
da536844 MA |
1125 | if (owned) |
1126 | dsl_dataset_disown(ds, FTAG); | |
1127 | else | |
1128 | dsl_dataset_rele(ds, FTAG); | |
1129 | return (err); | |
13fe0198 MA |
1130 | } |
1131 | ||
5dc8b736 | 1132 | static int |
2aa34383 DK |
1133 | dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t uncompressed, |
1134 | uint64_t compressed, boolean_t stream_compressed, uint64_t *sizep) | |
5dc8b736 MG |
1135 | { |
1136 | int err; | |
2aa34383 | 1137 | uint64_t size; |
5dc8b736 MG |
1138 | /* |
1139 | * Assume that space (both on-disk and in-stream) is dominated by | |
1140 | * data. We will adjust for indirect blocks and the copies property, | |
1141 | * but ignore per-object space used (eg, dnodes and DRR_OBJECT records). | |
1142 | */ | |
1143 | ||
2aa34383 DK |
1144 | uint64_t recordsize; |
1145 | uint64_t record_count; | |
dd429b46 PD |
1146 | objset_t *os; |
1147 | VERIFY0(dmu_objset_from_ds(ds, &os)); | |
2aa34383 DK |
1148 | |
1149 | /* Assume all (uncompressed) blocks are recordsize. */ | |
dd429b46 PD |
1150 | if (os->os_phys->os_type == DMU_OST_ZVOL) { |
1151 | err = dsl_prop_get_int_ds(ds, | |
1152 | zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &recordsize); | |
1153 | } else { | |
1154 | err = dsl_prop_get_int_ds(ds, | |
1155 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), &recordsize); | |
1156 | } | |
2aa34383 DK |
1157 | if (err != 0) |
1158 | return (err); | |
1159 | record_count = uncompressed / recordsize; | |
1160 | ||
1161 | /* | |
1162 | * If we're estimating a send size for a compressed stream, use the | |
1163 | * compressed data size to estimate the stream size. Otherwise, use the | |
1164 | * uncompressed data size. | |
1165 | */ | |
1166 | size = stream_compressed ? compressed : uncompressed; | |
1167 | ||
5dc8b736 MG |
1168 | /* |
1169 | * Subtract out approximate space used by indirect blocks. | |
1170 | * Assume most space is used by data blocks (non-indirect, non-dnode). | |
2aa34383 | 1171 | * Assume no ditto blocks or internal fragmentation. |
5dc8b736 MG |
1172 | * |
1173 | * Therefore, space used by indirect blocks is sizeof(blkptr_t) per | |
2aa34383 | 1174 | * block. |
5dc8b736 | 1175 | */ |
2aa34383 | 1176 | size -= record_count * sizeof (blkptr_t); |
5dc8b736 MG |
1177 | |
1178 | /* Add in the space for the record associated with each block. */ | |
2aa34383 | 1179 | size += record_count * sizeof (dmu_replay_record_t); |
5dc8b736 MG |
1180 | |
1181 | *sizep = size; | |
1182 | ||
1183 | return (0); | |
1184 | } | |
1185 | ||
13fe0198 | 1186 | int |
2aa34383 DK |
1187 | dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, |
1188 | boolean_t stream_compressed, uint64_t *sizep) | |
330d06f9 | 1189 | { |
330d06f9 | 1190 | int err; |
2aa34383 | 1191 | uint64_t uncomp, comp; |
13fe0198 | 1192 | |
fd0fd646 | 1193 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); |
330d06f9 MA |
1194 | |
1195 | /* tosnap must be a snapshot */ | |
0c66c32d | 1196 | if (!ds->ds_is_snapshot) |
2e528b49 | 1197 | return (SET_ERROR(EINVAL)); |
330d06f9 | 1198 | |
71e2fe41 AG |
1199 | /* fromsnap, if provided, must be a snapshot */ |
1200 | if (fromds != NULL && !fromds->ds_is_snapshot) | |
1201 | return (SET_ERROR(EINVAL)); | |
1202 | ||
6f1ffb06 MA |
1203 | /* |
1204 | * fromsnap must be an earlier snapshot from the same fs as tosnap, | |
1205 | * or the origin's fs. | |
1206 | */ | |
da536844 | 1207 | if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0)) |
2e528b49 | 1208 | return (SET_ERROR(EXDEV)); |
330d06f9 | 1209 | |
2aa34383 | 1210 | /* Get compressed and uncompressed size estimates of changed data. */ |
330d06f9 | 1211 | if (fromds == NULL) { |
2aa34383 DK |
1212 | uncomp = dsl_dataset_phys(ds)->ds_uncompressed_bytes; |
1213 | comp = dsl_dataset_phys(ds)->ds_compressed_bytes; | |
330d06f9 | 1214 | } else { |
2aa34383 | 1215 | uint64_t used; |
330d06f9 | 1216 | err = dsl_dataset_space_written(fromds, ds, |
2aa34383 | 1217 | &used, &comp, &uncomp); |
13fe0198 | 1218 | if (err != 0) |
330d06f9 MA |
1219 | return (err); |
1220 | } | |
1221 | ||
2aa34383 DK |
1222 | err = dmu_adjust_send_estimate_for_indirects(ds, uncomp, comp, |
1223 | stream_compressed, sizep); | |
dd429b46 PD |
1224 | /* |
1225 | * Add the size of the BEGIN and END records to the estimate. | |
1226 | */ | |
1227 | *sizep += 2 * sizeof (dmu_replay_record_t); | |
5dc8b736 MG |
1228 | return (err); |
1229 | } | |
330d06f9 | 1230 | |
2aa34383 DK |
1231 | struct calculate_send_arg { |
1232 | uint64_t uncompressed; | |
1233 | uint64_t compressed; | |
1234 | }; | |
1235 | ||
5dc8b736 MG |
1236 | /* |
1237 | * Simple callback used to traverse the blocks of a snapshot and sum their | |
2aa34383 | 1238 | * uncompressed and compressed sizes. |
5dc8b736 MG |
1239 | */ |
1240 | /* ARGSUSED */ | |
1241 | static int | |
1242 | dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
1243 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) | |
1244 | { | |
2aa34383 | 1245 | struct calculate_send_arg *space = arg; |
5dc8b736 | 1246 | if (bp != NULL && !BP_IS_HOLE(bp)) { |
2aa34383 DK |
1247 | space->uncompressed += BP_GET_UCSIZE(bp); |
1248 | space->compressed += BP_GET_PSIZE(bp); | |
5dc8b736 MG |
1249 | } |
1250 | return (0); | |
1251 | } | |
1252 | ||
1253 | /* | |
1254 | * Given a desination snapshot and a TXG, calculate the approximate size of a | |
1255 | * send stream sent from that TXG. from_txg may be zero, indicating that the | |
1256 | * whole snapshot will be sent. | |
1257 | */ | |
1258 | int | |
1259 | dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg, | |
2aa34383 | 1260 | boolean_t stream_compressed, uint64_t *sizep) |
5dc8b736 | 1261 | { |
5dc8b736 | 1262 | int err; |
2aa34383 | 1263 | struct calculate_send_arg size = { 0 }; |
5dc8b736 | 1264 | |
fd0fd646 | 1265 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); |
5dc8b736 MG |
1266 | |
1267 | /* tosnap must be a snapshot */ | |
1268 | if (!dsl_dataset_is_snapshot(ds)) | |
1269 | return (SET_ERROR(EINVAL)); | |
1270 | ||
1271 | /* verify that from_txg is before the provided snapshot was taken */ | |
1272 | if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) { | |
1273 | return (SET_ERROR(EXDEV)); | |
1274 | } | |
330d06f9 | 1275 | /* |
5dc8b736 MG |
1276 | * traverse the blocks of the snapshot with birth times after |
1277 | * from_txg, summing their uncompressed size | |
330d06f9 | 1278 | */ |
5dc8b736 MG |
1279 | err = traverse_dataset(ds, from_txg, TRAVERSE_POST, |
1280 | dmu_calculate_send_traversal, &size); | |
2aa34383 | 1281 | |
5dc8b736 | 1282 | if (err) |
330d06f9 | 1283 | return (err); |
330d06f9 | 1284 | |
2aa34383 DK |
1285 | err = dmu_adjust_send_estimate_for_indirects(ds, size.uncompressed, |
1286 | size.compressed, stream_compressed, sizep); | |
5dc8b736 | 1287 | return (err); |
330d06f9 MA |
1288 | } |
1289 | ||
13fe0198 MA |
1290 | typedef struct dmu_recv_begin_arg { |
1291 | const char *drba_origin; | |
1292 | dmu_recv_cookie_t *drba_cookie; | |
1293 | cred_t *drba_cred; | |
19580676 | 1294 | uint64_t drba_snapobj; |
13fe0198 | 1295 | } dmu_recv_begin_arg_t; |
34dc7c2f | 1296 | |
34dc7c2f | 1297 | static int |
13fe0198 MA |
1298 | recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, |
1299 | uint64_t fromguid) | |
34dc7c2f | 1300 | { |
34dc7c2f | 1301 | uint64_t val; |
13fe0198 MA |
1302 | int error; |
1303 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
34dc7c2f | 1304 | |
13fe0198 MA |
1305 | /* temporary clone name must not exist */ |
1306 | error = zap_lookup(dp->dp_meta_objset, | |
d683ddbb | 1307 | dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, |
13fe0198 MA |
1308 | 8, 1, &val); |
1309 | if (error != ENOENT) | |
1310 | return (error == 0 ? EBUSY : error); | |
1311 | ||
572e2857 | 1312 | /* new snapshot name must not exist */ |
13fe0198 | 1313 | error = zap_lookup(dp->dp_meta_objset, |
d683ddbb JG |
1314 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, |
1315 | drba->drba_cookie->drc_tosnap, 8, 1, &val); | |
13fe0198 MA |
1316 | if (error != ENOENT) |
1317 | return (error == 0 ? EEXIST : error); | |
572e2857 | 1318 | |
788eb90c JJ |
1319 | /* |
1320 | * Check snapshot limit before receiving. We'll recheck again at the | |
1321 | * end, but might as well abort before receiving if we're already over | |
1322 | * the limit. | |
1323 | * | |
1324 | * Note that we do not check the file system limit with | |
1325 | * dsl_dir_fscount_check because the temporary %clones don't count | |
1326 | * against that limit. | |
1327 | */ | |
1328 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, | |
1329 | NULL, drba->drba_cred); | |
1330 | if (error != 0) | |
1331 | return (error); | |
1332 | ||
13fe0198 | 1333 | if (fromguid != 0) { |
19580676 | 1334 | dsl_dataset_t *snap; |
d683ddbb | 1335 | uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; |
19580676 MA |
1336 | |
1337 | /* Find snapshot in this dir that matches fromguid. */ | |
1338 | while (obj != 0) { | |
1339 | error = dsl_dataset_hold_obj(dp, obj, FTAG, | |
1340 | &snap); | |
1341 | if (error != 0) | |
1342 | return (SET_ERROR(ENODEV)); | |
1343 | if (snap->ds_dir != ds->ds_dir) { | |
1344 | dsl_dataset_rele(snap, FTAG); | |
1345 | return (SET_ERROR(ENODEV)); | |
1346 | } | |
d683ddbb | 1347 | if (dsl_dataset_phys(snap)->ds_guid == fromguid) |
19580676 | 1348 | break; |
d683ddbb | 1349 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; |
19580676 MA |
1350 | dsl_dataset_rele(snap, FTAG); |
1351 | } | |
1352 | if (obj == 0) | |
2e528b49 | 1353 | return (SET_ERROR(ENODEV)); |
34dc7c2f | 1354 | |
19580676 MA |
1355 | if (drba->drba_cookie->drc_force) { |
1356 | drba->drba_snapobj = obj; | |
1357 | } else { | |
1358 | /* | |
1359 | * If we are not forcing, there must be no | |
1360 | * changes since fromsnap. | |
1361 | */ | |
1362 | if (dsl_dataset_modified_since_snap(ds, snap)) { | |
428870ff | 1363 | dsl_dataset_rele(snap, FTAG); |
19580676 | 1364 | return (SET_ERROR(ETXTBSY)); |
428870ff | 1365 | } |
19580676 | 1366 | drba->drba_snapobj = ds->ds_prev->ds_object; |
428870ff | 1367 | } |
19580676 MA |
1368 | |
1369 | dsl_dataset_rele(snap, FTAG); | |
428870ff | 1370 | } else { |
cf50a2b0 AG |
1371 | /* if full, then must be forced */ |
1372 | if (!drba->drba_cookie->drc_force) | |
1373 | return (SET_ERROR(EEXIST)); | |
1374 | /* start from $ORIGIN@$ORIGIN, if supported */ | |
1375 | drba->drba_snapobj = dp->dp_origin_snap != NULL ? | |
1376 | dp->dp_origin_snap->ds_object : 0; | |
428870ff | 1377 | } |
34dc7c2f | 1378 | |
34dc7c2f | 1379 | return (0); |
13fe0198 MA |
1380 | |
1381 | } | |
1382 | ||
1383 | static int | |
1384 | dmu_recv_begin_check(void *arg, dmu_tx_t *tx) | |
1385 | { | |
1386 | dmu_recv_begin_arg_t *drba = arg; | |
1387 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1388 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
1389 | uint64_t fromguid = drrb->drr_fromguid; | |
1390 | int flags = drrb->drr_flags; | |
1391 | int error; | |
9b67f605 | 1392 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); |
13fe0198 MA |
1393 | dsl_dataset_t *ds; |
1394 | const char *tofs = drba->drba_cookie->drc_tofs; | |
1395 | ||
1396 | /* already checked */ | |
1397 | ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); | |
47dfff3b | 1398 | ASSERT(!(featureflags & DMU_BACKUP_FEATURE_RESUMING)); |
13fe0198 MA |
1399 | |
1400 | if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == | |
1401 | DMU_COMPOUNDSTREAM || | |
1402 | drrb->drr_type >= DMU_OST_NUMTYPES || | |
1403 | ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) | |
2e528b49 | 1404 | return (SET_ERROR(EINVAL)); |
13fe0198 MA |
1405 | |
1406 | /* Verify pool version supports SA if SA_SPILL feature set */ | |
9b67f605 MA |
1407 | if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && |
1408 | spa_version(dp->dp_spa) < SPA_VERSION_SA) | |
1409 | return (SET_ERROR(ENOTSUP)); | |
1410 | ||
47dfff3b MA |
1411 | if (drba->drba_cookie->drc_resumable && |
1412 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EXTENSIBLE_DATASET)) | |
1413 | return (SET_ERROR(ENOTSUP)); | |
1414 | ||
9b67f605 MA |
1415 | /* |
1416 | * The receiving code doesn't know how to translate a WRITE_EMBEDDED | |
2aa34383 | 1417 | * record to a plain WRITE record, so the pool must have the |
9b67f605 MA |
1418 | * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED |
1419 | * records. Same with WRITE_EMBEDDED records that use LZ4 compression. | |
1420 | */ | |
1421 | if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && | |
1422 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) | |
1423 | return (SET_ERROR(ENOTSUP)); | |
2aa34383 | 1424 | if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && |
9b67f605 | 1425 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) |
2e528b49 | 1426 | return (SET_ERROR(ENOTSUP)); |
13fe0198 | 1427 | |
f1512ee6 MA |
1428 | /* |
1429 | * The receiving code doesn't know how to translate large blocks | |
1430 | * to smaller ones, so the pool must have the LARGE_BLOCKS | |
73aac4aa NB |
1431 | * feature enabled if the stream has LARGE_BLOCKS. Same with |
1432 | * large dnodes. | |
f1512ee6 MA |
1433 | */ |
1434 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
1435 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) | |
1436 | return (SET_ERROR(ENOTSUP)); | |
50c957f7 NB |
1437 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_DNODE) && |
1438 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_DNODE)) | |
1439 | return (SET_ERROR(ENOTSUP)); | |
1440 | ||
13fe0198 MA |
1441 | error = dsl_dataset_hold(dp, tofs, FTAG, &ds); |
1442 | if (error == 0) { | |
1443 | /* target fs already exists; recv into temp clone */ | |
1444 | ||
1445 | /* Can't recv a clone into an existing fs */ | |
e6d3a843 | 1446 | if (flags & DRR_FLAG_CLONE || drba->drba_origin) { |
13fe0198 | 1447 | dsl_dataset_rele(ds, FTAG); |
2e528b49 | 1448 | return (SET_ERROR(EINVAL)); |
13fe0198 MA |
1449 | } |
1450 | ||
1451 | error = recv_begin_check_existing_impl(drba, ds, fromguid); | |
1452 | dsl_dataset_rele(ds, FTAG); | |
1453 | } else if (error == ENOENT) { | |
1454 | /* target fs does not exist; must be a full backup or clone */ | |
eca7b760 | 1455 | char buf[ZFS_MAX_DATASET_NAME_LEN]; |
13fe0198 MA |
1456 | |
1457 | /* | |
1458 | * If it's a non-clone incremental, we are missing the | |
1459 | * target fs, so fail the recv. | |
1460 | */ | |
fcff0f35 PD |
1461 | if (fromguid != 0 && !(flags & DRR_FLAG_CLONE || |
1462 | drba->drba_origin)) | |
2e528b49 | 1463 | return (SET_ERROR(ENOENT)); |
13fe0198 | 1464 | |
e6d3a843 PD |
1465 | /* |
1466 | * If we're receiving a full send as a clone, and it doesn't | |
1467 | * contain all the necessary free records and freeobject | |
1468 | * records, reject it. | |
1469 | */ | |
1470 | if (fromguid == 0 && drba->drba_origin && | |
1471 | !(flags & DRR_FLAG_FREERECORDS)) | |
1472 | return (SET_ERROR(EINVAL)); | |
1473 | ||
13fe0198 | 1474 | /* Open the parent of tofs */ |
eca7b760 | 1475 | ASSERT3U(strlen(tofs), <, sizeof (buf)); |
13fe0198 MA |
1476 | (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); |
1477 | error = dsl_dataset_hold(dp, buf, FTAG, &ds); | |
1478 | if (error != 0) | |
1479 | return (error); | |
1480 | ||
788eb90c JJ |
1481 | /* |
1482 | * Check filesystem and snapshot limits before receiving. We'll | |
1483 | * recheck snapshot limits again at the end (we create the | |
1484 | * filesystems and increment those counts during begin_sync). | |
1485 | */ | |
1486 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, | |
1487 | ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred); | |
1488 | if (error != 0) { | |
1489 | dsl_dataset_rele(ds, FTAG); | |
1490 | return (error); | |
1491 | } | |
1492 | ||
1493 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, | |
1494 | ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); | |
1495 | if (error != 0) { | |
1496 | dsl_dataset_rele(ds, FTAG); | |
1497 | return (error); | |
1498 | } | |
1499 | ||
13fe0198 MA |
1500 | if (drba->drba_origin != NULL) { |
1501 | dsl_dataset_t *origin; | |
1502 | error = dsl_dataset_hold(dp, drba->drba_origin, | |
1503 | FTAG, &origin); | |
1504 | if (error != 0) { | |
1505 | dsl_dataset_rele(ds, FTAG); | |
1506 | return (error); | |
1507 | } | |
0c66c32d | 1508 | if (!origin->ds_is_snapshot) { |
13fe0198 MA |
1509 | dsl_dataset_rele(origin, FTAG); |
1510 | dsl_dataset_rele(ds, FTAG); | |
2e528b49 | 1511 | return (SET_ERROR(EINVAL)); |
13fe0198 | 1512 | } |
e6d3a843 PD |
1513 | if (dsl_dataset_phys(origin)->ds_guid != fromguid && |
1514 | fromguid != 0) { | |
13fe0198 MA |
1515 | dsl_dataset_rele(origin, FTAG); |
1516 | dsl_dataset_rele(ds, FTAG); | |
2e528b49 | 1517 | return (SET_ERROR(ENODEV)); |
13fe0198 MA |
1518 | } |
1519 | dsl_dataset_rele(origin, FTAG); | |
1520 | } | |
1521 | dsl_dataset_rele(ds, FTAG); | |
1522 | error = 0; | |
1523 | } | |
1524 | return (error); | |
34dc7c2f BB |
1525 | } |
1526 | ||
34dc7c2f | 1527 | static void |
13fe0198 | 1528 | dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1529 | { |
13fe0198 MA |
1530 | dmu_recv_begin_arg_t *drba = arg; |
1531 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
47dfff3b | 1532 | objset_t *mos = dp->dp_meta_objset; |
13fe0198 MA |
1533 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; |
1534 | const char *tofs = drba->drba_cookie->drc_tofs; | |
1535 | dsl_dataset_t *ds, *newds; | |
34dc7c2f | 1536 | uint64_t dsobj; |
13fe0198 | 1537 | int error; |
47dfff3b | 1538 | uint64_t crflags = 0; |
13fe0198 | 1539 | |
47dfff3b MA |
1540 | if (drrb->drr_flags & DRR_FLAG_CI_DATA) |
1541 | crflags |= DS_FLAG_CI_DATASET; | |
34dc7c2f | 1542 | |
13fe0198 MA |
1543 | error = dsl_dataset_hold(dp, tofs, FTAG, &ds); |
1544 | if (error == 0) { | |
1545 | /* create temporary clone */ | |
19580676 MA |
1546 | dsl_dataset_t *snap = NULL; |
1547 | if (drba->drba_snapobj != 0) { | |
1548 | VERIFY0(dsl_dataset_hold_obj(dp, | |
1549 | drba->drba_snapobj, FTAG, &snap)); | |
1550 | } | |
13fe0198 | 1551 | dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, |
19580676 | 1552 | snap, crflags, drba->drba_cred, tx); |
6b42ea85 PD |
1553 | if (drba->drba_snapobj != 0) |
1554 | dsl_dataset_rele(snap, FTAG); | |
13fe0198 MA |
1555 | dsl_dataset_rele(ds, FTAG); |
1556 | } else { | |
1557 | dsl_dir_t *dd; | |
1558 | const char *tail; | |
1559 | dsl_dataset_t *origin = NULL; | |
1560 | ||
1561 | VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); | |
1562 | ||
1563 | if (drba->drba_origin != NULL) { | |
1564 | VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, | |
1565 | FTAG, &origin)); | |
1566 | } | |
1567 | ||
1568 | /* Create new dataset. */ | |
1569 | dsobj = dsl_dataset_create_sync(dd, | |
1570 | strrchr(tofs, '/') + 1, | |
1571 | origin, crflags, drba->drba_cred, tx); | |
1572 | if (origin != NULL) | |
1573 | dsl_dataset_rele(origin, FTAG); | |
1574 | dsl_dir_rele(dd, FTAG); | |
1575 | drba->drba_cookie->drc_newfs = B_TRUE; | |
1576 | } | |
1577 | VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds)); | |
1578 | ||
47dfff3b MA |
1579 | if (drba->drba_cookie->drc_resumable) { |
1580 | uint64_t one = 1; | |
1581 | uint64_t zero = 0; | |
1582 | ||
1583 | dsl_dataset_zapify(newds, tx); | |
1584 | if (drrb->drr_fromguid != 0) { | |
1585 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_FROMGUID, | |
1586 | 8, 1, &drrb->drr_fromguid, tx)); | |
1587 | } | |
1588 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TOGUID, | |
1589 | 8, 1, &drrb->drr_toguid, tx)); | |
1590 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TONAME, | |
1591 | 1, strlen(drrb->drr_toname) + 1, drrb->drr_toname, tx)); | |
1592 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OBJECT, | |
1593 | 8, 1, &one, tx)); | |
1594 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OFFSET, | |
1595 | 8, 1, &zero, tx)); | |
1596 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_BYTES, | |
1597 | 8, 1, &zero, tx)); | |
2aa34383 DK |
1598 | if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & |
1599 | DMU_BACKUP_FEATURE_LARGE_BLOCKS) { | |
1600 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_LARGEBLOCK, | |
02730c33 | 1601 | 8, 1, &one, tx)); |
2aa34383 | 1602 | } |
47dfff3b MA |
1603 | if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & |
1604 | DMU_BACKUP_FEATURE_EMBED_DATA) { | |
1605 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_EMBEDOK, | |
1606 | 8, 1, &one, tx)); | |
1607 | } | |
2aa34383 DK |
1608 | if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & |
1609 | DMU_BACKUP_FEATURE_COMPRESSED) { | |
1610 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_COMPRESSOK, | |
1611 | 8, 1, &one, tx)); | |
1612 | } | |
47dfff3b MA |
1613 | } |
1614 | ||
13fe0198 | 1615 | dmu_buf_will_dirty(newds->ds_dbuf, tx); |
d683ddbb | 1616 | dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; |
34dc7c2f | 1617 | |
428870ff BB |
1618 | /* |
1619 | * If we actually created a non-clone, we need to create the | |
1620 | * objset in our new dataset. | |
1621 | */ | |
cc9bb3e5 | 1622 | rrw_enter(&newds->ds_bp_rwlock, RW_READER, FTAG); |
13fe0198 | 1623 | if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) { |
428870ff | 1624 | (void) dmu_objset_create_impl(dp->dp_spa, |
13fe0198 | 1625 | newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); |
428870ff | 1626 | } |
cc9bb3e5 | 1627 | rrw_exit(&newds->ds_bp_rwlock, FTAG); |
34dc7c2f | 1628 | |
13fe0198 | 1629 | drba->drba_cookie->drc_ds = newds; |
428870ff | 1630 | |
13fe0198 | 1631 | spa_history_log_internal_ds(newds, "receive", tx, ""); |
34dc7c2f BB |
1632 | } |
1633 | ||
47dfff3b MA |
1634 | static int |
1635 | dmu_recv_resume_begin_check(void *arg, dmu_tx_t *tx) | |
1636 | { | |
1637 | dmu_recv_begin_arg_t *drba = arg; | |
1638 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1639 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
1640 | int error; | |
1641 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); | |
1642 | dsl_dataset_t *ds; | |
1643 | const char *tofs = drba->drba_cookie->drc_tofs; | |
47dfff3b MA |
1644 | uint64_t val; |
1645 | ||
eca7b760 IK |
1646 | /* 6 extra bytes for /%recv */ |
1647 | char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; | |
1648 | ||
47dfff3b MA |
1649 | /* already checked */ |
1650 | ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); | |
1651 | ASSERT(featureflags & DMU_BACKUP_FEATURE_RESUMING); | |
1652 | ||
1653 | if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == | |
1654 | DMU_COMPOUNDSTREAM || | |
1655 | drrb->drr_type >= DMU_OST_NUMTYPES) | |
1656 | return (SET_ERROR(EINVAL)); | |
1657 | ||
1658 | /* Verify pool version supports SA if SA_SPILL feature set */ | |
1659 | if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && | |
1660 | spa_version(dp->dp_spa) < SPA_VERSION_SA) | |
1661 | return (SET_ERROR(ENOTSUP)); | |
1662 | ||
1663 | /* | |
1664 | * The receiving code doesn't know how to translate a WRITE_EMBEDDED | |
1665 | * record to a plain WRITE record, so the pool must have the | |
1666 | * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED | |
1667 | * records. Same with WRITE_EMBEDDED records that use LZ4 compression. | |
1668 | */ | |
1669 | if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && | |
1670 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) | |
1671 | return (SET_ERROR(ENOTSUP)); | |
2aa34383 | 1672 | if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && |
47dfff3b MA |
1673 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) |
