]>
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 | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
23 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. | |
24 | * Copyright (c) 2011, 2015 by Delphix. All rights reserved. | |
25 | * Copyright (c) 2014, Joyent, Inc. All rights reserved. | |
26 | * Copyright 2014 HybridCluster. All rights reserved. | |
27 | * Copyright 2016 RackTop Systems. | |
28 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. | |
d8d418ff | 29 | * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved. |
03916905 PD |
30 | */ |
31 | ||
32 | #include <sys/dmu.h> | |
33 | #include <sys/dmu_impl.h> | |
34 | #include <sys/dmu_tx.h> | |
35 | #include <sys/dbuf.h> | |
36 | #include <sys/dnode.h> | |
37 | #include <sys/zfs_context.h> | |
38 | #include <sys/dmu_objset.h> | |
39 | #include <sys/dmu_traverse.h> | |
40 | #include <sys/dsl_dataset.h> | |
41 | #include <sys/dsl_dir.h> | |
42 | #include <sys/dsl_prop.h> | |
43 | #include <sys/dsl_pool.h> | |
44 | #include <sys/dsl_synctask.h> | |
45 | #include <sys/spa_impl.h> | |
46 | #include <sys/zfs_ioctl.h> | |
47 | #include <sys/zap.h> | |
48 | #include <sys/zio_checksum.h> | |
49 | #include <sys/zfs_znode.h> | |
50 | #include <zfs_fletcher.h> | |
51 | #include <sys/avl.h> | |
52 | #include <sys/ddt.h> | |
53 | #include <sys/zfs_onexit.h> | |
54 | #include <sys/dmu_recv.h> | |
55 | #include <sys/dsl_destroy.h> | |
56 | #include <sys/blkptr.h> | |
57 | #include <sys/dsl_bookmark.h> | |
58 | #include <sys/zfeature.h> | |
59 | #include <sys/bqueue.h> | |
60 | #include <sys/zvol.h> | |
61 | #include <sys/policy.h> | |
62 | ||
63 | int zfs_recv_queue_length = SPA_MAXBLOCKSIZE; | |
64 | ||
65 | static char *dmu_recv_tag = "dmu_recv_tag"; | |
66 | const char *recv_clone_name = "%recv"; | |
67 | ||
68 | static void byteswap_record(dmu_replay_record_t *drr); | |
69 | ||
70 | typedef struct dmu_recv_begin_arg { | |
71 | const char *drba_origin; | |
72 | dmu_recv_cookie_t *drba_cookie; | |
73 | cred_t *drba_cred; | |
74 | dsl_crypto_params_t *drba_dcp; | |
03916905 PD |
75 | } dmu_recv_begin_arg_t; |
76 | ||
77 | static int | |
78 | recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, | |
79 | uint64_t fromguid, uint64_t featureflags) | |
80 | { | |
81 | uint64_t val; | |
d8d418ff | 82 | uint64_t children; |
03916905 PD |
83 | int error; |
84 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
85 | boolean_t encrypted = ds->ds_dir->dd_crypto_obj != 0; | |
86 | boolean_t raw = (featureflags & DMU_BACKUP_FEATURE_RAW) != 0; | |
87 | boolean_t embed = (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) != 0; | |
88 | ||
89 | /* temporary clone name must not exist */ | |
90 | error = zap_lookup(dp->dp_meta_objset, | |
91 | dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, | |
92 | 8, 1, &val); | |
93 | if (error != ENOENT) | |
94 | return (error == 0 ? EBUSY : error); | |
95 | ||
96 | /* new snapshot name must not exist */ | |
97 | error = zap_lookup(dp->dp_meta_objset, | |
98 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, | |
99 | drba->drba_cookie->drc_tosnap, 8, 1, &val); | |
100 | if (error != ENOENT) | |
101 | return (error == 0 ? EEXIST : error); | |
102 | ||
d8d418ff | 103 | /* must not have children if receiving a ZVOL */ |
104 | error = zap_count(dp->dp_meta_objset, | |
105 | dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, &children); | |
106 | if (error != 0) | |
107 | return (error); | |
108 | if (drba->drba_cookie->drc_drrb->drr_type != DMU_OST_ZFS && | |
109 | children > 0) | |
110 | return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); | |
111 | ||
03916905 PD |
112 | /* |
113 | * Check snapshot limit before receiving. We'll recheck again at the | |
114 | * end, but might as well abort before receiving if we're already over | |
115 | * the limit. | |
116 | * | |
117 | * Note that we do not check the file system limit with | |
118 | * dsl_dir_fscount_check because the temporary %clones don't count | |
119 | * against that limit. | |
120 | */ | |
121 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, | |
122 | NULL, drba->drba_cred); | |
123 | if (error != 0) | |
124 | return (error); | |
125 | ||
126 | if (fromguid != 0) { | |
127 | dsl_dataset_t *snap; | |
128 | uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; | |
129 | ||
f00ab3f2 | 130 | /* Can't raw receive on top of an unencrypted dataset */ |
03916905 PD |
131 | if (!encrypted && raw) |
132 | return (SET_ERROR(EINVAL)); | |
133 | ||
134 | /* Encryption is incompatible with embedded data */ | |
135 | if (encrypted && embed) | |
136 | return (SET_ERROR(EINVAL)); | |
137 | ||
138 | /* Find snapshot in this dir that matches fromguid. */ | |
139 | while (obj != 0) { | |
140 | error = dsl_dataset_hold_obj(dp, obj, FTAG, | |
141 | &snap); | |
142 | if (error != 0) | |
143 | return (SET_ERROR(ENODEV)); | |
144 | if (snap->ds_dir != ds->ds_dir) { | |
145 | dsl_dataset_rele(snap, FTAG); | |
146 | return (SET_ERROR(ENODEV)); | |
147 | } | |
148 | if (dsl_dataset_phys(snap)->ds_guid == fromguid) | |
149 | break; | |
150 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; | |
151 | dsl_dataset_rele(snap, FTAG); | |
152 | } | |
153 | if (obj == 0) | |
154 | return (SET_ERROR(ENODEV)); | |
155 | ||
156 | if (drba->drba_cookie->drc_force) { | |
f00ab3f2 | 157 | drba->drba_cookie->drc_fromsnapobj = obj; |
03916905 PD |
158 | } else { |
159 | /* | |
160 | * If we are not forcing, there must be no | |
161 | * changes since fromsnap. | |
162 | */ | |
163 | if (dsl_dataset_modified_since_snap(ds, snap)) { | |
164 | dsl_dataset_rele(snap, FTAG); | |
165 | return (SET_ERROR(ETXTBSY)); | |
166 | } | |
f00ab3f2 TC |
167 | drba->drba_cookie->drc_fromsnapobj = |
168 | ds->ds_prev->ds_object; | |
03916905 PD |
169 | } |
170 | ||
171 | dsl_dataset_rele(snap, FTAG); | |
172 | } else { | |
173 | /* if full, then must be forced */ | |
174 | if (!drba->drba_cookie->drc_force) | |
175 | return (SET_ERROR(EEXIST)); | |
176 | ||
177 | /* | |
178 | * We don't support using zfs recv -F to blow away | |
179 | * encrypted filesystems. This would require the | |
180 | * dsl dir to point to the old encryption key and | |
181 | * the new one at the same time during the receive. | |
182 | */ | |
183 | if ((!encrypted && raw) || encrypted) | |
184 | return (SET_ERROR(EINVAL)); | |
185 | ||
186 | /* | |
187 | * Perform the same encryption checks we would if | |
188 | * we were creating a new dataset from scratch. | |
189 | */ | |
190 | if (!raw) { | |
191 | boolean_t will_encrypt; | |
192 | ||
193 | error = dmu_objset_create_crypt_check( | |
194 | ds->ds_dir->dd_parent, drba->drba_dcp, | |
195 | &will_encrypt); | |
196 | if (error != 0) | |
197 | return (error); | |
198 | ||
199 | if (will_encrypt && embed) | |
200 | return (SET_ERROR(EINVAL)); | |
201 | } | |
202 | ||
f00ab3f2 | 203 | drba->drba_cookie->drc_fromsnapobj = 0; |
03916905 PD |
204 | } |
205 | ||
206 | return (0); | |
207 | ||
208 | } | |
209 | ||
210 | static int | |
211 | dmu_recv_begin_check(void *arg, dmu_tx_t *tx) | |
212 | { | |
213 | dmu_recv_begin_arg_t *drba = arg; | |
214 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
215 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
216 | uint64_t fromguid = drrb->drr_fromguid; | |
217 | int flags = drrb->drr_flags; | |
218 | ds_hold_flags_t dsflags = 0; | |
219 | int error; | |
220 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); | |
221 | dsl_dataset_t *ds; | |
222 | const char *tofs = drba->drba_cookie->drc_tofs; | |
223 | ||
224 | /* already checked */ | |
225 | ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); | |
226 | ASSERT(!(featureflags & DMU_BACKUP_FEATURE_RESUMING)); | |
227 | ||
228 | if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == | |
229 | DMU_COMPOUNDSTREAM || | |
230 | drrb->drr_type >= DMU_OST_NUMTYPES || | |
231 | ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) | |
232 | return (SET_ERROR(EINVAL)); | |
233 | ||
234 | /* Verify pool version supports SA if SA_SPILL feature set */ | |
235 | if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && | |
236 | spa_version(dp->dp_spa) < SPA_VERSION_SA) | |
237 | return (SET_ERROR(ENOTSUP)); | |
238 | ||
239 | if (drba->drba_cookie->drc_resumable && | |
240 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EXTENSIBLE_DATASET)) | |
241 | return (SET_ERROR(ENOTSUP)); | |
242 | ||
243 | /* | |
244 | * The receiving code doesn't know how to translate a WRITE_EMBEDDED | |
245 | * record to a plain WRITE record, so the pool must have the | |
246 | * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED | |
247 | * records. Same with WRITE_EMBEDDED records that use LZ4 compression. | |
248 | */ | |
249 | if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && | |
250 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) | |
251 | return (SET_ERROR(ENOTSUP)); | |
252 | if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && | |
253 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) | |
254 | return (SET_ERROR(ENOTSUP)); | |
255 | ||
256 | /* | |
257 | * The receiving code doesn't know how to translate large blocks | |
258 | * to smaller ones, so the pool must have the LARGE_BLOCKS | |
259 | * feature enabled if the stream has LARGE_BLOCKS. Same with | |
260 | * large dnodes. | |
261 | */ | |
262 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
263 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) | |
264 | return (SET_ERROR(ENOTSUP)); | |
265 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_DNODE) && | |
266 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_DNODE)) | |
267 | return (SET_ERROR(ENOTSUP)); | |
268 | ||
269 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
270 | /* raw receives require the encryption feature */ | |
271 | if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ENCRYPTION)) | |
272 | return (SET_ERROR(ENOTSUP)); | |
273 | ||
274 | /* embedded data is incompatible with encryption and raw recv */ | |
275 | if (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) | |
276 | return (SET_ERROR(EINVAL)); | |
277 | } else { | |
278 | dsflags |= DS_HOLD_FLAG_DECRYPT; | |
279 | } | |
280 | ||
281 | error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); | |
282 | if (error == 0) { | |
283 | /* target fs already exists; recv into temp clone */ | |
284 | ||
285 | /* Can't recv a clone into an existing fs */ | |
286 | if (flags & DRR_FLAG_CLONE || drba->drba_origin) { | |
287 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
288 | return (SET_ERROR(EINVAL)); | |
289 | } | |
290 | ||
291 | error = recv_begin_check_existing_impl(drba, ds, fromguid, | |
292 | featureflags); | |
293 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
294 | } else if (error == ENOENT) { | |
295 | /* target fs does not exist; must be a full backup or clone */ | |
296 | char buf[ZFS_MAX_DATASET_NAME_LEN]; | |
d8d418ff | 297 | objset_t *os; |
03916905 PD |
298 | |
299 | /* | |
300 | * If it's a non-clone incremental, we are missing the | |
301 | * target fs, so fail the recv. | |
302 | */ | |
303 | if (fromguid != 0 && !(flags & DRR_FLAG_CLONE || | |
304 | drba->drba_origin)) | |
305 | return (SET_ERROR(ENOENT)); | |
306 | ||
307 | /* | |
308 | * If we're receiving a full send as a clone, and it doesn't | |
309 | * contain all the necessary free records and freeobject | |
310 | * records, reject it. | |
311 | */ | |
312 | if (fromguid == 0 && drba->drba_origin && | |
313 | !(flags & DRR_FLAG_FREERECORDS)) | |
314 | return (SET_ERROR(EINVAL)); | |
315 | ||
316 | /* Open the parent of tofs */ | |
317 | ASSERT3U(strlen(tofs), <, sizeof (buf)); | |
318 | (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); | |
319 | error = dsl_dataset_hold_flags(dp, buf, dsflags, FTAG, &ds); | |
320 | if (error != 0) | |
321 | return (error); | |
322 | ||
323 | if ((featureflags & DMU_BACKUP_FEATURE_RAW) == 0 && | |
324 | drba->drba_origin == NULL) { | |
325 | boolean_t will_encrypt; | |
326 | ||
327 | /* | |
328 | * Check that we aren't breaking any encryption rules | |
329 | * and that we have all the parameters we need to | |
330 | * create an encrypted dataset if necessary. If we are | |
331 | * making an encrypted dataset the stream can't have | |
332 | * embedded data. | |
333 | */ | |
334 | error = dmu_objset_create_crypt_check(ds->ds_dir, | |
335 | drba->drba_dcp, &will_encrypt); | |
336 | if (error != 0) { | |
337 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
338 | return (error); | |
339 | } | |
340 | ||
341 | if (will_encrypt && | |
342 | (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)) { | |
343 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
344 | return (SET_ERROR(EINVAL)); | |
345 | } | |
346 | } | |
347 | ||
348 | /* | |
349 | * Check filesystem and snapshot limits before receiving. We'll | |
350 | * recheck snapshot limits again at the end (we create the | |
351 | * filesystems and increment those counts during begin_sync). | |
352 | */ | |
353 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, | |
354 | ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred); | |
355 | if (error != 0) { | |
356 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
357 | return (error); | |
358 | } | |
359 | ||
360 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, | |
361 | ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); | |
362 | if (error != 0) { | |
363 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
364 | return (error); | |
365 | } | |
366 | ||
d8d418ff | 367 | /* can't recv below anything but filesystems (eg. no ZVOLs) */ |
368 | error = dmu_objset_from_ds(ds, &os); | |
369 | if (error != 0) { | |
370 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
371 | return (error); | |
372 | } | |
373 | if (dmu_objset_type(os) != DMU_OST_ZFS) { | |
374 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
375 | return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); | |
376 | } | |
377 | ||
03916905 PD |
378 | if (drba->drba_origin != NULL) { |
379 | dsl_dataset_t *origin; | |
380 | ||
381 | error = dsl_dataset_hold_flags(dp, drba->drba_origin, | |
382 | dsflags, FTAG, &origin); | |
383 | if (error != 0) { | |
384 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
385 | return (error); | |
386 | } | |
387 | if (!origin->ds_is_snapshot) { | |
388 | dsl_dataset_rele_flags(origin, dsflags, FTAG); | |
389 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
390 | return (SET_ERROR(EINVAL)); | |
