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