]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - zfs/module/zfs/dmu_send.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:


6a349c6600d123c41f4b82b89abbc47b20978da4
[mirror_ubuntu-zesty-kernel.git] / zfs / 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 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
28 */
29
30 #include <sys/dmu.h>
31 #include <sys/dmu_impl.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/dbuf.h>
34 #include <sys/dnode.h>
35 #include <sys/zfs_context.h>
36 #include <sys/dmu_objset.h>
37 #include <sys/dmu_traverse.h>
38 #include <sys/dsl_dataset.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/dsl_prop.h>
41 #include <sys/dsl_pool.h>
42 #include <sys/dsl_synctask.h>
43 #include <sys/spa_impl.h>
44 #include <sys/zfs_ioctl.h>
45 #include <sys/zap.h>
46 #include <sys/zio_checksum.h>
47 #include <sys/zfs_znode.h>
48 #include <zfs_fletcher.h>
49 #include <sys/avl.h>
50 #include <sys/ddt.h>
51 #include <sys/zfs_onexit.h>
52 #include <sys/dmu_send.h>
53 #include <sys/dsl_destroy.h>
54 #include <sys/blkptr.h>
55 #include <sys/dsl_bookmark.h>
56 #include <sys/zfeature.h>
57 #include <sys/zvol.h>
58
59 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
60 int zfs_send_corrupt_data = B_FALSE;
61
62 static char *dmu_recv_tag = "dmu_recv_tag";
63 static const char *recv_clone_name = "%recv";
64
65 typedef struct dump_bytes_io {
66 dmu_sendarg_t *dbi_dsp;
67 void *dbi_buf;
68 int dbi_len;
69 } dump_bytes_io_t;
70
71 static void
72 dump_bytes_strategy(void *arg)
73 {
74 dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg;
75 dmu_sendarg_t *dsp = dbi->dbi_dsp;
76 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
77 ssize_t resid; /* have to get resid to get detailed errno */
78 ASSERT0(dbi->dbi_len % 8);
79
80 fletcher_4_incremental_native(dbi->dbi_buf, dbi->dbi_len, &dsp->dsa_zc);
81 dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
82 (caddr_t)dbi->dbi_buf, dbi->dbi_len,
83 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
84
85 mutex_enter(&ds->ds_sendstream_lock);
86 *dsp->dsa_off += dbi->dbi_len;
87 mutex_exit(&ds->ds_sendstream_lock);
88 }
89
90 static int
91 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
92 {
93 dump_bytes_io_t dbi;
94
95 dbi.dbi_dsp = dsp;
96 dbi.dbi_buf = buf;
97 dbi.dbi_len = len;
98
99 /*
100 * The vn_rdwr() call is performed in a taskq to ensure that there is
101 * always enough stack space to write safely to the target filesystem.
102 * The ZIO_TYPE_FREE threads are used because there can be a lot of
103 * them and they are used in vdev_file.c for a similar purpose.
104 */
105 spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE,
106 ZIO_TASKQ_ISSUE, dump_bytes_strategy, &dbi, TQ_SLEEP);
107
108 return (dsp->dsa_err);
109 }
110
111 static int
112 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
113 uint64_t length)
114 {
115 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
116
117 /*
118 * When we receive a free record, dbuf_free_range() assumes
119 * that the receiving system doesn't have any dbufs in the range
120 * being freed. This is always true because there is a one-record
121 * constraint: we only send one WRITE record for any given
122 * object+offset. We know that the one-record constraint is
123 * true because we always send data in increasing order by
124 * object,offset.
125 *
126 * If the increasing-order constraint ever changes, we should find
127 * another way to assert that the one-record constraint is still
128 * satisfied.
129 */
130 ASSERT(object > dsp->dsa_last_data_object ||
131 (object == dsp->dsa_last_data_object &&
132 offset > dsp->dsa_last_data_offset));
133
134 /*
135 * If we are doing a non-incremental send, then there can't
136 * be any data in the dataset we're receiving into. Therefore
137 * a free record would simply be a no-op. Save space by not
138 * sending it to begin with.
139 */
140 if (!dsp->dsa_incremental)
141 return (0);
142
143 if (length != -1ULL && offset + length < offset)
144 length = -1ULL;
145
146 /*
147 * If there is a pending op, but it's not PENDING_FREE, push it out,
148 * since free block aggregation can only be done for blocks of the
149 * same type (i.e., DRR_FREE records can only be aggregated with
150 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
151 * aggregated with other DRR_FREEOBJECTS records.
152 */
153 if (dsp->dsa_pending_op != PENDING_NONE &&
154 dsp->dsa_pending_op != PENDING_FREE) {
155 if (dump_bytes(dsp, dsp->dsa_drr,
156 sizeof (dmu_replay_record_t)) != 0)
157 return (SET_ERROR(EINTR));
158 dsp->dsa_pending_op = PENDING_NONE;
159 }
160
161 if (dsp->dsa_pending_op == PENDING_FREE) {
162 /*
163 * There should never be a PENDING_FREE if length is -1
164 * (because dump_dnode is the only place where this
165 * function is called with a -1, and only after flushing
166 * any pending record).
167 */
168 ASSERT(length != -1ULL);
169 /*
170 * Check to see whether this free block can be aggregated
171 * with pending one.
172 */
173 if (drrf->drr_object == object && drrf->drr_offset +
174 drrf->drr_length == offset) {
175 drrf->drr_length += length;
176 return (0);
177 } else {
178 /* not a continuation. Push out pending record */
179 if (dump_bytes(dsp, dsp->dsa_drr,
180 sizeof (dmu_replay_record_t)) != 0)
181 return (SET_ERROR(EINTR));
182 dsp->dsa_pending_op = PENDING_NONE;
183 }
184 }
185 /* create a FREE record and make it pending */
186 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
187 dsp->dsa_drr->drr_type = DRR_FREE;
188 drrf->drr_object = object;
189 drrf->drr_offset = offset;
190 drrf->drr_length = length;
191 drrf->drr_toguid = dsp->dsa_toguid;
192 if (length == -1ULL) {
193 if (dump_bytes(dsp, dsp->dsa_drr,
194 sizeof (dmu_replay_record_t)) != 0)
195 return (SET_ERROR(EINTR));
196 } else {
197 dsp->dsa_pending_op = PENDING_FREE;
198 }
199
200 return (0);
201 }
202
203 static int
204 dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type,
205 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
206 {
207 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
208
209 /*
210 * We send data in increasing object, offset order.
211 * See comment in dump_free() for details.
212 */
213 ASSERT(object > dsp->dsa_last_data_object ||
214 (object == dsp->dsa_last_data_object &&
215 offset > dsp->dsa_last_data_offset));
216 dsp->dsa_last_data_object = object;
217 dsp->dsa_last_data_offset = offset + blksz - 1;
218
219 /*
220 * If there is any kind of pending aggregation (currently either
221 * a grouping of free objects or free blocks), push it out to
222 * the stream, since aggregation can't be done across operations
223 * of different types.
224 */
225 if (dsp->dsa_pending_op != PENDING_NONE) {
226 if (dump_bytes(dsp, dsp->dsa_drr,
227 sizeof (dmu_replay_record_t)) != 0)
228 return (SET_ERROR(EINTR));
229 dsp->dsa_pending_op = PENDING_NONE;
230 }
231 /* write a DATA record */
232 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
233 dsp->dsa_drr->drr_type = DRR_WRITE;
234 drrw->drr_object = object;
235 drrw->drr_type = type;
236 drrw->drr_offset = offset;
237 drrw->drr_length = blksz;
238 drrw->drr_toguid = dsp->dsa_toguid;
239 if (bp == NULL || BP_IS_EMBEDDED(bp)) {
240 /*
241 * There's no pre-computed checksum for partial-block
242 * writes or embedded BP's, so (like
243 * fletcher4-checkummed blocks) userland will have to
244 * compute a dedup-capable checksum itself.
