]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/dmu_send.c
projects / mirror_zfs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Illumos 5027 - zfs large block support
[mirror_zfs.git] / module / zfs / dmu_send.c
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
27 */
28
29 #include <sys/dmu.h>
30 #include <sys/dmu_impl.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/dbuf.h>
33 #include <sys/dnode.h>
34 #include <sys/zfs_context.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/dsl_dir.h>
39 #include <sys/dsl_prop.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_synctask.h>
42 #include <sys/spa_impl.h>
43 #include <sys/zfs_ioctl.h>
44 #include <sys/zap.h>
45 #include <sys/zio_checksum.h>
46 #include <sys/zfs_znode.h>
47 #include <zfs_fletcher.h>
48 #include <sys/avl.h>
49 #include <sys/ddt.h>
50 #include <sys/zfs_onexit.h>
51 #include <sys/dmu_send.h>
52 #include <sys/dsl_destroy.h>
53 #include <sys/blkptr.h>
54 #include <sys/dsl_bookmark.h>
55 #include <sys/zfeature.h>
56
57 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
58 int zfs_send_corrupt_data = B_FALSE;
59
60 static char *dmu_recv_tag = "dmu_recv_tag";
61 static const char *recv_clone_name = "%recv";
62
63 typedef struct dump_bytes_io {
64 dmu_sendarg_t *dbi_dsp;
65 void *dbi_buf;
66 int dbi_len;
67 } dump_bytes_io_t;
68
69 static void
70 dump_bytes_strategy(void *arg)
71 {
72 dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg;
73 dmu_sendarg_t *dsp = dbi->dbi_dsp;
74 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
75 ssize_t resid; /* have to get resid to get detailed errno */
76 ASSERT0(dbi->dbi_len % 8);
77
78 fletcher_4_incremental_native(dbi->dbi_buf, dbi->dbi_len, &dsp->dsa_zc);
79 dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
80 (caddr_t)dbi->dbi_buf, dbi->dbi_len,
81 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
82
83 mutex_enter(&ds->ds_sendstream_lock);
84 *dsp->dsa_off += dbi->dbi_len;
85 mutex_exit(&ds->ds_sendstream_lock);
86 }
87
88 static int
89 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
90 {
91 dump_bytes_io_t dbi;
92
93 dbi.dbi_dsp = dsp;
94 dbi.dbi_buf = buf;
95 dbi.dbi_len = len;
96
97 /*
98 * The vn_rdwr() call is performed in a taskq to ensure that there is
99 * always enough stack space to write safely to the target filesystem.
100 * The ZIO_TYPE_FREE threads are used because there can be a lot of
101 * them and they are used in vdev_file.c for a similar purpose.
102 */
103 spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE,
104 ZIO_TASKQ_ISSUE, dump_bytes_strategy, &dbi, TQ_SLEEP);
105
106 return (dsp->dsa_err);
107 }
108
109 static int
110 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
111 uint64_t length)
112 {
113 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
114
115 /*
116 * When we receive a free record, dbuf_free_range() assumes
117 * that the receiving system doesn't have any dbufs in the range
118 * being freed. This is always true because there is a one-record
119 * constraint: we only send one WRITE record for any given
120 * object+offset. We know that the one-record constraint is
121 * true because we always send data in increasing order by
122 * object,offset.
123 *
124 * If the increasing-order constraint ever changes, we should find
125 * another way to assert that the one-record constraint is still
126 * satisfied.
127 */
128 ASSERT(object > dsp->dsa_last_data_object ||
129 (object == dsp->dsa_last_data_object &&
130 offset > dsp->dsa_last_data_offset));
131
132 /*
133 * If we are doing a non-incremental send, then there can't
134 * be any data in the dataset we're receiving into. Therefore
135 * a free record would simply be a no-op. Save space by not
136 * sending it to begin with.
137 */
138 if (!dsp->dsa_incremental)
139 return (0);
140
141 if (length != -1ULL && offset + length < offset)
142 length = -1ULL;
143
144 /*
145 * If there is a pending op, but it's not PENDING_FREE, push it out,
146 * since free block aggregation can only be done for blocks of the
147 * same type (i.e., DRR_FREE records can only be aggregated with
148 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
149 * aggregated with other DRR_FREEOBJECTS records.
150 */
151 if (dsp->dsa_pending_op != PENDING_NONE &&
152 dsp->dsa_pending_op != PENDING_FREE) {
153 if (dump_bytes(dsp, dsp->dsa_drr,
154 sizeof (dmu_replay_record_t)) != 0)
155 return (SET_ERROR(EINTR));
156 dsp->dsa_pending_op = PENDING_NONE;
157 }
158
159 if (dsp->dsa_pending_op == PENDING_FREE) {
160 /*
161 * There should never be a PENDING_FREE if length is -1
162 * (because dump_dnode is the only place where this
163 * function is called with a -1, and only after flushing
164 * any pending record).
165 */
166 ASSERT(length != -1ULL);
167 /*
168 * Check to see whether this free block can be aggregated
169 * with pending one.
170 */
171 if (drrf->drr_object == object && drrf->drr_offset +
172 drrf->drr_length == offset) {
173 drrf->drr_length += length;
174 return (0);
175 } else {
176 /* not a continuation. Push out pending record */
177 if (dump_bytes(dsp, dsp->dsa_drr,
178 sizeof (dmu_replay_record_t)) != 0)
179 return (SET_ERROR(EINTR));
180 dsp->dsa_pending_op = PENDING_NONE;
181 }
182 }
183 /* create a FREE record and make it pending */
184 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
185 dsp->dsa_drr->drr_type = DRR_FREE;
186 drrf->drr_object = object;
187 drrf->drr_offset = offset;
188 drrf->drr_length = length;
189 drrf->drr_toguid = dsp->dsa_toguid;
190 if (length == -1ULL) {
191 if (dump_bytes(dsp, dsp->dsa_drr,
192 sizeof (dmu_replay_record_t)) != 0)
193 return (SET_ERROR(EINTR));
194 } else {
195 dsp->dsa_pending_op = PENDING_FREE;
196 }
197
198 return (0);
199 }
200
201 static int
202 dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type,
203 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
204 {
205 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
206
207 /*
208 * We send data in increasing object, offset order.
209 * See comment in dump_free() for details.
210 */
211 ASSERT(object > dsp->dsa_last_data_object ||
212 (object == dsp->dsa_last_data_object &&
213 offset > dsp->dsa_last_data_offset));
214 dsp->dsa_last_data_object = object;
215 dsp->dsa_last_data_offset = offset + blksz - 1;
216
217 /*
218 * If there is any kind of pending aggregation (currently either
219 * a grouping of free objects or free blocks), push it out to
220 * the stream, since aggregation can't be done across operations
221 * of different types.
222 */
223 if (dsp->dsa_pending_op != PENDING_NONE) {
224 if (dump_bytes(dsp, dsp->dsa_drr,
225 sizeof (dmu_replay_record_t)) != 0)
226 return (SET_ERROR(EINTR));
227 dsp->dsa_pending_op = PENDING_NONE;
228 }
229 /* write a DATA record */
230 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
231 dsp->dsa_drr->drr_type = DRR_WRITE;
232 drrw->drr_object = object;
233 drrw->drr_type = type;
234 drrw->drr_offset = offset;
235 drrw->drr_length = blksz;
236 drrw->drr_toguid = dsp->dsa_toguid;
237 if (bp == NULL || BP_IS_EMBEDDED(bp)) {
238 /*
239 * There's no pre-computed checksum for partial-block
240 * writes or embedded BP's, so (like
241 * fletcher4-checkummed blocks) userland will have to
242 * compute a dedup-capable checksum itself.
