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) 2013 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 #include <sys/dmu_send.h>
50 #include <sys/dsl_destroy.h>
51
52 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
53 int zfs_send_corrupt_data = B_FALSE;
54
55 static char *dmu_recv_tag = "dmu_recv_tag";
56 static const char *recv_clone_name = "%recv";
57
58 static int
59 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
60 {
61 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
62 ssize_t resid; /* have to get resid to get detailed errno */
63 ASSERT0(len % 8);
64
65 fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
66 dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
67 (caddr_t)buf, len,
68 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
69
70 mutex_enter(&ds->ds_sendstream_lock);
71 *dsp->dsa_off += len;
72 mutex_exit(&ds->ds_sendstream_lock);
73
74 return (dsp->dsa_err);
75 }
76
77 static int
78 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
79 uint64_t length)
80 {
81 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
82
83 if (length != -1ULL && offset + length < offset)
84 length = -1ULL;
85
86 /*
87 * If there is a pending op, but it's not PENDING_FREE, push it out,
88 * since free block aggregation can only be done for blocks of the
89 * same type (i.e., DRR_FREE records can only be aggregated with
90 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
91 * aggregated with other DRR_FREEOBJECTS records.
92 */
93 if (dsp->dsa_pending_op != PENDING_NONE &&
94 dsp->dsa_pending_op != PENDING_FREE) {
95 if (dump_bytes(dsp, dsp->dsa_drr,
96 sizeof (dmu_replay_record_t)) != 0)
97 return (SET_ERROR(EINTR));
98 dsp->dsa_pending_op = PENDING_NONE;
99 }
100
101 if (dsp->dsa_pending_op == PENDING_FREE) {
102 /*
103 * There should never be a PENDING_FREE if length is -1
104 * (because dump_dnode is the only place where this
105 * function is called with a -1, and only after flushing
106 * any pending record).
107 */
108 ASSERT(length != -1ULL);
109 /*
110 * Check to see whether this free block can be aggregated
111 * with pending one.
112 */
113 if (drrf->drr_object == object && drrf->drr_offset +
114 drrf->drr_length == offset) {
115 drrf->drr_length += length;
116 return (0);
117 } else {
118 /* not a continuation. Push out pending record */
119 if (dump_bytes(dsp, dsp->dsa_drr,
120 sizeof (dmu_replay_record_t)) != 0)
121 return (SET_ERROR(EINTR));
122 dsp->dsa_pending_op = PENDING_NONE;
123 }
124 }
125 /* create a FREE record and make it pending */
126 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
127 dsp->dsa_drr->drr_type = DRR_FREE;
128 drrf->drr_object = object;
129 drrf->drr_offset = offset;
130 drrf->drr_length = length;
131 drrf->drr_toguid = dsp->dsa_toguid;
132 if (length == -1ULL) {
133 if (dump_bytes(dsp, dsp->dsa_drr,
134 sizeof (dmu_replay_record_t)) != 0)
135 return (SET_ERROR(EINTR));
136 } else {
137 dsp->dsa_pending_op = PENDING_FREE;
138 }
139
140 return (0);
141 }
142
143 static int
144 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
145 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
146 {
147 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
148
149
150 /*
151 * If there is any kind of pending aggregation (currently either
152 * a grouping of free objects or free blocks), push it out to
153 * the stream, since aggregation can't be done across operations
154 * of different types.
155 */
156 if (dsp->dsa_pending_op != PENDING_NONE) {
157 if (dump_bytes(dsp, dsp->dsa_drr,
158 sizeof (dmu_replay_record_t)) != 0)
159 return (SET_ERROR(EINTR));
160 dsp->dsa_pending_op = PENDING_NONE;
161 }
162 /* write a DATA record */
163 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
164 dsp->dsa_drr->drr_type = DRR_WRITE;
165 drrw->drr_object = object;
166 drrw->drr_type = type;
167 drrw->drr_offset = offset;
168 drrw->drr_length = blksz;
169 drrw->drr_toguid = dsp->dsa_toguid;
170 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
171 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
172 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
173 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
174 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
175 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
176 drrw->drr_key.ddk_cksum = bp->blk_cksum;
177
178 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
179 return (SET_ERROR(EINTR));
180 if (dump_bytes(dsp, data, blksz) != 0)
181 return (SET_ERROR(EINTR));
182 return (0);
183 }
184
185 static int
186 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
187 {
188 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
189
190 if (dsp->dsa_pending_op != PENDING_NONE) {
191 if (dump_bytes(dsp, dsp->dsa_drr,
192 sizeof (dmu_replay_record_t)) != 0)
193 return (SET_ERROR(EINTR));
194 dsp->dsa_pending_op = PENDING_NONE;
195 }
196
197 /* write a SPILL record */
198 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
199 dsp->dsa_drr->drr_type = DRR_SPILL;
200 drrs->drr_object = object;
201 drrs->drr_length = blksz;
202 drrs->drr_toguid = dsp->dsa_toguid;
203
204 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
205 return (SET_ERROR(EINTR));
206 if (dump_bytes(dsp, data, blksz))
207 return (SET_ERROR(EINTR));
208 return (0);
209 }
210
211 static int
212 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
213 {
214 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
215
216 /*
217 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
218 * push it out, since free block aggregation can only be done for
219 * blocks of the same type (i.e., DRR_FREE records can only be
220 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
221 * can only be aggregated with other DRR_FREEOBJECTS records.