245 */
246 drrw->drr_checksumtype = ZIO_CHECKSUM_OFF;
247 } else {
248 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
249 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
250 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
251 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
252 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
253 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
254 drrw->drr_key.ddk_cksum = bp->blk_cksum;
255 }
256
257 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
258 return (SET_ERROR(EINTR));
259 if (dump_bytes(dsp, data, blksz) != 0)
260 return (SET_ERROR(EINTR));
261 return (0);
262 }
263
264 static int
265 dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
266 int blksz, const blkptr_t *bp)
267 {
268 char buf[BPE_PAYLOAD_SIZE];
269 struct drr_write_embedded *drrw =
270 &(dsp->dsa_drr->drr_u.drr_write_embedded);
271
272 if (dsp->dsa_pending_op != PENDING_NONE) {
273 if (dump_bytes(dsp, dsp->dsa_drr,
274 sizeof (dmu_replay_record_t)) != 0)
275 return (EINTR);
276 dsp->dsa_pending_op = PENDING_NONE;
277 }
278
279 ASSERT(BP_IS_EMBEDDED(bp));
280
281 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
282 dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED;
283 drrw->drr_object = object;
284 drrw->drr_offset = offset;
285 drrw->drr_length = blksz;
286 drrw->drr_toguid = dsp->dsa_toguid;
287 drrw->drr_compression = BP_GET_COMPRESS(bp);
288 drrw->drr_etype = BPE_GET_ETYPE(bp);
289 drrw->drr_lsize = BPE_GET_LSIZE(bp);
290 drrw->drr_psize = BPE_GET_PSIZE(bp);
291
292 decode_embedded_bp_compressed(bp, buf);
293
294 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
295 return (EINTR);
296 if (dump_bytes(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0)
297 return (EINTR);
298 return (0);
299 }
300
301 static int
302 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
303 {
304 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
305
306 if (dsp->dsa_pending_op != PENDING_NONE) {
307 if (dump_bytes(dsp, dsp->dsa_drr,
308 sizeof (dmu_replay_record_t)) != 0)
309 return (SET_ERROR(EINTR));
310 dsp->dsa_pending_op = PENDING_NONE;
311 }
312
313 /* write a SPILL record */
314 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
315 dsp->dsa_drr->drr_type = DRR_SPILL;
316 drrs->drr_object = object;
317 drrs->drr_length = blksz;
318 drrs->drr_toguid = dsp->dsa_toguid;
319
320 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
321 return (SET_ERROR(EINTR));
322 if (dump_bytes(dsp, data, blksz))
323 return (SET_ERROR(EINTR));
324 return (0);
325 }
326
327 static int
328 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
329 {
330 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
331
332 /* See comment in dump_free(). */
333 if (!dsp->dsa_incremental)
334 return (0);
335
336 /*
337 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
338 * push it out, since free block aggregation can only be done for
339 * blocks of the same type (i.e., DRR_FREE records can only be
340 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
341 * can only be aggregated with other DRR_FREEOBJECTS records.
342 */
343 if (dsp->dsa_pending_op != PENDING_NONE &&
344 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
345 if (dump_bytes(dsp, dsp->dsa_drr,
346 sizeof (dmu_replay_record_t)) != 0)
347 return (SET_ERROR(EINTR));
348 dsp->dsa_pending_op = PENDING_NONE;
349 }
350 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
351 /*
352 * See whether this free object array can be aggregated
353 * with pending one
354 */
355 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
356 drrfo->drr_numobjs += numobjs;
357 return (0);
358 } else {
359 /* can't be aggregated. Push out pending record */
360 if (dump_bytes(dsp, dsp->dsa_drr,
361 sizeof (dmu_replay_record_t)) != 0)
362 return (SET_ERROR(EINTR));
363 dsp->dsa_pending_op = PENDING_NONE;
364 }
365 }
366
367 /* write a FREEOBJECTS record */
368 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
369 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
370 drrfo->drr_firstobj = firstobj;
371 drrfo->drr_numobjs = numobjs;
372 drrfo->drr_toguid = dsp->dsa_toguid;
373
374 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
375
376 return (0);
377 }
378
379 static int
380 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
381 {
382 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
383
384 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
385 return (dump_freeobjects(dsp, object, 1));
386
387 if (dsp->dsa_pending_op != PENDING_NONE) {
388 if (dump_bytes(dsp, dsp->dsa_drr,
389 sizeof (dmu_replay_record_t)) != 0)
390 return (SET_ERROR(EINTR));
391 dsp->dsa_pending_op = PENDING_NONE;
392 }
393
394 /* write an OBJECT record */
395 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
396 dsp->dsa_drr->drr_type = DRR_OBJECT;
397 drro->drr_object = object;
398 drro->drr_type = dnp->dn_type;
399 drro->drr_bonustype = dnp->dn_bonustype;
400 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
401 drro->drr_bonuslen = dnp->dn_bonuslen;
402 drro->drr_checksumtype = dnp->dn_checksum;
403 drro->drr_compress = dnp->dn_compress;
404 drro->drr_toguid = dsp->dsa_toguid;
405
406 if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
407 drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE)
408 drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE;
409
410 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
411 return (SET_ERROR(EINTR));
412
413 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
414 return (SET_ERROR(EINTR));
415
416 /* Free anything past the end of the file. */
417 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
418 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0)
419 return (SET_ERROR(EINTR));
420 if (dsp->dsa_err != 0)
421 return (SET_ERROR(EINTR));
422 return (0);
423 }
424
425 static boolean_t
426 backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp)
427 {
428 if (!BP_IS_EMBEDDED(bp))
429 return (B_FALSE);
430
431 /*
432 * Compression function must be legacy, or explicitly enabled.
433 */
434 if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS &&
435 !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4)))
436 return (B_FALSE);
437
438 /*
439 * Embed type must be explicitly enabled.
440 */
441 switch (BPE_GET_ETYPE(bp)) {
442 case BP_EMBEDDED_TYPE_DATA:
443 if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)
444 return (B_TRUE);
445 break;
446 default:
447 return (B_FALSE);
448 }
449 return (B_FALSE);
450 }
451
452 #define BP_SPAN(dnp, level) \
453 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
454 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
455
456 /* ARGSUSED */
457 static int
458 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
459 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
460 {
461 dmu_sendarg_t *dsp = arg;
462 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
463 int err = 0;
464
465 if (issig(JUSTLOOKING) && issig(FORREAL))
466 return (SET_ERROR(EINTR));
467
468 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
469 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
470 return (0);
471 } else if (zb->zb_level == ZB_ZIL_LEVEL) {
472 /*
473 * If we are sending a non-snapshot (which is allowed on
474 * read-only pools), it may have a ZIL, which must be ignored.
475 */
476 return (0);
477 } else if (BP_IS_HOLE(bp) &&
478 zb->zb_object == DMU_META_DNODE_OBJECT) {
479 uint64_t span = BP_SPAN(dnp, zb->zb_level);
480 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
481 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
482 } else if (BP_IS_HOLE(bp)) {
483 uint64_t span = BP_SPAN(dnp, zb->zb_level);
484 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
485 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
486 return (0);
487 } else if (type == DMU_OT_DNODE) {
488 dnode_phys_t *blk;
489 int i;
490 int blksz = BP_GET_LSIZE(bp);
491 arc_flags_t aflags = ARC_FLAG_WAIT;
492 arc_buf_t *abuf;
493
494 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
495 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
496 &aflags, zb) != 0)
497 return (SET_ERROR(EIO));
498
499 blk = abuf->b_data;
500 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
501 uint64_t dnobj = (zb->zb_blkid <<
502 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
503 err = dump_dnode(dsp, dnobj, blk+i);
504 if (err != 0)
505 break;
506 }
507 (void) arc_buf_remove_ref(abuf, &abuf);
508 } else if (type == DMU_OT_SA) {
509 arc_flags_t aflags = ARC_FLAG_WAIT;
510 arc_buf_t *abuf;
511 int blksz = BP_GET_LSIZE(bp);
512
513 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
514 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
515 &aflags, zb) != 0)
516 return (SET_ERROR(EIO));
517
518 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
519 (void) arc_buf_remove_ref(abuf, &abuf);
520 } else if (backup_do_embed(dsp, bp)) {
521 /* it's an embedded level-0 block of a regular object */
522 int blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
523 err = dump_write_embedded(dsp, zb->zb_object,
524 zb->zb_blkid * blksz, blksz, bp);
525 } else { /* it's a level-0 block of a regular object */
526 uint64_t offset;
527 arc_flags_t aflags = ARC_FLAG_WAIT;
528 arc_buf_t *abuf;
529 int blksz = BP_GET_LSIZE(bp);
530
531 ASSERT3U(blksz, ==, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
532 ASSERT0(zb->zb_level);
533 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
534 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
535 &aflags, zb) != 0) {
536 if (zfs_send_corrupt_data) {
537 uint64_t *ptr;
538 /* Send a block filled with 0x"zfs badd bloc" */
539 abuf = arc_buf_alloc(spa, blksz, &abuf,
540 ARC_BUFC_DATA);
541 for (ptr = abuf->b_data;
542 (char *)ptr < (char *)abuf->b_data + blksz;
543 ptr++)
544 *ptr = 0x2f5baddb10cULL;
545 } else {
546 return (SET_ERROR(EIO));
547 }
548 }
549
550 offset = zb->zb_blkid * blksz;
551
552 if (!(dsp->dsa_featureflags &
553 DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
554 blksz > SPA_OLD_MAXBLOCKSIZE) {
555 char *buf = abuf->b_data;
556 while (blksz > 0 && err == 0) {
557 int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE);
558 err = dump_write(dsp, type, zb->zb_object,
559 offset, n, NULL, buf);
560 offset += n;
561 buf += n;
562 blksz -= n;
563 }
564 } else {
565 err = dump_write(dsp, type, zb->zb_object,
566 offset, blksz, bp, abuf->b_data);
567 }
568 (void) arc_buf_remove_ref(abuf, &abuf);
569 }
570
571 ASSERT(err == 0 || err == EINTR);
572 return (err);
573 }
574
575 /*
576 * Releases dp using the specified tag.