243 */
244 drrw->drr_checksumtype = ZIO_CHECKSUM_OFF;
245 } else {
246 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
247 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
248 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
249 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
250 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
251 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
252 drrw->drr_key.ddk_cksum = bp->blk_cksum;
253 }
254
255 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
256 return (SET_ERROR(EINTR));
257 if (dump_bytes(dsp, data, blksz) != 0)
258 return (SET_ERROR(EINTR));
259 return (0);
260 }
261
262 static int
263 dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
264 int blksz, const blkptr_t *bp)
265 {
266 char buf[BPE_PAYLOAD_SIZE];
267 struct drr_write_embedded *drrw =
268 &(dsp->dsa_drr->drr_u.drr_write_embedded);
269
270 if (dsp->dsa_pending_op != PENDING_NONE) {
271 if (dump_bytes(dsp, dsp->dsa_drr,
272 sizeof (dmu_replay_record_t)) != 0)
273 return (EINTR);
274 dsp->dsa_pending_op = PENDING_NONE;
275 }
276
277 ASSERT(BP_IS_EMBEDDED(bp));
278
279 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
280 dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED;
281 drrw->drr_object = object;
282 drrw->drr_offset = offset;
283 drrw->drr_length = blksz;
284 drrw->drr_toguid = dsp->dsa_toguid;
285 drrw->drr_compression = BP_GET_COMPRESS(bp);
286 drrw->drr_etype = BPE_GET_ETYPE(bp);
287 drrw->drr_lsize = BPE_GET_LSIZE(bp);
288 drrw->drr_psize = BPE_GET_PSIZE(bp);
289
290 decode_embedded_bp_compressed(bp, buf);
291
292 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
293 return (EINTR);
294 if (dump_bytes(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0)
295 return (EINTR);
296 return (0);
297 }
298
299 static int
300 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
301 {
302 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
303
304 if (dsp->dsa_pending_op != PENDING_NONE) {
305 if (dump_bytes(dsp, dsp->dsa_drr,
306 sizeof (dmu_replay_record_t)) != 0)
307 return (SET_ERROR(EINTR));
308 dsp->dsa_pending_op = PENDING_NONE;
309 }
310
311 /* write a SPILL record */
312 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
313 dsp->dsa_drr->drr_type = DRR_SPILL;
314 drrs->drr_object = object;
315 drrs->drr_length = blksz;
316 drrs->drr_toguid = dsp->dsa_toguid;
317
318 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
319 return (SET_ERROR(EINTR));
320 if (dump_bytes(dsp, data, blksz))
321 return (SET_ERROR(EINTR));
322 return (0);
323 }
324
325 static int
326 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
327 {
328 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
329
330 /* See comment in dump_free(). */
331 if (!dsp->dsa_incremental)
332 return (0);
333
334 /*
335 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
336 * push it out, since free block aggregation can only be done for
337 * blocks of the same type (i.e., DRR_FREE records can only be
338 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
339 * can only be aggregated with other DRR_FREEOBJECTS records.
340 */
341 if (dsp->dsa_pending_op != PENDING_NONE &&
342 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
343 if (dump_bytes(dsp, dsp->dsa_drr,
344 sizeof (dmu_replay_record_t)) != 0)
345 return (SET_ERROR(EINTR));
346 dsp->dsa_pending_op = PENDING_NONE;
347 }
348 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
349 /*
350 * See whether this free object array can be aggregated
351 * with pending one
352 */
353 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
354 drrfo->drr_numobjs += numobjs;
355 return (0);
356 } else {
357 /* can't be aggregated. Push out pending record */
358 if (dump_bytes(dsp, dsp->dsa_drr,
359 sizeof (dmu_replay_record_t)) != 0)
360 return (SET_ERROR(EINTR));
361 dsp->dsa_pending_op = PENDING_NONE;
362 }
363 }
364
365 /* write a FREEOBJECTS record */
366 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
367 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
368 drrfo->drr_firstobj = firstobj;
369 drrfo->drr_numobjs = numobjs;
370 drrfo->drr_toguid = dsp->dsa_toguid;
371
372 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
373
374 return (0);
375 }
376
377 static int
378 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
379 {
380 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
381
382 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
383 return (dump_freeobjects(dsp, object, 1));
384
385 if (dsp->dsa_pending_op != PENDING_NONE) {
386 if (dump_bytes(dsp, dsp->dsa_drr,
387 sizeof (dmu_replay_record_t)) != 0)
388 return (SET_ERROR(EINTR));
389 dsp->dsa_pending_op = PENDING_NONE;
390 }
391
392 /* write an OBJECT record */
393 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
394 dsp->dsa_drr->drr_type = DRR_OBJECT;
395 drro->drr_object = object;
396 drro->drr_type = dnp->dn_type;
397 drro->drr_bonustype = dnp->dn_bonustype;
398 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
399 drro->drr_bonuslen = dnp->dn_bonuslen;
400 drro->drr_checksumtype = dnp->dn_checksum;
401 drro->drr_compress = dnp->dn_compress;
402 drro->drr_toguid = dsp->dsa_toguid;
403
404 if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
405 drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE)
406 drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE;
407
408 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
409 return (SET_ERROR(EINTR));
410
411 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
412 return (SET_ERROR(EINTR));
413
414 /* Free anything past the end of the file. */
415 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
416 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0)
417 return (SET_ERROR(EINTR));
418 if (dsp->dsa_err != 0)
419 return (SET_ERROR(EINTR));
420 return (0);
421 }
422
423 static boolean_t
424 backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp)
425 {
426 if (!BP_IS_EMBEDDED(bp))
427 return (B_FALSE);
428
429 /*
430 * Compression function must be legacy, or explicitly enabled.
431 */
432 if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS &&
433 !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4)))
434 return (B_FALSE);
435
436 /*
437 * Embed type must be explicitly enabled.
438 */
439 switch (BPE_GET_ETYPE(bp)) {
440 case BP_EMBEDDED_TYPE_DATA:
441 if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)
442 return (B_TRUE);
443 break;
444 default:
445 return (B_FALSE);
446 }
447 return (B_FALSE);
448 }
449
450 #define BP_SPAN(dnp, level) \
451 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
452 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
453
454 /* ARGSUSED */
455 static int
456 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
457 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
458 {
459 dmu_sendarg_t *dsp = arg;
460 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
461 int err = 0;
462
463 if (issig(JUSTLOOKING) && issig(FORREAL))
464 return (SET_ERROR(EINTR));
465
466 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
467 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
468 return (0);
469 } else if (zb->zb_level == ZB_ZIL_LEVEL) {
470 /*
471 * If we are sending a non-snapshot (which is allowed on
472 * read-only pools), it may have a ZIL, which must be ignored.
473 */
474 return (0);
475 } else if (BP_IS_HOLE(bp) &&
476 zb->zb_object == DMU_META_DNODE_OBJECT) {
477 uint64_t span = BP_SPAN(dnp, zb->zb_level);
478 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
479 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
480 } else if (BP_IS_HOLE(bp)) {
481 uint64_t span = BP_SPAN(dnp, zb->zb_level);
482 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
483 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
484 return (0);
485 } else if (type == DMU_OT_DNODE) {
486 dnode_phys_t *blk;
487 int i;
488 int blksz = BP_GET_LSIZE(bp);
489 uint32_t aflags = ARC_WAIT;
490 arc_buf_t *abuf;
491
492 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
493 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
494 &aflags, zb) != 0)
495 return (SET_ERROR(EIO));
496
497 blk = abuf->b_data;
498 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
499 uint64_t dnobj = (zb->zb_blkid <<
500 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
501 err = dump_dnode(dsp, dnobj, blk+i);
502 if (err != 0)
503 break;
504 }
505 (void) arc_buf_remove_ref(abuf, &abuf);
506 } else if (type == DMU_OT_SA) {
507 uint32_t aflags = ARC_WAIT;
508 arc_buf_t *abuf;
509 int blksz = BP_GET_LSIZE(bp);
510
511 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
512 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
513 &aflags, zb) != 0)
514 return (SET_ERROR(EIO));
515
516 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
517 (void) arc_buf_remove_ref(abuf, &abuf);
518 } else if (backup_do_embed(dsp, bp)) {
519 /* it's an embedded level-0 block of a regular object */
520 int blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
521 err = dump_write_embedded(dsp, zb->zb_object,
522 zb->zb_blkid * blksz, blksz, bp);
523 } else { /* it's a level-0 block of a regular object */
524 uint32_t aflags = ARC_WAIT;
525 uint64_t offset;
526 arc_buf_t *abuf;
527 int blksz = BP_GET_LSIZE(bp);
528
529 ASSERT3U(blksz, ==, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
530 ASSERT0(zb->zb_level);
531 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
532 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
533 &aflags, zb) != 0) {
534 if (zfs_send_corrupt_data) {
535 uint64_t *ptr;
536 /* Send a block filled with 0x"zfs badd bloc" */
537 abuf = arc_buf_alloc(spa, blksz, &abuf,
538 ARC_BUFC_DATA);
539 for (ptr = abuf->b_data;
540 (char *)ptr < (char *)abuf->b_data + blksz;
541 ptr++)
542 *ptr = 0x2f5baddb10cULL;
543 } else {
544 return (SET_ERROR(EIO));
545 }
546 }
547
548 offset = zb->zb_blkid * blksz;
549
550 if (!(dsp->dsa_featureflags &
551 DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
552 blksz > SPA_OLD_MAXBLOCKSIZE) {
553 char *buf = abuf->b_data;
554 while (blksz > 0 && err == 0) {
555 int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE);
556 err = dump_write(dsp, type, zb->zb_object,
557 offset, n, NULL, buf);
558 offset += n;
559 buf += n;
560 blksz -= n;
561 }
562 } else {
563 err = dump_write(dsp, type, zb->zb_object,
564 offset, blksz, bp, abuf->b_data);
565 }
566 (void) arc_buf_remove_ref(abuf, &abuf);
567 }
568
569 ASSERT(err == 0 || err == EINTR);
570 return (err);
571 }
572
573 /*
574 * Releases dp using the specified tag.
