]>
Commit | Line | Data |
---|---|---|
87d546d8 TG |
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 (c) 2011 by Delphix. All rights reserved. | |
24 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. | |
25 | * Copyright (c) 2014, Joyent, Inc. All rights reserved. | |
26 | * Copyright (c) 2011, 2014 by Delphix. All rights reserved. | |
27 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. | |
28 | */ | |
29 | ||
30 | #include <sys/dmu.h> | |
31 | #include <sys/dmu_impl.h> | |
32 | #include <sys/dmu_tx.h> | |
33 | #include <sys/dbuf.h> | |
34 | #include <sys/dnode.h> | |
35 | #include <sys/zfs_context.h> | |
36 | #include <sys/dmu_objset.h> | |
37 | #include <sys/dmu_traverse.h> | |
38 | #include <sys/dsl_dataset.h> | |
39 | #include <sys/dsl_dir.h> | |
40 | #include <sys/dsl_prop.h> | |
41 | #include <sys/dsl_pool.h> | |
42 | #include <sys/dsl_synctask.h> | |
43 | #include <sys/spa_impl.h> | |
44 | #include <sys/zfs_ioctl.h> | |
45 | #include <sys/zap.h> | |
46 | #include <sys/zio_checksum.h> | |
47 | #include <sys/zfs_znode.h> | |
48 | #include <zfs_fletcher.h> | |
49 | #include <sys/avl.h> | |
50 | #include <sys/ddt.h> | |
51 | #include <sys/zfs_onexit.h> | |
52 | #include <sys/dmu_send.h> | |
53 | #include <sys/dsl_destroy.h> | |
54 | #include <sys/blkptr.h> | |
55 | #include <sys/dsl_bookmark.h> | |
56 | #include <sys/zfeature.h> | |
57 | #include <sys/zvol.h> | |
58 | ||
59 | /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */ | |
60 | int zfs_send_corrupt_data = B_FALSE; | |
61 | ||
62 | static char *dmu_recv_tag = "dmu_recv_tag"; | |
63 | static const char *recv_clone_name = "%recv"; | |
64 | ||
65 | typedef struct dump_bytes_io { | |
66 | dmu_sendarg_t *dbi_dsp; | |
67 | void *dbi_buf; | |
68 | int dbi_len; | |
69 | } dump_bytes_io_t; | |
70 | ||
71 | static void | |
72 | dump_bytes_cb(void *arg) | |
73 | { | |
74 | dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg; | |
75 | dmu_sendarg_t *dsp = dbi->dbi_dsp; | |
76 | dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset; | |
77 | ssize_t resid; /* have to get resid to get detailed errno */ | |
78 | ASSERT0(dbi->dbi_len % 8); | |
79 | ||
80 | fletcher_4_incremental_native(dbi->dbi_buf, dbi->dbi_len, &dsp->dsa_zc); | |
81 | dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp, | |
82 | (caddr_t)dbi->dbi_buf, dbi->dbi_len, | |
83 | 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid); | |
84 | ||
85 | mutex_enter(&ds->ds_sendstream_lock); | |
86 | *dsp->dsa_off += dbi->dbi_len; | |
87 | mutex_exit(&ds->ds_sendstream_lock); | |
88 | } | |
89 | ||
90 | static int | |
91 | dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) | |
92 | { | |
93 | dump_bytes_io_t dbi; | |
94 | ||
95 | dbi.dbi_dsp = dsp; | |
96 | dbi.dbi_buf = buf; | |
97 | dbi.dbi_len = len; | |
98 | ||
99 | #if defined(HAVE_LARGE_STACKS) | |
100 | dump_bytes_cb(&dbi); | |
101 | #else | |
102 | /* | |
103 | * The vn_rdwr() call is performed in a taskq to ensure that there is | |
104 | * always enough stack space to write safely to the target filesystem. | |
105 | * The ZIO_TYPE_FREE threads are used because there can be a lot of | |
106 | * them and they are used in vdev_file.c for a similar purpose. | |
107 | */ | |
108 | spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE, | |
109 | ZIO_TASKQ_ISSUE, dump_bytes_cb, &dbi, TQ_SLEEP); | |
110 | #endif /* HAVE_LARGE_STACKS */ | |
111 | ||
112 | return (dsp->dsa_err); | |
113 | } | |
114 | ||
115 | static int | |
116 | dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, | |
117 | uint64_t length) | |
118 | { | |
119 | struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free); | |
120 | ||
121 | /* | |
122 | * When we receive a free record, dbuf_free_range() assumes | |
123 | * that the receiving system doesn't have any dbufs in the range | |
124 | * being freed. This is always true because there is a one-record | |
125 | * constraint: we only send one WRITE record for any given | |
126 | * object+offset. We know that the one-record constraint is | |
127 | * true because we always send data in increasing order by | |
128 | * object,offset. | |
129 | * | |
130 | * If the increasing-order constraint ever changes, we should find | |
131 | * another way to assert that the one-record constraint is still | |
132 | * satisfied. | |
133 | */ | |
134 | ASSERT(object > dsp->dsa_last_data_object || | |
135 | (object == dsp->dsa_last_data_object && | |
136 | offset > dsp->dsa_last_data_offset)); | |
137 | ||
138 | /* | |
139 | * If we are doing a non-incremental send, then there can't | |
140 | * be any data in the dataset we're receiving into. Therefore | |
141 | * a free record would simply be a no-op. Save space by not | |
142 | * sending it to begin with. | |
143 | */ | |
144 | if (!dsp->dsa_incremental) | |
145 | return (0); | |
146 | ||
147 | if (length != -1ULL && offset + length < offset) | |
148 | length = -1ULL; | |
149 | ||
150 | /* | |
151 | * If there is a pending op, but it's not PENDING_FREE, push it out, | |
152 | * since free block aggregation can only be done for blocks of the | |
153 | * same type (i.e., DRR_FREE records can only be aggregated with | |
154 | * other DRR_FREE records. DRR_FREEOBJECTS records can only be | |
155 | * aggregated with other DRR_FREEOBJECTS records. | |
156 | */ | |
157 | if (dsp->dsa_pending_op != PENDING_NONE && | |
158 | dsp->dsa_pending_op != PENDING_FREE) { | |
159 | if (dump_bytes(dsp, dsp->dsa_drr, | |
160 | sizeof (dmu_replay_record_t)) != 0) | |
161 | return (SET_ERROR(EINTR)); | |
162 | dsp->dsa_pending_op = PENDING_NONE; | |
163 | } | |
164 | ||
165 | if (dsp->dsa_pending_op == PENDING_FREE) { | |
166 | /* | |
167 | * There should never be a PENDING_FREE if length is -1 | |
168 | * (because dump_dnode is the only place where this | |
169 | * function is called with a -1, and only after flushing | |
170 | * any pending record). | |
171 | */ | |
172 | ASSERT(length != -1ULL); | |
173 | /* | |
174 | * Check to see whether this free block can be aggregated | |
175 | * with pending one. | |
176 | */ | |
177 | if (drrf->drr_object == object && drrf->drr_offset + | |
178 | drrf->drr_length == offset) { | |
179 | drrf->drr_length += length; | |
180 | return (0); | |
181 | } else { | |
182 | /* not a continuation. Push out pending record */ | |
183 | if (dump_bytes(dsp, dsp->dsa_drr, | |
184 | sizeof (dmu_replay_record_t)) != 0) | |
185 | return (SET_ERROR(EINTR)); | |
186 | dsp->dsa_pending_op = PENDING_NONE; | |
187 | } | |
188 | } | |
189 | /* create a FREE record and make it pending */ | |
190 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); | |
191 | dsp->dsa_drr->drr_type = DRR_FREE; | |
192 | drrf->drr_object = object; | |
193 | drrf->drr_offset = offset; | |
194 | drrf->drr_length = length; | |
195 | drrf->drr_toguid = dsp->dsa_toguid; | |
196 | if (length == -1ULL) { | |
197 | if (dump_bytes(dsp, dsp->dsa_drr, | |
198 | sizeof (dmu_replay_record_t)) != 0) | |
199 | return (SET_ERROR(EINTR)); | |
200 | } else { | |
201 | dsp->dsa_pending_op = PENDING_FREE; | |
202 | } | |
203 | ||
204 | return (0); | |
205 | } | |
206 | ||
207 | static int | |
208 | dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type, | |
209 | uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data) | |
210 | { | |
211 | struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write); | |
212 | ||
213 | /* | |
214 | * We send data in increasing object, offset order. | |
215 | * See comment in dump_free() for details. | |
216 | */ | |
217 | ASSERT(object > dsp->dsa_last_data_object || | |
218 | (object == dsp->dsa_last_data_object && | |
219 | offset > dsp->dsa_last_data_offset)); | |
220 | dsp->dsa_last_data_object = object; | |
221 | dsp->dsa_last_data_offset = offset + blksz - 1; | |
222 | ||
223 | /* | |
224 | * If there is any kind of pending aggregation (currently either | |
225 | * a grouping of free objects or free blocks), push it out to | |
226 | * the stream, since aggregation can't be done across operations | |
227 | * of different types. | |
228 | */ | |
229 | if (dsp->dsa_pending_op != PENDING_NONE) { | |
230 | if (dump_bytes(dsp, dsp->dsa_drr, | |
231 | sizeof (dmu_replay_record_t)) != 0) | |
232 | return (SET_ERROR(EINTR)); | |
233 | dsp->dsa_pending_op = PENDING_NONE; | |
234 | } | |
235 | /* write a DATA record */ | |
236 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); | |
237 | dsp->dsa_drr->drr_type = DRR_WRITE; | |
238 | drrw->drr_object = object; | |
239 | drrw->drr_type = type; | |
240 | drrw->drr_offset = offset; | |
241 | drrw->drr_length = blksz; | |
242 | drrw->drr_toguid = dsp->dsa_toguid; | |
243 | if (bp == NULL || BP_IS_EMBEDDED(bp)) { | |
244 | /* | |
245 | * There's no pre-computed checksum for partial-block | |
246 | * writes or embedded BP's, so (like | |
247 | * fletcher4-checkummed blocks) userland will have to | |
248 | * compute a dedup-capable checksum itself. | |
249 | */ | |
250 | drrw->drr_checksumtype = ZIO_CHECKSUM_OFF; | |
251 | } else { | |
252 | drrw->drr_checksumtype = BP_GET_CHECKSUM(bp); | |
253 | if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup) | |
254 | drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP; | |
255 | DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp)); | |
256 | DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp)); | |
257 | DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp)); | |
258 | drrw->drr_key.ddk_cksum = bp->blk_cksum; | |
259 | } | |
260 | ||
261 | if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) | |
262 | return (SET_ERROR(EINTR)); | |
263 | if (dump_bytes(dsp, data, blksz) != 0) | |
264 | return (SET_ERROR(EINTR)); | |
265 | return (0); | |
266 | } | |
267 | ||
268 | static int | |
269 | dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, | |
270 | int blksz, const blkptr_t *bp) | |
271 | { | |
272 | char buf[BPE_PAYLOAD_SIZE]; | |
273 | struct drr_write_embedded *drrw = | |
274 | &(dsp->dsa_drr->drr_u.drr_write_embedded); | |
275 | ||
276 | if (dsp->dsa_pending_op != PENDING_NONE) { | |
277 | if (dump_bytes(dsp, dsp->dsa_drr, | |
278 | sizeof (dmu_replay_record_t)) != 0) | |
279 | return (EINTR); | |
280 | dsp->dsa_pending_op = PENDING_NONE; | |
281 | } | |
282 | ||
