]>
Commit | Line | Data |
---|---|---|
03916905 PD |
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 | |
1d3ba0bf | 9 | * or https://opensource.org/licenses/CDDL-1.0. |
03916905 PD |
10 | * See the License for the specific language governing permissions |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
23 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. | |
196bee4c | 24 | * Copyright (c) 2011, 2020 by Delphix. All rights reserved. |
03916905 PD |
25 | * Copyright (c) 2014, Joyent, Inc. All rights reserved. |
26 | * Copyright 2014 HybridCluster. All rights reserved. | |
d8d418ff | 27 | * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved. |
10b3c7f5 MN |
28 | * Copyright (c) 2019, Klara Inc. |
29 | * Copyright (c) 2019, Allan Jude | |
e8cf3a4f AP |
30 | * Copyright (c) 2019 Datto Inc. |
31 | * Copyright (c) 2022 Axcient. | |
03916905 PD |
32 | */ |
33 | ||
e8cf3a4f | 34 | #include <sys/spa_impl.h> |
03916905 PD |
35 | #include <sys/dmu.h> |
36 | #include <sys/dmu_impl.h> | |
30af21b0 PD |
37 | #include <sys/dmu_send.h> |
38 | #include <sys/dmu_recv.h> | |
03916905 PD |
39 | #include <sys/dmu_tx.h> |
40 | #include <sys/dbuf.h> | |
41 | #include <sys/dnode.h> | |
42 | #include <sys/zfs_context.h> | |
43 | #include <sys/dmu_objset.h> | |
44 | #include <sys/dmu_traverse.h> | |
45 | #include <sys/dsl_dataset.h> | |
46 | #include <sys/dsl_dir.h> | |
47 | #include <sys/dsl_prop.h> | |
48 | #include <sys/dsl_pool.h> | |
49 | #include <sys/dsl_synctask.h> | |
03916905 PD |
50 | #include <sys/zfs_ioctl.h> |
51 | #include <sys/zap.h> | |
30af21b0 | 52 | #include <sys/zvol.h> |
03916905 PD |
53 | #include <sys/zio_checksum.h> |
54 | #include <sys/zfs_znode.h> | |
55 | #include <zfs_fletcher.h> | |
56 | #include <sys/avl.h> | |
57 | #include <sys/ddt.h> | |
58 | #include <sys/zfs_onexit.h> | |
03916905 PD |
59 | #include <sys/dsl_destroy.h> |
60 | #include <sys/blkptr.h> | |
61 | #include <sys/dsl_bookmark.h> | |
62 | #include <sys/zfeature.h> | |
63 | #include <sys/bqueue.h> | |
30af21b0 PD |
64 | #include <sys/objlist.h> |
65 | #ifdef _KERNEL | |
66 | #include <sys/zfs_vfsops.h> | |
67 | #endif | |
da92d5cb | 68 | #include <sys/zfs_file.h> |
03916905 | 69 | |
fdc2d303 RY |
70 | static uint_t zfs_recv_queue_length = SPA_MAXBLOCKSIZE; |
71 | static uint_t zfs_recv_queue_ff = 20; | |
72 | static uint_t zfs_recv_write_batch_size = 1024 * 1024; | |
e8cf3a4f | 73 | static int zfs_recv_best_effort_corrective = 0; |
03916905 | 74 | |
a926aab9 | 75 | static const void *const dmu_recv_tag = "dmu_recv_tag"; |
18168da7 | 76 | const char *const recv_clone_name = "%recv"; |
03916905 | 77 | |
30af21b0 PD |
78 | static int receive_read_payload_and_next_header(dmu_recv_cookie_t *ra, int len, |
79 | void *buf); | |
80 | ||
81 | struct receive_record_arg { | |
82 | dmu_replay_record_t header; | |
83 | void *payload; /* Pointer to a buffer containing the payload */ | |
84 | /* | |
ba67d821 | 85 | * If the record is a WRITE or SPILL, pointer to the abd containing the |
30af21b0 PD |
86 | * payload. |
87 | */ | |
ba67d821 | 88 | abd_t *abd; |
30af21b0 PD |
89 | int payload_size; |
90 | uint64_t bytes_read; /* bytes read from stream when record created */ | |
91 | boolean_t eos_marker; /* Marks the end of the stream */ | |
92 | bqueue_node_t node; | |
93 | }; | |
94 | ||
95 | struct receive_writer_arg { | |
96 | objset_t *os; | |
97 | boolean_t byteswap; | |
98 | bqueue_t q; | |
99 | ||
100 | /* | |
ba67d821 MA |
101 | * These three members are used to signal to the main thread when |
102 | * we're done. | |
30af21b0 PD |
103 | */ |
104 | kmutex_t mutex; | |
105 | kcondvar_t cv; | |
106 | boolean_t done; | |
107 | ||
108 | int err; | |
e8cf3a4f AP |
109 | const char *tofs; |
110 | boolean_t heal; | |
30af21b0 PD |
111 | boolean_t resumable; |
112 | boolean_t raw; /* DMU_BACKUP_FEATURE_RAW set */ | |
113 | boolean_t spill; /* DRR_FLAG_SPILL_BLOCK set */ | |
7bcb7f08 | 114 | boolean_t full; /* this is a full send stream */ |
30af21b0 PD |
115 | uint64_t last_object; |
116 | uint64_t last_offset; | |
117 | uint64_t max_object; /* highest object ID referenced in stream */ | |
118 | uint64_t bytes_read; /* bytes read when current record created */ | |
119 | ||
7261fc2e MA |
120 | list_t write_batch; |
121 | ||
30af21b0 PD |
122 | /* Encryption parameters for the last received DRR_OBJECT_RANGE */ |
123 | boolean_t or_crypt_params_present; | |
124 | uint64_t or_firstobj; | |
125 | uint64_t or_numslots; | |
126 | uint8_t or_salt[ZIO_DATA_SALT_LEN]; | |
127 | uint8_t or_iv[ZIO_DATA_IV_LEN]; | |
128 | uint8_t or_mac[ZIO_DATA_MAC_LEN]; | |
129 | boolean_t or_byteorder; | |
e8cf3a4f | 130 | zio_t *heal_pio; |
30af21b0 PD |
131 | }; |
132 | ||
03916905 PD |
133 | typedef struct dmu_recv_begin_arg { |
134 | const char *drba_origin; | |
135 | dmu_recv_cookie_t *drba_cookie; | |
136 | cred_t *drba_cred; | |
e59a377a | 137 | proc_t *drba_proc; |
03916905 | 138 | dsl_crypto_params_t *drba_dcp; |
03916905 PD |
139 | } dmu_recv_begin_arg_t; |
140 | ||
30af21b0 PD |
141 | static void |
142 | byteswap_record(dmu_replay_record_t *drr) | |
143 | { | |
144 | #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) | |
145 | #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) | |
146 | drr->drr_type = BSWAP_32(drr->drr_type); | |
147 | drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); | |
148 | ||
149 | switch (drr->drr_type) { | |
150 | case DRR_BEGIN: | |
151 | DO64(drr_begin.drr_magic); | |
152 | DO64(drr_begin.drr_versioninfo); | |
153 | DO64(drr_begin.drr_creation_time); | |
154 | DO32(drr_begin.drr_type); | |
155 | DO32(drr_begin.drr_flags); | |
156 | DO64(drr_begin.drr_toguid); | |
157 | DO64(drr_begin.drr_fromguid); | |
158 | break; | |
159 | case DRR_OBJECT: | |
160 | DO64(drr_object.drr_object); | |
161 | DO32(drr_object.drr_type); | |
162 | DO32(drr_object.drr_bonustype); | |
163 | DO32(drr_object.drr_blksz); | |
164 | DO32(drr_object.drr_bonuslen); | |
165 | DO32(drr_object.drr_raw_bonuslen); | |
166 | DO64(drr_object.drr_toguid); | |
167 | DO64(drr_object.drr_maxblkid); | |
168 | break; | |
169 | case DRR_FREEOBJECTS: | |
170 | DO64(drr_freeobjects.drr_firstobj); | |
171 | DO64(drr_freeobjects.drr_numobjs); | |
172 | DO64(drr_freeobjects.drr_toguid); | |
173 | break; | |
174 | case DRR_WRITE: | |
175 | DO64(drr_write.drr_object); | |
176 | DO32(drr_write.drr_type); | |
177 | DO64(drr_write.drr_offset); | |
178 | DO64(drr_write.drr_logical_size); | |
179 | DO64(drr_write.drr_toguid); | |
180 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write.drr_key.ddk_cksum); | |
181 | DO64(drr_write.drr_key.ddk_prop); | |
182 | DO64(drr_write.drr_compressed_size); | |
183 | break; | |
30af21b0 PD |
184 | case DRR_WRITE_EMBEDDED: |
185 | DO64(drr_write_embedded.drr_object); | |
186 | DO64(drr_write_embedded.drr_offset); | |
187 | DO64(drr_write_embedded.drr_length); | |
188 | DO64(drr_write_embedded.drr_toguid); | |
189 | DO32(drr_write_embedded.drr_lsize); | |
190 | DO32(drr_write_embedded.drr_psize); | |
191 | break; | |
192 | case DRR_FREE: | |
193 | DO64(drr_free.drr_object); | |
194 | DO64(drr_free.drr_offset); | |
195 | DO64(drr_free.drr_length); | |
196 | DO64(drr_free.drr_toguid); | |
197 | break; | |
198 | case DRR_SPILL: | |
199 | DO64(drr_spill.drr_object); | |
200 | DO64(drr_spill.drr_length); | |
201 | DO64(drr_spill.drr_toguid); | |
202 | DO64(drr_spill.drr_compressed_size); | |
203 | DO32(drr_spill.drr_type); | |
204 | break; | |
205 | case DRR_OBJECT_RANGE: | |
206 | DO64(drr_object_range.drr_firstobj); | |
207 | DO64(drr_object_range.drr_numslots); | |
208 | DO64(drr_object_range.drr_toguid); | |
209 | break; | |
210 | case DRR_REDACT: | |
211 | DO64(drr_redact.drr_object); | |
212 | DO64(drr_redact.drr_offset); | |
213 | DO64(drr_redact.drr_length); | |
214 | DO64(drr_redact.drr_toguid); | |
215 | break; | |
216 | case DRR_END: | |
217 | DO64(drr_end.drr_toguid); | |
218 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_end.drr_checksum); | |
219 | break; | |
220 | default: | |
221 | break; | |
222 | } | |
223 | ||
224 | if (drr->drr_type != DRR_BEGIN) { | |
225 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_checksum.drr_checksum); | |
226 | } | |
227 | ||
228 | #undef DO64 | |
229 | #undef DO32 | |
230 | } | |
231 | ||
232 | static boolean_t | |
233 | redact_snaps_contains(uint64_t *snaps, uint64_t num_snaps, uint64_t guid) | |
234 | { | |
235 | for (int i = 0; i < num_snaps; i++) { | |
236 | if (snaps[i] == guid) | |
237 | return (B_TRUE); | |
238 | } | |
239 | return (B_FALSE); | |
240 | } | |
241 | ||
242 | /* | |
243 | * Check that the new stream we're trying to receive is redacted with respect to | |
244 | * a subset of the snapshots that the origin was redacted with respect to. For | |
245 | * the reasons behind this, see the man page on redacted zfs sends and receives. | |
246 | */ | |
247 | static boolean_t | |
248 | compatible_redact_snaps(uint64_t *origin_snaps, uint64_t origin_num_snaps, | |
249 | uint64_t *redact_snaps, uint64_t num_redact_snaps) | |
250 | { | |
251 | /* | |
252 | * Short circuit the comparison; if we are redacted with respect to | |
253 | * more snapshots than the origin, we can't be redacted with respect | |
254 | * to a subset. | |
255 | */ | |
256 | if (num_redact_snaps > origin_num_snaps) { | |
257 | return (B_FALSE); | |
258 | } | |
259 | ||
260 | for (int i = 0; i < num_redact_snaps; i++) { | |
261 | if (!redact_snaps_contains(origin_snaps, origin_num_snaps, | |
262 | redact_snaps[i])) { | |
263 | return (B_FALSE); | |
264 | } | |
265 | } | |
266 | return (B_TRUE); | |
267 | } | |
268 | ||
269 | static boolean_t | |
270 | redact_check(dmu_recv_begin_arg_t *drba, dsl_dataset_t *origin) | |
271 | { | |
272 | uint64_t *origin_snaps; | |
273 | uint64_t origin_num_snaps; | |
274 | dmu_recv_cookie_t *drc = drba->drba_cookie; | |
275 | struct drr_begin *drrb = drc->drc_drrb; | |
276 | int featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); | |
277 | int err = 0; | |
278 | boolean_t ret = B_TRUE; | |
279 | uint64_t *redact_snaps; | |
280 | uint_t numredactsnaps; | |
281 | ||
282 | /* | |
283 | * If this is a full send stream, we're safe no matter what. | |
284 | */ | |
285 | if (drrb->drr_fromguid == 0) | |
286 | return (ret); | |
287 | ||
288 | VERIFY(dsl_dataset_get_uint64_array_feature(origin, | |
289 | SPA_FEATURE_REDACTED_DATASETS, &origin_num_snaps, &origin_snaps)); | |
290 | ||
291 | if (nvlist_lookup_uint64_array(drc->drc_begin_nvl, | |
292 | BEGINNV_REDACT_FROM_SNAPS, &redact_snaps, &numredactsnaps) == | |
293 | 0) { | |
294 | /* | |
295 | * If the send stream was sent from the redaction bookmark or | |
296 | * the redacted version of the dataset, then we're safe. Verify | |
297 | * that this is from the a compatible redaction bookmark or | |
298 | * redacted dataset. | |
299 | */ | |
300 | if (!compatible_redact_snaps(origin_snaps, origin_num_snaps, | |
301 | redact_snaps, numredactsnaps)) { | |
302 | err = EINVAL; | |
303 | } | |
304 | } else if (featureflags & DMU_BACKUP_FEATURE_REDACTED) { | |
305 | /* | |
306 | * If the stream is redacted, it must be redacted with respect | |
307 | * to a subset of what the origin is redacted with respect to. | |
308 | * See case number 2 in the zfs man page section on redacted zfs | |
309 | * send. | |
310 | */ | |
311 | err = nvlist_lookup_uint64_array(drc->drc_begin_nvl, | |
312 | BEGINNV_REDACT_SNAPS, &redact_snaps, &numredactsnaps); | |
313 | ||
314 | if (err != 0 || !compatible_redact_snaps(origin_snaps, | |
315 | origin_num_snaps, redact_snaps, numredactsnaps)) { | |
316 | err = EINVAL; | |
317 | } | |
318 | } else if (!redact_snaps_contains(origin_snaps, origin_num_snaps, | |
319 | drrb->drr_toguid)) { | |
320 | /* | |
321 | * If the stream isn't redacted but the origin is, this must be | |
322 | * one of the snapshots the origin is redacted with respect to. | |
323 | * See case number 1 in the zfs man page section on redacted zfs | |
324 | * send. | |
325 | */ | |
326 | err = EINVAL; | |
327 | } | |
328 | ||
329 | if (err != 0) | |
330 | ret = B_FALSE; | |
331 | return (ret); | |
332 | } | |
333 | ||
7bcb7f08 MA |
334 | /* |
335 | * If we previously received a stream with --large-block, we don't support | |
336 | * receiving an incremental on top of it without --large-block. This avoids | |
337 | * forcing a read-modify-write or trying to re-aggregate a string of WRITE | |
338 | * records. | |
339 | */ | |
340 | static int | |
341 | recv_check_large_blocks(dsl_dataset_t *ds, uint64_t featureflags) | |
342 | { | |
343 | if (dsl_dataset_feature_is_active(ds, SPA_FEATURE_LARGE_BLOCKS) && | |
344 | !(featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS)) | |
345 | return (SET_ERROR(ZFS_ERR_STREAM_LARGE_BLOCK_MISMATCH)); | |
346 | return (0); | |
347 | } | |
348 | ||
03916905 PD |
349 | static int |
350 | recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, | |
351 | uint64_t fromguid, uint64_t featureflags) | |
352 | { | |
e8cf3a4f | 353 | uint64_t obj; |
d8d418ff | 354 | uint64_t children; |
03916905 | 355 | int error; |
e8cf3a4f | 356 | dsl_dataset_t *snap; |
03916905 PD |
357 | dsl_pool_t *dp = ds->ds_dir->dd_pool; |
358 | boolean_t encrypted = ds->ds_dir->dd_crypto_obj != 0; | |
359 | boolean_t raw = (featureflags & DMU_BACKUP_FEATURE_RAW) != 0; | |
360 | boolean_t embed = (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) != 0; | |
361 | ||
ebeb6f23 | 362 | /* Temporary clone name must not exist. */ |
03916905 PD |
363 | error = zap_lookup(dp->dp_meta_objset, |
364 | dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, | |
e8cf3a4f | 365 | 8, 1, &obj); |
03916905 | 366 | if (error != ENOENT) |
30af21b0 | 367 | return (error == 0 ? SET_ERROR(EBUSY) : error); |
03916905 | 368 | |
ebeb6f23 AG |
369 | /* Resume state must not be set. */ |
370 | if (dsl_dataset_has_resume_receive_state(ds)) | |
371 | return (SET_ERROR(EBUSY)); | |
372 | ||
e8cf3a4f | 373 | /* New snapshot name must not exist if we're not healing it. */ |
03916905 PD |
374 | error = zap_lookup(dp->dp_meta_objset, |
375 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, | |
e8cf3a4f AP |
376 | drba->drba_cookie->drc_tosnap, 8, 1, &obj); |
377 | if (drba->drba_cookie->drc_heal) { | |
378 | if (error != 0) | |
379 | return (error); | |
380 | } else if (error != ENOENT) { | |
30af21b0 | 381 | return (error == 0 ? SET_ERROR(EEXIST) : error); |
e8cf3a4f | 382 | } |
03916905 | 383 | |
ebeb6f23 | 384 | /* Must not have children if receiving a ZVOL. */ |
d8d418ff | 385 | error = zap_count(dp->dp_meta_objset, |
386 | dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, &children); | |
387 | if (error != 0) | |
388 | return (error); | |
389 | if (drba->drba_cookie->drc_drrb->drr_type != DMU_OST_ZFS && | |
390 | children > 0) | |
391 | return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); | |
392 | ||
03916905 PD |
393 | /* |
394 | * Check snapshot limit before receiving. We'll recheck again at the | |
395 | * end, but might as well abort before receiving if we're already over | |
396 | * the limit. | |
397 | * | |
398 | * Note that we do not check the file system limit with | |
399 | * dsl_dir_fscount_check because the temporary %clones don't count | |
400 | * against that limit. | |
401 | */ | |
402 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, | |
e59a377a | 403 | NULL, drba->drba_cred, drba->drba_proc); |
03916905 PD |
404 | if (error != 0) |
405 | return (error); | |
406 | ||
e8cf3a4f AP |
407 | if (drba->drba_cookie->drc_heal) { |
408 | /* Encryption is incompatible with embedded data. */ | |
409 | if (encrypted && embed) | |
410 | return (SET_ERROR(EINVAL)); | |
411 | ||
412 | /* Healing is not supported when in 'force' mode. */ | |
413 | if (drba->drba_cookie->drc_force) | |
414 | return (SET_ERROR(EINVAL)); | |
415 | ||
416 | /* Must have keys loaded if doing encrypted non-raw recv. */ | |
417 | if (encrypted && !raw) { | |
418 | if (spa_keystore_lookup_key(dp->dp_spa, ds->ds_object, | |
419 | NULL, NULL) != 0) | |
420 | return (SET_ERROR(EACCES)); | |
421 | } | |
422 | ||
423 | error = dsl_dataset_hold_obj(dp, obj, FTAG, &snap); | |
424 | if (error != 0) | |
425 | return (error); | |
426 | ||
427 | /* | |
428 | * When not doing best effort corrective recv healing can only | |
429 | * be done if the send stream is for the same snapshot as the | |
430 | * one we are trying to heal. | |
431 | */ | |
432 | if (zfs_recv_best_effort_corrective == 0 && | |
433 | drba->drba_cookie->drc_drrb->drr_toguid != | |
434 | dsl_dataset_phys(snap)->ds_guid) { | |
435 | dsl_dataset_rele(snap, FTAG); | |
436 | return (SET_ERROR(ENOTSUP)); | |
437 | } | |
438 | dsl_dataset_rele(snap, FTAG); | |
439 | } else if (fromguid != 0) { | |
440 | /* Sanity check the incremental recv */ | |
03916905 PD |
441 | uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; |
442 | ||
30af21b0 | 443 | /* Can't perform a raw receive on top of a non-raw receive */ |
03916905 PD |
444 | if (!encrypted && raw) |
445 | return (SET_ERROR(EINVAL)); | |
446 | ||
447 | /* Encryption is incompatible with embedded data */ | |
448 | if (encrypted && embed) | |
449 | return (SET_ERROR(EINVAL)); | |
450 | ||
451 | /* Find snapshot in this dir that matches fromguid. */ | |
452 | while (obj != 0) { | |
453 | error = dsl_dataset_hold_obj(dp, obj, FTAG, | |
454 | &snap); | |
455 | if (error != 0) | |
456 | return (SET_ERROR(ENODEV)); | |
457 | if (snap->ds_dir != ds->ds_dir) { | |
458 | dsl_dataset_rele(snap, FTAG); | |
459 | return (SET_ERROR(ENODEV)); | |
460 | } | |
461 | if (dsl_dataset_phys(snap)->ds_guid == fromguid) | |
462 | break; | |
463 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; | |
464 | dsl_dataset_rele(snap, FTAG); | |
465 | } | |
466 | if (obj == 0) | |
467 | return (SET_ERROR(ENODEV)); | |
468 | ||
469 | if (drba->drba_cookie->drc_force) { | |
f00ab3f2 | 470 | drba->drba_cookie->drc_fromsnapobj = obj; |
03916905 PD |
471 | } else { |
472 | /* | |
473 | * If we are not forcing, there must be no | |
c08c30ed TC |
474 | * changes since fromsnap. Raw sends have an |
475 | * additional constraint that requires that | |
476 | * no "noop" snapshots exist between fromsnap | |
477 | * and tosnap for the IVset checking code to | |
478 | * work properly. | |
03916905 | 479 | */ |
c08c30ed TC |
480 | if (dsl_dataset_modified_since_snap(ds, snap) || |
481 | (raw && | |
482 | dsl_dataset_phys(ds)->ds_prev_snap_obj != | |
483 | snap->ds_object)) { | |
03916905 PD |
484 | dsl_dataset_rele(snap, FTAG); |
485 | return (SET_ERROR(ETXTBSY)); | |
486 | } | |
f00ab3f2 TC |
487 | drba->drba_cookie->drc_fromsnapobj = |
488 | ds->ds_prev->ds_object; | |
03916905 PD |
489 | } |
490 | ||
30af21b0 PD |
491 | if (dsl_dataset_feature_is_active(snap, |