575 */
576 static int
577 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
578 zfs_bookmark_phys_t *fromzb, boolean_t is_clone, boolean_t embedok,
579 boolean_t large_block_ok, int outfd, vnode_t *vp, offset_t *off)
580 {
581 objset_t *os;
582 dmu_replay_record_t *drr;
583 dmu_sendarg_t *dsp;
584 int err;
585 uint64_t fromtxg = 0;
586 uint64_t featureflags = 0;
587
588 err = dmu_objset_from_ds(ds, &os);
589 if (err != 0) {
590 dsl_pool_rele(dp, tag);
591 return (err);
592 }
593
594 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
595 drr->drr_type = DRR_BEGIN;
596 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
597 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
598 DMU_SUBSTREAM);
599
600 #ifdef _KERNEL
601 if (dmu_objset_type(os) == DMU_OST_ZFS) {
602 uint64_t version;
603 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
604 kmem_free(drr, sizeof (dmu_replay_record_t));
605 dsl_pool_rele(dp, tag);
606 return (SET_ERROR(EINVAL));
607 }
608 if (version >= ZPL_VERSION_SA) {
609 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
610 }
611 }
612 #endif
613
614 if (large_block_ok && ds->ds_large_blocks)
615 featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS;
616 if (embedok &&
617 spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) {
618 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA;
619 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
620 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA_LZ4;
621 } else {
622 embedok = B_FALSE;
623 }
624
625 DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo,
626 featureflags);
627
628 drr->drr_u.drr_begin.drr_creation_time =
629 dsl_dataset_phys(ds)->ds_creation_time;
630 drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
631 if (is_clone)
632 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
633 drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(ds)->ds_guid;
634 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
635 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
636
637 if (fromzb != NULL) {
638 drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid;
639 fromtxg = fromzb->zbm_creation_txg;
640 }
641 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
642 if (!ds->ds_is_snapshot) {
643 (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--",
644 sizeof (drr->drr_u.drr_begin.drr_toname));
645 }
646
647 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
648
649 dsp->dsa_drr = drr;
650 dsp->dsa_vp = vp;
651 dsp->dsa_outfd = outfd;
652 dsp->dsa_proc = curproc;
653 dsp->dsa_os = os;
654 dsp->dsa_off = off;
655 dsp->dsa_toguid = dsl_dataset_phys(ds)->ds_guid;
656 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
657 dsp->dsa_pending_op = PENDING_NONE;
658 dsp->dsa_incremental = (fromzb != NULL);
659 dsp->dsa_featureflags = featureflags;
660
661 mutex_enter(&ds->ds_sendstream_lock);
662 list_insert_head(&ds->ds_sendstreams, dsp);
663 mutex_exit(&ds->ds_sendstream_lock);
664
665 dsl_dataset_long_hold(ds, FTAG);
666 dsl_pool_rele(dp, tag);
667
668 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
669 err = dsp->dsa_err;
670 goto out;
671 }
672
673 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
674 backup_cb, dsp);
675
676 if (dsp->dsa_pending_op != PENDING_NONE)
677 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
678 err = SET_ERROR(EINTR);
679
680 if (err != 0) {
681 if (err == EINTR && dsp->dsa_err != 0)
682 err = dsp->dsa_err;
683 goto out;
684 }
685
686 bzero(drr, sizeof (dmu_replay_record_t));
687 drr->drr_type = DRR_END;
688 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
689 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
690
691 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
692 err = dsp->dsa_err;
693 goto out;
694 }
695
696 out:
697 mutex_enter(&ds->ds_sendstream_lock);
698 list_remove(&ds->ds_sendstreams, dsp);
699 mutex_exit(&ds->ds_sendstream_lock);
700
701 kmem_free(drr, sizeof (dmu_replay_record_t));
702 kmem_free(dsp, sizeof (dmu_sendarg_t));
703
704 dsl_dataset_long_rele(ds, FTAG);
705
706 return (err);
707 }
708
709 int
710 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
711 boolean_t embedok, boolean_t large_block_ok,
712 int outfd, vnode_t *vp, offset_t *off)
713 {
714 dsl_pool_t *dp;
715 dsl_dataset_t *ds;
716 dsl_dataset_t *fromds = NULL;
717 int err;
718
719 err = dsl_pool_hold(pool, FTAG, &dp);
720 if (err != 0)
721 return (err);
722
723 err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
724 if (err != 0) {
725 dsl_pool_rele(dp, FTAG);
726 return (err);
727 }
728
729 if (fromsnap != 0) {
730 zfs_bookmark_phys_t zb;
731 boolean_t is_clone;
732
733 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
734 if (err != 0) {
735 dsl_dataset_rele(ds, FTAG);
736 dsl_pool_rele(dp, FTAG);
737 return (err);
738 }
739 if (!dsl_dataset_is_before(ds, fromds, 0))
740 err = SET_ERROR(EXDEV);
741 zb.zbm_creation_time =
742 dsl_dataset_phys(fromds)->ds_creation_time;
743 zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg;
744 zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid;
745 is_clone = (fromds->ds_dir != ds->ds_dir);
746 dsl_dataset_rele(fromds, FTAG);
747 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
748 embedok, large_block_ok, outfd, vp, off);
749 } else {
750 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
751 embedok, large_block_ok, outfd, vp, off);
752 }
753 dsl_dataset_rele(ds, FTAG);
754 return (err);
755 }
756
757 int
758 dmu_send(const char *tosnap, const char *fromsnap,
759 boolean_t embedok, boolean_t large_block_ok,
760 int outfd, vnode_t *vp, offset_t *off)
761 {
762 dsl_pool_t *dp;
763 dsl_dataset_t *ds;
764 int err;
765 boolean_t owned = B_FALSE;
766
767 if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL)
768 return (SET_ERROR(EINVAL));
769
770 err = dsl_pool_hold(tosnap, FTAG, &dp);
771 if (err != 0)
772 return (err);
773
774 if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) {
775 /*
776 * We are sending a filesystem or volume. Ensure
777 * that it doesn't change by owning the dataset.
778 */
779 err = dsl_dataset_own(dp, tosnap, FTAG, &ds);
780 owned = B_TRUE;
781 } else {
782 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
783 }
784 if (err != 0) {
785 dsl_pool_rele(dp, FTAG);
786 return (err);
787 }
788
789 if (fromsnap != NULL) {
790 zfs_bookmark_phys_t zb;
791 boolean_t is_clone = B_FALSE;
792 int fsnamelen = strchr(tosnap, '@') - tosnap;
793
794 /*
795 * If the fromsnap is in a different filesystem, then
796 * mark the send stream as a clone.
797 */
798 if (strncmp(tosnap, fromsnap, fsnamelen) != 0 ||
799 (fromsnap[fsnamelen] != '@' &&
800 fromsnap[fsnamelen] != '#')) {
801 is_clone = B_TRUE;
802 }
803
804 if (strchr(fromsnap, '@')) {
805 dsl_dataset_t *fromds;
806 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
807 if (err == 0) {
808 if (!dsl_dataset_is_before(ds, fromds, 0))
809 err = SET_ERROR(EXDEV);
810 zb.zbm_creation_time =
811 dsl_dataset_phys(fromds)->ds_creation_time;
812 zb.zbm_creation_txg =
813 dsl_dataset_phys(fromds)->ds_creation_txg;
814 zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid;
815 is_clone = (ds->ds_dir != fromds->ds_dir);
816 dsl_dataset_rele(fromds, FTAG);
817 }
818 } else {
819 err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb);
820 }
821 if (err != 0) {
822 dsl_dataset_rele(ds, FTAG);
823 dsl_pool_rele(dp, FTAG);
824 return (err);
825 }
826 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
827 embedok, large_block_ok, outfd, vp, off);
828 } else {
829 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
830 embedok, large_block_ok, outfd, vp, off);
831 }
832 if (owned)
833 dsl_dataset_disown(ds, FTAG);
834 else
835 dsl_dataset_rele(ds, FTAG);
836 return (err);
837 }
838
839 int
840 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
841 {
842 int err;
843 uint64_t size, recordsize;
844 ASSERTV(dsl_pool_t *dp = ds->ds_dir->dd_pool);
845
846 ASSERT(dsl_pool_config_held(dp));
847
848 /* tosnap must be a snapshot */
849 if (!ds->ds_is_snapshot)
850 return (SET_ERROR(EINVAL));
851
852 /*
853 * fromsnap must be an earlier snapshot from the same fs as tosnap,
854 * or the origin's fs.
