1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 2012 by Delphix. All rights reserved.
25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26 */
27
28 #include <sys/dmu.h>
29 #include <sys/dmu_impl.h>
30 #include <sys/dmu_tx.h>
31 #include <sys/dbuf.h>
32 #include <sys/dnode.h>
33 #include <sys/zfs_context.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dsl_dataset.h>
37 #include <sys/dsl_dir.h>
38 #include <sys/dsl_prop.h>
39 #include <sys/dsl_pool.h>
40 #include <sys/dsl_synctask.h>
41 #include <sys/zfs_ioctl.h>
42 #include <sys/zap.h>
43 #include <sys/zio_checksum.h>
44 #include <sys/zfs_znode.h>
45 #include <zfs_fletcher.h>
46 #include <sys/avl.h>
47 #include <sys/ddt.h>
48 #include <sys/zfs_onexit.h>
49
50 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
51 int zfs_send_corrupt_data = B_FALSE;
52
53 static char *dmu_recv_tag = "dmu_recv_tag";
54
55 static int
56 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
57 {
58 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
59 ssize_t resid; /* have to get resid to get detailed errno */
60 ASSERT3U(len % 8, ==, 0);
61
62 fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
63 dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
64 (caddr_t)buf, len,
65 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
66
67 mutex_enter(&ds->ds_sendstream_lock);
68 *dsp->dsa_off += len;
69 mutex_exit(&ds->ds_sendstream_lock);
70
71 return (dsp->dsa_err);
72 }
73
74 static int
75 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
76 uint64_t length)
77 {
78 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
79
80 if (length != -1ULL && offset + length < offset)
81 length = -1ULL;
82
83 /*
84 * If there is a pending op, but it's not PENDING_FREE, push it out,
85 * since free block aggregation can only be done for blocks of the
86 * same type (i.e., DRR_FREE records can only be aggregated with
87 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
88 * aggregated with other DRR_FREEOBJECTS records.
89 */
90 if (dsp->dsa_pending_op != PENDING_NONE &&
91 dsp->dsa_pending_op != PENDING_FREE) {
92 if (dump_bytes(dsp, dsp->dsa_drr,
93 sizeof (dmu_replay_record_t)) != 0)
94 return (EINTR);
95 dsp->dsa_pending_op = PENDING_NONE;
96 }
97
98 if (dsp->dsa_pending_op == PENDING_FREE) {
99 /*
100 * There should never be a PENDING_FREE if length is -1
101 * (because dump_dnode is the only place where this
102 * function is called with a -1, and only after flushing
103 * any pending record).
104 */
105 ASSERT(length != -1ULL);
106 /*
107 * Check to see whether this free block can be aggregated
108 * with pending one.
109 */
110 if (drrf->drr_object == object && drrf->drr_offset +
111 drrf->drr_length == offset) {
112 drrf->drr_length += length;
113 return (0);
114 } else {
115 /* not a continuation. Push out pending record */
116 if (dump_bytes(dsp, dsp->dsa_drr,
117 sizeof (dmu_replay_record_t)) != 0)
118 return (EINTR);
119 dsp->dsa_pending_op = PENDING_NONE;
120 }
121 }
122 /* create a FREE record and make it pending */
123 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
124 dsp->dsa_drr->drr_type = DRR_FREE;
125 drrf->drr_object = object;
126 drrf->drr_offset = offset;
127 drrf->drr_length = length;
128 drrf->drr_toguid = dsp->dsa_toguid;
129 if (length == -1ULL) {
130 if (dump_bytes(dsp, dsp->dsa_drr,
131 sizeof (dmu_replay_record_t)) != 0)
132 return (EINTR);
133 } else {
134 dsp->dsa_pending_op = PENDING_FREE;
135 }
136
137 return (0);
138 }
139
140 static int
141 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
142 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
143 {
144 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
145
146
147 /*
148 * If there is any kind of pending aggregation (currently either
149 * a grouping of free objects or free blocks), push it out to
150 * the stream, since aggregation can't be done across operations
151 * of different types.
152 */
153 if (dsp->dsa_pending_op != PENDING_NONE) {
154 if (dump_bytes(dsp, dsp->dsa_drr,
155 sizeof (dmu_replay_record_t)) != 0)
156 return (EINTR);
157 dsp->dsa_pending_op = PENDING_NONE;
158 }
159 /* write a DATA record */
160 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
161 dsp->dsa_drr->drr_type = DRR_WRITE;
162 drrw->drr_object = object;
163 drrw->drr_type = type;
164 drrw->drr_offset = offset;
165 drrw->drr_length = blksz;
166 drrw->drr_toguid = dsp->dsa_toguid;
167 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
168 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
169 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
170 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
171 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
172 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
173 drrw->drr_key.ddk_cksum = bp->blk_cksum;
174
175 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
176 return (EINTR);
177 if (dump_bytes(dsp, data, blksz) != 0)
178 return (EINTR);
179 return (0);
180 }
181
182 static int
183 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
184 {
185 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
186
187 if (dsp->dsa_pending_op != PENDING_NONE) {
188 if (dump_bytes(dsp, dsp->dsa_drr,
189 sizeof (dmu_replay_record_t)) != 0)
190 return (EINTR);
191 dsp->dsa_pending_op = PENDING_NONE;
192 }
193
194 /* write a SPILL record */
195 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
196 dsp->dsa_drr->drr_type = DRR_SPILL;
197 drrs->drr_object = object;
198 drrs->drr_length = blksz;
199 drrs->drr_toguid = dsp->dsa_toguid;
200
201 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
202 return (EINTR);
203 if (dump_bytes(dsp, data, blksz))
204 return (EINTR);
205 return (0);
206 }
207
208 static int
209 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
210 {
211 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
212
213 /*
214 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
215 * push it out, since free block aggregation can only be done for
216 * blocks of the same type (i.e., DRR_FREE records can only be
217 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
218 * can only be aggregated with other DRR_FREEOBJECTS records.