492 | SPA_FEATURE_REDACTED_DATASETS) && !redact_check(drba, | |
493 | snap)) { | |
494 | dsl_dataset_rele(snap, FTAG); | |
495 | return (SET_ERROR(EINVAL)); | |
496 | } | |
497 | ||
7bcb7f08 MA |
498 | error = recv_check_large_blocks(snap, featureflags); |
499 | if (error != 0) { | |
500 | dsl_dataset_rele(snap, FTAG); | |
501 | return (error); | |
502 | } | |
503 | ||
03916905 PD |
504 | dsl_dataset_rele(snap, FTAG); |
505 | } else { | |
e8cf3a4f | 506 | /* If full and not healing then must be forced. */ |
03916905 PD |
507 | if (!drba->drba_cookie->drc_force) |
508 | return (SET_ERROR(EEXIST)); | |
509 | ||
510 | /* | |
511 | * We don't support using zfs recv -F to blow away | |
512 | * encrypted filesystems. This would require the | |
513 | * dsl dir to point to the old encryption key and | |
514 | * the new one at the same time during the receive. | |
515 | */ | |
516 | if ((!encrypted && raw) || encrypted) | |
517 | return (SET_ERROR(EINVAL)); | |
518 | ||
519 | /* | |
520 | * Perform the same encryption checks we would if | |
521 | * we were creating a new dataset from scratch. | |
522 | */ | |
523 | if (!raw) { | |
524 | boolean_t will_encrypt; | |
525 | ||
526 | error = dmu_objset_create_crypt_check( | |
527 | ds->ds_dir->dd_parent, drba->drba_dcp, | |
528 | &will_encrypt); | |
529 | if (error != 0) | |
530 | return (error); | |
531 | ||
532 | if (will_encrypt && embed) | |
533 | return (SET_ERROR(EINVAL)); | |
534 | } | |
03916905 PD |
535 | } |
536 | ||
537 | return (0); | |
03916905 PD |
538 | } |
539 | ||
30af21b0 PD |
540 | /* |
541 | * Check that any feature flags used in the data stream we're receiving are | |
542 | * supported by the pool we are receiving into. | |
543 | * | |
544 | * Note that some of the features we explicitly check here have additional | |
545 | * (implicit) features they depend on, but those dependencies are enforced | |
546 | * through the zfeature_register() calls declaring the features that we | |
547 | * explicitly check. | |
548 | */ | |
549 | static int | |
550 | recv_begin_check_feature_flags_impl(uint64_t featureflags, spa_t *spa) | |
551 | { | |
552 | /* | |
553 | * Check if there are any unsupported feature flags. | |
554 | */ | |
555 | if (!DMU_STREAM_SUPPORTED(featureflags)) { | |
556 | return (SET_ERROR(ZFS_ERR_UNKNOWN_SEND_STREAM_FEATURE)); | |
557 | } | |
558 | ||
559 | /* Verify pool version supports SA if SA_SPILL feature set */ | |
560 | if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && | |
561 | spa_version(spa) < SPA_VERSION_SA) | |
562 | return (SET_ERROR(ENOTSUP)); | |
563 | ||
564 | /* | |
10b3c7f5 MN |
565 | * LZ4 compressed, ZSTD compressed, embedded, mooched, large blocks, |
566 | * and large_dnodes in the stream can only be used if those pool | |
567 | * features are enabled because we don't attempt to decompress / | |
568 | * un-embed / un-mooch / split up the blocks / dnodes during the | |
569 | * receive process. | |
30af21b0 PD |
570 | */ |
571 | if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && | |
572 | !spa_feature_is_enabled(spa, SPA_FEATURE_LZ4_COMPRESS)) | |
573 | return (SET_ERROR(ENOTSUP)); | |
10b3c7f5 MN |
574 | if ((featureflags & DMU_BACKUP_FEATURE_ZSTD) && |
575 | !spa_feature_is_enabled(spa, SPA_FEATURE_ZSTD_COMPRESS)) | |
576 | return (SET_ERROR(ENOTSUP)); | |
30af21b0 PD |
577 | if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && |
578 | !spa_feature_is_enabled(spa, SPA_FEATURE_EMBEDDED_DATA)) | |
579 | return (SET_ERROR(ENOTSUP)); | |
580 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
581 | !spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) | |
582 | return (SET_ERROR(ENOTSUP)); | |
583 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_DNODE) && | |
584 | !spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) | |
585 | return (SET_ERROR(ENOTSUP)); | |
586 | ||
587 | /* | |
588 | * Receiving redacted streams requires that redacted datasets are | |
589 | * enabled. | |
590 | */ | |
591 | if ((featureflags & DMU_BACKUP_FEATURE_REDACTED) && | |
592 | !spa_feature_is_enabled(spa, SPA_FEATURE_REDACTED_DATASETS)) | |
593 | return (SET_ERROR(ENOTSUP)); | |
594 | ||
595 | return (0); | |
596 | } | |
597 | ||
03916905 PD |
598 | static int |
599 | dmu_recv_begin_check(void *arg, dmu_tx_t *tx) | |
600 | { | |
601 | dmu_recv_begin_arg_t *drba = arg; | |
602 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
603 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
604 | uint64_t fromguid = drrb->drr_fromguid; | |
605 | int flags = drrb->drr_flags; | |
40ab927a | 606 | ds_hold_flags_t dsflags = DS_HOLD_FLAG_NONE; |
03916905 | 607 | int error; |
30af21b0 | 608 | uint64_t featureflags = drba->drba_cookie->drc_featureflags; |
03916905 PD |
609 | dsl_dataset_t *ds; |
610 | const char *tofs = drba->drba_cookie->drc_tofs; | |
611 | ||
612 | /* already checked */ | |
613 | ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); | |
614 | ASSERT(!(featureflags & DMU_BACKUP_FEATURE_RESUMING)); | |
615 | ||
616 | if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == | |
617 | DMU_COMPOUNDSTREAM || | |
618 | drrb->drr_type >= DMU_OST_NUMTYPES || | |
619 | ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) | |
620 | return (SET_ERROR(EINVAL)); | |
621 | ||
30af21b0 PD |
622 | error = recv_begin_check_feature_flags_impl(featureflags, dp->dp_spa); |
623 | if (error != 0) | |
624 | return (error); | |
03916905 | 625 | |
30af21b0 | 626 | /* Resumable receives require extensible datasets */ |
03916905 PD |
627 | if (drba->drba_cookie->drc_resumable && |
628 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EXTENSIBLE_DATASET)) | |
629 | return (SET_ERROR(ENOTSUP)); | |
630 | ||
03916905 PD |
631 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { |
632 | /* raw receives require the encryption feature */ | |
633 | if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ENCRYPTION)) | |
634 | return (SET_ERROR(ENOTSUP)); | |
635 | ||
636 | /* embedded data is incompatible with encryption and raw recv */ | |
637 | if (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) | |
638 | return (SET_ERROR(EINVAL)); | |
caf9dd20 BB |
639 | |
640 | /* raw receives require spill block allocation flag */ | |
641 | if (!(flags & DRR_FLAG_SPILL_BLOCK)) | |
642 | return (SET_ERROR(ZFS_ERR_SPILL_BLOCK_FLAG_MISSING)); | |
03916905 | 643 | } else { |
68ddc06b AF |
644 | /* |
645 | * We support unencrypted datasets below encrypted ones now, | |
646 | * so add the DS_HOLD_FLAG_DECRYPT flag only if we are dealing | |
647 | * with a dataset we may encrypt. | |
648 | */ | |
211ec1b9 | 649 | if (drba->drba_dcp == NULL || |
68ddc06b AF |
650 | drba->drba_dcp->cp_crypt != ZIO_CRYPT_OFF) { |
651 | dsflags |= DS_HOLD_FLAG_DECRYPT; | |
652 | } | |
03916905 PD |
653 | } |
654 | ||
655 | error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); | |
656 | if (error == 0) { | |
657 | /* target fs already exists; recv into temp clone */ | |
658 | ||
659 | /* Can't recv a clone into an existing fs */ | |
660 | if (flags & DRR_FLAG_CLONE || drba->drba_origin) { | |
661 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
662 | return (SET_ERROR(EINVAL)); | |
663 | } | |
664 | ||
665 | error = recv_begin_check_existing_impl(drba, ds, fromguid, | |
666 | featureflags); | |
667 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
668 | } else if (error == ENOENT) { | |
669 | /* target fs does not exist; must be a full backup or clone */ | |
670 | char buf[ZFS_MAX_DATASET_NAME_LEN]; | |
d8d418ff | 671 | objset_t *os; |
03916905 | 672 | |
e8cf3a4f AP |
673 | /* healing recv must be done "into" an existing snapshot */ |
674 | if (drba->drba_cookie->drc_heal == B_TRUE) | |
675 | return (SET_ERROR(ENOTSUP)); | |
676 | ||
03916905 PD |
677 | /* |
678 | * If it's a non-clone incremental, we are missing the | |
679 | * target fs, so fail the recv. | |
680 | */ | |
30af21b0 | 681 | if (fromguid != 0 && !((flags & DRR_FLAG_CLONE) || |
03916905 PD |
682 | drba->drba_origin)) |
683 | return (SET_ERROR(ENOENT)); | |
684 | ||
685 | /* | |
686 | * If we're receiving a full send as a clone, and it doesn't | |
687 | * contain all the necessary free records and freeobject | |
688 | * records, reject it. | |
689 | */ | |
30af21b0 | 690 | if (fromguid == 0 && drba->drba_origin != NULL && |
03916905 PD |
691 | !(flags & DRR_FLAG_FREERECORDS)) |
692 | return (SET_ERROR(EINVAL)); | |
693 | ||
694 | /* Open the parent of tofs */ | |
695 | ASSERT3U(strlen(tofs), <, sizeof (buf)); | |
696 | (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); | |
da689887 | 697 | error = dsl_dataset_hold(dp, buf, FTAG, &ds); |
03916905 PD |
698 | if (error != 0) |
699 | return (error); | |
700 | ||
701 | if ((featureflags & DMU_BACKUP_FEATURE_RAW) == 0 && | |
702 | drba->drba_origin == NULL) { | |
703 | boolean_t will_encrypt; | |
704 | ||
705 | /* | |
706 | * Check that we aren't breaking any encryption rules | |
707 | * and that we have all the parameters we need to | |
708 | * create an encrypted dataset if necessary. If we are | |
709 | * making an encrypted dataset the stream can't have | |
710 | * embedded data. | |
711 | */ | |
712 | error = dmu_objset_create_crypt_check(ds->ds_dir, | |
713 | drba->drba_dcp, &will_encrypt); | |
714 | if (error != 0) { | |
da689887 | 715 | dsl_dataset_rele(ds, FTAG); |
03916905 PD |
716 | return (error); |
717 | } | |
718 | ||
719 | if (will_encrypt && | |
720 | (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)) { | |
da689887 | 721 | dsl_dataset_rele(ds, FTAG); |
03916905 PD |
722 | return (SET_ERROR(EINVAL)); |
723 | } | |
724 | } | |
725 | ||
726 | /* | |
727 | * Check filesystem and snapshot limits before receiving. We'll | |
728 | * recheck snapshot limits again at the end (we create the | |
729 | * filesystems and increment those counts during begin_sync). | |
730 | */ | |
731 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, | |
e59a377a MA |
732 | ZFS_PROP_FILESYSTEM_LIMIT, NULL, |
733 | drba->drba_cred, drba->drba_proc); | |
03916905 | 734 | if (error != 0) { |
da689887 | 735 | dsl_dataset_rele(ds, FTAG); |
03916905 PD |
736 | return (error); |
737 | } | |
738 | ||
739 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, | |
e59a377a MA |
740 | ZFS_PROP_SNAPSHOT_LIMIT, NULL, |
741 | drba->drba_cred, drba->drba_proc); | |
03916905 | 742 | if (error != 0) { |
da689887 | 743 | dsl_dataset_rele(ds, FTAG); |
03916905 PD |
744 | return (error); |
745 | } | |
746 | ||
d8d418ff | 747 | /* can't recv below anything but filesystems (eg. no ZVOLs) */ |
748 | error = dmu_objset_from_ds(ds, &os); | |
749 | if (error != 0) { | |
da689887 | 750 | dsl_dataset_rele(ds, FTAG); |
d8d418ff | 751 | return (error); |
752 | } | |
753 | if (dmu_objset_type(os) != DMU_OST_ZFS) { | |
da689887 | 754 | dsl_dataset_rele(ds, FTAG); |
d8d418ff | 755 | return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); |
756 | } | |
757 | ||
03916905 PD |
758 | if (drba->drba_origin != NULL) { |
759 | dsl_dataset_t *origin; | |
03916905 PD |
760 | error = dsl_dataset_hold_flags(dp, drba->drba_origin, |
761 | dsflags, FTAG, &origin); | |
762 | if (error != 0) { | |
da689887 | 763 | dsl_dataset_rele(ds, FTAG); |
03916905 PD |
764 | return (error); |
765 | } | |
766 | if (!origin->ds_is_snapshot) { | |
767 | dsl_dataset_rele_flags(origin, dsflags, FTAG); | |
da689887 | 768 | dsl_dataset_rele(ds, FTAG); |
03916905 PD |
769 | return (SET_ERROR(EINVAL)); |
770 | } | |
771 | if (dsl_dataset_phys(origin)->ds_guid != fromguid && | |
772 | fromguid != 0) { | |
773 | dsl_dataset_rele_flags(origin, dsflags, FTAG); | |
da689887 | 774 | dsl_dataset_rele(ds, FTAG); |
03916905 PD |
775 | return (SET_ERROR(ENODEV)); |
776 | } | |
30af21b0 | 777 | |
03916905 PD |
778 | if (origin->ds_dir->dd_crypto_obj != 0 && |
779 | (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)) { | |
780 | dsl_dataset_rele_flags(origin, dsflags, FTAG); | |
da689887 | 781 | dsl_dataset_rele(ds, FTAG); |
03916905 PD |
782 | return (SET_ERROR(EINVAL)); |
783 | } | |
30af21b0 PD |
784 | |
785 | /* | |
786 | * If the origin is redacted we need to verify that this | |
787 | * send stream can safely be received on top of the | |
788 | * origin. | |
789 | */ | |
790 | if (dsl_dataset_feature_is_active(origin, | |
791 | SPA_FEATURE_REDACTED_DATASETS)) { | |
792 | if (!redact_check(drba, origin)) { | |
793 | dsl_dataset_rele_flags(origin, dsflags, | |
794 | FTAG); | |
795 | dsl_dataset_rele_flags(ds, dsflags, | |
796 | FTAG); | |
797 | return (SET_ERROR(EINVAL)); | |
798 | } | |
799 | } | |
800 | ||
7bcb7f08 MA |
801 | error = recv_check_large_blocks(ds, featureflags); |
802 | if (error != 0) { | |
803 | dsl_dataset_rele_flags(origin, dsflags, FTAG); | |
804 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
805 | return (error); | |
806 | } | |
807 | ||
30af21b0 | 808 | dsl_dataset_rele_flags(origin, dsflags, FTAG); |
03916905 | 809 | } |
d8d418ff | 810 | |
da689887 | 811 | dsl_dataset_rele(ds, FTAG); |
03916905 PD |
812 | error = 0; |
813 | } | |
814 | return (error); | |
815 | } | |
816 | ||
817 | static void | |
818 | dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) | |
819 | { | |
820 | dmu_recv_begin_arg_t *drba = arg; | |
821 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
822 | objset_t *mos = dp->dp_meta_objset; | |
30af21b0 PD |
823 | dmu_recv_cookie_t *drc = drba->drba_cookie; |
824 | struct drr_begin *drrb = drc->drc_drrb; | |
825 | const char *tofs = drc->drc_tofs; | |
826 | uint64_t featureflags = drc->drc_featureflags; | |
03916905 PD |
827 | dsl_dataset_t *ds, *newds; |
828 | objset_t *os; | |
829 | uint64_t dsobj; | |
40ab927a | 830 | ds_hold_flags_t dsflags = DS_HOLD_FLAG_NONE; |
03916905 PD |
831 | int error; |
832 | uint64_t crflags = 0; | |
833 | dsl_crypto_params_t dummy_dcp = { 0 }; | |
834 | dsl_crypto_params_t *dcp = drba->drba_dcp; | |
835 | ||
836 | if (drrb->drr_flags & DRR_FLAG_CI_DATA) | |
837 | crflags |= DS_FLAG_CI_DATASET; | |
838 | ||
839 | if ((featureflags & DMU_BACKUP_FEATURE_RAW) == 0) | |
840 | dsflags |= DS_HOLD_FLAG_DECRYPT; | |
841 | ||
842 | /* | |
843 | * Raw, non-incremental recvs always use a dummy dcp with | |
844 | * the raw cmd set. Raw incremental recvs do not use a dcp | |
845 | * since the encryption parameters are already set in stone. | |
846 | */ | |
30af21b0 | 847 | if (dcp == NULL && drrb->drr_fromguid == 0 && |
03916905 PD |
848 | drba->drba_origin == NULL) { |
849 | ASSERT3P(dcp, ==, NULL); | |
850 | dcp = &dummy_dcp; | |
851 | ||
852 | if (featureflags & DMU_BACKUP_FEATURE_RAW) | |
853 | dcp->cp_cmd = DCP_CMD_RAW_RECV; | |
854 | } | |
855 | ||
856 | error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); | |
857 | if (error == 0) { | |
e8cf3a4f | 858 | /* Create temporary clone unless we're doing corrective recv */ |
03916905 PD |
859 | dsl_dataset_t *snap = NULL; |
860 | ||
f00ab3f2 | 861 | if (drba->drba_cookie->drc_fromsnapobj != 0) { |
03916905 | 862 | VERIFY0(dsl_dataset_hold_obj(dp, |
f00ab3f2 | 863 | drba->drba_cookie->drc_fromsnapobj, FTAG, &snap)); |
03916905 PD |
864 | ASSERT3P(dcp, ==, NULL); |
865 | } | |
e8cf3a4f AP |
866 | if (drc->drc_heal) { |
867 | /* When healing we want to use the provided snapshot */ | |
868 | VERIFY0(dsl_dataset_snap_lookup(ds, drc->drc_tosnap, | |
869 | &dsobj)); | |
870 | } else { | |
871 | dsobj = dsl_dataset_create_sync(ds->ds_dir, | |
872 | recv_clone_name, snap, crflags, drba->drba_cred, | |
873 | dcp, tx); | |
874 | } | |
f00ab3f2 | 875 | if (drba->drba_cookie->drc_fromsnapobj != 0) |
03916905 PD |
876 | dsl_dataset_rele(snap, FTAG); |
877 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
878 | } else { | |
879 | dsl_dir_t *dd; | |
880 | const char *tail; | |
881 | dsl_dataset_t *origin = NULL; | |
882 | ||
883 | VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); | |
884 | ||
885 | if (drba->drba_origin != NULL) { | |
886 | VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, | |
887 | FTAG, &origin)); | |
888 | ASSERT3P(dcp, ==, NULL); | |
889 | } | |
890 | ||
891 | /* Create new dataset. */ | |
892 | dsobj = dsl_dataset_create_sync(dd, strrchr(tofs, '/') + 1, | |
893 | origin, crflags, drba->drba_cred, dcp, tx); | |
894 | if (origin != NULL) | |
895 | dsl_dataset_rele(origin, FTAG); | |
896 | dsl_dir_rele(dd, FTAG); | |
30af21b0 PD |
897 | drc->drc_newfs = B_TRUE; |
898 | } | |
899 | VERIFY0(dsl_dataset_own_obj_force(dp, dsobj, dsflags, dmu_recv_tag, | |
900 | &newds)); | |
901 | if (dsl_dataset_feature_is_active(newds, | |
902 | SPA_FEATURE_REDACTED_DATASETS)) { | |
903 | /* | |
904 | * If the origin dataset is redacted, the child will be redacted | |
905 | * when we create it. We clear the new dataset's | |
906 | * redaction info; if it should be redacted, we'll fill | |
907 | * in its information later. | |
908 | */ | |
909 | dsl_dataset_deactivate_feature(newds, | |
910 | SPA_FEATURE_REDACTED_DATASETS, tx); | |
03916905 | 911 | } |
03916905 PD |
912 | VERIFY0(dmu_objset_from_ds(newds, &os)); |
913 | ||
30af21b0 | 914 | if (drc->drc_resumable) { |
03916905 PD |
915 | dsl_dataset_zapify(newds, tx); |
916 | if (drrb->drr_fromguid != 0) { | |
917 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_FROMGUID, | |
918 | 8, 1, &drrb->drr_fromguid, tx)); | |
919 | } | |
920 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TOGUID, | |
921 | 8, 1, &drrb->drr_toguid, tx)); | |
922 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TONAME, | |
923 | 1, strlen(drrb->drr_toname) + 1, drrb->drr_toname, tx)); | |
924 | uint64_t one = 1; | |
925 | uint64_t zero = 0; | |
926 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OBJECT, | |
927 | 8, 1, &one, tx)); | |
928 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OFFSET, | |
929 | 8, 1, &zero, tx)); | |
930 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_BYTES, | |
931 | 8, 1, &zero, tx)); | |
932 | if (featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) { | |
933 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_LARGEBLOCK, | |
934 | 8, 1, &one, tx)); | |
935 | } | |
936 | if (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) { | |
937 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_EMBEDOK, | |
938 | 8, 1, &one, tx)); | |
939 | } | |
940 | if (featureflags & DMU_BACKUP_FEATURE_COMPRESSED) { | |
941 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_COMPRESSOK, | |
942 | 8, 1, &one, tx)); | |
943 | } | |
944 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
945 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_RAWOK, | |
946 | 8, 1, &one, tx)); | |
947 | } | |
30af21b0 PD |
948 | |
949 | uint64_t *redact_snaps; | |
950 | uint_t numredactsnaps; | |
951 | if (nvlist_lookup_uint64_array(drc->drc_begin_nvl, | |
952 | BEGINNV_REDACT_FROM_SNAPS, &redact_snaps, | |
953 | &numredactsnaps) == 0) { | |