855 */
856 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
857 return (SET_ERROR(EXDEV));
858
859 /* Get uncompressed size estimate of changed data. */
860 if (fromds == NULL) {
861 size = dsl_dataset_phys(ds)->ds_uncompressed_bytes;
862 } else {
863 uint64_t used, comp;
864 err = dsl_dataset_space_written(fromds, ds,
865 &used, &comp, &size);
866 if (err != 0)
867 return (err);
868 }
869
870 /*
871 * Assume that space (both on-disk and in-stream) is dominated by
872 * data. We will adjust for indirect blocks and the copies property,
873 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
874 */
875
876 /*
877 * Subtract out approximate space used by indirect blocks.
878 * Assume most space is used by data blocks (non-indirect, non-dnode).
879 * Assume all blocks are recordsize. Assume ditto blocks and
880 * internal fragmentation counter out compression.
881 *
882 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
883 * block, which we observe in practice.
884 */
885 err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
886 if (err != 0)
887 return (err);
888 size -= size / recordsize * sizeof (blkptr_t);
889
890 /* Add in the space for the record associated with each block. */
891 size += size / recordsize * sizeof (dmu_replay_record_t);
892
893 *sizep = size;
894
895 return (0);
896 }
897
898 typedef struct dmu_recv_begin_arg {
899 const char *drba_origin;
900 dmu_recv_cookie_t *drba_cookie;
901 cred_t *drba_cred;
902 uint64_t drba_snapobj;
903 } dmu_recv_begin_arg_t;
904
905 static int
906 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
907 uint64_t fromguid)
908 {
909 uint64_t val;
910 int error;
911 dsl_pool_t *dp = ds->ds_dir->dd_pool;
912
913 /* temporary clone name must not exist */
914 error = zap_lookup(dp->dp_meta_objset,
915 dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name,
916 8, 1, &val);
917 if (error != ENOENT)
918 return (error == 0 ? EBUSY : error);
919
920 /* new snapshot name must not exist */
921 error = zap_lookup(dp->dp_meta_objset,
922 dsl_dataset_phys(ds)->ds_snapnames_zapobj,
923 drba->drba_cookie->drc_tosnap, 8, 1, &val);
924 if (error != ENOENT)
925 return (error == 0 ? EEXIST : error);
926
927 /*
928 * Check snapshot limit before receiving. We'll recheck again at the
929 * end, but might as well abort before receiving if we're already over
930 * the limit.
931 *
932 * Note that we do not check the file system limit with
933 * dsl_dir_fscount_check because the temporary %clones don't count
934 * against that limit.
935 */
936 error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT,
937 NULL, drba->drba_cred);
938 if (error != 0)
939 return (error);
940
941 if (fromguid != 0) {
942 dsl_dataset_t *snap;
943 uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
944
945 /* Find snapshot in this dir that matches fromguid. */
946 while (obj != 0) {
947 error = dsl_dataset_hold_obj(dp, obj, FTAG,
948 &snap);
949 if (error != 0)
950 return (SET_ERROR(ENODEV));
951 if (snap->ds_dir != ds->ds_dir) {
952 dsl_dataset_rele(snap, FTAG);
953 return (SET_ERROR(ENODEV));
954 }
955 if (dsl_dataset_phys(snap)->ds_guid == fromguid)
956 break;
957 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
958 dsl_dataset_rele(snap, FTAG);
959 }
960 if (obj == 0)
961 return (SET_ERROR(ENODEV));
962
963 if (drba->drba_cookie->drc_force) {
964 drba->drba_snapobj = obj;
965 } else {
966 /*
967 * If we are not forcing, there must be no
968 * changes since fromsnap.
969 */
970 if (dsl_dataset_modified_since_snap(ds, snap)) {
971 dsl_dataset_rele(snap, FTAG);
972 return (SET_ERROR(ETXTBSY));
973 }
974 drba->drba_snapobj = ds->ds_prev->ds_object;
975 }
976
977 dsl_dataset_rele(snap, FTAG);
978 } else {
979 /* if full, most recent snapshot must be $ORIGIN */
980 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= TXG_INITIAL)
981 return (SET_ERROR(ENODEV));
982 drba->drba_snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
983 }
984
985 return (0);
986
987 }
988
989 static int
990 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
991 {
992 dmu_recv_begin_arg_t *drba = arg;
993 dsl_pool_t *dp = dmu_tx_pool(tx);
994 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
995 uint64_t fromguid = drrb->drr_fromguid;
996 int flags = drrb->drr_flags;
997 int error;
998 uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
999 dsl_dataset_t *ds;
1000 const char *tofs = drba->drba_cookie->drc_tofs;
1001
1002 /* already checked */
1003 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
1004
1005 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
1006 DMU_COMPOUNDSTREAM ||
1007 drrb->drr_type >= DMU_OST_NUMTYPES ||
1008 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
1009 return (SET_ERROR(EINVAL));
1010
1011 /* Verify pool version supports SA if SA_SPILL feature set */
1012 if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
1013 spa_version(dp->dp_spa) < SPA_VERSION_SA)
1014 return (SET_ERROR(ENOTSUP));
1015
1016 /*
1017 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1018 * record to a plan WRITE record, so the pool must have the
1019 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1020 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1021 */
1022 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) &&
1023 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA))
1024 return (SET_ERROR(ENOTSUP));
1025 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) &&
1026 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
1027 return (SET_ERROR(ENOTSUP));
1028
1029 /*
1030 * The receiving code doesn't know how to translate large blocks
1031 * to smaller ones, so the pool must have the LARGE_BLOCKS
1032 * feature enabled if the stream has LARGE_BLOCKS.
1033 */
1034 if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
1035 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
1036 return (SET_ERROR(ENOTSUP));
1037
1038 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1039 if (error == 0) {
1040 /* target fs already exists; recv into temp clone */
1041
1042 /* Can't recv a clone into an existing fs */
1043 if (flags & DRR_FLAG_CLONE) {
1044 dsl_dataset_rele(ds, FTAG);
1045 return (SET_ERROR(EINVAL));
1046 }
1047
1048 error = recv_begin_check_existing_impl(drba, ds, fromguid);
1049 dsl_dataset_rele(ds, FTAG);
1050 } else if (error == ENOENT) {
1051 /* target fs does not exist; must be a full backup or clone */
1052 char buf[MAXNAMELEN];
1053
1054 /*
1055 * If it's a non-clone incremental, we are missing the
1056 * target fs, so fail the recv.
1057 */
1058 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
1059 return (SET_ERROR(ENOENT));
1060
1061 /* Open the parent of tofs */
1062 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
1063 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
1064 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
1065 if (error != 0)
1066 return (error);
1067
1068 /*
1069 * Check filesystem and snapshot limits before receiving. We'll
1070 * recheck snapshot limits again at the end (we create the
1071 * filesystems and increment those counts during begin_sync).