219 */
220 if (dsp->dsa_pending_op != PENDING_NONE &&
221 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
222 if (dump_bytes(dsp, dsp->dsa_drr,
223 sizeof (dmu_replay_record_t)) != 0)
224 return (EINTR);
225 dsp->dsa_pending_op = PENDING_NONE;
226 }
227 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
228 /*
229 * See whether this free object array can be aggregated
230 * with pending one
231 */
232 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
233 drrfo->drr_numobjs += numobjs;
234 return (0);
235 } else {
236 /* can't be aggregated. Push out pending record */
237 if (dump_bytes(dsp, dsp->dsa_drr,
238 sizeof (dmu_replay_record_t)) != 0)
239 return (EINTR);
240 dsp->dsa_pending_op = PENDING_NONE;
241 }
242 }
243
244 /* write a FREEOBJECTS record */
245 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
246 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
247 drrfo->drr_firstobj = firstobj;
248 drrfo->drr_numobjs = numobjs;
249 drrfo->drr_toguid = dsp->dsa_toguid;
250
251 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
252
253 return (0);
254 }
255
256 static int
257 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
258 {
259 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
260
261 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
262 return (dump_freeobjects(dsp, object, 1));
263
264 if (dsp->dsa_pending_op != PENDING_NONE) {
265 if (dump_bytes(dsp, dsp->dsa_drr,
266 sizeof (dmu_replay_record_t)) != 0)
267 return (EINTR);
268 dsp->dsa_pending_op = PENDING_NONE;
269 }
270
271 /* write an OBJECT record */
272 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
273 dsp->dsa_drr->drr_type = DRR_OBJECT;
274 drro->drr_object = object;
275 drro->drr_type = dnp->dn_type;
276 drro->drr_bonustype = dnp->dn_bonustype;
277 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
278 drro->drr_bonuslen = dnp->dn_bonuslen;
279 drro->drr_checksumtype = dnp->dn_checksum;
280 drro->drr_compress = dnp->dn_compress;
281 drro->drr_toguid = dsp->dsa_toguid;
282
283 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
284 return (EINTR);
285
286 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
287 return (EINTR);
288
289 /* free anything past the end of the file */
290 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
291 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
292 return (EINTR);
293 if (dsp->dsa_err)
294 return (EINTR);
295 return (0);
296 }
297
298 #define BP_SPAN(dnp, level) \
299 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
300 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
301
302 /* ARGSUSED */
303 static int
304 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
305 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
306 {
307 dmu_sendarg_t *dsp = arg;
308 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
309 int err = 0;
310
311 if (issig(JUSTLOOKING) && issig(FORREAL))
312 return (EINTR);
313
314 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
315 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
316 return (0);
317 } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) {
318 uint64_t span = BP_SPAN(dnp, zb->zb_level);
319 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
320 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
321 } else if (bp == NULL) {
322 uint64_t span = BP_SPAN(dnp, zb->zb_level);
323 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
324 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
325 return (0);
326 } else if (type == DMU_OT_DNODE) {
327 dnode_phys_t *blk;
328 int i;
329 int blksz = BP_GET_LSIZE(bp);
330 uint32_t aflags = ARC_WAIT;
331 arc_buf_t *abuf;
332
333 if (dsl_read(NULL, spa, bp, pbuf,
334 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
335 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
336 return (EIO);
337
338 blk = abuf->b_data;
339 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
340 uint64_t dnobj = (zb->zb_blkid <<
341 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
342 err = dump_dnode(dsp, dnobj, blk+i);
343 if (err)
344 break;
345 }
346 (void) arc_buf_remove_ref(abuf, &abuf);
347 } else if (type == DMU_OT_SA) {
348 uint32_t aflags = ARC_WAIT;
349 arc_buf_t *abuf;
350 int blksz = BP_GET_LSIZE(bp);
351
352 if (arc_read_nolock(NULL, spa, bp,
353 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
354 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
355 return (EIO);
356
357 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
358 (void) arc_buf_remove_ref(abuf, &abuf);
359 } else { /* it's a level-0 block of a regular object */
360 uint32_t aflags = ARC_WAIT;
361 arc_buf_t *abuf;
362 int blksz = BP_GET_LSIZE(bp);
363
364 if (dsl_read(NULL, spa, bp, pbuf,
365 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
366 ZIO_FLAG_CANFAIL, &aflags, zb) != 0) {
367 if (zfs_send_corrupt_data) {
368 /* Send a block filled with 0x"zfs badd bloc" */
369 abuf = arc_buf_alloc(spa, blksz, &abuf,
370 ARC_BUFC_DATA);
371 uint64_t *ptr;
372 for (ptr = abuf->b_data;
373 (char *)ptr < (char *)abuf->b_data + blksz;
374 ptr++)
375 *ptr = 0x2f5baddb10c;
376 } else {
377 return (EIO);
378 }
379 }
380
381 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
382 blksz, bp, abuf->b_data);
383 (void) arc_buf_remove_ref(abuf, &abuf);
384 }
385
386 ASSERT(err == 0 || err == EINTR);
387 return (err);
388 }
389
390 int
391 dmu_send(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
392 int outfd, vnode_t *vp, offset_t *off)
393 {
394 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
395 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
396 dmu_replay_record_t *drr;
397 dmu_sendarg_t *dsp;
398 int err;
399 uint64_t fromtxg = 0;
400
401 /* tosnap must be a snapshot */
402 if (ds->ds_phys->ds_next_snap_obj == 0)
403 return (EINVAL);
404
405 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
406 if (fromds && (ds->ds_dir != fromds->ds_dir ||
407 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
408 return (EXDEV);
409
410 if (fromorigin) {
411 dsl_pool_t *dp = ds->ds_dir->dd_pool;
412
413 if (fromsnap)
414 return (EINVAL);
415
416 if (dsl_dir_is_clone(ds->ds_dir)) {
