]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/dmu_send.c
projects / mirror_zfs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Illumos 3693 - restore_object uses at least two transactions to restore an object
[mirror_zfs.git] / module / zfs / dmu_send.c
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) 2012, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
27 */
28
29 #include <sys/dmu.h>
30 #include <sys/dmu_impl.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/dbuf.h>
33 #include <sys/dnode.h>
34 #include <sys/zfs_context.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/dsl_dir.h>
39 #include <sys/dsl_prop.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_synctask.h>
42 #include <sys/spa_impl.h>
43 #include <sys/zfs_ioctl.h>
44 #include <sys/zap.h>
45 #include <sys/zio_checksum.h>
46 #include <sys/zfs_znode.h>
47 #include <zfs_fletcher.h>
48 #include <sys/avl.h>
49 #include <sys/ddt.h>
50 #include <sys/zfs_onexit.h>
51 #include <sys/dmu_send.h>
52 #include <sys/dsl_destroy.h>
53 #include <sys/blkptr.h>
54 #include <sys/dsl_bookmark.h>
55 #include <sys/zfeature.h>
56
57 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
58 int zfs_send_corrupt_data = B_FALSE;
59
60 static char *dmu_recv_tag = "dmu_recv_tag";
61 static const char *recv_clone_name = "%recv";
62
63 typedef struct dump_bytes_io {
64 dmu_sendarg_t *dbi_dsp;
65 void *dbi_buf;
66 int dbi_len;
67 } dump_bytes_io_t;
68
69 static void
70 dump_bytes_strategy(void *arg)
71 {
72 dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg;
73 dmu_sendarg_t *dsp = dbi->dbi_dsp;
74 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
75 ssize_t resid; /* have to get resid to get detailed errno */
76 ASSERT0(dbi->dbi_len % 8);
77
78 fletcher_4_incremental_native(dbi->dbi_buf, dbi->dbi_len, &dsp->dsa_zc);
79 dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
80 (caddr_t)dbi->dbi_buf, dbi->dbi_len,
81 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
82
83 mutex_enter(&ds->ds_sendstream_lock);
84 *dsp->dsa_off += dbi->dbi_len;
85 mutex_exit(&ds->ds_sendstream_lock);
86 }
87
88 static int
89 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
90 {
91 dump_bytes_io_t dbi;
92
93 dbi.dbi_dsp = dsp;
94 dbi.dbi_buf = buf;
95 dbi.dbi_len = len;
96
97 /*
98 * The vn_rdwr() call is performed in a taskq to ensure that there is
99 * always enough stack space to write safely to the target filesystem.
100 * The ZIO_TYPE_FREE threads are used because there can be a lot of
101 * them and they are used in vdev_file.c for a similar purpose.
102 */
103 spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE,
104 ZIO_TASKQ_ISSUE, dump_bytes_strategy, &dbi, TQ_SLEEP);
105
106 return (dsp->dsa_err);
107 }
108
109 static int
110 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
111 uint64_t length)
112 {
113 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
114
115 /*
116 * When we receive a free record, dbuf_free_range() assumes
117 * that the receiving system doesn't have any dbufs in the range
118 * being freed. This is always true because there is a one-record
119 * constraint: we only send one WRITE record for any given
120 * object+offset. We know that the one-record constraint is
121 * true because we always send data in increasing order by
122 * object,offset.
123 *
124 * If the increasing-order constraint ever changes, we should find
125 * another way to assert that the one-record constraint is still
126 * satisfied.
127 */
128 ASSERT(object > dsp->dsa_last_data_object ||
129 (object == dsp->dsa_last_data_object &&
130 offset > dsp->dsa_last_data_offset));
131
132 /*
133 * If we are doing a non-incremental send, then there can't
134 * be any data in the dataset we're receiving into. Therefore
135 * a free record would simply be a no-op. Save space by not
136 * sending it to begin with.
137 */
138 if (!dsp->dsa_incremental)
139 return (0);
140
141 if (length != -1ULL && offset + length < offset)
142 length = -1ULL;
143
144 /*
145 * If there is a pending op, but it's not PENDING_FREE, push it out,
146 * since free block aggregation can only be done for blocks of the
147 * same type (i.e., DRR_FREE records can only be aggregated with
148 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
149 * aggregated with other DRR_FREEOBJECTS records.
150 */
151 if (dsp->dsa_pending_op != PENDING_NONE &&
152 dsp->dsa_pending_op != PENDING_FREE) {
153 if (dump_bytes(dsp, dsp->dsa_drr,
154 sizeof (dmu_replay_record_t)) != 0)
155 return (SET_ERROR(EINTR));
156 dsp->dsa_pending_op = PENDING_NONE;
157 }
158
159 if (dsp->dsa_pending_op == PENDING_FREE) {
160 /*
161 * There should never be a PENDING_FREE if length is -1
162 * (because dump_dnode is the only place where this
163 * function is called with a -1, and only after flushing
164 * any pending record).
165 */
166 ASSERT(length != -1ULL);
167 /*
168 * Check to see whether this free block can be aggregated
169 * with pending one.
170 */
171 if (drrf->drr_object == object && drrf->drr_offset +
172 drrf->drr_length == offset) {
173 drrf->drr_length += length;
174 return (0);
175 } else {
176 /* not a continuation. Push out pending record */
177 if (dump_bytes(dsp, dsp->dsa_drr,
178 sizeof (dmu_replay_record_t)) != 0)
179 return (SET_ERROR(EINTR));
180 dsp->dsa_pending_op = PENDING_NONE;
181 }
182 }
183 /* create a FREE record and make it pending */
184 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
185 dsp->dsa_drr->drr_type = DRR_FREE;
186 drrf->drr_object = object;
187 drrf->drr_offset = offset;
188 drrf->drr_length = length;
189 drrf->drr_toguid = dsp->dsa_toguid;
190 if (length == -1ULL) {
191 if (dump_bytes(dsp, dsp->dsa_drr,
192 sizeof (dmu_replay_record_t)) != 0)
193 return (SET_ERROR(EINTR));
194 } else {
195 dsp->dsa_pending_op = PENDING_FREE;
196 }
197
198 return (0);
199 }
200
201 static int
202 dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type,
203 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
204 {
205 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
206
207 /*
208 * We send data in increasing object, offset order.
209 * See comment in dump_free() for details.
210 */
211 ASSERT(object > dsp->dsa_last_data_object ||
212 (object == dsp->dsa_last_data_object &&
213 offset > dsp->dsa_last_data_offset));
214 dsp->dsa_last_data_object = object;
215 dsp->dsa_last_data_offset = offset + blksz - 1;
216
217 /*
218 * If there is any kind of pending aggregation (currently either
219 * a grouping of free objects or free blocks), push it out to
220 * the stream, since aggregation can't be done across operations
221 * of different types.
222 */
223 if (dsp->dsa_pending_op != PENDING_NONE) {
224 if (dump_bytes(dsp, dsp->dsa_drr,
225 sizeof (dmu_replay_record_t)) != 0)
226 return (SET_ERROR(EINTR));
227 dsp->dsa_pending_op = PENDING_NONE;
228 }
229 /* write a DATA record */
230 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
231 dsp->dsa_drr->drr_type = DRR_WRITE;
232 drrw->drr_object = object;
233 drrw->drr_type = type;
234 drrw->drr_offset = offset;
235 drrw->drr_length = blksz;
236 drrw->drr_toguid = dsp->dsa_toguid;
237 if (BP_IS_EMBEDDED(bp)) {
238 /*
239 * There's no pre-computed checksum of embedded BP's, so
240 * (like fletcher4-checkummed blocks) userland will have
241 * to compute a dedup-capable checksum itself.