283 | ASSERT(BP_IS_EMBEDDED(bp)); | |
284 | ||
285 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); | |
286 | dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED; | |
287 | drrw->drr_object = object; | |
288 | drrw->drr_offset = offset; | |
289 | drrw->drr_length = blksz; | |
290 | drrw->drr_toguid = dsp->dsa_toguid; | |
291 | drrw->drr_compression = BP_GET_COMPRESS(bp); | |
292 | drrw->drr_etype = BPE_GET_ETYPE(bp); | |
293 | drrw->drr_lsize = BPE_GET_LSIZE(bp); | |
294 | drrw->drr_psize = BPE_GET_PSIZE(bp); | |
295 | ||
296 | decode_embedded_bp_compressed(bp, buf); | |
297 | ||
298 | if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) | |
299 | return (EINTR); | |
300 | if (dump_bytes(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0) | |
301 | return (EINTR); | |
302 | return (0); | |
303 | } | |
304 | ||
305 | static int | |
306 | dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data) | |
307 | { | |
308 | struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill); | |
309 | ||
310 | if (dsp->dsa_pending_op != PENDING_NONE) { | |
311 | if (dump_bytes(dsp, dsp->dsa_drr, | |
312 | sizeof (dmu_replay_record_t)) != 0) | |
313 | return (SET_ERROR(EINTR)); | |
314 | dsp->dsa_pending_op = PENDING_NONE; | |
315 | } | |
316 | ||
317 | /* write a SPILL record */ | |
318 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); | |
319 | dsp->dsa_drr->drr_type = DRR_SPILL; | |
320 | drrs->drr_object = object; | |
321 | drrs->drr_length = blksz; | |
322 | drrs->drr_toguid = dsp->dsa_toguid; | |
323 | ||
324 | if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t))) | |
325 | return (SET_ERROR(EINTR)); | |
326 | if (dump_bytes(dsp, data, blksz)) | |
327 | return (SET_ERROR(EINTR)); | |
328 | return (0); | |
329 | } | |
330 | ||
331 | static int | |
332 | dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs) | |
333 | { | |
334 | struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects); | |
335 | ||
336 | /* See comment in dump_free(). */ | |
337 | if (!dsp->dsa_incremental) | |
338 | return (0); | |
339 | ||
340 | /* | |
341 | * If there is a pending op, but it's not PENDING_FREEOBJECTS, | |
342 | * push it out, since free block aggregation can only be done for | |
343 | * blocks of the same type (i.e., DRR_FREE records can only be | |
344 | * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records | |
345 | * can only be aggregated with other DRR_FREEOBJECTS records. | |
346 | */ | |
347 | if (dsp->dsa_pending_op != PENDING_NONE && | |
348 | dsp->dsa_pending_op != PENDING_FREEOBJECTS) { | |
349 | if (dump_bytes(dsp, dsp->dsa_drr, | |
350 | sizeof (dmu_replay_record_t)) != 0) | |
351 | return (SET_ERROR(EINTR)); | |
352 | dsp->dsa_pending_op = PENDING_NONE; | |
353 | } | |
354 | if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) { | |
355 | /* | |
356 | * See whether this free object array can be aggregated | |
357 | * with pending one | |
358 | */ | |
359 | if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) { | |
360 | drrfo->drr_numobjs += numobjs; | |
361 | return (0); | |
362 | } else { | |
363 | /* can't be aggregated. Push out pending record */ | |
364 | if (dump_bytes(dsp, dsp->dsa_drr, | |
365 | sizeof (dmu_replay_record_t)) != 0) | |
366 | return (SET_ERROR(EINTR)); | |
367 | dsp->dsa_pending_op = PENDING_NONE; | |
368 | } | |
369 | } | |
370 | ||
371 | /* write a FREEOBJECTS record */ | |
372 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); | |
373 | dsp->dsa_drr->drr_type = DRR_FREEOBJECTS; | |
374 | drrfo->drr_firstobj = firstobj; | |
375 | drrfo->drr_numobjs = numobjs; | |
376 | drrfo->drr_toguid = dsp->dsa_toguid; | |
377 | ||
378 | dsp->dsa_pending_op = PENDING_FREEOBJECTS; | |
379 | ||
380 | return (0); | |
381 | } | |
382 | ||
383 | static int | |
384 | dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp) | |
385 | { | |
386 | struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object); | |
387 | ||
388 | if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) | |
389 | return (dump_freeobjects(dsp, object, 1)); | |
390 | ||
391 | if (dsp->dsa_pending_op != PENDING_NONE) { | |
392 | if (dump_bytes(dsp, dsp->dsa_drr, | |
393 | sizeof (dmu_replay_record_t)) != 0) | |
394 | return (SET_ERROR(EINTR)); | |
395 | dsp->dsa_pending_op = PENDING_NONE; | |
396 | } | |
397 | ||
398 | /* write an OBJECT record */ | |
399 | bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); | |
400 | dsp->dsa_drr->drr_type = DRR_OBJECT; | |
401 | drro->drr_object = object; | |
402 | drro->drr_type = dnp->dn_type; | |
403 | drro->drr_bonustype = dnp->dn_bonustype; | |
404 | drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; | |
405 | drro->drr_bonuslen = dnp->dn_bonuslen; | |
406 | drro->drr_checksumtype = dnp->dn_checksum; | |
407 | drro->drr_compress = dnp->dn_compress; | |
408 | drro->drr_toguid = dsp->dsa_toguid; | |
409 | ||
410 | if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
411 | drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE) | |
412 | drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE; | |
413 | ||
414 | if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) | |
415 | return (SET_ERROR(EINTR)); | |
416 | ||
417 | if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0) | |
418 | return (SET_ERROR(EINTR)); | |
419 | ||
420 | /* Free anything past the end of the file. */ | |
421 | if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * | |
422 | (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0) | |
423 | return (SET_ERROR(EINTR)); | |
424 | if (dsp->dsa_err != 0) | |
425 | return (SET_ERROR(EINTR)); | |
426 | return (0); | |
427 | } | |
428 | ||
429 | static boolean_t | |
430 | backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp) | |
431 | { | |
432 | if (!BP_IS_EMBEDDED(bp)) | |
433 | return (B_FALSE); | |
434 | ||
435 | /* | |
436 | * Compression function must be legacy, or explicitly enabled. | |
437 | */ | |
438 | if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS && | |
439 | !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4))) | |
440 | return (B_FALSE); | |
441 | ||
442 | /* | |
443 | * Embed type must be explicitly enabled. | |
444 | */ | |
445 | switch (BPE_GET_ETYPE(bp)) { | |
446 | case BP_EMBEDDED_TYPE_DATA: | |
447 | if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) | |
448 | return (B_TRUE); | |
449 | break; | |
450 | default: | |
451 | return (B_FALSE); | |
452 | } | |
453 | return (B_FALSE); | |
454 | } | |
455 | ||
456 | #define BP_SPAN(dnp, level) \ | |
457 | (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \ | |
458 | (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) | |
459 | ||
460 | /* ARGSUSED */ | |
461 | static int | |
462 | backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
463 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) | |
464 | { | |
465 | dmu_sendarg_t *dsp = arg; | |
466 | dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; | |
467 | int err = 0; | |
468 | ||
469 | if (issig(JUSTLOOKING) && issig(FORREAL)) | |
470 | return (SET_ERROR(EINTR)); | |
471 | ||
472 | if (zb->zb_object != DMU_META_DNODE_OBJECT && | |
473 | DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { | |
474 | return (0); | |
475 | } else if (zb->zb_level == ZB_ZIL_LEVEL) { | |
476 | /* | |
477 | * If we are sending a non-snapshot (which is allowed on | |
478 | * read-only pools), it may have a ZIL, which must be ignored. | |
479 | */ | |
480 | return (0); | |
481 | } else if (BP_IS_HOLE(bp) && | |
482 | zb->zb_object == DMU_META_DNODE_OBJECT) { | |
483 | uint64_t span = BP_SPAN(dnp, zb->zb_level); | |
484 | uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; | |
485 | err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT); | |
486 | } else if (BP_IS_HOLE(bp)) { | |
487 | uint64_t span = BP_SPAN(dnp, zb->zb_level); | |
488 | err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span); | |
489 | } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { | |
490 | return (0); | |
491 | } else if (type == DMU_OT_DNODE) { | |
492 | dnode_phys_t *blk; | |
493 | int i; | |
494 | int blksz = BP_GET_LSIZE(bp); | |
495 | arc_flags_t aflags = ARC_FLAG_WAIT; | |
496 | arc_buf_t *abuf; | |
497 | ||
498 | if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, | |
499 | ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, | |
500 | &aflags, zb) != 0) | |
501 | return (SET_ERROR(EIO)); | |
502 | ||
503 | blk = abuf->b_data; | |
504 | for (i = 0; i < blksz >> DNODE_SHIFT; i++) { | |
505 | uint64_t dnobj = (zb->zb_blkid << | |
506 | (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i; | |
507 | err = dump_dnode(dsp, dnobj, blk+i); | |
508 | if (err != 0) | |
509 | break; | |
510 | } | |
511 | (void) arc_buf_remove_ref(abuf, &abuf); | |
512 | } else if (type == DMU_OT_SA) { | |
513 | arc_flags_t aflags = ARC_FLAG_WAIT; | |
514 | arc_buf_t *abuf; | |
515 | int blksz = BP_GET_LSIZE(bp); | |
516 | ||
517 | if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, | |
518 | ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, | |
519 | &aflags, zb) != 0) | |
520 | return (SET_ERROR(EIO)); | |
521 | ||
522 | err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data); | |
523 | (void) arc_buf_remove_ref(abuf, &abuf); | |
524 | } else if (backup_do_embed(dsp, bp)) { | |
525 | /* it's an embedded level-0 block of a regular object */ | |
526 | int blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; | |
527 | err = dump_write_embedded(dsp, zb->zb_object, | |
528 | zb->zb_blkid * blksz, blksz, bp); | |
529 | } else { /* it's a level-0 block of a regular object */ | |
530 | uint64_t offset; | |
531 | arc_flags_t aflags = ARC_FLAG_WAIT; | |
532 | arc_buf_t *abuf; | |
533 | int blksz = BP_GET_LSIZE(bp); | |
534 | ||
535 | ASSERT3U(blksz, ==, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
536 | ASSERT0(zb->zb_level); | |
537 | if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, | |
538 | ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, | |
539 | &aflags, zb) != 0) { | |
540 | if (zfs_send_corrupt_data) { | |
541 | uint64_t *ptr; | |
542 | /* Send a block filled with 0x"zfs badd bloc" */ | |
543 | abuf = arc_buf_alloc(spa, blksz, &abuf, | |
544 | ARC_BUFC_DATA); | |
545 | for (ptr = abuf->b_data; | |