391 | } | |
392 | if (dsl_dataset_phys(origin)->ds_guid != fromguid && | |
393 | fromguid != 0) { | |
394 | dsl_dataset_rele_flags(origin, dsflags, FTAG); | |
395 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
396 | return (SET_ERROR(ENODEV)); | |
397 | } | |
398 | if (origin->ds_dir->dd_crypto_obj != 0 && | |
399 | (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)) { | |
400 | dsl_dataset_rele_flags(origin, dsflags, FTAG); | |
401 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
402 | return (SET_ERROR(EINVAL)); | |
403 | } | |
404 | dsl_dataset_rele_flags(origin, | |
405 | dsflags, FTAG); | |
406 | } | |
d8d418ff | 407 | |
03916905 PD |
408 | dsl_dataset_rele_flags(ds, dsflags, FTAG); |
409 | error = 0; | |
410 | } | |
411 | return (error); | |
412 | } | |
413 | ||
414 | static void | |
415 | dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) | |
416 | { | |
417 | dmu_recv_begin_arg_t *drba = arg; | |
418 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
419 | objset_t *mos = dp->dp_meta_objset; | |
420 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
421 | const char *tofs = drba->drba_cookie->drc_tofs; | |
422 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); | |
423 | dsl_dataset_t *ds, *newds; | |
424 | objset_t *os; | |
425 | uint64_t dsobj; | |
426 | ds_hold_flags_t dsflags = 0; | |
427 | int error; | |
428 | uint64_t crflags = 0; | |
429 | dsl_crypto_params_t dummy_dcp = { 0 }; | |
430 | dsl_crypto_params_t *dcp = drba->drba_dcp; | |
431 | ||
432 | if (drrb->drr_flags & DRR_FLAG_CI_DATA) | |
433 | crflags |= DS_FLAG_CI_DATASET; | |
434 | ||
435 | if ((featureflags & DMU_BACKUP_FEATURE_RAW) == 0) | |
436 | dsflags |= DS_HOLD_FLAG_DECRYPT; | |
437 | ||
438 | /* | |
439 | * Raw, non-incremental recvs always use a dummy dcp with | |
440 | * the raw cmd set. Raw incremental recvs do not use a dcp | |
441 | * since the encryption parameters are already set in stone. | |
442 | */ | |
f00ab3f2 | 443 | if (dcp == NULL && drba->drba_cookie->drc_fromsnapobj == 0 && |
03916905 PD |
444 | drba->drba_origin == NULL) { |
445 | ASSERT3P(dcp, ==, NULL); | |
446 | dcp = &dummy_dcp; | |
447 | ||
448 | if (featureflags & DMU_BACKUP_FEATURE_RAW) | |
449 | dcp->cp_cmd = DCP_CMD_RAW_RECV; | |
450 | } | |
451 | ||
452 | error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); | |
453 | if (error == 0) { | |
454 | /* create temporary clone */ | |
455 | dsl_dataset_t *snap = NULL; | |
456 | ||
f00ab3f2 | 457 | if (drba->drba_cookie->drc_fromsnapobj != 0) { |
03916905 | 458 | VERIFY0(dsl_dataset_hold_obj(dp, |
f00ab3f2 | 459 | drba->drba_cookie->drc_fromsnapobj, FTAG, &snap)); |
03916905 PD |
460 | ASSERT3P(dcp, ==, NULL); |
461 | } | |
462 | ||
463 | dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, | |
464 | snap, crflags, drba->drba_cred, dcp, tx); | |
f00ab3f2 | 465 | if (drba->drba_cookie->drc_fromsnapobj != 0) |
03916905 PD |
466 | dsl_dataset_rele(snap, FTAG); |
467 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
468 | } else { | |
469 | dsl_dir_t *dd; | |
470 | const char *tail; | |
471 | dsl_dataset_t *origin = NULL; | |
472 | ||
473 | VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); | |
474 | ||
475 | if (drba->drba_origin != NULL) { | |
476 | VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, | |
477 | FTAG, &origin)); | |
478 | ASSERT3P(dcp, ==, NULL); | |
479 | } | |
480 | ||
481 | /* Create new dataset. */ | |
482 | dsobj = dsl_dataset_create_sync(dd, strrchr(tofs, '/') + 1, | |
483 | origin, crflags, drba->drba_cred, dcp, tx); | |
484 | if (origin != NULL) | |
485 | dsl_dataset_rele(origin, FTAG); | |
486 | dsl_dir_rele(dd, FTAG); | |
487 | drba->drba_cookie->drc_newfs = B_TRUE; | |
488 | } | |
489 | ||
490 | VERIFY0(dsl_dataset_own_obj(dp, dsobj, dsflags, dmu_recv_tag, &newds)); | |
491 | VERIFY0(dmu_objset_from_ds(newds, &os)); | |
492 | ||
493 | if (drba->drba_cookie->drc_resumable) { | |
494 | dsl_dataset_zapify(newds, tx); | |
495 | if (drrb->drr_fromguid != 0) { | |
496 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_FROMGUID, | |
497 | 8, 1, &drrb->drr_fromguid, tx)); | |
498 | } | |
499 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TOGUID, | |
500 | 8, 1, &drrb->drr_toguid, tx)); | |
501 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TONAME, | |
502 | 1, strlen(drrb->drr_toname) + 1, drrb->drr_toname, tx)); | |
503 | uint64_t one = 1; | |
504 | uint64_t zero = 0; | |
505 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OBJECT, | |
506 | 8, 1, &one, tx)); | |
507 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OFFSET, | |
508 | 8, 1, &zero, tx)); | |
509 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_BYTES, | |
510 | 8, 1, &zero, tx)); | |
511 | if (featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) { | |
512 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_LARGEBLOCK, | |
513 | 8, 1, &one, tx)); | |
514 | } | |
515 | if (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) { | |
516 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_EMBEDOK, | |
517 | 8, 1, &one, tx)); | |
518 | } | |
519 | if (featureflags & DMU_BACKUP_FEATURE_COMPRESSED) { | |
520 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_COMPRESSOK, | |
521 | 8, 1, &one, tx)); | |
522 | } | |
523 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
524 | VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_RAWOK, | |
525 | 8, 1, &one, tx)); | |
526 | } | |
527 | } | |
528 | ||
529 | /* | |
530 | * Usually the os->os_encrypted value is tied to the presence of a | |
531 | * DSL Crypto Key object in the dd. However, that will not be received | |
532 | * until dmu_recv_stream(), so we set the value manually for now. | |
533 | */ | |
534 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
535 | os->os_encrypted = B_TRUE; | |
536 | drba->drba_cookie->drc_raw = B_TRUE; | |
537 | } | |
538 | ||
539 | dmu_buf_will_dirty(newds->ds_dbuf, tx); | |
540 | dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; | |
541 | ||
542 | /* | |
543 | * If we actually created a non-clone, we need to create the objset | |
544 | * in our new dataset. If this is a raw send we postpone this until | |
545 | * dmu_recv_stream() so that we can allocate the metadnode with the | |
546 | * properties from the DRR_BEGIN payload. | |
547 | */ | |
548 | rrw_enter(&newds->ds_bp_rwlock, RW_READER, FTAG); | |
549 | if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds)) && | |
550 | (featureflags & DMU_BACKUP_FEATURE_RAW) == 0) { | |
551 | (void) dmu_objset_create_impl(dp->dp_spa, | |
552 | newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); | |
553 | } | |
554 | rrw_exit(&newds->ds_bp_rwlock, FTAG); | |
555 | ||
556 | drba->drba_cookie->drc_ds = newds; | |
557 | ||
558 | spa_history_log_internal_ds(newds, "receive", tx, ""); | |
559 | } | |
560 | ||
561 | static int | |
562 | dmu_recv_resume_begin_check(void *arg, dmu_tx_t *tx) | |
563 | { | |
564 | dmu_recv_begin_arg_t *drba = arg; | |
565 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
566 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
567 | int error; | |
568 | ds_hold_flags_t dsflags = 0; | |
569 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); | |
570 | dsl_dataset_t *ds; | |
571 | const char *tofs = drba->drba_cookie->drc_tofs; | |
572 | ||
573 | /* already checked */ | |
574 | ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); | |
575 | ASSERT(featureflags & DMU_BACKUP_FEATURE_RESUMING); | |
576 | ||
577 | if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == | |
578 | DMU_COMPOUNDSTREAM || | |
579 | drrb->drr_type >= DMU_OST_NUMTYPES) | |
580 | return (SET_ERROR(EINVAL)); | |
581 | ||
582 | /* Verify pool version supports SA if SA_SPILL feature set */ | |
583 | if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && | |
584 | spa_version(dp->dp_spa) < SPA_VERSION_SA) | |
585 | return (SET_ERROR(ENOTSUP)); | |
586 | ||
587 | /* | |
588 | * The receiving code doesn't know how to translate a WRITE_EMBEDDED | |
589 | * record to a plain WRITE record, so the pool must have the | |
590 | * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED | |
591 | * records. Same with WRITE_EMBEDDED records that use LZ4 compression. | |
592 | */ | |
593 | if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && | |
594 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) | |
595 | return (SET_ERROR(ENOTSUP)); | |
596 | if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && | |
597 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) | |
598 | return (SET_ERROR(ENOTSUP)); | |
599 | ||
600 | /* | |
601 | * The receiving code doesn't know how to translate large blocks | |
602 | * to smaller ones, so the pool must have the LARGE_BLOCKS | |
603 | * feature enabled if the stream has LARGE_BLOCKS. Same with | |
604 | * large dnodes. | |
605 | */ | |
606 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
607 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) | |
608 | return (SET_ERROR(ENOTSUP)); | |
609 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_DNODE) && | |
610 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_DNODE)) | |
611 | return (SET_ERROR(ENOTSUP)); | |
612 | ||
613 | /* 6 extra bytes for /%recv */ | |
614 | char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; | |
615 | (void) snprintf(recvname, sizeof (recvname), "%s/%s", | |
616 | tofs, recv_clone_name); | |
617 | ||
618 | if ((featureflags & DMU_BACKUP_FEATURE_RAW) == 0) | |
619 | dsflags |= DS_HOLD_FLAG_DECRYPT; | |
620 | ||
621 | if (dsl_dataset_hold_flags(dp, recvname, dsflags, FTAG, &ds) != 0) { | |
622 | /* %recv does not exist; continue in tofs */ | |
623 | error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); | |
624 | if (error != 0) | |
625 | return (error); | |
626 | } | |
627 | ||
628 | /* check that ds is marked inconsistent */ | |
629 | if (!DS_IS_INCONSISTENT(ds)) { | |
630 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
631 | return (SET_ERROR(EINVAL)); | |
632 | } | |
633 | ||
634 | /* check that there is resuming data, and that the toguid matches */ | |
635 | if (!dsl_dataset_is_zapified(ds)) { | |
636 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
637 | return (SET_ERROR(EINVAL)); | |
638 | } | |
639 | uint64_t val; | |
640 | error = zap_lookup(dp->dp_meta_objset, ds->ds_object, | |
641 | DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val); | |
642 | if (error != 0 || drrb->drr_toguid != val) { | |
643 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
644 | return (SET_ERROR(EINVAL)); | |
645 | } | |
646 | ||
647 | /* | |
648 | * Check if the receive is still running. If so, it will be owned. | |
649 | * Note that nothing else can own the dataset (e.g. after the receive | |
650 | * fails) because it will be marked inconsistent. | |
651 | */ | |
652 | if (dsl_dataset_has_owner(ds)) { | |
653 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
654 | return (SET_ERROR(EBUSY)); | |
655 | } | |
656 | ||
657 | /* There should not be any snapshots of this fs yet. */ | |
658 | if (ds->ds_prev != NULL && ds->ds_prev->ds_dir == ds->ds_dir) { | |
659 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
660 | return (SET_ERROR(EINVAL)); | |
661 | } | |
662 | ||
663 | /* | |
664 | * Note: resume point will be checked when we process the first WRITE | |
665 | * record. | |
666 | */ | |
667 | ||
668 | /* check that the origin matches */ | |
669 | val = 0; | |
670 | (void) zap_lookup(dp->dp_meta_objset, ds->ds_object, | |
671 | DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val); | |
672 | if (drrb->drr_fromguid != val) { | |
673 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
674 | return (SET_ERROR(EINVAL)); | |
675 | } | |
676 | ||
677 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
678 | return (0); | |
679 | } | |
680 | ||
681 | static void | |
682 | dmu_recv_resume_begin_sync(void *arg, dmu_tx_t *tx) | |
683 | { | |
684 | dmu_recv_begin_arg_t *drba = arg; | |
685 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
686 | const char *tofs = drba->drba_cookie->drc_tofs; | |
687 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
688 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); | |
689 | dsl_dataset_t *ds; | |
690 | objset_t *os; | |
691 | ds_hold_flags_t dsflags = 0; | |
692 | uint64_t dsobj; | |
693 | /* 6 extra bytes for /%recv */ | |
694 | char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; | |
695 | ||
696 | (void) snprintf(recvname, sizeof (recvname), "%s/%s", | |
697 | tofs, recv_clone_name); | |
698 | ||
699 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
700 | drba->drba_cookie->drc_raw = B_TRUE; | |
701 | } else { | |
702 | dsflags |= DS_HOLD_FLAG_DECRYPT; | |
703 | } | |
704 | ||
705 | if (dsl_dataset_hold_flags(dp, recvname, dsflags, FTAG, &ds) != 0) { | |
706 | /* %recv does not exist; continue in tofs */ | |
707 | VERIFY0(dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds)); | |
708 | drba->drba_cookie->drc_newfs = B_TRUE; | |
709 | } | |
710 | ||
711 | /* clear the inconsistent flag so that we can own it */ | |
712 | ASSERT(DS_IS_INCONSISTENT(ds)); | |
713 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
714 | dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; | |
715 | dsobj = ds->ds_object; | |
716 | dsl_dataset_rele_flags(ds, dsflags, FTAG); | |
717 | ||
718 | VERIFY0(dsl_dataset_own_obj(dp, dsobj, dsflags, dmu_recv_tag, &ds)); | |
719 | VERIFY0(dmu_objset_from_ds(ds, &os)); | |
720 | ||
721 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
722 | dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT; | |
723 | ||
724 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); | |
725 | ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds)) || | |
726 | drba->drba_cookie->drc_raw); | |
727 | rrw_exit(&ds->ds_bp_rwlock, FTAG); | |
728 | ||
729 | drba->drba_cookie->drc_ds = ds; | |
730 | ||
731 | spa_history_log_internal_ds(ds, "resume receive", tx, ""); | |
732 | } | |
733 | ||
734 | /* | |
735 | * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() | |
736 | * succeeds; otherwise we will leak the holds on the datasets. | |
737 | */ | |
738 | int | |
739 | dmu_recv_begin(char *tofs, char *tosnap, dmu_replay_record_t *drr_begin, | |
740 | boolean_t force, boolean_t resumable, nvlist_t *localprops, | |
741 | nvlist_t *hidden_args, char *origin, dmu_recv_cookie_t *drc) | |
742 | { | |
743 | dmu_recv_begin_arg_t drba = { 0 }; | |
744 | ||