954 | VERIFY0(zap_add(mos, dsobj, | |
955 | DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, | |
956 | sizeof (*redact_snaps), numredactsnaps, | |
957 | redact_snaps, tx)); | |
958 | } | |
03916905 PD |
959 | } |
960 | ||
961 | /* | |
962 | * Usually the os->os_encrypted value is tied to the presence of a | |
963 | * DSL Crypto Key object in the dd. However, that will not be received | |
964 | * until dmu_recv_stream(), so we set the value manually for now. | |
965 | */ | |
966 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
967 | os->os_encrypted = B_TRUE; | |
968 | drba->drba_cookie->drc_raw = B_TRUE; | |
969 | } | |
970 | ||
30af21b0 PD |
971 | if (featureflags & DMU_BACKUP_FEATURE_REDACTED) { |
972 | uint64_t *redact_snaps; | |
973 | uint_t numredactsnaps; | |
974 | VERIFY0(nvlist_lookup_uint64_array(drc->drc_begin_nvl, | |
975 | BEGINNV_REDACT_SNAPS, &redact_snaps, &numredactsnaps)); | |
976 | dsl_dataset_activate_redaction(newds, redact_snaps, | |
977 | numredactsnaps, tx); | |
978 | } | |
979 | ||
03916905 PD |
980 | dmu_buf_will_dirty(newds->ds_dbuf, tx); |
981 | dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; | |
982 | ||
983 | /* | |
984 | * If we actually created a non-clone, we need to create the objset | |
985 | * in our new dataset. If this is a raw send we postpone this until | |
986 | * dmu_recv_stream() so that we can allocate the metadnode with the | |
987 | * properties from the DRR_BEGIN payload. | |
988 | */ | |
989 | rrw_enter(&newds->ds_bp_rwlock, RW_READER, FTAG); | |
990 | if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds)) && | |
e8cf3a4f AP |
991 | (featureflags & DMU_BACKUP_FEATURE_RAW) == 0 && |
992 | !drc->drc_heal) { | |
03916905 PD |
993 | (void) dmu_objset_create_impl(dp->dp_spa, |
994 | newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); | |
995 | } | |
996 | rrw_exit(&newds->ds_bp_rwlock, FTAG); | |
997 | ||
998 | drba->drba_cookie->drc_ds = newds; | |
0fdd6106 | 999 | drba->drba_cookie->drc_os = os; |
03916905 | 1000 | |
74756182 | 1001 | spa_history_log_internal_ds(newds, "receive", tx, " "); |
03916905 PD |
1002 | } |
1003 | ||
1004 | static int | |
1005 | dmu_recv_resume_begin_check(void *arg, dmu_tx_t *tx) | |
1006 | { | |
1007 | dmu_recv_begin_arg_t *drba = arg; | |
30af21b0 | 1008 | dmu_recv_cookie_t *drc = drba->drba_cookie; |
03916905 | 1009 | dsl_pool_t *dp = dmu_tx_pool(tx); |
30af21b0 | 1010 | struct drr_begin *drrb = drc->drc_drrb; |
03916905 | 1011 | int error; |
40ab927a | 1012 | ds_hold_flags_t dsflags = DS_HOLD_FLAG_NONE; |
03916905 | 1013 | dsl_dataset_t *ds; |
30af21b0 | 1014 | const char *tofs = drc->drc_tofs; |
03916905 PD |
1015 | |
1016 | /* already checked */ | |
1017 | ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); | |
30af21b0 | 1018 | ASSERT(drc->drc_featureflags & DMU_BACKUP_FEATURE_RESUMING); |
03916905 PD |
1019 | |
1020 | if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == | |
1021 | DMU_COMPOUNDSTREAM || | |
1022 | drrb->drr_type >= DMU_OST_NUMTYPES) | |
1023 | return (SET_ERROR(EINVAL)); | |
1024 | ||
03916905 | 1025 | /* |
30af21b0 PD |
1026 | * This is mostly a sanity check since we should have already done these |
1027 | * checks during a previous attempt to receive the data. | |
03916905 | 1028 | */ |
30af21b0 PD |
1029 | error = recv_begin_check_feature_flags_impl(drc->drc_featureflags, |
1030 | dp->dp_spa); | |
1031 | if (error != 0) | |
1032 | return (error); | |
03916905 PD |
1033 | |
1034 | /* 6 extra bytes for /%recv */ | |
1035 | char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; | |
30af21b0 | 1036 | |
03916905 PD |
1037 | (void) snprintf(recvname, sizeof (recvname), "%s/%s", |
1038 | tofs, recv_clone_name); | |
1039 | ||
30af21b0 | 1040 | if (drc->drc_featureflags & DMU_BACKUP_FEATURE_RAW) { |
caf9dd20 BB |
1041 | /* raw receives require spill block allocation flag */ |
1042 | if (!(drrb->drr_flags & DRR_FLAG_SPILL_BLOCK)) | |
1043 | return (SET_ERROR(ZFS_ERR_SPILL_BLOCK_FLAG_MISSING)); | |
1044 | } else { | |
03916905 | 1045 | dsflags |= DS_HOLD_FLAG_DECRYPT; |
caf9dd20 | 1046 | } |
03916905 | 1047 | |
3ed9d688 | 1048 | boolean_t recvexist = B_TRUE; |
03916905 PD |
1049 | if (dsl_dataset_hold_flags(dp, recvname, dsflags, FTAG, &ds) != 0) { |
1050 | /* %recv does not exist; continue in tofs */ | |
3ed9d688 | 1051 | recvexist = B_FALSE; |
03916905 PD |
1052 | error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); |
1053 | if (error != 0) | |
1054 | return (error); | |
1055 | } | |
1056 | ||
3ed9d688 JP |
1057 | /* |
1058 | * Resume of full/newfs recv on existing dataset should be done with | |
1059 | * force flag | |
1060 | */ | |
1061 | if (recvexist && drrb->drr_fromguid == 0 && !drc->drc_force) { | |
1062 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1063 | return (SET_ERROR(ZFS_ERR_RESUME_EXISTS)); | |
1064 | } | |
1065 | ||
03916905 PD |
1066 | /* check that ds is marked inconsistent */ |
1067 | if (!DS_IS_INCONSISTENT(ds)) { | |
1068 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1069 | return (SET_ERROR(EINVAL)); | |
1070 | } | |
1071 | ||
1072 | /* check that there is resuming data, and that the toguid matches */ | |
1073 | if (!dsl_dataset_is_zapified(ds)) { | |
1074 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1075 | return (SET_ERROR(EINVAL)); | |
1076 | } | |
1077 | uint64_t val; | |
1078 | error = zap_lookup(dp->dp_meta_objset, ds->ds_object, | |
1079 | DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val); | |
1080 | if (error != 0 || drrb->drr_toguid != val) { | |
1081 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1082 | return (SET_ERROR(EINVAL)); | |
1083 | } | |
1084 | ||
1085 | /* | |
1086 | * Check if the receive is still running. If so, it will be owned. | |
1087 | * Note that nothing else can own the dataset (e.g. after the receive | |
1088 | * fails) because it will be marked inconsistent. | |
1089 | */ | |
1090 | if (dsl_dataset_has_owner(ds)) { | |
1091 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1092 | return (SET_ERROR(EBUSY)); | |
1093 | } | |
1094 | ||
1095 | /* There should not be any snapshots of this fs yet. */ | |
1096 | if (ds->ds_prev != NULL && ds->ds_prev->ds_dir == ds->ds_dir) { | |
1097 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1098 | return (SET_ERROR(EINVAL)); | |
1099 | } | |
1100 | ||
1101 | /* | |
1102 | * Note: resume point will be checked when we process the first WRITE | |
1103 | * record. | |
1104 | */ | |
1105 | ||
1106 | /* check that the origin matches */ | |
1107 | val = 0; | |
1108 | (void) zap_lookup(dp->dp_meta_objset, ds->ds_object, | |
1109 | DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val); | |
1110 | if (drrb->drr_fromguid != val) { | |
1111 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1112 | return (SET_ERROR(EINVAL)); | |
1113 | } | |
1114 | ||
aa646323 | 1115 | if (ds->ds_prev != NULL && drrb->drr_fromguid != 0) |
61152d10 TC |
1116 | drc->drc_fromsnapobj = ds->ds_prev->ds_object; |
1117 | ||
30af21b0 PD |
1118 | /* |
1119 | * If we're resuming, and the send is redacted, then the original send | |
1120 | * must have been redacted, and must have been redacted with respect to | |
1121 | * the same snapshots. | |
1122 | */ | |
1123 | if (drc->drc_featureflags & DMU_BACKUP_FEATURE_REDACTED) { | |
1124 | uint64_t num_ds_redact_snaps; | |
1125 | uint64_t *ds_redact_snaps; | |
1126 | ||
1127 | uint_t num_stream_redact_snaps; | |
1128 | uint64_t *stream_redact_snaps; | |
1129 | ||
1130 | if (nvlist_lookup_uint64_array(drc->drc_begin_nvl, | |
1131 | BEGINNV_REDACT_SNAPS, &stream_redact_snaps, | |
1132 | &num_stream_redact_snaps) != 0) { | |
1133 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1134 | return (SET_ERROR(EINVAL)); | |
1135 | } | |
1136 | ||
1137 | if (!dsl_dataset_get_uint64_array_feature(ds, | |
1138 | SPA_FEATURE_REDACTED_DATASETS, &num_ds_redact_snaps, | |
1139 | &ds_redact_snaps)) { | |
1140 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1141 | return (SET_ERROR(EINVAL)); | |
1142 | } | |
1143 | ||
1144 | for (int i = 0; i < num_ds_redact_snaps; i++) { | |
1145 | if (!redact_snaps_contains(ds_redact_snaps, | |
1146 | num_ds_redact_snaps, stream_redact_snaps[i])) { | |
1147 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1148 | return (SET_ERROR(EINVAL)); | |
1149 | } | |
1150 | } | |
1151 | } | |
7bcb7f08 MA |
1152 | |
1153 | error = recv_check_large_blocks(ds, drc->drc_featureflags); | |
1154 | if (error != 0) { | |
1155 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
1156 | return (error); | |
1157 | } | |
1158 | ||
03916905 PD |
1159 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
1160 | return (0); | |
1161 | } | |
1162 | ||
1163 | static void | |
1164 | dmu_recv_resume_begin_sync(void *arg, dmu_tx_t *tx) | |
1165 | { | |
1166 | dmu_recv_begin_arg_t *drba = arg; | |
1167 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1168 | const char *tofs = drba->drba_cookie->drc_tofs; | |
30af21b0 | 1169 | uint64_t featureflags = drba->drba_cookie->drc_featureflags; |
03916905 | 1170 | dsl_dataset_t *ds; |
40ab927a | 1171 | ds_hold_flags_t dsflags = DS_HOLD_FLAG_NONE; |
03916905 PD |
1172 | /* 6 extra bytes for /%recv */ |
1173 | char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; | |
1174 | ||
30af21b0 PD |
1175 | (void) snprintf(recvname, sizeof (recvname), "%s/%s", tofs, |
1176 | recv_clone_name); | |
03916905 PD |
1177 | |
1178 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
1179 | drba->drba_cookie->drc_raw = B_TRUE; | |
1180 | } else { | |
1181 | dsflags |= DS_HOLD_FLAG_DECRYPT; | |
1182 | } | |
1183 | ||
30af21b0 PD |
1184 | if (dsl_dataset_own_force(dp, recvname, dsflags, dmu_recv_tag, &ds) |
1185 | != 0) { | |
03916905 | 1186 | /* %recv does not exist; continue in tofs */ |
30af21b0 PD |
1187 | VERIFY0(dsl_dataset_own_force(dp, tofs, dsflags, dmu_recv_tag, |
1188 | &ds)); | |
03916905 PD |
1189 | drba->drba_cookie->drc_newfs = B_TRUE; |
1190 | } | |
1191 | ||
03916905 | 1192 | ASSERT(DS_IS_INCONSISTENT(ds)); |
03916905 PD |
1193 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
1194 | ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds)) || | |
1195 | drba->drba_cookie->drc_raw); | |
1196 | rrw_exit(&ds->ds_bp_rwlock, FTAG); | |
1197 | ||
1198 | drba->drba_cookie->drc_ds = ds; | |
0fdd6106 | 1199 | VERIFY0(dmu_objset_from_ds(ds, &drba->drba_cookie->drc_os)); |
61152d10 | 1200 | drba->drba_cookie->drc_should_save = B_TRUE; |
03916905 | 1201 | |
74756182 | 1202 | spa_history_log_internal_ds(ds, "resume receive", tx, " "); |
03916905 PD |
1203 | } |
1204 | ||
1205 | /* | |
1206 | * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() | |
1207 | * succeeds; otherwise we will leak the holds on the datasets. | |
1208 | */ | |
1209 | int | |
1210 | dmu_recv_begin(char *tofs, char *tosnap, dmu_replay_record_t *drr_begin, | |
e8cf3a4f | 1211 | boolean_t force, boolean_t heal, boolean_t resumable, nvlist_t *localprops, |
da92d5cb MM |
1212 | nvlist_t *hidden_args, char *origin, dmu_recv_cookie_t *drc, |
1213 | zfs_file_t *fp, offset_t *voffp) | |
03916905 PD |
1214 | { |
1215 | dmu_recv_begin_arg_t drba = { 0 }; | |
30af21b0 | 1216 | int err; |
03916905 | 1217 | |
861166b0 | 1218 | memset(drc, 0, sizeof (dmu_recv_cookie_t)); |
03916905 PD |
1219 | drc->drc_drr_begin = drr_begin; |
1220 | drc->drc_drrb = &drr_begin->drr_u.drr_begin; | |
1221 | drc->drc_tosnap = tosnap; | |
1222 | drc->drc_tofs = tofs; | |
1223 | drc->drc_force = force; | |
e8cf3a4f | 1224 | drc->drc_heal = heal; |
03916905 PD |
1225 | drc->drc_resumable = resumable; |
1226 | drc->drc_cred = CRED(); | |
e59a377a | 1227 | drc->drc_proc = curproc; |
03916905 PD |
1228 | drc->drc_clone = (origin != NULL); |
1229 | ||
1230 | if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { | |
1231 | drc->drc_byteswap = B_TRUE; | |
1232 | (void) fletcher_4_incremental_byteswap(drr_begin, | |
1233 | sizeof (dmu_replay_record_t), &drc->drc_cksum); | |
1234 | byteswap_record(drr_begin); | |
1235 | } else if (drc->drc_drrb->drr_magic == DMU_BACKUP_MAGIC) { | |
1236 | (void) fletcher_4_incremental_native(drr_begin, | |
1237 | sizeof (dmu_replay_record_t), &drc->drc_cksum); | |
1238 | } else { | |
1239 | return (SET_ERROR(EINVAL)); | |
1240 | } | |
1241 | ||
da92d5cb | 1242 | drc->drc_fp = fp; |
30af21b0 PD |
1243 | drc->drc_voff = *voffp; |
1244 | drc->drc_featureflags = | |
1245 | DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); | |
1246 | ||
1247 | uint32_t payloadlen = drc->drc_drr_begin->drr_payloadlen; | |
1248 | void *payload = NULL; | |
1249 | if (payloadlen != 0) | |
1250 | payload = kmem_alloc(payloadlen, KM_SLEEP); | |
1251 | ||
1252 | err = receive_read_payload_and_next_header(drc, payloadlen, | |
1253 | payload); | |
1254 | if (err != 0) { | |
1255 | kmem_free(payload, payloadlen); | |
1256 | return (err); | |
1257 | } | |
1258 | if (payloadlen != 0) { | |
1259 | err = nvlist_unpack(payload, payloadlen, &drc->drc_begin_nvl, | |
1260 | KM_SLEEP); | |
1261 | kmem_free(payload, payloadlen); | |
1262 | if (err != 0) { | |
1263 | kmem_free(drc->drc_next_rrd, | |
1264 | sizeof (*drc->drc_next_rrd)); | |
1265 | return (err); | |
1266 | } | |
1267 | } | |
1268 | ||
caf9dd20 BB |
1269 | if (drc->drc_drrb->drr_flags & DRR_FLAG_SPILL_BLOCK) |
1270 | drc->drc_spill = B_TRUE; | |
1271 | ||
03916905 PD |
1272 | drba.drba_origin = origin; |
1273 | drba.drba_cookie = drc; | |
1274 | drba.drba_cred = CRED(); | |
e59a377a | 1275 | drba.drba_proc = curproc; |
03916905 | 1276 | |
30af21b0 PD |
1277 | if (drc->drc_featureflags & DMU_BACKUP_FEATURE_RESUMING) { |
1278 | err = dsl_sync_task(tofs, | |
03916905 | 1279 | dmu_recv_resume_begin_check, dmu_recv_resume_begin_sync, |
30af21b0 PD |
1280 | &drba, 5, ZFS_SPACE_CHECK_NORMAL); |
1281 | } else { | |
03916905 PD |
1282 | /* |
1283 | * For non-raw, non-incremental, non-resuming receives the | |
1284 | * user can specify encryption parameters on the command line | |
1285 | * with "zfs recv -o". For these receives we create a dcp and | |
1286 | * pass it to the sync task. Creating the dcp will implicitly | |
1287 | * remove the encryption params from the localprops nvlist, | |
1288 | * which avoids errors when trying to set these normally | |
1289 | * read-only properties. Any other kind of receive that | |
1290 | * attempts to set these properties will fail as a result. | |
1291 | */ | |
1292 | if ((DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & | |
1293 | DMU_BACKUP_FEATURE_RAW) == 0 && | |
1294 | origin == NULL && drc->drc_drrb->drr_fromguid == 0) { | |
1295 | err = dsl_crypto_params_create_nvlist(DCP_CMD_NONE, | |
1296 | localprops, hidden_args, &drba.drba_dcp); | |
03916905 PD |
1297 | } |
1298 | ||
30af21b0 PD |
1299 | if (err == 0) { |
1300 | err = dsl_sync_task(tofs, | |
1301 | dmu_recv_begin_check, dmu_recv_begin_sync, | |
1302 | &drba, 5, ZFS_SPACE_CHECK_NORMAL); | |
1303 | dsl_crypto_params_free(drba.drba_dcp, !!err); | |
1304 | } | |
1305 | } | |
03916905 | 1306 | |
30af21b0 PD |
1307 | if (err != 0) { |
1308 | kmem_free(drc->drc_next_rrd, sizeof (*drc->drc_next_rrd)); | |
1309 | nvlist_free(drc->drc_begin_nvl); | |
03916905 | 1310 | } |
30af21b0 | 1311 | return (err); |
03916905 PD |
1312 | } |
1313 | ||
e8cf3a4f AP |
1314 | /* |
1315 | * Holds data need for corrective recv callback | |
1316 | */ | |
1317 | typedef struct cr_cb_data { | |
1318 | uint64_t size; | |
1319 | zbookmark_phys_t zb; | |
1320 | spa_t *spa; | |
1321 | } cr_cb_data_t; | |
1322 | ||
1323 | static void | |
1324 | corrective_read_done(zio_t *zio) | |
1325 | { | |
1326 | cr_cb_data_t *data = zio->io_private; | |
1327 | /* Corruption corrected; update error log if needed */ | |
1328 | if (zio->io_error == 0) | |
1329 | spa_remove_error(data->spa, &data->zb); | |
1330 | kmem_free(data, sizeof (cr_cb_data_t)); | |
1331 | abd_free(zio->io_abd); | |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * zio_rewrite the data pointed to by bp with the data from the rrd's abd. | |
1336 | */ | |
1337 | static int | |
1338 | do_corrective_recv(struct receive_writer_arg *rwa, struct drr_write *drrw, | |
1339 | struct receive_record_arg *rrd, blkptr_t *bp) | |
1340 | { | |
1341 | int err; | |
1342 | zio_t *io; | |
1343 | zbookmark_phys_t zb; | |
1344 | dnode_t *dn; | |
1345 | abd_t *abd = rrd->abd; | |
1346 | zio_cksum_t bp_cksum = bp->blk_cksum; | |
4938d01d | 1347 | zio_flag_t flags = ZIO_FLAG_SPECULATIVE | |
e8cf3a4f AP |
1348 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_RETRY | ZIO_FLAG_CANFAIL; |
1349 | ||
1350 | if (rwa->raw) | |
1351 | flags |= ZIO_FLAG_RAW; | |
1352 | ||
1353 | err = dnode_hold(rwa->os, drrw->drr_object, FTAG, &dn); | |
1354 | if (err != 0) | |
1355 | return (err); | |
1356 | SET_BOOKMARK(&zb, dmu_objset_id(rwa->os), drrw->drr_object, 0, | |
1357 | dbuf_whichblock(dn, 0, drrw->drr_offset)); | |
1358 | dnode_rele(dn, FTAG); | |
1359 | ||
1360 | if (!rwa->raw && DRR_WRITE_COMPRESSED(drrw)) { | |
1361 | /* Decompress the stream data */ | |
1362 | abd_t *dabd = abd_alloc_linear( | |
1363 | drrw->drr_logical_size, B_FALSE); | |
1364 | err = zio_decompress_data(drrw->drr_compressiontype, | |
1365 | abd, abd_to_buf(dabd), abd_get_size(abd), | |
1366 | abd_get_size(dabd), NULL); | |
1367 | ||
1368 | if (err != 0) { | |
1369 | abd_free(dabd); | |
1370 | return (err); | |
1371 | } | |
1372 | /* Swap in the newly decompressed data into the abd */ | |
1373 | abd_free(abd); | |
1374 | abd = dabd; | |
1375 | } | |
1376 | ||
1377 | if (!rwa->raw && BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF) { | |
1378 | /* Recompress the data */ | |
1379 | abd_t *cabd = abd_alloc_linear(BP_GET_PSIZE(bp), | |
1380 | B_FALSE); | |
1381 | uint64_t csize = zio_compress_data(BP_GET_COMPRESS(bp), | |
1382 | abd, abd_to_buf(cabd), abd_get_size(abd), | |
1383 | rwa->os->os_complevel); | |
1384 | abd_zero_off(cabd, csize, BP_GET_PSIZE(bp) - csize); | |
1385 | /* Swap in newly compressed data into the abd */ | |
1386 | abd_free(abd); | |
1387 | abd = cabd; | |
1388 | flags |= ZIO_FLAG_RAW_COMPRESS; | |
1389 | } | |
1390 | ||
1391 | /* | |
1392 | * The stream is not encrypted but the data on-disk is. | |
1393 | * We need to re-encrypt the buf using the same | |
1394 | * encryption type, salt, iv, and mac that was used to encrypt | |
1395 | * the block previosly. | |
1396 | */ | |
1397 | if (!rwa->raw && BP_USES_CRYPT(bp)) { | |
1398 | dsl_dataset_t *ds; | |
1399 | dsl_crypto_key_t *dck = NULL; | |
1400 | uint8_t salt[ZIO_DATA_SALT_LEN]; | |
1401 | uint8_t iv[ZIO_DATA_IV_LEN]; | |
1402 | uint8_t mac[ZIO_DATA_MAC_LEN]; | |
1403 | boolean_t no_crypt = B_FALSE; | |
1404 | dsl_pool_t *dp = dmu_objset_pool(rwa->os); | |
1405 | abd_t *eabd = abd_alloc_linear(BP_GET_PSIZE(bp), B_FALSE); | |
1406 | ||
1407 | zio_crypt_decode_params_bp(bp, salt, iv); | |