417 rw_enter(&dp->dp_config_rwlock, RW_READER);
418 err = dsl_dataset_hold_obj(dp,
419 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
420 rw_exit(&dp->dp_config_rwlock);
421 if (err)
422 return (err);
423 } else {
424 fromorigin = B_FALSE;
425 }
426 }
427
428
429 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
430 drr->drr_type = DRR_BEGIN;
431 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
432 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
433 DMU_SUBSTREAM);
434
435 #ifdef _KERNEL
436 if (dmu_objset_type(tosnap) == DMU_OST_ZFS) {
437 uint64_t version;
438 if (zfs_get_zplprop(tosnap, ZFS_PROP_VERSION, &version) != 0) {
439 kmem_free(drr, sizeof (dmu_replay_record_t));
440 return (EINVAL);
441 }
442 if (version == ZPL_VERSION_SA) {
443 DMU_SET_FEATUREFLAGS(
444 drr->drr_u.drr_begin.drr_versioninfo,
445 DMU_BACKUP_FEATURE_SA_SPILL);
446 }
447 }
448 #endif
449
450 drr->drr_u.drr_begin.drr_creation_time =
451 ds->ds_phys->ds_creation_time;
452 drr->drr_u.drr_begin.drr_type = tosnap->os_phys->os_type;
453 if (fromorigin)
454 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
455 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
456 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
457 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
458
459 if (fromds)
460 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
461 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
462
463 if (fromds)
464 fromtxg = fromds->ds_phys->ds_creation_txg;
465 if (fromorigin)
466 dsl_dataset_rele(fromds, FTAG);
467
468 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
469
470 dsp->dsa_drr = drr;
471 dsp->dsa_vp = vp;
472 dsp->dsa_outfd = outfd;
473 dsp->dsa_proc = curproc;
474 dsp->dsa_os = tosnap;
475 dsp->dsa_off = off;
476 dsp->dsa_toguid = ds->ds_phys->ds_guid;
477 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
478 dsp->dsa_pending_op = PENDING_NONE;
479
480 mutex_enter(&ds->ds_sendstream_lock);
481 list_insert_head(&ds->ds_sendstreams, dsp);
482 mutex_exit(&ds->ds_sendstream_lock);
483
484 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
485 err = dsp->dsa_err;
486 goto out;
487 }
488
489 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
490 backup_cb, dsp);
491
492 if (dsp->dsa_pending_op != PENDING_NONE)
493 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
494 err = EINTR;
495
496 if (err) {
497 if (err == EINTR && dsp->dsa_err)
498 err = dsp->dsa_err;
499 goto out;
500 }
501
502 bzero(drr, sizeof (dmu_replay_record_t));
503 drr->drr_type = DRR_END;
504 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
505 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
506
507 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
508 err = dsp->dsa_err;
509 goto out;
510 }
511
512 out:
513 mutex_enter(&ds->ds_sendstream_lock);
514 list_remove(&ds->ds_sendstreams, dsp);
515 mutex_exit(&ds->ds_sendstream_lock);
516
517 kmem_free(drr, sizeof (dmu_replay_record_t));
518 kmem_free(dsp, sizeof (dmu_sendarg_t));
519
520 return (err);
521 }
522
523 int
524 dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
525 uint64_t *sizep)
526 {
527 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
528 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
529 dsl_pool_t *dp = ds->ds_dir->dd_pool;
530 int err;
531 uint64_t size;
532
533 /* tosnap must be a snapshot */
534 if (ds->ds_phys->ds_next_snap_obj == 0)
535 return (EINVAL);
536
537 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
538 if (fromds && (ds->ds_dir != fromds->ds_dir ||
539 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
540 return (EXDEV);
541
542 if (fromorigin) {
543 if (fromsnap)
544 return (EINVAL);
545
546 if (dsl_dir_is_clone(ds->ds_dir)) {
547 rw_enter(&dp->dp_config_rwlock, RW_READER);
548 err = dsl_dataset_hold_obj(dp,
549 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
550 rw_exit(&dp->dp_config_rwlock);
551 if (err)
552 return (err);
553 } else {
554 fromorigin = B_FALSE;
555 }
556 }
557
558 /* Get uncompressed size estimate of changed data. */
559 if (fromds == NULL) {
560 size = ds->ds_phys->ds_uncompressed_bytes;
561 } else {
562 uint64_t used, comp;
563 err = dsl_dataset_space_written(fromds, ds,
564 &used, &comp, &size);
565 if (fromorigin)
566 dsl_dataset_rele(fromds, FTAG);
567 if (err)
568 return (err);
569 }
570
571 /*
572 * Assume that space (both on-disk and in-stream) is dominated by
573 * data. We will adjust for indirect blocks and the copies property,
574 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
575 */
576
577 /*
578 * Subtract out approximate space used by indirect blocks.
579 * Assume most space is used by data blocks (non-indirect, non-dnode).
580 * Assume all blocks are recordsize. Assume ditto blocks and
581 * internal fragmentation counter out compression.
582 *
583 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
584 * block, which we observe in practice.
585 */
586 uint64_t recordsize;
587 rw_enter(&dp->dp_config_rwlock, RW_READER);
588 err = dsl_prop_get_ds(ds, "recordsize",
589 sizeof (recordsize), 1, &recordsize, NULL);
590 rw_exit(&dp->dp_config_rwlock);
591 if (err)
592 return (err);
593 size -= size / recordsize * sizeof (blkptr_t);
594
595 /* Add in the space for the record associated with each block. */
596 size += size / recordsize * sizeof (dmu_replay_record_t);
597
598 *sizep = size;
599
600 return (0);
601 }
602
603 struct recvbeginsyncarg {
604 const char *tofs;
605 const char *tosnap;
606 dsl_dataset_t *origin;
607 uint64_t fromguid;
608 dmu_objset_type_t type;
609 void *tag;
610 boolean_t force;
611 uint64_t dsflags;
612 char clonelastname[MAXNAMELEN];
613 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
614 cred_t *cr;
615 };
616
617 /* ARGSUSED */
618 static int
619 recv_new_check(void *arg1, void *arg2, dmu_tx_t *tx)
620 {
621 dsl_dir_t *dd = arg1;
622 struct recvbeginsyncarg *rbsa = arg2;
623 objset_t *mos = dd->dd_pool->dp_meta_objset;
624 uint64_t val;
625 int err;
626
627 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
628 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
629
630 if (err != ENOENT)
631 return (err ? err : EEXIST);