745 | bzero(drc, sizeof (dmu_recv_cookie_t)); | |
746 | drc->drc_drr_begin = drr_begin; | |
747 | drc->drc_drrb = &drr_begin->drr_u.drr_begin; | |
748 | drc->drc_tosnap = tosnap; | |
749 | drc->drc_tofs = tofs; | |
750 | drc->drc_force = force; | |
751 | drc->drc_resumable = resumable; | |
752 | drc->drc_cred = CRED(); | |
753 | drc->drc_clone = (origin != NULL); | |
754 | ||
755 | if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { | |
756 | drc->drc_byteswap = B_TRUE; | |
757 | (void) fletcher_4_incremental_byteswap(drr_begin, | |
758 | sizeof (dmu_replay_record_t), &drc->drc_cksum); | |
759 | byteswap_record(drr_begin); | |
760 | } else if (drc->drc_drrb->drr_magic == DMU_BACKUP_MAGIC) { | |
761 | (void) fletcher_4_incremental_native(drr_begin, | |
762 | sizeof (dmu_replay_record_t), &drc->drc_cksum); | |
763 | } else { | |
764 | return (SET_ERROR(EINVAL)); | |
765 | } | |
766 | ||
767 | drba.drba_origin = origin; | |
768 | drba.drba_cookie = drc; | |
769 | drba.drba_cred = CRED(); | |
770 | ||
771 | if (DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & | |
772 | DMU_BACKUP_FEATURE_RESUMING) { | |
773 | return (dsl_sync_task(tofs, | |
774 | dmu_recv_resume_begin_check, dmu_recv_resume_begin_sync, | |
775 | &drba, 5, ZFS_SPACE_CHECK_NORMAL)); | |
776 | } else { | |
777 | int err; | |
778 | ||
779 | /* | |
780 | * For non-raw, non-incremental, non-resuming receives the | |
781 | * user can specify encryption parameters on the command line | |
782 | * with "zfs recv -o". For these receives we create a dcp and | |
783 | * pass it to the sync task. Creating the dcp will implicitly | |
784 | * remove the encryption params from the localprops nvlist, | |
785 | * which avoids errors when trying to set these normally | |
786 | * read-only properties. Any other kind of receive that | |
787 | * attempts to set these properties will fail as a result. | |
788 | */ | |
789 | if ((DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & | |
790 | DMU_BACKUP_FEATURE_RAW) == 0 && | |
791 | origin == NULL && drc->drc_drrb->drr_fromguid == 0) { | |
792 | err = dsl_crypto_params_create_nvlist(DCP_CMD_NONE, | |
793 | localprops, hidden_args, &drba.drba_dcp); | |
794 | if (err != 0) | |
795 | return (err); | |
796 | } | |
797 | ||
798 | err = dsl_sync_task(tofs, | |
799 | dmu_recv_begin_check, dmu_recv_begin_sync, | |
800 | &drba, 5, ZFS_SPACE_CHECK_NORMAL); | |
801 | dsl_crypto_params_free(drba.drba_dcp, !!err); | |
802 | ||
803 | return (err); | |
804 | } | |
805 | } | |
806 | ||
807 | struct receive_record_arg { | |
808 | dmu_replay_record_t header; | |
809 | void *payload; /* Pointer to a buffer containing the payload */ | |
810 | /* | |
811 | * If the record is a write, pointer to the arc_buf_t containing the | |
812 | * payload. | |
813 | */ | |
814 | arc_buf_t *arc_buf; | |
815 | int payload_size; | |
816 | uint64_t bytes_read; /* bytes read from stream when record created */ | |
817 | boolean_t eos_marker; /* Marks the end of the stream */ | |
818 | bqueue_node_t node; | |
819 | }; | |
820 | ||
821 | struct receive_writer_arg { | |
822 | objset_t *os; | |
823 | boolean_t byteswap; | |
824 | bqueue_t q; | |
825 | ||
826 | /* | |
827 | * These three args are used to signal to the main thread that we're | |
828 | * done. | |
829 | */ | |
830 | kmutex_t mutex; | |
831 | kcondvar_t cv; | |
832 | boolean_t done; | |
833 | ||
834 | int err; | |
835 | /* A map from guid to dataset to help handle dedup'd streams. */ | |
836 | avl_tree_t *guid_to_ds_map; | |
837 | boolean_t resumable; | |
838 | boolean_t raw; | |
839 | uint64_t last_object; | |
840 | uint64_t last_offset; | |
841 | uint64_t max_object; /* highest object ID referenced in stream */ | |
842 | uint64_t bytes_read; /* bytes read when current record created */ | |
843 | ||
844 | /* Encryption parameters for the last received DRR_OBJECT_RANGE */ | |
845 | boolean_t or_crypt_params_present; | |
846 | uint64_t or_firstobj; | |
847 | uint64_t or_numslots; | |
848 | uint8_t or_salt[ZIO_DATA_SALT_LEN]; | |
849 | uint8_t or_iv[ZIO_DATA_IV_LEN]; | |
850 | uint8_t or_mac[ZIO_DATA_MAC_LEN]; | |
851 | boolean_t or_byteorder; | |
852 | }; | |
853 | ||
854 | struct objlist { | |
855 | list_t list; /* List of struct receive_objnode. */ | |
856 | /* | |
857 | * Last object looked up. Used to assert that objects are being looked | |
858 | * up in ascending order. | |
859 | */ | |
860 | uint64_t last_lookup; | |
861 | }; | |
862 | ||
863 | struct receive_objnode { | |
864 | list_node_t node; | |
865 | uint64_t object; | |
866 | }; | |
867 | ||
868 | struct receive_arg { | |
869 | objset_t *os; | |
870 | vnode_t *vp; /* The vnode to read the stream from */ | |
871 | uint64_t voff; /* The current offset in the stream */ | |
872 | uint64_t bytes_read; | |
873 | /* | |
874 | * A record that has had its payload read in, but hasn't yet been handed | |
875 | * off to the worker thread. | |
876 | */ | |
877 | struct receive_record_arg *rrd; | |
878 | /* A record that has had its header read in, but not its payload. */ | |
879 | struct receive_record_arg *next_rrd; | |
880 | zio_cksum_t cksum; | |
881 | zio_cksum_t prev_cksum; | |
882 | int err; | |
883 | boolean_t byteswap; | |
884 | boolean_t raw; | |
885 | uint64_t featureflags; | |
886 | /* Sorted list of objects not to issue prefetches for. */ | |
887 | struct objlist ignore_objlist; | |
888 | }; | |
889 | ||
890 | typedef struct guid_map_entry { | |
891 | uint64_t guid; | |
892 | boolean_t raw; | |
893 | dsl_dataset_t *gme_ds; | |
894 | avl_node_t avlnode; | |
895 | } guid_map_entry_t; | |
896 | ||
897 | static int | |
898 | guid_compare(const void *arg1, const void *arg2) | |
899 | { | |
900 | const guid_map_entry_t *gmep1 = (const guid_map_entry_t *)arg1; | |
901 | const guid_map_entry_t *gmep2 = (const guid_map_entry_t *)arg2; | |
902 | ||
903 | return (AVL_CMP(gmep1->guid, gmep2->guid)); | |
904 | } | |
905 | ||
906 | static void | |
907 | free_guid_map_onexit(void *arg) | |
908 | { | |
909 | avl_tree_t *ca = arg; | |
910 | void *cookie = NULL; | |
911 | guid_map_entry_t *gmep; | |
912 | ||
913 | while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { | |
914 | ds_hold_flags_t dsflags = DS_HOLD_FLAG_DECRYPT; | |
915 | ||
916 | if (gmep->raw) { | |
917 | gmep->gme_ds->ds_objset->os_raw_receive = B_FALSE; | |
918 | dsflags &= ~DS_HOLD_FLAG_DECRYPT; | |
919 | } | |
920 | ||
921 | dsl_dataset_disown(gmep->gme_ds, dsflags, gmep); | |
922 | kmem_free(gmep, sizeof (guid_map_entry_t)); | |
923 | } | |
924 | avl_destroy(ca); | |
925 | kmem_free(ca, sizeof (avl_tree_t)); | |
926 | } | |
927 | ||
928 | static int | |
929 | receive_read(struct receive_arg *ra, int len, void *buf) | |
930 | { | |
931 | int done = 0; | |
932 | ||
933 | /* | |
934 | * The code doesn't rely on this (lengths being multiples of 8). See | |
935 | * comment in dump_bytes. | |
936 | */ | |
937 | ASSERT(len % 8 == 0 || | |
938 | (ra->featureflags & DMU_BACKUP_FEATURE_RAW) != 0); | |
939 | ||
940 | while (done < len) { | |
941 | ssize_t resid; | |
942 | ||
943 | ra->err = vn_rdwr(UIO_READ, ra->vp, | |
944 | (char *)buf + done, len - done, | |
945 | ra->voff, UIO_SYSSPACE, FAPPEND, | |
946 | RLIM64_INFINITY, CRED(), &resid); | |
947 | ||
948 | if (resid == len - done) { | |
949 | /* | |
950 | * Note: ECKSUM indicates that the receive | |
951 | * was interrupted and can potentially be resumed. | |
952 | */ | |
953 | ra->err = SET_ERROR(ECKSUM); | |
954 | } | |
955 | ra->voff += len - done - resid; | |
956 | done = len - resid; | |
957 | if (ra->err != 0) | |
958 | return (ra->err); | |
959 | } | |
960 | ||
961 | ra->bytes_read += len; | |
962 | ||
963 | ASSERT3U(done, ==, len); | |
964 | return (0); | |
965 | } | |
966 | ||
967 | noinline static void | |
968 | byteswap_record(dmu_replay_record_t *drr) | |
969 | { | |
970 | #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) | |
971 | #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) | |
972 | drr->drr_type = BSWAP_32(drr->drr_type); | |
973 | drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); | |
974 | ||
975 | switch (drr->drr_type) { | |
976 | case DRR_BEGIN: | |
977 | DO64(drr_begin.drr_magic); | |
978 | DO64(drr_begin.drr_versioninfo); | |
979 | DO64(drr_begin.drr_creation_time); | |
980 | DO32(drr_begin.drr_type); | |
981 | DO32(drr_begin.drr_flags); | |
982 | DO64(drr_begin.drr_toguid); | |
983 | DO64(drr_begin.drr_fromguid); | |
984 | break; | |
985 | case DRR_OBJECT: | |
986 | DO64(drr_object.drr_object); | |
987 | DO32(drr_object.drr_type); | |
988 | DO32(drr_object.drr_bonustype); | |
989 | DO32(drr_object.drr_blksz); | |
990 | DO32(drr_object.drr_bonuslen); | |
991 | DO32(drr_object.drr_raw_bonuslen); | |
992 | DO64(drr_object.drr_toguid); | |
993 | DO64(drr_object.drr_maxblkid); | |
994 | break; | |
995 | case DRR_FREEOBJECTS: | |
996 | DO64(drr_freeobjects.drr_firstobj); | |
997 | DO64(drr_freeobjects.drr_numobjs); | |
998 | DO64(drr_freeobjects.drr_toguid); | |
999 | break; | |
1000 | case DRR_WRITE: | |
1001 | DO64(drr_write.drr_object); | |
1002 | DO32(drr_write.drr_type); | |
1003 | DO64(drr_write.drr_offset); | |
1004 | DO64(drr_write.drr_logical_size); | |
1005 | DO64(drr_write.drr_toguid); | |
1006 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write.drr_key.ddk_cksum); | |
1007 | DO64(drr_write.drr_key.ddk_prop); | |
1008 | DO64(drr_write.drr_compressed_size); | |
1009 | break; | |
1010 | case DRR_WRITE_BYREF: | |
1011 | DO64(drr_write_byref.drr_object); | |
1012 | DO64(drr_write_byref.drr_offset); | |
1013 | DO64(drr_write_byref.drr_length); | |
1014 | DO64(drr_write_byref.drr_toguid); | |
1015 | DO64(drr_write_byref.drr_refguid); | |
1016 | DO64(drr_write_byref.drr_refobject); | |
1017 | DO64(drr_write_byref.drr_refoffset); | |
1018 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write_byref. | |
1019 | drr_key.ddk_cksum); | |
1020 | DO64(drr_write_byref.drr_key.ddk_prop); | |
1021 | break; | |
1022 | case DRR_WRITE_EMBEDDED: | |
1023 | DO64(drr_write_embedded.drr_object); | |
1024 | DO64(drr_write_embedded.drr_offset); | |
1025 | DO64(drr_write_embedded.drr_length); | |
1026 | DO64(drr_write_embedded.drr_toguid); | |
1027 | DO32(drr_write_embedded.drr_lsize); | |
1028 | DO32(drr_write_embedded.drr_psize); | |
1029 | break; | |
1030 | case DRR_FREE: | |
1031 | DO64(drr_free.drr_object); | |
1032 | DO64(drr_free.drr_offset); | |
1033 | DO64(drr_free.drr_length); | |
1034 | DO64(drr_free.drr_toguid); | |
1035 | break; | |
1036 | case DRR_SPILL: | |
1037 | DO64(drr_spill.drr_object); | |
1038 | DO64(drr_spill.drr_length); | |
1039 | DO64(drr_spill.drr_toguid); | |
1040 | DO64(drr_spill.drr_compressed_size); | |
1041 | DO32(drr_spill.drr_type); | |
1042 | break; | |
1043 | case DRR_OBJECT_RANGE: | |
1044 | DO64(drr_object_range.drr_firstobj); | |
1045 | DO64(drr_object_range.drr_numslots); | |
1046 | DO64(drr_object_range.drr_toguid); | |
1047 | break; | |
1048 | case DRR_END: | |
1049 | DO64(drr_end.drr_toguid); | |
1050 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_end.drr_checksum); | |
1051 | break; | |
1052 | default: | |
1053 | break; | |
1054 | } | |
1055 | ||
1056 | if (drr->drr_type != DRR_BEGIN) { | |
1057 | ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_checksum.drr_checksum); | |
1058 | } | |
1059 | ||
1060 | #undef DO64 | |
1061 | #undef DO32 | |
1062 | } | |
1063 | ||
1064 | static inline uint8_t | |
1065 | deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) | |
1066 | { | |
1067 | if (bonus_type == DMU_OT_SA) { | |
1068 | return (1); | |
1069 | } else { | |
1070 | return (1 + | |
1071 | ((DN_OLD_MAX_BONUSLEN - | |
1072 | MIN(DN_OLD_MAX_BONUSLEN, bonus_size)) >> SPA_BLKPTRSHIFT)); | |
1073 | } | |
1074 | } | |
1075 | ||
1076 | static void | |
1077 | save_resume_state(struct receive_writer_arg *rwa, | |
1078 | uint64_t object, uint64_t offset, dmu_tx_t *tx) | |
1079 | { | |
1080 | int txgoff = dmu_tx_get_txg(tx) & TXG_MASK; | |
1081 | ||
1082 | if (!rwa->resumable) | |
1083 | return; | |
1084 | ||
1085 | /* | |
1086 | * We use ds_resume_bytes[] != 0 to indicate that we need to | |
1087 | * update this on disk, so it must not be 0. | |
1088 | */ | |
1089 | ASSERT(rwa->bytes_read != 0); | |
1090 | ||
1091 | /* | |
1092 | * We only resume from write records, which have a valid | |
1093 | * (non-meta-dnode) object number. | |
1094 | */ | |
1095 | ASSERT(object != 0); | |
1096 | ||
1097 | /* | |
1098 | * For resuming to work correctly, we must receive records in order, | |
1099 | * sorted by object,offset. This is checked by the callers, but | |
1100 | * assert it here for good measure. | |
1101 | */ | |
1102 | ASSERT3U(object, >=, rwa->os->os_dsl_dataset->ds_resume_object[txgoff]); | |
1103 | ASSERT(object != rwa->os->os_dsl_dataset->ds_resume_object[txgoff] || | |
1104 | offset >= rwa->os->os_dsl_dataset->ds_resume_offset[txgoff]); | |
1105 | ASSERT3U(rwa->bytes_read, >=, | |
1106 | rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff]); | |
1107 | ||
1108 | rwa->os->os_dsl_dataset->ds_resume_object[txgoff] = object; | |
1109 | rwa->os->os_dsl_dataset->ds_resume_offset[txgoff] = offset; | |
1110 | rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff] = rwa->bytes_read; | |
1111 | } | |
1112 | ||
1113 | noinline static int | |
1114 | receive_object(struct receive_writer_arg *rwa, struct drr_object *drro, | |
1115 | void *data) | |
1116 | { | |
1117 | dmu_object_info_t doi; | |
1118 | dmu_tx_t *tx; | |
1119 | uint64_t object; | |
1120 | int err; | |
1121 | uint8_t dn_slots = drro->drr_dn_slots != 0 ? | |
1122 | drro->drr_dn_slots : DNODE_MIN_SLOTS; | |
1123 | ||
1124 | if (drro->drr_type == DMU_OT_NONE || | |
1125 | !DMU_OT_IS_VALID(drro->drr_type) || | |
1126 | !DMU_OT_IS_VALID(drro->drr_bonustype) || | |
1127 | drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || | |
1128 | drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || | |
1129 | P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || | |
1130 | drro->drr_blksz < SPA_MINBLOCKSIZE || | |
1131 | drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(rwa->os)) || | |
1132 | drro->drr_bonuslen > | |
1133 | DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa->os))) || | |
1134 | dn_slots > | |
1135 | (spa_maxdnodesize(dmu_objset_spa(rwa->os)) >> DNODE_SHIFT)) { | |
1136 | return (SET_ERROR(EINVAL)); | |
1137 | } | |
1138 | ||
1139 | if (rwa->raw) { | |
1140 | /* | |
1141 | * We should have received a DRR_OBJECT_RANGE record | |