577 */
578 static int
579 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
580 zfs_bookmark_phys_t *fromzb, boolean_t is_clone, boolean_t embedok,
581 boolean_t large_block_ok, int outfd, vnode_t *vp, offset_t *off)
582 {
583 objset_t *os;
584 dmu_replay_record_t *drr;
585 dmu_sendarg_t *dsp;
586 int err;
587 uint64_t fromtxg = 0;
588 uint64_t featureflags = 0;
589
590 err = dmu_objset_from_ds(ds, &os);
591 if (err != 0) {
592 dsl_pool_rele(dp, tag);
593 return (err);
594 }
595
596 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
597 drr->drr_type = DRR_BEGIN;
598 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
599 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
600 DMU_SUBSTREAM);
601
602 #ifdef _KERNEL
603 if (dmu_objset_type(os) == DMU_OST_ZFS) {
604 uint64_t version;
605 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
606 kmem_free(drr, sizeof (dmu_replay_record_t));
607 dsl_pool_rele(dp, tag);
608 return (SET_ERROR(EINVAL));
609 }
610 if (version >= ZPL_VERSION_SA) {
611 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
612 }
613 }
614 #endif
615
616 if (large_block_ok && ds->ds_large_blocks)
617 featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS;
618 if (embedok &&
619 spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) {
620 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA;
621 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
622 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA_LZ4;
623 } else {
624 embedok = B_FALSE;
625 }
626
627 DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo,
628 featureflags);
629
630 drr->drr_u.drr_begin.drr_creation_time =
631 dsl_dataset_phys(ds)->ds_creation_time;
632 drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
633 if (is_clone)
634 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
635 drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(ds)->ds_guid;
636 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
637 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
638
639 if (fromzb != NULL) {
640 drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid;
641 fromtxg = fromzb->zbm_creation_txg;
642 }
643 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
644 if (!ds->ds_is_snapshot) {
645 (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--",
646 sizeof (drr->drr_u.drr_begin.drr_toname));
647 }
648
649 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
650
651 dsp->dsa_drr = drr;
652 dsp->dsa_vp = vp;
653 dsp->dsa_outfd = outfd;
654 dsp->dsa_proc = curproc;
655 dsp->dsa_os = os;
656 dsp->dsa_off = off;
657 dsp->dsa_toguid = dsl_dataset_phys(ds)->ds_guid;
658 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
659 dsp->dsa_pending_op = PENDING_NONE;
660 dsp->dsa_incremental = (fromzb != NULL);
661 dsp->dsa_featureflags = featureflags;
662
663 mutex_enter(&ds->ds_sendstream_lock);
664 list_insert_head(&ds->ds_sendstreams, dsp);
665 mutex_exit(&ds->ds_sendstream_lock);
666
667 dsl_dataset_long_hold(ds, FTAG);
668 dsl_pool_rele(dp, tag);
669
670 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
671 err = dsp->dsa_err;
672 goto out;
673 }
674
675 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
676 backup_cb, dsp);
677
678 if (dsp->dsa_pending_op != PENDING_NONE)
679 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
680 err = SET_ERROR(EINTR);
681
682 if (err != 0) {
683 if (err == EINTR && dsp->dsa_err != 0)
684 err = dsp->dsa_err;
685 goto out;
686 }
687
688 bzero(drr, sizeof (dmu_replay_record_t));
689 drr->drr_type = DRR_END;
690 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
691 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
692
693 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
694 err = dsp->dsa_err;
695 goto out;
696 }
697
698 out:
699 mutex_enter(&ds->ds_sendstream_lock);
700 list_remove(&ds->ds_sendstreams, dsp);
701 mutex_exit(&ds->ds_sendstream_lock);
702
703 kmem_free(drr, sizeof (dmu_replay_record_t));
704 kmem_free(dsp, sizeof (dmu_sendarg_t));
705
706 dsl_dataset_long_rele(ds, FTAG);
707
708 return (err);
709 }
710
711 int
712 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
713 boolean_t embedok, boolean_t large_block_ok,
714 int outfd, vnode_t *vp, offset_t *off)
715 {
716 dsl_pool_t *dp;
717 dsl_dataset_t *ds;
718 dsl_dataset_t *fromds = NULL;
719 int err;
720
721 err = dsl_pool_hold(pool, FTAG, &dp);
722 if (err != 0)
723 return (err);
724
725 err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
726 if (err != 0) {
727 dsl_pool_rele(dp, FTAG);
728 return (err);
729 }
730
731 if (fromsnap != 0) {
732 zfs_bookmark_phys_t zb;
733 boolean_t is_clone;
734
735 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
736 if (err != 0) {
737 dsl_dataset_rele(ds, FTAG);
738 dsl_pool_rele(dp, FTAG);
739 return (err);
740 }
741 if (!dsl_dataset_is_before(ds, fromds, 0))
742 err = SET_ERROR(EXDEV);
743 zb.zbm_creation_time =
744 dsl_dataset_phys(fromds)->ds_creation_time;
745 zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg;
746 zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid;
747 is_clone = (fromds->ds_dir != ds->ds_dir);
748 dsl_dataset_rele(fromds, FTAG);
749 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
750 embedok, large_block_ok, outfd, vp, off);
751 } else {
752 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
753 embedok, large_block_ok, outfd, vp, off);
754 }
755 dsl_dataset_rele(ds, FTAG);
756 return (err);
757 }
758
759 int
760 dmu_send(const char *tosnap, const char *fromsnap,
761 boolean_t embedok, boolean_t large_block_ok,
762 int outfd, vnode_t *vp, offset_t *off)
763 {
764 dsl_pool_t *dp;
765 dsl_dataset_t *ds;
766 int err;
767 boolean_t owned = B_FALSE;
768
769 if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL)
770 return (SET_ERROR(EINVAL));
771
772 err = dsl_pool_hold(tosnap, FTAG, &dp);
773 if (err != 0)
774 return (err);
775
776 if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) {
777 /*
778 * We are sending a filesystem or volume. Ensure
779 * that it doesn't change by owning the dataset.
780 */
781 err = dsl_dataset_own(dp, tosnap, FTAG, &ds);
782 owned = B_TRUE;
783 } else {
784 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
785 }
786 if (err != 0) {
787 dsl_pool_rele(dp, FTAG);
788 return (err);
789 }
790
791 if (fromsnap != NULL) {
792 zfs_bookmark_phys_t zb;
793 boolean_t is_clone = B_FALSE;
794 int fsnamelen = strchr(tosnap, '@') - tosnap;
795
796 /*
797 * If the fromsnap is in a different filesystem, then
798 * mark the send stream as a clone.
799 */
800 if (strncmp(tosnap, fromsnap, fsnamelen) != 0 ||
801 (fromsnap[fsnamelen] != '@' &&
802 fromsnap[fsnamelen] != '#')) {
803 is_clone = B_TRUE;
804 }
805
806 if (strchr(fromsnap, '@')) {
807 dsl_dataset_t *fromds;
808 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
809 if (err == 0) {
810 if (!dsl_dataset_is_before(ds, fromds, 0))
811 err = SET_ERROR(EXDEV);
812 zb.zbm_creation_time =
813 dsl_dataset_phys(fromds)->ds_creation_time;
814 zb.zbm_creation_txg =
815 dsl_dataset_phys(fromds)->ds_creation_txg;
816 zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid;
817 is_clone = (ds->ds_dir != fromds->ds_dir);
818 dsl_dataset_rele(fromds, FTAG);
819 }
820 } else {
821 err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb);
822 }
823 if (err != 0) {
824 dsl_dataset_rele(ds, FTAG);
825 dsl_pool_rele(dp, FTAG);
826 return (err);
827 }
828 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
829 embedok, large_block_ok, outfd, vp, off);
830 } else {
831 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
832 embedok, large_block_ok, outfd, vp, off);
833 }
834 if (owned)
835 dsl_dataset_disown(ds, FTAG);
836 else
837 dsl_dataset_rele(ds, FTAG);
838 return (err);
839 }
840
841 static int
842 dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t size,
843 uint64_t *sizep)
844 {
845 int err;
846 /*
847 * Assume that space (both on-disk and in-stream) is dominated by
848 * data. We will adjust for indirect blocks and the copies property,
849 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
850 */
851
852 /*
853 * Subtract out approximate space used by indirect blocks.
854 * Assume most space is used by data blocks (non-indirect, non-dnode).
855 * Assume all blocks are recordsize. Assume ditto blocks and
856 * internal fragmentation counter out compression.
857 *
858 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
859 * block, which we observe in practice.