222 */
223 if (dsp->dsa_pending_op != PENDING_NONE &&
224 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
225 if (dump_bytes(dsp, dsp->dsa_drr,
226 sizeof (dmu_replay_record_t)) != 0)
227 return (SET_ERROR(EINTR));
228 dsp->dsa_pending_op = PENDING_NONE;
229 }
230 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
231 /*
232 * See whether this free object array can be aggregated
233 * with pending one
234 */
235 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
236 drrfo->drr_numobjs += numobjs;
237 return (0);
238 } else {
239 /* can't be aggregated. Push out pending record */
240 if (dump_bytes(dsp, dsp->dsa_drr,
241 sizeof (dmu_replay_record_t)) != 0)
242 return (SET_ERROR(EINTR));
243 dsp->dsa_pending_op = PENDING_NONE;
244 }
245 }
246
247 /* write a FREEOBJECTS record */
248 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
249 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
250 drrfo->drr_firstobj = firstobj;
251 drrfo->drr_numobjs = numobjs;
252 drrfo->drr_toguid = dsp->dsa_toguid;
253
254 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
255
256 return (0);
257 }
258
259 static int
260 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
261 {
262 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
263
264 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
265 return (dump_freeobjects(dsp, object, 1));
266
267 if (dsp->dsa_pending_op != PENDING_NONE) {
268 if (dump_bytes(dsp, dsp->dsa_drr,
269 sizeof (dmu_replay_record_t)) != 0)
270 return (SET_ERROR(EINTR));
271 dsp->dsa_pending_op = PENDING_NONE;
272 }
273
274 /* write an OBJECT record */
275 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
276 dsp->dsa_drr->drr_type = DRR_OBJECT;
277 drro->drr_object = object;
278 drro->drr_type = dnp->dn_type;
279 drro->drr_bonustype = dnp->dn_bonustype;
280 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
281 drro->drr_bonuslen = dnp->dn_bonuslen;
282 drro->drr_checksumtype = dnp->dn_checksum;
283 drro->drr_compress = dnp->dn_compress;
284 drro->drr_toguid = dsp->dsa_toguid;
285
286 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
287 return (SET_ERROR(EINTR));
288
289 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
290 return (SET_ERROR(EINTR));
291
292 /* free anything past the end of the file */
293 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
294 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
295 return (SET_ERROR(EINTR));
296 if (dsp->dsa_err != 0)
297 return (SET_ERROR(EINTR));
298 return (0);
299 }
300
301 #define BP_SPAN(dnp, level) \
302 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
303 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
304
305 /* ARGSUSED */
306 static int
307 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
308 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
309 {
310 dmu_sendarg_t *dsp = arg;
311 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
312 int err = 0;
313
314 if (issig(JUSTLOOKING) && issig(FORREAL))
315 return (SET_ERROR(EINTR));
316
317 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
318 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
319 return (0);
320 } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) {
321 uint64_t span = BP_SPAN(dnp, zb->zb_level);
322 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
323 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
324 } else if (bp == NULL) {
325 uint64_t span = BP_SPAN(dnp, zb->zb_level);
326 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
327 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
328 return (0);
329 } else if (type == DMU_OT_DNODE) {
330 dnode_phys_t *blk;
331 int i;
332 int blksz = BP_GET_LSIZE(bp);
333 uint32_t aflags = ARC_WAIT;
334 arc_buf_t *abuf;
335
336 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
337 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
338 &aflags, zb) != 0)
339 return (SET_ERROR(EIO));
340
341 blk = abuf->b_data;
342 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
343 uint64_t dnobj = (zb->zb_blkid <<
344 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
345 err = dump_dnode(dsp, dnobj, blk+i);
346 if (err != 0)
347 break;
348 }
349 (void) arc_buf_remove_ref(abuf, &abuf);
350 } else if (type == DMU_OT_SA) {
351 uint32_t aflags = ARC_WAIT;
352 arc_buf_t *abuf;
353 int blksz = BP_GET_LSIZE(bp);
354
355 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
356 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
357 &aflags, zb) != 0)
358 return (SET_ERROR(EIO));
359
360 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
361 (void) arc_buf_remove_ref(abuf, &abuf);
362 } else { /* it's a level-0 block of a regular object */
363 uint32_t aflags = ARC_WAIT;
364 arc_buf_t *abuf;
365 int blksz = BP_GET_LSIZE(bp);
366
367 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
368 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
369 &aflags, zb) != 0) {
370 if (zfs_send_corrupt_data) {
371 /* Send a block filled with 0x"zfs badd bloc" */
372 abuf = arc_buf_alloc(spa, blksz, &abuf,
373 ARC_BUFC_DATA);
374 uint64_t *ptr;
375 for (ptr = abuf->b_data;
376 (char *)ptr < (char *)abuf->b_data + blksz;
377 ptr++)
378 *ptr = 0x2f5baddb10c;
379 } else {
380 return (SET_ERROR(EIO));
381 }
382 }
383
384 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
385 blksz, bp, abuf->b_data);
386 (void) arc_buf_remove_ref(abuf, &abuf);
387 }
388
389 ASSERT(err == 0 || err == EINTR);
390 return (err);
391 }
392
393 /*
394 * Releases dp, ds, and fromds, using the specified tag.
395 */
396 static int
397 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
398 dsl_dataset_t *fromds, int outfd, vnode_t *vp, offset_t *off)
399 {
400 objset_t *os;
401 dmu_replay_record_t *drr;
402 dmu_sendarg_t *dsp;
403 int err;
404 uint64_t fromtxg = 0;
405
406 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds)) {
407 dsl_dataset_rele(fromds, tag);
408 dsl_dataset_rele(ds, tag);
409 dsl_pool_rele(dp, tag);
410 return (SET_ERROR(EXDEV));
411 }
412
413 err = dmu_objset_from_ds(ds, &os);
414 if (err != 0) {
415 if (fromds != NULL)
416 dsl_dataset_rele(fromds, tag);
417 dsl_dataset_rele(ds, tag);