1142 | * containing this block and stored it in rwa. | |
1143 | */ | |
1144 | if (drro->drr_object < rwa->or_firstobj || | |
1145 | drro->drr_object >= rwa->or_firstobj + rwa->or_numslots || | |
1146 | drro->drr_raw_bonuslen < drro->drr_bonuslen || | |
1147 | drro->drr_indblkshift > SPA_MAXBLOCKSHIFT || | |
1148 | drro->drr_nlevels > DN_MAX_LEVELS || | |
1149 | drro->drr_nblkptr > DN_MAX_NBLKPTR || | |
1150 | DN_SLOTS_TO_BONUSLEN(dn_slots) < | |
1151 | drro->drr_raw_bonuslen) | |
1152 | return (SET_ERROR(EINVAL)); | |
1153 | } else { | |
1154 | if (drro->drr_flags != 0 || drro->drr_raw_bonuslen != 0 || | |
1155 | drro->drr_indblkshift != 0 || drro->drr_nlevels != 0 || | |
1156 | drro->drr_nblkptr != 0) | |
1157 | return (SET_ERROR(EINVAL)); | |
1158 | } | |
1159 | ||
1160 | err = dmu_object_info(rwa->os, drro->drr_object, &doi); | |
1161 | if (err != 0 && err != ENOENT && err != EEXIST) | |
1162 | return (SET_ERROR(EINVAL)); | |
1163 | ||
1164 | if (drro->drr_object > rwa->max_object) | |
1165 | rwa->max_object = drro->drr_object; | |
1166 | ||
1167 | /* | |
1168 | * If we are losing blkptrs or changing the block size this must | |
1169 | * be a new file instance. We must clear out the previous file | |
1170 | * contents before we can change this type of metadata in the dnode. | |
1171 | * Raw receives will also check that the indirect structure of the | |
1172 | * dnode hasn't changed. | |
1173 | */ | |
1174 | if (err == 0) { | |
1175 | uint32_t indblksz = drro->drr_indblkshift ? | |
1176 | 1ULL << drro->drr_indblkshift : 0; | |
1177 | int nblkptr = deduce_nblkptr(drro->drr_bonustype, | |
1178 | drro->drr_bonuslen); | |
c2c6eadf | 1179 | boolean_t did_free = B_FALSE; |
03916905 PD |
1180 | |
1181 | object = drro->drr_object; | |
1182 | ||
369aa501 | 1183 | /* nblkptr should be bounded by the bonus size and type */ |
03916905 PD |
1184 | if (rwa->raw && nblkptr != drro->drr_nblkptr) |
1185 | return (SET_ERROR(EINVAL)); | |
1186 | ||
369aa501 TC |
1187 | /* |
1188 | * Check for indicators that the object was freed and | |
1189 | * reallocated. For all sends, these indicators are: | |
1190 | * - A changed block size | |
1191 | * - A smaller nblkptr | |
1192 | * - A changed dnode size | |
1193 | * For raw sends we also check a few other fields to | |
1194 | * ensure we are preserving the objset structure exactly | |
1195 | * as it was on the receive side: | |
1196 | * - A changed indirect block size | |
1197 | * - A smaller nlevels | |
1198 | */ | |
03916905 PD |
1199 | if (drro->drr_blksz != doi.doi_data_block_size || |
1200 | nblkptr < doi.doi_nblkptr || | |
1201 | dn_slots != doi.doi_dnodesize >> DNODE_SHIFT || | |
1202 | (rwa->raw && | |
1203 | (indblksz != doi.doi_metadata_block_size || | |
1204 | drro->drr_nlevels < doi.doi_indirection))) { | |
1205 | err = dmu_free_long_range(rwa->os, | |
1206 | drro->drr_object, 0, DMU_OBJECT_END); | |
1207 | if (err != 0) | |
1208 | return (SET_ERROR(EINVAL)); | |
c2c6eadf TC |
1209 | else |
1210 | did_free = B_TRUE; | |
03916905 PD |
1211 | } |
1212 | ||
1213 | /* | |
1214 | * The dmu does not currently support decreasing nlevels | |
369aa501 TC |
1215 | * or changing the number of dnode slots on an object. For |
1216 | * non-raw sends, this does not matter and the new object | |
1217 | * can just use the previous one's nlevels. For raw sends, | |
1218 | * however, the structure of the received dnode (including | |
1219 | * nlevels and dnode slots) must match that of the send | |
1220 | * side. Therefore, instead of using dmu_object_reclaim(), | |
1221 | * we must free the object completely and call | |
1222 | * dmu_object_claim_dnsize() instead. | |
03916905 PD |
1223 | */ |
1224 | if ((rwa->raw && drro->drr_nlevels < doi.doi_indirection) || | |
1225 | dn_slots != doi.doi_dnodesize >> DNODE_SHIFT) { | |
1226 | err = dmu_free_long_object(rwa->os, drro->drr_object); | |
1227 | if (err != 0) | |
1228 | return (SET_ERROR(EINVAL)); | |
1229 | ||
1230 | txg_wait_synced(dmu_objset_pool(rwa->os), 0); | |
1231 | object = DMU_NEW_OBJECT; | |
1232 | } | |
369aa501 TC |
1233 | |
1234 | /* | |
1235 | * For raw receives, free everything beyond the new incoming | |
1236 | * maxblkid. Normally this would be done with a DRR_FREE | |
1237 | * record that would come after this DRR_OBJECT record is | |
1238 | * processed. However, for raw receives we manually set the | |
1239 | * maxblkid from the drr_maxblkid and so we must first free | |
1240 | * everything above that blkid to ensure the DMU is always | |
c2c6eadf TC |
1241 | * consistent with itself. We will never free the first block |
1242 | * of the object here because a maxblkid of 0 could indicate | |
1243 | * an object with a single block or one with no blocks. This | |
1244 | * free may be skipped when dmu_free_long_range() was called | |
1245 | * above since it covers the entire object's contents. | |
369aa501 | 1246 | */ |
c2c6eadf | 1247 | if (rwa->raw && object != DMU_NEW_OBJECT && !did_free) { |
369aa501 | 1248 | err = dmu_free_long_range(rwa->os, drro->drr_object, |
c2c6eadf | 1249 | (drro->drr_maxblkid + 1) * doi.doi_data_block_size, |
369aa501 TC |
1250 | DMU_OBJECT_END); |
1251 | if (err != 0) | |
1252 | return (SET_ERROR(EINVAL)); | |
1253 | } | |
03916905 PD |
1254 | } else if (err == EEXIST) { |
1255 | /* | |
1256 | * The object requested is currently an interior slot of a | |
1257 | * multi-slot dnode. This will be resolved when the next txg | |
1258 | * is synced out, since the send stream will have told us | |
1259 | * to free this slot when we freed the associated dnode | |
1260 | * earlier in the stream. | |
1261 | */ | |
1262 | txg_wait_synced(dmu_objset_pool(rwa->os), 0); | |
b92f5d9f BB |
1263 | |
1264 | if (dmu_object_info(rwa->os, drro->drr_object, NULL) != ENOENT) | |
1265 | return (SET_ERROR(EINVAL)); | |
1266 | ||
1267 | /* object was freed and we are about to allocate a new one */ | |
1268 | object = DMU_NEW_OBJECT; | |
03916905 PD |
1269 | } else { |
1270 | /* object is free and we are about to allocate a new one */ | |
1271 | object = DMU_NEW_OBJECT; | |
1272 | } | |
1273 | ||
1274 | /* | |
1275 | * If this is a multi-slot dnode there is a chance that this | |
1276 | * object will expand into a slot that is already used by | |
1277 | * another object from the previous snapshot. We must free | |
1278 | * these objects before we attempt to allocate the new dnode. | |
1279 | */ | |
1280 | if (dn_slots > 1) { | |
1281 | boolean_t need_sync = B_FALSE; | |
1282 | ||
1283 | for (uint64_t slot = drro->drr_object + 1; | |
1284 | slot < drro->drr_object + dn_slots; | |
1285 | slot++) { | |
1286 | dmu_object_info_t slot_doi; | |
1287 | ||
1288 | err = dmu_object_info(rwa->os, slot, &slot_doi); | |
1289 | if (err == ENOENT || err == EEXIST) | |
1290 | continue; | |
1291 | else if (err != 0) | |
1292 | return (err); | |
1293 | ||
1294 | err = dmu_free_long_object(rwa->os, slot); | |
03916905 PD |
1295 | if (err != 0) |
1296 | return (err); | |
1297 | ||
1298 | need_sync = B_TRUE; | |
1299 | } | |
1300 | ||
1301 | if (need_sync) | |
1302 | txg_wait_synced(dmu_objset_pool(rwa->os), 0); | |
1303 | } | |
1304 | ||
1305 | tx = dmu_tx_create(rwa->os); | |
1306 | dmu_tx_hold_bonus(tx, object); | |
1307 | dmu_tx_hold_write(tx, object, 0, 0); | |
1308 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1309 | if (err != 0) { | |
1310 | dmu_tx_abort(tx); | |
1311 | return (err); | |
1312 | } | |
1313 | ||
1314 | if (object == DMU_NEW_OBJECT) { | |
1315 | /* currently free, want to be allocated */ | |
1316 | err = dmu_object_claim_dnsize(rwa->os, drro->drr_object, | |
1317 | drro->drr_type, drro->drr_blksz, | |
1318 | drro->drr_bonustype, drro->drr_bonuslen, | |
1319 | dn_slots << DNODE_SHIFT, tx); | |
1320 | } else if (drro->drr_type != doi.doi_type || | |
1321 | drro->drr_blksz != doi.doi_data_block_size || | |
1322 | drro->drr_bonustype != doi.doi_bonus_type || | |
1323 | drro->drr_bonuslen != doi.doi_bonus_size) { | |
1324 | /* currently allocated, but with different properties */ | |
1325 | err = dmu_object_reclaim_dnsize(rwa->os, drro->drr_object, | |
1326 | drro->drr_type, drro->drr_blksz, | |
1327 | drro->drr_bonustype, drro->drr_bonuslen, | |
1328 | dn_slots << DNODE_SHIFT, tx); | |
1329 | } | |
3fa93bb8 | 1330 | |
03916905 PD |
1331 | if (err != 0) { |
1332 | dmu_tx_commit(tx); | |
1333 | return (SET_ERROR(EINVAL)); | |
1334 | } | |
1335 | ||
1336 | if (rwa->or_crypt_params_present) { | |
1337 | /* | |
1338 | * Set the crypt params for the buffer associated with this | |
1339 | * range of dnodes. This causes the blkptr_t to have the | |
1340 | * same crypt params (byteorder, salt, iv, mac) as on the | |
1341 | * sending side. | |
1342 | * | |
1343 | * Since we are committing this tx now, it is possible for | |
1344 | * the dnode block to end up on-disk with the incorrect MAC, | |
1345 | * if subsequent objects in this block are received in a | |
1346 | * different txg. However, since the dataset is marked as | |
1347 | * inconsistent, no code paths will do a non-raw read (or | |
1348 | * decrypt the block / verify the MAC). The receive code and | |
1349 | * scrub code can safely do raw reads and verify the | |
1350 | * checksum. They don't need to verify the MAC. | |
1351 | */ | |
1352 | dmu_buf_t *db = NULL; | |
1353 | uint64_t offset = rwa->or_firstobj * DNODE_MIN_SIZE; | |
1354 | ||
1355 | err = dmu_buf_hold_by_dnode(DMU_META_DNODE(rwa->os), | |
1356 | offset, FTAG, &db, DMU_READ_PREFETCH | DMU_READ_NO_DECRYPT); | |
1357 | if (err != 0) { | |
1358 | dmu_tx_commit(tx); | |
1359 | return (SET_ERROR(EINVAL)); | |
1360 | } | |
1361 | ||
1362 | dmu_buf_set_crypt_params(db, rwa->or_byteorder, | |
1363 | rwa->or_salt, rwa->or_iv, rwa->or_mac, tx); | |
1364 | ||
1365 | dmu_buf_rele(db, FTAG); | |
1366 | ||
1367 | rwa->or_crypt_params_present = B_FALSE; | |
1368 | } | |
1369 | ||
1370 | dmu_object_set_checksum(rwa->os, drro->drr_object, | |
1371 | drro->drr_checksumtype, tx); | |
1372 | dmu_object_set_compress(rwa->os, drro->drr_object, | |
1373 | drro->drr_compress, tx); | |
1374 | ||
1375 | /* handle more restrictive dnode structuring for raw recvs */ | |
1376 | if (rwa->raw) { | |
1377 | /* | |
369aa501 TC |
1378 | * Set the indirect block size, block shift, nlevels. |
1379 | * This will not fail because we ensured all of the | |
1380 | * blocks were freed earlier if this is a new object. | |
1381 | * For non-new objects block size and indirect block | |
1382 | * shift cannot change and nlevels can only increase. | |
03916905 PD |
1383 | */ |
1384 | VERIFY0(dmu_object_set_blocksize(rwa->os, drro->drr_object, | |
1385 | drro->drr_blksz, drro->drr_indblkshift, tx)); | |
1386 | VERIFY0(dmu_object_set_nlevels(rwa->os, drro->drr_object, | |
1387 | drro->drr_nlevels, tx)); | |
369aa501 TC |
1388 | |
1389 | /* | |
c2c6eadf TC |
1390 | * Set the maxblkid. This will always succeed because |
1391 | * we freed all blocks beyond the new maxblkid above. | |
369aa501 | 1392 | */ |
03916905 PD |
1393 | VERIFY0(dmu_object_set_maxblkid(rwa->os, drro->drr_object, |
1394 | drro->drr_maxblkid, tx)); | |
1395 | } | |
1396 | ||
1397 | if (data != NULL) { | |
1398 | dmu_buf_t *db; | |
6955b401 | 1399 | dnode_t *dn; |
03916905 PD |
1400 | uint32_t flags = DMU_READ_NO_PREFETCH; |
1401 | ||
1402 | if (rwa->raw) | |
1403 | flags |= DMU_READ_NO_DECRYPT; | |
1404 | ||
6955b401 BB |
1405 | VERIFY0(dnode_hold(rwa->os, drro->drr_object, FTAG, &dn)); |
1406 | VERIFY0(dmu_bonus_hold_by_dnode(dn, FTAG, &db, flags)); | |
1407 | ||
03916905 PD |
1408 | dmu_buf_will_dirty(db, tx); |
1409 | ||
1410 | ASSERT3U(db->db_size, >=, drro->drr_bonuslen); | |
1411 | bcopy(data, db->db_data, DRR_OBJECT_PAYLOAD_SIZE(drro)); | |
1412 | ||
1413 | /* | |
1414 | * Raw bonus buffers have their byteorder determined by the | |
1415 | * DRR_OBJECT_RANGE record. | |
1416 | */ | |
1417 | if (rwa->byteswap && !rwa->raw) { | |
1418 | dmu_object_byteswap_t byteswap = | |
1419 | DMU_OT_BYTESWAP(drro->drr_bonustype); | |
1420 | dmu_ot_byteswap[byteswap].ob_func(db->db_data, | |
1421 | DRR_OBJECT_PAYLOAD_SIZE(drro)); | |
1422 | } | |
1423 | dmu_buf_rele(db, FTAG); | |
6955b401 | 1424 | dnode_rele(dn, FTAG); |
03916905 PD |
1425 | } |
1426 | dmu_tx_commit(tx); | |
1427 | ||
1428 | return (0); | |
1429 | } | |
1430 | ||
1431 | /* ARGSUSED */ | |
1432 | noinline static int | |
1433 | receive_freeobjects(struct receive_writer_arg *rwa, | |
1434 | struct drr_freeobjects *drrfo) | |
1435 | { | |
1436 | uint64_t obj; | |
1437 | int next_err = 0; | |
1438 | ||
1439 | if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) | |
1440 | return (SET_ERROR(EINVAL)); | |
1441 | ||
1442 | for (obj = drrfo->drr_firstobj == 0 ? 1 : drrfo->drr_firstobj; | |
1443 | obj < drrfo->drr_firstobj + drrfo->drr_numobjs && next_err == 0; | |
1444 | next_err = dmu_object_next(rwa->os, &obj, FALSE, 0)) { | |
1445 | dmu_object_info_t doi; | |
1446 | int err; | |
1447 | ||
1448 | err = dmu_object_info(rwa->os, obj, &doi); | |
1449 | if (err == ENOENT) | |
1450 | continue; | |
1451 | else if (err != 0) | |
1452 | return (err); | |
1453 | ||
1454 | err = dmu_free_long_object(rwa->os, obj); | |
1455 | ||
1456 | if (err != 0) | |
1457 | return (err); | |
1458 | ||
1459 | if (obj > rwa->max_object) | |
1460 | rwa->max_object = obj; | |
1461 | } | |
1462 | if (next_err != ESRCH) | |
1463 | return (next_err); | |
1464 | return (0); | |
1465 | } | |
1466 | ||
1467 | noinline static int | |
1468 | receive_write(struct receive_writer_arg *rwa, struct drr_write *drrw, | |
1469 | arc_buf_t *abuf) | |
1470 | { | |
1471 | int err; | |
1472 | dmu_tx_t *tx; | |
1473 | dnode_t *dn; | |
1474 | ||
1475 | if (drrw->drr_offset + drrw->drr_logical_size < drrw->drr_offset || | |
1476 | !DMU_OT_IS_VALID(drrw->drr_type)) | |
1477 | return (SET_ERROR(EINVAL)); | |
1478 | ||
1479 | /* | |
1480 | * For resuming to work, records must be in increasing order | |
1481 | * by (object, offset). | |
1482 | */ | |
1483 | if (drrw->drr_object < rwa->last_object || | |
1484 | (drrw->drr_object == rwa->last_object && | |
1485 | drrw->drr_offset < rwa->last_offset)) { | |
1486 | return (SET_ERROR(EINVAL)); | |
1487 | } | |
1488 | rwa->last_object = drrw->drr_object; | |
1489 | rwa->last_offset = drrw->drr_offset; | |
1490 | ||
1491 | if (rwa->last_object > rwa->max_object) | |
1492 | rwa->max_object = rwa->last_object; | |
1493 | ||
1494 | if (dmu_object_info(rwa->os, drrw->drr_object, NULL) != 0) | |
1495 | return (SET_ERROR(EINVAL)); | |
1496 | ||
1497 | tx = dmu_tx_create(rwa->os); | |