546 | (char *)ptr < (char *)abuf->b_data + blksz; | |
547 | ptr++) | |
548 | *ptr = 0x2f5baddb10cULL; | |
549 | } else { | |
550 | return (SET_ERROR(EIO)); | |
551 | } | |
552 | } | |
553 | ||
554 | offset = zb->zb_blkid * blksz; | |
555 | ||
556 | if (!(dsp->dsa_featureflags & | |
557 | DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
558 | blksz > SPA_OLD_MAXBLOCKSIZE) { | |
559 | char *buf = abuf->b_data; | |
560 | while (blksz > 0 && err == 0) { | |
561 | int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE); | |
562 | err = dump_write(dsp, type, zb->zb_object, | |
563 | offset, n, NULL, buf); | |
564 | offset += n; | |
565 | buf += n; | |
566 | blksz -= n; | |
567 | } | |
568 | } else { | |
569 | err = dump_write(dsp, type, zb->zb_object, | |
570 | offset, blksz, bp, abuf->b_data); | |
571 | } | |
572 | (void) arc_buf_remove_ref(abuf, &abuf); | |
573 | } | |
574 | ||
575 | ASSERT(err == 0 || err == EINTR); | |
576 | return (err); | |
577 | } | |
578 | ||
579 | /* | |
580 | * Releases dp using the specified tag. | |
581 | */ | |
582 | static int | |
583 | dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds, | |
584 | zfs_bookmark_phys_t *fromzb, boolean_t is_clone, boolean_t embedok, | |
585 | boolean_t large_block_ok, int outfd, vnode_t *vp, offset_t *off) | |
586 | { | |
587 | objset_t *os; | |
588 | dmu_replay_record_t *drr; | |
589 | dmu_sendarg_t *dsp; | |
590 | int err; | |
591 | uint64_t fromtxg = 0; | |
592 | uint64_t featureflags = 0; | |
593 | ||
594 | err = dmu_objset_from_ds(ds, &os); | |
595 | if (err != 0) { | |
596 | dsl_pool_rele(dp, tag); | |
597 | return (err); | |
598 | } | |
599 | ||
600 | drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); | |
601 | drr->drr_type = DRR_BEGIN; | |
602 | drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; | |
603 | DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo, | |
604 | DMU_SUBSTREAM); | |
605 | ||
606 | #ifdef _KERNEL | |
607 | if (dmu_objset_type(os) == DMU_OST_ZFS) { | |
608 | uint64_t version; | |
609 | if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { | |
610 | kmem_free(drr, sizeof (dmu_replay_record_t)); | |
611 | dsl_pool_rele(dp, tag); | |
612 | return (SET_ERROR(EINVAL)); | |
613 | } | |
614 | if (version >= ZPL_VERSION_SA) { | |
615 | featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; | |
616 | } | |
617 | } | |
618 | #endif | |
619 | ||
620 | if (large_block_ok && ds->ds_large_blocks) | |
621 | featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS; | |
622 | if (embedok && | |
623 | spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) { | |
624 | featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA; | |
625 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) | |
626 | featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA_LZ4; | |
627 | } else { | |
628 | embedok = B_FALSE; | |
629 | } | |
630 | ||
631 | DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo, | |
632 | featureflags); | |
633 | ||
634 | drr->drr_u.drr_begin.drr_creation_time = | |
635 | dsl_dataset_phys(ds)->ds_creation_time; | |
636 | drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); | |
637 | if (is_clone) | |
638 | drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; | |
639 | drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(ds)->ds_guid; | |
640 | if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) | |
641 | drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; | |
642 | ||
643 | if (fromzb != NULL) { | |
644 | drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid; | |
645 | fromtxg = fromzb->zbm_creation_txg; | |
646 | } | |
647 | dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname); | |
648 | if (!ds->ds_is_snapshot) { | |
649 | (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--", | |
650 | sizeof (drr->drr_u.drr_begin.drr_toname)); | |
651 | } | |
652 | ||
653 | dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); | |
654 | ||
655 | dsp->dsa_drr = drr; | |
656 | dsp->dsa_vp = vp; | |
657 | dsp->dsa_outfd = outfd; | |
658 | dsp->dsa_proc = curproc; | |
659 | dsp->dsa_os = os; | |
660 | dsp->dsa_off = off; | |
661 | dsp->dsa_toguid = dsl_dataset_phys(ds)->ds_guid; | |
662 | ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0); | |
663 | dsp->dsa_pending_op = PENDING_NONE; | |
664 | dsp->dsa_incremental = (fromzb != NULL); | |
665 | dsp->dsa_featureflags = featureflags; | |
666 | ||
667 | mutex_enter(&ds->ds_sendstream_lock); | |
668 | list_insert_head(&ds->ds_sendstreams, dsp); | |
669 | mutex_exit(&ds->ds_sendstream_lock); | |
670 | ||
671 | dsl_dataset_long_hold(ds, FTAG); | |
672 | dsl_pool_rele(dp, tag); | |
673 | ||
674 | if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) { | |
675 | err = dsp->dsa_err; | |
676 | goto out; | |
677 | } | |
678 | ||
679 | err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH, | |
680 | backup_cb, dsp); | |
681 | ||
682 | if (dsp->dsa_pending_op != PENDING_NONE) | |
683 | if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) | |
684 | err = SET_ERROR(EINTR); | |
685 | ||
686 | if (err != 0) { | |
687 | if (err == EINTR && dsp->dsa_err != 0) | |
688 | err = dsp->dsa_err; | |
689 | goto out; | |
690 | } | |
691 | ||
692 | bzero(drr, sizeof (dmu_replay_record_t)); | |
693 | drr->drr_type = DRR_END; | |
694 | drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc; | |
695 | drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid; | |
696 | ||
697 | if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) { | |
698 | err = dsp->dsa_err; | |
699 | goto out; | |
700 | } | |
701 | ||
702 | out: | |
703 | mutex_enter(&ds->ds_sendstream_lock); | |
704 | list_remove(&ds->ds_sendstreams, dsp); | |
705 | mutex_exit(&ds->ds_sendstream_lock); | |
706 | ||
707 | kmem_free(drr, sizeof (dmu_replay_record_t)); | |
708 | kmem_free(dsp, sizeof (dmu_sendarg_t)); | |
709 | ||
710 | dsl_dataset_long_rele(ds, FTAG); | |
711 | ||
712 | return (err); | |
713 | } | |
714 | ||
715 | int | |
716 | dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, | |
717 | boolean_t embedok, boolean_t large_block_ok, | |
718 | int outfd, vnode_t *vp, offset_t *off) | |
719 | { | |
720 | dsl_pool_t *dp; | |
721 | dsl_dataset_t *ds; | |
722 | dsl_dataset_t *fromds = NULL; | |
723 | int err; | |
724 | ||
725 | err = dsl_pool_hold(pool, FTAG, &dp); | |
726 | if (err != 0) | |
727 | return (err); | |
728 | ||
729 | err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds); | |
730 | if (err != 0) { | |
731 | dsl_pool_rele(dp, FTAG); | |
732 | return (err); | |
733 | } | |
734 | ||
735 | if (fromsnap != 0) { | |
736 | zfs_bookmark_phys_t zb; | |
737 | boolean_t is_clone; | |
738 | ||
739 | err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); | |
740 | if (err != 0) { | |
741 | dsl_dataset_rele(ds, FTAG); | |
742 | dsl_pool_rele(dp, FTAG); | |
743 | return (err); | |
744 | } | |
745 | if (!dsl_dataset_is_before(ds, fromds, 0)) | |
746 | err = SET_ERROR(EXDEV); | |
747 | zb.zbm_creation_time = | |
748 | dsl_dataset_phys(fromds)->ds_creation_time; | |
749 | zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg; | |
750 | zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; | |
751 | is_clone = (fromds->ds_dir != ds->ds_dir); | |
752 | dsl_dataset_rele(fromds, FTAG); | |
753 | err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, | |
754 | embedok, large_block_ok, outfd, vp, off); | |
755 | } else { | |
756 | err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, | |
757 | embedok, large_block_ok, outfd, vp, off); | |
758 | } | |
759 | dsl_dataset_rele(ds, FTAG); | |
760 | return (err); | |
761 | } | |
762 | ||
763 | int | |
764 | dmu_send(const char *tosnap, const char *fromsnap, | |
765 | boolean_t embedok, boolean_t large_block_ok, | |
766 | int outfd, vnode_t *vp, offset_t *off) | |
767 | { | |
768 | dsl_pool_t *dp; | |
769 | dsl_dataset_t *ds; | |
770 | int err; | |
771 | boolean_t owned = B_FALSE; | |
772 | ||
773 | if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL) | |
774 | return (SET_ERROR(EINVAL)); | |
775 | ||
776 | err = dsl_pool_hold(tosnap, FTAG, &dp); | |
777 | if (err != 0) | |
778 | return (err); | |
779 | ||
780 | if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) { | |
781 | /* | |
782 | * We are sending a filesystem or volume. Ensure | |
783 | * that it doesn't change by owning the dataset. | |
784 | */ | |
785 | err = dsl_dataset_own(dp, tosnap, FTAG, &ds); | |
786 | owned = B_TRUE; | |
787 | } else { | |
788 | err = dsl_dataset_hold(dp, tosnap, FTAG, &ds); | |
789 | } | |
790 | if (err != 0) { | |
791 | dsl_pool_rele(dp, FTAG); | |
792 | return (err); | |
793 | } | |
794 | ||
795 | if (fromsnap != NULL) { | |
796 | zfs_bookmark_phys_t zb; | |
797 | boolean_t is_clone = B_FALSE; | |
798 | int fsnamelen = strchr(tosnap, '@') - tosnap; | |
799 | ||
800 | /* | |
801 | * If the fromsnap is in a different filesystem, then | |
802 | * mark the send stream as a clone. | |
803 | */ | |
804 | if (strncmp(tosnap, fromsnap, fsnamelen) != 0 || | |
805 | (fromsnap[fsnamelen] != '@' && | |
806 | fromsnap[fsnamelen] != '#')) { | |
807 | is_clone = B_TRUE; | |
808 | } | |
809 | ||
810 | if (strchr(fromsnap, '@')) { | |
811 | dsl_dataset_t *fromds; | |
812 | err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); | |
813 | if (err == 0) { | |
814 | if (!dsl_dataset_is_before(ds, fromds, 0)) | |
815 | err = SET_ERROR(EXDEV); | |
816 | zb.zbm_creation_time = | |
817 | dsl_dataset_phys(fromds)->ds_creation_time; | |
818 | zb.zbm_creation_txg = | |
819 | dsl_dataset_phys(fromds)->ds_creation_txg; | |
820 | zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; | |
821 | is_clone = (ds->ds_dir != fromds->ds_dir); | |
822 | dsl_dataset_rele(fromds, FTAG); | |
823 | } | |
824 | } else { | |
825 | err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb); | |
826 | } | |
827 | if (err != 0) { | |
828 | dsl_dataset_rele(ds, FTAG); | |
829 | dsl_pool_rele(dp, FTAG); | |
830 | return (err); | |
831 | } | |
832 | err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, | |
833 | embedok, large_block_ok, outfd, vp, off); | |
834 | } else { | |
835 | err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, | |