1674 | return (SET_ERROR(ENOTSUP)); | |
1675 | ||
73aac4aa NB |
1676 | /* |
1677 | * The receiving code doesn't know how to translate large blocks | |
1678 | * to smaller ones, so the pool must have the LARGE_BLOCKS | |
1679 | * feature enabled if the stream has LARGE_BLOCKS. Same with | |
1680 | * large dnodes. | |
1681 | */ | |
1682 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
1683 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) | |
1684 | return (SET_ERROR(ENOTSUP)); | |
1685 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_DNODE) && | |
1686 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_DNODE)) | |
1687 | return (SET_ERROR(ENOTSUP)); | |
1688 | ||
47dfff3b MA |
1689 | (void) snprintf(recvname, sizeof (recvname), "%s/%s", |
1690 | tofs, recv_clone_name); | |
1691 | ||
1692 | if (dsl_dataset_hold(dp, recvname, FTAG, &ds) != 0) { | |
1693 | /* %recv does not exist; continue in tofs */ | |
1694 | error = dsl_dataset_hold(dp, tofs, FTAG, &ds); | |
1695 | if (error != 0) | |
1696 | return (error); | |
1697 | } | |
1698 | ||
1699 | /* check that ds is marked inconsistent */ | |
1700 | if (!DS_IS_INCONSISTENT(ds)) { | |
1701 | dsl_dataset_rele(ds, FTAG); | |
1702 | return (SET_ERROR(EINVAL)); | |
1703 | } | |
1704 | ||
1705 | /* check that there is resuming data, and that the toguid matches */ | |
1706 | if (!dsl_dataset_is_zapified(ds)) { | |
1707 | dsl_dataset_rele(ds, FTAG); | |
1708 | return (SET_ERROR(EINVAL)); | |
1709 | } | |
1710 | error = zap_lookup(dp->dp_meta_objset, ds->ds_object, | |
1711 | DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val); | |
1712 | if (error != 0 || drrb->drr_toguid != val) { | |
1713 | dsl_dataset_rele(ds, FTAG); | |
1714 | return (SET_ERROR(EINVAL)); | |
1715 | } | |
1716 | ||
1717 | /* | |
1718 | * Check if the receive is still running. If so, it will be owned. | |
1719 | * Note that nothing else can own the dataset (e.g. after the receive | |
1720 | * fails) because it will be marked inconsistent. | |
1721 | */ | |
1722 | if (dsl_dataset_has_owner(ds)) { | |
1723 | dsl_dataset_rele(ds, FTAG); | |
1724 | return (SET_ERROR(EBUSY)); | |
1725 | } | |
1726 | ||
1727 | /* There should not be any snapshots of this fs yet. */ | |
1728 | if (ds->ds_prev != NULL && ds->ds_prev->ds_dir == ds->ds_dir) { | |
1729 | dsl_dataset_rele(ds, FTAG); | |
1730 | return (SET_ERROR(EINVAL)); | |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * Note: resume point will be checked when we process the first WRITE | |
1735 | * record. | |
1736 | */ | |
1737 | ||
1738 | /* check that the origin matches */ | |
1739 | val = 0; | |
1740 | (void) zap_lookup(dp->dp_meta_objset, ds->ds_object, | |
1741 | DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val); | |
1742 | if (drrb->drr_fromguid != val) { | |
1743 | dsl_dataset_rele(ds, FTAG); | |
1744 | return (SET_ERROR(EINVAL)); | |
1745 | } | |
1746 | ||
1747 | dsl_dataset_rele(ds, FTAG); | |
1748 | return (0); | |
1749 | } | |
1750 | ||
1751 | static void | |
1752 | dmu_recv_resume_begin_sync(void *arg, dmu_tx_t *tx) | |
1753 | { | |
1754 | dmu_recv_begin_arg_t *drba = arg; | |
1755 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1756 | const char *tofs = drba->drba_cookie->drc_tofs; | |
1757 | dsl_dataset_t *ds; | |
1758 | uint64_t dsobj; | |
eca7b760 IK |
1759 | /* 6 extra bytes for /%recv */ |
1760 | char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; | |
47dfff3b MA |
1761 | |
1762 | (void) snprintf(recvname, sizeof (recvname), "%s/%s", | |
1763 | tofs, recv_clone_name); | |
1764 | ||
1765 | if (dsl_dataset_hold(dp, recvname, FTAG, &ds) != 0) { | |
1766 | /* %recv does not exist; continue in tofs */ | |
1767 | VERIFY0(dsl_dataset_hold(dp, tofs, FTAG, &ds)); | |
1768 | drba->drba_cookie->drc_newfs = B_TRUE; | |
1769 | } | |
1770 | ||
1771 | /* clear the inconsistent flag so that we can own it */ | |
1772 | ASSERT(DS_IS_INCONSISTENT(ds)); | |
1773 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
1774 | dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; | |
1775 | dsobj = ds->ds_object; | |
1776 | dsl_dataset_rele(ds, FTAG); | |
1777 | ||
1778 | VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &ds)); | |
1779 | ||
1780 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
1781 | dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT; | |
1782 | ||
cc9bb3e5 | 1783 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
47dfff3b | 1784 | ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds))); |
cc9bb3e5 | 1785 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
47dfff3b MA |
1786 | |
1787 | drba->drba_cookie->drc_ds = ds; | |
1788 | ||
1789 | spa_history_log_internal_ds(ds, "resume receive", tx, ""); | |
1790 | } | |
1791 | ||
34dc7c2f BB |
1792 | /* |
1793 | * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() | |
1794 | * succeeds; otherwise we will leak the holds on the datasets. | |
1795 | */ | |
1796 | int | |
47dfff3b MA |
1797 | dmu_recv_begin(char *tofs, char *tosnap, dmu_replay_record_t *drr_begin, |
1798 | boolean_t force, boolean_t resumable, char *origin, dmu_recv_cookie_t *drc) | |
34dc7c2f | 1799 | { |
13fe0198 | 1800 | dmu_recv_begin_arg_t drba = { 0 }; |
34dc7c2f BB |
1801 | |
1802 | bzero(drc, sizeof (dmu_recv_cookie_t)); | |
47dfff3b MA |
1803 | drc->drc_drr_begin = drr_begin; |
1804 | drc->drc_drrb = &drr_begin->drr_u.drr_begin; | |
34dc7c2f | 1805 | drc->drc_tosnap = tosnap; |
13fe0198 | 1806 | drc->drc_tofs = tofs; |
34dc7c2f | 1807 | drc->drc_force = force; |
47dfff3b | 1808 | drc->drc_resumable = resumable; |
788eb90c | 1809 | drc->drc_cred = CRED(); |
34dc7c2f | 1810 | |
47dfff3b | 1811 | if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { |
13fe0198 | 1812 | drc->drc_byteswap = B_TRUE; |
a6255b7f | 1813 | (void) fletcher_4_incremental_byteswap(drr_begin, |
13fe0198 | 1814 | sizeof (dmu_replay_record_t), &drc->drc_cksum); |
47dfff3b MA |
1815 | byteswap_record(drr_begin); |
1816 | } else if (drc->drc_drrb->drr_magic == DMU_BACKUP_MAGIC) { | |
a6255b7f | 1817 | (void) fletcher_4_incremental_native(drr_begin, |
13fe0198 | 1818 | sizeof (dmu_replay_record_t), &drc->drc_cksum); |
47dfff3b MA |
1819 | } else { |
1820 | return (SET_ERROR(EINVAL)); | |
34dc7c2f BB |
1821 | } |
1822 | ||
13fe0198 MA |
1823 | drba.drba_origin = origin; |
1824 | drba.drba_cookie = drc; | |
1825 | drba.drba_cred = CRED(); | |
1826 | ||
47dfff3b MA |
1827 | if (DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & |
1828 | DMU_BACKUP_FEATURE_RESUMING) { | |
1829 | return (dsl_sync_task(tofs, | |
1830 | dmu_recv_resume_begin_check, dmu_recv_resume_begin_sync, | |
1831 | &drba, 5, ZFS_SPACE_CHECK_NORMAL)); | |
1832 | } else { | |
1833 | return (dsl_sync_task(tofs, | |
1834 | dmu_recv_begin_check, dmu_recv_begin_sync, | |
1835 | &drba, 5, ZFS_SPACE_CHECK_NORMAL)); | |
1836 | } | |
34dc7c2f BB |
1837 | } |
1838 | ||
fcff0f35 PD |
1839 | struct receive_record_arg { |
1840 | dmu_replay_record_t header; | |
1841 | void *payload; /* Pointer to a buffer containing the payload */ | |
1842 | /* | |
1843 | * If the record is a write, pointer to the arc_buf_t containing the | |
1844 | * payload. | |
1845 | */ | |
1846 | arc_buf_t *write_buf; | |
1847 | int payload_size; | |
47dfff3b | 1848 | uint64_t bytes_read; /* bytes read from stream when record created */ |
fcff0f35 PD |
1849 | boolean_t eos_marker; /* Marks the end of the stream */ |
1850 | bqueue_node_t node; | |
1851 | }; | |
1852 | ||
1853 | struct receive_writer_arg { | |
37f8a883 | 1854 | objset_t *os; |
13fe0198 | 1855 | boolean_t byteswap; |
fcff0f35 | 1856 | bqueue_t q; |
47dfff3b | 1857 | |
fcff0f35 PD |
1858 | /* |
1859 | * These three args are used to signal to the main thread that we're | |
1860 | * done. | |
1861 | */ | |
1862 | kmutex_t mutex; | |
1863 | kcondvar_t cv; | |
1864 | boolean_t done; | |
47dfff3b | 1865 | |
fcff0f35 PD |
1866 | int err; |
1867 | /* A map from guid to dataset to help handle dedup'd streams. */ | |
1868 | avl_tree_t *guid_to_ds_map; | |
47dfff3b MA |
1869 | boolean_t resumable; |
1870 | uint64_t last_object, last_offset; | |
1871 | uint64_t bytes_read; /* bytes read when current record created */ | |
fcff0f35 | 1872 | }; |
37f8a883 | 1873 | |
e6d3a843 PD |
1874 | struct objlist { |
1875 | list_t list; /* List of struct receive_objnode. */ | |
1876 | /* | |
1877 | * Last object looked up. Used to assert that objects are being looked | |
1878 | * up in ascending order. | |
1879 | */ | |
1880 | uint64_t last_lookup; | |
1881 | }; | |
1882 | ||
1883 | struct receive_objnode { | |
1884 | list_node_t node; | |
1885 | uint64_t object; | |
1886 | }; | |
1887 | ||
fcff0f35 PD |
1888 | struct receive_arg { |
1889 | objset_t *os; | |
1890 | vnode_t *vp; /* The vnode to read the stream from */ | |
1891 | uint64_t voff; /* The current offset in the stream */ | |
47dfff3b | 1892 | uint64_t bytes_read; |
fcff0f35 PD |
1893 | /* |
1894 | * A record that has had its payload read in, but hasn't yet been handed | |
1895 | * off to the worker thread. | |
1896 | */ | |
1897 | struct receive_record_arg *rrd; | |
1898 | /* A record that has had its header read in, but not its payload. */ | |
1899 | struct receive_record_arg *next_rrd; | |
34dc7c2f | 1900 | zio_cksum_t cksum; |
37f8a883 | 1901 | zio_cksum_t prev_cksum; |
fcff0f35 PD |
1902 | int err; |
1903 | boolean_t byteswap; | |
1904 | /* Sorted list of objects not to issue prefetches for. */ | |
e6d3a843 | 1905 | struct objlist ignore_objlist; |
34dc7c2f BB |
1906 | }; |
1907 | ||
428870ff BB |
1908 | typedef struct guid_map_entry { |
1909 | uint64_t guid; | |
1910 | dsl_dataset_t *gme_ds; | |
1911 | avl_node_t avlnode; | |
1912 | } guid_map_entry_t; | |
1913 | ||
1914 | static int | |
1915 | guid_compare(const void *arg1, const void *arg2) | |
1916 | { | |
ee36c709 GN |
1917 | const guid_map_entry_t *gmep1 = (const guid_map_entry_t *)arg1; |
1918 | const guid_map_entry_t *gmep2 = (const guid_map_entry_t *)arg2; | |
428870ff | 1919 | |
ee36c709 | 1920 | return (AVL_CMP(gmep1->guid, gmep2->guid)); |
428870ff BB |
1921 | } |
1922 | ||
572e2857 BB |
1923 | static void |
1924 | free_guid_map_onexit(void *arg) | |
1925 | { | |
1926 | avl_tree_t *ca = arg; | |
1927 | void *cookie = NULL; | |
1928 | guid_map_entry_t *gmep; | |
1929 | ||
1930 | while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { | |
13fe0198 | 1931 | dsl_dataset_long_rele(gmep->gme_ds, gmep); |
7ec09286 | 1932 | dsl_dataset_rele(gmep->gme_ds, gmep); |
572e2857 BB |
1933 | kmem_free(gmep, sizeof (guid_map_entry_t)); |
1934 | } | |
1935 | avl_destroy(ca); | |
1936 | kmem_free(ca, sizeof (avl_tree_t)); | |
1937 | } | |
1938 | ||
37f8a883 | 1939 | static int |
fcff0f35 | 1940 | receive_read(struct receive_arg *ra, int len, void *buf) |
34dc7c2f | 1941 | { |
34dc7c2f BB |
1942 | int done = 0; |
1943 | ||
f8866f8a ER |
1944 | /* |
1945 | * The code doesn't rely on this (lengths being multiples of 8). See | |
1946 | * comment in dump_bytes. | |
1947 | */ | |
c99c9001 | 1948 | ASSERT0(len % 8); |
34dc7c2f BB |
1949 | |
1950 | while (done < len) { | |
1951 | ssize_t resid; | |
1952 | ||
1953 | ra->err = vn_rdwr(UIO_READ, ra->vp, | |
37f8a883 | 1954 | (char *)buf + done, len - done, |
34dc7c2f BB |
1955 | ra->voff, UIO_SYSSPACE, FAPPEND, |
1956 | RLIM64_INFINITY, CRED(), &resid); | |
1957 | ||
47dfff3b MA |
1958 | if (resid == len - done) { |
1959 | /* | |
1960 | * Note: ECKSUM indicates that the receive | |
1961 | * was interrupted and can potentially be resumed. | |
1962 | */ | |
1963 | ra->err = SET_ERROR(ECKSUM); | |
1964 | } | |
34dc7c2f BB |
1965 | ra->voff += len - done - resid; |
1966 | done = len - resid; | |
13fe0198 | 1967 | if (ra->err != 0) |
37f8a883 | 1968 | return (ra->err); |