860 */
861 uint64_t recordsize;
862 err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
863 if (err != 0)
864 return (err);
865 size -= size / recordsize * sizeof (blkptr_t);
866
867 /* Add in the space for the record associated with each block. */
868 size += size / recordsize * sizeof (dmu_replay_record_t);
869
870 *sizep = size;
871
872 return (0);
873 }
874
875 int
876 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
877 {
878 int err;
879 uint64_t size;
880
881 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
882
883 /* tosnap must be a snapshot */
884 if (!ds->ds_is_snapshot)
885 return (SET_ERROR(EINVAL));
886
887 /* fromsnap, if provided, must be a snapshot */
888 if (fromds != NULL && !fromds->ds_is_snapshot)
889 return (SET_ERROR(EINVAL));
890
891 /*
892 * fromsnap must be an earlier snapshot from the same fs as tosnap,
893 * or the origin's fs.
894 */
895 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
896 return (SET_ERROR(EXDEV));
897
898 /* Get uncompressed size estimate of changed data. */
899 if (fromds == NULL) {
900 size = dsl_dataset_phys(ds)->ds_uncompressed_bytes;
901 } else {
902 uint64_t used, comp;
903 err = dsl_dataset_space_written(fromds, ds,
904 &used, &comp, &size);
905 if (err != 0)
906 return (err);
907 }
908
909 err = dmu_adjust_send_estimate_for_indirects(ds, size, sizep);
910 return (err);
911 }
912
913 /*
914 * Simple callback used to traverse the blocks of a snapshot and sum their
915 * uncompressed size
916 */
917 /* ARGSUSED */
918 static int
919 dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
920 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
921 {
922 uint64_t *spaceptr = arg;
923 if (bp != NULL && !BP_IS_HOLE(bp)) {
924 *spaceptr += BP_GET_UCSIZE(bp);
925 }
926 return (0);
927 }
928
929 /*
930 * Given a desination snapshot and a TXG, calculate the approximate size of a
931 * send stream sent from that TXG. from_txg may be zero, indicating that the
932 * whole snapshot will be sent.
933 */
934 int
935 dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg,
936 uint64_t *sizep)
937 {
938 int err;
939 uint64_t size = 0;
940
941 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
942
943 /* tosnap must be a snapshot */
944 if (!dsl_dataset_is_snapshot(ds))
945 return (SET_ERROR(EINVAL));
946
947 /* verify that from_txg is before the provided snapshot was taken */
948 if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) {
949 return (SET_ERROR(EXDEV));
950 }
951 /*
952 * traverse the blocks of the snapshot with birth times after
953 * from_txg, summing their uncompressed size
954 */
955 err = traverse_dataset(ds, from_txg, TRAVERSE_POST,
956 dmu_calculate_send_traversal, &size);
957 if (err)
958 return (err);
959
960 err = dmu_adjust_send_estimate_for_indirects(ds, size, sizep);
961 return (err);
962 }
963
964 typedef struct dmu_recv_begin_arg {
965 const char *drba_origin;
966 dmu_recv_cookie_t *drba_cookie;
967 cred_t *drba_cred;
968 uint64_t drba_snapobj;
969 } dmu_recv_begin_arg_t;
970
971 static int
972 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
973 uint64_t fromguid)
974 {
975 uint64_t val;
976 int error;
977 dsl_pool_t *dp = ds->ds_dir->dd_pool;
978
979 /* temporary clone name must not exist */
980 error = zap_lookup(dp->dp_meta_objset,
981 dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name,
982 8, 1, &val);
983 if (error != ENOENT)
984 return (error == 0 ? EBUSY : error);
985
986 /* new snapshot name must not exist */
987 error = zap_lookup(dp->dp_meta_objset,
988 dsl_dataset_phys(ds)->ds_snapnames_zapobj,
989 drba->drba_cookie->drc_tosnap, 8, 1, &val);
990 if (error != ENOENT)
991 return (error == 0 ? EEXIST : error);
992
993 /*
994 * Check snapshot limit before receiving. We'll recheck again at the
995 * end, but might as well abort before receiving if we're already over
996 * the limit.
997 *
998 * Note that we do not check the file system limit with
999 * dsl_dir_fscount_check because the temporary %clones don't count
1000 * against that limit.
1001 */
1002 error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT,
1003 NULL, drba->drba_cred);
1004 if (error != 0)
1005 return (error);
1006
1007 if (fromguid != 0) {
1008 dsl_dataset_t *snap;
1009 uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1010
1011 /* Find snapshot in this dir that matches fromguid. */
1012 while (obj != 0) {
1013 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1014 &snap);
1015 if (error != 0)
1016 return (SET_ERROR(ENODEV));
1017 if (snap->ds_dir != ds->ds_dir) {
1018 dsl_dataset_rele(snap, FTAG);
1019 return (SET_ERROR(ENODEV));
1020 }
1021 if (dsl_dataset_phys(snap)->ds_guid == fromguid)
1022 break;
1023 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
1024 dsl_dataset_rele(snap, FTAG);
1025 }
1026 if (obj == 0)
1027 return (SET_ERROR(ENODEV));
1028
1029 if (drba->drba_cookie->drc_force) {
1030 drba->drba_snapobj = obj;
1031 } else {
1032 /*
1033 * If we are not forcing, there must be no
1034 * changes since fromsnap.
1035 */
1036 if (dsl_dataset_modified_since_snap(ds, snap)) {
1037 dsl_dataset_rele(snap, FTAG);
1038 return (SET_ERROR(ETXTBSY));
1039 }
1040 drba->drba_snapobj = ds->ds_prev->ds_object;
1041 }
1042
1043 dsl_dataset_rele(snap, FTAG);
1044 } else {
1045 /* if full, then must be forced */
1046 if (!drba->drba_cookie->drc_force)
1047 return (SET_ERROR(EEXIST));
1048 /* start from $ORIGIN@$ORIGIN, if supported */
1049 drba->drba_snapobj = dp->dp_origin_snap != NULL ?
1050 dp->dp_origin_snap->ds_object : 0;
1051 }
1052
1053 return (0);
1054
1055 }
1056
1057 static int
1058 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
1059 {
1060 dmu_recv_begin_arg_t *drba = arg;
1061 dsl_pool_t *dp = dmu_tx_pool(tx);
1062 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1063 uint64_t fromguid = drrb->drr_fromguid;
1064 int flags = drrb->drr_flags;
1065 int error;
1066 uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
1067 dsl_dataset_t *ds;
1068 const char *tofs = drba->drba_cookie->drc_tofs;
1069
1070 /* already checked */
1071 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
1072
1073 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
1074 DMU_COMPOUNDSTREAM ||
1075 drrb->drr_type >= DMU_OST_NUMTYPES ||
1076 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
1077 return (SET_ERROR(EINVAL));
1078
1079 /* Verify pool version supports SA if SA_SPILL feature set */
1080 if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
1081 spa_version(dp->dp_spa) < SPA_VERSION_SA)
1082 return (SET_ERROR(ENOTSUP));
1083
1084 /*
1085 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1086 * record to a plan WRITE record, so the pool must have the
1087 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1088 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1089 */
1090 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) &&
1091 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA))
1092 return (SET_ERROR(ENOTSUP));
1093 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) &&
1094 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
1095 return (SET_ERROR(ENOTSUP));
1096
1097 /*
1098 * The receiving code doesn't know how to translate large blocks
1099 * to smaller ones, so the pool must have the LARGE_BLOCKS
1100 * feature enabled if the stream has LARGE_BLOCKS.
1101 */
1102 if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
1103 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
1104 return (SET_ERROR(ENOTSUP));
1105
1106 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1107 if (error == 0) {
1108 /* target fs already exists; recv into temp clone */
1109
1110 /* Can't recv a clone into an existing fs */
1111 if (flags & DRR_FLAG_CLONE) {
1112 dsl_dataset_rele(ds, FTAG);
1113 return (SET_ERROR(EINVAL));
1114 }
1115
1116 error = recv_begin_check_existing_impl(drba, ds, fromguid);
1117 dsl_dataset_rele(ds, FTAG);
1118 } else if (error == ENOENT) {
1119 /* target fs does not exist; must be a full backup or clone */
1120 char buf[MAXNAMELEN];
1121
1122 /*
1123 * If it's a non-clone incremental, we are missing the
1124 * target fs, so fail the recv.
1125 */
1126 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
1127 return (SET_ERROR(ENOENT));
1128
1129 /* Open the parent of tofs */
1130 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
1131 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
1132 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
1133 if (error != 0)
1134 return (error);
1135
1136 /*
1137 * Check filesystem and snapshot limits before receiving. We'll
1138 * recheck snapshot limits again at the end (we create the
1139 * filesystems and increment those counts during begin_sync).