836 | embedok, large_block_ok, outfd, vp, off); | |
837 | } | |
838 | if (owned) | |
839 | dsl_dataset_disown(ds, FTAG); | |
840 | else | |
841 | dsl_dataset_rele(ds, FTAG); | |
842 | return (err); | |
843 | } | |
844 | ||
845 | static int | |
846 | dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t size, | |
847 | uint64_t *sizep) | |
848 | { | |
849 | int err; | |
850 | /* | |
851 | * Assume that space (both on-disk and in-stream) is dominated by | |
852 | * data. We will adjust for indirect blocks and the copies property, | |
853 | * but ignore per-object space used (eg, dnodes and DRR_OBJECT records). | |
854 | */ | |
855 | ||
856 | /* | |
857 | * Subtract out approximate space used by indirect blocks. | |
858 | * Assume most space is used by data blocks (non-indirect, non-dnode). | |
859 | * Assume all blocks are recordsize. Assume ditto blocks and | |
860 | * internal fragmentation counter out compression. | |
861 | * | |
862 | * Therefore, space used by indirect blocks is sizeof(blkptr_t) per | |
863 | * block, which we observe in practice. | |
864 | */ | |
865 | uint64_t recordsize; | |
866 | err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize); | |
867 | if (err != 0) | |
868 | return (err); | |
869 | size -= size / recordsize * sizeof (blkptr_t); | |
870 | ||
871 | /* Add in the space for the record associated with each block. */ | |
872 | size += size / recordsize * sizeof (dmu_replay_record_t); | |
873 | ||
874 | *sizep = size; | |
875 | ||
876 | return (0); | |
877 | } | |
878 | ||
879 | int | |
880 | dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep) | |
881 | { | |
882 | int err; | |
883 | uint64_t size; | |
884 | ||
885 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); | |
886 | ||
887 | /* tosnap must be a snapshot */ | |
888 | if (!ds->ds_is_snapshot) | |
889 | return (SET_ERROR(EINVAL)); | |
890 | ||
891 | /* fromsnap, if provided, must be a snapshot */ | |
892 | if (fromds != NULL && !fromds->ds_is_snapshot) | |
893 | return (SET_ERROR(EINVAL)); | |
894 | ||
895 | /* | |
896 | * fromsnap must be an earlier snapshot from the same fs as tosnap, | |
897 | * or the origin's fs. | |
898 | */ | |
899 | if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0)) | |
900 | return (SET_ERROR(EXDEV)); | |
901 | ||
902 | /* Get uncompressed size estimate of changed data. */ | |
903 | if (fromds == NULL) { | |
904 | size = dsl_dataset_phys(ds)->ds_uncompressed_bytes; | |
905 | } else { | |
906 | uint64_t used, comp; | |
907 | err = dsl_dataset_space_written(fromds, ds, | |
908 | &used, &comp, &size); | |
909 | if (err != 0) | |
910 | return (err); | |
911 | } | |
912 | ||
913 | err = dmu_adjust_send_estimate_for_indirects(ds, size, sizep); | |
914 | return (err); | |
915 | } | |
916 | ||
917 | /* | |
918 | * Simple callback used to traverse the blocks of a snapshot and sum their | |
919 | * uncompressed size | |
920 | */ | |
921 | /* ARGSUSED */ | |
922 | static int | |
923 | dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, | |
924 | const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) | |
925 | { | |
926 | uint64_t *spaceptr = arg; | |
927 | if (bp != NULL && !BP_IS_HOLE(bp)) { | |
928 | *spaceptr += BP_GET_UCSIZE(bp); | |
929 | } | |
930 | return (0); | |
931 | } | |
932 | ||
933 | /* | |
934 | * Given a desination snapshot and a TXG, calculate the approximate size of a | |
935 | * send stream sent from that TXG. from_txg may be zero, indicating that the | |
936 | * whole snapshot will be sent. | |
937 | */ | |
938 | int | |
939 | dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg, | |
940 | uint64_t *sizep) | |
941 | { | |
942 | int err; | |
943 | uint64_t size = 0; | |
944 | ||
945 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); | |
946 | ||
947 | /* tosnap must be a snapshot */ | |
948 | if (!dsl_dataset_is_snapshot(ds)) | |
949 | return (SET_ERROR(EINVAL)); | |
950 | ||
951 | /* verify that from_txg is before the provided snapshot was taken */ | |
952 | if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) { | |
953 | return (SET_ERROR(EXDEV)); | |
954 | } | |
955 | /* | |
956 | * traverse the blocks of the snapshot with birth times after | |
957 | * from_txg, summing their uncompressed size | |
958 | */ | |
959 | err = traverse_dataset(ds, from_txg, TRAVERSE_POST, | |
960 | dmu_calculate_send_traversal, &size); | |
961 | if (err) | |
962 | return (err); | |
963 | ||
964 | err = dmu_adjust_send_estimate_for_indirects(ds, size, sizep); | |
965 | return (err); | |
966 | } | |
967 | ||
968 | typedef struct dmu_recv_begin_arg { | |
969 | const char *drba_origin; | |
970 | dmu_recv_cookie_t *drba_cookie; | |
971 | cred_t *drba_cred; | |
972 | uint64_t drba_snapobj; | |
973 | } dmu_recv_begin_arg_t; | |
974 | ||
975 | static int | |
976 | recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, | |
977 | uint64_t fromguid) | |
978 | { | |
979 | uint64_t val; | |
980 | int error; | |
981 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
982 | ||
983 | /* temporary clone name must not exist */ | |
984 | error = zap_lookup(dp->dp_meta_objset, | |
985 | dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, | |
986 | 8, 1, &val); | |
987 | if (error != ENOENT) | |
988 | return (error == 0 ? EBUSY : error); | |
989 | ||
990 | /* new snapshot name must not exist */ | |
991 | error = zap_lookup(dp->dp_meta_objset, | |
992 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, | |
993 | drba->drba_cookie->drc_tosnap, 8, 1, &val); | |
994 | if (error != ENOENT) | |
995 | return (error == 0 ? EEXIST : error); | |
996 | ||
997 | /* | |
998 | * Check snapshot limit before receiving. We'll recheck again at the | |
999 | * end, but might as well abort before receiving if we're already over | |
1000 | * the limit. | |
1001 | * | |
1002 | * Note that we do not check the file system limit with | |
1003 | * dsl_dir_fscount_check because the temporary %clones don't count | |
1004 | * against that limit. | |
1005 | */ | |
1006 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, | |
1007 | NULL, drba->drba_cred); | |
1008 | if (error != 0) | |
1009 | return (error); | |
1010 | ||
1011 | if (fromguid != 0) { | |
1012 | dsl_dataset_t *snap; | |
1013 | uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; | |
1014 | ||
1015 | /* Find snapshot in this dir that matches fromguid. */ | |
1016 | while (obj != 0) { | |
1017 | error = dsl_dataset_hold_obj(dp, obj, FTAG, | |
1018 | &snap); | |
1019 | if (error != 0) | |
1020 | return (SET_ERROR(ENODEV)); | |
1021 | if (snap->ds_dir != ds->ds_dir) { | |
1022 | dsl_dataset_rele(snap, FTAG); | |
1023 | return (SET_ERROR(ENODEV)); | |
1024 | } | |
1025 | if (dsl_dataset_phys(snap)->ds_guid == fromguid) | |
1026 | break; | |
1027 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; | |
1028 | dsl_dataset_rele(snap, FTAG); | |
1029 | } | |
1030 | if (obj == 0) | |
1031 | return (SET_ERROR(ENODEV)); | |
1032 | ||
1033 | if (drba->drba_cookie->drc_force) { | |
1034 | drba->drba_snapobj = obj; | |
1035 | } else { | |
1036 | /* | |
1037 | * If we are not forcing, there must be no | |
1038 | * changes since fromsnap. | |
1039 | */ | |
1040 | if (dsl_dataset_modified_since_snap(ds, snap)) { | |
1041 | dsl_dataset_rele(snap, FTAG); | |
1042 | return (SET_ERROR(ETXTBSY)); | |
1043 | } | |
1044 | drba->drba_snapobj = ds->ds_prev->ds_object; | |
1045 | } | |
1046 | ||
1047 | dsl_dataset_rele(snap, FTAG); | |
1048 | } else { | |
1049 | /* if full, then must be forced */ | |
1050 | if (!drba->drba_cookie->drc_force) | |
1051 | return (SET_ERROR(EEXIST)); | |
1052 | /* start from $ORIGIN@$ORIGIN, if supported */ | |
1053 | drba->drba_snapobj = dp->dp_origin_snap != NULL ? | |
1054 | dp->dp_origin_snap->ds_object : 0; | |
1055 | } | |
1056 | ||
1057 | return (0); | |
1058 | ||
1059 | } | |
1060 | ||
1061 | static int | |
1062 | dmu_recv_begin_check(void *arg, dmu_tx_t *tx) | |
1063 | { | |
1064 | dmu_recv_begin_arg_t *drba = arg; | |
1065 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1066 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
1067 | uint64_t fromguid = drrb->drr_fromguid; | |
1068 | int flags = drrb->drr_flags; | |
1069 | int error; | |
1070 | uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); | |
1071 | dsl_dataset_t *ds; | |
1072 | const char *tofs = drba->drba_cookie->drc_tofs; | |
1073 | ||
1074 | /* already checked */ | |
1075 | ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); | |
1076 | ||
1077 | if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == | |
1078 | DMU_COMPOUNDSTREAM || | |
1079 | drrb->drr_type >= DMU_OST_NUMTYPES || | |
1080 | ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) | |
1081 | return (SET_ERROR(EINVAL)); | |
1082 | ||
1083 | /* Verify pool version supports SA if SA_SPILL feature set */ | |
1084 | if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && | |
1085 | spa_version(dp->dp_spa) < SPA_VERSION_SA) | |
1086 | return (SET_ERROR(ENOTSUP)); | |
1087 | ||
1088 | /* | |
1089 | * The receiving code doesn't know how to translate a WRITE_EMBEDDED | |
1090 | * record to a plan WRITE record, so the pool must have the | |
1091 | * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED | |
1092 | * records. Same with WRITE_EMBEDDED records that use LZ4 compression. | |
1093 | */ | |
1094 | if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && | |
1095 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) | |
1096 | return (SET_ERROR(ENOTSUP)); | |
1097 | if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) && | |
1098 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) | |
1099 | return (SET_ERROR(ENOTSUP)); | |
1100 | ||
1101 | /* | |
1102 | * The receiving code doesn't know how to translate large blocks | |
1103 | * to smaller ones, so the pool must have the LARGE_BLOCKS | |
1104 | * feature enabled if the stream has LARGE_BLOCKS. | |
1105 | */ | |
1106 | if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
1107 | !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) | |
1108 | return (SET_ERROR(ENOTSUP)); | |
1109 | ||
1110 | error = dsl_dataset_hold(dp, tofs, FTAG, &ds); | |
1111 | if (error == 0) { | |
1112 | /* target fs already exists; recv into temp clone */ | |
1113 | ||
1114 | /* Can't recv a clone into an existing fs */ | |