34dc7c2f BB |
1969 | } |
1970 | ||
47dfff3b MA |
1971 | ra->bytes_read += len; |
1972 | ||
34dc7c2f | 1973 | ASSERT3U(done, ==, len); |
37f8a883 | 1974 | return (0); |
34dc7c2f BB |
1975 | } |
1976 | ||
60948de1 | 1977 | noinline static void |
37f8a883 | 1978 | byteswap_record(dmu_replay_record_t *drr) |
34dc7c2f BB |
1979 | { |
1980 | #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) | |
1981 | #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) | |
1982 | drr->drr_type = BSWAP_32(drr->drr_type); | |
1983 | drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); | |
37f8a883 | 1984 | |
34dc7c2f BB |
1985 | switch (drr->drr_type) { |
1986 | case DRR_BEGIN: | |
1987 | DO64(drr_begin.drr_magic); | |
428870ff | 1988 | DO64(drr_begin.drr_versioninfo); |
34dc7c2f BB |
1989 | DO64(drr_begin.drr_creation_time); |
1990 | DO32(drr_begin.drr_type); | |
1991 | DO32(drr_begin.drr_flags); | |
1992 | DO64(drr_begin.drr_toguid); | |
1993 | DO64(drr_begin.drr_fromguid); | |
1994 | break; | |
1995 | case DRR_OBJECT: | |
1996 | DO64(drr_object.drr_object); | |
34dc7c2f BB |
1997 | DO32(drr_object.drr_type); |
1998 | DO32(drr_object.drr_bonustype); | |
1999 | DO32(drr_object.drr_blksz); | |
2000 | DO32(drr_object.drr_bonuslen); | |
428870ff | 2001 | DO64(drr_object.drr_toguid); |
34dc7c2f BB |
2002 | break; |
2003 | case DRR_FREEOBJECTS: | |
2004 | DO64(drr_freeobjects.drr_firstobj); | |
2005 | DO64(drr_freeobjects.drr_numobjs); | |
428870ff | 2006 | DO64(drr_freeobjects.drr_toguid); |
34dc7c2f BB |
2007 | break; |
2008 | case DRR_WRITE: | |
2009 | DO64(drr_write.drr_object); | |
2010 | DO32(drr_write.drr_type); | |
2011 | DO64(drr_write.drr_offset); | |
2aa34383 | 2012 | DO64(drr_write.drr_logical_size); |
428870ff | 2013 | DO64(drr_write.drr_toguid); |
37f8a883 | 2014 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write.drr_key.ddk_cksum); |
428870ff | 2015 | DO64(drr_write.drr_key.ddk_prop); |
2aa34383 | 2016 | DO64(drr_write.drr_compressed_size); |
428870ff BB |
2017 | break; |
2018 | case DRR_WRITE_BYREF: | |
2019 | DO64(drr_write_byref.drr_object); | |
2020 | DO64(drr_write_byref.drr_offset); | |
2021 | DO64(drr_write_byref.drr_length); | |
2022 | DO64(drr_write_byref.drr_toguid); | |
2023 | DO64(drr_write_byref.drr_refguid); | |
2024 | DO64(drr_write_byref.drr_refobject); | |
2025 | DO64(drr_write_byref.drr_refoffset); | |
37f8a883 MA |
2026 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write_byref. |
2027 | drr_key.ddk_cksum); | |
428870ff | 2028 | DO64(drr_write_byref.drr_key.ddk_prop); |
34dc7c2f | 2029 | break; |
9b67f605 MA |
2030 | case DRR_WRITE_EMBEDDED: |
2031 | DO64(drr_write_embedded.drr_object); | |
2032 | DO64(drr_write_embedded.drr_offset); | |
2033 | DO64(drr_write_embedded.drr_length); | |
2034 | DO64(drr_write_embedded.drr_toguid); | |
2035 | DO32(drr_write_embedded.drr_lsize); | |
2036 | DO32(drr_write_embedded.drr_psize); | |
2037 | break; | |
34dc7c2f BB |
2038 | case DRR_FREE: |
2039 | DO64(drr_free.drr_object); | |
2040 | DO64(drr_free.drr_offset); | |
2041 | DO64(drr_free.drr_length); | |
428870ff BB |
2042 | DO64(drr_free.drr_toguid); |
2043 | break; | |
2044 | case DRR_SPILL: | |
2045 | DO64(drr_spill.drr_object); | |
2046 | DO64(drr_spill.drr_length); | |
2047 | DO64(drr_spill.drr_toguid); | |
34dc7c2f BB |
2048 | break; |
2049 | case DRR_END: | |
428870ff | 2050 | DO64(drr_end.drr_toguid); |
37f8a883 | 2051 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_end.drr_checksum); |
34dc7c2f | 2052 | break; |
e75c13c3 BB |
2053 | default: |
2054 | break; | |
34dc7c2f | 2055 | } |
37f8a883 MA |
2056 | |
2057 | if (drr->drr_type != DRR_BEGIN) { | |
2058 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_checksum.drr_checksum); | |
2059 | } | |
2060 | ||
34dc7c2f BB |
2061 | #undef DO64 |
2062 | #undef DO32 | |
2063 | } | |
2064 | ||
6c59307a MA |
2065 | static inline uint8_t |
2066 | deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) | |
2067 | { | |
2068 | if (bonus_type == DMU_OT_SA) { | |
2069 | return (1); | |
2070 | } else { | |
2071 | return (1 + | |
50c957f7 NB |
2072 | ((DN_OLD_MAX_BONUSLEN - |
2073 | MIN(DN_OLD_MAX_BONUSLEN, bonus_size)) >> SPA_BLKPTRSHIFT)); | |
6c59307a MA |
2074 | } |
2075 | } | |
2076 | ||
47dfff3b MA |
2077 | static void |
2078 | save_resume_state(struct receive_writer_arg *rwa, | |
2079 | uint64_t object, uint64_t offset, dmu_tx_t *tx) | |
2080 | { | |
2081 | int txgoff = dmu_tx_get_txg(tx) & TXG_MASK; | |
2082 | ||
2083 | if (!rwa->resumable) | |
2084 | return; | |
2085 | ||
2086 | /* | |
2087 | * We use ds_resume_bytes[] != 0 to indicate that we need to | |
2088 | * update this on disk, so it must not be 0. | |
2089 | */ | |
2090 | ASSERT(rwa->bytes_read != 0); | |
2091 | ||
2092 | /* | |
2093 | * We only resume from write records, which have a valid | |
2094 | * (non-meta-dnode) object number. | |
2095 | */ | |
2096 | ASSERT(object != 0); | |
2097 | ||
2098 | /* | |
2099 | * For resuming to work correctly, we must receive records in order, | |
2100 | * sorted by object,offset. This is checked by the callers, but | |
2101 | * assert it here for good measure. | |
2102 | */ | |
2103 | ASSERT3U(object, >=, rwa->os->os_dsl_dataset->ds_resume_object[txgoff]); | |
2104 | ASSERT(object != rwa->os->os_dsl_dataset->ds_resume_object[txgoff] || | |
2105 | offset >= rwa->os->os_dsl_dataset->ds_resume_offset[txgoff]); | |
2106 | ASSERT3U(rwa->bytes_read, >=, | |
2107 | rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff]); | |
2108 | ||
2109 | rwa->os->os_dsl_dataset->ds_resume_object[txgoff] = object; | |
2110 | rwa->os->os_dsl_dataset->ds_resume_offset[txgoff] = offset; | |
2111 | rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff] = rwa->bytes_read; | |
2112 | } | |
2113 | ||
60948de1 | 2114 | noinline static int |
fcff0f35 | 2115 | receive_object(struct receive_writer_arg *rwa, struct drr_object *drro, |
4ea3f864 | 2116 | void *data) |
34dc7c2f | 2117 | { |
6c59307a | 2118 | dmu_object_info_t doi; |
34dc7c2f | 2119 | dmu_tx_t *tx; |
6c59307a MA |
2120 | uint64_t object; |
2121 | int err; | |
34dc7c2f | 2122 | |
34dc7c2f | 2123 | if (drro->drr_type == DMU_OT_NONE || |
9ae529ec CS |
2124 | !DMU_OT_IS_VALID(drro->drr_type) || |
2125 | !DMU_OT_IS_VALID(drro->drr_bonustype) || | |
428870ff | 2126 | drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || |
34dc7c2f BB |
2127 | drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || |
2128 | P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || | |
2129 | drro->drr_blksz < SPA_MINBLOCKSIZE || | |
fcff0f35 | 2130 | drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(rwa->os)) || |
50c957f7 | 2131 | drro->drr_bonuslen > |
73aac4aa NB |
2132 | DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa->os))) || |
2133 | drro->drr_dn_slots > | |
2134 | (spa_maxdnodesize(dmu_objset_spa(rwa->os)) >> DNODE_SHIFT)) { | |
2e528b49 | 2135 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
2136 | } |
2137 | ||
fcff0f35 | 2138 | err = dmu_object_info(rwa->os, drro->drr_object, &doi); |
9babb374 BB |
2139 | |
2140 | if (err != 0 && err != ENOENT) | |
2e528b49 | 2141 | return (SET_ERROR(EINVAL)); |
6c59307a | 2142 | object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT; |
9babb374 | 2143 | |
6c59307a MA |
2144 | /* |
2145 | * If we are losing blkptrs or changing the block size this must | |
2146 | * be a new file instance. We must clear out the previous file | |
2147 | * contents before we can change this type of metadata in the dnode. | |
2148 | */ | |
2149 | if (err == 0) { | |
2150 | int nblkptr; | |
2151 | ||
2152 | nblkptr = deduce_nblkptr(drro->drr_bonustype, | |
2153 | drro->drr_bonuslen); | |
2154 | ||
2155 | if (drro->drr_blksz != doi.doi_data_block_size || | |
2156 | nblkptr < doi.doi_nblkptr) { | |
fcff0f35 | 2157 | err = dmu_free_long_range(rwa->os, drro->drr_object, |
6c59307a MA |
2158 | 0, DMU_OBJECT_END); |
2159 | if (err != 0) | |
2160 | return (SET_ERROR(EINVAL)); | |
34dc7c2f | 2161 | } |
6c59307a MA |
2162 | } |
2163 | ||
fcff0f35 | 2164 | tx = dmu_tx_create(rwa->os); |
6c59307a MA |
2165 | dmu_tx_hold_bonus(tx, object); |
2166 | err = dmu_tx_assign(tx, TXG_WAIT); | |
2167 | if (err != 0) { | |
2168 | dmu_tx_abort(tx); | |
2169 | return (err); | |
2170 | } | |
2171 | ||
2172 | if (object == DMU_NEW_OBJECT) { | |
2173 | /* currently free, want to be allocated */ | |
50c957f7 | 2174 | err = dmu_object_claim_dnsize(rwa->os, drro->drr_object, |
34dc7c2f | 2175 | drro->drr_type, drro->drr_blksz, |
50c957f7 NB |
2176 | drro->drr_bonustype, drro->drr_bonuslen, |
2177 | drro->drr_dn_slots << DNODE_SHIFT, tx); | |
6c59307a MA |
2178 | } else if (drro->drr_type != doi.doi_type || |
2179 | drro->drr_blksz != doi.doi_data_block_size || | |
2180 | drro->drr_bonustype != doi.doi_bonus_type || | |
2181 | drro->drr_bonuslen != doi.doi_bonus_size) { | |
2182 | /* currently allocated, but with different properties */ | |
fcff0f35 | 2183 | err = dmu_object_reclaim(rwa->os, drro->drr_object, |
34dc7c2f | 2184 | drro->drr_type, drro->drr_blksz, |
6c59307a | 2185 | drro->drr_bonustype, drro->drr_bonuslen, tx); |
34dc7c2f | 2186 | } |
13fe0198 | 2187 | if (err != 0) { |
6c59307a | 2188 | dmu_tx_commit(tx); |
2e528b49 | 2189 | return (SET_ERROR(EINVAL)); |
428870ff | 2190 | } |
9babb374 | 2191 | |
fcff0f35 | 2192 | dmu_object_set_checksum(rwa->os, drro->drr_object, |
37f8a883 | 2193 | drro->drr_checksumtype, tx); |
fcff0f35 | 2194 | dmu_object_set_compress(rwa->os, drro->drr_object, |
37f8a883 | 2195 | drro->drr_compress, tx); |
34dc7c2f | 2196 | |
b128c09f | 2197 | if (data != NULL) { |
34dc7c2f | 2198 | dmu_buf_t *db; |
b128c09f | 2199 | |
fcff0f35 | 2200 | VERIFY0(dmu_bonus_hold(rwa->os, drro->drr_object, FTAG, &db)); |
34dc7c2f BB |
2201 | dmu_buf_will_dirty(db, tx); |
2202 | ||
2203 | ASSERT3U(db->db_size, >=, drro->drr_bonuslen); | |
34dc7c2f | 2204 | bcopy(data, db->db_data, drro->drr_bonuslen); |
fcff0f35 | 2205 | if (rwa->byteswap) { |
9ae529ec CS |
2206 | dmu_object_byteswap_t byteswap = |
2207 | DMU_OT_BYTESWAP(drro->drr_bonustype); | |
2208 | dmu_ot_byteswap[byteswap].ob_func(db->db_data, | |
34dc7c2f BB |
2209 | drro->drr_bonuslen); |
2210 | } | |
2211 | dmu_buf_rele(db, FTAG); | |
2212 | } | |
2213 | dmu_tx_commit(tx); | |
47dfff3b | 2214 | |
34dc7c2f BB |
2215 | return (0); |
2216 | } | |
2217 | ||
2218 | /* ARGSUSED */ | |
60948de1 | 2219 | noinline static int |
fcff0f35 | 2220 | receive_freeobjects(struct receive_writer_arg *rwa, |
34dc7c2f BB |
2221 | struct drr_freeobjects *drrfo) |
2222 | { | |
2223 | uint64_t obj; | |
e6d3a843 | 2224 | int next_err = 0; |
34dc7c2f BB |
2225 | |
2226 | if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) | |
2e528b49 | 2227 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 2228 | |
50c957f7 | 2229 | for (obj = drrfo->drr_firstobj == 0 ? 1 : drrfo->drr_firstobj; |
e6d3a843 PD |
2230 | obj < drrfo->drr_firstobj + drrfo->drr_numobjs && next_err == 0; |
2231 | next_err = dmu_object_next(rwa->os, &obj, FALSE, 0)) { | |
50c957f7 | 2232 | dmu_object_info_t doi; |
34dc7c2f BB |
2233 | int err; |
2234 | ||
50c957f7 NB |
2235 | err = dmu_object_info(rwa->os, obj, &doi); |
2236 | if (err == ENOENT) { | |
2237 | obj++; | |
34dc7c2f | 2238 | continue; |
50c957f7 NB |
2239 | } else if (err != 0) { |
2240 | return (err); | |
2241 | } | |
34dc7c2f | 2242 | |
fcff0f35 | 2243 | err = dmu_free_long_object(rwa->os, obj); |
13fe0198 | 2244 | if (err != 0) |
34dc7c2f | 2245 | return (err); |
34dc7c2f | 2246 | } |
e6d3a843 PD |
2247 | if (next_err != ESRCH) |
2248 | return (next_err); | |
34dc7c2f BB |
2249 | return (0); |
2250 | } | |
2251 | ||
60948de1 | 2252 | noinline static int |
fcff0f35 | 2253 | receive_write(struct receive_writer_arg *rwa, struct drr_write *drrw, |