1140 */
1141 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1142 ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred);
1143 if (error != 0) {
1144 dsl_dataset_rele(ds, FTAG);
1145 return (error);
1146 }
1147
1148 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1149 ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred);
1150 if (error != 0) {
1151 dsl_dataset_rele(ds, FTAG);
1152 return (error);
1153 }
1154
1155 if (drba->drba_origin != NULL) {
1156 dsl_dataset_t *origin;
1157 error = dsl_dataset_hold(dp, drba->drba_origin,
1158 FTAG, &origin);
1159 if (error != 0) {
1160 dsl_dataset_rele(ds, FTAG);
1161 return (error);
1162 }
1163 if (!origin->ds_is_snapshot) {
1164 dsl_dataset_rele(origin, FTAG);
1165 dsl_dataset_rele(ds, FTAG);
1166 return (SET_ERROR(EINVAL));
1167 }
1168 if (dsl_dataset_phys(origin)->ds_guid != fromguid) {
1169 dsl_dataset_rele(origin, FTAG);
1170 dsl_dataset_rele(ds, FTAG);
1171 return (SET_ERROR(ENODEV));
1172 }
1173 dsl_dataset_rele(origin, FTAG);
1174 }
1175 dsl_dataset_rele(ds, FTAG);
1176 error = 0;
1177 }
1178 return (error);
1179 }
1180
1181 static void
1182 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
1183 {
1184 dmu_recv_begin_arg_t *drba = arg;
1185 dsl_pool_t *dp = dmu_tx_pool(tx);
1186 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1187 const char *tofs = drba->drba_cookie->drc_tofs;
1188 dsl_dataset_t *ds, *newds;
1189 uint64_t dsobj;
1190 int error;
1191 uint64_t crflags;
1192
1193 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
1194 DS_FLAG_CI_DATASET : 0;
1195
1196 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1197 if (error == 0) {
1198 /* create temporary clone */
1199 dsl_dataset_t *snap = NULL;
1200 if (drba->drba_snapobj != 0) {
1201 VERIFY0(dsl_dataset_hold_obj(dp,
1202 drba->drba_snapobj, FTAG, &snap));
1203 }
1204 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
1205 snap, crflags, drba->drba_cred, tx);
1206 dsl_dataset_rele(snap, FTAG);
1207 dsl_dataset_rele(ds, FTAG);
1208 } else {
1209 dsl_dir_t *dd;
1210 const char *tail;
1211 dsl_dataset_t *origin = NULL;
1212
1213 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
1214
1215 if (drba->drba_origin != NULL) {
1216 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
1217 FTAG, &origin));
1218 }
1219
1220 /* Create new dataset. */
1221 dsobj = dsl_dataset_create_sync(dd,
1222 strrchr(tofs, '/') + 1,
1223 origin, crflags, drba->drba_cred, tx);
1224 if (origin != NULL)
1225 dsl_dataset_rele(origin, FTAG);
1226 dsl_dir_rele(dd, FTAG);
1227 drba->drba_cookie->drc_newfs = B_TRUE;
1228 }
1229 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
1230
1231 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
1232 DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
1233 !newds->ds_large_blocks) {
1234 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1235 newds->ds_large_blocks = B_TRUE;
1236 }
1237
1238 dmu_buf_will_dirty(newds->ds_dbuf, tx);
1239 dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT;
1240
1241 /*
1242 * If we actually created a non-clone, we need to create the
1243 * objset in our new dataset.
1244 */
1245 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
1246 (void) dmu_objset_create_impl(dp->dp_spa,
1247 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1248 }
1249
1250 drba->drba_cookie->drc_ds = newds;
1251
1252 spa_history_log_internal_ds(newds, "receive", tx, "");
1253 }
1254
1255 /*
1256 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1257 * succeeds; otherwise we will leak the holds on the datasets.
1258 */
1259 int
1260 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1261 boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1262 {
1263 dmu_recv_begin_arg_t drba = { 0 };
1264 dmu_replay_record_t *drr;
1265
1266 bzero(drc, sizeof (dmu_recv_cookie_t));
1267 drc->drc_drrb = drrb;
1268 drc->drc_tosnap = tosnap;
1269 drc->drc_tofs = tofs;
1270 drc->drc_force = force;
1271 drc->drc_cred = CRED();
1272
1273 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1274 drc->drc_byteswap = B_TRUE;
1275 else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1276 return (SET_ERROR(EINVAL));
1277
1278 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1279 drr->drr_type = DRR_BEGIN;
1280 drr->drr_u.drr_begin = *drc->drc_drrb;
1281 if (drc->drc_byteswap) {
1282 fletcher_4_incremental_byteswap(drr,
1283 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1284 } else {
1285 fletcher_4_incremental_native(drr,
1286 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1287 }
1288 kmem_free(drr, sizeof (dmu_replay_record_t));
1289
1290 if (drc->drc_byteswap) {
1291 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1292 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1293 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1294 drrb->drr_type = BSWAP_32(drrb->drr_type);
1295 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1296 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1297 }
1298
1299 drba.drba_origin = origin;
1300 drba.drba_cookie = drc;
1301 drba.drba_cred = CRED();
1302
1303 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1304 &drba, 5, ZFS_SPACE_CHECK_NORMAL));
1305 }
1306
1307 struct restorearg {
1308 int err;
1309 boolean_t byteswap;
1310 vnode_t *vp;
1311 char *buf;
1312 uint64_t voff;
1313 int bufsize; /* amount of memory allocated for buf */
1314 zio_cksum_t cksum;
1315 avl_tree_t *guid_to_ds_map;
1316 };
1317
1318 typedef struct guid_map_entry {
1319 uint64_t guid;
1320 dsl_dataset_t *gme_ds;
1321 avl_node_t avlnode;
1322 } guid_map_entry_t;
1323
1324 static int
1325 guid_compare(const void *arg1, const void *arg2)
1326 {
1327 const guid_map_entry_t *gmep1 = arg1;
1328 const guid_map_entry_t *gmep2 = arg2;
1329
1330 if (gmep1->guid < gmep2->guid)
1331 return (-1);
1332 else if (gmep1->guid > gmep2->guid)
1333 return (1);
1334 return (0);
1335 }
1336
1337 static void
1338 free_guid_map_onexit(void *arg)
1339 {
1340 avl_tree_t *ca = arg;
1341 void *cookie = NULL;
1342 guid_map_entry_t *gmep;
1343
1344 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1345 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1346 dsl_dataset_rele(gmep->gme_ds, gmep);
1347 kmem_free(gmep, sizeof (guid_map_entry_t));
1348 }
1349 avl_destroy(ca);
1350 kmem_free(ca, sizeof (avl_tree_t));
1351 }
1352
1353 static void *
1354 restore_read(struct restorearg *ra, int len, char *buf)
1355 {
1356 int done = 0;
1357
1358 if (buf == NULL)
1359 buf = ra->buf;
1360
1361 /* some things will require 8-byte alignment, so everything must */
1362 ASSERT0(len % 8);
1363 ASSERT3U(len, <=, ra->bufsize);
1364
1365 while (done < len) {
1366 ssize_t resid;
1367
1368 ra->err = vn_rdwr(UIO_READ, ra->vp,
1369 buf + done, len - done,
1370 ra->voff, UIO_SYSSPACE, FAPPEND,
1371 RLIM64_INFINITY, CRED(), &resid);
1372
1373 if (resid == len - done)
1374 ra->err = SET_ERROR(EINVAL);
1375 ra->voff += len - done - resid;
1376 done = len - resid;
1377 if (ra->err != 0)
1378 return (NULL);
1379 }
1380
1381 ASSERT3U(done, ==, len);
1382 if (ra->byteswap)
1383 fletcher_4_incremental_byteswap(buf, len, &ra->cksum);
1384 else
1385 fletcher_4_incremental_native(buf, len, &ra->cksum);
1386 return (buf);
1387 }
1388
1389 noinline static void
1390 backup_byteswap(dmu_replay_record_t *drr)
1391 {
1392 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1393 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1394 drr->drr_type = BSWAP_32(drr->drr_type);
1395 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1396 switch (drr->drr_type) {
1397 case DRR_BEGIN:
1398 DO64(drr_begin.drr_magic);
1399 DO64(drr_begin.drr_versioninfo);
1400 DO64(drr_begin.drr_creation_time);
1401 DO32(drr_begin.drr_type);
1402 DO32(drr_begin.drr_flags);
1403 DO64(drr_begin.drr_toguid);
1404 DO64(drr_begin.drr_fromguid);
1405 break;
1406 case DRR_OBJECT:
1407 DO64(drr_object.drr_object);
1408 DO32(drr_object.drr_type);
1409 DO32(drr_object.drr_bonustype);
1410 DO32(drr_object.drr_blksz);
1411 DO32(drr_object.drr_bonuslen);
1412 DO64(drr_object.drr_toguid);
1413 break;
1414 case DRR_FREEOBJECTS:
1415 DO64(drr_freeobjects.drr_firstobj);
1416 DO64(drr_freeobjects.drr_numobjs);
1417 DO64(drr_freeobjects.drr_toguid);
1418 break;
1419 case DRR_WRITE:
1420 DO64(drr_write.drr_object);
1421 DO32(drr_write.drr_type);