242 */
243 drrw->drr_checksumtype = ZIO_CHECKSUM_OFF;
244 } else {
245 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
246 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
247 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
248 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
249 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
250 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
251 drrw->drr_key.ddk_cksum = bp->blk_cksum;
252 }
253
254 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
255 return (SET_ERROR(EINTR));
256 if (dump_bytes(dsp, data, blksz) != 0)
257 return (SET_ERROR(EINTR));
258 return (0);
259 }
260
261 static int
262 dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
263 int blksz, const blkptr_t *bp)
264 {
265 char buf[BPE_PAYLOAD_SIZE];
266 struct drr_write_embedded *drrw =
267 &(dsp->dsa_drr->drr_u.drr_write_embedded);
268
269 if (dsp->dsa_pending_op != PENDING_NONE) {
270 if (dump_bytes(dsp, dsp->dsa_drr,
271 sizeof (dmu_replay_record_t)) != 0)
272 return (EINTR);
273 dsp->dsa_pending_op = PENDING_NONE;
274 }
275
276 ASSERT(BP_IS_EMBEDDED(bp));
277
278 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
279 dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED;
280 drrw->drr_object = object;
281 drrw->drr_offset = offset;
282 drrw->drr_length = blksz;
283 drrw->drr_toguid = dsp->dsa_toguid;
284 drrw->drr_compression = BP_GET_COMPRESS(bp);
285 drrw->drr_etype = BPE_GET_ETYPE(bp);
286 drrw->drr_lsize = BPE_GET_LSIZE(bp);
287 drrw->drr_psize = BPE_GET_PSIZE(bp);
288
289 decode_embedded_bp_compressed(bp, buf);
290
291 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
292 return (EINTR);
293 if (dump_bytes(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0)
294 return (EINTR);
295 return (0);
296 }
297
298 static int
299 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
300 {
301 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
302
303 if (dsp->dsa_pending_op != PENDING_NONE) {
304 if (dump_bytes(dsp, dsp->dsa_drr,
305 sizeof (dmu_replay_record_t)) != 0)
306 return (SET_ERROR(EINTR));
307 dsp->dsa_pending_op = PENDING_NONE;
308 }
309
310 /* write a SPILL record */
311 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
312 dsp->dsa_drr->drr_type = DRR_SPILL;
313 drrs->drr_object = object;
314 drrs->drr_length = blksz;
315 drrs->drr_toguid = dsp->dsa_toguid;
316
317 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
318 return (SET_ERROR(EINTR));
319 if (dump_bytes(dsp, data, blksz))
320 return (SET_ERROR(EINTR));
321 return (0);
322 }
323
324 static int
325 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
326 {
327 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
328
329 /* See comment in dump_free(). */
330 if (!dsp->dsa_incremental)
331 return (0);
332
333 /*
334 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
335 * push it out, since free block aggregation can only be done for
336 * blocks of the same type (i.e., DRR_FREE records can only be
337 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
338 * can only be aggregated with other DRR_FREEOBJECTS records.
339 */
340 if (dsp->dsa_pending_op != PENDING_NONE &&
341 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
342 if (dump_bytes(dsp, dsp->dsa_drr,
343 sizeof (dmu_replay_record_t)) != 0)
344 return (SET_ERROR(EINTR));
345 dsp->dsa_pending_op = PENDING_NONE;
346 }
347 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
348 /*
349 * See whether this free object array can be aggregated
350 * with pending one
351 */
352 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
353 drrfo->drr_numobjs += numobjs;
354 return (0);
355 } else {
356 /* can't be aggregated. Push out pending record */
357 if (dump_bytes(dsp, dsp->dsa_drr,
358 sizeof (dmu_replay_record_t)) != 0)
359 return (SET_ERROR(EINTR));
360 dsp->dsa_pending_op = PENDING_NONE;
361 }
362 }
363
364 /* write a FREEOBJECTS record */
365 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
366 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
367 drrfo->drr_firstobj = firstobj;
368 drrfo->drr_numobjs = numobjs;
369 drrfo->drr_toguid = dsp->dsa_toguid;
370
371 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
372
373 return (0);
374 }
375
376 static int
377 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
378 {
379 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
380
381 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
382 return (dump_freeobjects(dsp, object, 1));
383
384 if (dsp->dsa_pending_op != PENDING_NONE) {
385 if (dump_bytes(dsp, dsp->dsa_drr,
386 sizeof (dmu_replay_record_t)) != 0)
387 return (SET_ERROR(EINTR));
388 dsp->dsa_pending_op = PENDING_NONE;
389 }
390
391 /* write an OBJECT record */
392 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
393 dsp->dsa_drr->drr_type = DRR_OBJECT;
394 drro->drr_object = object;
395 drro->drr_type = dnp->dn_type;
396 drro->drr_bonustype = dnp->dn_bonustype;
397 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
398 drro->drr_bonuslen = dnp->dn_bonuslen;
399 drro->drr_checksumtype = dnp->dn_checksum;
400 drro->drr_compress = dnp->dn_compress;
401 drro->drr_toguid = dsp->dsa_toguid;
402
403 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
404 return (SET_ERROR(EINTR));
405
406 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
407 return (SET_ERROR(EINTR));
408
409 /* Free anything past the end of the file. */
410 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
411 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0)
412 return (SET_ERROR(EINTR));
413 if (dsp->dsa_err != 0)
414 return (SET_ERROR(EINTR));
415 return (0);
416 }
417
418 static boolean_t
419 backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp)
420 {
421 if (!BP_IS_EMBEDDED(bp))
422 return (B_FALSE);
423
424 /*
425 * Compression function must be legacy, or explicitly enabled.
426 */
427 if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS &&
428 !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4)))
429 return (B_FALSE);
430
431 /*
432 * Embed type must be explicitly enabled.
433 */
434 switch (BPE_GET_ETYPE(bp)) {
435 case BP_EMBEDDED_TYPE_DATA:
436 if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)
437 return (B_TRUE);
438 break;
439 default:
440 return (B_FALSE);
441 }
442 return (B_FALSE);
443 }
444
445 #define BP_SPAN(dnp, level) \
446 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
447 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
448
449 /* ARGSUSED */
450 static int
451 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
452 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
453 {
454 dmu_sendarg_t *dsp = arg;
455 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
456 int err = 0;
457
458 if (issig(JUSTLOOKING) && issig(FORREAL))
459 return (SET_ERROR(EINTR));
460
461 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
462 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
463 return (0);
464 } else if (zb->zb_level == ZB_ZIL_LEVEL) {
465 /*
466 * If we are sending a non-snapshot (which is allowed on
467 * read-only pools), it may have a ZIL, which must be ignored.
468 */
469 return (0);
470 } else if (BP_IS_HOLE(bp) &&
471 zb->zb_object == DMU_META_DNODE_OBJECT) {
472 uint64_t span = BP_SPAN(dnp, zb->zb_level);
473 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
474 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
475 } else if (BP_IS_HOLE(bp)) {
476 uint64_t span = BP_SPAN(dnp, zb->zb_level);
477 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
478 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
479 return (0);
480 } else if (type == DMU_OT_DNODE) {
481 dnode_phys_t *blk;
482 int i;
483 int blksz = BP_GET_LSIZE(bp);
484 uint32_t aflags = ARC_WAIT;
485 arc_buf_t *abuf;
486
487 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
488 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
489 &aflags, zb) != 0)
490 return (SET_ERROR(EIO));
491
492 blk = abuf->b_data;
493 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
494 uint64_t dnobj = (zb->zb_blkid <<
495 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
496 err = dump_dnode(dsp, dnobj, blk+i);
497 if (err != 0)
498 break;
499 }
500 (void) arc_buf_remove_ref(abuf, &abuf);
501 } else if (type == DMU_OT_SA) {
502 uint32_t aflags = ARC_WAIT;
503 arc_buf_t *abuf;
504 int blksz = BP_GET_LSIZE(bp);
505
506 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
507 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
508 &aflags, zb) != 0)
509 return (SET_ERROR(EIO));
510
511 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
512 (void) arc_buf_remove_ref(abuf, &abuf);
513 } else if (backup_do_embed(dsp, bp)) {
514 /* it's an embedded level-0 block of a regular object */
515 int blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
516 err = dump_write_embedded(dsp, zb->zb_object,
517 zb->zb_blkid * blksz, blksz, bp);
518 } else { /* it's a level-0 block of a regular object */
519 uint32_t aflags = ARC_WAIT;
520 arc_buf_t *abuf;
521 int blksz = BP_GET_LSIZE(bp);
522
523 ASSERT3U(blksz, ==, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
524 ASSERT0(zb->zb_level);
525 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
526 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
527 &aflags, zb) != 0) {
528 if (zfs_send_corrupt_data) {
529 uint64_t *ptr;
530 /* Send a block filled with 0x"zfs badd bloc" */
531 abuf = arc_buf_alloc(spa, blksz, &abuf,
532 ARC_BUFC_DATA);
533 for (ptr = abuf->b_data;
534 (char *)ptr < (char *)abuf->b_data + blksz;
535 ptr++)
536 *ptr = 0x2f5baddb10cULL;
537 } else {
538 return (SET_ERROR(EIO));
539 }
540 }
541
542 err = dump_write(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
543 blksz, bp, abuf->b_data);
544 (void) arc_buf_remove_ref(abuf, &abuf);
545 }
546
547 ASSERT(err == 0 || err == EINTR);
548 return (err);
549 }
550
551 /*
552 * Releases dp using the specified tag.