418 dsl_pool_rele(dp, tag);
419 return (err);
420 }
421
422 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
423 drr->drr_type = DRR_BEGIN;
424 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
425 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
426 DMU_SUBSTREAM);
427
428 #ifdef _KERNEL
429 if (dmu_objset_type(os) == DMU_OST_ZFS) {
430 uint64_t version;
431 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
432 kmem_free(drr, sizeof (dmu_replay_record_t));
433 if (fromds != NULL)
434 dsl_dataset_rele(fromds, tag);
435 dsl_dataset_rele(ds, tag);
436 dsl_pool_rele(dp, tag);
437 return (SET_ERROR(EINVAL));
438 }
439 if (version >= ZPL_VERSION_SA) {
440 DMU_SET_FEATUREFLAGS(
441 drr->drr_u.drr_begin.drr_versioninfo,
442 DMU_BACKUP_FEATURE_SA_SPILL);
443 }
444 }
445 #endif
446
447 drr->drr_u.drr_begin.drr_creation_time =
448 ds->ds_phys->ds_creation_time;
449 drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
450 if (fromds != NULL && ds->ds_dir != fromds->ds_dir)
451 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
452 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
453 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
454 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
455
456 if (fromds != NULL)
457 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
458 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
459
460 if (fromds != NULL) {
461 fromtxg = fromds->ds_phys->ds_creation_txg;
462 dsl_dataset_rele(fromds, tag);
463 fromds = NULL;
464 }
465
466 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
467
468 dsp->dsa_drr = drr;
469 dsp->dsa_vp = vp;
470 dsp->dsa_outfd = outfd;
471 dsp->dsa_proc = curproc;
472 dsp->dsa_os = os;
473 dsp->dsa_off = off;
474 dsp->dsa_toguid = ds->ds_phys->ds_guid;
475 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
476 dsp->dsa_pending_op = PENDING_NONE;
477
478 mutex_enter(&ds->ds_sendstream_lock);
479 list_insert_head(&ds->ds_sendstreams, dsp);
480 mutex_exit(&ds->ds_sendstream_lock);
481
482 dsl_dataset_long_hold(ds, FTAG);
483 dsl_pool_rele(dp, tag);
484
485 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
486 err = dsp->dsa_err;
487 goto out;
488 }
489
490 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
491 backup_cb, dsp);
492
493 if (dsp->dsa_pending_op != PENDING_NONE)
494 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
495 err = SET_ERROR(EINTR);
496
497 if (err != 0) {
498 if (err == EINTR && dsp->dsa_err != 0)
499 err = dsp->dsa_err;
500 goto out;
501 }
502
503 bzero(drr, sizeof (dmu_replay_record_t));
504 drr->drr_type = DRR_END;
505 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
506 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
507
508 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
509 err = dsp->dsa_err;
510 goto out;
511 }
512
513 out:
514 mutex_enter(&ds->ds_sendstream_lock);
515 list_remove(&ds->ds_sendstreams, dsp);
516 mutex_exit(&ds->ds_sendstream_lock);
517
518 kmem_free(drr, sizeof (dmu_replay_record_t));
519 kmem_free(dsp, sizeof (dmu_sendarg_t));
520
521 dsl_dataset_long_rele(ds, FTAG);
522 dsl_dataset_rele(ds, tag);
523
524 return (err);
525 }
526
527 int
528 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
529 int outfd, vnode_t *vp, offset_t *off)
530 {
531 dsl_pool_t *dp;
532 dsl_dataset_t *ds;
533 dsl_dataset_t *fromds = NULL;
534 int err;
535
536 err = dsl_pool_hold(pool, FTAG, &dp);
537 if (err != 0)
538 return (err);
539
540 err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
541 if (err != 0) {
542 dsl_pool_rele(dp, FTAG);
543 return (err);
544 }
545
546 if (fromsnap != 0) {
547 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
548 if (err != 0) {
549 dsl_dataset_rele(ds, FTAG);
550 dsl_pool_rele(dp, FTAG);
551 return (err);
552 }
553 }
554
555 return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, vp, off));
556 }
557
558 int
559 dmu_send(const char *tosnap, const char *fromsnap,
560 int outfd, vnode_t *vp, offset_t *off)
561 {
562 dsl_pool_t *dp;
563 dsl_dataset_t *ds;
564 dsl_dataset_t *fromds = NULL;
565 int err;
566
567 if (strchr(tosnap, '@') == NULL)
568 return (SET_ERROR(EINVAL));
569 if (fromsnap != NULL && strchr(fromsnap, '@') == NULL)
570 return (SET_ERROR(EINVAL));
571
572 err = dsl_pool_hold(tosnap, FTAG, &dp);
573 if (err != 0)
574 return (err);
575
576 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
577 if (err != 0) {
578 dsl_pool_rele(dp, FTAG);
579 return (err);
580 }
581
582 if (fromsnap != NULL) {
583 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
584 if (err != 0) {
585 dsl_dataset_rele(ds, FTAG);
586 dsl_pool_rele(dp, FTAG);
587 return (err);
588 }
589 }
590 return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, vp, off));
591 }
592
593 int
594 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
595 {
596 dsl_pool_t *dp = ds->ds_dir->dd_pool;
597 int err;
598 uint64_t size;
599
600 ASSERT(dsl_pool_config_held(dp));
601
602 /* tosnap must be a snapshot */
603 if (!dsl_dataset_is_snapshot(ds))
604 return (SET_ERROR(EINVAL));
605
606 /*
607 * fromsnap must be an earlier snapshot from the same fs as tosnap,
608 * or the origin's fs.
609 */
610 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds))
611 return (SET_ERROR(EXDEV));
612
613 /* Get uncompressed size estimate of changed data. */
614 if (fromds == NULL) {
615 size = ds->ds_phys->ds_uncompressed_bytes;
616 } else {
617 uint64_t used, comp;
618 err = dsl_dataset_space_written(fromds, ds,
619 &used, &comp, &size);
620 if (err != 0)
621 return (err);
622 }
623
624 /*
625 * Assume that space (both on-disk and in-stream) is dominated by
626 * data. We will adjust for indirect blocks and the copies property,
627 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
628 */
629
630 /*
631 * Subtract out approximate space used by indirect blocks.
632 * Assume most space is used by data blocks (non-indirect, non-dnode).
633 * Assume all blocks are recordsize. Assume ditto blocks and
634 * internal fragmentation counter out compression.
635 *
636 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
637 * block, which we observe in practice.