632
633 if (rbsa->origin) {
634 /* make sure it's a snap in the same pool */
635 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
636 return (EXDEV);
637 if (!dsl_dataset_is_snapshot(rbsa->origin))
638 return (EINVAL);
639 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
640 return (ENODEV);
641 }
642
643 return (0);
644 }
645
646 static void
647 recv_new_sync(void *arg1, void *arg2, dmu_tx_t *tx)
648 {
649 dsl_dir_t *dd = arg1;
650 struct recvbeginsyncarg *rbsa = arg2;
651 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
652 uint64_t dsobj;
653
654 /* Create and open new dataset. */
655 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
656 rbsa->origin, flags, rbsa->cr, tx);
657 VERIFY(0 == dsl_dataset_own_obj(dd->dd_pool, dsobj,
658 B_TRUE, dmu_recv_tag, &rbsa->ds));
659
660 if (rbsa->origin == NULL) {
661 (void) dmu_objset_create_impl(dd->dd_pool->dp_spa,
662 rbsa->ds, &rbsa->ds->ds_phys->ds_bp, rbsa->type, tx);
663 }
664
665 spa_history_log_internal(LOG_DS_REPLAY_FULL_SYNC,
666 dd->dd_pool->dp_spa, tx, "dataset = %lld", dsobj);
667 }
668
669 /* ARGSUSED */
670 static int
671 recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
672 {
673 dsl_dataset_t *ds = arg1;
674 struct recvbeginsyncarg *rbsa = arg2;
675 int err;
676 uint64_t val;
677
678 /* must not have any changes since most recent snapshot */
679 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
680 return (ETXTBSY);
681
682 /* new snapshot name must not exist */
683 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
684 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
685 if (err == 0)
686 return (EEXIST);
687 if (err != ENOENT)
688 return (err);
689
690 if (rbsa->fromguid) {
691 /* if incremental, most recent snapshot must match fromguid */
692 if (ds->ds_prev == NULL)
693 return (ENODEV);
694
695 /*
696 * most recent snapshot must match fromguid, or there are no
697 * changes since the fromguid one
698 */
699 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid) {
700 uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth;
701 uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj;
702 while (obj != 0) {
703 dsl_dataset_t *snap;
704 err = dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
705 obj, FTAG, &snap);
706 if (err)
707 return (ENODEV);
708 if (snap->ds_phys->ds_creation_txg < birth) {
709 dsl_dataset_rele(snap, FTAG);
710 return (ENODEV);
711 }
712 if (snap->ds_phys->ds_guid == rbsa->fromguid) {
713 dsl_dataset_rele(snap, FTAG);
714 break; /* it's ok */
715 }
716 obj = snap->ds_phys->ds_prev_snap_obj;
717 dsl_dataset_rele(snap, FTAG);
718 }
719 if (obj == 0)
720 return (ENODEV);
721 }
722 } else {
723 /* if full, most recent snapshot must be $ORIGIN */
724 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
725 return (ENODEV);
726 }
727
728 /* temporary clone name must not exist */
729 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
730 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
731 rbsa->clonelastname, 8, 1, &val);
732 if (err == 0)
733 return (EEXIST);
734 if (err != ENOENT)
735 return (err);
736
737 return (0);
738 }
739
740 /* ARGSUSED */
741 static void
742 recv_existing_sync(void *arg1, void *arg2, dmu_tx_t *tx)
743 {
744 dsl_dataset_t *ohds = arg1;
745 struct recvbeginsyncarg *rbsa = arg2;
746 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
747 dsl_dataset_t *cds;
748 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
749 uint64_t dsobj;
750
751 /* create and open the temporary clone */
752 dsobj = dsl_dataset_create_sync(ohds->ds_dir, rbsa->clonelastname,
753 ohds->ds_prev, flags, rbsa->cr, tx);
754 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, B_TRUE, dmu_recv_tag, &cds));
755
756 /*
757 * If we actually created a non-clone, we need to create the
758 * objset in our new dataset.
759 */
760 if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds))) {
761 (void) dmu_objset_create_impl(dp->dp_spa,
762 cds, dsl_dataset_get_blkptr(cds), rbsa->type, tx);
763 }
764
765 rbsa->ds = cds;
766
767 spa_history_log_internal(LOG_DS_REPLAY_INC_SYNC,
768 dp->dp_spa, tx, "dataset = %lld", dsobj);
769 }
770
771 static boolean_t
772 dmu_recv_verify_features(dsl_dataset_t *ds, struct drr_begin *drrb)
773 {
774 int featureflags;
775
776 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
777
778 /* Verify pool version supports SA if SA_SPILL feature set */
779 return ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
780 (spa_version(dsl_dataset_get_spa(ds)) < SPA_VERSION_SA));
781 }
782
783 /*
784 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
785 * succeeds; otherwise we will leak the holds on the datasets.
786 */
787 int
788 dmu_recv_begin(char *tofs, char *tosnap, char *top_ds, struct drr_begin *drrb,
789 boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
790 {
791 int err = 0;
792 boolean_t byteswap;
793 struct recvbeginsyncarg rbsa = { 0 };
794 uint64_t versioninfo;
795 int flags;
796 dsl_dataset_t *ds;
797
798 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
799 byteswap = FALSE;
800 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
801 byteswap = TRUE;
802 else
803 return (EINVAL);
804
805 rbsa.tofs = tofs;
806 rbsa.tosnap = tosnap;
807 rbsa.origin = origin ? origin->os_dsl_dataset : NULL;
808 rbsa.fromguid = drrb->drr_fromguid;
809 rbsa.type = drrb->drr_type;
810 rbsa.tag = FTAG;
811 rbsa.dsflags = 0;
812 rbsa.cr = CRED();
813 versioninfo = drrb->drr_versioninfo;
814 flags = drrb->drr_flags;
815
816 if (byteswap) {
817 rbsa.type = BSWAP_32(rbsa.type);
818 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
819 versioninfo = BSWAP_64(versioninfo);
820 flags = BSWAP_32(flags);
821 }
822
823 if (DMU_GET_STREAM_HDRTYPE(versioninfo) == DMU_COMPOUNDSTREAM ||
824 rbsa.type >= DMU_OST_NUMTYPES ||
825 ((flags & DRR_FLAG_CLONE) && origin == NULL))
826 return (EINVAL);
827
828 if (flags & DRR_FLAG_CI_DATA)
829 rbsa.dsflags = DS_FLAG_CI_DATASET;
830
831 bzero(drc, sizeof (dmu_recv_cookie_t));
832 drc->drc_drrb = drrb;
833 drc->drc_tosnap = tosnap;
834 drc->drc_top_ds = top_ds;
835 drc->drc_force = force;
836
837 /*
838 * Process the begin in syncing context.