1115 | if (flags & DRR_FLAG_CLONE) { | |
1116 | dsl_dataset_rele(ds, FTAG); | |
1117 | return (SET_ERROR(EINVAL)); | |
1118 | } | |
1119 | ||
1120 | error = recv_begin_check_existing_impl(drba, ds, fromguid); | |
1121 | dsl_dataset_rele(ds, FTAG); | |
1122 | } else if (error == ENOENT) { | |
1123 | /* target fs does not exist; must be a full backup or clone */ | |
1124 | char buf[MAXNAMELEN]; | |
1125 | ||
1126 | /* | |
1127 | * If it's a non-clone incremental, we are missing the | |
1128 | * target fs, so fail the recv. | |
1129 | */ | |
1130 | if (fromguid != 0 && !(flags & DRR_FLAG_CLONE)) | |
1131 | return (SET_ERROR(ENOENT)); | |
1132 | ||
1133 | /* Open the parent of tofs */ | |
1134 | ASSERT3U(strlen(tofs), <, MAXNAMELEN); | |
1135 | (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); | |
1136 | error = dsl_dataset_hold(dp, buf, FTAG, &ds); | |
1137 | if (error != 0) | |
1138 | return (error); | |
1139 | ||
1140 | /* | |
1141 | * Check filesystem and snapshot limits before receiving. We'll | |
1142 | * recheck snapshot limits again at the end (we create the | |
1143 | * filesystems and increment those counts during begin_sync). | |
1144 | */ | |
1145 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, | |
1146 | ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred); | |
1147 | if (error != 0) { | |
1148 | dsl_dataset_rele(ds, FTAG); | |
1149 | return (error); | |
1150 | } | |
1151 | ||
1152 | error = dsl_fs_ss_limit_check(ds->ds_dir, 1, | |
1153 | ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); | |
1154 | if (error != 0) { | |
1155 | dsl_dataset_rele(ds, FTAG); | |
1156 | return (error); | |
1157 | } | |
1158 | ||
1159 | if (drba->drba_origin != NULL) { | |
1160 | dsl_dataset_t *origin; | |
1161 | error = dsl_dataset_hold(dp, drba->drba_origin, | |
1162 | FTAG, &origin); | |
1163 | if (error != 0) { | |
1164 | dsl_dataset_rele(ds, FTAG); | |
1165 | return (error); | |
1166 | } | |
1167 | if (!origin->ds_is_snapshot) { | |
1168 | dsl_dataset_rele(origin, FTAG); | |
1169 | dsl_dataset_rele(ds, FTAG); | |
1170 | return (SET_ERROR(EINVAL)); | |
1171 | } | |
1172 | if (dsl_dataset_phys(origin)->ds_guid != fromguid) { | |
1173 | dsl_dataset_rele(origin, FTAG); | |
1174 | dsl_dataset_rele(ds, FTAG); | |
1175 | return (SET_ERROR(ENODEV)); | |
1176 | } | |
1177 | dsl_dataset_rele(origin, FTAG); | |
1178 | } | |
1179 | dsl_dataset_rele(ds, FTAG); | |
1180 | error = 0; | |
1181 | } | |
1182 | return (error); | |
1183 | } | |
1184 | ||
1185 | static void | |
1186 | dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) | |
1187 | { | |
1188 | dmu_recv_begin_arg_t *drba = arg; | |
1189 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1190 | struct drr_begin *drrb = drba->drba_cookie->drc_drrb; | |
1191 | const char *tofs = drba->drba_cookie->drc_tofs; | |
1192 | dsl_dataset_t *ds, *newds; | |
1193 | uint64_t dsobj; | |
1194 | int error; | |
1195 | uint64_t crflags; | |
1196 | ||
1197 | crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ? | |
1198 | DS_FLAG_CI_DATASET : 0; | |
1199 | ||
1200 | error = dsl_dataset_hold(dp, tofs, FTAG, &ds); | |
1201 | if (error == 0) { | |
1202 | /* create temporary clone */ | |
1203 | dsl_dataset_t *snap = NULL; | |
1204 | if (drba->drba_snapobj != 0) { | |
1205 | VERIFY0(dsl_dataset_hold_obj(dp, | |
1206 | drba->drba_snapobj, FTAG, &snap)); | |
1207 | } | |
1208 | dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, | |
1209 | snap, crflags, drba->drba_cred, tx); | |
1210 | dsl_dataset_rele(snap, FTAG); | |
1211 | dsl_dataset_rele(ds, FTAG); | |
1212 | } else { | |
1213 | dsl_dir_t *dd; | |
1214 | const char *tail; | |
1215 | dsl_dataset_t *origin = NULL; | |
1216 | ||
1217 | VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); | |
1218 | ||
1219 | if (drba->drba_origin != NULL) { | |
1220 | VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, | |
1221 | FTAG, &origin)); | |
1222 | } | |
1223 | ||
1224 | /* Create new dataset. */ | |
1225 | dsobj = dsl_dataset_create_sync(dd, | |
1226 | strrchr(tofs, '/') + 1, | |
1227 | origin, crflags, drba->drba_cred, tx); | |
1228 | if (origin != NULL) | |
1229 | dsl_dataset_rele(origin, FTAG); | |
1230 | dsl_dir_rele(dd, FTAG); | |
1231 | drba->drba_cookie->drc_newfs = B_TRUE; | |
1232 | } | |
1233 | VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds)); | |
1234 | ||
1235 | if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & | |
1236 | DMU_BACKUP_FEATURE_LARGE_BLOCKS) && | |
1237 | !newds->ds_large_blocks) { | |
1238 | dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); | |
1239 | newds->ds_large_blocks = B_TRUE; | |
1240 | } | |
1241 | ||
1242 | dmu_buf_will_dirty(newds->ds_dbuf, tx); | |
1243 | dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; | |
1244 | ||
1245 | /* | |
1246 | * If we actually created a non-clone, we need to create the | |
1247 | * objset in our new dataset. | |
1248 | */ | |
1249 | if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) { | |
1250 | (void) dmu_objset_create_impl(dp->dp_spa, | |
1251 | newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); | |
1252 | } | |
1253 | ||
1254 | drba->drba_cookie->drc_ds = newds; | |
1255 | ||
1256 | spa_history_log_internal_ds(newds, "receive", tx, ""); | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() | |
1261 | * succeeds; otherwise we will leak the holds on the datasets. | |
1262 | */ | |
1263 | int | |
1264 | dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb, | |
1265 | boolean_t force, char *origin, dmu_recv_cookie_t *drc) | |
1266 | { | |
1267 | dmu_recv_begin_arg_t drba = { 0 }; | |
1268 | dmu_replay_record_t *drr; | |
1269 | ||
1270 | bzero(drc, sizeof (dmu_recv_cookie_t)); | |
1271 | drc->drc_drrb = drrb; | |
1272 | drc->drc_tosnap = tosnap; | |
1273 | drc->drc_tofs = tofs; | |
1274 | drc->drc_force = force; | |
1275 | drc->drc_cred = CRED(); | |
1276 | ||
1277 | if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) | |
1278 | drc->drc_byteswap = B_TRUE; | |
1279 | else if (drrb->drr_magic != DMU_BACKUP_MAGIC) | |
1280 | return (SET_ERROR(EINVAL)); | |
1281 | ||
1282 | drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); | |
1283 | drr->drr_type = DRR_BEGIN; | |
1284 | drr->drr_u.drr_begin = *drc->drc_drrb; | |
1285 | if (drc->drc_byteswap) { | |
1286 | fletcher_4_incremental_byteswap(drr, | |
1287 | sizeof (dmu_replay_record_t), &drc->drc_cksum); | |
1288 | } else { | |
1289 | fletcher_4_incremental_native(drr, | |
1290 | sizeof (dmu_replay_record_t), &drc->drc_cksum); | |
1291 | } | |
1292 | kmem_free(drr, sizeof (dmu_replay_record_t)); | |
1293 | ||
1294 | if (drc->drc_byteswap) { | |
1295 | drrb->drr_magic = BSWAP_64(drrb->drr_magic); | |
1296 | drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo); | |
1297 | drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); | |
1298 | drrb->drr_type = BSWAP_32(drrb->drr_type); | |
1299 | drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); | |
1300 | drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); | |
1301 | } | |
1302 | ||
1303 | drba.drba_origin = origin; | |
1304 | drba.drba_cookie = drc; | |
1305 | drba.drba_cred = CRED(); | |
1306 | ||
1307 | return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync, | |
1308 | &drba, 5, ZFS_SPACE_CHECK_NORMAL)); | |
1309 | } | |
1310 | ||
1311 | struct restorearg { | |
1312 | int err; | |
1313 | boolean_t byteswap; | |
1314 | vnode_t *vp; | |
1315 | char *buf; | |
1316 | uint64_t voff; | |
1317 | int bufsize; /* amount of memory allocated for buf */ | |
1318 | zio_cksum_t cksum; | |
1319 | avl_tree_t *guid_to_ds_map; | |
1320 | }; | |
1321 | ||
1322 | typedef struct guid_map_entry { | |
1323 | uint64_t guid; | |
1324 | dsl_dataset_t *gme_ds; | |
1325 | avl_node_t avlnode; | |
1326 | } guid_map_entry_t; | |
1327 | ||
1328 | static int | |
1329 | guid_compare(const void *arg1, const void *arg2) | |
1330 | { | |
1331 | const guid_map_entry_t *gmep1 = arg1; | |
1332 | const guid_map_entry_t *gmep2 = arg2; | |
1333 | ||
1334 | if (gmep1->guid < gmep2->guid) | |
1335 | return (-1); | |
1336 | else if (gmep1->guid > gmep2->guid) | |
1337 | return (1); | |
1338 | return (0); | |
1339 | } | |
1340 | ||
1341 | static void | |
1342 | free_guid_map_onexit(void *arg) | |
1343 | { | |
1344 | avl_tree_t *ca = arg; | |
1345 | void *cookie = NULL; | |
1346 | guid_map_entry_t *gmep; | |
1347 | ||
1348 | while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { | |
1349 | dsl_dataset_long_rele(gmep->gme_ds, gmep); | |
1350 | dsl_dataset_rele(gmep->gme_ds, gmep); | |
1351 | kmem_free(gmep, sizeof (guid_map_entry_t)); | |
1352 | } | |
1353 | avl_destroy(ca); | |
1354 | kmem_free(ca, sizeof (avl_tree_t)); | |
1355 | } | |
1356 | ||
1357 | static void * | |
1358 | restore_read(struct restorearg *ra, int len, char *buf) | |
1359 | { | |
1360 | int done = 0; | |
1361 | ||
1362 | if (buf == NULL) | |
1363 | buf = ra->buf; | |
1364 | ||
1365 | /* some things will require 8-byte alignment, so everything must */ | |
1366 | ASSERT0(len % 8); | |
1367 | ASSERT3U(len, <=, ra->bufsize); | |
1368 | ||
1369 | while (done < len) { | |
1370 | ssize_t resid; | |
1371 | ||
1372 | ra->err = vn_rdwr(UIO_READ, ra->vp, | |
1373 | buf + done, len - done, | |
1374 | ra->voff, UIO_SYSSPACE, FAPPEND, | |
1375 | RLIM64_INFINITY, CRED(), &resid); | |
1376 | ||
1377 | if (resid == len - done) | |
1378 | ra->err = SET_ERROR(EINVAL); | |
1379 | ra->voff += len - done - resid; | |
1380 | done = len - resid; | |
1381 | if (ra->err != 0) | |
1382 | return (NULL); | |
1383 | } | |
1384 | ||
1385 | ASSERT3U(done, ==, len); | |
1386 | if (ra->byteswap) | |
1387 | fletcher_4_incremental_byteswap(buf, len, &ra->cksum); | |
1388 | else | |
1389 | fletcher_4_incremental_native(buf, len, &ra->cksum); | |
1390 | return (buf); | |
1391 | } | |
1392 | ||
1393 | noinline static void | |
1394 | backup_byteswap(dmu_replay_record_t *drr) | |
1395 | { | |
1396 | #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) | |
1397 | #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) | |
1398 | drr->drr_type = BSWAP_32(drr->drr_type); | |
1399 | drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); | |
1400 | switch (drr->drr_type) { | |
1401 | case DRR_BEGIN: | |