638 */
639 uint64_t recordsize;
640 err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
641 if (err != 0)
642 return (err);
643 size -= size / recordsize * sizeof (blkptr_t);
644
645 /* Add in the space for the record associated with each block. */
646 size += size / recordsize * sizeof (dmu_replay_record_t);
647
648 *sizep = size;
649
650 return (0);
651 }
652
653 typedef struct dmu_recv_begin_arg {
654 const char *drba_origin;
655 dmu_recv_cookie_t *drba_cookie;
656 cred_t *drba_cred;
657 } dmu_recv_begin_arg_t;
658
659 static int
660 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
661 uint64_t fromguid)
662 {
663 uint64_t val;
664 int error;
665 dsl_pool_t *dp = ds->ds_dir->dd_pool;
666
667 /* must not have any changes since most recent snapshot */
668 if (!drba->drba_cookie->drc_force &&
669 dsl_dataset_modified_since_lastsnap(ds))
670 return (SET_ERROR(ETXTBSY));
671
672 /* temporary clone name must not exist */
673 error = zap_lookup(dp->dp_meta_objset,
674 ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name,
675 8, 1, &val);
676 if (error != ENOENT)
677 return (error == 0 ? EBUSY : error);
678
679 /* new snapshot name must not exist */
680 error = zap_lookup(dp->dp_meta_objset,
681 ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap,
682 8, 1, &val);
683 if (error != ENOENT)
684 return (error == 0 ? EEXIST : error);
685
686 if (fromguid != 0) {
687 /* if incremental, most recent snapshot must match fromguid */
688 if (ds->ds_prev == NULL)
689 return (SET_ERROR(ENODEV));
690
691 /*
692 * most recent snapshot must match fromguid, or there are no
693 * changes since the fromguid one
694 */
695 if (ds->ds_prev->ds_phys->ds_guid != fromguid) {
696 uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth;
697 uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj;
698 while (obj != 0) {
699 dsl_dataset_t *snap;
700 error = dsl_dataset_hold_obj(dp, obj, FTAG,
701 &snap);
702 if (error != 0)
703 return (SET_ERROR(ENODEV));
704 if (snap->ds_phys->ds_creation_txg < birth) {
705 dsl_dataset_rele(snap, FTAG);
706 return (SET_ERROR(ENODEV));
707 }
708 if (snap->ds_phys->ds_guid == fromguid) {
709 dsl_dataset_rele(snap, FTAG);
710 break; /* it's ok */
711 }
712 obj = snap->ds_phys->ds_prev_snap_obj;
713 dsl_dataset_rele(snap, FTAG);
714 }
715 if (obj == 0)
716 return (SET_ERROR(ENODEV));
717 }
718 } else {
719 /* if full, most recent snapshot must be $ORIGIN */
720 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
721 return (SET_ERROR(ENODEV));
722 }
723
724 return (0);
725
726 }
727
728 static int
729 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
730 {
731 dmu_recv_begin_arg_t *drba = arg;
732 dsl_pool_t *dp = dmu_tx_pool(tx);
733 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
734 uint64_t fromguid = drrb->drr_fromguid;
735 int flags = drrb->drr_flags;
736 int error;
737 dsl_dataset_t *ds;
738 const char *tofs = drba->drba_cookie->drc_tofs;
739
740 /* already checked */
741 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
742
743 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
744 DMU_COMPOUNDSTREAM ||
745 drrb->drr_type >= DMU_OST_NUMTYPES ||
746 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
747 return (SET_ERROR(EINVAL));
748
749 /* Verify pool version supports SA if SA_SPILL feature set */
750 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
751 DMU_BACKUP_FEATURE_SA_SPILL) &&
752 spa_version(dp->dp_spa) < SPA_VERSION_SA) {
753 return (SET_ERROR(ENOTSUP));
754 }
755
756 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
757 if (error == 0) {
758 /* target fs already exists; recv into temp clone */
759
760 /* Can't recv a clone into an existing fs */
761 if (flags & DRR_FLAG_CLONE) {
762 dsl_dataset_rele(ds, FTAG);
763 return (SET_ERROR(EINVAL));
764 }
765
766 error = recv_begin_check_existing_impl(drba, ds, fromguid);
767 dsl_dataset_rele(ds, FTAG);
768 } else if (error == ENOENT) {
769 /* target fs does not exist; must be a full backup or clone */
770 char buf[MAXNAMELEN];
771
772 /*
773 * If it's a non-clone incremental, we are missing the
774 * target fs, so fail the recv.
775 */
776 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
777 return (SET_ERROR(ENOENT));
778
779 /* Open the parent of tofs */
780 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
781 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
782 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
783 if (error != 0)
784 return (error);
785
786 if (drba->drba_origin != NULL) {
787 dsl_dataset_t *origin;
788 error = dsl_dataset_hold(dp, drba->drba_origin,
789 FTAG, &origin);
790 if (error != 0) {
791 dsl_dataset_rele(ds, FTAG);
792 return (error);
793 }
794 if (!dsl_dataset_is_snapshot(origin)) {
795 dsl_dataset_rele(origin, FTAG);
796 dsl_dataset_rele(ds, FTAG);
797 return (SET_ERROR(EINVAL));
798 }
799 if (origin->ds_phys->ds_guid != fromguid) {
800 dsl_dataset_rele(origin, FTAG);
801 dsl_dataset_rele(ds, FTAG);
802 return (SET_ERROR(ENODEV));
803 }
804 dsl_dataset_rele(origin, FTAG);
805 }
806 dsl_dataset_rele(ds, FTAG);
807 error = 0;
808 }
809 return (error);
810 }
811
812 static void
813 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
814 {
815 dmu_recv_begin_arg_t *drba = arg;
816 dsl_pool_t *dp = dmu_tx_pool(tx);
817 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
818 const char *tofs = drba->drba_cookie->drc_tofs;
819 dsl_dataset_t *ds, *newds;
820 uint64_t dsobj;
821 int error;
822 uint64_t crflags;
823
824 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
825 DS_FLAG_CI_DATASET : 0;
826
827 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
828 if (error == 0) {
829 /* create temporary clone */
830 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
831 ds->ds_prev, crflags, drba->drba_cred, tx);
832 dsl_dataset_rele(ds, FTAG);
833 } else {
834 dsl_dir_t *dd;
835 const char *tail;
836 dsl_dataset_t *origin = NULL;
837
838 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
839
840 if (drba->drba_origin != NULL) {
841 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
842 FTAG, &origin));
843 }
844
845 /* Create new dataset. */
846 dsobj = dsl_dataset_create_sync(dd,
847 strrchr(tofs, '/') + 1,
848 origin, crflags, drba->drba_cred, tx);
849 if (origin != NULL)
850 dsl_dataset_rele(origin, FTAG);
851 dsl_dir_rele(dd, FTAG);
852 drba->drba_cookie->drc_newfs = B_TRUE;
853 }
854 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
855
856 dmu_buf_will_dirty(newds->ds_dbuf, tx);
857 newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
858
859 /*
860 * If we actually created a non-clone, we need to create the
861 * objset in our new dataset.
862 */
863 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
864 (void) dmu_objset_create_impl(dp->dp_spa,
865 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
866 }
867
868 drba->drba_cookie->drc_ds = newds;
869
870 spa_history_log_internal_ds(newds, "receive", tx, "");
871 }
872
873 /*
874 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
875 * succeeds; otherwise we will leak the holds on the datasets.