839 */
840
841 /* open the dataset we are logically receiving into */
842 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
843 if (err == 0) {
844 if (dmu_recv_verify_features(ds, drrb)) {
845 dsl_dataset_rele(ds, dmu_recv_tag);
846 return (ENOTSUP);
847 }
848 /* target fs already exists; recv into temp clone */
849
850 /* Can't recv a clone into an existing fs */
851 if (flags & DRR_FLAG_CLONE) {
852 dsl_dataset_rele(ds, dmu_recv_tag);
853 return (EINVAL);
854 }
855
856 /* must not have an incremental recv already in progress */
857 if (!mutex_tryenter(&ds->ds_recvlock)) {
858 dsl_dataset_rele(ds, dmu_recv_tag);
859 return (EBUSY);
860 }
861
862 /* tmp clone name is: tofs/%tosnap" */
863 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
864 "%%%s", tosnap);
865 rbsa.force = force;
866 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
867 recv_existing_check, recv_existing_sync, ds, &rbsa, 5);
868 if (err) {
869 mutex_exit(&ds->ds_recvlock);
870 dsl_dataset_rele(ds, dmu_recv_tag);
871 return (err);
872 }
873 drc->drc_logical_ds = ds;
874 drc->drc_real_ds = rbsa.ds;
875 } else if (err == ENOENT) {
876 /* target fs does not exist; must be a full backup or clone */
877 char *cp;
878
879 /*
880 * If it's a non-clone incremental, we are missing the
881 * target fs, so fail the recv.
882 */
883 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE))
884 return (ENOENT);
885
886 /* Open the parent of tofs */
887 cp = strrchr(tofs, '/');
888 *cp = '\0';
889 err = dsl_dataset_hold(tofs, FTAG, &ds);
890 *cp = '/';
891 if (err)
892 return (err);
893
894 if (dmu_recv_verify_features(ds, drrb)) {
895 dsl_dataset_rele(ds, FTAG);
896 return (ENOTSUP);
897 }
898
899 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
900 recv_new_check, recv_new_sync, ds->ds_dir, &rbsa, 5);
901 dsl_dataset_rele(ds, FTAG);
902 if (err)
903 return (err);
904 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
905 drc->drc_newfs = B_TRUE;
906 }
907
908 return (err);
909 }
910
911 struct restorearg {
912 int err;
913 int byteswap;
914 vnode_t *vp;
915 char *buf;
916 uint64_t voff;
917 int bufsize; /* amount of memory allocated for buf */
918 zio_cksum_t cksum;
919 avl_tree_t *guid_to_ds_map;
920 };
921
922 typedef struct guid_map_entry {
923 uint64_t guid;
924 dsl_dataset_t *gme_ds;
925 avl_node_t avlnode;
926 } guid_map_entry_t;
927
928 static int
929 guid_compare(const void *arg1, const void *arg2)
930 {
931 const guid_map_entry_t *gmep1 = arg1;
932 const guid_map_entry_t *gmep2 = arg2;
933
934 if (gmep1->guid < gmep2->guid)
935 return (-1);
936 else if (gmep1->guid > gmep2->guid)
937 return (1);
938 return (0);
939 }
940
941 static void
942 free_guid_map_onexit(void *arg)
943 {
944 avl_tree_t *ca = arg;
945 void *cookie = NULL;
946 guid_map_entry_t *gmep;
947
948 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
949 dsl_dataset_rele(gmep->gme_ds, ca);
950 kmem_free(gmep, sizeof (guid_map_entry_t));
951 }
952 avl_destroy(ca);
953 kmem_free(ca, sizeof (avl_tree_t));
954 }
955
956 static void *
957 restore_read(struct restorearg *ra, int len)
958 {
959 void *rv;
960 int done = 0;
961
962 /* some things will require 8-byte alignment, so everything must */
963 ASSERT3U(len % 8, ==, 0);
964
965 while (done < len) {
966 ssize_t resid;
967
968 ra->err = vn_rdwr(UIO_READ, ra->vp,
969 (caddr_t)ra->buf + done, len - done,
970 ra->voff, UIO_SYSSPACE, FAPPEND,
971 RLIM64_INFINITY, CRED(), &resid);
972
973 if (resid == len - done)
974 ra->err = EINVAL;
975 ra->voff += len - done - resid;
976 done = len - resid;
977 if (ra->err)
978 return (NULL);
979 }
980
981 ASSERT3U(done, ==, len);
982 rv = ra->buf;
983 if (ra->byteswap)
984 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
985 else
986 fletcher_4_incremental_native(rv, len, &ra->cksum);
987 return (rv);
988 }
989
990 static void
991 backup_byteswap(dmu_replay_record_t *drr)
992 {
993 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
994 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
995 drr->drr_type = BSWAP_32(drr->drr_type);
996 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
997 switch (drr->drr_type) {
998 case DRR_BEGIN:
999 DO64(drr_begin.drr_magic);
1000 DO64(drr_begin.drr_versioninfo);
1001 DO64(drr_begin.drr_creation_time);
1002 DO32(drr_begin.drr_type);
1003 DO32(drr_begin.drr_flags);
1004 DO64(drr_begin.drr_toguid);
1005 DO64(drr_begin.drr_fromguid);
1006 break;
1007 case DRR_OBJECT:
1008 DO64(drr_object.drr_object);
1009 /* DO64(drr_object.drr_allocation_txg); */
1010 DO32(drr_object.drr_type);
1011 DO32(drr_object.drr_bonustype);
1012 DO32(drr_object.drr_blksz);
1013 DO32(drr_object.drr_bonuslen);
1014 DO64(drr_object.drr_toguid);
1015 break;
1016 case DRR_FREEOBJECTS:
1017 DO64(drr_freeobjects.drr_firstobj);
1018 DO64(drr_freeobjects.drr_numobjs);
1019 DO64(drr_freeobjects.drr_toguid);
1020 break;
1021 case DRR_WRITE:
1022 DO64(drr_write.drr_object);
1023 DO32(drr_write.drr_type);
1024 DO64(drr_write.drr_offset);
1025 DO64(drr_write.drr_length);
1026 DO64(drr_write.drr_toguid);
1027 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1028 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1029 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1030 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1031 DO64(drr_write.drr_key.ddk_prop);
1032 break;
1033 case DRR_WRITE_BYREF:
1034 DO64(drr_write_byref.drr_object);
1035 DO64(drr_write_byref.drr_offset);
1036 DO64(drr_write_byref.drr_length);
1037 DO64(drr_write_byref.drr_toguid);
1038 DO64(drr_write_byref.drr_refguid);
1039 DO64(drr_write_byref.drr_refobject);
1040 DO64(drr_write_byref.drr_refoffset);