553 */
554 static int
555 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
556 zfs_bookmark_phys_t *fromzb, boolean_t is_clone, boolean_t embedok,
557 int outfd, vnode_t *vp, offset_t *off)
558 {
559 objset_t *os;
560 dmu_replay_record_t *drr;
561 dmu_sendarg_t *dsp;
562 int err;
563 uint64_t fromtxg = 0;
564 uint64_t featureflags = 0;
565
566 err = dmu_objset_from_ds(ds, &os);
567 if (err != 0) {
568 dsl_pool_rele(dp, tag);
569 return (err);
570 }
571
572 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
573 drr->drr_type = DRR_BEGIN;
574 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
575 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
576 DMU_SUBSTREAM);
577
578 #ifdef _KERNEL
579 if (dmu_objset_type(os) == DMU_OST_ZFS) {
580 uint64_t version;
581 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
582 kmem_free(drr, sizeof (dmu_replay_record_t));
583 dsl_pool_rele(dp, tag);
584 return (SET_ERROR(EINVAL));
585 }
586 if (version >= ZPL_VERSION_SA) {
587 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
588 }
589 }
590 #endif
591
592 if (embedok &&
593 spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) {
594 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA;
595 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
596 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA_LZ4;
597 } else {
598 embedok = B_FALSE;
599 }
600
601 DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo,
602 featureflags);
603
604 drr->drr_u.drr_begin.drr_creation_time =
605 ds->ds_phys->ds_creation_time;
606 drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
607 if (is_clone)
608 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
609 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
610 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
611 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
612
613 if (fromzb != NULL) {
614 drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid;
615 fromtxg = fromzb->zbm_creation_txg;
616 }
617 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
618 if (!dsl_dataset_is_snapshot(ds)) {
619 (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--",
620 sizeof (drr->drr_u.drr_begin.drr_toname));
621 }
622
623 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
624
625 dsp->dsa_drr = drr;
626 dsp->dsa_vp = vp;
627 dsp->dsa_outfd = outfd;
628 dsp->dsa_proc = curproc;
629 dsp->dsa_os = os;
630 dsp->dsa_off = off;
631 dsp->dsa_toguid = ds->ds_phys->ds_guid;
632 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
633 dsp->dsa_pending_op = PENDING_NONE;
634 dsp->dsa_incremental = (fromzb != NULL);
635 dsp->dsa_featureflags = featureflags;
636
637 mutex_enter(&ds->ds_sendstream_lock);
638 list_insert_head(&ds->ds_sendstreams, dsp);
639 mutex_exit(&ds->ds_sendstream_lock);
640
641 dsl_dataset_long_hold(ds, FTAG);
642 dsl_pool_rele(dp, tag);
643
644 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
645 err = dsp->dsa_err;
646 goto out;
647 }
648
649 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
650 backup_cb, dsp);
651
652 if (dsp->dsa_pending_op != PENDING_NONE)
653 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
654 err = SET_ERROR(EINTR);
655
656 if (err != 0) {
657 if (err == EINTR && dsp->dsa_err != 0)
658 err = dsp->dsa_err;
659 goto out;
660 }
661
662 bzero(drr, sizeof (dmu_replay_record_t));
663 drr->drr_type = DRR_END;
664 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
665 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
666
667 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
668 err = dsp->dsa_err;
669 goto out;
670 }
671
672 out:
673 mutex_enter(&ds->ds_sendstream_lock);
674 list_remove(&ds->ds_sendstreams, dsp);
675 mutex_exit(&ds->ds_sendstream_lock);
676
677 kmem_free(drr, sizeof (dmu_replay_record_t));
678 kmem_free(dsp, sizeof (dmu_sendarg_t));
679
680 dsl_dataset_long_rele(ds, FTAG);
681
682 return (err);
683 }
684
685 int
686 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
687 boolean_t embedok, int outfd, vnode_t *vp, offset_t *off)
688 {
689 dsl_pool_t *dp;
690 dsl_dataset_t *ds;
691 dsl_dataset_t *fromds = NULL;
692 int err;
693
694 err = dsl_pool_hold(pool, FTAG, &dp);
695 if (err != 0)
696 return (err);
697
698 err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
699 if (err != 0) {
700 dsl_pool_rele(dp, FTAG);
701 return (err);
702 }
703
704 if (fromsnap != 0) {
705 zfs_bookmark_phys_t zb;
706 boolean_t is_clone;
707
708 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
709 if (err != 0) {
710 dsl_dataset_rele(ds, FTAG);
711 dsl_pool_rele(dp, FTAG);
712 return (err);
713 }
714 if (!dsl_dataset_is_before(ds, fromds, 0))
715 err = SET_ERROR(EXDEV);
716 zb.zbm_creation_time = fromds->ds_phys->ds_creation_time;
717 zb.zbm_creation_txg = fromds->ds_phys->ds_creation_txg;
718 zb.zbm_guid = fromds->ds_phys->ds_guid;
719 is_clone = (fromds->ds_dir != ds->ds_dir);
720 dsl_dataset_rele(fromds, FTAG);
721 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, embedok,
722 outfd, vp, off);
723 } else {
724 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, embedok,
725 outfd, vp, off);
726 }
727 dsl_dataset_rele(ds, FTAG);
728 return (err);
729 }
730
731 int
732 dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok,
733 int outfd, vnode_t *vp, offset_t *off)
734 {
735 dsl_pool_t *dp;
736 dsl_dataset_t *ds;
737 int err;
738 boolean_t owned = B_FALSE;
739
740 if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL)
741 return (SET_ERROR(EINVAL));
742
743 err = dsl_pool_hold(tosnap, FTAG, &dp);
744 if (err != 0)
745 return (err);
746
747 if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) {
748 /*
749 * We are sending a filesystem or volume. Ensure
750 * that it doesn't change by owning the dataset.
751 */
752 err = dsl_dataset_own(dp, tosnap, FTAG, &ds);
753 owned = B_TRUE;
754 } else {
755 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
756 }
757 if (err != 0) {
758 dsl_pool_rele(dp, FTAG);
759 return (err);
760 }
761
762 if (fromsnap != NULL) {
763 zfs_bookmark_phys_t zb;
764 boolean_t is_clone = B_FALSE;
765 int fsnamelen = strchr(tosnap, '@') - tosnap;
766
767 /*
768 * If the fromsnap is in a different filesystem, then
769 * mark the send stream as a clone.
770 */
771 if (strncmp(tosnap, fromsnap, fsnamelen) != 0 ||
772 (fromsnap[fsnamelen] != '@' &&
773 fromsnap[fsnamelen] != '#')) {
774 is_clone = B_TRUE;
775 }
776
777 if (strchr(fromsnap, '@')) {
778 dsl_dataset_t *fromds;
779 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
780 if (err == 0) {
781 if (!dsl_dataset_is_before(ds, fromds, 0))
782 err = SET_ERROR(EXDEV);
783 zb.zbm_creation_time =
784 fromds->ds_phys->ds_creation_time;
785 zb.zbm_creation_txg =
786 fromds->ds_phys->ds_creation_txg;
787 zb.zbm_guid = fromds->ds_phys->ds_guid;
788 is_clone = (ds->ds_dir != fromds->ds_dir);
789 dsl_dataset_rele(fromds, FTAG);
790 }
791 } else {
792 err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb);
793 }
794 if (err != 0) {
795 dsl_dataset_rele(ds, FTAG);
796 dsl_pool_rele(dp, FTAG);
797 return (err);
798 }
799 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, embedok,
800 outfd, vp, off);
801 } else {
802 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, embedok,
803 outfd, vp, off);
804 }
805 if (owned)
806 dsl_dataset_disown(ds, FTAG);
807 else
808 dsl_dataset_rele(ds, FTAG);
809 return (err);
810 }
811
812 int
813 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
814 {
815 int err;
816 uint64_t size, recordsize;
817 ASSERTV(dsl_pool_t *dp = ds->ds_dir->dd_pool);
818
819 ASSERT(dsl_pool_config_held(dp));
820
821 /* tosnap must be a snapshot */
822 if (!dsl_dataset_is_snapshot(ds))
823 return (SET_ERROR(EINVAL));
824
825 /*
826 * fromsnap must be an earlier snapshot from the same fs as tosnap,
827 * or the origin's fs.