1408 | zio_crypt_decode_mac_bp(bp, mac); | |
1409 | ||
1410 | dsl_pool_config_enter(dp, FTAG); | |
1411 | err = dsl_dataset_hold_flags(dp, rwa->tofs, | |
1412 | DS_HOLD_FLAG_DECRYPT, FTAG, &ds); | |
1413 | if (err != 0) { | |
1414 | dsl_pool_config_exit(dp, FTAG); | |
1415 | abd_free(eabd); | |
1416 | return (SET_ERROR(EACCES)); | |
1417 | } | |
1418 | ||
1419 | /* Look up the key from the spa's keystore */ | |
1420 | err = spa_keystore_lookup_key(rwa->os->os_spa, | |
1421 | zb.zb_objset, FTAG, &dck); | |
1422 | if (err != 0) { | |
1423 | dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, | |
1424 | FTAG); | |
1425 | dsl_pool_config_exit(dp, FTAG); | |
1426 | abd_free(eabd); | |
1427 | return (SET_ERROR(EACCES)); | |
1428 | } | |
1429 | ||
1430 | err = zio_do_crypt_abd(B_TRUE, &dck->dck_key, | |
1431 | BP_GET_TYPE(bp), BP_SHOULD_BYTESWAP(bp), salt, iv, | |
1432 | mac, abd_get_size(abd), abd, eabd, &no_crypt); | |
1433 | ||
1434 | spa_keystore_dsl_key_rele(rwa->os->os_spa, dck, FTAG); | |
1435 | dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); | |
1436 | dsl_pool_config_exit(dp, FTAG); | |
1437 | ||
1438 | ASSERT0(no_crypt); | |
1439 | if (err != 0) { | |
1440 | abd_free(eabd); | |
1441 | return (err); | |
1442 | } | |
1443 | /* Swap in the newly encrypted data into the abd */ | |
1444 | abd_free(abd); | |
1445 | abd = eabd; | |
1446 | ||
1447 | /* | |
1448 | * We want to prevent zio_rewrite() from trying to | |
1449 | * encrypt the data again | |
1450 | */ | |
1451 | flags |= ZIO_FLAG_RAW_ENCRYPT; | |
1452 | } | |
1453 | rrd->abd = abd; | |
1454 | ||
1455 | io = zio_rewrite(NULL, rwa->os->os_spa, bp->blk_birth, bp, abd, | |
1456 | BP_GET_PSIZE(bp), NULL, NULL, ZIO_PRIORITY_SYNC_WRITE, flags, &zb); | |
1457 | ||
1458 | ASSERT(abd_get_size(abd) == BP_GET_LSIZE(bp) || | |
1459 | abd_get_size(abd) == BP_GET_PSIZE(bp)); | |
1460 | ||
1461 | /* compute new bp checksum value and make sure it matches the old one */ | |
1462 | zio_checksum_compute(io, BP_GET_CHECKSUM(bp), abd, abd_get_size(abd)); | |
1463 | if (!ZIO_CHECKSUM_EQUAL(bp_cksum, io->io_bp->blk_cksum)) { | |
1464 | zio_destroy(io); | |
1465 | if (zfs_recv_best_effort_corrective != 0) | |
1466 | return (0); | |
1467 | return (SET_ERROR(ECKSUM)); | |
1468 | } | |
1469 | ||
1470 | /* Correct the corruption in place */ | |
1471 | err = zio_wait(io); | |
1472 | if (err == 0) { | |
1473 | cr_cb_data_t *cb_data = | |
1474 | kmem_alloc(sizeof (cr_cb_data_t), KM_SLEEP); | |
1475 | cb_data->spa = rwa->os->os_spa; | |
1476 | cb_data->size = drrw->drr_logical_size; | |
1477 | cb_data->zb = zb; | |
1478 | /* Test if healing worked by re-reading the bp */ | |
1479 | err = zio_wait(zio_read(rwa->heal_pio, rwa->os->os_spa, bp, | |
1480 | abd_alloc_for_io(drrw->drr_logical_size, B_FALSE), | |
1481 | drrw->drr_logical_size, corrective_read_done, | |
1482 | cb_data, ZIO_PRIORITY_ASYNC_READ, flags, NULL)); | |
1483 | } | |
1484 | if (err != 0 && zfs_recv_best_effort_corrective != 0) | |
1485 | err = 0; | |
1486 | ||
1487 | return (err); | |
1488 | } | |
1489 | ||
03916905 | 1490 | static int |
30af21b0 | 1491 | receive_read(dmu_recv_cookie_t *drc, int len, void *buf) |
03916905 PD |
1492 | { |
1493 | int done = 0; | |
1494 | ||
1495 | /* | |
1496 | * The code doesn't rely on this (lengths being multiples of 8). See | |
1497 | * comment in dump_bytes. | |
1498 | */ | |
1499 | ASSERT(len % 8 == 0 || | |
30af21b0 | 1500 | (drc->drc_featureflags & DMU_BACKUP_FEATURE_RAW) != 0); |
03916905 PD |
1501 | |
1502 | while (done < len) { | |
1503 | ssize_t resid; | |
8b240f14 MA |
1504 | zfs_file_t *fp = drc->drc_fp; |
1505 | int err = zfs_file_read(fp, (char *)buf + done, | |
da92d5cb | 1506 | len - done, &resid); |
03916905 PD |
1507 | if (resid == len - done) { |
1508 | /* | |
7145123b PD |
1509 | * Note: ECKSUM or ZFS_ERR_STREAM_TRUNCATED indicates |
1510 | * that the receive was interrupted and can | |
1511 | * potentially be resumed. | |
03916905 | 1512 | */ |
8b240f14 | 1513 | err = SET_ERROR(ZFS_ERR_STREAM_TRUNCATED); |
03916905 | 1514 | } |
30af21b0 | 1515 | drc->drc_voff += len - done - resid; |
03916905 | 1516 | done = len - resid; |
8b240f14 MA |
1517 | if (err != 0) |
1518 | return (err); | |
03916905 PD |
1519 | } |
1520 | ||
30af21b0 | 1521 | drc->drc_bytes_read += len; |
03916905 PD |
1522 | |
1523 | ASSERT3U(done, ==, len); | |
1524 | return (0); | |
1525 | } | |
1526 | ||
03916905 PD |
1527 | static inline uint8_t |
1528 | deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) | |
1529 | { | |
1530 | if (bonus_type == DMU_OT_SA) { | |
1531 | return (1); | |
1532 | } else { | |
1533 | return (1 + | |
1534 | ((DN_OLD_MAX_BONUSLEN - | |
1535 | MIN(DN_OLD_MAX_BONUSLEN, bonus_size)) >> SPA_BLKPTRSHIFT)); | |
1536 | } | |
1537 | } | |
1538 | ||
1539 | static void | |
1540 | save_resume_state(struct receive_writer_arg *rwa, | |
1541 | uint64_t object, uint64_t offset, dmu_tx_t *tx) | |
1542 | { | |
1543 | int txgoff = dmu_tx_get_txg(tx) & TXG_MASK; | |
1544 | ||
1545 | if (!rwa->resumable) | |
1546 | return; | |
1547 | ||
1548 | /* | |
1549 | * We use ds_resume_bytes[] != 0 to indicate that we need to | |
1550 | * update this on disk, so it must not be 0. | |
1551 | */ | |
1552 | ASSERT(rwa->bytes_read != 0); | |
1553 | ||
1554 | /* | |
1555 | * We only resume from write records, which have a valid | |
1556 | * (non-meta-dnode) object number. | |
1557 | */ | |
1558 | ASSERT(object != 0); | |
1559 | ||
1560 | /* | |
1561 | * For resuming to work correctly, we must receive records in order, | |
1562 | * sorted by object,offset. This is checked by the callers, but | |
1563 | * assert it here for good measure. | |
1564 | */ | |
1565 | ASSERT3U(object, >=, rwa->os->os_dsl_dataset->ds_resume_object[txgoff]); | |
1566 | ASSERT(object != rwa->os->os_dsl_dataset->ds_resume_object[txgoff] || | |
1567 | offset >= rwa->os->os_dsl_dataset->ds_resume_offset[txgoff]); | |
1568 | ASSERT3U(rwa->bytes_read, >=, | |
1569 | rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff]); | |
1570 | ||
1571 | rwa->os->os_dsl_dataset->ds_resume_object[txgoff] = object; | |
1572 | rwa->os->os_dsl_dataset->ds_resume_offset[txgoff] = offset; | |
1573 | rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff] = rwa->bytes_read; | |
1574 | } | |
1575 | ||
7bcb7f08 MA |
1576 | static int |
1577 | receive_object_is_same_generation(objset_t *os, uint64_t object, | |
1578 | dmu_object_type_t old_bonus_type, dmu_object_type_t new_bonus_type, | |
1579 | const void *new_bonus, boolean_t *samegenp) | |
1580 | { | |
1581 | zfs_file_info_t zoi; | |
1582 | int err; | |
1583 | ||
1584 | dmu_buf_t *old_bonus_dbuf; | |
1585 | err = dmu_bonus_hold(os, object, FTAG, &old_bonus_dbuf); | |
1586 | if (err != 0) | |
1587 | return (err); | |
1588 | err = dmu_get_file_info(os, old_bonus_type, old_bonus_dbuf->db_data, | |
1589 | &zoi); | |
1590 | dmu_buf_rele(old_bonus_dbuf, FTAG); | |
1591 | if (err != 0) | |
1592 | return (err); | |
1593 | uint64_t old_gen = zoi.zfi_generation; | |
1594 | ||
1595 | err = dmu_get_file_info(os, new_bonus_type, new_bonus, &zoi); | |
1596 | if (err != 0) | |
1597 | return (err); | |
1598 | uint64_t new_gen = zoi.zfi_generation; | |
1599 | ||
1600 | *samegenp = (old_gen == new_gen); | |
1601 | return (0); | |
1602 | } | |
1603 | ||
1604 | static int | |
1605 | receive_handle_existing_object(const struct receive_writer_arg *rwa, | |
1606 | const struct drr_object *drro, const dmu_object_info_t *doi, | |
1607 | const void *bonus_data, | |
1608 | uint64_t *object_to_hold, uint32_t *new_blksz) | |
1609 | { | |
1610 | uint32_t indblksz = drro->drr_indblkshift ? | |
1611 | 1ULL << drro->drr_indblkshift : 0; | |
1612 | int nblkptr = deduce_nblkptr(drro->drr_bonustype, | |
1613 | drro->drr_bonuslen); | |
1614 | uint8_t dn_slots = drro->drr_dn_slots != 0 ? | |
1615 | drro->drr_dn_slots : DNODE_MIN_SLOTS; | |
1616 | boolean_t do_free_range = B_FALSE; | |
1617 | int err; | |
1618 | ||
1619 | *object_to_hold = drro->drr_object; | |
1620 | ||
1621 | /* nblkptr should be bounded by the bonus size and type */ | |
1622 | if (rwa->raw && nblkptr != drro->drr_nblkptr) | |
1623 | return (SET_ERROR(EINVAL)); | |
1624 | ||
1625 | /* | |
1626 | * After the previous send stream, the sending system may | |
1627 | * have freed this object, and then happened to re-allocate | |
1628 | * this object number in a later txg. In this case, we are | |
1629 | * receiving a different logical file, and the block size may | |
1630 | * appear to be different. i.e. we may have a different | |
1631 | * block size for this object than what the send stream says. | |
1632 | * In this case we need to remove the object's contents, | |
1633 | * so that its structure can be changed and then its contents | |
1634 | * entirely replaced by subsequent WRITE records. | |
1635 | * | |
1636 | * If this is a -L (--large-block) incremental stream, and | |
1637 | * the previous stream was not -L, the block size may appear | |
1638 | * to increase. i.e. we may have a smaller block size for | |
1639 | * this object than what the send stream says. In this case | |
1640 | * we need to keep the object's contents and block size | |
1641 | * intact, so that we don't lose parts of the object's | |
1642 | * contents that are not changed by this incremental send | |
1643 | * stream. | |
1644 | * | |
1645 | * We can distinguish between the two above cases by using | |
1646 | * the ZPL's generation number (see | |
1647 | * receive_object_is_same_generation()). However, we only | |
1648 | * want to rely on the generation number when absolutely | |
1649 | * necessary, because with raw receives, the generation is | |
1650 | * encrypted. We also want to minimize dependence on the | |
1651 | * ZPL, so that other types of datasets can also be received | |
1652 | * (e.g. ZVOLs, although note that ZVOLS currently do not | |
1653 | * reallocate their objects or change their structure). | |
1654 | * Therefore, we check a number of different cases where we | |
1655 | * know it is safe to discard the object's contents, before | |
1656 | * using the ZPL's generation number to make the above | |
1657 | * distinction. | |
1658 | */ | |
1659 | if (drro->drr_blksz != doi->doi_data_block_size) { | |
1660 | if (rwa->raw) { | |
1661 | /* | |
1662 | * RAW streams always have large blocks, so | |
1663 | * we are sure that the data is not needed | |
1664 | * due to changing --large-block to be on. | |
1665 | * Which is fortunate since the bonus buffer | |
1666 | * (which contains the ZPL generation) is | |
1667 | * encrypted, and the key might not be | |
1668 | * loaded. | |
1669 | */ | |
1670 | do_free_range = B_TRUE; | |
1671 | } else if (rwa->full) { | |
1672 | /* | |
1673 | * This is a full send stream, so it always | |
1674 | * replaces what we have. Even if the | |
1675 | * generation numbers happen to match, this | |
1676 | * can not actually be the same logical file. | |
1677 | * This is relevant when receiving a full | |
1678 | * send as a clone. | |
1679 | */ | |
1680 | do_free_range = B_TRUE; | |
1681 | } else if (drro->drr_type != | |
1682 | DMU_OT_PLAIN_FILE_CONTENTS || | |
1683 | doi->doi_type != DMU_OT_PLAIN_FILE_CONTENTS) { | |
1684 | /* | |
1685 | * PLAIN_FILE_CONTENTS are the only type of | |
1686 | * objects that have ever been stored with | |
1687 | * large blocks, so we don't need the special | |
1688 | * logic below. ZAP blocks can shrink (when | |
1689 | * there's only one block), so we don't want | |
1690 | * to hit the error below about block size | |
1691 | * only increasing. | |
1692 | */ | |
1693 | do_free_range = B_TRUE; | |
1694 | } else if (doi->doi_max_offset <= | |
1695 | doi->doi_data_block_size) { | |
1696 | /* | |
1697 | * There is only one block. We can free it, | |
1698 | * because its contents will be replaced by a | |
1699 | * WRITE record. This can not be the no-L -> | |
1700 | * -L case, because the no-L case would have | |
1701 | * resulted in multiple blocks. If we | |
1702 | * supported -L -> no-L, it would not be safe | |
1703 | * to free the file's contents. Fortunately, | |
1704 | * that is not allowed (see | |
1705 | * recv_check_large_blocks()). | |
1706 | */ | |
1707 | do_free_range = B_TRUE; | |
1708 | } else { | |
1709 | boolean_t is_same_gen; | |
1710 | err = receive_object_is_same_generation(rwa->os, | |
1711 | drro->drr_object, doi->doi_bonus_type, | |
1712 | drro->drr_bonustype, bonus_data, &is_same_gen); | |
1713 | if (err != 0) | |
1714 | return (SET_ERROR(EINVAL)); | |
1715 | ||
1716 | if (is_same_gen) { | |
1717 | /* | |
1718 | * This is the same logical file, and | |
1719 | * the block size must be increasing. | |
1720 | * It could only decrease if | |
1721 | * --large-block was changed to be | |
1722 | * off, which is checked in | |
1723 | * recv_check_large_blocks(). | |
1724 | */ | |
1725 | if (drro->drr_blksz <= | |
1726 | doi->doi_data_block_size) | |
1727 | return (SET_ERROR(EINVAL)); | |
1728 | /* | |
1729 | * We keep the existing blocksize and | |
1730 | * contents. | |
1731 | */ | |
1732 | *new_blksz = | |
1733 | doi->doi_data_block_size; | |
1734 | } else { | |
1735 | do_free_range = B_TRUE; | |
1736 | } | |
1737 | } | |
1738 | } | |
1739 | ||
1740 | /* nblkptr can only decrease if the object was reallocated */ | |
1741 | if (nblkptr < doi->doi_nblkptr) | |
1742 | do_free_range = B_TRUE; | |
1743 | ||
1744 | /* number of slots can only change on reallocation */ | |
1745 | if (dn_slots != doi->doi_dnodesize >> DNODE_SHIFT) | |
1746 | do_free_range = B_TRUE; | |
1747 | ||
1748 | /* | |
1749 | * For raw sends we also check a few other fields to | |
1750 | * ensure we are preserving the objset structure exactly | |
1751 | * as it was on the receive side: | |
1752 | * - A changed indirect block size | |
1753 | * - A smaller nlevels | |
1754 | */ | |
1755 | if (rwa->raw) { | |
1756 | if (indblksz != doi->doi_metadata_block_size) | |
1757 | do_free_range = B_TRUE; | |
1758 | if (drro->drr_nlevels < doi->doi_indirection) | |
1759 | do_free_range = B_TRUE; | |
1760 | } | |
1761 | ||
1762 | if (do_free_range) { | |
1763 | err = dmu_free_long_range(rwa->os, drro->drr_object, | |
1764 | 0, DMU_OBJECT_END); | |
1765 | if (err != 0) | |
1766 | return (SET_ERROR(EINVAL)); | |
1767 | } | |
1768 | ||
1769 | /* | |
1770 | * The dmu does not currently support decreasing nlevels | |
1771 | * or changing the number of dnode slots on an object. For | |
1772 | * non-raw sends, this does not matter and the new object | |
1773 | * can just use the previous one's nlevels. For raw sends, | |
1774 | * however, the structure of the received dnode (including | |
1775 | * nlevels and dnode slots) must match that of the send | |
1776 | * side. Therefore, instead of using dmu_object_reclaim(), | |
1777 | * we must free the object completely and call | |
1778 | * dmu_object_claim_dnsize() instead. | |
1779 | */ | |
1780 | if ((rwa->raw && drro->drr_nlevels < doi->doi_indirection) || | |
1781 | dn_slots != doi->doi_dnodesize >> DNODE_SHIFT) { | |
1782 | err = dmu_free_long_object(rwa->os, drro->drr_object); | |
1783 | if (err != 0) | |
1784 | return (SET_ERROR(EINVAL)); | |
1785 | ||
1786 | txg_wait_synced(dmu_objset_pool(rwa->os), 0); | |
1787 | *object_to_hold = DMU_NEW_OBJECT; | |
1788 | } | |
1789 | ||
1790 | /* | |
1791 | * For raw receives, free everything beyond the new incoming | |
1792 | * maxblkid. Normally this would be done with a DRR_FREE | |
1793 | * record that would come after this DRR_OBJECT record is | |
1794 | * processed. However, for raw receives we manually set the | |
1795 | * maxblkid from the drr_maxblkid and so we must first free | |
1796 | * everything above that blkid to ensure the DMU is always | |
1797 | * consistent with itself. We will never free the first block | |
1798 | * of the object here because a maxblkid of 0 could indicate | |
1799 | * an object with a single block or one with no blocks. This | |
1800 | * free may be skipped when dmu_free_long_range() was called | |
1801 | * above since it covers the entire object's contents. | |
1802 | */ | |
1803 | if (rwa->raw && *object_to_hold != DMU_NEW_OBJECT && !do_free_range) { | |
1804 | err = dmu_free_long_range(rwa->os, drro->drr_object, | |
1805 | (drro->drr_maxblkid + 1) * doi->doi_data_block_size, | |
1806 | DMU_OBJECT_END); | |
1807 | if (err != 0) | |
1808 | return (SET_ERROR(EINVAL)); | |
1809 | } | |
1810 | return (0); | |
1811 | } | |
1812 | ||
03916905 PD |
1813 | noinline static int |
1814 | receive_object(struct receive_writer_arg *rwa, struct drr_object *drro, | |
1815 | void *data) | |
1816 | { | |
1817 | dmu_object_info_t doi; | |
1818 | dmu_tx_t *tx; | |
03916905 | 1819 | int err; |
7bcb7f08 | 1820 | uint32_t new_blksz = drro->drr_blksz; |
03916905 PD |
1821 | uint8_t dn_slots = drro->drr_dn_slots != 0 ? |
1822 | drro->drr_dn_slots : DNODE_MIN_SLOTS; | |
1823 | ||
1824 | if (drro->drr_type == DMU_OT_NONE || | |
1825 | !DMU_OT_IS_VALID(drro->drr_type) || | |
1826 | !DMU_OT_IS_VALID(drro->drr_bonustype) || | |
1827 | drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || | |
1828 | drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || | |
1829 | P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || | |
1830 | drro->drr_blksz < SPA_MINBLOCKSIZE || | |
1831 | drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(rwa->os)) || | |
1832 | drro->drr_bonuslen > | |
1833 | DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa->os))) || | |
1834 | dn_slots > | |
30af21b0 | 1835 | (spa_maxdnodesize(dmu_objset_spa(rwa->os)) >> DNODE_SHIFT)) { |
03916905 PD |
1836 | return (SET_ERROR(EINVAL)); |
1837 | } | |
1838 | ||
1839 | if (rwa->raw) { | |
1840 | /* | |
1841 | * We should have received a DRR_OBJECT_RANGE record | |
1842 | * containing this block and stored it in rwa. | |
1843 | */ | |
1844 | if (drro->drr_object < rwa->or_firstobj || | |
1845 | drro->drr_object >= rwa->or_firstobj + rwa->or_numslots || | |
1846 | drro->drr_raw_bonuslen < drro->drr_bonuslen || | |
1847 | drro->drr_indblkshift > SPA_MAXBLOCKSHIFT || | |
1848 | drro->drr_nlevels > DN_MAX_LEVELS || | |
1849 | drro->drr_nblkptr > DN_MAX_NBLKPTR || | |
1850 | DN_SLOTS_TO_BONUSLEN(dn_slots) < | |
1851 | drro->drr_raw_bonuslen) | |
1852 | return (SET_ERROR(EINVAL)); | |
1853 | } else { | |
caf9dd20 BB |
1854 | /* |
1855 | * The DRR_OBJECT_SPILL flag is valid when the DRR_BEGIN | |
1856 | * record indicates this by setting DRR_FLAG_SPILL_BLOCK. | |
1857 | */ | |
1858 | if (((drro->drr_flags & ~(DRR_OBJECT_SPILL))) || | |
1859 | (!rwa->spill && DRR_OBJECT_HAS_SPILL(drro->drr_flags))) { | |
1860 | return (SET_ERROR(EINVAL)); | |
1861 | } | |
1862 | ||
1863 | if (drro->drr_raw_bonuslen != 0 || drro->drr_nblkptr != 0 || | |
1864 | drro->drr_indblkshift != 0 || drro->drr_nlevels != 0) { | |