876 */
877 int
878 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
879 boolean_t force, char *origin, dmu_recv_cookie_t *drc)
880 {
881 dmu_recv_begin_arg_t drba = { 0 };
882 dmu_replay_record_t *drr;
883
884 bzero(drc, sizeof (dmu_recv_cookie_t));
885 drc->drc_drrb = drrb;
886 drc->drc_tosnap = tosnap;
887 drc->drc_tofs = tofs;
888 drc->drc_force = force;
889
890 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
891 drc->drc_byteswap = B_TRUE;
892 else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
893 return (SET_ERROR(EINVAL));
894
895 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
896 drr->drr_type = DRR_BEGIN;
897 drr->drr_u.drr_begin = *drc->drc_drrb;
898 if (drc->drc_byteswap) {
899 fletcher_4_incremental_byteswap(drr,
900 sizeof (dmu_replay_record_t), &drc->drc_cksum);
901 } else {
902 fletcher_4_incremental_native(drr,
903 sizeof (dmu_replay_record_t), &drc->drc_cksum);
904 }
905 kmem_free(drr, sizeof (dmu_replay_record_t));
906
907 if (drc->drc_byteswap) {
908 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
909 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
910 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
911 drrb->drr_type = BSWAP_32(drrb->drr_type);
912 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
913 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
914 }
915
916 drba.drba_origin = origin;
917 drba.drba_cookie = drc;
918 drba.drba_cred = CRED();
919
920 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
921 &drba, 5));
922 }
923
924 struct restorearg {
925 int err;
926 boolean_t byteswap;
927 vnode_t *vp;
928 char *buf;
929 uint64_t voff;
930 int bufsize; /* amount of memory allocated for buf */
931 zio_cksum_t cksum;
932 avl_tree_t *guid_to_ds_map;
933 };
934
935 typedef struct guid_map_entry {
936 uint64_t guid;
937 dsl_dataset_t *gme_ds;
938 avl_node_t avlnode;
939 } guid_map_entry_t;
940
941 static int
942 guid_compare(const void *arg1, const void *arg2)
943 {
944 const guid_map_entry_t *gmep1 = arg1;
945 const guid_map_entry_t *gmep2 = arg2;
946
947 if (gmep1->guid < gmep2->guid)
948 return (-1);
949 else if (gmep1->guid > gmep2->guid)
950 return (1);
951 return (0);
952 }
953
954 static void
955 free_guid_map_onexit(void *arg)
956 {
957 avl_tree_t *ca = arg;
958 void *cookie = NULL;
959 guid_map_entry_t *gmep;
960
961 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
962 dsl_dataset_long_rele(gmep->gme_ds, gmep);
963 dsl_dataset_rele(gmep->gme_ds, gmep);
964 kmem_free(gmep, sizeof (guid_map_entry_t));
965 }
966 avl_destroy(ca);
967 kmem_free(ca, sizeof (avl_tree_t));
968 }
969
970 static void *
971 restore_read(struct restorearg *ra, int len)
972 {
973 void *rv;
974 int done = 0;
975
976 /* some things will require 8-byte alignment, so everything must */
977 ASSERT0(len % 8);
978
979 while (done < len) {
980 ssize_t resid;
981
982 ra->err = vn_rdwr(UIO_READ, ra->vp,
983 (caddr_t)ra->buf + done, len - done,
984 ra->voff, UIO_SYSSPACE, FAPPEND,
985 RLIM64_INFINITY, CRED(), &resid);
986
987 if (resid == len - done)
988 ra->err = SET_ERROR(EINVAL);
989 ra->voff += len - done - resid;
990 done = len - resid;
991 if (ra->err != 0)
992 return (NULL);
993 }
994
995 ASSERT3U(done, ==, len);
996 rv = ra->buf;
997 if (ra->byteswap)
998 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
999 else
1000 fletcher_4_incremental_native(rv, len, &ra->cksum);
1001 return (rv);
1002 }
1003
1004 static void
1005 backup_byteswap(dmu_replay_record_t *drr)
1006 {
1007 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1008 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1009 drr->drr_type = BSWAP_32(drr->drr_type);
1010 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1011 switch (drr->drr_type) {
1012 case DRR_BEGIN:
1013 DO64(drr_begin.drr_magic);
1014 DO64(drr_begin.drr_versioninfo);
1015 DO64(drr_begin.drr_creation_time);
1016 DO32(drr_begin.drr_type);
1017 DO32(drr_begin.drr_flags);
1018 DO64(drr_begin.drr_toguid);
1019 DO64(drr_begin.drr_fromguid);
1020 break;
1021 case DRR_OBJECT:
1022 DO64(drr_object.drr_object);
1023 /* DO64(drr_object.drr_allocation_txg); */
1024 DO32(drr_object.drr_type);
1025 DO32(drr_object.drr_bonustype);
1026 DO32(drr_object.drr_blksz);
1027 DO32(drr_object.drr_bonuslen);
1028 DO64(drr_object.drr_toguid);
1029 break;
1030 case DRR_FREEOBJECTS:
1031 DO64(drr_freeobjects.drr_firstobj);
1032 DO64(drr_freeobjects.drr_numobjs);
1033 DO64(drr_freeobjects.drr_toguid);
1034 break;
1035 case DRR_WRITE:
1036 DO64(drr_write.drr_object);
1037 DO32(drr_write.drr_type);
1038 DO64(drr_write.drr_offset);
1039 DO64(drr_write.drr_length);
1040 DO64(drr_write.drr_toguid);
1041 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1042 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1043 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1044 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1045 DO64(drr_write.drr_key.ddk_prop);
1046 break;
1047 case DRR_WRITE_BYREF:
1048 DO64(drr_write_byref.drr_object);
1049 DO64(drr_write_byref.drr_offset);
1050 DO64(drr_write_byref.drr_length);
1051 DO64(drr_write_byref.drr_toguid);
1052 DO64(drr_write_byref.drr_refguid);
1053 DO64(drr_write_byref.drr_refobject);
1054 DO64(drr_write_byref.drr_refoffset);
1055 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1056 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1057 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1058 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1059 DO64(drr_write_byref.drr_key.ddk_prop);
1060 break;
1061 case DRR_FREE:
1062 DO64(drr_free.drr_object);