1072 */
1073 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1074 ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred);
1075 if (error != 0) {
1076 dsl_dataset_rele(ds, FTAG);
1077 return (error);
1078 }
1079
1080 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1081 ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred);
1082 if (error != 0) {
1083 dsl_dataset_rele(ds, FTAG);
1084 return (error);
1085 }
1086
1087 if (drba->drba_origin != NULL) {
1088 dsl_dataset_t *origin;
1089 error = dsl_dataset_hold(dp, drba->drba_origin,
1090 FTAG, &origin);
1091 if (error != 0) {
1092 dsl_dataset_rele(ds, FTAG);
1093 return (error);
1094 }
1095 if (!origin->ds_is_snapshot) {
1096 dsl_dataset_rele(origin, FTAG);
1097 dsl_dataset_rele(ds, FTAG);
1098 return (SET_ERROR(EINVAL));
1099 }
1100 if (dsl_dataset_phys(origin)->ds_guid != fromguid) {
1101 dsl_dataset_rele(origin, FTAG);
1102 dsl_dataset_rele(ds, FTAG);
1103 return (SET_ERROR(ENODEV));
1104 }
1105 dsl_dataset_rele(origin, FTAG);
1106 }
1107 dsl_dataset_rele(ds, FTAG);
1108 error = 0;
1109 }
1110 return (error);
1111 }
1112
1113 static void
1114 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
1115 {
1116 dmu_recv_begin_arg_t *drba = arg;
1117 dsl_pool_t *dp = dmu_tx_pool(tx);
1118 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1119 const char *tofs = drba->drba_cookie->drc_tofs;
1120 dsl_dataset_t *ds, *newds;
1121 uint64_t dsobj;
1122 int error;
1123 uint64_t crflags;
1124
1125 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
1126 DS_FLAG_CI_DATASET : 0;
1127
1128 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1129 if (error == 0) {
1130 /* create temporary clone */
1131 dsl_dataset_t *snap = NULL;
1132 if (drba->drba_snapobj != 0) {
1133 VERIFY0(dsl_dataset_hold_obj(dp,
1134 drba->drba_snapobj, FTAG, &snap));
1135 }
1136 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
1137 snap, crflags, drba->drba_cred, tx);
1138 dsl_dataset_rele(snap, FTAG);
1139 dsl_dataset_rele(ds, FTAG);
1140 } else {
1141 dsl_dir_t *dd;
1142 const char *tail;
1143 dsl_dataset_t *origin = NULL;
1144
1145 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
1146
1147 if (drba->drba_origin != NULL) {
1148 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
1149 FTAG, &origin));
1150 }
1151
1152 /* Create new dataset. */
1153 dsobj = dsl_dataset_create_sync(dd,
1154 strrchr(tofs, '/') + 1,
1155 origin, crflags, drba->drba_cred, tx);
1156 if (origin != NULL)
1157 dsl_dataset_rele(origin, FTAG);
1158 dsl_dir_rele(dd, FTAG);
1159 drba->drba_cookie->drc_newfs = B_TRUE;
1160 }
1161 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
1162
1163 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
1164 DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
1165 !newds->ds_large_blocks) {
1166 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1167 newds->ds_large_blocks = B_TRUE;
1168 }
1169
1170 dmu_buf_will_dirty(newds->ds_dbuf, tx);
1171 dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT;
1172
1173 /*
1174 * If we actually created a non-clone, we need to create the
1175 * objset in our new dataset.
1176 */
1177 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
1178 (void) dmu_objset_create_impl(dp->dp_spa,
1179 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1180 }
1181
1182 drba->drba_cookie->drc_ds = newds;
1183
1184 spa_history_log_internal_ds(newds, "receive", tx, "");
1185 }
1186
1187 /*
1188 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1189 * succeeds; otherwise we will leak the holds on the datasets.
1190 */
1191 int
1192 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1193 boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1194 {
1195 dmu_recv_begin_arg_t drba = { 0 };
1196 dmu_replay_record_t *drr;
1197
1198 bzero(drc, sizeof (dmu_recv_cookie_t));
1199 drc->drc_drrb = drrb;
1200 drc->drc_tosnap = tosnap;
1201 drc->drc_tofs = tofs;
1202 drc->drc_force = force;
1203 drc->drc_cred = CRED();
1204
1205 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1206 drc->drc_byteswap = B_TRUE;
1207 else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1208 return (SET_ERROR(EINVAL));
1209
1210 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1211 drr->drr_type = DRR_BEGIN;
1212 drr->drr_u.drr_begin = *drc->drc_drrb;
1213 if (drc->drc_byteswap) {
1214 fletcher_4_incremental_byteswap(drr,
1215 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1216 } else {
1217 fletcher_4_incremental_native(drr,
1218 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1219 }
1220 kmem_free(drr, sizeof (dmu_replay_record_t));
1221
1222 if (drc->drc_byteswap) {
1223 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1224 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1225 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1226 drrb->drr_type = BSWAP_32(drrb->drr_type);
1227 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1228 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1229 }
1230
1231 drba.drba_origin = origin;
1232 drba.drba_cookie = drc;
1233 drba.drba_cred = CRED();
1234
1235 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1236 &drba, 5, ZFS_SPACE_CHECK_NORMAL));
1237 }
1238
1239 struct restorearg {
1240 int err;
1241 boolean_t byteswap;
1242 vnode_t *vp;
1243 char *buf;
1244 uint64_t voff;
1245 int bufsize; /* amount of memory allocated for buf */
1246 zio_cksum_t cksum;
1247 avl_tree_t *guid_to_ds_map;
1248 };
1249
1250 typedef struct guid_map_entry {
1251 uint64_t guid;
1252 dsl_dataset_t *gme_ds;
1253 avl_node_t avlnode;
1254 } guid_map_entry_t;
1255
1256 static int
1257 guid_compare(const void *arg1, const void *arg2)
1258 {
1259 const guid_map_entry_t *gmep1 = arg1;
1260 const guid_map_entry_t *gmep2 = arg2;
1261
1262 if (gmep1->guid < gmep2->guid)
1263 return (-1);
1264 else if (gmep1->guid > gmep2->guid)
1265 return (1);
1266 return (0);
1267 }
1268
1269 static void
1270 free_guid_map_onexit(void *arg)
1271 {
1272 avl_tree_t *ca = arg;
1273 void *cookie = NULL;
1274 guid_map_entry_t *gmep;
1275
1276 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1277 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1278 dsl_dataset_rele(gmep->gme_ds, gmep);
1279 kmem_free(gmep, sizeof (guid_map_entry_t));
1280 }
1281 avl_destroy(ca);
1282 kmem_free(ca, sizeof (avl_tree_t));
1283 }
1284
1285 static void *
1286 restore_read(struct restorearg *ra, int len, char *buf)
1287 {
1288 int done = 0;
1289
1290 if (buf == NULL)
1291 buf = ra->buf;
1292
1293 /* some things will require 8-byte alignment, so everything must */
1294 ASSERT0(len % 8);
1295 ASSERT3U(len, <=, ra->bufsize);
1296
1297 while (done < len) {
1298 ssize_t resid;
1299
1300 ra->err = vn_rdwr(UIO_READ, ra->vp,
1301 buf + done, len - done,
1302 ra->voff, UIO_SYSSPACE, FAPPEND,
1303 RLIM64_INFINITY, CRED(), &resid);
1304
1305 if (resid == len - done)
1306 ra->err = SET_ERROR(EINVAL);
1307 ra->voff += len - done - resid;
1308 done = len - resid;
1309 if (ra->err != 0)
1310 return (NULL);
1311 }
1312
1313 ASSERT3U(done, ==, len);
1314 if (ra->byteswap)
1315 fletcher_4_incremental_byteswap(buf, len, &ra->cksum);
1316 else
1317 fletcher_4_incremental_native(buf, len, &ra->cksum);
1318 return (buf);
1319 }
1320
1321 noinline static void
1322 backup_byteswap(dmu_replay_record_t *drr)
1323 {
1324 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1325 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1326 drr->drr_type = BSWAP_32(drr->drr_type);
1327 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1328 switch (drr->drr_type) {
1329 case DRR_BEGIN:
1330 DO64(drr_begin.drr_magic);
1331 DO64(drr_begin.drr_versioninfo);
1332 DO64(drr_begin.drr_creation_time);
1333 DO32(drr_begin.drr_type);
1334 DO32(drr_begin.drr_flags);
1335 DO64(drr_begin.drr_toguid);
1336 DO64(drr_begin.drr_fromguid);
1337 break;
1338 case DRR_OBJECT:
1339 DO64(drr_object.drr_object);
1340 DO32(drr_object.drr_type);
1341 DO32(drr_object.drr_bonustype);
1342 DO32(drr_object.drr_blksz);
1343 DO32(drr_object.drr_bonuslen);
1344 DO64(drr_object.drr_toguid);
1345 break;
1346 case DRR_FREEOBJECTS:
1347 DO64(drr_freeobjects.drr_firstobj);
1348 DO64(drr_freeobjects.drr_numobjs);
1349 DO64(drr_freeobjects.drr_toguid);
1350 break;
1351 case DRR_WRITE:
1352 DO64(drr_write.drr_object);
1353 DO32(drr_write.drr_type);
1354 DO64(drr_write.drr_offset);
1355 DO64(drr_write.drr_length);
1356 DO64(drr_write.drr_toguid);
1357 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1358 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1359 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1360 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1361 DO64(drr_write.drr_key.ddk_prop);
1362 break;
1363 case DRR_WRITE_BYREF:
1364 DO64(drr_write_byref.drr_object);
1365 DO64(drr_write_byref.drr_offset);
1366 DO64(drr_write_byref.drr_length);
1367 DO64(drr_write_byref.drr_toguid);
1368 DO64(drr_write_byref.drr_refguid);
1369 DO64(drr_write_byref.drr_refobject);
1370 DO64(drr_write_byref.drr_refoffset);
1371 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1372 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1373 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1374 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1375 DO64(drr_write_byref.drr_key.ddk_prop);
1376 break;
1377 case DRR_WRITE_EMBEDDED:
1378 DO64(drr_write_embedded.drr_object);
1379 DO64(drr_write_embedded.drr_offset);
1380 DO64(drr_write_embedded.drr_length);
1381 DO64(drr_write_embedded.drr_toguid);
1382 DO32(drr_write_embedded.drr_lsize);
1383 DO32(drr_write_embedded.drr_psize);
1384 break;
1385 case DRR_FREE:
1386 DO64(drr_free.drr_object);
1387 DO64(drr_free.drr_offset);
1388 DO64(drr_free.drr_length);
1389 DO64(drr_free.drr_toguid);
1390 break;
1391 case DRR_SPILL:
1392 DO64(drr_spill.drr_object);
1393 DO64(drr_spill.drr_length);
1394 DO64(drr_spill.drr_toguid);
1395 break;
1396 case DRR_END:
1397 DO64(drr_end.drr_checksum.zc_word[0]);
1398 DO64(drr_end.drr_checksum.zc_word[1]);
1399 DO64(drr_end.drr_checksum.zc_word[2]);
1400 DO64(drr_end.drr_checksum.zc_word[3]);
1401 DO64(drr_end.drr_toguid);
1402 break;
1403 default:
1404 break;
1405 }
1406 #undef DO64
1407 #undef DO32
1408 }
1409
1410 static inline uint8_t
1411 deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size)
1412 {
1413 if (bonus_type == DMU_OT_SA) {
1414 return (1);
1415 } else {
1416 return (1 +
1417 ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT));
1418 }
1419 }
1420
1421 noinline static int
1422 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1423 {
1424 dmu_object_info_t doi;
1425 dmu_tx_t *tx;
1426 void *data = NULL;
1427 uint64_t object;
1428 int err;
1429
1430 if (drro->drr_type == DMU_OT_NONE ||
1431 !DMU_OT_IS_VALID(drro->drr_type) ||
1432 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1433 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1434 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1435 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1436 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1437 drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(os)) ||
1438 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1439 return (SET_ERROR(EINVAL));
1440 }
1441
1442 err = dmu_object_info(os, drro->drr_object, &doi);
1443
1444 if (err != 0 && err != ENOENT)
1445 return (SET_ERROR(EINVAL));
1446 object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT;
1447
1448 if (drro->drr_bonuslen) {
1449 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8), NULL);
1450 if (ra->err != 0)
1451 return (ra->err);
1452 }
1453
1454 /*
1455 * If we are losing blkptrs or changing the block size this must
1456 * be a new file instance. We must clear out the previous file
1457 * contents before we can change this type of metadata in the dnode.