03916905 | 1865 | return (SET_ERROR(EINVAL)); |
caf9dd20 | 1866 | } |
03916905 PD |
1867 | } |
1868 | ||
1869 | err = dmu_object_info(rwa->os, drro->drr_object, &doi); | |
30af21b0 | 1870 | |
03916905 PD |
1871 | if (err != 0 && err != ENOENT && err != EEXIST) |
1872 | return (SET_ERROR(EINVAL)); | |
1873 | ||
1874 | if (drro->drr_object > rwa->max_object) | |
1875 | rwa->max_object = drro->drr_object; | |
1876 | ||
1877 | /* | |
1878 | * If we are losing blkptrs or changing the block size this must | |
1879 | * be a new file instance. We must clear out the previous file | |
1880 | * contents before we can change this type of metadata in the dnode. | |
1881 | * Raw receives will also check that the indirect structure of the | |
1882 | * dnode hasn't changed. | |
1883 | */ | |
7bcb7f08 | 1884 | uint64_t object_to_hold; |
03916905 | 1885 | if (err == 0) { |
7bcb7f08 MA |
1886 | err = receive_handle_existing_object(rwa, drro, &doi, data, |
1887 | &object_to_hold, &new_blksz); | |
2a493a4c RY |
1888 | if (err != 0) |
1889 | return (err); | |
03916905 PD |
1890 | } else if (err == EEXIST) { |
1891 | /* | |
1892 | * The object requested is currently an interior slot of a | |
1893 | * multi-slot dnode. This will be resolved when the next txg | |
1894 | * is synced out, since the send stream will have told us | |
1895 | * to free this slot when we freed the associated dnode | |
1896 | * earlier in the stream. | |
1897 | */ | |
1898 | txg_wait_synced(dmu_objset_pool(rwa->os), 0); | |
b92f5d9f BB |
1899 | |
1900 | if (dmu_object_info(rwa->os, drro->drr_object, NULL) != ENOENT) | |
1901 | return (SET_ERROR(EINVAL)); | |
1902 | ||
1903 | /* object was freed and we are about to allocate a new one */ | |
7bcb7f08 | 1904 | object_to_hold = DMU_NEW_OBJECT; |
03916905 PD |
1905 | } else { |
1906 | /* object is free and we are about to allocate a new one */ | |
7bcb7f08 | 1907 | object_to_hold = DMU_NEW_OBJECT; |
03916905 PD |
1908 | } |
1909 | ||
1910 | /* | |
1911 | * If this is a multi-slot dnode there is a chance that this | |
1912 | * object will expand into a slot that is already used by | |
1913 | * another object from the previous snapshot. We must free | |
1914 | * these objects before we attempt to allocate the new dnode. | |
1915 | */ | |
1916 | if (dn_slots > 1) { | |
1917 | boolean_t need_sync = B_FALSE; | |
1918 | ||
1919 | for (uint64_t slot = drro->drr_object + 1; | |
1920 | slot < drro->drr_object + dn_slots; | |
1921 | slot++) { | |
1922 | dmu_object_info_t slot_doi; | |
1923 | ||
1924 | err = dmu_object_info(rwa->os, slot, &slot_doi); | |
1925 | if (err == ENOENT || err == EEXIST) | |
1926 | continue; | |
1927 | else if (err != 0) | |
1928 | return (err); | |
1929 | ||
1930 | err = dmu_free_long_object(rwa->os, slot); | |
03916905 PD |
1931 | if (err != 0) |
1932 | return (err); | |
1933 | ||
1934 | need_sync = B_TRUE; | |
1935 | } | |
1936 | ||
1937 | if (need_sync) | |
1938 | txg_wait_synced(dmu_objset_pool(rwa->os), 0); | |
1939 | } | |
1940 | ||
1941 | tx = dmu_tx_create(rwa->os); | |
7bcb7f08 MA |
1942 | dmu_tx_hold_bonus(tx, object_to_hold); |
1943 | dmu_tx_hold_write(tx, object_to_hold, 0, 0); | |
03916905 PD |
1944 | err = dmu_tx_assign(tx, TXG_WAIT); |
1945 | if (err != 0) { | |
1946 | dmu_tx_abort(tx); | |
1947 | return (err); | |
1948 | } | |
1949 | ||
7bcb7f08 | 1950 | if (object_to_hold == DMU_NEW_OBJECT) { |
caf9dd20 | 1951 | /* Currently free, wants to be allocated */ |
03916905 | 1952 | err = dmu_object_claim_dnsize(rwa->os, drro->drr_object, |
7bcb7f08 | 1953 | drro->drr_type, new_blksz, |
03916905 PD |
1954 | drro->drr_bonustype, drro->drr_bonuslen, |
1955 | dn_slots << DNODE_SHIFT, tx); | |
1956 | } else if (drro->drr_type != doi.doi_type || | |
7bcb7f08 | 1957 | new_blksz != doi.doi_data_block_size || |
03916905 PD |
1958 | drro->drr_bonustype != doi.doi_bonus_type || |
1959 | drro->drr_bonuslen != doi.doi_bonus_size) { | |
caf9dd20 | 1960 | /* Currently allocated, but with different properties */ |
03916905 | 1961 | err = dmu_object_reclaim_dnsize(rwa->os, drro->drr_object, |
7bcb7f08 | 1962 | drro->drr_type, new_blksz, |
03916905 | 1963 | drro->drr_bonustype, drro->drr_bonuslen, |
caf9dd20 BB |
1964 | dn_slots << DNODE_SHIFT, rwa->spill ? |
1965 | DRR_OBJECT_HAS_SPILL(drro->drr_flags) : B_FALSE, tx); | |
1966 | } else if (rwa->spill && !DRR_OBJECT_HAS_SPILL(drro->drr_flags)) { | |
1967 | /* | |
1968 | * Currently allocated, the existing version of this object | |
1969 | * may reference a spill block that is no longer allocated | |
1970 | * at the source and needs to be freed. | |
1971 | */ | |
1972 | err = dmu_object_rm_spill(rwa->os, drro->drr_object, tx); | |
03916905 | 1973 | } |
3fa93bb8 | 1974 | |
03916905 PD |
1975 | if (err != 0) { |
1976 | dmu_tx_commit(tx); | |
1977 | return (SET_ERROR(EINVAL)); | |
1978 | } | |
1979 | ||
1980 | if (rwa->or_crypt_params_present) { | |
1981 | /* | |
1982 | * Set the crypt params for the buffer associated with this | |
1983 | * range of dnodes. This causes the blkptr_t to have the | |
1984 | * same crypt params (byteorder, salt, iv, mac) as on the | |
1985 | * sending side. | |
1986 | * | |
1987 | * Since we are committing this tx now, it is possible for | |
1988 | * the dnode block to end up on-disk with the incorrect MAC, | |
1989 | * if subsequent objects in this block are received in a | |
1990 | * different txg. However, since the dataset is marked as | |
1991 | * inconsistent, no code paths will do a non-raw read (or | |
1992 | * decrypt the block / verify the MAC). The receive code and | |
1993 | * scrub code can safely do raw reads and verify the | |
1994 | * checksum. They don't need to verify the MAC. | |
1995 | */ | |
1996 | dmu_buf_t *db = NULL; | |
1997 | uint64_t offset = rwa->or_firstobj * DNODE_MIN_SIZE; | |
1998 | ||
1999 | err = dmu_buf_hold_by_dnode(DMU_META_DNODE(rwa->os), | |
2000 | offset, FTAG, &db, DMU_READ_PREFETCH | DMU_READ_NO_DECRYPT); | |
2001 | if (err != 0) { | |
2002 | dmu_tx_commit(tx); | |
2003 | return (SET_ERROR(EINVAL)); | |
2004 | } | |
2005 | ||
2006 | dmu_buf_set_crypt_params(db, rwa->or_byteorder, | |
2007 | rwa->or_salt, rwa->or_iv, rwa->or_mac, tx); | |
2008 | ||
2009 | dmu_buf_rele(db, FTAG); | |
2010 | ||
2011 | rwa->or_crypt_params_present = B_FALSE; | |
2012 | } | |
2013 | ||
2014 | dmu_object_set_checksum(rwa->os, drro->drr_object, | |
2015 | drro->drr_checksumtype, tx); | |
2016 | dmu_object_set_compress(rwa->os, drro->drr_object, | |
2017 | drro->drr_compress, tx); | |
2018 | ||
2019 | /* handle more restrictive dnode structuring for raw recvs */ | |
2020 | if (rwa->raw) { | |
2021 | /* | |
369aa501 TC |
2022 | * Set the indirect block size, block shift, nlevels. |
2023 | * This will not fail because we ensured all of the | |
2024 | * blocks were freed earlier if this is a new object. | |
2025 | * For non-new objects block size and indirect block | |
2026 | * shift cannot change and nlevels can only increase. | |
03916905 | 2027 | */ |
7bcb7f08 | 2028 | ASSERT3U(new_blksz, ==, drro->drr_blksz); |
03916905 PD |
2029 | VERIFY0(dmu_object_set_blocksize(rwa->os, drro->drr_object, |
2030 | drro->drr_blksz, drro->drr_indblkshift, tx)); | |
2031 | VERIFY0(dmu_object_set_nlevels(rwa->os, drro->drr_object, | |
2032 | drro->drr_nlevels, tx)); | |
369aa501 TC |
2033 | |
2034 | /* | |
c2c6eadf TC |
2035 | * Set the maxblkid. This will always succeed because |
2036 | * we freed all blocks beyond the new maxblkid above. | |
369aa501 | 2037 | */ |
03916905 PD |
2038 | VERIFY0(dmu_object_set_maxblkid(rwa->os, drro->drr_object, |
2039 | drro->drr_maxblkid, tx)); | |
2040 | } | |
2041 | ||
2042 | if (data != NULL) { | |
2043 | dmu_buf_t *db; | |
6955b401 | 2044 | dnode_t *dn; |
03916905 PD |
2045 | uint32_t flags = DMU_READ_NO_PREFETCH; |
2046 | ||
2047 | if (rwa->raw) | |
2048 | flags |= DMU_READ_NO_DECRYPT; | |
2049 | ||
6955b401 BB |
2050 | VERIFY0(dnode_hold(rwa->os, drro->drr_object, FTAG, &dn)); |
2051 | VERIFY0(dmu_bonus_hold_by_dnode(dn, FTAG, &db, flags)); | |
2052 | ||
03916905 PD |
2053 | dmu_buf_will_dirty(db, tx); |
2054 | ||
2055 | ASSERT3U(db->db_size, >=, drro->drr_bonuslen); | |
861166b0 | 2056 | memcpy(db->db_data, data, DRR_OBJECT_PAYLOAD_SIZE(drro)); |
03916905 PD |
2057 | |
2058 | /* | |
2059 | * Raw bonus buffers have their byteorder determined by the | |
2060 | * DRR_OBJECT_RANGE record. | |
2061 | */ | |
2062 | if (rwa->byteswap && !rwa->raw) { | |
2063 | dmu_object_byteswap_t byteswap = | |
2064 | DMU_OT_BYTESWAP(drro->drr_bonustype); | |
2065 | dmu_ot_byteswap[byteswap].ob_func(db->db_data, | |
2066 | DRR_OBJECT_PAYLOAD_SIZE(drro)); | |
2067 | } | |
2068 | dmu_buf_rele(db, FTAG); | |
6955b401 | 2069 | dnode_rele(dn, FTAG); |
03916905 PD |
2070 | } |
2071 | dmu_tx_commit(tx); | |
2072 | ||
2073 | return (0); | |
2074 | } | |
2075 | ||
03916905 PD |
2076 | noinline static int |
2077 | receive_freeobjects(struct receive_writer_arg *rwa, | |
2078 | struct drr_freeobjects *drrfo) | |
2079 | { | |
2080 | uint64_t obj; | |
2081 | int next_err = 0; | |
2082 | ||
2083 | if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) | |
2084 | return (SET_ERROR(EINVAL)); | |
2085 | ||
2086 | for (obj = drrfo->drr_firstobj == 0 ? 1 : drrfo->drr_firstobj; | |
30af21b0 PD |
2087 | obj < drrfo->drr_firstobj + drrfo->drr_numobjs && |
2088 | obj < DN_MAX_OBJECT && next_err == 0; | |
03916905 PD |
2089 | next_err = dmu_object_next(rwa->os, &obj, FALSE, 0)) { |
2090 | dmu_object_info_t doi; | |
2091 | int err; | |
2092 | ||
2093 | err = dmu_object_info(rwa->os, obj, &doi); | |
2094 | if (err == ENOENT) | |
2095 | continue; | |
2096 | else if (err != 0) | |
2097 | return (err); | |
2098 | ||
2099 | err = dmu_free_long_object(rwa->os, obj); | |
2100 | ||
2101 | if (err != 0) | |
2102 | return (err); | |
03916905 PD |
2103 | } |
2104 | if (next_err != ESRCH) | |
2105 | return (next_err); | |
2106 | return (0); | |
2107 | } | |
2108 | ||
7261fc2e MA |
2109 | /* |
2110 | * Note: if this fails, the caller will clean up any records left on the | |
2111 | * rwa->write_batch list. | |
2112 | */ | |
2113 | static int | |
2114 | flush_write_batch_impl(struct receive_writer_arg *rwa) | |
03916905 | 2115 | { |
03916905 | 2116 | dnode_t *dn; |
7261fc2e MA |
2117 | int err; |
2118 | ||
2119 | if (dnode_hold(rwa->os, rwa->last_object, FTAG, &dn) != 0) | |
2120 | return (SET_ERROR(EINVAL)); | |
2121 | ||
2122 | struct receive_record_arg *last_rrd = list_tail(&rwa->write_batch); | |
2123 | struct drr_write *last_drrw = &last_rrd->header.drr_u.drr_write; | |
2124 | ||
2125 | struct receive_record_arg *first_rrd = list_head(&rwa->write_batch); | |
2126 | struct drr_write *first_drrw = &first_rrd->header.drr_u.drr_write; | |
2127 | ||
2128 | ASSERT3U(rwa->last_object, ==, last_drrw->drr_object); | |
2129 | ASSERT3U(rwa->last_offset, ==, last_drrw->drr_offset); | |
2130 | ||
2131 | dmu_tx_t *tx = dmu_tx_create(rwa->os); | |
2132 | dmu_tx_hold_write_by_dnode(tx, dn, first_drrw->drr_offset, | |
2133 | last_drrw->drr_offset - first_drrw->drr_offset + | |
2134 | last_drrw->drr_logical_size); | |
2135 | err = dmu_tx_assign(tx, TXG_WAIT); | |
2136 | if (err != 0) { | |
2137 | dmu_tx_abort(tx); | |
2138 | dnode_rele(dn, FTAG); | |
2139 | return (err); | |
2140 | } | |
2141 | ||
2142 | struct receive_record_arg *rrd; | |
2143 | while ((rrd = list_head(&rwa->write_batch)) != NULL) { | |
2144 | struct drr_write *drrw = &rrd->header.drr_u.drr_write; | |
ba67d821 | 2145 | abd_t *abd = rrd->abd; |
7261fc2e MA |
2146 | |
2147 | ASSERT3U(drrw->drr_object, ==, rwa->last_object); | |
2148 | ||
ba67d821 MA |
2149 | if (drrw->drr_logical_size != dn->dn_datablksz) { |
2150 | /* | |
2151 | * The WRITE record is larger than the object's block | |
2152 | * size. We must be receiving an incremental | |
2153 | * large-block stream into a dataset that previously did | |
2154 | * a non-large-block receive. Lightweight writes must | |
2155 | * be exactly one block, so we need to decompress the | |
2156 | * data (if compressed) and do a normal dmu_write(). | |
2157 | */ | |
7bcb7f08 | 2158 | ASSERT3U(drrw->drr_logical_size, >, dn->dn_datablksz); |
ba67d821 MA |
2159 | if (DRR_WRITE_COMPRESSED(drrw)) { |
2160 | abd_t *decomp_abd = | |
2161 | abd_alloc_linear(drrw->drr_logical_size, | |
2162 | B_FALSE); | |
2163 | ||
2164 | err = zio_decompress_data( | |
2165 | drrw->drr_compressiontype, | |
2166 | abd, abd_to_buf(decomp_abd), | |
2167 | abd_get_size(abd), | |
2168 | abd_get_size(decomp_abd), NULL); | |
2169 | ||
2170 | if (err == 0) { | |
2171 | dmu_write_by_dnode(dn, | |
2172 | drrw->drr_offset, | |
2173 | drrw->drr_logical_size, | |
2174 | abd_to_buf(decomp_abd), tx); | |
2175 | } | |
2176 | abd_free(decomp_abd); | |
2177 | } else { | |
2178 | dmu_write_by_dnode(dn, | |
2179 | drrw->drr_offset, | |
2180 | drrw->drr_logical_size, | |
2181 | abd_to_buf(abd), tx); | |
2182 | } | |
2183 | if (err == 0) | |
2184 | abd_free(abd); | |
2185 | } else { | |
2186 | zio_prop_t zp; | |
2187 | dmu_write_policy(rwa->os, dn, 0, 0, &zp); | |
2188 | ||
4938d01d | 2189 | zio_flag_t zio_flags = 0; |
ba67d821 MA |
2190 | |
2191 | if (rwa->raw) { | |
2192 | zp.zp_encrypt = B_TRUE; | |
2193 | zp.zp_compress = drrw->drr_compressiontype; | |
2194 | zp.zp_byteorder = ZFS_HOST_BYTEORDER ^ | |
2195 | !!DRR_IS_RAW_BYTESWAPPED(drrw->drr_flags) ^ | |
2196 | rwa->byteswap; | |
861166b0 | 2197 | memcpy(zp.zp_salt, drrw->drr_salt, |
ba67d821 | 2198 | ZIO_DATA_SALT_LEN); |
861166b0 | 2199 | memcpy(zp.zp_iv, drrw->drr_iv, |
ba67d821 | 2200 | ZIO_DATA_IV_LEN); |
861166b0 | 2201 | memcpy(zp.zp_mac, drrw->drr_mac, |
ba67d821 MA |
2202 | ZIO_DATA_MAC_LEN); |
2203 | if (DMU_OT_IS_ENCRYPTED(zp.zp_type)) { | |
2204 | zp.zp_nopwrite = B_FALSE; | |
2205 | zp.zp_copies = MIN(zp.zp_copies, | |
2206 | SPA_DVAS_PER_BP - 1); | |
2207 | } | |
2208 | zio_flags |= ZIO_FLAG_RAW; | |
2209 | } else if (DRR_WRITE_COMPRESSED(drrw)) { | |
2210 | ASSERT3U(drrw->drr_compressed_size, >, 0); | |
2211 | ASSERT3U(drrw->drr_logical_size, >=, | |
2212 | drrw->drr_compressed_size); | |
2213 | zp.zp_compress = drrw->drr_compressiontype; | |
2214 | zio_flags |= ZIO_FLAG_RAW_COMPRESS; | |
2215 | } else if (rwa->byteswap) { | |
2216 | /* | |
2217 | * Note: compressed blocks never need to be | |
2218 | * byteswapped, because WRITE records for | |
2219 | * metadata blocks are never compressed. The | |
2220 | * exception is raw streams, which are written | |
2221 | * in the original byteorder, and the byteorder | |
2222 | * bit is preserved in the BP by setting | |
2223 | * zp_byteorder above. | |
2224 | */ | |
2225 | dmu_object_byteswap_t byteswap = | |
2226 | DMU_OT_BYTESWAP(drrw->drr_type); | |
2227 | dmu_ot_byteswap[byteswap].ob_func( | |
2228 | abd_to_buf(abd), | |
2229 | DRR_WRITE_PAYLOAD_SIZE(drrw)); | |
2230 | } | |
7bcb7f08 MA |
2231 | |
2232 | /* | |
ba67d821 MA |
2233 | * Since this data can't be read until the receive |
2234 | * completes, we can do a "lightweight" write for | |
2235 | * improved performance. | |
7bcb7f08 | 2236 | */ |
ba67d821 MA |
2237 | err = dmu_lightweight_write_by_dnode(dn, |
2238 | drrw->drr_offset, abd, &zp, zio_flags, tx); | |
7bcb7f08 MA |
2239 | } |
2240 | ||
7261fc2e MA |
2241 | if (err != 0) { |
2242 | /* | |
2243 | * This rrd is left on the list, so the caller will | |
ba67d821 | 2244 | * free it (and the abd). |
7261fc2e MA |
2245 | */ |
2246 | break; | |
2247 | } | |
2248 | ||
2249 | /* | |
2250 | * Note: If the receive fails, we want the resume stream to | |
2251 | * start with the same record that we last successfully | |
2252 | * received (as opposed to the next record), so that we can | |
2253 | * verify that we are resuming from the correct location. | |
2254 | */ | |
2255 | save_resume_state(rwa, drrw->drr_object, drrw->drr_offset, tx); | |
2256 | ||
2257 | list_remove(&rwa->write_batch, rrd); | |
2258 | kmem_free(rrd, sizeof (*rrd)); | |
2259 | } | |
2260 | ||
2261 | dmu_tx_commit(tx); | |
2262 | dnode_rele(dn, FTAG); | |
2263 | return (err); | |
2264 | } | |
2265 | ||
2266 | noinline static int | |
2267 | flush_write_batch(struct receive_writer_arg *rwa) | |
2268 | { | |
2269 | if (list_is_empty(&rwa->write_batch)) | |
2270 | return (0); | |
2271 | int err = rwa->err; | |
2272 | if (err == 0) | |
2273 | err = flush_write_batch_impl(rwa); | |
2274 | if (err != 0) { | |
2275 | struct receive_record_arg *rrd; | |
2276 | while ((rrd = list_remove_head(&rwa->write_batch)) != NULL) { | |
ba67d821 | 2277 | abd_free(rrd->abd); |
7261fc2e MA |
2278 | kmem_free(rrd, sizeof (*rrd)); |
2279 | } | |
2280 | } | |
2281 | ASSERT(list_is_empty(&rwa->write_batch)); | |
2282 | return (err); | |
2283 | } | |
2284 | ||
2285 | noinline static int | |
2286 | receive_process_write_record(struct receive_writer_arg *rwa, | |
2287 | struct receive_record_arg *rrd) | |
2288 | { | |
2289 | int err = 0; | |
2290 | ||
2291 | ASSERT3U(rrd->header.drr_type, ==, DRR_WRITE); | |
2292 | struct drr_write *drrw = &rrd->header.drr_u.drr_write; | |
03916905 PD |
2293 | |
2294 | if (drrw->drr_offset + drrw->drr_logical_size < drrw->drr_offset || | |
2295 | !DMU_OT_IS_VALID(drrw->drr_type)) | |
2296 | return (SET_ERROR(EINVAL)); | |
2297 | ||
e8cf3a4f AP |
2298 | if (rwa->heal) { |
2299 | blkptr_t *bp; | |
2300 | dmu_buf_t *dbp; | |
2301 | dnode_t *dn; | |
2302 | int flags = DB_RF_CANFAIL; | |
2303 | ||
2304 | if (rwa->raw) | |
2305 | flags |= DB_RF_NO_DECRYPT; | |
2306 | ||
2307 | if (rwa->byteswap) { | |
2308 | dmu_object_byteswap_t byteswap = | |
2309 | DMU_OT_BYTESWAP(drrw->drr_type); | |
2310 | dmu_ot_byteswap[byteswap].ob_func(abd_to_buf(rrd->abd), | |
2311 | DRR_WRITE_PAYLOAD_SIZE(drrw)); | |
2312 | } | |
2313 | ||
2314 | err = dmu_buf_hold_noread(rwa->os, drrw->drr_object, | |
2315 | drrw->drr_offset, FTAG, &dbp); | |
2316 | if (err != 0) | |
2317 | return (err); | |
2318 | ||
2319 | /* Try to read the object to see if it needs healing */ | |
2320 | err = dbuf_read((dmu_buf_impl_t *)dbp, NULL, flags); | |
2321 | /* | |
2322 | * We only try to heal when dbuf_read() returns a ECKSUMs. | |
2323 | * Other errors (even EIO) get returned to caller. | |
2324 | * EIO indicates that the device is not present/accessible, | |
2325 | * so writing to it will likely fail. | |
2326 | * If the block is healthy, we don't want to overwrite it | |
2327 | * unnecessarily. | |
2328 | */ | |
2329 | if (err != ECKSUM) { | |
2330 | dmu_buf_rele(dbp, FTAG); | |
2331 | return (err); | |
2332 | } | |
2333 | dn = dmu_buf_dnode_enter(dbp); | |
2334 | /* Make sure the on-disk block and recv record sizes match */ | |
2335 | if (drrw->drr_logical_size != | |
2336 | dn->dn_datablkszsec << SPA_MINBLOCKSHIFT) { | |
2337 | err = ENOTSUP; | |
2338 | dmu_buf_dnode_exit(dbp); | |
2339 | dmu_buf_rele(dbp, FTAG); | |
2340 | return (err); | |
2341 | } | |