828 */
829 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
830 return (SET_ERROR(EXDEV));
831
832 /* Get uncompressed size estimate of changed data. */
833 if (fromds == NULL) {
834 size = ds->ds_phys->ds_uncompressed_bytes;
835 } else {
836 uint64_t used, comp;
837 err = dsl_dataset_space_written(fromds, ds,
838 &used, &comp, &size);
839 if (err != 0)
840 return (err);
841 }
842
843 /*
844 * Assume that space (both on-disk and in-stream) is dominated by
845 * data. We will adjust for indirect blocks and the copies property,
846 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
847 */
848
849 /*
850 * Subtract out approximate space used by indirect blocks.
851 * Assume most space is used by data blocks (non-indirect, non-dnode).
852 * Assume all blocks are recordsize. Assume ditto blocks and
853 * internal fragmentation counter out compression.
854 *
855 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
856 * block, which we observe in practice.
857 */
858 err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
859 if (err != 0)
860 return (err);
861 size -= size / recordsize * sizeof (blkptr_t);
862
863 /* Add in the space for the record associated with each block. */
864 size += size / recordsize * sizeof (dmu_replay_record_t);
865
866 *sizep = size;
867
868 return (0);
869 }
870
871 typedef struct dmu_recv_begin_arg {
872 const char *drba_origin;
873 dmu_recv_cookie_t *drba_cookie;
874 cred_t *drba_cred;
875 uint64_t drba_snapobj;
876 } dmu_recv_begin_arg_t;
877
878 static int
879 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
880 uint64_t fromguid)
881 {
882 uint64_t val;
883 int error;
884 dsl_pool_t *dp = ds->ds_dir->dd_pool;
885
886 /* temporary clone name must not exist */
887 error = zap_lookup(dp->dp_meta_objset,
888 ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name,
889 8, 1, &val);
890 if (error != ENOENT)
891 return (error == 0 ? EBUSY : error);
892
893 /* new snapshot name must not exist */
894 error = zap_lookup(dp->dp_meta_objset,
895 ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap,
896 8, 1, &val);
897 if (error != ENOENT)
898 return (error == 0 ? EEXIST : error);
899
900 if (fromguid != 0) {
901 dsl_dataset_t *snap;
902 uint64_t obj = ds->ds_phys->ds_prev_snap_obj;
903
904 /* Find snapshot in this dir that matches fromguid. */
905 while (obj != 0) {
906 error = dsl_dataset_hold_obj(dp, obj, FTAG,
907 &snap);
908 if (error != 0)
909 return (SET_ERROR(ENODEV));
910 if (snap->ds_dir != ds->ds_dir) {
911 dsl_dataset_rele(snap, FTAG);
912 return (SET_ERROR(ENODEV));
913 }
914 if (snap->ds_phys->ds_guid == fromguid)
915 break;
916 obj = snap->ds_phys->ds_prev_snap_obj;
917 dsl_dataset_rele(snap, FTAG);
918 }
919 if (obj == 0)
920 return (SET_ERROR(ENODEV));
921
922 if (drba->drba_cookie->drc_force) {
923 drba->drba_snapobj = obj;
924 } else {
925 /*
926 * If we are not forcing, there must be no
927 * changes since fromsnap.
928 */
929 if (dsl_dataset_modified_since_snap(ds, snap)) {
930 dsl_dataset_rele(snap, FTAG);
931 return (SET_ERROR(ETXTBSY));
932 }
933 drba->drba_snapobj = ds->ds_prev->ds_object;
934 }
935
936 dsl_dataset_rele(snap, FTAG);
937 } else {
938 /* if full, most recent snapshot must be $ORIGIN */
939 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
940 return (SET_ERROR(ENODEV));
941 drba->drba_snapobj = ds->ds_phys->ds_prev_snap_obj;
942 }
943
944 return (0);
945
946 }
947
948 static int
949 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
950 {
951 dmu_recv_begin_arg_t *drba = arg;
952 dsl_pool_t *dp = dmu_tx_pool(tx);
953 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
954 uint64_t fromguid = drrb->drr_fromguid;
955 int flags = drrb->drr_flags;
956 int error;
957 uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
958 dsl_dataset_t *ds;
959 const char *tofs = drba->drba_cookie->drc_tofs;
960
961 /* already checked */
962 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
963
964 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
965 DMU_COMPOUNDSTREAM ||
966 drrb->drr_type >= DMU_OST_NUMTYPES ||
967 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
968 return (SET_ERROR(EINVAL));
969
970 /* Verify pool version supports SA if SA_SPILL feature set */
971 if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
972 spa_version(dp->dp_spa) < SPA_VERSION_SA)
973 return (SET_ERROR(ENOTSUP));
974
975 /*
976 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
977 * record to a plan WRITE record, so the pool must have the
978 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
979 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
980 */
981 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) &&
982 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA))
983 return (SET_ERROR(ENOTSUP));
984 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) &&
985 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
986 return (SET_ERROR(ENOTSUP));
987
988 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
989 if (error == 0) {
990 /* target fs already exists; recv into temp clone */
991
992 /* Can't recv a clone into an existing fs */
993 if (flags & DRR_FLAG_CLONE) {
994 dsl_dataset_rele(ds, FTAG);
995 return (SET_ERROR(EINVAL));
996 }
997
998 error = recv_begin_check_existing_impl(drba, ds, fromguid);
999 dsl_dataset_rele(ds, FTAG);
1000 } else if (error == ENOENT) {
1001 /* target fs does not exist; must be a full backup or clone */
1002 char buf[MAXNAMELEN];
1003
1004 /*
1005 * If it's a non-clone incremental, we are missing the
1006 * target fs, so fail the recv.
1007 */
1008 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
1009 return (SET_ERROR(ENOENT));
1010
1011 /* Open the parent of tofs */
1012 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
1013 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
1014 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
1015 if (error != 0)
1016 return (error);
1017
1018 if (drba->drba_origin != NULL) {
1019 dsl_dataset_t *origin;
1020 error = dsl_dataset_hold(dp, drba->drba_origin,
1021 FTAG, &origin);
1022 if (error != 0) {
1023 dsl_dataset_rele(ds, FTAG);
1024 return (error);
1025 }
1026 if (!dsl_dataset_is_snapshot(origin)) {
1027 dsl_dataset_rele(origin, FTAG);
1028 dsl_dataset_rele(ds, FTAG);
1029 return (SET_ERROR(EINVAL));
1030 }
1031 if (origin->ds_phys->ds_guid != fromguid) {
1032 dsl_dataset_rele(origin, FTAG);
1033 dsl_dataset_rele(ds, FTAG);
1034 return (SET_ERROR(ENODEV));
1035 }
1036 dsl_dataset_rele(origin, FTAG);
1037 }
1038 dsl_dataset_rele(ds, FTAG);
1039 error = 0;
1040 }
1041 return (error);
1042 }
1043
1044 static void
1045 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
1046 {
1047 dmu_recv_begin_arg_t *drba = arg;
1048 dsl_pool_t *dp = dmu_tx_pool(tx);
1049 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1050 const char *tofs = drba->drba_cookie->drc_tofs;
1051 dsl_dataset_t *ds, *newds;
1052 uint64_t dsobj;
1053 int error;
1054 uint64_t crflags;
1055
1056 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
1057 DS_FLAG_CI_DATASET : 0;
1058
1059 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1060 if (error == 0) {
1061 /* create temporary clone */
1062 dsl_dataset_t *snap = NULL;
1063 if (drba->drba_snapobj != 0) {
1064 VERIFY0(dsl_dataset_hold_obj(dp,
1065 drba->drba_snapobj, FTAG, &snap));
1066 }
1067 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
1068 snap, crflags, drba->drba_cred, tx);
1069 dsl_dataset_rele(snap, FTAG);
1070 dsl_dataset_rele(ds, FTAG);
1071 } else {
1072 dsl_dir_t *dd;
1073 const char *tail;
1074 dsl_dataset_t *origin = NULL;
1075
1076 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
1077
1078 if (drba->drba_origin != NULL) {
1079 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
1080 FTAG, &origin));
1081 }
1082
1083 /* Create new dataset. */
1084 dsobj = dsl_dataset_create_sync(dd,
1085 strrchr(tofs, '/') + 1,
1086 origin, crflags, drba->drba_cred, tx);
1087 if (origin != NULL)
1088 dsl_dataset_rele(origin, FTAG);
1089 dsl_dir_rele(dd, FTAG);
1090 drba->drba_cookie->drc_newfs = B_TRUE;
1091 }
1092 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
1093
1094 dmu_buf_will_dirty(newds->ds_dbuf, tx);
1095 newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1096
1097 /*
1098 * If we actually created a non-clone, we need to create the
1099 * objset in our new dataset.