1458 */
1459 if (err == 0) {
1460 int nblkptr;
1461
1462 nblkptr = deduce_nblkptr(drro->drr_bonustype,
1463 drro->drr_bonuslen);
1464
1465 if (drro->drr_blksz != doi.doi_data_block_size ||
1466 nblkptr < doi.doi_nblkptr) {
1467 err = dmu_free_long_range(os, drro->drr_object,
1468 0, DMU_OBJECT_END);
1469 if (err != 0)
1470 return (SET_ERROR(EINVAL));
1471 }
1472 }
1473
1474 tx = dmu_tx_create(os);
1475 dmu_tx_hold_bonus(tx, object);
1476 err = dmu_tx_assign(tx, TXG_WAIT);
1477 if (err != 0) {
1478 dmu_tx_abort(tx);
1479 return (err);
1480 }
1481
1482 if (object == DMU_NEW_OBJECT) {
1483 /* currently free, want to be allocated */
1484 err = dmu_object_claim(os, drro->drr_object,
1485 drro->drr_type, drro->drr_blksz,
1486 drro->drr_bonustype, drro->drr_bonuslen, tx);
1487 } else if (drro->drr_type != doi.doi_type ||
1488 drro->drr_blksz != doi.doi_data_block_size ||
1489 drro->drr_bonustype != doi.doi_bonus_type ||
1490 drro->drr_bonuslen != doi.doi_bonus_size) {
1491 /* currently allocated, but with different properties */
1492 err = dmu_object_reclaim(os, drro->drr_object,
1493 drro->drr_type, drro->drr_blksz,
1494 drro->drr_bonustype, drro->drr_bonuslen, tx);
1495 }
1496 if (err != 0) {
1497 dmu_tx_commit(tx);
1498 return (SET_ERROR(EINVAL));
1499 }
1500
1501 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1502 tx);
1503 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1504
1505 if (data != NULL) {
1506 dmu_buf_t *db;
1507
1508 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1509 dmu_buf_will_dirty(db, tx);
1510
1511 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1512 bcopy(data, db->db_data, drro->drr_bonuslen);
1513 if (ra->byteswap) {
1514 dmu_object_byteswap_t byteswap =
1515 DMU_OT_BYTESWAP(drro->drr_bonustype);
1516 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1517 drro->drr_bonuslen);
1518 }
1519 dmu_buf_rele(db, FTAG);
1520 }
1521 dmu_tx_commit(tx);
1522 return (0);
1523 }
1524
1525 /* ARGSUSED */
1526 noinline static int
1527 restore_freeobjects(struct restorearg *ra, objset_t *os,
1528 struct drr_freeobjects *drrfo)
1529 {
1530 uint64_t obj;
1531
1532 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1533 return (SET_ERROR(EINVAL));
1534
1535 for (obj = drrfo->drr_firstobj;
1536 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1537 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1538 int err;
1539
1540 if (dmu_object_info(os, obj, NULL) != 0)
1541 continue;
1542
1543 err = dmu_free_long_object(os, obj);
1544 if (err != 0)
1545 return (err);
1546 }
1547 return (0);
1548 }
1549
1550 noinline static int
1551 restore_write(struct restorearg *ra, objset_t *os,
1552 struct drr_write *drrw)
1553 {
1554 dmu_tx_t *tx;
1555 dmu_buf_t *bonus;
1556 arc_buf_t *abuf;
1557 void *data;
1558 int err;
1559
1560 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1561 !DMU_OT_IS_VALID(drrw->drr_type))
1562 return (SET_ERROR(EINVAL));
1563
1564 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1565 return (SET_ERROR(EINVAL));
1566
1567 if (dmu_bonus_hold(os, drrw->drr_object, FTAG, &bonus) != 0)
1568 return (SET_ERROR(EINVAL));
1569
1570 abuf = dmu_request_arcbuf(bonus, drrw->drr_length);
1571
1572 data = restore_read(ra, drrw->drr_length, abuf->b_data);
1573 if (data == NULL) {
1574 dmu_return_arcbuf(abuf);
1575 dmu_buf_rele(bonus, FTAG);
1576 return (ra->err);
1577 }
1578
1579 tx = dmu_tx_create(os);
1580
1581 dmu_tx_hold_write(tx, drrw->drr_object,
1582 drrw->drr_offset, drrw->drr_length);
1583 err = dmu_tx_assign(tx, TXG_WAIT);
1584 if (err != 0) {
1585 dmu_return_arcbuf(abuf);
1586 dmu_buf_rele(bonus, FTAG);
1587 dmu_tx_abort(tx);
1588 return (err);
1589 }
1590 if (ra->byteswap) {
1591 dmu_object_byteswap_t byteswap =
1592 DMU_OT_BYTESWAP(drrw->drr_type);
1593 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1594 }
1595 dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx);
1596 dmu_tx_commit(tx);
1597 dmu_buf_rele(bonus, FTAG);
1598 return (0);
1599 }
1600
1601 /*
1602 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1603 * streams to refer to a copy of the data that is already on the
1604 * system because it came in earlier in the stream. This function
1605 * finds the earlier copy of the data, and uses that copy instead of
1606 * data from the stream to fulfill this write.
1607 */
1608 static int
1609 restore_write_byref(struct restorearg *ra, objset_t *os,
1610 struct drr_write_byref *drrwbr)
1611 {
1612 dmu_tx_t *tx;
1613 int err;
1614 guid_map_entry_t gmesrch;
1615 guid_map_entry_t *gmep;
1616 avl_index_t where;
1617 objset_t *ref_os = NULL;
1618 dmu_buf_t *dbp;
1619
1620 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1621 return (SET_ERROR(EINVAL));
1622
1623 /*
1624 * If the GUID of the referenced dataset is different from the
1625 * GUID of the target dataset, find the referenced dataset.