1041 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1042 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1043 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1044 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1045 DO64(drr_write_byref.drr_key.ddk_prop);
1046 break;
1047 case DRR_FREE:
1048 DO64(drr_free.drr_object);
1049 DO64(drr_free.drr_offset);
1050 DO64(drr_free.drr_length);
1051 DO64(drr_free.drr_toguid);
1052 break;
1053 case DRR_SPILL:
1054 DO64(drr_spill.drr_object);
1055 DO64(drr_spill.drr_length);
1056 DO64(drr_spill.drr_toguid);
1057 break;
1058 case DRR_END:
1059 DO64(drr_end.drr_checksum.zc_word[0]);
1060 DO64(drr_end.drr_checksum.zc_word[1]);
1061 DO64(drr_end.drr_checksum.zc_word[2]);
1062 DO64(drr_end.drr_checksum.zc_word[3]);
1063 DO64(drr_end.drr_toguid);
1064 break;
1065 }
1066 #undef DO64
1067 #undef DO32
1068 }
1069
1070 static int
1071 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1072 {
1073 int err;
1074 dmu_tx_t *tx;
1075 void *data = NULL;
1076
1077 if (drro->drr_type == DMU_OT_NONE ||
1078 !DMU_OT_IS_VALID(drro->drr_type) ||
1079 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1080 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1081 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1082 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1083 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1084 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1085 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1086 return (EINVAL);
1087 }
1088
1089 err = dmu_object_info(os, drro->drr_object, NULL);
1090
1091 if (err != 0 && err != ENOENT)
1092 return (EINVAL);
1093
1094 if (drro->drr_bonuslen) {
1095 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1096 if (ra->err)
1097 return (ra->err);
1098 }
1099
1100 if (err == ENOENT) {
1101 /* currently free, want to be allocated */
1102 tx = dmu_tx_create(os);
1103 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1104 err = dmu_tx_assign(tx, TXG_WAIT);
1105 if (err) {
1106 dmu_tx_abort(tx);
1107 return (err);
1108 }
1109 err = dmu_object_claim(os, drro->drr_object,
1110 drro->drr_type, drro->drr_blksz,
1111 drro->drr_bonustype, drro->drr_bonuslen, tx);
1112 dmu_tx_commit(tx);
1113 } else {
1114 /* currently allocated, want to be allocated */
1115 err = dmu_object_reclaim(os, drro->drr_object,
1116 drro->drr_type, drro->drr_blksz,
1117 drro->drr_bonustype, drro->drr_bonuslen);
1118 }
1119 if (err) {
1120 return (EINVAL);
1121 }
1122
1123 tx = dmu_tx_create(os);
1124 dmu_tx_hold_bonus(tx, drro->drr_object);
1125 err = dmu_tx_assign(tx, TXG_WAIT);
1126 if (err) {
1127 dmu_tx_abort(tx);
1128 return (err);
1129 }
1130
1131 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1132 tx);
1133 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1134
1135 if (data != NULL) {
1136 dmu_buf_t *db;
1137
1138 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1139 dmu_buf_will_dirty(db, tx);
1140
1141 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1142 bcopy(data, db->db_data, drro->drr_bonuslen);
1143 if (ra->byteswap) {
1144 dmu_object_byteswap_t byteswap =
1145 DMU_OT_BYTESWAP(drro->drr_bonustype);
1146 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1147 drro->drr_bonuslen);
1148 }
1149 dmu_buf_rele(db, FTAG);
1150 }
1151 dmu_tx_commit(tx);
1152 return (0);
1153 }
1154
1155 /* ARGSUSED */
1156 static int
1157 restore_freeobjects(struct restorearg *ra, objset_t *os,
1158 struct drr_freeobjects *drrfo)
1159 {
1160 uint64_t obj;
1161
1162 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1163 return (EINVAL);
1164
1165 for (obj = drrfo->drr_firstobj;
1166 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1167 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1168 int err;
1169
1170 if (dmu_object_info(os, obj, NULL) != 0)
1171 continue;
1172
1173 err = dmu_free_object(os, obj);
1174 if (err)
1175 return (err);
1176 }
1177 return (0);
1178 }
1179
1180 static int
1181 restore_write(struct restorearg *ra, objset_t *os,
1182 struct drr_write *drrw)
1183 {
1184 dmu_tx_t *tx;
1185 void *data;
1186 int err;
1187
1188 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1189 !DMU_OT_IS_VALID(drrw->drr_type))
1190 return (EINVAL);
1191
1192 data = restore_read(ra, drrw->drr_length);
1193 if (data == NULL)
1194 return (ra->err);
1195
1196 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1197 return (EINVAL);
1198
1199 tx = dmu_tx_create(os);
1200
1201 dmu_tx_hold_write(tx, drrw->drr_object,
1202 drrw->drr_offset, drrw->drr_length);
1203 err = dmu_tx_assign(tx, TXG_WAIT);
1204 if (err) {
1205 dmu_tx_abort(tx);
1206 return (err);
1207 }
1208 if (ra->byteswap) {
1209 dmu_object_byteswap_t byteswap =
1210 DMU_OT_BYTESWAP(drrw->drr_type);
1211 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1212 }
1213 dmu_write(os, drrw->drr_object,
1214 drrw->drr_offset, drrw->drr_length, data, tx);
1215 dmu_tx_commit(tx);
1216 return (0);
1217 }
1218
1219 /*
1220 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1221 * streams to refer to a copy of the data that is already on the
1222 * system because it came in earlier in the stream. This function
1223 * finds the earlier copy of the data, and uses that copy instead of
1224 * data from the stream to fulfill this write.
1225 */
1226 static int
1227 restore_write_byref(struct restorearg *ra, objset_t *os,
1228 struct drr_write_byref *drrwbr)
1229 {
1230 dmu_tx_t *tx;
1231 int err;
1232 guid_map_entry_t gmesrch;
1233 guid_map_entry_t *gmep;
1234 avl_index_t where;
1235 objset_t *ref_os = NULL;
1236 dmu_buf_t *dbp;
1237
1238 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1239 return (EINVAL);
1240
1241 /*
1242 * If the GUID of the referenced dataset is different from the
1243 * GUID of the target dataset, find the referenced dataset.