2342 | /* Get the block pointer for the corrupted block */ | |
2343 | bp = dmu_buf_get_blkptr(dbp); | |
2344 | err = do_corrective_recv(rwa, drrw, rrd, bp); | |
2345 | dmu_buf_dnode_exit(dbp); | |
2346 | dmu_buf_rele(dbp, FTAG); | |
2347 | return (err); | |
2348 | } | |
2349 | ||
03916905 PD |
2350 | /* |
2351 | * For resuming to work, records must be in increasing order | |
2352 | * by (object, offset). | |
2353 | */ | |
2354 | if (drrw->drr_object < rwa->last_object || | |
2355 | (drrw->drr_object == rwa->last_object && | |
2356 | drrw->drr_offset < rwa->last_offset)) { | |
2357 | return (SET_ERROR(EINVAL)); | |
2358 | } | |
7261fc2e MA |
2359 | |
2360 | struct receive_record_arg *first_rrd = list_head(&rwa->write_batch); | |
2361 | struct drr_write *first_drrw = &first_rrd->header.drr_u.drr_write; | |
2362 | uint64_t batch_size = | |
2363 | MIN(zfs_recv_write_batch_size, DMU_MAX_ACCESS / 2); | |
2364 | if (first_rrd != NULL && | |
2365 | (drrw->drr_object != first_drrw->drr_object || | |
2366 | drrw->drr_offset >= first_drrw->drr_offset + batch_size)) { | |
2367 | err = flush_write_batch(rwa); | |
2368 | if (err != 0) | |
2369 | return (err); | |
2370 | } | |
2371 | ||
03916905 PD |
2372 | rwa->last_object = drrw->drr_object; |
2373 | rwa->last_offset = drrw->drr_offset; | |
2374 | ||
2375 | if (rwa->last_object > rwa->max_object) | |
2376 | rwa->max_object = rwa->last_object; | |
2377 | ||
7261fc2e | 2378 | list_insert_tail(&rwa->write_batch, rrd); |
03916905 | 2379 | /* |
7261fc2e MA |
2380 | * Return EAGAIN to indicate that we will use this rrd again, |
2381 | * so the caller should not free it | |
03916905 | 2382 | */ |
7261fc2e | 2383 | return (EAGAIN); |
03916905 PD |
2384 | } |
2385 | ||
03916905 PD |
2386 | static int |
2387 | receive_write_embedded(struct receive_writer_arg *rwa, | |
2388 | struct drr_write_embedded *drrwe, void *data) | |
2389 | { | |
2390 | dmu_tx_t *tx; | |
2391 | int err; | |
2392 | ||
2393 | if (drrwe->drr_offset + drrwe->drr_length < drrwe->drr_offset) | |
2394 | return (SET_ERROR(EINVAL)); | |
2395 | ||
2396 | if (drrwe->drr_psize > BPE_PAYLOAD_SIZE) | |
2397 | return (SET_ERROR(EINVAL)); | |
2398 | ||
2399 | if (drrwe->drr_etype >= NUM_BP_EMBEDDED_TYPES) | |
2400 | return (SET_ERROR(EINVAL)); | |
2401 | if (drrwe->drr_compression >= ZIO_COMPRESS_FUNCTIONS) | |
2402 | return (SET_ERROR(EINVAL)); | |
2403 | if (rwa->raw) | |
2404 | return (SET_ERROR(EINVAL)); | |
2405 | ||
2406 | if (drrwe->drr_object > rwa->max_object) | |
2407 | rwa->max_object = drrwe->drr_object; | |
2408 | ||
2409 | tx = dmu_tx_create(rwa->os); | |
2410 | ||
2411 | dmu_tx_hold_write(tx, drrwe->drr_object, | |
2412 | drrwe->drr_offset, drrwe->drr_length); | |
2413 | err = dmu_tx_assign(tx, TXG_WAIT); | |
2414 | if (err != 0) { | |
2415 | dmu_tx_abort(tx); | |
2416 | return (err); | |
2417 | } | |
2418 | ||
2419 | dmu_write_embedded(rwa->os, drrwe->drr_object, | |
2420 | drrwe->drr_offset, data, drrwe->drr_etype, | |
2421 | drrwe->drr_compression, drrwe->drr_lsize, drrwe->drr_psize, | |
2422 | rwa->byteswap ^ ZFS_HOST_BYTEORDER, tx); | |
2423 | ||
2424 | /* See comment in restore_write. */ | |
2425 | save_resume_state(rwa, drrwe->drr_object, drrwe->drr_offset, tx); | |
2426 | dmu_tx_commit(tx); | |
2427 | return (0); | |
2428 | } | |
2429 | ||
2430 | static int | |
2431 | receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs, | |
ba67d821 | 2432 | abd_t *abd) |
03916905 | 2433 | { |
03916905 PD |
2434 | dmu_buf_t *db, *db_spill; |
2435 | int err; | |
03916905 PD |
2436 | |
2437 | if (drrs->drr_length < SPA_MINBLOCKSIZE || | |
2438 | drrs->drr_length > spa_maxblocksize(dmu_objset_spa(rwa->os))) | |
2439 | return (SET_ERROR(EINVAL)); | |
2440 | ||
caf9dd20 BB |
2441 | /* |
2442 | * This is an unmodified spill block which was added to the stream | |
2443 | * to resolve an issue with incorrectly removing spill blocks. It | |
2444 | * should be ignored by current versions of the code which support | |
2445 | * the DRR_FLAG_SPILL_BLOCK flag. | |
2446 | */ | |
2447 | if (rwa->spill && DRR_SPILL_IS_UNMODIFIED(drrs->drr_flags)) { | |
ba67d821 | 2448 | abd_free(abd); |
caf9dd20 BB |
2449 | return (0); |
2450 | } | |
2451 | ||
03916905 PD |
2452 | if (rwa->raw) { |
2453 | if (!DMU_OT_IS_VALID(drrs->drr_type) || | |
2454 | drrs->drr_compressiontype >= ZIO_COMPRESS_FUNCTIONS || | |
2455 | drrs->drr_compressed_size == 0) | |
2456 | return (SET_ERROR(EINVAL)); | |
03916905 PD |
2457 | } |
2458 | ||
2459 | if (dmu_object_info(rwa->os, drrs->drr_object, NULL) != 0) | |
2460 | return (SET_ERROR(EINVAL)); | |
2461 | ||
2462 | if (drrs->drr_object > rwa->max_object) | |
2463 | rwa->max_object = drrs->drr_object; | |
2464 | ||
2465 | VERIFY0(dmu_bonus_hold(rwa->os, drrs->drr_object, FTAG, &db)); | |
2466 | if ((err = dmu_spill_hold_by_bonus(db, DMU_READ_NO_DECRYPT, FTAG, | |
2467 | &db_spill)) != 0) { | |
2468 | dmu_buf_rele(db, FTAG); | |
2469 | return (err); | |
2470 | } | |
2471 | ||
ba67d821 | 2472 | dmu_tx_t *tx = dmu_tx_create(rwa->os); |
03916905 PD |
2473 | |
2474 | dmu_tx_hold_spill(tx, db->db_object); | |
2475 | ||
2476 | err = dmu_tx_assign(tx, TXG_WAIT); | |
2477 | if (err != 0) { | |
2478 | dmu_buf_rele(db, FTAG); | |
2479 | dmu_buf_rele(db_spill, FTAG); | |
2480 | dmu_tx_abort(tx); | |
2481 | return (err); | |
2482 | } | |
2483 | ||
caf9dd20 BB |
2484 | /* |
2485 | * Spill blocks may both grow and shrink. When a change in size | |
2486 | * occurs any existing dbuf must be updated to match the logical | |
2487 | * size of the provided arc_buf_t. | |
2488 | */ | |
2489 | if (db_spill->db_size != drrs->drr_length) { | |
2490 | dmu_buf_will_fill(db_spill, tx); | |
ba67d821 | 2491 | VERIFY0(dbuf_spill_set_blksz(db_spill, |
03916905 | 2492 | drrs->drr_length, tx)); |
caf9dd20 | 2493 | } |
03916905 | 2494 | |
ba67d821 MA |
2495 | arc_buf_t *abuf; |
2496 | if (rwa->raw) { | |
2497 | boolean_t byteorder = ZFS_HOST_BYTEORDER ^ | |
2498 | !!DRR_IS_RAW_BYTESWAPPED(drrs->drr_flags) ^ | |
2499 | rwa->byteswap; | |
2500 | ||
2501 | abuf = arc_loan_raw_buf(dmu_objset_spa(rwa->os), | |
2502 | drrs->drr_object, byteorder, drrs->drr_salt, | |
2503 | drrs->drr_iv, drrs->drr_mac, drrs->drr_type, | |
2504 | drrs->drr_compressed_size, drrs->drr_length, | |
2505 | drrs->drr_compressiontype, 0); | |
2506 | } else { | |
2507 | abuf = arc_loan_buf(dmu_objset_spa(rwa->os), | |
2508 | DMU_OT_IS_METADATA(drrs->drr_type), | |
2509 | drrs->drr_length); | |
2510 | if (rwa->byteswap) { | |
2511 | dmu_object_byteswap_t byteswap = | |
2512 | DMU_OT_BYTESWAP(drrs->drr_type); | |
2513 | dmu_ot_byteswap[byteswap].ob_func(abd_to_buf(abd), | |
2514 | DRR_SPILL_PAYLOAD_SIZE(drrs)); | |
2515 | } | |
03916905 PD |
2516 | } |
2517 | ||
861166b0 | 2518 | memcpy(abuf->b_data, abd_to_buf(abd), DRR_SPILL_PAYLOAD_SIZE(drrs)); |
ba67d821 | 2519 | abd_free(abd); |
03916905 PD |
2520 | dbuf_assign_arcbuf((dmu_buf_impl_t *)db_spill, abuf, tx); |
2521 | ||
2522 | dmu_buf_rele(db, FTAG); | |
2523 | dmu_buf_rele(db_spill, FTAG); | |
2524 | ||
2525 | dmu_tx_commit(tx); | |
2526 | return (0); | |
2527 | } | |
2528 | ||
03916905 PD |
2529 | noinline static int |
2530 | receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf) | |
2531 | { | |
2532 | int err; | |
2533 | ||
30af21b0 | 2534 | if (drrf->drr_length != -1ULL && |
03916905 PD |
2535 | drrf->drr_offset + drrf->drr_length < drrf->drr_offset) |
2536 | return (SET_ERROR(EINVAL)); | |
2537 | ||
2538 | if (dmu_object_info(rwa->os, drrf->drr_object, NULL) != 0) | |
2539 | return (SET_ERROR(EINVAL)); | |
2540 | ||
2541 | if (drrf->drr_object > rwa->max_object) | |
2542 | rwa->max_object = drrf->drr_object; | |
2543 | ||
2544 | err = dmu_free_long_range(rwa->os, drrf->drr_object, | |
2545 | drrf->drr_offset, drrf->drr_length); | |
2546 | ||
2547 | return (err); | |
2548 | } | |
2549 | ||
2550 | static int | |
2551 | receive_object_range(struct receive_writer_arg *rwa, | |
2552 | struct drr_object_range *drror) | |
2553 | { | |
2554 | /* | |
2555 | * By default, we assume this block is in our native format | |
2556 | * (ZFS_HOST_BYTEORDER). We then take into account whether | |
2557 | * the send stream is byteswapped (rwa->byteswap). Finally, | |
2558 | * we need to byteswap again if this particular block was | |
2559 | * in non-native format on the send side. | |
2560 | */ | |
2561 | boolean_t byteorder = ZFS_HOST_BYTEORDER ^ rwa->byteswap ^ | |
2562 | !!DRR_IS_RAW_BYTESWAPPED(drror->drr_flags); | |
2563 | ||
2564 | /* | |
2565 | * Since dnode block sizes are constant, we should not need to worry | |
2566 | * about making sure that the dnode block size is the same on the | |
2567 | * sending and receiving sides for the time being. For non-raw sends, | |
2568 | * this does not matter (and in fact we do not send a DRR_OBJECT_RANGE | |
2569 | * record at all). Raw sends require this record type because the | |
2570 | * encryption parameters are used to protect an entire block of bonus | |
2571 | * buffers. If the size of dnode blocks ever becomes variable, | |
2572 | * handling will need to be added to ensure that dnode block sizes | |
2573 | * match on the sending and receiving side. | |
2574 | */ | |
2575 | if (drror->drr_numslots != DNODES_PER_BLOCK || | |
2576 | P2PHASE(drror->drr_firstobj, DNODES_PER_BLOCK) != 0 || | |
2577 | !rwa->raw) | |
2578 | return (SET_ERROR(EINVAL)); | |
2579 | ||
2580 | if (drror->drr_firstobj > rwa->max_object) | |
2581 | rwa->max_object = drror->drr_firstobj; | |
2582 | ||
2583 | /* | |
2584 | * The DRR_OBJECT_RANGE handling must be deferred to receive_object() | |
2585 | * so that the block of dnodes is not written out when it's empty, | |
2586 | * and converted to a HOLE BP. | |
2587 | */ | |
2588 | rwa->or_crypt_params_present = B_TRUE; | |
2589 | rwa->or_firstobj = drror->drr_firstobj; | |
2590 | rwa->or_numslots = drror->drr_numslots; | |
861166b0 AZ |
2591 | memcpy(rwa->or_salt, drror->drr_salt, ZIO_DATA_SALT_LEN); |
2592 | memcpy(rwa->or_iv, drror->drr_iv, ZIO_DATA_IV_LEN); | |
2593 | memcpy(rwa->or_mac, drror->drr_mac, ZIO_DATA_MAC_LEN); | |
03916905 PD |
2594 | rwa->or_byteorder = byteorder; |
2595 | ||
2596 | return (0); | |
2597 | } | |
2598 | ||
30af21b0 PD |
2599 | /* |
2600 | * Until we have the ability to redact large ranges of data efficiently, we | |
2601 | * process these records as frees. | |
2602 | */ | |
30af21b0 PD |
2603 | noinline static int |
2604 | receive_redact(struct receive_writer_arg *rwa, struct drr_redact *drrr) | |
2605 | { | |
2606 | struct drr_free drrf = {0}; | |
2607 | drrf.drr_length = drrr->drr_length; | |
2608 | drrf.drr_object = drrr->drr_object; | |
2609 | drrf.drr_offset = drrr->drr_offset; | |
2610 | drrf.drr_toguid = drrr->drr_toguid; | |
2611 | return (receive_free(rwa, &drrf)); | |
2612 | } | |
2613 | ||
03916905 PD |
2614 | /* used to destroy the drc_ds on error */ |
2615 | static void | |
2616 | dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) | |
2617 | { | |
2618 | dsl_dataset_t *ds = drc->drc_ds; | |
40ab927a | 2619 | ds_hold_flags_t dsflags; |
03916905 | 2620 | |
40ab927a | 2621 | dsflags = (drc->drc_raw) ? DS_HOLD_FLAG_NONE : DS_HOLD_FLAG_DECRYPT; |
03916905 PD |
2622 | /* |
2623 | * Wait for the txg sync before cleaning up the receive. For | |
2624 | * resumable receives, this ensures that our resume state has | |
2625 | * been written out to disk. For raw receives, this ensures | |
2626 | * that the user accounting code will not attempt to do anything | |
2627 | * after we stopped receiving the dataset. | |
2628 | */ | |
2629 | txg_wait_synced(ds->ds_dir->dd_pool, 0); | |
2630 | ds->ds_objset->os_raw_receive = B_FALSE; | |
2631 | ||
2632 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); | |
61152d10 TC |
2633 | if (drc->drc_resumable && drc->drc_should_save && |
2634 | !BP_IS_HOLE(dsl_dataset_get_blkptr(ds))) { | |
03916905 PD |
2635 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
2636 | dsl_dataset_disown(ds, dsflags, dmu_recv_tag); | |
2637 | } else { | |
2638 | char name[ZFS_MAX_DATASET_NAME_LEN]; | |
2639 | rrw_exit(&ds->ds_bp_rwlock, FTAG); | |
2640 | dsl_dataset_name(ds, name); | |
2641 | dsl_dataset_disown(ds, dsflags, dmu_recv_tag); | |
e8cf3a4f AP |
2642 | if (!drc->drc_heal) |
2643 | (void) dsl_destroy_head(name); | |
03916905 PD |
2644 | } |
2645 | } | |
2646 | ||
2647 | static void | |
30af21b0 | 2648 | receive_cksum(dmu_recv_cookie_t *drc, int len, void *buf) |
03916905 | 2649 | { |
30af21b0 PD |
2650 | if (drc->drc_byteswap) { |
2651 | (void) fletcher_4_incremental_byteswap(buf, len, | |
2652 | &drc->drc_cksum); | |
03916905 | 2653 | } else { |
30af21b0 | 2654 | (void) fletcher_4_incremental_native(buf, len, &drc->drc_cksum); |
03916905 PD |
2655 | } |
2656 | } | |
2657 | ||
2658 | /* | |
2659 | * Read the payload into a buffer of size len, and update the current record's | |
2660 | * payload field. | |
30af21b0 PD |
2661 | * Allocate drc->drc_next_rrd and read the next record's header into |
2662 | * drc->drc_next_rrd->header. | |
03916905 PD |
2663 | * Verify checksum of payload and next record. |
2664 | */ | |
2665 | static int | |
30af21b0 | 2666 | receive_read_payload_and_next_header(dmu_recv_cookie_t *drc, int len, void *buf) |
03916905 PD |
2667 | { |
2668 | int err; | |
03916905 PD |
2669 | |
2670 | if (len != 0) { | |
2671 | ASSERT3U(len, <=, SPA_MAXBLOCKSIZE); | |
30af21b0 | 2672 | err = receive_read(drc, len, buf); |
03916905 PD |
2673 | if (err != 0) |
2674 | return (err); | |
30af21b0 | 2675 | receive_cksum(drc, len, buf); |
03916905 PD |
2676 | |
2677 | /* note: rrd is NULL when reading the begin record's payload */ | |
30af21b0 PD |
2678 | if (drc->drc_rrd != NULL) { |
2679 | drc->drc_rrd->payload = buf; | |
2680 | drc->drc_rrd->payload_size = len; | |
2681 | drc->drc_rrd->bytes_read = drc->drc_bytes_read; | |
03916905 | 2682 | } |
960347d3 TC |
2683 | } else { |
2684 | ASSERT3P(buf, ==, NULL); | |
03916905 PD |
2685 | } |
2686 | ||
30af21b0 | 2687 | drc->drc_prev_cksum = drc->drc_cksum; |
03916905 | 2688 | |
30af21b0 PD |
2689 | drc->drc_next_rrd = kmem_zalloc(sizeof (*drc->drc_next_rrd), KM_SLEEP); |
2690 | err = receive_read(drc, sizeof (drc->drc_next_rrd->header), | |
2691 | &drc->drc_next_rrd->header); | |
2692 | drc->drc_next_rrd->bytes_read = drc->drc_bytes_read; | |
03916905 PD |
2693 | |
2694 | if (err != 0) { | |
30af21b0 PD |
2695 | kmem_free(drc->drc_next_rrd, sizeof (*drc->drc_next_rrd)); |
2696 | drc->drc_next_rrd = NULL; | |
03916905 PD |
2697 | return (err); |
2698 | } | |
30af21b0 PD |
2699 | if (drc->drc_next_rrd->header.drr_type == DRR_BEGIN) { |
2700 | kmem_free(drc->drc_next_rrd, sizeof (*drc->drc_next_rrd)); | |
2701 | drc->drc_next_rrd = NULL; | |
03916905 PD |
2702 | return (SET_ERROR(EINVAL)); |
2703 | } | |
2704 | ||
2705 | /* | |
2706 | * Note: checksum is of everything up to but not including the | |
2707 | * checksum itself. | |
2708 | */ | |
2709 | ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), | |
2710 | ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); | |
30af21b0 | 2711 | receive_cksum(drc, |
03916905 | 2712 | offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), |
30af21b0 | 2713 | &drc->drc_next_rrd->header); |
03916905 | 2714 | |
30af21b0 PD |
2715 | zio_cksum_t cksum_orig = |
2716 | drc->drc_next_rrd->header.drr_u.drr_checksum.drr_checksum; | |
2717 | zio_cksum_t *cksump = | |
2718 | &drc->drc_next_rrd->header.drr_u.drr_checksum.drr_checksum; | |
03916905 | 2719 | |
30af21b0 PD |
2720 | if (drc->drc_byteswap) |
2721 | byteswap_record(&drc->drc_next_rrd->header); | |
03916905 PD |
2722 | |
2723 | if ((!ZIO_CHECKSUM_IS_ZERO(cksump)) && | |
30af21b0 PD |
2724 | !ZIO_CHECKSUM_EQUAL(drc->drc_cksum, *cksump)) { |
2725 | kmem_free(drc->drc_next_rrd, sizeof (*drc->drc_next_rrd)); | |
2726 | drc->drc_next_rrd = NULL; | |
03916905 PD |
2727 | return (SET_ERROR(ECKSUM)); |
2728 | } | |
2729 | ||
30af21b0 | 2730 | receive_cksum(drc, sizeof (cksum_orig), &cksum_orig); |
03916905 PD |
2731 | |
2732 | return (0); | |
2733 | } | |
2734 | ||
03916905 PD |
2735 | /* |
2736 | * Issue the prefetch reads for any necessary indirect blocks. | |
2737 | * | |
2738 | * We use the object ignore list to tell us whether or not to issue prefetches | |
2739 | * for a given object. We do this for both correctness (in case the blocksize | |
2740 | * of an object has changed) and performance (if the object doesn't exist, don't | |
2741 | * needlessly try to issue prefetches). We also trim the list as we go through | |
2742 | * the stream to prevent it from growing to an unbounded size. | |
2743 | * | |
2744 | * The object numbers within will always be in sorted order, and any write | |
2745 | * records we see will also be in sorted order, but they're not sorted with | |
2746 | * respect to each other (i.e. we can get several object records before | |
2747 | * receiving each object's write records). As a result, once we've reached a | |
2748 | * given object number, we can safely remove any reference to lower object | |
2749 | * numbers in the ignore list. In practice, we receive up to 32 object records | |
2750 | * before receiving write records, so the list can have up to 32 nodes in it. | |
2751 | */ | |
03916905 | 2752 | static void |
30af21b0 PD |
2753 | receive_read_prefetch(dmu_recv_cookie_t *drc, uint64_t object, uint64_t offset, |
2754 | uint64_t length) | |
03916905 | 2755 | { |
30af21b0 PD |
2756 | if (!objlist_exists(drc->drc_ignore_objlist, object)) { |
2757 | dmu_prefetch(drc->drc_os, object, 1, offset, length, | |
03916905 PD |
2758 | ZIO_PRIORITY_SYNC_READ); |
2759 | } | |
2760 | } | |
2761 | ||
2762 | /* | |
2763 | * Read records off the stream, issuing any necessary prefetches. | |
2764 | */ | |
2765 | static int | |
30af21b0 | 2766 | receive_read_record(dmu_recv_cookie_t *drc) |
03916905 PD |
2767 | { |
2768 | int err; | |
2769 | ||
30af21b0 | 2770 | switch (drc->drc_rrd->header.drr_type) { |
03916905 PD |
2771 | case DRR_OBJECT: |
2772 | { | |
30af21b0 PD |
2773 | struct drr_object *drro = |
2774 | &drc->drc_rrd->header.drr_u.drr_object; | |
03916905 | 2775 | uint32_t size = DRR_OBJECT_PAYLOAD_SIZE(drro); |
960347d3 | 2776 | void *buf = NULL; |
03916905 PD |
2777 | dmu_object_info_t doi; |
2778 | ||
960347d3 TC |
2779 | if (size != 0) |
2780 | buf = kmem_zalloc(size, KM_SLEEP); | |
2781 | ||
30af21b0 | 2782 | err = receive_read_payload_and_next_header(drc, size, buf); |
03916905 PD |
2783 | if (err != 0) { |
2784 | kmem_free(buf, size); | |
2785 | return (err); | |
2786 | } | |
30af21b0 | 2787 | err = dmu_object_info(drc->drc_os, drro->drr_object, &doi); |
03916905 PD |
2788 | /* |
2789 | * See receive_read_prefetch for an explanation why we're | |
2790 | * storing this object in the ignore_obj_list. | |
2791 | */ | |
2792 | if (err == ENOENT || err == EEXIST || | |