4ea3f864 | 2254 | arc_buf_t *abuf) |
34dc7c2f BB |
2255 | { |
2256 | dmu_tx_t *tx; | |
88904bb3 | 2257 | dmu_buf_t *bonus; |
34dc7c2f BB |
2258 | int err; |
2259 | ||
2aa34383 | 2260 | if (drrw->drr_offset + drrw->drr_logical_size < drrw->drr_offset || |
9ae529ec | 2261 | !DMU_OT_IS_VALID(drrw->drr_type)) |
2e528b49 | 2262 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 2263 | |
47dfff3b MA |
2264 | /* |
2265 | * For resuming to work, records must be in increasing order | |
2266 | * by (object, offset). | |
2267 | */ | |
2268 | if (drrw->drr_object < rwa->last_object || | |
2269 | (drrw->drr_object == rwa->last_object && | |
2270 | drrw->drr_offset < rwa->last_offset)) { | |
2271 | return (SET_ERROR(EINVAL)); | |
2272 | } | |
2273 | rwa->last_object = drrw->drr_object; | |
2274 | rwa->last_offset = drrw->drr_offset; | |
2275 | ||
fcff0f35 | 2276 | if (dmu_object_info(rwa->os, drrw->drr_object, NULL) != 0) |
88904bb3 MA |
2277 | return (SET_ERROR(EINVAL)); |
2278 | ||
fcff0f35 | 2279 | tx = dmu_tx_create(rwa->os); |
34dc7c2f BB |
2280 | |
2281 | dmu_tx_hold_write(tx, drrw->drr_object, | |
2aa34383 | 2282 | drrw->drr_offset, drrw->drr_logical_size); |
34dc7c2f | 2283 | err = dmu_tx_assign(tx, TXG_WAIT); |
13fe0198 | 2284 | if (err != 0) { |
34dc7c2f BB |
2285 | dmu_tx_abort(tx); |
2286 | return (err); | |
2287 | } | |
fcff0f35 | 2288 | if (rwa->byteswap) { |
9ae529ec CS |
2289 | dmu_object_byteswap_t byteswap = |
2290 | DMU_OT_BYTESWAP(drrw->drr_type); | |
37f8a883 | 2291 | dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, |
2aa34383 | 2292 | DRR_WRITE_PAYLOAD_SIZE(drrw)); |
9ae529ec | 2293 | } |
37f8a883 | 2294 | |
2aa34383 | 2295 | /* use the bonus buf to look up the dnode in dmu_assign_arcbuf */ |
fcff0f35 | 2296 | if (dmu_bonus_hold(rwa->os, drrw->drr_object, FTAG, &bonus) != 0) |
37f8a883 | 2297 | return (SET_ERROR(EINVAL)); |
88904bb3 | 2298 | dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx); |
47dfff3b MA |
2299 | |
2300 | /* | |
2301 | * Note: If the receive fails, we want the resume stream to start | |
2302 | * with the same record that we last successfully received (as opposed | |
2303 | * to the next record), so that we can verify that we are | |
2304 | * resuming from the correct location. | |
2305 | */ | |
2306 | save_resume_state(rwa, drrw->drr_object, drrw->drr_offset, tx); | |
34dc7c2f | 2307 | dmu_tx_commit(tx); |
88904bb3 | 2308 | dmu_buf_rele(bonus, FTAG); |
47dfff3b | 2309 | |
34dc7c2f BB |
2310 | return (0); |
2311 | } | |
2312 | ||
428870ff BB |
2313 | /* |
2314 | * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed | |
2315 | * streams to refer to a copy of the data that is already on the | |
2316 | * system because it came in earlier in the stream. This function | |
2317 | * finds the earlier copy of the data, and uses that copy instead of | |
2318 | * data from the stream to fulfill this write. | |
2319 | */ | |
2320 | static int | |
fcff0f35 PD |
2321 | receive_write_byref(struct receive_writer_arg *rwa, |
2322 | struct drr_write_byref *drrwbr) | |
428870ff BB |
2323 | { |
2324 | dmu_tx_t *tx; | |
2325 | int err; | |
2326 | guid_map_entry_t gmesrch; | |
2327 | guid_map_entry_t *gmep; | |
9b67f605 | 2328 | avl_index_t where; |
428870ff BB |
2329 | objset_t *ref_os = NULL; |
2330 | dmu_buf_t *dbp; | |
2331 | ||
2332 | if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) | |
2e528b49 | 2333 | return (SET_ERROR(EINVAL)); |
428870ff BB |
2334 | |
2335 | /* | |
2336 | * If the GUID of the referenced dataset is different from the | |
2337 | * GUID of the target dataset, find the referenced dataset. | |
2338 | */ | |
2339 | if (drrwbr->drr_toguid != drrwbr->drr_refguid) { | |
2340 | gmesrch.guid = drrwbr->drr_refguid; | |
fcff0f35 | 2341 | if ((gmep = avl_find(rwa->guid_to_ds_map, &gmesrch, |
428870ff | 2342 | &where)) == NULL) { |
2e528b49 | 2343 | return (SET_ERROR(EINVAL)); |
428870ff BB |
2344 | } |
2345 | if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) | |
2e528b49 | 2346 | return (SET_ERROR(EINVAL)); |
428870ff | 2347 | } else { |
fcff0f35 | 2348 | ref_os = rwa->os; |
428870ff BB |
2349 | } |
2350 | ||
c65aa5b2 BB |
2351 | err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, |
2352 | drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH); | |
9b67f605 | 2353 | if (err != 0) |
428870ff BB |
2354 | return (err); |
2355 | ||
fcff0f35 | 2356 | tx = dmu_tx_create(rwa->os); |
428870ff BB |
2357 | |
2358 | dmu_tx_hold_write(tx, drrwbr->drr_object, | |
2359 | drrwbr->drr_offset, drrwbr->drr_length); | |
2360 | err = dmu_tx_assign(tx, TXG_WAIT); | |
13fe0198 | 2361 | if (err != 0) { |
428870ff BB |
2362 | dmu_tx_abort(tx); |
2363 | return (err); | |
2364 | } | |
fcff0f35 | 2365 | dmu_write(rwa->os, drrwbr->drr_object, |
428870ff BB |
2366 | drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); |
2367 | dmu_buf_rele(dbp, FTAG); | |
47dfff3b MA |
2368 | |
2369 | /* See comment in restore_write. */ | |
2370 | save_resume_state(rwa, drrwbr->drr_object, drrwbr->drr_offset, tx); | |
428870ff BB |
2371 | dmu_tx_commit(tx); |
2372 | return (0); | |
2373 | } | |
2374 | ||
9b67f605 | 2375 | static int |
fcff0f35 | 2376 | receive_write_embedded(struct receive_writer_arg *rwa, |
47dfff3b | 2377 | struct drr_write_embedded *drrwe, void *data) |
9b67f605 MA |
2378 | { |
2379 | dmu_tx_t *tx; | |
2380 | int err; | |
9b67f605 | 2381 | |
47dfff3b | 2382 | if (drrwe->drr_offset + drrwe->drr_length < drrwe->drr_offset) |
ecb2b7dc | 2383 | return (SET_ERROR(EINVAL)); |
9b67f605 | 2384 | |
47dfff3b | 2385 | if (drrwe->drr_psize > BPE_PAYLOAD_SIZE) |
ecb2b7dc | 2386 | return (SET_ERROR(EINVAL)); |
9b67f605 | 2387 | |
47dfff3b | 2388 | if (drrwe->drr_etype >= NUM_BP_EMBEDDED_TYPES) |
ecb2b7dc | 2389 | return (SET_ERROR(EINVAL)); |
47dfff3b | 2390 | if (drrwe->drr_compression >= ZIO_COMPRESS_FUNCTIONS) |
ecb2b7dc | 2391 | return (SET_ERROR(EINVAL)); |
9b67f605 | 2392 | |
fcff0f35 | 2393 | tx = dmu_tx_create(rwa->os); |
9b67f605 | 2394 | |
47dfff3b MA |
2395 | dmu_tx_hold_write(tx, drrwe->drr_object, |
2396 | drrwe->drr_offset, drrwe->drr_length); | |
9b67f605 MA |
2397 | err = dmu_tx_assign(tx, TXG_WAIT); |
2398 | if (err != 0) { | |
2399 | dmu_tx_abort(tx); | |
2400 | return (err); | |
2401 | } | |
2402 | ||
47dfff3b MA |
2403 | dmu_write_embedded(rwa->os, drrwe->drr_object, |
2404 | drrwe->drr_offset, data, drrwe->drr_etype, | |
2405 | drrwe->drr_compression, drrwe->drr_lsize, drrwe->drr_psize, | |
fcff0f35 | 2406 | rwa->byteswap ^ ZFS_HOST_BYTEORDER, tx); |
9b67f605 | 2407 | |
47dfff3b MA |
2408 | /* See comment in restore_write. */ |
2409 | save_resume_state(rwa, drrwe->drr_object, drrwe->drr_offset, tx); | |
9b67f605 MA |
2410 | dmu_tx_commit(tx); |
2411 | return (0); | |
2412 | } | |
2413 | ||
428870ff | 2414 | static int |
fcff0f35 PD |
2415 | receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs, |
2416 | void *data) | |
428870ff BB |
2417 | { |
2418 | dmu_tx_t *tx; | |
428870ff BB |
2419 | dmu_buf_t *db, *db_spill; |
2420 | int err; | |
2421 | ||
2422 | if (drrs->drr_length < SPA_MINBLOCKSIZE || | |
fcff0f35 | 2423 | drrs->drr_length > spa_maxblocksize(dmu_objset_spa(rwa->os))) |
2e528b49 | 2424 | return (SET_ERROR(EINVAL)); |
428870ff | 2425 | |
fcff0f35 | 2426 | if (dmu_object_info(rwa->os, drrs->drr_object, NULL) != 0) |
2e528b49 | 2427 | return (SET_ERROR(EINVAL)); |
428870ff | 2428 | |
fcff0f35 | 2429 | VERIFY0(dmu_bonus_hold(rwa->os, drrs->drr_object, FTAG, &db)); |
428870ff BB |
2430 | if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) { |
2431 | dmu_buf_rele(db, FTAG); | |
2432 | return (err); | |
2433 | } | |
2434 | ||
fcff0f35 | 2435 | tx = dmu_tx_create(rwa->os); |
428870ff BB |
2436 | |
2437 | dmu_tx_hold_spill(tx, db->db_object); | |
2438 | ||
2439 | err = dmu_tx_assign(tx, TXG_WAIT); | |
13fe0198 | 2440 | if (err != 0) { |
428870ff BB |
2441 | dmu_buf_rele(db, FTAG); |
2442 | dmu_buf_rele(db_spill, FTAG); | |
2443 | dmu_tx_abort(tx); | |
2444 | return (err); | |
2445 | } | |
2446 | dmu_buf_will_dirty(db_spill, tx); | |
2447 | ||
2448 | if (db_spill->db_size < drrs->drr_length) | |
2449 | VERIFY(0 == dbuf_spill_set_blksz(db_spill, | |
2450 | drrs->drr_length, tx)); | |
2451 | bcopy(data, db_spill->db_data, drrs->drr_length); | |
2452 | ||
2453 | dmu_buf_rele(db, FTAG); | |
2454 | dmu_buf_rele(db_spill, FTAG); | |
2455 | ||
2456 | dmu_tx_commit(tx); | |
2457 | return (0); | |
2458 | } | |
2459 | ||
34dc7c2f | 2460 | /* ARGSUSED */ |
60948de1 | 2461 | noinline static int |
fcff0f35 | 2462 | receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf) |
34dc7c2f | 2463 | { |
34dc7c2f BB |
2464 | int err; |
2465 | ||
2466 | if (drrf->drr_length != -1ULL && | |
2467 | drrf->drr_offset + drrf->drr_length < drrf->drr_offset) | |
2e528b49 | 2468 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 2469 | |
fcff0f35 | 2470 | if (dmu_object_info(rwa->os, drrf->drr_object, NULL) != 0) |
2e528b49 | 2471 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 2472 | |
fcff0f35 | 2473 | err = dmu_free_long_range(rwa->os, drrf->drr_object, |
34dc7c2f | 2474 | drrf->drr_offset, drrf->drr_length); |
fcff0f35 | 2475 | |
34dc7c2f BB |
2476 | return (err); |
2477 | } | |
2478 | ||
13fe0198 MA |
2479 | /* used to destroy the drc_ds on error */ |
2480 | static void | |
2481 | dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) | |
2482 | { | |
47dfff3b MA |
2483 | if (drc->drc_resumable) { |
2484 | /* wait for our resume state to be written to disk */ | |
2485 | txg_wait_synced(drc->drc_ds->ds_dir->dd_pool, 0); | |
2486 | dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); | |
2487 | } else { | |
eca7b760 | 2488 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
47dfff3b MA |
2489 | dsl_dataset_name(drc->drc_ds, name); |
2490 | dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); | |
2491 | (void) dsl_destroy_head(name); | |
2492 | } | |
13fe0198 MA |
2493 | } |
2494 | ||
37f8a883 | 2495 | static void |
fcff0f35 | 2496 | receive_cksum(struct receive_arg *ra, int len, void *buf) |
37f8a883 MA |
2497 | { |
2498 | if (ra->byteswap) { | |
a6255b7f | 2499 | (void) fletcher_4_incremental_byteswap(buf, len, &ra->cksum); |
37f8a883 | 2500 | } else { |
a6255b7f | 2501 | (void) fletcher_4_incremental_native(buf, len, &ra->cksum); |
37f8a883 MA |
2502 | } |
2503 | } | |
2504 | ||
2505 | /* | |
fcff0f35 PD |
2506 | * Read the payload into a buffer of size len, and update the current record's |
2507 | * payload field. | |
2508 | * Allocate ra->next_rrd and read the next record's header into | |
2509 | * ra->next_rrd->header. | |
37f8a883 MA |
2510 | * Verify checksum of payload and next record. |
2511 | */ | |
2512 | static int | |
fcff0f35 | 2513 | receive_read_payload_and_next_header(struct receive_arg *ra, int len, void *buf) |
37f8a883 MA |
2514 | { |
2515 | int err; | |
2516 | zio_cksum_t cksum_orig; | |
2517 | zio_cksum_t *cksump; | |
2518 | ||
2519 | if (len != 0) { | |
fcff0f35 | 2520 | ASSERT3U(len, <=, SPA_MAXBLOCKSIZE); |
47dfff3b | 2521 | err = receive_read(ra, len, buf); |
37f8a883 MA |
2522 | if (err != 0) |
2523 | return (err); | |
47dfff3b MA |
2524 | receive_cksum(ra, len, buf); |
2525 | ||
2526 | /* note: rrd is NULL when reading the begin record's payload */ | |
2527 | if (ra->rrd != NULL) { | |
2528 | ra->rrd->payload = buf; | |
2529 | ra->rrd->payload_size = len; | |
2530 | ra->rrd->bytes_read = ra->bytes_read; | |
2531 | } | |
37f8a883 MA |
2532 | } |
2533 | ||
2534 | ra->prev_cksum = ra->cksum; | |
2535 | ||
fcff0f35 PD |
2536 | ra->next_rrd = kmem_zalloc(sizeof (*ra->next_rrd), KM_SLEEP); |
2537 | err = receive_read(ra, sizeof (ra->next_rrd->header), | |
2538 | &ra->next_rrd->header); | |
47dfff3b | 2539 | ra->next_rrd->bytes_read = ra->bytes_read; |