1402 | DO64(drr_begin.drr_magic); | |
1403 | DO64(drr_begin.drr_versioninfo); | |
1404 | DO64(drr_begin.drr_creation_time); | |
1405 | DO32(drr_begin.drr_type); | |
1406 | DO32(drr_begin.drr_flags); | |
1407 | DO64(drr_begin.drr_toguid); | |
1408 | DO64(drr_begin.drr_fromguid); | |
1409 | break; | |
1410 | case DRR_OBJECT: | |
1411 | DO64(drr_object.drr_object); | |
1412 | DO32(drr_object.drr_type); | |
1413 | DO32(drr_object.drr_bonustype); | |
1414 | DO32(drr_object.drr_blksz); | |
1415 | DO32(drr_object.drr_bonuslen); | |
1416 | DO64(drr_object.drr_toguid); | |
1417 | break; | |
1418 | case DRR_FREEOBJECTS: | |
1419 | DO64(drr_freeobjects.drr_firstobj); | |
1420 | DO64(drr_freeobjects.drr_numobjs); | |
1421 | DO64(drr_freeobjects.drr_toguid); | |
1422 | break; | |
1423 | case DRR_WRITE: | |
1424 | DO64(drr_write.drr_object); | |
1425 | DO32(drr_write.drr_type); | |
1426 | DO64(drr_write.drr_offset); | |
1427 | DO64(drr_write.drr_length); | |
1428 | DO64(drr_write.drr_toguid); | |
1429 | DO64(drr_write.drr_key.ddk_cksum.zc_word[0]); | |
1430 | DO64(drr_write.drr_key.ddk_cksum.zc_word[1]); | |
1431 | DO64(drr_write.drr_key.ddk_cksum.zc_word[2]); | |
1432 | DO64(drr_write.drr_key.ddk_cksum.zc_word[3]); | |
1433 | DO64(drr_write.drr_key.ddk_prop); | |
1434 | break; | |
1435 | case DRR_WRITE_BYREF: | |
1436 | DO64(drr_write_byref.drr_object); | |
1437 | DO64(drr_write_byref.drr_offset); | |
1438 | DO64(drr_write_byref.drr_length); | |
1439 | DO64(drr_write_byref.drr_toguid); | |
1440 | DO64(drr_write_byref.drr_refguid); | |
1441 | DO64(drr_write_byref.drr_refobject); | |
1442 | DO64(drr_write_byref.drr_refoffset); | |
1443 | DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]); | |
1444 | DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]); | |
1445 | DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]); | |
1446 | DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]); | |
1447 | DO64(drr_write_byref.drr_key.ddk_prop); | |
1448 | break; | |
1449 | case DRR_WRITE_EMBEDDED: | |
1450 | DO64(drr_write_embedded.drr_object); | |
1451 | DO64(drr_write_embedded.drr_offset); | |
1452 | DO64(drr_write_embedded.drr_length); | |
1453 | DO64(drr_write_embedded.drr_toguid); | |
1454 | DO32(drr_write_embedded.drr_lsize); | |
1455 | DO32(drr_write_embedded.drr_psize); | |
1456 | break; | |
1457 | case DRR_FREE: | |
1458 | DO64(drr_free.drr_object); | |
1459 | DO64(drr_free.drr_offset); | |
1460 | DO64(drr_free.drr_length); | |
1461 | DO64(drr_free.drr_toguid); | |
1462 | break; | |
1463 | case DRR_SPILL: | |
1464 | DO64(drr_spill.drr_object); | |
1465 | DO64(drr_spill.drr_length); | |
1466 | DO64(drr_spill.drr_toguid); | |
1467 | break; | |
1468 | case DRR_END: | |
1469 | DO64(drr_end.drr_checksum.zc_word[0]); | |
1470 | DO64(drr_end.drr_checksum.zc_word[1]); | |
1471 | DO64(drr_end.drr_checksum.zc_word[2]); | |
1472 | DO64(drr_end.drr_checksum.zc_word[3]); | |
1473 | DO64(drr_end.drr_toguid); | |
1474 | break; | |
1475 | default: | |
1476 | break; | |
1477 | } | |
1478 | #undef DO64 | |
1479 | #undef DO32 | |
1480 | } | |
1481 | ||
1482 | static inline uint8_t | |
1483 | deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) | |
1484 | { | |
1485 | if (bonus_type == DMU_OT_SA) { | |
1486 | return (1); | |
1487 | } else { | |
1488 | return (1 + | |
1489 | ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT)); | |
1490 | } | |
1491 | } | |
1492 | ||
1493 | noinline static int | |
1494 | restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) | |
1495 | { | |
1496 | dmu_object_info_t doi; | |
1497 | dmu_tx_t *tx; | |
1498 | void *data = NULL; | |
1499 | uint64_t object; | |
1500 | int err; | |
1501 | ||
1502 | if (drro->drr_type == DMU_OT_NONE || | |
1503 | !DMU_OT_IS_VALID(drro->drr_type) || | |
1504 | !DMU_OT_IS_VALID(drro->drr_bonustype) || | |
1505 | drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || | |
1506 | drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || | |
1507 | P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || | |
1508 | drro->drr_blksz < SPA_MINBLOCKSIZE || | |
1509 | drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(os)) || | |
1510 | drro->drr_bonuslen > DN_MAX_BONUSLEN) { | |
1511 | return (SET_ERROR(EINVAL)); | |
1512 | } | |
1513 | ||
1514 | err = dmu_object_info(os, drro->drr_object, &doi); | |
1515 | ||
1516 | if (err != 0 && err != ENOENT) | |
1517 | return (SET_ERROR(EINVAL)); | |
1518 | object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT; | |
1519 | ||
1520 | if (drro->drr_bonuslen) { | |
1521 | data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8), NULL); | |
1522 | if (ra->err != 0) | |
1523 | return (ra->err); | |
1524 | } | |
1525 | ||
1526 | /* | |
1527 | * If we are losing blkptrs or changing the block size this must | |
1528 | * be a new file instance. We must clear out the previous file | |
1529 | * contents before we can change this type of metadata in the dnode. | |
1530 | */ | |
1531 | if (err == 0) { | |
1532 | int nblkptr; | |
1533 | ||
1534 | nblkptr = deduce_nblkptr(drro->drr_bonustype, | |
1535 | drro->drr_bonuslen); | |
1536 | ||
1537 | if (drro->drr_blksz != doi.doi_data_block_size || | |
1538 | nblkptr < doi.doi_nblkptr) { | |
1539 | err = dmu_free_long_range(os, drro->drr_object, | |
1540 | 0, DMU_OBJECT_END); | |
1541 | if (err != 0) | |
1542 | return (SET_ERROR(EINVAL)); | |
1543 | } | |
1544 | } | |
1545 | ||
1546 | tx = dmu_tx_create(os); | |
1547 | dmu_tx_hold_bonus(tx, object); | |
1548 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1549 | if (err != 0) { | |
1550 | dmu_tx_abort(tx); | |
1551 | return (err); | |
1552 | } | |
1553 | ||
1554 | if (object == DMU_NEW_OBJECT) { | |
1555 | /* currently free, want to be allocated */ | |
1556 | err = dmu_object_claim(os, drro->drr_object, | |
1557 | drro->drr_type, drro->drr_blksz, | |
1558 | drro->drr_bonustype, drro->drr_bonuslen, tx); | |
1559 | } else if (drro->drr_type != doi.doi_type || | |
1560 | drro->drr_blksz != doi.doi_data_block_size || | |
1561 | drro->drr_bonustype != doi.doi_bonus_type || | |
1562 | drro->drr_bonuslen != doi.doi_bonus_size) { | |
1563 | /* currently allocated, but with different properties */ | |
1564 | err = dmu_object_reclaim(os, drro->drr_object, | |
1565 | drro->drr_type, drro->drr_blksz, | |
1566 | drro->drr_bonustype, drro->drr_bonuslen, tx); | |
1567 | } | |
1568 | if (err != 0) { | |
1569 | dmu_tx_commit(tx); | |
1570 | return (SET_ERROR(EINVAL)); | |
1571 | } | |
1572 | ||
1573 | dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype, | |
1574 | tx); | |
1575 | dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx); | |
1576 | ||
1577 | if (data != NULL) { | |
1578 | dmu_buf_t *db; | |
1579 | ||
1580 | VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db)); | |
1581 | dmu_buf_will_dirty(db, tx); | |
1582 | ||
1583 | ASSERT3U(db->db_size, >=, drro->drr_bonuslen); | |
1584 | bcopy(data, db->db_data, drro->drr_bonuslen); | |
1585 | if (ra->byteswap) { | |
1586 | dmu_object_byteswap_t byteswap = | |
1587 | DMU_OT_BYTESWAP(drro->drr_bonustype); | |
1588 | dmu_ot_byteswap[byteswap].ob_func(db->db_data, | |
1589 | drro->drr_bonuslen); | |
1590 | } | |
1591 | dmu_buf_rele(db, FTAG); | |
1592 | } | |
1593 | dmu_tx_commit(tx); | |
1594 | return (0); | |
1595 | } | |
1596 | ||
1597 | /* ARGSUSED */ | |
1598 | noinline static int | |
1599 | restore_freeobjects(struct restorearg *ra, objset_t *os, | |
1600 | struct drr_freeobjects *drrfo) | |
1601 | { | |
1602 | uint64_t obj; | |
1603 | ||
1604 | if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) | |
1605 | return (SET_ERROR(EINVAL)); | |
1606 | ||
1607 | for (obj = drrfo->drr_firstobj; | |
1608 | obj < drrfo->drr_firstobj + drrfo->drr_numobjs; | |
1609 | (void) dmu_object_next(os, &obj, FALSE, 0)) { | |
1610 | int err; | |
1611 | ||
1612 | if (dmu_object_info(os, obj, NULL) != 0) | |
1613 | continue; | |
1614 | ||
1615 | err = dmu_free_long_object(os, obj); | |
1616 | if (err != 0) | |
1617 | return (err); | |
1618 | } | |
1619 | return (0); | |
1620 | } | |
1621 | ||
1622 | noinline static int | |
1623 | restore_write(struct restorearg *ra, objset_t *os, | |
1624 | struct drr_write *drrw) | |
1625 | { | |
1626 | dmu_tx_t *tx; | |
1627 | dmu_buf_t *bonus; | |
1628 | arc_buf_t *abuf; | |
1629 | void *data; | |
1630 | int err; | |
1631 | ||
1632 | if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset || | |
1633 | !DMU_OT_IS_VALID(drrw->drr_type)) | |
1634 | return (SET_ERROR(EINVAL)); | |
1635 | ||
1636 | if (dmu_object_info(os, drrw->drr_object, NULL) != 0) | |
1637 | return (SET_ERROR(EINVAL)); | |
1638 | ||
1639 | if (dmu_bonus_hold(os, drrw->drr_object, FTAG, &bonus) != 0) | |
1640 | return (SET_ERROR(EINVAL)); | |
1641 | ||
1642 | abuf = dmu_request_arcbuf(bonus, drrw->drr_length); | |
1643 | ||
1644 | data = restore_read(ra, drrw->drr_length, abuf->b_data); | |
1645 | if (data == NULL) { | |
1646 | dmu_return_arcbuf(abuf); | |
1647 | dmu_buf_rele(bonus, FTAG); | |
1648 | return (ra->err); | |
1649 | } | |
1650 | ||
1651 | tx = dmu_tx_create(os); | |
1652 | ||
1653 | dmu_tx_hold_write(tx, drrw->drr_object, | |
1654 | drrw->drr_offset, drrw->drr_length); | |
1655 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1656 | if (err != 0) { | |
1657 | dmu_return_arcbuf(abuf); | |
1658 | dmu_buf_rele(bonus, FTAG); | |
1659 | dmu_tx_abort(tx); | |
1660 | return (err); | |
1661 | } | |
1662 | if (ra->byteswap) { | |
1663 | dmu_object_byteswap_t byteswap = | |
1664 | DMU_OT_BYTESWAP(drrw->drr_type); | |
1665 | dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length); | |
1666 | } | |
1667 | dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx); | |
1668 | dmu_tx_commit(tx); | |
1669 | dmu_buf_rele(bonus, FTAG); | |
1670 | return (0); | |
1671 | } | |
1672 | ||
1673 | /* | |
1674 | * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed | |
1675 | * streams to refer to a copy of the data that is already on the | |
1676 | * system because it came in earlier in the stream. This function | |
1677 | * finds the earlier copy of the data, and uses that copy instead of | |
1678 | * data from the stream to fulfill this write. | |
1679 | */ | |
1680 | static int | |