1626 */
1627 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1628 gmesrch.guid = drrwbr->drr_refguid;
1629 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1630 &where)) == NULL) {
1631 return (SET_ERROR(EINVAL));
1632 }
1633 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1634 return (SET_ERROR(EINVAL));
1635 } else {
1636 ref_os = os;
1637 }
1638
1639 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1640 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1641 if (err != 0)
1642 return (err);
1643
1644 tx = dmu_tx_create(os);
1645
1646 dmu_tx_hold_write(tx, drrwbr->drr_object,
1647 drrwbr->drr_offset, drrwbr->drr_length);
1648 err = dmu_tx_assign(tx, TXG_WAIT);
1649 if (err != 0) {
1650 dmu_tx_abort(tx);
1651 return (err);
1652 }
1653 dmu_write(os, drrwbr->drr_object,
1654 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1655 dmu_buf_rele(dbp, FTAG);
1656 dmu_tx_commit(tx);
1657 return (0);
1658 }
1659
1660 static int
1661 restore_write_embedded(struct restorearg *ra, objset_t *os,
1662 struct drr_write_embedded *drrwnp)
1663 {
1664 dmu_tx_t *tx;
1665 int err;
1666 void *data;
1667
1668 if (drrwnp->drr_offset + drrwnp->drr_length < drrwnp->drr_offset)
1669 return (EINVAL);
1670
1671 if (drrwnp->drr_psize > BPE_PAYLOAD_SIZE)
1672 return (EINVAL);
1673
1674 if (drrwnp->drr_etype >= NUM_BP_EMBEDDED_TYPES)
1675 return (EINVAL);
1676 if (drrwnp->drr_compression >= ZIO_COMPRESS_FUNCTIONS)
1677 return (EINVAL);
1678
1679 data = restore_read(ra, P2ROUNDUP(drrwnp->drr_psize, 8), NULL);
1680 if (data == NULL)
1681 return (ra->err);
1682
1683 tx = dmu_tx_create(os);
1684
1685 dmu_tx_hold_write(tx, drrwnp->drr_object,
1686 drrwnp->drr_offset, drrwnp->drr_length);
1687 err = dmu_tx_assign(tx, TXG_WAIT);
1688 if (err != 0) {
1689 dmu_tx_abort(tx);
1690 return (err);
1691 }
1692
1693 dmu_write_embedded(os, drrwnp->drr_object,
1694 drrwnp->drr_offset, data, drrwnp->drr_etype,
1695 drrwnp->drr_compression, drrwnp->drr_lsize, drrwnp->drr_psize,
1696 ra->byteswap ^ ZFS_HOST_BYTEORDER, tx);
1697
1698 dmu_tx_commit(tx);
1699 return (0);
1700 }
1701
1702 static int
1703 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1704 {
1705 dmu_tx_t *tx;
1706 void *data;
1707 dmu_buf_t *db, *db_spill;
1708 int err;
1709
1710 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1711 drrs->drr_length > spa_maxblocksize(dmu_objset_spa(os)))
1712 return (SET_ERROR(EINVAL));
1713
1714 data = restore_read(ra, drrs->drr_length, NULL);
1715 if (data == NULL)
1716 return (ra->err);
1717
1718 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1719 return (SET_ERROR(EINVAL));
1720
1721 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1722 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1723 dmu_buf_rele(db, FTAG);
1724 return (err);
1725 }
1726
1727 tx = dmu_tx_create(os);
1728
1729 dmu_tx_hold_spill(tx, db->db_object);
1730
1731 err = dmu_tx_assign(tx, TXG_WAIT);
1732 if (err != 0) {
1733 dmu_buf_rele(db, FTAG);
1734 dmu_buf_rele(db_spill, FTAG);
1735 dmu_tx_abort(tx);
1736 return (err);
1737 }
1738 dmu_buf_will_dirty(db_spill, tx);
1739
1740 if (db_spill->db_size < drrs->drr_length)
1741 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1742 drrs->drr_length, tx));
1743 bcopy(data, db_spill->db_data, drrs->drr_length);
1744
1745 dmu_buf_rele(db, FTAG);
1746 dmu_buf_rele(db_spill, FTAG);
1747
1748 dmu_tx_commit(tx);
1749 return (0);
1750 }
1751
1752 /* ARGSUSED */
1753 noinline static int
1754 restore_free(struct restorearg *ra, objset_t *os,
1755 struct drr_free *drrf)
1756 {
1757 int err;
1758
1759 if (drrf->drr_length != -1ULL &&
1760 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1761 return (SET_ERROR(EINVAL));
1762
1763 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1764 return (SET_ERROR(EINVAL));
1765
1766 err = dmu_free_long_range(os, drrf->drr_object,
1767 drrf->drr_offset, drrf->drr_length);
1768 return (err);
1769 }
1770
1771 /* used to destroy the drc_ds on error */
1772 static void
1773 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1774 {
1775 char name[MAXNAMELEN];
1776 dsl_dataset_name(drc->drc_ds, name);
1777 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1778 (void) dsl_destroy_head(name);
1779 }
1780
1781 /*
1782 * NB: callers *must* call dmu_recv_end() if this succeeds.
1783 */
1784 int
1785 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1786 int cleanup_fd, uint64_t *action_handlep)
1787 {
1788 struct restorearg ra = { 0 };
1789 dmu_replay_record_t *drr;
1790 objset_t *os;
1791 zio_cksum_t pcksum;
1792 int featureflags;
1793
1794 ra.byteswap = drc->drc_byteswap;
1795 ra.cksum = drc->drc_cksum;
1796 ra.vp = vp;
1797 ra.voff = *voffp;
1798 ra.bufsize = SPA_MAXBLOCKSIZE;
1799 ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
1800
1801 /* these were verified in dmu_recv_begin */
1802 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1803 DMU_SUBSTREAM);
1804 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1805
1806 /*
1807 * Open the objset we are modifying.
1808 */
1809 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1810
1811 ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT);
1812
1813 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1814
1815 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1816 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1817 minor_t minor;
1818
1819 if (cleanup_fd == -1) {
1820 ra.err = SET_ERROR(EBADF);
1821 goto out;
1822 }
1823 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1824 if (ra.err != 0) {
1825 cleanup_fd = -1;
1826 goto out;
1827 }
1828
1829 if (*action_handlep == 0) {
1830 ra.guid_to_ds_map =
1831 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1832 avl_create(ra.guid_to_ds_map, guid_compare,
1833 sizeof (guid_map_entry_t),
1834 offsetof(guid_map_entry_t, avlnode));
1835 ra.err = zfs_onexit_add_cb(minor,
1836 free_guid_map_onexit, ra.guid_to_ds_map,
1837 action_handlep);
1838 if (ra.err != 0)
1839 goto out;
1840 } else {
1841 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1842 (void **)&ra.guid_to_ds_map);
1843 if (ra.err != 0)
1844 goto out;
1845 }
1846
1847 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1848 }
1849
1850 /*
1851 * Read records and process them.
1852 */
1853 pcksum = ra.cksum;
1854 while (ra.err == 0 &&
1855 NULL != (drr = restore_read(&ra, sizeof (*drr), NULL))) {
1856 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1857 ra.err = SET_ERROR(EINTR);
1858 goto out;
1859 }
1860
1861 if (ra.byteswap)
1862 backup_byteswap(drr);
1863
1864 switch (drr->drr_type) {
1865 case DRR_OBJECT:
1866 {
1867 /*
1868 * We need to make a copy of the record header,
1869 * because restore_{object,write} may need to
1870 * restore_read(), which will invalidate drr.
1871 */
1872 struct drr_object drro = drr->drr_u.drr_object;
1873 ra.err = restore_object(&ra, os, &drro);
1874 break;
1875 }
1876 case DRR_FREEOBJECTS:
1877 {
1878 struct drr_freeobjects drrfo =
1879 drr->drr_u.drr_freeobjects;
1880 ra.err = restore_freeobjects(&ra, os, &drrfo);
1881 break;
1882 }
1883 case DRR_WRITE:
1884 {
1885 struct drr_write drrw = drr->drr_u.drr_write;
1886 ra.err = restore_write(&ra, os, &drrw);
1887 break;
1888 }
1889 case DRR_WRITE_BYREF:
1890 {
1891 struct drr_write_byref drrwbr =
1892 drr->drr_u.drr_write_byref;
1893 ra.err = restore_write_byref(&ra, os, &drrwbr);
1894 break;
1895 }
1896 case DRR_WRITE_EMBEDDED:
1897 {
1898 struct drr_write_embedded drrwe =
1899 drr->drr_u.drr_write_embedded;
1900 ra.err = restore_write_embedded(&ra, os, &drrwe);
1901 break;
1902 }
1903 case DRR_FREE:
1904 {
1905 struct drr_free drrf = drr->drr_u.drr_free;
1906 ra.err = restore_free(&ra, os, &drrf);
1907 break;
1908 }
1909 case DRR_END:
1910 {
1911 struct drr_end drre = drr->drr_u.drr_end;
1912 /*
1913 * We compare against the *previous* checksum
1914 * value, because the stored checksum is of
1915 * everything before the DRR_END record.