fcff0f35 PD |
2540 | if (err != 0) { |
2541 | kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); | |
2542 | ra->next_rrd = NULL; | |
37f8a883 | 2543 | return (err); |
fcff0f35 PD |
2544 | } |
2545 | if (ra->next_rrd->header.drr_type == DRR_BEGIN) { | |
2546 | kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); | |
2547 | ra->next_rrd = NULL; | |
37f8a883 | 2548 | return (SET_ERROR(EINVAL)); |
fcff0f35 | 2549 | } |
37f8a883 MA |
2550 | |
2551 | /* | |
2552 | * Note: checksum is of everything up to but not including the | |
2553 | * checksum itself. | |
2554 | */ | |
2555 | ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), | |
2556 | ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); | |
fcff0f35 | 2557 | receive_cksum(ra, |
37f8a883 | 2558 | offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), |
fcff0f35 | 2559 | &ra->next_rrd->header); |
37f8a883 | 2560 | |
fcff0f35 PD |
2561 | cksum_orig = ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; |
2562 | cksump = &ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; | |
37f8a883 MA |
2563 | |
2564 | if (ra->byteswap) | |
fcff0f35 | 2565 | byteswap_record(&ra->next_rrd->header); |
37f8a883 MA |
2566 | |
2567 | if ((!ZIO_CHECKSUM_IS_ZERO(cksump)) && | |
fcff0f35 PD |
2568 | !ZIO_CHECKSUM_EQUAL(ra->cksum, *cksump)) { |
2569 | kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); | |
2570 | ra->next_rrd = NULL; | |
37f8a883 | 2571 | return (SET_ERROR(ECKSUM)); |
fcff0f35 | 2572 | } |
37f8a883 | 2573 | |
fcff0f35 | 2574 | receive_cksum(ra, sizeof (cksum_orig), &cksum_orig); |
37f8a883 MA |
2575 | |
2576 | return (0); | |
2577 | } | |
2578 | ||
e6d3a843 PD |
2579 | static void |
2580 | objlist_create(struct objlist *list) | |
2581 | { | |
2582 | list_create(&list->list, sizeof (struct receive_objnode), | |
2583 | offsetof(struct receive_objnode, node)); | |
2584 | list->last_lookup = 0; | |
2585 | } | |
2586 | ||
2587 | static void | |
2588 | objlist_destroy(struct objlist *list) | |
2589 | { | |
2590 | struct receive_objnode *n; | |
2591 | ||
2592 | for (n = list_remove_head(&list->list); | |
2593 | n != NULL; n = list_remove_head(&list->list)) { | |
2594 | kmem_free(n, sizeof (*n)); | |
2595 | } | |
2596 | list_destroy(&list->list); | |
2597 | } | |
2598 | ||
2599 | /* | |
2600 | * This function looks through the objlist to see if the specified object number | |
2601 | * is contained in the objlist. In the process, it will remove all object | |
2602 | * numbers in the list that are smaller than the specified object number. Thus, | |
2603 | * any lookup of an object number smaller than a previously looked up object | |
2604 | * number will always return false; therefore, all lookups should be done in | |
2605 | * ascending order. | |
2606 | */ | |
2607 | static boolean_t | |
2608 | objlist_exists(struct objlist *list, uint64_t object) | |
2609 | { | |
2610 | struct receive_objnode *node = list_head(&list->list); | |
2611 | ASSERT3U(object, >=, list->last_lookup); | |
2612 | list->last_lookup = object; | |
2613 | while (node != NULL && node->object < object) { | |
2614 | VERIFY3P(node, ==, list_remove_head(&list->list)); | |
2615 | kmem_free(node, sizeof (*node)); | |
2616 | node = list_head(&list->list); | |
2617 | } | |
2618 | return (node != NULL && node->object == object); | |
2619 | } | |
2620 | ||
2621 | /* | |
2622 | * The objlist is a list of object numbers stored in ascending order. However, | |
2623 | * the insertion of new object numbers does not seek out the correct location to | |
2624 | * store a new object number; instead, it appends it to the list for simplicity. | |
2625 | * Thus, any users must take care to only insert new object numbers in ascending | |
2626 | * order. | |
2627 | */ | |
2628 | static void | |
2629 | objlist_insert(struct objlist *list, uint64_t object) | |
2630 | { | |
2631 | struct receive_objnode *node = kmem_zalloc(sizeof (*node), KM_SLEEP); | |
2632 | node->object = object; | |
2633 | #ifdef ZFS_DEBUG | |
2634 | { | |
2635 | struct receive_objnode *last_object = list_tail(&list->list); | |
2636 | uint64_t last_objnum = (last_object != NULL ? last_object->object : 0); | |
2637 | ASSERT3U(node->object, >, last_objnum); | |
2638 | } | |
2639 | #endif | |
2640 | list_insert_tail(&list->list, node); | |
2641 | } | |
2642 | ||
fcff0f35 PD |
2643 | /* |
2644 | * Issue the prefetch reads for any necessary indirect blocks. | |
2645 | * | |
2646 | * We use the object ignore list to tell us whether or not to issue prefetches | |
2647 | * for a given object. We do this for both correctness (in case the blocksize | |
2648 | * of an object has changed) and performance (if the object doesn't exist, don't | |
2649 | * needlessly try to issue prefetches). We also trim the list as we go through | |
2650 | * the stream to prevent it from growing to an unbounded size. | |
2651 | * | |
2652 | * The object numbers within will always be in sorted order, and any write | |
2653 | * records we see will also be in sorted order, but they're not sorted with | |
2654 | * respect to each other (i.e. we can get several object records before | |
2655 | * receiving each object's write records). As a result, once we've reached a | |
2656 | * given object number, we can safely remove any reference to lower object | |
2657 | * numbers in the ignore list. In practice, we receive up to 32 object records | |
2658 | * before receiving write records, so the list can have up to 32 nodes in it. | |
2659 | */ | |
2660 | /* ARGSUSED */ | |
2661 | static void | |
2662 | receive_read_prefetch(struct receive_arg *ra, | |
2663 | uint64_t object, uint64_t offset, uint64_t length) | |
2664 | { | |
e6d3a843 | 2665 | if (!objlist_exists(&ra->ignore_objlist, object)) { |
fcff0f35 PD |
2666 | dmu_prefetch(ra->os, object, 1, offset, length, |
2667 | ZIO_PRIORITY_SYNC_READ); | |
2668 | } | |
2669 | } | |
2670 | ||
2671 | /* | |
2672 | * Read records off the stream, issuing any necessary prefetches. | |
2673 | */ | |
37f8a883 | 2674 | static int |
fcff0f35 | 2675 | receive_read_record(struct receive_arg *ra) |
37f8a883 MA |
2676 | { |
2677 | int err; | |
2678 | ||
fcff0f35 | 2679 | switch (ra->rrd->header.drr_type) { |
37f8a883 MA |
2680 | case DRR_OBJECT: |
2681 | { | |
fcff0f35 PD |
2682 | struct drr_object *drro = &ra->rrd->header.drr_u.drr_object; |
2683 | uint32_t size = P2ROUNDUP(drro->drr_bonuslen, 8); | |
2684 | void *buf = kmem_zalloc(size, KM_SLEEP); | |
2685 | dmu_object_info_t doi; | |
2686 | err = receive_read_payload_and_next_header(ra, size, buf); | |
2687 | if (err != 0) { | |
2688 | kmem_free(buf, size); | |
37f8a883 | 2689 | return (err); |
fcff0f35 PD |
2690 | } |
2691 | err = dmu_object_info(ra->os, drro->drr_object, &doi); | |
2692 | /* | |
2693 | * See receive_read_prefetch for an explanation why we're | |
2694 | * storing this object in the ignore_obj_list. | |
2695 | */ | |
2696 | if (err == ENOENT || | |
2697 | (err == 0 && doi.doi_data_block_size != drro->drr_blksz)) { | |
e6d3a843 | 2698 | objlist_insert(&ra->ignore_objlist, drro->drr_object); |
fcff0f35 PD |
2699 | err = 0; |
2700 | } | |
2701 | return (err); | |
37f8a883 MA |
2702 | } |
2703 | case DRR_FREEOBJECTS: | |
2704 | { | |
fcff0f35 PD |
2705 | err = receive_read_payload_and_next_header(ra, 0, NULL); |
2706 | return (err); | |
37f8a883 MA |
2707 | } |
2708 | case DRR_WRITE: | |
2709 | { | |
fcff0f35 | 2710 | struct drr_write *drrw = &ra->rrd->header.drr_u.drr_write; |
2aa34383 DK |
2711 | arc_buf_t *abuf; |
2712 | boolean_t is_meta = DMU_OT_IS_METADATA(drrw->drr_type); | |
2713 | if (DRR_WRITE_COMPRESSED(drrw)) { | |
2714 | ASSERT3U(drrw->drr_compressed_size, >, 0); | |
2715 | ASSERT3U(drrw->drr_logical_size, >=, | |
2716 | drrw->drr_compressed_size); | |
2717 | ASSERT(!is_meta); | |
2718 | abuf = arc_loan_compressed_buf( | |
2719 | dmu_objset_spa(ra->os), | |
2720 | drrw->drr_compressed_size, drrw->drr_logical_size, | |
2721 | drrw->drr_compressiontype); | |
2722 | } else { | |
2723 | abuf = arc_loan_buf(dmu_objset_spa(ra->os), | |
2724 | is_meta, drrw->drr_logical_size); | |
2725 | } | |
37f8a883 | 2726 | |
fcff0f35 | 2727 | err = receive_read_payload_and_next_header(ra, |
2aa34383 | 2728 | DRR_WRITE_PAYLOAD_SIZE(drrw), abuf->b_data); |
fcff0f35 | 2729 | if (err != 0) { |
37f8a883 | 2730 | dmu_return_arcbuf(abuf); |
fcff0f35 PD |
2731 | return (err); |
2732 | } | |
2733 | ra->rrd->write_buf = abuf; | |
2734 | receive_read_prefetch(ra, drrw->drr_object, drrw->drr_offset, | |
2aa34383 | 2735 | drrw->drr_logical_size); |
37f8a883 MA |
2736 | return (err); |
2737 | } | |
2738 | case DRR_WRITE_BYREF: | |
2739 | { | |
fcff0f35 PD |
2740 | struct drr_write_byref *drrwb = |
2741 | &ra->rrd->header.drr_u.drr_write_byref; | |
2742 | err = receive_read_payload_and_next_header(ra, 0, NULL); | |
2743 | receive_read_prefetch(ra, drrwb->drr_object, drrwb->drr_offset, | |
2744 | drrwb->drr_length); | |
2745 | return (err); | |
37f8a883 MA |
2746 | } |
2747 | case DRR_WRITE_EMBEDDED: | |
2748 | { | |
2749 | struct drr_write_embedded *drrwe = | |
fcff0f35 PD |
2750 | &ra->rrd->header.drr_u.drr_write_embedded; |
2751 | uint32_t size = P2ROUNDUP(drrwe->drr_psize, 8); | |
2752 | void *buf = kmem_zalloc(size, KM_SLEEP); | |
2753 | ||
2754 | err = receive_read_payload_and_next_header(ra, size, buf); | |
2755 | if (err != 0) { | |
2756 | kmem_free(buf, size); | |
37f8a883 | 2757 | return (err); |
fcff0f35 PD |
2758 | } |
2759 | ||
2760 | receive_read_prefetch(ra, drrwe->drr_object, drrwe->drr_offset, | |
2761 | drrwe->drr_length); | |
2762 | return (err); | |
37f8a883 MA |
2763 | } |
2764 | case DRR_FREE: | |
2765 | { | |
fcff0f35 PD |
2766 | /* |
2767 | * It might be beneficial to prefetch indirect blocks here, but | |
2768 | * we don't really have the data to decide for sure. | |
2769 | */ | |
2770 | err = receive_read_payload_and_next_header(ra, 0, NULL); | |
2771 | return (err); | |
37f8a883 MA |
2772 | } |
2773 | case DRR_END: | |
2774 | { | |
fcff0f35 | 2775 | struct drr_end *drre = &ra->rrd->header.drr_u.drr_end; |
37f8a883 | 2776 | if (!ZIO_CHECKSUM_EQUAL(ra->prev_cksum, drre->drr_checksum)) |
47dfff3b | 2777 | return (SET_ERROR(ECKSUM)); |
37f8a883 MA |
2778 | return (0); |
2779 | } | |
2780 | case DRR_SPILL: | |
2781 | { | |
fcff0f35 PD |
2782 | struct drr_spill *drrs = &ra->rrd->header.drr_u.drr_spill; |
2783 | void *buf = kmem_zalloc(drrs->drr_length, KM_SLEEP); | |
2784 | err = receive_read_payload_and_next_header(ra, drrs->drr_length, | |
2785 | buf); | |
37f8a883 | 2786 | if (err != 0) |
fcff0f35 PD |
2787 | kmem_free(buf, drrs->drr_length); |
2788 | return (err); | |
2789 | } | |
2790 | default: | |
2791 | return (SET_ERROR(EINVAL)); | |
2792 | } | |
2793 | } | |
2794 | ||
2795 | /* | |
2796 | * Commit the records to the pool. | |
2797 | */ | |
2798 | static int | |
2799 | receive_process_record(struct receive_writer_arg *rwa, | |
2800 | struct receive_record_arg *rrd) | |
2801 | { | |
2802 | int err; | |
2803 | ||
47dfff3b MA |
2804 | /* Processing in order, therefore bytes_read should be increasing. */ |
2805 | ASSERT3U(rrd->bytes_read, >=, rwa->bytes_read); | |
2806 | rwa->bytes_read = rrd->bytes_read; | |
2807 | ||
fcff0f35 PD |
2808 | switch (rrd->header.drr_type) { |
2809 | case DRR_OBJECT: | |
2810 | { | |
2811 | struct drr_object *drro = &rrd->header.drr_u.drr_object; | |
2812 | err = receive_object(rwa, drro, rrd->payload); | |
2813 | kmem_free(rrd->payload, rrd->payload_size); | |
2814 | rrd->payload = NULL; | |
2815 | return (err); | |
2816 | } | |
2817 | case DRR_FREEOBJECTS: | |
2818 | { | |
2819 | struct drr_freeobjects *drrfo = | |
2820 | &rrd->header.drr_u.drr_freeobjects; | |
2821 | return (receive_freeobjects(rwa, drrfo)); | |