1063 DO64(drr_free.drr_offset);
1064 DO64(drr_free.drr_length);
1065 DO64(drr_free.drr_toguid);
1066 break;
1067 case DRR_SPILL:
1068 DO64(drr_spill.drr_object);
1069 DO64(drr_spill.drr_length);
1070 DO64(drr_spill.drr_toguid);
1071 break;
1072 case DRR_END:
1073 DO64(drr_end.drr_checksum.zc_word[0]);
1074 DO64(drr_end.drr_checksum.zc_word[1]);
1075 DO64(drr_end.drr_checksum.zc_word[2]);
1076 DO64(drr_end.drr_checksum.zc_word[3]);
1077 DO64(drr_end.drr_toguid);
1078 break;
1079 }
1080 #undef DO64
1081 #undef DO32
1082 }
1083
1084 static int
1085 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1086 {
1087 int err;
1088 dmu_tx_t *tx;
1089 void *data = NULL;
1090
1091 if (drro->drr_type == DMU_OT_NONE ||
1092 !DMU_OT_IS_VALID(drro->drr_type) ||
1093 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1094 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1095 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1096 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1097 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1098 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1099 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1100 return (SET_ERROR(EINVAL));
1101 }
1102
1103 err = dmu_object_info(os, drro->drr_object, NULL);
1104
1105 if (err != 0 && err != ENOENT)
1106 return (SET_ERROR(EINVAL));
1107
1108 if (drro->drr_bonuslen) {
1109 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1110 if (ra->err != 0)
1111 return (ra->err);
1112 }
1113
1114 if (err == ENOENT) {
1115 /* currently free, want to be allocated */
1116 tx = dmu_tx_create(os);
1117 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1118 err = dmu_tx_assign(tx, TXG_WAIT);
1119 if (err != 0) {
1120 dmu_tx_abort(tx);
1121 return (err);
1122 }
1123 err = dmu_object_claim(os, drro->drr_object,
1124 drro->drr_type, drro->drr_blksz,
1125 drro->drr_bonustype, drro->drr_bonuslen, tx);
1126 dmu_tx_commit(tx);
1127 } else {
1128 /* currently allocated, want to be allocated */
1129 err = dmu_object_reclaim(os, drro->drr_object,
1130 drro->drr_type, drro->drr_blksz,
1131 drro->drr_bonustype, drro->drr_bonuslen);
1132 }
1133 if (err != 0) {
1134 return (SET_ERROR(EINVAL));
1135 }
1136
1137 tx = dmu_tx_create(os);
1138 dmu_tx_hold_bonus(tx, drro->drr_object);
1139 err = dmu_tx_assign(tx, TXG_WAIT);
1140 if (err != 0) {
1141 dmu_tx_abort(tx);
1142 return (err);
1143 }
1144
1145 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1146 tx);
1147 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1148
1149 if (data != NULL) {
1150 dmu_buf_t *db;
1151
1152 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1153 dmu_buf_will_dirty(db, tx);
1154
1155 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1156 bcopy(data, db->db_data, drro->drr_bonuslen);
1157 if (ra->byteswap) {
1158 dmu_object_byteswap_t byteswap =
1159 DMU_OT_BYTESWAP(drro->drr_bonustype);
1160 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1161 drro->drr_bonuslen);
1162 }
1163 dmu_buf_rele(db, FTAG);
1164 }
1165 dmu_tx_commit(tx);
1166 return (0);
1167 }
1168
1169 /* ARGSUSED */
1170 static int
1171 restore_freeobjects(struct restorearg *ra, objset_t *os,
1172 struct drr_freeobjects *drrfo)
1173 {
1174 uint64_t obj;
1175
1176 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1177 return (SET_ERROR(EINVAL));
1178
1179 for (obj = drrfo->drr_firstobj;
1180 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1181 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1182 int err;
1183
1184 if (dmu_object_info(os, obj, NULL) != 0)
1185 continue;
1186
1187 err = dmu_free_object(os, obj);
1188 if (err != 0)
1189 return (err);
1190 }
1191 return (0);
1192 }
1193
1194 static int
1195 restore_write(struct restorearg *ra, objset_t *os,
1196 struct drr_write *drrw)
1197 {
1198 dmu_tx_t *tx;
1199 void *data;
1200 int err;
1201
1202 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1203 !DMU_OT_IS_VALID(drrw->drr_type))
1204 return (SET_ERROR(EINVAL));
1205
1206 data = restore_read(ra, drrw->drr_length);
1207 if (data == NULL)
1208 return (ra->err);
1209
1210 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1211 return (SET_ERROR(EINVAL));
1212
1213 tx = dmu_tx_create(os);
1214
1215 dmu_tx_hold_write(tx, drrw->drr_object,
1216 drrw->drr_offset, drrw->drr_length);
1217 err = dmu_tx_assign(tx, TXG_WAIT);
1218 if (err != 0) {
1219 dmu_tx_abort(tx);
1220 return (err);
1221 }
1222 if (ra->byteswap) {
1223 dmu_object_byteswap_t byteswap =
1224 DMU_OT_BYTESWAP(drrw->drr_type);
1225 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1226 }
1227 dmu_write(os, drrw->drr_object,
1228 drrw->drr_offset, drrw->drr_length, data, tx);
1229 dmu_tx_commit(tx);
1230 return (0);
1231 }
1232
1233 /*
1234 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1235 * streams to refer to a copy of the data that is already on the
1236 * system because it came in earlier in the stream. This function
1237 * finds the earlier copy of the data, and uses that copy instead of
1238 * data from the stream to fulfill this write.
1239 */
1240 static int
1241 restore_write_byref(struct restorearg *ra, objset_t *os,
1242 struct drr_write_byref *drrwbr)
1243 {
1244 dmu_tx_t *tx;
1245 int err;
1246 guid_map_entry_t gmesrch;
1247 guid_map_entry_t *gmep;
1248 avl_index_t where;
1249 objset_t *ref_os = NULL;
1250 dmu_buf_t *dbp;
1251
1252 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1253 return (SET_ERROR(EINVAL));
1254
1255 /*
1256 * If the GUID of the referenced dataset is different from the
1257 * GUID of the target dataset, find the referenced dataset.