2793 | (err == 0 && doi.doi_data_block_size != drro->drr_blksz)) { | |
30af21b0 PD |
2794 | objlist_insert(drc->drc_ignore_objlist, |
2795 | drro->drr_object); | |
03916905 PD |
2796 | err = 0; |
2797 | } | |
2798 | return (err); | |
2799 | } | |
2800 | case DRR_FREEOBJECTS: | |
2801 | { | |
30af21b0 | 2802 | err = receive_read_payload_and_next_header(drc, 0, NULL); |
03916905 PD |
2803 | return (err); |
2804 | } | |
2805 | case DRR_WRITE: | |
2806 | { | |
30af21b0 | 2807 | struct drr_write *drrw = &drc->drc_rrd->header.drr_u.drr_write; |
ba67d821 MA |
2808 | int size = DRR_WRITE_PAYLOAD_SIZE(drrw); |
2809 | abd_t *abd = abd_alloc_linear(size, B_FALSE); | |
2810 | err = receive_read_payload_and_next_header(drc, size, | |
2811 | abd_to_buf(abd)); | |
03916905 | 2812 | if (err != 0) { |
ba67d821 | 2813 | abd_free(abd); |
03916905 PD |
2814 | return (err); |
2815 | } | |
ba67d821 | 2816 | drc->drc_rrd->abd = abd; |
30af21b0 | 2817 | receive_read_prefetch(drc, drrw->drr_object, drrw->drr_offset, |
03916905 PD |
2818 | drrw->drr_logical_size); |
2819 | return (err); | |
2820 | } | |
03916905 PD |
2821 | case DRR_WRITE_EMBEDDED: |
2822 | { | |
2823 | struct drr_write_embedded *drrwe = | |
30af21b0 | 2824 | &drc->drc_rrd->header.drr_u.drr_write_embedded; |
03916905 PD |
2825 | uint32_t size = P2ROUNDUP(drrwe->drr_psize, 8); |
2826 | void *buf = kmem_zalloc(size, KM_SLEEP); | |
2827 | ||
30af21b0 | 2828 | err = receive_read_payload_and_next_header(drc, size, buf); |
03916905 PD |
2829 | if (err != 0) { |
2830 | kmem_free(buf, size); | |
2831 | return (err); | |
2832 | } | |
2833 | ||
30af21b0 | 2834 | receive_read_prefetch(drc, drrwe->drr_object, drrwe->drr_offset, |
03916905 PD |
2835 | drrwe->drr_length); |
2836 | return (err); | |
2837 | } | |
2838 | case DRR_FREE: | |
30af21b0 | 2839 | case DRR_REDACT: |
03916905 PD |
2840 | { |
2841 | /* | |
2842 | * It might be beneficial to prefetch indirect blocks here, but | |
2843 | * we don't really have the data to decide for sure. | |
2844 | */ | |
30af21b0 | 2845 | err = receive_read_payload_and_next_header(drc, 0, NULL); |
03916905 PD |
2846 | return (err); |
2847 | } | |
2848 | case DRR_END: | |
2849 | { | |
30af21b0 PD |
2850 | struct drr_end *drre = &drc->drc_rrd->header.drr_u.drr_end; |
2851 | if (!ZIO_CHECKSUM_EQUAL(drc->drc_prev_cksum, | |
2852 | drre->drr_checksum)) | |
03916905 PD |
2853 | return (SET_ERROR(ECKSUM)); |
2854 | return (0); | |
2855 | } | |
2856 | case DRR_SPILL: | |
2857 | { | |
30af21b0 | 2858 | struct drr_spill *drrs = &drc->drc_rrd->header.drr_u.drr_spill; |
ba67d821 MA |
2859 | int size = DRR_SPILL_PAYLOAD_SIZE(drrs); |
2860 | abd_t *abd = abd_alloc_linear(size, B_FALSE); | |
2861 | err = receive_read_payload_and_next_header(drc, size, | |
2862 | abd_to_buf(abd)); | |
30af21b0 | 2863 | if (err != 0) |
ba67d821 | 2864 | abd_free(abd); |
30af21b0 | 2865 | else |
ba67d821 | 2866 | drc->drc_rrd->abd = abd; |
03916905 PD |
2867 | return (err); |
2868 | } | |
2869 | case DRR_OBJECT_RANGE: | |
2870 | { | |
30af21b0 | 2871 | err = receive_read_payload_and_next_header(drc, 0, NULL); |
03916905 | 2872 | return (err); |
30af21b0 | 2873 | |
03916905 PD |
2874 | } |
2875 | default: | |
2876 | return (SET_ERROR(EINVAL)); | |
2877 | } | |
2878 | } | |
2879 | ||
30af21b0 PD |
2880 | |
2881 | ||
03916905 PD |
2882 | static void |
2883 | dprintf_drr(struct receive_record_arg *rrd, int err) | |
2884 | { | |
2885 | #ifdef ZFS_DEBUG | |
2886 | switch (rrd->header.drr_type) { | |
2887 | case DRR_OBJECT: | |
2888 | { | |
2889 | struct drr_object *drro = &rrd->header.drr_u.drr_object; | |
2890 | dprintf("drr_type = OBJECT obj = %llu type = %u " | |
2891 | "bonustype = %u blksz = %u bonuslen = %u cksumtype = %u " | |
2892 | "compress = %u dn_slots = %u err = %d\n", | |
8e739b2c RE |
2893 | (u_longlong_t)drro->drr_object, drro->drr_type, |
2894 | drro->drr_bonustype, drro->drr_blksz, drro->drr_bonuslen, | |
03916905 PD |
2895 | drro->drr_checksumtype, drro->drr_compress, |
2896 | drro->drr_dn_slots, err); | |
2897 | break; | |
2898 | } | |
2899 | case DRR_FREEOBJECTS: | |
2900 | { | |
2901 | struct drr_freeobjects *drrfo = | |
2902 | &rrd->header.drr_u.drr_freeobjects; | |
2903 | dprintf("drr_type = FREEOBJECTS firstobj = %llu " | |
2904 | "numobjs = %llu err = %d\n", | |
8e739b2c RE |
2905 | (u_longlong_t)drrfo->drr_firstobj, |
2906 | (u_longlong_t)drrfo->drr_numobjs, err); | |
03916905 PD |
2907 | break; |
2908 | } | |
2909 | case DRR_WRITE: | |
2910 | { | |
2911 | struct drr_write *drrw = &rrd->header.drr_u.drr_write; | |
2912 | dprintf("drr_type = WRITE obj = %llu type = %u offset = %llu " | |
5dbf8b4e | 2913 | "lsize = %llu cksumtype = %u flags = %u " |
03916905 | 2914 | "compress = %u psize = %llu err = %d\n", |
8e739b2c RE |
2915 | (u_longlong_t)drrw->drr_object, drrw->drr_type, |
2916 | (u_longlong_t)drrw->drr_offset, | |
2917 | (u_longlong_t)drrw->drr_logical_size, | |
2918 | drrw->drr_checksumtype, drrw->drr_flags, | |
2919 | drrw->drr_compressiontype, | |
2920 | (u_longlong_t)drrw->drr_compressed_size, err); | |
03916905 PD |
2921 | break; |
2922 | } | |
2923 | case DRR_WRITE_BYREF: | |
2924 | { | |
2925 | struct drr_write_byref *drrwbr = | |
2926 | &rrd->header.drr_u.drr_write_byref; | |
2927 | dprintf("drr_type = WRITE_BYREF obj = %llu offset = %llu " | |
2928 | "length = %llu toguid = %llx refguid = %llx " | |
2929 | "refobject = %llu refoffset = %llu cksumtype = %u " | |
5dbf8b4e | 2930 | "flags = %u err = %d\n", |
8e739b2c RE |
2931 | (u_longlong_t)drrwbr->drr_object, |
2932 | (u_longlong_t)drrwbr->drr_offset, | |
2933 | (u_longlong_t)drrwbr->drr_length, | |
2934 | (u_longlong_t)drrwbr->drr_toguid, | |
2935 | (u_longlong_t)drrwbr->drr_refguid, | |
2936 | (u_longlong_t)drrwbr->drr_refobject, | |
2937 | (u_longlong_t)drrwbr->drr_refoffset, | |
2938 | drrwbr->drr_checksumtype, drrwbr->drr_flags, err); | |
03916905 PD |
2939 | break; |
2940 | } | |
2941 | case DRR_WRITE_EMBEDDED: | |
2942 | { | |
2943 | struct drr_write_embedded *drrwe = | |
2944 | &rrd->header.drr_u.drr_write_embedded; | |
2945 | dprintf("drr_type = WRITE_EMBEDDED obj = %llu offset = %llu " | |
2946 | "length = %llu compress = %u etype = %u lsize = %u " | |
2947 | "psize = %u err = %d\n", | |
8e739b2c RE |
2948 | (u_longlong_t)drrwe->drr_object, |
2949 | (u_longlong_t)drrwe->drr_offset, | |
2950 | (u_longlong_t)drrwe->drr_length, | |
03916905 PD |
2951 | drrwe->drr_compression, drrwe->drr_etype, |
2952 | drrwe->drr_lsize, drrwe->drr_psize, err); | |
2953 | break; | |
2954 | } | |
2955 | case DRR_FREE: | |
2956 | { | |
2957 | struct drr_free *drrf = &rrd->header.drr_u.drr_free; | |
2958 | dprintf("drr_type = FREE obj = %llu offset = %llu " | |
2959 | "length = %lld err = %d\n", | |
8e739b2c RE |
2960 | (u_longlong_t)drrf->drr_object, |
2961 | (u_longlong_t)drrf->drr_offset, | |
2962 | (longlong_t)drrf->drr_length, | |
03916905 PD |
2963 | err); |
2964 | break; | |
2965 | } | |
2966 | case DRR_SPILL: | |
2967 | { | |
2968 | struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; | |
2969 | dprintf("drr_type = SPILL obj = %llu length = %llu " | |
8e739b2c RE |
2970 | "err = %d\n", (u_longlong_t)drrs->drr_object, |
2971 | (u_longlong_t)drrs->drr_length, err); | |
03916905 PD |
2972 | break; |
2973 | } | |
5dbf8b4e TC |
2974 | case DRR_OBJECT_RANGE: |
2975 | { | |
2976 | struct drr_object_range *drror = | |
2977 | &rrd->header.drr_u.drr_object_range; | |
2978 | dprintf("drr_type = OBJECT_RANGE firstobj = %llu " | |
2979 | "numslots = %llu flags = %u err = %d\n", | |
8e739b2c RE |
2980 | (u_longlong_t)drror->drr_firstobj, |
2981 | (u_longlong_t)drror->drr_numslots, | |
5dbf8b4e TC |
2982 | drror->drr_flags, err); |
2983 | break; | |
2984 | } | |
03916905 PD |
2985 | default: |
2986 | return; | |
2987 | } | |
2988 | #endif | |
2989 | } | |
2990 | ||
2991 | /* | |
2992 | * Commit the records to the pool. | |
2993 | */ | |
2994 | static int | |
2995 | receive_process_record(struct receive_writer_arg *rwa, | |
2996 | struct receive_record_arg *rrd) | |
2997 | { | |
2998 | int err; | |
2999 | ||
3000 | /* Processing in order, therefore bytes_read should be increasing. */ | |
3001 | ASSERT3U(rrd->bytes_read, >=, rwa->bytes_read); | |
3002 | rwa->bytes_read = rrd->bytes_read; | |
3003 | ||
e8cf3a4f AP |
3004 | /* We can only heal write records; other ones get ignored */ |
3005 | if (rwa->heal && rrd->header.drr_type != DRR_WRITE) { | |
3006 | if (rrd->abd != NULL) { | |
3007 | abd_free(rrd->abd); | |
3008 | rrd->abd = NULL; | |
3009 | } else if (rrd->payload != NULL) { | |
3010 | kmem_free(rrd->payload, rrd->payload_size); | |
3011 | rrd->payload = NULL; | |
3012 | } | |
3013 | return (0); | |
3014 | } | |
3015 | ||
3016 | if (!rwa->heal && rrd->header.drr_type != DRR_WRITE) { | |
7261fc2e MA |
3017 | err = flush_write_batch(rwa); |
3018 | if (err != 0) { | |
ba67d821 MA |
3019 | if (rrd->abd != NULL) { |
3020 | abd_free(rrd->abd); | |
3021 | rrd->abd = NULL; | |
7261fc2e MA |
3022 | rrd->payload = NULL; |
3023 | } else if (rrd->payload != NULL) { | |
3024 | kmem_free(rrd->payload, rrd->payload_size); | |
3025 | rrd->payload = NULL; | |
3026 | } | |
3027 | ||
3028 | return (err); | |
3029 | } | |
3030 | } | |
3031 | ||
03916905 PD |
3032 | switch (rrd->header.drr_type) { |
3033 | case DRR_OBJECT: | |
3034 | { | |
3035 | struct drr_object *drro = &rrd->header.drr_u.drr_object; | |
3036 | err = receive_object(rwa, drro, rrd->payload); | |
3037 | kmem_free(rrd->payload, rrd->payload_size); | |
3038 | rrd->payload = NULL; | |
3039 | break; | |
3040 | } | |
3041 | case DRR_FREEOBJECTS: | |
3042 | { | |
3043 | struct drr_freeobjects *drrfo = | |
3044 | &rrd->header.drr_u.drr_freeobjects; | |
3045 | err = receive_freeobjects(rwa, drrfo); | |
3046 | break; | |
3047 | } | |
3048 | case DRR_WRITE: | |
3049 | { | |
7261fc2e | 3050 | err = receive_process_write_record(rwa, rrd); |
e8cf3a4f AP |
3051 | if (rwa->heal) { |
3052 | /* | |
3053 | * If healing - always free the abd after processing | |
3054 | */ | |
3055 | abd_free(rrd->abd); | |
3056 | rrd->abd = NULL; | |
3057 | } else if (err != EAGAIN) { | |
7261fc2e | 3058 | /* |
e8cf3a4f AP |
3059 | * On success, a non-healing |
3060 | * receive_process_write_record() returns | |
7261fc2e MA |
3061 | * EAGAIN to indicate that we do not want to free |
3062 | * the rrd or arc_buf. | |
3063 | */ | |
3064 | ASSERT(err != 0); | |
ba67d821 MA |
3065 | abd_free(rrd->abd); |
3066 | rrd->abd = NULL; | |
7261fc2e | 3067 | } |
03916905 PD |
3068 | break; |
3069 | } | |
03916905 PD |
3070 | case DRR_WRITE_EMBEDDED: |
3071 | { | |
3072 | struct drr_write_embedded *drrwe = | |
3073 | &rrd->header.drr_u.drr_write_embedded; | |
3074 | err = receive_write_embedded(rwa, drrwe, rrd->payload); | |
3075 | kmem_free(rrd->payload, rrd->payload_size); | |
3076 | rrd->payload = NULL; | |
3077 | break; | |
3078 | } | |
3079 | case DRR_FREE: | |
3080 | { | |
3081 | struct drr_free *drrf = &rrd->header.drr_u.drr_free; | |
3082 | err = receive_free(rwa, drrf); | |
3083 | break; | |
3084 | } | |
3085 | case DRR_SPILL: | |
3086 | { | |
3087 | struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; | |
ba67d821 | 3088 | err = receive_spill(rwa, drrs, rrd->abd); |
03916905 | 3089 | if (err != 0) |
ba67d821 MA |
3090 | abd_free(rrd->abd); |
3091 | rrd->abd = NULL; | |
03916905 PD |
3092 | rrd->payload = NULL; |
3093 | break; | |
3094 | } | |
3095 | case DRR_OBJECT_RANGE: | |
3096 | { | |
3097 | struct drr_object_range *drror = | |
3098 | &rrd->header.drr_u.drr_object_range; | |
5dbf8b4e TC |
3099 | err = receive_object_range(rwa, drror); |
3100 | break; | |
03916905 | 3101 | } |
30af21b0 PD |
3102 | case DRR_REDACT: |
3103 | { | |
3104 | struct drr_redact *drrr = &rrd->header.drr_u.drr_redact; | |
3105 | err = receive_redact(rwa, drrr); | |
3106 | break; | |
3107 | } | |
03916905 | 3108 | default: |
5dbf8b4e | 3109 | err = (SET_ERROR(EINVAL)); |
03916905 PD |
3110 | } |
3111 | ||
3112 | if (err != 0) | |
3113 | dprintf_drr(rrd, err); | |
3114 | ||
3115 | return (err); | |
3116 | } | |
3117 | ||
3118 | /* | |
3119 | * dmu_recv_stream's worker thread; pull records off the queue, and then call | |
3120 | * receive_process_record When we're done, signal the main thread and exit. | |
3121 | */ | |
460748d4 | 3122 | static __attribute__((noreturn)) void |
03916905 PD |
3123 | receive_writer_thread(void *arg) |
3124 | { | |
3125 | struct receive_writer_arg *rwa = arg; | |
3126 | struct receive_record_arg *rrd; | |
3127 | fstrans_cookie_t cookie = spl_fstrans_mark(); | |
3128 | ||
3129 | for (rrd = bqueue_dequeue(&rwa->q); !rrd->eos_marker; | |
3130 | rrd = bqueue_dequeue(&rwa->q)) { | |
3131 | /* | |
3132 | * If there's an error, the main thread will stop putting things | |
3133 | * on the queue, but we need to clear everything in it before we | |
3134 | * can exit. | |
3135 | */ | |
7261fc2e | 3136 | int err = 0; |
03916905 | 3137 | if (rwa->err == 0) { |
7261fc2e | 3138 | err = receive_process_record(rwa, rrd); |
ba67d821 MA |
3139 | } else if (rrd->abd != NULL) { |
3140 | abd_free(rrd->abd); | |
3141 | rrd->abd = NULL; | |
03916905 PD |
3142 | rrd->payload = NULL; |
3143 | } else if (rrd->payload != NULL) { | |
3144 | kmem_free(rrd->payload, rrd->payload_size); | |
3145 | rrd->payload = NULL; | |
3146 | } | |
7261fc2e MA |
3147 | /* |
3148 | * EAGAIN indicates that this record has been saved (on | |
3149 | * raw->write_batch), and will be used again, so we don't | |
3150 | * free it. | |
e8cf3a4f | 3151 | * When healing data we always need to free the record. |
7261fc2e | 3152 | */ |
e8cf3a4f | 3153 | if (err != EAGAIN || rwa->heal) { |
1b9cd1a9 MA |
3154 | if (rwa->err == 0) |
3155 | rwa->err = err; | |
7261fc2e MA |
3156 | kmem_free(rrd, sizeof (*rrd)); |
3157 | } | |
03916905 PD |
3158 | } |
3159 | kmem_free(rrd, sizeof (*rrd)); | |
7261fc2e | 3160 | |
e8cf3a4f AP |
3161 | if (rwa->heal) { |
3162 | zio_wait(rwa->heal_pio); | |
3163 | } else { | |
3164 | int err = flush_write_batch(rwa); | |
3165 | if (rwa->err == 0) | |
3166 | rwa->err = err; | |
3167 | } | |
03916905 PD |
3168 | mutex_enter(&rwa->mutex); |
3169 | rwa->done = B_TRUE; | |
3170 | cv_signal(&rwa->cv); | |
3171 | mutex_exit(&rwa->mutex); | |
3172 | spl_fstrans_unmark(cookie); | |
3173 | thread_exit(); | |
3174 | } | |
3175 | ||
3176 | static int | |
30af21b0 | 3177 | resume_check(dmu_recv_cookie_t *drc, nvlist_t *begin_nvl) |
03916905 PD |
3178 | { |
3179 | uint64_t val; | |
30af21b0 PD |
3180 | objset_t *mos = dmu_objset_pool(drc->drc_os)->dp_meta_objset; |
3181 | uint64_t dsobj = dmu_objset_id(drc->drc_os); | |
03916905 PD |
3182 | uint64_t resume_obj, resume_off; |
3183 | ||
3184 | if (nvlist_lookup_uint64(begin_nvl, | |
3185 | "resume_object", &resume_obj) != 0 || | |
3186 | nvlist_lookup_uint64(begin_nvl, | |
3187 | "resume_offset", &resume_off) != 0) { | |
3188 | return (SET_ERROR(EINVAL)); | |
3189 | } | |
3190 | VERIFY0(zap_lookup(mos, dsobj, | |
3191 | DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val)); | |
3192 | if (resume_obj != val) | |
3193 | return (SET_ERROR(EINVAL)); | |
3194 | VERIFY0(zap_lookup(mos, dsobj, | |
3195 | DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val)); | |
3196 | if (resume_off != val) | |
3197 | return (SET_ERROR(EINVAL)); | |
3198 | ||
3199 | return (0); | |
3200 | } | |
3201 | ||
3202 | /* | |
3203 | * Read in the stream's records, one by one, and apply them to the pool. There | |
3204 | * are two threads involved; the thread that calls this function will spin up a | |
3205 | * worker thread, read the records off the stream one by one, and issue | |
3206 | * prefetches for any necessary indirect blocks. It will then push the records | |
3207 | * onto an internal blocking queue. The worker thread will pull the records off | |
3208 | * the queue, and actually write the data into the DMU. This way, the worker | |
3209 | * thread doesn't have to wait for reads to complete, since everything it needs | |
3210 | * (the indirect blocks) will be prefetched. | |
3211 | * | |
3212 | * NB: callers *must* call dmu_recv_end() if this succeeds. | |
3213 | */ | |
3214 | int | |
196bee4c | 3215 | dmu_recv_stream(dmu_recv_cookie_t *drc, offset_t *voffp) |
03916905 PD |
3216 | { |
3217 | int err = 0; | |
30af21b0 | 3218 | struct receive_writer_arg *rwa = kmem_zalloc(sizeof (*rwa), KM_SLEEP); |
03916905 | 3219 | |
fce29d6a PZ |
3220 | if (dsl_dataset_has_resume_receive_state(drc->drc_ds)) { |
3221 | uint64_t bytes = 0; | |
03916905 PD |
3222 | (void) zap_lookup(drc->drc_ds->ds_dir->dd_pool->dp_meta_objset, |
3223 | drc->drc_ds->ds_object, DS_FIELD_RESUME_BYTES, | |
30af21b0 PD |
3224 | sizeof (bytes), 1, &bytes); |
3225 | drc->drc_bytes_read += bytes; | |
03916905 PD |
3226 | } |
3227 | ||
30af21b0 | 3228 | drc->drc_ignore_objlist = objlist_create(); |
03916905 PD |
3229 | |
3230 | /* these were verified in dmu_recv_begin */ | |
3231 | ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, | |
3232 | DMU_SUBSTREAM); | |
3233 | ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); | |
3234 | ||
03916905 | 3235 | ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); |
30af21b0 PD |
3236 | ASSERT0(drc->drc_os->os_encrypted && |
3237 | (drc->drc_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)); | |
03916905 | 3238 | |
03916905 | 3239 | /* handle DSL encryption key payload */ |
30af21b0 | 3240 | if (drc->drc_featureflags & DMU_BACKUP_FEATURE_RAW) { |
03916905 PD |
3241 | nvlist_t *keynvl = NULL; |
3242 | ||
30af21b0 | 3243 | ASSERT(drc->drc_os->os_encrypted); |
03916905 PD |
3244 | ASSERT(drc->drc_raw); |
3245 | ||
30af21b0 PD |
3246 | err = nvlist_lookup_nvlist(drc->drc_begin_nvl, "crypt_keydata", |
3247 | &keynvl); | |
03916905 PD |
3248 | if (err != 0) |
3249 | goto out; | |
3250 | ||
e8cf3a4f AP |
3251 | if (!drc->drc_heal) { |
3252 | /* | |
3253 | * If this is a new dataset we set the key immediately. | |
3254 | * Otherwise we don't want to change the key until we | |
3255 | * are sure the rest of the receive succeeded so we | |
3256 | * stash the keynvl away until then. | |
3257 | */ | |
3258 | err = dsl_crypto_recv_raw(spa_name(drc->drc_os->os_spa), | |
3259 | drc->drc_ds->ds_object, drc->drc_fromsnapobj, | |
3260 | drc->drc_drrb->drr_type, keynvl, drc->drc_newfs); | |
3261 | if (err != 0) | |
3262 | goto out; | |
3263 | } | |
03916905 | 3264 | |
f00ab3f2 TC |
3265 | /* see comment in dmu_recv_end_sync() */ |
3266 | drc->drc_ivset_guid = 0; | |
3267 | (void) nvlist_lookup_uint64(keynvl, "to_ivset_guid", | |
3268 | &drc->drc_ivset_guid); | |
3269 | ||
03916905 PD |
3270 | if (!drc->drc_newfs) |
3271 | drc->drc_keynvl = fnvlist_dup(keynvl); | |
3272 | } | |
3273 | ||
30af21b0 PD |
3274 | if (drc->drc_featureflags & DMU_BACKUP_FEATURE_RESUMING) { |
3275 | err = resume_check(drc, drc->drc_begin_nvl); | |