1422 DO64(drr_write.drr_offset);
1423 DO64(drr_write.drr_length);
1424 DO64(drr_write.drr_toguid);
1425 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1426 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1427 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1428 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1429 DO64(drr_write.drr_key.ddk_prop);
1430 break;
1431 case DRR_WRITE_BYREF:
1432 DO64(drr_write_byref.drr_object);
1433 DO64(drr_write_byref.drr_offset);
1434 DO64(drr_write_byref.drr_length);
1435 DO64(drr_write_byref.drr_toguid);
1436 DO64(drr_write_byref.drr_refguid);
1437 DO64(drr_write_byref.drr_refobject);
1438 DO64(drr_write_byref.drr_refoffset);
1439 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1440 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1441 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1442 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1443 DO64(drr_write_byref.drr_key.ddk_prop);
1444 break;
1445 case DRR_WRITE_EMBEDDED:
1446 DO64(drr_write_embedded.drr_object);
1447 DO64(drr_write_embedded.drr_offset);
1448 DO64(drr_write_embedded.drr_length);
1449 DO64(drr_write_embedded.drr_toguid);
1450 DO32(drr_write_embedded.drr_lsize);
1451 DO32(drr_write_embedded.drr_psize);
1452 break;
1453 case DRR_FREE:
1454 DO64(drr_free.drr_object);
1455 DO64(drr_free.drr_offset);
1456 DO64(drr_free.drr_length);
1457 DO64(drr_free.drr_toguid);
1458 break;
1459 case DRR_SPILL:
1460 DO64(drr_spill.drr_object);
1461 DO64(drr_spill.drr_length);
1462 DO64(drr_spill.drr_toguid);
1463 break;
1464 case DRR_END:
1465 DO64(drr_end.drr_checksum.zc_word[0]);
1466 DO64(drr_end.drr_checksum.zc_word[1]);
1467 DO64(drr_end.drr_checksum.zc_word[2]);
1468 DO64(drr_end.drr_checksum.zc_word[3]);
1469 DO64(drr_end.drr_toguid);
1470 break;
1471 default:
1472 break;
1473 }
1474 #undef DO64
1475 #undef DO32
1476 }
1477
1478 static inline uint8_t
1479 deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size)
1480 {
1481 if (bonus_type == DMU_OT_SA) {
1482 return (1);
1483 } else {
1484 return (1 +
1485 ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT));
1486 }
1487 }
1488
1489 noinline static int
1490 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1491 {
1492 dmu_object_info_t doi;
1493 dmu_tx_t *tx;
1494 void *data = NULL;
1495 uint64_t object;
1496 int err;
1497
1498 if (drro->drr_type == DMU_OT_NONE ||
1499 !DMU_OT_IS_VALID(drro->drr_type) ||
1500 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1501 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1502 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1503 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1504 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1505 drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(os)) ||
1506 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1507 return (SET_ERROR(EINVAL));
1508 }
1509
1510 err = dmu_object_info(os, drro->drr_object, &doi);
1511
1512 if (err != 0 && err != ENOENT)
1513 return (SET_ERROR(EINVAL));
1514 object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT;
1515
1516 if (drro->drr_bonuslen) {
1517 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8), NULL);
1518 if (ra->err != 0)
1519 return (ra->err);
1520 }
1521
1522 /*
1523 * If we are losing blkptrs or changing the block size this must
1524 * be a new file instance. We must clear out the previous file
1525 * contents before we can change this type of metadata in the dnode.
1526 */
1527 if (err == 0) {
1528 int nblkptr;
1529
1530 nblkptr = deduce_nblkptr(drro->drr_bonustype,
1531 drro->drr_bonuslen);
1532
1533 if (drro->drr_blksz != doi.doi_data_block_size ||
1534 nblkptr < doi.doi_nblkptr) {
1535 err = dmu_free_long_range(os, drro->drr_object,
1536 0, DMU_OBJECT_END);
1537 if (err != 0)
1538 return (SET_ERROR(EINVAL));
1539 }
1540 }
1541
1542 tx = dmu_tx_create(os);
1543 dmu_tx_hold_bonus(tx, object);
1544 err = dmu_tx_assign(tx, TXG_WAIT);
1545 if (err != 0) {
1546 dmu_tx_abort(tx);
1547 return (err);
1548 }
1549
1550 if (object == DMU_NEW_OBJECT) {
1551 /* currently free, want to be allocated */
1552 err = dmu_object_claim(os, drro->drr_object,
1553 drro->drr_type, drro->drr_blksz,
1554 drro->drr_bonustype, drro->drr_bonuslen, tx);
1555 } else if (drro->drr_type != doi.doi_type ||
1556 drro->drr_blksz != doi.doi_data_block_size ||
1557 drro->drr_bonustype != doi.doi_bonus_type ||
1558 drro->drr_bonuslen != doi.doi_bonus_size) {
1559 /* currently allocated, but with different properties */
1560 err = dmu_object_reclaim(os, drro->drr_object,
1561 drro->drr_type, drro->drr_blksz,
1562 drro->drr_bonustype, drro->drr_bonuslen, tx);
1563 }
1564 if (err != 0) {
1565 dmu_tx_commit(tx);
1566 return (SET_ERROR(EINVAL));
1567 }
1568
1569 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1570 tx);
1571 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1572
1573 if (data != NULL) {
1574 dmu_buf_t *db;
1575
1576 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1577 dmu_buf_will_dirty(db, tx);
1578
1579 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1580 bcopy(data, db->db_data, drro->drr_bonuslen);
1581 if (ra->byteswap) {
1582 dmu_object_byteswap_t byteswap =
1583 DMU_OT_BYTESWAP(drro->drr_bonustype);
1584 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1585 drro->drr_bonuslen);
1586 }
1587 dmu_buf_rele(db, FTAG);
1588 }
1589 dmu_tx_commit(tx);
1590 return (0);
1591 }
1592
1593 /* ARGSUSED */
1594 noinline static int
1595 restore_freeobjects(struct restorearg *ra, objset_t *os,
1596 struct drr_freeobjects *drrfo)
1597 {
1598 uint64_t obj;
1599
1600 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1601 return (SET_ERROR(EINVAL));
1602
1603 for (obj = drrfo->drr_firstobj;
1604 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1605 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1606 int err;
1607
1608 if (dmu_object_info(os, obj, NULL) != 0)
1609 continue;
1610
1611 err = dmu_free_long_object(os, obj);
1612 if (err != 0)
1613 return (err);
1614 }
1615 return (0);
1616 }
1617
1618 noinline static int
1619 restore_write(struct restorearg *ra, objset_t *os,
1620 struct drr_write *drrw)
1621 {
1622 dmu_tx_t *tx;
1623 dmu_buf_t *bonus;
1624 arc_buf_t *abuf;
1625 void *data;
1626 int err;
1627
1628 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1629 !DMU_OT_IS_VALID(drrw->drr_type))
1630 return (SET_ERROR(EINVAL));
1631
1632 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1633 return (SET_ERROR(EINVAL));
1634
1635 if (dmu_bonus_hold(os, drrw->drr_object, FTAG, &bonus) != 0)
1636 return (SET_ERROR(EINVAL));
1637
1638 abuf = dmu_request_arcbuf(bonus, drrw->drr_length);
1639
1640 data = restore_read(ra, drrw->drr_length, abuf->b_data);
1641 if (data == NULL) {
1642 dmu_return_arcbuf(abuf);
1643 dmu_buf_rele(bonus, FTAG);
1644 return (ra->err);
1645 }
1646
1647 tx = dmu_tx_create(os);
1648
1649 dmu_tx_hold_write(tx, drrw->drr_object,
1650 drrw->drr_offset, drrw->drr_length);
1651 err = dmu_tx_assign(tx, TXG_WAIT);
1652 if (err != 0) {
1653 dmu_return_arcbuf(abuf);
1654 dmu_buf_rele(bonus, FTAG);
1655 dmu_tx_abort(tx);
1656 return (err);
1657 }
1658 if (ra->byteswap) {
1659 dmu_object_byteswap_t byteswap =
1660 DMU_OT_BYTESWAP(drrw->drr_type);
1661 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1662 }
1663 dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx);
1664 dmu_tx_commit(tx);
1665 dmu_buf_rele(bonus, FTAG);
1666 return (0);
1667 }
1668
1669 /*
1670 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1671 * streams to refer to a copy of the data that is already on the
1672 * system because it came in earlier in the stream. This function
1673 * finds the earlier copy of the data, and uses that copy instead of
1674 * data from the stream to fulfill this write.
1675 */
1676 static int
1677 restore_write_byref(struct restorearg *ra, objset_t *os,
1678 struct drr_write_byref *drrwbr)
1679 {
1680 dmu_tx_t *tx;
1681 int err;
1682 guid_map_entry_t gmesrch;
1683 guid_map_entry_t *gmep;
1684 avl_index_t where;
1685 objset_t *ref_os = NULL;
1686 dmu_buf_t *dbp;
1687
1688 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1689 return (SET_ERROR(EINVAL));
1690
1691 /*
1692 * If the GUID of the referenced dataset is different from the
1693 * GUID of the target dataset, find the referenced dataset.