1498 | dmu_tx_hold_write(tx, drrw->drr_object, | |
1499 | drrw->drr_offset, drrw->drr_logical_size); | |
1500 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1501 | if (err != 0) { | |
1502 | dmu_tx_abort(tx); | |
1503 | return (err); | |
1504 | } | |
1505 | ||
1506 | if (rwa->byteswap && !arc_is_encrypted(abuf) && | |
1507 | arc_get_compression(abuf) == ZIO_COMPRESS_OFF) { | |
1508 | dmu_object_byteswap_t byteswap = | |
1509 | DMU_OT_BYTESWAP(drrw->drr_type); | |
1510 | dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, | |
1511 | DRR_WRITE_PAYLOAD_SIZE(drrw)); | |
1512 | } | |
1513 | ||
1514 | VERIFY0(dnode_hold(rwa->os, drrw->drr_object, FTAG, &dn)); | |
305781da TC |
1515 | err = dmu_assign_arcbuf_by_dnode(dn, drrw->drr_offset, abuf, tx); |
1516 | if (err != 0) { | |
1517 | dnode_rele(dn, FTAG); | |
1518 | dmu_tx_commit(tx); | |
1519 | return (err); | |
1520 | } | |
03916905 PD |
1521 | dnode_rele(dn, FTAG); |
1522 | ||
1523 | /* | |
1524 | * Note: If the receive fails, we want the resume stream to start | |
1525 | * with the same record that we last successfully received (as opposed | |
1526 | * to the next record), so that we can verify that we are | |
1527 | * resuming from the correct location. | |
1528 | */ | |
1529 | save_resume_state(rwa, drrw->drr_object, drrw->drr_offset, tx); | |
1530 | dmu_tx_commit(tx); | |
1531 | ||
1532 | return (0); | |
1533 | } | |
1534 | ||
1535 | /* | |
1536 | * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed | |
1537 | * streams to refer to a copy of the data that is already on the | |
1538 | * system because it came in earlier in the stream. This function | |
1539 | * finds the earlier copy of the data, and uses that copy instead of | |
1540 | * data from the stream to fulfill this write. | |
1541 | */ | |
1542 | static int | |
1543 | receive_write_byref(struct receive_writer_arg *rwa, | |
1544 | struct drr_write_byref *drrwbr) | |
1545 | { | |
1546 | dmu_tx_t *tx; | |
1547 | int err; | |
1548 | guid_map_entry_t gmesrch; | |
1549 | guid_map_entry_t *gmep; | |
1550 | avl_index_t where; | |
1551 | objset_t *ref_os = NULL; | |
1552 | int flags = DMU_READ_PREFETCH; | |
1553 | dmu_buf_t *dbp; | |
1554 | ||
1555 | if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) | |
1556 | return (SET_ERROR(EINVAL)); | |
1557 | ||
1558 | /* | |
1559 | * If the GUID of the referenced dataset is different from the | |
1560 | * GUID of the target dataset, find the referenced dataset. | |
1561 | */ | |
1562 | if (drrwbr->drr_toguid != drrwbr->drr_refguid) { | |
1563 | gmesrch.guid = drrwbr->drr_refguid; | |
1564 | if ((gmep = avl_find(rwa->guid_to_ds_map, &gmesrch, | |
1565 | &where)) == NULL) { | |
1566 | return (SET_ERROR(EINVAL)); | |
1567 | } | |
1568 | if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) | |
1569 | return (SET_ERROR(EINVAL)); | |
1570 | } else { | |
1571 | ref_os = rwa->os; | |
1572 | } | |
1573 | ||
1574 | if (drrwbr->drr_object > rwa->max_object) | |
1575 | rwa->max_object = drrwbr->drr_object; | |
1576 | ||
1577 | if (rwa->raw) | |
1578 | flags |= DMU_READ_NO_DECRYPT; | |
1579 | ||
1580 | /* may return either a regular db or an encrypted one */ | |
1581 | err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, | |
1582 | drrwbr->drr_refoffset, FTAG, &dbp, flags); | |
1583 | if (err != 0) | |
1584 | return (err); | |
1585 | ||
1586 | tx = dmu_tx_create(rwa->os); | |
1587 | ||
1588 | dmu_tx_hold_write(tx, drrwbr->drr_object, | |
1589 | drrwbr->drr_offset, drrwbr->drr_length); | |
1590 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1591 | if (err != 0) { | |
1592 | dmu_tx_abort(tx); | |
1593 | return (err); | |
1594 | } | |
1595 | ||
1596 | if (rwa->raw) { | |
1597 | dmu_copy_from_buf(rwa->os, drrwbr->drr_object, | |
1598 | drrwbr->drr_offset, dbp, tx); | |
1599 | } else { | |
1600 | dmu_write(rwa->os, drrwbr->drr_object, | |
1601 | drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); | |
1602 | } | |
1603 | dmu_buf_rele(dbp, FTAG); | |
1604 | ||
1605 | /* See comment in restore_write. */ | |
1606 | save_resume_state(rwa, drrwbr->drr_object, drrwbr->drr_offset, tx); | |
1607 | dmu_tx_commit(tx); | |
1608 | return (0); | |
1609 | } | |
1610 | ||
1611 | static int | |
1612 | receive_write_embedded(struct receive_writer_arg *rwa, | |
1613 | struct drr_write_embedded *drrwe, void *data) | |
1614 | { | |
1615 | dmu_tx_t *tx; | |
1616 | int err; | |
1617 | ||
1618 | if (drrwe->drr_offset + drrwe->drr_length < drrwe->drr_offset) | |
1619 | return (SET_ERROR(EINVAL)); | |
1620 | ||
1621 | if (drrwe->drr_psize > BPE_PAYLOAD_SIZE) | |
1622 | return (SET_ERROR(EINVAL)); | |
1623 | ||
1624 | if (drrwe->drr_etype >= NUM_BP_EMBEDDED_TYPES) | |
1625 | return (SET_ERROR(EINVAL)); | |
1626 | if (drrwe->drr_compression >= ZIO_COMPRESS_FUNCTIONS) | |
1627 | return (SET_ERROR(EINVAL)); | |
1628 | if (rwa->raw) | |
1629 | return (SET_ERROR(EINVAL)); | |
1630 | ||
1631 | if (drrwe->drr_object > rwa->max_object) | |
1632 | rwa->max_object = drrwe->drr_object; | |
1633 | ||
1634 | tx = dmu_tx_create(rwa->os); | |
1635 | ||
1636 | dmu_tx_hold_write(tx, drrwe->drr_object, | |
1637 | drrwe->drr_offset, drrwe->drr_length); | |
1638 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1639 | if (err != 0) { | |
1640 | dmu_tx_abort(tx); | |
1641 | return (err); | |
1642 | } | |
1643 | ||
1644 | dmu_write_embedded(rwa->os, drrwe->drr_object, | |
1645 | drrwe->drr_offset, data, drrwe->drr_etype, | |
1646 | drrwe->drr_compression, drrwe->drr_lsize, drrwe->drr_psize, | |
1647 | rwa->byteswap ^ ZFS_HOST_BYTEORDER, tx); | |
1648 | ||
1649 | /* See comment in restore_write. */ | |
1650 | save_resume_state(rwa, drrwe->drr_object, drrwe->drr_offset, tx); | |
1651 | dmu_tx_commit(tx); | |
1652 | return (0); | |
1653 | } | |
1654 | ||
1655 | static int | |
1656 | receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs, | |
1657 | arc_buf_t *abuf) | |
1658 | { | |
1659 | dmu_tx_t *tx; | |
1660 | dmu_buf_t *db, *db_spill; | |
1661 | int err; | |
1662 | uint32_t flags = 0; | |
1663 | ||
1664 | if (drrs->drr_length < SPA_MINBLOCKSIZE || | |
1665 | drrs->drr_length > spa_maxblocksize(dmu_objset_spa(rwa->os))) | |
1666 | return (SET_ERROR(EINVAL)); | |
1667 | ||
1668 | if (rwa->raw) { | |
1669 | if (!DMU_OT_IS_VALID(drrs->drr_type) || | |
1670 | drrs->drr_compressiontype >= ZIO_COMPRESS_FUNCTIONS || | |
1671 | drrs->drr_compressed_size == 0) | |
1672 | return (SET_ERROR(EINVAL)); | |
1673 | ||
1674 | flags |= DMU_READ_NO_DECRYPT; | |
1675 | } | |
1676 | ||
1677 | if (dmu_object_info(rwa->os, drrs->drr_object, NULL) != 0) | |
1678 | return (SET_ERROR(EINVAL)); | |
1679 | ||
1680 | if (drrs->drr_object > rwa->max_object) | |
1681 | rwa->max_object = drrs->drr_object; | |
1682 | ||
1683 | VERIFY0(dmu_bonus_hold(rwa->os, drrs->drr_object, FTAG, &db)); | |
1684 | if ((err = dmu_spill_hold_by_bonus(db, DMU_READ_NO_DECRYPT, FTAG, | |
1685 | &db_spill)) != 0) { | |
1686 | dmu_buf_rele(db, FTAG); | |
1687 | return (err); | |
1688 | } | |
1689 | ||
1690 | tx = dmu_tx_create(rwa->os); | |
1691 | ||
1692 | dmu_tx_hold_spill(tx, db->db_object); | |
1693 | ||
1694 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1695 | if (err != 0) { | |
1696 | dmu_buf_rele(db, FTAG); | |
1697 | dmu_buf_rele(db_spill, FTAG); | |
1698 | dmu_tx_abort(tx); | |
1699 | return (err); | |
1700 | } | |
1701 | ||
1702 | if (db_spill->db_size < drrs->drr_length) | |
1703 | VERIFY(0 == dbuf_spill_set_blksz(db_spill, | |
1704 | drrs->drr_length, tx)); | |
1705 | ||
1706 | if (rwa->byteswap && !arc_is_encrypted(abuf) && | |
1707 | arc_get_compression(abuf) == ZIO_COMPRESS_OFF) { | |
1708 | dmu_object_byteswap_t byteswap = | |
1709 | DMU_OT_BYTESWAP(drrs->drr_type); | |
1710 | dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, | |
1711 | DRR_SPILL_PAYLOAD_SIZE(drrs)); | |
1712 | } | |
1713 | ||
1714 | dbuf_assign_arcbuf((dmu_buf_impl_t *)db_spill, abuf, tx); | |
1715 | ||
1716 | dmu_buf_rele(db, FTAG); | |
1717 | dmu_buf_rele(db_spill, FTAG); | |
1718 | ||
1719 | dmu_tx_commit(tx); | |
1720 | return (0); | |
1721 | } | |
1722 | ||
1723 | /* ARGSUSED */ | |
1724 | noinline static int | |
1725 | receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf) | |
1726 | { | |
1727 | int err; | |
1728 | ||
1729 | if (drrf->drr_length != DMU_OBJECT_END && | |
1730 | drrf->drr_offset + drrf->drr_length < drrf->drr_offset) | |
1731 | return (SET_ERROR(EINVAL)); | |
1732 | ||
1733 | if (dmu_object_info(rwa->os, drrf->drr_object, NULL) != 0) | |
1734 | return (SET_ERROR(EINVAL)); | |
1735 | ||
1736 | if (drrf->drr_object > rwa->max_object) | |
1737 | rwa->max_object = drrf->drr_object; | |
1738 | ||
1739 | err = dmu_free_long_range(rwa->os, drrf->drr_object, | |
1740 | drrf->drr_offset, drrf->drr_length); | |
1741 | ||
1742 | return (err); | |
1743 | } | |
1744 | ||
1745 | static int | |
1746 | receive_object_range(struct receive_writer_arg *rwa, | |
1747 | struct drr_object_range *drror) | |
1748 | { | |
1749 | /* | |
1750 | * By default, we assume this block is in our native format | |
1751 | * (ZFS_HOST_BYTEORDER). We then take into account whether | |
1752 | * the send stream is byteswapped (rwa->byteswap). Finally, | |
1753 | * we need to byteswap again if this particular block was | |
1754 | * in non-native format on the send side. | |
1755 | */ | |
1756 | boolean_t byteorder = ZFS_HOST_BYTEORDER ^ rwa->byteswap ^ | |
1757 | !!DRR_IS_RAW_BYTESWAPPED(drror->drr_flags); | |
1758 | ||
1759 | /* | |
1760 | * Since dnode block sizes are constant, we should not need to worry | |
1761 | * about making sure that the dnode block size is the same on the | |
1762 | * sending and receiving sides for the time being. For non-raw sends, | |
1763 | * this does not matter (and in fact we do not send a DRR_OBJECT_RANGE | |
1764 | * record at all). Raw sends require this record type because the | |
1765 | * encryption parameters are used to protect an entire block of bonus | |
1766 | * buffers. If the size of dnode blocks ever becomes variable, | |
1767 | * handling will need to be added to ensure that dnode block sizes | |
1768 | * match on the sending and receiving side. | |
1769 | */ | |
1770 | if (drror->drr_numslots != DNODES_PER_BLOCK || | |
1771 | P2PHASE(drror->drr_firstobj, DNODES_PER_BLOCK) != 0 || | |
1772 | !rwa->raw) | |
1773 | return (SET_ERROR(EINVAL)); | |
1774 | ||
1775 | if (drror->drr_firstobj > rwa->max_object) | |
1776 | rwa->max_object = drror->drr_firstobj; | |
1777 | ||
1778 | /* | |
1779 | * The DRR_OBJECT_RANGE handling must be deferred to receive_object() | |
1780 | * so that the block of dnodes is not written out when it's empty, | |
1781 | * and converted to a HOLE BP. | |
1782 | */ | |
1783 | rwa->or_crypt_params_present = B_TRUE; | |
1784 | rwa->or_firstobj = drror->drr_firstobj; | |
1785 | rwa->or_numslots = drror->drr_numslots; | |
1786 | bcopy(drror->drr_salt, rwa->or_salt, ZIO_DATA_SALT_LEN); | |
1787 | bcopy(drror->drr_iv, rwa->or_iv, ZIO_DATA_IV_LEN); | |
1788 | bcopy(drror->drr_mac, rwa->or_mac, ZIO_DATA_MAC_LEN); | |
1789 | rwa->or_byteorder = byteorder; | |
1790 | ||
1791 | return (0); | |
1792 | } | |
1793 | ||
1794 | /* used to destroy the drc_ds on error */ | |
1795 | static void | |
1796 | dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) | |
1797 | { | |
1798 | dsl_dataset_t *ds = drc->drc_ds; | |
1799 | ds_hold_flags_t dsflags = (drc->drc_raw) ? 0 : DS_HOLD_FLAG_DECRYPT; | |
1800 | ||
1801 | /* | |
1802 | * Wait for the txg sync before cleaning up the receive. For | |
1803 | * resumable receives, this ensures that our resume state has | |
1804 | * been written out to disk. For raw receives, this ensures | |
1805 | * that the user accounting code will not attempt to do anything | |
1806 | * after we stopped receiving the dataset. | |
1807 | */ | |
1808 | txg_wait_synced(ds->ds_dir->dd_pool, 0); | |
1809 | ds->ds_objset->os_raw_receive = B_FALSE; | |
1810 | ||
1811 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); | |
1812 | if (drc->drc_resumable && !BP_IS_HOLE(dsl_dataset_get_blkptr(ds))) { | |
1813 | rrw_exit(&ds->ds_bp_rwlock, FTAG); | |
1814 | dsl_dataset_disown(ds, dsflags, dmu_recv_tag); | |
1815 | } else { | |
1816 | char name[ZFS_MAX_DATASET_NAME_LEN]; | |
1817 | rrw_exit(&ds->ds_bp_rwlock, FTAG); | |
1818 | dsl_dataset_name(ds, name); | |
1819 | dsl_dataset_disown(ds, dsflags, dmu_recv_tag); | |
1820 | (void) dsl_destroy_head(name); | |
1821 | } | |
1822 | } | |
1823 | ||
1824 | static void | |
1825 | receive_cksum(struct receive_arg *ra, int len, void *buf) | |
1826 | { | |
1827 | if (ra->byteswap) { | |
1828 | (void) fletcher_4_incremental_byteswap(buf, len, &ra->cksum); | |
1829 | } else { | |
1830 | (void) fletcher_4_incremental_native(buf, len, &ra->cksum); | |
1831 | } | |
1832 | } | |
1833 | ||
1834 | /* | |
1835 | * Read the payload into a buffer of size len, and update the current record's | |
1836 | * payload field. | |
1837 | * Allocate ra->next_rrd and read the next record's header into | |
1838 | * ra->next_rrd->header. | |
1839 | * Verify checksum of payload and next record. | |
1840 | */ | |
1841 | static int | |
1842 | receive_read_payload_and_next_header(struct receive_arg *ra, int len, void *buf) | |
1843 | { | |
1844 | int err; | |
1845 | zio_cksum_t cksum_orig; | |
1846 | zio_cksum_t *cksump; | |
1847 | ||
1848 | if (len != 0) { | |
1849 | ASSERT3U(len, <=, SPA_MAXBLOCKSIZE); | |
1850 | err = receive_read(ra, len, buf); | |
1851 | if (err != 0) | |
1852 | return (err); | |
1853 | receive_cksum(ra, len, buf); | |
1854 | ||
1855 | /* note: rrd is NULL when reading the begin record's payload */ | |
1856 | if (ra->rrd != NULL) { | |
1857 | ra->rrd->payload = buf; | |
1858 | ra->rrd->payload_size = len; | |
1859 | ra->rrd->bytes_read = ra->bytes_read; | |
1860 | } | |
960347d3 TC |
1861 | } else { |
1862 | ASSERT3P(buf, ==, NULL); | |
03916905 PD |
1863 | } |
1864 | ||
1865 | ra->prev_cksum = ra->cksum; | |
1866 | ||
1867 | ra->next_rrd = kmem_zalloc(sizeof (*ra->next_rrd), KM_SLEEP); | |
1868 | err = receive_read(ra, sizeof (ra->next_rrd->header), | |
1869 | &ra->next_rrd->header); | |
1870 | ra->next_rrd->bytes_read = ra->bytes_read; | |
1871 | ||
1872 | if (err != 0) { | |
1873 | kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); | |
1874 | ra->next_rrd = NULL; | |
1875 | return (err); | |
1876 | } | |
1877 | if (ra->next_rrd->header.drr_type == DRR_BEGIN) { | |
1878 | kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); | |