1244 */
1245 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1246 gmesrch.guid = drrwbr->drr_refguid;
1247 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1248 &where)) == NULL) {
1249 return (EINVAL);
1250 }
1251 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1252 return (EINVAL);
1253 } else {
1254 ref_os = os;
1255 }
1256
1257 if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1258 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1259 return (err);
1260
1261 tx = dmu_tx_create(os);
1262
1263 dmu_tx_hold_write(tx, drrwbr->drr_object,
1264 drrwbr->drr_offset, drrwbr->drr_length);
1265 err = dmu_tx_assign(tx, TXG_WAIT);
1266 if (err) {
1267 dmu_tx_abort(tx);
1268 return (err);
1269 }
1270 dmu_write(os, drrwbr->drr_object,
1271 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1272 dmu_buf_rele(dbp, FTAG);
1273 dmu_tx_commit(tx);
1274 return (0);
1275 }
1276
1277 static int
1278 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1279 {
1280 dmu_tx_t *tx;
1281 void *data;
1282 dmu_buf_t *db, *db_spill;
1283 int err;
1284
1285 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1286 drrs->drr_length > SPA_MAXBLOCKSIZE)
1287 return (EINVAL);
1288
1289 data = restore_read(ra, drrs->drr_length);
1290 if (data == NULL)
1291 return (ra->err);
1292
1293 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1294 return (EINVAL);
1295
1296 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1297 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1298 dmu_buf_rele(db, FTAG);
1299 return (err);
1300 }
1301
1302 tx = dmu_tx_create(os);
1303
1304 dmu_tx_hold_spill(tx, db->db_object);
1305
1306 err = dmu_tx_assign(tx, TXG_WAIT);
1307 if (err) {
1308 dmu_buf_rele(db, FTAG);
1309 dmu_buf_rele(db_spill, FTAG);
1310 dmu_tx_abort(tx);
1311 return (err);
1312 }
1313 dmu_buf_will_dirty(db_spill, tx);
1314
1315 if (db_spill->db_size < drrs->drr_length)
1316 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1317 drrs->drr_length, tx));
1318 bcopy(data, db_spill->db_data, drrs->drr_length);
1319
1320 dmu_buf_rele(db, FTAG);
1321 dmu_buf_rele(db_spill, FTAG);
1322
1323 dmu_tx_commit(tx);
1324 return (0);
1325 }
1326
1327 /* ARGSUSED */
1328 static int
1329 restore_free(struct restorearg *ra, objset_t *os,
1330 struct drr_free *drrf)
1331 {
1332 int err;
1333
1334 if (drrf->drr_length != -1ULL &&
1335 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1336 return (EINVAL);
1337
1338 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1339 return (EINVAL);
1340
1341 err = dmu_free_long_range(os, drrf->drr_object,
1342 drrf->drr_offset, drrf->drr_length);
1343 return (err);
1344 }
1345
1346 /*
1347 * NB: callers *must* call dmu_recv_end() if this succeeds.
1348 */
1349 int
1350 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1351 int cleanup_fd, uint64_t *action_handlep)
1352 {
1353 struct restorearg ra = { 0 };
1354 dmu_replay_record_t *drr;
1355 objset_t *os;
1356 zio_cksum_t pcksum;
1357 int featureflags;
1358
1359 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1360 ra.byteswap = TRUE;
1361
1362 {
1363 /* compute checksum of drr_begin record */
1364 dmu_replay_record_t *drr;
1365 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1366
1367 drr->drr_type = DRR_BEGIN;
1368 drr->drr_u.drr_begin = *drc->drc_drrb;
1369 if (ra.byteswap) {
1370 fletcher_4_incremental_byteswap(drr,
1371 sizeof (dmu_replay_record_t), &ra.cksum);
1372 } else {
1373 fletcher_4_incremental_native(drr,
1374 sizeof (dmu_replay_record_t), &ra.cksum);
1375 }
1376 kmem_free(drr, sizeof (dmu_replay_record_t));
1377 }
1378
1379 if (ra.byteswap) {
1380 struct drr_begin *drrb = drc->drc_drrb;
1381 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1382 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1383 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1384 drrb->drr_type = BSWAP_32(drrb->drr_type);
1385 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1386 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1387 }
1388
1389 ra.vp = vp;
1390 ra.voff = *voffp;
1391 ra.bufsize = 1<<20;
1392 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1393
1394 /* these were verified in dmu_recv_begin */
1395 ASSERT(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo) ==
1396 DMU_SUBSTREAM);
1397 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1398
1399 /*
1400 * Open the objset we are modifying.
1401 */
1402 VERIFY(dmu_objset_from_ds(drc->drc_real_ds, &os) == 0);
1403
1404 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1405
1406 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1407
1408 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1409 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1410 minor_t minor;
1411
1412 if (cleanup_fd == -1) {
1413 ra.err = EBADF;
1414 goto out;
1415 }
1416 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1417 if (ra.err) {
1418 cleanup_fd = -1;
1419 goto out;
1420 }
1421
1422 if (*action_handlep == 0) {
1423 ra.guid_to_ds_map =
1424 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1425 avl_create(ra.guid_to_ds_map, guid_compare,
1426 sizeof (guid_map_entry_t),
1427 offsetof(guid_map_entry_t, avlnode));
1428 ra.err = zfs_onexit_add_cb(minor,
1429 free_guid_map_onexit, ra.guid_to_ds_map,
1430 action_handlep);
1431 if (ra.err)
1432 goto out;
1433 } else {
1434 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1435 (void **)&ra.guid_to_ds_map);
1436 if (ra.err)
1437 goto out;
1438 }
1439
1440 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1441 }
1442
1443 /*
1444 * Read records and process them.
1445 */
1446 pcksum = ra.cksum;
1447 while (ra.err == 0 &&
1448 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1449 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1450 ra.err = EINTR;
1451 goto out;
1452 }
1453
1454 if (ra.byteswap)
1455 backup_byteswap(drr);
1456
1457 switch (drr->drr_type) {
1458 case DRR_OBJECT:
1459 {
1460 /*
1461 * We need to make a copy of the record header,
1462 * because restore_{object,write} may need to
1463 * restore_read(), which will invalidate drr.