1694 */
1695 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1696 gmesrch.guid = drrwbr->drr_refguid;
1697 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1698 &where)) == NULL) {
1699 return (SET_ERROR(EINVAL));
1700 }
1701 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1702 return (SET_ERROR(EINVAL));
1703 } else {
1704 ref_os = os;
1705 }
1706
1707 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1708 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1709 if (err != 0)
1710 return (err);
1711
1712 tx = dmu_tx_create(os);
1713
1714 dmu_tx_hold_write(tx, drrwbr->drr_object,
1715 drrwbr->drr_offset, drrwbr->drr_length);
1716 err = dmu_tx_assign(tx, TXG_WAIT);
1717 if (err != 0) {
1718 dmu_tx_abort(tx);
1719 return (err);
1720 }
1721 dmu_write(os, drrwbr->drr_object,
1722 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1723 dmu_buf_rele(dbp, FTAG);
1724 dmu_tx_commit(tx);
1725 return (0);
1726 }
1727
1728 static int
1729 restore_write_embedded(struct restorearg *ra, objset_t *os,
1730 struct drr_write_embedded *drrwnp)
1731 {
1732 dmu_tx_t *tx;
1733 int err;
1734 void *data;
1735
1736 if (drrwnp->drr_offset + drrwnp->drr_length < drrwnp->drr_offset)
1737 return (EINVAL);
1738
1739 if (drrwnp->drr_psize > BPE_PAYLOAD_SIZE)
1740 return (EINVAL);
1741
1742 if (drrwnp->drr_etype >= NUM_BP_EMBEDDED_TYPES)
1743 return (EINVAL);
1744 if (drrwnp->drr_compression >= ZIO_COMPRESS_FUNCTIONS)
1745 return (EINVAL);
1746
1747 data = restore_read(ra, P2ROUNDUP(drrwnp->drr_psize, 8), NULL);
1748 if (data == NULL)
1749 return (ra->err);
1750
1751 tx = dmu_tx_create(os);
1752
1753 dmu_tx_hold_write(tx, drrwnp->drr_object,
1754 drrwnp->drr_offset, drrwnp->drr_length);
1755 err = dmu_tx_assign(tx, TXG_WAIT);
1756 if (err != 0) {
1757 dmu_tx_abort(tx);
1758 return (err);
1759 }
1760
1761 dmu_write_embedded(os, drrwnp->drr_object,
1762 drrwnp->drr_offset, data, drrwnp->drr_etype,
1763 drrwnp->drr_compression, drrwnp->drr_lsize, drrwnp->drr_psize,
1764 ra->byteswap ^ ZFS_HOST_BYTEORDER, tx);
1765
1766 dmu_tx_commit(tx);
1767 return (0);
1768 }
1769
1770 static int
1771 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1772 {
1773 dmu_tx_t *tx;
1774 void *data;
1775 dmu_buf_t *db, *db_spill;
1776 int err;
1777
1778 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1779 drrs->drr_length > spa_maxblocksize(dmu_objset_spa(os)))
1780 return (SET_ERROR(EINVAL));
1781
1782 data = restore_read(ra, drrs->drr_length, NULL);
1783 if (data == NULL)
1784 return (ra->err);
1785
1786 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1787 return (SET_ERROR(EINVAL));
1788
1789 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1790 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1791 dmu_buf_rele(db, FTAG);
1792 return (err);
1793 }
1794
1795 tx = dmu_tx_create(os);
1796
1797 dmu_tx_hold_spill(tx, db->db_object);
1798
1799 err = dmu_tx_assign(tx, TXG_WAIT);
1800 if (err != 0) {
1801 dmu_buf_rele(db, FTAG);
1802 dmu_buf_rele(db_spill, FTAG);
1803 dmu_tx_abort(tx);
1804 return (err);
1805 }
1806 dmu_buf_will_dirty(db_spill, tx);
1807
1808 if (db_spill->db_size < drrs->drr_length)
1809 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1810 drrs->drr_length, tx));
1811 bcopy(data, db_spill->db_data, drrs->drr_length);
1812
1813 dmu_buf_rele(db, FTAG);
1814 dmu_buf_rele(db_spill, FTAG);
1815
1816 dmu_tx_commit(tx);
1817 return (0);
1818 }
1819
1820 /* ARGSUSED */
1821 noinline static int
1822 restore_free(struct restorearg *ra, objset_t *os,
1823 struct drr_free *drrf)
1824 {
1825 int err;
1826
1827 if (drrf->drr_length != -1ULL &&
1828 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1829 return (SET_ERROR(EINVAL));
1830
1831 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1832 return (SET_ERROR(EINVAL));
1833
1834 err = dmu_free_long_range(os, drrf->drr_object,
1835 drrf->drr_offset, drrf->drr_length);
1836 return (err);
1837 }
1838
1839 /* used to destroy the drc_ds on error */
1840 static void
1841 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1842 {
1843 char name[MAXNAMELEN];
1844 dsl_dataset_name(drc->drc_ds, name);
1845 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1846 (void) dsl_destroy_head(name);
1847 }
1848
1849 /*
1850 * NB: callers *must* call dmu_recv_end() if this succeeds.
1851 */
1852 int
1853 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1854 int cleanup_fd, uint64_t *action_handlep)
1855 {
1856 struct restorearg ra = { 0 };
1857 dmu_replay_record_t *drr;
1858 objset_t *os;
1859 zio_cksum_t pcksum;
1860 int featureflags;
1861
1862 ra.byteswap = drc->drc_byteswap;
1863 ra.cksum = drc->drc_cksum;
1864 ra.vp = vp;
1865 ra.voff = *voffp;
1866 ra.bufsize = SPA_MAXBLOCKSIZE;
1867 ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
1868
1869 /* these were verified in dmu_recv_begin */
1870 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1871 DMU_SUBSTREAM);
1872 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1873
1874 /*
1875 * Open the objset we are modifying.
1876 */
1877 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1878
1879 ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT);
1880
1881 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1882
1883 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1884 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1885 minor_t minor;
1886
1887 if (cleanup_fd == -1) {
1888 ra.err = SET_ERROR(EBADF);
1889 goto out;
1890 }
1891 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1892 if (ra.err != 0) {
1893 cleanup_fd = -1;
1894 goto out;
1895 }
1896
1897 if (*action_handlep == 0) {
1898 ra.guid_to_ds_map =
1899 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1900 avl_create(ra.guid_to_ds_map, guid_compare,
1901 sizeof (guid_map_entry_t),
1902 offsetof(guid_map_entry_t, avlnode));
1903 ra.err = zfs_onexit_add_cb(minor,
1904 free_guid_map_onexit, ra.guid_to_ds_map,
1905 action_handlep);
1906 if (ra.err != 0)
1907 goto out;
1908 } else {
1909 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1910 (void **)&ra.guid_to_ds_map);
1911 if (ra.err != 0)
1912 goto out;
1913 }
1914
1915 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1916 }
1917
1918 /*
1919 * Read records and process them.
1920 */
1921 pcksum = ra.cksum;
1922 while (ra.err == 0 &&
1923 NULL != (drr = restore_read(&ra, sizeof (*drr), NULL))) {
1924 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1925 ra.err = SET_ERROR(EINTR);
1926 goto out;
1927 }
1928
1929 if (ra.byteswap)
1930 backup_byteswap(drr);
1931
1932 switch (drr->drr_type) {
1933 case DRR_OBJECT:
1934 {
1935 /*
1936 * We need to make a copy of the record header,
1937 * because restore_{object,write} may need to
1938 * restore_read(), which will invalidate drr.
1939 */
1940 struct drr_object drro = drr->drr_u.drr_object;
1941 ra.err = restore_object(&ra, os, &drro);
1942 break;
1943 }
1944 case DRR_FREEOBJECTS:
1945 {
1946 struct drr_freeobjects drrfo =
1947 drr->drr_u.drr_freeobjects;
1948 ra.err = restore_freeobjects(&ra, os, &drrfo);
1949 break;
1950 }
1951 case DRR_WRITE:
1952 {
1953 struct drr_write drrw = drr->drr_u.drr_write;
1954 ra.err = restore_write(&ra, os, &drrw);
1955 break;
1956 }
1957 case DRR_WRITE_BYREF:
1958 {
1959 struct drr_write_byref drrwbr =
1960 drr->drr_u.drr_write_byref;
1961 ra.err = restore_write_byref(&ra, os, &drrwbr);
1962 break;
1963 }
1964 case DRR_WRITE_EMBEDDED:
1965 {
1966 struct drr_write_embedded drrwe =
1967 drr->drr_u.drr_write_embedded;
1968 ra.err = restore_write_embedded(&ra, os, &drrwe);
1969 break;
1970 }
1971 case DRR_FREE:
1972 {
1973 struct drr_free drrf = drr->drr_u.drr_free;
1974 ra.err = restore_free(&ra, os, &drrf);
1975 break;
1976 }
1977 case DRR_END:
1978 {
1979 struct drr_end drre = drr->drr_u.drr_end;
1980 /*
1981 * We compare against the *previous* checksum
1982 * value, because the stored checksum is of
1983 * everything before the DRR_END record.