1879 | ra->next_rrd = NULL; | |
1880 | return (SET_ERROR(EINVAL)); | |
1881 | } | |
1882 | ||
1883 | /* | |
1884 | * Note: checksum is of everything up to but not including the | |
1885 | * checksum itself. | |
1886 | */ | |
1887 | ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), | |
1888 | ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); | |
1889 | receive_cksum(ra, | |
1890 | offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), | |
1891 | &ra->next_rrd->header); | |
1892 | ||
1893 | cksum_orig = ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; | |
1894 | cksump = &ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; | |
1895 | ||
1896 | if (ra->byteswap) | |
1897 | byteswap_record(&ra->next_rrd->header); | |
1898 | ||
1899 | if ((!ZIO_CHECKSUM_IS_ZERO(cksump)) && | |
1900 | !ZIO_CHECKSUM_EQUAL(ra->cksum, *cksump)) { | |
1901 | kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); | |
1902 | ra->next_rrd = NULL; | |
1903 | return (SET_ERROR(ECKSUM)); | |
1904 | } | |
1905 | ||
1906 | receive_cksum(ra, sizeof (cksum_orig), &cksum_orig); | |
1907 | ||
1908 | return (0); | |
1909 | } | |
1910 | ||
1911 | static void | |
1912 | objlist_create(struct objlist *list) | |
1913 | { | |
1914 | list_create(&list->list, sizeof (struct receive_objnode), | |
1915 | offsetof(struct receive_objnode, node)); | |
1916 | list->last_lookup = 0; | |
1917 | } | |
1918 | ||
1919 | static void | |
1920 | objlist_destroy(struct objlist *list) | |
1921 | { | |
1922 | for (struct receive_objnode *n = list_remove_head(&list->list); | |
1923 | n != NULL; n = list_remove_head(&list->list)) { | |
1924 | kmem_free(n, sizeof (*n)); | |
1925 | } | |
1926 | list_destroy(&list->list); | |
1927 | } | |
1928 | ||
1929 | /* | |
1930 | * This function looks through the objlist to see if the specified object number | |
1931 | * is contained in the objlist. In the process, it will remove all object | |
1932 | * numbers in the list that are smaller than the specified object number. Thus, | |
1933 | * any lookup of an object number smaller than a previously looked up object | |
1934 | * number will always return false; therefore, all lookups should be done in | |
1935 | * ascending order. | |
1936 | */ | |
1937 | static boolean_t | |
1938 | objlist_exists(struct objlist *list, uint64_t object) | |
1939 | { | |
1940 | struct receive_objnode *node = list_head(&list->list); | |
1941 | ASSERT3U(object, >=, list->last_lookup); | |
1942 | list->last_lookup = object; | |
1943 | while (node != NULL && node->object < object) { | |
1944 | VERIFY3P(node, ==, list_remove_head(&list->list)); | |
1945 | kmem_free(node, sizeof (*node)); | |
1946 | node = list_head(&list->list); | |
1947 | } | |
1948 | return (node != NULL && node->object == object); | |
1949 | } | |
1950 | ||
1951 | /* | |
1952 | * The objlist is a list of object numbers stored in ascending order. However, | |
1953 | * the insertion of new object numbers does not seek out the correct location to | |
1954 | * store a new object number; instead, it appends it to the list for simplicity. | |
1955 | * Thus, any users must take care to only insert new object numbers in ascending | |
1956 | * order. | |
1957 | */ | |
1958 | static void | |
1959 | objlist_insert(struct objlist *list, uint64_t object) | |
1960 | { | |
1961 | struct receive_objnode *node = kmem_zalloc(sizeof (*node), KM_SLEEP); | |
1962 | node->object = object; | |
1963 | #ifdef ZFS_DEBUG | |
1964 | { | |
1965 | struct receive_objnode *last_object = list_tail(&list->list); | |
1966 | uint64_t last_objnum = (last_object != NULL ? last_object->object : 0); | |
1967 | ASSERT3U(node->object, >, last_objnum); | |
1968 | } | |
1969 | #endif | |
1970 | list_insert_tail(&list->list, node); | |
1971 | } | |
1972 | ||
1973 | /* | |
1974 | * Issue the prefetch reads for any necessary indirect blocks. | |
1975 | * | |
1976 | * We use the object ignore list to tell us whether or not to issue prefetches | |
1977 | * for a given object. We do this for both correctness (in case the blocksize | |
1978 | * of an object has changed) and performance (if the object doesn't exist, don't | |
1979 | * needlessly try to issue prefetches). We also trim the list as we go through | |
1980 | * the stream to prevent it from growing to an unbounded size. | |
1981 | * | |
1982 | * The object numbers within will always be in sorted order, and any write | |
1983 | * records we see will also be in sorted order, but they're not sorted with | |
1984 | * respect to each other (i.e. we can get several object records before | |
1985 | * receiving each object's write records). As a result, once we've reached a | |
1986 | * given object number, we can safely remove any reference to lower object | |
1987 | * numbers in the ignore list. In practice, we receive up to 32 object records | |
1988 | * before receiving write records, so the list can have up to 32 nodes in it. | |
1989 | */ | |
1990 | /* ARGSUSED */ | |
1991 | static void | |
1992 | receive_read_prefetch(struct receive_arg *ra, | |
1993 | uint64_t object, uint64_t offset, uint64_t length) | |
1994 | { | |
1995 | if (!objlist_exists(&ra->ignore_objlist, object)) { | |
1996 | dmu_prefetch(ra->os, object, 1, offset, length, | |
1997 | ZIO_PRIORITY_SYNC_READ); | |
1998 | } | |
1999 | } | |
2000 | ||
2001 | /* | |
2002 | * Read records off the stream, issuing any necessary prefetches. | |
2003 | */ | |
2004 | static int | |
2005 | receive_read_record(struct receive_arg *ra) | |
2006 | { | |
2007 | int err; | |
2008 | ||
2009 | switch (ra->rrd->header.drr_type) { | |
2010 | case DRR_OBJECT: | |
2011 | { | |
2012 | struct drr_object *drro = &ra->rrd->header.drr_u.drr_object; | |
2013 | uint32_t size = DRR_OBJECT_PAYLOAD_SIZE(drro); | |
960347d3 | 2014 | void *buf = NULL; |
03916905 PD |
2015 | dmu_object_info_t doi; |
2016 | ||
960347d3 TC |
2017 | if (size != 0) |
2018 | buf = kmem_zalloc(size, KM_SLEEP); | |
2019 | ||
03916905 PD |
2020 | err = receive_read_payload_and_next_header(ra, size, buf); |
2021 | if (err != 0) { | |
2022 | kmem_free(buf, size); | |
2023 | return (err); | |
2024 | } | |
2025 | err = dmu_object_info(ra->os, drro->drr_object, &doi); | |
2026 | /* | |
2027 | * See receive_read_prefetch for an explanation why we're | |
2028 | * storing this object in the ignore_obj_list. | |
2029 | */ | |
2030 | if (err == ENOENT || err == EEXIST || | |
2031 | (err == 0 && doi.doi_data_block_size != drro->drr_blksz)) { | |
2032 | objlist_insert(&ra->ignore_objlist, drro->drr_object); | |
2033 | err = 0; | |
2034 | } | |
2035 | return (err); | |
2036 | } | |
2037 | case DRR_FREEOBJECTS: | |
2038 | { | |
2039 | err = receive_read_payload_and_next_header(ra, 0, NULL); | |
2040 | return (err); | |
2041 | } | |
2042 | case DRR_WRITE: | |
2043 | { | |
2044 | struct drr_write *drrw = &ra->rrd->header.drr_u.drr_write; | |
2045 | arc_buf_t *abuf; | |
2046 | boolean_t is_meta = DMU_OT_IS_METADATA(drrw->drr_type); | |
2047 | ||
2048 | if (ra->raw) { | |
2049 | boolean_t byteorder = ZFS_HOST_BYTEORDER ^ | |
2050 | !!DRR_IS_RAW_BYTESWAPPED(drrw->drr_flags) ^ | |
2051 | ra->byteswap; | |
2052 | ||
2053 | abuf = arc_loan_raw_buf(dmu_objset_spa(ra->os), | |
2054 | drrw->drr_object, byteorder, drrw->drr_salt, | |
2055 | drrw->drr_iv, drrw->drr_mac, drrw->drr_type, | |
2056 | drrw->drr_compressed_size, drrw->drr_logical_size, | |
2057 | drrw->drr_compressiontype); | |
2058 | } else if (DRR_WRITE_COMPRESSED(drrw)) { | |
2059 | ASSERT3U(drrw->drr_compressed_size, >, 0); | |
2060 | ASSERT3U(drrw->drr_logical_size, >=, | |
2061 | drrw->drr_compressed_size); | |
2062 | ASSERT(!is_meta); | |
2063 | abuf = arc_loan_compressed_buf( | |
2064 | dmu_objset_spa(ra->os), | |
2065 | drrw->drr_compressed_size, drrw->drr_logical_size, | |
2066 | drrw->drr_compressiontype); | |
2067 | } else { | |
2068 | abuf = arc_loan_buf(dmu_objset_spa(ra->os), | |
2069 | is_meta, drrw->drr_logical_size); | |
2070 | } | |
2071 | ||
2072 | err = receive_read_payload_and_next_header(ra, | |
2073 | DRR_WRITE_PAYLOAD_SIZE(drrw), abuf->b_data); | |
2074 | if (err != 0) { | |
2075 | dmu_return_arcbuf(abuf); | |
2076 | return (err); | |
2077 | } | |
2078 | ra->rrd->arc_buf = abuf; | |
2079 | receive_read_prefetch(ra, drrw->drr_object, drrw->drr_offset, | |
2080 | drrw->drr_logical_size); | |
2081 | return (err); | |
2082 | } | |
2083 | case DRR_WRITE_BYREF: | |
2084 | { | |
2085 | struct drr_write_byref *drrwb = | |
2086 | &ra->rrd->header.drr_u.drr_write_byref; | |
2087 | err = receive_read_payload_and_next_header(ra, 0, NULL); | |
2088 | receive_read_prefetch(ra, drrwb->drr_object, drrwb->drr_offset, | |
2089 | drrwb->drr_length); | |
2090 | return (err); | |
2091 | } | |
2092 | case DRR_WRITE_EMBEDDED: | |
2093 | { | |
2094 | struct drr_write_embedded *drrwe = | |
2095 | &ra->rrd->header.drr_u.drr_write_embedded; | |
2096 | uint32_t size = P2ROUNDUP(drrwe->drr_psize, 8); | |
2097 | void *buf = kmem_zalloc(size, KM_SLEEP); | |
2098 | ||
2099 | err = receive_read_payload_and_next_header(ra, size, buf); | |
2100 | if (err != 0) { | |
2101 | kmem_free(buf, size); | |
2102 | return (err); | |
2103 | } | |
2104 | ||
2105 | receive_read_prefetch(ra, drrwe->drr_object, drrwe->drr_offset, | |
2106 | drrwe->drr_length); | |
2107 | return (err); | |
2108 | } | |
2109 | case DRR_FREE: | |
2110 | { | |
2111 | /* | |
2112 | * It might be beneficial to prefetch indirect blocks here, but | |
2113 | * we don't really have the data to decide for sure. | |
2114 | */ | |
2115 | err = receive_read_payload_and_next_header(ra, 0, NULL); | |
2116 | return (err); | |
2117 | } | |
2118 | case DRR_END: | |
2119 | { | |
2120 | struct drr_end *drre = &ra->rrd->header.drr_u.drr_end; | |
2121 | if (!ZIO_CHECKSUM_EQUAL(ra->prev_cksum, drre->drr_checksum)) | |
2122 | return (SET_ERROR(ECKSUM)); | |
2123 | return (0); | |
2124 | } | |
2125 | case DRR_SPILL: | |
2126 | { | |
2127 | struct drr_spill *drrs = &ra->rrd->header.drr_u.drr_spill; | |
2128 | arc_buf_t *abuf; | |
2129 | int len = DRR_SPILL_PAYLOAD_SIZE(drrs); | |
2130 | ||
2131 | /* DRR_SPILL records are either raw or uncompressed */ | |
2132 | if (ra->raw) { | |
2133 | boolean_t byteorder = ZFS_HOST_BYTEORDER ^ | |
2134 | !!DRR_IS_RAW_BYTESWAPPED(drrs->drr_flags) ^ | |
2135 | ra->byteswap; | |
2136 | ||
2137 | abuf = arc_loan_raw_buf(dmu_objset_spa(ra->os), | |
2138 | dmu_objset_id(ra->os), byteorder, drrs->drr_salt, | |
2139 | drrs->drr_iv, drrs->drr_mac, drrs->drr_type, | |
2140 | drrs->drr_compressed_size, drrs->drr_length, | |
2141 | drrs->drr_compressiontype); | |
2142 | } else { | |
2143 | abuf = arc_loan_buf(dmu_objset_spa(ra->os), | |
2144 | DMU_OT_IS_METADATA(drrs->drr_type), | |
2145 | drrs->drr_length); | |
2146 | } | |
2147 | ||
2148 | err = receive_read_payload_and_next_header(ra, len, | |
2149 | abuf->b_data); | |
2150 | if (err != 0) { | |
2151 | dmu_return_arcbuf(abuf); | |
2152 | return (err); | |
2153 | } | |
2154 | ra->rrd->arc_buf = abuf; | |
2155 | return (err); | |
2156 | } | |
2157 | case DRR_OBJECT_RANGE: | |
2158 | { | |
2159 | err = receive_read_payload_and_next_header(ra, 0, NULL); | |
2160 | return (err); | |
2161 | } | |
2162 | default: | |
2163 | return (SET_ERROR(EINVAL)); | |
2164 | } | |
2165 | } | |
2166 | ||
2167 | static void | |
2168 | dprintf_drr(struct receive_record_arg *rrd, int err) | |
2169 | { | |
2170 | #ifdef ZFS_DEBUG | |
2171 | switch (rrd->header.drr_type) { | |
2172 | case DRR_OBJECT: | |
2173 | { | |
2174 | struct drr_object *drro = &rrd->header.drr_u.drr_object; | |
2175 | dprintf("drr_type = OBJECT obj = %llu type = %u " | |
2176 | "bonustype = %u blksz = %u bonuslen = %u cksumtype = %u " | |
2177 | "compress = %u dn_slots = %u err = %d\n", | |
2178 | drro->drr_object, drro->drr_type, drro->drr_bonustype, | |
2179 | drro->drr_blksz, drro->drr_bonuslen, | |
2180 | drro->drr_checksumtype, drro->drr_compress, | |
2181 | drro->drr_dn_slots, err); | |
2182 | break; | |
2183 | } | |
2184 | case DRR_FREEOBJECTS: | |
2185 | { | |
2186 | struct drr_freeobjects *drrfo = | |
2187 | &rrd->header.drr_u.drr_freeobjects; | |
2188 | dprintf("drr_type = FREEOBJECTS firstobj = %llu " | |
2189 | "numobjs = %llu err = %d\n", | |
2190 | drrfo->drr_firstobj, drrfo->drr_numobjs, err); | |
2191 | break; | |
2192 | } | |
2193 | case DRR_WRITE: | |
2194 | { | |
2195 | struct drr_write *drrw = &rrd->header.drr_u.drr_write; | |
2196 | dprintf("drr_type = WRITE obj = %llu type = %u offset = %llu " | |
5dbf8b4e | 2197 | "lsize = %llu cksumtype = %u flags = %u " |
03916905 PD |
2198 | "compress = %u psize = %llu err = %d\n", |
2199 | drrw->drr_object, drrw->drr_type, drrw->drr_offset, | |
2200 | drrw->drr_logical_size, drrw->drr_checksumtype, | |
2201 | drrw->drr_flags, drrw->drr_compressiontype, | |
2202 | drrw->drr_compressed_size, err); | |
2203 | break; | |
2204 | } | |
2205 | case DRR_WRITE_BYREF: | |
2206 | { | |
2207 | struct drr_write_byref *drrwbr = | |
2208 | &rrd->header.drr_u.drr_write_byref; | |
2209 | dprintf("drr_type = WRITE_BYREF obj = %llu offset = %llu " | |
2210 | "length = %llu toguid = %llx refguid = %llx " | |
2211 | "refobject = %llu refoffset = %llu cksumtype = %u " | |
5dbf8b4e | 2212 | "flags = %u err = %d\n", |
03916905 PD |
2213 | drrwbr->drr_object, drrwbr->drr_offset, |
2214 | drrwbr->drr_length, drrwbr->drr_toguid, | |
2215 | drrwbr->drr_refguid, drrwbr->drr_refobject, | |
2216 | drrwbr->drr_refoffset, drrwbr->drr_checksumtype, | |
2217 | drrwbr->drr_flags, err); | |
2218 | break; | |
2219 | } | |
2220 | case DRR_WRITE_EMBEDDED: | |
2221 | { | |
2222 | struct drr_write_embedded *drrwe = | |
2223 | &rrd->header.drr_u.drr_write_embedded; | |
2224 | dprintf("drr_type = WRITE_EMBEDDED obj = %llu offset = %llu " | |
2225 | "length = %llu compress = %u etype = %u lsize = %u " | |
2226 | "psize = %u err = %d\n", | |
2227 | drrwe->drr_object, drrwe->drr_offset, drrwe->drr_length, | |
2228 | drrwe->drr_compression, drrwe->drr_etype, | |
2229 | drrwe->drr_lsize, drrwe->drr_psize, err); | |
2230 | break; | |
2231 | } | |
2232 | case DRR_FREE: | |
2233 | { | |
2234 | struct drr_free *drrf = &rrd->header.drr_u.drr_free; | |
2235 | dprintf("drr_type = FREE obj = %llu offset = %llu " | |
2236 | "length = %lld err = %d\n", | |
2237 | drrf->drr_object, drrf->drr_offset, drrf->drr_length, | |
2238 | err); | |
2239 | break; | |
2240 | } | |
2241 | case DRR_SPILL: | |
2242 | { | |
2243 | struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; | |