1916 */
1917 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1918 ra.err = SET_ERROR(ECKSUM);
1919 goto out;
1920 }
1921 case DRR_SPILL:
1922 {
1923 struct drr_spill drrs = drr->drr_u.drr_spill;
1924 ra.err = restore_spill(&ra, os, &drrs);
1925 break;
1926 }
1927 default:
1928 ra.err = SET_ERROR(EINVAL);
1929 goto out;
1930 }
1931 pcksum = ra.cksum;
1932 }
1933 ASSERT(ra.err != 0);
1934
1935 out:
1936 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1937 zfs_onexit_fd_rele(cleanup_fd);
1938
1939 if (ra.err != 0) {
1940 /*
1941 * destroy what we created, so we don't leave it in the
1942 * inconsistent restoring state.
1943 */
1944 dmu_recv_cleanup_ds(drc);
1945 }
1946
1947 vmem_free(ra.buf, ra.bufsize);
1948 *voffp = ra.voff;
1949 return (ra.err);
1950 }
1951
1952 static int
1953 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1954 {
1955 dmu_recv_cookie_t *drc = arg;
1956 dsl_pool_t *dp = dmu_tx_pool(tx);
1957 int error;
1958
1959 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1960
1961 if (!drc->drc_newfs) {
1962 dsl_dataset_t *origin_head;
1963
1964 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1965 if (error != 0)
1966 return (error);
1967 if (drc->drc_force) {
1968 /*
1969 * We will destroy any snapshots in tofs (i.e. before
1970 * origin_head) that are after the origin (which is
1971 * the snap before drc_ds, because drc_ds can not
1972 * have any snaps of its own).
1973 */
1974 uint64_t obj;
1975
1976 obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj;
1977 while (obj !=
1978 dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) {
1979 dsl_dataset_t *snap;
1980 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1981 &snap);
1982 if (error != 0)
1983 return (error);
1984 if (snap->ds_dir != origin_head->ds_dir)
1985 error = SET_ERROR(EINVAL);
1986 if (error == 0) {
1987 error = dsl_destroy_snapshot_check_impl(
1988 snap, B_FALSE);
1989 }
1990 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
1991 dsl_dataset_rele(snap, FTAG);
1992 if (error != 0)
1993 return (error);
1994 }
1995 }
1996 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1997 origin_head, drc->drc_force, drc->drc_owner, tx);
1998 if (error != 0) {
1999 dsl_dataset_rele(origin_head, FTAG);
2000 return (error);
2001 }
2002 error = dsl_dataset_snapshot_check_impl(origin_head,
2003 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
2004 dsl_dataset_rele(origin_head, FTAG);
2005 if (error != 0)
2006 return (error);
2007
2008 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
2009 } else {
2010 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
2011 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
2012 }
2013 return (error);
2014 }
2015
2016 static void
2017 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
2018 {
2019 dmu_recv_cookie_t *drc = arg;
2020 dsl_pool_t *dp = dmu_tx_pool(tx);
2021
2022 spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
2023 tx, "snap=%s", drc->drc_tosnap);
2024
2025 if (!drc->drc_newfs) {
2026 dsl_dataset_t *origin_head;
2027
2028 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
2029 &origin_head));
2030
2031 if (drc->drc_force) {
2032 /*
2033 * Destroy any snapshots of drc_tofs (origin_head)
2034 * after the origin (the snap before drc_ds).
2035 */
2036 uint64_t obj;
2037
2038 obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj;
2039 while (obj !=
2040 dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) {
2041 dsl_dataset_t *snap;
2042 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
2043 &snap));
2044 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
2045 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
2046 dsl_destroy_snapshot_sync_impl(snap,
2047 B_FALSE, tx);
2048 dsl_dataset_rele(snap, FTAG);
2049 }
2050 }
2051 VERIFY3P(drc->drc_ds->ds_prev, ==,
2052 origin_head->ds_prev);
2053
2054 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
2055 origin_head, tx);
2056 dsl_dataset_snapshot_sync_impl(origin_head,
2057 drc->drc_tosnap, tx);
2058
2059 /* set snapshot's creation time and guid */
2060 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
2061 dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time =
2062 drc->drc_drrb->drr_creation_time;
2063 dsl_dataset_phys(origin_head->ds_prev)->ds_guid =
2064 drc->drc_drrb->drr_toguid;
2065 dsl_dataset_phys(origin_head->ds_prev)->ds_flags &=
2066 ~DS_FLAG_INCONSISTENT;
2067
2068 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2069 dsl_dataset_phys(origin_head)->ds_flags &=
2070 ~DS_FLAG_INCONSISTENT;
2071
2072 dsl_dataset_rele(origin_head, FTAG);
2073 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
2074
2075 if (drc->drc_owner != NULL)
2076 VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
2077 } else {
2078 dsl_dataset_t *ds = drc->drc_ds;
2079
2080 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
2081
2082 /* set snapshot's creation time and guid */
2083 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2084 dsl_dataset_phys(ds->ds_prev)->ds_creation_time =
2085 drc->drc_drrb->drr_creation_time;
2086 dsl_dataset_phys(ds->ds_prev)->ds_guid =
2087 drc->drc_drrb->drr_toguid;
2088 dsl_dataset_phys(ds->ds_prev)->ds_flags &=
2089 ~DS_FLAG_INCONSISTENT;
2090
2091 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2092 dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT;
2093 }
2094 drc->drc_newsnapobj = dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj;
2095 /*
2096 * Release the hold from dmu_recv_begin. This must be done before
2097 * we return to open context, so that when we free the dataset's dnode,
2098 * we can evict its bonus buffer.
2099 */
2100 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
2101 drc->drc_ds = NULL;
2102 }
2103
2104 static int
2105 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
2106 {
2107 dsl_pool_t *dp;
2108 dsl_dataset_t *snapds;
2109 guid_map_entry_t *gmep;
2110 int err;
2111
2112 ASSERT(guid_map != NULL);
2113
2114 err = dsl_pool_hold(name, FTAG, &dp);
2115 if (err != 0)
2116 return (err);
2117 gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
2118 err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
2119 if (err == 0) {
2120 gmep->guid = dsl_dataset_phys(snapds)->ds_guid;
2121 gmep->gme_ds = snapds;
2122 avl_add(guid_map, gmep);
2123 dsl_dataset_long_hold(snapds, gmep);
2124 } else {
2125 kmem_free(gmep, sizeof (*gmep));
2126 }
2127
2128 dsl_pool_rele(dp, FTAG);
2129 return (err);
2130 }
2131
2132 static int dmu_recv_end_modified_blocks = 3;
2133
2134 static int
2135 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
2136 {
2137 int error;
2138
2139 #ifdef _KERNEL
2140 char *name;
2141
2142 /*
2143 * We will be destroying the ds; make sure its origin is unmounted if
2144 * necessary.
2145 */
2146 name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2147 dsl_dataset_name(drc->drc_ds, name);
2148 zfs_destroy_unmount_origin(name);
2149 kmem_free(name, MAXNAMELEN);
2150 #endif
2151
2152 error = dsl_sync_task(drc->drc_tofs,
2153 dmu_recv_end_check, dmu_recv_end_sync, drc,
2154 dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2155
2156 if (error != 0)
2157 dmu_recv_cleanup_ds(drc);
2158 return (error);
2159 }
2160
2161 static int
2162 dmu_recv_new_end(dmu_recv_cookie_t *drc)
2163 {
2164 int error;
2165
2166 error = dsl_sync_task(drc->drc_tofs,
2167 dmu_recv_end_check, dmu_recv_end_sync, drc,
2168 dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2169
2170 if (error != 0) {
2171 dmu_recv_cleanup_ds(drc);
2172 } else if (drc->drc_guid_to_ds_map != NULL) {
2173 (void) add_ds_to_guidmap(drc->drc_tofs,
2174 drc->drc_guid_to_ds_map,
2175 drc->drc_newsnapobj);
2176 }
2177 return (error);
2178 }
2179
2180 int
2181 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
2182 {
2183 drc->drc_owner = owner;
2184
2185 if (drc->drc_newfs)
2186 return (dmu_recv_new_end(drc));
2187 else
2188 return (dmu_recv_existing_end(drc));
2189 }
2190
2191 /*
2192 * Return TRUE if this objset is currently being received into.
2193 */
2194 boolean_t
2195 dmu_objset_is_receiving(objset_t *os)
2196 {
2197 return (os->os_dsl_dataset != NULL &&
2198 os->os_dsl_dataset->ds_owner == dmu_recv_tag);
2199 }
2200
2201 #if defined(_KERNEL)
2202 module_param(zfs_send_corrupt_data, int, 0644);
2203 MODULE_PARM_DESC(zfs_send_corrupt_data, "Allow sending corrupt data");
2204 #endif
ZFS On Linux mirror for PVE