1984 */
1985 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1986 ra.err = SET_ERROR(ECKSUM);
1987 goto out;
1988 }
1989 case DRR_SPILL:
1990 {
1991 struct drr_spill drrs = drr->drr_u.drr_spill;
1992 ra.err = restore_spill(&ra, os, &drrs);
1993 break;
1994 }
1995 default:
1996 ra.err = SET_ERROR(EINVAL);
1997 goto out;
1998 }
1999 pcksum = ra.cksum;
2000 }
2001 ASSERT(ra.err != 0);
2002
2003 out:
2004 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
2005 zfs_onexit_fd_rele(cleanup_fd);
2006
2007 if (ra.err != 0) {
2008 /*
2009 * destroy what we created, so we don't leave it in the
2010 * inconsistent restoring state.
2011 */
2012 dmu_recv_cleanup_ds(drc);
2013 }
2014
2015 vmem_free(ra.buf, ra.bufsize);
2016 *voffp = ra.voff;
2017 return (ra.err);
2018 }
2019
2020 static int
2021 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
2022 {
2023 dmu_recv_cookie_t *drc = arg;
2024 dsl_pool_t *dp = dmu_tx_pool(tx);
2025 int error;
2026
2027 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
2028
2029 if (!drc->drc_newfs) {
2030 dsl_dataset_t *origin_head;
2031
2032 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
2033 if (error != 0)
2034 return (error);
2035 if (drc->drc_force) {
2036 /*
2037 * We will destroy any snapshots in tofs (i.e. before
2038 * origin_head) that are after the origin (which is
2039 * the snap before drc_ds, because drc_ds can not
2040 * have any snaps of its own).
2041 */
2042 uint64_t obj;
2043
2044 obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj;
2045 while (obj !=
2046 dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) {
2047 dsl_dataset_t *snap;
2048 error = dsl_dataset_hold_obj(dp, obj, FTAG,
2049 &snap);
2050 if (error != 0)
2051 break;
2052 if (snap->ds_dir != origin_head->ds_dir)
2053 error = SET_ERROR(EINVAL);
2054 if (error == 0) {
2055 error = dsl_destroy_snapshot_check_impl(
2056 snap, B_FALSE);
2057 }
2058 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
2059 dsl_dataset_rele(snap, FTAG);
2060 if (error != 0)
2061 break;
2062 }
2063 if (error != 0) {
2064 dsl_dataset_rele(origin_head, FTAG);
2065 return (error);
2066 }
2067 }
2068 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
2069 origin_head, drc->drc_force, drc->drc_owner, tx);
2070 if (error != 0) {
2071 dsl_dataset_rele(origin_head, FTAG);
2072 return (error);
2073 }
2074 error = dsl_dataset_snapshot_check_impl(origin_head,
2075 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
2076 dsl_dataset_rele(origin_head, FTAG);
2077 if (error != 0)
2078 return (error);
2079
2080 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
2081 } else {
2082 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
2083 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
2084 }
2085 return (error);
2086 }
2087
2088 static void
2089 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
2090 {
2091 dmu_recv_cookie_t *drc = arg;
2092 dsl_pool_t *dp = dmu_tx_pool(tx);
2093
2094 spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
2095 tx, "snap=%s", drc->drc_tosnap);
2096
2097 if (!drc->drc_newfs) {
2098 dsl_dataset_t *origin_head;
2099
2100 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
2101 &origin_head));
2102
2103 if (drc->drc_force) {
2104 /*
2105 * Destroy any snapshots of drc_tofs (origin_head)
2106 * after the origin (the snap before drc_ds).
2107 */
2108 uint64_t obj;
2109
2110 obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj;
2111 while (obj !=
2112 dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) {
2113 dsl_dataset_t *snap;
2114 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
2115 &snap));
2116 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
2117 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
2118 dsl_destroy_snapshot_sync_impl(snap,
2119 B_FALSE, tx);
2120 dsl_dataset_rele(snap, FTAG);
2121 }
2122 }
2123 VERIFY3P(drc->drc_ds->ds_prev, ==,
2124 origin_head->ds_prev);
2125
2126 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
2127 origin_head, tx);
2128 dsl_dataset_snapshot_sync_impl(origin_head,
2129 drc->drc_tosnap, tx);
2130
2131 /* set snapshot's creation time and guid */
2132 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
2133 dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time =
2134 drc->drc_drrb->drr_creation_time;
2135 dsl_dataset_phys(origin_head->ds_prev)->ds_guid =
2136 drc->drc_drrb->drr_toguid;
2137 dsl_dataset_phys(origin_head->ds_prev)->ds_flags &=
2138 ~DS_FLAG_INCONSISTENT;
2139
2140 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2141 dsl_dataset_phys(origin_head)->ds_flags &=
2142 ~DS_FLAG_INCONSISTENT;
2143
2144 dsl_dataset_rele(origin_head, FTAG);
2145 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
2146
2147 if (drc->drc_owner != NULL)
2148 VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
2149 } else {
2150 dsl_dataset_t *ds = drc->drc_ds;
2151
2152 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
2153
2154 /* set snapshot's creation time and guid */
2155 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2156 dsl_dataset_phys(ds->ds_prev)->ds_creation_time =
2157 drc->drc_drrb->drr_creation_time;
2158 dsl_dataset_phys(ds->ds_prev)->ds_guid =
2159 drc->drc_drrb->drr_toguid;
2160 dsl_dataset_phys(ds->ds_prev)->ds_flags &=
2161 ~DS_FLAG_INCONSISTENT;
2162
2163 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2164 dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT;
2165 }
2166 drc->drc_newsnapobj = dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj;
2167 zvol_create_minors(dp->dp_spa, drc->drc_tofs, B_TRUE);
2168 /*
2169 * Release the hold from dmu_recv_begin. This must be done before
2170 * we return to open context, so that when we free the dataset's dnode,
2171 * we can evict its bonus buffer.
2172 */
2173 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
2174 drc->drc_ds = NULL;
2175 }
2176
2177 static int
2178 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
2179 {
2180 dsl_pool_t *dp;
2181 dsl_dataset_t *snapds;
2182 guid_map_entry_t *gmep;
2183 int err;
2184
2185 ASSERT(guid_map != NULL);
2186
2187 err = dsl_pool_hold(name, FTAG, &dp);
2188 if (err != 0)
2189 return (err);
2190 gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
2191 err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
2192 if (err == 0) {
2193 gmep->guid = dsl_dataset_phys(snapds)->ds_guid;
2194 gmep->gme_ds = snapds;
2195 avl_add(guid_map, gmep);
2196 dsl_dataset_long_hold(snapds, gmep);
2197 } else {
2198 kmem_free(gmep, sizeof (*gmep));
2199 }
2200
2201 dsl_pool_rele(dp, FTAG);
2202 return (err);
2203 }
2204
2205 static int dmu_recv_end_modified_blocks = 3;
2206
2207 static int
2208 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
2209 {
2210 int error;
2211
2212 #ifdef _KERNEL
2213 char *name;
2214
2215 /*
2216 * We will be destroying the ds; make sure its origin is unmounted if
2217 * necessary.
2218 */
2219 name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2220 dsl_dataset_name(drc->drc_ds, name);
2221 zfs_destroy_unmount_origin(name);
2222 kmem_free(name, MAXNAMELEN);
2223 #endif
2224
2225 error = dsl_sync_task(drc->drc_tofs,
2226 dmu_recv_end_check, dmu_recv_end_sync, drc,
2227 dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2228
2229 if (error != 0)
2230 dmu_recv_cleanup_ds(drc);
2231 return (error);
2232 }
2233
2234 static int
2235 dmu_recv_new_end(dmu_recv_cookie_t *drc)
2236 {
2237 int error;
2238
2239 error = dsl_sync_task(drc->drc_tofs,
2240 dmu_recv_end_check, dmu_recv_end_sync, drc,
2241 dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2242
2243 if (error != 0) {
2244 dmu_recv_cleanup_ds(drc);
2245 } else if (drc->drc_guid_to_ds_map != NULL) {
2246 (void) add_ds_to_guidmap(drc->drc_tofs,
2247 drc->drc_guid_to_ds_map,
2248 drc->drc_newsnapobj);
2249 }
2250 return (error);
2251 }
2252
2253 int
2254 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
2255 {
2256 drc->drc_owner = owner;
2257
2258 if (drc->drc_newfs)
2259 return (dmu_recv_new_end(drc));
2260 else
2261 return (dmu_recv_existing_end(drc));
2262 }
2263
2264 /*
2265 * Return TRUE if this objset is currently being received into.
2266 */
2267 boolean_t
2268 dmu_objset_is_receiving(objset_t *os)
2269 {
2270 return (os->os_dsl_dataset != NULL &&
2271 os->os_dsl_dataset->ds_owner == dmu_recv_tag);
2272 }
2273
2274 #if defined(_KERNEL)
2275 module_param(zfs_send_corrupt_data, int, 0644);
2276 MODULE_PARM_DESC(zfs_send_corrupt_data, "Allow sending corrupt data");
2277 #endif
ubuntu-zesty-kernel mirror