1100 */
1101 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
1102 (void) dmu_objset_create_impl(dp->dp_spa,
1103 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1104 }
1105
1106 drba->drba_cookie->drc_ds = newds;
1107
1108 spa_history_log_internal_ds(newds, "receive", tx, "");
1109 }
1110
1111 /*
1112 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1113 * succeeds; otherwise we will leak the holds on the datasets.
1114 */
1115 int
1116 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1117 boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1118 {
1119 dmu_recv_begin_arg_t drba = { 0 };
1120 dmu_replay_record_t *drr;
1121
1122 bzero(drc, sizeof (dmu_recv_cookie_t));
1123 drc->drc_drrb = drrb;
1124 drc->drc_tosnap = tosnap;
1125 drc->drc_tofs = tofs;
1126 drc->drc_force = force;
1127
1128 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1129 drc->drc_byteswap = B_TRUE;
1130 else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1131 return (SET_ERROR(EINVAL));
1132
1133 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1134 drr->drr_type = DRR_BEGIN;
1135 drr->drr_u.drr_begin = *drc->drc_drrb;
1136 if (drc->drc_byteswap) {
1137 fletcher_4_incremental_byteswap(drr,
1138 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1139 } else {
1140 fletcher_4_incremental_native(drr,
1141 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1142 }
1143 kmem_free(drr, sizeof (dmu_replay_record_t));
1144
1145 if (drc->drc_byteswap) {
1146 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1147 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1148 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1149 drrb->drr_type = BSWAP_32(drrb->drr_type);
1150 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1151 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1152 }
1153
1154 drba.drba_origin = origin;
1155 drba.drba_cookie = drc;
1156 drba.drba_cred = CRED();
1157
1158 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1159 &drba, 5));
1160 }
1161
1162 struct restorearg {
1163 int err;
1164 boolean_t byteswap;
1165 vnode_t *vp;
1166 char *buf;
1167 uint64_t voff;
1168 int bufsize; /* amount of memory allocated for buf */
1169 zio_cksum_t cksum;
1170 avl_tree_t *guid_to_ds_map;
1171 };
1172
1173 typedef struct guid_map_entry {
1174 uint64_t guid;
1175 dsl_dataset_t *gme_ds;
1176 avl_node_t avlnode;
1177 } guid_map_entry_t;
1178
1179 static int
1180 guid_compare(const void *arg1, const void *arg2)
1181 {
1182 const guid_map_entry_t *gmep1 = arg1;
1183 const guid_map_entry_t *gmep2 = arg2;
1184
1185 if (gmep1->guid < gmep2->guid)
1186 return (-1);
1187 else if (gmep1->guid > gmep2->guid)
1188 return (1);
1189 return (0);
1190 }
1191
1192 static void
1193 free_guid_map_onexit(void *arg)
1194 {
1195 avl_tree_t *ca = arg;
1196 void *cookie = NULL;
1197 guid_map_entry_t *gmep;
1198
1199 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1200 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1201 dsl_dataset_rele(gmep->gme_ds, gmep);
1202 kmem_free(gmep, sizeof (guid_map_entry_t));
1203 }
1204 avl_destroy(ca);
1205 kmem_free(ca, sizeof (avl_tree_t));
1206 }
1207
1208 static void *
1209 restore_read(struct restorearg *ra, int len)
1210 {
1211 void *rv;
1212 int done = 0;
1213
1214 /* some things will require 8-byte alignment, so everything must */
1215 ASSERT0(len % 8);
1216
1217 while (done < len) {
1218 ssize_t resid;
1219
1220 ra->err = vn_rdwr(UIO_READ, ra->vp,
1221 (caddr_t)ra->buf + done, len - done,
1222 ra->voff, UIO_SYSSPACE, FAPPEND,
1223 RLIM64_INFINITY, CRED(), &resid);
1224
1225 if (resid == len - done)
1226 ra->err = SET_ERROR(EINVAL);
1227 ra->voff += len - done - resid;
1228 done = len - resid;
1229 if (ra->err != 0)
1230 return (NULL);
1231 }
1232
1233 ASSERT3U(done, ==, len);
1234 rv = ra->buf;
1235 if (ra->byteswap)
1236 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1237 else
1238 fletcher_4_incremental_native(rv, len, &ra->cksum);
1239 return (rv);
1240 }
1241
1242 noinline static void
1243 backup_byteswap(dmu_replay_record_t *drr)
1244 {
1245 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1246 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1247 drr->drr_type = BSWAP_32(drr->drr_type);
1248 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1249 switch (drr->drr_type) {
1250 case DRR_BEGIN:
1251 DO64(drr_begin.drr_magic);
1252 DO64(drr_begin.drr_versioninfo);
1253 DO64(drr_begin.drr_creation_time);
1254 DO32(drr_begin.drr_type);
1255 DO32(drr_begin.drr_flags);
1256 DO64(drr_begin.drr_toguid);
1257 DO64(drr_begin.drr_fromguid);
1258 break;
1259 case DRR_OBJECT:
1260 DO64(drr_object.drr_object);
1261 DO32(drr_object.drr_type);
1262 DO32(drr_object.drr_bonustype);
1263 DO32(drr_object.drr_blksz);
1264 DO32(drr_object.drr_bonuslen);
1265 DO64(drr_object.drr_toguid);
1266 break;
1267 case DRR_FREEOBJECTS:
1268 DO64(drr_freeobjects.drr_firstobj);
1269 DO64(drr_freeobjects.drr_numobjs);
1270 DO64(drr_freeobjects.drr_toguid);
1271 break;
1272 case DRR_WRITE:
1273 DO64(drr_write.drr_object);
1274 DO32(drr_write.drr_type);
1275 DO64(drr_write.drr_offset);
1276 DO64(drr_write.drr_length);
1277 DO64(drr_write.drr_toguid);
1278 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1279 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1280 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1281 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1282 DO64(drr_write.drr_key.ddk_prop);
1283 break;
1284 case DRR_WRITE_BYREF:
1285 DO64(drr_write_byref.drr_object);
1286 DO64(drr_write_byref.drr_offset);
1287 DO64(drr_write_byref.drr_length);
1288 DO64(drr_write_byref.drr_toguid);
1289 DO64(drr_write_byref.drr_refguid);
1290 DO64(drr_write_byref.drr_refobject);
1291 DO64(drr_write_byref.drr_refoffset);
1292 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1293 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1294 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1295 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1296 DO64(drr_write_byref.drr_key.ddk_prop);
1297 break;
1298 case DRR_WRITE_EMBEDDED:
1299 DO64(drr_write_embedded.drr_object);
1300 DO64(drr_write_embedded.drr_offset);
1301 DO64(drr_write_embedded.drr_length);
1302 DO64(drr_write_embedded.drr_toguid);
1303 DO32(drr_write_embedded.drr_lsize);
1304 DO32(drr_write_embedded.drr_psize);
1305 break;
1306 case DRR_FREE:
1307 DO64(drr_free.drr_object);
1308 DO64(drr_free.drr_offset);
1309 DO64(drr_free.drr_length);
1310 DO64(drr_free.drr_toguid);
1311 break;
1312 case DRR_SPILL:
1313 DO64(drr_spill.drr_object);
1314 DO64(drr_spill.drr_length);
1315 DO64(drr_spill.drr_toguid);
1316 break;
1317 case DRR_END:
1318 DO64(drr_end.drr_checksum.zc_word[0]);
1319 DO64(drr_end.drr_checksum.zc_word[1]);
1320 DO64(drr_end.drr_checksum.zc_word[2]);
1321 DO64(drr_end.drr_checksum.zc_word[3]);
1322 DO64(drr_end.drr_toguid);
1323 break;
1324 default:
1325 break;
1326 }
1327 #undef DO64
1328 #undef DO32
1329 }
1330
1331 static inline uint8_t
1332 deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size)
1333 {
1334 if (bonus_type == DMU_OT_SA) {
1335 return (1);
1336 } else {
1337 return (1 +
1338 ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT));
1339 }
1340 }
1341
1342 noinline static int
1343 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1344 {
1345 dmu_object_info_t doi;
1346 dmu_tx_t *tx;
1347 void *data = NULL;
1348 uint64_t object;
1349 int err;
1350
1351 if (drro->drr_type == DMU_OT_NONE ||
1352 !DMU_OT_IS_VALID(drro->drr_type) ||
1353 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1354 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1355 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1356 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1357 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1358 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1359 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1360 return (SET_ERROR(EINVAL));
1361 }
1362
1363 err = dmu_object_info(os, drro->drr_object, &doi);
1364
1365 if (err != 0 && err != ENOENT)
1366 return (SET_ERROR(EINVAL));
1367 object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT;
1368
1369 if (drro->drr_bonuslen) {
1370 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1371 if (ra->err != 0)
1372 return (ra->err);
1373 }
1374
1375 /*
1376 * If we are losing blkptrs or changing the block size this must
1377 * be a new file instance. We must clear out the previous file
1378 * contents before we can change this type of metadata in the dnode.