2822 | } | |
2823 | case DRR_WRITE: | |
2824 | { | |
2825 | struct drr_write *drrw = &rrd->header.drr_u.drr_write; | |
2826 | err = receive_write(rwa, drrw, rrd->write_buf); | |
2827 | /* if receive_write() is successful, it consumes the arc_buf */ | |
2828 | if (err != 0) | |
2829 | dmu_return_arcbuf(rrd->write_buf); | |
2830 | rrd->write_buf = NULL; | |
2831 | rrd->payload = NULL; | |
2832 | return (err); | |
2833 | } | |
2834 | case DRR_WRITE_BYREF: | |
2835 | { | |
2836 | struct drr_write_byref *drrwbr = | |
2837 | &rrd->header.drr_u.drr_write_byref; | |
2838 | return (receive_write_byref(rwa, drrwbr)); | |
2839 | } | |
2840 | case DRR_WRITE_EMBEDDED: | |
2841 | { | |
2842 | struct drr_write_embedded *drrwe = | |
2843 | &rrd->header.drr_u.drr_write_embedded; | |
2844 | err = receive_write_embedded(rwa, drrwe, rrd->payload); | |
2845 | kmem_free(rrd->payload, rrd->payload_size); | |
2846 | rrd->payload = NULL; | |
2847 | return (err); | |
2848 | } | |
2849 | case DRR_FREE: | |
2850 | { | |
2851 | struct drr_free *drrf = &rrd->header.drr_u.drr_free; | |
2852 | return (receive_free(rwa, drrf)); | |
2853 | } | |
2854 | case DRR_SPILL: | |
2855 | { | |
2856 | struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; | |
2857 | err = receive_spill(rwa, drrs, rrd->payload); | |
2858 | kmem_free(rrd->payload, rrd->payload_size); | |
2859 | rrd->payload = NULL; | |
2860 | return (err); | |
37f8a883 MA |
2861 | } |
2862 | default: | |
2863 | return (SET_ERROR(EINVAL)); | |
2864 | } | |
2865 | } | |
2866 | ||
34dc7c2f | 2867 | /* |
fcff0f35 PD |
2868 | * dmu_recv_stream's worker thread; pull records off the queue, and then call |
2869 | * receive_process_record When we're done, signal the main thread and exit. | |
2870 | */ | |
2871 | static void | |
2872 | receive_writer_thread(void *arg) | |
2873 | { | |
2874 | struct receive_writer_arg *rwa = arg; | |
2875 | struct receive_record_arg *rrd; | |
3e635ac1 TC |
2876 | fstrans_cookie_t cookie = spl_fstrans_mark(); |
2877 | ||
fcff0f35 PD |
2878 | for (rrd = bqueue_dequeue(&rwa->q); !rrd->eos_marker; |
2879 | rrd = bqueue_dequeue(&rwa->q)) { | |
2880 | /* | |
2881 | * If there's an error, the main thread will stop putting things | |
2882 | * on the queue, but we need to clear everything in it before we | |
2883 | * can exit. | |
2884 | */ | |
2885 | if (rwa->err == 0) { | |
2886 | rwa->err = receive_process_record(rwa, rrd); | |
2887 | } else if (rrd->write_buf != NULL) { | |
2888 | dmu_return_arcbuf(rrd->write_buf); | |
2889 | rrd->write_buf = NULL; | |
2890 | rrd->payload = NULL; | |
2891 | } else if (rrd->payload != NULL) { | |
2892 | kmem_free(rrd->payload, rrd->payload_size); | |
2893 | rrd->payload = NULL; | |
2894 | } | |
2895 | kmem_free(rrd, sizeof (*rrd)); | |
2896 | } | |
2897 | kmem_free(rrd, sizeof (*rrd)); | |
2898 | mutex_enter(&rwa->mutex); | |
2899 | rwa->done = B_TRUE; | |
2900 | cv_signal(&rwa->cv); | |
2901 | mutex_exit(&rwa->mutex); | |
3e635ac1 | 2902 | spl_fstrans_unmark(cookie); |
34a6b428 | 2903 | thread_exit(); |
fcff0f35 PD |
2904 | } |
2905 | ||
47dfff3b MA |
2906 | static int |
2907 | resume_check(struct receive_arg *ra, nvlist_t *begin_nvl) | |
2908 | { | |
2909 | uint64_t val; | |
2910 | objset_t *mos = dmu_objset_pool(ra->os)->dp_meta_objset; | |
2911 | uint64_t dsobj = dmu_objset_id(ra->os); | |
2912 | uint64_t resume_obj, resume_off; | |
2913 | ||
2914 | if (nvlist_lookup_uint64(begin_nvl, | |
2915 | "resume_object", &resume_obj) != 0 || | |
2916 | nvlist_lookup_uint64(begin_nvl, | |
2917 | "resume_offset", &resume_off) != 0) { | |
2918 | return (SET_ERROR(EINVAL)); | |
2919 | } | |
2920 | VERIFY0(zap_lookup(mos, dsobj, | |
2921 | DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val)); | |
2922 | if (resume_obj != val) | |
2923 | return (SET_ERROR(EINVAL)); | |
2924 | VERIFY0(zap_lookup(mos, dsobj, | |
2925 | DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val)); | |
2926 | if (resume_off != val) | |
2927 | return (SET_ERROR(EINVAL)); | |
2928 | ||
2929 | return (0); | |
2930 | } | |
2931 | ||
fcff0f35 PD |
2932 | /* |
2933 | * Read in the stream's records, one by one, and apply them to the pool. There | |
2934 | * are two threads involved; the thread that calls this function will spin up a | |
2935 | * worker thread, read the records off the stream one by one, and issue | |
2936 | * prefetches for any necessary indirect blocks. It will then push the records | |
2937 | * onto an internal blocking queue. The worker thread will pull the records off | |
2938 | * the queue, and actually write the data into the DMU. This way, the worker | |
2939 | * thread doesn't have to wait for reads to complete, since everything it needs | |
2940 | * (the indirect blocks) will be prefetched. | |
2941 | * | |
34dc7c2f BB |
2942 | * NB: callers *must* call dmu_recv_end() if this succeeds. |
2943 | */ | |
2944 | int | |
572e2857 BB |
2945 | dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, |
2946 | int cleanup_fd, uint64_t *action_handlep) | |
34dc7c2f | 2947 | { |
37f8a883 | 2948 | int err = 0; |
04bc4610 NB |
2949 | struct receive_arg *ra; |
2950 | struct receive_writer_arg *rwa; | |
428870ff | 2951 | int featureflags; |
47dfff3b MA |
2952 | uint32_t payloadlen; |
2953 | void *payload; | |
2954 | nvlist_t *begin_nvl = NULL; | |
34dc7c2f | 2955 | |
04bc4610 NB |
2956 | ra = kmem_zalloc(sizeof (*ra), KM_SLEEP); |
2957 | rwa = kmem_zalloc(sizeof (*rwa), KM_SLEEP); | |
2958 | ||
2959 | ra->byteswap = drc->drc_byteswap; | |
2960 | ra->cksum = drc->drc_cksum; | |
2961 | ra->vp = vp; | |
2962 | ra->voff = *voffp; | |
47dfff3b MA |
2963 | |
2964 | if (dsl_dataset_is_zapified(drc->drc_ds)) { | |
2965 | (void) zap_lookup(drc->drc_ds->ds_dir->dd_pool->dp_meta_objset, | |
2966 | drc->drc_ds->ds_object, DS_FIELD_RESUME_BYTES, | |
2967 | sizeof (ra->bytes_read), 1, &ra->bytes_read); | |
2968 | } | |
2969 | ||
e6d3a843 | 2970 | objlist_create(&ra->ignore_objlist); |
34dc7c2f BB |
2971 | |
2972 | /* these were verified in dmu_recv_begin */ | |
13fe0198 | 2973 | ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, |
428870ff | 2974 | DMU_SUBSTREAM); |
13fe0198 | 2975 | ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); |
34dc7c2f BB |
2976 | |
2977 | /* | |
2978 | * Open the objset we are modifying. | |
2979 | */ | |
04bc4610 | 2980 | VERIFY0(dmu_objset_from_ds(drc->drc_ds, &ra->os)); |
34dc7c2f | 2981 | |
d683ddbb | 2982 | ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); |
34dc7c2f | 2983 | |
428870ff BB |
2984 | featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); |
2985 | ||
2986 | /* if this stream is dedup'ed, set up the avl tree for guid mapping */ | |
2987 | if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { | |
572e2857 BB |
2988 | minor_t minor; |
2989 | ||
2990 | if (cleanup_fd == -1) { | |
04bc4610 | 2991 | ra->err = SET_ERROR(EBADF); |
572e2857 BB |
2992 | goto out; |
2993 | } | |
04bc4610 NB |
2994 | ra->err = zfs_onexit_fd_hold(cleanup_fd, &minor); |
2995 | if (ra->err != 0) { | |
572e2857 BB |
2996 | cleanup_fd = -1; |
2997 | goto out; | |
2998 | } | |
2999 | ||
3000 | if (*action_handlep == 0) { | |
04bc4610 | 3001 | rwa->guid_to_ds_map = |
572e2857 | 3002 | kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); |
04bc4610 | 3003 | avl_create(rwa->guid_to_ds_map, guid_compare, |
572e2857 BB |
3004 | sizeof (guid_map_entry_t), |
3005 | offsetof(guid_map_entry_t, avlnode)); | |
37f8a883 | 3006 | err = zfs_onexit_add_cb(minor, |
04bc4610 | 3007 | free_guid_map_onexit, rwa->guid_to_ds_map, |
572e2857 | 3008 | action_handlep); |
04bc4610 | 3009 | if (ra->err != 0) |
572e2857 BB |
3010 | goto out; |
3011 | } else { | |
37f8a883 | 3012 | err = zfs_onexit_cb_data(minor, *action_handlep, |
04bc4610 NB |
3013 | (void **)&rwa->guid_to_ds_map); |
3014 | if (ra->err != 0) | |
572e2857 BB |
3015 | goto out; |
3016 | } | |
8d35c149 | 3017 | |
04bc4610 | 3018 | drc->drc_guid_to_ds_map = rwa->guid_to_ds_map; |
428870ff BB |
3019 | } |
3020 | ||
47dfff3b MA |
3021 | payloadlen = drc->drc_drr_begin->drr_payloadlen; |
3022 | payload = NULL; | |
3023 | if (payloadlen != 0) | |
3024 | payload = kmem_alloc(payloadlen, KM_SLEEP); | |
3025 | ||
3026 | err = receive_read_payload_and_next_header(ra, payloadlen, payload); | |
3027 | if (err != 0) { | |
3028 | if (payloadlen != 0) | |
3029 | kmem_free(payload, payloadlen); | |
37f8a883 | 3030 | goto out; |
47dfff3b MA |
3031 | } |
3032 | if (payloadlen != 0) { | |
3033 | err = nvlist_unpack(payload, payloadlen, &begin_nvl, KM_SLEEP); | |
3034 | kmem_free(payload, payloadlen); | |
3035 | if (err != 0) | |
3036 | goto out; | |
3037 | } | |
3038 | ||
3039 | if (featureflags & DMU_BACKUP_FEATURE_RESUMING) { | |
3040 | err = resume_check(ra, begin_nvl); | |
3041 | if (err != 0) | |
3042 | goto out; | |
3043 | } | |
37f8a883 | 3044 | |
04bc4610 | 3045 | (void) bqueue_init(&rwa->q, zfs_recv_queue_length, |
fcff0f35 | 3046 | offsetof(struct receive_record_arg, node)); |
04bc4610 NB |
3047 | cv_init(&rwa->cv, NULL, CV_DEFAULT, NULL); |
3048 | mutex_init(&rwa->mutex, NULL, MUTEX_DEFAULT, NULL); | |
3049 | rwa->os = ra->os; | |
3050 | rwa->byteswap = drc->drc_byteswap; | |
47dfff3b | 3051 | rwa->resumable = drc->drc_resumable; |
fcff0f35 | 3052 | |
04bc4610 | 3053 | (void) thread_create(NULL, 0, receive_writer_thread, rwa, 0, curproc, |
fcff0f35 PD |
3054 | TS_RUN, minclsyspri); |
3055 | /* | |
04bc4610 | 3056 | * We're reading rwa->err without locks, which is safe since we are the |
fcff0f35 PD |
3057 | * only reader, and the worker thread is the only writer. It's ok if we |
3058 | * miss a write for an iteration or two of the loop, since the writer | |
3059 | * thread will keep freeing records we send it until we send it an eos | |
3060 | * marker. | |
3061 | * | |
04bc4610 | 3062 | * We can leave this loop in 3 ways: First, if rwa->err is |
fcff0f35 PD |
3063 | * non-zero. In that case, the writer thread will free the rrd we just |
3064 | * pushed. Second, if we're interrupted; in that case, either it's the | |
47dfff3b | 3065 | * first loop and ra->rrd was never allocated, or it's later and ra->rrd |
fcff0f35 PD |
3066 | * has been handed off to the writer thread who will free it. Finally, |
3067 | * if receive_read_record fails or we're at the end of the stream, then | |
04bc4610 | 3068 | * we free ra->rrd and exit. |
fcff0f35 | 3069 | */ |
04bc4610 | 3070 | while (rwa->err == 0) { |
34dc7c2f | 3071 | if (issig(JUSTLOOKING) && issig(FORREAL)) { |
37f8a883 MA |
3072 | err = SET_ERROR(EINTR); |
3073 | break; | |
34dc7c2f BB |
3074 | } |
3075 | ||
04bc4610 NB |
3076 | ASSERT3P(ra->rrd, ==, NULL); |
3077 | ra->rrd = ra->next_rrd; | |
3078 | ra->next_rrd = NULL; | |
3079 | /* Allocates and loads header into ra->next_rrd */ | |
3080 | err = receive_read_record(ra); | |
34dc7c2f | 3081 | |
04bc4610 NB |
3082 | if (ra->rrd->header.drr_type == DRR_END || err != 0) { |
3083 | kmem_free(ra->rrd, sizeof (*ra->rrd)); | |
3084 | ra->rrd = NULL; | |
428870ff | 3085 | break; |
fcff0f35 PD |
3086 | } |
3087 | ||
04bc4610 NB |
3088 | bqueue_enqueue(&rwa->q, ra->rrd, |
3089 | sizeof (struct receive_record_arg) + ra->rrd->payload_size); | |
3090 | ra->rrd = NULL; | |
fcff0f35 | 3091 | } |
04bc4610 NB |
3092 | if (ra->next_rrd == NULL) |
3093 | ra->next_rrd = kmem_zalloc(sizeof (*ra->next_rrd), KM_SLEEP); | |
3094 | ra->next_rrd->eos_marker = B_TRUE; | |
3095 | bqueue_enqueue(&rwa->q, ra->next_rrd, 1); | |
fcff0f35 | 3096 | |
04bc4610 NB |
3097 | mutex_enter(&rwa->mutex); |
3098 | while (!rwa->done) { | |
3099 | cv_wait(&rwa->cv, &rwa->mutex); | |
34dc7c2f | 3100 | } |
04bc4610 | 3101 | mutex_exit(&rwa->mutex); |
fcff0f35 | 3102 | |
04bc4610 NB |
3103 | cv_destroy(&rwa->cv); |
3104 | mutex_destroy(&rwa->mutex); | |