2244 | dprintf("drr_type = SPILL obj = %llu length = %llu " | |
2245 | "err = %d\n", drrs->drr_object, drrs->drr_length, err); | |
2246 | break; | |
2247 | } | |
5dbf8b4e TC |
2248 | case DRR_OBJECT_RANGE: |
2249 | { | |
2250 | struct drr_object_range *drror = | |
2251 | &rrd->header.drr_u.drr_object_range; | |
2252 | dprintf("drr_type = OBJECT_RANGE firstobj = %llu " | |
2253 | "numslots = %llu flags = %u err = %d\n", | |
2254 | drror->drr_firstobj, drror->drr_numslots, | |
2255 | drror->drr_flags, err); | |
2256 | break; | |
2257 | } | |
03916905 PD |
2258 | default: |
2259 | return; | |
2260 | } | |
2261 | #endif | |
2262 | } | |
2263 | ||
2264 | /* | |
2265 | * Commit the records to the pool. | |
2266 | */ | |
2267 | static int | |
2268 | receive_process_record(struct receive_writer_arg *rwa, | |
2269 | struct receive_record_arg *rrd) | |
2270 | { | |
2271 | int err; | |
2272 | ||
2273 | /* Processing in order, therefore bytes_read should be increasing. */ | |
2274 | ASSERT3U(rrd->bytes_read, >=, rwa->bytes_read); | |
2275 | rwa->bytes_read = rrd->bytes_read; | |
2276 | ||
2277 | switch (rrd->header.drr_type) { | |
2278 | case DRR_OBJECT: | |
2279 | { | |
2280 | struct drr_object *drro = &rrd->header.drr_u.drr_object; | |
2281 | err = receive_object(rwa, drro, rrd->payload); | |
2282 | kmem_free(rrd->payload, rrd->payload_size); | |
2283 | rrd->payload = NULL; | |
2284 | break; | |
2285 | } | |
2286 | case DRR_FREEOBJECTS: | |
2287 | { | |
2288 | struct drr_freeobjects *drrfo = | |
2289 | &rrd->header.drr_u.drr_freeobjects; | |
2290 | err = receive_freeobjects(rwa, drrfo); | |
2291 | break; | |
2292 | } | |
2293 | case DRR_WRITE: | |
2294 | { | |
2295 | struct drr_write *drrw = &rrd->header.drr_u.drr_write; | |
2296 | err = receive_write(rwa, drrw, rrd->arc_buf); | |
2297 | /* if receive_write() is successful, it consumes the arc_buf */ | |
2298 | if (err != 0) | |
2299 | dmu_return_arcbuf(rrd->arc_buf); | |
2300 | rrd->arc_buf = NULL; | |
2301 | rrd->payload = NULL; | |
2302 | break; | |
2303 | } | |
2304 | case DRR_WRITE_BYREF: | |
2305 | { | |
2306 | struct drr_write_byref *drrwbr = | |
2307 | &rrd->header.drr_u.drr_write_byref; | |
2308 | err = receive_write_byref(rwa, drrwbr); | |
2309 | break; | |
2310 | } | |
2311 | case DRR_WRITE_EMBEDDED: | |
2312 | { | |
2313 | struct drr_write_embedded *drrwe = | |
2314 | &rrd->header.drr_u.drr_write_embedded; | |
2315 | err = receive_write_embedded(rwa, drrwe, rrd->payload); | |
2316 | kmem_free(rrd->payload, rrd->payload_size); | |
2317 | rrd->payload = NULL; | |
2318 | break; | |
2319 | } | |
2320 | case DRR_FREE: | |
2321 | { | |
2322 | struct drr_free *drrf = &rrd->header.drr_u.drr_free; | |
2323 | err = receive_free(rwa, drrf); | |
2324 | break; | |
2325 | } | |
2326 | case DRR_SPILL: | |
2327 | { | |
2328 | struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; | |
2329 | err = receive_spill(rwa, drrs, rrd->arc_buf); | |
2330 | /* if receive_spill() is successful, it consumes the arc_buf */ | |
2331 | if (err != 0) | |
2332 | dmu_return_arcbuf(rrd->arc_buf); | |
2333 | rrd->arc_buf = NULL; | |
2334 | rrd->payload = NULL; | |
2335 | break; | |
2336 | } | |
2337 | case DRR_OBJECT_RANGE: | |
2338 | { | |
2339 | struct drr_object_range *drror = | |
2340 | &rrd->header.drr_u.drr_object_range; | |
5dbf8b4e TC |
2341 | err = receive_object_range(rwa, drror); |
2342 | break; | |
03916905 PD |
2343 | } |
2344 | default: | |
5dbf8b4e | 2345 | err = (SET_ERROR(EINVAL)); |
03916905 PD |
2346 | } |
2347 | ||
2348 | if (err != 0) | |
2349 | dprintf_drr(rrd, err); | |
2350 | ||
2351 | return (err); | |
2352 | } | |
2353 | ||
2354 | /* | |
2355 | * dmu_recv_stream's worker thread; pull records off the queue, and then call | |
2356 | * receive_process_record When we're done, signal the main thread and exit. | |
2357 | */ | |
2358 | static void | |
2359 | receive_writer_thread(void *arg) | |
2360 | { | |
2361 | struct receive_writer_arg *rwa = arg; | |
2362 | struct receive_record_arg *rrd; | |
2363 | fstrans_cookie_t cookie = spl_fstrans_mark(); | |
2364 | ||
2365 | for (rrd = bqueue_dequeue(&rwa->q); !rrd->eos_marker; | |
2366 | rrd = bqueue_dequeue(&rwa->q)) { | |
2367 | /* | |
2368 | * If there's an error, the main thread will stop putting things | |
2369 | * on the queue, but we need to clear everything in it before we | |
2370 | * can exit. | |
2371 | */ | |
2372 | if (rwa->err == 0) { | |
2373 | rwa->err = receive_process_record(rwa, rrd); | |
2374 | } else if (rrd->arc_buf != NULL) { | |
2375 | dmu_return_arcbuf(rrd->arc_buf); | |
2376 | rrd->arc_buf = NULL; | |
2377 | rrd->payload = NULL; | |
2378 | } else if (rrd->payload != NULL) { | |
2379 | kmem_free(rrd->payload, rrd->payload_size); | |
2380 | rrd->payload = NULL; | |
2381 | } | |
2382 | kmem_free(rrd, sizeof (*rrd)); | |
2383 | } | |
2384 | kmem_free(rrd, sizeof (*rrd)); | |
2385 | mutex_enter(&rwa->mutex); | |
2386 | rwa->done = B_TRUE; | |
2387 | cv_signal(&rwa->cv); | |
2388 | mutex_exit(&rwa->mutex); | |
2389 | spl_fstrans_unmark(cookie); | |
2390 | thread_exit(); | |
2391 | } | |
2392 | ||
2393 | static int | |
2394 | resume_check(struct receive_arg *ra, nvlist_t *begin_nvl) | |
2395 | { | |
2396 | uint64_t val; | |
2397 | objset_t *mos = dmu_objset_pool(ra->os)->dp_meta_objset; | |
2398 | uint64_t dsobj = dmu_objset_id(ra->os); | |
2399 | uint64_t resume_obj, resume_off; | |
2400 | ||
2401 | if (nvlist_lookup_uint64(begin_nvl, | |
2402 | "resume_object", &resume_obj) != 0 || | |
2403 | nvlist_lookup_uint64(begin_nvl, | |
2404 | "resume_offset", &resume_off) != 0) { | |
2405 | return (SET_ERROR(EINVAL)); | |
2406 | } | |
2407 | VERIFY0(zap_lookup(mos, dsobj, | |
2408 | DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val)); | |
2409 | if (resume_obj != val) | |
2410 | return (SET_ERROR(EINVAL)); | |
2411 | VERIFY0(zap_lookup(mos, dsobj, | |
2412 | DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val)); | |
2413 | if (resume_off != val) | |
2414 | return (SET_ERROR(EINVAL)); | |
2415 | ||
2416 | return (0); | |
2417 | } | |
2418 | ||
2419 | /* | |
2420 | * Read in the stream's records, one by one, and apply them to the pool. There | |
2421 | * are two threads involved; the thread that calls this function will spin up a | |
2422 | * worker thread, read the records off the stream one by one, and issue | |
2423 | * prefetches for any necessary indirect blocks. It will then push the records | |
2424 | * onto an internal blocking queue. The worker thread will pull the records off | |
2425 | * the queue, and actually write the data into the DMU. This way, the worker | |
2426 | * thread doesn't have to wait for reads to complete, since everything it needs | |
2427 | * (the indirect blocks) will be prefetched. | |
2428 | * | |
2429 | * NB: callers *must* call dmu_recv_end() if this succeeds. | |
2430 | */ | |
2431 | int | |
2432 | dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, | |
2433 | int cleanup_fd, uint64_t *action_handlep) | |
2434 | { | |
2435 | int err = 0; | |
2436 | struct receive_arg *ra; | |
2437 | struct receive_writer_arg *rwa; | |
2438 | int featureflags; | |
2439 | uint32_t payloadlen; | |
2440 | void *payload; | |
2441 | nvlist_t *begin_nvl = NULL; | |
2442 | ||
2443 | ra = kmem_zalloc(sizeof (*ra), KM_SLEEP); | |
2444 | rwa = kmem_zalloc(sizeof (*rwa), KM_SLEEP); | |
2445 | ||
2446 | ra->byteswap = drc->drc_byteswap; | |
2447 | ra->raw = drc->drc_raw; | |
2448 | ra->cksum = drc->drc_cksum; | |
2449 | ra->vp = vp; | |
2450 | ra->voff = *voffp; | |
2451 | ||
2452 | if (dsl_dataset_is_zapified(drc->drc_ds)) { | |
2453 | (void) zap_lookup(drc->drc_ds->ds_dir->dd_pool->dp_meta_objset, | |
2454 | drc->drc_ds->ds_object, DS_FIELD_RESUME_BYTES, | |
2455 | sizeof (ra->bytes_read), 1, &ra->bytes_read); | |
2456 | } | |
2457 | ||
2458 | objlist_create(&ra->ignore_objlist); | |
2459 | ||
2460 | /* these were verified in dmu_recv_begin */ | |
2461 | ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, | |
2462 | DMU_SUBSTREAM); | |
2463 | ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); | |
2464 | ||
2465 | /* | |
2466 | * Open the objset we are modifying. | |
2467 | */ | |
2468 | VERIFY0(dmu_objset_from_ds(drc->drc_ds, &ra->os)); | |
2469 | ||
2470 | ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); | |
2471 | ||
2472 | featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); | |
2473 | ra->featureflags = featureflags; | |
2474 | ||
2475 | ASSERT0(ra->os->os_encrypted && | |
2476 | (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)); | |
2477 | ||
2478 | /* if this stream is dedup'ed, set up the avl tree for guid mapping */ | |
2479 | if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { | |
2480 | minor_t minor; | |
2481 | ||
2482 | if (cleanup_fd == -1) { | |
2483 | err = SET_ERROR(EBADF); | |
2484 | goto out; | |
2485 | } | |
2486 | err = zfs_onexit_fd_hold(cleanup_fd, &minor); | |
2487 | if (err != 0) { | |
2488 | cleanup_fd = -1; | |
2489 | goto out; | |
2490 | } | |
2491 | ||
2492 | if (*action_handlep == 0) { | |
2493 | rwa->guid_to_ds_map = | |
2494 | kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); | |
2495 | avl_create(rwa->guid_to_ds_map, guid_compare, | |
2496 | sizeof (guid_map_entry_t), | |
2497 | offsetof(guid_map_entry_t, avlnode)); | |
2498 | err = zfs_onexit_add_cb(minor, | |
2499 | free_guid_map_onexit, rwa->guid_to_ds_map, | |
2500 | action_handlep); | |
2501 | if (err != 0) | |
2502 | goto out; | |
2503 | } else { | |
2504 | err = zfs_onexit_cb_data(minor, *action_handlep, | |
2505 | (void **)&rwa->guid_to_ds_map); | |
2506 | if (err != 0) | |
2507 | goto out; | |
2508 | } | |
2509 | ||
2510 | drc->drc_guid_to_ds_map = rwa->guid_to_ds_map; | |
2511 | } | |
2512 | ||
2513 | payloadlen = drc->drc_drr_begin->drr_payloadlen; | |
2514 | payload = NULL; | |
2515 | if (payloadlen != 0) | |
2516 | payload = kmem_alloc(payloadlen, KM_SLEEP); | |
2517 | ||
2518 | err = receive_read_payload_and_next_header(ra, payloadlen, payload); | |
2519 | if (err != 0) { | |
2520 | if (payloadlen != 0) | |
2521 | kmem_free(payload, payloadlen); | |
2522 | goto out; | |
2523 | } | |
2524 | if (payloadlen != 0) { | |
2525 | err = nvlist_unpack(payload, payloadlen, &begin_nvl, KM_SLEEP); | |
2526 | kmem_free(payload, payloadlen); | |
2527 | if (err != 0) | |
2528 | goto out; | |
2529 | } | |
2530 | ||
2531 | /* handle DSL encryption key payload */ | |
2532 | if (featureflags & DMU_BACKUP_FEATURE_RAW) { | |
2533 | nvlist_t *keynvl = NULL; | |
2534 | ||
2535 | ASSERT(ra->os->os_encrypted); | |
2536 | ASSERT(drc->drc_raw); | |
2537 | ||
2538 | err = nvlist_lookup_nvlist(begin_nvl, "crypt_keydata", &keynvl); | |
2539 | if (err != 0) | |
2540 | goto out; | |
2541 | ||
2542 | /* | |
2543 | * If this is a new dataset we set the key immediately. | |
2544 | * Otherwise we don't want to change the key until we | |
2545 | * are sure the rest of the receive succeeded so we stash | |
2546 | * the keynvl away until then. | |
2547 | */ | |
2548 | err = dsl_crypto_recv_raw(spa_name(ra->os->os_spa), | |
f00ab3f2 TC |
2549 | drc->drc_ds->ds_object, drc->drc_fromsnapobj, |
2550 | drc->drc_drrb->drr_type, keynvl, drc->drc_newfs); | |
03916905 PD |
2551 | if (err != 0) |
2552 | goto out; | |
2553 | ||
f00ab3f2 TC |
2554 | /* see comment in dmu_recv_end_sync() */ |
2555 | drc->drc_ivset_guid = 0; | |
2556 | (void) nvlist_lookup_uint64(keynvl, "to_ivset_guid", | |
2557 | &drc->drc_ivset_guid); | |
2558 | ||
03916905 PD |
2559 | if (!drc->drc_newfs) |
2560 | drc->drc_keynvl = fnvlist_dup(keynvl); | |
2561 | } | |
2562 | ||
2563 | if (featureflags & DMU_BACKUP_FEATURE_RESUMING) { | |
2564 | err = resume_check(ra, begin_nvl); | |
2565 | if (err != 0) | |
2566 | goto out; | |
2567 | } | |
2568 | ||
2569 | (void) bqueue_init(&rwa->q, | |
2570 | MAX(zfs_recv_queue_length, 2 * zfs_max_recordsize), | |
2571 | offsetof(struct receive_record_arg, node)); | |
2572 | cv_init(&rwa->cv, NULL, CV_DEFAULT, NULL); | |
2573 | mutex_init(&rwa->mutex, NULL, MUTEX_DEFAULT, NULL); | |
2574 | rwa->os = ra->os; | |
2575 | rwa->byteswap = drc->drc_byteswap; | |
2576 | rwa->resumable = drc->drc_resumable; | |
2577 | rwa->raw = drc->drc_raw; | |
2578 | rwa->os->os_raw_receive = drc->drc_raw; | |
2579 | ||
2580 | (void) thread_create(NULL, 0, receive_writer_thread, rwa, 0, curproc, | |
2581 | TS_RUN, minclsyspri); | |
2582 | /* | |
2583 | * We're reading rwa->err without locks, which is safe since we are the | |
2584 | * only reader, and the worker thread is the only writer. It's ok if we | |
2585 | * miss a write for an iteration or two of the loop, since the writer | |
2586 | * thread will keep freeing records we send it until we send it an eos | |
2587 | * marker. | |
2588 | * | |
2589 | * We can leave this loop in 3 ways: First, if rwa->err is | |
2590 | * non-zero. In that case, the writer thread will free the rrd we just | |
2591 | * pushed. Second, if we're interrupted; in that case, either it's the | |
2592 | * first loop and ra->rrd was never allocated, or it's later and ra->rrd | |
2593 | * has been handed off to the writer thread who will free it. Finally, | |
2594 | * if receive_read_record fails or we're at the end of the stream, then | |
2595 | * we free ra->rrd and exit. | |
2596 | */ | |
2597 | while (rwa->err == 0) { | |
2598 | if (issig(JUSTLOOKING) && issig(FORREAL)) { | |
2599 | err = SET_ERROR(EINTR); | |
2600 | break; | |
2601 | } | |
2602 | ||
2603 | ASSERT3P(ra->rrd, ==, NULL); | |
2604 | ra->rrd = ra->next_rrd; | |
2605 | ra->next_rrd = NULL; | |
2606 | /* Allocates and loads header into ra->next_rrd */ | |
2607 | err = receive_read_record(ra); | |
2608 | ||
2609 | if (ra->rrd->header.drr_type == DRR_END || err != 0) { | |
2610 | kmem_free(ra->rrd, sizeof (*ra->rrd)); | |
2611 | ra->rrd = NULL; | |
2612 | break; | |
2613 | } | |
2614 | ||
2615 | bqueue_enqueue(&rwa->q, ra->rrd, | |
2616 | sizeof (struct receive_record_arg) + ra->rrd->payload_size); | |
2617 | ra->rrd = NULL; | |
2618 | } | |
f00ab3f2 TC |
2619 | ASSERT3P(ra->rrd, ==, NULL); |
2620 | ra->rrd = kmem_zalloc(sizeof (*ra->rrd), KM_SLEEP); | |
2621 | ra->rrd->eos_marker = B_TRUE; | |
2622 | bqueue_enqueue(&rwa->q, ra->rrd, 1); | |
03916905 PD |
2623 | |
2624 | mutex_enter(&rwa->mutex); | |
2625 | while (!rwa->done) { | |
2626 | cv_wait(&rwa->cv, &rwa->mutex); | |
2627 | } | |
2628 | mutex_exit(&rwa->mutex); | |
2629 | ||
2630 | /* | |
2631 | * If we are receiving a full stream as a clone, all object IDs which | |