1379 */
1380 if (err == 0) {
1381 int nblkptr;
1382
1383 nblkptr = deduce_nblkptr(drro->drr_bonustype,
1384 drro->drr_bonuslen);
1385
1386 if (drro->drr_blksz != doi.doi_data_block_size ||
1387 nblkptr < doi.doi_nblkptr) {
1388 err = dmu_free_long_range(os, drro->drr_object,
1389 0, DMU_OBJECT_END);
1390 if (err != 0)
1391 return (SET_ERROR(EINVAL));
1392 }
1393 }
1394
1395 tx = dmu_tx_create(os);
1396 dmu_tx_hold_bonus(tx, object);
1397 err = dmu_tx_assign(tx, TXG_WAIT);
1398 if (err != 0) {
1399 dmu_tx_abort(tx);
1400 return (err);
1401 }
1402
1403 if (object == DMU_NEW_OBJECT) {
1404 /* currently free, want to be allocated */
1405 err = dmu_object_claim(os, drro->drr_object,
1406 drro->drr_type, drro->drr_blksz,
1407 drro->drr_bonustype, drro->drr_bonuslen, tx);
1408 } else if (drro->drr_type != doi.doi_type ||
1409 drro->drr_blksz != doi.doi_data_block_size ||
1410 drro->drr_bonustype != doi.doi_bonus_type ||
1411 drro->drr_bonuslen != doi.doi_bonus_size) {
1412 /* currently allocated, but with different properties */
1413 err = dmu_object_reclaim(os, drro->drr_object,
1414 drro->drr_type, drro->drr_blksz,
1415 drro->drr_bonustype, drro->drr_bonuslen, tx);
1416 }
1417 if (err != 0) {
1418 dmu_tx_commit(tx);
1419 return (SET_ERROR(EINVAL));
1420 }
1421
1422 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1423 tx);
1424 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1425
1426 if (data != NULL) {
1427 dmu_buf_t *db;
1428
1429 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1430 dmu_buf_will_dirty(db, tx);
1431
1432 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1433 bcopy(data, db->db_data, drro->drr_bonuslen);
1434 if (ra->byteswap) {
1435 dmu_object_byteswap_t byteswap =
1436 DMU_OT_BYTESWAP(drro->drr_bonustype);
1437 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1438 drro->drr_bonuslen);
1439 }
1440 dmu_buf_rele(db, FTAG);
1441 }
1442 dmu_tx_commit(tx);
1443 return (0);
1444 }
1445
1446 /* ARGSUSED */
1447 noinline static int
1448 restore_freeobjects(struct restorearg *ra, objset_t *os,
1449 struct drr_freeobjects *drrfo)
1450 {
1451 uint64_t obj;
1452
1453 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1454 return (SET_ERROR(EINVAL));
1455
1456 for (obj = drrfo->drr_firstobj;
1457 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1458 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1459 int err;
1460
1461 if (dmu_object_info(os, obj, NULL) != 0)
1462 continue;
1463
1464 err = dmu_free_long_object(os, obj);
1465 if (err != 0)
1466 return (err);
1467 }
1468 return (0);
1469 }
1470
1471 noinline static int
1472 restore_write(struct restorearg *ra, objset_t *os,
1473 struct drr_write *drrw)
1474 {
1475 dmu_tx_t *tx;
1476 void *data;
1477 int err;
1478
1479 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1480 !DMU_OT_IS_VALID(drrw->drr_type))
1481 return (SET_ERROR(EINVAL));
1482
1483 data = restore_read(ra, drrw->drr_length);
1484 if (data == NULL)
1485 return (ra->err);
1486
1487 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1488 return (SET_ERROR(EINVAL));
1489
1490 tx = dmu_tx_create(os);
1491
1492 dmu_tx_hold_write(tx, drrw->drr_object,
1493 drrw->drr_offset, drrw->drr_length);
1494 err = dmu_tx_assign(tx, TXG_WAIT);
1495 if (err != 0) {
1496 dmu_tx_abort(tx);
1497 return (err);
1498 }
1499 if (ra->byteswap) {
1500 dmu_object_byteswap_t byteswap =
1501 DMU_OT_BYTESWAP(drrw->drr_type);
1502 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1503 }
1504 dmu_write(os, drrw->drr_object,
1505 drrw->drr_offset, drrw->drr_length, data, tx);
1506 dmu_tx_commit(tx);
1507 return (0);
1508 }
1509
1510 /*
1511 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1512 * streams to refer to a copy of the data that is already on the
1513 * system because it came in earlier in the stream. This function
1514 * finds the earlier copy of the data, and uses that copy instead of
1515 * data from the stream to fulfill this write.
1516 */
1517 static int
1518 restore_write_byref(struct restorearg *ra, objset_t *os,
1519 struct drr_write_byref *drrwbr)
1520 {
1521 dmu_tx_t *tx;
1522 int err;
1523 guid_map_entry_t gmesrch;
1524 guid_map_entry_t *gmep;
1525 avl_index_t where;
1526 objset_t *ref_os = NULL;
1527 dmu_buf_t *dbp;
1528
1529 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1530 return (SET_ERROR(EINVAL));
1531
1532 /*
1533 * If the GUID of the referenced dataset is different from the
1534 * GUID of the target dataset, find the referenced dataset.