1258 */
1259 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1260 gmesrch.guid = drrwbr->drr_refguid;
1261 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1262 &where)) == NULL) {
1263 return (SET_ERROR(EINVAL));
1264 }
1265 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1266 return (SET_ERROR(EINVAL));
1267 } else {
1268 ref_os = os;
1269 }
1270
1271 if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1272 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1273 return (err);
1274
1275 tx = dmu_tx_create(os);
1276
1277 dmu_tx_hold_write(tx, drrwbr->drr_object,
1278 drrwbr->drr_offset, drrwbr->drr_length);
1279 err = dmu_tx_assign(tx, TXG_WAIT);
1280 if (err != 0) {
1281 dmu_tx_abort(tx);
1282 return (err);
1283 }
1284 dmu_write(os, drrwbr->drr_object,
1285 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1286 dmu_buf_rele(dbp, FTAG);
1287 dmu_tx_commit(tx);
1288 return (0);
1289 }
1290
1291 static int
1292 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1293 {
1294 dmu_tx_t *tx;
1295 void *data;
1296 dmu_buf_t *db, *db_spill;
1297 int err;
1298
1299 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1300 drrs->drr_length > SPA_MAXBLOCKSIZE)
1301 return (SET_ERROR(EINVAL));
1302
1303 data = restore_read(ra, drrs->drr_length);
1304 if (data == NULL)
1305 return (ra->err);
1306
1307 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1308 return (SET_ERROR(EINVAL));
1309
1310 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1311 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1312 dmu_buf_rele(db, FTAG);
1313 return (err);
1314 }
1315
1316 tx = dmu_tx_create(os);
1317
1318 dmu_tx_hold_spill(tx, db->db_object);
1319
1320 err = dmu_tx_assign(tx, TXG_WAIT);
1321 if (err != 0) {
1322 dmu_buf_rele(db, FTAG);
1323 dmu_buf_rele(db_spill, FTAG);
1324 dmu_tx_abort(tx);
1325 return (err);
1326 }
1327 dmu_buf_will_dirty(db_spill, tx);
1328
1329 if (db_spill->db_size < drrs->drr_length)
1330 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1331 drrs->drr_length, tx));
1332 bcopy(data, db_spill->db_data, drrs->drr_length);
1333
1334 dmu_buf_rele(db, FTAG);
1335 dmu_buf_rele(db_spill, FTAG);
1336
1337 dmu_tx_commit(tx);
1338 return (0);
1339 }
1340
1341 /* ARGSUSED */
1342 static int
1343 restore_free(struct restorearg *ra, objset_t *os,
1344 struct drr_free *drrf)
1345 {
1346 int err;
1347
1348 if (drrf->drr_length != -1ULL &&
1349 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1350 return (SET_ERROR(EINVAL));
1351
1352 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1353 return (SET_ERROR(EINVAL));
1354
1355 err = dmu_free_long_range(os, drrf->drr_object,
1356 drrf->drr_offset, drrf->drr_length);
1357 return (err);
1358 }
1359
1360 /* used to destroy the drc_ds on error */
1361 static void
1362 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1363 {
1364 char name[MAXNAMELEN];
1365 dsl_dataset_name(drc->drc_ds, name);
1366 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1367 (void) dsl_destroy_head(name);
1368 }
1369
1370 /*
1371 * NB: callers *must* call dmu_recv_end() if this succeeds.
1372 */
1373 int
1374 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1375 int cleanup_fd, uint64_t *action_handlep)
1376 {
1377 struct restorearg ra = { 0 };
1378 dmu_replay_record_t *drr;
1379 objset_t *os;
1380 zio_cksum_t pcksum;
1381 int featureflags;
1382
1383 ra.byteswap = drc->drc_byteswap;
1384 ra.cksum = drc->drc_cksum;
1385 ra.vp = vp;
1386 ra.voff = *voffp;
1387 ra.bufsize = 1<<20;
1388 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1389
1390 /* these were verified in dmu_recv_begin */
1391 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1392 DMU_SUBSTREAM);
1393 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1394
1395 /*
1396 * Open the objset we are modifying.
1397 */
1398 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1399
1400 ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1401
1402 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1403
1404 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1405 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1406 minor_t minor;
1407
1408 if (cleanup_fd == -1) {
1409 ra.err = SET_ERROR(EBADF);
1410 goto out;
1411 }
1412 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1413 if (ra.err != 0) {
1414 cleanup_fd = -1;
1415 goto out;
1416 }
1417
1418 if (*action_handlep == 0) {
1419 ra.guid_to_ds_map =
1420 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1421 avl_create(ra.guid_to_ds_map, guid_compare,
1422 sizeof (guid_map_entry_t),
1423 offsetof(guid_map_entry_t, avlnode));
1424 ra.err = zfs_onexit_add_cb(minor,
1425 free_guid_map_onexit, ra.guid_to_ds_map,
1426 action_handlep);
1427 if (ra.err != 0)
1428 goto out;
1429 } else {
1430 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1431 (void **)&ra.guid_to_ds_map);
1432 if (ra.err != 0)
1433 goto out;
1434 }
1435
1436 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1437 }
1438
1439 /*
1440 * Read records and process them.
1441 */
1442 pcksum = ra.cksum;
1443 while (ra.err == 0 &&
1444 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1445 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1446 ra.err = SET_ERROR(EINTR);
1447 goto out;
1448 }
1449
1450 if (ra.byteswap)
1451 backup_byteswap(drr);
1452
1453 switch (drr->drr_type) {
1454 case DRR_OBJECT:
1455 {
1456 /*
1457 * We need to make a copy of the record header,
1458 * because restore_{object,write} may need to
1459 * restore_read(), which will invalidate drr.
1460 */
1461 struct drr_object drro = drr->drr_u.drr_object;
1462 ra.err = restore_object(&ra, os, &drro);
1463 break;
1464 }
1465 case DRR_FREEOBJECTS:
1466 {
1467 struct drr_freeobjects drrfo =
1468 drr->drr_u.drr_freeobjects;
1469 ra.err = restore_freeobjects(&ra, os, &drrfo);
1470 break;
1471 }
1472 case DRR_WRITE:
1473 {
1474 struct drr_write drrw = drr->drr_u.drr_write;
1475 ra.err = restore_write(&ra, os, &drrw);
1476 break;
1477 }
1478 case DRR_WRITE_BYREF:
1479 {
1480 struct drr_write_byref drrwbr =
1481 drr->drr_u.drr_write_byref;
1482 ra.err = restore_write_byref(&ra, os, &drrwbr);
1483 break;
1484 }
1485 case DRR_FREE:
1486 {
1487 struct drr_free drrf = drr->drr_u.drr_free;
1488 ra.err = restore_free(&ra, os, &drrf);
1489 break;
1490 }
1491 case DRR_END:
1492 {
1493 struct drr_end drre = drr->drr_u.drr_end;
1494 /*
1495 * We compare against the *previous* checksum
1496 * value, because the stored checksum is of
1497 * everything before the DRR_END record.