03916905 PD |
3276 | if (err != 0) |
3277 | goto out; | |
3278 | } | |
3279 | ||
61152d10 TC |
3280 | /* |
3281 | * If we failed before this point we will clean up any new resume | |
3282 | * state that was created. Now that we've gotten past the initial | |
3283 | * checks we are ok to retain that resume state. | |
3284 | */ | |
3285 | drc->drc_should_save = B_TRUE; | |
3286 | ||
30af21b0 | 3287 | (void) bqueue_init(&rwa->q, zfs_recv_queue_ff, |
03916905 PD |
3288 | MAX(zfs_recv_queue_length, 2 * zfs_max_recordsize), |
3289 | offsetof(struct receive_record_arg, node)); | |
3290 | cv_init(&rwa->cv, NULL, CV_DEFAULT, NULL); | |
3291 | mutex_init(&rwa->mutex, NULL, MUTEX_DEFAULT, NULL); | |
30af21b0 | 3292 | rwa->os = drc->drc_os; |
03916905 | 3293 | rwa->byteswap = drc->drc_byteswap; |
e8cf3a4f AP |
3294 | rwa->heal = drc->drc_heal; |
3295 | rwa->tofs = drc->drc_tofs; | |
03916905 PD |
3296 | rwa->resumable = drc->drc_resumable; |
3297 | rwa->raw = drc->drc_raw; | |
caf9dd20 | 3298 | rwa->spill = drc->drc_spill; |
7bcb7f08 | 3299 | rwa->full = (drc->drc_drr_begin->drr_u.drr_begin.drr_fromguid == 0); |
03916905 | 3300 | rwa->os->os_raw_receive = drc->drc_raw; |
e8cf3a4f AP |
3301 | if (drc->drc_heal) { |
3302 | rwa->heal_pio = zio_root(drc->drc_os->os_spa, NULL, NULL, | |
3303 | ZIO_FLAG_GODFATHER); | |
3304 | } | |
7261fc2e MA |
3305 | list_create(&rwa->write_batch, sizeof (struct receive_record_arg), |
3306 | offsetof(struct receive_record_arg, node.bqn_node)); | |
03916905 PD |
3307 | |
3308 | (void) thread_create(NULL, 0, receive_writer_thread, rwa, 0, curproc, | |
3309 | TS_RUN, minclsyspri); | |
3310 | /* | |
3311 | * We're reading rwa->err without locks, which is safe since we are the | |
3312 | * only reader, and the worker thread is the only writer. It's ok if we | |
3313 | * miss a write for an iteration or two of the loop, since the writer | |
3314 | * thread will keep freeing records we send it until we send it an eos | |
3315 | * marker. | |
3316 | * | |
3317 | * We can leave this loop in 3 ways: First, if rwa->err is | |
3318 | * non-zero. In that case, the writer thread will free the rrd we just | |
3319 | * pushed. Second, if we're interrupted; in that case, either it's the | |
30af21b0 PD |
3320 | * first loop and drc->drc_rrd was never allocated, or it's later, and |
3321 | * drc->drc_rrd has been handed off to the writer thread who will free | |
3322 | * it. Finally, if receive_read_record fails or we're at the end of the | |
3323 | * stream, then we free drc->drc_rrd and exit. | |
03916905 PD |
3324 | */ |
3325 | while (rwa->err == 0) { | |
3326 | if (issig(JUSTLOOKING) && issig(FORREAL)) { | |
3327 | err = SET_ERROR(EINTR); | |
3328 | break; | |
3329 | } | |
3330 | ||
30af21b0 PD |
3331 | ASSERT3P(drc->drc_rrd, ==, NULL); |
3332 | drc->drc_rrd = drc->drc_next_rrd; | |
3333 | drc->drc_next_rrd = NULL; | |
3334 | /* Allocates and loads header into drc->drc_next_rrd */ | |
3335 | err = receive_read_record(drc); | |
03916905 | 3336 | |
30af21b0 PD |
3337 | if (drc->drc_rrd->header.drr_type == DRR_END || err != 0) { |
3338 | kmem_free(drc->drc_rrd, sizeof (*drc->drc_rrd)); | |
3339 | drc->drc_rrd = NULL; | |
03916905 PD |
3340 | break; |
3341 | } | |
3342 | ||
30af21b0 PD |
3343 | bqueue_enqueue(&rwa->q, drc->drc_rrd, |
3344 | sizeof (struct receive_record_arg) + | |
3345 | drc->drc_rrd->payload_size); | |
3346 | drc->drc_rrd = NULL; | |
03916905 | 3347 | } |
30af21b0 PD |
3348 | |
3349 | ASSERT3P(drc->drc_rrd, ==, NULL); | |
3350 | drc->drc_rrd = kmem_zalloc(sizeof (*drc->drc_rrd), KM_SLEEP); | |
3351 | drc->drc_rrd->eos_marker = B_TRUE; | |
3352 | bqueue_enqueue_flush(&rwa->q, drc->drc_rrd, 1); | |
03916905 PD |
3353 | |
3354 | mutex_enter(&rwa->mutex); | |
3355 | while (!rwa->done) { | |
30af21b0 PD |
3356 | /* |
3357 | * We need to use cv_wait_sig() so that any process that may | |
3358 | * be sleeping here can still fork. | |
3359 | */ | |
3360 | (void) cv_wait_sig(&rwa->cv, &rwa->mutex); | |
03916905 PD |
3361 | } |
3362 | mutex_exit(&rwa->mutex); | |
3363 | ||
3364 | /* | |
3365 | * If we are receiving a full stream as a clone, all object IDs which | |
3366 | * are greater than the maximum ID referenced in the stream are | |
3367 | * by definition unused and must be freed. | |
3368 | */ | |
3369 | if (drc->drc_clone && drc->drc_drrb->drr_fromguid == 0) { | |
3370 | uint64_t obj = rwa->max_object + 1; | |
3371 | int free_err = 0; | |
3372 | int next_err = 0; | |
3373 | ||
3374 | while (next_err == 0) { | |
3375 | free_err = dmu_free_long_object(rwa->os, obj); | |
3376 | if (free_err != 0 && free_err != ENOENT) | |
3377 | break; | |
3378 | ||
3379 | next_err = dmu_object_next(rwa->os, &obj, FALSE, 0); | |
3380 | } | |
3381 | ||
3382 | if (err == 0) { | |
3383 | if (free_err != 0 && free_err != ENOENT) | |
3384 | err = free_err; | |
3385 | else if (next_err != ESRCH) | |
3386 | err = next_err; | |
3387 | } | |
3388 | } | |
3389 | ||
3390 | cv_destroy(&rwa->cv); | |
3391 | mutex_destroy(&rwa->mutex); | |
3392 | bqueue_destroy(&rwa->q); | |
7261fc2e | 3393 | list_destroy(&rwa->write_batch); |
03916905 PD |
3394 | if (err == 0) |
3395 | err = rwa->err; | |
3396 | ||
3397 | out: | |
f00ab3f2 TC |
3398 | /* |
3399 | * If we hit an error before we started the receive_writer_thread | |
3400 | * we need to clean up the next_rrd we create by processing the | |
3401 | * DRR_BEGIN record. | |
3402 | */ | |
30af21b0 PD |
3403 | if (drc->drc_next_rrd != NULL) |
3404 | kmem_free(drc->drc_next_rrd, sizeof (*drc->drc_next_rrd)); | |
f00ab3f2 | 3405 | |
ec213971 MA |
3406 | /* |
3407 | * The objset will be invalidated by dmu_recv_end() when we do | |
3408 | * dsl_dataset_clone_swap_sync_impl(). | |
3409 | */ | |
3410 | drc->drc_os = NULL; | |
3411 | ||
30af21b0 PD |
3412 | kmem_free(rwa, sizeof (*rwa)); |
3413 | nvlist_free(drc->drc_begin_nvl); | |
03916905 PD |
3414 | |
3415 | if (err != 0) { | |
3416 | /* | |
3417 | * Clean up references. If receive is not resumable, | |
3418 | * destroy what we created, so we don't leave it in | |
3419 | * the inconsistent state. | |
3420 | */ | |
3421 | dmu_recv_cleanup_ds(drc); | |
3422 | nvlist_free(drc->drc_keynvl); | |
3423 | } | |
3424 | ||
30af21b0 PD |
3425 | objlist_destroy(drc->drc_ignore_objlist); |
3426 | drc->drc_ignore_objlist = NULL; | |
3427 | *voffp = drc->drc_voff; | |
03916905 PD |
3428 | return (err); |
3429 | } | |
3430 | ||
3431 | static int | |
3432 | dmu_recv_end_check(void *arg, dmu_tx_t *tx) | |
3433 | { | |
3434 | dmu_recv_cookie_t *drc = arg; | |
3435 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
3436 | int error; | |
3437 | ||
3438 | ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); | |
3439 | ||
e8cf3a4f AP |
3440 | if (drc->drc_heal) { |
3441 | error = 0; | |
3442 | } else if (!drc->drc_newfs) { | |
03916905 PD |
3443 | dsl_dataset_t *origin_head; |
3444 | ||
3445 | error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); | |
3446 | if (error != 0) | |
3447 | return (error); | |
3448 | if (drc->drc_force) { | |
3449 | /* | |
3450 | * We will destroy any snapshots in tofs (i.e. before | |
3451 | * origin_head) that are after the origin (which is | |
3452 | * the snap before drc_ds, because drc_ds can not | |
3453 | * have any snaps of its own). | |
3454 | */ | |
3455 | uint64_t obj; | |
3456 | ||
3457 | obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
3458 | while (obj != | |
3459 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { | |
3460 | dsl_dataset_t *snap; | |
3461 | error = dsl_dataset_hold_obj(dp, obj, FTAG, | |
3462 | &snap); | |
3463 | if (error != 0) | |
3464 | break; | |
3465 | if (snap->ds_dir != origin_head->ds_dir) | |
3466 | error = SET_ERROR(EINVAL); | |
3467 | if (error == 0) { | |
3468 | error = dsl_destroy_snapshot_check_impl( | |
3469 | snap, B_FALSE); | |
3470 | } | |
3471 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; | |
3472 | dsl_dataset_rele(snap, FTAG); | |
3473 | if (error != 0) | |
3474 | break; | |
3475 | } | |
3476 | if (error != 0) { | |
3477 | dsl_dataset_rele(origin_head, FTAG); | |
3478 | return (error); | |
3479 | } | |
3480 | } | |
3481 | if (drc->drc_keynvl != NULL) { | |
3482 | error = dsl_crypto_recv_raw_key_check(drc->drc_ds, | |
3483 | drc->drc_keynvl, tx); | |
3484 | if (error != 0) { | |
3485 | dsl_dataset_rele(origin_head, FTAG); | |
3486 | return (error); | |
3487 | } | |
3488 | } | |
3489 | ||
3490 | error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, | |
3491 | origin_head, drc->drc_force, drc->drc_owner, tx); | |
3492 | if (error != 0) { | |
3493 | dsl_dataset_rele(origin_head, FTAG); | |
3494 | return (error); | |
3495 | } | |
3496 | error = dsl_dataset_snapshot_check_impl(origin_head, | |
e59a377a MA |
3497 | drc->drc_tosnap, tx, B_TRUE, 1, |
3498 | drc->drc_cred, drc->drc_proc); | |
03916905 PD |
3499 | dsl_dataset_rele(origin_head, FTAG); |
3500 | if (error != 0) | |
3501 | return (error); | |
3502 | ||
3503 | error = dsl_destroy_head_check_impl(drc->drc_ds, 1); | |
3504 | } else { | |
3505 | error = dsl_dataset_snapshot_check_impl(drc->drc_ds, | |
e59a377a MA |
3506 | drc->drc_tosnap, tx, B_TRUE, 1, |
3507 | drc->drc_cred, drc->drc_proc); | |
03916905 PD |
3508 | } |
3509 | return (error); | |
3510 | } | |
3511 | ||
3512 | static void | |
3513 | dmu_recv_end_sync(void *arg, dmu_tx_t *tx) | |
3514 | { | |
3515 | dmu_recv_cookie_t *drc = arg; | |
3516 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
3517 | boolean_t encrypted = drc->drc_ds->ds_dir->dd_crypto_obj != 0; | |
e8cf3a4f | 3518 | uint64_t newsnapobj = 0; |
03916905 PD |
3519 | |
3520 | spa_history_log_internal_ds(drc->drc_ds, "finish receiving", | |
3521 | tx, "snap=%s", drc->drc_tosnap); | |
3522 | drc->drc_ds->ds_objset->os_raw_receive = B_FALSE; | |
3523 | ||
e8cf3a4f AP |
3524 | if (drc->drc_heal) { |
3525 | if (drc->drc_keynvl != NULL) { | |
3526 | nvlist_free(drc->drc_keynvl); | |
3527 | drc->drc_keynvl = NULL; | |
3528 | } | |
3529 | } else if (!drc->drc_newfs) { | |
03916905 PD |
3530 | dsl_dataset_t *origin_head; |
3531 | ||
3532 | VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, | |
3533 | &origin_head)); | |
3534 | ||
3535 | if (drc->drc_force) { | |
3536 | /* | |
3537 | * Destroy any snapshots of drc_tofs (origin_head) | |
3538 | * after the origin (the snap before drc_ds). | |
3539 | */ | |
3540 | uint64_t obj; | |
3541 | ||
3542 | obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
3543 | while (obj != | |
3544 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { | |
3545 | dsl_dataset_t *snap; | |
3546 | VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, | |
3547 | &snap)); | |
3548 | ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); | |
3549 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; | |
3550 | dsl_destroy_snapshot_sync_impl(snap, | |
3551 | B_FALSE, tx); | |
3552 | dsl_dataset_rele(snap, FTAG); | |
3553 | } | |
3554 | } | |
3555 | if (drc->drc_keynvl != NULL) { | |
3556 | dsl_crypto_recv_raw_key_sync(drc->drc_ds, | |
3557 | drc->drc_keynvl, tx); | |
3558 | nvlist_free(drc->drc_keynvl); | |
3559 | drc->drc_keynvl = NULL; | |
3560 | } | |
3561 | ||
30af21b0 PD |
3562 | VERIFY3P(drc->drc_ds->ds_prev, ==, |
3563 | origin_head->ds_prev); | |
03916905 PD |
3564 | |
3565 | dsl_dataset_clone_swap_sync_impl(drc->drc_ds, | |
3566 | origin_head, tx); | |
0fdd6106 MA |
3567 | /* |
3568 | * The objset was evicted by dsl_dataset_clone_swap_sync_impl, | |
3569 | * so drc_os is no longer valid. | |
3570 | */ | |
3571 | drc->drc_os = NULL; | |
3572 | ||
03916905 PD |
3573 | dsl_dataset_snapshot_sync_impl(origin_head, |
3574 | drc->drc_tosnap, tx); | |
3575 | ||
3576 | /* set snapshot's creation time and guid */ | |
3577 | dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); | |
3578 | dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = | |
3579 | drc->drc_drrb->drr_creation_time; | |
3580 | dsl_dataset_phys(origin_head->ds_prev)->ds_guid = | |
3581 | drc->drc_drrb->drr_toguid; | |
3582 | dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= | |
3583 | ~DS_FLAG_INCONSISTENT; | |
3584 | ||
3585 | dmu_buf_will_dirty(origin_head->ds_dbuf, tx); | |
3586 | dsl_dataset_phys(origin_head)->ds_flags &= | |
3587 | ~DS_FLAG_INCONSISTENT; | |
3588 | ||
196bee4c | 3589 | newsnapobj = |
03916905 PD |
3590 | dsl_dataset_phys(origin_head)->ds_prev_snap_obj; |
3591 | ||
3592 | dsl_dataset_rele(origin_head, FTAG); | |
3593 | dsl_destroy_head_sync_impl(drc->drc_ds, tx); | |
3594 | ||
3595 | if (drc->drc_owner != NULL) | |
3596 | VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); | |
3597 | } else { | |
3598 | dsl_dataset_t *ds = drc->drc_ds; | |
3599 | ||
3600 | dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); | |
3601 | ||
3602 | /* set snapshot's creation time and guid */ | |
3603 | dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
3604 | dsl_dataset_phys(ds->ds_prev)->ds_creation_time = | |
3605 | drc->drc_drrb->drr_creation_time; | |
3606 | dsl_dataset_phys(ds->ds_prev)->ds_guid = | |
3607 | drc->drc_drrb->drr_toguid; | |
3608 | dsl_dataset_phys(ds->ds_prev)->ds_flags &= | |
3609 | ~DS_FLAG_INCONSISTENT; | |
3610 | ||
3611 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
3612 | dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; | |
3613 | if (dsl_dataset_has_resume_receive_state(ds)) { | |
3614 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3615 | DS_FIELD_RESUME_FROMGUID, tx); | |
3616 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3617 | DS_FIELD_RESUME_OBJECT, tx); | |
3618 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3619 | DS_FIELD_RESUME_OFFSET, tx); | |
3620 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3621 | DS_FIELD_RESUME_BYTES, tx); | |
3622 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3623 | DS_FIELD_RESUME_TOGUID, tx); | |
3624 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
3625 | DS_FIELD_RESUME_TONAME, tx); | |
30af21b0 PD |
3626 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, |
3627 | DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, tx); | |
03916905 | 3628 | } |
196bee4c | 3629 | newsnapobj = |
03916905 PD |
3630 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; |
3631 | } | |
f00ab3f2 TC |
3632 | |
3633 | /* | |
3634 | * If this is a raw receive, the crypt_keydata nvlist will include | |
3635 | * a to_ivset_guid for us to set on the new snapshot. This value | |
3636 | * will override the value generated by the snapshot code. However, | |
3637 | * this value may not be present, because older implementations of | |
3638 | * the raw send code did not include this value, and we are still | |
3639 | * allowed to receive them if the zfs_disable_ivset_guid_check | |
3640 | * tunable is set, in which case we will leave the newly-generated | |
3641 | * value. | |
3642 | */ | |
e8cf3a4f | 3643 | if (!drc->drc_heal && drc->drc_raw && drc->drc_ivset_guid != 0) { |
196bee4c | 3644 | dmu_object_zapify(dp->dp_meta_objset, newsnapobj, |
f00ab3f2 | 3645 | DMU_OT_DSL_DATASET, tx); |
196bee4c | 3646 | VERIFY0(zap_update(dp->dp_meta_objset, newsnapobj, |
f00ab3f2 TC |
3647 | DS_FIELD_IVSET_GUID, sizeof (uint64_t), 1, |
3648 | &drc->drc_ivset_guid, tx)); | |
3649 | } | |
3650 | ||
03916905 PD |
3651 | /* |
3652 | * Release the hold from dmu_recv_begin. This must be done before | |
3653 | * we return to open context, so that when we free the dataset's dnode | |
3654 | * we can evict its bonus buffer. Since the dataset may be destroyed | |
3655 | * at this point (and therefore won't have a valid pointer to the spa) | |
3656 | * we release the key mapping manually here while we do have a valid | |
3657 | * pointer, if it exists. | |
3658 | */ | |
3659 | if (!drc->drc_raw && encrypted) { | |
3660 | (void) spa_keystore_remove_mapping(dmu_tx_pool(tx)->dp_spa, | |
3661 | drc->drc_ds->ds_object, drc->drc_ds); | |
3662 | } | |
3663 | dsl_dataset_disown(drc->drc_ds, 0, dmu_recv_tag); | |
3664 | drc->drc_ds = NULL; | |
3665 | } | |
3666 | ||
03916905 PD |
3667 | static int dmu_recv_end_modified_blocks = 3; |
3668 | ||
3669 | static int | |
3670 | dmu_recv_existing_end(dmu_recv_cookie_t *drc) | |
3671 | { | |
3672 | #ifdef _KERNEL | |
3673 | /* | |
3674 | * We will be destroying the ds; make sure its origin is unmounted if | |
3675 | * necessary. | |
3676 | */ | |
3677 | char name[ZFS_MAX_DATASET_NAME_LEN]; | |
3678 | dsl_dataset_name(drc->drc_ds, name); | |
3679 | zfs_destroy_unmount_origin(name); | |
3680 | #endif | |
3681 | ||
3682 | return (dsl_sync_task(drc->drc_tofs, | |
3683 | dmu_recv_end_check, dmu_recv_end_sync, drc, | |
3684 | dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); | |
3685 | } | |
3686 | ||
3687 | static int | |
3688 | dmu_recv_new_end(dmu_recv_cookie_t *drc) | |
3689 | { | |
3690 | return (dsl_sync_task(drc->drc_tofs, | |
3691 | dmu_recv_end_check, dmu_recv_end_sync, drc, | |
3692 | dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); | |
3693 | } | |
3694 | ||
3695 | int | |
3696 | dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) | |
3697 | { | |
3698 | int error; | |
3699 | ||
3700 | drc->drc_owner = owner; | |
3701 | ||
3702 | if (drc->drc_newfs) | |
3703 | error = dmu_recv_new_end(drc); | |
3704 | else | |
3705 | error = dmu_recv_existing_end(drc); | |
3706 | ||
3707 | if (error != 0) { | |
3708 | dmu_recv_cleanup_ds(drc); | |
3709 | nvlist_free(drc->drc_keynvl); | |
e8cf3a4f | 3710 | } else if (!drc->drc_heal) { |
ec213971 MA |
3711 | if (drc->drc_newfs) { |
3712 | zvol_create_minor(drc->drc_tofs); | |
3713 | } | |
3714 | char *snapname = kmem_asprintf("%s@%s", | |
3715 | drc->drc_tofs, drc->drc_tosnap); | |
3716 | zvol_create_minor(snapname); | |
3717 | kmem_strfree(snapname); | |
03916905 PD |
3718 | } |
3719 | return (error); | |
3720 | } | |
3721 | ||
3722 | /* | |
3723 | * Return TRUE if this objset is currently being received into. | |
3724 | */ | |
3725 | boolean_t | |
3726 | dmu_objset_is_receiving(objset_t *os) | |
3727 | { | |
3728 | return (os->os_dsl_dataset != NULL && | |
3729 | os->os_dsl_dataset->ds_owner == dmu_recv_tag); | |
3730 | } | |
3731 | ||
fdc2d303 | 3732 | ZFS_MODULE_PARAM(zfs_recv, zfs_recv_, queue_length, UINT, ZMOD_RW, |
03fdcb9a | 3733 | "Maximum receive queue length"); |
30af21b0 | 3734 | |
fdc2d303 | 3735 | ZFS_MODULE_PARAM(zfs_recv, zfs_recv_, queue_ff, UINT, ZMOD_RW, |
03fdcb9a | 3736 | "Receive queue fill fraction"); |
7261fc2e | 3737 | |
fdc2d303 | 3738 | ZFS_MODULE_PARAM(zfs_recv, zfs_recv_, write_batch_size, UINT, ZMOD_RW, |
7261fc2e | 3739 | "Maximum amount of writes to batch into one transaction"); |
e8cf3a4f AP |
3740 | |
3741 | ZFS_MODULE_PARAM(zfs_recv, zfs_recv_, best_effort_corrective, INT, ZMOD_RW, | |
3742 | "Ignore errors during corrective receive"); | |
3743 | /* END CSTYLED */ |