1535 */
1536 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1537 gmesrch.guid = drrwbr->drr_refguid;
1538 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1539 &where)) == NULL) {
1540 return (SET_ERROR(EINVAL));
1541 }
1542 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1543 return (SET_ERROR(EINVAL));
1544 } else {
1545 ref_os = os;
1546 }
1547
1548 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1549 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1550 if (err != 0)
1551 return (err);
1552
1553 tx = dmu_tx_create(os);
1554
1555 dmu_tx_hold_write(tx, drrwbr->drr_object,
1556 drrwbr->drr_offset, drrwbr->drr_length);
1557 err = dmu_tx_assign(tx, TXG_WAIT);
1558 if (err != 0) {
1559 dmu_tx_abort(tx);
1560 return (err);
1561 }
1562 dmu_write(os, drrwbr->drr_object,
1563 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1564 dmu_buf_rele(dbp, FTAG);
1565 dmu_tx_commit(tx);
1566 return (0);
1567 }
1568
1569 static int
1570 restore_write_embedded(struct restorearg *ra, objset_t *os,
1571 struct drr_write_embedded *drrwnp)
1572 {
1573 dmu_tx_t *tx;
1574 int err;
1575 void *data;
1576
1577 if (drrwnp->drr_offset + drrwnp->drr_length < drrwnp->drr_offset)
1578 return (EINVAL);
1579
1580 if (drrwnp->drr_psize > BPE_PAYLOAD_SIZE)
1581 return (EINVAL);
1582
1583 if (drrwnp->drr_etype >= NUM_BP_EMBEDDED_TYPES)
1584 return (EINVAL);
1585 if (drrwnp->drr_compression >= ZIO_COMPRESS_FUNCTIONS)
1586 return (EINVAL);
1587
1588 data = restore_read(ra, P2ROUNDUP(drrwnp->drr_psize, 8));
1589 if (data == NULL)
1590 return (ra->err);
1591
1592 tx = dmu_tx_create(os);
1593
1594 dmu_tx_hold_write(tx, drrwnp->drr_object,
1595 drrwnp->drr_offset, drrwnp->drr_length);
1596 err = dmu_tx_assign(tx, TXG_WAIT);
1597 if (err != 0) {
1598 dmu_tx_abort(tx);
1599 return (err);
1600 }
1601
1602 dmu_write_embedded(os, drrwnp->drr_object,
1603 drrwnp->drr_offset, data, drrwnp->drr_etype,
1604 drrwnp->drr_compression, drrwnp->drr_lsize, drrwnp->drr_psize,
1605 ra->byteswap ^ ZFS_HOST_BYTEORDER, tx);
1606
1607 dmu_tx_commit(tx);
1608 return (0);
1609 }
1610
1611 static int
1612 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1613 {
1614 dmu_tx_t *tx;
1615 void *data;
1616 dmu_buf_t *db, *db_spill;
1617 int err;
1618
1619 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1620 drrs->drr_length > SPA_MAXBLOCKSIZE)
1621 return (SET_ERROR(EINVAL));
1622
1623 data = restore_read(ra, drrs->drr_length);
1624 if (data == NULL)
1625 return (ra->err);
1626
1627 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1628 return (SET_ERROR(EINVAL));
1629
1630 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1631 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1632 dmu_buf_rele(db, FTAG);
1633 return (err);
1634 }
1635
1636 tx = dmu_tx_create(os);
1637
1638 dmu_tx_hold_spill(tx, db->db_object);
1639
1640 err = dmu_tx_assign(tx, TXG_WAIT);
1641 if (err != 0) {
1642 dmu_buf_rele(db, FTAG);
1643 dmu_buf_rele(db_spill, FTAG);
1644 dmu_tx_abort(tx);
1645 return (err);
1646 }
1647 dmu_buf_will_dirty(db_spill, tx);
1648
1649 if (db_spill->db_size < drrs->drr_length)
1650 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1651 drrs->drr_length, tx));
1652 bcopy(data, db_spill->db_data, drrs->drr_length);
1653
1654 dmu_buf_rele(db, FTAG);
1655 dmu_buf_rele(db_spill, FTAG);
1656
1657 dmu_tx_commit(tx);
1658 return (0);
1659 }
1660
1661 /* ARGSUSED */
1662 noinline static int
1663 restore_free(struct restorearg *ra, objset_t *os,
1664 struct drr_free *drrf)
1665 {
1666 int err;
1667
1668 if (drrf->drr_length != -1ULL &&
1669 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1670 return (SET_ERROR(EINVAL));
1671
1672 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1673 return (SET_ERROR(EINVAL));
1674
1675 err = dmu_free_long_range(os, drrf->drr_object,
1676 drrf->drr_offset, drrf->drr_length);
1677 return (err);
1678 }
1679
1680 /* used to destroy the drc_ds on error */
1681 static void
1682 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1683 {
1684 char name[MAXNAMELEN];
1685 dsl_dataset_name(drc->drc_ds, name);
1686 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1687 (void) dsl_destroy_head(name);
1688 }
1689
1690 /*
1691 * NB: callers *must* call dmu_recv_end() if this succeeds.
1692 */
1693 int
1694 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1695 int cleanup_fd, uint64_t *action_handlep)
1696 {
1697 struct restorearg ra = { 0 };
1698 dmu_replay_record_t *drr;
1699 objset_t *os;
1700 zio_cksum_t pcksum;
1701 int featureflags;
1702
1703 ra.byteswap = drc->drc_byteswap;
1704 ra.cksum = drc->drc_cksum;
1705 ra.vp = vp;
1706 ra.voff = *voffp;
1707 ra.bufsize = 1<<20;
1708 ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
1709
1710 /* these were verified in dmu_recv_begin */
1711 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1712 DMU_SUBSTREAM);
1713 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1714
1715 /*
1716 * Open the objset we are modifying.
1717 */
1718 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1719
1720 ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1721
1722 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1723
1724 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1725 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1726 minor_t minor;
1727
1728 if (cleanup_fd == -1) {
1729 ra.err = SET_ERROR(EBADF);
1730 goto out;
1731 }
1732 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1733 if (ra.err != 0) {
1734 cleanup_fd = -1;
1735 goto out;
1736 }
1737
1738 if (*action_handlep == 0) {
1739 ra.guid_to_ds_map =
1740 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1741 avl_create(ra.guid_to_ds_map, guid_compare,
1742 sizeof (guid_map_entry_t),
1743 offsetof(guid_map_entry_t, avlnode));
1744 ra.err = zfs_onexit_add_cb(minor,
1745 free_guid_map_onexit, ra.guid_to_ds_map,
1746 action_handlep);
1747 if (ra.err != 0)
1748 goto out;
1749 } else {
1750 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1751 (void **)&ra.guid_to_ds_map);
1752 if (ra.err != 0)
1753 goto out;
1754 }
1755
1756 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1757 }
1758
1759 /*
1760 * Read records and process them.
1761 */
1762 pcksum = ra.cksum;
1763 while (ra.err == 0 &&
1764 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1765 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1766 ra.err = SET_ERROR(EINTR);
1767 goto out;
1768 }
1769
1770 if (ra.byteswap)
1771 backup_byteswap(drr);
1772
1773 switch (drr->drr_type) {
1774 case DRR_OBJECT:
1775 {
1776 /*
1777 * We need to make a copy of the record header,
1778 * because restore_{object,write} may need to
1779 * restore_read(), which will invalidate drr.
1780 */
1781 struct drr_object drro = drr->drr_u.drr_object;
1782 ra.err = restore_object(&ra, os, &drro);
1783 break;
1784 }
1785 case DRR_FREEOBJECTS:
1786 {
1787 struct drr_freeobjects drrfo =
1788 drr->drr_u.drr_freeobjects;
1789 ra.err = restore_freeobjects(&ra, os, &drrfo);
1790 break;
1791 }
1792 case DRR_WRITE:
1793 {
1794 struct drr_write drrw = drr->drr_u.drr_write;
1795 ra.err = restore_write(&ra, os, &drrw);
1796 break;
1797 }
1798 case DRR_WRITE_BYREF:
1799 {
1800 struct drr_write_byref drrwbr =
1801 drr->drr_u.drr_write_byref;
1802 ra.err = restore_write_byref(&ra, os, &drrwbr);
1803 break;
1804 }
1805 case DRR_WRITE_EMBEDDED:
1806 {
1807 struct drr_write_embedded drrwe =
1808 drr->drr_u.drr_write_embedded;
1809 ra.err = restore_write_embedded(&ra, os, &drrwe);
1810 break;
1811 }
1812 case DRR_FREE:
1813 {
1814 struct drr_free drrf = drr->drr_u.drr_free;
1815 ra.err = restore_free(&ra, os, &drrf);
1816 break;
1817 }
1818 case DRR_END:
1819 {
1820 struct drr_end drre = drr->drr_u.drr_end;
1821 /*
1822 * We compare against the *previous* checksum
1823 * value, because the stored checksum is of
1824 * everything before the DRR_END record.