3105 | bqueue_destroy(&rwa->q); | |
fcff0f35 | 3106 | if (err == 0) |
04bc4610 | 3107 | err = rwa->err; |
34dc7c2f BB |
3108 | |
3109 | out: | |
47dfff3b | 3110 | nvlist_free(begin_nvl); |
572e2857 BB |
3111 | if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) |
3112 | zfs_onexit_fd_rele(cleanup_fd); | |
3113 | ||
37f8a883 | 3114 | if (err != 0) { |
34dc7c2f | 3115 | /* |
47dfff3b MA |
3116 | * Clean up references. If receive is not resumable, |
3117 | * destroy what we created, so we don't leave it in | |
3118 | * the inconsistent state. | |
34dc7c2f | 3119 | */ |
13fe0198 | 3120 | dmu_recv_cleanup_ds(drc); |
34dc7c2f BB |
3121 | } |
3122 | ||
04bc4610 | 3123 | *voffp = ra->voff; |
e6d3a843 | 3124 | objlist_destroy(&ra->ignore_objlist); |
04bc4610 NB |
3125 | kmem_free(ra, sizeof (*ra)); |
3126 | kmem_free(rwa, sizeof (*rwa)); | |
37f8a883 | 3127 | return (err); |
34dc7c2f BB |
3128 | } |
3129 | ||
34dc7c2f | 3130 | static int |
13fe0198 | 3131 | dmu_recv_end_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 3132 | { |
13fe0198 MA |
3133 | dmu_recv_cookie_t *drc = arg; |
3134 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
3135 | int error; | |
3136 | ||
3137 | ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); | |
3138 | ||
3139 | if (!drc->drc_newfs) { | |
3140 | dsl_dataset_t *origin_head; | |
3141 | ||
3142 | error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); | |
3143 | if (error != 0) | |
3144 | return (error); | |
19580676 MA |
3145 | if (drc->drc_force) { |
3146 | /* | |
3147 | * We will destroy any snapshots in tofs (i.e. before | |
3148 | * origin_head) that are after the origin (which is | |
3149 | * the snap before drc_ds, because drc_ds can not | |
3150 | * have any snaps of its own). | |
3151 | */ | |
d683ddbb JG |
3152 | uint64_t obj; |
3153 | ||
3154 | obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
3155 | while (obj != | |
3156 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { | |
19580676 MA |
3157 | dsl_dataset_t *snap; |
3158 | error = dsl_dataset_hold_obj(dp, obj, FTAG, | |
3159 | &snap); | |
3160 | if (error != 0) | |
b6640117 | 3161 | break; |
19580676 MA |
3162 | if (snap->ds_dir != origin_head->ds_dir) |
3163 | error = SET_ERROR(EINVAL); | |
3164 | if (error == 0) { | |
3165 | error = dsl_destroy_snapshot_check_impl( | |
3166 | snap, B_FALSE); | |
3167 | } | |
d683ddbb | 3168 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; |
19580676 MA |
3169 | dsl_dataset_rele(snap, FTAG); |
3170 | if (error != 0) | |
b6640117 AG |
3171 | break; |
3172 | } | |
3173 | if (error != 0) { | |
3174 | dsl_dataset_rele(origin_head, FTAG); | |
3175 | return (error); | |
19580676 MA |
3176 | } |
3177 | } | |
13fe0198 | 3178 | error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, |
831baf06 | 3179 | origin_head, drc->drc_force, drc->drc_owner, tx); |
13fe0198 MA |
3180 | if (error != 0) { |
3181 | dsl_dataset_rele(origin_head, FTAG); | |
3182 | return (error); | |
3183 | } | |
3184 | error = dsl_dataset_snapshot_check_impl(origin_head, | |
788eb90c | 3185 | drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); |
13fe0198 MA |
3186 | dsl_dataset_rele(origin_head, FTAG); |
3187 | if (error != 0) | |
3188 | return (error); | |
34dc7c2f | 3189 | |
13fe0198 MA |
3190 | error = dsl_destroy_head_check_impl(drc->drc_ds, 1); |
3191 | } else { | |
3192 | error = dsl_dataset_snapshot_check_impl(drc->drc_ds, | |
788eb90c | 3193 | drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); |
13fe0198 MA |
3194 | } |
3195 | return (error); | |
34dc7c2f BB |
3196 | } |
3197 | ||
3198 | static void | |
13fe0198 | 3199 | dmu_recv_end_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 3200 | { |
13fe0198 MA |
3201 | dmu_recv_cookie_t *drc = arg; |
3202 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
3203 | ||
3204 | spa_history_log_internal_ds(drc->drc_ds, "finish receiving", | |
3205 | tx, "snap=%s", drc->drc_tosnap); | |
3206 | ||
3207 | if (!drc->drc_newfs) { | |
3208 | dsl_dataset_t *origin_head; | |
3209 | ||
3210 | VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, | |
3211 | &origin_head)); | |
19580676 MA |
3212 | |
3213 | if (drc->drc_force) { | |
3214 | /* | |
3215 | * Destroy any snapshots of drc_tofs (origin_head) | |
3216 | * after the origin (the snap before drc_ds). | |
3217 | */ | |
d683ddbb JG |
3218 | uint64_t obj; |
3219 | ||
3220 | obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
3221 | while (obj != | |
3222 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { | |
19580676 MA |
3223 | dsl_dataset_t *snap; |
3224 | VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, | |
3225 | &snap)); | |
3226 | ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); | |
d683ddbb | 3227 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; |
19580676 MA |
3228 | dsl_destroy_snapshot_sync_impl(snap, |
3229 | B_FALSE, tx); | |
3230 | dsl_dataset_rele(snap, FTAG); | |
3231 | } | |
3232 | } | |
3233 | VERIFY3P(drc->drc_ds->ds_prev, ==, | |
3234 | origin_head->ds_prev); | |
3235 | ||
13fe0198 MA |
3236 | dsl_dataset_clone_swap_sync_impl(drc->drc_ds, |
3237 | origin_head, tx); | |
3238 | dsl_dataset_snapshot_sync_impl(origin_head, | |
3239 | drc->drc_tosnap, tx); | |
3240 | ||
3241 | /* set snapshot's creation time and guid */ | |
3242 | dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); | |
d683ddbb | 3243 | dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = |
13fe0198 | 3244 | drc->drc_drrb->drr_creation_time; |
d683ddbb | 3245 | dsl_dataset_phys(origin_head->ds_prev)->ds_guid = |
13fe0198 | 3246 | drc->drc_drrb->drr_toguid; |
d683ddbb | 3247 | dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= |
13fe0198 MA |
3248 | ~DS_FLAG_INCONSISTENT; |
3249 | ||
3250 | dmu_buf_will_dirty(origin_head->ds_dbuf, tx); | |
d683ddbb JG |
3251 | dsl_dataset_phys(origin_head)->ds_flags &= |
3252 | ~DS_FLAG_INCONSISTENT; | |
13fe0198 | 3253 | |
2e0e443a GM |
3254 | drc->drc_newsnapobj = |
3255 | dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
3256 | ||
13fe0198 MA |
3257 | dsl_dataset_rele(origin_head, FTAG); |
3258 | dsl_destroy_head_sync_impl(drc->drc_ds, tx); | |
831baf06 KW |
3259 | |
3260 | if (drc->drc_owner != NULL) | |
3261 | VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); | |
13fe0198 MA |
3262 | } else { |
3263 | dsl_dataset_t *ds = drc->drc_ds; | |
34dc7c2f | 3264 | |
13fe0198 | 3265 | dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); |
34dc7c2f | 3266 | |
13fe0198 MA |
3267 | /* set snapshot's creation time and guid */ |
3268 | dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
d683ddbb | 3269 | dsl_dataset_phys(ds->ds_prev)->ds_creation_time = |
13fe0198 | 3270 | drc->drc_drrb->drr_creation_time; |
d683ddbb JG |
3271 | dsl_dataset_phys(ds->ds_prev)->ds_guid = |
3272 | drc->drc_drrb->drr_toguid; | |
3273 | dsl_dataset_phys(ds->ds_prev)->ds_flags &= | |
3274 | ~DS_FLAG_INCONSISTENT; | |
34dc7c2f | 3275 | |
13fe0198 | 3276 | dmu_buf_will_dirty(ds->ds_dbuf, tx); |
d683ddbb | 3277 | dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; |
47dfff3b MA |
3278 | if (dsl_dataset_has_resume_receive_state(ds)) { |
3279 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3280 | DS_FIELD_RESUME_FROMGUID, tx); | |
3281 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3282 | DS_FIELD_RESUME_OBJECT, tx); | |
3283 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3284 | DS_FIELD_RESUME_OFFSET, tx); | |
3285 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3286 | DS_FIELD_RESUME_BYTES, tx); | |
3287 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3288 | DS_FIELD_RESUME_TOGUID, tx); | |
3289 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3290 | DS_FIELD_RESUME_TONAME, tx); | |
3291 | } | |
2e0e443a GM |
3292 | drc->drc_newsnapobj = |
3293 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; | |
13fe0198 | 3294 | } |
a0bd735a | 3295 | zvol_create_minors(dp->dp_spa, drc->drc_tofs, B_TRUE); |
13fe0198 MA |
3296 | /* |
3297 | * Release the hold from dmu_recv_begin. This must be done before | |
3298 | * we return to open context, so that when we free the dataset's dnode, | |
3299 | * we can evict its bonus buffer. | |
3300 | */ | |
3301 | dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); | |
3302 | drc->drc_ds = NULL; | |
34dc7c2f BB |
3303 | } |
3304 | ||
8d35c149 | 3305 | static int |
13fe0198 | 3306 | add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj) |
8d35c149 | 3307 | { |
13fe0198 | 3308 | dsl_pool_t *dp; |
8d35c149 AS |
3309 | dsl_dataset_t *snapds; |
3310 | guid_map_entry_t *gmep; | |
3311 | int err; | |
3312 | ||
3313 | ASSERT(guid_map != NULL); | |
3314 | ||
13fe0198 MA |
3315 | err = dsl_pool_hold(name, FTAG, &dp); |
3316 | if (err != 0) | |
3317 | return (err); | |
7ec09286 MA |
3318 | gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP); |
3319 | err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds); | |
8d35c149 | 3320 | if (err == 0) { |
d683ddbb | 3321 | gmep->guid = dsl_dataset_phys(snapds)->ds_guid; |
8d35c149 AS |
3322 | gmep->gme_ds = snapds; |
3323 | avl_add(guid_map, gmep); | |
13fe0198 | 3324 | dsl_dataset_long_hold(snapds, gmep); |
7ec09286 MA |
3325 | } else { |
3326 | kmem_free(gmep, sizeof (*gmep)); | |
8d35c149 AS |
3327 | } |
3328 | ||
13fe0198 | 3329 | dsl_pool_rele(dp, FTAG); |
8d35c149 AS |
3330 | return (err); |
3331 | } | |
3332 | ||
13fe0198 MA |
3333 | static int dmu_recv_end_modified_blocks = 3; |
3334 | ||
428870ff BB |
3335 | static int |
3336 | dmu_recv_existing_end(dmu_recv_cookie_t *drc) | |
34dc7c2f | 3337 | { |
13fe0198 | 3338 | #ifdef _KERNEL |
13fe0198 MA |
3339 | /* |
3340 | * We will be destroying the ds; make sure its origin is unmounted if | |
3341 | * necessary. | |
3342 | */ | |
eca7b760 | 3343 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
13fe0198 MA |
3344 | dsl_dataset_name(drc->drc_ds, name); |
3345 | zfs_destroy_unmount_origin(name); | |
13fe0198 | 3346 | #endif |
34dc7c2f | 3347 | |
2e0e443a | 3348 | return (dsl_sync_task(drc->drc_tofs, |
13fe0198 | 3349 | dmu_recv_end_check, dmu_recv_end_sync, drc, |
2e0e443a | 3350 | dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); |
34dc7c2f | 3351 | } |
428870ff BB |
3352 | |
3353 | static int | |
3354 | dmu_recv_new_end(dmu_recv_cookie_t *drc) | |
2e0e443a GM |
3355 | { |
3356 | return (dsl_sync_task(drc->drc_tofs, | |
3357 | dmu_recv_end_check, dmu_recv_end_sync, drc, | |
3358 | dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); | |
3359 | } | |
3360 | ||
3361 | int | |
3362 | dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) | |
428870ff | 3363 | { |
13fe0198 | 3364 | int error; |
428870ff | 3365 | |
2e0e443a GM |
3366 | drc->drc_owner = owner; |
3367 | ||
3368 | if (drc->drc_newfs) | |
3369 | error = dmu_recv_new_end(drc); | |
3370 | else | |
3371 | error = dmu_recv_existing_end(drc); | |
428870ff | 3372 | |
13fe0198 MA |
3373 | if (error != 0) { |
3374 | dmu_recv_cleanup_ds(drc); | |
3375 | } else if (drc->drc_guid_to_ds_map != NULL) { | |
3376 | (void) add_ds_to_guidmap(drc->drc_tofs, | |
3377 | drc->drc_guid_to_ds_map, | |
3378 | drc->drc_newsnapobj); | |
428870ff | 3379 | } |
13fe0198 | 3380 | return (error); |
428870ff BB |
3381 | } |
3382 | ||
ea97f8ce MA |
3383 | /* |
3384 | * Return TRUE if this objset is currently being received into. | |
3385 | */ | |
3386 | boolean_t | |
3387 | dmu_objset_is_receiving(objset_t *os) | |
3388 | { | |
3389 | return (os->os_dsl_dataset != NULL && | |
3390 | os->os_dsl_dataset->ds_owner == dmu_recv_tag); | |
3391 | } | |
fd8febbd TF |
3392 | |
3393 | #if defined(_KERNEL) | |
3394 | module_param(zfs_send_corrupt_data, int, 0644); | |
3395 | MODULE_PARM_DESC(zfs_send_corrupt_data, "Allow sending corrupt data"); | |
3396 | #endif |