2632 | * are greater than the maximum ID referenced in the stream are | |
2633 | * by definition unused and must be freed. | |
2634 | */ | |
2635 | if (drc->drc_clone && drc->drc_drrb->drr_fromguid == 0) { | |
2636 | uint64_t obj = rwa->max_object + 1; | |
2637 | int free_err = 0; | |
2638 | int next_err = 0; | |
2639 | ||
2640 | while (next_err == 0) { | |
2641 | free_err = dmu_free_long_object(rwa->os, obj); | |
2642 | if (free_err != 0 && free_err != ENOENT) | |
2643 | break; | |
2644 | ||
2645 | next_err = dmu_object_next(rwa->os, &obj, FALSE, 0); | |
2646 | } | |
2647 | ||
2648 | if (err == 0) { | |
2649 | if (free_err != 0 && free_err != ENOENT) | |
2650 | err = free_err; | |
2651 | else if (next_err != ESRCH) | |
2652 | err = next_err; | |
2653 | } | |
2654 | } | |
2655 | ||
2656 | cv_destroy(&rwa->cv); | |
2657 | mutex_destroy(&rwa->mutex); | |
2658 | bqueue_destroy(&rwa->q); | |
2659 | if (err == 0) | |
2660 | err = rwa->err; | |
2661 | ||
2662 | out: | |
f00ab3f2 TC |
2663 | /* |
2664 | * If we hit an error before we started the receive_writer_thread | |
2665 | * we need to clean up the next_rrd we create by processing the | |
2666 | * DRR_BEGIN record. | |
2667 | */ | |
2668 | if (ra->next_rrd != NULL) | |
2669 | kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); | |
2670 | ||
03916905 PD |
2671 | nvlist_free(begin_nvl); |
2672 | if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) | |
2673 | zfs_onexit_fd_rele(cleanup_fd); | |
2674 | ||
2675 | if (err != 0) { | |
2676 | /* | |
2677 | * Clean up references. If receive is not resumable, | |
2678 | * destroy what we created, so we don't leave it in | |
2679 | * the inconsistent state. | |
2680 | */ | |
2681 | dmu_recv_cleanup_ds(drc); | |
2682 | nvlist_free(drc->drc_keynvl); | |
2683 | } | |
2684 | ||
2685 | *voffp = ra->voff; | |
2686 | objlist_destroy(&ra->ignore_objlist); | |
2687 | kmem_free(ra, sizeof (*ra)); | |
2688 | kmem_free(rwa, sizeof (*rwa)); | |
2689 | return (err); | |
2690 | } | |
2691 | ||
2692 | static int | |
2693 | dmu_recv_end_check(void *arg, dmu_tx_t *tx) | |
2694 | { | |
2695 | dmu_recv_cookie_t *drc = arg; | |
2696 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
2697 | int error; | |
2698 | ||
2699 | ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); | |
2700 | ||
2701 | if (!drc->drc_newfs) { | |
2702 | dsl_dataset_t *origin_head; | |
2703 | ||
2704 | error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); | |
2705 | if (error != 0) | |
2706 | return (error); | |
2707 | if (drc->drc_force) { | |
2708 | /* | |
2709 | * We will destroy any snapshots in tofs (i.e. before | |
2710 | * origin_head) that are after the origin (which is | |
2711 | * the snap before drc_ds, because drc_ds can not | |
2712 | * have any snaps of its own). | |
2713 | */ | |
2714 | uint64_t obj; | |
2715 | ||
2716 | obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
2717 | while (obj != | |
2718 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { | |
2719 | dsl_dataset_t *snap; | |
2720 | error = dsl_dataset_hold_obj(dp, obj, FTAG, | |
2721 | &snap); | |
2722 | if (error != 0) | |
2723 | break; | |
2724 | if (snap->ds_dir != origin_head->ds_dir) | |
2725 | error = SET_ERROR(EINVAL); | |
2726 | if (error == 0) { | |
2727 | error = dsl_destroy_snapshot_check_impl( | |
2728 | snap, B_FALSE); | |
2729 | } | |
2730 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; | |
2731 | dsl_dataset_rele(snap, FTAG); | |
2732 | if (error != 0) | |
2733 | break; | |
2734 | } | |
2735 | if (error != 0) { | |
2736 | dsl_dataset_rele(origin_head, FTAG); | |
2737 | return (error); | |
2738 | } | |
2739 | } | |
2740 | if (drc->drc_keynvl != NULL) { | |
2741 | error = dsl_crypto_recv_raw_key_check(drc->drc_ds, | |
2742 | drc->drc_keynvl, tx); | |
2743 | if (error != 0) { | |
2744 | dsl_dataset_rele(origin_head, FTAG); | |
2745 | return (error); | |
2746 | } | |
2747 | } | |
2748 | ||
2749 | error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, | |
2750 | origin_head, drc->drc_force, drc->drc_owner, tx); | |
2751 | if (error != 0) { | |
2752 | dsl_dataset_rele(origin_head, FTAG); | |
2753 | return (error); | |
2754 | } | |
2755 | error = dsl_dataset_snapshot_check_impl(origin_head, | |
2756 | drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); | |
2757 | dsl_dataset_rele(origin_head, FTAG); | |
2758 | if (error != 0) | |
2759 | return (error); | |
2760 | ||
2761 | error = dsl_destroy_head_check_impl(drc->drc_ds, 1); | |
2762 | } else { | |
2763 | error = dsl_dataset_snapshot_check_impl(drc->drc_ds, | |
2764 | drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); | |
2765 | } | |
2766 | return (error); | |
2767 | } | |
2768 | ||
2769 | static void | |
2770 | dmu_recv_end_sync(void *arg, dmu_tx_t *tx) | |
2771 | { | |
2772 | dmu_recv_cookie_t *drc = arg; | |
2773 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
2774 | boolean_t encrypted = drc->drc_ds->ds_dir->dd_crypto_obj != 0; | |
2775 | ||
2776 | spa_history_log_internal_ds(drc->drc_ds, "finish receiving", | |
2777 | tx, "snap=%s", drc->drc_tosnap); | |
2778 | drc->drc_ds->ds_objset->os_raw_receive = B_FALSE; | |
2779 | ||
2780 | if (!drc->drc_newfs) { | |
2781 | dsl_dataset_t *origin_head; | |
2782 | ||
2783 | VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, | |
2784 | &origin_head)); | |
2785 | ||
2786 | if (drc->drc_force) { | |
2787 | /* | |
2788 | * Destroy any snapshots of drc_tofs (origin_head) | |
2789 | * after the origin (the snap before drc_ds). | |
2790 | */ | |
2791 | uint64_t obj; | |
2792 | ||
2793 | obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
2794 | while (obj != | |
2795 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { | |
2796 | dsl_dataset_t *snap; | |
2797 | VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, | |
2798 | &snap)); | |
2799 | ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); | |
2800 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; | |
2801 | dsl_destroy_snapshot_sync_impl(snap, | |
2802 | B_FALSE, tx); | |
2803 | dsl_dataset_rele(snap, FTAG); | |
2804 | } | |
2805 | } | |
2806 | if (drc->drc_keynvl != NULL) { | |
2807 | dsl_crypto_recv_raw_key_sync(drc->drc_ds, | |
2808 | drc->drc_keynvl, tx); | |
2809 | nvlist_free(drc->drc_keynvl); | |
2810 | drc->drc_keynvl = NULL; | |
2811 | } | |
2812 | ||
2813 | VERIFY3P(drc->drc_ds->ds_prev, ==, origin_head->ds_prev); | |
2814 | ||
2815 | dsl_dataset_clone_swap_sync_impl(drc->drc_ds, | |
2816 | origin_head, tx); | |
2817 | dsl_dataset_snapshot_sync_impl(origin_head, | |
2818 | drc->drc_tosnap, tx); | |
2819 | ||
2820 | /* set snapshot's creation time and guid */ | |
2821 | dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); | |
2822 | dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = | |
2823 | drc->drc_drrb->drr_creation_time; | |
2824 | dsl_dataset_phys(origin_head->ds_prev)->ds_guid = | |
2825 | drc->drc_drrb->drr_toguid; | |
2826 | dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= | |
2827 | ~DS_FLAG_INCONSISTENT; | |
2828 | ||
2829 | dmu_buf_will_dirty(origin_head->ds_dbuf, tx); | |
2830 | dsl_dataset_phys(origin_head)->ds_flags &= | |
2831 | ~DS_FLAG_INCONSISTENT; | |
2832 | ||
2833 | drc->drc_newsnapobj = | |
2834 | dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
2835 | ||
2836 | dsl_dataset_rele(origin_head, FTAG); | |
2837 | dsl_destroy_head_sync_impl(drc->drc_ds, tx); | |
2838 | ||
2839 | if (drc->drc_owner != NULL) | |
2840 | VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); | |
2841 | } else { | |
2842 | dsl_dataset_t *ds = drc->drc_ds; | |
2843 | ||
2844 | dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); | |
2845 | ||
2846 | /* set snapshot's creation time and guid */ | |
2847 | dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
2848 | dsl_dataset_phys(ds->ds_prev)->ds_creation_time = | |
2849 | drc->drc_drrb->drr_creation_time; | |
2850 | dsl_dataset_phys(ds->ds_prev)->ds_guid = | |
2851 | drc->drc_drrb->drr_toguid; | |
2852 | dsl_dataset_phys(ds->ds_prev)->ds_flags &= | |
2853 | ~DS_FLAG_INCONSISTENT; | |
2854 | ||
2855 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
2856 | dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; | |
2857 | if (dsl_dataset_has_resume_receive_state(ds)) { | |
2858 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
2859 | DS_FIELD_RESUME_FROMGUID, tx); | |
2860 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
2861 | DS_FIELD_RESUME_OBJECT, tx); | |
2862 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
2863 | DS_FIELD_RESUME_OFFSET, tx); | |
2864 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
2865 | DS_FIELD_RESUME_BYTES, tx); | |
2866 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
2867 | DS_FIELD_RESUME_TOGUID, tx); | |
2868 | (void) zap_remove(dp->dp_meta_objset, ds->ds_object, | |
2869 | DS_FIELD_RESUME_TONAME, tx); | |
2870 | } | |
2871 | drc->drc_newsnapobj = | |
2872 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; | |
2873 | } | |
f00ab3f2 TC |
2874 | |
2875 | /* | |
2876 | * If this is a raw receive, the crypt_keydata nvlist will include | |
2877 | * a to_ivset_guid for us to set on the new snapshot. This value | |
2878 | * will override the value generated by the snapshot code. However, | |
2879 | * this value may not be present, because older implementations of | |
2880 | * the raw send code did not include this value, and we are still | |
2881 | * allowed to receive them if the zfs_disable_ivset_guid_check | |
2882 | * tunable is set, in which case we will leave the newly-generated | |
2883 | * value. | |
2884 | */ | |
2885 | if (drc->drc_raw && drc->drc_ivset_guid != 0) { | |
2886 | dmu_object_zapify(dp->dp_meta_objset, drc->drc_newsnapobj, | |
2887 | DMU_OT_DSL_DATASET, tx); | |
2888 | VERIFY0(zap_update(dp->dp_meta_objset, drc->drc_newsnapobj, | |
2889 | DS_FIELD_IVSET_GUID, sizeof (uint64_t), 1, | |
2890 | &drc->drc_ivset_guid, tx)); | |
2891 | } | |
2892 | ||
03916905 PD |
2893 | zvol_create_minors(dp->dp_spa, drc->drc_tofs, B_TRUE); |
2894 | ||
2895 | /* | |
2896 | * Release the hold from dmu_recv_begin. This must be done before | |
2897 | * we return to open context, so that when we free the dataset's dnode | |
2898 | * we can evict its bonus buffer. Since the dataset may be destroyed | |
2899 | * at this point (and therefore won't have a valid pointer to the spa) | |
2900 | * we release the key mapping manually here while we do have a valid | |
2901 | * pointer, if it exists. | |
2902 | */ | |
2903 | if (!drc->drc_raw && encrypted) { | |
2904 | (void) spa_keystore_remove_mapping(dmu_tx_pool(tx)->dp_spa, | |
2905 | drc->drc_ds->ds_object, drc->drc_ds); | |
2906 | } | |
2907 | dsl_dataset_disown(drc->drc_ds, 0, dmu_recv_tag); | |
2908 | drc->drc_ds = NULL; | |
2909 | } | |
2910 | ||
2911 | static int | |
2912 | add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj, | |
2913 | boolean_t raw) | |
2914 | { | |
2915 | dsl_pool_t *dp; | |
2916 | dsl_dataset_t *snapds; | |
2917 | guid_map_entry_t *gmep; | |
2918 | objset_t *os; | |
2919 | ds_hold_flags_t dsflags = (raw) ? 0 : DS_HOLD_FLAG_DECRYPT; | |
2920 | int err; | |
2921 | ||
2922 | ASSERT(guid_map != NULL); | |
2923 | ||
2924 | err = dsl_pool_hold(name, FTAG, &dp); | |
2925 | if (err != 0) | |
2926 | return (err); | |
2927 | gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP); | |
2928 | err = dsl_dataset_own_obj(dp, snapobj, dsflags, gmep, &snapds); | |
2929 | if (err == 0) { | |
2930 | /* | |
2931 | * If this is a deduplicated raw send stream, we need | |
2932 | * to make sure that we can still read raw blocks from | |
2933 | * earlier datasets in the stream, so we set the | |
2934 | * os_raw_receive flag now. | |
2935 | */ | |
2936 | if (raw) { | |
2937 | err = dmu_objset_from_ds(snapds, &os); | |
2938 | if (err != 0) { | |
2939 | dsl_dataset_disown(snapds, dsflags, FTAG); | |
2940 | dsl_pool_rele(dp, FTAG); | |
2941 | kmem_free(gmep, sizeof (*gmep)); | |
2942 | return (err); | |
2943 | } | |
2944 | os->os_raw_receive = B_TRUE; | |
2945 | } | |
2946 | ||
2947 | gmep->raw = raw; | |
2948 | gmep->guid = dsl_dataset_phys(snapds)->ds_guid; | |
2949 | gmep->gme_ds = snapds; | |
2950 | avl_add(guid_map, gmep); | |
2951 | } else { | |
2952 | kmem_free(gmep, sizeof (*gmep)); | |
2953 | } | |
2954 | ||
2955 | dsl_pool_rele(dp, FTAG); | |
2956 | return (err); | |
2957 | } | |
2958 | ||
2959 | static int dmu_recv_end_modified_blocks = 3; | |
2960 | ||
2961 | static int | |
2962 | dmu_recv_existing_end(dmu_recv_cookie_t *drc) | |
2963 | { | |
2964 | #ifdef _KERNEL | |
2965 | /* | |
2966 | * We will be destroying the ds; make sure its origin is unmounted if | |
2967 | * necessary. | |
2968 | */ | |
2969 | char name[ZFS_MAX_DATASET_NAME_LEN]; | |
2970 | dsl_dataset_name(drc->drc_ds, name); | |
2971 | zfs_destroy_unmount_origin(name); | |
2972 | #endif | |
2973 | ||
2974 | return (dsl_sync_task(drc->drc_tofs, | |
2975 | dmu_recv_end_check, dmu_recv_end_sync, drc, | |
2976 | dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); | |
2977 | } | |
2978 | ||
2979 | static int | |
2980 | dmu_recv_new_end(dmu_recv_cookie_t *drc) | |
2981 | { | |
2982 | return (dsl_sync_task(drc->drc_tofs, | |
2983 | dmu_recv_end_check, dmu_recv_end_sync, drc, | |
2984 | dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); | |
2985 | } | |
2986 | ||
2987 | int | |
2988 | dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) | |
2989 | { | |
2990 | int error; | |
2991 | ||
2992 | drc->drc_owner = owner; | |
2993 | ||
2994 | if (drc->drc_newfs) | |
2995 | error = dmu_recv_new_end(drc); | |
2996 | else | |
2997 | error = dmu_recv_existing_end(drc); | |
2998 | ||
2999 | if (error != 0) { | |
3000 | dmu_recv_cleanup_ds(drc); | |
3001 | nvlist_free(drc->drc_keynvl); | |
3002 | } else if (drc->drc_guid_to_ds_map != NULL) { | |
3003 | (void) add_ds_to_guidmap(drc->drc_tofs, drc->drc_guid_to_ds_map, | |
3004 | drc->drc_newsnapobj, drc->drc_raw); | |
3005 | } | |
3006 | return (error); | |
3007 | } | |
3008 | ||
3009 | /* | |
3010 | * Return TRUE if this objset is currently being received into. | |
3011 | */ | |
3012 | boolean_t | |
3013 | dmu_objset_is_receiving(objset_t *os) | |
3014 | { | |
3015 | return (os->os_dsl_dataset != NULL && | |
3016 | os->os_dsl_dataset->ds_owner == dmu_recv_tag); | |
3017 | } | |
3018 | ||
3019 | #if defined(_KERNEL) | |
3020 | module_param(zfs_recv_queue_length, int, 0644); | |
3021 | MODULE_PARM_DESC(zfs_recv_queue_length, "Maximum receive queue length"); | |
3022 | #endif |