1464 */
1465 struct drr_object drro = drr->drr_u.drr_object;
1466 ra.err = restore_object(&ra, os, &drro);
1467 break;
1468 }
1469 case DRR_FREEOBJECTS:
1470 {
1471 struct drr_freeobjects drrfo =
1472 drr->drr_u.drr_freeobjects;
1473 ra.err = restore_freeobjects(&ra, os, &drrfo);
1474 break;
1475 }
1476 case DRR_WRITE:
1477 {
1478 struct drr_write drrw = drr->drr_u.drr_write;
1479 ra.err = restore_write(&ra, os, &drrw);
1480 break;
1481 }
1482 case DRR_WRITE_BYREF:
1483 {
1484 struct drr_write_byref drrwbr =
1485 drr->drr_u.drr_write_byref;
1486 ra.err = restore_write_byref(&ra, os, &drrwbr);
1487 break;
1488 }
1489 case DRR_FREE:
1490 {
1491 struct drr_free drrf = drr->drr_u.drr_free;
1492 ra.err = restore_free(&ra, os, &drrf);
1493 break;
1494 }
1495 case DRR_END:
1496 {
1497 struct drr_end drre = drr->drr_u.drr_end;
1498 /*
1499 * We compare against the *previous* checksum
1500 * value, because the stored checksum is of
1501 * everything before the DRR_END record.
1502 */
1503 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1504 ra.err = ECKSUM;
1505 goto out;
1506 }
1507 case DRR_SPILL:
1508 {
1509 struct drr_spill drrs = drr->drr_u.drr_spill;
1510 ra.err = restore_spill(&ra, os, &drrs);
1511 break;
1512 }
1513 default:
1514 ra.err = EINVAL;
1515 goto out;
1516 }
1517 pcksum = ra.cksum;
1518 }
1519 ASSERT(ra.err != 0);
1520
1521 out:
1522 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1523 zfs_onexit_fd_rele(cleanup_fd);
1524
1525 if (ra.err != 0) {
1526 /*
1527 * destroy what we created, so we don't leave it in the
1528 * inconsistent restoring state.
1529 */
1530 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1531
1532 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1533 B_FALSE);
1534 if (drc->drc_real_ds != drc->drc_logical_ds) {
1535 mutex_exit(&drc->drc_logical_ds->ds_recvlock);
1536 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
1537 }
1538 }
1539
1540 kmem_free(ra.buf, ra.bufsize);
1541 *voffp = ra.voff;
1542 return (ra.err);
1543 }
1544
1545 struct recvendsyncarg {
1546 char *tosnap;
1547 uint64_t creation_time;
1548 uint64_t toguid;
1549 };
1550
1551 static int
1552 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1553 {
1554 dsl_dataset_t *ds = arg1;
1555 struct recvendsyncarg *resa = arg2;
1556
1557 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1558 }
1559
1560 static void
1561 recv_end_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1562 {
1563 dsl_dataset_t *ds = arg1;
1564 struct recvendsyncarg *resa = arg2;
1565
1566 dsl_dataset_snapshot_sync(ds, resa->tosnap, tx);
1567
1568 /* set snapshot's creation time and guid */
1569 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1570 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1571 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1572 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1573
1574 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1575 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1576 }
1577
1578 static int
1579 add_ds_to_guidmap(avl_tree_t *guid_map, dsl_dataset_t *ds)
1580 {
1581 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1582 uint64_t snapobj = ds->ds_phys->ds_prev_snap_obj;
1583 dsl_dataset_t *snapds;
1584 guid_map_entry_t *gmep;
1585 int err;
1586
1587 ASSERT(guid_map != NULL);
1588
1589 rw_enter(&dp->dp_config_rwlock, RW_READER);
1590 err = dsl_dataset_hold_obj(dp, snapobj, guid_map, &snapds);
1591 if (err == 0) {
1592 gmep = kmem_alloc(sizeof (guid_map_entry_t), KM_SLEEP);
1593 gmep->guid = snapds->ds_phys->ds_guid;
1594 gmep->gme_ds = snapds;
1595 avl_add(guid_map, gmep);
1596 }
1597
1598 rw_exit(&dp->dp_config_rwlock);
1599 return (err);
1600 }
1601
1602 static int
1603 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1604 {
1605 struct recvendsyncarg resa;
1606 dsl_dataset_t *ds = drc->drc_logical_ds;
1607 int err, myerr;
1608
1609 /*
1610 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1611 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1612 * can close it.
1613 */
1614 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1615
1616 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1617 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1618 drc->drc_force);
1619 if (err)
1620 goto out;
1621 } else {
1622 mutex_exit(&ds->ds_recvlock);
1623 dsl_dataset_rele(ds, dmu_recv_tag);
1624 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1625 B_FALSE);
1626 return (EBUSY);
1627 }
1628
1629 resa.creation_time = drc->drc_drrb->drr_creation_time;
1630 resa.toguid = drc->drc_drrb->drr_toguid;
1631 resa.tosnap = drc->drc_tosnap;
1632
1633 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1634 recv_end_check, recv_end_sync, ds, &resa, 3);
1635 if (err) {
1636 /* swap back */
1637 (void) dsl_dataset_clone_swap(drc->drc_real_ds, ds, B_TRUE);
1638 }
1639
1640 out:
1641 mutex_exit(&ds->ds_recvlock);
1642 if (err == 0 && drc->drc_guid_to_ds_map != NULL)
1643 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1644 dsl_dataset_disown(ds, dmu_recv_tag);
1645 myerr = dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, B_FALSE);
1646 ASSERT3U(myerr, ==, 0);
1647 return (err);
1648 }
1649
1650 static int
1651 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1652 {
1653 struct recvendsyncarg resa;
1654 dsl_dataset_t *ds = drc->drc_logical_ds;
1655 int err;
1656
1657 /*
1658 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1659 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1660 * can close it.
1661 */
1662 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1663
1664 resa.creation_time = drc->drc_drrb->drr_creation_time;
1665 resa.toguid = drc->drc_drrb->drr_toguid;
1666 resa.tosnap = drc->drc_tosnap;
1667
1668 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1669 recv_end_check, recv_end_sync, ds, &resa, 3);
1670 if (err) {
1671 /* clean up the fs we just recv'd into */
1672 (void) dsl_dataset_destroy(ds, dmu_recv_tag, B_FALSE);
1673 } else {
1674 if (drc->drc_guid_to_ds_map != NULL)
1675 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1676 /* release the hold from dmu_recv_begin */
1677 dsl_dataset_disown(ds, dmu_recv_tag);
1678 }
1679 return (err);
1680 }
1681
1682 int
1683 dmu_recv_end(dmu_recv_cookie_t *drc)
1684 {
1685 if (drc->drc_logical_ds != drc->drc_real_ds)
1686 return (dmu_recv_existing_end(drc));
1687 else
1688 return (dmu_recv_new_end(drc));
1689 }