1681 | restore_write_byref(struct restorearg *ra, objset_t *os, | |
1682 | struct drr_write_byref *drrwbr) | |
1683 | { | |
1684 | dmu_tx_t *tx; | |
1685 | int err; | |
1686 | guid_map_entry_t gmesrch; | |
1687 | guid_map_entry_t *gmep; | |
1688 | avl_index_t where; | |
1689 | objset_t *ref_os = NULL; | |
1690 | dmu_buf_t *dbp; | |
1691 | ||
1692 | if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) | |
1693 | return (SET_ERROR(EINVAL)); | |
1694 | ||
1695 | /* | |
1696 | * If the GUID of the referenced dataset is different from the | |
1697 | * GUID of the target dataset, find the referenced dataset. | |
1698 | */ | |
1699 | if (drrwbr->drr_toguid != drrwbr->drr_refguid) { | |
1700 | gmesrch.guid = drrwbr->drr_refguid; | |
1701 | if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch, | |
1702 | &where)) == NULL) { | |
1703 | return (SET_ERROR(EINVAL)); | |
1704 | } | |
1705 | if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) | |
1706 | return (SET_ERROR(EINVAL)); | |
1707 | } else { | |
1708 | ref_os = os; | |
1709 | } | |
1710 | ||
1711 | err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, | |
1712 | drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH); | |
1713 | if (err != 0) | |
1714 | return (err); | |
1715 | ||
1716 | tx = dmu_tx_create(os); | |
1717 | ||
1718 | dmu_tx_hold_write(tx, drrwbr->drr_object, | |
1719 | drrwbr->drr_offset, drrwbr->drr_length); | |
1720 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1721 | if (err != 0) { | |
1722 | dmu_tx_abort(tx); | |
1723 | return (err); | |
1724 | } | |
1725 | dmu_write(os, drrwbr->drr_object, | |
1726 | drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); | |
1727 | dmu_buf_rele(dbp, FTAG); | |
1728 | dmu_tx_commit(tx); | |
1729 | return (0); | |
1730 | } | |
1731 | ||
1732 | static int | |
1733 | restore_write_embedded(struct restorearg *ra, objset_t *os, | |
1734 | struct drr_write_embedded *drrwnp) | |
1735 | { | |
1736 | dmu_tx_t *tx; | |
1737 | int err; | |
1738 | void *data; | |
1739 | ||
1740 | if (drrwnp->drr_offset + drrwnp->drr_length < drrwnp->drr_offset) | |
1741 | return (EINVAL); | |
1742 | ||
1743 | if (drrwnp->drr_psize > BPE_PAYLOAD_SIZE) | |
1744 | return (EINVAL); | |
1745 | ||
1746 | if (drrwnp->drr_etype >= NUM_BP_EMBEDDED_TYPES) | |
1747 | return (EINVAL); | |
1748 | if (drrwnp->drr_compression >= ZIO_COMPRESS_FUNCTIONS) | |
1749 | return (EINVAL); | |
1750 | ||
1751 | data = restore_read(ra, P2ROUNDUP(drrwnp->drr_psize, 8), NULL); | |
1752 | if (data == NULL) | |
1753 | return (ra->err); | |
1754 | ||
1755 | tx = dmu_tx_create(os); | |
1756 | ||
1757 | dmu_tx_hold_write(tx, drrwnp->drr_object, | |
1758 | drrwnp->drr_offset, drrwnp->drr_length); | |
1759 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1760 | if (err != 0) { | |
1761 | dmu_tx_abort(tx); | |
1762 | return (err); | |
1763 | } | |
1764 | ||
1765 | dmu_write_embedded(os, drrwnp->drr_object, | |
1766 | drrwnp->drr_offset, data, drrwnp->drr_etype, | |
1767 | drrwnp->drr_compression, drrwnp->drr_lsize, drrwnp->drr_psize, | |
1768 | ra->byteswap ^ ZFS_HOST_BYTEORDER, tx); | |
1769 | ||
1770 | dmu_tx_commit(tx); | |
1771 | return (0); | |
1772 | } | |
1773 | ||
1774 | static int | |
1775 | restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs) | |
1776 | { | |
1777 | dmu_tx_t *tx; | |
1778 | void *data; | |
1779 | dmu_buf_t *db, *db_spill; | |
1780 | int err; | |
1781 | ||
1782 | if (drrs->drr_length < SPA_MINBLOCKSIZE || | |
1783 | drrs->drr_length > spa_maxblocksize(dmu_objset_spa(os))) | |
1784 | return (SET_ERROR(EINVAL)); | |
1785 | ||
1786 | data = restore_read(ra, drrs->drr_length, NULL); | |
1787 | if (data == NULL) | |
1788 | return (ra->err); | |
1789 | ||
1790 | if (dmu_object_info(os, drrs->drr_object, NULL) != 0) | |
1791 | return (SET_ERROR(EINVAL)); | |
1792 | ||
1793 | VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db)); | |
1794 | if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) { | |
1795 | dmu_buf_rele(db, FTAG); | |
1796 | return (err); | |
1797 | } | |
1798 | ||
1799 | tx = dmu_tx_create(os); | |
1800 | ||
1801 | dmu_tx_hold_spill(tx, db->db_object); | |
1802 | ||
1803 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1804 | if (err != 0) { | |
1805 | dmu_buf_rele(db, FTAG); | |
1806 | dmu_buf_rele(db_spill, FTAG); | |
1807 | dmu_tx_abort(tx); | |
1808 | return (err); | |
1809 | } | |
1810 | dmu_buf_will_dirty(db_spill, tx); | |
1811 | ||
1812 | if (db_spill->db_size < drrs->drr_length) | |
1813 | VERIFY(0 == dbuf_spill_set_blksz(db_spill, | |
1814 | drrs->drr_length, tx)); | |
1815 | bcopy(data, db_spill->db_data, drrs->drr_length); | |
1816 | ||
1817 | dmu_buf_rele(db, FTAG); | |
1818 | dmu_buf_rele(db_spill, FTAG); | |
1819 | ||
1820 | dmu_tx_commit(tx); | |
1821 | return (0); | |
1822 | } | |
1823 | ||
1824 | /* ARGSUSED */ | |
1825 | noinline static int | |
1826 | restore_free(struct restorearg *ra, objset_t *os, | |
1827 | struct drr_free *drrf) | |
1828 | { | |
1829 | int err; | |
1830 | ||
1831 | if (drrf->drr_length != -1ULL && | |
1832 | drrf->drr_offset + drrf->drr_length < drrf->drr_offset) | |
1833 | return (SET_ERROR(EINVAL)); | |
1834 | ||
1835 | if (dmu_object_info(os, drrf->drr_object, NULL) != 0) | |
1836 | return (SET_ERROR(EINVAL)); | |
1837 | ||
1838 | err = dmu_free_long_range(os, drrf->drr_object, | |
1839 | drrf->drr_offset, drrf->drr_length); | |
1840 | return (err); | |
1841 | } | |
1842 | ||
1843 | /* used to destroy the drc_ds on error */ | |
1844 | static void | |
1845 | dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) | |
1846 | { | |
1847 | char name[MAXNAMELEN]; | |
1848 | dsl_dataset_name(drc->drc_ds, name); | |
1849 | dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); | |
1850 | (void) dsl_destroy_head(name); | |
1851 | } | |
1852 | ||
1853 | /* | |
1854 | * NB: callers *must* call dmu_recv_end() if this succeeds. | |
1855 | */ | |
1856 | int | |
1857 | dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, | |
1858 | int cleanup_fd, uint64_t *action_handlep) | |
1859 | { | |
1860 | struct restorearg ra = { 0 }; | |
1861 | dmu_replay_record_t *drr; | |
1862 | objset_t *os; | |
1863 | zio_cksum_t pcksum; | |
1864 | int featureflags; | |
1865 | ||
1866 | ra.byteswap = drc->drc_byteswap; | |
1867 | ra.cksum = drc->drc_cksum; | |
1868 | ra.vp = vp; | |
1869 | ra.voff = *voffp; | |
1870 | ra.bufsize = SPA_MAXBLOCKSIZE; | |
1871 | ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP); | |
1872 | ||
1873 | /* these were verified in dmu_recv_begin */ | |
1874 | ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, | |
1875 | DMU_SUBSTREAM); | |
1876 | ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); | |
1877 | ||
1878 | /* | |
1879 | * Open the objset we are modifying. | |
1880 | */ | |
1881 | VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os)); | |
1882 | ||
1883 | ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); | |
1884 | ||
1885 | featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); | |
1886 | ||
1887 | /* if this stream is dedup'ed, set up the avl tree for guid mapping */ | |
1888 | if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { | |
1889 | minor_t minor; | |
1890 | ||
1891 | if (cleanup_fd == -1) { | |
1892 | ra.err = SET_ERROR(EBADF); | |
1893 | goto out; | |
1894 | } | |
1895 | ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor); | |
1896 | if (ra.err != 0) { | |
1897 | cleanup_fd = -1; | |
1898 | goto out; | |
1899 | } | |
1900 | ||
1901 | if (*action_handlep == 0) { | |
1902 | ra.guid_to_ds_map = | |
1903 | kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); | |
1904 | avl_create(ra.guid_to_ds_map, guid_compare, | |
1905 | sizeof (guid_map_entry_t), | |
1906 | offsetof(guid_map_entry_t, avlnode)); | |
1907 | ra.err = zfs_onexit_add_cb(minor, | |
1908 | free_guid_map_onexit, ra.guid_to_ds_map, | |
1909 | action_handlep); | |
1910 | if (ra.err != 0) | |
1911 | goto out; | |
1912 | } else { | |
1913 | ra.err = zfs_onexit_cb_data(minor, *action_handlep, | |
1914 | (void **)&ra.guid_to_ds_map); | |
1915 | if (ra.err != 0) | |
1916 | goto out; | |
1917 | } | |
1918 | ||
1919 | drc->drc_guid_to_ds_map = ra.guid_to_ds_map; | |
1920 | } | |
1921 | ||
1922 | /* | |
1923 | * Read records and process them. | |
1924 | */ | |
1925 | pcksum = ra.cksum; | |
1926 | while (ra.err == 0 && | |
1927 | NULL != (drr = restore_read(&ra, sizeof (*drr), NULL))) { | |
1928 | if (issig(JUSTLOOKING) && issig(FORREAL)) { | |
1929 | ra.err = SET_ERROR(EINTR); | |
1930 | goto out; | |
1931 | } | |
1932 | ||
1933 | if (ra.byteswap) | |
1934 | backup_byteswap(drr); | |
1935 | ||
1936 | switch (drr->drr_type) { | |
1937 | case DRR_OBJECT: | |
1938 | { | |
1939 | /* | |
1940 | * We need to make a copy of the record header, | |
1941 | * because restore_{object,write} may need to | |
1942 | * restore_read(), which will invalidate drr. | |
1943 | */ | |
1944 | struct drr_object drro = drr->drr_u.drr_object; | |
1945 | ra.err = restore_object(&ra, os, &drro); | |
1946 | break; | |
1947 | } | |
1948 | case DRR_FREEOBJECTS: | |
1949 | { | |
1950 | struct drr_freeobjects drrfo = | |
1951 | drr->drr_u.drr_freeobjects; | |
1952 | ra.err = restore_freeobjects(&ra, os, &drrfo); | |
1953 | break; | |
1954 | } | |
1955 | case DRR_WRITE: | |
1956 | { | |
1957 | struct drr_write drrw = drr->drr_u.drr_write; | |
1958 | ra.err = restore_write(&ra, os, &drrw); | |
1959 | break; | |
1960 | } | |
1961 | case DRR_WRITE_BYREF: | |
1962 | { | |
1963 | struct drr_write_byref drrwbr = | |
1964 | drr->drr_u.drr_write_byref; | |
1965 | ra.err = restore_write_byref(&ra, os, &drrwbr); | |
1966 | break; | |
1967 | } | |
1968 | case DRR_WRITE_EMBEDDED: | |
1969 | { | |
1970 | struct drr_write_embedded drrwe = | |
1971 | drr->drr_u.drr_write_embedded; | |
1972 | ra.err = restore_write_embedded(&ra, os, &drrwe); | |
1973 | break; | |
1974 | } | |
1975 | case DRR_FREE: | |
1976 | { | |
1977 | struct drr_free drrf = drr->drr_u.drr_free; | |
1978 | ra.err = restore_free(&ra, os, &drrf); | |
1979 | break; | |
1980 | } | |
1981 | case DRR_END: | |
1982 | { | |
1983 | struct drr_end drre = drr->drr_u.drr_end; | |
1984 | /* | |
1985 | * We compare against the *previous* checksum | |
1986 | * value, because the stored checksum is of | |
