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