1825 */
1826 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1827 ra.err = SET_ERROR(ECKSUM);
1828 goto out;
1829 }
1830 case DRR_SPILL:
1831 {
1832 struct drr_spill drrs = drr->drr_u.drr_spill;
1833 ra.err = restore_spill(&ra, os, &drrs);
1834 break;
1835 }
1836 default:
1837 ra.err = SET_ERROR(EINVAL);
1838 goto out;
1839 }
1840 pcksum = ra.cksum;
1841 }
1842 ASSERT(ra.err != 0);
1843
1844 out:
1845 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1846 zfs_onexit_fd_rele(cleanup_fd);
1847
1848 if (ra.err != 0) {
1849 /*
1850 * destroy what we created, so we don't leave it in the
1851 * inconsistent restoring state.
1852 */
1853 dmu_recv_cleanup_ds(drc);
1854 }
1855
1856 vmem_free(ra.buf, ra.bufsize);
1857 *voffp = ra.voff;
1858 return (ra.err);
1859 }
1860
1861 static int
1862 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1863 {
1864 dmu_recv_cookie_t *drc = arg;
1865 dsl_pool_t *dp = dmu_tx_pool(tx);
1866 int error;
1867
1868 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1869
1870 if (!drc->drc_newfs) {
1871 dsl_dataset_t *origin_head;
1872
1873 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1874 if (error != 0)
1875 return (error);
1876 if (drc->drc_force) {
1877 /*
1878 * We will destroy any snapshots in tofs (i.e. before
1879 * origin_head) that are after the origin (which is
1880 * the snap before drc_ds, because drc_ds can not
1881 * have any snaps of its own).
1882 */
1883 uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1884 while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1885 dsl_dataset_t *snap;
1886 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1887 &snap);
1888 if (error != 0)
1889 return (error);
1890 if (snap->ds_dir != origin_head->ds_dir)
1891 error = SET_ERROR(EINVAL);
1892 if (error == 0) {
1893 error = dsl_destroy_snapshot_check_impl(
1894 snap, B_FALSE);
1895 }
1896 obj = snap->ds_phys->ds_prev_snap_obj;
1897 dsl_dataset_rele(snap, FTAG);
1898 if (error != 0)
1899 return (error);
1900 }
1901 }
1902 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1903 origin_head, drc->drc_force, drc->drc_owner, tx);
1904 if (error != 0) {
1905 dsl_dataset_rele(origin_head, FTAG);
1906 return (error);
1907 }
1908 error = dsl_dataset_snapshot_check_impl(origin_head,
1909 drc->drc_tosnap, tx, B_TRUE);
1910 dsl_dataset_rele(origin_head, FTAG);
1911 if (error != 0)
1912 return (error);
1913
1914 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1915 } else {
1916 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1917 drc->drc_tosnap, tx, B_TRUE);
1918 }
1919 return (error);
1920 }
1921
1922 static void
1923 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1924 {
1925 dmu_recv_cookie_t *drc = arg;
1926 dsl_pool_t *dp = dmu_tx_pool(tx);
1927
1928 spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1929 tx, "snap=%s", drc->drc_tosnap);
1930
1931 if (!drc->drc_newfs) {
1932 dsl_dataset_t *origin_head;
1933
1934 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1935 &origin_head));
1936
1937 if (drc->drc_force) {
1938 /*
1939 * Destroy any snapshots of drc_tofs (origin_head)
1940 * after the origin (the snap before drc_ds).
1941 */
1942 uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1943 while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1944 dsl_dataset_t *snap;
1945 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
1946 &snap));
1947 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
1948 obj = snap->ds_phys->ds_prev_snap_obj;
1949 dsl_destroy_snapshot_sync_impl(snap,
1950 B_FALSE, tx);
1951 dsl_dataset_rele(snap, FTAG);
1952 }
1953 }
1954 VERIFY3P(drc->drc_ds->ds_prev, ==,
1955 origin_head->ds_prev);
1956
1957 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1958 origin_head, tx);
1959 dsl_dataset_snapshot_sync_impl(origin_head,
1960 drc->drc_tosnap, tx);
1961
1962 /* set snapshot's creation time and guid */
1963 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1964 origin_head->ds_prev->ds_phys->ds_creation_time =
1965 drc->drc_drrb->drr_creation_time;
1966 origin_head->ds_prev->ds_phys->ds_guid =
1967 drc->drc_drrb->drr_toguid;
1968 origin_head->ds_prev->ds_phys->ds_flags &=
1969 ~DS_FLAG_INCONSISTENT;
1970
1971 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1972 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1973
1974 dsl_dataset_rele(origin_head, FTAG);
1975 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1976
1977 if (drc->drc_owner != NULL)
1978 VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
1979 } else {
1980 dsl_dataset_t *ds = drc->drc_ds;
1981
1982 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1983
1984 /* set snapshot's creation time and guid */
1985 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1986 ds->ds_prev->ds_phys->ds_creation_time =
1987 drc->drc_drrb->drr_creation_time;
1988 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1989 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1990
1991 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1992 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1993 }
1994 drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1995 /*
1996 * Release the hold from dmu_recv_begin. This must be done before
1997 * we return to open context, so that when we free the dataset's dnode,
1998 * we can evict its bonus buffer.
1999 */
2000 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
2001 drc->drc_ds = NULL;
2002 }
2003
2004 static int
2005 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
2006 {
2007 dsl_pool_t *dp;
2008 dsl_dataset_t *snapds;
2009 guid_map_entry_t *gmep;
2010 int err;
2011
2012 ASSERT(guid_map != NULL);
2013
2014 err = dsl_pool_hold(name, FTAG, &dp);
2015 if (err != 0)
2016 return (err);
2017 gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
2018 err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
2019 if (err == 0) {
2020 gmep->guid = snapds->ds_phys->ds_guid;
2021 gmep->gme_ds = snapds;
2022 avl_add(guid_map, gmep);
2023 dsl_dataset_long_hold(snapds, gmep);
2024 } else {
2025 kmem_free(gmep, sizeof (*gmep));
2026 }
2027
2028 dsl_pool_rele(dp, FTAG);
2029 return (err);
2030 }
2031
2032 static int dmu_recv_end_modified_blocks = 3;
2033
2034 static int
2035 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
2036 {
2037 int error;
2038
2039 #ifdef _KERNEL
2040 char *name;
2041
2042 /*
2043 * We will be destroying the ds; make sure its origin is unmounted if
2044 * necessary.
2045 */
2046 name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2047 dsl_dataset_name(drc->drc_ds, name);
2048 zfs_destroy_unmount_origin(name);
2049 kmem_free(name, MAXNAMELEN);
2050 #endif
2051
2052 error = dsl_sync_task(drc->drc_tofs,
2053 dmu_recv_end_check, dmu_recv_end_sync, drc,
2054 dmu_recv_end_modified_blocks);
2055
2056 if (error != 0)
2057 dmu_recv_cleanup_ds(drc);
2058 return (error);
2059 }
2060
2061 static int
2062 dmu_recv_new_end(dmu_recv_cookie_t *drc)
2063 {
2064 int error;
2065
2066 error = dsl_sync_task(drc->drc_tofs,
2067 dmu_recv_end_check, dmu_recv_end_sync, drc,
2068 dmu_recv_end_modified_blocks);
2069
2070 if (error != 0) {
2071 dmu_recv_cleanup_ds(drc);
2072 } else if (drc->drc_guid_to_ds_map != NULL) {
2073 (void) add_ds_to_guidmap(drc->drc_tofs,
2074 drc->drc_guid_to_ds_map,
2075 drc->drc_newsnapobj);
2076 }
2077 return (error);
2078 }
2079
2080 int
2081 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
2082 {
2083 drc->drc_owner = owner;
2084
2085 if (drc->drc_newfs)
2086 return (dmu_recv_new_end(drc));
2087 else
2088 return (dmu_recv_existing_end(drc));
2089 }
2090
2091 /*
2092 * Return TRUE if this objset is currently being received into.
2093 */
2094 boolean_t
2095 dmu_objset_is_receiving(objset_t *os)
2096 {
2097 return (os->os_dsl_dataset != NULL &&
2098 os->os_dsl_dataset->ds_owner == dmu_recv_tag);
2099 }
2100
2101 #if defined(_KERNEL)
2102 module_param(zfs_send_corrupt_data, int, 0644);
2103 MODULE_PARM_DESC(zfs_send_corrupt_data, "Allow sending corrupt data");
2104 #endif
ZFS On Linux mirror for PVE