1987 | * everything before the DRR_END record. | |
1988 | */ | |
1989 | if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum)) | |
1990 | ra.err = SET_ERROR(ECKSUM); | |
1991 | goto out; | |
1992 | } | |
1993 | case DRR_SPILL: | |
1994 | { | |
1995 | struct drr_spill drrs = drr->drr_u.drr_spill; | |
1996 | ra.err = restore_spill(&ra, os, &drrs); | |
1997 | break; | |
1998 | } | |
1999 | default: | |
2000 | ra.err = SET_ERROR(EINVAL); | |
2001 | goto out; | |
2002 | } | |
2003 | pcksum = ra.cksum; | |
2004 | } | |
2005 | ASSERT(ra.err != 0); | |
2006 | ||
2007 | out: | |
2008 | if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) | |
2009 | zfs_onexit_fd_rele(cleanup_fd); | |
2010 | ||
2011 | if (ra.err != 0) { | |
2012 | /* | |
2013 | * destroy what we created, so we don't leave it in the | |
2014 | * inconsistent restoring state. | |
2015 | */ | |
2016 | dmu_recv_cleanup_ds(drc); | |
2017 | } | |
2018 | ||
2019 | vmem_free(ra.buf, ra.bufsize); | |
2020 | *voffp = ra.voff; | |
2021 | return (ra.err); | |
2022 | } | |
2023 | ||
2024 | static int | |
2025 | dmu_recv_end_check(void *arg, dmu_tx_t *tx) | |
2026 | { | |
2027 | dmu_recv_cookie_t *drc = arg; | |
2028 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
2029 | int error; | |
2030 | ||
2031 | ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); | |
2032 | ||
2033 | if (!drc->drc_newfs) { | |
2034 | dsl_dataset_t *origin_head; | |
2035 | ||
2036 | error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); | |
2037 | if (error != 0) | |
2038 | return (error); | |
2039 | if (drc->drc_force) { | |
2040 | /* | |
2041 | * We will destroy any snapshots in tofs (i.e. before | |
2042 | * origin_head) that are after the origin (which is | |
2043 | * the snap before drc_ds, because drc_ds can not | |
2044 | * have any snaps of its own). | |
2045 | */ | |
2046 | uint64_t obj; | |
2047 | ||
2048 | obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
2049 | while (obj != | |
2050 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { | |
2051 | dsl_dataset_t *snap; | |
2052 | error = dsl_dataset_hold_obj(dp, obj, FTAG, | |
2053 | &snap); | |
2054 | if (error != 0) | |
2055 | break; | |
2056 | if (snap->ds_dir != origin_head->ds_dir) | |
2057 | error = SET_ERROR(EINVAL); | |
2058 | if (error == 0) { | |
2059 | error = dsl_destroy_snapshot_check_impl( | |
2060 | snap, B_FALSE); | |
2061 | } | |
2062 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; | |
2063 | dsl_dataset_rele(snap, FTAG); | |
2064 | if (error != 0) | |
2065 | break; | |
2066 | } | |
2067 | if (error != 0) { | |
2068 | dsl_dataset_rele(origin_head, FTAG); | |
2069 | return (error); | |
2070 | } | |
2071 | } | |
2072 | error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, | |
2073 | origin_head, drc->drc_force, drc->drc_owner, tx); | |
2074 | if (error != 0) { | |
2075 | dsl_dataset_rele(origin_head, FTAG); | |
2076 | return (error); | |
2077 | } | |
2078 | error = dsl_dataset_snapshot_check_impl(origin_head, | |
2079 | drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); | |
2080 | dsl_dataset_rele(origin_head, FTAG); | |
2081 | if (error != 0) | |
2082 | return (error); | |
2083 | ||
2084 | error = dsl_destroy_head_check_impl(drc->drc_ds, 1); | |
2085 | } else { | |
2086 | error = dsl_dataset_snapshot_check_impl(drc->drc_ds, | |
2087 | drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); | |
2088 | } | |
2089 | return (error); | |
2090 | } | |
2091 | ||
2092 | static void | |
2093 | dmu_recv_end_sync(void *arg, dmu_tx_t *tx) | |
2094 | { | |
2095 | dmu_recv_cookie_t *drc = arg; | |
2096 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
2097 | ||
2098 | spa_history_log_internal_ds(drc->drc_ds, "finish receiving", | |
2099 | tx, "snap=%s", drc->drc_tosnap); | |
2100 | ||
2101 | if (!drc->drc_newfs) { | |
2102 | dsl_dataset_t *origin_head; | |
2103 | ||
2104 | VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, | |
2105 | &origin_head)); | |
2106 | ||
2107 | if (drc->drc_force) { | |
2108 | /* | |
2109 | * Destroy any snapshots of drc_tofs (origin_head) | |
2110 | * after the origin (the snap before drc_ds). | |
2111 | */ | |
2112 | uint64_t obj; | |
2113 | ||
2114 | obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; | |
2115 | while (obj != | |
2116 | dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { | |
2117 | dsl_dataset_t *snap; | |
2118 | VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, | |
2119 | &snap)); | |
2120 | ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); | |
2121 | obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; | |
2122 | dsl_destroy_snapshot_sync_impl(snap, | |
2123 | B_FALSE, tx); | |
2124 | dsl_dataset_rele(snap, FTAG); | |
2125 | } | |
2126 | } | |
2127 | VERIFY3P(drc->drc_ds->ds_prev, ==, | |
2128 | origin_head->ds_prev); | |
2129 | ||
2130 | dsl_dataset_clone_swap_sync_impl(drc->drc_ds, | |
2131 | origin_head, tx); | |
2132 | dsl_dataset_snapshot_sync_impl(origin_head, | |
2133 | drc->drc_tosnap, tx); | |
2134 | ||
2135 | /* set snapshot's creation time and guid */ | |
2136 | dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); | |
2137 | dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = | |
2138 | drc->drc_drrb->drr_creation_time; | |
2139 | dsl_dataset_phys(origin_head->ds_prev)->ds_guid = | |
2140 | drc->drc_drrb->drr_toguid; | |
2141 | dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= | |
2142 | ~DS_FLAG_INCONSISTENT; | |
2143 | ||
2144 | dmu_buf_will_dirty(origin_head->ds_dbuf, tx); | |
2145 | dsl_dataset_phys(origin_head)->ds_flags &= | |
2146 | ~DS_FLAG_INCONSISTENT; | |
2147 | ||
2148 | dsl_dataset_rele(origin_head, FTAG); | |
2149 | dsl_destroy_head_sync_impl(drc->drc_ds, tx); | |
2150 | ||
2151 | if (drc->drc_owner != NULL) | |
2152 | VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); | |
2153 | } else { | |
2154 | dsl_dataset_t *ds = drc->drc_ds; | |
2155 | ||
2156 | dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); | |
2157 | ||
2158 | /* set snapshot's creation time and guid */ | |
2159 | dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
2160 | dsl_dataset_phys(ds->ds_prev)->ds_creation_time = | |
2161 | drc->drc_drrb->drr_creation_time; | |
2162 | dsl_dataset_phys(ds->ds_prev)->ds_guid = | |
2163 | drc->drc_drrb->drr_toguid; | |
2164 | dsl_dataset_phys(ds->ds_prev)->ds_flags &= | |
2165 | ~DS_FLAG_INCONSISTENT; | |
2166 | ||
2167 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
2168 | dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; | |
2169 | } | |
2170 | drc->drc_newsnapobj = dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; | |
2171 | zvol_create_minors(dp->dp_spa, drc->drc_tofs, B_TRUE); | |
2172 | /* | |
2173 | * Release the hold from dmu_recv_begin. This must be done before | |
2174 | * we return to open context, so that when we free the dataset's dnode, | |
2175 | * we can evict its bonus buffer. | |
2176 | */ | |
2177 | dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); | |
2178 | drc->drc_ds = NULL; | |
2179 | } | |
2180 | ||
2181 | static int | |
2182 | add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj) | |
2183 | { | |
2184 | dsl_pool_t *dp; | |
2185 | dsl_dataset_t *snapds; | |
2186 | guid_map_entry_t *gmep; | |
2187 | int err; | |
2188 | ||
2189 | ASSERT(guid_map != NULL); | |
2190 | ||
2191 | err = dsl_pool_hold(name, FTAG, &dp); | |
2192 | if (err != 0) | |
2193 | return (err); | |
2194 | gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP); | |
2195 | err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds); | |
2196 | if (err == 0) { | |
2197 | gmep->guid = dsl_dataset_phys(snapds)->ds_guid; | |
2198 | gmep->gme_ds = snapds; | |
2199 | avl_add(guid_map, gmep); | |
2200 | dsl_dataset_long_hold(snapds, gmep); | |
2201 | } else { | |
2202 | kmem_free(gmep, sizeof (*gmep)); | |
2203 | } | |
2204 | ||
2205 | dsl_pool_rele(dp, FTAG); | |
2206 | return (err); | |
2207 | } | |
2208 | ||
2209 | static int dmu_recv_end_modified_blocks = 3; | |
2210 | ||
2211 | static int | |
2212 | dmu_recv_existing_end(dmu_recv_cookie_t *drc) | |
2213 | { | |
2214 | int error; | |
2215 | ||
2216 | #ifdef _KERNEL | |
2217 | char *name; | |
2218 | ||
2219 | /* | |
2220 | * We will be destroying the ds; make sure its origin is unmounted if | |
2221 | * necessary. | |
2222 | */ | |
2223 | name = kmem_alloc(MAXNAMELEN, KM_SLEEP); | |
2224 | dsl_dataset_name(drc->drc_ds, name); | |
2225 | zfs_destroy_unmount_origin(name); | |
2226 | kmem_free(name, MAXNAMELEN); | |
2227 | #endif | |
2228 | ||
2229 | error = dsl_sync_task(drc->drc_tofs, | |
2230 | dmu_recv_end_check, dmu_recv_end_sync, drc, | |
2231 | dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL); | |
2232 | ||
2233 | if (error != 0) | |
2234 | dmu_recv_cleanup_ds(drc); | |
2235 | return (error); | |
2236 | } | |
2237 | ||
2238 | static int | |
2239 | dmu_recv_new_end(dmu_recv_cookie_t *drc) | |
2240 | { | |
2241 | int error; | |
2242 | ||
2243 | error = dsl_sync_task(drc->drc_tofs, | |
2244 | dmu_recv_end_check, dmu_recv_end_sync, drc, | |
2245 | dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL); | |
2246 | ||
2247 | if (error != 0) { | |
2248 | dmu_recv_cleanup_ds(drc); | |
2249 | } else if (drc->drc_guid_to_ds_map != NULL) { | |
2250 | (void) add_ds_to_guidmap(drc->drc_tofs, | |
2251 | drc->drc_guid_to_ds_map, | |
2252 | drc->drc_newsnapobj); | |
2253 | } | |
2254 | return (error); | |
2255 | } | |
2256 | ||
2257 | int | |
2258 | dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) | |
2259 | { | |
2260 | drc->drc_owner = owner; | |
2261 | ||
2262 | if (drc->drc_newfs) | |
2263 | return (dmu_recv_new_end(drc)); | |
2264 | else | |
2265 | return (dmu_recv_existing_end(drc)); | |
2266 | } | |
2267 | ||
2268 | /* | |
2269 | * Return TRUE if this objset is currently being received into. | |
2270 | */ | |
2271 | boolean_t | |
2272 | dmu_objset_is_receiving(objset_t *os) | |
2273 | { | |
2274 | return (os->os_dsl_dataset != NULL && | |
2275 | os->os_dsl_dataset->ds_owner == dmu_recv_tag); | |
2276 | } | |
2277 | ||
2278 | #if defined(_KERNEL) | |
2279 | module_param(zfs_send_corrupt_data, int, 0644); | |
2280 | MODULE_PARM_DESC(zfs_send_corrupt_data, "Allow sending corrupt data"); | |
2281 | #endif |