1498 */
1499 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1500 ra.err = SET_ERROR(ECKSUM);
1501 goto out;
1502 }
1503 case DRR_SPILL:
1504 {
1505 struct drr_spill drrs = drr->drr_u.drr_spill;
1506 ra.err = restore_spill(&ra, os, &drrs);
1507 break;
1508 }
1509 default:
1510 ra.err = SET_ERROR(EINVAL);
1511 goto out;
1512 }
1513 pcksum = ra.cksum;
1514 }
1515 ASSERT(ra.err != 0);
1516
1517 out:
1518 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1519 zfs_onexit_fd_rele(cleanup_fd);
1520
1521 if (ra.err != 0) {
1522 /*
1523 * destroy what we created, so we don't leave it in the
1524 * inconsistent restoring state.
1525 */
1526 dmu_recv_cleanup_ds(drc);
1527 }
1528
1529 kmem_free(ra.buf, ra.bufsize);
1530 *voffp = ra.voff;
1531 return (ra.err);
1532 }
1533
1534 static int
1535 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1536 {
1537 dmu_recv_cookie_t *drc = arg;
1538 dsl_pool_t *dp = dmu_tx_pool(tx);
1539 int error;
1540
1541 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1542
1543 if (!drc->drc_newfs) {
1544 dsl_dataset_t *origin_head;
1545
1546 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1547 if (error != 0)
1548 return (error);
1549 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1550 origin_head, drc->drc_force, drc->drc_owner, tx);
1551 if (error != 0) {
1552 dsl_dataset_rele(origin_head, FTAG);
1553 return (error);
1554 }
1555 error = dsl_dataset_snapshot_check_impl(origin_head,
1556 drc->drc_tosnap, tx, B_TRUE);
1557 dsl_dataset_rele(origin_head, FTAG);
1558 if (error != 0)
1559 return (error);
1560
1561 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1562 } else {
1563 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1564 drc->drc_tosnap, tx, B_TRUE);
1565 }
1566 return (error);
1567 }
1568
1569 static void
1570 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1571 {
1572 dmu_recv_cookie_t *drc = arg;
1573 dsl_pool_t *dp = dmu_tx_pool(tx);
1574
1575 spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1576 tx, "snap=%s", drc->drc_tosnap);
1577
1578 if (!drc->drc_newfs) {
1579 dsl_dataset_t *origin_head;
1580
1581 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1582 &origin_head));
1583 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1584 origin_head, tx);
1585 dsl_dataset_snapshot_sync_impl(origin_head,
1586 drc->drc_tosnap, tx);
1587
1588 /* set snapshot's creation time and guid */
1589 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1590 origin_head->ds_prev->ds_phys->ds_creation_time =
1591 drc->drc_drrb->drr_creation_time;
1592 origin_head->ds_prev->ds_phys->ds_guid =
1593 drc->drc_drrb->drr_toguid;
1594 origin_head->ds_prev->ds_phys->ds_flags &=
1595 ~DS_FLAG_INCONSISTENT;
1596
1597 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1598 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1599
1600 dsl_dataset_rele(origin_head, FTAG);
1601 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1602
1603 if (drc->drc_owner != NULL)
1604 VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
1605 } else {
1606 dsl_dataset_t *ds = drc->drc_ds;
1607
1608 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1609
1610 /* set snapshot's creation time and guid */
1611 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1612 ds->ds_prev->ds_phys->ds_creation_time =
1613 drc->drc_drrb->drr_creation_time;
1614 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1615 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1616
1617 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1618 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1619 }
1620 drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1621 /*
1622 * Release the hold from dmu_recv_begin. This must be done before
1623 * we return to open context, so that when we free the dataset's dnode,
1624 * we can evict its bonus buffer.
1625 */
1626 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1627 drc->drc_ds = NULL;
1628 }
1629
1630 static int
1631 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
1632 {
1633 dsl_pool_t *dp;
1634 dsl_dataset_t *snapds;
1635 guid_map_entry_t *gmep;
1636 int err;
1637
1638 ASSERT(guid_map != NULL);
1639
1640 err = dsl_pool_hold(name, FTAG, &dp);
1641 if (err != 0)
1642 return (err);
1643 gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
1644 err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
1645 if (err == 0) {
1646 gmep->guid = snapds->ds_phys->ds_guid;
1647 gmep->gme_ds = snapds;
1648 avl_add(guid_map, gmep);
1649 dsl_dataset_long_hold(snapds, gmep);
1650 } else {
1651 kmem_free(gmep, sizeof (*gmep));
1652 }
1653
1654 dsl_pool_rele(dp, FTAG);
1655 return (err);
1656 }
1657
1658 static int dmu_recv_end_modified_blocks = 3;
1659
1660 static int
1661 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1662 {
1663 int error;
1664 char name[MAXNAMELEN];
1665
1666 #ifdef _KERNEL
1667 /*
1668 * We will be destroying the ds; make sure its origin is unmounted if
1669 * necessary.
1670 */
1671 dsl_dataset_name(drc->drc_ds, name);
1672 zfs_destroy_unmount_origin(name);
1673 #endif
1674
1675 error = dsl_sync_task(drc->drc_tofs,
1676 dmu_recv_end_check, dmu_recv_end_sync, drc,
1677 dmu_recv_end_modified_blocks);
1678
1679 if (error != 0)
1680 dmu_recv_cleanup_ds(drc);
1681 return (error);
1682 }
1683
1684 static int
1685 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1686 {
1687 int error;
1688
1689 error = dsl_sync_task(drc->drc_tofs,
1690 dmu_recv_end_check, dmu_recv_end_sync, drc,
1691 dmu_recv_end_modified_blocks);
1692
1693 if (error != 0) {
1694 dmu_recv_cleanup_ds(drc);
1695 } else if (drc->drc_guid_to_ds_map != NULL) {
1696 (void) add_ds_to_guidmap(drc->drc_tofs,
1697 drc->drc_guid_to_ds_map,
1698 drc->drc_newsnapobj);
1699 }
1700 return (error);
1701 }
1702
1703 int
1704 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
1705 {
1706 drc->drc_owner = owner;
1707
1708 if (drc->drc_newfs)
1709 return (dmu_recv_new_end(drc));
1710 else
1711 return (dmu_recv_existing_end(drc));
1712 }